Locality: a case study from Äiwoo
by
Giovanni Roversi
B.A., University of Oslo, 2017
M.A., University of Oslo, 2019
Submitted to the Department of Linguistics and Philosophy
in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY IN LINGUISTICS
at the
MASSACHUSETTS INSTITUTE OF TECHNOLOGY
August 2025
© 2025 Giovanni Roversi. This work is licensed under a CC BY-NC-ND 4.0 license.
The author hereby grants to MIT a nonexclusive, worldwide, irrevocable, royalty-free
license to exercise any and all rights under copyright, including to reproduce, preserve,
distribute and publicly display copies of the thesis, or release the thesis under an
open-access license.
Authored by: Giovanni Roversi
Department of Linguistics and Philosophy
August 22, 2025
Certified by: Norvin Richards
Dean’s Distinguished Professor of Linguistics
Thesis Supervisor
Accepted by: Danny Fox
Anshen-Chomsky Professor of Language & Thought
Linguistics Section head
Locality: a case study from Äiwoo
by
Giovanni Roversi
Submitted to the Department of Linguistics and Philosophy
on August 22, 2025 in partial fulfillment of the requirements for the degree of
DOCTOR OF PHILOSOPHY IN LINGUISTICS
ABSTRACT
This thesis explores issues in the syntax of Äiwoo, an understudied Oceanic language from
the Solomon Islands. This language showcases an intricate system of clausal alternations,
where several parameters vary independently from each other: verbal morphology, word
order, and possibilities of Ā-extraction (i.e., which argument(s) may or may not be ex-
tracted). As the title indicates, this thesis is set up as a case study, in a sort of “learn by
doing” fashion. I attempt to build a descriptively and explanatory adequate formal model
of the clausal alternation system in Äiwoo, within a Minimalist framework. By doing so,
I examine what a theory of grammar must look like for the model to work as intended.
Building such amodel of Äiwoo teaches us something about a number of central issues
in syntactic theory such as the locality of movement, the A/Ā-distinction, and the syntax
of Austronesian languages specifically. I show that conjoining van Urk’s (2015) theory
of the A/Ā-distinction and to the independent idea of featurally relativized probes (Béjar
2003), we predict the existence of “non-local A-movement”, that is, movement with the
binding-theoretical properties of A-movement that nonetheless does not obey strict DP-
locality, and crucially without the need of invoking the notion of “mixed A/Ā-movement”.
I show that two instances of this predicted kind of movement are attested in the Äiwoo
clause: movement to both spec,TP and spec,CP can target either the subject or – non-
locally – the object, depending on their features. I also propose that features assigned
by a probe to a goal (“goal-flagging”; Deal to appear, Clem & Deal 2024) can be further
manipulated by the syntax, being searched for by a higher probe.
Further, Äiwoo shows an interesting instantiation of the Austronesian “pivot-only” Ā-
extraction restriction, in that it comes with a series of exceptions. I argue that in Äiwoo,
this restriction is caused by an Ā-intervention effect, contra analyses of this phenomenon
in other Austronesian languages based on phasehood (Rackowski & Richards 2005, Er-
lewine & C. Lim 2023, Hsieh 2025, a.o.) or DP-intervention (Aldridge 2004, 2008, a.o.).
Notably, as soon as the highest DP in a clause does not carry Ā-features, the restriction
vanishes, thus allowing for the “exceptional” extraction of lower arguments.
Thesis Supervisor: Norvin Richards
Title: Dean’s Distinguished Professor of Linguistics
ACKNOWLEDGMENTS
We come to this place… for magic.
— Nicole Kidman, about the 8th floor
of the MIT Stata Center, probably
This dissertation is too long. Therefore, it’s only fair to balance that with an equally too
long acknowledgment section. Luckily, writing the acknowledgments is by far the most
pleasant, most entertaining, and most fun part of this whole process – if anything, disser-
tations should involveway less contentful text andwaymore acknowledgments. I imagine
that this must be very similar to how writing Christmas cards feels for those people who
do write Christmas cards to their dear and loved ones (I’ve never written one, but this
is how I think I would feel if I did). Like with Christmas cards, there’s a few obligations
(people I must thank, out of tradition/politeness/respect/diplomacy) – and for what my
word’s worth, all the people who would fall into this class would’ve still been included in
these acknowledgments even in the absence of said obligations. Aside from that, there’s
really no set boundary on how many people, or what kind of people, I’m allowed to in-
clude in my acknowledgments. And I intend to make full use of this lack of boundaries:
there’s gonna be a lot of Christmas cards. The other nice part about writing Christmas
cards/thanking people here is that in both cases, the receiver can’t say no. You can’t send
a card back (I mean, I guess technically you can, but let’s be very serious now). And if
you’ve been thanked here, even if you’re the kind of person who doesn’t usually like get-
ting compliments and thanks, sorry, tough luck: there’s literally nothing you can do about
it, the compliment is now in print and archived for eternity in some MIT database.
That said, I can now start thanking people. The sequence is going to be the traditional
one. For the academia people, you’re first. For family people, you’re after that. For my
friends, skip to the bottom of page 6. (I low-key love that I have a verbose preamble even
for my acknowledgments – absolutely not foreboding for the rest of the dissertation.)
Doing a PhD is hard1. It’s also somewhat extra hard if your PhD starts during a global
pandemic and finishes during uh, turbulent times. Not that this really has much to do with
everything else, but I just wanted to mention it so that you’d feel bad for me. However,
1 I don’t know if other jobs are harder. I haven’t tried those. Not really looking forward to finding out.
4
doing a PhD was made much, much easier by the people on my committee. There’s four
of them and any order would come with unintended implications, so let’s just neutrally
pick reverse alphabetical order; it’s as good an order as any.
Norvin Richards is first up. I’ve never really left ameetingwith Norvin in anything less
than a lovely mood – and I can’t stress this enough: that says a lot, given what goes on in
Äiwoo (and given the nature of a PhD program too). His expertise on Tagalog and Dinka
– which is so strikingly syntactically similar to Äiwoo – was invaluable. Often he would
quickly work out the predictions of what I was trying to say, and immediately advise
me on how those could be tested. Equally invaluable was his skill at spotting meaningful
parallels from the chaotic data I’d present to him to other phenomena in other languages
discussed in other literature, connections that I would never have made myself. At those
times when I was getting bogged down into the depth of the mechanical technicalities of
the framework (which has been often), Norvin helped a lot in lifting my gaze up to abstract
away from the specific implementational details and rather think about the general shape
of the analysis, and its deeper conceptual content. All of this was incredibly motivating.
David Pesetsky helped me make connections I wouldn’t have thought of to both older
and recent literature, and this was often very helpful in avoiding various kinds of wheel
reinventions. Something that would often happen in our meetings is that David would
impromptu suggest an alternative analysis to the one I was trying to present to him, and
though this alternative might sometimes not be great on its own (his words, not mine),
it often prompted me to reframe the question from a different and fruitful perspective.
David also never failed to let me know when I was selling myself short, and always made
me see when what I was saying could actually have a larger and more meaningful impact
than I was imagining.
Athulya Aravind very explicitly made sure that I would remember that {a/the} im-
plementation is not (the most interesting part of) {an/the} analysis. Losing the forest for
the trees was a real and frequent danger throughout this process, and it was unvaluable to
have a reminder that there’s an important distinction between what’s a purely contingent
detail that will change in 5, 10, or 15 years, vs. what is a deeper and more long-lasting in-
sight that will hopefully still teach us something once our theory improves. Also, though
I don’t think this dissertation as it is now is a paramount example of the clearest writing
and exposition by any stretch of imagination, it was much worse before Athulya got in-
volved. Despite the complexity, I’ve tried to make it as roadmappy2 as possible, and that’s
largely thanks to her insistence.
Finally, I want to thank Amy Rose Deal for welcoming me to UC Berkeley for a lovely
Fall semester, for throwing a party for me when I left, and for agreeing on joining this
committee. At the very start of the project, she helped me see how all of these ideas
would fit into a single coherent narrative that could become a dissertation – one that was
not just worthy of writing, but hopefully also worthy of reading. During the research
work then, I need to thank her for repeatedly telling me to stop what I was doing. That
is: she kept me on the right track on the many, many occasions I was getting lost in
2 Adjective credit to Keely New.
Acknowledgments 5
various minute details. While I was losing my mind on trying to get data covering every
single possible combination of various parametersi that I didn’t know whether i would
ultimately matter or not, Amy Rose told me to pause that completely, go read what the
theories on the market are, see what works and what does not, how and why, and where
things break. Needless to say, I wouldn’t have made much progress without that.
Outside of the committee proper, I want to thank some other (temporary, alas) MIT
faculty: Ksenia Ershova for followingme at early stages of the process (please come back!),
and Peter Grishin for always letting me chat about the various intricacies I met (also
please come back!). My bothering linguists did not stop at MIT, and a lot of people out
there should by now be somewhere between fearful and annoyed if they hear the word
“Äiwoo” ever again (they might). Obviously there is Åshild Næss, without whom I would
never even have heard about Äiwoo in the first place, and who encouraged me to work
on this language as a lost MA student (and made it possible by sharing her data with
me for years). There’s Rajesh Bhatt, Henrison Hsieh, Victoria Chen, mitcho, the Berkeley
syntax faculty (Justin Royer, Line Mikkelsen, Peter Jenks), Gary Thoms, Michelle Yuan,
Coppe van Urk, Matt Hewett, Sandhya Sundaresan, Karlos Arregi, David Adger, Patrick
Elliott, and probably more that I’m forgetting. All of these people have at one time or
another been a precious sounding board (whether willingly or by coercion), and all their
contribution is in here somewhere. Finally, but definitely not least importantly, I must
thank the biggest Äiwoo expert of them all: my consultant Luke Gitakulu, who has been
demonstrated incredible patience, generosity, and linguistic expertise in sharing with his
language with me. Wâkâlou-mana, Luke.
Faculty are not the only people who helped me out with this whole thing. My days
at MIT started not at MIT at all, while I was still in Oslo due to the covid pandemic,
and that was so for my entire first year. My cohort mates Levi Driscoll, Katie Martin, Jad
Wehbe and Devon Denny did a fantastic job at keeping me in good humor despite the
very unusual circumstance of having to face the classic first year struggle all on Zoom
and in isolation (though for an entire year I had no idea how tall any of them were).
After that rather strange first year, I finally got to move to Boston, and the rest of the MIT
student body opened their arms to me. Keely Newmust be thanked first – I think that goes
without saying (she baked the cake for my defense, after all. And reminded me of every
single imaginable deadline I was about to miss). My officemates Peter Grishin, Agnes Bi,
Cora Lesure, Eunsun Jou, Anastasia Tsilia, Zhouyi Sun, Taieba Tawakoli, Bingzi Yu, and
Chelsea Tang have had to deal with my finger tapping, so they must be awarded a medal.
The list doesn’t stop there: there’s (been) Stan, Filipe, Adele, Margie (who single-handedly
no less than revolutionized my English in ways hitherto unexplored), Yash, Omri, Ido,
Christopher, Shrayana, Lorenzo, Ford, and many more. MIT (and Boston in general) can
be a rather dreary place at times, and I really have to thank all of you for lifting my spirits.
I don’t think I’m in any risk of running out of linguists to thank, but I would be re-
miss to limit these mentions to only MIT people (we already have a reputation, so maybe
let’s not do that). Thanks to Amy Rose, who as mentioned accepted to have me at UC
Berkeley for the Fall of 2023, I was able to go inconvenience even more people than just at
6
MIT. Luckily, the Berkeley linguistics department was incredibly welcoming to me, and
I couldn’t have asked for a better host department: I want to thank Akil, Shweta, Becky,
Måsi, Rhosean, Katie, Wesley, and more (Franco: you’re coming up but among the friends
instead of the academics, because you’re more important than to just be relegated to being
a linguist. Be patient for once.). I also want to thank the AFLA contingent (Henrison, Jed,
Alessa, DK), and I finally, all the other linguists met here and there who have entertained
me, enriched me, and kept me sane: thank you Lena, Yiannis, Dan (Greeson), Jelly, Yvette,
Tamisha, Jessica, Iman (my one and only zdōra sister), and the ones I’m forgetting (I feel
I’m ending every single paragraph like this, but it’s fine).
There’s people outside of academia who also need to be thanked. As it goes at every
respectable award show: I would like to thank my mom and my dad, my brother, and
my grandparents, for always believing in me and supporting me, and I would not stand
on this stage today if it hadn’t been for them. Or something like that, I’ve never won an
award. Though that was a standard script, it doesn’t make it any less true: my family has
relentlessly supported me in so many ways through the years that it feels silly to even try
to enumerate them. It is absolutely true in a very literal sense that I would not be here
without them, and that all I’ve accomplished goes in some measure back to them. Grazie
grazie grazie. As if one family wasn’t enough, I’m also lucky enough to have a second
family in Norway, who though I’ve seen less often in the past few years (unnskyld!), never
fails tomakeme feel warm andwelcome as thewayward bonus son I am. Especially during
my first year, when I was still in Norway, their support was invaluable, and having the
possibility of just visiting them at a moment’s notice was so incredibly precious when
things got rough (which they did, at the time). Lars, Marianne, Julie, Martin: tusen takk.
The last group of people I feel very strongly about including in here is also a family of
sort – they collectively very much function as one3. People who know me know that I am
not a person who is often particularly proud of much of what I do. However, this group of
people – this is something I am proud of, because it really does take a village, and I’m both
proud and extremely grateful for having been part of this village. There’s something very
particular that only a solid circle of individuals who might all be called one or multiple
slurs can do for each other. Like my family/-ies, this extra family also consists of people
who, fortunately for them, know absolutely nothing about what’s in the 300+ pages of
this document – and I really hope that it stays that way: no linguistics should be inflicted
upon them. They have done nothing wrong to deserve that, and in fact quite the opposite,
they have more than deserved to stay blissfully out of the academia of it all.
When I moved to Boston in August 2021, I was coming out of a series of frankly awful
personal circumstances that had been lasting for a very long time, and I had only barely
begun the long process of shaking them off and growing around them. In the very first
week after my moving, Magdiel and Albie Lorenzo Martinez Tsoi had no hesitation, and
3 Note to the readers who are not familiar with the English variety I’m silently code-switching to in these last
few paragraphs: here, kinship is not transitive. If someonei is my mother and someone elsej is my daughter,
this does not entail that theyi are theirj grandmother. Also, ‘she/her’ and ‘mother/daughter/sister/aunt’ are
all gender-neutral. I will not elaborate, nor accept any questions.
Acknowledgments 7
immediately took under their wings what was at the time a very lost stranger, with no
clear sense of who I was, nor who I was going to be. Ever since, and even after their moving
to Chicago, my moms have never showed me anything else than generosity, kindness,
patience, and mentorship (and, impossible to omit, they also let me spend precious time
with their dog/my niece Pika and the most recently arrived, potato-shaped menace Yoshi).
They have always been an incredibly strong and consistent support system for me to rely
on when times got rough. And even more importantly – they sawmy qualities way before
I saw them myself, they showed me how to build a support system for myself, how to put
in the effort to let friendships flourish and thrive, and they let me understand that I was
worthy of that. I genuinely believe that I would not be as fully realized a person as I am
today had it not been for them, and for all the times and ways they have showed up for
me through these last four years without ever asking for anything in return, aside from
the occasional tray(s) of lasagne or tiramisù. Thank you, and see you soon.
There’s more in the aforementioned village, the one that it takes, than just my moms.
All those words I spent above about them could equally be said about any of the people I
mention below, but it would get repetitive – please y’all, listen, I know it looks like they’re
getting a full paragraph and you’re not, but hear me out: don’t be messy, and just know
that I mean every single one of those words about all of you too. There’s John Fernandez
and Kuresse Bolds, two of the kindest and funniest people I know, sharing my interests,
letting me incoherently yap at lengths on their astoundingly comfortable couch, or taking
me in for the nightwhen I forgotmy house keys. There’s BrianGonzalez and ElieMehanna
(and Lemon!), always taking the initiative of getting me out of the house and inviting me
to functions, always introducing me to someone new, always making sure I was having
a good time. There’s Jordan Veiga, who would call me a new name every single day, who
would cook for me and give me free newly-concocted drinks and spend hours on the
phone to keep me company while allowing for and encouraging the utmost unhingedry
(I feel you’re gonna like this word). There’s Victor Flucus (and Rosemary, and the much
missed Patchouli), who would have the enthusiasm and spontaneity to improvise last-
minute plans, who would never fail to make me feel safe, seen, and proud of myself (and
God knows how hard that can be), and who would always have a kind word for me and
was never shy about it.
There’s Juan Vásquez, with whom I was simultaneously able to share both complaints
and concerns about our academic lives, and other kinds of thoughts that absolutely should
not be put on print here. Elliptically speaking, he should be thanked for bringing me
on some of the best vacation trips I’ve ever been on (in fact – a special shout out to
Hotel [redacted] in Zipolite for giving me the pool and cocktails that fueled the original
idea of C probing for [i |v]). And as anticipated: my beautiful, suspiciously tall daughter
Franco Liu (and my niece Anna), who was already named above, should really figure in
this section, as more often than not I was trying my best to talk about everything else
than linguistics with him. The way he helped me with all the linguistics in here, though
very real, actually pales compared to how he helped me by being my friend in the most
precious way possible. I’m not gonna mention specific facts or actions, because he knows
8
them all (because I tell him constantly). Suffice it to say: the patience he has for me and
the way I behave is honestly concerning, and I couldn’t thank him enough. Thank you
bby, there’s not enough parmigiana di melanzane in the world I could make for you to
thank you. There really is so many more; there’s Miguel and Aaron (and Koko), there’s
Tina, there’s Nyx, there’s Danny, there’s Mike, there’s Dennis, there’s – you get the gist.
To all of you: you have always kept me as happy as I could ever ask to be, you’ve
offered me help when I needed it, company when I needed it, entertainment, laughs, and
took my brain off this whole linguistic stuff that is ultimately, truly, not all that important.
Without you, this dissertation might perhaps still have been finished, but it would’ve been
written by a much, much sadder human being. Thank you.
CONTENTS (AT A GLANCE)
Chapter 1 What this dissertation is about 21
A case study about locality from an Oceanic language
Chapter 2 The Äiwoo language: background 45
Some necessary information
Chapter 3 Äiwoo clause structure 85
Deriving canonical transitive clauses
Chapter 4 The Ā-extraction restriction, and the logic of an exception 147
Why only certain arguments can be extracted (apart from when it isn’t so)
Chapter 5 The φ-indexing system 209
Morphosyntactic properties of φ-prefixes and φ-suffixes
Chapter 6 Non-canonical UV clauses 249
Completing the Word Order and Extraction Puzzles
Chapter 7 Conclusion and prospects 317
It’s just relativized minimality all the way down
Appendix A Notes about Circumstantial Voice 321
A construction that is not really a third voice at all
Appendix B A “case-agreement” analysis 343
How a potential alternative analysis of voice morphology would have to work
CONTENTS
Acknowledgments 3
Contents (at a glance) 9
Contents 10
Glossary 17
Typographical conventions 19
Chapter 1 What this dissertation is about 21
1.1 Introduction 21
1.2 Two puzzles: word order and extractability 22
1.3 What do we learn? 24
1.3.1 The syntax of Austronesian voice systems 24
1.3.2 The locality ofmovement, syntactic inversion, andĀ-extraction restrictions 29
1.3.3 Lexical DPs, pronouns, φ-features, and nominal licensing 32
1.4 Data and methodology 36
1.5 Theoretical assumptions 38
1.6 Overview of the dissertation 41
Chapter 2 The Äiwoo language: background 45
2.1 Introduction 45
Contents 11
2.2 Geographic and linguistic situation 45
2.3 Earlier literature 50
2.4 Phonology and morphophonology 51
2.4.1 Segmental inventory 51
2.4.2 Essential phonological patterns 53
2.4.3 Essential morphophonological patterns 59
2.5 Morphosyntax in the nominal domain 62
2.5.1 The number system 62
2.5.2 Structure of nominal constituents 67
2.5.3 Short notes about definiteness and specificity 69
2.5.4 A short note about null arguments 72
2.6 (Some) morphosyntax in the verbal domain 74
2.6.1 Aspectual/modal prefixes 75
2.6.2 Directionals 77
2.6.3 The post-verbal TAM complex 79
2.6.3.1 Tense/aspect/polarity semantics 80
2.6.3.2 The allomorphy of consonant-initial enclitics 82
2.6.3.3 The null suffixes -∅n and -∅ŋ 83
Chapter 3 Äiwoo clause structure 85
3.1 Introduction and empirical domain 85
3.2 Basic word order: Äiwoo as a V2 language 86
3.3 Morphosyntactic correlates of voice 89
3.3.1 Voice morphology 90
3.3.2 Presence of prefixal φ-agreement signals AV 92
3.3.3 Interim summary 93
3.4 Meta-theory of Austronesian voice: cause or effect 94
3.5 Deriving canonical clauses 97
3.5.1 The vP domain: voice is determined low 98
3.5.1.1 The position of voice morphology and the Mirror Principle 98
3.5.1.2 UV morphology is object agreement; AV is default 100
3.5.2 The TP domain 106
3.5.2.1 The position between V and TAM is spec,TP 106
12 Contents
3.5.2.2 Moving nominals to spec,TP (a first pass) 108
3.5.3 The CP domain: VIA fronting 109
3.5.3.1 C agrees with what v has agreed with (but it has a plan B) 111
3.5.3.2 The PlanA/Plan B logic,Multitasking, and theoretical parsimony 115
3.5.4 Putting it all together: step-by-step derivations for AV and UV clauses 116
3.5.4.1 UV derivation 116
3.5.4.2 AV derivation 119
3.5.5 Prefixal φ-agreement is in C, and gets impoverished in UV 120
3.6 A note about intransitives 126
3.7 The A/Ā-properties of fronting to spec,CP 131
3.7.1 VIA fronting to spec,CP is non-local A-movement 131
3.7.2 The locality of (all) movement is featurally relativized 137
3.7.3 Potential parallels 141
Chapter 4 The Ā-extraction restriction, and the logic of an exception 147
4.1 {Over/Pre}view 147
4.2 Setting the stage: Äiwoo relative clauses 150
4.2.1 Empirical description 150
4.2.2 Relativization is an Ā-dependency 153
4.2.3 The syntactic structure(s) of Äiwoo RCs 154
4.3 The Austronesian Ā-Extraction Restriction 159
4.3.1 What can be relativized when 159
4.3.2 Developing a theory of the extraction restriction 163
4.3.2.1 Against a phase-based account of Äiwoo extraction 164
4.3.2.2 An (Ā-)intervention theory of Äiwoo extraction 167
4.4 UV-only verbs: a description 173
4.4.1 An idiosyncratic lexical class 173
4.4.2 Proving the non-existence of AV forms 174
4.4.3 UV-only verbs are not just syncretic for voice 176
4.4.4 UV-only verbs in the Austronesian context 177
4.5 Exceptional non-VIA extraction 179
4.5.1 UV-only verbs allow subject extraction 179
4.5.2 Analysis: selecting a special v head 182
Contents 13
4.5.2.1 Walking backwards: how the story must go 183
4.5.2.2 Forcing UV/object agreement with selection 186
4.5.2.3 Deriving the lack of an extraction restriction 187
4.6 Summary 189
4.A A note about mi= and other “relativizers” 190
4.A.1 Distribution 190
4.A.2 (Non-)complementarity with a head noun 192
4.A.3 Other properties 194
4.A.3.1 Optionality 194
4.A.3.2 Restrictiveness 196
4.B A note about other Ā-dependencies 197
4.B.1 Topicalization is not movement-based 197
4.B.2 Focus is done with pseudo-clefts (so it’s just relativization) 198
4.C Verbs oscillating between UV-only and voice-syncretic: an acquisition problem? 203
4.C.1 Voice-syncretic verbs: how to establish that an AV form exists 204
4.C.2 UV-only verbs: how to establish that an AV form does not exist 205
Chapter 5 The φ-indexing system 209
5.1 Introduction 209
5.1.1 Domain of discussion 210
5.1.2 Agreement markers, pronouns, and all that’s in between 211
5.1.2.1 (Morpho)phonology is not (morpho)syntax 214
5.1.2.2 Morphophonologically, all Äiwoo φ-indexers are affixes 215
5.1.2.3 Syntactic evidence: co-occurrence with arguments 218
5.2 Φ-prefixes and φ-suffixes 221
5.2.1 The paradigms 221
5.2.2 Prefixes and suffixes are syntactically different: a preview 223
5.3 The φ-prefixes: obligatory co-occurrence 223
5.3.1 The data 223
5.3.2 An argument for the presence of null pronouns 224
5.3.2.1 Showing the existence of null pronouns in spec,CP 224
5.3.2.2 The position of the subject pronouns 226
5.3.3 Summary: the φ-prefixes are agreement markers 227
14 Contents
5.4 The φ-suffixes 228
5.4.1 Distribution: co-occurrence of φ-suffixes and arguments 229
5.4.2 The φ-suffixes are pronouns 232
5.5 The morphology of Äiwoo pronouns 235
5.5.1 Decomposing pronouns: short and long forms 235
5.5.2 Spelling out a non-constituent 239
5.6 Chapter summary 247
Chapter 6 Non-canonical UV clauses 249
6.1 Completing the Puzzles 249
6.2 Classes of nominals in Äiwoo syntax 251
6.3 Four types of UV clauses and their distribution 252
6.3.1 UVplain: canonical UV clauses 256
6.3.2 UVgu clauses: objects in spec,TP 256
6.3.3 UVSVO clauses: UV with SVO order 259
6.3.4 UV1>2 clauses: a subtype of UVSVO 261
6.3.5 Interim summary 263
6.4 Pronominal objects and subject extraction from UV 265
6.4.1 The parallel with UVonly clauses: obligatory UV and free extraction 266
6.4.2 Voice and extraction are independent from word order 270
6.4.3 Interim summary, desiderata, and preview 272
6.5 Deriving non-canonical UV clauses 275
6.5.1 Separating our concerns: voice/extraction properties and word order 275
6.5.1.1 The featural make-up of Äiwoo nominals 277
6.5.1.2 v agrees with Ā-marked DPs and pronouns 279
6.5.1.3 T raises the closest φ+P (but it has a plan B) 283
6.5.1.4 Dynamic interaction features as a per-probe property 288
6.5.2 Putting it all together again: step-by-step derivations 292
6.5.2.1 Deriving UVgu clauses 292
6.5.2.2 Deriving UVSVO clauses 297
6.5.2.3 Deriving UV1>2 clauses 301
6.5.2.4 Deriving UVplain clauses, updated 305
6.5.2.5 Deriving AV clauses, updated 308
Contents 15
6.6 Summary 312
6.A The caveat: plural lexical DPs 313
Chapter 7 Conclusion and prospects 317
7.1 Summary of the findings 317
7.2 Open questions 318
Appendix A Notes on Circumstantial Voice 321
A.1 Introduction 321
A.2 Descriptive properties 321
A.2.1 Word order, verbal morphology, and argument structure 321
A.2.2 Ā-extraction properties 327
A.2.3 Why “Circumstantial Voice” is not a voice 329
A.3 Towards an analysis of Circumstantial Voice 332
A.3.1 Circumstantial Voice as preposition stranding 332
A.3.2 Problems and incorrect predictions 338
Appendix B A “case-agreement” analysis 343
B.1 Object agreement, case agreement, and second-cycle effects 343
B.1.1 A competing theory of Austronesian voice 343
B.1.2 “Case agreement” without case? 345
B.2 An attempt at an implementation 346
Bibliography 351

GLOSSARY
1 first person F feminine
2 second person FOC focus
3 third person FUT future
12 first person inclusive GEN genitive
A agent HAB habitual
ABS absolutive HESIT hesitation
ACC accusative IMP imperative
ADDR addressee INCL inclusive
ADJ adjective IND indicative
ADV adverbial INDEF indefinite
APPL applicative INTR intransitive
APPR apprehensive IPFV imperfective
ASP aspect IRR irrealis
AUG augmented number KÄ Attitude predicate kä/kâ
AUTH author LK linker
AV actor voice LOC locative
CAUS causative LV locative voice
COLL collective M masculine
COM comitative MIN minimal number
CV circumstantial voice NEG negative
DAT dative NOM nominative
DEF definite NPIV non-pivot
DEIC deictic NVOL non-volitional/abilitative
DET determiner OBJ object
DIR directional OBL oblique
DIST distal P patient
DOM differential object marking PART participant
DU dual PFV perfective
ERG ergative PIV pivot
EXCL exclusive PL plural
18 Glossary
PN proper name PV patient voice
POSS possessive REL relative
PREP preposition SBJV subjunctive
PRF perfect SG singular
PROG progressive SUBJ subject
PROH prohibitive TAM tense/aspect/mood
PRON pronoun TEMP temporal
PROX proximal TOP topic
PRS present TRI trial number
PRT particle UA unit-augmented number
PST past UV undergoer voice
TYPOGRAPHICAL CONVENTIONS
(A) Movement:
a. X Y: movement
b. X 7 Y: failed/impossible movement
(B) Agree:
a. X Y: yes interaction, no satisfaction
b. X Y: yes interaction, yes satisfaction
c. X Y: no interaction, no satisfaction
d. X Y: no interaction, yes satisfaction
(C) a. Movement: b. Failed/impossible movement:
X X
tX 7 tX
(D) a. +INT, −SAT: c. −INT, −SAT:
X X
Y Y
b. +INT, +SAT: d. −INT, +SAT:
X X
Y Y


CHAPTER 1
WHAT THIS DISSERTATION IS ABOUT
1.1 INTRODUCTION
The kind of theoretical questions this dissertation explores revolves around the locality
profiles of different types of movement. To elucidate these theoretical issues empirically,
I use as a case study the syntax of Äiwoo, an understudied Oceanic language from the
Solomon Islands. In this language we see an intricate pattern of multiple clause-internal
A-movements. Moreover, this system interacts in non-trivial ways with the syntax of
Ā-extraction, sometimes restricting which argument(s) can be Ā-extracted from certain
types of clauses, and sometimes not. At a bigger-picture level, the whole dissertation can
be thought of as building on van Urk (2015), supporting the argument that the locality
of movement is solely determined by Relativized Minimality (Rizzi 1990). By conjoining
these proposals to the independently motivated notion of featurally relativized probes
(Béjar 2003 et seq.), I argue that we can derive the precise patterns of locality and inter-
vention we observe – including some that I show were implicitly predicted to exist by our
theory.
Äiwoo’s syntactic properties make it an interesting language to test these theoretical
proposals on, offering a particularly clear view on a series of questions that have oth-
erwise received diverging answers in the literature. These concern, among others, the
syntax of Austronesian languages and their voice systems, the locality properties of both
A- and Ā-movement, the relation between intra-clausal A-movement and the nature of
Ā-extraction restrictions, and the internal and external syntax of different categories of
nominals. Precisely because the Äiwoo dataset is rather complex and involves several in-
teracting moving parts, this lets us tease apart analytical options that – given a simpler
pattern – would be equally plausible on the theoretical side, but otherwise empirically
indistinguishable.
22 1.2. Two puzzles: word order and extractability
In the next section (§1.2), I present the two central puzzles that will constitute the
empirical domain of inquiry of the dissertation. In §1.3 I present howÄiwoo contributes to
answering the theoretical questions mentioned above. In §1.4 I discuss how the empirical
data at the basis of this dissertation was collected. The following section (§1.5) contains
a discussion of the theoretical framework I assume in this dissertation. Finally, in §1.6 I
give an overview of the following chapters.
1.2 TWO PUZZLES: WORD ORDER AND EXTRACTABILITY
The Äiwoo language presents a complex system of clausal alternations. Any declarative
clause will fall into one of several types, all of which differ in their word order patterns and
their Ā-extraction profiles, that is, which argument(s) can or cannot be extracted. This em-
pirical landscape presents us with two puzzles, highly intertwined with each other: why
and how does the word order change, and how is this related to Ā-extraction possibilities?
Let us now try and examine the two puzzles separately, to get a first-glance idea of the
kind of empirical problem this dissertation aims to solve.
THE WORD ORDER PUZZLE: VARIABLE, YET STRICT A striking property of Äiwoo at
first glance is its variety of word order patterns. In this language, word order is variable, in
the sense that clauses can look a variety of different ways. Just as an illustration, consider
the four sentences in (1). These are all simple transitive, monoclausal declaratives. For
ease of tracking, I have colored subjects and related morphemes in red, and objects and
related morphemes in blue; I will use this same color scheme throughout the dissertation.
We can immediately see the variability in word order. In (1a), we have what looks like
a relatively run-of-the-mill SVO clause. The subject is preverbal (mikilitei ‘fishermen’),
the verb carries agreement with it (lu- 3AUG), and the object is postverbal (nulei ‘crab’) –
nothing shocking so far. However, this changes in (1b): now the object is preverbal, the
verb ‘bring’ has a different stem (tu instead of tou) and it lacks agreement, and the subject
intervenes between the verb and the future marker. Further, consider (1c): we still have
a preverbal object, the tu-stem of ‘bring’, and no prefixal subject agreement, but now we
have some suffixes on the verb indexing the object (-gu-de ‘OBJ-12AUG’), and the subject
is to the right of the future marker instead of to its left. Finally, in (1d), we see SVO word
order again, but now with the tu-stem of ‘bring’ instead of the tou-stem; moreover, there
is no prefixal agreement, but a suffix indexing the subject instead (-i ‘3AUG’).
(1) Examples of different clausal word orders in Äiwoo: (250717)
a. mikilitei ku-lu-tou-mä=kaa nulei ngâ nuwopa
fishermen IPFV-3AUG-bring.AV-DIR1=FUT crab to house
‘The fishermen will bring crab(s) home’
b. nulei ku-tu-mä mikilitei=kaa ngâ nuwopa
crab IPFV-bring.UV-DIR1 fishermen=FUT to house
‘The fishermen will bring the crab(s) home’
Chapter 1. What this dissertation is about 23
c. (iumu) ku-tu-mä-gu-mu=waa mikilitei ngâ nuwopa
2MIN IPFV-bring.UV-DIR1-OBJ-2MIN=FUT fishermen to house
‘The fishermen will bring you home’
d. (ijii) ku-tu-mä-i=laa iumu ngâ nuwopa
3AUG IPFV-bring.UV-DIR1-3AUG=FUT 2MIN to house
‘They will bring you home’
The variation shown in (1) is not completely chaotic or unconstrained. What varies is
the position of the arguments and the verbal morphology (the stem morphology, and the
presence/position of agreement-like affixes). Other aspects remain constant: importantly,
we see a V2 effect, with exactly one argument always occurring to the left of the verb.
The robust V2 effect is a symptom of the fact that despite its variability, word order
in Äiwoo is strict. No changes can be made to any of the sentences in (1) without result-
ing in ungrammaticality. In other words, these different word orders are not optionally
available alternatives. For each sentence, only the given combination of word order and
verbal morphology is grammatical, and any of the other three options will be impossi-
ble. We thus have a rather rigidly regimented system: depending on some factors that we
will need to understand, Äiwoo clauses may end up looking quite different, but there’s
always only one correct answer. This is our first puzzle: what determines what an Äiwoo
clause will look like? What factors govern the distribution of the different word orders,
and of the varying verbal morphology? How do the arguments end up in their different
positions, and why there and not elsewhere?
THEEXTRACTIONPUZZLE: SOMETIMES RESTRICTED, SOMETIMESNOT Our second puz-
zle concerns patterns of Ā-extraction. Depending on the type of clause, some arguments
may be Ā-extracted (e.g., relativized), but not others. To observe the scope of the problem,
an overview of different word order and Ā-extractability profiles is shown in (2); note that
this table includes types of clauses that were not shown in (1)1. A superscript (3/7) on
an argument indicates whether it can or cannot be Ā-extracted; “NA” means that the ar-
gument in question cannot be extracted, but for independent/orthogonal reasons. A few
more notational remarks: (i) “V.AV” and “V.UV” identify the two different stems that tran-
sitive verbs can show, like tou and tu for ‘bring’ across the examples in (1); (ii) the symbol
“φS-” represents prefixal subject agreement, like lu- 3AUG in (1a); (iii) “-πS/O” represents a
pronominal suffix indexing the features of the subject or object respectively.
1 The shorthand names in the first column (“UVplain”, “UVgu”, etc.) are just arbitrary mnemonics for conve-
nience, and they are of no great importance now. I will use them throughout this chapter to handily refer
to particular clause types, but the actual motivations behind the different shorthands will become clear in
the chapters to follow.
24 1.3. What do we learn?
(2) Alternations of word order and Ā-extractability:
Clause type: Word order:
a. AV: S3 φS-V.AV =TAM O
7
b. UV 3 7plain: O V.UV S =TAM
c. UVonly: O
3 V.UV S3=TAM
d. UV NAinâ: O V.UV S
3=TAM
e. UVgu: O
NA V.UV -πO=TAM S
3
f. UV : SNASVO V.UV -π
NA
S =TAM O
In the upper third of the table (2a,b), for “AV” and “UVplain” clauses, we see that only
the highest argument can be extracted – respectively the subject and the object. Then, in
the middle third (2c,d) we see clauses that have the same word order as the UVplain ones
in (2b), but here, extraction is less restricted: the subject can be extracted, despite being
lower than the object. Finally, in the bottom third (2e,f ), we see clauses with different
word orders, and another case of extraction being possible from a lower position across a
higher argument, i.e., the subject in “UVgu” clauses (2e).
Importantly, extractability and word order are only partially correlated. For example,
the subject of a “UVplain” clause (2b) cannot be extracted, but in other types of clauses, the
subject is in the exact same position and yet it can be extracted (2c,d). Furthermore, the
subject in (2e) can be extracted despite being in a lower position – but this is not shared
with the object in (2a). Our goal will therefore be to devise a model that simultaneously
accounts for the two puzzles: the word order alternation patterns, and how these correlate
(or not) with restrictions on Ā-extraction.
1.3 WHAT DO WE LEARN?
1.3.1 THE SYNTAX OF AUSTRONESIAN VOICE SYSTEMS
The two empirical puzzles outlined above showcase a complex interplay of several syn-
tactic phenomena, including word order, Ā-extraction, and Austronesian voice. I will start
discussing the contributions of this dissertation from the latter. Austronesian languages
are notorious for – among other things – their “voice systems”, which we can describe in
coarsely simplified terms as follows. Any verb can take several morphological forms (dif-
ferent “voices”), and each form corresponds to a different argument in the clause becoming
“special”. Depending on the specific language, “special” can concretely mean e.g. being in
a particular position in the clause, showing a particular case marking, etc. This “special”
argument has variously – and at times confusingly – been labelled the “subject”, “trigger”,
“topic”, “focus”, and “pivot” (see Chen & McDonnell 2019 and references therein). In this
section, I will use the term “pivot” for consistency.
Chapter 1. What this dissertation is about 25
An example of this kind of system is given in (3), from Paiwan (Formosan). Here, the
voice morphology of the verb correlates with which argument takes the marker a (glossed
as ‘PIV’); I use color-coding and boldface to highlight the correlation. In the Actor Voice
(3a), the verb qalup ‘hunt’ is marked by the infix ⟨m⟩, and the subject/agent/external ar-
gument is the pivot. In the Undergoer Voice (3b)2, marked by the suffix -en, now the direct
object/internal argument/patient/theme takes the PIV marking, and the agent is marked
as non-pivot (“NPIV1”). In the Locative Voice (3c) the verb has a suffix -an, and now the
locative argument is the pivot; both the subject and the direct object are marked as non-
pivots (with two different case markers, “NPIV1” and “NPIV2”). Finally, in Circumstantial
Voice (3d), the verb has a prefix si-, and the pivot is the instrumental argument.
(3) An example of an Austronesian voice system (Paiwan) (elaborated from Ferrell
1979: 202, Chen & McDonnell 2019: 174):
a. Actor Voice:
q⟨m⟩alup a caucau tua vavuy i gadu tua vuluq
⟨AV⟩hunt PIV man NPIV2 pig LOC mountain OBL spear
‘The man hunts wild pigs in the mountains with a spear’
b. Undergoer Voice:
qalup-en nua caucau a vavuy i gadu tua vuluq
hunt-UV NPIV1 man PIV pig LOC mountain OBL spear
‘The man hunts wild pigs in the mountains with a spear’
c. Locative Voice:
qalup-an nua caucau tua vavuy a gadu tua vuluq
hunt-LV NPIV1 man NPIV2 pig PIV mountain OBL spear
‘The man hunts wild pigs in the mountains with a spear’
d. Circumstantial Voice:
si-qalup nua caucau tua vavuy i gadu a vuluq
CV-hunt NPIV1 man NPIV2 pig LOC mountain PIV spear
‘The man hunts wild pigs in the mountains with a spear’
This system of clausal organization has been object of a rich literature, and has been inter-
preted and analyzed in a number of ways. For some, these different “voices” reflect actual
argument structure alternations, with one argument per clause being able to become “the
subject” in certain senses (Payne 1982, De Guzman 1988, Gerdts 1988, Aldridge 2004, 2008,
2017, a.o.). In this way, “Austronesian voice” as a system pertains to the A-syntax of the
language. For others, there is no argument structure alternation, and the verbal morphol-
ogy we call “voice” actually tracks notions related to information structure and discourse,
such as topicality (e.g., the “pivot” is conceptually similar to a topic; Richards 2000, Pear-
son 2001, 2005, Rackowski 2002, Rackowski & Richards 2005, van Urk 2015, V. Chen 2017a,
2 The cited sources use the term “Patient Voice” (glossed PV) instead of “Undergoer Voice”. Terminological
variation of this kind is abundant in the Austronesianist literature. I use the term “Undergoer Voice” here
simply for ease of comparability with Äiwoo, with no deeper significance.
26 1.3. What do we learn?
Hsieh 2020, a.o.). For these kinds of approaches, “Austronesian voice” is more connected
to the Ā-syntax of the language.
In addition to the system of clausal alternations, Austronesian voice systems also often
show a notorious Ā-extraction restriction (Schachter & Otanes 1972, Keenan & Comrie
1977, Aldridge 2004, Rackowski & Richards 2005, van Urk 2015, Hsieh 2020, V. Chen 2023,
a.m.o.). Descriptively, only the pivot argument in any clause can be Ā-extracted, or framed
the other way around, to Ā-extract an argument one must use the voice that would make
that argument the pivot. Concretely: subject extraction is only possible with AV, object
extraction is only possible with UV, etc.
This pivot-only extraction restriction has also been analyzed in a number of ways3. For
some approaches, these languages feature a “highest DP-only” condition on Ā-extraction
(Nakamura 1996, Aldridge 2004, 2008, a.o.). The pivot always ends up as the structurally
highest nominal in a clause, and no lower arguments can undergo Ā-movement across it.
This kind of locality profile is different from that of canonical Ā-movement, which is usu-
ally not blocked by intervening DPs as long as they don’t have Ā-features (consider, e.g.,
any object wh-question in English, where the object can be wh-moved to spec,CP despite
the subject being higher; see also Branan & Erlewine 2024). For others, this Ā-extraction
restriction is a byproduct of phasehood (Rackowski & Richards 2005, Erlewine & C. Lim
2023, Hsieh 2025, a.o.). In any clause, only the pivot is above the clause-internal phase
boundary, whereas all other arguments are below it. As a result, only the pivot is accessi-
ble for Ā-extraction.
Having established some background, we can now bring Äiwoo back into the picture.
At first glance, what we see is a relatively traditional or “well-behaved” system, with the
typical characteristics of what we would expect from an Austronesian language. Äiwoo
has a binary opposition between “Actor Voice”, where the “special” argument is the sub-
ject/agent/external argument, and “Undergoer Voice”, where the “special” argument is the
object/theme/undergoer/internal argument (at first analyzed as such in Næss 2015b)4. The
opposition between Actor Voice and Undergoer Voice in Äiwoo is perhaps morphologi-
cally unusual in being realized largely through idiosyncratic stem alternations (e.g. tou~tu
‘bring.AV~UV’, but togo~togulo ‘hit.AV~UV’, vängä~ngä ‘eat.AV~UV’, etc.) rather than with
dedicated affixes as in other better-studied Austronesian languages, such as Paiwan in (3).
However, syntactically it behaves more or less as one would expect. Compare, e.g., Actor
Voice and “UVplain” clauses (2a,b), repeated in (4). Although Äiwoo doesn’t have any case
marking, in both kinds of clauses the pivot argument is identified by being in a consistent
dedicated position (preverbal). We also have an Ā-extraction restriction: only the subject
can be extracted from Actor Voice, and only the object can be extracted from Undergoer
3 Although prominent, the two family of approaches discussed here don’t exhaust the range of proposals that
have been put forward, especially outside of Minimalist/generative syntax; see e.g. Himmelmann (2002),
W. A. Foley (2008), and Riesberg (2014) for other ideas.
4 A third construction, “Circumstantial Voice”, is not in paradigmatic alternation with these two, and has
different properties that make it less clearly part of the voice system, despite its label (from Næss 2015b). I
will not focus on Circumstantial Voice in this dissertation. See Appendix A for some notes.
Chapter 1. What this dissertation is about 27
Voice. For ease of exposition, I don’t actually show this here, but merely indicate it with
superscripts; see chapter 4.
(4) The basic contrast between Actor Voice and Undergoer Voice: (250717)
a. mikilitei3 ku-lu-tou-mä=kaa nulei 7 ngâ nuwopa (AV)
fishermen IPFV-3AUG-bring.AV-DIR1=FUT crab to house
‘The fishermen will bring crab(s) home’
b. nulei3 ku-tu-mä mikilitei 7=kaa ngâ nuwopa (UV)
crab IPFV-bring.UV-DIR1 fishermen=FUT to house
‘The fishermen will bring the crab(s) home’
However, once one scratches the surface, the Äiwoo system also shows some addi-
tional complications that are not found in other Austronesian languages. As we saw from
our description of the Word Order Puzzle, word order varies both across and within the
two voices, without a one-to-one correlation: in certain clauses (“UVSVO”, (2f )) the verb
has UV morphology, but the argument in the preverbal “pivot” position is the subject,
not the object. Moreover, there are transitive verbs that do not participate in the voice
alternation system at all (“UV-only verbs”): they only exist in Undergoer Voice, and com-
pletely lack an Actor Voice form. Finally, Ā-extraction is sometimes indeed restricted in
the usual Austronesian “pivot-only” fashion (2a,b), yet other times it is not, and non-pivot
arguments can be extracted (2c–e).
The analysis we will build works, in very simplified terms, as follows. I will argue that
“Undergoer Voice” in Äiwoo is the morphological spell-out of v entering in an agreement
relationship with the object. Only certain classes of objects (to be defined) are agreed
with by v, and not others. When object agreement does not take place, Actor Voice is
inserted as the default/elsewhere form of the verb. The contrast between Actor Voice and
Undergoer Voice in Äiwoowill then be tied to theĀ-system of the language, but – crucially
– only indirectly. Under standard conditions (2a,b), this contrast tracks the distribution of
Ā-features. If the object carries Ā-features, v agrees with it and therefore we get UV;
else, there is no object agreement, and we get AV. These are also the cases where we see
an Ā-extraction restriction. In other words: as long as both voices are an option, each
shows a restricted extraction profile. In other circumstances, however, the system “forces
its hand”. For various reasons to be defined, in these cases the object is always agreed
with. As a result, UV is the only possible outcome, no matter where Ā-features may or
may not be (2c–f ). Precisely in these cases, the Ā-extraction restriction ceases to apply:
now any argument can be extracted (modulo orthogonal reasons). The general contour of
the analysis is schematized informally in (5).
28 1.3. What do we learn?
(5) Informal description of the analysis:
What word order and
Ā-extraction properties
will an Äiwoo clause have?
Sometimes Other times
(to be defined): (to be defined):
Ā-features matter Ā-features don’t matter
If the object has Ā-features: Otherwise: Regardless of which
object agreement no object agreement argument(s) have Ā-features:
always object agreement
Undergoer Voice Actor Voice Always Undergoer Voice
3 Voice alternation 7 Voice alternation
3 Ā-extraction restriction 7 Ā-extraction restriction
= AV, UVplain (2a,b) = UVonly, UVinâ, UVgu, UVSVO (2c–f )
I will argue that the specific incarnation of the Austronesian Ā-extraction restriction
we see in Äiwoo is not due to phasehood (unlike for Rackowski & Richards 2005, Er-
lewine & C. Lim 2023, Hsieh 2025, a.o.), and it is not a “highest DP-only” restriction either
(unlike for Aldridge 2004, 2008, a.o.), but rather it is due toĀ-intervention. In the sub-part
of the systemwhere a voice alternation is available (left side of the schema in (5)), the high-
est argument in a clause will be highest because it carries Ā-features, and therefore it will
intervene for the extraction of any lower argument; this is where the extraction restriction
applies. Viceversa, in those sub-parts of the system where only one voice is possible (right
side of the schema in (5)), the highest argument in a clause will end up in that position
even when it does not carry Ā-features; therefore, there will be no Ā-intervention, and
extraction will be free.
Äiwoo shows us an example of an Austronesian voice system where “voice” is con-
nected to both the A- and Ā-syntax of the language. The voice contrast in Äiwoo clearly
feeds the A-syntax, in the sense that it (partially) controls word order, i.e., where differ-
ent arguments (A-)move. In addition, voice is also influenced by the Ā-syntax of the lan-
guage, seen as it sometimes tracks the distribution of Ā-features on different arguments.
However, this can be “overwritten” by other concerns, in which case voice stops being
reflective of where Ā-features are. The coming and going of the Ā-extraction restriction
Chapter 1. What this dissertation is about 29
in Äiwoo is thus due precisely to the tie between voice and the distribution of Ā-features
being only indirect, and not one-to-one.
1.3.2 THE LOCALITY OF MOVEMENT, SYNTACTIC INVERSION, AND
Ā-EXTRACTION RESTRICTIONS
Zooming out from the Austronesian-specific aspects of our puzzles, Äiwoo also teaches us
something about the locality of syntactic movement in general. A rich literature has dis-
cussed potential connections between two phenomena: “syntactic inversion” or “leapfrog-
ging”, (that is, the A-movement of a lower argument above a higher one), and restrictions
on Ā-extraction (Bittner 1994, Bobaljik 1995, Bittner & Hale 1996a,b, McGinnis 1998, Coon
et al. 2014, 2021, a.o). This has often been investigated in the context of languages with
“syntactic ergativity”, here intended in the narrow sense of a ban on the Ā-extraction of
transitive subjects. This is the case, for example, in “high Absolutive” Mayan languages
(Coon et al. 2014, 2021, Assmann et al. 2015, Royer 2022), Inuit languages (Bittner 1994,
Bittner & Hale 1996a,b, Yuan 2018, 2022, though see Deal et al. 2024 for a contrasting anal-
ysis), and even under some theories of Austronesian voice (Aldridge 2004, 2008), among
others. Abstracting away from a lot of details, the general pattern is schematized in (6).
The object A-moves to a position above the subject. Simultaneously, it is impossible to
Ā-extract the subject. This represents what Deal (2016b) labels the “standard theory of
syntactic ergativity”: the A-movement of the (absolutive) object above the subject some-
how blocks the Ā-extraction of the (ergative) subject.
(6) “Standard theory of syntactic ergativity:”
C[uĀ] … [ Obj (ABS) [ Subj (ERG) v … tObj
non-local
! non-local
3!
7 A-movtĀ-movt
Two aspects of this type of proposal are particularly interesting: the connection between
these two movements, and their (im)possibility (independently from each other). Let us
start from the former issue.
A type of critique that has been raised against this kind of approach is precisely that
this posited connection is not in fact motivated; see e.g. Deal et al. (2024) and references
therein. The presence of the object above the subject should not, by itself, bring about
restrictions on Ā-extracting the subject. This is because canonical Ā-movement does not
need to be strictly local, and can typically skip intervening nominals. This kind of pro-
posed intervention effect would thus require additional assumptions. For example, one
would need to assume that in the relevant languages, Ā-extraction is idiosyncratically re-
stricted to the highest DP (“Ā-probing for the closest DP”; see Branan & Erlewine 2024). In
principle then, these two phenomena need not be connected: one could have (i) “syntactic
inversion” without any restrictions on Ā-extraction, as proposed for West Circassian (Er-
shova 2019, 2022, 2024), or (ii) the opposite, an Ā-extraction restriction with no syntactic
30 1.3. What do we learn?
inversion, as proposed for e.g. Tongan (Otsuka 2006, 2010), Nukuoro (Drummond 2023),
Kalaallisut (Deal et al. 2024), and Northern Tujia (Liu 2025).
Aside from the posited connection between them, these two movements have interest-
ing locality properties in and of themselves, which should be investigated independently.
The first has already been mentioned: we observe a restriction on Ā-extraction. Another
way to frame this is that although Ā-movement should in principle be possible non-locally,
here surprisingly it is not. The second is the opposite kind of problem, that is, the possi-
bility of A-moving the object across the subject. This kind of A-movement has received
a couple handy monickers, e.g. “leapfrogging” (Bobaljik 1995, McGinnis 1998), “syntac-
tic inversion”, etc. Once again, its locality profile is surprising: canonical A-movement
is strictly local and cannot skip intervening nominals, but here, this is possible. In other
words, in a derivation like (6) we have both surprisingly strictly local Ā-movement, and
surprisingly non-local A-movement.
Äiwoo is an interesting case study with respect to this discussion, because it offers a
fine-grained dataset where Austronesian voice, syntactic inversion, and Ā-extraction re-
strictions can be observed operating independently from each other. Consider again the
overview in (7), now augmented with two columns. Within one and the same language,
we see Undergoer Voice both with and without accompanying inversion (7b–e) vs. (7f ),
and simultaneously, we have inversion with or without an accompaning Ā-extraction re-
striction (7b) vs. (7c–e). (The fact that the object is raised past the subject by A-movement
in (7b–e) is demonstrated in chapter 3, §3.7.)
(7) Alternations of word order and Ā-extractability:
Syntactic Ā-extraction
Clause type: Word order:
inversion? restriction?
a. AV: S3 φS-V.AV =TAM O
7 no yes
b. UV 3plain: O V.UV S
7 =TAM yes yes
c. UV 3 3only: O V.UV S =TAM yes no
d. UV NAinâ: O V.UV S
3=TAM yes no
e. UV NA 3gu: O V.UV -πO=TAM S yes no
f. UV NASVO: S V.UV -π =TAM O
NA
S no (NA)
The fact that these phenomena come apart from each other in Äiwoo lets us gain insights
into the particular locality profiles of the different types of movements in question. To re-
peat, the two cases under discussion are surprisingly local (restricted) Ā-movement (7a,b),
and surprisingly non-local A-movement (7b–e). I adopt a conceptually unified approach
to both of these. What we will see is essentially a “it’s just Relativized Minimality all the
way down” type of proposal.
Let us start from restricted/surprisingly local Ā-movement, already discussed in the
previous section. I propose that despite appearances, in Äiwoo it is in fact not “surpris-
ingly local” at all. As mentioned above, I will argue that the Ā-extraction restriction found
UVplain clauses and, more indirectly, AV ones (7a,b) is not a case of Ā-extraction being
Chapter 1. What this dissertation is about 31
restricted to the highest DP, or an effect of phasehood, but an Ā-intervention effect. The
general idea is schematized in (8). The impossibility of Ā-moving a lower argument across
a higher one in (7a,b) is then not due the presence of a higher nominal in and of itself,
but to the presence of Ā-features on the higher argument (8a). The probe triggering Ā-
movement will find the Ā-features on the higher DP1, and stop. The lower DP2 cannot be
reached, and therefore not Ā-moved either. This predicts that when these features aren’t
present on DP1 (7c–f ), the extraction restriction should indeed disappear (8b). Therefore,
theÄiwoo pattern actually respects the canonical locality profile of Ā-movement, based on
Relativized Minimality: constituents carrying Ā-features are interveners for Ā-movement
(8a), but constituents without these features are not (8b).
(8) a. The highest DP has Ā-features: “Ā-extraction restriction”
7
[ probe[uĀ] … [ DP1[Ā] … [ DP2[Ā] …
7!
b. The highest DP does not have Ā-features: no extraction restriction
3
[ DP2[Ā] probe[uĀ] … [ DP1[ ] … [ DP2[Ā] …
3!
The second type of movement with a surprising locality profile is non-local A-movement,
which we will see two instances of. In both, the object moves across the subject: (i) in one
case, to the preverbal position (7b–d); (ii) in the other, to the position between the verb
and the TAM particles (7e). I will argue that this kind of movement should not be thought
of as a case of “mixed A/Ā-movement” (van Urk 2015 et seq.), or at least not in the literal
sense of a probe simultaneously searching for both φ and Ā-features, and that in fact, it
should not be considered a novel or surprising type of movement at all. I argue that this
kind of movement is already predicted to exist by our theory, and its existence in Äiwoo is
thus a welcome confirmation of this prediction. Specifically, its existence is predicted by
the union of two independently motivated theoretical concepts: (i) featurally relativized
probes (Béjar 2003 et seq.), and (ii) the featural theory of the A/Ā-distinction (van Urk
2015), ultimately grounded in featural Relativized Minimality (Rizzi 1990).
The core idea, schematized in (9), is simple, and conceptually isomorphic to the one
presented above for the Ā-extraction facts. If a probe triggering A-movement searches for
[φ], a feature that all nominals have, then every nominal will count as potential goals,
and thus interveners (van Urk 2015 et seq.). This gets us the canonical “no skips” locality
profile of A-movement (9a): because every nominal is a valid goal, only the closest one to
the probe can be agreed with and (A-)moved. However, consider what would happen, in
contrast, if a probe triggering movement is featurally “picky” (9b) – i.e., it is relativized
to a particular feature that only some nominals have, e.g. [PARTICIPANT]. In this case, a
nominal that doesn’t have the relevant feature will not constitute a valid goal. Therefore,
a nominal that is less local to the probe may be agreed with and moved. Depending on
32 1.3. What do we learn?
the specific feature(s) being probed for, this movement may have the binding-theoretical
properties of canonical A-movement despite its laxer locality profile.
(9) a. “Flat” probe: strictly local A-movement (= classic A-movt)
7!
[ DP1[φ, …] probe[uφ] [ DP1[φ, …] … [ DP2[φ, …] …
b. “Picky” probe: non-local A-movement
3!
[ DP2[φ, PART] probe[uPART] [ DP1[φ] … [ DP2[φ, PART] …
3!
Importantly, this is not a case of “mixed A/Ā-movement” in any clear sense: Ā-features
just don’t play any role in this at all. I will show that both cases of non-local A-movement
in Äiwoo are precisely of this kind. Äiwoo then fills a gap in our typology of movement,
validating a particular prediction made by the combination of two pre-existing and inde-
pendently needed theoretical ingredients.
1.3.3 LEXICAL DPS, PRONOUNS, φ-FEATURES, AND NOMINAL LICENSING
The notion of featurally motivated non-local A-movement that we just discussed provides
a good segue to another area of syntactic theory that Äiwoo lets us learn something about.
This is the relation between the internal and external syntax of different types of nominals:
lexical DPs, pronouns, their φ-features, and their licensing needs (or lack thereof).
Let us start from their external syntax. Different corners of Äiwoo grammar make
quite fine-grained contrasts between how they treat different types of nominals. We will
see that v and T make two cross-cutting distinctions. First, v draws a cut between how it
interacts with lexical DPs (nominals with lexical content: “a/the teacher”, “fish”, “Mary”,
etc.) and pronouns. Specifically, v will only agree with lexical DP objects when they carry
Ā-features5, but it will always agree with pronominal objects, whether or not they have
Ā-features. In contrast to v, T is not interested in the categorial distinction between lexical
DPs and pronouns. Rather, T draws an orthogonal cut between nominals that are 3MIN
(≈ 3SG; see chapter 2, §2.5.1), vs. nominals with “marked” φ-features (i.e., anything else
than 3MIN: 1st/2nd person pronouns, and 3AUG pronouns)6. This featural distinction is
at the base of the hierarchy effect that results in one of the two instances of non-local
A-movement of the object across the subject, discussed above.
The two cross-cutting distinctionsmade by v and T are schematized in (10). The crucial
difference between the two groupings lies in how pronominal 3MIN arguments (‘he/she’)
are treated. For the purposes of v, they pattern together with the other pronouns, and
5 As mentioned above, this is actually a simplification; see fn. ??.
6 See the discussion in chapter 6, appendix 6.A for where plural DPs fit into this picture.
Chapter 1. What this dissertation is about 33
unlike lexical DPs. For T, instead, they pattern unlike the other more featurally marked
pronouns, and identically to lexical DPs.
(10) Groups of nominals as treated by v and T:
Distinction made by v:
lexical DPs vs. pronouns
3MIN 1st/2nd/
Lexical DPs
pro. 3AUG pro.
Distinction made by T:
“unmarked” vs. “marked” φ-features
These two orthogonal cuts, i.e. the way v and T interact with the different groups of
nominals, is what ultimately results in the complex word order alternation pattern at the
core of one of our puzzles. In other words, to know what an Äiwoo transitive clause will
look like – in terms of the typology of clause types set up in (7): their word order, verbal
morphology, and Ā-extraction possibilities – what we need to know is the categorial and
φ-features of both arguments7.
Of course, word order alternations based on distinctions between different types of
nominals are nothing new. Consider, for example, Object Shift in Mainland Scandinavian
languages (where pronominal objects obligatorily shift, but lexical DP ones do not), or
object cliticization in Romance languages (where – simplifying extremely – pronominal
objects obligatorily cliticize onto the verb, while lexical DP objects remain postverbal)8. A
classic type of explanation for these kind of patterns centers around the idea that pronom-
inal elements (clitics, “weak” pronouns, etc.) are “deficient” in certain ways, and therefore
they must move to fix this deficiency in order for the derivation to converge (Roberts &
Shlonsky 1996, Cardinaletti & Starke 1999, Holmberg 1999, Koopman 1999, a.o.). This kind
of explanation is based on the concept of “Greed” (Chomsky 1993, 1995): these pronouns
have certain particular needs and must move to meet these needs (these could be prosodic
needs, a need for licensing/abstract Case assignment, a need to escape the domain of ex-
istential closure, etc.). Crucially, in this kind of approach, lexical DPs don’t have the same
kind of needs, and thus they don’t move in the same way.
7 This makes the Äiwoo word order alternation pattern reminiscent of “global case splits” (Keine 2010, Georgi
2012, Bárány 2017, Yuan 2020, Bárány & Sheehan 2024, Clem & Deal 2024, a.o.). These are systems where in
order to correctly predict the case properties (or agreement profile, or final landing site, etc.) of a particular
argument in a clause, one must know the properties of multiple arguments. Äiwoo does not have any mor-
phological case (see chapter 2, §2.5), but its word order alternations can be cast in the same kind of logic:
“iff the subject has [properties α] and the object has [properties β], then the clause will be of [type γ]”. See
chapter 6 for discussion.
8 Word order alternations based on φ-features are rarer, but they are also attested; see chapter 3, §3.7.3.
34 1.3. What do we learn?
This type of approach – and the general notion of Greed – has been criticized (Sichel
2001, 2002, Sichel & Toosarvandani 2020, 2024, a.o.). Under Greed-based analyses, the
movement of these pronouns takes place for the sakes of the moving element (i.e., the pro-
nouns’ particular requirements). In contrast, many other movement phenomena (e.g. wh-
movement, etc.) are known to take place for the sakes of the landing position, not of the
moving element. Consider for example wh-movement in English. In general, a wh-word
moves to spec,CP, but in multiple wh-questions only the highest wh-word will raise, and
all others remain in situ. This shows that wh-words have no inherent need to move in and
of themselves: they will move if something makes them, but nothing will necessarily go
wrong if they happen not to move. This, Sichel (2001, 2002) and Sichel & Toosarvandani
(2020, 2024) argue, is a problem of theoretical parsimony: we have two different reasons
why movement may be triggered, but all movement seems to otherwise obey the same
constraints (locality, the Extension Condition, etc.). Therefore, it would be better if we
could do with just one, and they propose that the various attested patterns where pro-
nouns move differently from lexical DPs can be reanalyzed without any Greed, but purely
in terms of probe-goal attraction mechanisms – like all other movement.
From the perspective of this debate, my analysis of the Word Order Puzzle in Äiwoo
supports this latter line of work, as a sort of proof of concept: even for a rather intricate
dataset involving different intra-clausal movements, a fully Greed-free analysis is not only
possible, but better than a Greed-based one. In fact, strikingly absent from the whole
model I build is the concept of nominal licensing (Vergnaud 1977/2008, et seq.). This notion
(however one wants to implement it) does not seem to play any particular role in Äiwoo
syntax. We will see that depending on the type of clause, both subjects and objects may
surface in any of the three available argument positions (preverbal, between the verb and
the TAM particles, and postverbal). This is shown in (11)–(12); the structural labels and
traces will be motivated throughout the following chapters.
(11) The subject may end up surfacing in any argument position: (250728)
a. [CP ijii ku-lu-tou-mä [TP tS=kaa [vP tS sii
3AUG IPFV-3AUG-bring.AV-DIR1 =FUT fish
‘They will bring fish’
b. [CP sii ku-tu-mä [TP Jon =kaa [vP tS tO
fish IPFV-bring.UV-DIR1 John=FUT
‘John will bring the fish’
c. [CP ( iumu) ku-tu-mä [TP -gu-mu=waa [vP Jon tO
2MIN IPFV-bring.UV-DIR1 -OBJ-2MIN=FUT John
‘John will bring you’
(12) The object may end up surfacing in any argument position:
a. [CP sii ku-tu-mä [TP -i=laa [vP tS tO (250728)
fish IPFV-bring.UV-DIR1 -3AUG=FUT
‘They will bring the fish’
Chapter 1. What this dissertation is about 35
b. [CP nuwopa to ki-te-kä [TP -gu-i =laa=kä [vP Mary tO
house my IPFV-see.UV-DIR3 -OBJ-3AUG=FUT=CV Mary
‘Mary will see them in my house’9 (240724)
c. [CP ( iumu) ku-tu-mä [TP -mu=waa [vP tS ijii (250728)
2MIN IPFV-bring.UV-DIR1 -2MIN=FUT 3AUG
‘You will bring them’
Above, we discussed how wh-words may remain in situ if their movement happens to
be blocked by something else, and how this shows that they have no inherent need to
move anywhere; if they happen not to move, everything is fine. The data in (11)–(12)
makes the same point for arguments of transitive clauses in Äiwoo. A subject or an object,
whatever its category or φ-features, does not need to move to any particular position for
the derivation to converge10. In other words, this is a system where nominals have no
apparent requirements of their own to be agreed with or moved (or licensed). Rather,
nominals just get moved around by different probes, each with particular featural desires.
Whatever does not get moved may either be moved later on by a higher probe, or remain
where it is, without further consequences.
Aside from the external syntax of different types of nominals in Äiwoo, their internal
structure is also interesting in its own respects. Particularly, I will focus on the internal
syntax and morphology of pronouns. For example, a 3AUG pronoun is realized as the stan-
dalone word ijii in the preverbal position (11a) or the postverbal one (12c). However, in
the position between the verb and the TAM particles we don’t see the full pronoun form,
but rather a suffix -i (12a), or -gu-i when this pronoun is an object (12b). These forms are
part of a full paradigm of verbal suffixes that look like a shorter/reduced version of the
corresponding full pronoun; compare e.g. the full 2MIN pronoun iumu to the 2MIN suffix
-mu (12c). Moreover, we have cases where both the full pronoun and the corresponding
suffix are realized simultaneously (11c)–(12c).
In this domain, Äiwoo contributes to the theoretical discussion revolving around pro-
nouns, clitics, and agreement. A rich literature has inquired into how to determinewhether
a particular morpheme in a given language is an instance of one or another of these cate-
gories (Woolford 2008, Nevins 2011, Kramer 2014, Baker & Kramer 2018, Yuan 2018, 2020,
2021, a.m.o.). Moreover, it has also been debated whether a morpheme’s morphophono-
logical properties are sufficiently informative to diagnose its underlying syntactic nature,
with both positive answers (Zwicky & Pullum 1983, and partially Nevins 2011, a.o.) and
negative ones (Yuan 2021, Akkuş et al. 2025, a.o.).
I argue that Äiwoo supports the latter camp: a morpheme’s morphophonology cannot
be taken as decisive evidence for a particular syntactic analysis. In other words, “looks
9 Sentence (12b) is an example of a clause in “Circumstantial Voice”; I won’t treat this kind of construction
in this dissertation (see Appendix A). However, picking this was necessary to show an example of a clause
where the object is only surfacing in the position between the verb and the TAM material, but not also
further up in the preverbal position, as in (11c).
10 These six sentences (11)–(12) do not exhaust all the possible combinations of different types of subjects/
objects. They were just picked for the sake of exposition. See chapter 6.
36 1.4. Data and methodology
can be deceiving”. In chapter 5 I will argue that despite these suffixes’ morphophonol-
ogy, they are syntactically full pronouns, that occupy a specifier position (as indicated
by the bracketing); they do not spell out agreement, and they are not clitics. The reason
for the full form vs. suffixal contrast, I propose, is purely morphological. A pronoun in
the position between the verb and the TAM particles undergoes obligatory post-syntactic
Local Dislocation to its left, becoming thus prosodically incorporated into the verb word.
This step of Local Dislocation may also end up forcing the overt pronunciation of a lower
copy in a movement chain, which would have otherwise been deleted. This accounts for
the cases of “doubling”, where the same argument is pronounced simultaneously as a full
pronoun and as a suffix (12c).
1.4 DATA AND METHODOLOGY
The Reef Islands, where Äiwoo is spoken, are a rather remote location (see chapter 2,
§2.2), and traveling there involves considerable logistical difficulties (sheer distance and
cost, poor infrastructure, unreliable flight and boat connections, etc.). As a result, the kind
of targeted elicitation that is necessary for the type of research in this dissertation was
simply not possible until recently, when the internet connection on the islands improved
and remote fieldwork over Zoom became possible. The bulk of the data in this dissertation
was thus collected through fieldwork with a native Äiwoo speaker, Luke Gitakulu.
Luke is from the Otelo village in the Reef Islands, and currently lives on the islands.
For several decades, Luke has been a language expert, collaborator, and consultant on
a variety of projects related to linguistic research, documentation, and translation. He
has earlier collaborated with Åshild Næss during her fieldwork in the Reef Islands (see
chapter 2, §2.3), and is collaborating on translating the Bible to Äiwoo. In addition to
Äiwoo, Luke also speaks English and Pijin/Solomon Islands Pidgin; the medium language
in the fieldwork process was English. Luke is also literate in Äiwoo, which is uncommon
in the community (see chapter 2, §2.2)
Starting in May 2023, Luke and I have been holding approximately weekly meetings
on Zoom, of one hour length; every meeting was recorded. Luke was compensated $ 32
USD an hour, plus the price for the internet connection at the local internet café in the
Reef Islands. The meetings involved a screen-shared Google Docs document, where Luke
could see what I was writing, both in English and in Äiwoo, and correct my transcrip-
tions. After the meeting, the resulting file would be tagged for what linguistic topics were
covered (e.g. relativization, word order, pronominal morphology, voice alternations, etc.),
and exported as a .pdf file for archival. Then, I created a copy of the raw file where I would
provide interlinear glosses for every sentence; this was also archived as a .pdf file. Both
the raw and annotated transcription files, in addition to the Zoom recordings, are stored in
a Google Drive folder. With Luke’s consent, this folder has been made available to Åshild
Næss and her collaborators, to aid with their research on Äiwoo.
Chapter 1. What this dissertation is about 37
The methodology employed in our meetings was the relatively standard variety of
tasks used in linguistic elicitation-based fieldwork. These included translation tasks (from
English to Äiwoo and viceversa), acceptability judgments in a particular context or for
a particular reading, etc. One limitation of this dissertation is that the Äiwoo data was
collected in collaboration with only one speaker (namely Luke); for logistical reasons,
the plan of consulting more speakers had to be postponed. To partially make up for this,
every judgment was replicated across multiple sessions and with multiple lexicalizations.
Throughout the dissertation, every sentence (or group of sentences) I elicited is labelled
with a six digit code, representing the date of elicitation in YYMMDD format; the label will
be placed wherever is most typographically convenient, and whether it refers to a single
sentence of a group of sentences should hopefully be clear in the relevant contexts. For
example, both sentences in (13) were elicited on June 22, 2023. The only exceptions will
be examples consisting of a single word, reported in isolation; these will be unlabelled, as
they are amply attested in the corpus.
(13) [Context: a house has just been sold.] (230622)
a. i-ve ibesi
ASP-buy.UV friend.1MIN
‘My friend bought it’
b. ?* ibesi i-veve elicitation date
(YYMMDD)
friend.1MIN ASP-buy.AV
Intended: ‘My friend bought it’
In addition to my own fieldwork, twomore data sources were consulted. One is the corpus
collected by Åshild Næss over several fieldwork trips from 2004 to 201511. This consists of
recorded and transcribed natural speech by a number of Äiwoo speakers, including Luke.
The corpus contains a 72 texts of widely varying length (from 82 to 8079 words), for a
total of about 72 000 words. The texts are mostly of narrative and procedural genre (i.e.,
explaining how to do something, e.g. a particular fishing technique, building a house, etc.).
The second source is the Äiwoo translation of the Gospels of Mark, Luke, and Matthew,
developed by John Rentz (at SIL) in collaboration with Luke and other Äiwoo speakers.
Sentences from the Bible translation are marked with a label indicating the book, chapter
and verse of origin, e.g. “MK 9:41” for Mark, chapter 9, verse 41 (“MK” = Mark, “LU” =
Luke, “MT” = Matthew). Sentences from Næss’ corpus are labelled with the name of the
text they belong to and their line index, e.g. “Nyeliwânubona toponu 141”.
11 The data collection that led to the creation of this corpus was funded by the Research Council of Norway,
grant no. 148717, and the Endangered Languages Documentation Programme, grant no. SG0308. I kindly
thank Åshild Næss for making it available to me.
38 1.5. Theoretical assumptions
1.5 THEORETICAL ASSUMPTIONS
The theoretical framework this dissertation assumes is a relatively standard generative
one, generally based on the Minimalist Program (Chomsky 1995, 2000, 2001) combined
with a strictly non-lexicalist model of word building (i.e., DistributedMorphology; Halle &
Marantz 1993, Marantz 1995, 1997, Harley & Noyer 1999). I want to focus this section on
two particular aspects of the theoretical apparatus that will be important in the develop-
ment of the model of Äiwoo syntax I propose.
The first is a relatively standard assumption within Minimalis, which Deal (to appear)
labels the “Merge-Agree Hypothesis”, that is, the idea that the only two syntactic oper-
ations in natural language are Merge and Agree. The former applies only very locally,
whereas the latter regulates dependencies between elements at a distance. This goes to-
gether with (or rather, entails) the idea that all syntactic (phrasal) movement is feature-
driven, that is, it is always triggered by a step of Agree (14)12. Formally, “movement” is
then just a shorthand term for the sequence of operations Agree + (Internal/re-)Merge.
(See below for a discussion of how exactly Agree works.)
(14) All movement is always preceded by Agree:
FP
XP
F …
PROBE: [X]
…
XP …
Within Minimalist literature, this is nothing new, but rather a widely shared and de facto
standard assumption13. However, I take a strict interpretation of this idea to entail that
some typical descriptions of movement phenomena cannot be held to happen literally as
described, at least in a mechanic sense. For example, under a strictly feature-driven view
movement, a DP cannot move to a particular position “to get (abstract) case/Case”, or “to
escape from its phase” – at least not in the literal, teleological sense of these statements.
Mechanically speaking, a constituent is only ever moved because a probe agreed with it.
This concretely means that there must exist a particular featural configuration that makes
this probe-goal relationship possible, crucially respecting relativized minimality.
12 I remain agnostic as to whether head movement is also feature-driven, and how exactly it is implemented.
See Harley (2013) and references therein for discussion.
13 However widely held, this assumption has not been free of critiques. Chomsky (2000) already raised the
possibility that maybe not all movement was feature-driven, mentioning stylistic fronting as a potential
exception; see also Syed & Simpson (2024) for a concrete proposal.
Chapter 1. What this dissertation is about 39
This theoretical context makes it important to precisely define exactly how Agree
works. In this dissertation I adopt the interaction-satisfaction theory of Agree (Deal 2015b,
2024a, to appear); see e.g. Clem (2019, 2022, 2023), Roversi (2020), Oxford (2024), Branan &
Erlewine (2024), Scott (2023), Clem & Deal (2024), and Yuan (2024), a.o., for various appli-
cations. In this model, every probe comes with two specifications: its interaction set, and
its satisfaction set. The interaction set specifies what feature(s) the probe “searches for”:
upon finding a constituent with at least one of these features, the probe interacts with this
constituent, and copies back its features14. The satisfaction set specifies what feature(s)
halt the probe: upon finding a node in the tree with at least one of these features, the probe
is satisfied, and stops its operations. Interaction and satisfaction are independent of each
other: a probe may interact with a goal and not be satisfied, or it might be satisfied by
a goal without interacting with it. This latter scenario would result in the probe halting
with no features being copied (at least from the specific goal that satisfied it).
Three more parts of the Agree mechanism need to be discussed: movement, and goal
flagging. As for movement: any feature in the interaction or satisfaction set may be fur-
ther specified as triggering movement (notated with a diacritic M). If a probe finds a target
carrying a feature FM that is part of either its interaction or satisfaction set, that target is
moved to the probe’s specifier. This could be informally thought of as an equivalent of
an “EPP feature”. As for goal flagging, this is the name given in Deal (to appear) to the
structural change that a probe causes on a goal it either interacted with or was satisfied
by. This idea (that upon Agree, information is transferred both from the goal to the probe
but also from the probe to the goal) is not new, and was already present in the theory since
Chomsky (2001) and Pesetsky & Torrego (2001, 2007). Concretely, if a probe H interacts
with a constituent XP, XP receives an “interaction flag feature” [i|H]; if a probe H is sat-
isfied by a constituent XP, XP receives a “satisfaction flag feature” [s|H]; this notation is
adopted from Deal (to appear)15.
Before proceeding to an example, some notational remarks. All agreement relations
(successful or not) are notated in trees with a dashed line (solid arrow lines notate move-
ment). For probes, Deal (2015b, 2024a, to appear) uses the notation [INT: α, SAT: β], but
in the trees I shorten this to just [I: α, S: β] for typographical reasons16. Furthermore, I
notate interaction and satisfaction independently on the dashed lines. If interaction takes
place, this line ends in a full circle (X Y). If there is no interaction, it ends in an open
14 I follow Grishin (2023) and Deal & Royer (2025) – though departing from some of the literature cited above
– in assuming that when a probe interacts with a goal, all of the goal’s features are copied onto the probe,
and not just those that triggered the interaction.
15 See Clem & Deal (2024) for further developments of this idea, where a goal can receive features via flagging
that a goal had copied onto itself earlier from a different constituent. This enrichment is not necessary for
the model of Äiwoo syntax I build, so I don’t include it.In general, Äiwoo syntax doesn’t seem to manipulate
satisfaction flag features at all, so I will not include them in the trees.
16 For the same reason, when the two sets have the same featural specification, I will use the condensed nota-
tion [I+S: α]. Note that if the interaction and satisfaction specifications are identical and there is movement,
[I: αM, S: α] is empirically indistinguishable from [I: α, S: αM]. I notate both as [I+S: αM].
40 1.5. Theoretical assumptions
circle (X Y). If a probe is satisfied by a target, this is shown by adding a bar before
the circle (X Y, X Y; respectively with/without interaction).
Consider now how this Agree mechanism works with the following abstract example
(15). The probe H is specified as having the feature [F] as its interaction feature, and the
feature [G] as its satisfaction feature; furthermore, its interaction feature carries the move-
ment diacritic M. Immediately after it is merged, H starts a search, examining each node in
its c-command domain17. XP doesn’t have either [F] nor [G], so it is not interacted with
and it does not satisfy the probe; nothing happens, and the probe continues searching18.
Further down, the probe finds YP; this constituent carries [F], the interaction feature of
the probe. Three things happen: (i) every feature from YP is copied onto the probe (⟨{F,
K}⟩); (ii) YP receives the interaction flag feature [i|H]; (iii) because the interaction feature
is specified with a movement instruction, YP moves to spec,HP. However, YP does not
have the probe’s satisfaction feature [G], so the probe still continues searching. Further
down the probe finds ZP, which carries both [F] and [G]. Because of [F], the probe inter-
acts with ZP, repeating the steps described above19. Finally, because of [G], the probe is
satisfied, and it stops its search altogether. The lower potential goal WP is never reached.
See also Deal (to appear) for a more formally described algorithm.
(15) HP
ZP
[F, G, i|H, s|H]
YP
[F, K, i|H]
[ H⟨{F, K}, {F, ]G}⟩
I: [F]M
S: [G] XP
[K]
YP
[F, K]
ZP
[F, G]
WP
[F, G, K]
This tree shows three possible outcomes: no interaction/no satisfaction (H–XP), yes in-
teraction/no satisfaction (H–YP), yes interaction/yes satisfaction (H–ZP). The fourth pos-
sibility is no interaction/yes satisfaction, shown in (16). Here, the probe H does not copy
17 See Branan & Erlewine (to appear) for a formal definition of the search procedure.
18 Strictly speaking, this is true for every node in the tree aside from YP andWP, including maximal projections,
intermediate bar-level projections, etc. I only notate this for XP for simplicity.
19 Here I show ZP as moving to an outer specifier above YP, i.e. not tucking in; this is not important, and
tucking in could also be assumed.
Chapter 1. What this dissertation is about 41
anything back from XP (because it doesn’t have [F]), but it is satisfied by it (because it
has [G]). Therefore, the probe stops at XP, and YP is never reached. (The satisfaction flag
feature [s|H] is added onto XP after the Agree step).
(16) HP
[ H⟨∅⟩ ]
I: [F]M
S: [G] XP
[G, K (+ s|H)] YP
[F]
In the trees throughout the dissertation, for simplicity, I will only notate the features that
are relevant to the discussion on hand. For example, even though a probe copies all fea-
tures from a goal upon interaction (e.g., ⟨{φ, N, D, PL, FOC, …}⟩, I will just write ⟨φ⟩ if this
is the relevant feature under discussion. Similarly, I will only notate flag features when
relevant, and ignore them otherwise.
One final theoretical remark is that in this theory of Agree there is no Activity Condi-
tion (Chomsky 2000, 2001). This means that multiple probes can agree with the same con-
stituent without any issues, as long as the featural relations in place allow for it. I consider
this an advantage, as the Activity Condition has been shown to face severe empirical chal-
lenges, such as the existence of hyperraising (Rodrigues 2004, Nevins 2004, Halpert 2019,
Fong 2019, Zyman 2023, a.o.), of cases where multiple heads show φ-agreement with the
same DP (Carstens 2001, 2010, 2011, Polinsky & Potsdam 2001, Bhatt 2005b, Carstens &
Diercks 2013, Deal 2015a, a.o.), and cases where a single DP is assigned multiple cases
(Bejar & Massam 1999, Richards 2013, Pesetsky 2013, a.o.). See also Deal (to appear) for a
discussion of the Activity Condition.
1.6 OVERVIEW OF THE DISSERTATION
As presented above, the main empirical goal of this dissertation is to build a model of
the syntax of Äiwoo clauses that accounts for both the attested word order alternation
patterns (the “Word Order Puzzle”), and their partial correlation to the patterns of Ā-
extractability (the “Extraction Puzzle”). The data at the base of this discussion is rather
rich and complex, with several of moving parts. This presents a problem for the exposi-
tion: in a sense, to understand any part of themodel, onewould need to know every part of
the data. As a consequence, there will be several points throughout the dissertation where
a particular analytical move may seem too complicated, unmotivated or unjustified given
the data under discussion at that point, but it will ultimately be necessary to accomo-
date the rest of the complexity of the system (which has not been presented yet). This
42 1.6. Overview of the dissertation
will be pointed out whenever relevant throughout the discussion. Below, I give succinct
summaries of the different chapters.
CHAPTER 2 — “THE ÄIWOO LANGUAGE: BACKGROUND” Here I present some back-
ground information about the Äiwoo language. This includes facts about its geography,
its speaker population, and its phylogenetic position within the larger Austronesian lan-
guage family. I also include a very simple description of its phonology and some salient
phonological/morpho-phonological patterns. Finally, I present some of the morphosyntax
of the nominal domain and the verbal domain; for the latter, this chapter covers certain
aspects that are not discussed elsewhere in the dissertation.
CHAPTER 3— “ÄIWOO CLAUSE STRUCTURE” This is the first of the fourmain contentful
chapters, where we will actually develop answers to our main puzzles. The focus of this
chapter is addressing the Word Order Puzzle, for the time being limiting ourselves to
the “canonical clauses”, in Actor Voice and Undergoer Voice – that is, the first two rows
of the table in (7). Part of the core questions this chapter answers are the classic ones
that a model of the clausal syntax of any language must answer: where are the different
arguments, how did they get there, and why? Moreover, specifically for Äiwoo, we must
have an answer to why a verb can surface with different verb stems, andwhy φ-agreement
seems to only be present in one case but not the other.
At a more descriptive level, I will argue for the following novel claims: (i) Äiwoo is
a V2 language; (ii) the preverbal position for arguments is spec,CP; (iii) the position for
arguments between the verb and the TAM material is spec,TP; (iv) the stem alternation is
governed by a system of object agreement: if v agrees with the object, the verb stem will
be spelled out in a particular form (“Undergoer Voice”), whereas if this does not happen,
the verb stem will be spelled out in a default form (“Actor Voice”); (v) v agrees with the
object if the latter carries Ā-features.
At a more theoretical level, an important claim is that in Äiwoo, either the subject (in
Actor Voice) or the object (in Undergoer Voice) are moved to spec,CP through potentially
non-local A-movement. I propose that this is not an instance of “mixed A/Ā-movement”, in
the sense that there is no Ā-ingredient to this movement at all, other than the fact that it
has a less strict locality profile than typical A-movement (C can raise the object to spec,CP,
skipping the subject). As mentioned earlier (§1.3.2), I will argue that this kind of non-local
A-movement is not an innovation, but that its existence is in fact already predicted by the
union of two pre-existing parts of our theory, both independently motivated and needed:
the notion of featurally relativized probes, and Relativized Minimality.
CHAPTER 4 — “THE Ā-EXTRACTION RESTRICTION, AND THE LOGIC OF AN EXCEPTION”
After addressing part of theWord Order Puzzle by establishing basic derivation for canon-
ical AV and UV clauses, we move on to addressing the first part of the Extraction Puzzle.
Here, I cover the top half of our overview table (7a–c): AV, UVplain, and UVonly clauses.
This chapter’s goal is to answer two questions: (i) the Austronesian Ā-extraction restric-
tion: why can only the subject be extracted from an AV clause, and only the object from
Chapter 1. What this dissertation is about 43
a UV clause? (ii) why is it exceptionally possible to extract the subject of a UV clause in
the presence of certain particular verbs (“UV-only”)?
To briefly repeat what was presented in §§1.3.1–1.3.2, I will argue that the Äiwoo
manifestation of the Austronesian Ā-extraction restriction is a case of Ā-intervention. In
AV and UVplain clauses, the highest argument (respectively the subject and the object)
carries Ā-features, and these will intervene for the Ā-extraction of any lower argument.
I argue that “UV-only” verbs idiosyncratically select for a different vH head. This vH is
different from ordinary v in that it will always agree with the any object, regardless of
Ā-features; this is why these verbs simply do not have a AV form. In this case, the entire
derivation will proceed as for a canonical UVplain clause, but in the end, the object in
spec,CP will not carry Ā-features. Therefore, it will not pose an intervention problem,
and the subject may thus be exceptionally extracted.
CHAPTER 5 — “THE φ-INDEXING SYSTEM” This is a descriptive detour from the main
narrative, focusing on the two paradigms of φ-indexing affixes: the φ-prefixes found on
AV verbs (and intransitives), and the φ-suffixes found on UV verbs. Contra the existing
descriptive literature, I will argue that these morphemes’ similar morphophonological
properties (they are all affixes, though on opposite sides) hide a fundamental syntactic
difference. On one hand, the φ-prefixes are pure agreement markers. In contrast, the φ-
suffixes are not agreement nor clitics, but they are amorphologically reduced spell-out of a
pronoun occupying spec,TP (see §1.3.3 above). Establishing this fact is necessary to really
understand the full empirical dimension of the Word Order Puzzle: without knowing the
syntactic nature of these affixes, we cannot determine where the different arguments are
in the different clause types.
CHAPTER 6 — “NON-CANONICAL UV CLAUSES” We now know enough to tackle the
more complex part of the Word Order Puzzle, that is, the system of intricate word order
alternations found in UV. The first part of the chapter is descriptive in character, go-
ing through the different clause types and clearly exposing their word order properties.
I then move to the Ā-extraction aspect: these clauses all show exceptions to the classic
Ā-extraction restriction. I will show that this is not caused by their special word order,
but rather, it is a result of the object in these clauses always being a pronoun. This is
supported by the existence of clauses (“UVinâ”) that have the same word order as UVplain
clauses, but without an extraction restriction.
To derive the word order, verbal morphology, and Ā-extraction properties of all of
these clauses I will makeminimal amendments to the probes posited for v and T in chapter
3. First, I propose that v agrees with QP objects and pronominal objects. In this respect, Äi-
woo is brought closer to many languages where these two categories of objects are treated
alike by some syntactic phenomena: for example, both Ā-marked objects and pronominal
objects often shift to a higher position in the vP. Because a pronominal object is agreed
with by v (thus triggering UV) whether or not it has Ā-features, we derive the same sus-
pension of the Ā-extraction restriction that we see with UV-only verbs. Then, I revise the
probe on T to account for the hierarchy effect we observe in the calculus determining
44 1.6. Overview of the dissertation
what moves to spec,TP. Like C, T also can trigger non-local A-movement, raising the ob-
ject to spec,TP across the intervening subject if it has the right features that the probe on
T is interested in. Finally, to recapitulate, I offer annotated step-by-step derivations of all
clause types.
CHAPTER 7 — “CONCLUSION AND PROSPECTS” The title of this chapter is as self-ex-
planatory as it could possibly be. I first summarize the main findings of the dissertation
and their theoretical implications, and then present some loose threads that I hope further
research will be able to shed more light on.
APPENDIX A — “NOTES ON CIRCUMSTANTIAL VOICE” In this appendix I offer some
preliminary notes on a third “voice” construction in the language. I discuss how it could
be integrated into the analysis proposed in this dissertation, and where the model breaks,
making wrong predictions.
APPENDIX B — “A “CASE-AGREEMENT” ANALYSIS” In this second appendix I take in-
spiration from analyses of Tagalog like Rackowski (2002), Rackowski & Richards (2005),
and Hsieh (2020), and I sketch what an analysis of the Äiwoo system in terms of “case
agreement” would have to look like. Ultimately, I find this analysis needlessly more com-
plex than the one I develop in chapter 3, to no clear empirical gain.
CHAPTER 2
THE ÄIWOO LANGUAGE:
BACKGROUND
2.1 INTRODUCTION
This chapter contains an introduction to the Äiwoo language. Part of the goal is to try
to cover the facts that are essential for the reader to follow the rest of the dissertation.
Moreover, due to Äiwoo’s under-researched status, I also attempt to include extra infor-
mation and observations about the language that would otherwise not fit elsewhere in
the dissertation. In a sense, parts of this chapter are also intended to serve as a (small)
grammar sketch, since there is none published at the moment. I hope this will be useful
and interesting to other linguists, both those specifically focusing on Oceanic languages
and to a more general scholarly audience.
First (§2.2), I present some background about the geographic, linguistic, and phylo-
genetic situation of Äiwoo. Then (§2.3) I cover the small amount of published literature
about this language. In §2.4 I offer a very simple description of the phonology of the
language, and of some salient morphophonological phenomena. Finally, §§2.5–2.6 cover
some morphosyntactic facts pertaining to the nominal and verbal domain respectively.
For a reader who is only interested in what is strictly necessary to follow the rest of the
dissertation, I would advise focusing on §2.5.1 and 2.6.
2.2 GEOGRAPHIC AND LINGUISTIC SITUATION
Äiwoo (ISO 639-3 code [nfl]) is an Oceanic language spoken in the Reef Islands, in the
Temotu province of the Solomon Islands, in the southwest Pacific Ocean (see figures 2.1–
2.2). There is also an Äiwoo-speaking diaspora community in Honiara, the country’s cap-
46 2.2. Geographic and linguistic situation
ital (Boerger et al. 2012, Sauppe et al. 2023). Earlier literature also referred to this language
as “Äiwo”, “Äiŷwoo”, or “Reefs”.
Du� Islands
Reef Islands
Santa Cruz
Bougainville
Papua New Guinea Solomon Islands
M  e  l  a  n  e  s  i  a FijiVanuatu
Australia
20°
0 250 500
Kilometres New Caledonia
150° 160° 170° 180°
Figure 2.1: The Reef Islands in their geographical context, in the window in the
north-east corner (map elaborated from Ross & Næss 2007: 457).
Figure 2.2 zooms in on the Reef Islands. This archipelago consists of a chain of coral
islands arranged around a coral reef (te akau loa ‘the long reef’ in its Vaeakau-Taumako
name; see below). The eastern part, the Main Reefs, consists of larger islands, and this is
where Äiwoo is spoken. The Outer Reefs are smaller and more scattered low-lying atolls,
and are poorer in natural resources.
Nukapu The Reef Islands
0 km 5
Pileni
Outer Reefs Makalumu Nifiloli
Te Akau Loa Main
Fenualoa Reefs
Lomlom
Nupani
Naloko Matema Ngawa
0 2
km
Figure 2.2: The Reef Islands (Næss & Hovdhaugen 2011: 2).
Chapter 2. The Äiwoo language: background 47
Five more Austronesian languages are spoken in this area, with various degree of
linguistic contact with Äiwoo. The geographically closest neighbor is Vaeakau-Taumako
[piv] (previously referred to as “Pileni”), a Polynesian outliner – i.e., also Oceanic/Austro-
nesian, but only very distantly related to Äiwoo (see figure 2.4 below for the language
family tree). Vaeakau-Taumako is spoken in the Outer Reefs and in the Duff Islands, about
100 km to the northeast of the Reef Islands (shown in figure 2.1). Äiwoo has a large number
of Vaeakau-Taumako loanwords, indicating a long history of language contact (Næss &
Jenny 2011)1. On the island of Santa Cruz, about 70 km south of the Reef Islands, four
languages are spoken: Natügu [ntu], Nalögo [nlz], Noipä [npx] and Engdewu [ngr]. The
geographic distribution of these languages is shown in figure 2.32.
Vaeakau-
Taumako
Äiwoo
Reef Islands
Santa Cruz
Natügu
Solomon Islands
Engdewu
Nalögo
Nalögo Engdewu
Figure 2.3: Languages spoken in the Reefs Island and Santa Cruz
(map elaborated from Boerger et al. 2012: 112)
The four Santa Cruz languages are the most closely genetically related to Äiwoo and
show several similarities in their grammar, although they are not mutually intelligible
with it. Natügu, Nalögo, and Noipä are somewhat closer to each other, forming a dialect
continuum, but they are not mutually intelligible with each other; Engdewu is more dis-
tant from the other three. Finally, Pijin (Solomon Islands Pidgin) and to a lesser degree
1 The first archeological signs of Polynesian settlements in the area date to around 1 200 CE (Næss & Jenny
2011). In contrast, present-day Äiwoo speakers are assumed to descend from a group that arrived to the
Reef Islands at least 3 200 years ago (the “Lapita cultural complex”), likely from the Bismarck Islands, the
area where Proto-Oceanic has been argued to have been spoken around 3 500 years ago (Spriggs 1997).
2 This map does not include Noipä, as it was only identified in 2015 (Boerger 2017). According to Alfarano &
Boerger (2025: fn. 3), Noipä is only spoken “near the coastal intersection of Natügu and Nalögo in a single
village of fewer than fifty adult speakers”. In the map in Boerger et al. (2012), the name “Nagu” was used for
the Engdewu language; this was updated here according to the preferences of the community (Vaa 2013).
48 2.2. Geographic and linguistic situation
English are also spoken by parts of the local population. See Boerger et al. (2012) for more
details.
Äiwoo is the biggest language in its region in terms of number of speakers. The most
recent census (from 2019) reports 9 632 speakers (Solomon Islands National Statistics Of-
fice 2023: 113). Earlier, the number of speakers was reported to be around 8 400 (Næss
2006, Ross & Næss 2007, et seq.), so the speaker population is increasing. Table 2.1 reports
estimated numbers for the speaker populations of all the Austronesian languages spoken
in this area (i.e., all languages mentioned above minus English and Pijin)3. These figures
are based on the 2019 census for Äiwoo, Alfarano & Boerger (2025) for the four Santa Cruz
languages, and Næss & Hovdhaugen (2011) for Vaeakau-Taumako.
Table 2.1: Approximate speaker populations of the
Austronesian languages of the Reef Islands and Santa Cruz
Language Number of speakers
Äiwoo 9 632
Natügu 4 000–5 000
Nalögo 1 600
Engdewu 200
Noipä 40
Vaeakau-Taumako 1 600
Being the largest language in the area, Äiwoo is “reportedly expanding at the expense
of other local languages, and as such it appears to be in no immediate danger of extinction”
(Boerger et al. 2012: 145). According to the same source, Äiwoo is consistently used at
home and at village-level meetings and gatherings, and for announcements at church,
although sermons and service are held in English and Pijin (a Bible translation to Äiwoo is
in progress). The intergenerational transmission seems to be robust at the time of writing,
with children in the Reef Islands growing up speaking Äiwoo (Åshild Næss, p.c.). There
seems to be some degree of lexical replacement, with younger speakers using more Pijin
loanwords than older speakers do. Other than at church, Pijin and English also dominate
in education contexts. As a result, literacy in Äiwoo is rather rare, partly also due to the
coexistence of competing orthographic standards developed at different times. Boerger
et al. (2012) thus assess Äiwoo as belonging to level 6a – “vigorous” – of the EGIDS scale
(Extended Graded Intergenerational Disruption Scale; Lewis & Simons 2010).
The phylogenetic affiliation of Äiwoo and its neighbors was a debated matter for
around three decades, largely due to scarce documentation. In earlier work, Äiwoo and
the Santa Cruz languages were argued to be either “Papuan” (a purely geographical term,
not identifying an actual phylogenetic grouping but simply non-Austronesian), or at best
3 This figure for Noipä (approximately 40 speakers) is from Alfarano & Boerger (2025: 48). However, the
official request for a new ISO language code claims “several hundred speakers” (Boerger 2017: 1). I choose
to report the more conservative figure here as it is more updated.
Chapter 2. The Äiwoo language: background 49
of some sort of mixed Papuan-Austronesian origin (Codrington 1885, Wurm 1972, Lincoln
1978, Wurm 1978, 1981a,b, 1982, 1991). However, it was ultimately shown in Næss (2006),
Ross & Næss (2007), and Næss & Boerger (2008) that Äiwoo and the Santa Cruz languages
belong to a first-order subgroup of the Oceanic branch of Austronesian, labelled “Temotu”.
The position of Äiwoo within the Austronesian language family is shown in figure 2.4
(elaborated from Ross & Næss 2007, Blust 2013, Ross et al. 2016, Lackey & Boerger 2021).
For simplicity, the internal structure of all other (sub-)branches is severely simplified.
Austronesian
languages
Formosan Malayo-Polynesian
languages
languages
Western Malayo-Polynesian Central-Eastern
languages Malayo-Polynesian
languages
Tagalog, Cebuano, Indonesian,
Malay, Malagasy, Chamorro, …
South Halmahera- Oceanic
West New Guinea
languages
Yapese, Admiralty languages, St. Temotu Micronesian Central Pacific
Matthias languages, Western Oceanic languages languages languages
linkage, Southeast Solomonic languages,
Southern Oceanic languages
… …
Reef-Santa Cruz Utupua Vanikoro
languages languages languages … Polynesian
languages
Santa Cruz Äiwoo Tongan, Niuean, Samoan, Māori,
languages Hawai’ian, Vaeakau-Taumako, …
Natügu, Nalögo,
Noipä, Engdewu
Figure 2.4: Phylogenetic affiliation of Äiwoo within the Austronesian language family
From a diachronic point of view, one of the most surprising aspects of Äiwoo being an
Oceanic language is that it shows a symmetrical voice system – the syntax of which will
50 2.3. Earlier literature
be the main empirical focus of this dissertation. This kind of system is reconstructed as
going back to Proto-Austronesian, but it was only retained in the Formosan and Western-
Malayo-Polynesian branches (Blust 2015, Blust & Chen 2017, V. Chen 2017a, Kroeger &
Riesberg 2024, a.m.o.). The more notorious examples of such systems are indeed found in
Tagalog, Indonesian, Malagasy, etc. In contrast, Proto-Oceanic has been argued to have
lost this system completely (Lynch et al. 2002: 61–62, Evans 2003, M. Ross 2004b, Ross et al.
2016: 26–27). It remains therefore an open question whether what ought to be revised is
the current reconstruction of the syntax of Proto-Oceanic, or the placement of the Temotu
sub-group within the language family (Næss 2015b).
2.3 EARLIER LITERATURE
Äiwoo is a severely underdocumented language. The older (pre-2000) literature cited
above was exclusively based on scarce wordlists and similar poor documentation, and
has been entirely superseded by the more recent work. The available literature is largely
centered around work by Åshild Næss and collaborators, carried out from a descriptive
and typological/functionalist lens, and based on a corpus of transcribed natural speech
collected by Næss over several field trips to the Reef Islands between 2004 and 2015.
Næss (2006), Ross & Næss (2007), and Næss & Boerger (2008), cited above, estab-
lish Äiwoo as an Oceanic language, based on morphological evidence from the “bound
nouns” (which I analyze here as relativizers; see chapter 4, appendix 4.B), the pronominal
paradigms, and verbal morphology. Næss (2012a) describes a particular kind of bi-eventive
serial verb construction. Næss (2013) analyzes the alternation between certain clause types
as a diachronically transitional stage between symmetrical voice systems (found in West-
ern Malayo-Polynesian languages) and the systems characteristic of Oceanic languages,
whichmorphologically overtly contrast between intransitive and transitive predicates and
clauses. This idea was later revised in Næss (2015a) and especially Næss (2015b), the latter
of which firmly establishes Äiwoo as a symmetrical voice language. Some more recent
papers all describe a particular corner of Äiwoo grammar, respectively: the morphologi-
cal status of bound nouns (Næss 2017b); certain kinds of plural morphology (Næss 2018a);
negation (Roversi & Næss 2019); the encoding of spatial relations (Næss 2018b); some de-
tails of voice and valence-related morphology (Næss 2021a); comparatives (Næss 2021b);
nominal compounds and some related constructions (Næss 2023a); a potential pluractional
prefix and its interactions with the voice system (Næss 2023b). Most recently, Næss (2024)
presents a comparison of Äiwoo’s symmetrical voice system to the transitivity opposition
found in Engdewu, and Holmen & Næss (2025) examine the diachrony of the shift from
symmetrical voice to transitivity opposition. There is no published reference grammar
for Äiwoo, although one is currently in progress at the time of writing (Næss in prep).
However, an Äiwoo-English dictionary is available (Næss 2017a). In addition, there ex-
ist a few master’s theses on Äiwoo (Frostad 2006, Vaa 2006, Roversi 2019, Holmen 2020,
Viken 2024), and a psycholinguistic processing experiment (Sauppe et al. 2023).
Chapter 2. The Äiwoo language: background 51
From a formal/generative perspective, the only work done on Äiwoo is my own (with
collaborators). Roversi (2020) describes the system of “object agreement” found on UV
verbs (although this was based on a faulty empirical generalization due to scarcity of
data, and is revised in chapters 5–6). Wu et al. (2023) contain a short analysis of the voice
concord system described in chapter 3 (§3.3.1). Roversi (2024a) analyzes the interaction
of Condition C and a selective anti-cataphora effect. Roversi (2024b) offers an analysis
of Äiwoo clause structure and of UV-only verbs, which will be the topic of chapters 3–
4. Roversi (2025a) describes a case of non-local A-movement which will be the topic of
chapter 6, and supersedes Roversi (2020). Finally, Roversi (2025b) reanalyzes the grammar
of possession in the language.
Beyond Äiwoo, the other languages spoken in this area are even less researched. There
is a grammar of Vaeakau-Taumako (Næss & Hovdhaugen 2011) and a handful of descrip-
tive articles about some parts of its grammar (Næss 2000, 2002, 2004a,b, 2011a,b, 2012b)
and its linguistic contact with Äiwoo (Næss & Hovdhaugen 2007, Næss & Jenny 2011). En-
gdewu has been described in a PhD dissertation grammar (Vaa 2013). The same is true for
Nalögo (Alfarano 2021), which also enjoys a couple more publications of descriptive and
comparative/historical character (Alfarano 2023, Alfarano & Boerger 2022, 2025). Finally,
there is some descriptive literature on Natügu (Næss & Boerger 2008, van den Berg &
Boerger 2011, Boerger 2019, 2022, Lackey & Boerger 2021), and a wordlist with a request
to the ISO to recognize Noipä with its own distinct language code (Boerger 2017).
2.4 PHONOLOGY AND MORPHOPHONOLOGY
This whole section is heavily based on Næss (in prep.).
2.4.1 SEGMENTAL INVENTORY
The phonemic inventory of Äiwoo is given in tables 2.2–2.3, elaborated from Ross & Næss
(2007) and Næss (2017a, in prep.).
Table 2.2: Consonant phonemes of Äiwoo
Bilabial Labiodental Coronal Palatal Velar
Voiceless stops p(w) t k
Voiced stops mb(w) nd ɲdʒ ŋg
Nasal stops m(w) n ɲ ŋ
Fricatives s
Approximants (w) ʋ
Sonorants l
52 2.4. Phonology and morphophonology
Table 2.3: Vowel phonemes of Äiwoo
i u
e o
æ~ɛ ɑ~ɒ~ɔ
a
It is unclear whether the labio-velarized versions of the bilabial series are phonem-
ically contrastive with respect to the plain versions (e.g., /pw/ vs. /p/, etc.). The labio-
velarized consonants only surface before low vowels (/æ, a, ɒ/). For many words, the
labio-velarized and plain versions freely alternate. For example, my consultant accepts
both [nuwopwa~nuwopa] as equally possible pronunciations of ‘house’, [mwana~mana]
for ‘very’, [lopwɒ~lopɒ] for ‘story’, etc., and both versions are often found in recordings
and texts. However, not all words with a plain bilabial consonant can also freely have
a labio-velarized one instead; for example, the negative particle ba is always [mba] and
never *[mbwa], paveli ‘garden’ is always [paveli] and not *[pwaveli], etc. The only min-
imal pair I am aware of is bwää [mbwæː] ‘ocean, deep sea’ vs. bää [mbæː] ‘lean’ (Åshild
Næss, p.c.). It is also unclear whether /w/ and /ʋ/ are distinct phonemes, or allophones of
the same underlying consonant. Note that all voiced stops are consistently pre-nasalized.
Vowel length is phonemically contrastive in the language. Minimal pairs for each
vowel are given in (17). All these pairs feature monosyllabic words to avoid the potential
confound of stress, as there are some poorly understood interactions between stress and
vowel lengthening. Note that ngâgu ‘to me’ and ngââgu ‘bush, inland’ are monosyllabic
due to the final /u/ being reduced; this process is covered below.
(17) Minimal pairs for vowel length (Næss in prep.):
Contrast Minimal pair
/i – iː/ bi [mbi] ‘bake.UV’ bii [mbiː] ‘spit.on.UV’
/e – eː/ te [te] ‘see.UV’ tee [teː] ‘fish.AV’
/æ – æː/ kä [kæ] ‘say, think, want’ kää [kæː] ‘know.UV’
/a – aː/ da [nda] ‘forest’ daa [ndaː] ‘bottom’
/ɒ – ɒː/ ngâgu [ŋɒŋkwh] ‘to me’ ngââgu [ŋɒːŋkwh] ‘bush, inland’
/o – oː/ do [ndo] ‘pay.UV’ doo [ndoː] ‘what’
/u – uː/ pu [pu] ‘go’ puu [puː] ‘gun’
The spelling system I use in this dissertation is the one adopted in the dictionary
(Næss 2017a) and in almost all other literature about Äiwoo from Næss (2006) and on.
Most consonant and vowel letters have their IPA value, and vowel length is marked by
doubling the vowel letter (as can be seen in (17)). The exceptions are listed in table 2.4. In
Chapter 2. The Äiwoo language: background 53
the phonetic transcription here I include morphophonological details that have not been
discussed yet (stress and high vowel deletion)4.
Table 2.4: Correspondences between spelling and IPA values
Spelling IPA Example
⟨b⟩ /mb/ bei /mbéi/ [mbéi] ‘be sick’
⟨bw⟩ /mbw/ bwää /mbwǽː/ [mbwǽː] ‘ocean, open sea’
⟨d⟩ /nd/ dâwââ /ndɑwɑ́ː/ [ndɑwɑ́ː] ‘tie.AV’
⟨g⟩ /ŋg/ gilaki /ŋgiláki/ [ŋglákjh] ‘boy, male child’
⟨ny⟩ /ɲ/ nyetââ /ɲetɑ́ː/ [ɲeɾɑ́ː] ‘mirror’
⟨ng⟩ /ŋ/ ngege /ŋéŋge/ [ŋéŋge] ‘immediately’
⟨j⟩ /ɲdʒ/ jelâ /ɲdʒélɑ/ [ɲdʒélɑ] ‘thing’
⟨v⟩ /ʋ/ vävaaviâ /ʋæʋáːʋiɑ/ [ʋæʋáːʋjɑ] ‘dress.UV’
⟨ä⟩ /æ~ɛ/ kä /kǽ/ [kǽ] ‘say, want, think’
⟨â⟩ /ɑ~ɒ~ɔ/ ngâgu /ŋɑ́ŋgu/ [ŋɑ́ŋkwh] ‘to me’
2.4.2 ESSENTIAL PHONOLOGICAL PATTERNS
For the purposes of this dissertation, only a few phonological details will be important. I
proceed to discuss a few of them below.
CONSONANT ALLOPHONY The voiceless coronal stop /t/ is often realized as a tap or trill
/ɾ~r/ in intervocalic position (e.g. butete ‘potato’ /mbutéte/ > [mbuɾéɾe]). Certain conso-
nant contrasts are neutralized before /i/: /t/ and /s/ are neutralized to /s/, /d/ and /dʒ/ are
neutralized to /dʒ/, and /n/ and /ɲ/ are neutralized to /ɲ/. For example, the verb ‘carry (on
one’s head)’ has the forms gote in Actor Voice and gosi in Undergoer Voice. The latter pre-
sumably contains a suffix -i at some underlying morphophonological level, which deletes
the final /e/ and mutates the /t/ into /s/. The same is true for the adverbial modifier ‘back,
again’, which is ute in Actor Voice and usi in Undergoer Voice.
VOWEL HARMONY Certain morphemes show a vowel harmony-like alternation. For ex-
ample, the third person directional -kä can either surface as -kä /kæ/ or -kâ /kɒ/ (18),
depending on whether what follows contains a front vowel or a back vowel respectively.
(18) a. i-tu-kä-i b. i-tu-kâ-mu
ASP-bring.UV-DIR3-3AUG ASP-bring.UV-DIR3-2MIN
‘They brought (it) there’ ‘You brought (it) there’ (*itukämu)
4 I depart from standard IPA in notating stress with an acute accent instead of the / ˈ/ diacritic. This is just for
convenience, as the latter is often poorly legible when next to pre-nasalization diacritics, which are common:
c.f. [mbéi] vs [ˈmbei]. Secondary stress will be notated with a grave accent when relevant, e.g. /kièpaʋímu/.
54 2.4. Phonology and morphophonology
This process is morpheme-specific, and does not apply to all forms containing /æ/ or /ɒ/.
For example, the first person directional is always -mä, and never harmonizes (19). The
same is true for the second person directional -wâ (20).
(19) a. i-tu-mä-i b. i-tu-mä-no
ASP-bring.UV-DIR1-3AUG ASP-bring.UV-DIR1-1MIN
‘They brought (it) here’ ‘I brought (it) here’ (*itumâno)
(20) a. i-tu-wâ-i b. i-tu-wâ-no
ASP-bring.UV-DIR2-3AUG ASP-bring.UV-DIR2-1MIN
‘They brought (it) there [towards ‘I brought (it) there [towards you]’
you]’ (*ituwäi)
For the morphemes that do harmonize, this process is obligatorily triggered by suffixes,
like -no 1MIN in (18b). However, it is only optionally triggered by enclitics like the TAM
particles (see §2.6.3 below). For example, the perfect particle =to only optionally triggers
vowel harmony on -kä; either forms are attested/acceptable (21)5.
(21) i-tou-{kâ/kä}=to (250717)
ASP-bring.AV-DIR3=PRF
‘S/he has brought (something) there’
An even closer minimal pair is shown in (22). Äiwoo has both a suffix and an enclitic
whose form is /ŋgu/, respectively an object-marking suffix and a negative particle. Despite
being segmentally identical, the suffix obligatorily triggers vowel harmony (22a), whereas
the enclitic only optionally does so (22b). (The null -∅n 3MIN suffix might be a potential
confound in (22b); see fn. 5.)
(22) Suffixes obligatorily trigger vowel harmony, enclitics only optionally do so:
(250717)
a. i-tu-{kâ/*kä}-gu-∅ŋ ⟹ /3itukɑŋgu, 7itukæŋgu/
ASP-bring.UV-DIR3-OBJ-1MIN
‘S/he brought me there’
b. ba i-tu-{kâ/kä}-∅n=gu ⟹ /3itukɑŋgu, 3itukæŋgu/
NEG ASP-bring.UV-DIR3-3MIN=NEG
‘S/he didn’t bring (it) there’
5 The back-harmonized form i-tou-kâ=towas judged negatively in one elicitation session (2025-07-28), though
similar forms are amply attested in the corpus. The fact that (21) is an Actor Voice form, as opposed to
Undergoer Voice in (18b), is orthogonal to the issue. The same optionality arises in UV (i), although in this
case there would also be a null suffix (-∅n 3MIN), which might potentially constitute an additional confound.
(i) i-tu- {kâ /kä }-∅n=to (250717)
ASP-bring.UV-DIR3-3MIN=PRF
‘S/he has brought (it) there’
Chapter 2. The Äiwoo language: background 55
The following forms (both functional and lexical ones) are known to show this harmony
pattern: -kä~kâ DIR3,nä~nâ- IRR, wâ~vä- CAUS, kä~kâ ‘say/think/want’, wä~wâ ‘go’, and
possibly also ngä~ngâ ‘eat.UV’, though with this root the alternation seems (impressionis-
tically) less frequent, with the front vowel form (ngä) predominating. Following the liter-
ature, I will consistently use the form with /æ/ as the citation form of these morphemes,
as this is the one found when they are not followed by anything in the same phonological
word (e.g. itukä Luk ‘Luke brought (it) there’.
Another harmony-like process affects the prefix ki~ku- ‘imperfective’ and the 3AUG
agreement prefix li~lu-. Once again, the vowel depends on the presence of back vowels
and/or labial consonants in the following syllable (23).
(23) a. ki-li-vei b. ku-lu-woi
IPFV-3AUG-weave.AV IPFV-3AUG-plant.AV
‘They are weaving’ ‘They are planting’
Unlike in the case of /æ~ɑ/ vowel harmony discussed above, there seems to be some degree
of optionality to this process. For example, the sequence IPFV-3AUG-bring.AV ‘they are
bringing’ has been volunteered by my consultant both as ki-li-tou and ku-lu-tou.
OPTIONAL METATHESIS/COPYING OF /u/ When /u/ is in a post-stressed position, it can
frequently (but optionally) undergo a phonological process that could be described in two
ways. One could either say that /u/ metathesizes with the preceding consonant, or alter-
natively, copying of /u/, then followed by deletion of the original /u/ due to high vowel
deletion (see below). This can in turn feed other phonological processes, like glide forma-
tion (/iu/ > [ju]). The examples in (24) are from Næss (in prep), with some annotations
added; those in (25) are from my own data.
(24) Optional metathesis of /u/:
a. i-du ⟹ [júndw̥] < /indu/; non-metathesized: [índ̥w]
ASP-finish
b. i-tu ⟹ [jútwh] < /itu/; non-metathesized: [ítwh]
ASP-bring.UV
c. ba ku-lu-pu-mä-le=gu ⟹ [kulupmæléuŋg̥w] < /kulupumæleŋgu/
NEG IPFV-3AUG-go-DIR1-UA=NEG
d. me-ku-wo-ute ⟹ [mèukwoúte] < /mekuwoute/
1AUG-IPFV-go-again
e. i-pulo=to=wâ ⟹ [jùploróɑ] < /ipulotowɑ/
ASP-burn=PRF=DIST
(25) a. pesigiläi-du ⟹ [pesiŋglæjúndw̥] < /pesiŋgilæjndu/ (240721)
boys-all
b. i-tu-i ⟹ [jutwíː] < /itui/   (240805)
ASP-bring.UV
56 2.4. Phonology and morphophonology
STRESS In general, stress in Äiwoo is trochaic. Seemingly, a trochee is counted as being
formed by two morae; each vowel is a mora, so that a syllable with two vowels or a long
vowel forms a trochee by itself. For example, the verb ‘fish with line’ has the forms eta in
Actor Voice and etai in Undergoer Voice. Both forms have two syllables, but the latter has
a heavy (bimoraic) final syllable, and as a result stress assignment is different (26).
(26) a. /eta/ ‘fish with line.AV’ b. /etai/ ‘fish with line.UV’
(φe.ta) > [é.ta] e.(φ tai) > [e.tái]
Feet are right-aligned with the edge of the phonological word. The rightmost foot gets
the primary stress, and every preceding foot gets a secondary stress6:
(27) a. eagovä ‘disease’
/eaŋgoʋæ/ > (φea)(
ŋ
φ go.ʋæ) > [èa
ŋgóʋæ]
b. ki-lolopâ-päko-mana ‘s/he is speaking very well’
IPFV-speak-good-very
/ki-lolopɑ-pæko-mana/ > klo(φ lopɑ)(φpæko)(φmana) > [klolòpɑpæ̀komána]
Both suffixes and enclitics seem to be counted as part of the phonological word for the
calculation of stress assignment, although there are unresolved questions as to exactly
how this process works. The following facts are at least descriptively true. If there is one
single monomoraic suffix/enclitic, this is ignored for the purposes of stress assignment,
and ends up as extrametrical:
(28) a. i-epavi-no ‘I cooked (it)’
ASP-cook.UV-1MIN
/i-epaʋi-no/ > je(φpaʋi)no > [jepáʋino]
NOT: (φ jepa)(φʋino) > *[jèpaʋíno]
b. i-epavi-∅n=to ‘S/he has cooked (it)’
ASP-cook.UV-3MIN=PRF
/i-epaʋi=to/ > je(φpaʋi)to > [jepáʋiɾo]
NOT: (φ jepa)(φʋito) > *[jèpaʋíɾo]
However, if there is either a sequence of two monomoraic suffixes/enclitics, or one bi-
moraic suffix/enclitic, these forms their own foot, and thus receive primary stress (since
they are at the right edge of the word):
(29) a. Two monomoraic suffixes:
i-epavi-kâ-no ‘I cooked (it) for them’
ASP-cook.UV-DIR3-1MIN
/i-epaʋi-kɑ-no/ > je(φpaʋi)(φkɑno) > [jepàʋikɑ́no]
NOT: (φ jepa)(φʋikɑ)no > *[jèpaʋíkɑno]
6 This is a simplification, but it will suffice for the purposes of this section. Morphological factors like root
structure play a factor in stress assignment. More details can be found in Næss (in prep.).
Chapter 2. The Äiwoo language: background 57
b. One bimoraic suffix:
i-epavi-ngopu ‘We are cooking (it)’
ASP-cook.UV-1AUG
/i-epaʋi-ŋopu/ > je(φpaʋi)(φŋopu) > [jepàʋiŋóp
wh]7
(30) a. Two monomoraic enclitics:
i-epave=to=wâ ‘S/he has cooked (something)’
NEG ASP-cook.AV=PRF=DIST 
/i-epaʋe=to=wɑ/ > je(φpaʋe)(φ towɑ) > [jepàʋeɾówɑ]
NOT: (φ jepa)(φʋeto)wɑ > *[jèpaʋéɾowɑ]
b. One bimoraic enclitic:
i-epave=kaa
ASP-cook.AV=FUT 
/i-epaʋe=kaː/ > je(φpaʋe)(φkaː) > [jepàʋekáː]
If there are three monomoraic suffixes/enclitics, then the last one is extrametrical, and the
first two form their own foot:
(31) i-epavi-kâ-no=to ‘I have cooked (it) for them’
ASP-cook.UV-DIR3-1MIN=PRF
/i-epaʋi-kɑ-no=to/ > je(φpaʋi)(φkɑno)to > [jepàʋikɑ́noɾo]
NOT: (φ jepa)(φʋikɑ)(φnoto) > *[jèpaʋìkɑnóɾo]
However, in a context with onemonomoraic suffix and either twomonomoraic enclitics or
one bimoraic one, the enclitics form their own foot and the suffix is seemingly “skipped”,
remaining thus extrametrical word-internally:
(32) a. i-epavi-no=to=wâ ‘I have cooked (it)’
ASP-cook.UV-1MIN=PRF=DIST
/i-epaʋi-no=to=wɑ/ > je(φpaʋi)no(φ towɑ) > [jepàʋinoɾówɑ]
NOT: > (φ jepa)(φʋino)(φ towɑ) > *[jèpaʋìnoɾówɑ]
b. ki-epavi-no=ngaa ‘I will cook (it)’
IPFV-cook.UV-1MIN=FUT
/ki-epaʋi-no=ŋaː/ > kje(φpaʋi)no(φŋaː) > [kjepàʋinoŋáː]
NOT: (φkjepa)(φʋino)(φŋaː) > *[kjèpaʋìnoŋáː]
HIGH VOWEL DELETION Finally, another prominent process – mentioned in passing
above – is the reduction of high vowels /i, u/. First, when preceding a vowel, /i, u/ are
reduced to a glide [j, w]; e.g. nulie [núlje] ‘pana’, leluwo [lélwo] ‘dance’. An underlying
sequence of /u-i/ is realized with the /u/ undergoing glide formation and the /i/ being
lengthened, e.g. tu-i ‘bring.UV-3AUG’ [twíː]. More strikingly, when they are unstressed,
high vowels are either devoiced or fully deleted, often leaving a palatalized/labio-velarized
coarticulation on the preceding consonant: laki ‘small’ /laki/ > [lákjh], nenu ‘coconut’
7 With high vowel deletion; see below.
58 2.4. Phonology and morphophonology
/nenu/ > [nénw]. When the deleted high vowel follows a voiceless stop, this is articulated
as aspirated; when it follows a voiced stop, this is devoiced and often aspirated whenword-
final, but it retains pre-nasalization. Compare nuku ‘my leg’ /nuku/ > [núkwh] vs. nugu
‘POSS.TOOL.1MIN’ /nuŋgu/ > [núŋkwh].
High vowel deletion is also the only source of surface consonant clusters and coda
consonants in the language. The only allowed underlying syllable structures are CV, VV
and Vː (e.g. bi ‘bake.UV’, ea ‘bad’, ii ‘peel.UV’). However, high vowel deletion can give rise
to both onset consonant clusters and coda consonants, e.g. ginoji ‘our son’ /ŋginoɲdʒi/
> [ŋknóɲtʃ]. This vowel deletion process is sometimes avoided in very slow and careful
speech, with the vowel being pronounced as fully voiced, but this is rather exceptional.
For example, my consultant would only pronounce these vowels fully sometimes when
repeating a word back to me slowly, to help me transcribe it correctly. High vowel deletion
is so pervasive that whenÄiwoo speakerswho are not trained in the standard orthography
(that is, most Äiwoo speakers; cf. §2.2) use it in writing in informal contexts (e.g.‚ on social
media), they will omit the vowels elided by this process. For example, mi=päko [mpæko]
‘the good one’ will be spelled as ⟨mpako⟩8.
High vowel deletion interacts with stress in interesting ways. Descriptively, only non-
word-initial unstressed high vowels are deleted. This would entail that stress assignment
precedes high vowel deletion. And indeed, in most cases this seems to be true. Consider the
dataset in (33). In (33a), first the bisyllabic word is parsed into one foot, stress is assigned to
the leftmost mora, and as a result the final /u/ is stressless, so it deletes. In (33b), however,
we’ve added a suffix -mu. Now, the foot is shifted to the right, so that the first /u/ is now
stressed, and therefore does not delete. Adding one more suffix changes things again: now
we have two whole feet (33c). As a consequence, the first and the third /u/ get deleted, but
not the middle one, because it is stressed. The surface forms are thus correctly predicted
by ordering stress assignment before vowel deletion.
(33) a. [ŋɑ́ŋkwh] /ŋɑŋgu/ (underlying form)
ngâgu ( ŋɑŋφ gu) (subdivision into feet)
to.1MIN ŋɑ́ŋgu (stress assignment)
‘To me’ ŋɑ́ŋkwh 3 (high vowel deletion)
b. [ŋɑŋgúmw] /ŋɑŋgumu/ (underlying form)
ngâgu-mu ŋɑ( ŋφ gumu) (subdivision into feet)
to-2MIN ŋɑŋgúmu (stress assignment)
‘To you’ ŋɑŋgúmw 3 (high vowel deletion)
c. [mba ŋɑ̀ŋkwmúŋkwh] /ŋɑŋgumuŋgu/ (underlying form)
ba ngâgu-mu=gu ( ŋ ŋφŋɑ gu)(φmu gu) (subdivision into feet)
NEG to-2MIN=NEG ŋɑ̀ŋgumúŋgu (stress assignment)
‘Not to you’ ŋɑ̀ŋkwmúŋkwh 3 (high vowel deletion)
8 In this kind of context, speakers will also stick to the symbols available on English keyboards and thus omit
the diacritics on ⟨ä⟩ and ⟨â⟩.
Chapter 2. The Äiwoo language: background 59
However, the stress pattern of certain words seems to give rise to an ordering paradox,
where vowel deletion must precede stress assignment in order to get the correct surface
form. Consider (34). The surface form of itogulo ‘S/he hit (it)’ is [itóŋglo], with the under-
lying /u/ getting deleted. However, if metrification and stress assignment precede high
vowel deletion, we would incorrectly predict *[ìtoŋgúlo] (34a). Instead, to get to the actual
form we would need high vowel deletion to feed parsing into feet and stress assignment
(34b), which contradicts the ordering established in (33).
(34) a. [itóŋglo] /itoŋgulo/ (underlying form)
i-togulo (φ ito)(
ŋ
φ gulo) (subdivision into feet)
ASP-hit.UV ìtoŋgúlo (stress assignment)
‘S/he hit (it)’ ìtoŋgúlo 7 ([u] is stressed, can’t delete)
b. /itoŋgulo/ (underlying form)
itoŋglo (high vowel deletion!)
i( toŋφ glo) (subdivision into feet)
itóŋglo 3 (stress assignment)
Another similar case is given in (35). Here again, ordering stress assignment before vowel
deletion would predict the wrong surface stress pattern, *[inúmbo] rather than the ex-
isting [ínmbo]. This case, however, has an additional complication: the prenasalization
component of the voiced stop has unclear syllabicity status.
(35) a. [ínmbo] /inumbo/ (underlying form)
i-nubo i(φnu
mbo) (subdivision into feet)
ASP-die inúmbo (stress assignment)
‘S/he died’ inúmbo 7 ([u] is stressed, can’t delete)
b. /inumbo/ (underlying form)
inmbo (high vowel deletion!)
( inmφ bo) (subdivision into feet)
ínmbo 3 (stress assignment)
I leave the development of a full account of Äiwoo stress and vowel deletion as an inter-
esting problem open for future research.
2.4.3 ESSENTIAL MORPHOPHONOLOGICAL PATTERNS
A few interactions at the morphology-phonology interface are worthy of mention. A first
one concerns the 3AUG agreement prefix li~lu-. When this precedes forms beginning with
/ʋ/ or /w/, these can instead optionally begin with /p(w)/ (remember that /ʋ/ is spelled as
⟨v⟩ in the orthography). This holds both for lexical roots, like vängä~pängä ‘eat.AV’ in (36),
and functional morphemes, like the causative prefix wâ~pâ~pwâ- in (37).
60 2.4. Phonology and morphophonology
(36) a. i-ki-vängä b. ki-li- {pängä/vängä }
1MIN-IPFV-eat.AV IPFV-3AUG-eat.AV
‘I am eating’ ‘They are eating’
(37) a. i-ku-wâ-nubo b. ku-lu- {p(w)â/wâ }-nubo
1MIN-IPFV-CAUS-die IPFV-3AUG-CAUS-die
‘I am killing’ ‘They are killing’
Ross & Næss (2007) suggest a diachronic explanation for this. Proto-Oceanic *p consis-
tently lenites to /ʋ/ before vowels in Äiwoo (and deletes before /u/): Proto-Oceanic *paŋan
‘eat’ > Äiwoo vängä ‘eat.AV’; Proto-Oceanic *pa- ‘causative’ > Äiwoo wâ- ‘causative’.
Somehow, the presence of this particular agreement prefix would at least optionally trig-
ger the appearance of the unlenited forms pängä, pâ~pwâ-. However, because the di-
achronic reconstruction of Äiwoo and Proto-Reef-Santa Cruz is still in progress, it’s not
clear that all modern Äiwoo forms that show a /{ʋ/w}~p/ alternation actually go back to
Proto-Oceanic forms with *p-. This alternation is also interesting on the synchronic side,
as the choice of a particular stem allomorph (pängä rather than default vängä) seems to be
triggered by a structurally higher affix (i.e., this is outward-sensitive allomorphy; Bobaljik
2000, 2012, a.m.o). I do not attempt to give a synchronic analysis of this alternation.
The next two phenomena I discuss concern some aspectual/modal prefixes. First, the
aspectual prefix i- obligatorily disappears when it is not word-initial (38). In a verb with
a 3MIN subject, i- will be word-initial, and therefore it must be pronounced (38a). How-
ever, if preceded by anything within the same phonological word, i- obligatorily deletes;
(38b,c) show this with respectively a subject agreement prefix and a relativizer9. Note
that there would be nothing phonologically ill-formed about *deibakisi, *meibakisi (cf.mei
‘sleep.IMP’), so this must be a morphological constraint. As a partial exception, the 3AUG
subject prefix li~lu- is consistently ordered to the right of the aspectual prefixes instead of
to their left (38d); in this case, i- is curiously optional. Contrast this with the imperfective
prefix ki-, which goes in the same morphological slot but is always realized faithfully (39).
(38) ASP i- only surfaces when word-initial: (250717)
a. *(i-) bakisi c. me= (*i-) bakisi 
ASP-run REL:person=ASP-run
‘S/he ran’ (i- obligatory) ‘The person who ran’ (i- impossible)
b. de- (*i-) bakisi d. (i-) li-bakisi
12AUG-ASP-run ASP-3AUG-run
‘We ran’ (i- impossible) ‘They ran’ (i- optional)
(39) IPFV ki- never disappears: (250717)
a. *(ki-) bakisi b. de- *(ki-) bakisi
IPFV-run 12AUG-IPFV-run
‘S/he is running’ ‘We are running’
9 For details about φ-agreement and relativizers, see chapter 5 and chapter 4, appendix 4.A respectively.
Chapter 2. The Äiwoo language: background 61
c. me= *(ki-) bakisi  d. *(ki-) li-bakisi
REL:person=IPFV-run IPFV-3AUG-run
‘The person who is running’ ‘They are running’
The last morphophonological phenomenon is an interaction between the aspectual/modal
prefixes nä- ‘irrealis’ and de- ‘apprehensive’, and the 1MIN subject agreement prefix i-. As
we have seen in (38b)–(39b), the normal position of an agreement prefix is to the left
of the aspectual/modal prefixes (with the only exception of 3AUG li-). As for the 1MIN
prefix i-, when the verb bears imperfective aspect, indeed we see the standard order φ-
aspect-verb (40a). However, something different happens with the prefixes nä- ‘irrealis’
or de- ‘apprehensive’ instead of the imperfective10. In combination with 1MIN i-, these
two markers create portmanteaux forms where there is no overt i- prefix but the initial
consonant of the modal prefix marker is palatalized (40b,c). As a contrast, (41b,c) show
the non-fused realization of these two modal prefixes when combined with a different
φ-agreement prefix, here mu- 2MIN.
(40) 1MIN i- forms portmanteaux with IRR nä- and APPR de-: (250717)
a. kâ-no=ngä i-ki- basiki=kaa
KÄ-1MIN=CV 1MIN-IPFV-run=FUT
‘I think that I will run’
b. kâ-no=ngä nyä- basiki=kaa
KÄ-1MIN=CV 1MIN.IRR-run=FUT
‘I want to run’ (lit. ‘I want that I will run’)
c. kâ-no=ngä je- basiki=io=waa
KÄ-1MIN=CV 1MIN.APPR-run=PROH=FUT
‘I don’t want to run’ (lit. ‘I want that I won’t run’)
(41) Other (non-1MIN) prefixes don’t form any portmanteaux: (250717)
a. kâ-no=ngä mu-ki- basiki=kaa
KÄ-1MIN=CV 2MIN-IPFV-run=FUT
‘I think that you will run’
b. kâ-no=ngä mu-nä- basiki=kaa
KÄ-1MIN=CV 2MIN-IRR-run=FUT
‘I want you to run’ (lit. ‘I want that you will run’)
c. kâ-no=ngä mu-de- basiki=io=waa
KÄ-1MIN=CV 2MIN-APPR-run=PROH=FUT
‘I don’t want you to run’ (lit. ‘I want that you won’t run’)
In a sense, it seems that 1MIN i- is being shifted rightward by a short distance: it infixes
after (or metathesizes with) the first consonant of the modal prefix, triggering its palatal-
ization and then disappearing (42). This is the regularly expected phonological outcome,
10 Although these markers can be found in matrix clauses, they are most common under the attitude predi-
cate kä~kâ. This verb can mean ‘think, say’ if the embedded clause has imperfective or general aspect (i-,
mentioned above), and ‘want/not want’ if the embedded clause has irrealis or apprehensive respectively.
62 2.5. Morphosyntax in the nominal domain
since /n/ and /ɲ/ are neutralized to /ɲ/ before /i/, and /d/ and /dʒ/ are neutralized to /dʒ/
in the same environment (see §2.4 above).
(42) a. i-nä-basiki ⟹ n⟨i⟩ä-basiki ⟹ nyä-basiki /ɲæ-mbasiki/
1MIN-IRR-run IRR⟨1MIN⟩-run
b. i-de-basiki ⟹ d⟨i⟩e-basiki ⟹ je-basiki /ɲdʒe-basiki/
1MIN-APPR-run APPR⟨1MIN⟩-run
I assume that some kind of morphotactic constraint can be specified to model the idiosyn-
cratic linearization of this particular φ-prefix, although I leave its precise formulation out-
side the scope of this work.
2.5 MORPHOSYNTAX IN THE NOMINAL DOMAIN
2.5.1 THE NUMBER SYSTEM
NUMBER ON PRONOUNS Typically for an Oceanic language, Äiwoo has a rich pronom-
inal inventory, making four person contrasts and three number contrasts. The number
system is a typologically unusual one, with the three categories minimal, augmented,
and unit-augmented (Næss 2006, Ross & Næss 2007, et seq.). The full paradigm is shown
in (43), with English translations and what the equivalent gloss would be in a singular-
plural(-dual-trial) system; see chapter 5, §5.5 for a more detailed morphosyntactic analysis
of these forms.
(43) Paradigm of full pronouns:
Form Translation Equivalent
1MIN iu ‘I’ 1SG
12MIN iuji ‘You and I; we two.INCL’ 1DU.INCL
2MIN iumu ‘You’ 2SG
3MIN inâ/ine ‘He, she’ (respectively DIST/PROX) 3SG
1AUG iungo(pu) ‘Me and others; we.EXCL (≥ 3)’ 1PL.EXCL
12AUG iude ‘You, me, and others; we.INCL.PL (≥ 4)’ 1PL.INCL
2AUG imi ‘Y’all (≥ 3)’ 2PL
3AUG ijii ‘They (≥ 3)’ 3PL
1UA iungole ‘Me and someone else; we two.EXCL’ 1DU.EXCL
12UA iudele ‘You, me, and someone else; we three.INCL’ 1TRI.INCL
2UA imile ‘You two’ 2DU
3UA ijiile ‘They two’ 3DU
An informal/intuitive way to describe such a system is to say that here, the first person
inclusive has a “singular” form, meaning ‘you and I’ (12MIN iuji). Of course, this is not
singular in the strict sense, since its reference is not atomic but refers to two individuals;
Chapter 2. The Äiwoo language: background 63
hence, the terms used are ‘minimal’ and ‘augmented’ rather than ‘singular’ and ‘plural’. To
this, Äiwoo also adds the category ‘unit-augmented’, whose reference is that of the corre-
sponding minimal form “plus one”. These forms would thus correspond to duals for first
exclusive, second, and third person, but to a trial for first person inclusive (12UA iudele),
since this refers to three individuals. This category is not only unified by its semantics but
also its morphology: the unit-augmented forms are consistently built on the augmented
forms with the addition of a suffix -le. For more information about minimal—augmented—
unit-augmented systems, see Cysouw (2003) and Harbour (2016), among others. See also
Thomas (1955) and McKay (1978) for early descriptions of this paradigmatic structure.
NUMBER ON NOUNS The rich system of person and number distinctions made on pro-
nouns contrasts with that found in the nominal system, which shows little to no number
distinctions (Næss 2018a). At a general level, Äiwoo nouns are largely number-neutral, in
the sense that the same form can express both atomic and non-atomic reference. How-
ever, nouns can be divided into three categories in terms of how they behave with respect
to number, distinguished by animacy. Their differences are previewed in (44); by “number
morphology” here I mean morphology on the noun itself.
(44) Behavior of different classes of nouns with respect to number:
Number morphology Verbal agreement
Inanimates 7 (7/??)
Non-human animates 7 3 (obligatory)
Humans 3 (optional) 3 (obligatory)
First, inanimate nouns never take any number morphology, and they are also generally
incompatible with 3AUG verbal agreement (45) (though sporadic exceptions are attested,
cf. (88b) below). Here, the agreement prefix was rejected on the modifier dâu ‘many’ (note
that this is an intransitive verb in a relative clause structure; lit. ‘[those books [that are
many]]’), and the consultant expressed an intuition that the presence of agreement didn’t
fit the inanimate noun. Impressionistically, nouns referring to small animals (e.g., ants)
also pattern this way.
(45) Inanimates are incompatible with number agreement on verbs: (240805)
buk (*li-)dâu eângâ ku-tu-kä mikilivaavee=kaa ngä skul
book 3AUG-many that IPFV-bring.UV-DIR3 teachers=FUT to school
‘The teachers will bring all those books to school’
Native speaker’s comment: “li- always refers to people, human beings”
Nouns referring to non-human animates (at least those of a larger size, like dogs, pigs,
etc.) are again number-neutral, but they obligatorily trigger verbal agreement depending
on their reference. In both sentences in (46a), the noun kuli ‘dog(s)’ has the same form.
However, when there is no 3AUG agreement on the verb the only available reading is
atomic (46a); viceversa, when the verb has the 3AUG prefix li- the noun can only be read
as non-atomic (46b).
64 2.5. Morphosyntax in the nominal domain
(46) Animate nouns: number-neutral, obligatory verbal agreement (250717)
a. kuli ki-bakisi b. kuli ki-li-bakisi
dog IPFV-run dog IPFV-3AUG-run
3SG: ‘The dog is running’ 7 SG: ‘The dog is running’
7 PL: ‘The dogs are running’ 3PL: ‘The dogs are running’
This type of contrast, however, is contingent on the availability of agreement for the argu-
ment in question. In syntactic positions that don’t trigger agreement, we see full number-
neutrality, so that the sentence is simply ambiguous. This is the case, for example, for
subjects of Undergoer Voice verbs (47a), or objects of Actor Voice verbs (47b). See chapter
5 for when agreement is and is not possible.
(47) Nouns in non-agreeing positions are fully number-neutral: (250717)
a. devalili i-mangi kuli
child ASP-bite.UV dog
‘The dog(s) bit the child’
b. i-ku-wâpunâ-kä kuli
1MIN-IPFV-feed.AV-DIR3 dog
‘I am feeding (a) dog(s)’
In other cases, the atomic and non-atomic readings can be disambiguated not by agree-
ment, but by adding an optional 3AUG pronoun in the post-verbal position; this is possible
for fronted objects of Undergoer Voice verbs (48)11. This pronoun is optional in the sense
that a non-atomic (plural) reading is possible even in its absence. In chapter 4 I suggest
this can be analyzed as an optional realization of the object’s lower copy, which receives
a reduced spell-out as a pronoun; see van Urk (2018) for a similar phenomenon in Dinka.
(48) poi i-dââ-no (ijii) (240201)
dog ASP-tie.UV–1MIN 3AUG
With ijii: ‘I tied the pigs/*pig’
Without ijii: ‘I tied the pig(s)’
Finally, nouns referring to humans are similar to non-human animates in that they are
number-neutral but trigger obligatory agreement – where syntactically available – if their
reference is non-atomic (49). However, if agreement is not possible, both atomic and non-
atomic readings are available (50).
11 Contra Næss (2018a), I don’t analyze the ijii found in a similar position in possessive constructions like (i)
as constituting a form of number marking. In Roversi (2025b) I show that the structure of possessed DPs
contain a relative clause: the possessum is the relativized object, POSS is a transitive verb, and the possessor
is the embedded subject. More literally, (i) is thus “the dogs [that we have ]”. The pronoun ijii here is
then just the standard realization of the object in a transitive sentence with this structure, cf. (ii).
(i) kuli no-de ijii (ii) ku-potaa-de ijii
dog POSS.UV-12AUG 3AUG IPFV-search.UV-12AUG 3AUG
‘Our dogs’ ‘We are looking for them’
Chapter 2. The Äiwoo language: background 65
(49) If agreement is available, it tracks atomicity: (250717)
a. devalili ki-plei-lâoo ile ngä paveli enge
child IPFV-play-always PROX in garden this
3SG: ‘The child is always playing in the garden’
7 PL: ‘The children are always playing in the garden’
b. devalili ki-li-plei-lâoo ile ngä paveli enge
child IPFV-3AUG-play-always PROX in garden this
7 SG: ‘The child is always playing in the garden’
3PL: ‘The children are always playing in the garden’
(50) If agreement is not available, the noun is number-neutral: (250717)
Pita i-tou-kä devalili ngä skul
Peter ASP-bring.AV-DIR3 child to school
‘Peter brought (a) child(ren) to school’
However, differently from non-human animates, human nouns can optionally show “num-
ber morphology”. This consists of the marker pe=, which I analyze in chapter 4, appendix
4.A as a relativizer (“The people [who…]”). This marker is only compatible with human
nouns, and a noun with this marker obligatorily triggers plural agreement (when avail-
able). However, it is optional: a noun can have non-atomic reference evenwithout it. These
properties are shown in (51)–(52) for nouns in both agreeing and non-agreeing positions.
(51) A noun marked with pe= must trigger verbal agreement: (250717)
pe=devalili ku-*(lu-)mâea
REL:people=child IPFV-3AUG-laugh
‘The children are laughing’
(52) Pe= is optional: (240919)
a. i-ku-potaa nurse
1MIN-IPFV-search.AV nurse
‘I am looking for a nurse/nurses’
Comment: “It could be one nurse or many”
b. i-ku-potaa pe=nurse
1MIN-IPFV-search.AV REL:people=nurse
‘I am looking for nurses’ 
Comment: “[This only means] more than one nurse”
Nouns marked by pe= don’t always have strictly non-atomic reference, however. In at
least some non-downward-entailing environments, they can have both atomic and non-
atomic readings, like English plurals12. Consider the question/answer pair in (53). The
question contains the form pedevalili ‘children’. However, given the response pattern, this
sentence clearly must be asking “do you have 𝑛 ≥ 1 children?” and not “do you have 𝑛 > 1
children?”. Therefore, the form pedevalili has a so-called “inclusive plural” readings, since
here it can refer to both atoms and non-atoms.
12 The environments this has been tested for are polar questions and under clausal negation.
66 2.5. Morphosyntax in the nominal domain
(53) Nounswith pe= have “inclusive plural” readings in (at least some) non-downward
entailing environments: (240919)
Q: pe=devalili no-mu i-to?
REL:people=child POSS-2MIN ASP-exist
‘Do you have children?’ (lit. ‘Do children of yours exist?’)
A: {#ba / 3ee }, nou nyigi
no yes POSS.1MIN one
‘#No/3Yes, I have one’
Comment: “ba, nou nyigi [‘No, I have one’] could be a good answer if the ques-
tion is ‘do you have many children?’”
Äiwoo thus constitutes an example of a languagewith number-neutral nouns and optional
plural marking (at least for a subset of its nouns), whose plural forms nonetheless allow
inclusive plural readings. This was assumed not to be possible (Rullmann & You 2006,
Bale et al. 2011), and constitutes a problem for certain theories of the semantics of number
(Sauerland et al. 2005, Farkas & de Swart 2010). However, in addition to Äiwoo, the same
pattern has also been reported for Southern Tiwa (Dawson 2024), Buryat (Bylinina &
Podobryaev 2020), and Balinese (Arka & Dalrymple 2021). I leave it to further research to
investigate the semantics of number in Äiwoo.
The reason why above I put “plural morphology” in scare quotes when introducing
pe= is that aside from its semantics, I don’t analyze pe= as a nominal prefix restricted to
human nouns (contra Næss 2018a), because it can attach to clearly phrasal constituents.
For example, it can attach to a PP describing a location (54a), or also to toponyms (54b).
As such, pe= is part of a group of relativizers; see chapter 4, appendix 4.A.
(54) a. pe=[ngâ nuumä eângâ] b. pe=[Honiara]
REL:people=PREP village that REL:people=Honiara
‘People in/from that village’ ‘People from Honiara’
Another way of overtly marking non-atomic readings for human nouns is with another
relativizing constructions, consisting of the generic relativizer mi= (see again 4.A). The
“singular” counterpart of these expressions usually uses the relativizerme= instead. These
forms are often used to express reference to e.g. professions (55)–(56), or other salient
human groups, such as ethnic or religious groups (57).
(55) a. me=ki-vaavee b. mi=ki-li-vaavee
REL:person=IPFV-show REL=IPFV-3AUG-show
‘Teacher’, lit. ‘Person who shows’ ‘Teachers’, lit. ‘People who show’
(56) a. me=ki-tei b. mi=ki-li-tei
REL:person=IPFV-fish REL=IPFV-3AUG-fish
‘Fisherman’, lit. ‘Person who fishes’ ‘Fishermen’, lit. ‘People who fish’
(57) a. mi=lu-wopa b. mi=li-Jiu
REL=3AUG-white REL=IPFV-Jew
‘White people’ ‘Jews, Jewish people’
Chapter 2. The Äiwoo language: background 67
Superficially, the “plural” forms with mi= have similar distributional properties to nouns
marked with pe=, though this has not been investigated in detail. The two markers are
also not in complementary distribution: for example, both pe=nurse and mi=ki-li-nurse
are possible for ‘nurses’. It is not clear to me what the possible semantic difference(s)
might be between these two forms.
2.5.2 STRUCTURE OF NOMINAL CONSTITUENTS
Nominal constituents in Äiwoo don’t have particularly rich morphology or structure.
There are no case distinctions, unlike other Austronesian languages with similar voice
systems (e.g. Tagalog; Schachter & Otanes 1972, a.o.). The noun is always leftmost in
a nominal, and it may be followed by a relative clause, the demonstratives enge/eângâ
(proximal/distal)13, and the deictic particles =Ce/Câ (proximal/distal) (58).
(58) [DP nuwopa mi=mâeâ enge=ke] nä-ve-no (231019)
house REL=new this=PROX IRR-buy.UV-1MIN
‘I would buy this new house/these new houses’
Sometimes, nouns can also be modified by the quantifier dä ‘some’ (59), although this
more frequently appears cliticized in a position the immediate left of the TAM sequence
(60). The distribution of dä is not well understood.
(59) Dä ‘some’ modifying a noun:
a. [dä sigiwâu] lâ i-lää-kä=to dowâlili ngâgo-∅n=nâ
some young.man DIST ASP-give.AV-DIR3=PRF child to-3MIN=DIST
‘A young man had given her a child’ (Usaliki 275)
b. [numotäpi dä] lamaa i-luwa-kä-∅n, …
betelnut some PRT ASP-take.UV-DIR3-3MIN
‘If he takes some betelnuts, …’ (Naive 28)
(60) Dä ‘some’ cliticized in the verb complex: (240909)
a. mikilivaavee ki-li-tou-mä=dä=naa nupwää ngä skul
teachers IPFV-3AUG-bring.AV-DIR1=some=FUT clothes to school
‘Some (of the) teachers will bring clothes to school’
b. nupwää eângâ ku-tu-mä mikilivaavee=dä=naa ngä skul
clothes that IPFV-bring.UV-DIR1 teachers=some=FUT to school
‘The teachers will bring some of those clothes to school’
Apart from dä ‘some’, the only elements that may precede the noun are the deictic markers
ile/ilâ PROX/DIST (61a). These forms can stand by themselves, in which case they seem to
convey either locational or temporal deixis, i.e., ‘here/there’ or ‘now/then’. However, these
can also attach to both DPs (61a) and PPs (61b). In the latter case, they cannot intervene
between the preposition and its complement noun phrase (61b).
13 These also have phonologically reduced forms ee /eââ.
68 2.5. Morphosyntax in the nominal domain
(61) Ile PROX precedes either a DP or a PP:
a. ile nubole enge i-kili Mary bugulo (230608)
PROX taro this ASP-dig.UV Mary yesterday
‘Mary harvested this taro yesterday’
b. i-ki-vängä-vesi sii 3{ile } ngä *{ile } täpilo enge (240215)
1MIN-IPFV-eat.AV-always fish PROX PREP PROX bowl this
‘I always eat fish in this bowl’
A variety of modifiers that in other languages are found on nominals, in Äiwoo must
be expressed in the form of a relative clause modifying the noun instead. This applies
to “adjectives” and numerals, both of which are – structurally – intransitive verbs. For
adjectives, this was shown for mâeâ ‘new’ in (58), where to modify the noun nuwopa
‘house’ it had to be accompanied by the relativizer mi=; nuwopa mimâeâ ‘new house’ is
thus, more literally, ‘(a/the) house that is new’. Often, however, this relativizer is optional.
The same applies to numerals (62a) and the quantifier dâu ‘many’ (62b).
(62) Numerals and dâu ‘many’ are intransitive verbs:
a. sime (mi=)li-eve eângâ
person REL=3AUG-three that
‘Those three men’
b. sime li-dâu
person 3AUG-many
‘Many people’
Possession, as mentioned above, is also realized structurally with a relative clause. The
so-called possessive classifiers are, really, Undergoer Voice transitive verbs, of which the
possessum is the relativized object, and the possessor is the (embedded) subject (63); see
Roversi (2025b) for arguments.
(63) Possessives are also relative clauses: (230601)
poii [RC i no Pita]
pig POSS.UV Peter
‘Peter’s pig’; lit. ‘The pig that Peter has/that is Peter’s’
Finally, PPs also cannot be embedded into a noun phrase directly in Äiwoo (64). The in-
tended meaning must instead be conveyed with a relative clause structure (65).
(64) PPs cannot be embedded inside nominals directly: (250116)
a. * [DP penyibe [PP go nupo]] ku-wâ=naa bulaape
man with net IPFV-go=FUT tomorrow
Intended: ‘The man with the net will leave tomorrow’
b. * nuwopa eângâ ki-ve [DP penyibe [PP go hat]]=kaa=kâ
house that IPFV-buy.UV man with hat=FUT=DIST
Intended: ‘The man with the hat will buy that house’
Chapter 2. The Äiwoo language: background 69
(65) A relative clause must be used instead: (250116)
a. [DP penyibe mi=[RC ki-gote hat]] ku-wâ=naa bulaape
man REL= IPFV-carry.AV hat IPFV-go=FUT tomorrow
‘The man wearing a hat will leave tomorrow’
b. nuwopa eângâ ki-ve [DP penyibe mi=[RC ki-gote hat]]=kaa=kâ
house that IPFV-buy.UV man REL= IPFV-carry.AV hat=FUT=DIST
‘The man wearing a hat will buy that house’
The respective order of multiple modifiers is shown in (66) (though remember that both
“adjectives”, “numerals” and “possessives” are, structurally, relative clauses).
(66) Order within the noun phrase: N-“Adj”-“Num”-Dem-“Poss”
a. nyidâbu elo nyigi eângâ nogo-i
day big one that POSS-3AUG
‘That one big day of theirs’ (Reported to me by Åshild Næss, p.c., from an
Äiwoo language workshop held in Honiara)
b. pesigiläi li-vili eââ nogo-i
boys 3AUG-five that POSS-3AUG
‘Those five sons of theirs’ (Næss in prep)
2.5.3 SHORT NOTES ABOUT DEFINITENESS AND SPECIFICITY
As should be clear from the discussion above, Äiwoo does not have articles, neither def-
inite nor indefinite. No in-depth study has been carried out to date about the semantic
interpretation of bare nouns in Äiwoo. I offer here some rather preliminary notes, which
will hopefully be helpful as a starting point for further research.
One aspect where Äiwoo seems to show similarities to other Austronesian languages
is how the definiteness/specificity value of objects are tied to the contrast between AV
and UV. In general, AV is more or less incompatible with definite and specific indefinite
objects (where “specific” is intended as “wide scope indefinites”), and viceversa, UV is in-
compatible with non-specific indefinite objects. This contrast shows two different surface
manifestations. First, consider the case where an object is only interpretable as definite/
specific due to some overt morphology, or in virtue of having a (contextually) unique ref-
erent. For example, if the object is either a lexical DP carrying an overt demonstrative, a
proper name, or a pronoun, AV is rejected (67), and UV is the only possibility (68)14.
(67) AV is incompatible with definite objects: (250807)
?* mikilitei ku-lu-tou-ute-mä=kaa {devalili eângâ /
fishermen IPFV-3AUG-bring.AV-back.AV-DIR1=FUT child that
Mary / iude }
Mary 12AUG
Intended: ‘The fishermen will bring back {that child/Mary/us}’
14 See chapter 6 for why the word order in UV is different when the object is a pronoun (68b).
70 2.5. Morphosyntax in the nominal domain
(68) UV is required instead: (250807)
a. {devalili eângâ / Mary } ku-tu-usi-mä mikilitei=kaa
child those Mary IPFV-bring.UV-back.UV-DIR1 fishermen=FUT
‘The fishermen will bring back that child/Mary’
b. (iude) ku-tu-usi-mä-gu-de=ngaa mikilitei
12AUG IPFV-bring.UV-back.UV-DIR1-OBJ-12AUG=FUT fishermen
‘The fishermen will bring us back’
Objects that are in principle compatible with both specific and non-specific readings (i.e.,
bare noun objects) are grammatical in both voices, but their interpretation then crucially
depends on the clause’s voice. Note that bare nouns are possible in any argument position,
both in Actor Voice andUndergoer Voice; many instances of both can be found throughout
the examples above. In AV, a bare noun object can only be interpreted as a non-specific
indefinite; in UV, only as either a definite or a specific indefinite. This is shown by the
contrast below. In (69), the second part of the sentence forces a non-specific indefinite
reading of the bare noun object dokta ‘doctor’; in this context, only AV is possible, and UV
is rejected15. Viceversa, in (70) the second part of the sentences forces a specific reading
of the bare noun object, and now only UV is possible, and not AV.
(69) Non-specific indefinite object: 3 AV, 7 UV (250325)
a. milinurse ku-lu-potaa-ute dokta, mo ba
nurses IPFV-3AUG-search.AV-again.AV doctor but NEG
i-te-kä-i=dä=gu
ASP-see.UV-DIR3-3AUG=some=NEG
‘The nurses are looking for [a doctor]non-spec again, but they haven’t found
one yet’
b. # dokta ku-potaa-usi milinurse, mo ba
doctor IPFV-search.UV-again.UV nurses but NEG
i-te-kä-i=dä=gu
ASP-see.UV-DIR3-3AUG=some=NEG
Intended: ‘The nurses are looking for [a doctor]non-spec again, but they haven’t
found one yet’
(70) Specific object: 3 UV, 7 UV (250325)
a. dokta ku-potaa-usi milinurse, mo ∅3MIN ba
doctor IPFV-search.UV-again.UV nurses but NEG
ku-mo=gu ngâ nuwopaa nyibei elenge
IPFV-stay=NEG in house.of place.sick today
‘The nurses are looking for a doctori again, but hei is not at the hospital
today’
15 See chapter 3, §3.3.3 for how we know whether the verb forms in these sentences are AV or UV.
Chapter 2. The Äiwoo language: background 71
b. # milinurse ku-lu-potaa-ute dokta, mo ∅3MIN ba
nurses IPFV-3AUG-search.AV-again.AV doctor but NEG
ku-mo=gu ngâ nuwopaa nyibei elenge
IPFV-stay=NEG in house.of place.sick today
Intended: ‘The nurses are looking for a doctori again, but hei is not at the
hospital today’
This restricted connection between the specificity value of an object and a clause’s voice
is summarized in (71). The general picture is consistent with what is known for some
other Austronesian languages, e.g. Tagalog (Rackowski 2002, Aldridge 2004, Rackowski &
Richards 2005, Latrouite 2011, Sabbagh 2016, J. N. Collins 2019).
(71) Summary of the connection between specificity of objects and voice:
Actor Voice Undergoer Voice
Overtly definite objects: 7 3
only non-specific only specific
Bare noun objects:
interpretation interpretation
Although the restriction I just presented holds as a general rule, the empirical facts ne-
cessitate further inquiry. For example, a few exceptions are attested in the corpus, such
as (72), where we see the clearly definite object ilâ poi eângâ ‘that pig’ (whose referent is
introduced in the preceding clause) in an Actor Voice clause.
(72) An Actor Voice clause with a definite object (Næss 2015b: 280):
poi le ku-bo-kele=to ngä sâpulâu=ke. Sime=kâ li-välâ=to
pig PROX IPFV-?-carve-PRF in men’s.house=PROX person=DIST 3AUG-gather=PRF
ngä sâpulâu go ki-li-vängä=kaa ilâ poi eângâ
in men’s.house for IPFV-3AUG-eat.AV=FUT DIST pig that
‘Pigs would be carved up in the single men’s house. People would gather in the
single men’s house to eat that pig.’
Consider also the contrast in (73). Here, my consultant judged both sentences as perfectly
acceptable, but expressed an intuition that the AV object has a mass interpretation (intu-
itively more akin to “all this fish”), while the UV object has a count interpretation (i.e. “this
one fish here”).
(73) a. mikilitei ki-li-tâbonge sii enge (241015)
fishermen IPFV-3AUG-cut.AV fish this
‘The fishermen are cutting this fish’
Comment: “[This means] more than one fish, a pile of fish”
b. sii enge ki-tâbu mikilitei
fish this IPFV-cut.UV fishermen
‘The fishermen are cutting this fish’
Comment: “Sounds like only one fish”
72 2.5. Morphosyntax in the nominal domain
Moreover, this connection between specificity and voice only holds in mono-clausal con-
texts. As mentioned in the introduction chapter, Äiwoo has a restriction on Ā-extraction
that may at times force either one voice or the other, depending on which argument is
being extracted; see chapter 4. In the context of subject extraction, particularly, AV is
generally obligatory16. In this case, the restriction against having definite objects in AV
disappears. In an informal sense, the need of respecting the Ā-extraction restriction (im-
posing AV here) is more important than the need of having UV for a definite object. This
same interaction between the definiteness/specificity of objects, voice, and Ā-extraction
is also observed in Tagalog (Rackowski & Richards 2005).
(74) Subject extraction forces AV, even with a definite object: (250728)
a. sime (mi=)[RC ku-tou-mä=kaa buk eângâ ]
person REL= IPFV-bring.AV-DIR1=FUT book that
‘The person who will bring that book (here)’
b. * sime (mi=)[RC buk eângâ ku-tu-mä-∅n=naa]
person REL= book that IPFV-bring.UV-DIR1-3MIN=FUT
Intended: ‘The person who will bring that book (here)’
2.5.4 A SHORT NOTE ABOUT NULL ARGUMENTS
Earlier literature described Äiwoo as a language with completely free argument drop,
solely restricted by context: “any argument NP can be omitted if it is retrievable from
context” (Næss 2015a: 83); “[t]here is, in fact, no constraint on which arguments may be
omitted in Äiwoo, under coordination or otherwise, beyond a general notion of recover-
ability from the discourse context” (Næss 2015b: 288, fn. 11). However, I want to argue
against this description. The amended generalization I propose instead is shown in (75).
(75) Distribution of null arguments in Äiwoo:
a. Any 3MIN argument can always be null, in any position.
b. A non-3MIN argument can only be null in the preverbal position, and must be
overt otherwise.
Let me first demonstrate the first part of the generalization, about 3MIN arguments. For
this specific type of nominal, Äiwoo does indeed show what essentially is free argument
drop, compatibly with the earlier descriptions cited above. Examples of null 3MIN subjects
and objects across AV and UV are shown in (76a–d)17,18, and (76e) is an example of a null
3MIN argument in the complement of a preposition (though see fn. 18).
16 At least in canonical transitive clauses; see chapters 4 and 6 for exceptions.
17 The example in (76b) contains a subject relative clause to circumvent the restriction against definite objects
in AV (see §2.5.3).
18 In (76c), the 3MIN subject is not actually null, since it is represented by a floating nasal feature; see §2.6.3.3.
Chapter 2. The Äiwoo language: background 73
(76) 3MIN arguments can be null anywhere:
a. ∅3MIN ki-vängä=kaa sii (AV subject)
IPFV-eat.AV=FUT fish
‘S/he will eat fish’ (250728)
b. [penyibe [RC i-veve ∅3MIN ]]=kâ mekivaavee (AV object)
man ASP-buy.AV =DIST teacher
‘The man who bought it is a teacher’ (230622)
c. ∅ n3MIN ki-ngä -∅ =naa (UV object and subject)
IPFV-eat.UV-3MIN=FUT
‘S/he will eat it’ (250728)
d. ∅2MIN ku-tu-kâ-gu-mu=waa ∅3MIN (UVgu subject)
IPFV-bring.UV-DIR3-OBJ-2MIN=FUT
‘S/he will bring you’
e. mekivaavee i-lää-kä buk ngâgo -∅n (Complement of P)
teacher ASP-give.AV-DIR3 book to-3MIN
‘The teacher gave him/her a book’
Any argument that is not 3MIN, however, shows a very different pattern. As stated in
(75b), these arguments can only ever be null in the preverbal position (77). In this case,
the argument will most often be null, and it will only be pronounced overtly for particular
information-structural effects (e.g. contrastive topic or focus).
(77) Non-3MIN arguments can be null in the preverbal position: (250728)
a. ∅12AUG /iude de-ki-vängä=kaa sii (AV subject)
12AUG 12AUG-IPFV-eat.AV=FUT fish
‘We will eat fish’
b. ∅12AUG /iude ku-tu-mä-gu-de=ngaa Jon (UVgu object)
12AUG IPFV-bring.UV-DIR1-OBJ-12AUG=FUT John
‘John will bring us’
c. ∅12AUG /iude ku-tu-mä-de=ngaa ijii (UVSVO subject)
12AUG IPFV-bring.UV-DIR1-12AUG=FUT 3AUG
‘We will bring them’
In any other position of the Äiwoo clause, however, non-3MIN arguments must be overt
(78). This is true both for arguments in the position between the verb and the TAM par-
ticles (78a,b), for arguments after the TAM particles (78c), and for complements of prepo-
sitions (78d)19.
19 Due to independent factors, the three cases listed in (78) are the only relevant combinations of arguments
and voice. For example, in AV there never is any argument in the position between the verb and the TAM
particles, and pronominal objects are impossible (see §2.5.3).
74 2.6. (Some) morphosyntax in the verbal domain
(78) Non-3MIN arguments must be overt anywhere else than preverbally: (250728)
a. sii ki-ngä *(-de) =ngaa (UV subject)
fish IPFV-eat.UV-12AUG=FUT
‘We will eat the fish’ (bad without -de)
b. ( iude) ku-tu-mä *(-gu-de) =ngaa Jon (UVgu object)
12AUG IPFV-bring.UV-DIR1-OBJ-12AUG=FUT John
‘John will bring us’ (bad without -gu-de)
c. ( ijii) ku-tu-mä *(-i) =laa *(iude) (UVSVO subject/object)
3AUG IPFV-bring.UV-DIR1-3AUG=FUT 12AUG
‘They will bring us’ (bad without -i or iude)
d. mekivaavee i-lää-kä buk ngâgu -*(de) (Complement of P)
teacher ASP-give.AV-DIR3 book to-12AUG
‘The teacher gave him/her a book’ (bad without -de)
In the earlier literature, the fact that the φ-suffixes in these sentences must be overt was
not interpreted as counterevidence to the “free argument drop” analysis, because these
were analyzed as agreement markers. Therefore, the putative corresponding argument
was thought to be null. (It was not known at all that a postverbal pronoun like iude in
(78c) had to be overt.) However, in chapter 5 I show that these suffixes are not agreement
markers but just phonologically reduced pronouns. Therefore, their obligatory overtness
is significative: we now know that an argument in this position must be overt, in contrast
to preverbal arguments.
2.6 (SOME) MORPHOSYNTAX IN THE VERBAL DOMAIN
Contrasting with the relatively simple nominal domain, the verbal domain is where the
greater part of grammatical complexity in Äiwoo lies. The schema in (79) represents an
abstract general “map” of the verb complex. From left to right, we find (i) a prefixal slot
for φ-agreement (only found for Actor Voice and intransitive verbs; see chapter 3); (ii) a
slot for prefixes conveying aspect and mood semantics; (iii) the stem, which itself may
be morphologically complex; (iv) a slot for directional suffixes; (v) a sequence of enclitic
particles with polarity related to TAM and semantics; a nominal may in certain cases in-
tervene between the verb and these particles. See also Næss & Boerger (2008) and Roversi
(2019) for more details.
(79) The Äiwoo verb complex:
(φ-) ASP/MOOD- STEM -DIR [DP]=TAM
MAIN STEM
(CAUS-) STEM (-APPL) (MODIFIERS)
Chapter 2. The Äiwoo language: background 75
The morphological structure of the stem is covered largely in chapter 3, and the complex
facts about φ-agreement are discussed throughout chapters 3 and 5. In the rest of this
section, I only cover those aspects of verbal morphosyntax that are not addressed in de-
tail in the rest of the dissertation. This will be the aspectual/modal prefixes (§2.6.1), the
directionals suffixes (§2.6.2), and the enclitic TAM complex (§2.6.3).
2.6.1 ASPECTUAL/MODAL PREFIXES
This morphological slot can contain one of four prefixes: i- ‘unmarked aspect’, ki- ‘im-
perfective’ (and its phonological variant ku-), nä- ‘irrealis’, and de- ‘apprehensive’. The
first two are chiefly found in “realis” or non-modal contexts, as opposed to the latter two.
No in-depth study of the aspectual/modal semantics of these prefixes has been conducted
as yet, so what I report below is simply what is known at this stage. None of these pre-
fixes seems to encode any information about tense; all four are in principle compatible
with present, past, and future statements (though there are some interactions with the
postverbal TAM particles; see below).
The imperfective prefix ki- is found in e.g. progressive (80a), habitual (80b), and generic
statements (80c). Moreover, it seems often obligatory when the future marker =Caa is
present in the postverbal enclitic complex (81), at least in non-modal contexts.
(80) Some areas of use of ki- IPFV:
a. nubole ki-epavi Mary (230706)
taro IPFV-cook.UV Mary
‘Mary is/was cooking the taro’
b. i-ki-epave-lâoo sii (230525)
1MIN-IPFV-cook.AV-always fish
‘I always cook fish’
c. i-ku-mo ngâ nuwopa i-lââ ibesi (240411)
1MIN-IPFV-stay in house ASP-build.UV friend.1MIN
‘I live in the house that my friend built’
(81) Future =Caa requires IPFV: (240411)
nuwopa eângâ {*i- /3ki- }ve-∅n=naa
house that ASP- IPFV-buy.UV-3MIN=FUT
‘He will buy that house’
The “unmarked aspect” prefix i-, in contrast, seems to be more of a default marker; alter-
natively, we just don’t currently understand what its semantics are. A non-stative verb
with i- is most commonly translated as past tense (82), but this prefix is also compatible
with statives (83). As mentioned in §2.4.3, this prefix disappears when not word-initial
(e.g., when preceded by an agreement prefix or a relativizer).
(82) Mary i-lâwâle-kâ-no (230525)
Mary ASP-help.UV-DIR3-1MIN
‘I helped Mary’
76 2.6. (Some) morphosyntax in the verbal domain
(83) lopwâ enge i-kää-päko-i Peluwoo
story this ASP-know.UV-good-UV Reef.Islanders
‘The people from the Reef Islands know this story well’
The two prefixes nä- ‘irrealis’ and de- ‘apprehensive’ have modal semantics, although
their precise meaning is not known yet. The irrealis prefix nä- is, at least, used for deontic
necessity (84), but not deontic weak necessity (85) nor epistemic necessity (86). These
contexts are adapted from Vander Klok (2022).
(84) Deontic necessity: nä- ‘irrealis’ (250401)
[Context: the laws state that when you ride a motor bike, …]
a. helmet nä-vitee-mu ngâ nuwotaa-mu
helmet IRR-put.on.UV-2MIN on head-2MIN
‘You must wear a helmet on your head’
b. # helmet i-vitee-mu ngâ nuwotaa-mu
helmet ASP-put.on.UV-2MIN on head-2MIN
Intended: ‘You must wear a helmet on your head’
Comment: “This sounds like past tense, ‘you put on a helmet’”
c. # helmet ki-vitee-mu ngâ nuwotaa-mu
helmet IPFV-put.on.UV-2MIN on head-2MIN
Intended: ‘You must wear a helmet on your head’
Comment: “This sounds like ‘you are wearing’. Nä- sounds like an order”
(85) Deontic weak necessity: iki ‘fit, appropriate’, not nä- ‘irrealis’ (250401)
[Context: Anna is not yet used to riding a motorbike, she just started learning to
ride one month ago. Her friend suggests that/gives her advice to:]
a. Anna iki ku-wo-meloo
Anna fit IPFV-go-slow
‘Anna should drive slowly’
b. # Anna nâ-wo-meloo
Anna IRR-go-slow
Intended: ‘Anna should drive slowly’
Comment: “[This means] ‘Anna must go slow’, not just an advice, it’s an
obligation/order20”
(86) Epistemic necessity: unmarked? (250401)
[Context: Mary routinely has coffee at Lisa’s house everyday. Even if she’s sick,
she doesn’t miss a day! It’s not obligatory for Mary; she just goes for coffee there
all the time. It’s coffee time now, so…]
20 Sentence (85b) was judged acceptable in a context where, for example, Anna is driving past a school yard
where children are playing, so the speed limit is lower/the laws state that she must drive slowly. This
confirms its use for deontic (strong) necessity.
Chapter 2. The Äiwoo language: background 77
a. # ilâ=kâ Mary nâ-mo ngâ nuwopa tä Lisa
DIST=DIST Mary IRR-stay in house POSS Lisa
Intended: ‘Mary must be at Lisa’s house’
b. ilâ=kâ Mary ku-mo ngâ nuwopa tä Lisa
DIST=DIST Mary IPFV-stay in house POSS Lisa
‘Mary must be at Lisa’s house’
Aside from main clauses, the irrealis prefix nä- is also required in the complement clause
of the attitude predicate kä when this means ‘want’ (87a), and in purpose clauses (87b).
(87) a. Complement of ‘want’: (240530)
kä mekivaavee=kä [pedevalili nä-li-gâwââ-du(=kaa) buk]
want teacher=CV children IRR-3AUG-read.AV-all(=FUT) book
‘The teacher wants all the children to read a book’
b. Purpose clause:
pedevalili li-momalee-eke [go nä-li-kei-mä nubole]
children 3AUG-wake.up-fast for IRR-3AUG-harvest.AV-DIR1 taro
(250401)
‘The children woke up early to harvest taro (for me)’
Finally, the “apprehensive” prefix de- is found combined with the “prohibitive” enclitic
=eo~io in negative imperatives (88a), and as the negative counterpart of nä- in negated
desires, that is, under kä ‘want’ (88b). Interestingly, combining nä- ‘irrealis’ with the usual
negation strategy ba VERB=gu (see below) seems to be impossible (88c).
(88) a. de-lâbu-mu=io iumu! (250311)
APPR-cut.UV-2MIN=PROH 2MIN
‘Don’t cut yourself!’
b. kâ-no=ngä [nyenââ eângâ de-li-eoboli=io] (240222)
want-1MIN=CV tree that APPR-3AUG-fall=PROH
‘I don’t want those trees do fall’; lit. ‘I want that those trees don’t fall’
c. * kâ-no=ngä [nyenââ eângâ ba nä-li-eoboli=gu] (240222)
want-1MIN=CV tree that NEG IRR-3AUG-fall=NEG
Intended: ‘I don’t want those trees do fall’
2.6.2 DIRECTIONALS
The three directional suffixes are -mä ‘towards speaker’, -wâ ‘towards addressee’, and -
kä~kâ ‘away from speaker/addressee’21. These suffixes are always found at the very right
21 The 1st and 2nd person forms have common cognates across Oceanic and have been reconstructed as re-
flexes of the Proto-Oceanic verbs *mai ‘towards deictic center’ and *uatu ‘towards addressee’ (Ross & Næss
2007, M. Ross 2004a); the 3rd person form might be an innovation (Ross & Næss 2007). A fourth form -ngo
‘towards 1st person’ is only found in imperatives:
78 2.6. (Some) morphosyntax in the verbal domain
edge of the (possibly complex) verb stem. Often, they indicate the actual physical direc-
tion of motion. For example, consider the following minimal pairs. Both sentences in (89)
convey the same truth-conditional meaning. However, the 2nd person directional -wâ in
(89b) makes this sentence only appropriate in a context where the addressee is also in the
same location as Mary/as where the basket is being brought. The same point is illustrated
in (90), this time contrasting between the 3rd person and 1st person directionals.
(89) 3rd vs. 2nd person directional: (231019)
a. nyibä i-tu-kâ-no go Mary
basket ASP-bring.UV-DIR3-1MIN to Mary
‘I brought the basket to Mary’
b. nyibä i-tu-wâ-no go Mary
basket ASP-bring.UV-DIR2-1MIN to Mary
‘I brought the basket to Mary [in a location where you were also present]’
(90) 3rd vs. 1st person directional: (231019)
a. nyibä i-tu-kâ-mu ngâ nuwopa to
basket ASP-bring.UV-DIR3-2MIN to house POSS.1MIN
‘You brought the basket to my house’
Comment: “I’m somewhere else, not at home, but you brought it there”
b. nyibä i-tu-mä-mu ngâ nuwopa to
basket ASP-bring.UV-DIR1-2MIN to house POSS.1MIN
‘You brought the basket to my house’
Comment: “It means that I’m at home when you bring it”
Aside from uses that convey a more literal/physical direction of motion, the directional
suffixes can also often indicate “direction” in a much more metaphorical sense. For exam-
ple, they can indicate the presence of a recipient or benefactory role, as in (91).
(91) bulaape pedevalili ki-li-kei-mä=kaa nubole (240530)
tomorrow children IPFV-3AUG-dig.AV-DIR1=FUT taro
‘Tomorrow the children will harvest taro [for me, to my benefit, or my taro]’
Moreover, directional suffixes seem to be obligatory with certain verbs, such as verbs of
transfer (e.g. ‘give’) and verbs of gaze (‘see’, ‘look at’). In these cases, the directionals
“correspond” to one of the actual syntactic arguments in the clause (the recipient of verbs
of transfer, or the object of verbs of gaze), matching its person. However, due to their usage
in examples like (89)–(90), I refrain from analyzing directionals as constituting genuine
syntactic agreement.
(92) i-eâmoli*(-mä)-i iu ngâ nuwopa eângâ (230706)
ASP-look.at.UV*(-DIR1)-3AUG 1MIN in house that
‘They found me in that house’ (bad without -mä)
(i) po-ngo!
go-DIR1
‘Come here!’ (Næss in prep)
Chapter 2. The Äiwoo language: background 79
Interestingly, when a verb that requires a directional is used in a reflexive construction,
the directional suffix used is always the third person one. Compare, for example, the non-
reflexive (93a) to the reflexive (93b). In the former, only the 1st person directional is pos-
sible, presumably due to the object being 1st person (inclusive). However, in the reflexive
sentence (93b) the object is still 1st person inclusive, but now the directional suffix must
be the 3rd person one. (See chapters 5–6 for a discussion of the other morphosyntactic
differences between these two sentences).
(93) a. Non-reflexive: DIR tracks the object (241015)
∅12AUG i-te- {mä/*kä }-gu-de Jon
ASP-see.UV-{DIR1/*DIR3}-OBJ-12AUG John
‘John saw us’
b. Reflexive: DIR must be 3rd person
∅12AUG i-te- {kä/*mä }-de iude
ASP-see.UV-{DIR3/*DIR1}-12AUG 12AUG
‘We saw ourselves’
2.6.3 THE POST-VERBAL TAM COMPLEX
Outside of the verb complex proper lies a sequence of particles, which may phonologi-
cally encliticize to the verb itself (thus becoming part of the same phonological word) or
to an intervening DP. With the exception of the “circumstantial voice” marker =Cä (see
Appendix A), all of these have semantics connected to tense/aspect(/modality) and nega-
tion, although as for the aspect/mood prefixes, their precise semantic contribution is not
known. This sequence consists of five “slots”, as schematized in (94).
(94) The post-verbal TAM clitic sequence:
to
⎧ ⎫
⎪ PRF ⎪ gu Câ
⎪ jo ⎪
⎧⎪ NEG ⎪
⎫ ⎧ ⎫
Caa Cä ⎪DIST⎪
V (DP) = = ={ } ={ } =
⎨ ? ⎬ ⎨ (eo) ⎬ FUT CV ⎨ Ce
⎪ ⎪ ⎪ ⎪ ⎪
⎬
⎪ (eo)
⎪
⎪ ⎩(PROH)⎭ ⎩PROX⎭
⎩(PROH)⎭
Any sentence may contain one or more particles from (94), although only one per slot; in
other words, particles that are assigned to the same “slot” in (94) are in complementary
distribution. It is not currently known whether =eo ‘prohibitive’ can co-occur with either
of =to/jo or =gu, hence its placement in parentheses. For completeness, in addition to the
ones shown here, two more forms seem to belong to this sequence, although they are very
scarcely attested in the corpus and have not been investigated in any detail. These are =ta
and =ja (95), of unclear semantic contribution.
(95) a. ji-nä-savele-kä=ta ngâ nuwopa
12MIN-IRR-stroll-DIR3=TA to house
‘Let’s just go to the house’ (Moliki 355)
80 2.6. (Some) morphosyntax in the verbal domain
b. ile nâbu, ee, wagu-usi=ja ngâ nââu-de
PROX song HESIT tell.UV-again.UV=JA in language-12AUG
‘This song, uhm, say it again in our language’ (Dancing leluwo 194)
2.6.3.1 TENSE/ASPECT/POLARITY SEMANTICS
The three leftmost “slots” in the TAM sequence have more clearly to do with the semantic
categories of tense, aspect, and polarity. A sentence showcasing all three is given in (96).
(96) nuwopa eângâ ba ki-ve John =to=gu=naa ngä taun (250717)
house that NEG IPFV-buy.UV John=PRF=NEG=FUT in town
‘John won’t already have bought that house in town’
The easiest one to present is the negative enclitic =gu. Negation in Äiwoo is bipartite,
consisting of a preverbal particle ba and the marker =gu in the post-verbal TAM sequence
(97). See Roversi & Næss (2019) for more details about negation, and for discussion of
sporadic cases where either ba or =gu appear by themselves.
(97) Examples of sentences with negation (ba … =gu): (250807)
a. Jon ba i-mâea=gu
John NEG ASP-laugh=NEG
‘John did not laugh’
b. sii ba ki-ngä Jon=gu=naa
fish NEG IPFV-eat.UV John=NEG=FUT
‘John will not eat the fish’
In the (linearly) first “slot” we find the marker =to , with semantics connected to the per-
fect/perfective aspect, and the meaning of ‘already’22. In the following examples, I show
how =to is compatible with several of the typical functions of the English perfect and of
‘already’ (see e.g. Krajinović et al. 2023, and references therein):
(98) =to is compatible with several functions of the English perfect: (240919)
a. Resultative (perfect of result):
George i-wo-mä(=to)
George ASP-go-DIR1=PRF
‘George has arrived (now)’; =to is optional
b. Experiential (existential perfect):
Ingland i-leaa-kâ-no*(=to)
England ASP-reach.UV-DIR3-1MIN=PRF
‘I have been to England’; =to is obligatory
22 I gloss =to as PRF ‘perfect’ mostly for expository purposes, and this should not be taken as a strong analytical
commitment. More research will be needed to elucidate the precise semantic contribution of =to, and of the
other particles in this slot.
Chapter 2. The Äiwoo language: background 81
c. Perfect of anteriority (past):
[Context: you and I were supposed to meet, but I had to leave earlier]
mu-wo-maa mo iu=nge i-wâ*(=to)
2MIN-go-LOC.DIST but 1MIN=PROX 1MIN-go=PRF
‘You arrived but I (had) already left’; =to is obligatory
d. Perfect of anteriority (future):
[Context: you’re supposed to arrive tonight at 10pm, but I have to leave earlier
than that]
mu-ku-wo-mä=kaa=kâ mo iu=nge i-wâ*(=to)(=waa)
2MIN-IPFV-go-DIR1=FUT=DIST but 1MIN=PROX 1MIN-go=PRF=FUT
‘You will come, but I will already have left’; =to is obligatory (but =waa is
optional)
(99) =to has the same “earlier-than-expected” inference as already: (240919)
George i-wo-mä=to
George ASP-go-DIR1=PRF
‘George has (already) arrived’
3 Earlier-than-expected context: we expected George to arrive at 5pm, but he
arrived at 1pm instead. (‘George has already arrived’)
7 Later-than-expected context: we expected George to arrive at 1pm, but he ar-
rived at 5pm instead. (can’t mean ‘George has arrived now/finally’)
In the same slot we find the marker =jo (100), which has been glossed in earlier literature
as “progressive” (Næss & Boerger 2008), although it is not entirely clear to me that this is
an accurate label, since e.g. (100b) does not indicate that the act of folding is ongoing. I
leave investigating the semantics of =jo to future research.
(100) Næss & Boerger (2008: 190–191; glosses adapted):
a. i-lobâku-usi-∅n=to
ASP-fold.UV-again.UV-3MIN=PRF
‘S/he had folded it again (the object was presented to me in a folded state)’
b. i-lobâku-usi-∅n=jo
ASP-fold.UV-again.UV-3MIN=JO
‘S/he folded it again (I saw the act of folding)’
Finally, the particle =Caa is glossed as ‘future’ because it is (in general) incompatible with
past tense reference (101a). However, this is not completely true: in addition to conveying
future tense it is also attested in contexts with non-future reference, in which case its
usage seems to be connected to habituality (101b).
(101) a. * bugulo=kâ sii eângâ ki-epavi-no=ngaa (250717)
yesterday=DIST fish that IPFV-cook.UV-1MIN=FUT
Intended: ‘*Yesterday I will cook that fish’ (grammatical with bulaape ‘to-
morrow’ instead of bugulo ‘yesterday’)
82 2.6. (Some) morphosyntax in the verbal domain
b. i-vevaale-kä=naa tumä pelivano i-ipolâ-mä
ASP-wait-DIR3=FUT husband ASP-step.out-DIR1
ä i-li-pängä-le=naa=kâ
and ASP-3AUG-eat.AV-UA=FUT=DIST
‘Shewaited/wouldwait until her husband came back, and then they ate/would
eat (habitually)’ (Moliki 20–21)
When =Caa has actual future tense semantics, it is usually not compatible with the “un-
marked aspect” prefix i-, but requires imperfective ki- (102) (or nä- in irrealis contexts,
e.g. as the complement of ‘want’ etc.; see §2.6.1 above). However, the combination of i-
and =Caa is possible if the perfect marker =to is present too, e.g. (98d) above, or when
=Caa is used for habitual readings (101).
(102) Future =Caa requires IPFV: (240411)
nuwopa eângâ {*i- /3ki- }ve-∅n=naa
house that ASP- IPFV-buy.UV-3MIN=FUT
‘He will buy that house’
2.6.3.2 THE ALLOMORPHY OF CONSONANT-INITIAL ENCLITICS
The four enclitic particles =Caa ‘future’, =Cä ‘circumstantial voice’, =Câ ‘distal’ and =Ce
‘proximal’ share the same allomorphy pattern, where what varies is their initial conso-
nant. For all of them, the default form is /k/-initial (i.e., =kaa, =kä, =kâ, =ke). However, the
first consonant will change depending on their phonological host, that is, on the element
that is immediately linearly adjacent to their left. In addition to the default k-form, there
are four or five possibilities for what the first consonant may be (depending on one’s anal-
ysis): =ngV, =lV, =nV, =wV, and possibly =V with no consonant23. I show each of these in
(103), using the future particle =Caa as an example.
(103) a. ki-epavi-no =ngaa d. ki-epavi-mu =waa
IPFV-cook.UV-1MIN=FUT IPFV-cook.UV-2MIN=FUT
‘I will cook (it)’ ‘You will cook (it)’
b. ki-epavi-i =laa e. ki-epavi-mi =aa
IPFV-cook.UV-3AUG=FUT IPFV-cook.UV-2AUG=FUT
‘They will cook (it)’ ‘Y’all will cook (it)’
c. ki-epavi-∅n =naa f. ki-epavi Mary =kaa
IPFV-cook.UV-3MIN=FUT IPFV-cook.UV Mary=FUT
‘S/he will cook (it)’ ‘Mary will cook (it)’
The examples in (103), chosen to create simple minimal comparisons, might give the im-
pression that the initial consonant of these clitics depends on the person/number features
23 It is possible that the =wV form in (103d) is underlyingly really just =V (cf. (103e), and the /w/ is inserted
for hiatus resolution. In contrast to the other variants, whose distribution is not phonologically coherent,
=wV forms are only found attached to morphemes ending in back vowels (/u, o/).
Chapter 2. The Äiwoo language: background 83
of their host (e.g., =ngV after 1st person, =lV after 3AUG, etc.). However, this is not the
case. The distribution of the various alternants is rather idiosyncratic, and cannot be re-
duced to either phonological or semantic principles. For example, the =lV form is found
after a 3AUG suffix but also after the two question words doo ‘what’ and iie ‘who’ (but
not other question words). The =nV form, in addition to surfacing after the 3MIN -∅n (see
below for this notation), is also found after e.g. the unit-augmented suffix -le, the negative
particle =gu, and the specific lexical roots wä ‘go’ and ngä ‘still, yet’. For more exhaustive
lists of which forms are attested in what contexts, see Roversi (2019: 24–25).
Moreover, note that linear adjacency is the determining factor for what consonant a
particle will take, and not any form of syntactic/structural relationship. For example, in
(104) the distal particle =Câ follows the entire DP mitoguloi iu ‘the people who hit me’,
which contains a relative clause whose rightmost word is the embedded object iu ‘me’.
As a consequence, the particle surfaces in its ng-form, and not e.g. in the l-form that one
would expect from a 3AUG host (103b).
(104) [mi=[togulo-i iu]] =ngâ li-ebulou-mana (230608)
REL=hit.UV-3AUG 1MIN=DIST 3AUG-long-very
‘The people who hit me were very tall’
2.6.3.3 THE NULL SUFFIXES -∅n AND -∅ŋ
The pattern of allomorphy of these consonant-initial enclitics reveals the existence of two
particular suffixes, which despite being segmentally null themselves, cause a particular
allomorph of these enclitics to surface. I notate these as -∅n (3MIN subjects) and -∅ŋ (1MIN
objects). As should hopefully be clear from my notation, the 3MIN suffix -∅n selects the
n-initial form of a following consonant-alternating enclitic (105a), and the 1MIN suffix -∅ŋ
selects the ng-initial form (105b); the default form =kaa is given for comparison (105c).
(105) (250717)
a. 3MIN -∅n selects the n-form:
sii ku-tu-kä-∅n=naa
fish IPFV-bring.UV-DIR3-3MIN=FUT
‘S/he will bring the fish’
b. 1MIN -∅ŋ selects the ng-form:
( iu) ku-tu-kâ-gu-∅ŋ=ngaa Mary
1MIN IPFV-bring.UV-DIR3-OBJ-1MIN=FUT Mary
‘Mary will bring me’
c. Comparison baseline: default =kaa
sii ku-tu-kä Mary=kaa
fish IPFV-bring.UV-DIR3 Mary=FUT
‘Mary will bring the fish’
One could analyze these two suffixes as solely consisting of a floating/unassociated feature
or bundle of features ([nasal] for -∅n; [nasal, velar] for -∅ŋ), but no segmental content.
84 2.6. (Some) morphosyntax in the verbal domain
When followed by an enclitic whose first consonant is under-specified, this feature would
then get associated to that consonantal slot, giving the correct realization. Of course, if
these suffixes are not followed by one of the consonant-alternating enclitics, then the
unassociated features remain floating, and the suffixes’ presence is just completely ob-
scured (106). See Roversi (2020) for further discussion.
(106) If followed by anything else, -∅n and -∅ŋ have no visible effect: (250717)
a. sii i-tu-kä-∅n=jo
fish ASP-bring.UV-DIR3-3MIN=TAM
‘S/he has brought the fish’
b. ( iu) i-tu-kâ-gu-∅ŋ=jo ∅3MIN
1MIN ASP-bring.UV-DIR3-OBJ-1MIN=TAM Mary
‘S/he has brought me’
CHAPTER 3
ÄIWOO CLAUSE STRUCTURE
3.1 INTRODUCTION AND EMPIRICAL DOMAIN
The focus of this chapter is the “canonical clauses” of Äiwoo, that is, ordinary AV and UV
transitive clauses. These are the first two (unshaded) rows in our overview of different
clause types (107a,b), repeated from chapter 1.
(107) Overview of clausal patterns:
Clause type: Word order:
a. AV: S3 φS-V.AV =TAM O
7
b. UV 3plain: O V.UV S
7 =TAM
c. UVonly: O
3 V.UV S3=TAM
d. UV NA 3inâ: O V.UV S =TAM
e. UV : ONA V.UV -π =TAM S3gu O
f. UV NASVO: S V.UV -πS =TAM O
NA
This chapter focuses on the “Word Order Puzzle” outlined in chapter 1, for now limiting
the domain of discussion to these two clause types. The aim is to build a model of these
clauses, deriving their word order properties, and the alternation in verbal morphology.
Of course, for this model to be of any value, it needs to be “future-proof” with respect
to the rest of the analysis, so to speak. Therefore, our model will also need to (i) lay the
groundwork for later correctly deriving the Ā-extraction properties of these clauses (chap-
ter 4), and (ii) be able to accommodate an expansion to account for the alternative word
orders (chapter 6). This means that at different points through this chapter, sometimes I
will choose a seemingly more complicated alternative (in terms of analysis or implemen-
tation) over a simpler one, because the simpler one would later end up being incompatible
86 3.2. Basic word order: Äiwoo as a V2 language
with other aspects of the language that have not been introduced yet. This will be flagged
and justified at the relevant junctures.
The rest of this chapter follows this outline. The first two sections (§§3.2–3.3) are
descriptive, examining respectively word order facts (the V2 effect) and verbal voice mor-
phology. After we have a surface description of the clauses we are dealing with, in §3.4 I
take a temporary step back, and situate our discussion of voice in Äiwoo into a broader
landscape of theories of Austronesian voice. With this foundation, §3.5 develops a full
model of canonical Äiwoo clauses in AV and UV: how and when is voice established,
when and where and why do arguments move, etc. Because the model is relatively com-
plex with quite a few moving parts, §3.5.4 simply presents full annotated derivations,
with no prose around, summarizing and illustrating how the system works. As a final
note, §3.6 discusses how intransitive clauses can be captured in this system. In deriving
the main clausal structure, we will observe a first instance of non-local A-movement (i.e.,
A-movement of a lower nominal across a higher one, seemingly violating the typical strict
locality of A-movement), and wewill see that our theory is already equipped to derive this
without adding any new mechanisms.
3.2 BASIC WORD ORDER: ÄIWOO AS A V2 LANGUAGE
The goal of this descriptive section is to establish Äiwoo as a V2 language, as anticipated
in chapter 1. This aspect of the language’s syntax has gone unrecognized in the descrip-
tive literature so far, because the V2 effect is often obscured on the surface by independent
confounds. Establishing V2 as a core property of Äiwoo syntax will have important con-
sequences down the line for our analysis of the language, informing us about the position
of the different constituents.
In every clause – both matrix and embedded ones – one nominal precedes the verb. In
(almost) all cases, this is what I purely descriptively label the “Voice-Indexed Argument”
(VIA): the subject in AV, and the object in UV1,2. This means that a clause’s word order will
be different depending on its voice. In this respect, Äiwoo is similar to many Indonesian-
type languages (Arka 2002, Himmelmann 2002, 2005). Moreover, this specific combination
of V2 word order and Austronesian voice system is also found in the Nilotic languages
Dinka (van Urk 2015) and Shilluk (Remijsen & Ayoker 2018). On the other hand, this sets
Äiwoo apart from many other (Western) Austronesian languages like Tagalog and some
Formosan languages, which are largely predicate-initial and often show relatively free
word order amongst the nominal arguments following the predicate (although some of
these have VIA/pivot-last order, like Malagasy, Atayal, and Seediq; see Blust 2013: §7.3).
1 This label is inspired by Næss (2015b), which calls this the “Voice-Selected Argument”. The word “selected”
would be misleading given the model I propose for the voice alternation – hence the change. The term
“pivot”, often used in the Austronesian literature, will receive later a narrower technical definition (§3.5.1).
2 The only exception is one of the aforementioned non-canonical UV clause types, discussed in chapter 6.
Chapter 3. Äiwoo clause structure 87
The general word order is schematized in (108). As we can see, we have three linearly
defined positions where arguments can be. In the preverbal slot we find the VIA. Then, in
the slot between the verb and the TAM particles we find non-VIA subjects. Finally, in the
slot after the TAM particles we find non-VIA objects.
(108) Basic word orders:
AV: S V =TAM O (… PPs and adjuncts …)
UV: O V S =TAM (… PPs and adjuncts …)
Concrete examples of both word orders are given in (109). These sentences have very sim-
ilar truth-conditional content, with the main difference being the information-structural
status of the arguments, and the specificity of the object (narrow-scope/non-specific in-
definite in AV, wide-scope/specific indefinite or definite in UV); see chapter 2, §2.5.3. Aside
from word order, the other main difference is the verb stem: ‘eat’ is realized as vängä in
AV (109a), but as ngä in UV (109b).
(109) VIA V (S) =TAM (O) (adjuncts) (230706)
a. Anna ki- vängä =kaa sii ngä täpilo enge
Anna IPFV-eat.AV =FUT fish in bowl this
‘Anna will eat fish in this bowl’
b. sii ki- ngä Anna=kaa ngä täpilo enge
fish IPFV-eat.UV Anna=FUT in bowl this
‘Anna will eat the fish in this bowl’
Consistently, the verb immediately follows the first nominal constituent (the VIA). Only
few things can intervene between the VIA and the verb: the negation particle ba, and a
few adverbials with TAM-related meaning. These are shown for both voices in (110)-(111)
respectively3. Negation is bipartite, consisting of the preverbal particle ba and of another
morpheme =gu in the post-verbal =TAM series (Roversi & Næss 2019).
(110) The negation particle ba intervenes between the VIA and the verb: (250717)
a. John ba ki-veve=gu=naa nuwopa (AV)
John NEG IPFV-buy.AV=NEG=FUT house
‘John will not buy a house’
b. nuwopa enge ba ki-ve John=gu=naa (UV)
house this NEG IPFV-buy.UV John=NEG=FUT
‘John will not buy this house’
3 The English translation in (111) uses ‘just’ in the temporal sense of “just now”, not as “merely, only”. Ad-
verbials like lewâu ‘just’ in (111) are often built on the deictic particles le /lâ ‘PROX/DIST’ (le-wâu PROX-first
≈ ‘just’). These simplex particles can also be used in this position between the VIA and the verb, and they
are highly frequent in natural speech. Their exact contribution is poorly understood.
88 3.2. Basic word order: Äiwoo as a V2 language
(111) Adverbials intervening between the VIA and the verb: (240523)
a. Mary lewâu i-epave sii(=ke) (AV)
Mary just ASP-cook.AV fish=PROX
‘Mary just cooked fish’
b. sii eângâ lewâu i-epavi Mary(=ke) (UV)
fish that just ASP-cook.UV Mary=PROX
‘Mary just cooked that fish’
Importantly, I intend “V2” here as a structural notion, and not one that will necessarily
(or even frequently) be apparent on the surface string. Two factors conspire to often hide
the V2 effect on the surface: null arguments in the preverbal position, and topicalization.
APPARENT VIOLATION OF V2: PREVERBAL NULL ARGUMENTS In both of the clauses
in (109), the preverbal argument – the subject in AV, and the object in UV – can be null
(112). The resulting strings are thus verb-initial on the surface. (See also chapter 2, §2.5.4.)
(112) Preverbal arguments can be null: (250717)
a. ∅3MIN ki-vängä=kaa sii ngä täpilo enge
IPFV-eat.AV=FUT fish in bowl this
‘S/he will eat fish in this bowl’
b. ∅3MIN ki-ngä Anna=kaa ngä täpilo enge
IPFV-eat.UV Anna=FUT in bowl this
‘Anna will eat it in this bowl’
Defending the assumption that there is a null argument filling the preverbal position in
(112) would take us too far a field, and require a lotmore knowledge about different aspects
of Äiwoo syntax than we have now. Therefore, I defer this discussion to chapter 5 (§5.3.2).
APPARENT VIOLATION OF V2: LEFT-PERIPHERAL TOPICS The other phenomenon that
often obscures the V2 effect inÄiwoo is frequent topicalization in the left periphery. Unlike
in Germanic V2 languages, in Äiwoo the preverbal VIA slot can be preceded by certain
adverbials and (potentially multiple) topicalized nominals. All of these “don’t count” for
the V2 effect. Hence, in many clauses the verb won’t necessarily be exactly in the second
position on the surface string.
Topicalized constituents are separated from the main clause by prosodic breaks and/or
the particles =Câ/=Ce (glossed as ‘DIST/PROX’), which have a rather complex and not well-
understood distribution. Sentence (113), for example, contains two topicalized nominals.
When a constituent that would otherwise correspond to the VIA is topicalized, as above,
I assume that the VIA position is occupied by a null pronoun (∅3MIN here).
(113) Multiple topicalized constituents can precede the VIA: (231019)
[pedevalili eângâ]TOP=kâ [nubotage]TOP=kâ ∅3MIN i-kää-päko-i-i
children that=DIST path=DIST ASP-know.UV-good-UV-3AUG
‘Those children, the road, they know it very well’
Chapter 3. Äiwoo clause structure 89
Going forward, it will be important for building ourmodel to be able to determinewhether
a preverbal nominal is in the actual VIA position, or if it is a left-peripheral topic. Certain
adverbials can reliably be used to diagnose this. Expressions like ‘tomorrow’, ‘yesterday’,
‘every day’, etc. can be topicalized in the left periphery. They can be freely ordered with
other topics, but they must precede the VIA, so that they cannot intervene between the
VIA and the verb. In other words, if a nominal precedes one of those adverbials, then it
must be a topic.
To illustrate how this diagnostic works, consider the sentences in (114). These are all
UV clauses with ‘John’, the object, as the VIA. When the adverbial dâbu dâuwângâ ‘every
day’ precedes ‘John’ (114a), then the object is the VIA. If ‘John’ is also a topic, then it
will be marked as such by a prosodic break and/or the =DIST marker =Câ, as in (114b).
Crucially, however, if the object precedes the adverbial (114c,d), then we know that the
nominal must be topicalized, and not in the VIA position. If ‘John’ is not appropriately
marked as a topic, the sentence is rejected (114d). (In all these sentences, the temporal
adverbial can optionally also be marked with =kâ, and the same judgments replicate.)
(114) Adverbials cannot intervene between the VSA and the verb: (230727)
a. [dâbu dâuwângâ]TOP [John]VIA ki-te-kâ-no
day every John IPFV-see.UV-DIR3-1MIN
‘Every day I see John’
b. [dâbu dâuwângâ]TOP [John]TOP=kâ ∅3MIN ki-te-kâ-no
day every John=DIST IPFV-see.UV-DIR3-1MIN
‘Every day, John, I see him’
c. [John]TOP=kâ [dâbu dâuwângâ]TOP ∅3MIN ki-te-kâ-no
John=DIST day every IPFV-see.UV-DIR3-1MIN
‘John, I see him every day’
d. * [John]VIA [dâbu dâuwângâ]TOP ki-te-kâ-no
John day every IPFV-see.UV-DIR3-1MIN
Comment: “if you don’t say John kâ you need a comma after ‘John’, a break”
In the following chapters, this kind of ordering test will come in hand to better understand
the non-canonical word orders.
3.3 MORPHOSYNTACTIC CORRELATES OF VOICE
In the previous section, we saw how word order is determined by voice. In this section
I will introduce two more morphosyntactic correlates of voice: verbal stem morphology
and voice concord, and the presence of φ-agreement prefixes on the verb. These three
factors always reliably pattern together4, giving us reliable indicators of what voice any
given clause is in.
4 Again, with the sole exception of certain non-canonical UV clauses; see chapter 6.
90 3.3. Morphosyntactic correlates of voice
3.3.1 VOICE MORPHOLOGY
Voice morphology manifests itself on verb forms in two different ways. The first and most
evident one is stem allomorphy. The second one is a pattern of voice concord on verbal
modifiers. As we will see, stem allomorphy is often uninformative on its own, whereas
the concord pattern will more reliably help us to disambiguate between the two voices.
STEM ALLOMORPHY IS SUPPLETIVE In better-studied Austronesian languages, voice is
morphologically realized by more or less neatly segmentable affixes5, although their dis-
tribution can be rather complex. In Tagalog, for example, different voices will be marked
with prefixes, suffixes, or infixes (Schachter & Otanes 1972, a.o.). A sample paradigm from
Tagalog is given in (115). Here, we can see that Actor Voice is marked with the infix ⟨um⟩,
(115a), Undergoer Voice is marked by a null suffix (115b)6, Locative Voice uses a suffix
-an (115c), and Circumstantial Voice with the prefix i- (115d). Similar paradigms can also
be found for e.g. Formosan languages and Indonesian-type languages (Levin & Polinsky
2019), languages of Sulawesi like Mandar (Brodkin 2022), etc.
(115) Voice morphology in Tagalog (Rackowski & Richards 2005: 566; glosses adapted
from Hsieh 2019: 528 and Nie & Hsieh 2024)
a. b ⟨um⟩ ili ang bata ng tela sa palengke para sa nanay
⟨AV.PFV⟩buy NOM child GEN cloth OBL market for OBL mother
‘The child bought cloth at the market for Mother’
b. b⟨in⟩ili -∅ ng bata ang tela sa palengke para sa nanay
⟨PFV⟩buy-UV GEN child NOM cloth OBL market for OBL mother
‘The child bought the cloth at the market for Mother’
c. b⟨in⟩ilh -an ng bata ng tela ang palengke para sa nanay
⟨PFV⟩buy-LV GEN child GEN cloth NOM market for OBL mother
‘The child bought cloth at the market for Mother’
d. i- b⟨in⟩ili ng bata ng tela sa palengke ang nanay
CV-⟨PFV⟩buy GEN child GEN cloth OBL market NOM mother
‘The child bought cloth at the market for Mother’
In Äiwoo, however, the morphological landscape is much less regular. Here, voice mor-
phology is strongly idiosyncratic, often realized via stem suppletion. Essentially, one can
describe (almost) every verb as having two stems, one for AV and one for UV7. A sample of
alternating verb forms is shown in (116). This set is not exhaustive, but it should suffice in
5 This is admittedly more evidently the case for a language like Tagalog than e.g. for Malagasy, where voice
morphology shows rather intricate interactions with TAM morphology and verb stems; see Pearson (2001).
6 The cited literature on Tagalog uses the term ‘Patient Voice’; here I am replacing it with ‘Undergoer Voice’
for ease of comparability to Äiwoo. This is a purely terminological choice, of no analytical consequence.
7 In the first work recognizing a symmetrical voice system in Äiwoo, Næss (2015b) treats these as lexical
doublets, and labels these stems as “A-verbs” and “O-verbs”, a conventioned followed by the dictionary
(Næss 2017b). In later literature (Roversi 2019, Næss 2021b, 2023b, et seq.), these language-specific labels
were abandoned, and the alternation was reanalyzed as stem allomorphy.
Chapter 3. Äiwoo clause structure 91
showing that the stem alternations are mostly irregular, in the sense that there is no pre-
dictable morphophonological way to derive one form from the other. As a consequence,
a verb stem alone is often uninformative as to its voice properties: upon encountering a
new verb by itself, we can’t really know whether we are looking at an AV or a UV form.
(116) Voice-based stem alternations in Äiwoo:
a. epavi epave ‘cook.UV/AV’ f. lâbu lâbonge ‘cut.UV/AV’
b. lââ lâwââ ‘build.UV/AV’ g. kili kei ‘dig.UV/AV’
c. tu tou ‘bring.UV/AV’ h. pââ pä ‘steal.UV/AV’
d. togulo togo ‘hit.UV/AV’ i. ve veve ‘buy.UV/AV’
e. ngä vängä ‘eat.UV/AV’ j. kali ekäi ‘sing.UV/AV’
k. la lää ‘give.UV/AV’ l. nu nu ‘drink.UV/AV’
Some of these alternations are more frequent than others. A considerable number of verbs
follow the pattern in (116a), ending in /i/ in their UV form and in a different vowel (often
/e/) in their AV form. Another (smaller) class of verbs follows a pattern like (116b,c), show-
ing what looks like an infix ⟨âw~o⟩ after the first consonant in their AV form (possibly a
reflex of Proto-Malayo-Polynesian/Proto-Austronesian *⟨um⟩, also seen in the Tagalog AV
example above (115a); Næss 2013). However, a considerable number of verbs simply show
idiosyncratic stem suppletion. Some verbs also show an identical stem for both voices,
such as nu ‘drink’ (116l)8. For a more exhaustive treatment, the reader is referred to Næss
(2015b, 2021b), and Roversi (2019: §3.2.2). Throughout the dissertation, I will gloss voice
as fused to the stem.
VOICE CONCORD ON VERBAL MODIFIERS Fortunately, another morphological correlate
of voice can help us to reliably tell the two voices apart. This is a pattern of concord on
verbal modifiers, described in Roversi (2019), Næss (2021b), and Wu et al. (2023) (see also
an earlier description, although in different terms, in Næss & Boerger 2008).
When a modifier is attached to a verbal stem, its morphology will be dependent on
the voice properties of the stem it is attached to. In the examples below, I box the entire
complex stem, consisting of the main stem (leftmost) and its modifiers. When AV (or in-
transitive) stems takemodifiers, nothing special happens (117a). However, whenUV stems
take modifiers, every single one of them must take a suffix -i (118a). The mismatched al-
ternatives (-i suffixes in AV, and their absence in UV) are ungrammatical (117b)-(118b).
(117) AV: no voice concord (240425)
a. i-ki- lâwââ-päko-mana =kaa nuwopa
1MIN-IPFV-build.AV-good-very=FUT house
‘I will build a house/houses very well’
b. * i-ki- lâwââ-päko-i-mana-i =kaa nuwopa
1MIN-IPFV-build.AV-good-UV-very-UV=FUT house
8 This is importantly different from verbs that only have UV; see chapter 4.
92 3.3. Morphosyntactic correlates of voice
(118) UV: obligatory voice concord (240425)
a. nuwopa eângâ ki- lââ-päko-i-mana-i -no=ngaa
house that IPFV-build.UV-good-UV-very-UV-1MIN=FUT
‘I will build that house/those houses very well’
b. * nuwopa eângâ ki- lââ-päko-mana -no=ngaa
house that IPFV-build.UV-good-very-1MIN=FUT
In Roversi (2019) and Wu et al. (2023) this pattern was analyzed as voice concord: all
modifiers have to “match” the main stem’s voice (see §3.5.1.2 for an implementation).
This is a very robust phenomenon; however, a couple notes are in order for completeness.
The -i suffix has an allomorph -nyi/nyii consistently selected by certain modifiers, like
ngege ‘immediately, straight away’ (119a). Other modifiers yet never take any concord,
like eopu ‘also’ (119b). It’s not currently clear what controls either of these facts. See Næss
(2021b), Roversi (2019: §3.3.1) for lists of which modifiers show these alternative patterns.
(119) a. Voice concord as -nyii: (250717)
Anna i- weevä-ngege-nyii -kâ-no
Anna ASP-visit.UV-immediately-UV-DIR3-1MIN
‘I visited Anna straight away’
b. No voice concord:
Anna i- weevä-eopu -kâ-no
Anna ASP-visit.UV-also-DIR3-1MIN
‘I also visited Anna’
3.3.2 PRESENCE OF PREFIXAL φ-AGREEMENT SIGNALS AV
The final reliable correlate of voice is the pattern of φ-indexing on verbs. Verbs in Actor
Voice show an obligatory agreement prefix indexing the subject’s φ-features, li- 3AUG in
(120a). In Undergoer Voice, this is impossible (120b); a form like *ki-li-ngä is ill-formed9.
(120) a. AV: prefixal φ-agreement (250717)
pedevalili ki- li- vängä=kaa sii
children IPFV-3AUG-eat.AV=FUT fish
‘The children will eat fish’
b. UV: no φ-agreement
sii enge ki-ngä pedevalili=kaa
fish this IPFV-eat.UV children=FUT
‘The children will eat this fish’
In all earlier work on Äiwoo (Næss 2006 et seq.), UV verbs were described as carrying a
φ-agreement suffix instead of a prefix, based on data like (121). Here, we see no prefixal
marker, as expected, but we also see the suffixal marker -de 12AUG indexing the subject’s
9 There exists a prefix li-, homophonous with the 3AUG agreement prefix li-, that is used for generic agents.
Unlike 3AUG agreement li-, generic agent li- is compatible with UV verbs (Næss 2015b: fn. 10).
Chapter 3. Äiwoo clause structure 93
φ-features. Incidentally, this sentence also shows that UV verbs don’t carry agreement
with the preverbal object: pedevalili ‘children’ is plural, but there is no li- 3AUG prefix10.
(121) pedevalili ku-tu-kä -de =ngaa ngä skul (250717)
children IPFV-bring.UV-DIR3-12AUG=FUT to school
‘We will bring the children to school’
However, in chapter 5 I show that this kind of suffixal markers and the prefixal ones in
AV are of different syntactic nature. The AV prefixes are actual φ-agreement with the sub-
ject. The suffixes found in UV, contrarily, are not agreement, but phonologically reduced
pronouns in an argument position, i.e., the same structural position where pedevalili ‘chil-
dren’ is in (120b). I refer the reader to chapter 5 for the arguments. To summarize, I argue
that the difference between φ-marking in AV and UV should not be characterized as in
terms of prefixes vs. suffixes, contra Næss (2006, 2015b, 2021b) and Roversi (2019, 2020,
2025b) a.o., but in terms of φ-agreement prefixes in AV vs. no agreement in UV.
3.3.3 INTERIM SUMMARY
In this section I have shown three reliable morphosyntactic correlates of voice in Äiwoo,
repeated in (122).
(122) Morphosyntactic correlates of voice:
a. Word order
b. Verbal stem morphology and voice concord
c. Presence of prefixal φ-agreement
I illustrate all three in the minimal pair below. In (123a), we see SVO order, the verb ‘bring’
in its AV stem tou and nomorphology on themodifier eke ‘quick(ly)’, and the prefixal 3AUG
marker lu-. In (123b), conversely, we see OVS order, the UV stem tu and the UV concord
suffix -nyii on the modifier, and no prefixal agreement.
(123) a. AV: SVO order, no voice concord, φ-agreement (250717)
pedevalili ku-lu-tou-eke-mä=kaa dekilingä ngâ nuwopa
children IPFV-3AUG-bring.AV-quick-DIR1=FUT food to house
‘The children will bring food to the house quickly’
b. UV: OVS order, voice concord, no φ-agreement
dekilingä ku-tu-eke-nyii-mä pedevalili=kaa ngâ nuwopa
food IPFV-bring.UV-quick-UV-DIR1 children=FUT to house
‘The children will bring the food to the house quickly’
10 In this respect Äiwoo is different from Dinka, where prefixal agreement on verbs in every voice always
tracks the preverbal argument, regardless of grammatical relations (van Urk 2015).
94 3.4. Meta-theory of Austronesian voice: cause or effect
3.4 META-THEORY OF AUSTRONESIAN VOICE: CAUSE OR EFFECT
We now have a descriptive lay of the land for the sentences under discussion: Actor Voice
clauses look like (123a), and Undergoer Voice clauses look like (123b). Before moving on to
deriving the syntax of these clauses, it will be useful to take a step back and reflect on what
kind of systems these notions (Actor Voice, Undergoer Voice) belong to. Within the liter-
ature about Austronesian languages, indeed, a lot has been said about what Austronesian
voice is.
Boiling a vast array of nuanced complications down to its very core traits, a crucial
desideratum is to capture the correlation abstractly illustrated in (124). In every sentence,
the verb will surface with some particular morphology (represented by the abstract affixes
-α, -β, -γ). This verbal morphology will then consistently correlate with one specific DP
being “special” in some way, represented here as boxed (concretely, this might be a partic-
ular position in the clause, a particular case marking, particular syntactic privileges, etc.).
For now, I deliberately leave it very vague exactly what “special” means, as this depends
on the specific proposal and the specific language in question.
(124) Austronesian voice systems, abstractly:
V-α ⇔ DP1 DP2 DP3 …
V-β ⇔ DP1 DP2 DP3 …
V-γ ⇔ DP1 DP2 DP3 …
Two broader families of approaches to Austronesian voice can be identified, differing in
the direction of the posited cause-effect relation between the sides of this two-ways corre-
lation. On one hand, we can imagine that the choice of one specific set of verbal morphol-
ogy is what causes one particular DP to be “special”. On the other hand, we could look at
this this the other way around, and think that the choice to make a given DP “special” is
what results in a particular morphological outcome on the verb. Let us now spell this out
more concretely.
For the first kind of approaches, voice morphology reflects an argument structure
alternation (Chen & McDonnell 2019, V. Chen 2025): voice is the cause of the various
syntactic properties of each kind of clause. Under this view, these languages have a series
of lexically distinct flavors of transitive v heads (vAV, vUV, etc.), each with its own different
syntactic properties and its own morphological realization. A speaker of such a language
simply decides to pick one of these heads for the numeration of a sentence, and the rest
of the morphosyntactic properties of the sentence follow from this choice. Crucially, the
argument structure of the sentence is altered in specific ways depending on what v head
was chosen, resulting in one particular DP being “special”.
For the other family of approaches, however, voice morphology reflects the effect of
certain syntactic events that take place during the derivation. Distinctly from the previous
kind of proposals, importantly, there is no argument structure alternation: there is only
one transitive v head. Therefore, v itself cannot be what makes one of the DPs special,
since v doesn’t vary in any way. Rather, in every sentence one DP is born with special
Chapter 3. Äiwoo clause structure 95
properties (at the choice of the speaker), and this sets in motion a chain of syntactic events
that will result in an overt morphological reflex on the verb of the identity of this DP. I
summarize the differences in (125).
(125) Voice as cause: Voice as effect:
Argument structure alternation No argument structure alternation
Several flavors of transitive v heads Only one transitive v head
Numeration: speaker chooses which v Numeration: speaker chooses which
head to use DP is the special one
Choice of v results in one particular Choice of special DP results in partic-
DP being special ular voice morphology (and syntax)
Selected references: Payne (1982), De Guzman Selected references: Chung (1994), Richards
(1988), Gerdts (1988), Guilfoyle et al. (1992), (2000), Pearson (2001, 2005), Rackowski (2002),
Mithun (1994), Liao (2004), Aldridge (2004, Rackowski & Richards (2005), V. Chen (2017a,
2008, 2017), H. Y. Chang (2011), Legate (2012, 2023, 2025), van Urk (2015), van Urk &
2014), Erlewine (2018), and Erlewine & Som- Richards (2015), Erlewine et al. (2017), and
merlot (2023a,b) Hsieh (2020, 2025)
To give a clearer illustration of the difference between these two approaches, I will now
compare two very simplified sketches of influential proposals, one of each type. For both,
I will abstract away from a lot of the complexity and fine nuances of the specific analyses,
and will deviate from their letter, but will preserve and zoom in on the crucial cause-effect
relationship.
A typical voice-as-cause proposal is Aldridge (2004, 2008). Under this analysis, the
speaker chooses whether to merge vAV or vUV. Each of these will be spelled out differently,
as they are lexically distinct heads. If the speaker chooses to merge vAV (126a), nothing
particularly interesting happens. If, on the other hand, the speaker chooses to merge vUV
(126b), this will have the effect of moving the object past the agent, to an outer specifier11.
After this step, the object is now the highest DP in the clause, and this will have various
kinds of syntactic repercussions, since it’s effectively an argument structure alternation.
This is often a first step in the so-called “standard theory of syntactic ergativity” (Deal
2016b), where theA-movement of the object across the subject ultimately ends up blocking
Ā-extraction of the latter. See §1.3.2, and §4.3.2.
11 Aldridge formalizes this with an EPP feature present on vUV but absent from vAV, but several different im-
plementations could achieve the same result.
96 3.4. Meta-theory of Austronesian voice: cause or effect
(126) A “voice-as-cause” proposal (Aldridge 2004, 2008):
a. Actor Voice: b. Undergoer Voice:
vAVP vUVP
Subject
vAV VP Object
Subject
v
V Object UV
VP
V tO
Compare this now to a voice-as-effect approach, using Rackowski & Richards (2005) as an
example (again, in a rather simplified illustration). As mentioned earlier, here the choice
point in the hands of the speaker is not the lexical identity of which v head to merge: there
is only one transitive v head in the lexicon. Rather, an independently existing process of
object shift will move a definite/specific object past the subject, to an outer specifier (127b),
whereas indefinite/non-specific ones stay low (127a). The morphological spell-out of v is
then dictated by case agreement between v and the argument in its outermost specifier.
(127) A “voice-as-effect” approach (Rackowski & Richards 2005):
a. Actor Voice: b. Undergoer Voice:
vP ⇔ AV vP ⇔ UV
Subject
v VP Object
[+DEF] Subject
v VP
V Object
[−DEF]
V Object
[+DEF]
Note that one aspect these two kinds of analyses have in common is that voice is deter-
mined early in the derivation, whether by a lexical choice (merging vAV or vUV) or by a step
of object shift/agreement in the vP domain. In other voice-as-effect approaches, instead,
voice is determined late in the derivation, in the CP domain (V. Chen 2023).
In this dissertation, I will defend a type of voice-as-effect analysis of Äiwoo, adopting
ideas from Rackowski & Richards (2005), V. Chen (2017a, 2023), and Hsieh (2020, 2025).
I will argue that, essentially, UV morphology represents object agreement, and AV mor-
phology is the default/elsewhere case. More concretely, v agrees with certain classes of
direct objects (to be defined), but not with others. This step of agreement – or rather, the
object’s features copied back onto v – is then spelled out as “UV morphology”. If this does
not happen, then “AV morphology” is inserted instead. Effectively, AV is a morphosyn-
tactic default.
Chapter 3. Äiwoo clause structure 97
(128) Preview of the analysis of Äiwoo:
a. Actor Voice: no object agreement b. Undergoer Voice: object agreement
vP⇔AV vP ⟨φO⟩ ⇔UV
Subject Subject
v VP v ⟨φO⟩ VP
V Object V Object
3.5 DERIVING CANONICAL CLAUSES
We are now ready to begin to address the “WordOrder Puzzle” (howword order and verbal
voice morphology are derived), for now in canonical AV and UV clauses. In this section
I develop an analysis of these Äiwoo clauses based on standard Minimalist assumptions
(see chapter 1, §1.5). Such an analysis must minimally answer the following questions:
(129) a. What is voice morphology (a spell-out of), and how does it come to be?
b. How is word order determined? More specifically:
i. How, when, and why do the arguments move to their surface position?
ii. How does the VIA-first V2 word order arise?
iii. How do the properties of these movements fit within the A/Ā-distinction?
c. Are argument movement and voice morphology two separate processes, or
just one? How are they related?
For ease of reference, from now on I will name the process by which an argument moves
past another “(syntactic) inversion”. Concretely, in UV we observe inversion since the ob-
ject moves from its base-generated position c-commanded by the subject (by standard
UTAH assumptions; Baker 1988) to a higher position where it c-commands the subject.
As pointed out in this chapter’s introduction, a model of the Äiwoo clause is only good
if beyond getting the word order facts right, it also makes correct predictions with respect
to other aspects of the syntax of the language. Based on word order facts and on evidence
from patterns of Ā-extraction (which will be the focus of chapter 4), I will develop and
defend an analysis of the syntax of the Äiwoo clause with the following core features:
(130) Core aspects of the proposal:
a. Voice-as-effect: voice morphology is the spell-out of object agreement by v,
and this happens low/early
b. The inversion step happens high/late: the object undergoes non-local A-move-
ment to spec,CP
c. The V2 order arises from the verb raising to C by (long) head movement
98 3.5. Deriving canonical clauses
To give a more concrete preview of where we are heading, the structure I will propose
for AV and UV clauses is schematically represented in (131). In both, the verb first head-
moves into v and then undergoes long head-movement to C, skipping the TAM particles
in T. In AV (131a), there is no agreement between v and the object; this will result in the
verb being spelled out with default (= AV) morphology. The subject raises first to spec,TP
and then to spec,CP. In UV (131b), v agrees with the object; this is what will be spelled
out as UV morphology on the verb. Then, the subject moves to spec,TP, just like in AV.
However, this time it is the object that moves to spec,CP, across the subject.
(131) General structure of Äiwoo clauses:
a. Actor Voice:
[CP S V.AV [TP tS =TAM [vP tS tV+v tV O ]]]
b. Undergoer Voice:
[CP O V.UV [TP S =TAM [vP tS tV+v tV tO ]]]
From this sketch, we can already observe the core aspects of the proposal outlined in
(130). Voice is determined early, in the vP domain: UV if there is object agreement, AV
otherwise. Then, inversion happens late: in UV, the object does not raise past the subject
until the CP domain. Finally, the V2 effect is derived by head-movement of the verb into
C. This analysis will show some important similarities to the one developed in Pearson
(2001, 2005) for Malagasy and especially in van Urk (2015) for Dinka, particularly in the
derivation of the V2 effect and the resulting combination of A/Ā-properties of argument
fronting to spec,CP. This will be shown to be essentially pure A-movement but with less
strict locality conditions than usual, since it can skip past an intervening argument. In
other words, this will be an instance of non-local A-movement (§3.7).
I will now discuss my analysis of the syntax of the Äiwoo clause. I will proceed bottom-
up, examining first the lower vP domain (§3.5.1), then the middle portion of the clause (§
3.5.2), and finally the higher domain (§3.5.3). For clarity, in §3.5.4 I give full annotated
step-by-step derivations representing the entire system developed here. Finally, in §3.5.5
I discuss the pattern of φ-agreement.
3.5.1 THE vP DOMAIN: VOICE IS DETERMINED LOW
3.5.1.1 THE POSITION OF VOICE MORPHOLOGY AND THE MIRROR PRINCIPLE
If we pursue an analysis where voice is something that happens as an effect of the com-
bination of various syntactic factors, we then need to determine when exactly it hap-
pens throughout the course of the derivation. The Mirror Principle (Baker 1985, 1988)
has been invoked in V. Chen (2023) as a clue. In Tagalog and the Formosan languages,
voice morphology surfaces as rather external, or far away from the verb stem. For exam-
ple, in Puyuma, Paiwan, and Tagalog, voice morphology surfaces outside Aspect (132),
being either infixed or prefixed to progressive, habitual, and future morphology respec-
Chapter 3. Äiwoo clause structure 99
tively. Moreover, also in Puyuma voice shows allomorphy conditioned by Mood (133): the
Locative Voice suffix is -ay in the indicative (133a), but -i in the imperative (133b).
(132) Voice morphology is outside Aspect:
a. d⟨em⟩a-deru i atrung dra patraka (Puyuma)
⟨AV⟩PROG-cook PN.PIV Atrung INDEF.ACC meat
‘Atrung is cooking meat’ (A. H.-c. Chang 2006: 54, as cited in V. Chen 2023: 10)
b. s⟨em⟩iu-siup ti zepul nu s⟨em⟩iaw (Paiwan)
⟨AV⟩HAB-suck PN.PIV Zepul IRR.TEMP ⟨AV⟩soup
‘Zepul sucks (it) when she eats soup’ (A. H.-c. Chang 2006: 54, as cited in V.
Chen 2023: 10)
c. i-pa~pa-bili ko kay Crispin ang tsokolate (Tagalog)
CV-FUT~CAUS-buy 1SG.GEN OBL.P Crispin NOM chocolate
‘I will make Crispin buy the chocolate’ (Hsieh 2020: 35)
(133) Voice inflects for Mood: (V. Chen 2023: 10) (Puyuma)
a. ku-beray-ay i senten dra paysu
1SG.NOM-give-LV.IND PN.PIV Senten INDEF.ACC money
‘I gave Senten money’
b. beray-i i senten dra paysu!
give-LV.IMP PN.PIV Senten INDEF.ACC money
‘Give Senten money!’
V. Chen (2023) takes this to signify that what is spelled out as voice morphology is some-
thing that happens relatively late in the derivation12.
However, in Äiwoo the situation looks quite different. As we saw in §3.3.1, voice is
mostly reflected in Äiwoo by a system of stem alternations: (almost) every verb has two
stems, one per voice, and the relationship between the two forms is largely idiosyncratic/
suppletive. Consider again the various alternations in (134), repeated from above.
(134) Voice-based stem alternations in Äiwoo:
a. epavi epave ‘cook.UV/AV’ f. lâbu lâbonge ‘cut.UV/AV’
b. lââ lâwââ ‘build.UV/AV’ g. kili kei ‘dig.UV/AV’
c. tu tou ‘bring.UV/AV’ h. pââ pä ‘steal.UV/AV’
d. togulo togo ‘hit.UV/AV’ i. ve veve ‘buy.UV/AV’
e. ngä vängä ‘eat.UV/AV’ j. kali ekäi ‘sing.UV/AV’
k. la lää ‘give.UV/AV’ l. nu nu ‘drink.UV/AV’
By the same Mirror Principle logic, this morphological fact would teach us that voice in
Äiwoo is determined early in the derivation, not late, as it’s very close to the verb stem.
In fact, it’s internal to the stem, and most importantly on the inside of any other inflec-
tional category (e.g., aspect and agreement). Two influential analyses of Tagalog propose
12 Although seeWu (2024) for arguments against Chen’s analysis, and a proposal for Seediq (Formosan) where
voice is low. In this proposal, voice is above inner aspect (Travis 2010), but crucially lower than other Infl-
region aspectual heads.
100 3.5. Deriving canonical clauses
that voice morphology is established early in the derivation, and for both analyses, voice
morphology reflects case agreement between v (or Voice) and one DP per sentence. For
Rackowski & Richards (2005), exactly which DP v agrees with depends on an independent
process of object shift, determined by definiteness/specificity. For Hsieh (2020, 2025), this
is instead determined by v agreeing with different arguments based on the distribution of
Ā-features13. Here, what I propose is a variant of Hsieh’s model.
3.5.1.2 UV MORPHOLOGY IS OBJECT AGREEMENT; AV IS DEFAULT
SETTING THE STAGE Following Hsieh (2020, 2025), I assume that DPs in a sentence can
be base-generated with an Ā-feature [PIV(OT)] (see also Kroeger 1993: chap. 3, Richards
2000, V. Chen 2017a: chap. 4). I will consistently call the argument carrying [PIV] the
pivot. In the canonical clauses under discussion here, the pivot will always coincide with
the VIA; these two notions will importantly come apart in the next chapters, when we
turn to non-canonical clauses and cases of non-VIA extraction. This [PIV] feature has
somewhat vague information-structural correlates14, and is essentially syntactically free,
in the sense that the speaker freely chooses to place it on one nominal per sentence. In
keeping with the idea that [PIV] is syntactically free, I posit no restriction on how many
nominals per clause may or may not carry [PIV]: there may be one, or zero, or several15.
A piece of evidence that [PIV] is a information-structure-related Ā-feature comes from
acceptability patterns in question/answer pairs. To a subject question, only a UV answer is
acceptable (135a); viceversa, to an object question, only an AV answer is acceptable (136a).
(135) Subject question: 3 UV, 7 AV  (136) Object question: 3 AV, 7 UV
Q: “Who ate (the) mango?” (240909) Q: “What did Anna eat?” (240909)
a. mangoPIV i-ngä Anna a. AnnaPIV i-vängä mango
mango ASP-eat.UV Anna Anna ASP-eat.AV mango
‘Anna ate the mango’ ‘Anna ate (a) mango’
b. # AnnaPIV i-vängä mango b. # mangoPIV i-ngä Anna
Anna ASP-eat.AV mango mango ASP-eat.UV Anna
At the very least, this reveals that our stipulated Ā-feature [PIV] has at least one trait in
common with a bona fide Ā-feature like [TOP(IC)]: both are incompatible with the kind
13 The proposal in V. Chen (2017b, 2023) is similar to Hsieh’s (2020) in that verbal voice is determined on the
basis of Ā-features, but it differs from his proposal on the timing of this. For Hsieh, this happens early (in
the vP domain); for V. Chen, this happens later in the derivation (at the CP level, approximately).
14 The mechanisms and motives behind voice choice in Äiwoo are not well understood. The reader is referred
to Holmen (2020) and Holmen & Næss (2025) for corpus studies trying to elucidate which factors are or are
not important to determine what voice will be chosen by a speaker for any given sentence.
15 This eliminates a stipulation made in Hsieh (2020, 2025), where one and only one nominal per clause must
carry [PIV]. We will see that the system will still predict the correct results even without this stipulation.
Chapter 3. Äiwoo clause structure 101
of focus that is involved in answering questions. In English, for example, topicalizing the
constituent that would answer the question is similarly infelicitous16:
(137) a. Q: “Who ate the mango?”
# A: (As for) Anna, she ate the mango
b. Q: “What did Anna eat?”
# A: (As for) the mango, Anna ate it
However, [PIV] cannot have the exact same semantic/pragmatic import as [TOP]. For ex-
ample, certain kinds of non-referential nominals (e.g. quantified expressions) cannot be
topicalized in many languages, for example Italian (138a) (Rizzi 1986 et seq.), whereas they
can be pivots in Äiwoo (138b). Whatever it is that makes [TOP] incompatible with a quan-
tified expression in Italian (see Chierchia 2025 for a discussion), [PIV] must not share this
property.
(138) a. Italian: a quantified expression cannot be a topic
* [tutti gli studenti]TOP, li ho portati al concerto
all the students them have.1SG brought.PL to.the concert
Intended: ‘(As for) every student/all students, I brought them to the concert’
b. Äiwoo: a quantified expression can be the pivot (250717)
[pedevalili dâuwângâ]PIV i-tu-du-kâ-no ngä skul
children all ASP-bring.UV-all-DIR3-1MIN to school
‘I brought every child/all children to school’
For the purposes of this dissertation, I follow Hsieh (2020, 2025) in only stipulating that
there exists a syntactic Ā-feature [PIV], with a semantic/pragmatic contribution similar
to [TOP] (yet not identical), and with completely free syntactic distribution (contra Hsieh
2020, 2025). I leave it to further research to shedmore light onwhat exactly the interpretive
import of [PIV] is, and what the constraints on its distribution are (if any).
IMPLEMENTING THE SYSTEM At this point, we have spelled out concretely what it
means for an argument to be “the pivot”: a pivot argument is a nominal that carries the
Ā-feature [PIV]. Now, we need to make sure that verbal morphology will be sensitive to
the presence vs. absence of this feature, to capture the contrast between AV and UV. I pro-
pose that when the object carries Ā-features, v agrees with it, and copies back its features;
this is then spelled out as UV morphology. Let us now go through the derivation to see
how exactly this happens.
16 For what it’s worth, the same judgments also hold for at least Italian and Norwegian, two languages where
topicalization is impressionistically more productive/frequent than in English.
102 3.5. Deriving canonical clauses
I propose that the v head carries a probe with the specification [INT: Ā; SAT: D]17,18.
If the object has Ā-features (139a), v will agree with it, and thus copy back its features
(the subscripted ⟨φO⟩). If, on the other hand, the object is just a normal (Ā-featureless)
DP (139b), v will not copy back any features (because the interaction condition, [Ā], is
not met), but it will stop probing (because the satisfaction condition, [D], is met)19. For
comparison’s sake, I have represented this case as onewhere [PIV] is on the subject. This is
not important, as the subject does not get agreed with by v; the same result would obtain
if neither argument carried [PIV].
(139) a. DP[PIV] object: object agreement b. DP object: no object agreement
vP ⟨φO⟩ vP
DP DP[PIV]
[ v ⟨φ] ⟩ VP [ v ] VPO
I: [Ā] I: [Ā]
S: [D] S: [D]
V DP[PIV] V DP
I assume that the verb head-moves into v (not represented in these trees for simplicity),
and the thus formed complex head is then mapped onto the correct stem for each verb.
Simply put, the stem alternation is conditioned allomorphy. If a root is in the context of a
v with φ-features on it, the UV exponent will be inserted; else, the AV exponent will, as a
default form. Some sample Vocabulary Insertion rules are given in (140).
17 Why the satisfaction condition is set as [D] will become clear in chapter 6, where I discuss non-canonical
UV clauses with pronominal objects. The probe specification as described here will also be partially revised
to account for the particular interactions between pronominal objects and Ā-extractability patterns. For
now, this is sufficient.
18 The idea of “Ā-agreement” is not new. Aside from the context of Austronesian voice, this has been pro-
posed for several languages, though it is often labelled “wh-agreement” (since it’s often analyzed in the
context of wh-questions). Some examples are Chamorro (Austronesian; Chung 1994), Abaza (Northwest
Caucasian; O’Herin 1993, 2002, Arkadiev & Caponigro 2021), Kinande (Bantu; Baker 2003), SanMartín Peras
Mixtec (Oto-Manguean; Ostrove 2018), Khanty (Uralic; Colley & Privoznov 2020), Ripano (Italo-Romance;
D’Alessandro 2020), Tibetan (Sino-Tibetan; Erlewine 2020), Kawahíva (Tupi-Guaraní; dos Santos 2023), and
Tira (Niger-Congo; Jenks 2023).
19 The system described here only applies to transitive verbs. For completeness, intransitives are treated sep-
arately, in §3.6, where a minor amendment is made to the system.
Chapter 3. Äiwoo clause structure 103
(140) Some Vocabulary Insertion rules for voice-conditioned stem allomorphy:
a. √COOK ⇔ epavi / v ⟨φ⟩ c. √BRING ⇔ tu / v ⟨φ⟩
√COOK ⇔ epave (elsewhere) √BRING ⇔ tou (elsewhere)
b. √HIT ⇔ togulo / v ⟨φ⟩ d. √EAT ⇔ ngä / v ⟨φ⟩
√HIT ⇔ togo (elsewhere) √EAT ⇔ vängä (elsewhere)
This system can easily be extended to capture the pattern of voice concord described in
§3.3.1. As a reminder: adverbial modifiers suffixed to UV stems must take an additional
suffix -i/-nyii, depending on the specific lexical item (141a). This never happens when the
same adverbial modifiers are suffixed to AV stems (141b).
(141) Voice concord in UV: (240425)
a. nuwopa eângâ ki- lââ-päko-i-mana-i -no=ngaa
house that IPFV-build.UV-good-UV-very-UV-1MIN=FUT
‘I will build that house/those houses very well’
b. i-ki- lâwââ-päko-mana =kaa nuwopa
1MIN-IPFV-build.AV-good-very=FUT house
‘I will build a house/houses very well’
For concreteness, I assume that these structures instantiate V-V (or root-root) compound-
ing. The complex V head might be merged as such in-situ, as sketched in (142), or created
through successive-cyclic V-to-V head movement; the difference between these options
is immaterial20. All three stems in this complex form then undergo the same kind of po-
sitional allomorphy in the context of v (143)21⟨φ⟩ .
(142) vP (143) a. √BUILD ⇔ lââ / v ⟨φ⟩
√BUILD ⇔ lâwââ (elsewhere)
Subj b. √GOOD ⇔ päko-i / v ⟨φ⟩
√GOOD ⇔ päko (elsewhere)v
VP
c. √VERY ⇔ mana-i / v ⟨φ⟩
V v ⟨φO⟩ √VERY ⇔ mana (elsewhere)
tV DP[PIV]
√ V
BUILD √ V √V
GOOD VERY
20 I don’t take the fact that [very well] forms a complex modifier of [build] to signify that these structures are
formed by phrasal (VP) movement, contra Hedding & Yuan (2023).
21 This is a relatively rough sketch. A more well-developed morphological analysis should hopefully offer a
principled explanation for the fact that whereas many transitive stems show (near-)suppletion for voice
or other idiosyncratic allomorphy patterns, adverbial modifiers like päko ‘good’ and mana ‘very’ just take
a consistent suffix -i, with no suppletion (see Wu et al. 2023 for an argument that -i might be the default
morphological realization of “UV”, although with different syntactic assumptions as to what “Undergoer
Voice” is than what I propose here). For the purposes of this dissertation, however, this suffices, and I leave
the development of a more sophisticated morphological analysis to future work.
104 3.5. Deriving canonical clauses
CONCEPTUAL PARALLELS The “UV as object agreement” system just outlined bears a
resemblance to the case agreement mechanism proposed in Hsieh (2020, 2025), and also
to earlier case agreement proposals like Rackowski & Richards (2005). In Hsieh’s (2020)
analysis, v agrees with whichever nominals carries the [PIV] feature, and copies back
its case feature (NOM or ACC); this is then spelled out as either AV or UV morphology.
This analysis is based on Tagalog, a language with overt case marking, so it’s somewhat
natural to posit that voice morphology is an overt reflex of agreement between v and
the case features of the pivot. However, Äiwoo doesn’t have any case morphology, so the
direct evidence behind such an analysis would be weaker. More problematically, this step
of agreement by v happens rather early in the derivation – as soon as v is merged, even
before the external argument is present. Therefore, it would not be obvious how to model
first case assignment and then further copying of case features back to a probe at this
early stage of the derivation22.
As a side note, one could also imagine that when v agrees with the object (139a), the
object is additionally shifted to an inner specifier of vP, below the subject; not-agreed-with
AV objects (139b) remain in situ (in a manner reminiscent of Rackowski & Richards 2005).
This step of movement would crucially have to not bring the object above the subject, for
reasons that will become clear in §3.5.223:
(144) a. Undergoer Voice: DP[PIV] objects shift
[TP … T [vP DP DP[PIV] v ⟨φ ⟩ [O VP V tO
b. Actor Voice: DP objects remain in situ
[TP … T [vP DP[PIV] v [VP V DP
This object shift idea would be theoretically attractive, as it would bring Äiwoo closer
to what we know about movement of objects to the phase edge in many languages. Ā-
marked objects (henceforth, “DPĀ”) are among the typical classes of nominals that could
be described as undergoing object shift, since they often move successive-cyclically to
the vP phase edge on their way to further movement to spec,CP (see e.g. van Urk 2020,
Keine & Zeijlstra 2025 for recent overviews with focus on the clause-internal vP phase)24.
22 An analysis of Äiwoo that is closer to the case-agreement models cited above can be developed, but in order
to correctly model the complexity of the whole system, one needs to make additional assumptions about
the inner workings of agreement. For concreteness, what such a system would have to look like is laid out
in Appendix B, but I do not assume this here as I consider the current analysis more parsimonious since it
necessitates fewer assumptions.
23 Formally, this could either be done by Tucking In (Richards 1997), or by ordering the step of Internal Merge
of the object before the External Merge of the subject.
24 The other typical class of objects that shift are pronouns (e.g. in Scandinavian languages; see Vikner 2017
and references therein). Indeed, we will see in chapter 6 that pronominal objects in Äiwoo are treated by v
just like Ā-carrying objects, and they are always agreed with regardless of Ā-features. All in all, the class of
objects that v agrees with in Äiwoo (thus triggering UV morphology) is precisely the union of two classes
of objects that in many languages are shifted to the vP phase edge.
Chapter 3. Äiwoo clause structure 105
However, because all PPs and adjuncts are linearized to the right in Äiwoo, this step of
object shift would be string-vacuous, and it has therefore proven quite hard to find any
independent evidence of it. Therefore, I don’t incorporate object shift into my analysis. In
a sense, given the framework adopted here where agreement is a necessary precondition
for movement, my analysis involves the first step of “object shift” (v agreeing with the
direct object), but not the second (the object actually moving). Notably, even if it is indeed
discovered that string-vacuous object shift does take place in Äiwoo as described here,
nothing would have to change in the system of probing and agreement described in this
dissertation.
In the “UV as object agreement” approach outlined here, the two voices are not on
equal standing (like they are in Hsieh’s 2020 proposal). Rather, UV is morphologically
marked, whereas AV is really a default/elsewhere spell-out. This is in line with a number
of analyses of other Austronesian languages, where it has been argued that Actor Voice
is the a morphosyntactic “default”, or “lack of voice” (T. Chen 2010, Levin 2015, Wurm-
brand 2016, Wurmbrand & Shimamura 2017, H. Y. Chang 2017, Wu et al. 2023). Moreover,
my analysis essentially makes “Undergoer Voice” a sort of head-marking counterpart
of Differential Object Marking (DOM; Bossong 1991, Aissen 2003, et seq.). In canonical
(dependent-marking) DOM patterns, certain types of “special” objects are singled out and
carry special morphology. In this version, the special marking takes place on the verb
taking said special objects as its complements, in the form of UV morphology.
So far, we have captured the correlation between voice morphology and pivothood
by following Hsieh (2020, 2025) in assuming that it’s grounded in agreement between v
and a nominal carrying Ā-features, but implementing in it a slightly different way. In my
analysis, only UV is a marked category, spelling out agreement between v and a DP[PIV]
object; AV is the morphology that surfaces when v doesn’t agree with anything (rather
than when it agrees with the subject, as in Hsieh’s model). This difference in implementa-
tion will become crucial throughout the next chapters in addressing both the Word Order
Puzzle and the Extraction Puzzle.
My analysis diverges from both Hsieh (2020, 2025) and earlier case-agreement-based
proposals like Rackowski & Richards (2005) in what happens immediately after this agree-
ment step. In those proposals, whichever argument v agrees with always ends up as the
highest argument in the vP phase. In the case of AV, the subject is already the highest ar-
gument, so nothing special happens. In the case of UV, the object gets moved to a position
above the subject, something that would be crucial to the rest of the derivation proposed
in those models. For Äiwoo, instead, I propose that there is no syntactic inversion at this
stage. In other words, the object does not raise across the subject. Even if we incorporate
object shift into our analysis, as discussed above, this posited movement step would bring
the object to an inner specifier of vP, still below the subject. The only aspect that gets
established at this point of the derivation is the morphological spell-out of the verb. The
main argument for not positing inversion this early comes from what goes on in the TAM
region of the clausal spine, which will be the topic of the next section.
106 3.5. Deriving canonical clauses
3.5.2 THE TP DOMAIN
Let us remind ourselves of the word order facts we’re trying to model. Every clause is
VIA-initial, and the VIA immediately precedes the verb. However, the non-VIA arguments
(objects in AV, subjects in UV) are not in the same linear position. AV objects are to the
right of the series of TAM particles (145a). In contrast, UV subjects are between the verb
and the TAM particles (145b).
(145) VIA V (S) =TAM (O) (adjuncts) (230706)
a. Anna ki-vängä =kaa sii ngä täpilo enge
Anna IPFV-eat.AV =FUT fish in bowl this
‘Anna will eat fish in this bowl’
b. sii ki-ngä Anna =kaa ngä täpilo enge
fish IPFV-eat.UV Anna =FUT in bowl this
‘Anna will eat the fish in this bowl’
In this section, I show that the position where UV subjects are – ‘Anna’ in (145b) – is a
derived position that the subject moves to. I will argue that this is a specifier position in
the T/Infl region of the clause, which I identify with spec,TP for simplicity. (This labeling
choice, i.e. this position being exactly spec,TP, is not particularly crucial; it is only im-
portant to me that this is a specifier position of a projection higher than the vP domain
and below the CP domain.) The two main pieces of evidence will be (i) the TAM-related
semantics of the particles in this position; (ii) the existence of a lower position where
subjects can sometimes surface.
3.5.2.1 THE POSITION BETWEEN V AND TAM IS SPEC,TP
The first argument for this position being spec,TP is once again based on Mirror Princi-
ple considerations. The set of particles that appear in this region consists of tense/aspect
particles, negation, and the CV marker (chapter 2, §2.6.3; see also Roversi 2019: §2.3.2).
Setting this last one aside, as it appears to have a different function than the others, (146)
is a sentence featuring the three main “slots” in the series, here illustrated by =to PRF, =gu
NEG and =Caa FUT.
(146) nuwopa eângâ ba ki-ve Pita =to=gu=naa ngä taun (250717)
house that NEG IPFV-buy.UV Peter=PRF=NEG=FUT in town
‘Peter won’t already have bought that house in town’
Some descriptive facts about the semantics of =to PRF and =Caa were described in chapter
2, §2.6.3. Because the semantics conveyed by the particles in this region are related to
tense and aspect (and polarity), it is reasonable to hold as a null hypothesis that they must
be in the general T/Infl-region, above the θ-domain (Ramchand & Svenonius 2014, a.m.o.).
If the subject – or any nominal, for that matter – surfaces to the left of these particles in
the T/Infl domain, then it would be hard to maintain that it’s still in its base-generated
position in spec,vP.
Chapter 3. Äiwoo clause structure 107
First, we must make sure that the position between the verb and the TAM particles is a
syntactically defined position. In other words, despite these particles being morphophono-
logical clitics, they are not linearized according to phonological principles. If we manipu-
late the phonological size of the nominal constituent filling this position, we observe that
this has no influence whatsoever on where the TAM particles are. The nominal in this slot
between the verb and the TAM particles can be as phonologically small as segmentally
null, only consisting of a floating [nasal] feature (147a). This nominal can also be very
large, containing an entire relative clause with its own arguments and adjuncts (222b,c).
Heavy DPs of this size in this position are somewhat uncommon, but they are not an
artifact of elicitation: sentence (147c) is attested in our natural speech corpus.
(147) The nominal between V and TAM can be of arbitrary phonological size: (240215)
a. nuwopa enge ba ki-ve [-∅n]=gu=naa
house this NEG IPFV-buy.UV-3MIN=NEG=FUT
‘S/he will not buy this house’
b. gino-mu ba ku-mangi [kuli (mi=)[ki-ngoduwâ-lâoo ngä paveli
son-2MIN NEG IPFV-bite.UV dog REL=IPFV-shout-always in garden
eângâ]]=gu=naa
that=NEG=FUT
‘The dog that’s always barking in that garden won’t bite your son’
c. ngamaa lâ ba i-kää [me=[ki-tokoli-woli-mä ngä
PRT DIST NEG ASP-know.UV REL:person=IPFV-sit-down-DIR1 in
botu]]=gu=nâ
boat=NEG=DIST
‘If the person sitting in the boat doesn’t know (it), …’ (Nyel. toponu 141)
Having established that the position between V and =TAM is a syntactically defined one,
we can now show that it is not the in-situ/base-generated position of the subject, but a
derived one. The most conclusive evidence to support this comes from a type of UV clause
whose properties will not be discussed until chapter 6. For now, I will use the sheer fact
that these clauses exist as an argument, and I ask the speaker to wait until chapter 6 to
see how they will fit into our model.
In this type of UV clause, the position between the verb and the TAM particles is oc-
cupied by the object, instead of the subject. Crucially, when the object is in this position,
the subject must exceptionally surface in a lower position, to the right of the TAM par-
ticles. In (148a), we see the pronominal marker -gu-de OBJ-12AUG occupying the position
between the verb and the TAM particles (and doubling the preverbal object pronoun iude
‘us.INCL’). Consequently, the subject ‘John’ comes after the negation and future particles.
Putting the subject in the canonical position between the verb and the TAM particles in
this context is ungrammatical (148b).
108 3.5. Deriving canonical clauses
(148) Evidence for a lower subject position: (240307)
a. (iude) ba ku-tu-kä-gu-de=gu=naa John ngä skul
12AUG NEG IPFV-bring.UV-DIR3-OBJ-12AUG=NEG=FUT John to school
‘John won’t bring us to school’
b. * (iude) ba ku-tu-kä-gu-de John =gu=naa   ngä skul
12AUG NEG IPFV-bring.UV-DIR3-OBJ-12AUG John=NEG=FUT to school
I assume that this lower position of the subject is its original base-generated one, in
spec,vP. Therefore, I argue that the structural position that linearly corresponds to the
slot between the verb and the TAM material is spec,TP.
3.5.2.2 MOVING NOMINALS TO SPEC,TP (A FIRST PASS)
Having established the nature of this position as spec,TP, this has important consequences
for our general analysis of the syntax of Äiwoo clauses. The most straight-forward one is
that in a very familiar way, Äiwoo subjects move to spec,TP. For a concrete implementa-
tion, I propose that T hosts a probe that simply raises the first/highest nominal it finds,
regardless of what v has agreed with25.
The existence of this step of subject movement is instructive with the respect to the
timing of the syntactic inversion between the object and the subject, which results in
OVS order in UV. If subject movement to spec,TP is local, this means that at this point
of the derivation the subject is still the highest argument. This entails that despite voice
morphology being established early in the derivation, inversion happens late.
Consider what an early-inversion approach would predict instead. In such a model,
adopted in several other influential analyses (Rackowski 2002, Aldridge 2004, 2008, Rack-
owski & Richards 2005, Erlewine 2018, Hsieh 2020, 2025, a.o.), v not only (case-)agrees
with the pivot/VIA but it also moves it to its outermost specifier, making it the highest
nominal in the θ-domain26. Specifically, in UV, the object moves past the subject:
(149) … [ T [vP Obj [vP Subj v
⇔UV [VP V tObj ]]]]
25 As stated, in this chapter I am exclusively concerned with deriving what I call canonical clauses. As we
observe from (148a), there are indeed sentences where I will argue that the object moves to spec,TP instead
of the subject. These are treated in chapter 6, where I will develop amore sophisticatedmodel of what kind of
constituents move to spec,TP and how this movement is derived. Crucially, that will not affect the argument
being made here. For expository reasons, therefore, I will develop the argument here ignoring those kinds
of sentences, and I ask the reader to trust that the system will still hold once we incorporate that added
complexity into our model. Ultimately, we will end up with a model where movement to spec,TP is mostly
local, with exactly one kind of well-motivated exception that we can capture with a neat generalization.
This is preferable to the opposite, that is, a model where movement to spec,TP is mostly non-local, and
where we are not able to draw a generalization about the natural class of nominals that move to spec,TP.
See chapter 6 for details.
26 Here I abstract away from the difference between voice-as-cause approaches, like Aldridge (2004), and voice-
as-effect ones, like Rackowski (2002).Whatmatters for now is that in both, syntactic inversion happens early
(at the vP/VoiceP level) rather than later.
Chapter 3. Äiwoo clause structure 109
Of course, the obvious problem now for Äiwoo is how to model the movement of the
subject to spec,TP, which in this case would be irredeemably non-local27. I propose that
(in the ordinary clauses under discussion) T instead acts very simply, and simply moves
the highest nominal it finds, which in both cases is the subject. This entails that syntactic
inversion happens late. Formally, the probe specification on T is [INT: φM; SAT: φ]. In
prose: it agrees with and moves the first φ-bearing constituents it finds, and immediately
stops. The steps of agreement and movement modeled so far are summarized in (150) for
both AV and UV28.
(150) a. Actor Voice: b. Undergoer Voice:
TP TP
DP DP
[PIV]
T vP⇔ AV T vP ⟨φO⟩ ⇔UV
[ =TAM ] [ =TAM ]M
I+S: [φ]M I+S: [φ]
t tSbj Sbj
[ v ] VP [ v ] VP
I: [Ā] I: [Ā]
S: [D] S: [D]
V DP V DP
[PIV]
3.5.3 THE CP DOMAIN: VIA FRONTING
In the previous section I have argued that subject movement to spec,TP happens locally,
and that therefore, at that point of the derivation – in either voice construction – the
subject is still the structurally highest DP. The question is now how to model the syntactic
inversion process which results in OVS word order in UV. Necessarily, there will have to
be a step of non-local movement, with the object being raised across the subject. We ought
to ask two questions:
27 One possible approach would be to posit that the probe on T is case-discriminating (see e.g. Otsuka 2006,
2010, Bobaljik 2008, Legate 2012, Deal 2016b, Drummond 2023, Akkuş et al. 2025, Deal et al. 2024, a.o.). If
our aim is for this probe to skip the higher object and move the subject non-locally, we could posit that it
only specifically targets nominative arguments, and not case-marked/accusative ones. However, this will
make wrong predictions for those non-canonical clauses where T visibly raises the object, and leaves the
subject in situ (148a). Therefore, we cannot adopt case discrimination as a way for the probe on T to trigger
non-local movement.
28 This proposal would also not change in any meaningful way if we incorporated object shift into the analysis
as described in §3.5.1.2. The putative step of object shift discussed above would bring the object to an inner
spec,vP, below the subject. Therefore, even with object shift the subject would still be the closest argument
to T, and the derivation would proceed as illustrated here.
110 3.5. Deriving canonical clauses
(151) a. Question 1: how does this non-local movement happen, mechanically?
b. Question 2: what kind of movement is it exactly (A, Ā, mixed)?
In the rest of this section, I will answer question 1, and thus build a complete model for
deriving AV and UV clauses. Question 2 is taken up in §3.7, where I show that the me-
chanics I posit here correctly predict the properties of this movement. We will see that the
process of VIA fronting will show properties of A-movement, despite being non-local (van
Urk 2015, et seq.). This specific combination of properties will be shown to be motivated
by the featural specification of the probe driving this movement.
Before that, a note is in order concerning the landing site of this movement. For con-
creteness and ease of exposition, following the spirit of the idea of Äiwoo as a V2 language,
I have been labelling this position as spec,CP, so that VIA fronting is triggered by a probe
on C. However, it is not crucial to the analysis that this head be identified specifically with
C. Another equally plausible option, for example, would be to label what I call here T and
C as two heads in an extended Infl-region of the spine, e.g. Infl1P (= TP here) and Infl2P (=
CP here). This idea might in fact have some merit. For example, it would explain the fact
that in embedded clauses containing a complementizer we can still see the same V2 word
order we see in main clauses (like Dinka according to van Urk 2015, but unlike Germanic
V2 languages, where embedded V2 is often highly restricted to certain specific kinds of
embedded clauses; see e.g. Julien 2007a, 2015, Bentzen et al. 2007, Heycock et al. 2010,
Bentzen 2014, a.o.). In (152), we can see that in the UV embedded clause, the fronted VIA
object buk ‘book’ follows the complementizer go. Depending on exactly what functional
head go spells out, this might indicate that the landing site of VIA fronting is not spec,CP
but a slightly lower projection.
(152) mekivaavee ki-giââ [CP go buk i-gââ-du Jen] (240530)
teacher IPFV-happy because book ASP-read.UV-all Jane
‘The teacher is happy because Jane read the whole book’
Here, I keep the labels TP and CP for simplicity, with the promissory note that this will
have to be made more precise in future research.
Consistently, I assume that the verb undergoes head movement to C. This will have
to be long head movement (Roberts 1994, 2010, Embick & Izvorski 1997, Rezac 2008,
Harizanov 2019, Harizanov & Gribanova 2019, Lee 2022, 2024), since it violates the Head
Movement Constraint (Travis 1984) by skipping the TAM material in the T/Infl region
(boxed):
(153) [CP FP[PIV] V+v+C [TP … =TAM [vP … tV+v [VP tV … ]]]]
An alternative analysis without non-local head movement is possible, if we assume that
the post-verbal TAMparticles are not heads of functional projections, but specifiers of null
heads (akin to sentential adverbials in Cinque 1999, et seq.). In that case, the verb could
undergo strictly local head movement through each head, and the TAM particles would
still correctly be predicted to appear in the attested position (154). (In this tree, both the
Chapter 3. Äiwoo clause structure 111
specific labels and the left/right distribution of heads in the complex head under C are
purely for exposition, and not a specific analytical commitment.)
(154) Alternative analysis without long head-movement:
CP
DP
[PIV]
C FP
C TAM2 DPSubj
∅ F TAM2P
TAM2 Neg
∅ PRF
=to t NegP
Neg TAM1 TAM2
ba
TAM1 v NEG
∅ =gu tNEG TAM1P
v V
FUT
=Caa tTAM1 vP
tSubj
tv VP
tV tO
Of course, assuming (154) instead of (153) would make several predictions concerning the
relative structural positions and semantic scope of the TAM and negation material, as
well as the landing site of the external argument. Throughout the rest of this dissertation,
I assume the long head movement-option for ease of exposition, with the note that this is
not a non-negotiable ingredient of the analysis.
3.5.3.1 C AGREES WITH WHAT v HAS AGREED WITH (BUT IT HAS A PLAN B)
To address our Word Order Puzzle, the first obvious desideratum is to capture the VIA-
first word order (OVS in UV; SVO in AV). This fact that can be restated as the following
extensionally equivalent correlation:
112 3.5. Deriving canonical clauses
(155) a. When v has agreed with an argument (the object, in UV clauses), C will front
that argument;
b. Else, C will front the highest argument it finds (the subject).
Let us first discuss the first sub-case, where C fronts a UV object. Later, we will extend the
discussion to what C does when it cannot find its first preferred target.
PLAN A: FRONT THE SAME ARGUMENT THAT v AGREED WITH Consider first the sub-
generalization in (155a). We can informally say that v and C “go together”: if v has agreed
with an argument, C wants that same argument too. Logically speaking, given the archi-
tecture of agreement, this can be done in two ways: either both v and C search for the
same feature(s), or C searches for whatever v happens to have agreed with. This latter op-
tion might seem less obviously possible, but it can be implemented with the mechanism
of goal flagging (Deal to appear; see chapter 1, §1.5). Whatever constituent v agrees with
will receive the interaction feature [i |v]29. Then, we need a probe on C that is specified
as [INT: i |vM; SAT: i |v]. Now, C will agree with and move the same constituent that was
agreed with by v30. This is, finally, the syntactic inversion step, where the object is raised
across the subject.
At this point, this goal flagging approach might seem unnecessarily cumbersome. A
simpler analysis, for example, would maintain that the probes on v and C go together
because they are searching for the same feature (i.e., [Ā], or even more specifically [PIV]);
this would at first glance get us the same effect. However, we will see in the following
chapters that adopting the goal flagging mechanism will be crucial to address the Extrac-
tion Puzzle, when we try to account for the Ā-extraction properties of different kinds of
Äiwoo clauses. For now, I ask the reader to temporarily trust that positing a probe on
C searching for [Ā] or [PIV] would make incorrect predictions about the Ā-extraction
patterns found in Äiwoo.
Here is how the goal flagging approach works, concretely. Consider the derivation of
a UV sentence, shown in the example below (156). First, v agrees with the DP[PIV] object,
copying its φ-features, and this is what will ultimately be spelled out as UV morphology
on the verb stem. As a consequence of this agreement step, the object now carries the
interaction flag feature [i |v]. Then, independently, T raises the subject to spec,TP. Finally,
C probes for an argument with the [i |v] feature. The closest such argument is the object,
and therefore C fronts the object across the subject. Note that this movement is non-local,
since C raises the object instead of the closer subject, but it is made possible by Relativized
Minimality (C wants a feature that the subject does not have, so the subject does not
intervene). We will see below that the goal flagging approach also correctly predicts the
cluster of properties of this movement to spec,CP. (For simplicity, again, I don’t represent
the head-movement of the verb, V-to-v-to-C.)
29 I leave open for now the question of exactly what kind of feature [i |v] may be. See §3.7.2 for a discussion.
30 If, for any reason, there happened to be more than one constituents with the [i |v] flag feature, C would
move the highest of them. This possibility will come in hand in chapter 6.
Chapter 3. Äiwoo clause structure 113
(156) C raises the argument v agreed with (to be slightly revised): UV derivation
CP
DP
[PIV, i|v] [ C ] TP
I+S: [i|v]M
DP
T vP ⟨φO⟩⇔UV
tS v VP
V DP[PIV, i|v]
I take no stance here as to whether vP constitutes a phase in Äiwoo, and if so, whether
an object fronted to spec,CP moves successive-cyclically through an inner spec,vP (below
the subject). As discussed in §3.5.1.2, this step of “object shift” would be theoretically
appealing, but there is no clear independent evidence for it, so I represent the movement
to spec,CP as happening in one fell swoop. Not much hinges on this, and an intermediate
landing site could be easily incorporated without any changes to the analysis.
PLAN B: JUST MOVE THE HIGHEST NOMINAL In AV clauses, we know that C fronts the
subject. These are also precisely the type of clauses where, by hypothesis, v didn’t agree
with anything (§3.5.1.2). Considering this together with the previous case, we want a sys-
tem where a probe on C will have a “primary target”, but if it can’t find that, it will fall
back to a less specific kind of target. In other words: if C can find a nominal flagged with
[i |v], it will raise that; if not, any nominal will do. As a result, the closest/highest one will
automatically be targeted – that is, the subject. To do so, we will only marginally revise
the system described in the previous section to incorporate this “plan B” logic. To this
purpose, I adopt an implementation developed in Scott (2021). The mechanism she pro-
poses is based on the idea ofMultitasking (van Urk & Richards 2015, Bossi & Diercks 2019)
and achieves the same results, but without the need of positing additional mechanisms/
operations in the grammar of Agree(ment).
The systemworks as follows.We assume that C has two ordered probes: the “primary”
probe described earlier, specified as [INT: i |vM; SAT: i |v], and a “secondary” probe with a
much more general featural specification [INT: φM; SAT: φ] (assuming that [φ] is present
on all nominals). In prose: the primary probe aims to move the highest constituent that
has been agreed with by v; the secondary probe just aims to move the highest nominal,
whatever that is. The primary probe acts first, and the secondary one acts later.
114 3.5. Deriving canonical clauses
Now we can see how the two probes will operate together. First, let’s consider again
the case of a UV derivation (157), where v has agreed with the object and flagged it with
[i |v]. Now, the primary probe on C will find this viable goal with [i |v], and move it to
its specifier. So far, this is exactly as described in the previous example. Consider now
how adding the secondary, less specific probe won’t change the derivation (which is a
good thing, since it already got the word order fact rights). Following the assumptions
of Cyclic Agree (Béjar & Rezac 2009), which ultimately just boil down to assuming bare
phrase structure (Chomsky 1995, 2000), the as-yet unsatisfied secondary probe reprojects
to the bar level. From here, the closest target it examines in its c-command domain is
the argument in the specifier position, that is, the one just raised by the primary probe.
Because the features sought by the secondary probe are a proper superset of those sought
by the first probe, this raised argument will always automatically also satisfy the second
probe, and nothing else will be moved. In other words, as long as there is an argument
with [i |v] that the main probe can raise, the secondary probe has no effect whatsoever.
(157) The secondary probe on C is satisfied by what the primary probe moved (UV):
CP
DP [ C′ ]
[PIV, i|v] I+S: [φ]M
[ C ] TP[I+S: [i|v]M]
I+S: [φ]M
DP
T vP
tS v VP
DP[PIV, i|v]
Consider now what happens instead if the primary probe can’t find a viable goal, that is,
if earlier in the derivation v didn’t agree with anything (and therefore, nothing carries the
[i |v] flag). This will be the case of AV clauses (158). The primary probe will find no suitable
goal, so it will just not agree with normove anything (represented by the crossmark). Now,
the secondary probe has its time to shine. Because of its very general featural specification,
it will simply move the highest nominal it can find: the subject. (Again, I’m representing
the subject here as carrying [PIV], but this is not crucial, hence the parentheses. What
matters is that the object does not carry [PIV].)
Chapter 3. Äiwoo clause structure 115
(158) The secondary probe on C only moves something if the primary one couldn’t
find anything: AV derivation
CP
DP
([PIV]) [[ C ]] TPI+S: [i|v]M 7
I+S: [φ]M
tS T vP
tS v VP
DP
To summarize, adding the secondary probe, and considering the exact mechanics of how
the two probes interact with each other, has the correct consequences. In a UV clause,
only the main probe will kick in, and raise the v-flagged argument (the object), despite it
being non-local to C. In a AV clause, where no argument carries [i |v], the “plan B” probe
kicks in instead, and the highest argument (the subject) is raised to spec,CP.
3.5.3.2 THE PLAN A/PLAN B LOGIC, MULTITASKING, AND THEORETICAL
PARSIMONY
The mechanism outlined here, originally developed in Scott (2021), derives this “Plan A/
Plan B” kind of system, where C in a sense has a preference list: if it can, it will target a
goal of type X, but if it cannot, a goal of type Y will suffice (with Y being a proper superset
of X). A skeptical reader might object that the complex system of multiple interacting
probes described above amounts to not more than a restatement of the facts, and that a
simpler solution would thus be preferable. However, I want to argue that Scott’s (2021)
system is superior to other alternatives on the grounds of theoretical parsimony.
Such an alternative proposed in earlier literature is van Urk & Richards’s (2015) notion
of Multitasking, illustrated in (159)31. Given a probe searching for features [α, β], and two
equidistant goals XP and YP (in multiple specifiers of the same projection FP), the probe
will agree with the goal that will check off most features, in this case YP. If YP had not
been there, then the probe would have agreed with XP, as this goal would at least have
checked off one of the features sought by the probe.
31 A conceptually similar proposal, formulated in terms of “Best Match”, was made in Coon & Bale (2014). The
technical details, however, differ from Multitasking.
116 3.5. Deriving canonical clauses
(159)
probe FP
[uα, β]
XP FP
[α]
YP F′
[α, β]
F …
This notion of “Multitasking” derives the same result as the two-probe system I adopted
from Scott (2021) for Äiwoo, and it does so with a single probe. However, this requires
the addition of several mechanisms to our theory of Agree(ment): (i) a notion of equidis-
tance; (ii) the ability for a probe to evaluate and compare several goals, and choose the
best matching one. This does result in a “simpler tree”, with fewer probes and less repro-
jection and search steps, but at the cost of a more complex theory. Scott’s (2021) proposal,
on the other hand, achieves the same descriptive power but only with pre-existing in-
gredients, that our theory of Agree(ment) independently already needs (probe ordering,
Bossi & Diercks 2019; Cyclic Agree, Béjar & Rezac 2009); therefore, the additional mech-
anism required by Multitasking are not needed. In this case, we have a “more complex
tree”, but the theoretical apparatus is kept simpler.
3.5.4 PUTTING IT ALL TOGETHER: STEP-BY-STEP DERIVATIONS FOR AV AND
UV CLAUSES
This is a good point to pause the introduction of new moving parts, and summarize what
we have so far. In this section, I just give annotated step-by-step derivations for entire AV
and UV clauses, combining everything that I presented above. In the trees below, I also
include an AspP projection between vP and TP, to which the verb head moves, to repre-
sent the aspectual prefixes on the verb. In the discussion above throughout the chapter
I excluded this to avoid unnecessary clutter in the trees, as it was not relevant; I include
it here for completeness. The only part I do not cover here is the presence of prefixal
φ-agreement on the verb in AV but not in UV, which is discussed in §3.5.5.
3.5.4.1 UV DERIVATION
EXAMPLE SENTENCE:
(160) nuwopa enge ki-ve Anna=kaa tS tO
house this IPFV-buy.UV Anna=FUT
‘Anna will buy this house’
Chapter 3. Äiwoo clause structure 117
WITHIN THE vP DOMAIN:
(161) vP
DP
Anna
v
VP
√V [ v ⟨φ]O⟩
BUY I: [Ā]
S: [D]
√V DP[PIV, i|v]
BUY
this house
a. v agrees with the Ā-marked object, copies back its φ-features, and is satisfied;
the object DP is flagged with [i |v]; no arguments move at this point.
b. The verb head-moves into v.
TP DOMAIN:
(162) TP
DP
Anna
[ T ] AspP
I+S: [φ]M
Asp vP
Asp v DP
IPFV v VP
√V v ⟨φO⟩
BUY tV DP
[PIV, i|v]
a. v head-moves into Asp (but not further up to T);
b. T raises the closest nominal it can find, i.e., the subject.
118 3.5. Deriving canonical clauses
CP DOMAIN:
(163) CP
DP [ C′ ]
I+S: [φ]M[PIV, i|v]
C TP
[ ] DPC Asp[ T AspPI+S: [i|v]M] Anna
I+S: [φ]M
Asp Asp vP
IPFV √V v ⟨φO⟩
BUY
tS tv VP
tV DP
[PIV, i|v]
a. Long head movement: Asp head-moves into C, skipping T.
b. The primary probe on C initiates a search, and finds the closest goal with [i |v],
i.e., the DP[PIV] object; the object is moved to spec,CP.
c. The secondary probe on C reprojects to the bar level, and initiates a search;
the closest viable argument is the just-moved DP in spec,CP, so the secondary
probe is satisfied as well.
SPELL-OUT OF THE VERB STEM:
(164) Stem allomorphy: UV
☞ √BUY ⇔ ve / v ⟨φ⟩
√BUY ⇔ veve (elsewhere)
Chapter 3. Äiwoo clause structure 119
3.5.4.2 AV DERIVATION
For concreteness, in the trees I represent the subject as carrying [PIV]. Nothing in this
derivation would change if the subject was a plain DP with no Ā-features, as long as the
object also does not carry [PIV].
EXAMPLE SENTENCE
(165) mikilitei ki-li-veve tS=kaa tS nuwopa
fishermen IPFV-3AUG-buy.AV =FUT house
‘The fishermen will buy a house’
WITHIN THE vP DOMAIN:
(166) vP
DP[PIV]
v
fishermen
[ ] VP√V v
BUY I: [Ā]
S: [D]
√V DP
BUY house
a. The probe on v does not interact with the object (because it doesn’t have [PIV]
or any other Ā-features), but it’s satisfied (because it has [D]). No features are
copied, and probing halts.
b. The verb head moves into v.
TP DOMAIN:
(167) TP
DP
[PIV] [ T ] AspP
I+S: [φ]M
Asp vP
Asp v DP
IPFV [PIV] v VP
√V v
BUY tV DP
120 3.5. Deriving canonical clauses
a. v head-moves into Asp;
b. T agrees with the closest nominal it can find (i.e., the subject), and raises it.
CP DOMAIN:
(168) CP
DP
[PIV]
C
TP
[ C ⟨φS⟩] Asp[I+S: [i|v]M] 7
I+S: [φ]M DP
Asp [PIV] T AspP
IPFV √V v
BUY Asp vP
tS tv VP
tV DP
house
a. Asp long-head-moves into C, skipping T;
b. The primary probe on C cannot find any goal carrying [i |v], so it doesn’t agree
with or move anything;
c. The secondary probe on C kicks in, and it agrees with and moves the closest
nominal, i.e., the subject; the subject’s features are copied onto C.
SPELL-OUT OF THE VERB STEM:
(169) Stem allomorphy: AV
√BUY ⇔ ve / v ⟨φ⟩
☞ √BUY ⇔ veve (elsewhere)
3.5.5 PREFIXAL φ-AGREEMENT IS IN C, AND GETS IMPOVERISHED IN UV
Let us briefly take stock on howwe are doing in addressing ourWord Order Puzzle. At this
point, our model accounts for (i) why (canonical) AV and UV clauses show their particular
Chapter 3. Äiwoo clause structure 121
word order patterns, and (ii) how this is correlated with the choice of a particular verbal
stem in either case. These are two of the three reliable morphosyntactic correlates of voice
presented in §3.3. The third and last one on our agenda is the presence of prefixal φ-
agreement in AV but not in UV, to which I turn now.
The phenomenon is concretely illustrated below. In AV (170), the verb must carry a
subject agreement prefix. The 3AUG prefix li~lu- is special in being positioned between the
aspect prefix and the stem (170a); every other φ-prefix goes before aspect (170b)32. In UV,
on the other hand, an agreeing prefix is completely impossible (171)33. To be sure, we can
also confirm that a UV verb carries no prefixal agreement with the argument in spec,CP,
even though it is the object: even when the object is plural (172), a prefix is still impossible
(remember that the alternation between ki-li- and ku-lu- is purely phonological; §2.4).
(170) AV: obligatory prefixal φ-agreement with the subject (250728)
a. mikilitei ku- *(lu-) tou-mä=kaa sii
fishermen IPFV-3AUG-bring.AV-DIR1=FUT fish
‘The fishermen will bring fish (here)’
b. ( iude) *(de-) ki-vängä=kaa sii
12AUG 12AUG-IPFV-eat.AV=FUT fish
‘We will eat fish’
(171) UV: no prefixal φ-agreement with the subject (250807)
a. sii ki- (*li-) ngä pedevalili=kaa
fish IPFV-(*3AUG-)eat.UV children=FUT
‘The children will eat the fish’
b. sii (*de-) ki-ngä-de=ngaa
fish (*12AUG-)IPFV-eat.UV-12AUG=FUT
‘We will eat the fish’
(172) UV: no prefixal φ-agreement with the object either (250807)
pedevalili ku- (*lu-) tu-mä-no=ngaa
children IPFV-(*3AUG-)bring.UV-DIR1-1MIN=FUT
‘I will bring the children back (here)’
Let us now consider what kind of analysis our model commits us to. We empirically know
that C always raises some argument, either the subject or the object. By assumption, if an
XP is moved anywhere, some probe has agreed with that XP, and also by assumption,
32 3MIN has no overt prefix. It’s therefore impossible to know whether a putative 3MIN prefix would be in the
same position as the 3AUG prefix, or the same as all the other ones. One could thus talk about this position
difference as drawing the cut between 3AUG vs. everything else, or alternatively, as between 3rd person
vs. 1st/2nd person. The difference is immaterial here.
33 As was pointed out earlier, in earlier literature UV was described as involving an agreement suffix, e.g. -de
12AUG in (171b). However, in chapter 5 I show that this is not an agreement suffix, but the actual subject
pronoun in spec,TP, spelled out in a phonologically reduced form. Therefore, the only difference in terms
of agreement between AV and UV is that the former involves an agreement prefix, and the latter involves
no agreement.
122 3.5. Deriving canonical clauses
having agreed with an XP means all of XP’s features have been copied onto the probe
(chapter 1, §1.5). We are thus forced to posit that the fronted argument’s φ-features are
always copied onto C (the probe driving the fronting). Concretely: in AV, C has copied the
features of the subject; in UV, those of the object. The question is now why these features
are only overtly spelled out in AV, and not in UV.
I propose the following analysis, where the difference in presence vs. absence of agree-
ment is morphological, not syntactic. In AV, the φ-features of the subject copied on C
are spelled out faithfully; this is the agreement prefix. As we just said, this agreement
step takes place in UV as well (between C and the object). However, in this case, the
φ-features copied on C undergo Impoverishment (Halle & Marantz 1993). Because this
Impoverishment rule applies only in UV, this can be cast a pattern that has been referred
to as Kinyalolo’s Constraint (Kinyalolo 1991, Carstens 2003, 2005, Baker 2010, Tyler &
Kastner 2022): two heads have agreed with the same argument, and one of them does not
get spelled out. In our concrete UV case, both v and C have agreed with the object/have
φ-features on them, so C does not get spelled out (173)34,35.
(173) Impoverishment rule for C:
C⟨φ⟩ ⇔ ∅ / v ⟨φ⟩
Concretely, I propose that the complex verb head is mapped onto its surface morpholog-
ical form as illustrated in the trees below (where squiggly lines indicate exponent inser-
tion). I include here again the AspP projection between TP and vP, as in the step-by-step
derivations above.
In an AV context (174), v has not copied any φ-features onto itself, so the AV stem
allomorph of the verb is inserted as the elsewhere case (174a). Again because v does not
carry any φ-features, C is not impoverished (174b). When the subject is any other φ-
category than 3AUG (e.g., 12AUG here), the φ-features copied onto C are spelled out to the
left of aspect, faithfully to the underlying structure (174c).
34 I assume that the domain within which morphological rules can be triggered is the complex head, even
though the two relevant (sub-)heads are not structurally or linearly adjacent (Embick 2010, Bobaljik 2000,
2012, Svenonius 2012, Moskal 2015, Merchant 2015, Choi & Harley 2019, a.m.o.). The reason for this will be
apparent considering the trees in (174)–(176).
35 It might be the case that the entire C head gets deleted morphologically, as written in (173), or that only the
φ-features on C get deleted (formally: C ⟨φ⟩ ⇔ C ⟨∅⟩ / v ⟨φ⟩), and C itself is just always phonologically null.
As far as I can tell, these two options are empirically indistinguishable.
Chapter 3. Äiwoo clause structure 123
(174) AV form, 12AUG subject: de-ki-vängä=kaa  
12AUG-IPFV-eat.AV=FUT
CP
DP
[12AUG]
C TP
t
C S⟨12AUG⟩ T AspP
Asp [FUT]
[IPFV] √V v
EAT tAsp vP
…
de- ki- vängä =kaa
a. Stem (voice allomorphy):
√EAT ⇔ ngä / v ⟨φ⟩
☞ √EAT ⇔ vängä (elsewhere)
b. C impoverishment: doesn’t apply
7 C⟨φ⟩ ⇔ ∅ / v ⟨φ⟩
c. VI rule for C-agreement:
C⟨12AUG⟩ ⇔ de-
The 3AUG prefix li- is unique in its placement to the right of aspect instead of to its left.
I assume this is a shallow, purely morphological fact. The underlying syntax is identi-
cal to the 12AUG subject case above, but the 3AUG prefix idiosyncratically undergoes a
step of local dislocation, swapping place with the aspect prefix, as shown below. Several
implementations are possible for this step of morphological displacement; here I merely
describe the empirical facts, without taking too much of a stance (175c)36.
36 A similar displacement rule is needed for the cases where the 1MIN prefix i- metathesizes into the irrealis
and apprehensive prefixes nä- and de-, palatalizing the first consonant and resulting in nyä- and je-; see
chapter 2, §2.4.
124 3.5. Deriving canonical clauses
(175) AV form, 3AUG subject: ki-li-vängä=kaa
IPFV-3AUG-eat.AV=FUT
CP
DP
[3AUG]
C TP
tS
C ⟨3AUG⟩ T AspP
Asp [FUT]
[IPFV] √V v
EAT tAsp vP
…
ki- li- vängä =kaa
a. Stem (voice allomorphy):
√EAT ⇔ ngä / v ⟨φ⟩
☞ √EAT ⇔ vängä (elsewhere)
b. C impoverishment: doesn’t apply
7 C⟨φ⟩ ⇔ ∅ / v ⟨φ⟩
c. Morphological displacement and VI rule for C-agreement:
C⟨3AUG⟩~Asp ⇒ Asp~C⟨3AUG⟩
C⟨3AUG⟩ ⇔ li-
As for the UV forms, the facts are similar, but with two crucial morphological differences.
Both have the same conditioning trigger, that is, the presence of φ-features on v. First, the
stem insertion rule is sensitive to this, and will thus insert the UV stem exponent rather
than the default AV one (176a). Second, (the φ-features on) C will be impoverished in this
same context (176b).
Chapter 3. Äiwoo clause structure 125
(176) UV form (any subject/object): ki-ngä [DP]=kaa
IPFV-eat.UV =FUT 
CP
DP
[PIV, i|v]
C TP
C ⟨φO⟩ DP
Asp
∅ [IPFV] √V v ⟨φO⟩ T AspP
EAT [FUT]
tAsp vP
ki- ngä =kaa … tO
a. Stem (voice allomorphy):
☞ √EAT ⇔ ngä / v ⟨φ⟩
√EAT ⇔ vängä (elsewhere)
b. C impoverishment: applies
☞ C⟨φ⟩ ⇔ ∅ / v ⟨φ⟩
There could potentially be a few imaginable alternatives to the “C-agreement plus impov-
erishment” approach drawn here, but they all have some important shortcomings. First,
we can’t blame the asymmetry in the presence of agreement between AV and UV to the
size/category of themoving nominal, since in both cases, what fronts is a DP (with or with-
out [PIV]). We also can’t hypothesize that the probe that fronts the subject in AV copies
φ-features, whereas the one that fronts the object in UV does not, since by assumption, all
movement involves agreement, and feature copying is coarse (see Deal & Royer 2025 for
this notion in an interaction/satisfaction architecture): any agreement involves the copy-
ing of all features from the agreed-with constituent onto the probe. Finally, we could try
to capitalize on the facts that under my analysis, C has two probes: an [i |v] probe and a
[φ] probe. In UV, the object is fronted by the [i |v] probe; in AV, the subject is ultimately
fronted by the [φ] probe (after the [i |v] was unable to find a viable target to move). We
could imagine then positing that only the [φ] probe has an overt morphological spell-out,
but not the [i |v] probe. The problem with this approach is that, in UV, the [φ] probe also
126 3.6. A note about intransitives
agrees with the fronted object (see the derivations above). This would wrongly predict
overt agreement – unless we posit another rule of Kinyalolo-type impoverishment when
both probes have agreed with the same argument, which would then be globally just as
complex as the system presented here.
3.6 A NOTE ABOUT INTRANSITIVES
Before concluding the more empirical part of this chapter, a minor amendment/addition
must be made to the model to correctly account for the behavior of intransitive clauses. In-
transitive verbs effectively look just like AV verbs but without an internal argument (177).
Specifically, they carry obligatory agreement prefixes (li- 3AUG), and adverbial modifiers
take no voice concord (in this case, showing the default/AV form ute ‘again’ rather than
the UV form usi). I have not been able to devise any diagnostics for unaccusativity, but
this description applies to verbs that one could assume would fall in both classes.
(177) Intransitives must have prefixal agreement, and cannot have voice concord:
a. (Putative) unergative: (241219)
kuli eângâ i-li-ngoduwâ-{ute/*usi}
dog that ASP-3AUG-shout-again.{AV/*UV}
‘Those dogs barked again’
b. (Putative) unaccusative:
kuli eângâ i-li-obu-{ute/*usi}
dog that ASP-3AUG-fall-again.{AV/*UV}
‘Those dogs fell again’
This morphological similarity between intransitive verbs and AV ones is very common
within Austronesian. In fact, in languages where voice morphology is segmentable, like
Tagalog, intransitives – unergatives and unaccusatives alike – often carry visible AV mor-
phology, and are incompatible with UV morphology (178).
(178) Tagalog intransitives: 3 AV, 7 UV (adapted from V. Chen 2025: 637)
a. Unergatives:
{3k⟨um⟩anta / *k⟨in⟩anta-∅ } si Aya
⟨AV.PFV⟩sing ⟨PFV⟩sing-UV PN.NOM Aya
‘Aya sang’
b. Unaccusatives:
{3d⟨um⟩ating / *d⟨in⟩ating-∅ } si Aya
⟨AV.PFV⟩arrive ⟨PFV⟩arrive-UV PN.NOM Aya
‘Aya arrived’
37 V. Chen (2025) only reported judgments for AV on unergatives, and AV/UV on unaccusatives. For the un-
grammaticality judgment about UV on an unergative (*kinanta), I thank Dave Kenneth Tayao Cayado (p.c.).
Chapter 3. Äiwoo clause structure 127
This fits well with proposals of the type being made here for Äiwoo, where AV is not an
equally-marked option as UV but a morphosyntactic default (T. Chen 2010, Levin 2015,
Wurmbrand 2016, Wurmbrand & Shimamura 2017, H. Y. Chang 2017, Wu et al. 2023).
Under this view, intransitives, since they only have one argument, “don’t play the voice
game”, and they therefore surface with default morphology – that is, AV morphology.
As things stand, however, the model I have described throughout this chapter makes
incorrect predictions for intransitives. Specifically, it wrongly predicts that they should
look like UV verbs, rather than AV verbs. Let us see why, considering the probe on v that
I have proposed in §3.5.1.2.
We have assumed that the distribution of [PIV] is syntactically completely free. There-
fore, in an intransitive clause, the sole argument will not always necessarily carry [PIV],
but it definitely can, so we need to account for that. For unaccusatives, in a context where
the internal argument carries [PIV], the probe on v would agree with the object (179a).
This would result in UV-style morphology (at least on the adverbial modifiers, even if the
main root √FALL itself doesn’t have a distinct UV allomorph), and in the disappearance of
prefixal φ-agreement in C. For unergatives (179b), the same problemwould arise, although
here the relevant agreement step would take place from the bar level instead38.
(179) a. Unaccusatives: predicted UV 7 b. Unergatives: predicted UV 7
vP ⟨φ⟩ ⇔ UV 7! vP ⟨φ⟩ ⇔ UV 7!
[ v ⟨φ]⟩ VP DP [ v′ ⟨φ]⟩I: [Ā] [PIV] I: [Ā]
S: [D] S: [D]
√V DP
FALL [PIV]
[ v ] √ V
I: [Ā] SHOUT
S: [D]
The simplest solution is to assume that intransitive roots select for a different v head
(Harley 1995, Marantz 1997, a.m.o.). This special vINTR head has no probe on it at all,
and therefore will never agree with any argument. The vINTR head may then select for
not merging any external argument (180a) or for merging an external argument (180b),
deriving the difference between unaccusatives and unergatives39.
38 This issue is already highlighted for Tagalog in Rackowski (2002), and forms part of her argument for why
voice reflects “case agreement” rather than “theta-role agreement” (contra Schachter &Otanes 1972, Kroeger
1993, a.o.). I thank Norvin Richards for bringing this to my attention.
39 Alternatively, one could imagine that Äiwoo is a language without real unergatives, as has been argued to be
the case for Ch’ol (Coon 2019). In this case, the inventory of v heads would contain not three different items
but two: the ordinary, subject-introducing v, which carries the Ā-probe, and the intransitive, unaccusative
v, which does not introduce an argument and does not carry a probe.
128 3.6. A note about intransitives
(180) a. Unaccusatives: b. Unergatives:
vPINTR[−AG] vPINTR[+AG]
vINTR[−AG] VP DP
(no probe) [PIV] vINTR[+AG] √ V
√V DP (no probe) SHOUT
FALL [PIV]
One possible argument for the presence of a lexically distinct, intransitive v head is the
behavior of morphological causatives. In Äiwoo, causatives are built with the prefix wâ-.
The sequencewâ-√STEM constitutes the AV version of the resulting causativized predicate
(181b). The UV version is consistently built by adding a suffix -eâ~wâ (181c)40. We can
confirm that these formations are respectively the AV and UV version of a causativized
predicate by observing the mutually exclusive presence/absence of prefixal φ-agreement
and voice concord. See also Næss & Boerger (2008), Næss (2015b), and Roversi (2019) for
a similar description. For a causativized predicate, whether in AV or UV, the added causer
is treated as the external argument/subject, and the original causee argument is treated
as the internal argument/object – regardless of whether the intransitive predicate was
unaccusative or unergative.
(181) a. Baseline intransitive:
sii ku-nubo=kaa (250807)
fish IPFV-die=FUT
‘The fish will die’
b. Causativized verb, AV form:
pedevalili ku-lu-wâ-nubo-ngege=kaa sii
children IPFV-3AUG-CAUS-die-immediately=FUT fish
‘The children will kill fish straight away’
c. Causativized verb, UV form:
sii enge ku-wâ-nubo-wâ-ngege-nyi pedevalili=kaa
fish this IPFV-CAUS-die-UV-immediately-UV children=FUT
‘The children will kill this/these fish straight away’
The relevance of morphological causatives is that the prefix wâ- specifically selects for
intransitive predicates. On the one hand, it is perfectly compatible with both (putative)
unaccusatives and unergatives (182a,b). However, importantly, this prefix completely re-
jects transitive predicates (182c). (In these examples, i use the UV version of the resulting
causativized predicates, but the same is true for the AV version).
40 These are just phonological variants of each other. A few verb stems optionally select for an allomorph -nâ,
seemingly just lexically/idiosyncratically.
Chapter 3. Äiwoo clause structure 129
(182) Causative wâ- is only compatible with intransitives, not with transitives:
a. CAUS + unaccusative: 3 (250807)
Mak ku-wâ-eobu-wâ-no=ngaa
Mark IPFV-CAUS-fall-UV-1MIN=FUT
‘I will make Mark fall’
b. CAUS + unergative: 3 (241015)
Mak ku-wâ-basiki-eâ-no=ngaa
Mark IPFV-CAUS-run-UV-1MIN=FUT
‘I will make Mark run’
c. CAUS + transitive: 7 (240411)
* Mak ku-wâ-{vängä/ngä}-eâ-no=ngaa sii
Mark IPFV-CAUS-{eat.AV/UV}-UV-1MIN=CV fish
Intended: ‘I will make Mark eat fish’
To account for this restriction, one could imagine the following approach. The causative
prefix wâ- is the spell-out of a particular v head whose external argument is semantically
interpreted as a causer. This vCAUS head lexically selects for a vPINTR complement, whether
it be unergative or unaccusative (183a,b). However, it cannot be merged on top of a regular
(transitive) vP complement (184), because this would violate its selectional requirement41.
41 A potential alternative that does not rest on having a special vINTR in the lexicon is possible if we follow
Neu (2023, to appear) in assuming that “causatives of unergatives” are, in a sense, an illusion. Under this
analysis, vCAUS consistently selects for a VP complement with an internal argument, even for those verb
roots which typically have an external argument (that is, unergatives). In other words, the only argument
of typically unergative roots would be exceptionally merged in the internal argument position in the context
of these causativized constructions. Neu (2023, to appear) argues that in Hindi, Turkish, and Sason Arabic,
the interpretation of the causee in these causatives of unergatives is indeedmore similar to that of an internal
argument (non-volitional, less agentive, etc.). If we pursue this approach, then we would account for the
intransitive-only selectional requirement of the Äiwoo causative without needing a special vINTR head, thus
rendering this argument moot.
130 3.6. A note about intransitives
(183) a. Causative of unaccusative (UV): wâ-obu-wâ
CAUS-fall-UV
vPCAUS
DP
(causer)
vCAUS ⟨φO⟩ vPINTR
[ vCA] tv VPUS ⟨φO⟩ vINTR INTR
I: [Ā]
S: [D]
√V v
t DP
INTR V [PIV]
FALL (causee)
i. vCAUS ⇔ wâ-
ii. √FALL ⇔ obu
iii. ☞ vINTR ⇔ -eâ~wâ / vCAUS ⟨φ⟩
vINTR ⇔∅ (elsewhere)
b. Causative of unergative (UV):
vPCAUS
DP
(causer)
vCAUS ⟨φO⟩ vPINTR
[ vCA]US ⟨φO⟩ vINTR DP[PIV]
I: [Ā] (causee) tvINTR tV
S: [D]
√V vINTR
RUN
Chapter 3. Äiwoo clause structure 131
(184) Causative of transitive: violation of c-selection
7 vPCAUS
DP
(causer)
vCAUS 7! vP
[•vPINTR•]
DP
v VP
√V DP
EAT
3.7 THE A/Ā-PROPERTIES OF FRONTING TO SPEC,CP
We have now developed an answer to (a sub-part of) the Word Order Puzzle, deriving the
word order and voice morphology properties of canonical AV and UV clauses. I will now
devote some time to examining the properties of the movement step of the VIA argument
to spec,CP, exploring where this fits along the canonical A/Ā-distinction. We will see
that empirically, Äiwoo is again strikingly reminiscent of Dinka as analyzed in van Urk
(2015). VIA fronting to spec,CP will be shown to have almost all properties of typical A-
movement, with the sole exception of its more relaxed locality profile (since it can skip
intervening nominals). For this, I will use themoniker “non-local A-movement” as a purely
descriptive label (see also the discussion about “leapfrogging” in chapter 1, §1.3.2). I will
argue that this kind of movement should not be thought of as an innovation, but that its
existence is in fact already predicted by the conjunction of pre-existing, independently
needed ingredients in our theory, specifically (i) featural Relativized Minimality (Rizzi
1990), and (ii) featurally relativized probes (Béjar 2003 et seq.).
First, in §3.7.1 I cover the empirical facts, showing that CP-fronting displays every
property of classic A-movement with the sole exception of strict locality. Then, in §3.7.2 I
discuss how our theory already predicts this to be possible, without the need for additional
mechanisms or amendments. Finally, §3.7.3 discusses some potential cross-linguistic par-
allels.
3.7.1 VIA FRONTING TO SPEC,CP IS NON-LOCAL A-MOVEMENT
A list of the differences between (canonical) A- and Ā-movement is given in (185)–(186)
(Postal 1971, Chomsky 1977, 1981, a.m.o.; this specific list is adapted from van Urk 2015: 23
and Richards 2014: 167–169). I have added a checkmark next to the properties that Äiwoo
132 3.7. The A/Ā-properties of fronting to spec,CP
CP-fronting showcases: (186a), and (185b–f). Φ-agreement (185b) ismore complex andwas
treated separately in the previous section, so I don’t rehash the discussion here. Finally, I
will not talk about parasitic gaps (185g)–(186g), as we don’t know how this phenomenon
works in Äiwoo or whether it exists in the first place.
(185) A-properties: (186) Ā-properties:
a. Local a. 3 Not strictly local
b. (3) Feeds φ-agreement b. Does not feed φ-agreements
c. 3 Restricted to nominals c. Not restricted to nominals
d. 3 No Condition C reconstruction d. Recontruction for Condition C
e. 3 No Weak Crossover e. Weak Crossover
f. 3 New binding antecedents f. No new binding antecedents
g. No parasitic gaps licensing g. Parasitic gap licensing
Descriptively then, VIA fronting is “non-local A-movement”: it has all the categorial and
binding-theoretical properties of A-movement, but it can skip intervening nominals.
LOCALITY As we established in §3.5.3, VIA fronting can clearly be non-local: in UV, C
agrees with the object and raises it (187). Crucially, the subject is skipped, despite being
closer to C.
(187) Fronting of the object to spec,CP in UV skips the intervening subject:
[CP O[i |v] C [ I+S: [i |v]M] [TP S =TAM [vP tS tV+v tV tO ]]]
3!
This contrasts with canonical A-movement, which is typically strictly local, and cannot
cross any intervening nominal:
(188) A-movement cannot skip intermediate nominals:
[the book] was given Anna t
7
CATEGORY RESTRICTION: ONLY NOMINALS In Äiwoo, the preverbal VIA position is
reserved for nominal constituents (189a), and constituents of other categories cannot be
in this position (189b,b). Äiwoo’s V2 effect is therefore different from that of Germanic V2
languages, and it is once again more similar to Dinka (van Urk 2015).
(189) Äiwoo pre-verbal position: 3 DP, 7 PPs, 7 AdvPs (240418)
a. [George]DP ki-te-vesi-i-kâ-no ngä taun dâbu dâuwângâ
George IPFV-see.UV-keep-UV-DIR3-1MIN in town day every
‘I see George in town every day’
b. * [ngä taun]PP ki-te-vesi-i-kâ-no George dâbu dâuwângâ
in town IPFV-see.UV-keep-UV-DIR3-1MIN George day every
Chapter 3. Äiwoo clause structure 133
c. * [dâbu dâuwângâ]AdvP ki-te-vesi-i-kâ-no George ngä taun
day every IPFV-see.UV-keep-UV-DIR3-1MIN George in town
In Germanic V2 languages, instead, fronting to spec,CP is pure Ā-movement (den Besten
1981, 1983, et seq.), and thereforewe don’t see this restriction (190); any kind of constituent
can be fronted. I use Norwegian here as an example.
(190) Norwegian pre-verbal position: 3 DPs, 3 PPs, 3 AdvPs
a. [Ana]DP ser jeg i by-en hver dag
Ana see I in city-DEF every day
‘I see Ana in town every day’
b. [i by-en]PP ser jeg Ana hver dag
in city-DEF see I Ana every day
c. [hver dag]AdvP ser jeg Ana i by-en
every day see I Ana in city-DEF
CONDITIONC RECONSTRUCTION For this test to be relevant at all, we need to first make
sure that Condition C is active in Äiwoo (191). Both these sentences are in AV, with the
subject in spec,CP. In (191a), the subject pronoun c-commands ‘Mark’, and the two cannot
co-refer. In (191b), however, the pronoun embedded within the subject (‘his daughter’42)
does not c-command ‘Mark’, and co-reference is possible. This is a canonical Condition C
effect.
(191) Condition C is active in Äiwoo (AV): (230615)
a. [ inâ/∅3MIN ]S ku-woi=kaa tS [nubole na Mak ]O ngä paveli
3MIN IPFV-plant.AV=FUT taro POSS Mark in garden
‘S/hej/*i will plant Marki’s taro in the garden’
b. [sipe -∅n ]S ku-woi=kaa tS [nubole na Mak ]O ngä paveli
daughter-3MIN IPFV-plant.AV=FUT taro POSS Mark in garden
‘Hisi daughter will plant Marki’s taro in the garden’
Now, we can observe that when an object fronts to spec,CP across the subject, it does not
reconstruct for Condition C (192). The R-expression Pita ‘Peter’ in the fronted object can
readily co-refer with the subject pronoun (cf. the literal translation, which is a Condition
C violation in English), although the subject c-commands the base-generated position of
the object (both from the subject’s original position and its derived one in spec,TP)43. This
datapoint also shows that CP-fronting in Äiwoo does not trigger Strong Crossover, which
42 See Roversi (2025b) for arguments that there is a null possessor pronoun here, and for the general syntax
of possessed DPs.
43 In addition to not showing Condition C reconstruction effects, VIA fronting of non-agents in Äiwoo shows
a puzzling yet very robust anti-cataphora effect, reminiscent of (though not identical to) the ones discussed
for Mayan languages in Royer (2025). The pattern can be described as follows. For any nominal fronted to
spec,CP in a non-AV clause, a pronoun contained within that nominal cannot co-refer with anything to its
right, regardless of the original c-command relations. An illustrative example (in UV) is given below.
134 3.7. The A/Ā-properties of fronting to spec,CP
indeed has been argued to be related to Condition C (see e.g. Keine & Bhatt to appear for
a recent argument).
(192) Äiwoo VIA-fronting: no Condition C reconstruction (230601)
[poi no Pita ] nO ki-dââ -∅ =naa tS tO
pig POSS Peter IPFV-tie.UV-3MIN=FUT
‘Peteri will tie hisi pig’; lit. ‘Hei will tie Peteri’s pig’
This behavior clearly contrasts with V2 fronting in Norwegian, which is Ā-movement and
therefore shows Condition C reconstruction effects.
(193) Norwegian V2-fronting: reconstruction for Condition C
[hus-et til Petter ]O kjøpte han tS tO for to år siden
house-DEF to Peter bought he for two years ago
‘Hej/*i bought Petteri’s house two years ago’
There is a potential confound with this Condition C data, however, that might make it
uninformative. In Roversi (2025b) I show that possessed DPs in Äiwoo always contain a
relative clause, where the possessee is the extracted argument (object), POSS is a UV verb
and ‘Peter’ is the embedded subject. The constituent poi no Pita in (192), more literally,
would roughly correspond to “[the pigi [RC that Peter has i]]”. The sub-constituent
containing ‘Peter’, therefore, is an adjunct (a relative clause contained within the larger
nominal). Therefore, the reason why we see no Condition C reconstruction effects might
be that the whole adjunct is be late-merged at the landing site (Lebeaux 1988, 1991, Fox
2002, S. Takahashi 2006, Takahashi & Hulsey 2009). If it is so, the subject pronoun never
effectively c-commands ‘Peter’, thus rendering our argument moot. Unfortunately, I have
not been able to find a way to avoid this confound, as I don’t know of any way to embed
an R-expression inside a nominal in Äiwoo without making it part of an adjunct44.
WEAK CROSSOVER In Äiwoo, VIA fronting to spec,CP does not trigger a WCO viola-
tion. In (194), the object iie ‘who’ is fronted to spec,CP, and it can still be bound by the
pronominal possessor embedded within the subject. Note that this is not wh-movement,
but just regular VIA fronting of the object in UV; in other words, this is awh-in situ clause.
(i) [poi no -∅n ]O i-dââ Pita tS tO (230601)
pig POSS-3MIN ASP-tie.UV Peter
‘Peteri tied his/herj/*i pig’ (only disjoint reading)
Comment: “No, that’s not Peter’s pig, that’s poi no somebody else”
The facts are rather complex, and not immediately relevant to the canonical Condition C effects under
discussion here. The interested reader is referred to Roversi (2024a).
44 For example, PPs cannot be attached directly inside DPs, and a relative clause must be used instead (see
chapter 2, §2.5). This just recreates the same problem of potential late merge.
Chapter 3. Äiwoo clause structure 135
(194) Äiwoo VIA fronting: no Weak Crossover violation (240215)
iie ku-tu-mä tumwä -∅n =naa? tS tO
who IPFV-bring.UV-DIR1 father-3MIN=FUT
‘Who are the individuals 𝑥 such that 𝑥 ’s father will bring 𝑥?’
Lit. ‘Whoi will his/heri father bring?’
For this phenomenon, the parallel with Germanic V2 fronting is somewhat imperfect. In
Germanic languages, V2 fronting has discourse effects similar to topicalization, and the
class of relevant expressions forWCO violations (quantified expressions,wh-phrases, etc.)
are not really topicalizable in the first place (Rizzi 1986 et seq.). However, at the very least
we can see that fronting a wh-word does indeed trigger a WCO violation, thus making
the bound reading impossible (195).
(195) Norwegian wh-fronting: Weak Crossover violation
hvem vil far-en hans ta med tO?
who will father-DEF his take with
‘Whoi will hisj/*i father bring?’
NEW ANTECEDENTS FOR BINDING Binding relations in Äiwoo are a complex topic to
investigate, because it’s often hard to control for the multiple movements that arguments
undergo in one single clause. However, it is possible to verify that a fronted VIA object can
bind into the subject. Note that I do not discuss anaphor binding, as sentences containing
anaphors have a different syntax altogether than regular non-reflexive sentences; see 5,
§5.3.2. I limit the discussion below to quantifier-variable binding.
First, we must establish some baselines about how (universal) quantification works
in Äiwoo. Universal quantification is often expressed with the morpheme du ‘all’ that at-
taches inside the verb complex, suffixed to the stem (196a) (Næss 2015a). It is also attested
as attaching directly to a noun, although this is not consistently accepted in elicitation
and it is rarer in the corpus; (196c) is part of an attested example. Note that unlike for
English ‘every’, and more similarly to ‘all’, a nominal quantified by du is grammatically
plural (as revealed by the obligatory 3AUG agreement prefix li- in (196a,b).
(196) Du ‘all’ attaches (mostly) to the verb: (250226)
a. pedevalili ki-li-bakisi- du -mä=kaa
children IPFV-3AUG-run-all-DIR1=FUT
‘All the children/Every child will run (towards here)’
b. ?? pedevalili- du ki-li-bakisi-mä=kaa
children-all IPFV-3AUG-run-DIR1=FUT
‘All the children/Every child will run (towards here)’
Native speaker’s comment: ‘[(196a)] sounds more natural’
c. nyenaa- du ngä paveli enge
tree-all in garden this
‘All the trees in this garden’ (Næss 2015a: 97)
136 3.7. The A/Ā-properties of fronting to spec,CP
From its position inside the verb, du ‘all’ can quantify either argument in a transitive
sentence, regardless of voice (197). I don’t have a concrete analysis as to how du ‘all’
comes to be affixed to the verb stem, and how it is interpreted as quantifying an argument.
Potentially, it could be a base-generated quantifier in the clausal spine, a la Fitzpatrick
(2006), which the verb picks up by head movement. Alternatively, it could perhaps have a
semantics akin to ‘completely’, and its interpretation as quantifying an argument would
be achieved only indirectly45.
(197) a. Association possibilities of du ‘all’ in AV: (250311)
pedevalili ki-li-gâwââ-du=kaa buk
children IPFV-3AUG-read.AV-all=FUT book
3 S∀: ‘All the children/every child will read a book’
3 O∀: ‘The children will read all books/every book’
b. Association possibilities of du ‘all’ in UV:
buk enge ki-gââ-du pedevalili=kaa
book this IPFV-read.UV-all children=FUT
3 O∀: ‘The children will read all these books’
3 S∀: ‘All the children/every child will read this book/these books’
Although much is unclear about the syntax and semantics of du ‘all’, we can use this
quantifier to establish that VIA fronting indeed generates new antecedent possibilities
for binding, like canonical A-movement and unlike canonical Ā-movement. In UV, the
universally quantified object fronted to spec,CP can bind a variable contained within the
subject (here, the possessor):
(198) UV: the fronted object can bind into the subject (250123)
3O can bind S
mikilibei ku-tu-usi-du-kä nurse nogo-du -i =laa
patients IPFV-bring.UV-back.UV-all-DIR3 nurse POSS-all-3AUG=FUT
tS tO ngâ nuwopaa nyibei
to house.of place.sick
‘Hisi nurse will bring every patienti back to the hospital’
Lit. ‘They-alli’s nurses will bring all patientsi back to the hospital’
Conversely, and as expected, this kind of binding is impossible in AV (199). Here, the
possessor pronoun embedded in the subject cannot be bound by the universally quantified
object, so that only a disjoint (non-bound) reading is available.
45 This is a potential confound for whether this diagnostic is actually informative in the first place. Another
possible confound is the fact that a DP quantified by du is plural rather than singular (like English ‘all’,
unlike English ‘every’). This might make it possible to generate “fake-bound” readings that are extensionally
equivalent to a true, covarying, bound reading, but achieve this via cumulativity rather than actual syntactic/
semantic binding. I thank Danny Fox for this observation, and leave this is as an open problem.
Chapter 3. Äiwoo clause structure 137
(199) AV: the object cannot bind into the subject (250123)
7O cannot bind S
nurse nogo -∅n ku-tou-ute-du-kä=naa tS mikilibei
nurse POSS-3MIN IPFV-bring.AV-back.AV-all-DIR3=FUT patients
ngâ nuwopaa nyibei
to house.of place.sick
‘Hisj/*i nurse will bring back every patienti to the hospital’
3.7.2 THE LOCALITY OF (ALL) MOVEMENT IS FEATURALLY RELATIVIZED
In the previous section we observed that movement to spec,CP in Äiwoo has almost all
the properties of A-movement – it can feed φ-agreement, it’s restricted to nominals, and
all the binding-theoretical properties – and yet, it can skip an intervening nominal. There-
fore, I described it as “non-local A-movement”. I argue that this should not be considered
an empirical or theoretical innovation, or even surprising at all. Rather, this kind of pat-
tern is in fact already predicted to be possible by the conjunction of two pre-existing and
independently motivated ingredients of our theory: (i) featural Relativized Minimality
(Rizzi 1990), and (ii) the concept of featurally relativized probes (Béjar 2003 et seq.). These
two, in addition to the assumption that all movement is triggered by a step of Agree (see
chapter 1, §1.5), predicts that precisely this kind of “non-local A-movement” should exist.
Äiwoo confirms this prediction46.
Deriving the canonically strict locality of A-movement from Relativized Minimality
was already a fundamental part of the featural theory of the A/Ā-distinction (van Urk
2015). According to this approach, “A-movement” and “Ā-movement” don’t actually ex-
ist as distinct primitives. Rather, there exists only one kind of phrasal movement, and
“A/Ā-movement” are just descriptive labels for frequent clusterings of independent prop-
erties. More specifically for our purposes, the locality profile of classic A-movement is
an epiphenomenon of the fact that the probes which drive (what we descriptively call)
“A-movement” search for φ-features. All nominals have φ-features, purely in virtue of
being nominals. Due to Relativized Minimality (Rizzi 1990), every nominal in a probe’s
c-command domain will constitute a valid goal for the probe, and therefore, every nom-
inal will be a potential intervener. The result is the familiar strictly local/“no skipping”
profile: only the nominal that is highest/closest to a probe can be targeted, or in other
words, the highest/closest nominal intervenes for the probing of every lower nominal.
This is schematized in (200).
(200) The strict locality of A-movement is due to φ-intervention:
7!
[ DP1[φ] probe [ I+S: [φ]M] [ DP1[φ] … [ DP2[φ] … [ DP3[φ] …
46 We will see in chapter 6 that Äiwoo actually showcases two instances of non-local A-movement: both the
one to spec,CP, discussed here, and also movement to spec,TP, which will be the object of that chapter.
138 3.7. The A/Ā-properties of fronting to spec,CP
Enter now the second ingredient, namely the idea that probes can be relativized as for
the specific features they search for. This has been established especially in the domain of
agreement in a very rich literature for the past two decades, starting at least with Béjar
(2003) (and Béjar & Rezac 2003, 2009, Nevins 2011, Deal 2015b, andmany others).We know
that, for example, agreement probes may search for specific φ-features, only agreeing
with, say, [PART] nominals (1st and 2nd person) but not 3rd person ones. As an illustration,
consider the following paradigm from Chirag Dargwa (201). If the subject is [PART], as in
(201a,b), the verb will agree with the subject. However, if the subject is 3rd person but the
object is [PART], then the verb will agree with the object instead (201c).
(201) Agreement for [PART] (Sumbatova 2011: 135): (Chirag Dargwa)
a. dicce {ʕu / it } r-iqqan-da
1SG.ERG 2SG.ABS 3SG.ABS F-lead-1
‘I lead you/her’
b. ʕicce du r-iqqan-de
2SG.ERG 1SG.ABS F-lead-2
‘You lead me’
c. ite du r-iqqan -da
3SG.ERG 1SG.ABS F-lead-1
‘S/he leads me’
This is captured by positing an agreement probe whose featural specification is not [φ],
but [PART]47. When the subject is 3rd person, it will not constitute a valid goal for the
probe, and will therefore be skipped. If the object then happens to be [PART], we see non-
local agreement: the verb agrees with the less-local object instead of the closer subject
(202) (I represent this schematically as head-initial for simplicity). In this case, there is no
intervention because of the featural specification of the probe.
(202) Non-local agreement for [PART]:
[ probe [ I+S: [PART]] [ Subj[3] … [ Obj[PART]
3!
The point of connection here is the idea that all movement is dependent on a step of agree-
ment. If agreement probes can do (202), then – in principle – probes driving movement
should be able to do (202) too. Putting together Relativized Minimality and the notion of
featurally relativized probes now generates a clear prediction. We should expect to find
patterns of movement that show no properties of Ā-movement, and yet can be non-local
(in the sense that a lower nominal can be picked over a higher one, which gets skipped). As
long as the probe driving the movement in question is featurally specific enough, a higher
argument might not constitute a valid intervener for a lower one. Essentially, what these
two ingredients together predict is the same pattern illustrated in (202) for φ-agreement,
but for syntactic movement instead.
47 I use the interaction/satisfaction notation here for consistency, but [uPART] would be equivalent.
Chapter 3. Äiwoo clause structure 139
Indeed, this is precisely what I propose we see for movement to spec,CP in Äiwoo.
The (primary) probe on C is featurally specific: it wants [i |v], rather than just [φ]. The
highest nominal carrying [i |v] will be agreed with and fronted to spec,CP; a higher nom-
inal without this feature (i.e., the subject in a UV clause) will not constitute a valid goal,
and not intervene (203).
(203) Non-local A-movement to spec,CP:
3!
[CP O[i |v] C [ I+S: [i |v]M] [TP S =TAM [vP tS tV+v tV tO ]]]
3!
This agreement-plus-movement step is thus not “strictly local” in the same way that clas-
sic A-movement is, but it is nonetheless local in this more specific, relativized way48. As
mentioned above, already in van Urk (2015), the strict locality of canonical A-movement
was argued to be an epiphenomenon of the featural specifications of the probes driving
such movement, i.e., a consequence of Relativized Minimality. Here, what I do is just re-
inforce this point. Because probes can have arbitrary featural specifications, we should
expect to find precisely this kind of “violation” of classic strict A-locality.
Of course, movements with the kind of properties we see for CP-fronting in Äiwoo are
nothing new. Perhaps especially since van Urk’s (2015) influential work, a rich literature
has analyzed cases of movement with similar profiles. This is the case for, among others,
movement to spec,CP in Dinka (van Urk 2015), non-canonical passives in Khanty (Col-
ley & Privoznov 2020), and bei-passives in Mandarin (F. Chen 2023); see Lohninger (2025)
for an overview (see also Webelhuth 1989, Mahajan 1990, Saito 1992, Miyagawa 1997,
McGinnis 1998, a.o., for earlier discussion of movements with mixed A/Ā-properties). Like
for CP-fronting in Äiwoo, these movements can be non-local/skip intervening nominals,
but they otherwise show categorial restrictions and binding-theoretical properties typ-
ical of A-movement. These movements have been analyzed as instances of “mixed A/
Ā-movement”, and this label is intended in a rather literal sense: the probes driving these
movements are simultaneously searching for φ-features and for some Ā-feature(s) (e.g.,
[TOP] or [FOC]), and this combination of features is what derives their properties. For ex-
ample, take the case of CP-fronting in Dinka, which is on the surface very similar to the
Äiwoo case. For van Urk (2015), the probe on C in Dinka is specified as searching for both
[φ] and [TOP], and it can therefore skip an intervening nominal without a [TOP] feature
(204). The resulting movement is therefore non-local.
(204) Dinka CP-fronting can skip an intervening DP without [TOP]:
3!
[CP DP[φ, TOP] C [ I+S: [φ+TOP]M] [TP DP[φ] T [vP  … DP[φ, TOP] …
3!
What I aim to add to this discussion is that the notion of “mixed A/Ā-probing” is not
strictly speaking necessary to derive movement patterns of this kind (that is, with A-
48 As described above (§3.5.3), an additional “plan B” probing step kicks in just in case the first round of [i |v]-
probing could not find anything. This is strictly speaking orthogonal to the issue under discussion here.
140 3.7. The A/Ā-properties of fronting to spec,CP
style categorial restrictions and binding-theoretical properties, but no strict locality). To
be clear, the analysis in terms of mixed/composite probes may of course be completely
correct for the languages mentioned above. Nonetheless, what Äiwoo shows us is that this
movement profile can arise even in the complete absence of any Ā-features on the probe,
purely as an epiphenomenon of Relativized Minimality and the independently needed
notion of featurally specific probes. CP-fronting in Äiwoo could in principle be thought
of as another instance of “mixed A/Ā-movement”; after all, this instance of movement is
strikingly similar to CP-fronting in Dinka (204). However, the probe I posit for Äiwoo C
is not actually interested in any Ā-features. Instead, it searches for [i |v] – which is only
indirectly correlated to the presence of Ā-features49.
Labeling Äiwoo CP-fronting as “mixed A/Ā-movement” would thus be somewhat un-
informative or at worst misleading, because there really is no Ā-ingredient to it whatso-
ever other than the locality profile – which, as we have seen, is independently derived
by probing for [i |v], which is not an Ā-feature (in the sense that it has no information-
structural/discourse import, it is not related to relativization or wh-movement, or any
other classic Ā-processes). The proposal that non-local A-movement patterns can arise
even in the complete absence of Ā-features will be strengthened in chapter 6, when tack-
ling the more complex part of the Word Order Puzzle (i.e., the word order alternation
found in non-canonical UV clauses). We will see that Äiwoo showcases another instance
of featurally motivated, non-local movement with A-type binding properties – this time
with a clear absence of any connection to Ā-features (or even potential “Ā-positions”) of
any kind. This will be the case of movement to spec,TP, which preferentially targets less
local nominals with “marked” φ-features over closer nominals with less marked features.
Our discussion thus far has established why Äiwoo CP-fronting shows the locality
properties it does: the probe on C searches for [i |v], a feature that not all nominals have,
and therefore the resulting movement can be non-local. This, however, leaves open the
question of why the other properties of this movement are the way they are. More specif-
ically: why does Äiwoo CP-fronting have A-style binding-theoretical properties? In the
original proposal of the featural theory of the A/Ā-distinction, van Urk (2015) argues that
the binding-theoretical differences between A- and Ā-movement derive from a difference
in the type of trace these movements leave behind, and how these different traces are in-
terpreted. In this model, movement created by probing for [φ]-features triggers abstrac-
tion over individuals, whereas probing for Ā-features results in abstraction over choice
functions, and the various binding-theoretical differences follow from this split (see van
Urk 2015: 37–51 for details).
Our question about the binding-theoretical properties of Äiwoo CP-fronting, then, be-
comes a question about what kind of feature [i |v] actually is, given the properties we ob-
serve in movement triggered by searching for this feature. Following van Urk (2015), what
49 For now, I have just asked the reader to temporarily trust that [i |v] as the probing feature is the right
analytical choice; the fact that C does not directly search for Ā-features has not been demonstrated yet.
This will be shown to be a fundamental part of the model in the next chapters (4 and 6), where we will see
cases where a nominal may have [i |v] and get raised to spec,CP despite, crucially, not having any Ā-features.
Chapter 3. Äiwoo clause structure 141
we learn is that a feature like [i |v] must somehow behave like φ-features in triggering
abstraction over individuals (i.e., generating movement with A-type binding-theoretical
properties). One possible way to think of this goes back to the idea of goal flagging as
implementing the bidirectional transfer of information under agreement between a probe
and a goal (Deal to appear, going back to e.g. Pesetsky & Torrego 2001, 2007). From this
perspective, “goal flagging” is essentially a close relative of, or really the same thing as,
case assignment; see Clem&Deal 2024 for a concrete proposal to analyze (dependent) case
assignment in terms of goal flagging. Then, “v agrees with the object and flags it with [i |v]”
would be just another way of saying “v assigns accusative case to the object” (although
Äiwoo has almost no have morphological case distinctions). If [i |v] is an equivalent of/
the same thing as case, our analysis could be reframed in perhaps more familiar terms
as follows: (i) v assigns accusative case to Ā-marked objects, while Ā-featureless objects
remain caseless (essentially, a DOM system); (ii) the probe on C is case-discriminating,
preferentially attracting an accusative-marked argument if there is one.
Independently from whether [i |v] is an incarnation of case or not, another relevant
aspect is the fact discussed by van Urk (2015) that Ā-features are optional features on
nominals, in the sense that the speaker themself decides to add them to a nominal or
not (depending on how they want to structure the discourse), whereas φ-features are
obligatory features on nominals. In this dichotomy, [i |v] is definitely closer to the latter
than to the former category. Although not all nominals have [i |v] (unlike [φ]), which do
and which do not is a purely mechanic outcome of the syntactic derivation (like case),
and not up to the speaker to decide. Or at least, not directly: a speaker can choose to put
Ā-features on an object, which in turn will trigger object agreement by v and therefore
cause [i |v] to be present on the object. However, in the next chapters we will see that this
indirect connection between a speaker’s information-structural choices and the presence
of [i |v] is very much intentional, as it will be important in deriving cases where the object
always gets moved to spec,CP, regardless of any information-structural factors.
3.7.3 POTENTIAL PARALLELS
Proposing that non-local A-movement can arise without any connection to Ā-features
(thus, not being an instance of genuinely “mixed A/Ā-movement” in the featural sense)
raises the question of what others patterns this type of analysis could be extended to50.
As briefly mentioned above, a large literature has been produced about apparent locality
violations in A-movement. In the rest of this section I discuss a series of phenomena from
other languages that may be re-analyzed with an Äiwoo-style story. The reader who is
not necessarily interested in the cross-linguistic extensions of this analysis will not need
to devote too much attention to this part.
Among these are cases of A-scrambling in Germanic languages, Hindi, Japanese, Ko-
rean, Mongolian, and other languages (Webelhuth 1989, Mahajan 1990, Saito 1992, Miya-
50 As anticipated, Äiwoo showcases another instance of non-local A-movement to spec,TP, which is clearly
driven by a φ-feature hierarchy effect; see chapter 6. In this section, I focus on parallels from other languages.
142 3.7. The A/Ā-properties of fronting to spec,CP
gawa 1997, McGinnis 1998, Ko 2014, Gong 2023, a.o.). Moreover, similar patterns have
been argued to obtain for locative inversion in English (e.g. C. Collins 1996, though see
Feldcamp 2025), and Generalized Inversion in Russian (Bailyn 2004, 2020). It is less clear
to me how several of these cases could be accounted for by an Äiwoo-style explanation,
since this crucially relies on there being some featural way to single out a lower nominal
to the exclusion of a higher one, at least without positing an ad-hoc scrambling feature
(e.g. Müller 1997, 1998, McGinnis 1998, Grewendorf & Sabel 1999; though see Erlewine &
J. Lim 2023 for a featural account of scrambling in Pangasinan).
These cases aside, there are other attested patterns (discussed under a variety of head-
ings) that might be thought of in terms of an Äiwoo-style analysis. A first example con-
cerns the ordering of Romance clitics, which often has been described as “templatic”
in the sense that it seems to vary arbitrarily both between and within languages (see
e.g. Pescarini 2023, a.m.o.). For example, in French a direct object clitic precedes an in-
direct object one if they are both third person (205), but the opposite order obtains if
the indirect object is 1st/2nd person (206). (These examples are form Crysmann & Luís
2024: 4.)
(205) 3 > 3: DO precedes IO (French)
a. Paul la lui présentera
Paul ACC.3SG.F DAT.3SG introduce.FUT.3SG
‘Paul will introduce her to him’
b. * Paul lui la présentera
Paul DAT.3SG ACC.3SG.F introduce.FUT.3SG
(206) 1/2 > 3: IO precedes DO
a. Paul me la présentera
Paul DAT.1SG ACC.3SG.F introduce.FUT.3SG
‘Paul will introduce her to me’
b. * Paul la me présentera
Paul ACC.1SG.F DAT.1SG introduce.FUT.3SG
Moreover, at least for Italian speakers who have a Weak PCC pattern (myself included;
see e.g. Bonet 1991, Bianchi 2006), for combinations of [PART] direct and indirect object,
the only possible linear order is 1st > 2nd person. This is ambiguous between the direct
object being 1st person and the indirect object being 2nd person, or viceversa (207a). The
opposite order is completely impossible, under any reading (207b).
(207) Italian (Weak PCC): 1st precedes 2nd
a. mi ti hanno presentato
1SG 2SG have.3PL introduced
3 ‘They introduced you to me’
3 ‘They introduced me to you’
b. * ti mi hanno presentato
2SG 1SG have.3PL introduced
Chapter 3. Äiwoo clause structure 143
If one assumes that (i) clitics are base-generated in argument position and then (A-)moved
(Kayne 1975 et seq.), and (ii) the linear order of clitics is established syntactically and not
morphophonologically, these could at least potentially be patterns amenable to an Äiwoo-
style story. For example, one could say that in French arguments carrying [PART] are
attracted by a higher probe than 3rd person ones, or that in Italian 1st person arguments
are attracted by a higher probe than 2nd person ones. See Bianchi (2006) a proposal along
these lines (at least in spirit)51 .
Another such potential case is reported in Myler (2017) for a number of Quechua
varieties, where a 2nd person object raise to a higher position than both 1st and 3rd person
ones, from where it ends up c-commanding the subject, and it can thus feed agreement
in a slot that is normally reserved for the subject. In (208a,b) we see the usual pattern:
the object clitic precedes the tense marker, which in turn precedes subject agreement52.
However, if the object is 2nd person (208c), now it follows the tense marker, and moreover,
the outermost slot that is normally reserved for subject agreement is now controlled by
the 2nd person object as well (leaving the 3rd person subject unexpressed). See Myler
(2017) for arguments that this pattern reflects actual syntactic movement.
(208) Cuzco Quechua: 2nd person arguments are raised higher (Myler 2017: 753):
a. maylla-wa-rqa-n
wash-1.OBJ-PST-3.SUBJ
‘S/he washed me’
b. maylla-∅-rwa-ni
wash-3.OBJ-PST-1.SUBJ
‘I washed him/her’
c. maylla-rqa-su-nki
wash-PST-2.OBJ-2.“SUBJ”
‘S/he washed you’
Another proposal that is more similar in letter to the one being made here is made in
Hammerly (to appear) and Hammerly & Mathieu (to appear) for Long-Distance Agree-
ment (LDA) in Border Lakes Ojibwe, under the term of “relativized EPP”. In this language,
matrix verbs can sometimes agree with an argument in the embedded clause. If the em-
bedded subject is 1st/2nd person and the embedded object is 3rd person, only the embed-
ded subject can control LDA (209a), but not the embedded object (209b)53. However, in
51 Moreover, clitic doubling in languages like Rioplatense Spanish, Greek, Bulgarian and Amharic (Suñer
1988, Anagnostopoulou 2017, Harizanov 2014, Kramer 2014) has been analyzed as showing properties of
A-movement, and since it can target an object from skipping the intervening subject, it could be thought of
as another instance of a movement of this kind.
52 SeeMyler (2017) for arguments why these morphemes are analyzed as clitics vs. agreement markers, though
see also chapter 5 for a critical assessment of these diagnostics.
53 Verbal agreement patterns in Algonquian languages are morphologically highly complex, so I refer the
reader to Hammerly & Mathieu (to appear) for details. Here I have simplified the glosses somewhat, and
added typographical representations of null pronouns for ease of exposition.
144 3.7. The A/Ā-properties of fronting to spec,CP
the opposite configuration, either argument can control LDA (210). The same is true for
combinations of 3rd and 1st person, not shown here (see Hammerly &Mathieu to appear).
(209) 2nd > 3rd: only subject LDA (Hammerly & Mathieu to appear: 9–10)
3
a. gi-gii-waabam-in [CP ∅2SG gii-miigwechiwi’-ad John ]
2-PST-see-2 PST-thank-2>3 John
‘I saw that you thanked John’
7
b. * in-gii-waabam-aa [CP ∅2SG gii-miigwechiwi’-ad John ]
1-PST-see-3 PST-thank-2>3 John
Intended: ‘I saw that you thanked John’
(210) 3rd > 2nd: subject or object LDA (Hammerly & Mathieu to appear: 10)
3
a. in-gii-waabam-aa [CP ikwe gii-miiwechiwi’-ik ∅2SG]
1-PST-see-3 woman PST-thank-3>2
‘I saw that the woman thanked you’
3
b. gi-gii-waabam-in [CP ikwe gii-miiwechiwi’-ik ∅2SG ]
2-PST-see-2 woman PST-thank-3>2
‘I saw that the woman thanked you’
Abstracting away from several complications, Hammerly (to appear) and Hammerly &
Mathieu (to appear) propose that LDA is consistently into a position they identify with
embedded spec,CP. C simply attracts the closest argument to its specifier, whichever that
may be. Crucially, they propose that a probe on Voice has an EPP feature relativized to
[PART], which specifically raises 1st/2nd person objects across subject, but not viceversa.
After this step, a [PART] object and the subject are both in spec,VoiceP, and thus equidistant
from C, which can therefore raise either without intervention problems. This results in
free LDA. The entire process is schematized in (211)–(212).
(211) Embedded 2 > 3: only subject LDA
a. verb-2SG [CP 2SG C … [VoiceP 2SG Voice [ I+S: [PART]M] … 3SG
7
b. * verb-3SG [CP 3SG C … [VoiceP 2SG Voice [ I+S: [PART]M] … 3SG
7
(212) Embedded 3 > 2: subject or object LDA
a. verb-3SG [CP 3SG C … [VoiceP 2SG 3SG Voice [ I+S: [PART]M] … 2SG
Chapter 3. Äiwoo clause structure 145
b. verb-2SG [CP 2SG C … [VoiceP 2SG 3SG Voice [ I+S: [PART]M] … 2SG
Although Hammerly & Mathieu’s notion of “relativized EPP” is not used to model a case
of non-local A-movement per se, it is nonetheless a fairly exact equivalent of what I am
proposing for Äiwoo: a movement-inducing probe that is relativized to a particular fea-
ture, and can therefore trigger movement of a lower nominal across a higher one without
appealing to any Ā-features.
Finally, another case of non-local A-movement that is featurally motivated is proposed
in Brodkin (2025) for Mandar, an Austronesian language of Sulawesi (Indonesia). Through
ordering effects and binding tests, Brodkin (2025) shows that in ditransitive clauses where
both objects are lexical DPs, the goal is base-generated above the theme, and is thenmoved
to an outer specifier of VoiceP, above the subject (213a). However, the probe triggering this
movement has a featural preference for pronouns: if the theme is a pronoun instead of a
lexical DP, then the theme is attracted to this position instead of the goal, which remains
low (213b).
(213) Mandar Voice moves a less-local pronoun over a closer DP:
a. [VoiceP GoalDP Subj Voice [ GoalDP … ObjDP
b. [VoiceP Objpronoun Subj Voice [ GoalDP … Objpronoun
3!
Although Brodkin’s technical implementation is somewhat different from mine, the spirit
is the same: a probe is relativized to agreeing with and moving arguments of a certain
type, and can therefore skip closer non-fitting nominals along the way.
A final case that can be explained in these terms is that of syntactic inversion in so-
called “High Absolutive” Mayan languages (Coon et al. 2014, Assmann et al. 2015, Coon
et al. 2021, Royer 2025, a.o.). In these languages, the object is consistently A-moved across
the subject (214), and this movement step feeds a variety of other syntactic phenomena
(controlling agreement, feeding/bleeding binding relationships, restricting Ā-extraction,
etc.). Importantly, this always happens regardless of the features of the two arguments.
(214) High Absolutive Mayan languages: the object A-moves across the subject
[v/VoiceP Obj Subj v/Voice [VP V Obj
A possible way to capture this pattern, illustrated in (215), is a variation of my proposal for
Äiwoo, where v agrees with Ā-marked objects, but not with objects without Ā-features54.
Suppose that in High Absolutive Mayan languages, instead, v always agrees with all ob-
jects, across the board. Then, a higher head (left unspecified as “F” below) searches for
[i |v], which by assumption will be present on all objects. This derives the desired out-
come: the object will consistently (A-)move across the subject.
54 I thank Amy Rose Deal and David Pesetsky for suggesting this possibility, independently from each other.
146 3.7. The A/Ā-properties of fronting to spec,CP
(215) Syntactic inversion through goal flagging:
FP
Obj
[i|v] [ F ] vP
I+S: [i|v]M
Subj
[ v ]
I+S: [φ] tObj
[i|v]
CHAPTER 4
THE Ā-EXTRACTION RESTRICTION,
AND THE LOGIC OF AN EXCEPTION
It’s not right, but it’s okay
— Whitney Houston
4.1 {OVER/PRE}VIEW
This chapter focuses on a well-trodden topic within Austronesian syntax: the Ā-extraction
restriction. A characteristic trait of many Austronesian languages is that there is a corre-
lation between the voice of a given sentence and what argument(s) may be Ā-extracted
(relativized, topicalized, focused, participate in wh-question formation, etc.). The first half
of this chapter (§§4.2–4.3) will focus on describing how this phenomenon manifests in
Äiwoo, and building a model that correctly derives a particularly interesting incarnation
of the extraction restriction, since in Äiwoo, in addition to its general profile we observe
a set of systematic exceptions. The latter half of this chapter (§§4.4–4.5) will then focus on
one particular category of exceptional Ā-extraction: the acceptability of extraction of the
subject of certain specific UV verbs. Considering once again our overview table (216), this
chapter concentrates on deriving the difference between the first two rows, where extrac-
tion is restricted, and the third, where it’s not. Below, I give an informal characterization
of the puzzle.
148 4.1. {Over/Pre}view
(216) Overview of clausal patterns and Ā-extraction profiles:
Clause type: Word order: Extract non-VIA?
a. AV: S3 φ 7S-V.AV =TAM O no
b. UVplain: O
3 V.UV S7 =TAM no
c. UVonly: O
3 V.UV S3=TAM yes
d. UVinâ: O
NA V.UV S3=TAM yes
e. UV NAgu: O V.UV -πO=TAM S
3 yes
f. UV NA NASVO: S V.UV -πS =TAM O NA
Three smaller appendices conclude this chapter. Appendix §4.A argues for the particular
analysis of relativizing morphemes assumed here contra the existing descriptive litera-
ture. Appendix §4.B is a note about topicalization and focus constructions, arguing that
the former is not movement-based and the latter is a sub-case of relativization. Finally,
appendix §4.C briefly discusses the cases of some verbs with oscillating behavior between
generating clauses of type (216b) or type (216c).
WHAT THE RULES SAY At first glance, Äiwoo is a relatively well-behaved Austrone-
sian language when it comes to relativization. In general, only the VIA of any clause can
undergo relativization, and the other argument(s) cannot (217).
(217) Only the VIA can be extracted:
a. [RC the girl cooked.AV the crab]
3
7
b. [RC the girl cooked.UV the crab]
7
3
Since this is a bidirectional correlation, one can also frame it the other way around: the
extracted argument must be the VIA of its clause (218).
(218) The extracted argument must be the VIA of its clause:
3 cooked.AV
a. The girl [RC that { } the crab ]7 cooked.UV
7 cooked.AV
b. The crab [RC that the girl { } ]3 cooked.UV
WHAT TO DO WHEN THE RULES ARE RIGGED AGAINST YOU We will see that Äiwoo
has a class of transitive verbs that are special in that they do not have an Actor Voice: their
AV form simply doesn’t exist. This is what I’ll call UV-only verbs. Now, we just saw above
how the system imposes a requirement that when a subject is extracted, AV is the only
Chapter 4. The Ā-extraction restriction, and the logic of an exception 149
possible option, and UV is ungrammatical. What is one to do then, when one wants to
relativize the subject of such a UV-only verb, for example te ‘see’? In this case, Äiwoo
syntax shows some understanding towards the speaker: in contrast to the ordinary cases,
non-VIA extraction is exceptionally grammatical (219).
(219) UV-only verbs allow subject extraction:
[RC the girl saw.UV the thief ]
3
3
a. 3 The thief [RC that the girl saw.UV ]
b. 3 The girl [RC that saw.UV the thief ]
DESIDERATA AND CONSEqUENCES On one hand, we want a model that is restrictive
enough to correctly predict the Ā-extraction restriction we find with most Äiwoo verbs
(217)-(218). On the other hand, we want our model to be flexible enough to capture this
systematic exception, where subjects can be extracted specifically from clauses contain-
ing UV-only verbs (219). Intuitively, this problem has an almost OT-like quality to it.
Metaphorically speaking, a constraint banning the existence of the AV form of certain
verbs is ranked higher than a constraint banning non-VIA extraction (“*see.AV≫ *subject-
extraction-from-UV”).
Such exceptions to the extraction restriction (that is, the possibility to extract non-
VIA/pivor arguments) have occasionally been discussed in the literature, but they “appear
in scattered sources and lack a unified analysis” (Hsieh 2025: 520). Erlewine & C. Lim
(2023) and Hsieh (2025), for example, discuss exceptional non-pivot Ā-extraction in Bikol
and Tagalog respectively (see also Ceña 1979, McGinn 1988, Nakamura 1996). However,
these languages are different fromÄiwoo in important respects, the primary one being not
having a lexically defined class of UV-only verbs that allow this. Furthermore, see Tanaka
et al. (2016), Pizarro-Guevara & Wagers (2018, 2020, 2021, 2024), and Pizarro-Guevara &
Garcia (2024) for experimental perspectives, focusing on Tagalog.
The non-absolute character of the Ā-extraction restriction in Äiwoo has important
consequences for our theory of this restriction. Particularly, I will argue that it constitutes
evidence for an approach where the extraction restriction in Äiwoo is a consequence of
specifically Ā-intervention, and it cannot arise due to phasehood (contra e.g. Rackowski &
Richards 2005, Hsieh 2020, 2025, Erlewine & Sommerlot 2023b for Tagalog and several
Malayic languages, a.o.) or other types of DP-intervention (contra e.g. Aldridge 2004, 2008,
Erlewine & C. Lim 2023 for Tagalog, Indonesian, and Bikol, a.o.).
150 4.2. Setting the stage: Äiwoo relative clauses
4.2 SETTING THE STAGE: ÄIWOO RELATIVE CLAUSES
4.2.1 EMPIRICAL DESCRIPTION
Throughout this chapter and the rest of the dissertation, I use relativization as the main
example of Ā-extraction (see this chapter’s appendix 4.B for why relativization and not
other types of Ā-dependencies). Therefore, it’s important to first establish an empirical
baseline of what relative clauses (RCs) in Äiwoo look like. A typical example is given in
(220). We can observe that (i) there is no obvious overt complementizer, aside perhaps
from the optional marker mi= at the left edge1; (ii) in the absence of mi=, the relative
clause (bracketed) is simply juxtaposed to the right of the head noun kuli ‘dog’, and (iii)
there is a gap inside the RC corresponding to the head noun.
(220) [kuli (mi=)[RC ki-ngoduwâ-lâoo ngä paveli eângâ]]=kâ no Mary
dog REL= IPFV-shout-always in garden that=DIST POSS Mary
‘[The dog that kept barking in that garden] is Mary’s’ (241015)
Because of the lack of embedding morphology – aside from the optional mi= – and the
most often RC-initial position of the gap, a DP containing a relative clause is often string-
identical to a main clause having the head noun as its VIA/pre-verbal argument. Compare
the bracketed portion of (220) to (221), which is just an entire (main) clause on its own,
with no extraction going on.
(221) kuli ki-ngoduwâ-lâoo ngä paveli eângâ (241015)
dog IPFV-shout-always in garden that
‘The dog kept barking in that garden’
First and foremost, even in the absence of mi= we can be sure that the bracketed portion
of (220) is a DP containing a relative clause because it has the external syntax of a DP with
respect to the rest of its larger clause. That entire string is being parsed as the subject of
a copular clause, marked as such by the particle =kâ. We can observe the same string be-
ing used in clearly defined argument positions in the next set of examples2. In (222a) it’s
the direct object of a UV clause, in the VIA position between a sentence-initial adverbial
(dâbu dâuwângâ ‘every day’) and the verb. In (222b) it’s the subject of an UV clause, be-
1 As a bookkeeping note going forward, the examples throughout the text will variably contain the marker
mi= or not, according to how they were produced in the elicitation context. For the sake of exposition,
throughout the text I will just assume these are (optional) relativizers, contra Næss (2017b). See appendix
4.B for a discussion of this and a few other similar markers, and arguments in favor of this analysis.
2 In natural speech (as attested from the corpus), speakers often avoid having heavy constituents like the
ones in (222a,b) in argument positions. This is because due to the syntax of arguments in Äiwoo, these
configurations will unavoidably involve a degree of center-embedding, which is known to be taxing to
process (Miller & Chomsky 1963). Instead, speakers will frequently topicalize the heavy DP containing
a relative clause, and then leave a null pronoun in the corresponding argument position. For example, a
slightly more natural version of (222b) would be (i); the null resumptive pronoun (the subject of ‘bite’) is
boxed.
Chapter 4. The Ā-extraction restriction, and the logic of an exception 151
tween the verb and the TAM/negation particles. Finally, in (222c) it’s the complement of
a preposition.
(222) DPs containing a RC in various argument positions: (241015)
a. Object of UV clause:
dâbu dâuwângâ [kuli (mi=)[RC ki-ngoduwâ-lâoo ngä paveli
day every dog REL= IPFV-shout-always in garden
eângâ]]Obj ku-wâpunâ-kâ-no, mo ba ku-mâlu=gu
that IPFV-feed.UV-DIR3-1MIN but NEG IPFV-stop=NEG
‘Every day I feed the dog that’s always barking in that garden, but it doesn’t
stop’
b. Subject of UV clause:
gino-mu ba ku-mangi [kuli (mi=)[RC ki-ngoduwâ-lâoo ngä
son-2MIN NEG IPFV-bite.UV dog REL= IPFV-shout-always in
paveli eângâ]]Subj=gu=naa
garden that=NEG=FUT
‘The dog that’s always barking in that garden won’t bite your son’
c. Complement of preposition:
bugulo i-tou-kä dekilingä go [kuli (mi=)[RC
yesterday 1MIN-bring.AV-DIR3 food to dog REL=
ki-ngoduwâ-lâoo ngä paveli eângâ]]
IPFV-shout-always in garden that
‘Yesterday I brought food to the dog that’s always barking in that garden’
The fact that the head noun is external to the RC is not obviously evident from these
examples, since the gap is always RC-initial and there is no overt complementizer (again,
aside from the optional mi=). Even having convinced ourselves that these are indeed DPs
containing RCs, we could imagine that they’re internally-headed relative clauses, with
the head noun being in its canonical argument position inside the RC. However, we can
demonstrate that the head noun is in fact external by adding other material at the start of
(i) [kuli (mi= )[RC ki-ngoduwâ-lâoo ngä paveli eângâ]]TOP=kâ, gino-mu
dog REL= IPFV-shout-always in garden that=DIST son-2MIN
ba ku-mangi -∅n =gu=naa
NEG IPFV-shout-always-3MIN=NEG=FUT
‘The dog that’s always barking in that garden, it won’t bite your son’ (241015)
However, this is clearly a performance-related effect. Sentences with a “reasonable” degree of center-
embedding like the ones above are accepted by my consultant as grammatical, since – despite being maybe
hard to parse – they don’t break any syntactic rules of the language. In fact, center-embedding structures
like these are attested in the natural speech corpus. One example is (ii), repeated from chapter 3, §3.5.2,
where the entire DP “the person sitting in the boat” stands between the verb and the negative particle =gu.
(ii) ngaama lâ ba i-kää [me=[RCki-tokoli-woli-mä ngä botu]]=gu=nâ
if DIST NEG ASP-know.UV REL:person=IPFV-sit-down-DIR1 in boat=NEG=DIST
‘If [the person sitting in the boat] doesn’t know (it), …’ (Nyeliwânubona toponu 141)
152 4.2. Setting the stage: Äiwoo relative clauses
the RC, to the left of the posited gap. As we have seen earlier (chapter 3, §3.2), adverbials
like dâbu dâuwângâ ‘every day’ can precede the VIA position, but not intervene between
the VIA and the verb (223a). In light of this, consider now (223b). If singedâ ‘girl’ was
internal to the RC, we would expect it to appear in the position where I have notated the
gap, and it would be ungrammatical to the left of the adverbial. Therefore, in (223b) we
can be sure that the adverbial marks the left edge of the RC, thus proving both that the
head noun is external and that there is indeed a gap in its correspondent position inside
the RC. The sentences in (224) make the same point, but with object extraction from a UV
clause instead of subject extraction from AV. Sentence (224b) was judged as only slightly
degraded, but possible.
(223) a. Monoclausal context (AV): *{DP} Adv 3{DP} V … (240418)
*{singedâ } dâbu dâuwângâ 3{singedâ } ki-tou-lâoo-kä flaua
girl day every girl IPFV-bring.AV-always-DIR3 flower
go mekivaavee nogo-∅n
to teacher POSS-3MIN
‘Every day the girl brings flowers to her teacher’
b. Subject relativization across a sentence-initial adverbial:
[singedâ [RC dâbu dâuwângâ ki-tou-lâoo-kä flaua
girl day every IPFV-bring.AV-always-DIR3 flower
go mekivaavee nogo-∅n]]=nâ i-te-kâ-no
to teacher POSS-3MIN=DIST ASP-see.UV-DIR3-1MIN
‘I met a girl who every day brings flowers to her teacher’
(224) a. Monoclausal context (UV): *{DP} Adv 3{DP} V … (240516)
*{buk } dâbu dâuwângâ 3{buk } ki-gââ pedevalili=kaa
book day every book IPFV-read.UV children=FUT
‘Every day the children will read the book’
b. Object relativization across a sentence-initial adverbial:
(?)kâ-no=ngä nyä-veve [buk [RC dâbu dâuwângâ
want-1MIN=CV 1MIN.IRR-buy.AV book day every
ki-gââ pedevalili=kaa]]
IPFV-read.UV children=FUT
‘I want to buy a book that every day the children will read’
Another strong piece of evidence for an analysis of the RC head as external comes from
the exceptional cases of non-VIA extraction, which – as anticipated – Äiwoo allows in
particular contexts. These will be discussed in depth in the second half of this chapter.
For now, their relevance is that in these contexts one can very clearly observe the RC-
external position of the head. In (225a), the head noun singedâ ‘girl’ corresponds to a gap
that is not sentence-initial inside the RC, thus clearly demonstrating that the head noun is
RC-external. Compare this to the baseline OVS order that the RC portion of (225a) would
have if it were just an independent clause (225b).
Chapter 4. The Ā-extraction restriction, and the logic of an exception 153
(225) a. Extraction across an overt DP shows that RCs are head-external: (250717)
kokä polis=kä nä-li-eeu-kä go [singedâ [RC mekipä
want police=CV IRR-3AUG-speak-DIR3 go girl thief
i-te-kä-∅n =to]]
ASP-see.UV-DIR3-3MIN =TAM
‘The police wants to talk to the girl who has seen the thief’
b. Baseline OVS order:
mekipä i-te-kä singedâ=to
thief ASP-see.UV-DIR3 girl=TAM
‘The girl has seen the thief’
4.2.2 RELATIVIZATION IS AN Ā-DEPENDENCY
After having examined what RCs look like, we need to ascertain that relativization is in
fact an Ā-dependency in Äiwoo. Indeed, relativization shows two important hallmarks of
Ā-dependencies cross-linguistically. First, it can cross clausal boundaries. The sentences
in (226) show long-distance relativization, for transitive subjects and objects; (227) is an
attested example of long-distance relativization of an intransitive subject3.
(226) Long-distance relativization: (240606)
a. [singedâ mi=[RC kâ-no=ngä [ ki-veve-wâ=kaa nupwää
woman REL= think-1MIN=CV IPFV-buy.AV-DIR2=FUT clothes
nâ-nugu-mu]]] i-wâ=to bugulo
IRR-POSS-2MIN ASP-go=PRF yesterday
‘The woman who I thought would buy clothes for you left yesterday’
b. dengâ [nupwää mi=[RC kâ-no=ngä [ ki-ve-wâ
that clothes REL= think-1MIN=CV IPFV-buy.UV-DIR2
Mary=kaa nâ-nugu-mu]]]
Mary=FUT IRR-POSS-2MIN
‘Those are the clothes that I thought Mary would buy for you’
(227) [me=[RC kä-i=lä [ ku-wopo-lâ-mä go King Deved]]]
REL:person= say-3AUG=CV IPFV-come-out-DIR1 from King David
‘The person who they say descends from King David’ (MT 12:23)
Despite being able to cross clausal boundaries, however, relativization is not completely
free. As we might expect from an Ā-dependency, relativization cannot cross syntactic
islands. More specifically, a relative clause constitutes itself an island for further rela-
tivization (228). Examples (228a,b) serve as a baseline to establish that relativizing both
3 I assume there is an additional gap as notated, to the left of kânongä ‘I think’, as a result of successive-
cyclic movement, though this is not crucial. The details of long-distance relativization in Äiwoo are being
investigated in ongoing work.
154 4.2. Setting the stage: Äiwoo relative clauses
the object and the subject from this kind of clause is independently grammatical4. The
crucial test case for islandhood is then (228c). Here, first we relativize the object from RC1
(‘the dog that the boys adopted’). Once we’ve done this, RC1 is an island for relativization,
so extracting the subject to create RC2 is impossible.
(228) a. Baseline 1: (240606) 
[kuli mi=[RC i-luwa-kä pesigiläi eângâ]]=kâ mi=olo
dog REL= ASP-take.UV-DIR3 boys that=DIST REL=big
‘The dog that those boys adopted was big’
b. Baseline 2:
[pesigiläi mi=[RC kuli eângâ i-luwa-kä -i ]]=lâ
boys REL= dog that ASP-take.UV-DIR3-3AUG=DIST
ki-li-vevaale-päko ngämi
IPFV-3AUG-take.care-good of.it
‘The boys who adopted that dog took good care of it’
c. Test case: a RC is itself an island for relativization
* [pesigiläi mi=[RC2 [kuli mi=[RC1 i-luwa-kä -i ]]=lâ
boys REL= dog REL= ASP-take.UV-DIR3-3AUG=DIST
mi=olo]]=kâ ki-li-vevaale-päko ngämi
REL=big=DIST IPFV-3AUG-take.care-good of.it
Intended: ‘The boysi who the dogj that theyi adopted was big treated itj well’
4.2.3 THE SYNTACTIC STRUCTURE(S) OF ÄIWOO RCS
A few types of analyses have been proposed for relative clauses. These can be chiefly di-
vided into three main groups. First, we can distinguish between analyses depending on
where the head of the RC is base-generated. For operator-movement and matching anal-
yses (229a,b), the head is base-generated outside of the RC, whereas for raising analyses
(229c), the head is base-generated internally to the RC and then moves to a position out-
side of it (Schachter 1973, Vergnaud 1974, Kayne 1994, Bianchi 1999, Bhatt 2002). Then,
within the first group we can distinguish two analyses depending on the nature of the
movement dependency established within the RC. For operator-movement analyses (Par-
tee 1975, Chomsky 1977, Jackendoff 1977), what undergoes movement is a null operator
or a relative pronoun (229a). For matching analyses (Chomsky 1965, Sauerland 1998, 2000,
2003, Cinque 2020), it’s a second instance of the head, which then undergoes deletion un-
der identity with the (RC-external) head (229b). All three trees in (229) are meant to be
heavily simplified illustrations, abstracting away from a lot of differences in the details of
the various analyses cited above.
4 This is because luwa ‘take’ is a “UV-only verb”, which therefore allows extraction of its subject despite being
UV; see §4.5. In this case, we see an “overt gap” (notated as -i ): the extraction site – in the red box –contains
an overt suffix, rather than being null like elsewhere. This is because a pronoun in this position (spec,TP)
must be overtly realized (as a suffix) even if it is a lower trace of movement, as a result of post-syntactic
Local Dislocation onto the verb. See chapter 5, §5.5 for discussion.
Chapter 4. The Ā-extraction restriction, and the logic of an exception 155
(229) a. Operator movement: b. Matching: c. Raising:
head CP head CP head CP
Opi head
C C
… ⟨Opi⟩ … C … ⟨head⟩ … … ⟨head⟩ …
Distinguishing between operator-movement and raising/matching analyses is often not
a trivial task. The main source of evidence comes from different kinds of connectivity
effects, that is, any signs that the head noun of the RC is present inside the RC at some
stage of the derivation. These connectivity effects may be of different kind: morphological
case patterns (inverse case attraction; Bianchi 1999, Deal 2016a, Abramovitz 2021), idiom
chunk interpretation (Schachter 1973, Jarvis 2025), semantic scope facts (e.g. Bhatt 2002),
binding-theoretical effects (Schachter 1973), among others.
However, when it comes to Äiwoo, all of these phenomena are hard or impossible to
investigate for independent reasons. Äiwoo does not have morphological case, we don’t
know of any NPIs, and we don’t know of any VP-idioms like ‘make headway’ whose parts
we could manipulate in the relevant way. Moreover, classic binding principles A/B/C do
not provide a good testing ground either, due to language-specific confounds. For example,
Condition C can be shown to apply in the canonical (structural) way; see chapter 3, §??. At
the same time, however, its effects are often obscured by an additional, linearly-defined
strict ban on cataphora targeting arguments in certain structural positions. Essentially,
from certain positions a pronoun embedded inside a larger DP cannot corefer to any R-
expression to its right, regardless of any structural factors. The patterns are quite complex,
and not of importance here; see Roversi (2024a) for details. However, this anti-cataphora
effect would get in the way of testing the possibilities for Condition C reconstruction
in relative clauses. As for binding principles A and B, the main issues are (i) there is no
morphologically dedicated anaphor, and the regular pronouns are used instead (lit. “hei
cut himi”; see a brief description in chapter 5, §5.3.2); (ii) reflexive contexts of this kind
result in sentences with pronominal objects, and these (“John saw him/himself/us/you/
...”) have rather different syntax and word order compared to sentences with lexical DP
objects (“John saw the dog”); these word order alternations will be the focus of chapter 6.
Setting aside these kinds of connectivity effects, then, what little evidence we have at
the current stage (to be presented below) seems to point towards the existence of, at least,
non-raising/operator movement-type RCs in Äiwoo. This, alone, does not necessarily ex-
clude the presence of other types of structures for Äiwoo RCs. That RCs are structurally
ambiguous and can be derived in multiple ways in the same language is already known to
be a possibility in several languages; among others, this case has been made for English
(Bhatt 2002, Hulsey & Sauerland 2006), Bulgarian (Krapova 2010), Georgian (S. Foley 2013),
Nez Perce (Deal 2016a), Hebrew (Sichel 2018), and Atchan (Jarvis 2025).
156 4.2. Setting the stage: Äiwoo relative clauses
As for non-raising derivations, the evidence comes from cases of what at least superfi-
cially looks like extraposition. I follow Fox & Nissenbaum (1999) and Hulsey & Sauerland
(2006) in assuming that extraposed RCs must have non-raising derivations; see the cited
works for the arguments behind this claim. Äiwoo speakers often produce structures like
the sentences in (230), that could potentially be analyzed as involving extraposition. Here,
the RCs (bracketed and labelled) are uttered at the end of the clause instead of appearing
to the immediate right of the head noun; (231) are the non-extraposed baselines for com-
parison. If we could confirm that these structures indeed involve genuine extraposition,
that would be an argument in favor of non-raising derivations being available for Äiwoo
RCs.
(230) Putative RC extraposition:
a. [nubole] i-epavi-no bugulo [RC i-tu-mä Mary]
taro ASP-cook.UV-1MIN yesterday ASP-bring.UV-DIR1 Mary
‘Yesterday I cooked the taro that Mary brought me’ (240411)
b. [butete] ki-ii Anna=kaa bulaape [RC i-kili-no
potato IPFV-peel.UV Anna=FUT tomorrow ASP-dig.UV-1MIN
pevaio=ke]
morning=PROX
‘Tomorrow Anna will peel the potatoes that I harvested this morning’ (250123)
(231) Non-extraposed baselines:
a. [nubole [RC i-tu-mä Mary]] i-epavi-no bugulo
taro ASP-bring.UV-DIR1 Mary ASP-cook.UV-1MIN yesterday
‘Yesterday I cooked the taro that Mary brought me’ (240411)
b. [butete [RC i-kili-no pevaio=ke]] ki-ii Anna=kaa
potato ASP-dig.UV-1MIN morning=PROX IPFV-peel.UV Anna=FUT
bulaape
tomorrow
‘Tomorrow Anna will peel the potatoes that I harvested this morning’ (250123)
At the very least, we can be sure that the RC portion that I am labelling as “extraposed” is
not simply stranded in situ by the fronting of the head noun. That is, we can exclude the
following analysis:
(232) Analysis to be excluded:
[nubole] i-epavi-no [nubole [RC i-tu-mä Mary]]
taro ASP-cooked.UV-1MIN taro ASP-brought.UV-DIR1 Mary
‘I cooked the taro that Mary brought me’
Remember that in Äiwoo, all sentential adjuncts (PPs and adverbs, chiefly) are linearized
to the right of a postverbal object. That is, any object in the vP domain must surface
immediately right adjacent to the TAM material, with nothing intervening. This is true
for objects in AV clauses (233a) and in UVSVO clauses (233b) (see chapter 6 for details about
these).
Chapter 4. The Ā-extraction restriction, and the logic of an exception 157
(233) Postverbal objects must always be right-adjacent to the TAM particles:
a. Jen ki-ebi=kaa 3{butete} ngä nyopwä *{butete} (230914)
Jane IPFV-bake.AV=FUT potato in oven potato
‘Jane will bake potatoes in the oven’
b. ( ijii) ku-potaa-i=laa 3{iude} ngâ nuumä *{iude} (250717)
3AUG IPFV-search.UV-3AUG=FUT 12AUG in town 12AUG
‘They will look for us in town’
Moreover, we can see this adjacency effect even in canonical UV clauses, where the object
is fronted to spec,CP. When the object is plural, a 3AUG pronoun ijii can optionally be
pronounced in the same position as the objects in (233). Note that nouns are number-
neutral in Äiwoo, so that this optional pronoun is the only overt signal of plurality for
kuli ‘dog’; see chapter 2, §2.5.1. I assume this optional pronoun is a reduced spell-out of
the lower copy of the object (a very similar phenomenon is known from Dinka; van Urk
2015, 2018). Once again, we see that this pronoun must be adjacent to the TAM particles,
and precede sentence adjuncts:
(234) kuli i-wâpunâ-kâ-no 3{(ijii)} bugulo *{ijii} (250717)
dog ASP-feed.UV-DIR3-1MIN 3AUG yesterday 3AUG
‘Yesterday I fed the dogs’
Consider now again one of the putative cases of RC-extraposition, repeated in (235). If the
bracketed RC portion here was stranded by the object after fronting to spec,CP, we would
expect that it should surface to the left of the adverbial bugulo ‘yesterday’, since that is
the in situ position of objects. Instead, we see that the RC follows the adverbial.
(235) [nubole] i-epavi-no bugulo [RC i-tu-mä Mary] (240411)
taro ASP-cook.UV-1MIN yesterday ASP-bring.UV-DIR1 Mary
‘Yesterday I cooked the taro that Mary brought me’
If confirmed, the presence of extraposition in Äiwoo would support the existence of non-
raising derivations for RCs. As highlighted above, however, this alone does not necessarily
exclude that raising derivations might be possible too, although we cannot confirm this
at the present time.
For concreteness, I will assume that RCs in Äiwoo have the following structure (236).
Inside the RC, the constituent undergoing relativization carries an Ā-feature [REL]. A rel-
ativizing head (“Rel”) takes CP as its complement. This head contains a probe searching
its c-command domain for a constituent carrying a [REL] feature, and it will move the goal
to its specifier ([INT: RELM]); why the satisfaction feature is set as [Ā] will become clear
in §4.3. Relativization of an argument is impossible when the relativizing probe cannot
agree with the argument in question.
158 4.2. Setting the stage: Äiwoo relative clauses
(236) General tree for Äiwoo RCs:
RelP
DP
[REL]
[ Rel ] CP
I: [REL]M
S: [Ā]
DP
[REL] C TP
…
I remain agnostic as to how exactly the moved DP[REL] constituent within the RC is con-
nected to the RC-external head noun. Because of the extraposition facts discussed above,
at the very least it should be possible that this RC-internal moved DP is just a relative op-
erator, and not the head noun itself. The posited Ā-feature [REL] may then be thought of as
the feature that is characteristic of/identifies relative operators. For simplicity, throughout
the examples below I will only mark the extraction site with a gap (notated as ) in the
(last) position of the relativized constituent before moving to spec,RelP – i.e., in spec,CP
here (236) – and not the final landing site in spec,RelP.
There is some possible evidence for the presence of this high Ā-position above spec,CP,
coming from certain effects in the syntax of wh-questions. In Äiwoo, leaving a wh-word
in situ is always possible. This is shown for objects and subjects in UV (237); see also Næss
(2015a).
(237) Object question: (230327)
a. doo i-ngä Jon(=kâ)? b. nubole i-ngä iie(=kâ)?
what ASP-eat.UV John=DIST taro ASP-eat.UV who=DIST
‘What did John eat?’ ‘Who ate the taro?’
However, this interacts with a strict ban on (linearly defined) cataphora that targets the
VIA position in non-AV clauses, that is, UV and CV; see Roversi (2024a) for more details.
For example, the possessor pronoun in (238) embedded inside the subject cannot corefer
with ‘Mary’, despite there being no c-command relation between them:
(238) [nubole [RC i-kili -∅n bugulo]] ki-epavi Mary (230608)
taro ASP-dig.UV-3MIN yesterday IPFV-cook.UV Mary
‘Maryi is cooking the taro that s/hej/*i harvested yesterday’
In light of this, consider now the following minimal pair. Sentence (239a) is a perfectly
grammatical wh-in situ question, cf. (237b) above. However, only a disjoint reading is
possible, because of the anti-cataphora effect. To get the intended bound reading, iie ‘who’
is fronted to a position before the object, in a seeming violation of V2 (239b). Now, there
is no cataphora, because the possessor pronoun follows ‘who’, so binding can succeed.
Chapter 4. The Ā-extraction restriction, and the logic of an exception 159
(239) a. # [buk no-∅n] ki-gââ iie=kaa? (240201)
book POSS-3MIN IPFV-read.UV who=FUT
‘Whoi will read hisj/*i book?’
b. iie [buk no-∅n] ki-gââ-∅n=naa?
who book POSS-3MIN IPFV-read.UV-3MIN=FUT
‘Whoi will read hisi book?’
Because the fronted constituent is a wh-word, (239b) cannot involve topicalization. I as-
sume that the position where ‘who’ exceptionally surfaces in (239b) is the high Ā-position
posited in (236), above spec,CP. Presumably, because wh-in situ is always possible in Äi-
woo, a wh-word will not need to be pronounced in its high position in a general case,
but the pronunciation of this copy might be required in certain circumstances to obviate
independent constraints.
4.3 THE AUSTRONESIAN Ā-EXTRACTION RESTRICTION
In this section, I will illustrate how the analysis of Äiwoo clauses developed in chapter
3 also explains the pattern of Ā-extraction in the language, without any extra necessary
mechanisms. The model I propose will be (in part) based on Ā-intervention: the highest
argument in a clause, in virtue of carrying [PIV], intervenes for Ā-extracting other lower
arguments. (In the other half of the chapter, we will then see how this kind of model also
opens the door to the specific kind of exceptions to the Ā-extraction restriction that Äiwoo
allows.) I will discuss how other theories that have been proposed of similar patterns
in other Austronesian language, respectively based on phasehood and DP-intervention,
would not be able to account for the Äiwoo facts.
4.3.1 WHAT CAN BE RELATIVIZED WHEN
Having established what relativization looks like in the language and that it is indeed
an Ā-movement dependency, we are now ready to examine which arguments can be rel-
ativized. Many Austronesian languages with the kind of voice system we see in Äiwoo
notoriously restrict relativization (and often other Ā-dependencies) to the VIA/pivot argu-
ment; see §4.3.2 for references. This is the so-called “pivot-only” Austronesian Extraction
Restriction. Given that in its most core manifestation the restriction is biconditional, one
can frame it two ways. On one hand, in (240), we can fix the argument we want to extract,
and ask “if I want to extract argument X, which voice(s) can I use?”. On the other hand,
in (241), we can fix the voice of a given clause, and ask “from a clause in voice X, which
argument(s) can I extract?”. These options are both shown schematically in (240)-(241).
160 4.3. The Austronesian Ā-Extraction Restriction
(240) The extracted argument must be the VIA of its clause:
3 cooked.AV
a. The girl [RC that { } the crab ]7 cooked.UV
7 cooked.AV
b. The crab [RC that the girl { } ]3 cooked.UV
(241) Alternative framing: from any clause, only the VIA can be extracted:
a. [RC the girl cooked.AV the crab]
3
7
b. [RC the girl cooked.UV the crab]
7
3
A concrete example of this restriction in another Austronesian language is given in (242)–
(244), adapted from Pearson (2005: 412–413). Malagasy, like Äiwoo, features three voices:
AV, UV, and CV (although differently from Äiwoo, Malagasy word order is pivot-final
rather than pivot-initial)5. I use the shorthand notation ‘X[Y’ to iconically represent ex-
traction of X from domain Y. For example, ‘S[AV’ should be read as “(Ā-)extraction of a
subject from an Actor Voice clause”; ‘O[UV’ means “(Ā-)extraction of an object from an
Undergoer Voice clause”, etc.
As anticipated, relativization is strictly pivot-only. From AV, only the subject can be
extracted (242a), and not the object or an applied argument (242b,c). From UV, only the
object can be extracted (243a), and not the subject6 or an applied argument (243b,c). Fi-
nally, from CV, only the applied argument – here, an instrument – can be extracted (244a),
and neither the subject nor the object can (244b,c)7. Similar paradigms can be found in the
literature for other languages with similar systems, like Tagalog (Rackowski 2002, Rack-
owski & Richards 2005, a.o.), Seediq (V. Chen 2023), Dinka (van Urk 2015), etc.
(242) Only the subject can be extracted from AV:a. 3 S [AV:
ny mpamboly [( izay) mamono ny akoho amin-ny antsy]
DET farmer REL AV.kill DET chicken with-DET knife
‘The farmer who is killing chickens with the knife’
5 Pearson (2005) and other Malagasy literature uses the term “trigger” where I use “pivot”, and consequently
uses the terms “Actor Trigger”, “Theme Trigger”, etc.; I have simply reglossed these for consistency. See
Pearson (2001: §§2.2.1, 2.4, 2005: §3.1) for details about the complex verbal morphology of Malagasy.
6 For some Austronesian languages, this is a simplification. In fact, subject extraction from UV is sometimes
possible, perhaps somewhat degraded but not ungrammatical. See the discussion about Tagalog non-pivot
subject extraction in 4.3.2.
7 For expository reasons, here I’m grouping the sentences by which voice the RC shows. This is a reordering
from Pearson (2005), which groups them by extracted argument (i.e., the three subject-extraction sentences
(242a)-(243b)-(244b) are grouped together, and so are all three object-extraction sentences, etc.).
Chapter 4. The Ā-extraction restriction, and the logic of an exception 161
b. 7 O[AV:
* ny akoho [( izay) mamono amin-ny antsy ny mpamboly]
DET chicken REL AV.kill with-DET knife DET farmer
Intended: ‘The chickens which the farmer is killing with the knife’c. 7 X [AV:
* ny antsy [( izay) mamono ny akoho (amin’) ny mpamboly]
DET knife REL AV.kill DET chicken with DET farmer
Intended: ‘The knife that the farmer is killing the chickens with’
(243) Only the object can be extracted from UV:
a. 3 O[UV:
ny akoho [( izay) vonoin’ ny mpamboly amin-ny antsy]
DET chicken REL UV.kill DET farmer with-DET knife
‘The chickens which the farmer is killing with the knife’b. 7 S [UV:
* ny mpamboly [( izay) vonoina amin-ny antsy ny akoho]
DET farmer REL UV.kill with-DET knife DET chicken
Intended: ‘The farmer who is killing chickens with the knife’c. 7 X [UV:
* ny antsy [( izay) vonoin’ ny mpamboly (amin’) ny akoho]
DET knife REL UV.kill DET farmer with DET chicken
Intended: ‘The knife that the farmer is killing the chickens with’
(244) Only the applied argument (X) can be extracted from CV:
a. 3 X[CV:
ny antsy [( izay) amonoan’ ny mpamboly ny akoho]
DET knife REL CV.kill DET farmer DET chicken
‘The knife that the farmer is killing the chickens with’b. 7 S [CV:
* ny mpamboly [( izay) amonoana ny akoho ny antsy]
DET farmer REL CV.kill DET chicken DET knife
Intended: ‘The farmer who is killing chickens with the knife’c. 7 O [CV:
* ny akoho [( izay) amonoan’ ny mpamboly ny antsy]
DET chicken REL CV.kill DET farmer DET knife
Intended: ‘The chickens which the farmer is killing with the knife’
Äiwoo behaves exactly how we would expect an Austronesian language to behave, per-
fectly replicating the Malagasy paradigm. Here, too, what can undergo relativization is
only the VIA in any given clause8. From an AV clause, only the subject can be extracted
8 As mentioned earlier in passing, in the examples in (228a,b), both the subject and the object could be ex-
tracted from the same UV clause. This is because the verb luwa ‘take’ is a UV-only verb, that simply does not
162 4.3. The Austronesian Ā-Extraction Restriction
(245a) and not the object (245b). From a UV clause, conversely, only the object can be
extracted (246a) and not the subject (246b).
(245) Only the subject can be extracted from AV: (240425)a. 3 S [AV:
[pedevalili (mi=)[RC ki-li-vängä=kaa sii]]=kâ
children REL= IPFV-3AUG-eat.AV=FUT fish=DIST
pelivano gisi
children brother.1MIN
‘The kids who will eat the fish are my brother’s children’9b. 7 O [AV:
* [sii (mi=)[RC pedevalili ki-li-vängä=kaa ]]=kâ
fish REL= children IPFV-3AUG-eat.AV=FUT =DIST
i-epavi-no
ASP-cook.UV-1MIN
Intended: ‘I cooked the fish that the children will eat’
(246) Only theobject can be extracted from UV: (240425)a. 3 O [UV:
[sii (mi=)[RC ki-ngä pedevalili=kaa]]=kâ i-epavi-no
fish REL= IPFV-eat.UV children=FUT=DIST ASP-cook.UV-1MIN
‘Icooked the fish that the children will eat’b. 7 S [UV:
* [pedevalili (mi=)[RC sii ki-ngä -i =laa]]=kâ pelivano
children REL= fish IPFV-eat.UV-3AUG=FUT=DIST children
gisi
brother.1MIN
Intended: ‘The kids who will eat the fish are my brother’s children’
These patterns are summarized in (247). Again, due to the one-to-one correspondence, one
can frame it in two logically equivalent ways: (i) only the subject can be extracted from
AV, only the object from UV, etc.; (ii) to extract a subject one needs AV and not UV/CV, to
extract an object one needs UV and not AV/CV, etc.
(247) Extract ↓ from →: AV UV
Subject 3 7
Object 7 3
have an AV form. These verbs and the Ā-extraction patterns they showcase will be the topic of the second
part of this chapter.
9 The nouns pedevalili and pelivano are both glossed as ‘children’, but the former is non-relational (‘kids, peo-
ple of very young age’), and the latter is relational/obligatorily possessed (‘someone’s children/offspring’).
Chapter 4. The Ā-extraction restriction, and the logic of an exception 163
Given the structural description of RCs given above (§4.2.3), we already have an account
for the impossibility of extracting an object from an AV clause, without any additional
mechanisms. If an object is going to be relativized, this means it must carry the Ā-feature
[REL]. But because it has an Ā-feature, this causes v to agree with it. Therefore, a clause
where the object undergoes (or tries to undergo) relativization will simply never be able to
be AV; that is just an underivable scenario. However, at this point we still cannot explain
why the subject of a UV clause cannot be extracted. We turn to this in the next section.
4.3.2 DEVELOPING A THEORY OF THE EXTRACTION RESTRICTION
Restrictions on Ā-extraction of the kind presented above have been discussed in a vast
literature, both when it comes to Austronesian languages (Schachter 1973, 1976, Kroeger
1993, Guilfoyle et al. 1992, Nakamura 1996, Aldridge 2004, 2008, 2017, Rackowski 2002,
Rackowski & Richards 2005, Erlewine et al. 2017, V. Chen 2017a, 2023, Erlewine 2018,
Hsieh 2020, 2025, Keine & Zeijlstra 2025, a.m.o), and languages from other families that
show similar patterns (Dinka < Nilotic, van Urk 2015; Kawahíva < Tupi-Guaraní, dos
Santos 2023). Whereas in chapter 3 we divided families of theoretical approaches to Aus-
tronesian voice into voice-as-cause vs. voice-as-effect, an orthogonal division can bemade
based on how these theories model the extraction restriction. Chiefly, two large-tent
camps can be identified. Some theories make crucial use of the concept of phasehood.
Others rather model this as an intervention effect, i.e., a byproduct of Relativized Mini-
mality.
Here are two abstract illustrations of both kinds of approaches. Since I assume that all
movement is triggered by a step of agreement, I represent this with a line targetting both
DPs coming from some unlabelled node higher in the structure. Remember that I use the
typographical convention of notating a successful agree step with a dashed line ending in
a solid circle, whereas an open circle signifies unsuccessful agreement.
Let’s consider a configuration with a higher DP1 and a lower DP2, and a phase bound-
ary in between them. Here, both kinds of theories make the same predictions: DP1 can be
extracted, and DP2 cannot. What is different between the two systems is the reason for
this restriction. In a phase-based approach (248a), DP2 can’t be agreed with and thereby
extracted because it’s below a phase boundary (typically, the clause-internal vP phase, as
shown here; I abstract away from whether or not a constituent in the phase edge is a pos-
sible goal). In an intervention approach (248b), on the other hand, DP2 can’t be agreed
with because DP1 is in the way, and a higher probe can’t see past it. This is standard Rela-
tivized Minimality, as long as DP1 and DP2 have some relevant feature(s) in common. In
other words, the problem for extracting DP2 is one of absolute locality for a phase-based
approach, whereas in an intervention approach the problem is one of relative locality.
164 4.3. The Austronesian Ā-Extraction Restriction
(248) a. Phase-based approaches:  b. Intervention approaches:
DP1 DP1
3 vP 3 vP
7
v v
DP2 DP2
7
For the time being, I remain deliberately vague about the exact feature(s) that are relevant
for the intervention effect in (248b). This will be object of in-depth discussion later. I
proceed now to argue that a phase-based approach is unfit to modeling the Äiwoo pattern,
and that therefore an intervention-based one will be superior.
4.3.2.1 AGAINST A PHASE-BASED ACCOUNT OF ÄIWOO EXTRACTION
In the preceding discussion, we have seen how a phase-based theory of the Austronesian
extraction restriction makes predictions as to which DPs should be available for extrac-
tion based on their absolute position in the clausal structure. The idea is simple: any DP
above a certain phase boundary can be extracted; any DP below the same boundary can-
not. Of course, if multiple DPs are above the phase boundary, then all of them should
be extractable. First, we will see how this prediction can be a feature and not a bug for
certain languages (Tagalog in Hsieh 2020, 2025; Bikol in Erlewine & C. Lim 2023). Then,
I will argue that this would be indeed be a bug for Äiwoo, and would make incorrect
predictions.
Hsieh’s (2020) work will constitute a good illustration of this prediction about the
ability to extract multiple DPs. I present here a simplified version of it, abstracting away
from some details; the reader is referred to the original work for the full discussion. First,
Hsieh (2020, 2025) models Tagalog as involving case-agreement between Voice (a head
above v) and the pivot argument, similarly to what we are doing here for Äiwoo, but
crucially involving early inversion rather than late. When Voice agrees with the pivot, it
moves it to its outermost specifier, rendering it the only DP available for extraction, due
to phasehood. In AV, for example, the subject is the only DP above the phase boundary,
so the object is too low to be extracted (249).
(249) Non-pivot extraction: 7 O[AV
*Obj [RC verb.AV [VoiceP Subj [vP tSubj … tObj ]]]
7
Chapter 4. The Ā-extraction restriction, and the logic of an exception 165
Hsieh (2020) then makes an empirical observation: in Tagalog, non-pivot extractions are
not all born alike. Extracting objects from AV is indeed ungrammatical (249)-(250a). More
interestingly, extracting subjects from UV10 is in many cases actually not judged by na-
tive speakers as ungrammatical, but as only somewhat degraded, and fundamentally ac-
ceptable (250b). This has also been confirmed experimentally in Tanaka et al. (2016) and
Pizarro-Guevara & Wagers (2018, 2020, 2021)11; see also Pizarro-Guevara & Garcia (2024).
(250) Asymmetry in Tagalog non-pivot Ā-extraction (Hsieh 2025: 535):a. 7 O [AV: ungrammatical
*mani=ng [RC mag-ha~hanap ang ibon ⟨ng mani⟩]
peanut=LK AV-FUT~search NOM bird GEN peanut
Intended: ‘Peanut that the bird will look for’b. 3 S [UV: somewhat degraded but attested and acceptable
? ibo[n]=ng [RC ha~hanap-in ⟨ng ibon⟩ ang mani]
bird=LK FUT~search-UV GEN bird NOM peanut
‘Bird that will look for the peanut’
Hsieh discusses how Tagalog also independently features an understudied phenomenon
which he labels Genitive Inversion. In this construction, what normally would’ve been an
agent in UV (which in Tagalog bears genitive case) is instead realized in a high preverbal
position, in oblique case (251). Importantly, Genitive Inversion is strictly limited to non-
pivot agents: themes in general and AV agents cannot participate in this process. See also
Hsieh (2023) for more details about Genitive Inversion.
(251) Genitive Inversion: [V DPGEN] ⟶ [DPOBL=LK V …]
aki[n]=ng bi~bilh-in ⟨ko⟩ ang damit
1SG.OBL=LK FUT~buy-UV 1SG.GEN NOM clothes
‘I will buy the clothes’ (Hsieh 2025: 537)
Finally, we now have the ingredients to model the possible non-pivot subject extraction(S [UV) in (250b). Simply put, Genitive Inversion can feed relativization. Hsieh argues
that Tagalog has two ways for an argument to get out of the lower phase. Standard pivot
movement can bring a DP above the phase boundary, which is why pivots can always be
extracted. Additionally, Genitive Inversion constitutes a separate way for (some) DPs to
be placed above the phase boundary. Crucially, once Genitive Inversion has taken place,
both the agent and the theme can be relativized, as both are above the phase boundary:
the theme due to being the pivot, and the agent due to Genitive Inversion.
10 I change Hsieh’s glosses from PV (“Patient Voice”) to UV, to ease comparison with Äiwoo. This is purely a
terminological choice, and of no theoretical substance.
11 Interestingly, in a follow-up study Pizarro-Guevara & Wagers (2024) find that there are essentially two pop-
ulations of Tagalog speakers, those who accept non-pivot subject extraction (S[UV) and those who don’t
(although this is a rather coarse simplification of their findings, which are more nuanced and distinguish
between relativization and wh-movement).
166 4.3. The Austronesian Ā-Extraction Restriction
(252) Extraction possibilities with genitive inversion: (Hsieh 2025: 540–542, adapted)a. 3 S [UV:
ibo[n]=ng [RC ha~hanap-in ang mani]
bird=LK FUT~search-UV NOM peanut
‘Birdthat will look for the peanut’b. 3 O [UV:
mani=ng [RC kanya=ng ha~hanap-in ]
peanut=LK 3SG.OBL=LK FUT~search-UV
‘Peanut that he will look for’
Let’s now examine how precisely what makes this analysis good for Tagalog, also makes it
unfit for Äiwoo. Importantly, the general clausal syntax of the two languages is different.
In Tagalog, only one argument – the pivot – is raised above the clause-internal phase
boundary by standard pivot movement. In Äiwoo, however, we saw in chapter 3 that the
subject always moves to spec,TP. Crucially this is also true when it’s not the VIA, like
in UV, as can be easily seen by its position to the left of TAM material. Under relatively
standard assumptions about which functional projections are phases (e.g. Chomsky 2001),
this would mean that in a UV clause both arguments are above the phase boundary. This
would then predict that in UV we should be able to relativize either arguments, contrary
to fact. The examples in (253) are repeated from above, adding the phase boundary under
discussion and the relevant trace of subject movement12.
(253) From UVyou can’t extract the S, despite it being above the phase boundary:a. 3 O [UV: (240425)
[sii [RC ki-ngä pelivanou=ngaa tS ]]
fish IPFV-eat.UV children.1MIN=FUT vP
‘The fish that my children will eat’
b. 7 S[UV:
* [pedevalili [RC sii ki-ngä -i =laa tS ]]
children fish IPFV-eat.UV-3AUG=FUT vP
Intended: ‘The kids who will eat the fish’
This alone is sufficient to rule out a phase-based analysis of the Ā-extraction restriction
we observe in Äiwoo. Moreover, in the next sections we will also see that indeed some
non-VIA (UV) subjects can be relativized (like in Tagalog, but with crucial differences
as to when and how). However, importantly there will be nothing different between the
structural position of these extractable non-VIA subjects vs. the canonical non-extractable
ones shown here. An abstract illustration of the problem is given in (254).
12 Even if we assumed that in Äiwoo TP constituted a phase boundary, in chapter 2 we will see instances
where subject extraction is allowed from an even lower position, i.e. the one marked by the subject trace in
(253) (spec,vP). This will constitute an even stronger argument against phasehood being the determining
factor behind the Ā-extraction restriction in Äiwoo.
Chapter 4. The Ā-extraction restriction, and the logic of an exception 167
(254) a. In most cases: 7 S[UV
* head noun [RC object verb.UV [TP =TAM [vP t Subj … ]]]
7
b. In some exceptional contexts: 3 S[UV
head noun [RC object verb.UV [TP =TAM [vP t Subj … ]]]
3
The precise nature and derivation of the constructions that allow non-VIA extraction will
be discussed below. However, their sheer existence shows that one can still detect an
asymmetry in extractability between certain arguments, despite them being in the same
identical structural positions. Therefore, phasehood and absolute locality cannot provide
a valid explanation of the Äiwoo extraction restriction.
4.3.2.2 AN (Ā-)INTERVENTION THEORY OF ÄIWOO EXTRACTION
In the last section we have seen how in cases where an argument can’t be Ā-extracted
in Äiwoo, the problem cannot be (only) its absolute position. We can now consider the
logical alternative: certain nominals can’t be extracted due to their relative position with
respect to other elements in the relevant domain. This is another widespread approach
to modelling extraction restrictions of various sorts, both the voice-conditioned one com-
monly found in Austronesian languages, and also what is often labelled “syntactic ergativ-
ity”, that is, the presence of Ā-extraction processes that cannot target ergative arguments
(transitive subjects), but can target absolutive ones (intransitive subjects and transitive
objects); see Deal (2016b) and references therein for an overview.
In theories of this kind, extraction of a nominal is blocked by the presence of an inter-
vener higher up in the structure – typically, another nominal. An influential analysis of
this kind of the Austronesian extraction restriction is that of Aldridge (2004, 2008). Here is
a rather simplified illustration. This is a voice-as-cause approach featuring early syntactic
inversion; however, the intervention idea is independent from both of these aspects of the
analysis. In AV (255a), nothing of note happens. Both arguments remain in the canoni-
cal position they’re base-generated in. In UV (255b), on the contrary, the object is raised
above the subject. This is the crucial step that makes extraction of the subject impossible
in UV. In both cases, the higher DP (the subject in AV, the object in UV) intervenes for
extracting the lower DP (the object in AV, the subject in UV).
168 4.3. The Austronesian Ā-Extraction Restriction
(255) The higher DP intervenes for extracting the lower one:
a. Actor Voice: b. Undergoer Voice:
… vAVP … vUVP
3 Subject 3
v VP ObjectAV
7 7 Subject
vUV VP
V Object
V tObj
Of course, the existence of syntactic inversion doesn’t alone explain why a lower DP
shouldn’t be Ā-extracted. Ā-movement is canonically non-local, and thus able to skip in-
tervening DPs (think of any non-subject questions in English, for example; see the dis-
cussion of the locality of A/Ā-movement in chapter 3, §3.7). Therefore, Ā-movement of
the lower DP is only blocked once we make the additional assumption that in a language
showcasing an extraction restriction of this kind, Ā-movement is somehow subject to
stricter locality constraints than we might be used to. Particularly, we need to assume
that Ā-movement can only target to the highest DP (Branan & Erlewine 2024). This is, in
a way, the logical opposite of the non-local A-movement discussed in chapter 3: it is Ā-
movement, but with a stricter locality profile than usual.
Perhaps because patterns of “surprisingly local” Ā-movement are so widespread cross-
linguistically, this intuition has been formalized in a lot of recent literature, often under the
headings of syntactic ergativity and/or mixed A/Ā-movement (see a.o. Legate 2014, van
Urk 2015, Baier 2018, Erlewine 2018, Ostrove 2018, Bossi & Diercks 2019, D’Alessandro
2020, Colley & Privoznov 2020, Scott 2021, Coon et al. 2021, Jarvis 2023, Jenks 2023,
Branan & Erlewine 2024, Lohninger 2025; see also Keenan & Comrie 1977 for an influ-
ential typological observation that can be framed in the same terms). Within the frame-
work I assume this dissertation, this pattern can easily be implemented, as the probe is
separately specified for what features it copies and what features stop it from searching
further. A probe that has Ā-features in its interaction condition, but φ-features (or [D],
e.g.) in its satisfaction condition, will derive precisely the desired kind of strictly local
Ā-movement. Consider the tree in (256), illustrating the mechanism just described. The
probe is seeking to copy back Ā-features, and move the goal that carries them. However,
upon meeting DP1, the probe has its satisfaction condition met ([D]), and therefore the
search immediately halts. The Ā-carrying DP2 is never found by the probe. This is pre-
cisely what Branan & Erlewine (2024) label “Ā-probing for the closest DP”. See also Deal
(2024b) for a more in-depth discussion of various types of movements with mixed A/Ā-
locality profiles, and how they can be modelled in this framework.
Chapter 4. The Ā-extraction restriction, and the logic of an exception 169
(256) Local-only Ā-movement:
[probe]
I: [Ā]M
S: [D]
DP1
SAT! [D] DP2 …
[Ā, D]
7
Having discussed how we can think about patterns of Ā-movement with the locality pro-
file canonical of A-movement, we now ought to ask whether this is the correct interpre-
tation of the extraction restriction in Äiwoo. Indeed, in all the cases of Ā-extraction we
have seen so far, the only extractable nominal was in fact the highest one in the clause (the
VIA in spec,CP). This is summarized in (257), where superscript 3/7 indicates whether
an argument can/cannot be Ā-extracted.
(257) Partial overview of clause patterns and extractability (canonical clauses):
Voice Word Order
AV S3 V =TAM O7
UV O3 V S7=TAM
On the one hand, we briefly saw earlier how our model already derives the impossibility
of object extraction from AV (the first row in (257)). The reasoning is shown in (258). If
the object is to be relativized, this means it carries an Ā-feature ([REL]). In turn, this will
cause v to agree with the object, thus giving us UV and not AV. In other words, whenever
the object is going to be relativized, AV just cannot arise in the first place.
(258) If the object is going to be relativized, AV cannot arise:
[ … [vP Subj v I: [Ā] [VP … Obj[REL] ]]]
[S: [D]]
However, we still need an explanation of why subjects cannot be extracted from UV
clauses. The fact alone that throughout (257), only the highest argument can be extracted
would support an Aldridge-style highest-DP-only approach. However, given the analysis
of the clausal structure of Äiwoo developed in chapter 3, we have a confounding factor.
Although we know that this intervention effect is caused by Relativized Minimality, we
don’t know exactly relativized to what. Here is why. For concreteness, consider the fol-
lowing general illustration of the problem in (259).
170 4.3. The Austronesian Ā-Extraction Restriction
(259) Relativization is impossible across the pivot DP:
[ Rel ]
I: [REL]M CP
S: […?]
SAT! DP1
[D, PIV] C …
7, DP2 …
but why? [D, REL]
Remember that we modelled pivothood as an Ā-feature [PIV], carried by an argument.
Specifically, in a UV clause, the object is a DP[PIV] that ends up in spec,CP. In general
then, we know that the highest nominal in every UV clause will also always carry an Ā-
feature (DP1 here, the pivot in spec,CP). By assumption, the constituent we are trying to
relativize also carries an Ā-feature, [REL], and the relativizing probe aims to move some-
thing carrying [REL]. Finally, we know from the empirical facts that DP2 (the lower one)
cannot be relativized.
This teaches us that something about DP1 makes it an intervener, that is, DP1 carries
some feature(s) that satisfy the relativization probe, so that this cannot reach the subject
(DP2) and move it. The problem, of course, is that we don’t know what the offending
feature(s) might be. Evidently, we have at least two candidate features. On one hand, this
could be an issue of DP-intervention: the probe has [SAT: D], and will be stopped by any
DP (as in the model discussed above). On the other hand, it could also be that this is an
issue of Ā-intervention: although DP1 was raised to spec,CP by (non-local) A-movement
(chapter 3, §3.7), it always carries an Ā-feature after all ([PIV]), and we could hypothesize
that this is what is stopping the probe. That is, the probe is specified as [SAT: Ā], where [Ā]
subsumes both [PIV] and [REL]; see e.g. Abels 2012, Aravind 2017, 2018, Callegari 2018).
If this problem is one of DP-intervention (as in Aldridge’s analysis), we would be
witnessing yet another case of Ā-movementwith the locality profile of A-movement – that
is, only the highest DP can move. If, on the other hand, this is a problem of Ā-intervention,
we would also have a different but familiar problem, as Ā-features are known to intervene
for each other. Consider for example the canonical effect of wh-superiority (260b), where
the higher wh-phrase makes it impossible to move the lower one.
(260) a. Who do you wonder [CP bought what]?
7
b. * What do you wonder [CP who bought ]?
Chapter 4. The Ā-extraction restriction, and the logic of an exception 171
The same intervention dynamic obtains between different types of Ā-movement, such
as wh-movement, focus movement, relativization, etc. (Abels 2012, Aravind 2017, 2018,
Callegari 2018). For example, in Italian, embedded foci are possible (261a). However, an
embedded focus will block wh-movement across it (261b). See Abels (2012) for arguments
why this should be regarded as an intervention effect, as opposed to a function of how
the left periphery of Italian CPs is organized (contra Rizzi 1997, 2004a,b)13.
(261) a. Embedded foci are possible:
credo [CP che [IL PREMIO NOBEL]FOC abbiano dato tFOC
believe.1SG that the prize Nobel have.SBJV.3PL given
a Rita]
to Rita
‘I think that they gave Rita THE NOBEL PRIZE’
b. Wh-movement cannot cross an embedded focus:
* [a chi]wh credi [CP che [IL PREMIO NOBEL]FOC abbiano
to who believe.2SG that the prize Nobel have.SBJV.3PL
dato tFOC twh]?
given
Intended: ‘Who do you think they gave THE NOBEL PRIZE to?’
I argue that this is precisely what goes wrong with non-VIA subject extraction (*S[UV)
in Äiwoo. The tree (262) illustrates this intervention effect. The DP[PIV] object in spec,CP
satisfies the relativizing probe before this can ever reach the subject in spec,TP. Therefore,
the subject cannot be extracted.
13 Note, however, that not all Ā-features intervene for each other. For example, wh-movement in Italian can
take place across a topicalized constituent (i) (the reported judgment is from the cited source, and I share
it). See Abels (2012), Aravind (2017, 2018), and Callegari (2018) for more discussion of which Ā-features
intervene for which, and what this teaches us about their feature hierarchy.
(i) ? non so [a chi]wh pensi [CP che, [queste cose]TOP, le dovremmo
NEG know.1SG to who think.2SG that these things them should.1PL
dire tTOP twh]
say
‘I don’t know to whom you think that, these things, we should say’ (Rizzi 2004a: 232)
172 4.3. The Austronesian Ā-Extraction Restriction
(262) [PIV] on the object Ā-intervenes for extracting the subject:
RelP
7
[ Rel ] CP
I: [REL]M
S: [Ā]
SAT! DP
[PIV] C TP
7 7! DP
[REL] …
Under this approach, it is the [PIV] feature on the higher argument that is blocking ex-
traction of a lower one, and not just the higher argument’s mere presence or category.
The crucial evidence will come from cases of exceptional non-VIA extraction. In special
circumstances, Äiwoo in fact does allow to extract a lower argument across a higher one.
Specifically, this applies to subjects of certain UV clauses. An example is provided in (263).
Here, the agent singedâ ‘girl’ is clearly being relativized across another argument, specif-
ically the object mekipä ‘thief’ in sentence-initial position. (See fn. 4 about the nature of
the gap in the red box.)
(263) Non-pivot extraction is sometimes possible: (250717)
singedâ [RC mekipä i-te-kä -∅n =jo]
girl thief ASP-see.UV-DIR3-3MIN=TAM
‘The girl who has seen the thief’
The properties and derivation of constructions like these will be the focus of the next
sections. For the time being, we can limit ourselves to take these as evidence that not
all nominals intervene for Ā-extraction in Äiwoo. In the canonical cases discussed above,
a VIA object blocks extraction of everything else, but here (263) we see how that is not
always the case. Therefore, we can exclude DP-intervention as the cause for the extraction
restriction in Äiwoo, and this alone provides support for our Ā-bar intervention approach.
Chapter 4. The Ā-extraction restriction, and the logic of an exception 173
4.4 UV-ONLY VERBS: A DESCRIPTION
This section presents an empirical description of what I call “UV-only verbs”, which simply
only exist in UV and lack an AV form altogether14. That some verbs of this kind exist in
Äiwoo has been briefly noted in passing in the descriptive literature (Næss 2015b: 279,
Næss 2024: 9). However, the existence of these verbs has important implications for any
syntactic analysis of Äiwoo, and of Austronesian languages in general. This is the first
piece of work examining these implications. The present section is largely descriptive,
and a reader who is more interested in the theoretical implications of the existence of
these verbs might want to just skim it, and proceed to the §4.5.
4.4.1 AN IDIOSYNCRATIC LEXICAL CLASS
First, let us establish what verbs we are talking about. At the time of writing, we have no
exhaustive list of UV-only verbs in Äiwoo. Or rather: we have a list of the ones we know,
but no guarantee that there are no others in the lexicon. Several of these have surfaced
during the elicitation process, and I have no reason not to expect that others might do so
in the future.
The known UV-only verbs are the following: kää ‘know’ (264a), te ‘see’ (264b), lâwâle
‘help’ (264c), nyida ‘love’ (264d), wagu ‘tell’ (264e), luwa ‘take’ (264f), tuwo ‘hold, grasp’,
siwo ‘hold, keep’. Moreover, Äiwoo has two applicatives -ive (general; (264i)) and -i (comi-
tative) that attach to intransitive stems. Any verb built with these applicatives is a UV-only
verb as well (264i,i). All the examples below contain negation and/or the future marker
=Caa, to show that the subject is in the typical UV position (between the verb and the
TAM particles).
(264) Examples of UV-only verbs (250717)
a. Jon ba i-kää Mary=gu
John NEG ASP-know.UV Mary=NEG
‘Mary doesn’t know John’
b. Jon ba ki-te-kä Mary=gu=naa ngâ nuumä
John NEG IPFV-see.UV-DIR3 Mary=NEG=FUT in village
‘Mary will not see John in town’
c. Jon ba ki-lâwâle-kä Mary=gu=naa ngâ pole
John NEG IPFV-help.UV-DIR3 Mary=NEG=FUT with work
‘Mary will not help John with the work’
d. Jon ba ki-nyida Mary=gu
John NEG IPFV-love.UV Mary=NEG
‘Mary doesn’t love John’
14 This is different from how so-called deponent verbs behave e.g. in Latin (see e.g. Weisser 2014, a.m.o.), since
those have passive morphology but otherwise active syntax and interpretation. UV-only verbs have not
only UV morphology but also UV syntax, as will be shown below.
174 4.4. UV-only verbs: a description
e. lopwâ ba ki-wagu-mä Mary=gu=naa bulaape
story NEG ASP-tell.UV-DIR1 Mary=NEG=FUT tomorrow
‘Mary will not tell me a/the story tomorrow’
f. nyibä nogo Mary ba i-luwa-kä Jon=gu   (230706)
basket POSS Mary NEG ASP-take.UV-DIR3 Jon=NEG
‘John didn’t take Mary’s basket’
g. mekipä ba ku-tuwo polis=gu=naa ngä skul
thief NEG IPFV-grab.UV police=NEG=FUT in school
‘The police will not catch/arrest the thief at school’
h. buk ba ki-siwo Mary=gu ngä nyimä
book NEG IPFV-hold.UV Mary=NEG in hand.3MIN
‘Mary is holding a/the book in her hand’
i. Jon ki-sââ-ive Mary=kaa go ku-pole-mana
John IPFV-proud-APPL Mary=NEG because IPFV-work-very
‘Mary will be proud of John because he works hard’
j. nubââ ba ki-eâ-i mekitei=gu=naa ngä bolevi
shark NEG ASP-paddle-COM fisherman=NEG=FUT in shore
‘The fisherman will not paddle (with) the shark to the shore’
The following verbs might also be UV-only, although their status is less clear:weevä ‘visit’,
komaa ‘invite’, and possibly potaa ‘search, look for’. The reason for the uncertainty is that
these show a behavior that overlaps both with UV-only verbs and with voice-syncretic
verbs; see §4.4.3 for the difference between these.
At the present state of research, all the known UV-only verbs don’t seem to obviously
belong to a coherent natural class at any syntactic or semantic level. They are better de-
scribed as constituting an idiosyncratic or arbitrary lexical class, and I will treat them as
such. Semantically, some of them are verbs might be characterized as “psych-verbs” (Bel-
letti & Rizzi 1988) in the sense that their subject is an experiencer (‘know’, ‘see’, ‘love’).
However, this does not extend to verbs like ‘grab/take’, ‘hold’, and ‘help’, which describe
physical actions. On the syntactic side, at least some of them (‘know’, ‘see’, ‘love’, ‘help’,
possibly ‘search’) are verbs that cross-linguistically often show peculiar case patterns, for
example taking dative-marked subjects or objects. For example, Icelandic hjálpa ‘help’
takes a dative direct object instead of an accusative one (Thráinsonn 1979, Sigurðsson
1991, 1992, a.m.o.). Once again, however, this property does not extend to all UV-only
verbs in Äiwoo. Hopefully, future work will be able to shed more light on what verbs are
UV-only, and on whether they constitute any natural class.
4.4.2 PROVING THE NON-EXISTENCE OF AV FORMS
I claim that AV forms of these verbs do not exist. Indeed, no AV form of these verbs is
known, but of course, absence of evidence is not evidence of absence. My claim, there-
fore, must be argued for somewhat indirectly. The main reason is the general irregularity
Chapter 4. The Ā-extraction restriction, and the logic of an exception 175
of Äiwoo voice morphology. In other Austronesian languages (e.g. Tagalog, Formosan lan-
guages, etc.), voice morphology is segmentable and regular. If we want to know whether
the AV form of a given stem exists, we simply attach an AV marker to that stem. In Äiwoo,
however, we don’t have this luxury. As described in chapter 3, §3.3, verbs mostly just have
two stems, one for AV and one for UV, with largely irregular/idiosyncratic alternations.
In short, there is simply no reliable way to derive either form from the other. Therefore,
something we cannot do is “add AV morphology” to verbs like kää ‘know’ or te ‘see’, be-
cause there just is no such morphology. As a result, we must pursue other ways to show
that these verbs really only exist in UV, and never in AV.
Since we cannot use stem morphology to reliably observe voice, we must rely on the
three other reliable indicators described in §3.3: word order, voice concord on modifiers,
and presence of prefixal φ-agreement. For our UV-only verbs, all three of these phenomena
point in the same direction: no AV version exists or is possible. First, considerword order.
As we should expect, only UV-style OVS is ever grammatical (265a), and never AV-style
SVO (265b)15. Because of the irregularity of Äiwoo voice morphology, of course we cannot
a priori assume that the AV form of te ‘see.UV’ would also be /te/. However, seen as there
is no reliable way to derive a putative AV form of this verb (which I claim does not exist),
throughout the examples below I use small caps (e.g. TE ‘see’, KÄÄ ‘know’) to mean “there
is no stem in the language that will make this sentence grammatical”.
(265) Word order: always like UV, never like AV (250728)
a. devalili ki-te-usi-kä mikilivaavee=kaa
child IPFV-see.UV-again.UV-DIR3 teachers=FUT
‘The teachers will see a/the child again’
b. * mikilivaavee ki-li-TE-ute-kä=naa devalili
teachers IPFV-3AUG-SEE.AV-again.AV-DIR3=FUT child
Intended: ‘The teachers will see a/the child again’
Speaker’s comment: ‘kiliutekä(naa) is not a word’
Consider now voice concord and prefixal φ-agreement, both described in §3.3. As for the
former: when adverbial modifiers are attached to a transitive verb, they must take a suffix
-i/nyii if the base verb is UV, and they cannot take this suffix if the base verb is AV. As
for the latter: AV verbs consistently take a subject-agreement prefix, whereas UV verbs
never do so. In fact, the latter take a subject-indexing suffix, although I show in chapter
5 that this is not syntactic agreement, but just a phonologically reduced subject pronoun
in spec,TP. Two baselines examples where both of these phenomena can be observed are
given in (266).
(266) a. nuwopa eângâ ki-lââ-päko-i-mana-i-no=ngaa (240425)
house that IPFV-build.UV-good-UV-very-UV-1MIN=FUT
‘I will build that house/those houses very well’
15 The only caveat will be non-canonical UV sentences with SVO order; these are treated in chapter 6, and are
independent from the issue of UV-only verbs vs. ordinary voice-alternating verbs.
176 4.4. UV-only verbs: a description
b. i-ki-lâwââ-päko-mana=kaa nuwopa
1MIN-IPFV-build.AV-good-very=FUT house
‘I will build a house/houses very well’
As can be seen from the examples in (265), UV-only verbs only ever show the UV pattern of
prefixal agreement and voice concord (265a), and never the AV one (265b), as expected16.
As a side note, in elicitation contexts native speakers will often volunteer the AV form
of a verb if I have mistakenly constructed it wrong. It is instructive, then, that the same
speakers reject any attempt at putting AV-style morphology on a UV-only verb stem,
simply judging the word ill-formed; see the comment in (265b).
To sum up, we have seen that UV-only verbs have only one stem, and show no sign
whatsoever of ever existing in AV. All possible morphosyntactic indicators (word order,
voice concord, prefixal φ-agreement) point in the same direction: these verbs just do not
have an AV form.
4.4.3 UV-ONLY VERBS ARE NOT JUST SYNCRETIC FOR VOICE
We must also exclude the possibility that these verbs might just be morphologically syn-
cretic, with both the AV and UV stem being homophonous. In fact, such voice-syncretic
verbs do exist in Äiwoo. However, they are different from UV-only verbs in the simple fact
that their AV form exists, despite their stem being identical to the UV one. For example, nu
‘drink’ has the same stem in both voices, but the three indicators considered above show
that an AV form clearly exists (267a): it can have SVO order, adverbial modifiers can take
no voice concord suffix (mana ‘very’), and it can take φ-prefixes (lu- 3AUG).
(267) ‘Drink’ has both voices, but with the same stem nu: (231012)
a. AV: SVO order, no voice concord, φ-prefixes
pedevalili ku-lu-nu-mana=kaa nuwoi
children IPFV-3AUG-drink.AV-very=FUT water
‘The children will drink a lot of water’
b. UV: OVS order, voice concord, no φ-prefixes
nuwoi ku-nu-mana-i pedevalili=kaa
water IPFV-drink.UV-very-UV children=FUT
‘The children will drink a lot of (the) water’
Furthermore, aside from these morphosyntactic indicators, we can tell that nu ‘drink’ has
both voices by syntactically imposing either AV or UV. Conforming to the Ā-extraction
restriction described in the first part of this chapter (§4.3), if we extract the subject only the
AV form is possible (268), and if we extract the object only the UV form is possible (269).
We will see in the next section that on the contrary, UV-only verbs stubbornly remain
UV even when the subject is extracted, thus exceptionally allowing non-VIA extraction.
16 The UV form usi of the adverbial modifier meaning ‘again’ is presumably segmentable as ute-i ‘again-UV’,
where ute is the AV/default form, with vowel hiatus resolution and regular t > s / i (see chapter 2, §2.4.3).
Chapter 4. The Ā-extraction restriction, and the logic of an exception 177
In other words, an equivalent of (268b) with a UV-only verb (e.g. ‘see’ instead of ‘drink’)
will be perfectly grammatical, and an equivalent of (268a) will be ungrammatical.(268) Subject extraction: 3 S [AV, 7 S [UV (231026)
a. [pedevalili (mi=)[RC ku-lu-nu-mana nuwoi]]=kâ
children REL= IPFV-3AUG-drink.AV-very water=DIST
ki-li-nudâ-mana
IPFV-3AUG-thirsty-very
‘The children who were drinking a lot of water were very thirsty’
b. * [pedevalili (mi=)[RC nuwoi ku-nu-mana-i -i ]]=lâ
children REL= water IPFV-drink.UV-very-UV-3AUG=DIST
Intended: ‘The children who were drinking a lot of water’
(269) Object extraction: 3 O[UV, 7 O[AV (231026)
a. [nuwoi (mi=)[RC ku-nu-mana-i pedevalili]]=kâ
water REL= IPFV-drink.UV-very-UV child=DIST
ba päko=gu
NEG good=NEG
‘The water that the children were drinking a lot (of) was not good’
b. * [nuwoi (mi=)[RC pedevalili ku-lu-nu-mana ]]=kâ
water REL= children IPFV-3AUG-drink.AV-very =DIST
Intended: ‘The water that the children were drinking a lot (of)’
As briefly mentioned in the introduction of this section, a few verbs show a sort of mixed
behavior between being voice-syncretic and UV-only. Three known ones are potaa ‘search,
look for’,weevä ‘visit’, and komaa ‘invite’. To avoid an unnecessary digression for the sake
of exposition, I treat these “oscillating” verbs in this chapter’s appendix 4.C.
4.4.4 UV-ONLY VERBS IN THE AUSTRONESIAN CONTEXT
The specific kind of strictly UV-only verbs that I describe here for Äiwoo has not been
reported to exist in any other Austronesian languages, to my knowledge, and as such it
constitutes a novel finding. However, somewhat similar phenomena have been reported
for a few languages, although they all differ from the Äiwoo case in certain respects.
Tagalog has some verbs that have been reported to show a preference for UV (Latrouite
2011 calls these “Undergoer-oriented verbs”). Latrouite (2011: 196–197) reports that Taga-
log speakers find the AV forms of these verbs “strongly dispreferred” (270). However,
unlike for Äiwoo UV-only verbs, once one extracts the subjects, AV is indeed the only
possible option (271).
178 4.4. UV-only verbs: a description
(270) Sira ‘broken’ is dispreferred in AV (Latrouite 2011: 197; glosses adapted):
a. ?? s⟨um⟩ira ang bata ng mesa
⟨AV.PFV⟩broken NOM child GEN table
Intended: ‘The child destroyed/broke a table’
b. s⟨in⟩ira-∅ ng bata ang mesa
⟨PFV⟩broken-UV GEN child NOM table
‘The child destroyed the table’
(271) Subject extraction imposes AV:
sino ang {3s⟨um⟩ira / *s⟨in⟩ira-∅ } sa mesa?
who NOM ⟨AV.PFV⟩broken ⟨PFV⟩broken-UV DOM table
‘Who destroyed/broke the table?’ (Henrison Hsieh, p.c.)
Moreover, it is not entirely clear how deep this “AV-dispreference” actually is in the first
place. For example, Henrison Hsieh (p.c.) reports to me that using the abilitative/non-
volitional AV form improves sentences like (270a), cf. (272a). The same is true for kita ‘see’,
another Tagalog verb that was reported by Latrouite (2011) as belonging to this class of
verbs. For two Tagalog speakers I was able to consult, sentences with the abilitative AV
form of this verb are unremarkably acceptable (272b)17.
(272) The abilitative/non-volitional AV form of these verbs is acceptable:
a. naka-sira ang bata ng mesa
AV.NVOL.PFV-broken NOM child GEN table 
‘The child {was able to/happened to} break a table’ (Henrison Hsieh, p.c.)
b. naka-kita ako ng pating
AV.NVOL.PFV-see 1SG.NOM GEN shark
‘I saw a shark’ (Henrison Hsieh, Jed-Sam Pizarro Guevara, p.c.)
Because of these aspects, these Tagalog verbs are not quite a close parallel to the Äiwoo
UV-only class. The same can be said for other syntactic contexts in Tagalog that allow
for non-pivot Ā-extraction, such as gerunds, recent perfective constructions, and excla-
mative adjective forms (Ceña 1979, McGinn 1988, Nakamura 1996, Schachter 1996, Hsieh
2020, 2025, a.o.). All of these are reduced forms in some sense and have been analyzed as
“voiceless” (concretely, for Hsieh 2025 they lack certain inflectional layers). This is not
true for Äiwoo UV-only verbs, which are not voiceless in any sense.
Two other relevant cases discussed in the literature are those of extraction in certain
comparative (273) and comitative (274) constructions in Tagalog (Ceña 1979, Nakamura
1996; the glosses throughout these examples are adapted from Hsieh 2020).
17 The verb kita ‘see’ obligatorily requires the nonvolitional/abilitative forms, also in non-AV contexts (i).
(i) na-kita=ko ang pating
UV.NVOL.PFV-see=1SG.GEN NOM shark
‘I saw the shark’ (Henrison Hsieh, p.c.)
Chapter 4. The Ā-extraction restriction, and the logic of an exception 179
(273) Comparative of equality:
a. Baseline (Ceña 1979: 123, Hsieh 2020: 279):
kasing-tangkad ng doktor ang anak
as.ADJ.as-tall GEN doctor NOM child
‘The child is as tall as the doctor’
b. Extraction of the non-pivot object of comparison: (Hsieh 2020: 279)
na-dismaya ang doktor na [kasing-tangkad ang bata]
PFV-disappoint NOM doctor LK as.ADJ.as-tall NOM child
‘The doctor whom the child is as tall as was disappointed’
(274) Comitative construction:
a. Baseline (Ceña 1979: 122):
ka-sama ng doktor ang anak
COM-with GEN doctor NOM child
‘The child is with the doctor’
b. Extraction of the non-pivot comitative argument (Ceña 1979: 122):
na-dismaya ang doktor na [ka-sama ang anak]
PFV-disappoint NOM doctor LK COM-with NOM child
‘The doctor who is with the child was disappointed’
Nakamura’s (1996) economy-based account of the general Tagalog extraction restriction is
based on minimizing the length of Ā-dependencies: simplifying coarsely, Ā-movement of
a constituent is forbidden when there exists an alternative derivation in which the same
movement chain would be shorter. For example, the reason why objects cannot be Ā-
extracted from AV clauses is that there is an alternative derivation (namely, that resulting
in a UV clause) where the Ā-movement chain would be shorter than in the AV case. For
Nakamura, then, these cases of non-pivot extraction (273)–(274) are conceptually similar
to what I argue for UV-only verbs in Äiwoo (though the implementation and technical
apparatus is, of course, rather different). For comparative and comitative constructions
there just is no such alternative/more economical derivation in the language, and there-
fore, Ā-extraction of a non-pivot is exceptionally allowed. The logic is the same as we
see in Äiwoo: non-VIA extraction is allowed precisely in those contexts where the voice
alternation is unavailable.
4.5 EXCEPTIONAL NON-VIA EXTRACTION
4.5.1 UV-ONLY VERBS ALLOW SUBJECT EXTRACTION
First, let us establish the empirical pattern. We can compare an ordinary transitive verb,
tu/tou ‘bring.UV/AV’, to a UV-only verb, te ‘see’. In all four sentences below (275)–(276) I
include an adverbial modifier, ngege ‘immediately, straight away’, so that we can easily
verify that the verb inside the RCs is in UV due to the voice concord suffix -nyi. For
180 4.5. Exceptional non-VIA extraction
ordinary transitive verbs, as we know (§4.3), from UV only the object can be extracted
(275a), and not the subject (275b). To do this, we need AV instead (275c).
(275) An ordinary transitive verb: ‘bring’ (250728)a. 3 O [UV:
kuli (mi=)[RC ku-tu-ngege-nyi-kä pesingedâ=kaa]
dog REL= IPFV-bring.UV-immediate-UV-DIR3 girls=FUT
‘A/The dog that the girls will bring straight away’
b. 7 S[UV:
* pesingedâ (mi=)[RC kuli ku-tu-ngege-nyi-kä -i =laa]
girls REL= dog IPFV-bring.UV-immediate-UV-DIR3-3AUG=FUT
Intended: ‘(The) girls who will bring the dog straight away’c. 3 S [AV:
pesingedâ (mi=)[RC ku-lu-tou-ngege-kä=naa kuli]
girls REL= IPFV-3AUG-bring.AV-immediate-DIR3=FUT dog
‘(The) girls who will bring the dog straight away’
If we switch the verb from ‘bring’ to the UV-only verb ‘see’, however, we see a clear differ-
ence. Unsurprisingly, extracting the object is still perfectly fine (276a). Crucially, however,
extracting the subject is also grammatical (276b), sharply contrasting with what is the case
for ordinary verbs (275b). In this case, we can also clearly see that we are extracting an
argument across a higher one, since the RC-internal direct object kuli ‘dog’ is – unexcep-
tionally – sitting in spec,CP, where it ordinarily belongs. Because ‘see’ is a UV-only verb,
a hypothetical AV form is impossible, even under subject extraction (276c).
(276) A UV-only verb: ‘see’ (250728)a. 3 O [UV:
kuli (mi=)[RC ki-te-ngege-nyi-kä pesingedâ=kaa]
dog REL= IPFV-see.UV-immediate-UV-DIR3 girls=FUT
‘A/The dog that the girls will see straight away’b. 3 S [UV: possible!
pesingedâ (mi=)[RC kuli ki-te-ngege-nyi-kä -i =laa]
girls REL= dog IPFV-see.UV-immediate-UV-DIR3-3AUG=FUT
‘(The) girls who will see the dog straight away’c. 7 S [AV: impossible, because the AV form of ‘see’ does not exist
* pesingedâ (mi=)[RC ki-li-TE-ngege-kä=naa kuli]
girls REL= IPFV-3AUG-SEE.AV-immediate-DIR3=FUT dog
Intended: ‘(The) girls who will see the dog straight away’
This behavior is not unique to te ‘see’, but extends to every UV-only verb. In (277)–(280) I
include some examples of subject extraction from a few others UV-only verbs; see §4.4.4
above for a complete list of the UV-only verbs we know of. In each pair of examples,
Chapter 4. The Ā-extraction restriction, and the logic of an exception 181
the a. sentence shows the grammatical subject relativization from a UV clause, and the
b. sentence shows the ungrammatical AV RC (which can be identified as AV from either
its word order, and/or prefixal φ-agreement, and/or lacking voice concord on modifiers).
For the ungrammatical AV ones, I only report the DP containing the RC, for conciseness.
(277) Kää ‘know’: (250728)a. 3 S [UV:
[isä devalili [RC ile lopwâ enge i-kää-päko-i -i ]]
women PROX story this ASP-know.UV-good-UV-3AUG
i-te-kâ-no=dä ijii
ASP-see.UV-DIR3-1MIN=some 3AUG
‘I’ve met some women who know this story well’b. 7 S [AV:
* [isä devalili [RC i-li-KÄÄ-päko ile lopwâ enge]]
women ASP-3AUG-KNOW.AV-good PROX story this
Intended: ‘Women who know this story well’
(278) Wagu ‘tell’: (250728)a. 3 S [UV:
[sime [RC lopwâ eângâ i-wagu-mä -i =lâ li-pe=Lende
person story that ASP-tell.UV-DIR1-3AUG=DIST 3AUG-COLL=S.Cruz
‘The people who told me that story were from Santa Cruz’b. 7 S [AV:
* [sime [RC i-lu-WAGU-mä lopwâ eângâ]]=kâ
person ASP-3AUG-TELL.AV-DIR1 story that=DIST
Intended: ‘(The) people who told me that story’
(279) Nyida ‘love’: (250728)a. 3 S [UV:
[gilaki mi=[RC isä-∅n ba ki-nyida-mana-i -∅n =gu]]=nâ
boy REL= mother-3MIN NEG IPFV-love.UV-very-UV-3MIN=NEG=DIST
mola nogo-∅n ea
manner POSS-3MIN bad
‘A boy who doesn’t love his mother very much has bad manners’ (lit. ‘A boy
who doesn’t love his mother very much, his manners are bad’)b. 7 S [AV:
* [gilaki mi=[RC ba ki-NYIDA-mana=gu isä-∅n]]=nâ
boy REL= NEG IPFV-LOVE.AV-very=NEG mother-3MIN=DIST
Intended: ‘A boy who doesn’t love his mother very much’
182 4.5. Exceptional non-VIA extraction
(280) Mâea-ive ‘laugh-APPL’ = ‘laugh at’: (250728)a. 3 S [UV:
[pedevalili [RC Jon ku-mâea-ive-mana-i -i ]]=lâ liluwaio
children John IPFV-laugh-APPL-very-UV-3AUG=DIST small.PL
‘The children who were laughing at John a lot were very small’
b. 7 S[AV:
* [pedevalili mi=[RC ku-lu-mâea-ive-mana Jon]]=kâ
children REL= IPFV-3AUG-laugh-APPL-very John=DIST
Intended: ‘The children who were laughing at John a lot’
4.5.2 ANALYSIS: SELECTING A SPECIAL v HEAD
Before we start discussing the analysis of this exceptional extraction pattern, I want to
immediately rule out one particular potential line of reasoning, due to a fundamental
conceptual issue. One thing we cannot say is something along the lines of “UV-only verbs
can only ever take objects with a [PIV] feature”. The reason why this is conceptually trou-
bling is that [PIV] is, by assumption, an Ā-feature, with a semantic/pragmatic contribution
connected to discourse and information structure. Ā-features are syntactically free, that
is, they are distributed purely according to the human free will of the speaker. Simply put,
whether a speaker decides to put [PIV] on the subject or on the object depends on what
they want their sentence “to be about” (for some as yet unclear definition of aboutness).
Stating a restriction of that sort would be equivalent to imagining, for example, that En-
glish might have certain lexically specific verbs that can only ever take object focus, and
never subject focus. This, of course, is nonsensical: whether a speaker wants to put focus
on the subject or the object should be a choice entirely in their hands, and not dictated by
the particular verb they decide to build their clause with18.
With this out of the way, let us now see how an analysis of UV-only verbs can be
developed. Our reasoning should go in the opposite direction, so to speak. It is not that
these particular lexical roots force a particular information structure. Rather, these par-
ticular lexical roots do not react to the information structure of the clause they are in like
ordinary Äiwoo verbs do, and therefore do not show a voice alternation.
18 It is true that natural language syntax does in fact put restrictions on what set of meanings a speaker may be
able to express; consider e.g. island violations. To offer an example that I actually witnessed happening, an
English-speaking physical therapist will be syntactically in trouble if they want to ask their patient ‘what
is the thing 𝑥 such that if you move 𝑥 it hurts’ (*whati does it hurt if you move i?) – a perfectly sensible
question given their profession and the conversation context. However, this is a ban on a certain type of
syntactic configuration, and not a ban on the combination of a particular lexical verb and a particular choice
of information structure/discourse structure. I assume natural languages do not/should not feature any such
restrictions. I thank David Pesetsky for raising this point.
Chapter 4. The Ā-extraction restriction, and the logic of an exception 183
4.5.2.1 WALKING BACKWARDS: HOW THE STORY MUST GO
To understand how and why UV-only verbs are special, let us retrace our steps and con-
sider what we said for ordinary voice-alternating verbs. For those, we said that the ex-
traction restriction is about Ā-intervention. Let us take examine a concrete case, that we
will then later compare to how our UV-only verbs must work.
According to the analysis developed in §4.3, in a UV clause, the subject (in spec,TP)
cannot be relativized because the higher object, in spec,CP, carries an Ā-feature [PIV] (281).
This feature satisfies the relativizing probe, which will therefore halt, and it will never be
able to reach the subject further down and move it.
(281) [PIV] on the object Ā-intervenes for extracting the subject:
RelP
7
[ Rel ] CP
I: [REL]M
S: [Ā]
SAT!
Object
[PIV] C TP
7!
7 Subject …
[REL]
Let us consider the causal chain of events that got us there. According to our analysis,
the object is high, in spec,CP, because C raised it. C raised the object because v agreed
with it earlier on. In turn, v agreed with it because it had an [Ā] feature [PIV], which will
ultimately cause an Ā-intervention problem. In other words, for all ordinary transitive
verbs, “being the pivot” (that is, carrying the [PIV] feature) is coextensional with “being
the VIA” (that is, the argument reflected by the voice morphology).
Let us now reconsider the important contrast under discussion in light of the Ā-
intervention approach. The object of an ordinary verb Ā-intervenes for extracting the
subject; somehow, the object of a UV-only verb does not. Other than this extraction pat-
tern, UV-only verbs and ordinary verbs in UV are completely identical. Therefore, we
want our derivation for UV clauses to look the same in both cases, with the object being
in spec,CP and the subject in spec,TP. Of course, Ā-intervention is caused by Ā-features:
namely, the [PIV] feature on the object in spec,CP. If that is the source of the intervention
effect, this must mean that only for UV-only verbs, somehow, the object gets raised to
spec,CP even without having an Ā-feature. That is, our aim should be to make sure the
derivation reaches the following result:
184 4.5. Exceptional non-VIA extraction
(282) A [PIV]-less object does not Ā-intervene for extracting the subject:
RelP
Subject
[ Rel ] CP
I: [REL]M
S: [Ā]
NO SAT!
Object
[∅] C TP
3! 3! Subject
[REL] …
Here, the object is still high in spec,CP but has no [PIV] feature – which is why the subject
is extractable. At this juncture, note that reaching this state of affairs is only made possible
by the otherwise seemingly unmotivated move we made in chapter 3 in positing that the
probe on C searches for [i |v], and not anything connected to Ā-features. For C to raise
the object, it must mean that v agreed with it. In our analysis, object agreement by v is
what causes verb stems to be spelled out with UV morphology. If these verbs do not have
an AV form, it means that in these cases, v must always agree with the object, no matter
what. Now, our goal is to understand why for these verbs, v will always agree with the
object even when it does not have [PIV].
Before continuing, I want to note that this idea of UV-only verbs somehow always
involving object agreement is intuitively appealing at a conceptual level. This point was
briefly made in passing in the introduction to this section, but we are now in a position to
appreciate it better. Ordinary voice-alternating verbs, in a sense, “care” about Ā-features:
depending on where the speakers chose to put these features, the verb will take one of
two different forms. UV-only verbs have only one form, and thus, if Ā-features are still
distributed freely (as they ought to be), this means that these verbs “don’t care” about
Ā-features. Regardless of what information structure the speaker wants to convey with
their sentence, these verbs can only ever be UV. These verbs show a “fake UV”, so to
speak: even if the speaker chose to put Ā-features ([PIV]) on the subject (or nowhere, for
that matter), these verbs will still look like UV ones – which, by hypothesis, means that
v still somehow agreed with the object despite it lacking [PIV]. Schematically, this can be
illustrated as follows:
Chapter 4. The Ā-extraction restriction, and the logic of an exception 185
(283) Ordinary verbs: voice alternation depends on where Ā-features are
Does the object have Ā-features?
Yes No
Object agreement No object agreement
UV morphology AV morphology (default)
Ā-marked object Subject moves
moves to spec,CP to spec,CP
Intervention effect: If the object had [REL],7 S [UV we would’ve had UV:7O [AV
(284) UV-only verbs: no voice alternation = Ā-features don’t matter
Does the object have Ā-features?
Yes No
Object agreement Somehow, still object
agreement nonetheless!
UV morphology UV morphology
Ā-marked object Non-Ā-marked object
moves to spec,CP moves to spec,CP
Intervention effect No intervention effect:7 S [UV 3 S [UV!
186 4.5. Exceptional non-VIA extraction
4.5.2.2 FORCING UV/OBJECT AGREEMENT WITH SELECTION
What we did in the previous section is, figuratively, to move the target of our question a
few nodes down in the tree: from “Why is this object in spec,CP if it doesn’t have [PIV]?”
to “Why do these verbs always agree with their object, even when it doesn’t have [PIV]?”.
We know that for ordinary, voice-alternating transitive verbs, v searches for [Ā], and
agrees with objects that carry any Ā-feature(s). We now need a way to make sure that for
UV-only verbs, v will agree with any object instead.
Here is how we do it. First, remember that in chapter 3, §3.6, we posited that intran-
sitives verb roots select for a special vINTR head, which – unlike the ordinary v head –
carries no probe at all (285a,b). As a result, intransitive verbs always surface with AV-
style morphology, φ-agreement, etc. Now, remember that UV-only verbs are, as far as we
can tell, a lexically idiosyncratic/arbitrary class: whether a verb is UV-only or not is only
dependent on the specific lexical root (§4.4). There is another syntactic phenomenon that
is notoriously arbitrary: selection. I propose that this specific set of verb roots is singled
out by selecting for a special type of v head, which I label vH (‘little v-star’), whose probe
has a different, less “picky” featural specification than the ordinary v head (285c)19.
(285) Inventory of v heads in Äiwoo:
a. v: I: [Ā][S: [D]] (selected by ordinary transitive verb roots)
b. vINTR: no probe (selected by intransitive verb roots)
c. vH: I: [φ][S: [D]] (selected by UV-only verb roots)
The interaction condition of vH is set to [φ], that is, a feature that absolutely all nominals
have: therefore, vH will always agree with its object, whether or not it has Ā-features. This
gives us the UV-only profile: because object agreement always takes place, AV is simply
impossible. In this respect, the syntactic behavior UV-only verbs could be similar to the
sketch idea proposed for “High Absolutive” Mayan languages in chapter 3, §3.7.3. There,
I suggested that the general syntactic inversion of the object across the subject found
in these languages could be derived by positing a v head that always agrees with any
object, instead of only the Ā-marked ones, then followed by a step of probing for [i |v] by
a higher head (like C does in Äiwoo). In other words, High Absolutive Mayan languages
would only have an equivalent of vH, whereas Äiwoo has both an Ā-seeking v and the
more general, all-object-agreeing vH.
19 Why the satisfaction condition of vH is set to [D], like for ordinary v, will become clear in chapter 6. Note
that there is no implied connection to Chomsky’s (1995) ‘v*’ notation, which would rather correspond do
my diacritic-less v here.
Chapter 4. The Ā-extraction restriction, and the logic of an exception 187
4.5.2.3 DERIVING THE LACK OF AN EXTRACTION RESTRICTION
In the last section we have established what exactly is special about UV-only verbs: they
select for vH, a head which always agrees with its object. This hypothesis correctly pre-
dicts that these verbs only have UV, and will never have an AV form.We will now see how
this also correctly predicts that clauses built around UV-only verbs (henceforth, UVonly
clauses) exceptionally allow for subject extraction, unlike other UV clauses (“UVplain”).
Let us consider the case of a UVonly clause where the speaker has decided not to put
[PIV] on the object (286), and wants to relativize the subject, which therefore carries [REL].
The verb ‘see’ selects for vH, which has [INT: φ], instead of the ordinary v with [INT: Ā].
Therefore, vH will interact with the object even when this does not have Ā-features, flag
it, and halt. In other words, whenever the verb ‘see’ is merged, because of its selectional
requirement there is no way to avoid object agreement, so this verb will always be in
UV. (I assume that the interaction flag assigned by vH is identical to the one assigned by
ordinary v, or at least that the two are similar enough that the probe on C can target
either.)
(286) ‘See’ selects for vH, and will always have object agreement:
vHP ⟨φO⟩ ⇔ UV
DP
[REL] [ vH ⟨]φO⟩ VP
I: [φ]
S: [D]
V DP
√ [i|v]
SEE
[•vH•]
Later, as usual, the subject will move to spec,TP. Finally, Cwill scan its c-command domain
for the closest goal bearing [i |v], i.e., the object, and raise it.
188 4.5. Exceptional non-VIA extraction
(287) With vH, an Ā-less object may end up being raised in spec,CP:
CP
DP[i|v]
[ C ] TP
I+S: [i|v]M
DP
[REL] T vHP
tDP … tDP[i|v]
We have now achieved the situation we wanted to derive (§4.5.2.1; example (282)). Now,
we have an object in spec,CP with no Ā-features, ultimately caused by the selectional
requirement of this specific verb. Therefore, we predict no extraction restriction, because
there is no Ā-intervention effect. Because the object has no Ā-features, the relativization
probe will simply skip it, and it will successfully extract the subject (288)20.
(288) No intervention effect/extraction restriction from UVonly clauses:
RelP
DP
[REL]
[ Rel ] CP
I: [Rel]M
S: [Ā]
DP[i|v]
C TP
3
t
3 DP
…
A potential cross-linguistic parallel to UV-only verbs can be found once we think of UV
as object agreement, as I’m doing here (see chapter 3, §3.5.1.2). In Bulgarian, object clitic
20 The derivation outlined here is one out of two possible ones that could happen for a UVonly clause. If we had
put the [PIV] feature on the object, the clause up until the CP level would look identical, but there would
be an Ā-intervention effect and thus, the subject would not be extractable. What I am claiming, therefore,
is not that UVonly clauses always allow for subject extraction. Rather, the idea is that for any surface string
reflecting a UVonly clause, there exists at least one possible derivation that will ultimately allow for subject
extraction.
Chapter 4. The Ā-extraction restriction, and the logic of an exception 189
doubling – which bears likes to object agreement – depends on a complex variety of fac-
tors such as the object’s definiteness, specificity, topicality, or a combination thereof (see
Harizanov 2014: 1037 and the references cited there). However, there also exist classes
of predicates that always, obligatorily require object clitic doubling, regardless of the ob-
ject’s properties (Krapova & Cinque 2008). These are more coherent natural classes than
the Äiwoo UV-only verbs (e.g. psych- and perception predicates, certain modal predicates,
etc.; see Krapova & Cinque 2008: 266-267 for a list)21. Syntactically, however, the paral-
lel holds: ordinary verbs will agree with their object only if it presents certain semantic/
pragmatic properties, but a particular set of verbs will always agree with their object no
matter what.
4.6 SUMMARY
In this chapter I have started addressing the Extraction Puzzle, temporarily limiting the
scope to canonical clauses and UVonly-clauses. In canonical AV and UVplain clauses, we
observe a classic Austronesian Ā-extraction restriction: only the VIA can be extracted
(the subject in AV, the object in UV). In UVonly-clauses, however, this restriction vanishes,
and non-VIA subject extraction from UV becomes possible.
I have argued that the Ā-extraction restriction in Äiwoo has diverse, non-uniform
causes (289). First, object extraction is impossible from AV (289a) because if the object is
to be extracted it means that it carries an Ā-feature ([REL]), and therefore v will agree
with it; in this case, AV just cannot arise in the first place. Then, I have argued that sub-
ject extraction from UV is impossible due to Ā-intervention (289b), and not phasehood
or DP-intervention, contra other analyses of similar patterns in Austronesian (e.g. Rack-
owski & Richards 2005, Hsieh 2020, 2025, Erlewine & Sommerlot 2023b for phase-based
approaches, and Aldridge 2004, 2008, Erlewine & C. Lim 2023 for DP-intervention ap-
proaches). TheVIA in canonical UV clauses is raised to spec,CP by (non-local) A-movement,
but it also carries the Ā-feature [PIV]; in other words, in these clauses, the VIA and the
pivot coincide. This creates an Ā-intervention effect, making it impossible to Ā-extract a
lower argument across the VIA.
(289) Subcases of the Ā-extraction restriction in Äiwoo:
Cause:
a. *O[AV: Underivableb. *S [UV: Ā-intervention
In the case of UV-only verbs, I have argued that the VIA does not necessarily coincide with
the pivot anymore. The feature [PIV] may well be on the subject (or nowhere), but the VIA
21 In addition to Bulgarian, similar classes of predicates also trigger obligatory clitic doubling in Albanian
(kallulli2000), Amharic (Kramer 2014), Greek (Anagnostopoulou 2003), Macedonian (Krapova & Cinque
2008), and Romanian (dobrovie-sorin1994).
190 4.A. A note about mi= and other “relativizers”
– the argument that v agrees with – will still be the object. The reason for this difference
from standard transitive clauses is that UV-only verb roots idiosyncratically select for
vH, which – unlike ordinary v – always agrees with its object no matter what. Even if a
speaker chooses not to put Ā-features on the object of such a verb, object agreement will
still take place, thus always giving UV. This way, the [i |v]-flagged object might end up
being raised to spec,CP even without having Ā-features. Therefore, there won’t be any
intervention effect, and Ā-extraction of the subject will be possible.
4.A A NOTE ABOUT mi= AND OTHER “RELATIVIZERS”
Some of the examples throughout this dissertation include an optional morpheme mi= at
the left edge of the RC, which I gloss ‘REL’. This marker is part of a small group of monosyl-
labic markers that Næss (2017b) labels “Generic Bound Nouns”. These aremorphologically
bound in the sense that they never appear on their own, but always in combination with
a host. I reproduce a list in (290)22,23. In this appendix, I argue that these are relativizers.
(290) Relativizers (elaborated from Næss 2017b: 919):
mi= ‘the one that …’
me= ‘the person who …’
de= ‘the thing that …’
pe= ‘the people who …’
nye/nyi= ‘the place where …; the way/manner in which …’
me= ‘the time when …’
4.A.1 DISTRIBUTION
All these markers are productively combined with predicates of any size to form relative
clauses: ‘the one that …’, ‘the person who…’, ‘the thing that…’, ‘the place where…’, etc. In
fact, Næss (2017b) raises the possibility of a relativizer analysis for these, although she does
not argue for it. Their distribution is determined syntactically, and not morpho(phono)-
logically: they attach at the left edge of a clausal constituent, namely, a RC. However, this
22 Næss (2017b) includes two more, gi- ‘man, male’ and si- ‘woman, female’. These are commonly found form-
ing parts of kinship terms (gite ‘man’s brother’ vs. site ‘woman’s sister’; gibo ‘nephew, grandson’ vs. sibo
‘niece, granddaughter’) and other human-referring terms (gilaki ‘boy’ vs. silaki ‘girl’, where gi- and si- are
possibly combined with laki ‘small’). In addition, they can be found combined with the demonstratives enge
(proximal) and eângâ (distal): ginenge/ginângâ ‘this/that man’, siwenge/siwângâ ‘this/that woman’. How-
ever, at least some of these forms may be synchronically unsegmentable/lexicalized, as these alternations
don’t seem to be productive (cf. gino ‘son’ but sipe ‘daughter’, rather than *sino). As these don’t seem to be
productively used to form relative clauses, I will exclude them from the discussion, and I will focus on the
ones that can clearly productively combine with verbs/predicates (290).
23 The relativizer me= ‘the time when’ is a novel discovery. It is completely homophonous with me= ‘the
person who’, though this is most likely just a coincidence.
Chapter 4. The Ā-extraction restriction, and the logic of an exception 191
might not be immediately evident. Most often, in natural speech data we will see these
relativizers combined with rather “small” hosts, typically a single phonological word:
(291) Examples from Næss (2017: 923):
a. mi=olo
 REL=big
‘A/the big one’
b. de=laki
REL:thing=small
‘A/the small thing; something small’
c. nye=maapi
REL:place=overgrown
‘Overgrown place, bush’
Common formations like these may have prompted the analysis of these morphemes in
Næss (2017b) as bound prefixes. However, these “small” forms are misleading with respect
to the relativizers’ actual combinatory possibilities. In fact, the constituent they can attach
to is clearly clause-sized, as it can include several DP arguments, TAM marking, temporal
adverbials, clausal adjuncts, and even nested embedded clauses (292d)–(293d).
(292) Attested examples (Næss 2017b: 926):
a. me=[wasele nyepolââ enge]
REL:person=make.AV world this
‘The one who made this world’
b. nye=[ki-vaabe ipebo-i-le=nä nänyi=kâ]
REL:place=IPFV-beat grandmother-3AUG-UA=CV coconut.fiber=DIST
‘The place where their grandmother was beating coconut fiber’
c. de=[wâte-wâ ngâgu-mu]
REL:thing=happen-DIR2 to-2MIN
‘The thing that happened to you’
d. me=[kä-i=lä [ku-wopo-lâ-mä go King Deved]]
REL:person=say-3AUG=CV IPFV-go-in-DIR1 PREP King David
‘The person who they say descends from King David’ (MT 12:23)
(293) Elicited examples:
a. me=[ku-tu-usi-mä pedevalili=kaa] (231109)
REL:person=IPFV-bring.UV-back.UV-DIR1 children=FUT
‘The person that the children will bring back’
b. i-wagu-mä-i ngâgu de=[wanyinyi-i bugulo] (240208)
ASP-tell.UV-DIR1-3AUG to.1MIN REL:thing=do.UV-3AUG yesterday
‘They told me what they did yesterday’
c. i-evä-kâ-no go Jen nye=[wä site-∅n=nä] (231026)
ASP-ask.UV-DIR3-1MIN to Jane REL:place=go sister-3MIN=CV
‘I asked Jane where her sister was’
192 4.A. A note about mi= and other “relativizers”
d. bââ de=[kä Mary=kä [nä-ve-∅n]] (240418)
not.exist REL:thing=want Mary=CV IRR-buy.UV-3MIN
‘There’s nothing that Mary wants to buy’
Even considering rather rich examples like these, it could seem like these morphemes
consistently attach to verbs/predicates, which is why Næss (2017b) (et seq.) glosses them
as (verbal) affixes. However, this is purely a consequence of the syntax of relativization
in Äiwoo. Since the RC-internal gap is most often sentence-initial, the first overt mor-
phophonological word of the RC will indeed be the verb. Instead, we can make sure that
these morphemes are attached not to a morphomorphological word but to a syntactic con-
stituent, that is, an entire RC24. To confirm that they are not restricted to attaching to the
verb, one can for example add negation inside the RC, and the relativizers will consistently
show up to the left of the pre-verbal negative particle ba:
(294) a. sime mi=[ba ku-wâ-une=gu]
person REL=NEG IPFV-CAUS-true=NEG
‘Person/people who don’t believe (in it)’ (Roversi & Næss 2019: 335)
b. singedâ mi=[ba i-lâwâle-mole-nyi-mä-gu-∅ŋ=gu] (240502)
girl REL=NEG ASP-help.UV-exactly-UV-DIR1-GU-1MIN=NEG
‘A/the girl who didn’t help me at all’
Even stronger evidence comes from those contexts where one can do non-VIA relativiza-
tion (discussed in §4.5.1). Here, the extracted DP it not sentence-initial. In these cases, we
can clearly see the relativizer morphemes attaching to the first phonological word at the
left edge of the RC, whatever that may be, and not to the verb:
(295) pesingedâ mi=[kuli no-mu i-te-usi-kä-i] (240328)
girls REL=dog POSS-2MIN ASP-see.UV-again.UV-DIR3-3AUG
‘(The) girls who saw your dog again’
4.A.2 (NON-)COMPLEMENTARITY WITH A HEAD NOUN
My labelling these morphemes ‘relativizers’ is meant to remain somewhat agnostic as to
their precise syntactic nature. A couple potential analyses are possible. They could be
Ā-complementizers (C-heads), with some semantic restrictions on the head noun they
may modify. Alternatively, they could be phonologically reduced nominal elements that
constitute the actual head of the RC. Furthermore, in principle we should not have any
expectations that all six of these are of the exact same nature. In fact, there seems to be
a distinction between mi= and the other more semantically contentful ones. The general
relativizer mi= routinely co-occurs with an overt head noun for the RC, as we can see in
(294)-(295) above and in many other examples throughout. This perhaps suggests an anal-
ysis of mi= as a pure Ā-complementizer. However, the other relativizers seem to be more
24 This is why I depart from Næss’s notation and gloss them as morphophonological clitics rather than affixes,
with the symbol ‘=’ rather than ‘-’.
Chapter 4. The Ā-extraction restriction, and the logic of an exception 193
or less in complementary distribution with an overt head noun25. Two sets of examples
are shown below for nye= ‘the place where…’ and de= ‘the thing that…’. In the (a) sen-
tences we see the relativizers combining with the RC, and no overt head noun. In the (b)
sentences, adding an overt head nounmakes the structure unacceptable. The (c) sentences
confirm that the head nouns themselves are not a problem, as long as the relativizers are
absent26.
(296) Nye= ‘the place where…’ is incompatible with an overt head noun: (240404)
a. [DP nye=[wâ-no=ngä ngä skul]RC]=kâ nyopu-mana
REL:place=go-1MIN=CV to school=DIST far-very
‘The place where I went to school is very far’
b. * [DP taun nye=[wâ-no=ngä ngä skul]RC]=kâ nyopu-mana
town REL:place=go-1MIN=CV to school=DIST far-very
Intended: ‘The town where I went to school is very far’
c. [DP taun [i-wâ-no=ngä ngä skul]RC]=kâ nyopu-mana
town ASP-go-1MIN=CV to school=DIST far-very
‘The town where I went to school is very far’
(297) De= ‘the thing that…’ is incompatible with an overt head noun: (240404)
a. [DP de=[ii-no=ngä butete]RC]=kâ ba mebädo=gu
REL:thing=peel.UV-1MIN=CV potato=DIST NEG sharp=NEG
‘The thing that I peeled the potatoes with was not sharp’
b. * [DP nuwoli de=[ii-no=ngä butete]RC]=kâ ba mebädo=gu
knife REL:thing=peel.UV-1MIN=CV potato=DIST NEG sharp=NEG
Intended: ‘The knife that I peeled the potatoes with was not sharp’
c. [DP nuwoli [i-ii-no=ngä butete]RC]=kâ ba mebädo=gu
knife ASP-peel.UV-1MIN=CV potato=DIST NEG sharp=NEG
‘The knife that I peeled the potatoes with was not sharp’
Contrary to e.g. de= ‘thing’ and nye= ‘place’, as established above mi= does in fact co-
occur with an overt head noun. This could be analyzed as an appositive construction
(Næss, p.c.): mi= is a semantically bleached nominal element that is the “real” head of the
RC (‘the one that…’), and stands in apposition to the more semantically contentful noun
to its left. A more literal translation of (295) would then be “(the) girls, the ones who saw
25 Rare cases of co-occurrence are sporadically attested (i). However, combinations like this are systematically
rejected in elicitation contexts. I don’t have an explanation for structures like this; perhaps, this could be
a case of a more syntactically loose apposition (‘our law, the one that Moses gave us’). Further research is
needed in this respect. However, for the cases wheremi= co-occurs with an overt relative head noun I argue
against an apposition analysis; see below.
(i) ile ngâ [molâ nugu-de de=[la-mä Moses]]=ke
here in law POSS-12AUG REL:thing=give.UV-DIR1 Moses=PROX
‘In our law that Moses gave us’ (MT 22:17)
26 The aspectual prefix i- systematically disappears when it’s not word-initial (chapter 2, §2.4.3). This is why
it’s visible in the (c) sentences but not in the (a,b) sentences.
194 4.A. A note about mi= and other “relativizers”
your dog again”. However, I believe we can show that this cannot be the correct analysis.
A characteristic trait of appositive RCs is that they can modify definite descriptions (298a),
but not quantified expressions (298b–d) (J. Ross 1967 et seq.).
(298) Appositive RCs cannot modify a quantified expression (Bhatt 2005a: 4):
a. She/The president/Anne, who was tired, …
b. * Susan interviewed every senator, who is crooked
c. * The doctor gave a lollipop to each child, who she examined
d. * No person, who knows everything, is perfect
Contrary to this, a mi=-marked relative clause in Äiwoo can be headed by a quantified
expression (299). This speaks against an analysis of mi=RCs as appositive, even when mi=
co-occurs with an overt noun.
(299) i-eeu-kä go [pesigiläi-du mi=[isä-i i-lâwâle-kä-i]RC]
ASP-speak-DIR3 to boys-all REL=mother-3AUG ASP-help.UV-DIR3-3AUG
‘I talked to every boyi who helped hisi mother’ (240721)
(more literally: ‘I talked to all the boysi who helped theiri mother’)
4.A.3 OTHER PROPERTIES
4.A.3.1 OPTIONALITY
As mentioned earlier, these relativizers are largely optional, and in most cases sentences
receive the same judgment by native speakers both with and without an overt relativizer.
One factor that seems to influence this judgment is that including an overt relativizer
disambiguates a string that would otherwise be syntactically ambiguous as containing
a RC, thereby preventing garden-path effects that may arise from not having any overt
marking of the RC boundary. For example, (300) (repeated from above) was judged to be
better with mi= than without, perhaps due to the fact that without mi= one ends up with
two DPs in a row, something that might be hard to parse in an out-of-the-blue elicitation
context.
(300) pesingedâ mi=[kuli no-mu i-te-usi-kä -i ] (240328)
girls REL=dog POSS-2MIN ASP-see.UV-again.UV-DIR3-3AUG
‘(The) girls who saw your dog again’
However, this seems to be a performance/processing type of constraint, and not a hard
and fast syntactic rule. In other cases, strings with two DPs in a row and no mi= were
judged to be acceptable (301).
(301) a. [sime [RC lopwâ eângâ i-wagu-mä -i ]]=lâ li-pe=Lende
person story that ASP-tell.UV-DIR1-3AUG=DIST 3AUG-REL:people=S.Cruz
‘The people who told me that story were from Santa Cruz’ (240208)
Chapter 4. The Ā-extraction restriction, and the logic of an exception 195
b. [mikilitei [RC nubââ i-eâ-i-to-mä -i ]]=lâ li-ngâgo-mana
fishermen shark ASP-paddle-COM-go.in-DIR1-3AUG=DIST 3AUG-strong-very
‘The fishermen who paddled back with the shark were very strong’ (230615)
c. [pedevalili [RC John ku-mâea-ive-mana-i -i ]]=lâ liluwaio
children John IPFV-laugh-APPL-very-UV-3AUG=DIST small.PL
‘The children who were laughing at John a lot were very small’ (250728)
Furthermore, mi= seems to be subject to haplology. To observe this, first note that many
common nouns (especially human-denoting ones) are morphologically complex, consist-
ing of (null-headed) RCs marked by mi= or one of the other relativizers27:
(302) a. mekivaavee ‘Teacher’ (303) a. mekibei ‘Patient, sick person’
me=ki-vaavee me=ki-bei
REL:person=IPFV-show REL:person=IPFV-sick
Lit. ‘person who shows’ Lit. ‘Person who is sick’
b. mikilivaavee ‘Teachers’ b. mikilibei ‘Patients, sick people’
mi=ki-li-vaavee mi=ki-li-bei
REL=IPFV-3AUG-show REL=IPFV-3AUG-sick
Lit. ‘Those who show’ Lit. ‘Those who are sick’
When a RC starts with a noun like mikilibei ‘patients’, mi= is avoided, probably because
this would result in a sequence of two mi=’s in a row (304). To further support the idea
that this is a haplology effect, consider also the minimal pair in (305). In (305a), where
mi= would attach to the nominal mekipä ‘thief, person who steals’, mi= is dispreferred –
presumably due to haplology with the immediately adjacent relativizer me=. When there
is no haplology issue, because mi= attaches to a proper noun (305b), mi= is then perfectly
acceptable (and in fact judged to be slightly better).
(304) dokta (?*mi=)[mi=ki-li-bei ku-wâpäkoeâ-gu-i=laa ] (240912)
doctor REL=REL=IPFV-3AUG-sick IPFV-cure.UV-GU-3AUG=FUT
‘The doctor who will cure the patients’
Comment: “mi-mikilibei doesn’t sound good, you’re repeating mi-”
(305) a. pesingedâ (??mi=)[me=ki-pä i-te-kä -i ] (240404)
girls REL=REL:person=IPFV-steal.AV ASP-see.UV-DIR3-3AUG
‘The girls who saw the thief’ (mi= is dispreferred)
b. pesingedâ ?(mi=)[Joseph i-te-kä -i ]
girls REL=Joseph ASP-see.UV-DIR3-3AUG
‘The girls who saw Joseph’ (mi= is preferred)
27 There is a strong tendency in my data for me= to show up in “singular” forms like (302a)-(303a), and mi= in
“plural” forms like (302b)-(303b) (the semantics of nominal number in Äiwoo are complex and poorly un-
derstood, hence the hedging; see chapter 2, §2.5.1). I don’t have an explanation for this. However, exceptions
exist in both directions, which may suggest it probably isn’t a hard syntactic fact.
196 4.A. A note about mi= and other “relativizers”
This haplology phenomenon also shows us that despite their frequency, nouns formed
with mi= or other relativizers are synchronically complex, and their internal structure is
accessible to and manipulable by speakers. Another piece of evidence for this comes from
the standard expression meaning ‘food’, which is again a relativized form (306). When
speakers wants to add a possessive to this, they will either treat the entire thing like a
normal noun and just add the possessive classifier appropriate for food (307a), or alterna-
tively, another option is to insert a 1MIN subject inside the RC, thus altering the form of
the verb that de= attaches to (307b).
(306) de=ki-li-ngä
REL:thing=IPFV-GEN.A-eat.UV
‘Food’; lit. ‘Thing(s) that one eats’
(307) “Possessed” forms (Næss 2017b: 925):
a. dekilingä nugo b. de=ki-ngâ-no
food POSS.1MIN REL:thing=IPFV-eat.UV-1MIN
‘My food’ ‘My food’; lit. ‘Thing(s) that I eat’
Throughout the dissertation, to simplify the glosses for readability I will not segment the
internal structure of nominal forms like mikilivaavee ‘teachers’ or dekilingä ‘food’ (as in
(307a)) when not immediately relevant.
4.A.3.2 RESTRICTIVENESS
An additional property of mi= is that it’s incompatible with non-restrictive RCs. When
the head noun is unique, so that restrictive modification is impossible, adding mi= is
not acceptable (308). This fits well with the descriptive characterization of mi= in Næss
(2017: 920): “forms in mi[=] are used to identify entities by picking them out from a larger
set” (emphasis in the original).
(308) [Context: I have only one sister] (240502)
siwou (#mi=)[ku-mo Honiara]=kâ i-weevä-mä-gu-∅ŋ
sister.1MIN REL=IPFV-stay Honiara=DIST ASP-visit.UV-DIR1-GU-1MIN
‘My sister, who lives in Honiara, came to visit me’
Comment: “mi= doesn’t fit in this context where there’s only one sister, if I say
mikumo it sounds like I have more than one sister”
In several cases where mi= was judged as optional, the variant including mi= received
comments that seemed to point towards this restrictive interpretation:
(309) [nubole (mi=)[tu-mä Mary]] ki-epavi-no=ngaa bulaape
taro REL=bring.UV-DIR1 Mary IPFV-cook.UV-1MIN=FUT tomorrow
‘Tomorrow I will cook the taro that Mary brought me’
Comment: “If you say mitumä it implies that there is some taro available to you
that Mary didn’t bring, but that someone else might have brought” (240418)
Chapter 4. The Ā-extraction restriction, and the logic of an exception 197
4.B A NOTE ABOUT OTHER Ā-DEPENDENCIES
The reason why I chose relativization as the main Ā-dependency under discussion is that
it simply is the currently best understood one. Other types of classic Ā-dependencies in
Äiwoo are either not movement-based at all (topicalization), can be reduced to relativiza-
tion (focus/clefting), or are simply not well-understood (wh-question formation). In this
appendix I will briefly go through the first two. I hope future research will shed light on
the syntax of wh-questions in Äiwoo.
4.B.1 TOPICALIZATION IS NOT MOVEMENT-BASED
As mentioned (3, §3.2), left-peripheral topicalized DPs are rampant in Äiwoo, to the point
that in the natural speech corpus, sentences with multiple DPs in the dedicated argument
positions are in fact rather rare. Speakers typically prefer to topicalize their DPs, and leave
pronominal forms in the argument positions (e.g. Annai, that bookj, shei bought itj). These
pronouns are most often null when morphosyntactically possible (i.e., for 3MIN referents;
non-3MIN pronouns must be overt).
However, Äiwoo topicalization shows no signs of being movement-derived. There
seems to be no locality limits or restrictions on what nominals can be topicalized. For
example, topicalization show no voice-related restrictions: any argument can be topical-
ized from any voice. The sentences in (310) provide AV and UV baselines without any
topicalization. In (311), we observe that a subject topic is allowed from both AV and UV.
In (312), conversely, we observe that an object topic is also allowed from both AV and
UV28.
(310) Baselines: (240530)
a. bulaape pedevalili ki-li-kei-mä=kaa nubole (AV)
tomorrow children IPFV-3AUG-dig.AV-DIR1=FUT taro
‘Tomorrow the children will harvest taro for me’
b. bulaape nubole ki-kili-mä pedevalili=kaa (UV)
tomorrow taro IPFV-dig.UV-DIR1 children=FUT
‘Tomorrow the children will harvest the taro for me’
(311) Subject topic: 3 AV, 3 UV (240530)
[Context: “what are the children doing tomorrow?”]
a. pedevalili=kâ, bulaape ∅3AUG ki-li-kei-mä=kaa nubole
children=DIST tomorrow IPFV-3AUG-dig.AV-DIR1=FUT taro
‘(As for) the childreni, tomorrow theyi will harvest taro for me’
28 In fact, (312a) is somewhat surprising not because an object is being topicalized from an AV sentence, but
because the result is an AV sentence containing what could be described as a pronominal object. This is
usually not possible; see chapter 2, §2.5.3. It is perhaps possible that there is in fact no object in this sentence,
and that kei ‘dig’ is being used intransitively.
198 4.B. A note about other Ā-dependencies
b. pedevalili=kâ, bulaape nubole ki-kili-mä- i =laa
children=DIST tomorrow taro IPFV-dig.UV-DIR1-3AUG=FUT
‘(As for) the childreni, tomorrow theyi will harvest the taro for me’
(312) Object topic: 3 AV, 3 UV (240530)
[Context: “who is picking up the taro tomorrow?”]
a. nubole=kâ, bulaape pedevalili ki-li-kei-mä=kaa ∅3MIN
taro=DIST tomorrow children IPFV-3AUG-dig.AV-DIR1=FUT
‘(As for) the taroi, tomorrow the children will harvest (iti) for me’
b. nubole=kâ, bulaape ∅3MIN ki-kili-mä pedevalili=kaa
taro=DIST tomorrow IPFV-dig.UV-DIR1 children=FUT
‘(As for) the taroi, tomorrow the children will harvest iti for me’
Moreover, topicalization also doesn’t show any island-sensitivity. While RCs constitute
islands for relativization, as shown above, topicalization is not sensitive to this kind of
boundary. In the two sentences below, a left-peripheral topic in the matrix clause is asso-
ciated with an argument (boxed) inside a relative clause. The fact that this is grammatical
confirms that this dependency is not movement-based.
(313) Topicalization is not sensitive to RC islands: (250116)
a. pedevalili=kâ, [nubole [RC ki-kili-mä -i =laa]]=kâ
children=DIST taro IPFV-dig.UV-DIR1-3AUG=FUT=DIST
ba-ngä ki-be
NEG-yet IPFV-ready
‘(As for) the childreni, the taro that theyi will harvest is not ready yet’
b. nubole=kâ, [pedevalili [RC ki-li-kei-mä=kaa ∅3MIN ]]=kâ
taro=DIST children IPFV-3AUG-dig.AV-DIR1=FUT =DIST
ku-lu-pole-eke
IPFV-3AUG-work-fast
‘(As for) the taroi, the children who will harvest iti work fast’
4.B.2 FOCUS IS DONE WITH PSEUDO-CLEFTS (SO IT’S JUST RELATIVIZATION)
The syntax of argument focus has been somewhat briefly described in passing in Næss
(2015b) a.o., in its core surface traits. Considering naturally attested examples like (314),
Næss claims that (i) a focused argument (boxed) is in clause-final position; (ii) the verb
does not agree with it. In ordinary non-focus contexts, the intransitive verb in (314a)
would’ve taken a 2MIN prefix mu-, and the AV verb in (314b) a 1MIN prefix i-.
(314) Næss (2015: 292; slightly adapted):
a. ku-mo ngâ nuumä=ke iumu
IPFV-stay in village=PROX 2MIN
‘You are the one who stays at home’
Chapter 4. The Ā-extraction restriction, and the logic of an exception 199
b. ba, dee sii=ee, ku-wânubo=kâ iu
no this fish=PROX IPFV-kill.AV=DIST 1MIN
‘No, this fish, I am the one who killed it’
I believe that this characterization, despite being descriptively correct, misses an impor-
tant generalization: argument focus constructions are (pseudo-)clefts29. Concretely, I
propose that these sentences have the following syntax (315a). The first part is a head-
less/null-headed relative clause30, followed by a deictic marker (DIST/PROX), and finally
the focused argument is actually a nominal predicate. As an illustration, (315b) shows the
same sentence in (314a) annotated according to this analysis.
(315) a. [∅ [RC …]]=DEIC [Pred DP]
b. Cleft analysis of (314a):
[∅ [RC ku-mo ngâ nuumä]]=ke [Pred iumu]
IPFV-stay in village=PROX 2MIN
‘[(The one) who stays at home] is [you]’
This analysis accounts for both surface-level properties of argument focus constructions
described above and by Næss. The focused argument is sentence-final rather than in its
ordinary position because it’s a predicate, not an argument of the clause. The verb lacks
ordinary agreement with the focused argument because this is not, in fact, an argument
of the verb at all; rather, a null 3MIN pronoun is (which has no overt agreement marker).
Concretely for the purposes of this chapter, the fact that focus constructions are clefts
means that there is no separate Ā-dependency related to focus; it just reduces to relative
clause formation.
In support of the cleft analysis, I present the following arguments, in order: (i) argu-
ment focus constructions have the syntax of clauses with nominal predicates (or “copular
clauses”); (ii) argument focus constructions are biclausal, with a matrix portion being a
copular clause and a RC-portion; (iii) argument focus constructions are compatible with
overt relativizers; (iv) argument focus constructions follow the Ā-extraction restriction
found with relativization; (v) rather than just lack of agreement, the RC-part of argument
focus constructions always have the syntax of a clause with a 3MIN argument in the place
of the focused argument, including changes in word order.
First, let’s consider what clauses with nominal predicates look like. Äiwoo has no
overt copula. However, the subject and predicate of such a sentence must be separated by
a deictic particle =Câ or =Ce (the former seems to be the default choice)31.
29 The same analysis is proposed for DP-focus constructions in Tagalog in Hsieh (2020).
30 Relative clauses without an overt head are independently possible in Äiwoo:
(i) a. Headless subject RC: (230622) b. Headless object RC: (230622)
[∅ [RC i-veve]]=kâ mekivaavee [∅ [RC i-ve-∅n]]=nâ mi=olo-mana
ASP-buy.AV=DIST teacher ASP-buy.UV-3MIN=DIST REL=big-very
‘The one who bought it is a teacher’ ‘The one s/he bought is very big’
31 The nature of this restriction is not entirely clear. Both in elicitation contexts and in the corpus, we find
cases where there is no =Câ/=Ce, but there is some other demonstrative material separating the two parts:
200 4.B. A note about other Ā-dependencies
(316) Jane*(=kâ) nurse (231012)
Jane*(=DIST) nurse
‘Jane is a nurse’ (ungrammatical without =kâ)
In cases of argument focus, we similarly find a deictic particle before the nominal predi-
cate (317), just like for copular clauses. Like with copular clauses, here too the particle is
necessary (though see fn. 31).
(317) [∅ [RC i-vängä nubole]]*(=kâ) Mary ngâpo (231130)
ASP-eat.AV taro*(=DIST) Mary only
‘Only Mary ate taro’ (lit. ‘[Who ate taro] is [only Mary]’)
Second, focus constructions contain two separate clausal domains, as is expected under
the cleft analysis. To prove this, we can observe that these constructions contain two
independent loci for negation (318). The matrix nominal predicate can be negated (318a),
or the RC portion of the pseudo-cleft can be negated (318b), or even both at once (318c).
The intended readings are hard to convey in English with the standard focus intonation,
so I use more literal paraphrases involving pseudo-clefts for clarity.
(318) a. Negation in the matrix clause: (241219)
[Context: someone toldme thatMary yesterday cooked crab, but I know they’re
wrong, she didn’t cook crab, she cooked something else.]
[∅ [RC i-epavi Mary bugulo]]=kâ ba nulei=gu
ASP-cook.UV Mary yesterday=DIST NEG crab=NEG
‘What Mary cooked yesterday is not crab’
b. Negation inside the RC:
[Context: yesterdayMary cookedmany different things, but the only thing she
didn’t cook was the crab.]
[∅ [RC ba i-epavi Mary=gu bugulo]]=kâ nulei
NEG ASP-cook.UV Mary=NEG yesterday=DIST crab
‘What Mary did not cook yesterday is crab’
c. Both at once:
[Context: someone told me that Mary yesterday cooked many different things,
(i) buk eângâ(=kâ) eobulou-mana (231214)
book that(=DIST) long-very
‘That book is very long’ (=kâ is optional)
Moreover, one can omit any deictic/demonstrative material when the subject is non-referential (in this case,
a generic/kind argument):
(ii) a. About snakes in general: (231214) b. [Context: you are pointing at a snake]
nyingidowe (=kâ ) ba päko=gu nyingidowe*(=kâ ) ba päko=gu
snake(=DIST) NEG good=NEG snake*(=DIST) NEG good=NEG
‘Snakes are not good’ (=kâ is optional) ‘The/that snake is not good’
Comment: “Without =kâ it means snakes in
general.”
Chapter 4. The Ā-extraction restriction, and the logic of an exception 201
but the only thing she did not cook was crab. But I know they’re wrong, the
only thing she didn’t cook is turtle.]
[∅ [RC ba i-epavi Mary=gu bugulo]]=kâ ba nulei=gu
NEG ASP-cook.UV Mary=NEG yesterday=DIST NEG crab=NEG
‘What Mary did not cook yesterday is not crab’
Third, if argument focus constructions are clefts, this means their left portion is a relative
clause. Therefore, we should expect them to be compatible with overt relativizers, at least
optionally. Although this is not attested in the corpus, and so it is probably rare in usage,
it is indeed possible:
(319) [ (de=)[RC ki-epavi Mary bugulo]]=kâ nulei (241219)
REL:thing= IPFV-cook.UV Mary yesterday=DIST crab
‘Mary was cooking CRAB yesterday’,
lit. ‘[The thing that Mary was cooking yesterday] is crab’
Fourth, if focus constructions contain relative clauses, we should expect to find the usual
Ā-extraction restriction, that is, the voice in the extraction site should depend on what
argument is being focused. This is also borne out. Just like for ordinary relativization,
subject focus requires AV (320), and object focus requires UV (321):
(320) Subject focus: 3 AV, 7 UV (241219)
a. [∅ [RC i-epave nulei bugulo]]=kâ Mary
ASP-cook.AV crab yesterday=DIST Mary
‘MARY cooked crab yesterday’ (lit. ‘[Who cooked crab yesterday] is [Mary]’)
b. * [∅ [RC nulei i-epavi -∅n bugulo]]=kâ Mary
crab ASP-cook.UV-3MIN yesterday=DIST Mary
Intended: ‘MARY cooked crab yesterday’32
(321) Object focus: 3 UV, 7 AV (241219)
a. [∅ [RC i-epavi Mary bugulo]]=kâ nulei
ASP-cook.UV Mary yesterday=DIST crab
‘Mary cooked CRAB yesterday’ (lit. ‘[WhatMary cooked yesterday] is [crab]’)
b. * [∅ [RC Mary i-epave bugulo]]=kâ nulei
Mary ASP-cook.AV yesterday=DIST crab
Intended: ‘Mary cooked CRAB yesterday’
We also find the same exceptions to the Ā-extraction restriction in the domain of focus
constructions as in that of relativization. For example, UV-only verbs allow subject focus
despite their being UV (322), unlike ordinary voice-alternating transitive verbs. Moreover,
32 This string is grammatical under an irrelevant parse, where nulei ‘crab’ is the head of an object RC:
(i) [nulei [RC i-epavi-∅n bugulo]]=kâ Mary (241219)
crab ASP-cook.UV-3MIN yesterday=DIST Mary
‘The crab that s/hei cooked yesterday is Maryj’ (i.e., the crab’s name is Mary)
202 4.B. A note about other Ā-dependencies
non-canonical “UVgu” clauses allow subject relativization despite being UV; this will be
discussed in chapter 6, §6.4. In the same way, these clauses also allow subject focus (323).
(322) Subject focus is grammatical with UV-only verbs33: (250116)
[∅ [RC lopwâ eângâ i-kää-päko-i -∅n ]]=nâ Mary, ba Sam=gu
story that ASP-know.UV-good-UV-3MIN=DIST Mary NEG Sam=NEG
‘MARY knows that story well, not Sam’
(323) Subject focus is grammatical from UVgu clauses: (241219)
[∅ [RC ∅12AUG i-tu-usi-kâ-gu-de ngä skul]]=kâ
ASP-bring.UV-back.UV-DIR3-OBJ-12AUG to school=DIST
mekivaavee
teacher
‘THE TEACHER brought us back to school’
Finally, because the focused argument is not part of the same clause where the verb is,
indeed the verb will not agree with it. However, we can push this further. As we know
from our Word Order Puzzle, Äiwoo shows complex word order alternations depending
on whether the subject and object are pronouns or lexical DPs, and what φ-features they
have. The relevance of this is that focus constructions not only lack agreement with the
focused argument, but their RC portion always looks like as if the focused argument was
3MIN. The details of the word order alternations are covered in chapter 6); for now, the
following contrast will suffice. In (324a) we see an ordinary UV clause. In (324b), we see
that the word order is different if the object is a 12AUG pronoun; this is a “UVgu” type of
clause, cf. chapter 6. Now, the subject (‘teacher’) is to the right of the future marker, and
the object is also indexed by a suffixal marker on the verb (-gu-de).
(324) a. Two 3MIN arguments: (250728)
devalili ku-potaa mekivaavee=kaa bulaape
child IPFV-search.UV teacher=FUT tomorrow
‘The teacher will look for the child tomorrow’
b. 3MIN subject, 12AUG object: different word order
(iude) ku-potaa-gu-de=ngaa mekivaavee bulaape
12AUG IPFV-search.UV-OBJ-12AUG=FUT teacher tomorrow
‘The teacher will look for us tomorrow’
33 There seems to be some degree of mixed behavior in this respect. Unlike with UV-only kää ‘know’ (322),
subject focus was judged as impossible with the UV-only verbs te ‘see’ (ia) and lâwâle ‘help’ (ib). Further
research is needed to understand the nature of this discrepancy.
(i) a. * [∅ [RC kuli i-te-kä -∅n ]]=nâ Mary, ba Sam=gu (250116)
dog ASP-see.UV-DIR3-3MIN=DIST Mary NEG Sam=NEG
Intended: ‘MARY saw the dog, not Sam’
b. * [∅ [RC iso-mu ki-lâwâle-kä -∅n =naa]]=kâ Mary, ba Sam=gu
mother-2MIN IPFV-help.UV-DIR3-3MIN=FUT=DIST Mary NEG Sam=NEG
Intended: ‘MARY will help your mother, not Sam’
Chapter 4. The Ā-extraction restriction, and the logic of an exception 203
Crucially, if we want to focus the pronominal object in (324b), the structure we get looks
quite different (325a). In fact, the RC portion of this sentence looks identical to (324a), and
using the word order in (324b) is ungrammatical (325b). This only makes sense if there
is no 12AUG pronominal argument in the RC-portion of this clause at all, and the object
argument in this portion of the clause is a 3MIN nominal.
(325) Focusing a 12AUG object: identical to sentences with two 3MIN arguments (324a)
a. [∅ [RC ku-potaa mekivaavee=kaa bulaape]]=kâ iude
IPFV-search.UV teacher=FUT tomorrow=DIST 12AUG
‘The teacher will look for US tomorrow’, (240919)
lit. ‘[Who the teacher will look for tomorrow] is [us]’
b. * ku-potaa-gu-de=ngaa mekivaavee bulaape=kâ iude
IPFV-search.UV-OBJ-12AUG=FUT teacher tomorrow=DIST 12AUG
Intended: ‘The teacher will look for US tomorrow’
4.C VERBS OSCILLATING BETWEEN UV-ONLY AND
VOICE-SYNCRETIC: AN ACqUISITION PROBLEM?
We saw in §4.4.3 that in addition to UV-only verbs and voice-syncretic verbs, Äiwoo also
has a small handful of verbs (at least potaa ‘search, look for’, komaa ‘invite’, weevä ‘visit’)
that seemingly live a double life, showing traits of both classes. Like voice-syncretic verbs,
these do have an AV form, identical to their UV one. However, like UV-only verbs, these
verbs are allowed to remain in UV even in the context of subject extraction. I speculate
that this kind of behavior might be the result of a poverty of the stimulus problem in the
acquisition phase. The reasoning at a glance is the following. Telling apart UV-only verbs
from voice-syncretic ones is a hard task, and it requires rather specific kind of evidence
in the input. For more frequent verbs, a child learner has enough evidence in their input
to settle this question, and will learn that certain verbs are UV-only (kää ‘know’, te ‘see’,
etc.), and others exist in both voices but are morphologically syncretic (nu ‘drink’, etc.).
For less-frequent verbs, the input available to a learner is too poor to settle the question,
and as a result, speakers end up oscillating in how they have analyzed these verbs (see
e.g. Han et al. 2007, 2016 for a similar proposal about the acquisition of string-vacuous
head movement in Korean). I will now explore this idea in more detail.
Let us imagine a child learner of Äiwoo, who hears a single verb stem 𝑉 , has learned its
meaning, and somehow knows that there is no alternative stem 𝑉 ′ spelling out a different
voice. The learner is now tasked with settling the question of whether 𝑉 is a UV-only verb,
or if it is a voice-syncretic one. This can be also framed as a slightly a different question:
does 𝑉 have an AV, or not? If it does not have one, then clearly 𝑉 is a UV-only verb (by
definition). If it does have one, then the AV form of 𝑉 must be identical to the UV form of
𝑉 , so 𝑉 is a voice-syncretic verb. However, it is not trivial to find an answer to the question
204 4.C. Verbs oscillating between UV-only and voice-syncretic: an acquisition problem?
of whether an AV form exists or not. Let us consider now what kind of input a learner
would need to settle this question.
4.C.1 VOICE-SYNCRETIC VERBS: HOW TO ESTABLISH THAT AN AV FORM
EXISTS
First, and independently fromwhat follows, note that UV is muchmore common in Äiwoo
natural speech than AV, approximately 85 % vs. 15 % (Holmen 2020, Holmen & Næss 2025).
As a result, the number of AV sentences in their input is already going to be rather low in
the first place. Referring to chapter 3, §3.3, remember that to know whether a given verb
form is in AV or UV, one can rely on three indicators: word order, prefixal φ-agreement,
and voice concord on modifiers. The main issue is that due to independent facts about
the language, even these morphological cues will largely be absent from a lot of natural
speech data, thus leaving the question often unsettled. In other words: in the most usage-
frequent contexts, AV and UV forms (and therefore, UV-only verbs and voice-syncretic
ones) will look exactly identical, and it’s only in rarer cases that the two cases will show
a surface-visible difference.
The concrete reasons for this frequent morphological ambiguity are: (i) 3rd person
arguments, especially 3MIN ones, are most often null, so that word order is not evident
from the surface string; (ii) 3MIN subjects are more common than non-3MIN subjects, and
3MIN arguments have no overt verbal agreement marker; (iii) verb formswith no adverbial
modifiers are more common than modified ones. Essentially, when a child is trying to
determine whether the AV form of any verb exists, the only kind of evidence that is going
to be definitely informative will need to be sentences with overt arguments, and/or non-
3MIN subjects, and/or containing adverbial modifiers.
Concrete examples that illustrate this problem are given in the table in (326). For each
sentence, the surface form of the whole clause is boxed in the first line. In the left column, I
show the more common kind of sentences, consisting of only one word (the verb), with all
null 3MIN arguments. In the right column, the sentences are richer, with overt arguments,
plural subjects triggering agreement in AV, and an adverbial modifier (eke ‘quickly’). Here,
we can see that in the more common contexts, a verb like nu ‘drink’ looks exactly identical
in both voices. Since all arguments are null, all φ-affixes are null, and there are no adverbs,
both the AV and UV forms are just inu (326a,b). In this kind scenario, given that UV is so
much more frequent than AV, it is perhaps not unreasonable to assume that both of these
forms are UV, and the existence of an AV form is still uncertain. Instead, morphologically
richer contexts provide sufficient information to see that this verb does indeed have both
voices. Once enough additional material is present, the learner will know that there clearly
exists an AV form of this verb ilunueke (326c), which is visibly different from its UV form
inuekenyii (326d).
Chapter 4. The Ā-extraction restriction, and the logic of an exception 205
(326) Morphological ambiguities in voice-syncretic verbs:
Common (poor) contexts: Rarer (rich) contexts:
no overt evidence of AV evidence of AV
AV: a. inu b. pedevalili ilunueke nuwoi
∅Subj i-nu ∅Obj pedevalili i-lu-nu-eke nuwoi
ASP-drink.AV children ASP-3AUG-drink.AV water
‘S/he drank (something)’ ‘The children drank water quickly’
UV: c. inu d. nuwoi inuekenyii pedevalili
∅Obj i-nu-∅nSubj nuwoi i-nu-eke-nyii pedevalili
ASP-drink.UV-3MIN water ASP-drink.UV-quick-UV children
‘S/he drank it’ ‘The children drank the water quickly’
For more frequent lexical items, we can probably assume that the kind of sentences con-
taining richer morphological clues are present in the input. Thus, the child will observe
clear undisputed AV forms, and that in and on itself settles the question: this is a voice-
syncretic verb.
4.C.2 UV-ONLY VERBS: HOW TO ESTABLISH THAT AN AV FORM DOES NOT
EXIST
Even for frequent verbs, however, how could the child ever settle the question in the op-
posite direction, that is, establish that some frequent verb is a UV-only verb and not just
a syncretic one? Conceivably, since AV is infrequent, they could simply not have heard
the AV usage of that verb yet. To reach the intended conclusion, they would need to ob-
serve a form that is unambiguously UV (because the AV form would look different) in a
syntactic context that would otherwise require the use of AV. Concretely, the only defini-
tive evidence for analyzing a verb as a UV-only one will come from subject relativization
contexts where at least some arguments are overt, the subject is plural, and/or there are
adverbial modifiers on the verb inside the RC. Here, I assume that the learner has already
acquired the Ā-extraction restriction, that is, AV is required under subject extraction and
UV is required under object extraction34.
To observe how it could work, consider the following examples. In the more frequent
contexts with poor morphological clues, both subject and object extraction look exactly
identical if the verb is voice-syncretic (326). The RC portion of these clauses (underlined in
the surface string) is completely string-identical, thereby providing no information what-
soever to the learner about whether the verb inu ‘drank’ is AV or UV in either case.
34 This is likely not an entirely innocent assumption, since we don’t know anything about the timing of ac-
quisition of different syntactic phenomena in Äiwoo.
206 4.C. Verbs oscillating between UV-only and voice-syncretic: an acquisition problem?
(326) a. Subject relativization: (250807)
devalili inu kunudâ
[devalili [RC i-nu ∅Obj]] ku-nudâ
child ASP-drink IPFV-thirsty
‘The child that drank (something) was thirsty’
b. Object relativization:
nuwoi inu ba päkogu
[nuwoi [ nRC i-nu-∅ ]] ba päko=gu
water ASP-drink-3MIN NEG good=NEG
‘The water that he/she/it drank wasn’t good’
As earlier, once the context contains richer material we can be sure that we are indeed
looking at an AV form (ilunueke ‘they drank quickly’) in the case of subject extraction,
and at a UV form (inuekenyii ‘they drank quickly’) in the case of object extraction:
(327) a. Subject relativization: (250807)
pedevalili ilunueke nuwoi kilinudâ
[pedevalili [RC i-lu-nu-eke nuwoi]] ki-li-nudâ
children ASP-3AUG-drink.AV-quick water IPFV-3AUG-thirsty
‘The children that drank water quickly were thirsty’
b. Object relativization:
nuwoi inuekenyii pedevalili ba päkogu
[nuwoi [RC i-nu-eke-nyii pedevalili]] ba päko=gu
water ASP-drink.UV-quick-UV children NEG good=NEG
‘The water that the children drank quickly was not good’
Only from these richer kind of examples a learner can ascertain that nu ‘drink’ does indeed
follow the Ā-extraction restriction, and that it thereby has both voices. Conversely, to
determine that luwa ‘take, grab’ is a UV-only verb, a morphologically poorer example
of subject extraction will not be enough, as this could also potentially be analyzed as a
syncretic AV form:
(328) Subject relativization (ambiguous between UV-only and voice-syncretic):
devalili iluwa kigiââ
‘The child that grabbed it was happy’
a. Parse if iluwa is a UV form (exceptional non-VIA extraction ⇒ UV-only):
[devalili [RC ∅Obj i-luwa -∅n ]] ki-giââ
child ASP-take.UV-3MIN IPFV-happy
‘The child that grabbed it was happy’
b. Parse if iluwa was a voice-syncretic AV form:
[devalili [RC i-luwa ∅Obj]] ki-giââ
child ASP-take.AV IPFV-happy
‘The child that grabbed it was happy’
Chapter 4. The Ā-extraction restriction, and the logic of an exception 207
Only richer data will let the learner establish that luwa is indeed a UV-only verb. In (329),
we see what cannot be anything else than a UV form, and yet subject extraction is hap-
pening; therefore, an AV form must not exist.
(329) Subject relativization (morphologically unambiguous: UV-only verb)
pedevalili buk iluwaekenyii kiligiââ
‘The children that grabbed the book quickly were happy’
[pedevalili [RC buk i-luwa-eke-nyi -i ]] ki-li-giââ (250728)
children book ASP-take.UV-quick-UV-3AUG IPFV-3AUG-happy
‘The children that grabbed the book quickly were happy’
I assume that the number of sentences with these properties in the child input cannot be
particularly high (although, to be fair, the learner does not need sentences containing all
types of clues at the same time; even only one will do, presumably). Precisely because
this is a matter of input data, we can expect that lexical frequency will have an effect.
For more frequent verbs, the child will be provided with enough informative examples,
so that they will reliably be able to acquire that ‘know’ and ‘see’ are UV-only, whereas
‘drink’ is voice-syncretic. On the other hand, for less frequent verbs, the input data will
be too scarce to make this decision, and as a result, verbs like ‘search’, ‘invite’, and ‘visit’
will oscillate between the two classes.

CHAPTER 5
THE φ-INDEXING SYSTEM
5.1 INTRODUCTION
Throughout chapters 3–4, we have addressed some parts of our Word Order Puzzle and
Extraction Puzzle. The general overview of different clause types and the Ā-extractability
of their arguments – indicated by superscripts – is repeated in (330)1. Specifically, we have
established a derivation that accounts for the word order and Ā-extraction properties of
AV, UVplain and UVonly clauses (330a–c). The proposed syntactic argument positions are
indicated by labelled brackets in the table header. The empirical focus of this chapter is
the system that governs how φ-features are indexed on the Äiwoo verb. To introduce this,
in this table I include two additional columns, reporting for each clause type (i) where
φ-affixes occur on the verb, and (ii) what argument(s) they index.
(330) Overview of clausal word orders and Ā-extractability profiles:
Word order: Position Indexed
Clause type:
[CP [TP [vP of φ-affix: argument:
a. AV: S3 φS-V.AV =TAM O
7 prefix S
b. UVplain: O
3 V.UV S7 =TAM suffix S
c. UV 3 3only: O V.UV S =TAM suffix S
d. UV : ONAinâ V.UV S
3 =TAM suffix S
e. UV NAgu: O V.UV -πO =TAM S
3 suffix O
f. UV NA NASVO: S V.UV -πS =TAM O suffix S
g. UV NA1>2: S V.UV -πS-π
NA
O =TAM suffix S+O
(3/7: can/cannot be Ā-extracted. NA: cannot be Ā-extracted, but for orthogonal reasons.)
1 This iteration of the table includes an extra row, labelled “UV1>2” (330g), that has not been shown before.
This will be shown to be a subtype of UVSVO (syntactically) in the next chapter.
210 5.1. Introduction
(The presence of subject-indexing φ-suffixes in UVplain, UVonly, and UVinâ clauses is not
apparent from the word order schema in the table; we will get to this in due time.)
This chapter takes a temporary step back from directly trying to solve the two larger
puzzles, and is largely descriptive instead. In order to precisely understand the word or-
der of the different types of clauses in (330), we must first ascertain in what syntactic
positions the subject and object are in each type. To do so, a necessary precondition is to
determine the syntactic nature of the different φ-affixes we observe. For example, if we
want to build a derivation for UVgu clauses (330e), we must understand whether the “πO”
morpheme between the verb and the TAM particles is an agreement marker (and thus not
too informative about word order), or some kind of pronominal clitic, or a full copy of the
object in spec,TP before moving up to spec,CP.
This chapter both adds to and amends the currently available descriptive literature
on this aspect of Äiwoo grammar (Næss 2006, 2015a,b, 2021b, 2024, Ross & Næss 2007,
Næss & Boerger 2008, Roversi 2019, 2020, 2025b, a.o). The analysis of Äiwoo’s complex φ-
indexing system proposed in this chapter will both inform, and be informed by, a global
analysis of Äiwoo clausal syntax – something which earlier analyses did not integrate
fully. Because Äiwoo is an understudied language, I take the time to empirically motivate
my analyses when they diverge from the available description.
For the readers who are not interested in the fine morphological details of φ-indexing
in Äiwoo, a cursory reading of this chapter should suffice. A very concise summary of the
main claims is given at the end of the next subsection.
5.1.1 DOMAIN OF DISCUSSION
Some properties of Äiwoo’s φ-indexing systemwere already seen in the previous chapters.
Just to offer a first-glance illustration of the variety and complexity of this system, consider
the examples in (331). We can observe subject-indexing prefixes in two different positions
(331a,b), but also subject-indexing suffixes combinedwith both OV(S) and SVOword order
(331c,d), object-indexing suffixes (331e), and both at once (331f). For each clause, in the
right margin I label what word order/clause type it exemplifies.
(331) Examples of φ-indexing suffixes: (250728)
a. ( iude) de- ku-tou-mä=kaa sii (AV)
12AUG 12AUG-IPFV-bring.AV-DIR1=FUT fish
‘We.INCL will bring fish’
b. ( ijii) ku- lu- tou-mä=kaa sii (AV)
3AUG IPFV-3AUG-bring.AV-DIR1=FUT fish
‘They will bring fish’
c. sii ku-tu-mä -i =laa (UVplain)
sii IPFV-bring.UV-DIR1-3AUG=FUT
‘They will bring the fish’
Chapter 5. The φ-indexing system 211
d. ( ijii) ku-tu-mä -i =laa iude (UVSVO)
3AUG IPFV-bring.UV-DIR1-3AUG=FUT 12AUG
‘They will bring us’
e. ( iude) ku-tu-mä -gu-de =ngaa John (UVgu)
12AUG IPFV-bring.UV-DIR1-OBJ-12AUG=FUT John
‘John will bring us’
f. ( iu) ku-tu-mä -nee-mu =waa (UV1>2)
1MIN IPFV-bring.UV-DIR1-1MIN-2MIN=FUT
‘I will bring you’
The second half of this introductory section (§5.1.2) is dedicated to presenting some the-
oretical grounding for the issues at matter: what φ-indexers are and can be, and what
care we must take when distinguishing between morphophonology and morphosyntax
(§5.1.2.1). Then, I confirm that all Äiwoo φ-indexers are morphophonologically affixes
(§5.1.2.2), and I discuss what kind of evidence we would need to determine the syntax
of these affixes (§5.1.2.3). The following sections present the full paradigms (§5.2), and
then I proceed to discuss how the syntactic properties of the prefixes and the suffixes are
different. Contra the earlier literature cited above, I will argue that the various affixes in
(331) are, despite their surface similarities, not syntactically homogenous. The prefixes
(331a,b) are agreement markers, that is, the spell-out of φ-features copied by C via agree-
ment (§5.3). The suffixes (331c–f), in contrast, do not instantiate agreement, but are actu-
ally pronouns in spec,TP, which get later phonologically reduced in the post-syntax (§5.4).
Finally, §5.5 accounts for when pronouns are realized in their full vs. reduced forms.
5.1.2 AGREEMENT MARKERS, PRONOUNS, AND ALL THAT’S IN BETWEEN
Throughout this chapter, the goal will be to determine the syntactic nature and properties
of the various forms boxed in (331). However, this task is often far from trivial when it
comes to morphemes whose main role is to index the φ-features of an argument. For
better-studied languages, the syntactic evidence is robust enough to make us confident
in calling ‘we’ in English a pronoun (332a) and the suffix -iamo in Italian an agreement
marker (332b), although they both index a 1PL subject.
(332) a. we often eat fish
b. mang -iamo spesso il pesce (Italian)
eat-1PL often the fish
‘We often eat fish’
Even in familiar languages, of course, things are not always as straight-forward. For exam-
ple, in both examples in (333), the form lo indexes a masculine 3SG direct object. However,
in Italian it cannot co-occur with a direct object DP (333a), whereas in Spanish it can
(333b). (See the vast literature about clitic doubling in Spanish; Jaeggli 1982, 1986 et seq.)
212 5.1. Introduction
(333) a. lo= ved-iamo sempre (*lui) (Italian)
3SG.M=see-1PL always him
‘We always see him’
b. siempre lo= ve-mos a él (Spanish)
always 3SG.M=see-1PL DOM him
‘We always see him’ (Juan Vásquez, p.c.)
The morphemes highlighted in (332)–(333) are generally held to reflect three possible
underlying syntactic configurations. Following Akkuş et al. (2025) I will use the umbrella
term ‘φ-indexer’ to neutrally subsume all three, which I proceed to discuss below. The
two most clear-cut ones are ‘agreement markers’ (334a) and ‘pronouns’ (334b); in both
trees, the element that gets spelled out as the surface form is highlighted in red. In the
first case (334a), the φ-indexer we see on the surface is the spell-out of features that have
been copied onto a functional head by a probe (often v, T, C, etc.) from a nominal. In the
second case (334b), the φ-indexer is simply a fully phrasal nominal constituent, occupying
a specifier position (I represent it here as having moved to its surface position, but this
is immaterial, and the same constituent in its base-generated position would still be a
pronoun). Notationally, I represent features inherently present on a nominal with brackets
([φi]), and features copied onto a probe via agreement in angle brackets (as elements in
an ordered list: ⟨φi, φj, …⟩).
(334) a. Agreement marker: b. Pronoun:
HP HP
… DP[φi]
[ H ⟨φ]i⟩ … [ H ⟨φi]⟩ …
I+S: [φ] I+S: [φ]M
… …
DP[φi] … DP[φi] …
I represent pronouns here as having the label ‘DP’ for simplicity, although there is a rich
literature about the categorial and/or size differences between pronominal arguments and
DPs containing a lexical noun (e.g. ‘they’ vs. ‘the children’). See a.o. Cardinaletti & Starke
(1999), Déchaine & Wiltschko (2002), and Sichel & Toosarvandani (2024), and the discus-
sion in §5.5 and chapter 6 for Äiwoo pronouns specifically.
Finally, and figuratively somewhere in between the two previous options, a φ-indexer
could be a (morphosyntactic) clitic. Although this term is used in at times somewhat
confusing and contradictory ways across the literature (see e.g. Julien 2007b, Bermúdez-
Otero & Payne 2011, a.o.), it is generally used to identify some kind of reduced nominal
structure, such as a D head rather than a fully phrasal DP, often forming a complex head
(at some level of the derivation) with another functional head in the clausal extended
projection (e.g. v, T, C). The literature about the (morpho)syntax of pronominal clitics
Chapter 5. The φ-indexing system 213
is all too vast to be summarized here; see Anagnostopoulou (2017) for a recent detailed
overview. What is different (among many things) between various approaches to the syn-
tax of clitics is how one gets to this configuration, derivationally speaking. For one family
of theories (see e.g. Rezac 2008, Roberts 2010), the clitic – represented as a D head in (335)
– is merged directly in its surface position in the narrow syntax, adjoined to the probing
head in question, as the result of a step of head-movement2.
(335) Narrow-syntactic “cliticization”:
HP
…
H …
Dφi [ H ⟨φ]⟩ …i
I+S: [φ]
D(P)[φi] …
I have labelled this descriptively “narrow-syntactic cliticization”, to convey the idea that
the clitic is generated as such already in the syntactic component. In an alternative ap-
proach, which we could call “post-syntactic cliticization” (see e.g. Harizanov 2014, a.o.),
what we have in the narrow syntax is instead a fully phrasal nominal constituent in a spec-
ifier position (334b), here repeated as (336a). It is only later, in the post-syntactic compo-
nent, that this nominal is morpho(phono)logically reduced (336b), through morphological
operations such as M[orphological]-merger, or Local Dislocation, or other similar con-
cepts (Marantz 1988, Embick & Noyer 2001, Matushansky 2006, Embick 2007, a.o.).
2 Depending on the specific approach, the element undergoing movement/cliticization could be either the
entire argument, or a smaller part of the whole argument (the “big DP hypothesis”; Torrego 1995, Uriagereka
1995, Belletti 1999, 2005, Cecchetto 2000, Arregi & Nevins 2012, a.o.). I also abstract away from whether
the whole DP is pronounced in its base position (thus resulting in “clitic doubling”, i.e., the simultaneous
pronunciation of both a clitic and its full argument associate) or not.
214 5.1. Introduction
(336) Post-syntactic cliticization:
a. In the narrow syntax: b. Post-syntactic reduction:
HP HP
DP[φi] DP[φi]
[ H ⟨φi]⟩ … …
I+S: [φ]M Dφi H ⟨φi⟩
… …
DP[φi] … tDPi …
In this chapter, I will argue that the suffixal φ-indexers in Äiwoo reflect precisely this
chain of events: in the narrow syntax they are pronouns, occupying spec,TP, and later
in the derivation they are morphologically reduced (see §5.5). However, to avoid confu-
sion, I will refrain from using the terms ‘clitic’ and ‘cliticization’, and will instead stick to
‘reduced pronoun’, ‘reduction’, and variants thereof. In a sense, there is no clear way in
which these φ-indexers are ‘clitics’: syntactically, they are fully pronouns (albeit phono-
logically reduced), and even morphophonologically, they are not clitics but affixes. I will
now proceed to substantiating this last claim.
5.1.2.1 (MORPHO)PHONOLOGY IS NOT (MORPHO)SYNTAX
Given the space of analytical options we just saw, identifying the underlying syntax of all
the various Äiwoo φ-indexers in (331) is not a straightforward matter. Indeed, a vast liter-
ature has approached this problem considering rich data from many different languages,
and trying to develop diagnostics and tests that will allow us to settle this matter for any
given φ-indexer (see, among others, Anagnostopoulou 2006, 2016, 2017, Woolford 2008,
Nevins 2011, Kramer 2014, Harizanov 2014, Compton 2016, Baker & Kramer 2018, Yuan
2018, 2020, 2021, Taghipour & Kahnemuyipour 2021, Paparounas & Salzmann 2024, Saab
2024). However, a lot of this literature suffers from a long-recognized overarching con-
ceptual issue: confounding (morpho)phonological properties with (morpho)syntactic ones
(cf. e.g. Yuan 2021). In a non-lexicalist framework like Distributed Morphology, the vari-
ous syntactic configurations showcased in (334)–(336) are not guaranteed in anyway to be
consistently reflected by distinct and non-overlapping morphophonological outcomes. In
other words, the mapping between syntax and morphology is not one-to-one but many-
to-many, allowing for various types of mismatches. Therefore, although a morpheme’s
morphophonological properties may be informative as an initial heuristic, they cannot
be relied on exclusively as diagnostics for syntactic structure. For more discussion of this
issue, see Yuan (2021) and Akkuş et al. (2025).
One example of a classic criterion used to distinguishes agreement markers from cl-
itics is the claim that the former may show allomorphy and idiosyncratic morphological
Chapter 5. The φ-indexing system 215
irregularities, whereas the latter should not (Zwicky & Pullum 1983; note that they use the
term ‘affix’ for what I’m calling ‘agreement marker’ here). Nevins (2011) criticizes this for
indentifying a frequent correlation rather than a real predictive factor, and weakens this
into a one-way prediction. He considers the specific case of interactions between T and
φ-indexers. For him, clitics should be tense-invariant, whereas agreement markers may
show tense-conditioned allomorphy. For example, in Italian subject agreement markers
show tense-conditioned allomorphy. In (337), the 1PL subject suffix is -iamo in the present
tense (337a), but it’s -emo in the future (337b). The object clitic lo=, on the other hand, re-
mains the same.
(337) a. lo=ved-iamo b. lo=ved-r-emo (Italian)
3SG.M=see-1PL(.PRS) 3SG.M=see-FUT-1PL
‘We see him/it’ ‘We will see him/it’
Nevins’s (2011) prediction is one-directional in the sense that if we see allomorphy, we can
conclude that we are looking at agreement, but we are not allowed to infer anything based
on the lack of allomorphy. However, as pointed out in Yuan (2021), this proposal fails in
the other direction too, due to the same issues that Nevins criticizes in Zwicky & Pullum’s
(1983) work. Although tense-(in)variance might be a frequent correlation and thereby a
useful heuristic, nothing in the theory really predicts this in a principled way. The syn-
tactic structure of clitics (335)–(336b) does allow the triggering of allomorphy between the
D head and e.g. T3. And indeed, Yuan (2021) points to the existence of counterexamples:
in several Irish varieties, pronominal clitics adjacent to verbs may constitute triggers for
allomorphy on both the clitic and/or the verb (see Bennett et al. 2019). As a further source
of counterexamples, Akkuş et al. (2025) examine the complex syntax of φ-indexing in So-
rani Kurdish (and other Kurdish varieties), and they show that the indexers’ phonological
status (affix vs. morphophonological clitic) is frequently mismatched with their syntactic
nature (agreement marker vs. a pronoun).
To reiterate, the general problem here is that determining whether a φ-indexer is an
agreement marker or a pronoun (or a clitic) is a syntactic matter. Therefore, this ques-
tion should be settled on syntactic grounds. Morphophonological considerations may be
useful heuristics and clues, but given the many-to-many mapping between syntax and
phonology, they cannot be relied on exclusively.
5.1.2.2 MORPHOPHONOLOGICALLY, ALL ÄIWOO φ-INDEXERS ARE AFFIXES
Before moving on to the discussion of the syntactic properties of the Äiwoo φ-indexing
system, let us now settle themorphophonological side of thematter. This is relatively easy:
all φ-indexers in Äiwoo (331) are simply affixes. To see this, it is instructive to compare
them to the TAM particles, which are morphophonological clitics (which should not be
3 For discussion about the locality constraints on allomorphy, specifically in Distributed Morphology, see
Bobaljik (2000, 2012), Embick (2010), Svenonius (2012), Moskal (2015), Moskal & Smith (2016), Merchant
(2015), Harðarson (2018), Choi & Harley (2019), Ganenkov (2020), and Paparounas (2023), a.o.
216 5.1. Introduction
confused with morphosyntactic clitics, as defined in (335) above). First, the TAM particles
show a lower degree of “selection” (Zwicky & Pullum 1983). The future marker =Caa
is shown here attaching to verbs (338a), nominals (338b), quantifiers (338c), other TAM
particles (338d), and certain adverbials (338e,f), although this last part of their distribution
is less well understood. Both prefixal and suffixal φ-indexers, on the other hand, can only
ever attach to a verb (339)4.
(338) TAM particles can attach to different types of hosts: (250728)
a. Anna ki-vängä =kaa sii
Anna IPFV-eat.AV=FUT fish
‘Anna will eat fish’
b. sii ki-ngä Anna =kaa
fish IPFV-eat.UV Anna=FUT
‘Anna will eat the fish’
c. sii ki-ngä Anna=dä =naa  
fish IPFV-eat.UV Anna=some=FUT
‘Anna will eat some (of the) fish’
d. nuwopa eângâ ki-ve-no=to =waa
house that IPFV-buy.UV-1MIN=PRF=FUT
‘I will buy/will have bought that house’
e. sii i-tâbu-mu=waa ponu lâwâu =kaa i-ki-epave=kâ (231109)
fish ASP-cut.UV-2MIN=FUT then ADV=FUT 1MIN-IPFV-co ok.AV=DIST
‘I will cook after you cut the fish’
(Perhaps more literally: ‘You will cut the fish, and then I will cook’)
f. Mary i-epave sup bugulo, eä Anna bulaape =kaa (250213)
Mary ASP-cook.AV soup yesterday and Anna tomorrow=FUT
‘Yesterday Mary cooked soup, and tomorrow Anna will’
(339) Φ-indexers can only attach to verbs: (250728)
a. de- ki-vängä=kaa sii
12AUG-IPFV-eat.AV=FUT fish
‘We will eat fish’ (any other position for de- is ungrammatical)
b. sii ki-ngä -de =dä=naa
fish IPFV-eat.UV-12AUG=some=FUT
‘We will eat some (of the) fish’ (any other position for -de is ungrammatical)
Moreover, the φ-indexers show a higher degree of phonological boundedness to their host
than the TAM particles, thereby supporting the idea that the former are morphophono-
logical affixes whereas the latter are morphophonological clitics. For example, remember
that certain morphemes in Äiwoo show a vowel alternation reminiscent of vowel har-
mony (chapter 2, §2.4). The 3rd person directional is realized with a back vowel -kâ when
4 See below for a discussion of the case where φ-suffixes are attached to possessive forms.
Chapter 5. The φ-indexing system 217
the following syllable contains a back vowel, and with a front vowel -kä otherwise. Thus,
when followed by the 1MIN and 2MIN suffixes -no and -mu, the directional is realized as
-kâ (340a,b). In contrast, the 12AUG and 3AUG suffixes are -de and -i, so preceding these,
the directional is realized as -kä (340c,d); the mismatched options are ill-formed. Crucially,
however, φ-indexers obligatorily trigger this phonological process, whereas the TAM par-
ticles only optionally do so. In (341) we see =to and =gu, both containing back vowels, and
in both cases the directional can be realized in its front vowel form -kä.
(340) Φ-indexers trigger vowel harmony on -kä DIR3: (250728)
a. sii ku-tu -kâ-no =ngaa
fish IPFV-bring.UV-DIR3-1MIN=FUT
‘I will bring the fish’ (*kutukäno=ngaa)
b. sii ku-tu -kâ-mu =waa
fish IPFV-bring.UV-DIR3-2MIN=FUT
‘You will bring the fish’ (*kutukämu=waa)
c. sii ku-tu -kä-de =ngaa
fish IPFV-bring.UV-DIR3-12AUG=FUT
‘We will bring the fish’ (*kutukâde=ngaa)
d. sii ku-tu -kä-i =laa
fish IPFV-bring.UV-DIR3-3AUG=FUT
‘They will bring the fish’ (*kutukâi=laa)
(341) TAM particles only optionally trigger vowel harmony on -kA DIR3:
a. i-tou -kä=to dekilingä   (250717)
1MIN-bring.AV-DIR3=PRF food
‘I have brought food’ (optionally also itoukâ=to)
b. ba i-tou -kä=gu dekilingä
NEG 1MIN-bring.AV-DIR3=NEG fish
‘I haven’t brought food’ (optionally also itoukâ=gu)
Furthermore, in certain paradigms, at least some of the suffixes show idiosyncratic al-
lomorphic interactions with the stem, providing thus further evidence for their mor-
phophonological nature as affixes. Roversi (2025b) shows that all possessive elements in
Äiwoo are, despite their looks, UV verbs, and that the paradigm of suffixal φ-indexers we
find on possessives is thus – syntactically – completely parallel to what we see on UV
verbs. For example, consider the paradigms in (342); I omit the unit-augmented forms for
simplicity. The suffixal paradigm on the UV verb ngä ‘eat’ is the standard one (see §5.2.1).
On the inalienably possessed form ‘mother’ and the two possessive classifiers for food
items and tools/utensils5, most of the suffixal exponents are the same as those found on
a UV verb. However, we also see no consistent exponent for 1MIN, and a variety of stem
alternation patterns conditioned by the presence of certain suffixes but not others.
5 See Roversi (2025b) for details about the possessive system of Äiwoo.
218 5.1. Introduction
(342) UV verb ‘mother’ POSS.FOOD POSS.TOOL
‘X is eating’ ‘X’s mother’ ‘X’s (food)’ ‘X’s (tool)’
1MIN ki-ngä-no iso nugo nugu
12MIN ki-ngä-ji iso-ji nä-ji nugu-ji
2MIN ki-ngä-mu iso-mu na-mu nugu-mu
3MIN ki-ngä-∅n isä-∅n na-∅n nogo-∅n
1AUG ki-ngä-ngopu iso-ngopu nugo-ngopu nugu-ngopu
12AUG ki-ngä-de iso-de nä-de nugu-de
2AUG ki-ngä-mi iso-mi nä-mi nugu-mi
3AUG ki-ngä-i isä-i na-i nogo-i
To summarize: all φ-indexers in Äiwoo can only attach to verbs and no other class, and
they show a high degree of morphophonological boundedness to their host (for example,
they can trigger vowel harmony and suppletion). Therefore, all Äiwoo φ-indexers are
morphophonological affixes.
5.1.2.3 SYNTACTIC EVIDENCE: CO-OCCURRENCE WITH ARGUMENTS
As discussed, however, these affixes’morphophonological properties can be a useful heuris-
tic as to their syntactic nature, but cannot – in a modular theory with Late Insertion – un-
ambiguously determine it. To establish what their syntax is, we need syntactic evidence.
Following Akkuş et al. (2025) (but really going back at least to Rizzi 1986), I propose that a
strong piece of syntactic evidence comes from patterns of co-occurrence (or lack thereof)
between φ-indexers and nominal arguments in the clause. Consider again three of the
structures described above, corresponding respectively to agreement markers (343a), pro-
nouns (343b), and clitics formed in the narrow-syntax (343c). (Recall that unlike narrow-
syntactic clitics, post-syntactically reduced pronouns are underlyingly the same as (343b),
and only get reduced to a structure like (343c) in the post-syntax.)
(343) a. Agreement marker: b. Pronoun: c. (Narrow-synt.) clitic:
HP HP HP
… DP[φi] …
[ H ⟨φ]i⟩ … [ H ⟨φiI+S: [φ] ]⟩ … H …I+S: [φ]M
… …
Dφi [ H ⟨φ]i⟩ …
I+S: [φ]
DP[φi] … DP[φi] … D(P)[φi] …
The structural differences between these three configurations make specific predictions as
to what we should expect in terms of two aspects: (i) co-occurrence between a φ-indexer
Chapter 5. The φ-indexing system 219
and a (coreferential, at least possibly overt) argument; (ii) whether or not the φ-indexer
in question occupies a specifier position. I will argue for the breakdown in (344).
(344) Does it co-occur Does it occupy a
with an argument? specifier position?
Agreement: yes no
(Narrow-syntactic) clitics: might no
Pronouns: no* yes
*Modulo the possibility of multiple copy spell-out.
Let us consider the first column. First, agreement markers (343a) ought to be able to co-
occur with their argument associate, which (modulo independent factors) should in prin-
ciple be able to be overt. They are, after all, only the spell-out of features copied onto a
functional head, so there should be no complementary distribution between these and the
nominal whose features have been copied. A pronoun (343b), on the other hand, should
not co-occur with a DP argument, since the pronoun itself is the argument in question.
The only scenario where a genuine pronoun could co-occur with a coreferent overt argu-
ment is in the case of multiple copy spell-out. For clitics, the landscape is slightly more
complex, since they are a “nominal” in some sense, but a rather reduced one (just a D
head). From the whole literature about clitic doubling (see e.g. Anagnostopoulou 2017)
we know that co-occurrence should in principle be possible: the nominal head spelling
out the clitic may be pronounced together with its fully phrasal DP associate.
Another syntactic difference between the three kinds of φ-indexers is whether or not
they occupy a specifier position. Agreement markers and (narrow-syntactic) clitics do not,
while pronouns do, since they are a fully phrasal constituent (even though they might
be reduced later, post-syntactically). This difference might result in different word order
effects, especially in a language with rather strict word order, like Äiwoo, where certain
specifier positions are required to be filled by a constituent (for example spec,CP, resulting
in a V2 effect, or spec,TP, etc.). Because agreement markers and clitics don’t occupy a
specifier position, in principle they will allow a specifier adjacent to them to be filled by
some other constituent, whereas this should be impossible with pronouns.
For example, Italian object clitics do not fill a specifier position. In (376a) the clitic la
‘her’ is not occupying spec,TP, and therefore, the subject can still raise to that position6.
(345) Italian object clitics do not occupy a specifier position:
a. [TP Ana la=porta [vP tS tO domani
Ana 3SG.F=brings tomorrow
‘Ana will bring her tomorrow’
6 I am presenting a rather simplified view here for the sake of the argument, making the assumptions that
(i) the finite verb is in T; (ii) subjects are base-generated in spec,vP; (iii) the preverbal subject position
is spec,TP. This is almost certainly too simplistic, cf. the discussion in e.g. Cardinaletti (1997, 2004) and
references therein. For simplicity, I am not considering the possibility of multiple specifiers, or that the site
of cliticization is a lower head than the one hosting the subject.
220 5.1. Introduction
b. TP
DP
Ana
T
vP
D V+v+T
la porta t v′S
tO … domani
If the morpheme la, on the other hand, was a full pronoun, it would occupy spec,TP. We
might then expect that this would block raising of the subject, contrary to fact (346)7.
(346) Cliticization of the object does not block raising of the subject:
a. * [TP la=porta [vP Ana tO domani
3SG.F=brings Ana tomorrow
Intended: ‘Anawill bring her tomorrow’ (ungrammatical under information-
structural neutral prosody)
b. Derivation that would generate the ungrammatical string:
TP
DP
la
V+v+T vP
porta
7
DP v′
Ana
tO … domani
In the rest of this chapter, I will argue in favor of the following claims: (i) the Äiwoo φ-
indexers are all morphophonological affixes, but they are syntactically hetereogenous: (ii)
the φ-prefixes are pure agreement markers; (iii) the φ-suffixes are pronouns in spec,TP,
spelled out in a phonologically shorter form due to post-syntactic operations.
7 Note that post-verbal subjects in Italian are possible under certain discourse conditions, and with marked
prosody. The VSO string in (346) is intended as ungrammatical when outside of these particular contexts,
with neutral prosody.
Chapter 5. The φ-indexing system 221
5.2 Φ-PREFIXES AND φ-SUFFIXES
5.2.1 THE PARADIGMS
In this section I will present the two main paradigms of affixal φ-indexers: the prefixal
and suffixal (subject) markers. These have been described in the present form since Næss
(2006). As described in chapter 3, the prefixal markers are used in AV and for intransitive
verbs, whereas the suffixal ones are used in UV (and CV), that is, whenever a subject DP
would occupy spec,TP – the position between the verb and the TAM markers. In (347)-
(348) I illustrate the complete person/number paradigms, using the voice-syncretic verb
potaa ‘search, look for’ (§4.4.4), for maximal minimality between the paradigms.
A few morphological details are worthy of comment. 1st person exclusive and 3rd
person forms are different between the two paradigms, whereas 1st+2nd (1st person in-
clusive) and 2nd person forms have identical exponents –modulo the different position, of
course. 3MIN is null in both paradigms, but the existence of a 3MIN suffix is revealed by the
fact that the immediately following future particle =Caa takes its /n/-initial allomorph, in-
stead of the default /k/-initial form (348g); cf. AV ku-potaa=kaa vs. UV ku-potaa-∅=naa)8.
It is impossible to tell whether or not a prefixal 3MIN marker exists, and what its position
would be (347g). Finally, as for 3AUG, the prefixal marker (347h) is to the right of the aspect
prefix, while all other prefixes are to its left. As discussed in §3.5.5, I take this to be just a
shallow morphological idiosyncrasy, and not reflective of a deeper syntactic difference.
(347) Prefixes: (250728)
1st person (exclusive): 2nd person:
a. i- ku-potaa=kaa sii e. mu- ku-potaa=kaa sii9
1MIN-IPFV-search.AV=FUT fish 2MIN-IPFV-search.AV=FUT fish
‘I will look for fish’ ‘You.SG will look for fish’
b. me- ku-potaa=kaa sii f. mi- ku-potaa=kaa sii
1AUG-IPFV-search.AV=FUT fish 2AUG-IPFV-search.AV=FUT fish
‘We.EXCL will look for fish’ ‘Y’all will look for fish’
1st+2nd person (inclusive): 3rd person:
c. ji- ku-potaa=kaa sii g. ku-potaa=kaa sii
12MIN-IPFV-search.AV=FUT fish IPFV-search.AV=FUT fish
‘You and I will look for fish’ ‘S/he will look for fish’
d. de- ku-potaa=kaa sii h. ku- lu- potaa=kaa sii
12AUG-IPFV-search.AV=FUT fish IPFV-3AUG-search.AV=FUT fish
‘We.INCL will look for fish’ ‘They will look for fish’
8 Remember that the initial consonant of this particle varies depending on the host immediately to its left. The
CV marker =Cä and the proximal/distal particles =Ce/=Câ follow the same pattern. See chapter 2, §2.6.3.2
and Roversi (2019: 24–25), Roversi (2020: 108) for details.
222 5.2. Φ-prefixes and φ-suffixes
(348) Suffixes: (250728)
1st person (exclusive): 2nd person:
a. sii ku-potaa -no =ngaa e. sii ku-potaa -mu =waa
fish IPFV-search.UV-1MIN=FUT fish IPFV-search.UV-2MIN=FUT
‘I will look for the fish’ ‘Y’all will look for the fish’
b. sii ku-potaa -ngopu =waa f. sii ku-potaa -mi =aa
fish IPFV-search.UV-1AUG=FUT fish IPFV-search.UV-2AUG=FUT
‘We.EXCL will look for the fish’ ‘You.PL will look for the fish’
1st+2nd person (inclusive): 3rd person:
c. sii ku-potaa -ji =ngaa g. sii ku-potaa -∅n =naa
fish IPFV-search.UV-12MIN=FUT fish IPFV-search.UV-3MIN=FUT
‘You and I will look for the fish’ ‘S/he will look for the fish’
d. sii ku-potaa -de =ngaa h. sii ku-potaa -i =laa
fish IPFV-search.UV-12AUG=FUT fish IPFV-search.UV-3AUG=FUT
‘We.INCL will look for the fish’ ‘They will look for the fish’
For completeness, I include here the unit-augmented forms. Informally speaking, these
can be thought of as the equivalent of a dual in a minimal-augmented system. Semanti-
cally, their reference is the same as the corresponding minimal form “plus one” (cf. the
English translations; see chapter 2, §2.5). Morphologically, they are consistently built with
the augmented form and an additional suffixal marker -le, for both the prefixal and the suf-
fixal paradigm. I will not focus on these forms in what follows, and leave the development
of a full model of them to future research.
(349) Prefixes: (250728)
a. me- ku-potaa -le =naa c. mi- ku-potaa -le =naa
1AUG-IPFV-search.AV-UA=FUT 2AUG-IPFV-search.AV-UA=FUT
‘We two.EXCL will look for’ ‘You two will look for’
b. de- ku-potaa -le =naa d. ku- lu- potaa -le =naa
12AUG-IPFV-search.AV-UA=FUT IPFV-3AUG-search.AV-UA=FUT
‘We three.INCL will look for’ ‘They two will look for’
9 The 2MIN prefix oscillates between mu- and mi-; my consultant typically accepts mu- as a perfectly good
alternative even when he produces mi- first himself. Both 2AUG affixes, in contrast, are always mi-/-mi,
and never *mu-/-mu. The vowel oscillation in 2MIN might be due to the process of high vowel deletion
(chapter 2, §2.4), whereby both mi- and mu- end up being pronounced in a near-identical way [mʲ~mʷ]).
As an alternative or additional factor, this oscillation might be due to a vowel-harmony-like process that
also affects the prefixes ki~ku- IPFV and li~lu- 3AUG (chapter 2, §2.4). Impressionistically, the IPFV and 3AUG
prefixes undergo this process fairly consistently, whereas this is much less the case for the 2MIN prefix. I
leave it to future research to determine the exact nature of this alternation. Note that the 2MIN suffix (348e),
on the contrary, is always realized as -mu, and never as *-mi.
Chapter 5. The φ-indexing system 223
(350) Suffixes: (250728)
a. ku-potaa -ngo-le =naa c. ku-potaa -mi-le =naa
IPFV-search.UV-1AUG-UA=FUT IPFV-search.UV-2AUG-UA=FUT
‘We two.EXCL will look for (it)’ ‘You two will look for (it)’
b. ku-potaa -de-le =naa d. ku-potaa -i-le =naa
IPFV-search.UV-12AUG-UA=FUT IPFV-search.UV-3AUG-UA=FUT
‘We three.INCL will look for (it)’ ‘They two will look for (it)’
5.2.2 PREFIXES AND SUFFIXES ARE SYNTACTICALLY DIFFERENT: A PREVIEW
Beyond the morphological differences between the two paradigms (position and expo-
nents), I will argue that there is an important syntactic difference between them. As an-
ticipated, the core argument relies on whether or not these markers can co-occur with a
corresponding overt subject argument. The generalization is stated in (351).
(351) Co-occurrence patterns:
a. The prefixal φ-indexers must co-occur with a subject argument, which may
optionally be overt.
b. The suffixal φ-indexers can’t co-occur with an (optionally) overt subject ar-
gument [to be revised].
In §5.3 I will show the evidence for the first part of this generalization (351a). In §5.4 I will
do the same for (351b). We will then see how this generalization, although useful, is not
precise enough, and we will revise it accordingly.
5.3 THE φ-PREFIXES: OBLIGATORY CO-OCCURRENCE
5.3.1 THE DATA
To show co-occurrence (or lack thereof) between a verbal φ-prefix and an overt subject,
only certain person/number categories will be useful. Specifically, we cannot use 3MIN
(either pronouns or lexical DPs), because there’s no overt verbal prefix. In contrast, 3AUG
is informative. Here, we can see that when the subject is plural – either a lexical DP (352a)
or a pronoun (352b) – the 3AUG prefix lu- is obligatory. When the subject is pronominal
(352b), a null form is the most frequent realization for the subject pronoun, but an overt
pronoun can be used under particular discourse conditions. In either case, if the verbal
agreement prefix is absent, the sentence is not acceptable (352c).
(352) a. mikilitei ku- lu- tou-mä=kaa sii (250728)
fishermen IPFV-3AUG-bring.AV-DIR1=FUT sii
‘The fishermen will bring fish’
b. ∅3AUG /ijii ku- lu- tou-mä=kaa sii
3AUG IPFV-3AUG-bring.AV-DIR1=FUT sii
‘They will bring fish’
224 5.3. The φ-prefixes: obligatory co-occurrence
c. * mikilitei/ijii/∅3AUG ku- tou-mä=kaa sii
fishermen/3AUG IPFV-bring.AV-DIR1=FUT fish
Intended: ‘The fishermen/they will bring fish’
The same pattern replicates for non-3rd person subjects. Like the subject pronoun in
(352b) above, the 12AUG subject pronoun in (353a) can be optionally realized (although
it will most frequently be null). The prefixal marker, however, cannot be dropped under
any circumstance (353b).
(353) a. ∅12AUG /iude de-ki-vängä=kaa sii (250728)
12AUG 12AUG-IPFV-eat.AV=FUT fish
‘We.INCL will eat fish’
b. *∅12AUG /iude ki-vängä=kaa sii
12AUG IPFV-eat.AV=FUT fish
5.3.2 AN ARGUMENT FOR THE PRESENCE OF NULL PRONOUNS
In the examples above, I have basedmy analysis on the assumption that whenwe don’t see
an overt pronoun in the preverbal position (spec,CP), there actually is a null one present;
that is, Äiwoo has “pro-drop”, at least in this position. A brief description of null argu-
menthood in Äiwoo was given in chapter 2, §2.5.4. Since Äiwoo is a underresearched
language, and the available descriptive literature does not not consider the possibility of
null arguments (Næss 2006 et seq.), at this juncture I want to motivate this carefully. .
5.3.2.1 SHOWING THE EXISTENCE OF NULL PRONOUNS IN SPEC,CP
One “useful” aspect of Äiwoo grammar in this respect is what I have claimed to be its
V2 property (chapter 3): one nominal argument – and only one – must be present before
the verb. However, one reason why Äiwoo was never described as V2 in earlier literature
is that descriptively, verb-initial strings do occur – and in fact extremely frequently so,
e.g. (353a) in the versionwithout iude ‘12AUG’.We should then ask the question of whether
our V2 analysis is simply off track, or whether there is a null pronoun filling the spec,CP
position and thus “counting” for the V2 effect. I will argue for the second answer.
The crucial argument is based on how Äiwoo syntax behaves when the preverbal po-
sition cannot be filled by something. The main piece of evidence comes from reflexive
sentences. Although a full analysis of reflexivity and anaphors in Äiwoo lies outside the
scope of this work, I will use some of their word order properties as an argument. Äiwoo
does not have morphologically dedicated anaphors, and simply uses the standard 3MIN
pronoun inâ (or more generally, the ordinary pronoun for any φ-category, i.e. iu 1MIN
for ‘myself’, iumu 2MIN for ‘yourself’, etc.). In this respect, Äiwoo is like other languages
where anaphors and (non-reflexive) pronouns are not morphologically distinct, such as
Frisian, Old English, etc. (see e.g. Reuland & Reinhart 1995, van Gelderen 2000). How-
ever, despite this pronoun’s morphology being ordinary, its syntax is not. Unlike other
Chapter 5. The φ-indexing system 225
UV objects, this pronoun must be post-verbal to be interpreted as an anaphor (354a)10.
In contrast, if this same pronoun is fronted to the pre-verbal position (354b) – where UV
objects usually are – the sentence is grammatical, but only with a non-reflexive reading.
Presumably, the anaphor needs to be c-commanded by the subject to be bound and inter-
preted correctly, and thus must remain low11.
(354) Inâ must be postverbal when used as an anaphor: (250728)
a. John ki-lâbu-∅n=naa inâ go nuwoli eângâ
John IPFV-cut.UV-3MIN=FUT 3MIN with knife that
‘John will cut himself with that knife’
b. inâ ki-lâbu John=kaa go nuwoli eângâ
3MIN IPFV-cut.UV John=FUT with knife that
7 ‘John will cut himself with that knife’
3 ‘Johni will cut him/herj/*i with that knife’
Pay attention now to the position of the subject, ‘John’, in (354a). This sentence is in UV,
as indicated by the verb stem lâbu ‘cut.UV’ instead of lâbonge ‘cut.AV’. Nonetheless, we see
that the subject is in spec,CP (the preverbal position) instead of the object12. Crucially, the
subject is not left in its normal UV position, between the verb and the TAM particles (355).
From there, it would presumably still be able to c-command and thus bind the anaphor,
and yet, the subject in this position is actually ungrammatical.
(355) VSO is impossible: the preverbal position must be filled (250728)
* ki-lâbu John=kaa inâ go nuwoli eângâ
IPFV-cut.UV John=FUT 3MIN with knife that
Intended: ‘John will cut himself with that knife’
I interpret this as a consequence of Äiwoo’s V2 effect. By V2 rules, spec,CP must be filled
by a nominal. These sentences contain two nominals: ‘John’ and the anaphor inâ. The
anaphor is not a viable candidate for spec,CP, as it would be in the wrong position to be
bound and interpreted correctly. Therefore, something else will have to plug the preverbal
position to satisfy the V2 requirement. Given the absence of other viable candidates, that
10 The same also holds for other φ-categories than 3MIN:
(i) ( iu ) ki-lâbu-no=ngaa iu go nuwoli eângâ (250728)
1MIN IPFV-cut.UV-1MIN=FUT 1MIN with knife that
‘I will cut myself with that knife’
11 This judgment is plausibly explained by the anti-cataphora ban on CP-fronted constituents in non-AV
clauses; see chapter 3, fn. 43. However, I don’t have a theory of how it might come to be that an anaphoric
inâ is not raised to spec,CP, but a non-anaphoric one is. The model presented in chapter 3 would wrongly
predict that the anaphoric inâ should be raised just like any other object, resulting in the word order in
(354b), so it needs an amendment. Given the modular framework assumed here, we cannot directly state a
condition of the type “C raises XP, unless this would result in XP not being interpreted correctly”. There-
fore, we would need to rather assume that anaphoric and non-anaphoric inâ, despite being morphologically
identical, are structurally/featurally different, so that only the non-anaphoric one can be raised by C. I leave
this issue to future research.
12 Other cases of UV clauses with SVO word order will be discussed in depth in chapter 6.
226 5.3. The φ-prefixes: obligatory co-occurrence
will – exceptionally – have to be the subject. We can also make sure that there is no
problem per se with the subject being in spec,TP (between the verb and =TAM), and in
fact this provides further support to our argument. If we add a third nominal argument
by using CV (356) (see Appendix A for details on this construction), now we have filled
spec,CP and satisfied the V2 requirement, and the subject is allowed to stay in its canonical
position, from where it can bind the anaphor.
(356) If the preverbal position is plugged by another DP, the subject can stay low:
nuwoli eângâ ki-lâbu John=kaa=kä inâ (250728)
knife that IPFV-cut.UV John=FUT=CV 3MIN
‘With that knife John will cut himself’
I take the ungrammaticality of (355) to be a strong argument for the presence of null
pronouns in spec,CP in sentences that are verb-initial on the surface, e.g. (352b)-(353a)
above. In the case of reflexive constructions, the nominal that “should” be raised to the
preverbal position (a UV object) cannot do so for independent reasons, so another nominal
will have to be raised instead, seemingly just to ensure that that position is filled. If those
null subject pronouns I notated in the AV sentences above were not present, we should
then expect some other nominal to be raised exceptionally (i.e., the object), but this is
impossible (357).
(357) “No null subject pronouns” wrongly predicts that an AV object should raise:
* sii de-ku-tou-mä=kaa tO (250728)
fish 12AUG-IPFV-bring.AV-DIR1=FUT
Intended: ‘We will bring fish’ (only grammatical order: dekutoumäkaa sii)
5.3.2.2 THE POSITION OF THE SUBJECT PRONOUNS
Having argued for the presence of a null pronoun filling spec,CP in (352b)-(353a), let’s
now do the same thing for the overt pronouns. We can show these are in fact in spec,CP
by testing their order with respect to certain adverbials. As shown in §3.2, adverbials like
‘yesterday’, ‘tomorrow’, ‘every day’, etc. can only precede the preverbal argument (the
VIA), and not intervene between the VIA and the verb. If an argument precedes these
adverbs, then it must be marked as a topic by either a prosodic break and/or a deictic
particle =Ce/=Câ (proximal/distal).
Consider now the following contrast. When the adverbial dâbu dâuwângâ ‘every day’
precedes the pronoun iu ‘I’, the latter can bemarked as a topic (358a) or left bare (358b); this
last sentence has at least one available structural parse where the pronoun is in spec,CP.
However, when the adverbial follows the pronoun, the pronoun must be marked as a topic
(359a), and not doing so is ungrammatical (359b).
(358) Adv > pronoun: the pronoun can but does not need to be marked as a topic
a. dâbu dâuwângâ iu=nge ∅1MIN i-ku-wânubo-lâoo sii
day every 1MIN=PROX 1MIN-IPFV-kill.AV-always fish
‘Every day, me here, I always catch fish’ (230707)
Chapter 5. The φ-indexing system 227
b. dâbu dâuwângâ iu i-ku-wânubo-lâoo sii
day every 1MIN 1MIN-IPFV-kill.AV-always fish
‘Every day I always catch fish’
(359) Pronoun > adv: the pronoun must be marked as a topic
a. iu=nge dâbu dâuwângâ ∅1MIN i-ku-wânubo-lâoo sii
1MIN=PROX day every 1MIN-IPFV-kill.AV-always fish
‘Me here, every day I always catch fish’
b. * iu dâbu dâuwângâ i-ku-wânubo-lâoo sii
1MIN day every 1MIN-IPFV-kill.AV-always fish
The logic of the argument is the following. The adverbial dâbu dâuwângâ ‘every day’
cannot intervene between the VIA and the verb. Therefore, if a nominal precedes the
adverbial, it must be a left-peripherical topic, and must be appropriately marked as such
– that’s why (359b) is unacceptable. The possibility of having an unmarked pronoun to
the right of this adverbial (358b) confirms that this pronoun can be in the canonical VIA
position, that is, spec,CP. (Under this analysis, (358a)–(359a) contain a ∅1MIN pronoun
between the adverbial and the verb, filling spec,CP, as notated here).
In the available literature, cases of cooccurrence of a subject pronoun and a verbal
prefix had already been documented on the basis of naturally attested examples (Næss
2015b: 292). However, it could not be established whether those pronouns were actual
VIAs (in spec,CP) or left-peripheral topics, as controlled elicited data and negative ev-
idence was unavailable. Moreover, it was not known that VSO order is ungrammatical
(see the discussion of reflexives above), and therefore we had no concrete evidence for
positing the presence of null subject pronouns.
Having showed that both null and overt subject pronouns in AV are in the canonical
VIA position, the alternation between null vs. overt pronouns (352b)-(353a) is then com-
pletely identical to what we observe in a better-studied pro-drop language like Italian.
The verb must always carry subject agreement (-iamo 1PL), and a subject pronoun is only
optionally realized (360).
(360) ∅1PL/noi and-iamo spesso a Chicago (Italian)
we go-1PL often to Chicago
‘We often go to Chicago’
5.3.3 SUMMARY: THE φ-PREFIXES ARE AGREEMENT MARKERS
In the previous section I showed that the prefixal series of φ-indexers is always obliga-
tory in the relevant contexts (that is, when it can appear in the first place: AV verbs and
intransitives). Even in the presence of an overt subject argument, whether a lexical DP
or a pronoun, the prefixes cannot be dropped. This supports the analysis I developed/
pre-figured in chapter 3, arguing that these prefixes are agreement markers.
Let us briefly remind ourselves again of how exactly it works. The derivation I pro-
posed for AV clauses in chapter 3 is schematized in (361); intransitives work in the same
228 5.4. The φ-suffixes
way, only with no object. At the last step, C copies the φ-features of the argument it raises
(see §3.5.3 for howC comes tomove the object and not the subject in UV). This is not really
any different from what happens in e.g. Italian, although on C rather than on T.
(361) [ DP[PIV] C ⟨φ ⟩ [ tS T [ tS v [V O]]]]S
At Vocabulary Insertion, the φ-features on C are realized as the prefixal series (362)13.
(362) a. C⟨1MIN⟩ ⇔ i- e. C⟨1AUG⟩ ⇔ me-
b. C⟨12MIN⟩ ⇔ ji- f. C⟨12AUG⟩ ⇔ de-
c. C⟨2MIN⟩ ⇔ mu- g. C⟨2AUG⟩ ⇔ mi-
d. (C⟨3MIN⟩ ⇔ ∅) h. C⟨3AUG⟩ ⇔ li~lu-
The fact that prefixal agreement is not realized at all in UV will be analyzed in §5.5.2 as a
result of a rule of Impoverishment, triggered by the presence of φ-features on v; see that
section for the full discussion. For the idiosyncratic linear placement of the 3AUG prefix
li~lu-, see chapter 3, §3.5.5. Moreover, for the allomorphic interactions between 1MIN i-
and the modal prefixes nä- (irrealis) and de- (apprehensive), see chapter 2, §2.4.
5.4 THE φ-SUFFIXES
I proceed now to discuss the properties of the suffixal series of φ-indexers. The main
syntactic difference between these and the prefixal markers is a near-opposite pattern of
co-occurrence with arguments. The prefixes always co-occur with a coreferent nominal
(which may be either overt or null). In contrast, in (§5.4.1) we will see that the suffixes can-
not co-occur with any nominal in spec,TP –whether coreferent or not. However, a φ-suffix
can co-occur with a coreferent argument, but only as long as this is in a higher position. I
will argue that despite the fact that these markers are suffixes at the morphophonological
level, their distribution indicates their syntactic nature not as agreement markers, but as
phonologically reduced pronouns (§5.4.2).
The long-story-short version of the argument goes as follows: (i) the φ-suffixes are in
complementary distribution with any nominal in the same slot, therefore they must fill a
specifier position, therefore they must be pronouns; (ii) surprisingly if they are pronouns,
however, they can co-occur with a coreferent argument in a higher position; (iii) therefore,
in these cases, the φ-suffixes must represent the obligatory spell-out of a lower copy (in
spec,TP) of the argument. In the next section (§5.5), we will see how an analysis of the
morphology of Äiwoo pronominal forms will also help us account for this pattern of lower
copy spell-out.
13 For the time being, I’m not making particularly sophisticated assumptions about how person and number
are represented in the syntax. The featural notation in (362) is purely meant as a simplified descriptive
shorthand. At this point, any theory of φ-features that can deal with a minimal-(unit-)augmented system
could be plugged in, be it privative (Harley & Ritter 2002, McGinnis 2005, a.o.) or binary (Nevins 2007,
Harbour 2016, a.o.). In chapter 6, §6.5.1.1 I’ll propose a more concrete theory of φ-features to model the
complex hierarchy effect connected to movement to spec,TP.
Chapter 5. The φ-indexing system 229
5.4.1 DISTRIBUTION: CO-OCCURRENCE OF φ-SUFFIXES AND ARGUMENTS
In this section I will describe the distribution of φ-suffixes. Their placement is constant:
they are suffixes on UV verbs. However, what is interesting is when they can and cannot
co-occur with different arguments in different positions. The picture we will arrive to is
summarized schematically in (363)–(364). On the one hand, wewill first see that φ-suffixes
(represented as “-φS/O”) are completely incompatible with a nominal occupying spec,TP.
In this position there can be either a φ-suffix or an argument (363a–c), but not both (363d,e)
– regardless of whether the two are coreferent or not (I don’t represent here every possible
combination, but only the “more plausible” ones, i.e., minimal variants of actually attested
structures). On the other hand, a φ-suffix can co-occur with an overt coreferent argument
as long as this is in a higher position (364). This happens in two non-canonical types of
UV clauses, whose syntax will be the focus of the next chapter.
(363) A φ-suffix is incompatible with any argument in spec,TP:
a. Obj V Subj=TAM
b. Obj V-φS =TAM
c. Obj V-φO =TAM Subj
d. *Obj V-φS Subj=TAM
e. *Obj V-φO Subj=TAM
(364) A φ-suffix can co-occur with a coreferent argument in spec,CP:
a. Obj V-φO =TAM Subj
b. Subj V-φS =TAM Obj
NO COOCCURRENCE 1: A φ-SUFFIX AND A CO-REFERENT ARGUMENT First, let us see
how a φ-suffix is in complementary distribution with a coreferent (overt) argument in
the same position. In the slot between the verb and the TAM particles, one can have an
overt subject DP (365a) or the suffixal marker -i 3AUG (365b), but crucially not both (365c).
Having neither is also impossible (365d).
(365) Cooccurrence of a φ-suffix and a subject DP in spec,TP is impossible: (250728)
a. sii ku-tu-mä mikilitei =kaa
fish IPFV-bring.UV-DIR1 fishermen=FUT
‘The fishermen will bring the fish’
b. sii ku-tu-mä -i =laa
fish IPFV-bring.UV-DIR1-3AUG=FUT
‘They will bring the fish’
c. * sii ku-tu-mä -i mikilitei =kaa
fish IPFV-bring.UV-DIR1-3AUG fishermen=FUT
d. * sii ku-tu-mä =Caa  
fish IPFV-bring.UV-DIR1- =FUT
(Only interpretable as containing a 3MIN subject, with =naa as the form of
the future marker: ku-tu-mä-∅n=naa ‘S/he will bring the fish’)
230 5.4. The φ-suffixes
This obviously contrasts with the obligatory co-occurrence pattern we saw for the prefixal
markers. The 3AUG case is repeated in (366), where the prefix lu- must co-occur with the
overt subject mikilitei ‘fishermen’.
(366) Cooccurrence of a φ-prefix and a subject DP is obligatory: (250728)
mikilitei ku- *(lu-) tou-mä=kaa sii
fishermen IPFV-*(3AUG-)bring.AV-DIR1=FUT fish
‘The fishermen will bring fish’ (only good with lu-)
When the subject is pronominal instead of a lexical DP, we observe that the suffixal φ-
marker is obligatory (367a,b). Interestingly, we also see that using a standalone pronoun in
this position is actually impossible, whether in addition to the suffixal marker or instead of
it (367c). This holds across all person/number categories. Two more examples are shown
for 3rd person pronominal subjects in (368)-(369), respectively 3AUG and 3MIN.
(367) a. sii ki-ngä -de =ngaa (250728)
fish IPFV-eat.UV-12AUG=FUT
‘We.INCL will eat the fish’
b. * sii ki-ngä =Caa
fish IPFV-eat.UV =FUT
(Only interpretable as having a 3MIN subject, with =naa: ki-ngä-∅n=naa)
c. * sii ki-ngä(-de) iude =ngaa
fish IPFV-eat.UV(-12AUG) 12AUG=FUT
Intended: ‘We.INCL will eat the fish’
(368) a. sii ki-ngä -i =laa (250728)
fish IPFV-eat.UV-3AUG=FUT
‘They will eat the fish’
b. * sii ki-ngä(-i) ijii =laa
fish IPFV-eat.UV(-3AUG) 3AUG=FUT
Intended: ‘They will eat the fish’
(369) a. sii ki-ngä -∅n =naa (250728)
fish IPFV-eat.UV-3MIN=FUT
‘S/he will eat the fish’
b. * sii ki-ngä(-∅n ) inâ =naa
fish IPFV-eat.UV(-3MIN) 3MIN=FUT
Intended: ‘S/he will eat the fish’
NO COOCCURRENCE 2: A φ-SUFFIX AND A DIFFERENT ARGUMENT A φ-suffix is incom-
patible with a nominal in this position even when the two realize/index different argu-
ments. This can be observed in a particular kind of UV clauses where the φ-suffix on
the verb indexes the object’s features, instead of the subject’s (370)14. The φ-suffix here is
14 See chapter 6, §6.3.2 for details about these clauses and the morpheme -gu.
Chapter 5. The φ-indexing system 231
obligatory, and cannot be dropped. Crucially, the subject John surfaces in a lower posi-
tion to the right of the TAM particles (370a), and it cannot be in its normal UV position
between the verb and the TAM particles – where the object suffix is (370b).
(370) If an object φ-suffix is present in spec,TP, the subject must be low: (240307)
a. (iude) ku-tu-kä -gu-de =ngaa John ngä skul
12AUG IPFV-bring.UV-DIR3-OBJ-12AUG=FUT John to school
‘John will bring us to school’
b. * (iude) ku-tu-kä -gu-de John =kaa  ngä skul
12AUG IPFV-bring.UV-DIR3-OBJ-12AUG John=FUT to school
POSSIBLE CO-OCCURRENCE: A φ-SUFFIX AND A HIGHER COREFERENT ARGUMENT A
second interesting aspect of this type of UV sentences is that the object pronoun in the
preverbal position can optionally be overt (370a). In that case, what we have is a φ-suffix
co-occurring with a coreferent argument, but the argument in question is in a higher
position (spec,CP). As mentioned, this “doubling” only takes place in non-canonical UV
clauses (see chapter 6 for their syntax). It happens for objects in UVgu clauses (371a) (re-
peated from above with different boxing), and for subjects in UVSVO clauses (371b). Note
that also here, the φ-suffixes cannot be dropped.
(371) A φ-suffix can co-occur with a coreferent argument if this is higher: (250728)
a. (iude) ku-tu-kä -gu-de =ngaa John ngä skul
12AUG IPFV-bring.UV-DIR3-OBJ-12AUG=FUT John to school
‘John will bring us to school’
b. (ijii) ku-tu-kä -i =laa iude ngä skul
3AUG IPFV-bring.UV-DIR3-3AUG=FUT 12AUG to school
‘They will bring us to school’
INTERIM SUMMARY The distribution of φ-suffixes is repeated in (372)–(373). They are
in complementary distribution with any nominal occupying this position, even if it is a
different argument (373). Descriptively, there seems to be one and only one “slot” between
the verb and the TAM particles, and only one argument – whether the subject or the
object – can occupy this position15. However, the φ-suffixes can co-occurwith a coreferent
argument, but only as long as this is in a higher position (373).
(372) A φ-suffix is incompatible with any argument in spec,TP:
a. Obj V Subj=TAM
b. Obj V-φS =TAM
c. Obj V-φO =TAM Subj
d. *Obj V-φS Subj=TAM
e. *Obj V-φO Subj=TAM
15 There will be one single very restricted exception to this generalization, where two arguments are realized
in spec,TP simultaneously. This is the case of UV1>2 clauses, discussed in chapter 6, §6.3.4.
232 5.4. The φ-suffixes
(373) A φ-suffix can co-occur with a coreferent argument in spec,CP:
a. Obj V-φO =TAM Subj
b. Subj V-φS =TAM Obj
5.4.2 THE φ-SUFFIXES ARE PRONOUNS
Let us consider again the analytical options we have for the Äiwoo φ-suffixes (374).
(374) Does it co-occur Does it occupy a
with an argument? specifier position?
Agreement: yes no
(Narrow-syntactic) clitics: might no
Pronouns: no* yes
*Modulo the possibility of multiple copy spell-out.
Given the complementary distribution facts in spec,TP (372), we can exclude the possibil-
ity that the φ-suffixes are pure agreement markers. An agreement analysis would need to
stipulate that the relevant probe only agreeswith null arguments (as has been proposed for
similar complementarity patterns in Irish and other Celtic languages; McCloskey & Hale
1984, Stump 1984, Legate 1999, Jouitteau & Rezac 2006). This, of course, would predict
that an agreed-with argument should indeed be null. However, this prediction is crucially
falsified by those cases where a φ-suffix can co-occur with an overt coreferent argument
as long as this is in a higher position (373)16. The analysis would need to be amended to
something like “agreement can only take place with null arguments, unless this argument
is then raised to a higher position, in which case it can (optionally) be overt again”. This is
a rather odd and cumbersome description, and hard to implement within the framework
adopted here. Moreover, it intuitively seems to be missing an important generalization:
cooccurrence between a φ-suffix and a coreferent argument is only possible when the two
are in different syntactic positions.
Having excluded an agreement analysis of the φ-suffixes, we are now left with two
options: either they are pronouns, or they are (narrow-syntactic) clitics. Here is where
the criterion of whether a φ-indexer occupies a specifier position becomes useful. Before
we move on, let us remind ourselves that regardless of what kind of analysis we want to
adopt for the φ-suffixes, it is independently true that the position between the verb and
the TAM particles can host what is uncontroversially a fully phrasal nominal constituent.
As shown in chapter 3, §3.5.2, an argument in this position can be of any phonological
size (375). Hence, there must exist a structural specifier position that is mapped out to this
linear position.
16 Interestingly, also in Irish, the complementarity between agreement suffixes and coreferent arguments
is also obviated by A-movement (Perry 2025 and references therein). However, it is not obviated by Ā-
movement. In Äiwoo, either kind of movement can result in the relevant kind of “doubling” pattern.
Chapter 5. The φ-indexing system 233
(375) The V =TAM position can host a phrasal nominal argument, of any size:
a. gino-mu ba ku-mangi [kuli]DP=gu=naa (250728)
son-2MIN NEG IPFV-bite.UV dog=NEG=FUT
‘The dog won’t bite your son’
b. gino-mu ba ku-mangi [kuli (mi=)[ki-ngoduwâ-lâoo ngä paveli
son-2MIN NEG IPFV-bite.UV dog REL=IPFV-shout-always in garden
eângâ]]DP=gu=naa
that=NEG=FUT
‘The dog that’s always barking in that garden won’t bite your son’ (241015)
As presented in §5.1.2, a clitic – by definition – does not occupy a specifier position,
whereas a pronoun does. This correctly predicts what happens in a language like Ital-
ian when pronominal objects cliticize. As we see in the data repeated below, because the
cliticized object la ‘her’ does not occupy a specifier, this position is still available for the
subject Ana to move to (376a). If la was a full pronoun, in contrast, it would fill up a spec-
ifier, and therefore – modulo the possibility of multiple specifiers – we might expect that
cliticizing the object would force the subject to stay low, contrary to fact (376b).
(376) Italian object clitics do not occupy a specifier position:
a. [TP Ana la= porta [vP tS tO domani
Ana 3SG.F=brings tomorrow
‘Ana will bring her tomorrow’
b. * [TP la= porta [vP Ana tO domani
3SG.F=brings Ana tomorrow
(Ungrammatical with information-structurally neutral prosody; see fn. 6)
Strikingly, what is ungrammatical in Italian (376b) is precisely what we observe in Äiwoo.
As we saw, when the object is indexed by a φ-feature on a UV verb, the subject cannot
raise to its usual position (in spec,TP), and must remain low instead. The minimal pair is
repeated in (377): ‘John’ must be to the right of the future marker, and not in its usual
position (to its left).
(377) If an object φ-suffix is present in spec,TP, the subject must be low: (240307)
a. (iude) ku-tu-kä-gu-de=ngaa John ngä skul
12AUG IPFV-bring.UV-DIR3-OBJ-12AUG=FUT John to school
‘John will bring us to school’
b. * (iude) ku-tu-kä-gu-de John =kaa  ngä skul
12AUG IPFV-bring.UV-DIR3-OBJ-12AUG John=FUT to school
Thus, the diagnosis is that φ-suffixes in Äiwoomust occupy a specifier position: they are in
a positionwhere fully phrasal nominal constituents can be, and they are in complementary
distribution with an argument in that same position (whether coreferent or not).
If we now consider our table summarizing the properties of agreement markers, clitics,
and pronouns again (374), we see that Äiwoo φ-suffixes show a somewhat mixed behav-
ior. On one hand, they clearly occupy a specifier position. This speaks in favor of them
234 5.4. The φ-suffixes
being pronouns, and not narrow-syntactic clitics17. However, we also see cases where a
φ-suffix co-occurs with a coreferent argument, which is not something pronouns typically
do (whereas clitic doubling patterns, of course, are familiar fare). However, importantly,
this kind of “doubling” only ever happens when the argument in question is in a higher
position. This peculiarity gives us a way to attack the puzzle. I propose that uniformly,
the φ-suffixes are pronouns: this explains why they occupy a specifier position, and thus
why multiple arguments cannot fill that position. In the “doubling” cases, I will argue that
the φ-suffixes are still pronouns, but they spell out a lower copy of movement (of either
the subject or the object), in spec,TP. Why this lower copy must be spelled out in this case
will become clear in the next section.
For the sake of explicitness, then: my analysis of the Äiwoo φ-suffixes as pronouns
treats the two sentences in (378) as syntactically identical, and only phonologically dif-
ferent. Their syntactic structure is shown in (379). In both cases, simply, the subject is in
spec,TP, even though Lisa contains the left edge of its own phonological word and the
12AUG prefix -de does not. Importantly, the phonological word spelled out as kupotaaden-
gaa /kupotàːndeŋáː/ does not neatly correspond to any syntactic constituent.
(378) a. Anna ku-potaa Lisa =kaa tS tO (250728)
Anna IPFV-search.UV Lisa=FUT
‘Lisa will look for Anna’
b. Anna ku-potaa -de =ngaa tS tO
Anna IPFV-search.UV-12AUG=FUT
‘We will look for Anna’
(379) Structure of (378a,b):
a. CP b. CP
Obj Obj
Anna Anna
C+Asp+V+v TP C+Asp+V+v TP
ku-potaa ku-potaa
IPFV-search.UV IPFV-search.UV
ω Subj
T AspP SubjLisa
=kaa -de T AspP
=FUT -12AUG =ngaa…
ω =FUTω …
If the suffix -de is really a full pronoun in spec,TP, the most obvious question concerns its
form: why is it -de and not iude, the usual form for a 12AUG (full) pronoun? This will be
17 This kind of analysis of agreement affixes as underlying pronouns has also been proposed for the com-
plementarity effects in Irish and Scottish Gaelic (Anderson 1982, Adger 2000, Ackema & Neeleman 2003,
Brennan 2009), Aleut (Yuan 2018), Washo (Arregi & Hanink 2022), and Sorani Kurdish (Akkuş et al. 2025).
Chapter 5. The φ-indexing system 235
the focus of the next section, where we will zoom in on the morphophonological aspects
of the mapping from the underlying syntactic structure to the surface form.
5.5 THE MORPHOLOGY OF ÄIWOO PRONOUNS
In the previous section, I have argued that the φ-suffixes on Äiwoo verbs, syntactically,
are pronouns in an argument position (spec,TP). This accounts for their complementary
distribution with other nominals in spec,TP. However, two outstanding questions remain:
(i) why do pronouns in this position obligatorily surface in a short/reduced suffixal form
and not the full/standalone form (e.g., -de 12AUG in (378b) rather than the full pronoun
iude)? (ii) for the “doubling” cases, why must a lower copy of movement of a pronoun in
this position obligatorily be realized?
I propose that the answer to both questions is not syntactic, but morphological in
nature: the full forms are simply the wrong spell-out for a pronoun in this position. De-
scriptively, Äiwoo pronouns come in two morphological shapes, a long/full one and a
short/reduced one – the φ-suffix. I suggest that a violable phonological/prosodic require-
ment imposes that a pronoun always become part of a larger phonological word to its
left whenever possible; this is when a pronoun is realized in its short form. This form of
“prosodic incorporation” will then force a this particular copy of movement to be spelled
out overtly, regardless of how the rest of the movement chain is pronounced. Wherever
this requirement cannot be satisfied, for example due to the pronoun’s syntactic posi-
tion, then the full form of the pronoun is inserted instead. All in all, the analysis I give
for Äiwoo φ-suffixes is akin of that proposed in Harizanov (2014) for Bulgarian clitics:
syntactically, they are a moved phrasal constituent in a specifier position, but they are
post-syntactically reduced to a shorter form.
5.5.1 DECOMPOSING PRONOUNS: SHORT AND LONG FORMS
The descriptive literature (Næss 2006, Ross & Næss 2007, a.o.) has already noted how
the full/standalone forms of Äiwoo pronouns are bimorphemic. All of them consist of a
(near-)constant base (mostly i~iu-, with one exception) and almost the same paradigm of
φ-suffixes that we just discussed, the one found on UV verbs (380)18. The only exceptions
18 There has been and still is some variation in the form of the “stem” of some of these full pronominal forms,
not shown in (380) for simplicity. It’s not clear if these forms are in dialectal/free variation (whether di-
achronically or synchronically), or what – if anything – governs the alternations. The 1MIN pronoun is
consistently iu, but when the =Ce proximal particle is attached, the result can be either iu=nge or, some-
what less commonly, i=nge. The 1AUG pronoun can be realized as either ingo, ingopu, iungo, iungopu; the
suffix can either be -ngo or -ngopu (or -ngee, but this has a different distribution; see the table footnote). The
2AUG pronoun is consistently imi; however, an alternative form iumi was reported in Tryon & Hackman
(1983: 290; ⟨yumi⟩ in the orthography used at the time). Furthermore, its unit-augmented form was listed
as both imile and iumile in field notes by Even Hovdhaugen in 2003 (transcribed respectively as ⟨imle⟩ and
⟨yumle⟩, with high vowel reduction targetting the /i/ before the -le suffix; Næss in prep). Finally, the 3AUG
236 5.5. The morphology of Äiwoo pronouns
are 1MIN iu, where the corresponding suffix -no is absent, and 3AUG, which shows a dif-
ferent base iji- (for the reasons behind this segmentation, see fn. 20). Although the 3MIN
pronouns inâ/ine (respectively distal and proximal) seem to be different, they can also be
analyzed as conforming to this pattern. These are in fact trimorphemic, consisting of the
base i-, the segmentally null 3MIN suffix -∅n, and the distal/proximal markers =Câ/=Ce
(which surface in their /n/-initial form because they follow the -∅n suffix): inâ is thus
segmentable as i-∅n=nâ (PRON-3MIN=DIST), and ine as i-∅n=ne (PRON-3MIN=PROX).
(380) Pronoun φ-suffix Pronoun φ-suffix
1MIN iu -no, -nee* 1AUG iu-ngo(pu) -ngo(pu), -ngee*
12MIN iu-ji -ji 12AUG iu-de -de
2MIN iu-mu -mu 2AUG i-mi -mi
3MIN inâ/ine -∅n 3AUG iji-i -i
* The forms -nee/-ngee for 1MIN/1AUG are only found preceding a 2nd person object suffix; see §6.3.4.
Another – stronger – argument for these pronominal forms being bimorphemic, other
than their morphological shape, is that the two hypothesized parts can actually be sepa-
rated. Specifically, the morpheme du ‘all’ can occur in between (381). This was reported
for 2AUG imi in (Næss 2018a: 34). With novel data, we can confirm that this behavior
extends to the other pronouns as well19,20.
(381) Du ‘all’ can separate the two parts of a full pronoun: (231026)
Unmodified pronoun: Modified pronoun:
1AUG iu-ngopu ‘We.EXCL’ iu-du-ngopu PRON-all-3AUG ‘All of us.EXCL’
12AUG iu-de ‘We.INCL’ iu-du-de PRON-all-12AUG ‘All of us.INCL’ 
2AUG i-mi ‘Y’all’ iu-du-mi PRON-all-2AUG ‘All of y’all’
3AUG iji-i ‘They’ iji-du-i PRON-all-3AUG ‘All of them’
Of course, minimal-number pronouns are semantically incompatible with ‘all’. However,
we can use other modifiers to show that these are also bimorphemic and separable, for
example eopu~epu ‘also’:
(382) iu-epu-mu Jises pe=Galilii eângâ ku-wobii-mu
PRON-also-2MIN Jesus REL:people=Galilee that IPFV-follow.UV-2MIN
‘You too were following Jesus of Galilee’ (MT 26:69)
pronoun is realized as either ijii or just jii; however, Wurm (1981: 84) also reports iujii (⟨yudyii⟩ in that
orthography), though my consultant does not use this form, and it is not attested in Næss’ corpus nor in
the more modern Bible translations.
19 The fact that 2AUG has i-mi but iu-du-mi is most plausibly phonological, involving metathesis and copying
of /u/; see chapter 2, §2.4.
20 The form ijidui ‘all of them’ is the reason behind analyzing 3AUG as having an exceptional pronominal stem
iji-. If the stem was just i- like for the other pronouns, we would expect *i-du-iji, contrary to fact.
Chapter 5. The φ-indexing system 237
To account for their bimorphemic character, I propose that Äiwoo pronouns have the
syntactic structure in (383), consisting of a lower layer where the φ-features reside (“φ(P)”;
I remain agnostic as to whether this is a terminal node or branching), and a higher layer
which I label πP. These specific labels are chosen for convenience, and nothing hinges too
much on them. I leave as an open question how exactly these labels may map onto the
projections/categories proposed by influential analyses of the inner structure of pronouns
(e.g. Cardinaletti & Starke 1999, Déchaine & Wiltschko 2002, a.o.). The only important
aspect for my analysis is the existence of two distinct structural layers21.
(383) Structure of an Äiwoo pronoun:
πP
π φ(P)
[12AUG]
I suggest that the π head is what is spelled out as the “stem” i~iu~iji-, and the φ-features
are consistently spelled out as the suffixal markers. In other words, the full form of a
pronoun (e.g. iu-de 12AUG) is the spell-out of both layers, whereas the short suffixal form
(-de 12AUG) is the spell-out of only the φ layer, with the π layer being null. The short form
will be inserted when the pronoun becomes part of a larger phonological word, whereas
the full form will be inserted when the pronoun cannot do so.
Let us see how this works concretely. For now, let’s just assume that pronouns in this
position become part of a larger phonological word, and temporarily sidestep the issue of
how or why (see §5.5.2). The π head undergoes allomorphy, conditioned by its position in
the phonological word22. If it is word-initial, it is spelled out as the base i~iu~iji- (384a)23.
Otherwise, when π is not word-initial, it is spelled out as ∅ (384b). Independently from
how π is realized, the φ portion of the pronoun receives a consistent spell-out (385). When
π is reduced to zero (when the pronoun – for some reason – becomes part of a larger
phonological word), φ will then remain the only overt part. This captures the alternation
between the short and long form of pronouns24.
21 In chapter 6, it will also become important for pronouns to have some kind of feature that lexical DPs don’t
share (e.g. [π]), so that pronouns are singled out as a featural natural class.
22 For now, it does not seem important that the conditioning factor for the allomorphy of π is word-initiality
as opposed to, say, being next to V specifically. This will become important in the next section.
23 This rule is a simplification. A more complex set of VI rules will be needed to account for the distribution
of the three variants of the pronominal base: iji- before 3AUG; iu- before 12MIN, 2MIN, optionally 1AUG, and
possibly 1MIN (see fn. 24); i- elsewhere (before 3MIN and 2AUG). See also fn. 18 for further morphological
variation in the form of the pronominal stem.
24 The only category where φ does not have a consistent morphological spell-out is 1MIN, which therefore
needs more complex VI rules (i). The 1MIN suffix for objects is -∅ŋ (ia); this only surfaces when π is spelled
out as -gu. For 1MIN subjects, the suffix -nee only appears when preceding a 2nd person object suffix (ib)
(see chapter 6, §6.3.4), while the default suffix form is -no (ic).
(i) a. φ[1MIN] ⇔ -∅ŋ / π[ACC]
238 5.5. The morphology of Äiwoo pronouns
(384) a. π ⇔ i~iu~iji- / #
b. π ⇔ ∅ (elsewhere)
(385) φ[12AUG] ⇔ -de
The only part these VI rules do not cover is the presence of the suffix -gu in all and only
the cases where the argument realized by a φ-suffix is an object25:
(386) ( iude) ku-potaa -gu-de =ngaa mekivaavee bulaape (250728)
12AUG IPFV-search.UV-OBJ-12AUG=FUT teacher tomorrow
‘The teacher will look for us tomorrow’
This is the only corner of Äiwoo grammar where we see any kind of overt morphological
reflex of “case”; full forms of all pronouns are always constant (e.g. iude 12AUG), regardless
of case or position. For concreteness, a simple implementation is to assume that an [ACC]
feature on a pronoun is targeted by a morphological rule of Fission (Halle 1997 et seq.)
that only applies in this context (387a)26. The ACC node is then spelled out as -gu (387b).
(387) a. Fission: φP[ACC, 12AUG] ⟹ φP / V
[ACC] φ[12AUG]
b. VI rules:
i. [ACC] ⇔ -gu / V
ii. φ[12AUG] ⇔ -de
We can find supporting evidence for this analysis of the φ-suffixes as a morphologically
shorter spell-out of full pronouns considering other environments where pronouns are
not linearly adjacent to verbs, and yet still appear in their short/suffixal form. This is the
case when a pronoun is the complement of a preposition. Consider (388a) as a baseline
example of a PP. When the complement of the preposition is a pronoun rather than a
lexical DP, only the shorter suffixal form -de is possible (388b), and the full form iude
b. φ[1MIN] ⇔ -nee / φ[ADDR]
c. φ[1MIN] ⇔ -no (elsewhere)
The 1MIN full pronoun form iu could be analyzed either as an unsegmentable portmanteaux spelling out
both π and φ, or – amending the VI rules above – as iu-∅ŋ (where the hyphen separates the pronominal
stem from the φ-suffix), or i-u; see also fn. 18. This also does not cover the variety of idiosyncratic/suppletive
1MIN forms found in the possessive paradigms; see Roversi (2025: appendix A).
25 More precisely, -gu appears when the argument in this position is not the same clause’s subject; see the
facts about CV in Appendix A.
26 I remain agnostic as to how exactly this case feature comes to be present on the pronoun, and where exactly
in the pronoun’s internal structure it resides. Any theory of morphological case assignment – via Agree by
functional heads (Vergnaud 1977/2008, Chomsky 2000, 2001), or Dependent Case (Yip et al. 1987, Marantz
1991) – should be compatible with this, with due adjustments.
Chapter 5. The φ-indexing system 239
(388c) is disallowed27,28. This parallels how the subject in (378b) can only be realized as
the suffix -de, and not as the full pronoun iude.
(388) Pronouns must surface in their short form as the complement of a preposition:
a. ngâgo Jon b. ngâgu-de c. * ngâg{u/o} iude
to John to-12AUG to 12AUG
‘To John’ (250728) ‘To us’ ‘To us’
Unlike in the context of verbs and arguments, where we have more confounding factors
and potential analyses (agreement, cliticization, etc.), there is no plausible syntactic differ-
ence between John in (388a) and the 12AUG pronoun in (388b); both are simply a comple-
ment to the preposition ngâgo. Therefore, the only reason why the pronoun occurs in its
short form in (388b) must be morphological in nature, and not syntactic. This is captured
by the Vocabulary Insertion rules proposed above: the pronoun ends up becoming part of
a larger phonological word (though exactly how remains to be seen), and therefore, the π
head is spelled out as ∅, and only the suffixal φ-portion remains overt.
5.5.2 SPELLING OUT A NON-CONSTITUENT
As mentioned above, this analysis entails that in many cases, a morphophonological verb
word in Äiwoo does not neatly correspond to any syntactic unit, but rather spells out
a syntactic non-constituent (Julien 2002, 2007b, a.m.o.). To give a concrete illustration,
the verb word in (389a) has the underlying syntactic representation in (389b). The whole
syntactic structure comprising two heads (C and T, with C being a complex head) and the
specifier between them (spec,TP) is mapped onto a single phonological word.
(389) a. ki-ngä-de=ngaa
IPFV-eat.UV-12AUG=FUT
‘We will eat (it)’
27 The final vowel alternation between ngâgo and ngâgu-de is purely phonological, and orthogonal to the issue.
28 Here the suffixal form of the pronoun is just -de 12AUG, and not -gu-de OBJ-12AUG. This could be explained
either by saying that (i) -gu is specifically a spell-out of accusative case, and prepositions assign a different
case, which remains morphologically null; or (ii) the segmentation of ngâgude (393b) should actually be
ngâ-gu-de ‘to-OBJ-12AUG’ (note that ngä~ngâ is an existing preposition in Äiwoo). If one goes for this latter
approach, one would have to explain the surface form of (388a) ngâgo Jon (though go Jon is a possible
alternative, with no apparent difference in meaning). I leave this as an open issue.
240 5.5. The morphology of Äiwoo pronouns
b. CP
…
C TP
C
πP
Asp
∅ IPFV √V v ⟨φ⟩
T AspP
EAT π φ(P) =FUT
[12AUG] …
ki-ngä-de=ngaa
IPFV-eat.UV-12AUG=FUT
/kiŋæ̀deŋáː/
Wemust then understand how andwhy exactly the πP constituent in spec,TP is reduced to
its short form and not mapped onto its own phonological word, that is, why we don’t get
*ki-ngä iude=ngaa (where iude is the full form of the 12AUG pronoun). For the purposes
of this work, in what follows I will limit the discussion to how the (sub-)word ki-ngä-
de is formed, to the exclusion of the phonological cliticization of the future particle =Caa.
Although the TAM enclitics are counted as part of the phonological word for the purposes
of stress assignment (see chapter 2, §2.4), they show signs of lesser morphophonological
integration with their host compared to φ-suffixes: as described in §5.1.2.2, they do not
obligatorily trigger vowel harmony and they don’t trigger any stem allomorphy, while
φ-suffixes do both. I assume that their cliticization is a more purely phonological process,
and not as tightly interwoven with the morphosyntax as is the case for the φ-suffixes.
Therefore, I will largely exclude them from the discussion below.
Here is a sketch of a proposal to account for this phenomenon. I argue that the reason
why a pronoun in spec,TP surfaces in its short form rather than the long one is the out-
come of a stipulated (violable) morphophonological/prosodic constraint, which imposes
that pronouns become part of larger phonological words to their left if possible. Below, we
will see how the interaction of this posited constraint and the Stray Affix Filter (Lasnik
1981, 1995, Baker 1988, Landau 2006) also helps us explain the cases of “doubling” dis-
cussed above, where a φ-suffix co-occurs with a coreferential overt argument in a higher
position.
First, let us state our constraint (390). I assume this to be an Optimality Theory-style
constraint operating at the interface between narrow syntax and prosody, regulating how
hierarchical syntactic structures are mapped onto surface phonological strings (c.f. Ben-
nett et al. 2016, a.m.o., or Bennett & Elfner 2019 for a recent overview).
Chapter 5. The φ-indexing system 241
(390) *(ω πP): A πP cannot be at the left edge of a phonological word.
In the spirit of OT, I assume that this constraint is violable: it ought to be satisfied, but if
for some reason it cannot, the outcome will still be a licit string. I remain rather agnostic
as to how exactly the calculus should be formalized. For concreteness, I assume that the
syntax can generate a variety of candidates that may or may not violate the constraint,
and these then enter a OT-style competition. The other necessary assumption, which is
relatively standard, is that a syntactic head, including complex ones, is what is – at least
by default – mapped onto a phonological word (Embick & Marantz 2008, Bobaljik 2012,
a.o.).
Adopting Harizanov’s (2014) analysis of clitic doubling in Bulgarian, I propose that the
strategy Äiwoo uses to avoid a violation of *(ω πP) is to apply a post-syntactic operation
that will make sure the πP becomes part of a larger (morpho)syntactic constituent, reat-
taching to its left. This kind of operation could be an instance of Local Dislocation (Embick
2007), which itself was proposed as a variant of M(orphological)-Merger (Marantz 1988,
Embick & Noyer 2001, Matushansky 2006, a.o.). The structural description for the kind of
Local Dislocation we need is given in (391). The constituent ZP, in spec,YP, is reattached
to form a complex head with the head immediately to its left (X0).
(391) Local Dislocation:
XP ⟹ XP
… …
X0 YP
0 YPX ZP
ZP
Y0 … ZP Y0 …
Crucially, this is a post-syntactic operation, different from movement/external merge. As
such, it does not leave a trace/lower copy; the locally dislocated constituent is simply
not present anymore in the original position (spec,YP), as indicated by the strike-out and
dashed tree branch. This, in conjunction to the assumption that the so-formed X0+ZP con-
stituent is mapped onto one morphophonological word, correctly predicts that an Äiwoo
pronoun in certain positions will be realized in its short form and not in its long form.
Before proceeding, note that for Harizanov (2014) this combination of syntactic config-
uration and post-syntactic Local Dislocation/M-merger is what gets spelled out in Bulgar-
ian as clitics. However, as mentioned in the introduction to this chapter (§5.1.2), I refrain
from using the term ‘clitic’ for the Äiwoo φ-suffixes, as I find it might be potentially con-
fusing: in the narrow syntax, they are simply pronouns, and after Local Dislocation, they
are spelled out as affixes, not morphophonological clitics.
The concrete illustration of how this process works for a pronoun in spec,TP is given
in (397). After Local Dislocation, the πP that was in spec,TP in the underlying syntactic
242 5.5. The morphology of Äiwoo pronouns
structure is reattached to the complex C head; this complex constituent is then spelled
out as one single morphophonological word (as shown by the parentheses)29. Now, the π
head is not word-initial, and therefore it is realized as null. As a result, the φ(P) portion is
the only part of the pronoun that remains overt.
(392) Applying Local Dislocation to the tree in (389b):
CP
…
TP
C πP πP
T AspP
C π φ(P) =FUT
Asp …
IPFV √V v ⟨φ⟩
ω EAT ∅
ki- ngä -de =ngaa
Because this is a process at the morphology/prosody interface, it can apply even if the
potential host to the left of a pronoun is not a verb; for example, it could be a preposition.
Remember that a pronoun in the complement of a preposition is obligatorily realized in
its short form (393b), and the full form is impossible (393c). After Local Dislocation applies
(394b), the π head is no longer word-initial, and therefore is realized as null.
(393) Pronouns must surface in their short form as the complement of a preposition:
a. ngâgo Jon b. ngâgu-de c. * ngâg{u/o} iude
to John to-12AUG to 12AUG
‘To John’ (250728) ‘To us’ ‘To us’
29 As mentioned above, I assume that the future particle =ngaa is phonologically encliticized to its left later in
the derivation, without first becoming part of a morphosyntactic constituent.
Chapter 5. The φ-indexing system 243
(394) a. Underlying structure of (393b): b. After Local Dislocation:
PP PP
P πP
πP
P πP
π φP
π φP
ω
ngâgu ∅ -de
What is going on, then, when an Äiwoo pronoun actually surfaces in its full form? I pro-
pose that these are contexts where Local Dislocation cannot apply, due to the syntactic
position of the pronoun. Äiwoo pronouns surface in their full forms in two positions: for
both subjects and objects in the preverbal position/spec,CP (395a,b), and for objects in the
postverbal/post-TAM position of UVSVO clauses (395b). I assume that the postverbal po-
sition in (395b) is the same as that of an object in AV, that is, the base-generated position
of an object (complement of V; see chapter 3, §3.5, and chapter 4, §4.2.3).
(395) a. Full pronouns in spec,CP: (250728)
iude ku-potaa-gu-de=ngaa mekivaavee
12AUG IPFV-search.UV-OBJ-12AUG=FUT teacher
‘The teacher will look for us’
b. Full pronouns in postverbal position in UVSVO clauses:
ijii ku-potaa-i=laa iude
3AUG IPFV-search.UV-3AUG=FUT 12AUG
‘They will look for us’
I assume that these are both cases where a pronoun has nothing to its left to undergo
Local Dislocation onto. For a pronoun in spec,CP, this is because there simply is no other
material to its left (and if there is, it will be part of a different phasal spell-out). The same
applies to a pronoun in the complement of V: the only potential material to its left would
be other lower copies of movement (which we need to assume cannot be hosts for Local
Dislocation, somehow), and the closest non-moved element would be T, which again is
part of a higher phase. I leave a full development of an implementation of this process as
an open issue for future research.
Let us now discuss the “doubling” cases, where the φ-suffix is obligatory even when
the higher copy of the pronoun (in spec,CP) is pronounced. In our analysis, this means
that the lower copy of the pronoun in spec,TP, for some reason, must be pronounced. I
follow Yuan (2025) in assuming that Local Dislocation (andM-merger, in her proposal) can
interact with the Stray Affix Filter (Lasnik 1981, 1995, Baker 1988). In its original formula-
tion, the Stray Affix Filter required that the host of affixal material be pronounced overtly.
244 5.5. The morphology of Äiwoo pronouns
As developed by Bobaljik (1995, 2002), Nunes (2004), and Landau (2006), this requirement
could then prevent a movement copy from being deleted, or in other words, force a (lower)
copy to be pronounced overtly. Once the copy of the pronoun in spec,TP is Locally Dislo-
cated onto C, it is then forced to be pronounced overtly despite originally being a lower
copy of movement, which the algorithm governing chain resolution would typically have
deleted. This can then result in the attested doubling pattern, where the copy in spec,CP
may optionally be pronounced overtly30, while the one in spec,TP must be pronounced
due to the Stray Affix Filter. This is conceptually similar to cases reported by Yuan (2025)
in Inuit languages where it seems that an incorporated DP is still syntactically active, and
can undergo movement.
The fact that a pronoun in spec,TP ends up becoming part of the same morphological
word as the verb in the post-syntax makes correct predictions with respect to two more
aspects of Äiwoo verbal morphology. The first is the fact that despite being pronouns at
the syntactic level, φ-suffixes are morphophonological affixes (§5.1.2.2): they can trigger
vowel harmony, they count for stress assignment, and they can even trigger suppletive
allomorphy on certain verbal stems. For example, in possessive constructions, 1MIN and
3MIN forms are often irregular/suppletive (§5.1.2.2). In Roversi (2025b) I argue that the
possessor is the subject in spec,TP of a relative clause modifying the possessum, and the
possessive classifiers are really just transitive UV verbs. Instead of showing the usual φ-
suffix -no for 1MIN, possessive classifiers often show idiosyncratic allomorphy patterns, for
example with stem alternations. For example, the food-related possessive usually surfaces
as a stem na- followed by the relevant φ-suffix (396a), but its 1MIN form is nugo rather than
*na-no (396b)31.
(396) a. [sii [RC na{-mu/-∅n /-i}]] (250728)
fish POSS.FOOD.UV{-2MIN/-3MIN/-3AUG}
‘{Your/his/her/their} fish’; lit. ‘The fish [that {you/s/he/they} POSS.as.food]’
b. [sii [RC nugo]]
fish POSS.FOOD.UV.1MIN
‘My fish’; lit. ‘The fish [that I POSS.as.food]’
This kind of allomorphy is not immediately predicted to be possible given the underlying
syntax in (389b), since the verb (in C) and the 1MIN pronoun (in spec,TP) do not form a
constituent. However, Local Dislocation creates a local enough domain for allomorphy to
be triggered (397), still under the assumption that the locality domain for allomorphy is
the complex head (Choi & Harley 2019). I leave the development of a full model of this
phenomenon to future research.
30 See chapter 2, §2.5.4 for the conditions on when pronominal arguments in Äiwoo can be null.
31 Interestingly, if the possessive verb has a modifier, the stem still shows its 1MIN allomorph, but now the
suffix -no is optional but preferred (i). I leave an analysis of this for future research.
(i) sii enge nugo, eä sii eângâ nugo-eopu?(-no ) (240909)
fish this POSS.FOOD.UV.1MIN and fish that POSS.FOOD.UV-also-1MIN
‘This fish is mine, and that fish is also mine’
Chapter 5. The φ-indexing system 245
(397) Applying m-merger to the tree in (389b):
CP
…
C TP
T …
C πP
π φP
C
Asp
IPFV √V v ⟨φ⟩
EAT
The fact that the pronoun becomes part of a complex head together with the verb helps
us account for a second, more crucial fact, something I have implicitly shown throughout,
but not commented on so far: the fact that on intransitive verbs and AV verbs there is
never any φ-suffix whatsoever. According to the syntactic derivation for Äiwoo clauses
developed in 3, the subject moves through spec,TP in these clauses too (398); intransitives
look the same, just without an object.
(398) [ DP[PIV] C ⟨φ ⟩ [ tS T [ tS v [V O]]]]S
Given the morphological steps proposed above, it is unclear why the copy of the subject in
spec,TP wouldn’t undergo Local Dislocation onto the verb in C, and thus be pronounced
as a φ-suffix, which is impossible (399). Instead, in AV and intransitive clauses, the linear
slot I have identified with spec,TP is consistently empty.
(399) No φ-suffixes on AV verbs and intransitives: (250728)
a. ( iude) de-ki-lâbonge (*-de) =kaa sii
12AUG 12AUG-IPFV-cut.AV-12AUG=FUT fish
‘We will cut fish’ (only good without -de)
b. ( iude) de-ki-obu (*-de) =kaa
12AUG 12AUG-IPFV-fall-12AUG=FUT
‘We will fall’ (only good without -de)
I propose that this fact may receive a morphological solution, in terms of Impoverishment.
The syntax (and early post-syntax) is kept uniform across all types of clauses: a pronoun
in spec,TP always undergoes Local Dislocation onto the verb in C, as described above.
However, what happens next varies. First, remember that a rule of Kinyalolo-style Im-
poverishment was how we derived the fact that AV and intransitive verbs carry prefixal
246 5.5. The morphology of Äiwoo pronouns
agreement, whereas UV verbs do not (chapter 3, §3.5.5). In any kind of clause, C agrees
with and copies the φ-features of the argument it moves (whichever that may be). How-
ever, only in UV and not in AV/intransitives, v also has copied φ-features. Therefore, in
UV the features on C undergo Impoverishment (400a), whereas in AV and intransitives
they survive, and get spelled out as prefixal agreement (400b).
(400) a. Undergoer Voice: C gets impoverished ⟹ no φ-prefixes
(i) Underlying syntax: [ C⟨φ⟩ … [ … v ⟨φ⟩ ]]
(ii) Impoverishment of C: 3 [ ∅ … [ … v ⟨φ⟩ ]]
b. Actor Voice: C does not get impoverished ⟹ φ-prefixes
(i) Underlying syntax: [ C⟨φ⟩ … [ … v ]]
(ii) Impoverishment of C: 7 [ C⟨φ⟩ … [ … v ]]
I assume that Local Dislocation precedes Impoverishment, which in turn precedes Vo-
cabulary Insertion, and that this evaluation is done bottom-up through the complex head
(Bobaljik 2012, a.o.). I suggest that after a pronoun (originally in spec,TP) is Locally Dis-
located onto the verb in C, this too can undergo another round of Impoverishment of the
same kind, sensitive to the presence of φ-features on C. Thus if the first round of Impov-
erishment applies, this will bleed the second round, as shown below. If C had been Impov-
erished (in UV), this will bleed Impoverishment on the pronoun, which will thus surface
overtly as a φ-suffix (401a). If, however, C was not impoverished and still has φ-features
on it, then it is the features on the pronoun (or the entire pronoun itself) that are deleted.
This is the case in AV and intransitive verbs, and therefore, we see no φ-suffix (401b). This
derives the complementary distribution of φ-prefixes and suffixes, which never co-occur.
(401) a. Undergoer Voice:
C gets impoverished, φ does not ⟹ no φ-prefixes, yes φ-suffixes
(i) After Local Dislocation: [[ C⟨φ⟩ … [ … v ⟨φ⟩ ]] φ ]
(ii) Impoverishment of C: 3 [[ ∅ … [ … v ⟨φ⟩ ]] φ ]
(iii) Impoverishment of φ: 7 [[ ∅ … [ … v ⟨φ⟩ ]] φ ]
b. Actor Voice:
C does not get impoverished, φ does ⟹ yes φ-prefixes, no φ-suffixes
(i) After Local Dislocation: [[ C⟨φ⟩ … [ … v ]] φ ]
(ii) Impoverishment of C: 7 [[ C⟨φ⟩ … [ … v ]] φ ]
(iii) Impoverishment of φ: 3 [[ C⟨φ⟩ … [ … v ]] ∅ ]
Admittedly, the morphological story sketched here is somewhat ad-hoc, and relies on
highly specific Impoverishment rules triggered within arbitrary domains. For example,
the Impoverishment rule for C makes reference to v, which is not structurally or linearly
adjacent to it but is in the same complex head. However, for some reason the Impov-
erishment rule for the pronoun is only sensitive to C, and not to v further down. For
similar interactions between different Impoverishment rules within the Algonquian verb,
see Sun & Grishin (2025).
Chapter 5. The φ-indexing system 247
5.6 CHAPTER SUMMARY
We can now take stock and remind ourselves of the conclusions reached so far. First, we
have shown that all φ-indexers in Äiwoo – both prefixes and suffixes – are morphophono-
logical affixes. However, the prefixes and the suffixes are syntactically of different nature.
The former constitute “pure” agreement markers, as evidenced by their obligatory co-
occurrence with overt DP arguments. The latter, on the contrary, are actually pronouns
occupying spec,TP, and they are only post-syntactically reduced to suffixes. Table 5.1 gives
an overview of the differences between how φ-indexing works in AV and UV. Finally, ta-
ble 5.2 shows the complete paradigms for φ-prefixes, suffixal pronouns, and full pronouns.
Table 5.1: Morphosyntactic differences between φ-indexing in AV and UV.
Actor Voice Undergoer Voice
Position: Prefix Suffix
i-ki-vängä=kaa sii sii ki-ngä-no=ngaa
1MIN-IPFV-eat.AV=FUT fish fish IPFV-eat.UV-1MIN=FUT
‘I will eat fish’ ‘I will eat the fish’
φ-indexer + DP? Obligatory Impossible (with DP in spec,TP)
mikilitei ki-*(li-)vängä=kaa… ki-ngä(*-li) mikilitei=kaa
fishermen IPFV-3AUG-eat.AV=FUT IPFV-eat.UV-3AUG fishermen=FUT
‘The fishermen will eat…’ ‘The fishermen will eat it’
Full pronouns? Possible Impossible (pronoun in spec,TP)
iude de-ki-vängä=kaa… ki-ngä(-de) (*iude)=ngaa
12AUG 12AUG-IPFV-eat.AV=FUT  IPFV-eat.UV-12AUG 12AUG=FUT
‘We will eat…’ ‘We will eat it’
Syntactic nature: Exponents of φ-agreement on C Reduced pronouns in spec,TP
248 5.6. Chapter summary
Table 5.2: Full paradigms for Äiwoo φ-indexers and pronouns.
Agreement Pronouns
prefixes Suffixal forms Full forms
1MIN i- -no, -nee*, -∅ŋ† iu
12MIN ji- -ji iuji
2MIN mu- -mu iumu
3MIN (∅) -∅n inâ/ine
1AUG me- -ngo(pu), -ngee* iungo(pu)
12AUG de- -de iude
2AUG mi- -mi imi
3AUG li~lu- -i ijii
* The forms -nee and -ngee only occur preceding a 2nd person object pronoun; see §6.3.4.
† The form -∅ŋ is specific for 1MIN objects; the other two forms are for subjects.
CHAPTER 6
NON-CANONICAL UV CLAUSES
6.1 COMPLETING THE PUZZLES
In the last chapter, I spent some time arguing that the φ-suffixes found on UV verbs are
not agreement, but full syntactic arguments: they are pronouns in spec,TP, although in a
phonologically shorter form. Establishing this was a necessary step for me now to develop
an analysis of certain kinds of UV clauses showing different word orders than the usual
one we have concentrated on so far (O V S=TAM). These clauses have been mentioned
several times throughout the dissertation. Often, I used their sheer existence as evidence
for or against certain analytical choices. For example, in discussing some phenomenon P, I
have often used rhetoric like: “just by looking at P alone, we could entertain Hypothesis 1 or
Hypothesis 2, and they would both work just as well. However, phenomenon Q also exists, and
that alone excludes Hypothesis 1, so wemust go with Hypothesis 2; stay tuned for what I’ll say
about Q” – where Q often was the non-canonical UV clauses discussed in this chapter. We
are now finally in a position to dedicate our full attention to them, and develop a model
that will derive both their word order properties and their extraction profiles.
To contextualize the new kinds of UV clauses we are about to introduce, consider the
ones we have seen on so far – what I call the “canonical” types of Äiwoo clauses. As the
reader will be familiar with by now, these have S V=TAM O order in AV (402a), and O V
S=TAM order in UV (402b).
(402) a. Mary ki-veve=kaa nuwopa (250807)
Mary IPFV-buy.AV=FUT house
‘Mary will buy a house’
b. nuwopa ki-ve Mary=kaa
house IPFV-buy.UV Mary=FUT
‘Mary will buy the house’
250 6.1. Completing the Puzzles
As for AV, this is both correct and comprehensive. As for UV, however, examples like
(402b) hide a considerably more complex reality; this is the core of the Word Order Puzzle.
Although O V S=TAM is indeed the default/canonical order for UV clauses, this is not the
whole story. In addition to (402b), three more types of UV clauses can be identified. These
differ from each other in word order; that is, which argument(s) are realized in spec,CP,
spec,TP, and the vP-internal position. Moreover, they differ in their Ā-extraction profiles.
Consider our usual overview table in (403). In chapters 3–4 we covered the upper portion
of the table (403a–c), deriving the word order and extraction properties of AV, UVplain, and
UVonly clauses. In this chapter, we will examine the bottom half (403d–g).
(403) Alternations of word order and Ā-extractability:
Word order:
Clause type:
[CP [TP [vP
a. AV: S3 φS-V.AV =TAM O
7
b. UV 3plain: O V.UV S
7 =TAM }3 done (chapters 3–4)
c. UV : O3only V.UV S
3 =TAM
d. UVinâ: O
NA V.UV S3 =TAM
e. UVgu: O
NA V.UV -π 3O =TAM S
f. UV NASVO: S V.UV -πS =TAM O
NA
g. UV NA1>2: S V.UV -πS-πO=TAM
First, in §6.2 I lay some conceptual and terminological groundwork that will be useful
going forward, about what classes of nominals are treated alike or differently by the part of
Äiwoo syntax that governs word order. With that in the background, §6.3 is a descriptive
section where I discuss the word order properties of UVgu, UVSVO, and UV1>2 clauses in
detail (403e–g), both establishing novel empirical generalizations and amending existing
ones from earlier literature. Then, in §6.4 I show how all these clauses also differ from the
ordinary ones in their Ā-extraction profiles. I illustrate their Ā-extraction possibilities in
§6.4.1, and in §6.4.2 I show how the determining factor behind their extraction patterns
cannot be word order alone by examining the case of UVinâ clauses (403d), which show
completely canonical word order but allow subject extraction. Finally, in §6.5 I show how
revising two probes in our model of the Äiwoo clause (v and T) will capture both the
word order and extraction properties of all these non-canonical UV clauses. We will see
that movement to spec,TP will constitute a second instance of non-local A-movement (in
addition to CP-fronting).
As we can already see from the table, the data is rather intricate – especially when one
starts considering the complex distribution of these four types of clauses. To aid presenta-
tion, throughout the chapter I will give frequent interim summaries, and try to represent
the data in a couple different ways, in the hope that at least one of them will be congenial
and intuitive to the reader. When working with such a complex dataset, it is unavoidable
that a formal analysis will have some degree of complexity as well. However, I will try to
make clear the difference between the strictly implementational formalisms (which may
Chapter 6. Non-canonical UV clauses 251
change in the future as the field progresses) vs. what I consider to be the deeper intuitions
guiding the analysis, which hopefully should be useful regardless of the specific formal
implementation I choose here.
6.2 CLASSES OF NOMINALS IN ÄIWOO SYNTAX
Talking about the system of clausal alternations in Äiwoo is hard not only because the
system itself is complex, but also because the classes of nominals that are grouped together
in displaying one behavior or another constitute somewhat unusual classes. Throughout
the rest of this chapter, we will see that Äiwoo syntax involves two phenomena, here
abstractly called P and Q, that cut the general domain of nominals in two different ways.
(For now, I remain silent aboutwhat exactly “Phenomenon P/Q” are; thesewill be the focus
of the rest of the chapter.) It is convenient, then, to first establish what these groupings
are. For the sake of exposition, it is also useful to not only have labels for the natural
classes themselves, but also for their complement sets, just to have a convenient way to
refer to both “all nominals that partake in process X” and “all nominals that do not partake
in process X” – whether the latter constitutes a natural class in any other sense or not.
The relevant groupings are summarized in (404). In this table, “1/2/3AUGπ” (the first
row) refers to the union of 1st person, 2nd person, and 3AUG pronominal arguments, i.e.,
every pronoun that is not 3MIN. Then, inâ is the overt 3MIN pronoun1. The notation
“∅3MIN” identifies null arguments that are featurally 3MIN. Finally, “lexical DPs” refers
to any nominal containing a lexical root, including proper names: ‘Mary’, ‘the children’,
‘three teachers’, ‘my dog(s)’, etc. Note that, of course, some of these classes overlap: for
example, all φ+Ps are also πPs/pronouns (but not viceversa), and all DPs are also φ0Ps (but
not viceversa).
(404) Classes of nominals in Äiwoo:
Nominal type Phenomenon P Phenomenon Q
1/2/3AUGπ 3 3 } “φ
+Ps”
} “πPs/pronouns”
inâ (3MINπ) 3 7
∅ } “φ0Ps”3MIN 7 7} “DPs”
Lexical DP 7 7
The first thing to notice is that across the whole language, null 3MIN arguments (∅3MIN)
are consistently treated the same way that lexical DPs are; in other words, they “count” as
lexical DPs. This could be interpreted as signifying that these null arguments are not, in
fact, a case of pro-drop, but rather of something more similar to Argument Ellipsis (Hoji
1998, Tomioka 2003, 2014, 2021, Saito 2007, D. Takahashi 2008a,b, a.m.o.). In other words,
1 This also includes its proximal counterpart ine, but I use inâ as the general label since this is the default
form (see §?? for details).
252 6.3. Four types of UV clauses and their distribution
what is dropped is not a pronoun (in the syntactic/featural sense), but a (lexical) DP. Teas-
ing apart cases of genuine pro-drop from Argument Ellipsis is a notoriously hard task, and
has been the focus of a rich literature (see citations above). As for Äiwoo, I don’t have con-
crete arguments to support this hypothesis, as the type of evidence that is typically used
to elucidate this question is not as yet available (this is largely semantic/interpretive evi-
dence: availability of strict/sloppy readings under ellipsis, “paycheck”/functional readings
of pronouns, N-bar ellipsis/one anaphora, necessity of a linguistic antecedent, etc.).
For the purpose of this dissertation, I will simply adopt this stipulation: ∅3MIN nom-
inals in Äiwoo are cases of Argument Ellipsis. In other words, they are elided DPs, and
not null pronouns. One piece of evidence that might point in this direction is that ∅3MIN
nominals have a different distribution than other null arguments (chapter 2, §2.5.4). They
can be in any position: spec,CP, spec,TP, in situ, complement of P, etc., and in any gram-
matical relation (subject, object). In contrast, other (non-3MIN) arguments that are clearly
pronominal can only ever be null in spec,CP, and must instead be overt instead anywhere
else. I leave it to future research to inquire into the interpretive properties of 3MIN argu-
ments, to shed light on their syntactic and semantic nature.
Having “unified” lexical DPs with ∅3MIN arguments, let us now look at the two dif-
ferent groupings Äiwoo syntax makes. The first, and easier cut to label is that between
rows 1/2 vs. rows 3/4. On one hand, we have every nominal that we would reasonably call
pronouns, and on the other, we have lexical DPs and ∅3MIN arguments. For simplicity, I
will call the former group “pronouns” and use the notation πPs (already used in chapter
5), and the latter group “DPs” (with the occasional reminder that this label also includes
∅3MIN arguments).
The second cut is between nominals that have “marked” φ-features (row 1), vs. those
that do not (rows 2–4). The first group consists of all participant pronouns (1st and 2nd
person, including 1st inclusive) plus 3AUG pronouns. The second group consists of the
union of lexical DPs, 3MINs, and the 3MIN overt pronoun inâ. It would be confusing to
call the first group “pronouns”, since inâ is not part of it here but rather patterns with the
complement set. Therefore, for the sake of exposition, I will use the labels “φ+P” and “φ0P”
(read “phi-plus P” and “phi-zero P”)2. This is meant to convey the intuition that the former
group has marked person/number features (PARTICIPANT, ADDRESSEE, AUGMENTED, etc.),
whereas the latter group does not. For now, I will take these as purely descriptive labels.
Later (§6.5.1.1), we will see how we can model the behavior of these groups of nominals
in a principled way by thinking about their featural make-up.
6.3 FOUR TYPES OF UV CLAUSES AND THEIR DISTRIBUTION
Here is the core of our Word Order Puzzle. A full overview of the four UV clause types
is given in (405), indicating for each type its word order, and which arguments occupy
which positions. Here, UVplain subsumes also clause types with the same word order but
2 I thank Zhouyi Sun for helping me come up with this terminology/labels, but he is not to blame for it.
Chapter 6. Non-canonical UV clauses 253
different extraction patterns (UVonly and UVinâ). The notation “πS/O” indicates a φ-indexer
from the suffixal paradigm discussed in chapter 5 – that is, a subject or object pronoun
in its short form. Parentheses around an argument signify that it can optionally be null.
Examples of all four types are given in (406); remember that the names/labels are purely
meant as convenient mnemonics. Each type will be discussed in depth in §§6.3.1–6.3.4.
(405) Overview of UV clause types:
“UVplain” “UVgu” “UVSVO” “UV1>2”
Word order: (O) V S=TAM (O) V-gu-πO=TAM S (S) V-πS=TAM O (S) V-πS-πO=TAM
vP: nothing Subj Obj nothing
Spec,TP: Subj Obj Subj Subj & Obj
Spec,CP: Obj  Obj Subj Subj
(406) a. UVplain: (250728)
John ku-tu-usi-kâ-mu=waa ngä party
John IPFV-bring.UV-back.UV-DIR3-2MIN=FUT to party
‘You will bring John back to the party’
b. UVgu:
∅2MIN /iumu ku-tu-usi-kâ-gu-mu=waa John ngä party
2MIN IPFV-bring.UV-back.UV-DIR3-OBJ-2MIN=FUT John to party
‘John will bring you back to the party’
c. UVSVO:
∅2MIN /iumu ku-tu-usi-kâ-mu=waa iu ngä party
2MIN IPFV-bring.UV-back.UV-DIR3-2MIN=FUT 1MIN to party
‘You will bring me back to the party’
d. UV1>2:
∅1MIN /iu ku-tu-usi-kä-nee-mu=waa ngä party
1MIN IPFV-bring.UV-back.UV-DIR3-1MIN-2MIN=FUT to party
‘I will bring you back to the party’
It is important to remember that these are not optional alternative word orders, but mutu-
ally exclusive: for any given sentence, only one of these pattern will be grammatical. What
determines which type of UV clause we will get are the φ-features of both the subject and
the object. No other factor plays a role: tense/aspect/modality, embedded vs. matrix status,
clause force (interrogative, declarative, etc.), information structure, etc. are all completely
orthogonal to this alternation. In other words: as long as we know that the subject has
φ-features X and the object has φ-features Y, we will be able to fully predict which clause
type we will observe.
After having glanced at what the four clause types look like, their distribution is shown
in (407). I exclude the reflexive combinations (1 > 1, 2 > 2) from the discussion; all 3 > 3
254 6.3. Four types of UV clauses and their distribution
combinations described here are intended as non-reflexive3. (In this table, UVinâ clauses
are subsumed under UVplain, since they share the same word order; this is because the
3MIN pronoun inâ is a φ0P.)
(407) Distribution of UV clause types:
φ+P
S↓, O→
1 2 3AUGπ φ
0P
1 — UV1>2 UVSVO UVplain
φ+P { 2 UVSVO — UVSVO UVplain
3AUGπ UVSVO UVSVO UVSVO UVplain
φ0P UVgu UVgu UVgu UVplain
Already from this table, we can observe a fundamental aspect determining the distribution
of the different clause types. The most important factor governing these alternations are
the categorial features of the two arguments. The mechanism controlling this alternation
is sensitive to the φ+P vs. φ0P distinction: (i) rightmost column: when the object is a
φ0P (lexical DP or 3MIN pronoun), we always get UVplain, no matter what the subject is;
(ii) bottom row: when the subject is a φ0P and the object is a φ+P (non-3MIN pronoun),
we get UVgu; (iii) top-left square: when both arguments are φ
+Ps, we get UVSVO; UV1>2
will be shown to be a sub-type of UVSVO (§6.3.4). The same information is illustrated
in a different way in the diagram in (408). This should be thought of as a (metaphorical)
algorithm/flowchart for determining what kind of clause we will have depending on what
the subject and object look like.
3 1st+2nd person (1st person inclusive) arguments behave across the board in the same way as 2nd person
arguments, so I simplify the table by omitting them. See a full version in Appendix 6.A.
Chapter 6. Non-canonical UV clauses 255
(408) Distribution of UV clause types (flowchart):
What category
is the object?
φ0P φ+P
(lexical DPs, (non-3MIN pronouns)
3MIN pronouns)
UVplain What category
is the subject?
φ0P φ+P
(lexical DPs, (non-3MIN pronouns)
3MIN pronouns)
UVgu Is the subject 1st person
and the object 2nd person?
No Yes
UVSVO UV1>2
This kind of clausal alternation in Äiwoo is reminiscent of what has been discussed un-
der the heading of “global case splits” (Keine 2010, Georgi 2012, Bárány 2017, Yuan 2020,
Bárány & Sheehan 2024, Clem & Deal 2024, a.o.). These are systems where at least de-
scriptively, in order to calculate the case properties, agreement profile, or movement pat-
tern of one argument in a (di-)transitive clause, it is not enough to know various proper-
ties of that argument (e.g. its base-generated position, φ-features, animacy, definiteness/
specificity, etc.), but one must know properties of multiple arguments. Abstracting away
from language-specific details, descriptions of global case splits follow this general struc-
ture: “iff the subject has [such and such properties] and the object has [such and such
properties], then [perform operation X]” (e.g., assign a certain case to the {subject/object},
or move the {subject/object} to spec,XP, or agree with the {subject/object}, etc.). Now, Äi-
woo does not have any morphological case (see chapter 2, §2.5), but its clausal alternation
system can be cast in the same kind of logic: “iff the subject has [properties α] and the
object has [properties β], then the clause will be of [type γ]”.
Before proceeding to the full discussion below, an important caveat is in order. In
what follows, I exclude plural lexical DPs (e.g. mikilivaavee ‘teachers’) from the analysis.
The reasons behind this exclusion would be hard to understand without having seen how
256 6.3. Four types of UV clauses and their distribution
the system works first, so for expository reasons, I relegate a discussion of this caveat to
appendix 6.A. The reader should just bear in mind that throughout the next few sections,
any mention of “(lexical) DP” is implicitly restricted to singular DPs.
6.3.1 UVplain: CANONICAL UV CLAUSES
This is the standard type of UV clause we have seen so far (henceforth, “UVplain”). The
properties of these clauses should be familiar by now: the object is in spec,CP, and the
subject is in spec,TP. The subject can either be any kind of nominal; if it’s a pronoun, it
will surface in its shorter suffixal form rather than the full form (409b).
(409) a. sii ku-tu-mä mikilitei =kaa (250728)
fish IPFV-bring.UV-DIR1 fishermen=FUT
‘The fishermen will bring the fish’
b. sii ku-tu-mä -i =laa
fish IPFV-bring.UV-DIR1-3AUG=FUT
‘They will bring the fish’
Although I have been treating this kind of UV clauses as the default type, their distribution
is effectively restricted to when the object is a φ0P, that is, either a lexical DP or a 3MIN
pronoun. Sentence (410) confirms that as for the object, both the null and the overt 3MIN
pronoun inâ behave identically to a lexical DP like ‘Anna’. Later (§6.4.2), we will see that
clauses with inâ have different extraction properties from the other two types of objects.
(410) ∅3MIN /inâ/Anna ku-tu-mä Lisa=kaa (250728)
3MIN/Anna IPFV-bring.UV-DIR1 Lisa=FUT
‘Lisa will bring him/her/Anna’
Whenever the object is a φ+P (a non-3MIN pronoun), different kinds of UV clause will
obtain. These are the non-canonical types of UV clauses, which we turn to now.
6.3.2 UVgu CLAUSES: OBJECTS IN SPEC,TP
In the first kind of non-canonical UV clauses, the word order is still OVS at a broader level,
but with clear differences from the standard type just described. Although some of these
clauses’ special properties have been mentioned in passing throughout the previous chap-
ters, here is a complete description. In (411) we observe that the object iude ‘us.INCL’ is
fronted to the canonical pre-verbal position, as we indeed would expect in UV. However,
the object pronoun is also realized in the slot that normally is reserved for UV subjects,
between the verb and the TAM particles (in its suffixal form -de). Moreover, this pronomi-
nal suffix is preceded by an extra morpheme, the namesake object marker -gu (see chapter
5, §5.5)4. Finally, the subject ‘John’ surfaces in a special low position, to the right of the
4 The paradigm of object-indexing pronominal suffixes found in this type of clauses is morphologically almost
identical to that of subject pronouns in UV. The only difference is the 1MIN exponent (see chapter 5, fn. 24,
Chapter 6. Non-canonical UV clauses 257
TAM particles (recall that in the canonical UV clauses described above, subjects are to the
left of these particles; (410)).
(411) (iude) ku-tu-usi-kä -gu-de =ngaa John ngä skul
12AUG IPFV-bring.UV-back.UV-DIR3-OBJ-12AUG=FUT John to school
‘John will bring us back to school’ (250728)
All these properties are obligatory. The subject cannot be in its ordinary UV position (be-
tween the verb and TAM), and the object pronoun in that position (including -gu) cannot
be dropped (412). As mentioned in §5.4.2, descriptively it seems as if the position between
the verb and the TAM particles has room for at most one argument. This supports our
analysis of this slot as a specifier position (spec,TP), which can(/must) host one argument:
if the object is in spec,TP – for whatever reason that might be – the subject cannot be
there too, and must stay low instead5. I assume this lower position is the base-generated
position of the subject, in spec,vP.
(412) * (iude) ku-tu-usi-kä(-gu-de) John=kaa ngä skul
12AUG IPFV-bring.UV-back.UV-DIR3-OBJ-12AUG John=FUT to school
Intended: ‘John will bring us back to school’ (250728)
Most frequently, these sentences look verb-initial on the surface string (hence my paren-
thesizing the full pronoun iude above). However, I analyze this as involving a null pronom-
inal object filling the preverbal position (spec,CP), thus counting for the V2 effect; see the
argument in §5.3.2. Indeed, these fronted null pronouns can optionally be realized overtly
under appropriate discourse conditions (e.g. contrastive focus). Furthermore, we can con-
firm that the preverbal full-form object pronoun in (411) is indeed in spec,CP, and not a
left-peripheral topic, by observing its ordering possibilities with respect to sentence ad-
verbs. In (413) that the object pronoun imi 2AUG must follow the adverb. This confirms
that this object pronoun is in spec,CP, and not topicalized. The order pronoun-adverb is
only acceptable if imi is marked as topicalized, by a prosodic break and/or the particle
=Câ DIST.
(413) *{imi } bulaape 3{imi } ku-tu-kä-gu-mi=aa John ngä skul
2AUG tomorrow 2AUG IPFV-bring.UV-DIR3-OBJ-2AUG=FUT John to school
‘Tomorrow John will bring y’all to school’ (240307)
and §5.2). For UV 1MIN subject pronouns, the exponent is -no (ia), or -nee in UV1>2 clauses (§6.3.4). For
1MIN objects, instead, it is ∅ŋ (ib); see chapter 2, §2.5 and Roversi (2020: §3.4) for this notation. For all other
categories than 1MIN, the subject and object suffixal pronouns are identical (modulo the presence of -gu).
(i) 1MIN has different exponents for subject and object suffixal pronouns: (250728)
a. John ku-tu-mä -no =ngaa ngä skul
John IPFV-bring.UV-DIR1-1MIN=FUT to school
‘I will bring John to school’
b. ( iu ) ku-tu-mä-gu -∅ŋ =ngaa John ngä skul
1MIN IPFV-bring.UV-DIR1-OBJ-1MIN=FUT John to school
‘John will bring me to school’ (*kutumä-gu-no=ngaa)
5 In §6.3.4 we will see the only exception to this generalization, i.e. “UV1>2” clauses.
258 6.3. Four types of UV clauses and their distribution
At the syntactic level, as mentioned earlier these clauses show a “doubling” pattern: the
object pronoun in spec,TP is obligatory, but the object is also present in spec,CP (whether
overt or null). This was analyzed in §5.4 as the result of obligatory spell-out of the lower
copy of the object pronoun (the one in spec,TP). Combining these observations, we get
to the following syntactic “map” (414). First, v agrees with the object (thus giving us UV
morphology), and then – exceptionally – T raises the object to spec,TP instead of the
subject, which stays in situ in spec,vP. Finally, C raises the object as well.
(414) Syntactic structure of UVgu clauses:
[CP Obj[i |v] C [TP Obj[i |v] T [vP Subj v ⟨φO⟩ … Obj[i |v]
Exactly how and why T comes to agree with and move the object instead of the subject
will be the main focus of §6.5.1.3. To preview the analysis, I will show that this is another
instance of non-local A-movement, like the one to spec,CP posited in chapter 3. In §6.5.1.3
we will revise the featural specification of the probe on T to derive this effect.
Having examined the morphosyntactic properties of UVgu clauses, we can now turn
to their distribution. We get a UVgu clause when the subject is a φ
0P and the object is
a φ+P (415). If the object is φ0P (regardless of the subject’s features), UVgu is impossible
(416a), and we will get a canonical UV clause instead (416b).
(415) φ0P subject > φ+P object: UVgu (250728)
(iude) ku-tu-kä-gu-de=ngaa ∅3MIN /John ngä skul
12AUG IPFV-bring.UV-OBJ-12AUG=FUT John to school
‘S/he/John will bring us to school’
(416) If the object is a φ0P, UVgu is impossible: (250728)
a. *∅3MIN /inâ/devalili ku-tu-kä-gu-∅n=naa John ngä skul
3MIN/child IPFV-bring.UV-DIR3-OBJ-3MIN=FUT John to school
Intended: ‘John will bring him/her/the child to school’
b. ∅3MIN /inâ/devalili ku-tu-kä John=kaa ngä skul
3MIN/child IPFV-bring.UV-DIR3 John=FUT to school
‘John will bring him/her/the child to school’
Certain properties of UVgu clauses had already been noted in the descriptive literature
(Næss 2006 et seq.). Specifically, this is true of (i) the presence of “object markers” or
“object agreement” on the verb and of the morpheme -gu (although this was analyzed
as a special 3MIN subject marker); (ii) the null 1MIN object exponent ∅ŋ (again, this also
wasn’t analyzed in these terms, but rather with -gu exceptionally being a portmanteau
3MIN>1MIN marker in this specific context). However, the other aspects discussed in this
section are novel findings: (i) the fact that an overt standalone pronoun can optionally
be realized in the preverbal pivot position, teaching us that these clauses also fit the V2
pattern and are essentially OV(O)S; (ii) the particular low position of the subject, to the
right of the =TAM slot.
Chapter 6. Non-canonical UV clauses 259
6.3.3 UVSVO CLAUSES: UV WITH SVO ORDER
Let us now turn to discussing the properties and distribution of UV clauses with SVO
order (shortened as “UVSVO”). These are particularly striking, as for the first time we see
a disconnect between the VIA, i.e. the argument that the voice morphology indexes (the
object), and what occupies the preverbal position (the subject). As we know, Äiwoo is
normally strictly VIA-initial: AV has SVOword order, UV has OVS. However, here we have
UV clauses with SVO word order. I will exclude from the discussion sentences containing
reflexives/anaphors, like the ones briefly discussed in passing in §5.3.2. Although their
word order and voice properties are the same as the ones under discussion here (UV with
SVO order), other properties are not, such as their distribution. The SVO order of reflexive
sentences seems to be connected to binding-theoretical concerns, so I will leave a full
analysis of those to future research.
An example of a UVSVO clause is given in (417a), repeated from above. The SVO word
order is fairly glaring: ijii ‘they’ occupies the pre-verbal position that is normally reserved
for VIA arguments (which here would be the object). However, the verbal morphology
signals that we are in UV, and not in AV. We see the UV stems tu and usi for ‘bring’ and
‘back’, we don’t see any prefixal agreement like wewould find in AV (li- 3AUG), andwe also
see the typical suffixal φ-indexers that we find in UV clauses. Using any type of AV-style
morphology (AV stems, prefixal φ-agreement) is impossible (417b).
(417) a. UVSVO clause: (250728)
(ijii) ku-tu-usi-mä-i=laa iude ngä skul
3AUG IPFV-bring.UV-back.UV-DIR1-3AUG=FUT 12AUG to school
‘They will bring us back to school’
b. AV morphology is not allowed:
* (ijii) ku-lu-tou-ute-mä=kaa iude ngä skul
3AUG IPFV-3AUG-bring.AV-back.AV-DIR1=FUT 12AUG to school
Intended: ‘They will bring us back to school’
Like for UVgu clauses, the preverbal pronoun will most often be null, and an overt pro-
noun is possible but typically only used for e.g. contrastive focus and similar information-
structural effects. By using our ordering diagnostic, we confirm that the preverbal subject
pronoun is in the position that is typically reserved for VIAs (spec,CP), and not a left-
peripheral topic: ijii ‘they’ must follow bulaape ‘tomorrow’ (418).
(418) *{ijii } bulaape 3{ijii } ku-potaa-i=laa iude (250728)
3AUG tomorrow 3AUG IPFV-search.UV-3AUG=FUT 12AUG
‘Tomorrow they will be looking for us’
Moreover, we can confirm that the postverbal pronominal objects in UVSVO clauses are in
the same position as AV objects. In both cases, they must be immediately right-adjacent
to the TAM particles, and any PP or adverbial must follow them (419).
260 6.3. Four types of UV clauses and their distribution
(419) Postverbal objects in UVSVO are in the same position as AV objects: (250728)
a. ( ijii) ku-tu-mä-i=laa 3{iude } ngä skul *{iude }
3AUG IPFV-bring.UV-DIR1-3AUG=FUT 12AUG to school 12AUG
‘They will bring us to school’
b. ( ijii) ku-lu-tou-mä=kaa 3{buk } ngä skul *{buk }
3AUG IPFV-3AUG-bring.AV-DIR1-3AUG=FUT book to school book
‘They will bring (a) book(s) to school’
In contrast to the subject pronoun in spec,CP, the postverbal object cannot be null; it must
be realized overtly. Similarly, and importantly, the suffixal subject pronoun is obligatory
as well. Dropping iude 12AUG in (417a) results in ungrammaticality, and the same is true
for removing -i 3AUG. Both these facts are illustrated in (420a,b) respectively.
(420) Both the suffixal subject pronoun and an overt object pronoun are obligatory:
a. * (ijii) ku-tu-kä-i=laa ∅12AUG ngä skul (250728)
3AUG IPFV-bring.UV-DIR1-3AUG=FUT to school
Intended: ‘They will bring us back to school’
b. * (ijii) ku-tu-kä =Caa iude ngä skul
3AUG IPFV-bring.UV-DIR3 =FUT 12AUG to school
Intended: ‘They will bring us to school’
Similarly to UVgu clauses, here as well we see another “doubling” pattern. The lower copy
of the subject, in spec,TP, is obligatorily realized as its suffixal form -i, and the subject is
optionally also realized in spec,CP in its full form ijii. Combining these observations, we
arrive to the following syntactic structure (421). Again, v agrees with the object, as always
in UV. Exceptionally, v here also agrees with the subject; this will be derived in §6.5.1.2.
Then, again ordinarily for UV clauses, the subject raises to spec,TP. The surprising step
is the last, where C raises the subject – instead of the object, as would usually happen.
This happens because v had agreed with the subject too, thus making it the [i |v]-bearing
argument closest to C.
(421) Syntactic structure of UVSVO clauses:
[CP Subj[i |v] C [TP Subj[i |v] T [vP Subj[i |v] v ⟨φ ⟩ … ObjO [i |v]
The distribution of UVSVO clauses once again depends on the φ-features of both argu-
ments. The overall empirical generalization is that we observe a UVSVO UV clause when
both arguments are φ+Ps, that is, non-3MIN pronouns6. The only exception to this gen-
eralization is configurations where the subject is 1st person (exclusive) and the object is
second person (“1>2”), which I turn to in the next section. Once again, remember that
3MIN pronouns behave like lexical DPs and unlike other pronouns: if the object is a 3MIN
6 As mentioned earlier, this does not cover reflexive sentences, which are UVSVO even when the subject is a
lexical DP (e.g. “John/the teacher saw himself”). See chapter 5, §5.3.2 for an example.
Chapter 6. Non-canonical UV clauses 261
pronoun, we don’t get UVSVO, but UVplain (422). Compare this to the case of e.g. a 2MIN
pronoun, which gives rise to UVSVO order and forbids UVplain order (423).
(422) Object is a 3MIN pronoun/DP (φ0P): UVplain (250728)
a. (inâ)/John ku-tu-kä-i=laa ngä skul 3 UVplain
3MIN/John IPFV-bring.UV-DIR3-3AUG=FUT to school
‘They will bring him/her/John to school’
b. * (ijii) ku-tu-kä-i=laa (inâ)/John ngä skul 7 UVSVO
3AUG IPFV-bring.UV-DIR3-3AUG=FUT 3MIN/John to school
Intended: ‘They will bring him/her/John to school’
(423) Object is a 2MIN pronoun (φ+P): UVSVO (250728)
a. (ijii) ku-tu-kä-i=laa iumu ngä skul 3 UVSVO
3AUG IPFV-bring.UV-DIR3-3AUG=FUT 2MIN to school
‘They will bring you to school’
b. * (iumu) ku-tu-kä-i=laa ngä skul 7 UVplain
2MIN IPFV-bring.UV-DIR3-3AUG=FUT to school
Intended: ‘They will bring you to school’
The following properties of UVSVO clauses had already been reported in the descriptive
literature (Næss 2006 et seq.): (i) the special post-verbal position of the object; (ii) the fact
that the object is always pronominal. This was analyzed in the opposite causal direction,
however: “if the object is a pronoun, it must be postverbal”. However, this framing is
incompatible with the novel finding that UVgu clauses also have pronominal objects: in
that case, the object is clearly preverbal despite being a pronoun. The following aspects of
UVSVO clauses are novel findings: (i) the fact that 3MIN subjects or objects never result in a
UVSVO clause; (ii) the preverbal position of the subject: the VOword order was established,
but we did not know if the complete word order was SVO or VSO; (iii) the possibility of
overt standalone pronouns in preverbal position, doubling the suffixal marker on the verb;
(iv) the fact that the object pronoun cannot be null/must be overtly realized.
6.3.4 UV1>2 CLAUSES: A SUBTYPE OF UVSVO
The final type of special UV clause (“UV1>2 clauses”) has a rather restricted distribution,
and it can be shown to be a sub-type of the UVSVO clause type. This kind of clause is found
only when the subject is first person (exclusive) and the object is second person; number
does not matter7. The possible combinations are illustrated in (424).
7 This generalization comes with a small caveat. If the subject is 1st person unit-augmented rather than
augmented, then we get a UVSVO sentence (ia), not UV1>2 (ib).
(i) 1st person unit-augmented subjects are incompatible with UV1>2: (250617)
a. ( iungo-le ) ku-tu-usi-kä-ngo-le=naa iumu ngä party
1AUG-UA IPFV-bring.UV-back.UV-DIR3-1AUG-UA=FUT 2MIN to party
‘We two.EXCL will bring you back to the party’
262 6.3. Four types of UV clauses and their distribution
(424) a. 1MIN > 2MIN: (250617)
(iu) ku-tu-usi-kä-nee-mu=waa ngä party
1MIN IPFV-bring.UV-back.UV-DIR3-1MIN-2MIN=FUT to party
‘I will bring you back to the party’
b. 1AUG > 2MIN:
(iungopu) ku-tu-usi-kä-ngee-mu=waa ngä party
1AUG IPFV-bring.UV-back.UV-DIR3-1AUG-2MIN=FUT to party
‘We will bring you back to the party’
c. 1MIN > 2AUG:
(iu) ku-tu-usi-kä-nee-mi=aa ngä party
1MIN IPFV-bring.UV-back.UV-DIR3-1MIN-2AUG=FUT to party
‘I will bring y’all back to the party’
d. 1AUG > 2AUG:
(iungopu) ku-tu-usi-kä-ngee-mi=aa ngä party
1AUG IPFV-bring.UV-back.UV-DIR3-1AUG-2AUG=FUT to party
‘We will bring y’all back to the party’
This is the only circumstance in Äiwoo where two arguments are realized in as spec,TP.
The suffixal 1MIN/1AUG subject pronouns have special allomorphs in this context: -nee
1MIN instead of usual -no, and -ngee 1AUG instead of usual -ngo(pu)8. Furthermore, as in
the previous cases we see that an overt standalone subject pronoun can be realized in
spec,CP. Thus, the word order of these clauses is SVO; this is yet another case of “dou-
bling”. This is also a novel finding, not having been described before in the published
literature. As in UVSVO clauses, a fronted subject pronoun is in the canonical preverbal
VIA position, despite being a UV subject (425a). Fronting the object pronoun is impossible
(425b). Moreover, using a standalone object pronoun is not possible even in the postverbal
position (426): the object can only be realized in spec,TP.
(425) Spec,CP is occupied by the subject, not the object: (240404)
a. *{iu } bulaape 3{iu } ki-lâwâle-wâ-nee-mu=waa
1MIN tomorrow 1MIN IPFV-help.UV-DIR2-1MIN-2MIN=FUT
‘Tomorrow I will help you’
b. * ( iungo-le ) ku-tu-usi-kä-ngo /ngee-le-mu=waa ngä party
1AUG-UA IPFV-bring.UV-back.UV-DIR3-1AUG-UA-2MIN=FUT to party
Intended: ‘We two.EXCL will bring you back to the party’
In contrast, a 2nd person unit-augmented object can still give rise to UV1>2, as long as the subject is 1st
person minimal or augmented (ii). I leave this issue to future research.
(ii) 2nd person unit-augmented objects are compatible with UV1>2: (250617)
( iu /iungopu ) ku-tu-usi-kä- {nee /ngee }-mi-le=naa ngä party
1MIN IPFV-bring.UV-back.UV-DIR3-{1MIN/1AUG }-2AUG-UA=FUT to party
‘I/We.EXCL will bring you two back to the party’
8 The existence of the 1AUG suffix -ngee and its use in this kind of sentences was first described in Roversi
(2025: appendix B). In earlier literature (e.g. Næss 2015a, 2018a, Roversi 2019), only the 1MIN > 2 cases had
been reported.
Chapter 6. Non-canonical UV clauses 263
b. * bulaape iumu ki-lâwâle-wâ-nee-mu=waa
tomorrow 2MIN IPFV-help.UV-DIR2-1MIN-2MIN=FUT
Intended: ‘Tomorrow I will help you’
Native speaker’s comment: “That is not how Äiwoo works”
(426) The object can only in spec,TP, not in situ: (240523)
* bulaape (iu) ki-lâwâle-wâ-nee-mu=waa iumu
tomorrow 1MIN IPFV-help.UV-DIR2-1MIN-2MIN=FUT 2MIN
Intended: ‘Tomorrow I will help you’
Given the position of the arguments, the syntactic derivation of these clauses is previewed
in (427). First, as in UVSVO clauses, v agrees with both arguments (for reasons to be un-
derstood). In keeping with my analysis of the φ-suffixes as pronouns occupying spec,TP
then, I propose that in this case both the subject and the object are raised by T to this
position (I assume that the object tucks in in an inner specifier; Richards 1997). Then, as
in UVSVO clauses, the subject is raised further to spec,CP.
(427) Syntactic structure of UV1>2 clauses:
[CP Subj[i |v] C [TP Subj[i |v] Obj[i |v] T [vP Subj[i |v] v ⟨φ ⟩ … ObjO [i |v]
I assume that the lower copy of the object (in situ) cannot be spelled out due to the usual
chain resolution mechanism, which deletes all lower copies of movement. Both pronouns
in spec,TP are forced to be pronounced due to the interaction of Local Dislocation and the
Stray Affix Filter, as described in 5, §5.5. For the object, this is the highest copy, which
presumably would have been realized overtly regardless; Local Dislocation forces the re-
alization of this pronoun in its short form. For the subject, Local Dislocation forces the
realization of the spec,TP copy despite there being a higher copy.
6.3.5 INTERIM SUMMARY
In this section I have presented all four types of UV clauses we find in Äiwoo, finally giving
the full empirical landscape of ourWord Order Puzzle. Below, I repeat (i) a comprehensive
overview of the four clause types (428); (ii) their distribution (429); (iii) sketch derivations,
only drawing the agreement steps by v for simplicity (430).
(428) Overview of UV clause types:
“UVplain” “UVgu” “UVSVO” “UV1>2”
Word order: (O) V S=TAM (O) V-gu-πO=TAM S (S) V-πS=TAM O (S) V-πS-πO=TAM
vP: nothing Subj Obj nothing
Spec,TP: Subj Obj Subj Subj & Obj
Spec,CP: Obj  Obj Subj Subj
264 6.3. Four types of UV clauses and their distribution
(429) Distribution of UV clause types:
φ+P
S↓, O→
1 2 3AUG φ0π P
1 — UV1>2 UVSVO UVplain
φ+P { 2 UVSVO — UVSVO UVplain
3AUGπ UVSVO UVSVO UVSVO UVplain
φ0P UVgu UVgu UVgu UVplain
(430) Sketch derivations:
a. UVplain: S moves to spec,TP; O moves to spec,CP
[CP Obj[i |v] C [TP Subj T [vP tS v … tO [i |v]
b. UVgu: O moves to spec,TP and spec,CP; S remains in situ
[CP Obj[i |v] C [TP Obj[i |v] T [vP Subj v … tO [i |v]
c. UVSVO: S moves to spec,TP and spec,CP; O remains in situ
[CP Subj[i |v] C [TP Subj[i |v] T [vP tS [i |v] v … Obj[i |v]
d. UV1>2: S and O move to spec,TP; S moves to spec,CP
[CP Subj[i |v] C [TP Subj[i |v] Obj[i |v] T [vP tS [i |v] v … tO [i |v]
To solve theWord Order Puzzle and correctly generate the derivations in (430), we need to
understand when which probes move which argument(s): what does v agree with, what
does T agree with/move, and what does C agree with/move? From the distribution in
(429) we can then draw a neat generalization that will help us answering these questions.
Non-canonical word orders (UVgu, UVSVO, UV1>2) arise whenever the object is a φ
+P, i.e.,
a nominal with “marked” φ-features (anything other than 3MIN). In contrast, φ0P objects
(lexical DPs and 3MIN pronouns) trigger canonical word order (UVplain).
As has been pointed out throughout this section, the existence of these four types of
patterns was already known in the literature. However, the available descriptions con-
tained a series of gaps concerning several aspects of the system. This is to my knowledge
the most in-depth study to date of the system of word order alternations in Äiwoo UV,
and should hopefully be useful also to readers who don’t necessarily subscribe to the the-
oretical framework and the variety of assumptions being made here (e.g., labelling the
preverbal position “spec,CP”, but also the notion of movement, the generative approach
at large, etc.).
Of course, all/any analyses are necessarily always theory-dependent, and “theory-
neutrality” is impossible. However, throughout this chapter and the last I have tried to
Chapter 6. Non-canonical UV clauses 265
make my arguments not theory-internal, but based on empirically observable realities of
the language. To make a concrete example, positing null pronouns in the preverbal po-
sition (5.3.2) was not done to make Äiwoo comply to some particular (generative) claim
about the reality of null arguments, or about how every clause must have a subject (or
direct object), or for some notion of theoretical/analytical elegance. Rather, I posited null
pronouns in that position because hypothesizing their presence leads to correct predic-
tions in terms of observable word order phenomena. An alternative analysis of these Äi-
woo patterns without null pronouns would leave these facts and correlations unexplained,
and would thus need to be supplied by extra assumptions and hypotheses to be equally
descriptively and explanatorily adequate.
6.4 PRONOMINAL OBJECTS AND SUBJECT EXTRACTION FROM UV
In the previous section, we got an empirical lay of the land of the Word Order Puzzle, and
observed how the φ+P vs. φ0P distinction is what controls the distribution of different
word orders. In this section, we turn to the Extraction Puzzle, that is, how and why these
various UV clause types differ in which arguments can or cannot be Ā-extracted. We will
now see a property that is not sensitive to the φ+P vs. φ0P distinction, but to the more
ordinary distinction between all pronouns (including 3MIN) vs. lexical DPs. Specifically,
we will see that pronominal objects not only can trigger special word orders, but they
have another “superpower”: they impose UV/ban AV, and asa result, subject extractionis exceptionally possible from UV clauses (shortened to “3S [UV” below). One way to
frame this is that metaphorically, a pronominal object automatically turns any transitive
verb into a UV-only verb (§4.5).
We can now thus put names on the two phenomena that were used in §6.2 to justify
the existence of these four classes of nominals. This is summarized again in (431).Whether
the object is a pronoun or a DP governs Ā-extractability. In contrast, whether the object
is a φ+P or a φ0P governs the word order alternation.
266 6.4. Pronominal objects and subject extraction from UV
(431) Classes of nominals in Äiwoo:
Voice and Ā-extractability:
lexical DPs vs. pronouns
DPs: πPs/pronouns:
3 both voices, Ban on AV,
Ā-restriction 3 S[UV
∅ 1st/2nd/Lexical DPs 3MIN inâ 3AUG pro.
φ0Ps: φ+Ps:
Canonical word order Special word orders
Word order alternation:
φ0Ps vs. φ+Ps
In the next sections, first (§6.4.1), I will demonstrate two properties of clauseswith pronom-
inal objects with respect to Ā-extractability: UV is the only option/AV is impossible, and
subject extraction is allowed from UV. Then (§6.4.2), I will show that this cluster of prop-
erties is sensitive to all pronouns, and not just to φ+Ps, by looking at what happens when
the object is inâ (the 3MIN overt pronoun). We will see that despite these clauses’ ordinary
(UVplain) word order, they have the same exceptional voice and extraction properties as
the non-canonical ones. Finally (§6.4.3), I will set up a series of desiderata for a model
deriving this pattern.
6.4.1 THE PARALLEL WITH UVONLY CLAUSES: OBLIGATORY UV AND FREE
EXTRACTION
First, I will show how pronominal objects generate clauses with several properties in com-
mon with UVonly clauses: (i) AV is impossible; (ii) subject extraction is allowed despite the
clause being UV.
Consider the first property (obligatory UV/impossible AV). Like in clauses built around
UV-only verbs, any clause with a pronominal object can only be in UV, and AV is just not
an option (432). This holds regardless of what kind of nominal the subject is. If it’s a φ0P,
we will get a UV +gu clause (432a); if it’s a φ P, we will get a UVSVO clause (432b). In either
case, AV is impossible (432c).
Chapter 6. Non-canonical UV clauses 267
(432) If the object is a pronoun, only UV is allowed: (250728)
a. (iude) ku-tu-usi-mä-gu-de=ngaa nurse (3UVgu)
12AUG IPFV-bring.UV-back.UV-DIR1-OBJ-12AUG=FUT nurse
‘The nurse will bring us back’
b. ( ijii) ku-tu-usi-mä-i=laa iude (3UVSVO)
3AUG IPFV-bring.UV-back.UV-DIR1-3AUG=FUT 12AUG
‘They will bring us back’
c. * ( ijii)/nurse ku-( lu)-tou-ute-mä=kaa iude (7AV)
3AUG/nurse IPFV-(3AUG)-bring.AV-back.AV-DIR1=FUT 12AUG
Intended: ‘They/The nurse will bring us back’
At first glance, we might imagine wanting to explain away the ban on pronominal objects
in AV as just a sub-case of the general restriction found on objects in AV clauses, discussed
in chapter 2, §2.5.3. Generally, an AV object must be a non-specific/narrow-scope indefi-
nite. Definite objects (or even specific indefinites) are not usually allowed in AV. Pronouns
are definite descriptions, so given this restriction we should indeed expect them to be bad
objects for AV clauses. However, this cannot be the same kind of effect. As mentioned in
§2.5.3, this specificity restriction only holds in the absence of Ā-extraction. If one extracts
the subject of a clause with a definite (DP) object, such as buk eângâ ‘that book’, the usual
extraction restriction applies: AV is obligatory (433a), and UV is impossible (433b). In a
sense, the need to comply to the Ā-extraction restriction “trumps” the semantic restric-tion AV objects. Described in metaphorical OT terms, the constraint “*S [UV” outranks
the constraint “*AV-with-O[+SPEC]”.
(433) With definite DP objects, subject extraction forces AV: (250728)
a. sime (mi=)[RC ku-tou-mä=kaa buk eângâ ]
person REL= IPFV-bring.AV-DIR1=FUT book that
‘The person who will bring that book (here)’
b. * sime (mi=)[ nRC buk eângâ ku-tu-mä -∅ =naa]
person REL= book that IPFV-bring.UV-DIR1-3MIN=FUT
Intended: ‘The person who will bring that book (here)’
Consider now what happens if the object is not a lexical DP, but a pronoun. I will mostly
use subject extraction from UVgu clauses as the case study since in these, the subject can
be a lexical DP, and therefore it is easy to create contexts where it is relativized9. Crucially,
9 In UVSVO clauses, in contrast, the subject is typically a pronoun (φ
+P). Extracting pronouns – or rather,
forming relative clauseswith a pronoun as the head (“they/we [who…]”) – is often restricted or at least shows
different properties from ordinary relativization (see e.g. Conrod 2019), and the facts about this phenomenon
inÄiwoo are not entirely clear. However, we are able to confirm subjects can be extracted fromUVSVO clauses
as well. Plural lexical DP subjects can sometimes generate UVSVO clauses; this is the caveat discussed in §6.A.
We can now observe that they too can be extracted (ia). This is perhaps less surprising under our theory
of Ā-extraction, since in these clauses, the subject is the highest argument anyway, but I include it here for
completeness. Just to be sure, AV here is impossible as well, because the object is a pronoun (ib).
268 6.4. Pronominal objects and subject extraction from UV
even under subject extraction, UV is the only option (434a), and AV is still banned (434b)10.
Comparing this to the case of definite DP objects, we could describe thiswith a three-tieredconstraint ranking: “*AV-with-OπP” ≫ “*S [UV” ≫ “*AV-with-O[+SPEC]”.(434) Pronominal object: S [UV is possible, and AV is impossible (250728)
a. mekivaavee (mi=)[RC ∅2MIN ku-tu-usi-mä-gu-mu=waa
teacher REL= IPFV-bring.UV-back.UV-DIR1-OBJ-2MIN=FUT
ngä skul]
to school
‘The teacher who will bring you back to school’
b. *mekivaavee (mi=)[RC ku-tou-ute-mä=kaa iumu
teacher REL= IPFV-bring.AV-back.AV-DIR1=FUT 2MIN
ngä skul]
to school
Intended: ‘The teacher who will bring you back to school’
For clarity, the pattern is summarized in (435). In monoclausal environment, definite ob-
ject DPs and pronominal objects pattern alike in only permitting UV, and not AV. How-
ever, they differ under subject extraction: for definite DP objects, the usual Ā-extraction
restriction triumphs, and only AV is allowed; for pronominal objects, AV is still stubbornly
impossible, so we can now extract a subject from UV.
(i) Subject extraction from UVSVO is possible: (250728)
a. pedevalili (mi= )[RC ku-tu-usi-mä-i=laa iumu ngä skul]
children REL= IPFV-bring.UV-again.UV-DIR1-3AUG=FUT 2MIN to school
‘The children who will bring you back to school’
b. * pedevalili (mi= )[RC ku-tou-ute-mä=kaa iumu ngä skul]
children REL= IPFV-bring.AV-again.AV-DIR1=FUT 2MIN to school
Intended: ‘The children who will bring you back to school’
10 I have only received exactly one positive acceptability judgment for a sentence like (434b), shown in (ia).
This is a null-headed RC (see chapter 4). The AV version was judged as equivalent to the UVSVO version (ib).
All other times I have tested AV structures like (ia) with my consultant, they were rejected. I hope further
empirical work with more speakers will clarify this discrepancy.
(i) a. Subject extraction from AV with a pronominal object: (230608)
[∅ mi=[RC li-togo iu]]=ngâ li-ebulou-mana
REL= 3AUG-hit.AV 1MIN=DIST 3AUG-long-very
‘Those/the people who hit me were very tall’
b. UVSVO equivalent:
[∅ mi=[RC togulo-i iu]]=ngâ li-ebulou-mana
REL= hit.UV-3AUG 1MIN=DIST 3AUG-long-very
‘Those/the people who hit me were very tall’
Chapter 6. Non-canonical UV clauses 269
(435) Differences between definite DP objects and pronominal objects:
Object: Monoclausal environment: Under subject extraction:
Def. DP: John will bring.{3UV/7AV} that book The man [who will bring.{7UV/3AV} that book]
Pronoun: John will bring.{3UV/7AV} you /us /… The man [who will bring.{3UV/7AV} you /us /…]
Having ruled out the definiteness effect for AV objects as a potential confound, we can
now see that UV clauses with pronominal objects share properties with UVonly clauses:
AV is impossible, and subject extraction from UV is exceptionally allowed. To restate our
informal intuition: a pronominal object can “turn any ordinary transitive verb into a UV-
only verb”.
FURTHER EVIDENCE FOR Ā-INTERVENTION: MAKING Ā-FEATURES VISIBLE In the
model I propose, the exceptional possibility of subject extraction in (434a) in explained
in terms of (the lack of) Ā-intervention. The core idea is that in contexts where both AV
and UV are possible, in UV the highest argument (the object in spec,CP) will have an Ā-
feature [PIV], and thus intervene for Ā-extraction of a lower argument. However, in other
contexts (UVonly, UVgu, etc.), the clause will look the same way regardless of where Ā-
features are. The object in spec,CP in (434a) doesn’t (necessarily) have Ā-features despite
being in spec,CP, and therefore subject extraction across it is possible.
The prediction of this model is that if we put Ā-features on the fronted object pronoun
in (434a), in principle this should create an Ā-intervention effect, and thus block subject
extraction – despite the fact that such a sentence would look identical to (434a). We can
test this by exploiting a peculiarity of Äiwoo grammar, that is, the distribution of null
arguments. Äiwoo allows pro-drop, but for non-3MIN pronouns, this is restricted to argu-
ments in spec,CP (3MIN arguments can be dropped in any position); see chapter 2, §2.5.4.
In this position, however, realizing a pronoun as null is the default option, and pronounc-
ing it overtly is only done in particular discourse/information-structural circumstances,
as mentioned above; see also chapter 2, §2.5.4 about the discourse effects of pro-drop. Be-
cause of this fact, I assume that /∅/ is the default pronunciation of e.g. the 12AUG pronoun
in spec,CP (436a), and iude is the pronunciation of this pronoun when it carries some kind
of Ā-feature, which I will call [FOC] for convenience (436b) (though nothing hinges on this
being specifically focus, as opposed to some other Ā-feature). In other words: the pres-
ence of an Ā-feature (“[FOC]”) on a pronoun in this position forces its overt pronunciation
(see chapter 5, §5.5 for an in-depth discussion of how Äiwoo pronouns get spelled out; I
notate [FOC] on the π head for simplicity, though this is not crucial). See also Szabolcsi
(2009) a.o. for discussion of how focus can force otherwise null elements to be pronounced
overtly, and in particular Herbeck (2015) for the idea that null vs. overt pronunciation of
a pronoun can be framed as allomorphy (conditioned by e.g. Ā-features) rather than an
underlying syntactic difference between the two.
270 6.4. Pronominal objects and subject extraction from UV
(436) Spell-out rules for pronouns in spec,CP:
a. Default spell-out: ∅ b. [FOC] forces overt spell-out:
πP ⇔ ∅ / C πP ⇔ iude / C
π φ(P) π φ(P)
[12AUG] [FOC] [12AUG]
This predicts that if we take a sentence like (434a) and minimally change it by making
the object pronoun in spec,CP overt, subject extraction should become impossible. This
is borne out (437). From this perspective, this otherwise somewhat puzzling fact provides
support for our Ā-intervention approach to the extraction restriction in Äiwoo
11.
(437) S [UVgu is only possible if the object in spec,CP is null: (250728) 
Mary ki-eeu-kä=naa go [mekivaavee [ 3RC ∅12AUG /*iude
Mary IPFV-speak-DIR3=FUT to teacher 12AUG
ku-tu-kä-gu-de=ngaa ngä skul]]
IPFV-bring.UV-DIR3-OBJ-12AUG=FUT to school
‘Mary will talk to the teacher who will bring us to school’
6.4.2 VOICE AND EXTRACTION ARE INDEPENDENT FROM WORD ORDER
In the previous section, we established a correlation between UV clauses with pronominal
objects, and the clustering of UVonly-like voice and extraction properties: AV is impossible,
and subject extraction is exceptionally allowed from UV. However, all the UV clauses with
pronominal objects we have seen so far also have special word orders, so we are left with
a possible confound. Are the voice/extraction properties of these clauses caused by the
pronominal objects themselves, or by the particular word orders these kind of objects
trigger? That is: are theWord Order Puzzle and the Extraction Puzzle reducible to a single
puzzle, or not?
Given our model, what we concretely need to know to answer this question is the fol-
lowing: are the voice/extraction properties also sensitive to the φ+P/φ0P distinction (like
word order is), or are they sensitive to the DP/pronoun distinction instead? Luckily, we
can tease these two alternatives apart. The evidence comes from a type of clause whose
object is a pronoun, but not a φ+P: the 3MIN overt pronoun inâ. These clauses have ordi-
nary UVplain word order (because their object is a φ
0P). Importantly, we will see that de-
spite their ordinary word order, inâ being a pronoun has an effect on the voice/extraction
system: here too AV is impossible, and subject extraction from UV is allowed.
11 The impossibility of having an overt pronoun in spec,CP could, in principle, be analyzed in different terms.
For example, one could say that in the case where subject extraction is possible, there just is no null ob-
ject pronoun in spec,CP at all. Presumably, under this kind of approach spec,CP would be occupied by the
extracted subject. However, to fit this kind of analysis with the rest of our model, we would need to con-
siderably complicate the agreement/movement pattern performed by C, with no clear explanatory gains.
Chapter 6. Non-canonical UV clauses 271
Let us now see the properties of these UV clauses whose object is the overt 3MIN pro-
noun inâ, henceforth “UV 12inâ clauses” . These clauses are strikingly similar to UVonly ones,
despite not containing UV-only verbs but ordinary, voice-alternating transitive verbs. As
we can see from (438), their word order is the canonical UVplain one, with the object in
spec,CP and the subject in spec,TP. This follows from our generalization that special word
orders are triggered by φ+P objects, which inâ is not.
(438) UVinâ: baseline example (231109)
inâ ku-tu-usi-kä pedevalili=kaa ngâ nuwopa
3MIN IPFV-bring.UV-back.UV-DIR3 children=FUT to house
‘The children will bring her back home’
However, despite the canonical word order, UVinâ clauses pattern with UVonly clauses and
other clauses with pronominal objects (UVgu/UVSVO) in forbidding AV (439), and allowing
subject extraction from UV (440a).
(439) Monoclausal environment with inâ as the object: only UV is possible (231012)
* pedevalili ku-lu-tou-ute-mä=kaa inâ ngâ nuwopa
children IPFV-3AUG-bring.AV-back.AV-DIR1=FUT 3MIN to house
Intended: ‘The children will bring her back home’
(440) Subject extraction with inâ as the object: only UV is possible (231012)
a. [pedevalili [RC inâ ku-tu-usi-kä -i =laa ngâ
children 3MIN IPFV-bring.UV-back.UV-DIR3-3AUG=FUT to
nuwopa]]=kâ, nubotage i-kää-päko-i-i
house=DIST road ASP-know.UV-good-UV-3AUG
‘The children who will bring her home, they know the road very well’
b. * [pedevalili [RC ku-lu-tou-ute-kä=naa inâ ngâ
children IPFV-3AUG-bring.AV-back.AV-DIR3=FUT 3MIN to
nuwopa]]
house
Intended: ‘The children who will bring her home’
Note that these properties do not hold if the object is a null 3MIN argument, as opposed
to the overt inâ. As described in §6.2, null ∅3MIN arguments behave/are treated like DPs,
not like pronouns. In this case, we get the canonical pattern: both AV and UV are allowed
(modulo the specificity effect on objects), and subject extraction imposes AV. Concretely,
if we take the sentences in (440) and replace inâ with its null counterpart, the judgments
reverse: now UV is impossible (441a), and AV is the only option (441b).
12 The same is true for the proximal 3MIN pronoun ine. The form inâ is distal, and the default one. Both forms
are gender-neutral; I arbitrarily use she/her in the English translations for convenience.
272 6.4. Pronominal objects and subject extraction from UV
(441) Subject extraction with ∅3MIN object: only AV is possible (231012)
a. * [pedevalili [RC ∅3MIN ku-tu-usi-kä -i =laa ngâ
children IPFV-bring.UV-back.UV-DIR3-3AUG=FUT to
nuwopa]]
house=DIST
Intended: ‘The children who will bring her home’
b. [pedevalili [RC ku-lu-tou-ute-kä=naa ∅3MIN
children IPFV-3AUG-bring.AV-back.AV-DIR3=FUT
ngâ nuwopa]]
to house
‘The children who will bring her home’
We can observe that inâ-objects have some properties in common with φ+P (non-3MIN)
pronominal objects, and other properties in common with φ0P objects. An inâ object is
grouped together with the other pronouns for whatever mechanism it is that forces UV/
bans AV and allows for subject extraction. However, it counts as a φ0P for the mechanism
determining word order.
6.4.3 INTERIM SUMMARY, DESIDERATA, AND PREVIEW
In keeping with our grand mission of answering both the Word Order Puzzle and the Ex-
traction Puzzle, let us now summarize the facts just described with the following typol-
ogy, once again repurposing our general overview table (442), to see how the two systems
(clausal word order and Ā-extraction) are only partially correlated. The notation ‘DP[Ā]’
signifies a DP carrying some Ā-feature(s), for example [PIV].
On one hand, we have canonical clauses, which show an AV/UV alternation. Precisely
in these cases where both voices are possible, we have restricted (VIA-only) Ā-extraction
(442a,b). Then, we have four clause types – here subsuming UV1>2 under UVSVO – where
there is no alternation (AV is impossible), and where the non-VIA subject can be extracted
(442c–f). Intuitively, this seems like a meaningful correlation: Ā-extraction in one voice
is restricted when we actually have an alternative voice option, but if only one voice is
possible, then extraction is free. For one of these restricted types (UVonly), the determining
factor is the presence of a particular kind of verb13. For the other three (UVinâ, UVgu,
UVSVO), the determining factor is the object being a pronoun. This is then cross-cut with
word order differences: UVgu and UVSVO have special word orders, whereas UVonly and
13 To be clear, this statement is implicitly restricted specifically to UVonly clauses with non-pronominal objects,
e.g. ‘the teacher sawMary’ (te ‘see.UV’ is a UV-only verb). Of course, UV-only verbs are perfectly compatible
with inâ-objects and φ+P objects (‘the teachers saw us’; ‘you saw me’, etc.). The resulting clause, like for
every other transitive verb, will be either a UVinâ, UVgu, or UVSVO (or UV1>2) clause. Therefore, such a clause
(UV-only verb + pronominal object) will be “special” for multiple reasons, and it wouldn’t allow us to isolate
the fact of interest, namely, that UVonly clauses have special properties even with DP objects, unlike UV
clauses with ordinary voice-alternating verbs.
Chapter 6. Non-canonical UV clauses 273
UVinâ clauses have the same word order as canonical UVplain clauses (what is labelled in
the table as “regular word order”).
(442) Overview of Äiwoo clause types:
Regular  Object  Extract Is another
Clause type: Word order:
word order? category? non-VIA? voice possible?
a. AV: S3 φS-V.AV =TAM O
7 yes φ0P DP no yes (→ UV)
b. UV 3 7 0plain: O V.UV S =TAM yes φ P DP[Ā] no yes (→ AV)
c. UV 3 3 0only: O V.UV S =TAM yes φ P DP[Ā] yes no (*AV)
d. UVinâ: O
NA V.UV S3 =TAM yes φ0P πP yes no (*AV)
e. UV : ONA V.UV -π =TAM S3gu O no φ
+P πP yes no (*AV)
f. UV NASVO: S V.UV -πS =TAM O
NA no φ+P πP yes no (*AV)
Looking at this table, we can see that our model needs to make two different cuts – namely
the two distinctions we have been talking about. First, we need something in common be-
tween UVonly clauses and all clauses with any pronominal objects (442c–f), including inâ:
in all of these, UV is the only possibility, AV is banned, and subject extraction is allowed
despite the clause being UV. Within these clause types, we then have a split between
clauses with regular word order (UVonly, UVinâ; (442c,d)) and clauses with special word
orders (UVgu, UVSVO; (442e,f)). Therefore, we need to make a different cut, and identify
some property that φ+P objects (nominals with “marked” φ-features, i.e. non-3MIN) have
in common, not shared by φ0P objects (lexical DPs and 3MIN pronouns). We can now for-
mulate our two Puzzles in much more specific terms:
(443) a. The Extraction Puzzle:Why do sentences with pronominal objects and UVonly-
sentences forbid AV, and allow subject extraction from UV?
b. TheWord Order Puzzle:Why do φ+P objects give rise to special word orders?
Luckily, we have already done the job of figuring out an answer to the first question, that
is, how to derive clauses where AV is impossible, and subjects can be extracted from UV.
This was the case of UVonly clauses (§4.5). For those, we pursued the following reasoning.
If AV is impossible/UV is obligatory, this means that v always agrees with the object. Be-
cause object agreement always happens, clearly it is not conditioned by Ā-features like
[PIV], which are syntactically free (that is, their distribution at the choice of the speaker).
Moreover, we also know that C always fronts whatever v agreed with; this is something
we will keep constant in deriving the alternative word orders. The only difference be-
tween ordinary voice-alternating verbs and UV-only verbs is that the latter select for vH;
unlike the ordinary v head, vH is specified as agreeing with any object. This means that in
UVonly clauses, an object can end up being fronted to spec,CP evenwithout any Ā-features.
Therefore, such an object won’t pose an Ā-intervention problem, and that is why subjects
can be extracted despite the clause being UV.
Because the voice and extraction properties of UV clauses with pronominal objects
are identical to those of UVonly clauses, it is most parsimonious to imagine that the un-
274 6.4. Pronominal objects and subject extraction from UV
derlying reason must the same. Therefore, our desideratum at the implementational level
is to make sure that v always agrees with a pronominal object (both φ+Ps and inâ),
regardless of whether it has Ā-features or not. Like for UV-only verbs, this “obligatory
object agreement” will have the desired effect. Any clause where the object is a pronoun
will obligatorily be UV, whether the object carries Ā-features or not. This will ultimately
suspend the Ā-restriction, since even if the object moves past the subject, there will not
be any Ā-intervention effect, so the subject can be extracted from UV. This gives us an
answer to the Extraction Puzzle.
For theWordOrder Puzzle, on the other hand, wewill need to understandwhat exactly
controls movement of the different arguments to their various landing sites. To preserve
our answer to the Extraction Puzzle, we will keep constant our idea of what C does: C
always raises the closest argument with [i |v] (i.e., agreed with by v). Because in UVSVO
clauses C attracts the subject, this entails that in this case v somehow agrees with the sub-
ject (whether in addition to or instead of the object), flagging it with [i |v]. Independently,
we will need to understand why it is that T generally raises the subject, but it can also
raise the object (in UVgu clauses), or both arguments (in UV1>2 clauses). To achieve these
results, we will need to revise two probe specifications, that on v and that on T (while
keeping the one on C constant).
The solution to our Word Order and Extraction Puzzles will thus lay in the complex
interaction between three probes in the clausal spine (v, T, and C), each sensitive to par-
ticular features and not others. I give a preview of what the final analysis will look like in
(444)–(445), for now remaining silent on the exact mechanical details of implementation.
(444) The Extraction Puzzle: why do sentences with pronominal objects and UVonly
sentences forbid AV, and allow subject extraction from UV?
a. The probe on v agrees with two classes of objects: those with Ā-features, and
all pronominal ones (φ+P-objects and inâ-objects), regardless of whether or
not the latter have Ā-features.
b. This results in obligatory UV whenever the object is a pronoun, making AV
impossible – exactly as for UV-only verbs.
c. C will then raise a v-flagged pronominal argument even if it doesn’t carry Ā-
features, thus making subject extraction from UV possible (no Ā-intervention).
(445) The Word Order Puzzle: why do φ+P objects give rise to special word orders?
a. In addition to consistently agreeing with all pronominal objects, if both argu-
ments are pronouns, v agrees with both.
b. The probe on T can trigger non-local A-movement, based on a φ-featural hi-
erarchy effect. It can raise the object, skipping the closer subject, if the former
has more “marked” φ-features than the latter.
c. This step of movement to spec,TP can in principle bring either [i |v]-flagged
argument closer to C, feeding CP-fronting.
Chapter 6. Non-canonical UV clauses 275
We are now ready to derive the extraction and word order properties of non-canonical
UV clauses, thus answering both puzzles.
6.5 DERIVING NON-CANONICAL UV CLAUSES
6.5.1 SEPARATING OUR CONCERNS: VOICE/EXTRACTION PROPERTIES AND
WORD ORDER
Let us now get into the details on how exactly the probes on v, T, and C interact. We have
two main aspects to focus on to derive the properties of the various types of UV clauses:
voice and Ā-extraction properties on one side, and word order on the other. We can thus
start subdividing the labor between the probes. The voice and Ā-extraction properties will
be controlled by v and C: v will agree with (and flag) various classes of arguments and not
others; C will raise the closest argument v agreed with, potentially creating an interven-
tion effect for Ā-extracting a lower argument (or not). The word order properties, on the
other hand, will be controlled by T and C (since v just agrees, but doesn’t move anything).
In this, T operates according to its own logic, without really paying any attention to what
v does. However, T may alter which argument is the closest [i |v]-flagged one to C, thus
potentially feeding CP-fronting. This is summarized in (446).
(446) Which probes control what:
a. Voice, Ā-extraction: v, C
b. Word order: T, C
To provide probe specifications that will get us the desired results, we then need to know
what exactly the three probes agree with (and move) in each type of clause. For T and C,
this is easy to know, since they move their goal to their specifiers, and we can directly
observe that. For v, the evidence is slightly more indirect. In AV clauses, we know by
hypothesis that v did not agree with anything. In any UV clause, in contrast, we know
that v has agreed with at least one argument (because the verb stem gets spelled out with
UV morphology). Moreover, by hypothesis, we know that whatever gets fronted by C
must have been agreed with by v.
Putting all this information together gives us the following landscape (447). For each
clause type, this table shows what we ultimately aim to derive: (i) what kind of subject
and object will result in this type of clause14; (ii) what argument(s) v, T, and C agree with
14 For simplicity, I only report in the table the combinations of subjects/objects that result in clauses with
unique properties, not reducible to any other clause type. For example, UV-only verbs can have any kind
of object: this could be either a lexical DP or a pronoun, with or without Ā-features (for example, there’s
no restriction against UV-only verbs taking pronominal objects, or against having Ā-features on pronouns).
Therefore, UVonly-clauses can in principle have a variety of objects. However, these clauses’ special proper-
ties only become evident when the object is a lexical DP without Ā-features. If the object of a UV-only verb
is a DP[Ā], the voice and extraction properties would be indistinguishable from those of a regular UVplain
276 6.5. Deriving non-canonical UV clauses
(and move); (iii) the resulting word order, including non-spelled-out traces (shaded); (iv)
whether the highest argument will have Ā-features, and thus create an intervention effect.
(447) Overview of agreement and movement patterns across clause types:
Type of Type of Agreement by: Ā-features
Clause type: Resulting word order: Ā-interv.?
subject: object: v T C on highest arg.?
a. AV: any ⎧DP ∅ S S S V.AV tS =TAM tS O not necessarily* no*
b. UVplain: any ⎪0 DP[Ā] O S O O V.UV S =TAM tS tO yes yesφ P
c. UVonly: any ⎨DP O S O O V.UV S =TAM tS tO not necessarily no⎪
d. UVinâ: any ⎩πP (inâ) O S O O V.UV S =TAM tS tO not necessarily no
e. UV : φ0gu P φ
+P O(+S)† O O O V.UV -πO =TAM S tO not necessarily no
f. UV : φ+P φ+SVO P O+S S S S V.UV -πS =TAM tS O not necessarily no
g. UV +1>2: φ P
‡ φ+P‡ O+S S+O S S V.UV -πS-πO=TAM tS tO not necessarily no
*Although in AV clauses there is no Ā-intervention (at least not necessarily), object extraction is still
impossible because if the object were extracted, the presence of the Ā-feature [REL] on it would au-
tomatically trigger object agreement and thus UV.
† It is impossible to determine with certainty whether in a UVgu clause, v agrees with the subject as
well; see §6.5.2.1 below for why.
‡ The subject and object must be respectively 1st and 2nd person to generate a UV1>2 clause.
This table is rather information-dense, so let us break it down, only considering the sub-
part that is immediately relevant for now. We will zoom in on the agreement patterns of
v, T, and C, across the various clause types. These are shown in the reduced table in (448).
In AV clauses, v agrees with nothing, and both T and C agree with and move the subject.
In UVplain, UVonly, and UVinâ clauses, v agrees with the object, T agrees with and moves
the subject, and C agrees with and moves the object. In UVgu clauses, all three probes
agree with the object. In UVSVO clauses, v agrees with both arguments, while T and C
agree with the subject. UV1>2 are a variant of UVSVO, where T agrees with and moves
both arguments.
(448) Agreement and movement patterns across clause types (simplified):
Type of Type of Agreement by:
Clause type:
subject: object: v T C
a. AV: any DP ∅ S S
b. UV 0plain/only: any φ P { DP[Ā] O S O
c. UVinâ: any πP (inâ) O S O
d. UV : φ0gu P φ
+P O(+S) O O
e. UV : φ+P φ+SVO P O+S S S
f. UV + +1>2: φ P(1st) φ P(2nd) O+S S+O S
clause. Similarly, if the object is a pronoun, we’ll get a UVinâ, UVgu, or UVSVO (or UV1>2) sentence, with the
properties that follow. See also fn. 13 for the same point.
Chapter 6. Non-canonical UV clauses 277
The same information is shown again in (449), in a different format. The way this table is
meant to be read is: given what the subject and object of a clause are, what do the three
probes agree with (and move)? Something that is easier to see from this visualization is
that clause type and voice are largely a function of what the object is; the subject does not
play as big of a role. The table can be “sliced” vertically according to what the object is.
If the object is just a DP with no Ā-features (first column), we get AV, no matter what the
subject is. If the object is a DP[Ā] or a πP (but still a φ
0P), we get clauses with canonical
UV order: these may be UVplain, UVonly (second column) or UVinâ (third column). Finally,
if the object is a φ+P, we get the non-canonical word orders (fourth column): UVgu if the
subject is a φ0P, or UVSVO if the subject is a φ
+P as well (UV1>2 would be a sub-case of
this red square).
(449) Agreement and movement patterns across subject/object types:
φ0P
S↓, O→
DP DP +[Ā] inâ φ P
v: ∅ v: O v: O v: O
DP T: S T: S T: S T: O
C: S C: O C: O C: O
v: ∅ v: O v: O v: O(+S)
φ0P DP[Ā] T: S T: S T: S T: O UVgu
C: S C: O C: O C: O
v: ∅ v: O v: O v: O(+S)
inâ T: S T: S T: S T: O
C: S C: O C: O C: O
v: ∅ v: O v: O v: O+S
φ+P T: S T: S T: S T: S(+O) UVSVO/1>2
C: S C: O C: O C: S
AV UVplain/only/inâ
6.5.1.1 THE FEATURAL MAKE-UP OF ÄIWOO NOMINALS
Before proceeding to giving a fully implemented revision of the probes on v and T, let us
first be explicit about what features are carried by different kinds of Äiwoo nominals. I
will assume the breakdown shown in (450). The goal of this featural make-up, of course,
is to make sure that the correct natural classes arise. Some of these featural assumptions
are somewhat non-standard, and ought to be commented on. Note that, however, much
of this list is negotiable and could be reformulated in a different feature system, as long as
the correct natural classes can arise. In this sense, these featural specifications should be
considered convenient shorthands, or a step towards a better/more explanatory theory.
278 6.5. Deriving non-canonical UV clauses
(450) Features on Äiwoo nominals:
∅3MIN: φ, D
φ0P Lexical DP: φ, D, (N)
inâ: φ, π, D
1MIN: φ, π, +PART, +AUTH, −ADDR, −AUG
12MIN: φ, π, +PART, +AUTH, +ADDR, −AUG
2MIN: φ, π, +PART, −AUTH, +ADDR, −AUG
φ+P 1AUG: φ, π, +PART, +AUTH, −ADDR, +AUG
12AUG: φ, π, +PART, +AUTH, +ADDR, +AUG
2AUG: φ, π, +PART, +AUTH, +ADDR, +AUG
3AUG: φ, π, −PART, −AUTH, −ADDR, +AUG
All (and only) nominals are defined as a class by having the feature [φ]. Then, pronouns
are singled out by a feature [π], which lexical DPs don’t have; viceversa, lexical DPs have
[D], and (most) pronouns don’t. The only form with both [π] and [D] is the overt 3MIN
pronoun inâ (and its proximal counterpart ine). Furthermore, any of these nominals can
also freely be merged with Ā-features on them, so that e.g. [PIV] could be added to the list
for all categories.
First, let us discuss “category” features, that is, [π] and [D]. We will see below that
we need probe specifications that refer to both the class of πPs (everything with [π]: all
φ+Ps and inâ) and φ0Ps (everything with [D]). Crucially, inâ is part of both classes. We
can be justified in assuming that inâ has [π] because, like all other pronouns, it shows the
pronominal stem i-, which we assumed to be the spell-out of the π head (chapter 5, §5.5),
Adding the feature [D], in contrast, is less empirically justified, but it will be necessary in
this model to get things to add up (see §6.5.1.2 for why).
As for person and number features, I adopt here a binary system (Noyer 1992, a.o.)
with [±PART, ±AUTH, ±ADDR]15. Importantly, I follow Grishin (2023, 2025) in assuming
that 3rd person is a feature, contra e.g. Harley & Ritter (2002), a.m.o. In this case, 3rd
person is represented as [−PART]. It will be important for me below that, perhaps some-
what oddly, only 3AUG has [−PART], and not 3MIN forms like inâ16. Perhaps, it could be
that the D head in Äiwoo is incompatible with [±PART], so that inâ cannot have this,
whereas other pronouns can; see Grishin (2025) for the empirical need of distinguishing
between [+PART], [−PART], and [ ] (absence of [±PART]). It is less crucial for the system
that [±AUTH,±ADDR,±AUG] be binary as opposed to privative, since Äiwoo grammar only
refers to the positive value; I notate them as binary for consistency.
15 This could, presumably, be reformulated in a system like Harbour (2016). Note that I am not trying to fit the
unit-augmented forms within the system; I leave this is as an open issue.
16 This oddity could be eliminated if one adopts a privative feature system where 3rd person is represented as
its own feature [3], as opposed to just being the absence of PART (Grishin 2023, Bondarenko 2020, a.o.). In
this case, both 3AUG and inâ would have [3]. This move, however, would force us to adopt a slightly more
complicated probe specification on T (see §6.5.1.3 for the current proposal). As far as I can see, the two
alternatives are somewhat equivalent, and I adopt the binary system purely for convenience.
Chapter 6. Non-canonical UV clauses 279
We can now move on to revising the probes on v and T to get the desired behavior,
that is, what they do and do not agree with (and move). The probe on C will remain the
same, searching for and moving the highest/closest [i |v]-carrying nominal.
6.5.1.2 v AGREES WITH Ā-MARKED DPS AND PRONOUNS
Throughout chapters 3–4we posited that v agreeswith objects carryingĀ-features (“DP[Ā]
objects”) – setting aside for now UV-only verbs and vH. In §6.4.1 we learned that UV is
consistently obligatory whenever the object is a pronoun. Therefore, this means that in
addition to Ā-marked objects, v also agrees with all pronominal objects (πPs). This cov-
ers both those pronouns that have marked φ-features (φ+Ps) and the 3MIN pronoun inâ,
but not ∅3MIN objects. To summarize, the intuition is that v agrees with two classes of
objects: Ā-marked ones, and pronouns. Moreover, we want the probe to agree with both
arguments only when they are both pronouns (which will result in a UVSVO/UV1>2 clause).
Concretely, I propose the following probe specification for Äiwoo v (451). Compared to
the one proposed in chapter 3, the only change was adding the disjunction “or π” to the
interaction condition. (We will see below that no changes are necessary in the probe spec-
ification for vH).
I: [Ā or π]
(451) Probe on v: [ ]
S: [D]
First, let us see how this specification gets us the desired results, again setting aside UV-
only verbs and vH for the time being. Because of its disjunctive interaction condition, v
will agree with the object if this carries either [Ā] or [π]; this captures the fact that all
Ā-marked objects and all pronominal objects trigger object agreement – that is, UV. The
satisfaction condition being set as [D] predicts that if the object is either a DP or inâ, this
will also halt the probe; note that it is possible for a goal to satisfy the probe without being
agreed with (this happens if the object is a bare DP, with no Ā-features). However, if the
probe has [π] but not [D] (like every pronoun apart from inâ), v will agree with the object
but not be satisfied. Therefore, it will reproject (by Cyclic Agree logic; chapter 3, §3.5.3.1)
and then potentially agree with the subject if it carries relevant features. The breakdown
of the logically possible combinations is given in the trees in (452). I notate disjunction as
“Ā/π” for simplicity, and I also add the interaction flag-feature [i |v] on all arguments that
v agrees with.
280 6.5. Deriving non-canonical UV clauses
(452) a. Object = DP: d. Object = πP (φ+P), subject = DP([Ā]):
no agr., halt; ⟹ AV agr., continue; ⟹ UVgu
vP vP ⟨φO, (φS)⟩
DP/πP([Ā]) DP([Ā])
[ ] [
v ′ ]
v VP I: [Ā/π]
S: [D]
I: [Ā/π]
S: [D]
V DP
[ v ] VP
I: [Ā/π]
b. Object = DP S: [D][Ā]: V πP
agr., halt; ⟹ UV [i|v]plain
vP ⟨φO⟩ e. Both subject and object = φ+P:
agr., continue; ⟹ UVSVO
DP/πP([Ā]) vP ⟨φO, φS⟩
[ v ] VP
I: [Ā/π]
S: [D] ′
V DP πP[i|v] v[i|v] [ ]I: [Ā/π]
[PIV] S: [D]
c. Object = inâ:
agr., halt; ⟹ UVinâ [ v ] VP
vP I: [Ā/π]⟨φO⟩ S: [D]
V πP[i|v]
DP/πP([Ā])
[ v ] VP
I: [Ā/π]
S: [D]
V πP[i|v]
π DP
Let us go through the space of possible derivations, depending on what the arguments
are. If the object is just a DP with no Ā-features (452a), v will not interact with it (that is,
there is no agreement), but it will be satisfied by it, and halt. This will ultimately give us
an AV clause. If the object is a DP carrying [PIV] (452b) – or any other Ā-feature, for that
matter – then v will agree with it and also halt, because the object has [D]. This will result
in a UVplain clause. Then, if the object is inâ (452c), it means it has both [π] and [D]; once
again, v will agree with it and halt. Crucially, [π] makes sure that object agreement will
always happen even if the object doesn’t have Ā-features: this will derive the fact that with
Chapter 6. Non-canonical UV clauses 281
inâ (and other pronominal objects), AV is impossible, and ultimately, subject extraction
will be possible from UV.
Let us now consider what happens if the object has [π], but not [D]; this is the case
of φ+P objects – that is, non-3MIN pronouns (452d,e). Again regardless of whether the
object has Ā-features or not, v will agree with it because it has [π] – giving us obligatory
UV. In this case, however, v is not satisfied, because it hasn’t found [D]. Therefore, it
will reproject to the bar level, and look at its sister node, the subject. If the subject is a
DP (452d), v will interact with it if it has Ā-features and then halt, or otherwise just halt
without interacting with the subject. If, however, the subject is a πP (452e), v will agree
with it as well17. Note that if the subject is a φ0P nominal with Ā-features (452d), due to
how the probing on T works, this step of subject agreement will effectively be completely
obscured; see below.
This gives us then the correct results.With this specification on v, wewill get UVwhen
the object is either an Ā-marked nominal or a pronoun. In the case of non-pronominal ob-
jects, agreement will be conditional on the presence of Ā-features. In the ordinary (non-
UVonly) case, what we get is a voice alternation correlating with the distribution of Ā-
features: this will ultimately cause the Ā-extraction restriction (see chapters 3–4). When-
ever the object is a pronoun (whether inâ or a φ+P), however, v will always agree with it
regardless of whether or not it has Ā-features, giving us obligatory UV and no extraction
restriction.
Let us now see how the probe specification proposed for vH in chapter 4 ([INT: φ; SAT:
D]) also still gives us the correct results despite this amendment to v. Remember that UV-
only verbs show the same exact set of word order alternations as all other transitive verbs
in the language; their only difference from ordinary verbs is that AV is not an option,
and consequently, subject extraction is allowed even with UVplain word order. In the case
of UV-only verbs, because Ā-features don’t figure in the probe specification of vH, their
distribution is essentially completely unconsequential. If the object is either a DP or inâ
(453a,b), vH will agree with it and immediately halt (due to its [D] feature); this will give us
the sameword order as in UVplain clauses. If, on the other hand, the object is a φ
+P (453c,d),
vH will agree with it without being satisfied, reproject, and agree with the subject as well.
Depending on which argument T moves to spec,TP, this may result in a clause with either
UVgu or UVSVO word order.
17 In this case too, the stem will be realized in its UV form, since the VI rule is sensitive to the presence of any
φ-features on v, not just specifically the object’s; see chapter 3, §3.5.1.
282 6.5. Deriving non-canonical UV clauses
(453) a. Object = DP: c. Object = πP (φ+P), subject = DP:
no agr., halt; ⟹ UVonly agr., continue; ⟹ UVgu
vHP vHP ⟨φO, φS⟩
DP/πP DP H ′[i|v] [ v[ ] ]vH VP I: [φ]S: [D]
I: [φ]
S: [D]
V DP [ vH ] VP
I: [φ]
S: [D]
V πP[i|v]
b. Object = inâ: d. Both subject and object = φ
+P:
agr., halt; ⟹ UVinâ agr., continue; ⟹ UVSVO
vHP ⟨φ ⟩ v
HP ⟨φO, φO S⟩
H ′
DP/πP πP[i|v] [ v ]
[ vH ] VP I: [φ]S: [D]
I: [φ]
S: [D]
V πP[i|v] [ vH ] VP
I: [φ]
π DP S: [D] V πP[i|v]
The mechanism proposed here is, of course, simply an implementation, and as mentioned
in the introduction of this chapter, the specific mechanic formalism will (or should) likely
be revised when our theory progresses. However, the conceptual underpinnings of the
system help bringing Äiwoo closer to many other languages, despite its initial strangeness
on the surface. On one hand, the classes of objects that v targets here (Ā-marked nomi-
nals and pronouns) are not necessarily a natural class in a coherent featural sense. On the
other hand, they are often treated by the syntax as if they were one class: pronouns and
Ā-marked objects are the typical categories of objects that in many languages show some
“special” behavior. For example, they might undergo object shift (like object pronouns in
Scandinavian; Holmberg 1986, Vikner 2017 a.o.), or get clitic-doubled (Anagnostopoulou
2017, a.o.), or trigger object agreement (Anagnostopoulou 2016, a.o.), or receive DOM
(Comrie 1986, Aissen 2003, a.o.), etc. See also Brodkin (2025) for another case of v (Voice
in his analysis) specifically agreeing with a similar disjunction of objects, implemented
in a way not too different from the one proposed here. Other than the empirical fact that
these two groups of objects often pattern alike in certain respects, to my knowledge we
Chapter 6. Non-canonical UV clauses 283
don’t know of a deeper explanation of this. Here, I implemented this common behavior
between Ā-marked nominals and pronouns by means of a featural disjunction in the prob-
ing mechanism. This is purely a mechanical device to make sure that both classes receive
the same treatment. Hopefully, further research will be able to shed light on why exactly
these two classes of objects often receive “special attention”.
6.5.1.3 T RAISES THE CLOSEST φ+P (BUT IT HAS A PLAN B)
In chapter 3, whenwe only considered canonical AV and UVplain clauses, we kept our story
simple: T always raises the closest nominal – the subject – to spec,TP. However, we know
now that this is not in fact always true. The summary of which argument(s) move(s) to
spec,TP in the different kinds of clauses is shown in (454); for now, I ignore the following
movement to spec,CP (shaded). In UVgu clauses (454b), the object moves to spec,TP, and
the subject stays low; note that this movement is non-local. In UV1>2 clauses (454d), both
arguments move to spec,TP.
(454) What raises to spec,TP, by clause type:
a. UVplain/only/inâ: T raises the subject
[CP Obj C [TP Subj T [vP tS v … tO
b. UVgu: T raises the object
[CP Obj C [TP Obj T [vP Subj v … tO
c. UVSVO: T raises the subject
[CP Subj C [TP Subj T [vP tS v … Obj
d. UV1>2: T raises both arguments
[CP Subj C [TP Subj Obj T [vP tS v … tO
Putting this together with how the different clause types are distributed according to the
φ-features of the two arguments, we reach the following state of affairs (455).
(455) What T agrees with and moves:
φ+P φ0P
S↓, O→
1 12 2 3AUGπ inâ DP
1 — — S+O S S S
⎧
+ ⎪ 12 — — — S S Sφ P
⎨ 2 S — — S S S⎪
⎩ 3AUGπ S S S S S S
0 inâ O O O O S Sφ P {
DP O O O O S S
284 6.5. Deriving non-canonical UV clauses
Temporarily setting aside the 1 > 2 combination, we can now draw a relatively simple gen-
eralization: T raises the closest φ+P; if there are no φ+Ps, it raises the closest nominal. If
the subject is a φ+P (top four rows), T simply agrees with andmoves the subject, and that’s
all; what the object is does not matter (456a,b). If the subject is a φ0P but the object is the
desirable φ+P (bottom two rows), then T will agree with and move the object, despite the
subject being closer (456c). This is another instance of non-local A-movement, once again
motivated by the “featural priority” of the lower argument over the higher one18. Only
when neither argument is a φ+P (i.e., when both arguments are φ0Ps), T “backtracks” onto
a less specific preference, and just raises the highest/closest nominal – that is, the subject
(456d).
(456) T raises the closest φ+P if it can find one:
a. φ+P > φ+P: T raises subj
b. φ+P > φ0P: T raises subj
c. φ0P > φ+P: T raises obj!
d. φ0P > φ0P: T raises subj (“plan B”)
Movement to spec,TP then shows two similarities to movement to spec,CP (chapter 3,
§3.5.3.1): (i) it is non-local A-movement: it prefers a more specific type of nominal, and
will rather move a lower such nominal over a closer, less featurally fitting one; (ii) it has
a plan B: if it cannot find its preferred type of nominal, it will resort to a much more
generic featural specification, and just move the closest nominal. Before we proceed to
the mechanical implementation, we can verify that the non-local movement of the object
to spec,TP (skipping the subject) shows at least one binding-theoretical property of A-
movement: it obviates Weak Crossover. In (457), the quantified pronominal object ijidui
‘all of them’ moves first to spec,TP (where it’s realized as the shortened object form -gu-i;
see §5.5) across the subject isäi ‘their mother’. Crucially, this sentence can have a bound
reading, showing that this step of movement is not sensitive to WCO19. (The object then
raises further to spec,CP, but this is irrelevant here.)
(457) [Context: we are talking about a group of girls who are coming back from a long
trip, and their mothers are missing them very much.] (250422)
3!
ijidui ki-te-usi-kä -gu-i =laa isä -i tObj
3AUG.all IPFV-see.UV-again.UV-DIR3-OBJ-3AUG=FUT mother-3AUG
3 Bound reading: ∀𝑥 , 𝑥 ’s mother will see 𝑥 again
≈ ‘Heri mother will see every onei of them again’
18 As a sidenote, the existence of non-local A-movement to spec,TP goes against a prediction made in
Lohninger (2025) that this kind of movement – though implementationally framed in a slightly different way
– should only be possible to phase edges (e.g., spec,CP or spec,vP). There is no evidence for the existence
of a phase boundary in this region of the Äiwoo clause.
19 See chapter 3, fn. 45 for discussion of a potential analytical confound that arises here just as in the context
of CP-fronting. Because the quantified moving element is grammatically plural, it might be that the WCO-
violating reading that is available for (457) is actually not derived via genuine binding, but rather through
cumulativity. I leave this possibility open for future research.
Chapter 6. Non-canonical UV clauses 285
From our earlier discussion about CP-fronting, we know how to derive this kind of non-
local A-movement with a plan B. T will have a primary probe that is featurally “picky”,
and a secondary, more generic probe that only kicks in when the primary probe has not
found anything; see chapter 3, §3.5.3 for the technical details. In the spirit of separating
the analytical intuition from the weeds of the mechanical implementation, the diagram in
(458) informally describes what T does, this time incorporating the 1>2 combination. A
reader who is not necessarily interested in the details of a particular theory of agreement
might want to focus on this diagram, and not the mechanical implementation below.
(458) Probing by T (informal):
What category
is the subject?
φ+P φ0P
Raise the subject What category
is the object?
Is the subject 1st person
and the object 2nd person? φ+P φ0P
Yes No Raise the object Plan B:
raise the subject
Raise the object too Stop
To implement this sequence of operations, I propose that Äiwoo T hosts the following
two probes (459)20. Note that “3AUG” here is a shorthand notation for the list of features
[−PART, −AUTH, −ADDR, +AUG]. In the trees below, for typographical reason, I will use the
condensed notation on the right-hand side.
(459) Probes on T:
I: [±PART↑]M I: [±PART↑]M
a. Primary probe: [ ] = [ ]
S: [+ADDR or 3AUG] S: [AD/3AUG]
I: [φ]M
b. Secondary probe: [ ] = [ I+S: [φ]M]
S: [φ]
20 This effectively supersedes the analysis and proposal in Roversi (2020), which was based on partially gappy
data. Specifically, it was not known at the time that 1AUG>2nd person combinations resulted in UV1>2 sen-
tences rather than UVSVO. This minor difference had implications for the formulation of the hierarchy effect.
286 6.5. Deriving non-canonical UV clauses
Let us break down the picky primary probe, themore interesting one21. Its interaction con-
dition is [±PART↑]: it will only agree with and move nominals that contain either [+PART]
or [−PART], but not those without (either value of) this feature. This means that at first
pass, T is only interested in φ+Ps; remember that only 1st person, 2nd person, and 3AUG
pronouns are specified for [±PART], but not DPs and the 3MIN pronoun inâ (see §6.5.1.1).
The symbol ↑ on the feature [±PART] signifies “dynamic interaction” (Deal 2024a): af-
ter T agrees with a goal with [+PART], the interaction condition becomes [+PART], and
viceversa for [−PART]22. The satisfaction condition is set to the disjunction of [+ADDR] or
[3AUG]; goals that lack either of these may be interacted with but they will not halt the
probe, which will continue searching. Importantly, this is the case for 1st person (exclu-
sive) arguments: they are [+PART], so they get interacted with, but they are [−ADDR], so
they do not stop the probe.
Let us now concretely see how this complicated probe (together with the simpler non-
picky probe) actually derives the attested pattern. Rather than going through every single
combination, I have set this up as an algorithm written in pseudo-code (460); see e.g. Deal
(to appear) for this way of using pseudo-code. The algorithm goes through different classes
of subjects, at each step highlighting and “checking off” the parts of the complete pattern
(455) that have been accounted for.
(460) Full algorithm for T probing:
 Primary probe. If subject has [−part] or equivalently [3aug]:
▷ Interact with subject, move subject, halt (SAT!)
S↓, O→ 1 12 2 3AUGπ inâ DP
1 — — S+O S S S
12 — — — S S S
2 S — — S S S
3AUGπ S S S S S S 3
inâ O O O O S S
DP O O O O S S
 Else, if subject has [+part] (S ∈ {{1/12/2}min, {1/12/2}aug})
▷ Interact with subject, move subject
▷ Dynamic interaction: copy [+part] onto the INT condition
21 This is admittedly a rather complex specification, which might seem purely ad-hoc to correctly capture the
data – and it is. However, the data pattern we are trying to capture is objectively complex, so a simpler
specification will not achieve descriptive adequacy. The existence of agreement/movement patterns of this
complexity can be thought of as supporting evidence for a more powerful theory of agreement like Deal’s
(2015) interaction/satifaction model, over more restrictive ones. For conjunction and disjunction in probing,
and specifically for person/number features in Deal’s framework, see e.g. Coon & Bale (2014), Roversi (2020),
Bondarenko & Zompì (2021), Scott (2021), Oxford (2024), and Deal (2024a), a.o.
22 Note that I assume here a slightly different version of dynamic interaction than that proposed in Deal (2024a)
and Deal & Royer (2025). There, being dynamic is a property of a feature on a goal, whereas here, I assume
that it is a property of a feature on the probe. See 6.5.1.4 for a discussion.
Chapter 6. Non-canonical UV clauses 287
▷ If subject has [+addr]: (S ∈ {{12/2}min, {12/2}aug})
▷ Halt (SAT!)
S↓, O→ 1 12 2 3AUGπ inâ DP
1 — — S+O S S S
12 — — — S S S
2 S — — S S S 3
3AUGπ S S S S S S
inâ O O O O S S
DP O O O O S S
▷ Else: (S ∈ 1min, 1aug)
▷ If object has [+part]: (O ∈ {2min, 2aug})
▷ Interact with object, move object
▷ Object has [+addr]: halt (SAT!)
▷ Else: (O ∈ {3aug.pro, inâ, DP})
▷ Do nothing; halt (end of c-command domain)
S↓, O→ 1 12 2 3AUGπ inâ DP
1 — —3 S+O S S S 3
12 — — — S S S
2 S — — S S S
3AUGπ S S S S S S
inâ O O O O S S
DP O O O O S S
 Else, if subject does not have [±part] (S ∈ {inâ, DP}):
▷ If object has [±part]:
▷ Interact with object, move object, halt
(end of c-command domain)
S↓, O→ 1 12 2 3AUGπ inâ DP
1 — — S+O S S S
12 — — — S S S
2 S — — S S S
3AUGπ S S S S S S
inâ O O O O S S
DP O O O O 3 S S
▷ Else: (O ∈ {inâ, DP})
▷ Don't interact with anything, halt
(end of c-command domain)
 Plan B-probe. If Plan A-probe moved any nominal:
▷ Interact with that nominal, halt (SAT!)
(= don't do anything)
288 6.5. Deriving non-canonical UV clauses
 Else, if Plan A-probe did not move anything:
▷ If subject has [phi]: interact with subject, move subject,
halt (SAT!)
S↓, O→ 1 12 2 3AUGπ inâ DP
1 — — S+O S S S
12 — — — S S S
2 S — — S S S
3AUGπ S S S S S S
inâ O O O O S S
DP O O O O S S 3
We have now seen how the interaction of two probes can derive the complex hierarchy
effect of what T raises to its specifier. Crucially, movement to spec,TP may alter which
[i |v]-flagged nominal will be closest to C, and therefore be further raised to spec,CP. For
example, in a UVgu sentence, the object has [i |v] because it is a pronoun. Even if the subject
also has [i |v] (due to how v works), the object will be raised to spec,TP, and thus C will
front it. In a UVSVO clause, on the other hand, both arguments have [i |v], but T raises the
subject. Therefore, C will front the subject.
6.5.1.4 DYNAMIC INTERACTION FEATURES AS A PER-PROBE PROPERTY
In the discussion of probing by T above I briefly mentioned a departure I make in a de-
tail of the mechanics of Agree outlined in Deal (2024a), Clem & Deal (2024), and Deal &
Royer (2025). Because this is not necessarily an innocent move, it is worth discussing. The
difference concerns the nature and distribution of dynamic interaction features. If an in-
teraction feature [F] is dynamic (notated [F↑]), it will be copied onto the interaction set of
the probe upon successful agree, so that further steps of agree will be restricted to goals
that also bear [F]. The question at hand concerns where exactly features can be dynamic,
or said otherwise, what kind of syntactic objects dynamic features can be a property of. I
contrast three possible views:
(461) Where are dynamic features?
a. Dynamicity as a property of goals. Two sub-possibilities:
i. Per feature: across a whole grammar, every instance of a given feature [F]
is dynamic [F↑]. For example, within the same language, it is not possible
that a 1st person pronoun has dynamic [(+)PART↑] while a 2nd person
pronoun has non-dynamic [(+)PART]. (Deal 2024a, Clem & Deal 2024)
ii. Per lexical item: dynamicity is a property of a feature on a particular lexical
item. For example, it is possible that a 1st person pronoun has [(+)PART↑],
while 2nd person has just [(+)PART]. (Entertained as a possibility in Deal &
Royer 2025: 1637–1638)
b. Dynamicity as a property of probes:
Whether a feature is dynamic or not is specified on a particular probe. Within
Chapter 6. Non-canonical UV clauses 289
the same grammar, it is possible that a probe (say, on v) will have [F] but an-
other one (say, on T) will have [F↑].
As indicated by the citations, the available literature has argued that dynamicity is a prop-
erty of features on agreement goals (461a). In contrast to this, in our discussion above I
modeled dynamicity as a property of agreement probes (461b). This second kind of the-
ory (461b) is strictly more powerful/expressive than the per-feature version of the first
(461a-i); let us see why.
Consider a language where three probes (P1, P2, P3) all dynamically interact with,
say, [PL] – that is: after they agree with a plural DP, their further search is restricted
to other goals carrying [PL], and will ignore other (singular) DPs. Under a per-feature
theory (461a-i), we would say that this not a coincidence: [PL↑] is a dynamic feature in
the language, and therefore every probe interacting with a nominal carrying [PL] will do
so dynamically. Under a per-probe theory (461a), this would be a coincidence: the only
reason why all three probes interact dynamically with [PL] is that all three probes just so
happen to be specified as having [INT: PL↑].
Because a per-probe theory can express everything a per-goal theory can, plus some
additional cases, an argument against the per-probe theory will need to say that it is too
powerful/unrestricted: the additional cases it predicts are not attested. This is what Deal &
Royer (2025) do with evidence from Mayan languages, where probes both in the nomi-
nal and the clausal domain all interact dynamically with the same features in the same
way. From this perspective, the per-probe theory would just have to say that the fact that
these probes act alike is a coincidence, which may be inelegant or unparsimonious. In
contrast, an argument in favor of the per-probe theory would be to show that those ad-
ditional scenarios that it predicts to be possible (and that are impossible under a per-goal
theory) actually do exist. To reiterate, these are cases where a probe P1 interacts dynami-
cally with, for example, [PL], while another probe P2 does not. This is simply not possible
under the per-goal theory (whether in the per-feature or per-item version). If [PL↑] is
specified as dynamic on the goals themselves, then every probe will have to interact with
it dynamically.
I argue that this is precisely the scenario we observe in Äiwoo, and that therefore
we empirically need the less restrictive per-probe theory. Specifically, I argue that T in
Äiwoo interacts dynamically with [±PART], while v does not. For T, we know that interac-
tion with [±PART] is dynamic (§6.5.1.3): after interacting with a [+PART] subject, [−PART]
objects are not considered (which means [+PART] was copied onto the interaction set of
the probe), and viceversa. For v, in contrast, this cannot be the case. The crucial cases
are those where both arguments are φ+P, but with differing [±PART] specifications, i.e.,
1/2 > 3AUG or 3AUG > 1/2; let’s consider the combination of 1MIN and 3AUG for concrete-
ness (462). If [±PART] interacted dynamically here (or even if either only [+PART] or only
[−PART] did), then we would predict that v will agree only with the object first, copy its
value of [±PART] into its interaction set, and then fail to interact with the subject due of
its opposite value. However, we know that this prediction is wrong: in all combinations
290 6.5. Deriving non-canonical UV clauses
of φ+P > φ+P, crucially regardless of [±PART] specifications, v consistently agrees with
both arguments (§6.5.1.2), which is what ultimately results in UVSVO clauses.
(462) v cannot interact dynamically with [±PART]:
a. 1MIN > 3AUG: 7 wrong prediction b. 3AUG > 1MIN: 7 wrong prediction
vP vP
πP [ v ′ ] πP [ v ′ ]
1MIN I: [−PART] 3AUG I: [+PART]
[+PART(↑), …] S: [D] [−PART(↑), …] S: [D]
[ v ] [ v ]
I: [Ā/π] πP I: [Ā/π] πP
S: [D] 3AUG S: [D] 1MIN
[−PART↑, …] [+PART↑, …]
In conclusion, I argue that although a per-probe theory of dynamic interaction is less re-
strictive than a per-goal theory, this additional power is empirically necessary, as shown
by the different probing patterns of v and T in Äiwoo. For completeness: a per-item the-
ory, as entertained by Deal & Royer (2025), is also more expressive/less restrictive than a
per-feature theory, but in different ways than a per-probe theory. Specifically, a per-item
theory predicts certain scenarios to be possible that are not predicted under a per-probe
theory. These would be cases where, say, a certain probe (or several probes, for that mat-
ter) interacts dynamically for [PL] with a 1PL goal, but not with a 2PL goal; Deal & Royer
(2025: 1637–1638) argue that the agreement patterns in several Mayan languages might be
analyzed this way. If such cases exist, and if it is true that the Äiwoo case is one that can
only be modelled with a per-probe theory of dynamicity, then we might want to carefully
reconsider the entire idea of dynamic interaction.
The general typology of scenarios predicted to be possible by these three theories is
shown in (463). In each case, I consider an abstract situation where two probes, P1 and
P2, interact with two goals, G1 and G2, each of which carries a feature [F] (and some
other features); to remain notationally neutral, I put the symbol for dynamic interaction
↑ on the dashed agreement line rather than on the goals or probes themselves. In a per-
goal, per-feature theory (463a), [F] is dynamic across the board, so both P1 and P2 will
interact dynamically with both G1 and G2. In a per-goal, per-item theory (463b), [F] may
be dynamic on one goal but not the other: both probes will interact dynamically with
G1, and neither will with G2. In a per-probe theory (463c), P1 will interact dynamically
with (all instances of) [F], whereas P2 will not. Scenarios that are not predicted to be
possible under any of these theories, as far as I can tell, are for example one where P1
only interacts dynamically with G1 and P2 only with G2 (463d-i), or one where P1 only
interacts dynamically with G1 and not G2, and P2 with neither (463d-ii).
Chapter 6. Non-canonical UV clauses 291
(463) a. Per-goal, per-feature: [F] is dynamic on all goals
G1
↑ [F, X, …] ↑
P1 P2
↑ ↑
G2
[F, Y, …]
b. Per-goal, per-item: [F] is dynamic on G1 but not on G2
G1
↑ [F, X, …] ↑
P1 P2
G2
[F, Y, …]
c. Per-probe: P1 interacts dynamically with [F], but not P2
G1
↑ [F, X, …]
P1 P2
↑
G2
[F, Y, …]
d. Not predicted by any of these theories:
i. G1 ii. G1
↑ [F, X, …] ↑ [F, X, …]
P1 P2 P1 P2
↑
G2 G2
[F, Y, …] [F, Y, …]
The breakdown ofwhich theory predicts which scenarios to be possible is shown as a Venn
diagram in (464). Both a per-probe theory and a per-item theory predict the across-the-
board scenario in (463a) to be possible, but each of them also predicts additional scenarios
that the other one does not.
(464) Prediction of the three theories:
per probe per feature per item
(463c) (463a) (463b)
292 6.5. Deriving non-canonical UV clauses
In this chapter, I have argued that an agreement pattern like (463c) is actually attested in
Äiwoo, and that therefore a per-probe theory is necessary. I hope future research will shed
light both on which empirical patterns are actually crosslinguistically attested, and also,
on the theoretical side, on the precise relation between the expressive powers of these
alternative theories.
6.5.2 PUTTING IT ALL TOGETHER AGAIN: STEP-BY-STEP DERIVATIONS
In this section, I put all the ingredients together, and provide full annotated derivations.
First I go through the non-canonical cases of UVgu, UVSVO and UV1>2 clauses (§§6.5.2.1–
6.5.2.3). For these three cases I use as an example derivation one where the object does not
carry [PIV], to show how the lack of an Ā-extraction restriction arises. Then, as a sanity
check, I reconsider the derivation of canonical UVplain and AV clauses (§§6.5.2.4–6.5.2.5)
to show that the amendments we have made to the original model from chapter 3 still
derive the correct results. Throughout, I remain agnostic as the precise syntactic nature
of the directional suffix -mä DIR1, representing it as part of the complex V head; this is
not crucial.
I do not also include a derivation for UVonly clauses, involving a UV-only verb root and
vH instead of the ordinary v head. This would proceed in the same way as either a UVplain,
UVgu, UVSVO, or UV1>2 derivation, depending on the features of the two arguments, but
with no necessity for Ā-features on the object in the first case (where it looks superficially
like a UVplain clause), and thus no extraction restriction. See §6.5.1.2 above for a discussion
of how the probe specification posited for vH still predicts the correct results in terms of
word order alternations.
6.5.2.1 DERIVING UVgu CLAUSES
DISTRIBUTION: The subject must be a φ0P; the object must be a φ+P.
EXAMPLE SENTENCE
(465) (iude) ku-tu-mä-gu-de=ngaa mekivaavee
12AUG IPFV-bring.UV-DIR1-OBJ-12AUG=FUT teacher
‘The teacher will bring us (here)’
Chapter 6. Non-canonical UV clauses 293
IN THE vP DOMAIN:
(466) vP ⟨φO, φS⟩
(φ0P =) DP ′[PIV, i|v] [ v ⟨φ]O, φS⟩
I: [Ā/π]
S: [D]
teacher
v VP
tV πP +[i|v] (= φ P)
V [ v ⟨φO]⟩
I: [Ā/π]
√ S: [D] π φ(P)
BRING DIR1 [12AUG]
 The (complex) verb head-moves into V.
 v agrees with the object because it has [π], whether or not it has Ā-features. The
object receives the interaction flag feature [i |v]. The object does not have [D], so it
does not satisfy the probe.
▷ Because [π] is present on the object, AV will be impossible.
 v reprojects to the bar level. Here I illustrate a case where the subject has [PIV], so
v agrees with the subject too (giving it [i |v]).
▷ Ultimately, this won’t matter: even if the subject was just an Ā-less DP and v
thus didn’t agree with it, the rest of the derivation would be identical.
294 6.5. Deriving non-canonical UV clauses
TP DOMAIN:
(467) TP
πP ′[i|v] [ T ]
I+S: [φ]M
π φ(P)
[12AUG]
[ T ] AspP
I:[ [±PART↑]
M
S: [AD/3AUG]]
I+S: [φ]M
Asp vP
Asp v
[IPFV] DP[PIV, i|v] tv VP
V v ⟨φO, φS⟩ teacher
tV πP
√
BRING DIR1
π φ(P)
[12AUG]
 The verb head-moves into Asp.
 T’s primary probe initiates a search.
▷ The subject does not have [±PART]; the probe does not interact with it and it
is not satisfied.
▷ The object also has [+PART]. The probe interacts with it and moves it to
spec,TP.
▷ In this case the object also has [+ADDR], so the probe is satisfied. (even
if it didn’t, nothing would change in the rest of the derivation).
 T’s secondary probe reprojects to T′, and initiates a search.
▷ The closest potential goal in its c-command domain is the just-moved object
in spec,TP.
Chapter 6. Non-canonical UV clauses 295
▷ The object has [φ], so the probe interacts with it (and perhaps moves it vac-
uously), and it is satisfied.
CP DOMAIN:
(468) CP
πP ′[i|v] [ C ]
I+S: [φ]M
π φ(P)
[12AUG]
C TP
[ C ⟨φO[ ]]
⟩ Asp πP[i|v]
I+S: [i|v]M
M T AspPI+S: [φ]
Asp
V v π φ(P)
[FUT]
[IPFV] ⟨φO, φS⟩ [12AUG] tAsp vP
√
BRING DIR1
DP[PIV, i|v] tv VP
teacher
…
 The verb head-moves from Asp to C, skipping T.
 The primary probe on C searches for [i |v]. Although both arguments have it, the
object is now closest to C after having been raised to spec,TP. C raises the object to
spec,CP.
▷ The secondary probe on C reprojects, and from the bar level it agrees with
the object in spec,CP again.
 The object is now in spec,CP even without having Ā-features. Therefore, there will
be no intervention effect.
296 6.5. Deriving non-canonical UV clauses
VERB SPELL-OUT:
 Post-syntactic Local Dislocation applies to the πP in spec,TP, but not to the higher
copy in spec,CP (§5.5.2).
(469) CP
πP[i|v]
π φ(P)
[12AUG]
TP
πP
C πP T AspP
[FUT]
…
π φ(P)
C ⟨φ ⟩ [ACC, 12AUG]O
∅
∅ Asp
[IPFV] √ v ⟨φO, φS⟩
BRING DIR1
ku- tu -mä -gu-de =ngaa
(470) Fission: φ(P)[ACC, 12AUG] ⟹ φ(P) / V
[ACC] φ[12AUG]
(471) Impoverishment:
a. C impoverishment: applies
☞ C⟨φ⟩ ⇔ ∅ / v ⟨φ⟩
b. φ(P) impoverishment: does not apply
7 φ(P) ⇔ ∅ / C⟨φ⟩
Chapter 6. Non-canonical UV clauses 297
(472) Vocabulary Insertion:
a. Stem (voice allomorphy):
☞ √BRING ⇔ tu / v ⟨φ⟩
√BRING ⇔ tou (elsewhere)
b. π is not word-initial, so it gets deleted:
π ⇔ i~iu- / #
☞ π ⇔ ∅ (elsewhere)
c. φ(P):
i. [ACC] ⇔ -gu / V
ii. φ[12AUG] ⇔ -de
6.5.2.2 DERIVING UVSVO CLAUSES
DISTRIBUTION: Both the subject and the object must be φ+Ps.
EXAMPLE SENTENCE
(473) (iu) ku-tu-mä-no=ngaa ijii
1MIN IPFV-bring.UV-DIR1-1MIN=FUT 3AUG
‘I will bring them (here)’
IN THE vP DOMAIN:
(474) vP ⟨φO, φS⟩
(φ+P =) πP[i|v] [ v ′ ⟨φ]O, φS⟩
I: [Ā/π]
S: [D]
π φ(P)
[1MIN]
v VP
tV πP[i|v] (= φ+P)
V [ v ⟨φO]⟩
I: [Ā/π]
√ S: [D] π φ(P)
BRING DIR1 [3AUG]
 The (complex) verb head-moves into V.
 v agrees with the object because it has [π], whether or not it has Ā-features. The
object receives the interaction flag feature [i |v]. The object does not have [D], so it
does not satisfy the probe.
298 6.5. Deriving non-canonical UV clauses
▷ Because [π] is present on the object, AV will be impossible.
 v reprojects to the bar level. The subject has [π], so v agrees with the subject too
(giving it [i |v]).
 v is not satisfied, but its c-command domain has been exhausted, so v just stops.
TP DOMAIN:
(475) TP
πP ′[i|v] [ T ]
I+S: [φ]M
π φ(P)
[1MIN]
[ T ] AspP
I: [±PART↑]M
S[: [AD/3AUG]]
I+S: [φ]M
Asp vP
Asp v
[IPFV] πP[i|v]
V v ⟨φ , φ ⟩ tv VPO S
√ π φ(P)
BRING DIR1 [1MIN] tV πP[i|v]
[3AUG]
 The verb head-moves into Asp.
 T’s primary probe initiates a search.
▷ The subject has [+PART]; the probe interacts with it. The interaction condition
is now [+PART], because of dynamic interaction.
▷ The subject does not have [ADDR] nor [3AUG], so the probe is not satisfied,
and it keeps searching.
Chapter 6. Non-canonical UV clauses 299
▷ The object does not have [+PART], so the probe does not interact with it. (The
object does have [3AUG], so it satisfies the probe, but this is inconsequential,
because the object doesn’t get raised.)
 T’s secondary probe reprojects to T′, and initiates a search.
▷ The closest potential goal in its c-command domain is the just-moved subject
in spec,TP.
▷ The object has [φ], so the probe interacts with it (and perhaps moves it vac-
uously), and it is satisfied.
 Note: the specific sequence of steps for T’s primary probe is dependent on the exact
combination of φ-features on the subject and the object. I chose here 1MIN > 3AUG as
an example to provide a minimal contrast with 1MIN > 2MIN, which triggers UV1>2;
see the next section, and (460) for a breakdown of all possible combinations.
CP DOMAIN:
(476) CP
πP ′[i|v] [ C ]
I+S: [φ]M
π φ(P)
[1MIN]
C TP
[ C ⟨φS⟩ Asp[I+S: [i|v]M]] πP[i|v] T AspP
I+S: [φ]M
Asp π φ(P) [FUT]
[IPFV] V v ⟨φO, φS⟩ [1MIN] tAsp vP
√
BRING DIR1 … πP[i|v, 3AUG]
300 6.5. Deriving non-canonical UV clauses
 The verb head-moves from Asp to C, skipping T.
 The primary probe on C searches for [i |v]. The closest argument with this feature
is the subject in spec,TP. C agrees with and fronts the subject.
▷ The secondary probe on C reprojects, and from the bar level it agrees with
the subject in spec,CP again.
VERB SPELL-OUT
 Post-syntactic Local Dislocation applies to the πP in spec,TP, but not to the higher
copy in spec,CP (§5.5.2).
(477) CP
πP[i|v]
π φ(P)
[1MIN] TP
πP
C πP T AspP
[FUT]
π φ(P) …
C ⟨φ ⟩ [1MIN]S
∅
∅ Asp
[IPFV] √ v ⟨φO, φS⟩
BRING DIR1
ku- tu -mä -no =ngaa
(478) Impoverishment:
a. C impoverishment: applies
☞ C⟨φ⟩ ⇔ ∅ / v ⟨φ⟩
b. φ(P) impoverishment: does not apply
7 φ(P) ⇔ ∅ / C⟨φ⟩
Chapter 6. Non-canonical UV clauses 301
(479) Vocabulary Insertion:
a. Stem (voice allomorphy):
☞ √BRING ⇔ tu / v ⟨φ⟩
√BRING ⇔ tou (elsewhere)
b. π is not word-initial, so it gets deleted:
π ⇔ i~iu- / #
☞ π ⇔ ∅ (elsewhere)
c. φ(P): φ[1MIN] ⇔ -no
6.5.2.3 DERIVING UV1>2 CLAUSES
DISTRIBUTION
 A sub-case of UVSVO. The subject must be 1st person (MIN or AUG), and the object
must be 2nd person (MIN or AUG).
EXAMPLE SENTENCE
(480) (iu) ku-tu-mä-nee-mu=waa
1MIN IPFV-bring.UV-DIR1-2MIN-2MIN=FUT
‘I will bring you (here)’
IN THE vP DOMAIN: IDENTICAL TO UVSVO
(481) vP ⟨φO, φS⟩
(φ+P =) πP ′[i|v] [ v ⟨φ]O, φS⟩
I: [Ā/π]
S: [D]
π φ(P)
[1MIN]
v VP
tV πP[i|v] (= φ+P)
V [ v ⟨φO]⟩
I: [Ā/π]
√ S: [D] π φ(P)
BRING DIR1 [2MIN]
 The (complex) verb head-moves into V.
302 6.5. Deriving non-canonical UV clauses
 v agrees with the object because it has [π], whether or not it has Ā-features. The
object receives the interaction flag feature [i |v]. The object does not have [D], so it
does not satisfy the probe.
▷ Because [π] is present on the object, AV will be impossible.
 v reprojects to the bar level. The subject has [π], so v agrees with the subject too
(giving it [i |v]).
 v is not satisfied, but its c-command domain has been exhausted, so v just stops.
TP DOMAIN:
(482) TP
πP[i|v] TP
π φ(P)
[1MIN]
πP[i|v] [ T′ ]
I+S: [φ]M
π φ(P)
[2MIN]
[ T ] AspP
I: [±PART↑]M
S[: [AD/3AUG]]
I+S: [φ]M
Asp vP
Asp v
[IPFV] πP[i|v]
V v ⟨φO, φ ⟩ tv VPS
√ π φ(P)
BRING DIR1 [1MIN] tV πP[i|v]
[2MIN]
Chapter 6. Non-canonical UV clauses 303
 The verb head-moves into Asp.
 T’s primary probe initiates a search.
▷ The subject has [+PART]; the probe interacts with it and moves it to spec,TP.
The probe’s interaction condition is now [+PART], because of dynamic in-
teraction. The subject does not have [ADDR] or [3AUG], so the probe is not
satisfied and keeps searching.
▷ The object has [+PART], so the probe interacts with it, and raises it to spec,TP
(tucking it in, under the subject). The object also has [ADDR], so the probe is
satisfied.
 T’s secondary probe reprojects to T′, and initiates a search.
▷ The closest potential goal in its c-command domain is either the subject or
the object, depending on one’s definition of c-command and “closest”; either
way, the secondary probe is vacuously satisfied.
CP DOMAIN:
(483) CP
πP[i|v] [ C′ ]
I+S: [φ]M
π φ(P)
[1MIN]
C TP
[ C ⟨φS⟩[ ]
Asp
I+S: [i|v]M] πP[i|v]
I+S: [φ]M
Asp πP
V v ⟨φ , φ ⟩ π φ(P)
[i|v]
[IPFV] O S T AspP
[1MIN]
√ π φ(P)
[FUT]
BRING DIR1 [2MIN] tAsp vP
…
304 6.5. Deriving non-canonical UV clauses
 The verb head-moves from Asp to C, skipping T.
 The primary probe on C searches for [i |v]. The closest argument with this feature
is the subject in spec,TP. C agrees with and fronts the subject.
▷ The secondary probe on C reprojects, and from the bar level it agrees with
the subject in spec,CP again.
VERB SPELL-OUT:
 Post-syntactic Local Dislocation applies to both πPs in spec,TP, but not to the higher
copy of the subject in spec,CP (§5.5.2).
▷ I represent it the constituency here as [[C πPS] πPO] for concreteness, but it
could equally well be [C [πPS πPO]].
(484) CP
πP[i|v]
π φ(P) TP
[1MIN]
πP πP
πP
T AspP
π φ(P) [FUT]
C πP [2MIN] …
∅
π φ(P)
C ⟨φ ⟩ [1MIN]S
∅
∅ Asp
[IPFV] √ v ⟨φO, φS⟩
BRING DIR1
ku- tu -mä -nee -mu =waa
Chapter 6. Non-canonical UV clauses 305
(485) Impoverishment:
a. C impoverishment: applies
☞ C⟨φ⟩ ⇔ ∅ / v ⟨φ⟩
b. φ(P) impoverishment: does not apply (to either)
7 φ(P) ⇔ ∅ / C⟨φ⟩
(486) Vocabulary Insertion:
a. Stem (voice allomorphy):
☞ √BRING ⇔ tu / v ⟨φ⟩
√BRING ⇔ tou (elsewhere)
b. Both π’s are not word-initial, so they get deleted:
π ⇔ i~iu- / #
☞ π ⇔ ∅ (elsewhere)
c. φ(P):
i. φ[1MIN] ⇔ -nee / φ[2MIN]
ii. φ[2MIN] ⇔ -mu
6.5.2.4 DERIVING UVPLAIN CLAUSES, UPDATED
DISTRIBUTION: The object must be a φ0P; it could be either a DP with Ā-features (as
shown here), or inâ, which has both [π] and [D] features. The subject could be any cate-
gory of nominal.
EXAMPLE SENTENCE:
(487) sii ku-tu-mä pedevalili=kaa
fish IPFV-bring.UV-DIR1 children=FUT
‘The children will bring the fish (here)’
IN THE vP DOMAIN:
(488) vP ⟨φO⟩
DP
children v VP
V [ v ⟨ ]⟩ t DP 0V [i|v] (= φ P)φO
I: [Ā/π] [PIV]
√ S: [D]
BRING DIR1 fish
306 6.5. Deriving non-canonical UV clauses
 The (complex) verb head-moves into V.
 v agrees with the object because it has [PIV]. The object receives the interaction flag
feature [i |v]. The object also has [D], so it satisfies the probe.
TP DOMAIN:
(489) TP
DP
children [ T ] AspP
I: [±PART↑]M 7
S:[[AD/3AUG]]
I+S: [φ]M
Asp vP
Asp v DP
[IPFV] tv VP
V v ⟨φO⟩ children tV DP[PIV, i|v]
√
BRING DIR1
fish
 The verb head-moves into Asp.
 T’s primary probe initiates a search.
▷ Neither the subject nor the object have [±PART]. The primary probe doesn’t
interact with anything and just stops (represented by 7).
 T’s secondary probe initiates a search.
▷ The subject has [φ] features, so the secondary probe interacts with it, raises
it to spec,TP, and stops.
 Note: if the subject was, e.g., a 1st person pronoun, the primary probe on T would
raise it and the secondary probe would be automatically satisfied from the bar level.
Crucially, the object will never get raised to spec,TP if it’s a φ0P.
Chapter 6. Non-canonical UV clauses 307
CP DOMAIN:
(490) CP
DP ′[PIV, i|v] [ C ]
I+S: [φ]M
fish
C TP
[ C ⟨φO]⟩ Asp DP[I+S: [i|v]M] T AspP
I+S: [φ]M children [FUT]
Asp
[IPFV] V v ⟨φO⟩ tAsp vP
√
BRING DIR1 tSubj …
… DP[PIV, i|v]
fish
 The verb head-moves from Asp into C, skipping T.
 The [i |v]-probe on C initiates a search. The closest argument with [i |v] is the object,
in its base position. C raises it to spec,CP, and the probe is satisfied.
▷ The secondary probe on C reprojects to the bar level and it is vacuously sat-
isfied by the just-moved object.
 The object in spec,CP carries the Ā-feature [PIV], so Ā-extracting the subject will be
impossible due to intervention.
SPELL-OUT OF THE VERB:
 No Local Dislocation applies (because there are no πPs).
308 6.5. Deriving non-canonical UV clauses
(491) CP
DP
fish
C TP
C ⟨φO⟩ DP
Asp
∅ [IPFV] V v ⟨φO⟩ children T AspP
[FUT]
√
BRING DIR1 tAsp vP
… tO
ku- tu -mä pedevalili =kaa
(492) C impoverishment: applies
☞ C⟨φ⟩ ⇔ ∅ / v ⟨φ⟩
(493) Stem (voice allomorphy):
☞ √BRING ⇔ tu / v ⟨φ⟩
√BRING ⇔ tou (elsewhere)
6.5.2.5 DERIVING AV CLAUSES, UPDATED
DISTRIBUTION: The object can only be a lexical DP, with no Ā-features. I represent the
subject here as carrying [PIV], but this is not necessary. I use a pronominal subject as an
example to show how the impoverishment mechanism works, but it could just as well be
a lexical DP.
EXAMPLE SENTENCE:
(494) (iude) de-ku-tou-mä=kaa sii
12AUG 12AUG-IPFV-bring.AV-DIR1=FUT fish
‘We will bring fish (here)’
Chapter 6. Non-canonical UV clauses 309
IN THE vP DOMAIN:
(495) vP
πP
π φ(P) v VP
[PIV] [12AUG]
V [ v ] tV DP
I: [Ā/π]
√ S: [D] fish
BRING DIR1
 The (complex) verb head-moves into V.
 The v probe initiates a search. It does not interact with the object, because it doesn’t
have [Ā] nor [π]. However, the object has [D], so v is satisfied and halts.
TP DOMAIN:
(496) TP
πP [ T′ ]
[PIV, 12AUG] I+S: [φ]M
[ T ] AspP
I:[ [±PART↑]
M
S: [AD/3AUG]]
I+S: [φ]M
Asp vP
Asp v πP
[IPFV] [PIV, 12AUG] tv VP
V v
tV DP
√
BRING DIR1
fish
310 6.5. Deriving non-canonical UV clauses
 The verb head-moves into Asp.
 T’s primary probe initiates a search.
▷ Neither the subject nor the object have [±PART]. The primary probe doesn’t
interact with anything and just stops (represented by 7).
 T’s secondary probe initiates a search.
▷ The subject has [φ] features, so the secondary probe interacts with it, raises
it to spec,TP, and stops.
 Note: if the subject was a φ0P, the primary probe on T would not find anything,
but the secondary one would move the subject anyway, as shown here, so nothing
would change.
CP DOMAIN:
(497) CP
πP
[PIV, 12AUG]
C TP
[ C ⟨φS⟩] Asp πP[I+S: [i|v]M] 7 [PIV, 12AUG] T AspP
I+S: [φ]M
Asp [FUT]
[IPFV] V v tAsp vP
√
BRING DIR1 … DP
 The [i |v]-probe on C initiates a search. Nothing in its c-command domain has this
feature, so this probe stops without agreeing with anything (7).
 The secondary probe just agrees with the closest constituent with [φ]-features, that
is, the subject in spec,TP. The subject is raised to spec,CP.
Chapter 6. Non-canonical UV clauses 311
 The subject has Ā-features ([PIV]), so it will block Ā-extraction of a lower argument.
▷ If the subject did not have Ā-features, in principle there would be no inter-
vention effect. However, if the object were to be Ā-extracted, it would have
Ā-features, which means we would never get AV to begin with.
SPELL-OUT OF THE VERB:
 Post-syntactic Local Dislocation applies to the πP in spec,TP, but not to the higher
copy in spec,CP (§5.5.2).
(498) CP
πP
[PIV, 12AUG]
TP
πP
C πP T AspP
[FUT]
π φ(P) …
C ⟨[12AUG]⟩ [PIV] [12AUG]
Asp ∅ ∅
[IPFV] √ v
BRING DIR1
de- ku- tou -mä =kaa
(499) Impoverishment:
a. C impoverishment: does not apply
7 C⟨φ⟩ ⇔ ∅ / v ⟨φ⟩
b. φ(P) impoverishment: applies
☞ φ(P) ⇔ ∅ / C⟨φ⟩
(500) Vocabulary Insertion:
a. Stem (voice allomorphy):
√BRING ⇔ tu / v ⟨φ⟩
☞ √BRING ⇔ tou (elsewhere)
312 6.6. Summary
b. Agreement on C:
C⟨[12AUG]⟩ ⇔ de-
6.6 SUMMARY
This chapter concludes the contentful part of this dissertation.We started in chapter 1with
two interconnected puzzles, the Word Order Puzzle, and the Extraction Puzzle: transitive
clauses showed a variety of different word orders and voice morphology, each with its
own Ā-extraction possibilities. In this chapter I have shown that this complex system
of clausal alternations ultimately arises from the interplay between three probes in the
Äiwoo clausal spine: v, T, and C. Crucially, all probe-goal interactions are strictly local, in
featurally relativized sense of “local”. Every probe will agree with (and sometimes move)
the closest goal with some property [α], and then either continue probing or halt. In no
part of the system is there any step of “global” calculus, that is, where a probe “waits” until
having examined the properties of multiple arguments to make a decision as to which one
to agree with (and move). Abstracting away from a lot of the implementational/technical
complexity, an informal summary of what these three heads do is shown in (501).
(501) Three probes and their operations:
a. v: agrees with Ā-marked objects and pronominal objects.
b. T: agrees with and raises the closest nominal with “marked” φ-features.
▷ If there is none, plan B: raise the closest nominal.
c. C: agrees with and raises the closest nominal that v agreed with.
▷ If there is none, plan B: raise the closest nominal.
The voice and Ā-extraction properties of any transitive clause are controlled by the object,
in a local interaction with v or vH (502). As we saw, v in Äiwoo shows a familiar type of
behavior: it agrees with two classes of objects, namely pronouns and nominals with Ā-
feaetures. We thus have a system that is only rather indirectly sensitive to the presence
of Ā-features on the object. If they are present, v will certainly agree with it. This will
give us a UV clause, and if the object ends up as the highest argument, its Ā-features
will make it an intervener for Ā-extracting any lower argument. If, however, the specific
lexical verb root selects for vH (which agrees with all objects), there will always be object
agreement no matter what category the object is. In this case, we will have UV, but with
no Ā-extraction restriction.
Chapter 6. Non-canonical UV clauses 313
(502) Voice and Ā-extraction properties are determined by the object and v/vH:
With v: With vH:
What category is the object? With any object:
DP[−Ā] DP[+Ā] πP[±Ā]
No agr. Obj agr. Obj agr. Obj agr.
⇒ AV ⇒ UV ⇒ UV ⇒ UV
Ā-features matter: Ā-features don’t matter:
3 Voice alternation 7 Voice alternation
3 Ā-extraction restriction 7 Ā-extraction restriction
The two higher heads, T and C, control the final word order that a transitive clause will
have. Although the probes on these two heads are “interested” in different features, they
show some mechanical similarities: they can both trigger non-local A-movement of a
lower argument across a higher one, and they both have a “plan B” in case their pri-
mary type of goal is not around. T shows a hierarchy effect, raising the closest argument
that has “marked” φ-features (i.e., anything other than 3MIN). This featural preference
may result in raising the object to spec,TP and leaving the subject low, given the right
combination of arguments. In turn, this may alter which [i |v]-flagged argument is closest
to C. If the argument that C ends up raising carries Ā-features ([PIV]), we will have an
intervention effect, and therefore an Ā-extraction restriction. Otherwise, there will be no
intervention, and Ā-extraction will be free.
6.A THE CAVEAT: PLURAL LEXICAL DPS
The discussion of the Word Order Puzzle, and specifically of the system of word order al-
ternations in §6.3 largely excluded one category of nominals: plural lexical DPs (by which
I simply mean lexical DPs with non-atomic reference). The reason why this exclusion was
made is that plural DPs ultimately don’t fit neatly into the φ+P vs. φ0P dichotomy, and
they behave sometimes as if they belonged to one group, sometimes to the other. Consider
the table in (503), an expanded version of the original overview table (407) making more
fine-grained distinctions. As above, “π” abbreviates “pronoun”, and DP stands for specif-
ically lexical DPs. Here, I have highlighted the row and column for plural DPs. The left
half of the 3AUGDP row has a split color-coding because these combinations of arguments
can result in either kind of clause. The data is rather complex, so for ease of exposition,
314 6.A. The caveat: plural lexical DPs
together with every minimal pair of sentences I will repeat the matrix table, highlighting
in turn the relevant cells under comparison.
(503) Complete distribution of UV clause types:
φ+P φ0P
S↓, O→
1 2 3AUGπ 3AUGDP 3MINπ 3MINDP
1 – UV1>2 UVSVO UVplain UVplain UVplain
φ+P { 2 UVSVO – UVSVO UVplain UVplain UVplain
3AUGπ UVSVO UVSVO UVSVO UVplain UVplain UVplain
3AUGDP UVSVO/gu UVSVO/gu UVSVO/gu UVgu UVplain UVplain
3MIN UV
φ0P { π gu
UVgu UVgu UVgu UVplain UVplain
3MINDP UVgu UVgu UVgu UVgu UVplain UVplain
As a note going forward, everything I say in this section only applies to human/animate
DPs. Inanimate DPs are completely number-neutral, never showing any kind of plural
morphology/morphosyntax in any part of the language (in other words, always behaving
syntactically as if they were singular, regardless of their reference); see chapter 2, §2.5.1.
Here, we observe a complex split behavior, where plural DPs behave sometimes like
φ0Ps (i.e., like other lexical DPs), and sometimes like φ+Ps (i.e.), like 3AUG and other pro-
nouns). First, consider plural DP objects, the highlighted column in (503). Let us first con-
sider the upper half of the table, where the subject is a φ+P; for example, a 2nd person
pronoun. In this case, plural DP objects behave like other φ0P objects, so we get UVplain
sentences (504a). Here, plural DP objects are different from 3AUG pronoun objects, which
in this case generate UVSVO sentences (504b).
(504) If the subject is a φ+P:
Objects: {plural DPs = φ0Ps} ≠ {φ+Ps}
+
S↓, O→ φ P φ
0P
1 2 3AUGπ 3AUGDP 3MINπ 3MINDP
1 – UV1>2 UVSVO UVplain UVplain UVplain
φ+P { 2 UVSVO – UVSVO (504b) UVplain (504a) UVplain UVplain (504a)
3AUGπ UVSVO UVSVO UVSVO UVplain UVp=lain UVplain3AUGDP UVSVO/gu UVSVO/gu UVSVO/gu ≠ UVgu UVplain UVplain
0 3MINφ P { π
UVgu UVgu UVgu UVgu UVplain UVplain
3MINDP UVgu UVgu UVgu UVgu UVplain UVplain
a. φ+Ps > {plural DPs, φ0Ps}: UVplain (250807)
John/pedevalili ku-tu-kâ-mu=waa ngä skul
John/children IPFV-bring.UV-DIR3-2MIN=FUT to school
‘You will bring John/the children to school’
b. φ+Ps > {3AUG pronouns}: UVSVO
( iumu) ku-tu-kâ-mu=waa ijii ngä skul
2MIN IPFV-bring.UV-DIR3-2MIN=FUT 3AUG to school
‘You will bring them to school’
Chapter 6. Non-canonical UV clauses 315
However, if the subject is a φ0P (lexical DP or 3MIN pronoun; bottom half of the table),
we observe a different pattern. Now, a plural DP object patterns together with other φ+P
arguments (505a), and differently from other (singular) DPs (505b).
(505) If the subject is a φ0P:
Objects: {plural DPs = φ+Ps} ≠ {φ0Ps}
+ 0
S↓, O→ φ P φ P
1 2 3AUGπ 3AUGDP 3MINπ 3MINDP
1 – UV1>2 UVSVO UVplain UVplain UVplain
φ+P { 2 UVSVO – UVSVO UVplain UVplain UVplain
3AUGπ UVSVO UVSVO UVSVO UVplain UVplain UVplain
3AUGDP UVSVO/gu UVSVO/gu UVSVO/gu = UVgu UVp≠lain UVplain
0 3MIN UV UV UV UV UV UVφ P { π gu gu gu gu plain plain3MINDP UVgu UVgu UVgu (505a) UVgu (505a) UVplain UVplain (505b)
a. φ0Ps > {plural DPs, 3AUG pronouns}: UVgu (250807)
∅3AUG /ijii/pedevalili ku-tu-kä-gu-i=laa John ngä skul
3AUG/children IPFV-bring.UV-DIR3-OBJ-3AUG=FUT John to school
‘John will bring them/the children to school’
b. φ0Ps > {singular DPs}: UVplain
devalili ku-tu-kä John=kaa ngä skul
child IPFV-bring.UV-DIR3 John=FUT to school
‘John will bring the child to school’
To summarize: plural DP objects behave like φ0Ps (like other singular DPs) if the subject
is a φ+P, but they behave like φ+Ps (like 3AUG pronouns) if the subject is a φ0P.
For plural DP subjects, the matter is even more complicated. First, if the object is a φ+P
(e.g. a 2nd person pronoun; left half of the table), we get two options – hence the split
color-coding. A plural DP subject can either behave like a φ0P (resulting in UVgu), or like
a φ+P (resulting in UVSVO). This seems to be optional, and the factors behind this choice
(if any) are not known.
(506) If the object is a φ+P:
Subject: {plural DPs = φ+Ps}, OR {plural DPs = φ0Ps}
φ+ 0S↓, O→ P φ P
1 2 3AUGπ 3AUGDP 3MINπ 3MINDP
1 – UV1>2 UVSVO UVplain UVplain UVplain
φ+P { 2 UVSVO – UVSVO UVplain UVplain UVplain
3AUGπ UVSVO UVSVO (506a) =UVSVO UVplain UVplain UVplain3AUGDP UVSVO/gu UVSVO/gu (506a,b) UVSVO/gu UVgu UVplain UVplain
0 3MINπ UVgu UVgu =UVgu UVgu UVφ P { plain UVplain3MINDP UVgu UVgu (506b) UVgu UVgu UVplain UVplain
a. Option 1: {plural DPs, 3AUG pronouns} > φ+Ps: UVSVO (250807)
∅3AUG /ijii/mikilivaavee ku-tu-kä-i=laa iumu ngä skul
3AUG/teachers IPFV-bring.UV-DIR3-3AUG=FUT 2MIN to school
‘They/the teachers will bring you to school’
316 6.A. The caveat: plural lexical DPs
b. Option 2: {plural DPs, φ0Ps} > φ+Ps: UVgu
( iumu) ku-tu-kâ-gu-mu=waa ∅3MIN /John/mikilivaavee ngä skul
2MIN IPFV-bring.UV-DIR3-OBJ-2MIN=FUT John/teachers to school
‘He/she/John/the teachers will bring you to school’
However, if both the subject and the object are plural DPs, then there is no optionality, and
UVgu is the only possible alternative (507a). UVplain and UVSVO are impossible (507b,c).
(507) Plural DP > Plural DP: (250807)
a. 3 UVgu:
pedevalili ku-tu-kä-gu-i=laa mikilivaavee ngä skul
children IPFV-bring.UV-DIR3-OBJ-3AUG=FUT teachers to school
‘The teachers will bring the children to school’
b. 7 UVplain:
* pedevalili ku-tu-kä mikilivaavee=kaa ngä skul
children IPFV-bring.UV-DIR3 teachers=FUT to school
Intended: ‘The teachers will bring the children to school’
c. 7 UVSVO:
*mikilivaavee ku-tu-kä-i=laa pedevalili ngä skul
teachers IPFV-bring.UV-DIR3-3AUG=FUT pedevalili to school
Intended: ‘The teachers will bring the children to school’
At present, I don’t have a proposal to explain the rather complex distribution of plural DP
arguments. One potential avenue might be to consider the internal syntax of plural DPs
as a possible factor in their external distribution. As discussed in chapter 2, §2.5.1, most
“plural DPs” are really relativized expressions, containing the relativizers mi= ‘the one(s)
that […]’ or pe= ‘people that […]’:
(508) a. mi=ki-li-vaavee
REL=IPFV-3AUG-show
‘Teachers’, lit. ‘People who show/teach’
b. pe=devalili
REL:people=child
‘Children’
A possible intuition to pursue is the following. If these “plural DP” nominals are really
headless (or null-headed) relative clauses, then the head of the constituent is not trivially
a lexical DP, but perhaps some kind of (null) pronominal element (“those who …”). Maybe,
the fact that these “plural DPs” have a distribution that is in between that of plural pro-
nouns and singular DPs might at least partially derive from the syntactic nature of the
head of the relative clause. However, it is not clear what would cause the split between
plural DP subjects vs. objects. I leave developing a full account of this complex pattern as
an open issue.
CHAPTER 7
CONCLUSION AND PROSPECTS
7.1 SUMMARY OF THE FINDINGS
This dissertation started with the promise of accounting for two intertwined puzzles: an
alternation in the word order and verbal morphology of transitive clauses, and an alter-
nation in what argument(s) can be Ā-extracted from the various types of clauses. The
summary of the various clause types is repeated in (509).
(509) Alternations of word order and Ā-extractability:
Clause type: Word order:
a. AV: S3 φS-V.AV =TAM O
7
b. UV 3plain: O V.UV S
7 =TAM
c. UV 3only: O V.UV S
3 =TAM
d. UVinâ: O
NA V.UV S3 =TAM
e. UV : ONAgu V.UV -πO =TAM S
3
f. UV NASVO: S V.UV -π =TAM O
NA
S
g. UV NA1>2: S V.UV -πS-πO=TAM
Devising a formal model that would generate all and only the attested clause types with
their specific Ā-extraction profile has let us gain new insights in a number of unrelated
areas of syntactic theory.
Some of these theoretical issues are more specific to the Austronesian literature. I have
proposed amodel of Austronesian voice where “Undergoer Voice” is really a form of object
agreement, which is only partially tied to both the Ā- and A-syntax of the language. Fur-
ther, I have offered an account of an interesting instantiation of the classic Austronesian
“pivot-only” Ā-extraction restriction, since in Äiwoo this comes with a series of excep-
318 7.2. Open questions
tions. Precisely because the restriction in Äiwoo only holds in certain contexts but dis-
appears in others, I have modeled it as an Ā-intervention effect: if the highest argument
carries Ā-features, extracting anything lower will be impossible; else, it will be possible.
This was prompted by the novel empirical discovery of UV-only transitive verbs, which
have not been reported for other Austronesian languages. Ultimately, this analysis of the
voice and extraction system is somewhat different from influential analyses of these phe-
nomena in other Austronesian languages, which have been framed as either a “highest
DP-only” restriction or an effect of phasehood (see chapter 4 for references).
Other issues this dissertation bears on are much more general in their theoretical
scope, and concern the notion of locality of agreement and movement. I have shown that
Äiwoo bears out an implicit prediction our theory already made, that is, the possibility of
featurally motivated, non-local A-movement. This should not be thought of as a differ-
ent type of movement from classic A/Ā-movement, or as a form of “mixed” movement.
Rather, I argued that the typical “no skips” locality profile of classic A-movement is purely
epiphenomenal, and that since probes can be relativized to only targeting goals with cer-
tain features and not others (Béjar 2003), it should be possible for A-movement-triggering
probes to also skip certain nominals in favor of others.
Finally, the model I have built of the word order alternation system entirely revolves
around the desires of probes to agree with (and move) certain types of arguments over
others. Importantly, no appeal is made to any kind of need of the arguments themselves
(i.e., a Greed-based analysis). Any kind of argument can ultimately end up remaining in
their base position (or any position, for that matter) and never get agreed with, if the
relevant probes find a better/closer goal. This is not an argument in and of itself against
the concept of nominal licensing, or a claim that it is not needed as part of the theory.
However, it might at least provide a sort of “proof of concept”: at least for Äiwoo, an
analysis based on licensing/Greed is not only not necessary, but also not desirable.
7.2 OPEN qUESTIONS
The analysis I propose, of course, leaves a number of issues open for future research.
Some of them are at a slightly more narrowly implementational level (e.g., what exactly is
the featural make-up of Äiwoo pronouns/nominals, given the hierarchy effect governing
raising to spec,TP?), while others have a deeper theoretical significance to them.
One such issue is the idea that C probes for [i |v], the interaction flag feature of v. On
one hand, this could be thought of as aminor technical innovation. The idea of information
transfer from a probe to a goal goes back to at least Chomsky (2000, 2001) and Pesetsky &
Torrego (2001, 2007), and the more specific concept of goal flagging features has been used
in Deal (2022, to appear), Clem &Deal (2024), andWu et al. (2023) to model morphological
effects. However, this is to my knowledge the first proposal where a flag feature, after
being assigned, is then later manipulated by the syntax itself, making something carrying
a flag feature into a possible target for movement. We could follow Clem & Deal 2024
Chapter 7. Conclusion and prospects 319
(but also Pesetsky & Torrego 2001, 2007, anachronistically) in thinking of goal flagging as
the mechanism behind case assignment. Under this view, “accusative case” or “nominative
case” are not more than the descriptive labels for the flag features assigned by respectively
v or T (typically, at least).
This would then bring the Äiwoo pattern of CP-fronting closer to known patterns
of case discrimination, where a probe may target arguments bearing certain cases but
not others. These are known for both pure agreement patterns (e.g. Hindi and Nepali;
Bobaljik 2008), Ā-movement (Tongan for Otsuka 2006, 2010, Chukchi for Polinsky 2016,
Deal 2016b, 2017, Nukuoro for Drummond 2023, a.o.), and also A-movement if cliticiza-
tion/clitic-doubling are examples of A-movement (see Deal 2025 and references therein).
For concreteness, let’s say that [i |v] is what we are used to calling “accusative case”: in
a classic Burzio-an way v assigns accusative case (Burzio 1986). In this case, Äiwoo CP-
fronting would constitute an example of “positive” case discrimination (Akkuş et al. 2025
call this ‘case targetting’), where a probe is specified as only targeting nominals with the
more “marked” case features rather than the less marked ones, at least under standard/
reasonable assumptions about the featural make-up of different cases (see Caha 2009, et
seq.). This would then be a counterexample to Bobaljik (2008), i.e., exactly the opposite of
the expected pattern. Any process that targets nominals with a more marked case value
should also target nominals with all less-marked cases, but here, we see the reverse. Akkuş
et al. (2025) argue that such patterns do in fact exist in a series of varieties of Kurdish, so
Äiwoo would constitute another example from an unrelated language family. It remains
to be seen whether (i) this expansion of the expressive power of the theory is actually
empirically warranted or not, and (ii) goal-flagging and case are really one and the same
phenomenon.
The potential relation between goal flagging and case (whether morphological case
or abstract Case) is also an interesting question from the perspective of another central
issue in this dissertation, namely that that of the locality of movement. Building on van
Urk (2015), I have supported his argument that locality is completely epiphenomenal with
respect to the A/Ā-distinction. More specifically, I have argued that once one connects van
Urk’s (2015) featural theory of the A/Ā-distinction and Béjar’s (2003) notion of featurally
relativized probes, one automatically predicts the existence of “non-local A-movement”.
I show that this kind of movement does in fact take place in Äiwoo, both to spec,TP (as
determined by a complex hierarchy effect governed by φ-features) and, crucially for the is-
sue at hand, to spec,CP (as determined by the flag feature [i |v]). Because raising to spec,CP
has the binding-theoretical properties of A-movement, this must mean that [i |v] has some-
thing in common with other features that canonically trigger this kind of movement, such
as φ-features and, more importantly, case/Case features. This fact gives at least prelimi-
nary supports to idea of connecting case and goal flagging, which I hope will be a fruitful
avenue for future research.
Finally, due to Äiwoo’s underresearched character, of course there still are a lot of
open questions that are purely empirical. Here, just as an example, I will name three (out
of many more). One such open question concerns the external syntax of plural lexical
320 7.2. Open questions
DPs, which is highly complex; sometimes, they are treated like other (non-plural) lex-
ical DPs, but other times as if they were pronouns. See chapter 6, appendix 6.A for a
breakdown of the known empirical facts, with no attempt at a solution. Another severely
under-explored area is the syntax of questions, and particularly of wh-questions. Unlike
in other Austronesian languages where they have featured prominently in the discussion
of voice and Ā-phenomena, wh-questions in Äiwoo seem to be not too tied to the voice
distinction at all. Finally, the syntax of “Circumstantial Voice” and its argument structure,
its relation the rest of the voice system, the clausal alternation system, and Ā-extraction
is still rather poorly understood. I put together some notes in Appendix A.
APPENDIX A
NOTES ON CIRCUMSTANTIAL VOICE
A.1 INTRODUCTION
This appendix gives a few notes about a kind of construction that has been labelled “Cir-
cumstantial Voice” (CV) in the literature (Næss 2015b et seq.). Although I will keep this
label for convenience, it will become clear that this is not a “voice” in the same way as
Actor Voice and Undergoer Voice, that is, it is not in paradigmatic alternation with them.
Therefore, CV in Äiwoo is different in this sense from its terminological equivalents in
other Austronesian languages, such as e.g. Circumstantial Voice and Locative Voice in
Tagalog (Schachter & Otanes 1972 et seq.), etc.
First, in §A.2 I will give a description of the argument structure, word order, ver-
bal morphology, and Ā-extraction properties of CV clauses in Äiwoo. The material I dis-
cuss here contains both facts already reported in the literature and novel findings. Then,
in §A.3 I discuss how integrating CV into the larger analysis of Äiwoo clauses and Ā-
extraction presents some interesting challenges. I will entertain a possible analysis, and
discuss where it breaks.
A.2 DESCRIPTIVE PROPERTIES
A.2.1 WORD ORDER, VERBAL MORPHOLOGY, AND ARGUMENT STRUCTURE
WORD ORDER The simplest way to describe at a glance what CV does is that it adds one
argument to the existing argument structure of a verb, and simultaneously makes it the
VIA (Næss 2015b et seq.). The added argument is typically some kind of “oblique”, such
as a locative, instrument, etc. (see below for more details). If the base verb is intransitive,
adding CV to it will result in a clause with two arguments; if it is transitive, then there will
322 A.2. Descriptive properties
be three arguments. This is the only way in Äiwoo to create a clause with three nominal/
DP arguments (though see §A.3 below for discussion of whether the added argument is
underlyingly a nominal). The general template of a CV clause is shown in (510), where
‘X’ is the additional argument contributed by CV; concrete examples will be given below.
(510) Abstract template of CV clauses:
X V S=TAM=Cä (O) (… PPs and adjuncts…)
In terms of word order, we can recognize the same three linear “slots” where arguments
can go that we find in the various kinds of intransitive and (mono-)transitive clauses
discussed throughout the dissertation. The preverbal VIA position (spec,CP) hosts the
added argument. The subject is between the verb and the TAM particles (spec,TP), as in
UVplain clauses. Finally, if there is an object, this is in the postverbal position. Additional
PPs and adjuncts are, as usual, linearized to the right.
VERBAL MORPHOLOGY Unlike AV and UV, CV itself is not morphologically marked by
stem suppletion, but by adding the enclitic marker =Cä at the right edge of the TAM se-
quence1. This shows that CV is not in paradigmatic alternationwith AV and UV, but rather
is “built on top” of them. To support this point, when CV is added on top of a transitive
verb, the base verb can take either its AV or its UV stem. Äiwoo CV is therefore func-
tionally similar to “Circumstantial Voice” in other Austronesian languages (e.g. Tagalog,
Malagasy, etc.), seen as it adds an “oblique” argument to the existing ones and promotes
it to VIA/pivot. Nonetheless, the Äiwoo case is syntactically rather different from those
other languages, since it is not clearly part of the voice paradigm but rather added on top
of it. In this respect, Äiwoo CV is at least on the surface more similar to the system of
applicatives found in Indonesian-type languages (Himmelmann & Riesberg 2013, Næss
2015b).
The data below shows concrete examples of CV sentences built on top of an intran-
sitive verb (511), a transitive AV verb (512), and a transitive UV verb (513). For each pair,
the (a) sentence is a non-CV baseline, i.e. either an intransitive or transitive clause with a
PP adjunct, and the (b) sentence is the corresponding CV version2. In each case, we find
that: (i) what corresponds to the adjunct PP in the baseline is now a DP (with no prepo-
sition) in the preverbal position; (ii) the enclitic CV marker =Cä is added after the future
marker; (iii) if there was subject agreement in the baseline, like for intransitives and AV
verbs (511a)-(512a), it disappears in the CV version (511b)-(512b).
(511) Intransitive (+ CV): (250814)
a. kuli ki-li-ngoduwâ=kaa ngä paveli eângâ INTR + PP
dog IPFV-3AUG-shout=FUT in garden that
‘The dogs will bark in that garden’
1 The first consonant of this particle varies following the same allomorphy pattern as future =Caa, distal =Câ
and proximal =Ce; see chapter 2, §2.6.3.2.
2 For exposition purposes, I limit the discussion for now to CV clauses where the added argument is a lexical
DP. See §A.3.2 for a discussion of what happens when the added argument is a pronoun/φ+P.
Appendix A. Notes on Circumstantial Voice 323
b. paveli eângâ ki-ngoduwâ kuli=kaa=kä INTR + CV
garden that IPFV-shout dog=FUT=CV
‘The dog(s) will bark in that garden’
(512) Transitive, AV (+ CV): (250814)
a. pedevalili ki-li-vängä=kaa sii ngä täpilo enge AV + PP
children IPFV-3AUG-eat.AV=FUT fish in bowl this
‘The children will eat fish in this bowl’
b. täpilo enge ki-vängä pedevalili=kaa=kä sii AV + CV
bowl this IPFV-eat.AV children=FUT=CV fish
‘The children will eat fish in this bowl’
(513) Transitive, UV (+ CV): (250814)
a. sii ki-ngä pedevalili=kaa ngä täpilo enge UV + PP
fish IPFV-eat.UV children=FUT in bowl this
‘The children will eat the fish in this bowl’
b. täpilo enge ki-ngä pedevalili=kaa=kä sii UV + CV
bowl this IPFV-eat.UV children=FUT=CV fish
‘The children will eat the fish in this bowl’
If the subject is a pronoun rather than a lexical DP, it will be realized in its short suffixal
form rather than the full pronoun. For UV, a subject pronoun in this position (spec,TP)
would already be realized in its short suffixal form, so nothing changes when CV is added
(514). However, this creates a morphologically unique situation where an intransitive or
AV verb takes φ-suffixes rather than the usual prefixal agreement (515)–(516); this is oth-
erwise impossible outside of CV.
(514) a. Baseline UV: φ-suffix, no φ-prefix (250814)
sii ki- ngä-de =ngaa ngä täpilo enge
fish IPFV-eat.UV-12AUG=FUT in bowl this
‘We will eat the fish in this bowl’
b. UV + CV: φ-suffix, no φ-prefix
täpilo enge ki- ngä-de =ngaa=kä sii
bowl this IPFV-eat.UV-12AUG=FUT=CV fish
‘We will eat the fish in this bowl’
(515) a. Baseline INTR: φ-prefix, no φ-suffix (250814)
∅3AUG ki- li-ngoduwâ =kaa ngä paveli eângâ
IPFV-3AUG-shout=FUT in garden that
‘They will bark in that garden’ (*ki(li)ngoduwâ-i=laa)
b. INTR + CV: φ-suffix, no φ-prefix
paveli eângâ ki- ngoduwâ-i =laa=kä
garden that IPFV-shout-3AUG=FUT=CV
‘They will bark in that garden’
324 A.2. Descriptive properties
(516) a. Baseline AV: φ-prefix, no φ-suffix (250814)
∅12AUG de-ki-vängä =kaa sii ngä täpilo enge
12AUG-IPFV-eat.AV=FUT fish in bowl this
‘We will eat fish in this bowl’ (*(de)kivängäde=ngaa)
b. AV + CV: φ-suffix, no φ-prefix
täpilo enge ki- vängä-de =ngaa=kä sii
bowl this IPFV-eat.AV-12AUG=FUT=CV fish
‘We will eat fish in this bowl’
Voice concord (see chapter 3, §3.3) is also found in CV. If an adverbial modifier is added
to a base UV verb in a CV construction, predictably, the voice concord morphology is
obligatory (517a). If a modifier is added do an AV verb or an intransitive verb, suprisingly,
the voice concord morphology is now optionally possible (517b,c), whereas it would be
impossible without CV. This phenomenon has not been described in the literature, and
constitutes thus a novel finding.
(517) a. UV + CV: voice concord obligatory (230727)
täpilo enge ki-ngä-lâoo*(-nyii)-no=ngä sii
bowl this IPFV-eat.UV-always-UV-1MIN=CV fish
‘In this bowl I always eat fish’ (with or without CV, -nyii is obligatory)
b. AV + CV: voice concord optional
täpilo enge ki-vängä-lâoo(-nyii)-no=ngä sii
bowl this IPFV-eat.AV-always-UV-1MIN=CV fish
‘In this bowl I always eat fish’ (without CV, -nyii is impossible)
c. Intransitive + CV: voice concord optional (230720)
paveli eângâ ki-ngoduwâ-lâoo(-nyii) kuli=kä
garden that IPFV-shout-always-UV dog=CV
‘The dog is always barking in that garden’ (without CV, -nyii is impossible)
The properties illustrated thus far are summarized in (518). Informally speaking, adding
CV to intransitive or AV verbs makes them “more UV-like”: prefixal agreement disappears,
suffixal pronouns in spec,TP arise, and voice concord becomes optionally possible.
(518) Comparison of baseline INTR, AV and UV to their version with CV added:
Prefixal Suffixal Voice
Word order
φ-agr.? subj. pron.? concord?
INTR: S φS-V =TAM (…PPs) yes no no
AV: S φS-V.AV =TAM O (…PPs) yes no no
UV: O V.UV S=TAM (…PPs) no yes yes
INTR+CV: X V S=TAM=CV (…PPs) no yes optional
AV+CV: X V.AV S=TAM=CV O (…PPs) no yes optional
UV+CV: X V.UV S=TAM=CV O (…PPs) no yes yes
Appendix A. Notes on Circumstantial Voice 325
All in all, Äiwoo CV has some similarities with CV in Dinka (van Urk 2015), but also
important differences. First, in Dinka the CV marker is a suffix on the raised verb (in C),
while in Äiwoo it is an enclitic in a lower region of the clause (either at the very low end
of the Infl/T domain, or at the very high end of the vP domain). Second, in Dinka the CV
suffix can only attach to verb stems with UV morphology; in Äiwoo, CV can go on top of
verb stems with either voice morphology.
ARGUMENT STRUCTURE In all the examples shown so far, the argument added by CV
and promoted to the preverbal position is a locative argument; another such example
is (519a). Another common type of argument introduced by CV is an instrument (519b).
Moreover, other types of “oblique” arguments (in a relatively loose definition) are possible,
for example a stimulus for certain psychological verbs (519c), the object of a warning/what
someone is being warned about (519d), an inanimate cause (519e), a time (519f). (In the
last two sentences, the CV-added argument is relativized or clefted; see chapter 4, §4.B.2
about clefting.)
(519) Possible types of arguments that can become VIAs in CV clauses:
a. Locative: (230907)
tebol enge i-tâbuwoli gino Anna=kä sii
table this ASP-cut.UV son Anna=CV fish
‘Anna’s son cut (the) fish on this table’
b. Instrument: (240215)
nuwoli i-lâbolâ-no=ngä nenu
knife ASP-open.UV-1MIN=CV coconut
‘I opened the coconut with the knife’
c. Psychological stimulus: (250206)
muli i-bou Anna=kä
sea.snake ASP-afraid Anna=CV
‘Anna is/was afraid of the sea snake’
d. Object of warning: (240222)
pesime eângâ i-wââ-mä John=kä (ngâgu)
people that ASP-warn-DIR1 John=CV to.1MIN
‘John warned me about those people’
e. Reason/inanimate cause: (250206)
[∅3MIN [RC i-nubo Sam=kä]]=nâ covid
ASP-die Sam=CV=DIST covid
‘[What Sam died of ] is covid’
f. Time: (250220)
ilâ i-wâte=kâ ngä [me=[RC ku-mo-no=ngä Honiara]]
DIST ASP-happen=DIST in REL:time= IPFV-stay-1MIN=CV Honiara
‘That happened at the time when I was living in Honiara’
326 A.2. Descriptive properties
One could observe that the CV marker =Cä is morphologically somewhat similar to the
locative preposition ngä. However, the correlation is not as water-tight. Äiwoo has chiefly
two prepositions: ngä is used for locative semantics (in, on, under, over, above, from, etc.),
and go for ‘for, to, with, etc.’. Both of these can introduce arguments that can potentially
become/correspond to CV VIAs. Above, we saw a locative PP (with the preposition ngä)
correspond to a locative CV VIA. However, an instrument argument (519b) and the stimu-
lus of ‘be afraid’ (519c) are introduced by the preposition go. Compare now the intransitive
+ PP sentences in (520) with their CV counterparts in (519b,c). Regardless of whether the
corresponding preposition for the fronted argument is ngä or go, the CV marker is always
=Cä.
(520) a. nenu i-lâbolâ-no go nuwoli (240215)
coconut ASP-open.UV-1MIN PREP knife
‘I opened the coconut with a/the knife’
b. Anna i-bou go muli (250206)
Anna ASP-afraid PREP sea.snake
‘Anna is afraid of the sea snake’
Finally, a couple clause-embedding attitude predicates always obligatorily take CV. This is
the case for kä ‘say, think, want’ (521a,b) and ko ‘say, want’ (521c). Here, it could perhaps
be said that the argument added/introduced by CV is the content of the attitude predicated
of the attitude holder.
(521) Two attitute predicates with obligatory CV marking: kä and ko
a. kä Mary=kä [CP site-∅n nâ-wâte] (230720)
KÄ Mary=CV sister-3MIN IRR-dance
‘Mary wants her sister to dance’
b. kä Mary=kä [CP site-∅n ku-wâte] (230720)
KÄ Mary=CV sister-3MIN IPFV-dance
‘Mary thinks/thought/said that her sister is/was dancing’
c. ko-kä polis=kä [CP nä-li-eeu-kä go pesingedâ] (231109)
want-DIR3 polis=CV IRR-3AUG-speak-DIR3 to girls
‘The police wants to talk to the girls’
Interestingly, certain types of arguments cannot become the VIA of a CV clause; this is
also a novel finding. This is true for typical “dative” arguments (recipients, benefactives),
e.g. with verbs like ‘give’ or ‘introduce’ (522a)–(522b). Moreover, the destination argument
of ‘bring’ also can’t become a CV VIA (523a). Compare this to an instrument argument of
the same verb, for which this is possible.
(522) a. * Jen ki-lää-kä mekivaavee=kaa=kä buk   (250813)
Jane IPFV-give.AV-DIR3 teacher=FUT-CV book
Intended: ‘The teacher gave Jane a book’
Appendix A. Notes on Circumstantial Voice 327
b. *Mary i-wagu-kâ-no=ngä Sam ngä party (240721)
Mary ASP-tell.UV-DIR3-1MIN=CV Sam at party
Intended: ‘I introduced Sam to Mary at the party’
(523) ‘Bring’ + CV: a destination cannot be a CV VIA, but an instrument can (231130)
a. * nyivängäna eângâ ki-tou-wâ-no=ngaa=kä dekilingä
party that IPFV-bring.AV-DIR2-1MIN=FUT=CV food
Intended: ‘I will bring food to that party’
b. nyibä mi=olo enge i-tou-kâ-no=ngä dekilingä
basket REL=big this ASP-bring.AV-DIR3-1MIN=CV food
‘I brought food in/with this big basket’
A.2.2 Ā-EXTRACTION PROPERTIES
In broad strokes, clauses with CV behave for the purposes of Ā-extraction as one would
expect from an Austronesian language. From a CV clause, only the added oblique argu-
ment (“X”) can be extracted (524a), and neither the subject nor the object can (524b,c).
This is true regardless of the voice morphology of the base verb to which CV is aded.
(524) Only theadded argument (X) can be extracted from CV:a. 3 X [CV: (231214)
i-ku-mo-kä vaakä ngâ [nuwopaa nyibei (mi=)[RC
 1MIN-IPFV-live-DIR3 close to house.of place.sick REL=
ku-{tou/tu }-mä pesingedâ=kaa=kä mikilibei]]
IPFV-bring.{AV/UV}-DIR1 women=FUT=CV patients
‘Ilive close to the hospital where the women will bring the patients’b. 7 S [CV: (240208)
* nâ-wâkâluwâ-kä-de [pesingedâ (mi=)[RC nuwopaa nyibei
IRR-thank.UV-DIR3-12AUG women REL= house.of place.sick
ku-{tou/tu }-kä-i =laa=kä mikilibei]]
IPFV-bring.{AV/UV}-DIR3-3AUG =FUT=CV patients
Intended: ‘We must thank the women who will bring the patients to the
hospital’
c. 7 O[CV: (231214)
* [mikilibei (mi=)[RC nuwopaa nyibei ku-{tou/tu }-mä
patients REL= house.of place.sick IPFV-bring.{AV/UV}-DIR1
pesingedâ=kaa=kä ]]=nâ ki-li-päko-ute=kaa
women=FUT=CV =DIST IPFV-3AUG-good-again=FUT
Intended: ‘The patients that the women will bring to the hospital will get
better’
The same is also true for clauses consisting of an intransitive verb and CV, both for (pu-
tative) unergatives and unaccusatives: only the added argument can be extracted (525a)–
328 A.2. Descriptive properties
(526a), and not the subject (525b)–(526b). To extract that argument, an intransitive clause
without CV must be used instead (527).
(525) Extraction from unergative + CV:a. 3 X [CV (230720)
[paveli [RC ki-ngoduwâ kuli=kä]]=nâ kâlâ ngâ nyângâ
garden IPFV-shout dog=CV=DIST there in place.that
‘The garden where the dog was barking is over there’
b. 7 S[CV
* [kuli [RC paveli eângâ ki-ngoduwâ-∅n =nä]]=nâ no Mary
dog garden that IPFV-shout-3MIN =CV=DIST POSS Mary
Intended: ‘The dog that was barking in that garden is Mary’s’
(526) Extraction from unaccusative + CV:
a. 3 X[CV (230727)
[numobâ [RC i-obu-to Mak=kä]]=nâ i-woli-mana nyalo
hole ASP-fall-in Mark=CV=DIST ASP-down-very deep
‘The hole where Mark fell goes down very deep’b. 7 S [CV
* [gilaki [RC numobâ eângâ i-obu-to-∅n =nä]]=nâ
boy hole that ASP-fall-in-3MIN =CV=DIST
nyisi-∅n ba i-ta=gu
body-3MIN NEG ASP-hurt=NEG
Intended: ‘The boy who fell in that hole didn’t get hurt’ (lit. ‘[The boy
[who…]], his body was not hurt’)
(527) Extraction of subject of intransitive: no CVa. From an unergative: S [ INTR (230720)
[kuli [RC ki-ngoduwâ ngä paveli eângâ]]=kâ no Mary
dog IPFV-shout in garden that=DIST POSS Mary
‘The dog that was barkingin that garden is Mary’s’b. From an unaccusative: S [ INTR (230727)
[gilaki [RC i-obu-to ngâ numobâ eângâ]]=kâ
boy ASP-fall-in in hole that=DIST
nyisi-∅n ba i-ta=gu 
body-3MIN NEG ASP-hurt=NEG
‘The boy who fell in that hole didn’t get hurt’
Moreover, the presence of CV “overrides” the special property of UV-only verbs. If one
adds CV to a UV-only verb, from the resulting clause still only the added argument can
be extracted (528a), and neither of the other two can (528b,c). This starkly contrasts with
the behavior of UV-only verbs without CV, since in that case they allow for extraction of
either the subject or the object (cf. chapter 4).
Appendix A. Notes on Circumstantial Voice 329
(528) Extraction from UV-only + CV: only the X argument (230914)a. 3 X [CV(UV-only):
i-wä ngä [paveli [RC i-te-kä mekivaavee=kä Jane]]
1MIN-go to garden ASP-see.UV-DIR3 teacher=CV Jane
‘Iwent to the garden where the teacher saw Jane’b. 7 S [CV(UV-only):
* me-wä-le mo [mekivaavee [RC paveli eângâ i-te-kä-∅n
1AUG-go-UA with teacher garden that ASP-see.UV-DIR3-3MIN
=nä Jen]]
=CV Jane
Intended: ‘I went with the teacher who saw Jane in that garden’c. 7 O [CV(UV-only):
* i-eeu-kä go [singedâ [RC paveli eângâ i-te-kä
ASP-speak-DIR3 to girl garden that ASP-see.UV-DIR3
mekivaavee=kä ]]
teacher=CV
Intended: ‘I talked to the girl that the teacher saw in that garden’
A similarity to the situation with UV-only verbs can be seen in the contexts of those
“CV-only” verbs mentioned above, e.g. the attitude predicate kä ‘say, think, want’. Per-
haps because this verb can never occur without CV, extracting its subject is exceptionally
grammatical (529). (Note that this is not a case of long-distance relativization: the ex-
tracted argument is the subject of the attitude predicate, not an argument of the more
deeply embedded CP.)
(529) Extracting the subject of kä ‘say, think, want’ is grammatical despite CV:
i-eeu-kä go … (241219)
1MIN-speak-DIR3 to …
‘I talked to …’
a. [pesime mi=[RC kä-i =lä [CP nuwopa tomu nä-ve-i=laa]]
people REL= KÄ-3AUG =CV house your IRR-buy.UV-3AUG=FUT
‘… the people who want to buy your house’
b. [pesime mi=[RC kä-i =lä [CP nuwopa tomu ki-ve-i=laa]]
people REL= KÄ-3AUG =CV house your IPFV-buy.UV-3AUG=FUT
‘… the people who think/said they’ll buy your house’
A.2.3 WHY “CIRCUMSTANTIAL VOICE” IS NOT A VOICE
Despite being evidently not in paradigmatic alternation with AV and UV, the construction
involving the fronting of an oblique DP and the marker =Cäwas analyzed in Næss (2015b)
as a voice (hence the label “Circumstantial Voice”), and not as an applicative. Here, I want
330 A.2. Descriptive properties
to argue that calling this construction a “voice” is a misnomer, though purely for conve-
nience of exposure I will retain the label “CV”. Themain two arguments in that paper were
(Næss 2015b: 295): (i) “the addition of =Cä neutralizes the difference between clauses with
[AV] and [UV]”3; (ii) “the addition of =Cä to a clause with an [AV] verb changes person
marking from prefixing to suffixing”.
As for argument (ii), I argue that this is a moot point, as it is based on a mis-analysis of
the φ-suffixes found on UV (and CV) verbs. On one hand, it is true that in clauses where
CV is added to either an AV or intransitive verb, the prefixal φ-agreement usually found
on those verbs disappears; in (A.3) I suggest that this may be due to the category of the
fronted pivot in CV clauses. Its “reappearance” as a suffix, however, is a red herring. As
argued for in depth in chapter 5, these suffixes are not actually agreement markers, unlike
the φ-prefixes, but rather phonologically reduced pronouns in spec,TP. Hence, their pres-
ence in CV clauses is purely a function of the position of the subject. In CV clauses, like
in UV ones (UVplain), the subject is in spec,TP (since spec,CP is plugged by the fronted
oblique argument). Therefore, when the subject is a pronoun in this position, the mor-
phology forces it to appear as a suffixal marker, but this does not mean that anything in
the agreement system has changed.
As for point (i) (the neutralization of the difference between AV and UV clauses), on
the other hand, I argue that this is empirically wrong, on the basis of novel data. In Næss
(2015b) it was observed that CV could be added on top of both AV and UV verb stems,
and in Roversi (2019) it was confirmed that this was true even for the same verb (e.g.,
‘weave.AV + CV’ and ‘weave.UV + CV’ were both attested in the corpus). However, it was
simply not known what the difference between these options was, if any. I have found
at least one such difference. Specifically, the interpretation of direct objects in AV+CV
vs. UV+CV clauses is different in the same way as between AV vs. UV clauses. Remember
from chapter 2, §2.5.3 that bare noun objects are interpreted differently in the two voices:
in AV, they are non-specific/narrow-scope indefinites (530), while in UV, they are specific/
wide-scope indefinites (531).
(530) Non-specific/narrow-scope indefinite object: 3 AV, 7 UV (250325)
a. milinurse ku-lu-potaa-ute dokta, mo ba
nurses IPFV-3AUG-search.AV-again.AV doctor but NEG
i-te-kä-i=dä=gu
ASP-see.UV-DIR3-3AUG=some=NEG
‘The nurses are looking for a doctor again, but they haven’t found one yet’
b. # dokta ku-potaa-usi milinurse, mo ba
doctor IPFV-search.UV-again.UV nurses but NEG
i-te-kä-i=dä=gu
ASP-see.UV-DIR3-3AUG=some=NEG
Intended: ‘The nurses are looking for a doctor again, but they haven’t found
one yet’
3 In that paper, AV and UV verb forms were analyzed as lexical doublets, labelled “A-verbs” and “O-verbs”.
Appendix A. Notes on Circumstantial Voice 331
(531) Specific object: 3 UV, 7 UV (250325)
a. dokta ku-potaa-usi milinurse, mo ∅3MIN ba
doctor IPFV-search.UV-again.UV nurses but NEG
ku-mo=gu ngâ nuwopaa nyibei elenge
IPFV-stay=NEG in house.of place.sick today
‘The nurses are looking for a doctori again, but hei is not at the hospital
today’
b. # milinurse ku-lu-potaa-ute dokta, mo ∅3MIN ba
nurses IPFV-3AUG-search.AV-again.AV doctor but NEG
ku-mo=gu ngâ nuwopaa nyibei elenge
IPFV-stay=NEG in house.of place.sick today
Intended: ‘The nurses are looking for a doctori again, but hei is not at the
hospital today’
Strikingly, this interpretational difference persists when adding CV on top of respectively
AV and UV verbs. When the intended reading of the object is that of a specific/wide-
scope indefinite, only UV+CV is possible (532a). Conversely, when the intended reading
is non-specific/narrow-scope, only AV+CV is possible (532b)4.
(532) a. Specific/wide-scope indefinite object: 3 UV+CV, 7 AV+CV (250220)
dee nye=[RC kä Sam=kä [nâ-{7tou/3tu}-mä-∅n=nä nurse
this REL:OBL= want Sam=CV IRR-bring.{7AV/3UV}-DIR1-3MIN=CV nurse
ngâ nuwopaa nyibei]]=kâ, mo ba ku-pole=gu bugulo
to house.of place.sick=DIST but NEG IPFV-work=NEG yesterday
‘This is (the reason) why Sam wanted to bring a nursei to the hospital, but shei
wasn’t working yesterday’
b. Non-specific/narrow-scope indefinite object: 3 AV+CV, 7 UV+CV
dee nye=[ nRC kä Sam=kä [nâ-{3tou/7tu}-mä-∅ =nä nurse
this REL:OBL= want Sam=CV IRR-bring.{3AV/7UV}-DIR1-3MIN=CV nurse
ngâ nuwopaa nyibei]]=kâ, mo ba i-te-kä-∅n=dä=gu
to house.of place.sick=DIST but NEG ASP-see.UV-DIR3-∅n=some=NEG
‘This is (the reason) why Sam wanted to bring a nurse to the hospital, but he
couldn’t find any’
As of now, I don’t have a theory of the contrast in the interpretation of bare noun objects
between plain AV and UV clauses (530)–(531). Accordingly, I also don’t have a theory of
this contrast in the more complex AV+CV vs. UV+CV contexts. However, the fact that a
4 These datapoints may present a couple confounds. First, they involve relativization of an oblique argument
(“the reason [RC why …]”) to force CV, but this should not be too problematic in how the direct object is
interpreted. More problematically, they involve long-distance relativization across the CV-only intensional
attitude predicate kä (here meaning ‘want’). This means that the direct object in (532a) actually scopes out
of the whole intensional predicate, rather than just scoping high in its own clause. Controlling for this will
be important in future research.
332 A.3. Towards an analysis of Circumstantial Voice
contrast exists in the first place proves Næss’s (2015) argument wrong: adding CV does
not neutralize the difference between AV and UV clauses.
A.3 TOWARDS AN ANALYSIS OF CIRCUMSTANTIAL VOICE
In this section I offer some promissory notes for what an analysis of Äiwoo CVmight look
like within the proposal in this dissertation. Some parts of the properties of CV clauses
follow quite elegantly from our model of the Äiwoo clause. However, we will also see that
some predictions our model makes are crucially wrong; I will leave these as open issues
for future research.
A.3.1 CIRCUMSTANTIAL VOICE AS PREPOSITION STRANDING
The kind of approach I aim to explore here is one where CV is essentially a form of prepo-
sition stranding. Under this idea, “Circumstantial Voice” is what happens when a speaker
decides to put the Ā-feature [PIV] not on a core argument of the clause, but on a nominal
that is embedded under a PP – that is, an adjunct in the VP domain. Because this nominal
is embedded under additional structure, the probes on v, T, and C will interact with it
somewhat differently, only being able to access parts of its features but not others. Below,
I will present a derivation bottom-up.
The structure of the vP domain in a CV clause is represented in (533). I will assume
that prepositions in Äiwoo assign a special case to their complement, here just labelled
“oblique” for simplicity; this will be morphologically null. On the technical implementa-
tion side, I will model this as such: a nominal that is the complement of a preposition in
Äiwoo carries an extra KP layer on top of the DP (this layer may be either selected for by
P, or countercyclically inserted, etc). Crucially, the K head is a phase head (as shown with
the arc), “shielding” the inner DP from interactions from higher probes.
(533) The vP domain in a CV clause:
vP
Subject
[ v ] VP
I: [Ā/π]
S: [D]
PP VP
P KP V Object
K DP
[PIV, OBL]
…
Appendix A. Notes on Circumstantial Voice 333
Let us now consider how, given this structure, the probe on v would interact with the
elements in its c-command domain5. The nominal embedded under the PP is closer to
v than the object (see e.g. Branan & Erlewine to appear for a discussion of search algo-
rithms), so v looks at it first. The probe finds an Ā-feature ([PIV]), so it interacts with it;
this KP is flagged with [i |v]. The entire set of features from the interacted-with KP that
are visible to v are copied onto the probe. However, importantly, the set that gets copied
does not contain any φ-features – because these are on the DP, crucially under the KP-
phase. Moreover, for the same reason, v doesn’t “see” the [D] feature, and it is therefore
not satisfied. After this first round of agreement, we have the state of affairs in (534).
(534) Stage 1: v interacts with the [PIV]-bearing oblique KP under the PP
[ v ⟨{P]IV, K}⟩ VP
I: [Ā/π]
S: [D]
PP VP
P KP[i|v] V Object
K DP
[PIV, OBL]
…
Because the probe on v has not been satisfied, it will now continue its search. The rest
continues as in any other (mono-)transitive clause in Äiwooo. If the object is a DP with
no Ā-features (535a), v will not interact with it, but it will be satisfied by it. If, instead, the
object is a DP with [PIV] (535b), v will both interact with it and be satisfied by it.
5 Again, for now I limit the discussion to sentences with no pronominal arguments, for simplicity. See below
for discussion of those cases.
334 A.3. Towards an analysis of Circumstantial Voice
(535) Stage 2:
a. If the object has no [PIV]: b. If the object has [PIV]:
no interaction, yes satisfaction yes interaction, yes satisfaction
[ v ⟨{P]IV, K}⟩ VP [ v ⟨{P]IV, K}, {PIV, D, φ}⟩ VP
I: [Ā/π] I: [Ā/π]
S: [D] S: [D]
PP VP PP VP
P KP V DP P KP[i|v] [i|v]
V DP[i|v]
[PIV]
K DP K DP
[PIV, OBL] [PIV, OBL]
This difference in what features v has copied back in (535a) vs. (535b) will ultimately
result in the difference between “AV + CV” vs. “UV + CV”. In (535b), v has φ-features on
it, among the members of the different sets it has copied. Therefore, when later on the
verb undergoes Vocabulary Insertion, the UV stem will be inserted, because the context
of application of this rule is met. In contrast, (535a), v has no φ-features on it. Therefore,
here the AV stem will be inserted.
At the TP level, nothing particularly interesting goes on (536). At least for the cases
where neither of the internal arguments is a pronoun/φ+P, T will follow the algorithm
described in chapter 6 and just raise the subject.
(536) T raises the subject to spec,TP:
TP
Subject
[ T ] vP
I: [[PART↑/PL]
M
7
S: [AD/3PL] ]
I+S: [φ]M
tS
v VP
⟨{PIV, K}, ({PIV, D, φ})⟩
V v PP VP
P KP tV DP[(PIV, i|v)]
[PIV, i|v]
Appendix A. Notes on Circumstantial Voice 335
Once C is merged, its [i |v]-probe will initiate a search (537). The closest argument to C
carrying this feature is the KP embedded under the preposition; this is true regardless of
whether or not the object also has [i |v]. Therefore, C raises the KP to spec,CP, extracting it
from under the PP, and thus stranding the preposition in situ; this is what will ultimately
be spelled out as the CV marker =Cä. Like for v, because the φ-features on this nominal
are buried inside the KP phase, C cannot copy them.
(537) C raises the KP to spec,CP:
CP
KP
[PIV, i|v] [C+V+v] TP
I+S: [i|v]M
Subj
T vP
tS tv VP
PP VP
P tKP tV DP[(PIV, i|v)]
Now, the KP in spec,CP has a [PIV] feature, and therefore, it will intervene for Ā-extracting
any lower argument. This correctly derives the restricted extraction profile described
in §A.2.2. We can also confirm that the nominal is spec,CP is raised by non-local A-
movement, just like for mono-transitive AV and UV clauses (chapter 3, §3.7). This move-
ment step does not reconstruct for Condition C (538a), and it is not sensitive to WCO
(538b).
(538) The fronted oblique nominal in a CV clause… (250617)
a. … does not reconstruct for Condition C:
nuwoli nogo Pita i-{lâbonge/lâbu }- ∅n =nä tX sii
knife POSS Peter ASP-cut.{AV/UV}-3MIN=CV fish
‘Peteri cut (the) fish with hisi knife’, lit. ‘Hei cut (the) fish with Peteri’s knife’
336 A.3. Towards an analysis of Circumstantial Voice
b. … does not induce a WCO violation:
penyibe dâuwângâ i-bou-du gino -i =lä tX
man every ASP-afraid-all son-3AUG=CV
3Bound reading: ∀ man 𝑥 , 𝑥 ’s son is afraid of 𝑥
Lit. ‘Hisi son is afraid of every mani’
Let us now consider how the different parts of this structure get spelled out onto their
surface exponents, distinguishing between the case where the object was a DP without
[PIV] (539) or with [PIV] (540). In either case, the KP in spec,CP is a nominal, with no
preposition, and the K head itself is phonologically null; therefore, this will just be spelled
out as a plain nominal. Also in either case, C has not copied any φ-features from the
argument it raised, because those were rendered invisible by the KP layer. The two cases,
however, differ in terms of what features v has copied: if the object is an [PIV]-less DP,
v has no φ-features on itself (539), whereas if the object is a DP with [PIV], it does (540).
This will prompt a differential spell-out, because the Vocabulary Insertion rule for the
verb stem is sensitive to this difference: in the first case, we will insert the AV stem; in
the second case, the UV stem. This gives us the difference between “AV + CV” on one
hand, and “UV + CV” on the other. Moreover, if the subject in spec,TP is a pronoun, it will
have to undergo Local Dislocation onto the verb (see chapter 5). Crucially, this happens
regardless of whether the verb stem takes its AV or its UV. Because there are no φ-features
on C, this now suffixal pronoun does not get impoverished, so we will always have φ-
suffixes even if the base verb is AV or intransitive. Finally, the stranded preposition needs
to be handled by a special Vocabulary Insertion rule, stating that a preposition without
an overt complement must be realized as the enclitic marker =Cä rather than its normal
exponent (either ngä or go; see above).
Appendix A. Notes on Circumstantial Voice 337
(539) Spell-out: AV + CV
a. CP
KP
TP
C πP πP
T …
π φ(P)
C [FUT][12AUG]
Asp … VP∅
∅ [IPFV] √V v
EAT ⟨{PIV, K}⟩
PP VP
P tKP V DP
ki- vängä -de =ngaa =kä
b. φ(P(-impoverishment: does not apply
7 φ ⇔ ∅ / C⟨φ⟩
c. VI rule for the stem (voice allomorphy):
√EAT ⇔ ngä / v ⟨φ⟩
☞ √EAT ⇔ vängä (elsewhere)
d. VI rule for C:
C⟨∅⟩ ⇔ ∅
e. VI rule for φP:
φ[12AUG] ⇔ -de
f. VI rule for P:
P ⇔ =Cä / tXP
338 A.3. Towards an analysis of Circumstantial Voice
(540) Spell-out: UV + CV
a. CP
KP
TP
C πP πP
T …
π φ(P)
C [FUT][12AUG]
Asp … VP∅
∅ [IPFV] √V v
EAT ⟨{PIV, K},
{PIV, D, φ}⟩ PP VP
P tKP V DP[PIV, i|v]
ki- ngä -de =ngaa =kä
b. φP-impoverishment: does not apply
7 φ ⇔ ∅ / C⟨φ⟩
c. VI rule for the stem (voice allomorphy):
☞ √EAT ⇔ ngä / v ⟨φ⟩
√EAT ⇔ vängä (elsewhere)
d. VI rule for C:
C⟨∅⟩ ⇔ ∅
e. VI rule for φP:
φ[12AUG] ⇔ -de
f. VI rule for P:
P ⇔ =Cä / tXP
A.3.2 PROBLEMS AND INCORRECT PREDICTIONS
This analysis of CV unfortunately faces some challenges. A first one is the possibility of
optional voice concord morphology on verbal modifiers when CV is added to a base AV
or intransitive verb (517b,c). Given the analysis of voice concord I proposed in chapter 3,
this shouldn’t be possible here. Moreover, the optionality itself is mysterious.
Appendix A. Notes on Circumstantial Voice 339
A more fundamental problem concerns sentences with CV where the direct object or
the applied argument is a pronoun (specifically, a φ+P). As we know, pronominal objects in
Äiwoo tend to cause word order changes, and this is true in CV clauses as well. However,
the theory proposed here generates incorrect predictions. Let us see why.
Consider first the case where the direct object is a pronoun/φ+P, and both other argu-
ments are lexical DPs/φ0Ps. We expect our probe on v to agree first with the applied KP,
not be satisfied (because D is “invisible” under the KP phase boundary), and then proceed
to agree with the direct object pronoun (541). Now both internal arguments have [i |v].
(541) [vP DP v I: [Ā/π] [VP [PP P KP[PIV, i |v] ] [VP … πP[i |v]
[S: [D] ]
Then, the probe on T would raise the direct object pronoun across the other two argu-
ments, since it is the preferred type of goal (542). This is the same phenomenon that we
observe in UVgu clauses (see chapter 6).
(542) [TP πP[i |v] T I: [±PART↑]M [vP DP v [VP [PP P KP[PIV, i |v] ] [VP … tO
[S: [AD/3AUG] ]
However, when we merge C is where things go wrong. Because of the syntactic inversion
step triggered by T, the argument carrying [i |v] closest to C is now the direct object.
Therefore, C should raise it to spec,CP (543), and leave the applied argument in situ.
(543) Predicted (incorrectly): C fronts the object
7 [CP πP[i |v] C [ I+S: [i |v]M] [TP tO T [vP DP … [VP [PP P KP[PIV, i |v] ] …
But this is not what we observe. Instead, C raises the applied argument (544a). The pre-
dicted scenario is ungrammatical (544b). Moreover, even if we somehow found a way for
C to exceptionally raise the applied argument instead of the direct object in spec,TP, the
position of the CV marker =Cä is problematic. If it really was simply a stranded preposi-
tion left in situ, first of all in this case we should not expect it to be stranded at all in the
first place, since its complement never gets moved anywhere. Moreover, even if it did get
stranded somehow, we should expect it to be to the right of the low subject. Instead, we
see that it surfaces to its left.
(544) a. C fronts the applied argument across the object in spec,TP (250311)
nuwopa to ki-te-kä-gu-i=laa=kä Mary tX tO
house POSS.1MIN IPFV-see.UV-DIR3-OBJ-3AUG=FUT=CV Mary
‘Mary will see them in my house’
340 A.3. Towards an analysis of Circumstantial Voice
b. Predicted surface form corresponding to (543):
* (ijii) ki-te-kä-gu-i=laa Mary {=kä / ngâ } nuwopa
3AUG IPFV-see.UV-DIR3-OBJ-3AUG=FUT Mary{=CV PREP} house
to tO
POSS.1MIN
Intended: ‘Mary will see them in that house’
This problem cannot be amended by changing our assumptions about the underlying
argument structure of CV, for example, by positing that the PP adjunct is actually base-
generated higher than the subject. We can show that this cannot be the case, by examining
what happens when the applied argument itself is a pronoun (φ+P). First, remember that
we modelled the hierarchy effect in T, simplifying coarsely, to achieve the effect in (545).
In prose: T raises the closest φ+P, and will only raise a φ0P if no φ+P is available. This
description only makes sense if the probe on T sees the subject first and then the object,
as it is reasonable to assume. Therefore, rather unsurprisingly, we know that the subject
is base-generated above the object.
(545) T-raising pattern (subject vs. direct object):
a. φ+P > φ+P: move subject
b. φ+P > φ0P: move subject
c. φ0P > φ+P: move object!
d. φ0P > φ0P: move subject
We can now use this to confirm that the subject is also base-generated above the applied
argument in a CV clause, because the same exact pattern obtains. By manipulating the
φ0P vs. φ+P category of the subject and the applied argument, we can recreate the same
paradigm with combinations of subjects and applied arguments (546) as with subjects and
direct objects (545). The concrete examples are given in (547).
(546) T-raising pattern (subject vs. applied argument):
a. φ+P > φ+P: move subject
b. φ+P > φ0P: move subject
c. φ0P > φ+P: move applied arg.!
d. φ0P > φ0P: move subject
(547) a. (iumu) i-bou -mu =wä tS iu (250220)
2MIN ASP-afraid-2MIN=CV 1MIN
‘You are afraid of me’
b. Sam i-bou -no =ngä tS tX
Sam ASP-afraid-1MIN=CV
‘I am afraid of Sam’
Appendix A. Notes on Circumstantial Voice 341
??
c. (iude) i-bou -gu-de =ngä Sam [PP P tX ]
12AUG ASP-afraid-OBJ-12AUG=CV Sam
‘Sam is afraid of us’
d. muli i-bou Anna =kä tS tX
sea.snake ASP-afraid Anna=CV
‘Anna is afraid of the sea snake’
This pattern confirms that the probe on T must have the subject as its closest argument,
and the applied argument must be lower. Therefore, sentences like (547c) are already puz-
zling: why is the CV marker =Cä to the left of the subject, if it “belongs to” the applied
argument, which we have shown must be lower than the subject?6
Moreover, the assumption we made earlier about the KP layer constituting a phase
boundary, encapsulating the features under it from being seen from higher probes, now
becomes a problem. Earlier, it was important that a [D] feature on the applied argument
would not be visible to the probe on v, since we had to prevent the probe from halting
immediately and thus making sure it could at least be able to interact with the direct
object. However, following the same assumption, this would mean that the φ-features
of the applied argument should also not be visible to T (since they, too, are under the
KP-phase boundary). But in turn, this would make it impossible for T to run its attested
hierarchy effect, and we should thus not expect the pattern in (546)–(547).
6 Note that the position of the CV marker under this analysis should be to the right of the subject trace, not
to its left as represented in (547). However, at least in those cases where the subject raises to spec,TP, this
structural difference would result in the same surface linear string, so that is less concerning.

APPENDIX B
A “CASE-AGREEMENT” ANALYSIS
B.1 OBJECT AGREEMENT, CASE AGREEMENT, AND
SECOND-CYCLE EFFECTS
B.1.1 A COMPETING THEORY OF AUSTRONESIAN VOICE
The analysis of Äiwoo voice morphology I developed in chapter 3 gives an asymmetric
status to AV and UV. In this model, UV is the spell-out of object agreement by v (548a),
while AV is a morphosyntactic default, i.e., what gets spelled out when no object agree-
ment takes place. The Ā-feature [PIV] is assumed to have completely free distribution:
there could be zero, one, or more instances of it per sentence. If the object has [PIV], v
agrees with it (setting aside the cases where the object is pronominal); else, there is no
object agreement. In either case, the subject may ormay not have [PIV], but nothingwould
change, because the [D]-feature on the object already satisfies the probe anyway.
(548) An object agreement analysis (chapter 3):
a. UV: object agreement b. AV: no agreement/default
[ Subj ([PIV]) v [ … Obj[PIV] [ Subj ([PIV]) v ⟨φ ⟩ [ … ObjO
As mentioned in chapter 3, this kind of model is inspired by but ultimately contrasts
with those developed for Tagalog in Rackowski (2002), Rackowski & Richards (2005), and
Hsieh (2020, 2025), a.o. For the sake of exposition, I am going to present a sort of analytical
Frankenstein, combining certain aspects of these approaches to highlight their common-
alities and ignoring others to make the analysis as easily comparable as possible to the
one I develop for Äiwoo; see fn. 2 for details. However, it ought to be clear that the exact
model I describe here is not proposed as such (in its letter) in any of the works cited above.
344 B.1. Object agreement, case agreement, and second-cycle effects
Although these approaches have various differences, what they have in common is that
they are based on the idea of “case agreement” (though only Rackowski 2002, Rackowski &
Richards 2005 commit to this explicitly, and not Hsieh 2020, 2025). The two voices1 are on
more symmetrical standing: in UV, v agrees with the object (549a); in AV, v agrees with
the subject (549b)2.
(549) A case agreement analysis (Hsieh 2020, 2025):
a. UV: agreement with object b. AV: agreement with subject
[ Subj v [ACC] [ … Obj[PIV] [ Subj[PIV] v [NOM] [ … Obj
Even if we don’t commit to calling this “case agreement” specifically (as Hsieh 2020, 2025
does not do, for example), or we don’t assume that Ā-features play a role (as Rackowski
2002, Rackowski & Richards 2005 don’t do), an essential difference from my approach still
remains. On one hand, we have a split between object agreement vs. no agreement; on
the other hand, we have a split between object agreement vs. subject agreement (550).
Under this second approach, the difference in verbal morphology between AV and UV
must result from a difference in which DP the verb agrees with; calling this “case” seems
like a reasonable choice.
(550) Undergoer Voice: Actor Voice:
Object agreement analysis: v agrees with object v agrees with nothing
Case agreement analysis: v agrees with object v agrees with subject 
It is then reasonable to ask the question of whether we should follow this literature and
give Äiwoo a case-agreement-style analysis, to see whether that gains us any insight com-
pared to the object agreement one developed in chapter 3. Of course, the devil is in the
details. As much as a case-agreement analysis of Äiwoo seems plausible at first, once one
attempts to actually develop a fully-formalized one, a series of problems arise.
1 Tagalog actually has four voices. I limit the discussion here to AV/UV, excluding the two applicative voices.
Note that in the Tagalog literature, “UV” is traditionally labelled “Patient Voice” (PV); I re-label it here just
for comparability with Äiwoo, with no theoretical differences.
2 To do more justice to the works cited above, here is the breakdown of what parts of the idealized model in
(549) come from where. Rackowski (2002) and Rackowski & Richards (2005) propose that a process of object
shift raises either the subject or the object to an outer spec,vP. Then, v agrees in case with the argument in
its outermost specifier: if this is the subject, we get AVmorphology; if it’s the object, we get UVmorphology.
Ā-features do not feature in this analysis. For Hsieh (2020, 2025), on the other hand, exactly one (and only
one) DP per clause carries the Ā-feature [PIV]. Then, v (or rather, an analogous head in the θ-domain) probes
for [PIV], and agrees with whichever argument carries it (and raises it to its specifier, making it the highest
argument; this part is not too important here). Hsieh argues that in this process, the [PIV]-carrying argument
is assigned nominative (which is then spelled out as ang in Tagalog), but he does not commit to exactly how
the verb is spelled out with either AV or UV morphology in the two cases. Despite all these differences, I am
synthesizing the various approaches to hone in on the case-agreement part: across both, if v agrees with
the subject we get AV, and if v agrees with the object, we get UV.
Appendix B. A “case-agreement” analysis 345
B.1.2 “CASE AGREEMENT” WITHOUT CASE?
The first, but not unsurmountable, problem comes right at the start, with invoking the
concept of case (whether abstract or morphological) as the potential difference between
v agreeing with the object vs. the subject. It is important in my model that the probe that
ends up being spelled out as voice morphology sees the object first, and only later the
subject (as Äiwoo voice is almost exclusively controlled by the properties of the object,
and almost never by the properties of the subject). Therefore, this probe must be low,
between the object and the subject. This is a very early point in the derivation, and it’s
not clear how or whether case would already be assigned. For example: would we say that
v assigns accusative to the object, and then copy this case feature back from it if the object
has [PIV]? How does the subject get nominative case, and how exactly can this happen
before v agrees with the subject? If we commit to the notion of case, we would need to
give a concrete answer to these questions.
Fortunately, I think this particular problem can be sidestepped if we accept being non-
committal enough about exactly what “case” is. Boil down the difference between the two
models as in (550), we see that the crucial difference is what happens in AV. What we aim
to develop is, at the core, an analysis where in UV v agrees with the object, and in AV, v
agrees with the subject. I suggest that if we wish to remain agnostic about how exactly
case assignment works in Äiwoo, we can redraw this same distinction without invoking
the notion of case. If we give up “case” as the distinguishing feature between subjects and
objects, we need to find another way to distinguish between what is otherwise just sets of
features copied onto a probe, with no record of origin. To maintain a difference between
these scenarios, we can use the concept of “second-cycle effect” (Béjar & Rezac 2009, et
seq.). The idea is simple enough: depending on whether a probe has agreed with a goal
on the first attempt or on a second one(or third, etc.), the morphological spell-out will be
different.
Remember, however, that our case-agreement analysis does not only need to account
for the AV/UV contrast, but ideally for thewhole gamut of combinations of word order and
voice found in Äiwoo. Crucially, we need to make sure we have an explanation for those
cases where the verb shows UV morphology, but the argument fronted to spec,CP is the
subject (UVSVO clauses). Maintaining the idea that C raises the closest argument flagged
by v, the desiderata for our analysis are summarized in (551). (Note that I am excluding
intransitive verbs from the analysis, holding onto the assumption that their vINTR simply
has no probe, as I did in chapter 3. This makes the comparison between the two competing
analyses more fair.)
(551) a. UVplain, UVgu:
i. UV morphology: v has agreed with the object
ii. The object is in spec,CP: the object must have [i |v]
⟹ v successfully agrees with the object only
b. AV:
i. AV morphology: v has agreed with the subject
346 B.2. An attempt at an implementation
ii. The subject is in spec,CP: the subject must have [i |v]
⟹ v successfully agrees with the subject only
c. UVSVO:
i. UV morphology: v has agreed with the object (at least)
ii. The subject is in spec,CP: the subject must have [i |v]
⟹ v successfully agrees with both the object and the subject
We thus need to make sure that the syntax can make sure these three scenarios take
place in the correct environments, and that the morphological spell-out of the verb can
distinguish between them (552). (For the time being, I’m ignoring the distribution of [PIV],
which arguments are DPs or pronouns, and when the probe gets satisfied.)
(552) a. [ Subj v [ … Obj[i |v] ⇔ UV morphology, obj. in spec,CP
b. [ Subj[i |v] v [ … Obj[i |v] ⇔ UV morphology, subj. in spec,CP
c. [ Subj[i |v] v [ … Obj ⇔ AV morphology, subj. in spec,CP
B.2 AN ATTEMPT AT AN IMPLEMENTATION
Having seen what we would need to derive for a case-agreement analysis of Äiwoo, let us
now try to implement it. Although the idea of second cycle effect has been in the literature
for a while (Béjar & Rezac 2009), I am not aware of a mechanical implementation of it,
at least not in the interaction/satisfaction model of Agree (whose expressive power we
know we need for Äiwoo anyway, due to the complexity of the hierarchy effect found in
movement to spec,TP). To distinguish whether a set of features has been copied back at
the first or second (or 𝑛th) step of agree, I will make an amendment to the model of agree
described in Deal (2015b, 2024a, to appear) and Deal & Royer (2025), a.o. Concretely, I will
assume (553).
(553) “Unsuccessful interaction leaves a record”:
Let FH be the interaction set of a probe H. When H examines a potential goal G:
a. If G contains a feature [F] ∈ FH, copy all G’s features onto the probe, as a set
in an ordered list.
 Result of successful interaction: H⟨{F, …}, …⟩
b. If G does not contain any feature in FH, add (an instance of) the empty set
to an ordered list.
 Result of unsuccessful interaction: H⟨∅, …⟩
The first part (553a) is not novel: if a probe successfully interacts with a goal, the goal’s
features are copied onto the probe. This is all traditional stuff. The innovation is in the
Appendix B. A “case-agreement” analysis 347
second clause (553b): if a probe does not successfully interact with a goal, it doesn’t just do
nothing; instead, it will keep a record of the fact that it did not agree with this goal3. With
this, we could get the desired distinctions (at least if we can find a plausible solution to
the considerable complications described in fn. 3, which alone is a non-trivial challenge).
The set of possible agreement outcomes would be the following:
(554) a. v ⟨{φ, …}⟩: ⇔UV morphology
b. v ⟨{φ, …}, {φ, …}⟩: ⇔UV morphology
c. v ⟨∅, {φ, …}⟩: ⇔AV morphology
We could then make our VI rules sensitive to these distinctions. For example, we could
impose that for a given root, the AV stem be inserted in the context of a v head whose
ordered list starts with∅ (555a), and make the UV stem the default exponent (555b). (The
reverse choice of marked vs. default exponent could be possible too, of course.)
(555) VI rules for stem allomorphy:
a. √EAT ⇔ vängä / v ⟨∅, …⟩
b. √EAT ⇔ ngä (elsewhere)
A second change we would need to make to our model is a small amendment to the
probe specification on v. The one defended in chapters 3–6 is [INT: Ā or π; SAT: D],
and assumes that the distribution of [PIV] is syntactically completely free. Crossing the
possible category of the two arguments and the possible positions of [PIV], we correctly
predict the following breakdown (556)4. (The difference inword order between the various
sub-cases of UVplain, UVgu, and UVSVO arise from the interactions with the probes on T
and C; see chapter 6.)
3 The implementation can almost certainly not work exactly as written in (553). The reason is that we have no
formally defined notion of “potential goal G”. In the way the Agree algorithm is formulated in Deal (2024a,
to appear), a probe checks every single node in its c-command domain (following some specified search
procedure, e.g. Branan & Erlewine to appear) for whether it contains any of its interaction features, until
it is satisfied (or until it exhausts its entire c-command domain). This means that, for example, a probe on
T will not just examine the subject and the object: it will examine AspP, Asp0, vP, the subject, v′, v0, VP,
V, and the object (and any other intermediate functional projections that may exist). Of course, this would
result in the addition of a large number of ∅s to the ordered list. If we want our morphology to be able
to distinguish between when v has successfully agreed only with the object vs. only with the subject, we
would need to make the morphological component powerful enough to be able to count the number of ∅s
in the list. Obviously, this is not ideal in the least. The only solution I can envision would be to add a notion
of “potential goal” to the theory of Agree (so that only certain “relevant” ∅s would get added, and not all
of them), which however would come with its own share of complications and ad-hocness.
4 For conciseness, I make the following simplifications. By “πP” here I mean specifically a φ+P (a pronoun
with non-3MIN features), and not a 3MIN pronoun. I represent any set of features copied by the probe simply
as “φ”, standing as an abbreviation for the whole set of features copied from the goal (e.g. {φ, N, D, PL, …}).
348 B.2. An attempt at an implementation
(556) Subject: Object: Outcome: [i |v]? Voice: spec,TP: spec,CP: Clause type:
a. DP v DP v ⟨ ⟩ — AV S S AV
b. DP[PIV] v DP v ⟨ ⟩ — AV S S AV
c. DP v DP[PIV] v ⟨φ⟩ O UV S O UVplain
d. DP[PIV] v DP[PIV] v ⟨φ⟩ O UV S O UVplain
e. πP v DP v ⟨ ⟩ — AV S S AV
f. πP[PIV] v DP v ⟨ ⟩ — AV S S AV
g. πP v DP[PIV] v ⟨φ⟩ O UV S O UVplain
h. πP[PIV] v DP[PIV] v ⟨φ⟩ O UV S O UVplain
i. DP v πP v ⟨φ⟩ O UV O O UVgu
j. DP[PIV] v πP v ⟨φ, φ⟩ O, S UV O O UVgu
k. DP v πP[PIV] v ⟨φ⟩ O UV O O UVgu
l. DP[PIV] v πP[PIV] v ⟨φ, φ⟩ O, S UV O O UVgu
m. πP v πP v ⟨φ, φ⟩ O, S UV S S UVSVO
n. πP[PIV] v πP v ⟨φ, φ⟩ O, S UV S S UVSVO
o. πP v πP[PIV] v ⟨φ, φ⟩ O, S UV S S UVSVO
p. πP[PIV] v πP[PIV] v ⟨φ, φ⟩ O, S UV S S UVSVO
If our aim is to develop a system where AV specifically features agreement with the sub-
ject, the best model I was able to devise that still captures the word order alternation and
extraction facts is the following. First, we need the revised agreement algorithm devised
above. Second, we need to change the specification of the probe on v to [INT: Ā or π; SAT:
D and Ā] (with an added conjunction in the satisfaction condition); why will become clear
by looking at the table. Finally, we must follow Hsieh (2020) in restricting the distribution
of [PIV] by stipulation; we must assume that [PIV] always occurs on exactly one argument
Appendix B. A “case-agreement” analysis 349
per sentence (not zero, and not more than one)5. With all of this in place, we would get
the following breakdown (557).
(557) Subject: Object: Outcome: [i |v]? Voice: spec,TP: spec,CP: Clause type:
a. DP[PIV] v DP v ⟨∅, φ⟩ S AV S S AV
b. DP v DP[PIV] v ⟨φ⟩ O UV S O UVplain
c. πP[PIV] v DP v ⟨∅, φ⟩ S AV S S AV
d. πP v DP[PIV] v ⟨φ⟩ O UV S O UVplain
e. DP[PIV] v πP v ⟨φ, φ⟩ O, S UV O O UVgu
f. DP v πP[PIV] v ⟨φ, ∅⟩ O UV O O UVgu
g. πP[PIV] v πP v ⟨φ, φ⟩ O, S UV S S UVSVO
h. πP v πP[PIV] v ⟨φ, φ⟩ O, S UV S S UVSVO
All in all, this kind of “case-agreement” analysis seems to me to be more complex than the
“object agreement” one defended in this dissertation, with no clear gain in either empirical
coverage, theoretical elegance or parsimony, or insightfulness. This kind of model needs
more complicated probe specifications, an unmotivated stipulation on the distribution of
Ā-features, and a revision of the agreement mechanism that faces considerable challenges
once one tries to work out the details (see fn. 3).
5 We could abandon this assumption if we assumed that AV morphology can arise from either v agreeing
with the subject, or from v not agreeing with anything at all. But at that point, we would be essentially
reintroducing the “object agreement vs. default” idea into the picture.

BIBLIOGRAPHY
Abels, Klaus. 2012. The Italian Left Periphery: A View from Locality. Linguistic Inquiry
43(2). 229–254. https://doi.org/10.1162/LING_a_00084.
Abramovitz, Rafael. 2021. Deconstructing Inverse Case Attraction. Ms., MIT. https://ling.
auf.net/lingbuzz/006146.
Ackema, Pete & Ad Neeleman. 2003. Context-Sensitive Spell-Out. Natural Language &
Linguistic Theory 21(4). 681–735. https://doi.org/10.1023/A:1025502221221.
Adger, David. 2000. Feature Checking Under Adjacency And Vso Clause Structure. In
Robert D. Borsley & Jaklin Kornfilt (eds.), The nature and function of syntactic cate-
gories, 79–100. Leiden: Brill.
Aissen, Judith. 2003. Differential ObjectMarking: Iconicity vs. Economy.Natural Language
& Linguistic Theory 21(3). 435–483. https://doi.org/10.1023/A:1024109008573.
Akkuş, Faruk, David Embick & Mohammed Salih. 2025. Case and the syntax of argument
indexation. Oxford: Oxford University Press.
Aldridge, Edith. 2004. Ergativity and word order in Austronesian languages. Ithaca, NY:
Cornell University dissertation.
Aldridge, Edith. 2008. Phase-based account of extraction in Indonesian. Lingua 118(10).
1440–1469. https://doi.org/10.1016/j.lingua.2007.08.006.
Aldridge, Edith. 2017. 𝜑-feature competition: A unified approach to the Austronesian ex-
traction restriction. Proceedings of CLS 52. 313–326.
Alfarano, Valentina. 2021. A grammar of Nalögo, an Oceanic language of Santa Cruz Island.
INALCO, Paris dissertation.
Alfarano, Valentina. 2023. Middle, Reflexive, and Reciprocal Constructions in Nalögo: A
Typological and Diachronic Account. Oceanic Linguistics. https://muse.jhu.edu/pub/
5/article/915664.
Alfarano, Valentina & Brenda H. Boerger. 2022. The prefix (v)ö- in Nalögo and Natügu
and its origin from POc *paRi-. Language & Linguistics in Melanesia 40. 31–59.
Alfarano, Valentina & Brenda H. Boerger. 2025. Verb Classes and Valency in Santa Cruz
and Proto-Oceanic: A Historical-Comparative Analysis. 64(1). 46–85.
Anagnostopoulou, Elena. 2003. The syntax of ditransitives: Evidence from clitics. Berlin:
Mouton de Gruyter.
352 Bibliography
Anagnostopoulou, Elena. 2006. Clitic Doubling. In Martin Everaert & Henk C. van Riems-
dijk (eds.), The Wiley Blackwell Companion to Syntax, 519–581. John Wiley & Sons,
Ltd. https://doi.org/10.1002/9781118358733.wbsyncom071.
Anagnostopoulou, Elena. 2016. Clitic Doubling and Object Agreement. In Susann Fis-
cher & Mario Navarro (eds.), Proceedings of the VII Nereus International Workshop, 11–
42. Konstanz: Fachbereich Sprachwissenschaft Universität Konstanz.
Anagnostopoulou, Elena. 2017. Clitic Doubling. In Martin Everaert & Henk C. van Riems-
dijk (eds.), The Wiley Blackwell Companion to Syntax, 2nd edn., 1–56. John Wiley &
Sons, Ltd. https://doi.org/10.1002/9781118358733.wbsyncom071.
Anderson, Stephen R. 1982. Where’s morphology? Linguistic Inquiry 13(4). 571–612.
Aravind, Athulya. 2017. Ā-Interactions and Feature-Geometries. In Claire Halpert, Hadas
Kotek & Coppe van Urk (eds.), A Pesky Set: Papers for David Pesetsky, 333–343. Cam-
bridge, MA: MIT Press.
Aravind, Athulya. 2018. Licensing long-distance wh-in-situ in Malayalam. Natural Lan-
guage & Linguistic Theory 36(1). 1–43. https://doi.org/10.1007/s11049-017-9371-2.
Arka, I Wayan. 2002. Voice systems in the Austronesian languages of Nusantara: typology,
symmetricality and Undergoer orientation. Linguistik Indonesia 21(1). 113–139.
Arka, I Wayan & Mary Dalrymple. 2021. Number in Balinese. In Patricia Cabredo
Hofherr & Jenny Doetjes (eds.), The Oxford Handbook of Grammatical Number, 682–
700. Oxford: Oxford University Press.
Arkadiev, Peter & Ivano Caponigro. 2021. Conveying content questions without wh-
words: evidence from Abaza. Proceedings of Sinn und Bedeutung 25. 73–94. https : / /
doi.org/10.18148/sub/2021.v25i0.925.
Arregi, Karlos & Emily Anne Hanink. 2022. Reverse weak PCC in Washo. Proceedings of
NELS 52(1). 43–56.
Arregi, Karlos & Andrew Nevins. 2012. Morphotactics: Basque auxiliaries and the structure
of spellout. Dordrecht: Springer. https://doi.org/10.1007/978-94-007-3889-8.
Assmann, Anke, Doreen Georgi, Fabian Heck, Gereon Müller & Philipp Weisser. 2015.
Ergatives Move Too Early: On an Instance of Opacity in Syntax. Syntax 18(4). 343–
387. https://doi.org/10.1111/synt.12034.
Baier, Nicholas. 2018. Anti-agreement. UC Berkeley dissertation.
Bailyn, John Frederick. 2004. Generalized Inversion.Natural Language & Linguistic Theory
22(1). 1–50. https://doi.org/10.1023/B:NALA.0000005556.40898.a5.
Bailyn, John Frederick. 2020. The Scrambling Paradox. Linguistic Inquiry 51(4). 635–669.
https://doi.org/10.1162/ling_a_00361.
Baker, Mark C. 1985. The mirror principle and morphosyntactic explanation. Linguistic
Inquiry 16(3). 373–416.
Baker, Mark C. 1988. Incorporation: a theory of grammatical function changing. Chicago:
University of Chicago Press.
Baker, Mark C. 2003. Agreement, dislocation, and partial configurationality. In Andrew
Carnie, Heidi Harley &MaryAnnWillie (eds.), Formal Approaches to Function in Gram-
Bibliography 353
mar: In honor of Eloise Jelinek, 107–132. Amsterdam: John Benjamins. https://doi.org/
10.1075/la.62.09bak.
Baker, Mark C. 2010. On parameters of agreement in Austronesian languages. In Raphael
Mercado, Eric Potsdam & Lisa Travis (eds.), Austronesian and theoretical linguistics,
345–374. Amsterdam: John Benjamins.
Baker, Mark C. & Ruth Kramer. 2018. Doubled clitics are pronouns: Amharic objects (and
beyond). Natural Language & Linguistic Theory 36(4). 1035–1088. https://doi.org/10.
1007/s11049-018-9401-8.
Bale, Alan, Michaël Gagnon & Hrayr Khanjian. 2011. Cross-linguistic representations of
numerals and number marking. Proceedings of SALT 20. 582–598.
Bárány, András. 2017. Person, case, and agreement: The morphosyntax of inverse agreement
and global case splits. Oxford: Oxford University Press.
Bárány, András & Michelle Sheehan. 2024. Challenges for dependent case. In Christina
Sevdali, Dionysios Mertyris & Elena Anagnostopoulou (eds.), The Place of Case in
Grammar, 0. Oxford University Press. https://doi .org/10.1093/oso/9780198865926.
003.0005.
Bejar, Susana & Diane Massam. 1999. Multiple Case Checking. Syntax 2(2). 65–79. https:
//doi.org/10.1111/1467-9612.00015.
Béjar, Susana. 2003. Phi-syntax: A theory of agreement. University of Toronto dissertation.
Béjar, Susana & Milan Rezac. 2003. Person licensing and the derivation of PCC effects.
In Ana Teresa Perez-Leroux & Yves Roberge (eds.), Romance linguistics: Theory and
acquisition, 49–62. Amsterdam: John Benjamins.
Béjar, Susana & Milan Rezac. 2009. Cyclic Agree. Linguistic Inquiry 40(1). 35–73. https :
//doi.org/10.1162/ling.2009.40.1.35.
Belletti, Adriana. 1999. Italian/Romance clitics: Structure and derivation. In Henk C. van
Riemsdijk (ed.),Clitics in the Languages of Europe, 543–580. Berlin: De GruyterMouton.
https://doi.org/10.1515/9783110804010.543.
Belletti, Adriana. 2005. Extended Doubling and the VP Periphery. Probus 17(1). 1–35. https:
//doi.org/10.1515/prbs.2005.17.1.1.
Belletti, Adriana & Luigi Rizzi. 1988. Psych-Verbs and -Theory. Natural Language & Lin-
guistic Theory 6(3). 291–352. http://www.jstor.org/stable/4047649.
Bennett, Ryan & Emily Elfner. 2019. The Syntax–Prosody Interface. Annual Review of Lin-
guistics 5(Volume 5, 2019). 151–171. https : / /doi . org /10 . 1146 /annurev- linguistics -
011718-012503.
Bennett, Ryan, Emily Elfner & James McCloskey. 2016. Lightest to the Right: An Ap-
parently Anomalous Displacement in Irish. Linguistic Inquiry 47(2). 169–234. https :
//doi.org/10.1162/LING_a_00209.
Bennett, Ryan, Emily Elfner & James McCloskey. 2019. Prosody, focus, and ellipsis in Irish.
Language 95(1). 66–106. https://muse.jhu.edu/pub/24/article/719234.
Bentzen, Kristine. 2014. Embedded Verb Second (V2). Nordic Atlas of Language Structures
(NALS) Journal 1. 211–224.
354 Bibliography
Bentzen, Kristine, Gunnar Hrafn Hrafnbjargarson, Hróarsdóttir Þorbjörg & Anna-Lena
Wiklund. 2007. The Tromsø guide to the force behind V2. Working Papers in Scandi-
navian Syntax 79. 93–118.
Bermúdez-Otero, Ricardo & John Payne. 2011. There are no special clitics. In Alexandra
Galani, Glyn Hicks & George Tsoulas (eds.), Morphology and its Interfaces, 57–96. John
Benjamins Publishing Company. https://www.degruyterbrill.com/document/doi/10.
1075/la.178.06ber/html.
Bhatt, Rajesh. 2002. The Raising Analysis of Relative Clauses: Evidence from Adjectival
Modification. Natural Language Semantics 10(1). 43–90. https://doi .org/10.1023/A:
1015536226396.
Bhatt, Rajesh. 2005a. Appositive Relative Clauses. Hand-out from a course on the Syntax &
Semantics of Nominal Modification; LOT Summer School, Universiteit Leiden.
Bhatt, Rajesh. 2005b. Long Distance Agreement in Hindi-Urdu. Natural Language & Lin-
guistic Theory 23(4). 757–807. https://doi.org/10.1007/s11049-004-4136-0.
Bianchi, Valentina. 1999. Consequences of antisymmetry: headed relative clauses. Berlin:
Mouton de Gruyter.
Bianchi, Valentina. 2006. On the syntax of personal arguments. Lingua 116(12). 2023–2067.
https://doi.org/10.1016/j.lingua.2005.05.002.
Bittner, Maria. 1994. Case, scope, and binding. Dordrecht: Kluwer.
Bittner, Maria & Ken Hale. 1996a. Ergativity: Toward a Theory of a Heterogeneous Class.
Linguistic Inquiry 27(4). 531–604. http://www.jstor.org/stable/4178953.
Bittner, Maria & Ken Hale. 1996b. The structural determination of case and agreement.
Linguistic Inquiry 27. 1–68.
Blust, Robert. 2013. The Austronesian Languages. Revised edition. Canberra: Pacific Lin-
guistics.
Blust, Robert. 2015. The Case-Markers of Proto-Austronesian. Oceanic Linguistics 54(2).
436–491. http://www.jstor.org/stable/43897710.
Blust, Robert & Victoria Chen. 2017. The pitfalls of negative evidence: ‘Nuclear Austrone-
sian’, ‘Ergative Austronesian’, and their progeny. Language & Linguistics 18(4). 577–
621.
Bobaljik, Jonathan David. 1995. Morphosyntax: the syntax of verbal inflection. MIT disser-
tation. https://dspace.mit.edu/handle/1721.1/11351.
Bobaljik, Jonathan David. 2000. The ins and outs of contextual allomorphy. In Kleanthes K.
Grohmann & Caro Struijke (eds.), The 1999 Maryland Mayfest on Morphology (Univer-
sity of Maryland Working Papers in Linguistics 10). College Park: University of Mary-
land.
Bobaljik, Jonathan David. 2002. A-Chains at the PF-Interface: Copies and ’Covert’ Move-
ment. Natural Language & Linguistic Theory 20(2). 197–267. http://www.jstor.org/
stable/4048028.
Bobaljik, Jonathan David. 2008. Where’s Phi? Agreement as a Postsyntactic Operation.
In Daniel Harbour, David Adger & Susana Bejar (eds.), Phi Theory: Phi-Features across
Bibliography 355
Modules and Interfaces, 295–328. Oxford University Press. https://doi.org/10.1093/oso/
9780199213764.003.0010.
Bobaljik, Jonathan David. 2012. Universals in Comparative Morphology: Suppletion, Su-
perlatives, and the Structure of Words. Cambridge, MA: MIT Press.
Boerger, Brenda H, Åshild Næss, Anders Vaa, Rachel Emerine & Angela Hoover. 2012.
Sociological factors in Reefs-Santa Cruz language vitality: a 40 year retrospective. In-
ternational Journal of the Sociology of Language 2012(214). 111–152. https://doi.org/
10.1515/ijsl-2012-0023.
Boerger, Brenda H. 2017. Change request for Noipä to receive a separate ISO code [npx].
ISO 639-3 change request. https://iso639-3.sil.org/request/2016-017.
Boerger, Brenda H. 2019. Marked use of personal directionals in Natügu narrative texts.
Proceedings of the Conference on Oceanic Linguistics (COOL) 10. 1–23. https://www.sil.
org/resources/publications/entry/82335.
Boerger, Brenda H. 2022. A grammar sketch of Natqgu [ntu]: An Oceanic language of
Santa Cruz, Solomon Islands. Language & Linguistics in Melanesia. Texts in the Lan-
guages of the Pacific (TILP) series.
Bondarenko, Tatiana. 2020. Feature gluttony for Algonquian: agreement in Pas-
samaquoddy.
Bondarenko, Tatiana & Stanislao Zompì. 2021. Leftover Agreement: Spelling Out
Kartvelian Number. Proceedings of WCCFL 38. 65–73.
Bonet, Eulàlia. 1991. Morphology after syntax: Pronominal clitics in Romance. MIT disser-
tation.
Bossi, Madeline & Michael Diercks. 2019. V1 in Kipsigis: Head movement and discourse-
based scrambling. Glossa: a journal of general linguistics 4(1). https://doi.org/10.5334/
gjgl.246.
Bossong, Georg. 1991. Differential Object marking in Romance and Beyond. In Dieter
Wanner & Douglas A. Kibbee (eds.), New Analyses in Romance Linguistics: Selected Pa-
pers from the XVIII Linguistic Symposium on Romance Languages, 143–170. John Ben-
jamins: Amsterdam.
Branan, Kenyon & Michael Yoshitaka Erlewine. 2024. Ā-Probing for the Closest DP. Lin-
guistic Inquiry 55(2). 375–401. https://doi.org/10.1162/ling_a_00459.
Branan, Kenyon & Michael Yoshitaka Erlewine. to appear. Locality and (minimal) search.
In Kleanthes K. Grohmann & Evelina Leivada (eds.), The Cambridge Handbook of the
Minimalist Program. Cambridge: Cambridge University Press.
Brennan, Jonathan. 2009. Pronouns, Inflection, and Irish Prepositions. NYU Working Pa-
pers in Linguistics 2. 1–32.
Brodkin, Dan. 2022. Two steps to high absolutive syntax: Austronesian voice and agent
focus in Mandar. Journal of East Asian Linguistics 31(4). 465–516. https://doi.org/10.
1007/s10831-022-09248-0.
Brodkin, Dan. 2025. Inversion, Priority, and A-Locality.
Burzio, Luigi. 1986. Italian Syntax: A Government-Binding Approach. Dordrecht: Reidel.
356 Bibliography
Bylinina, Lisa & Alexander Podobryaev. 2020. Plurality in Buriat and Structurally Con-
strained Alternatives. Journal of Semantics 37(1). 117–128. https://doi.org/10.1093/jos/
ffz017.
Caha, Pavel. 2009. The nanosyntax of case. University of Tromsø dissertation.
Callegari, Elena. 2018. Understanding Word Order in the Left Periphery. University of Oslo
dissertation.
Cardinaletti, Anna. 1997. Subjects and Clause Structure. In Liliane Haegeman (ed.), The
New Comparative Syntax, 33–63. New York: Longman.
Cardinaletti, Anna. 2004. Toward a Cartography of Subject Positions. In Luigi Rizzi (ed.),
The Structure of CP and IP: The Cartography of Syntactic Structures, Volume 2, 115–165.
Oxford: Oxford University Press. https://doi.org/10.1093/oso/9780195159486.003.0005.
Cardinaletti, Anna & Michal Starke. 1999. The typology of structural deficiency: A case
study of the three classes of pronouns. In Henk C. van Riemsdijk (ed.), Clitics in the
Languages of Europe, 145–234. Berlin: De Gruyter Mouton. https://doi.org/10.1515/
9783110804010.145.
Carstens, Vicki. 2001. Multiple Agreement and Case Deletion: Against 𝜑-incompleteness.
Syntax 4(3). 147–163. https://doi.org/10.1111/1467-9612.00042.
Carstens, Vicki. 2003. Rethinking Complementizer Agreement: Agree with a Case-
Checked Goal. Linguistic Inquiry 34(3). 393–412. https : / / doi . org / 10 . 1162 /
002438903322247533.
Carstens, Vicki. 2005. Agree and Epp in Bantu.Natural Language & Linguistic Theory 23(2).
219–279. https://doi.org/10.1007/s11049-004-0996-6.
Carstens, Vicki. 2010. Implications of grammatical gender for the theory of uninter-
pretable features. In Michael T. Putnam (ed.), Exploring Crash-Proof Grammars, 31–
58. Amsterdam: John Benjamins. https://doi.org/10.1075/lfab.3.03car.
Carstens, Vicki. 2011. Hyperactivity and Hyperagreement in Bantu. Lingua 121(5). 721–
741. https://doi.org/10.1016/j.lingua.2010.11.001.
Carstens, Vicki & Michael Diercks. 2013. Parameterizing case and activity: Hyperraising
in Bantu. Proceedings of NELS 40. 99–118.
Cecchetto, Carlo. 2000. Doubling structures and reconstruction. Probus 12(1). 93–126.
https://doi.org/10.1515/prbs.2000.12.1.93.
Ceña, Resty M. 1979. Tagalog counterexamples to the accessibility hierarchy. Studies in
Philippine Linguistics 3. 119–124. https://www.sil.org/resources/archives/25963.
Chang, Anna Hsiou-chuan. 2006.A reference grammar of Paiwan. Australian National Uni-
versity dissertation.
Chang, Henry Yungli. 2011. Triadic Encoding in Tsou. Language and Linguistics 12(4).
799–843.
Chang, Henry Yungli. 2017. The AV-only Restriction and Locality in Formosan Languages.
Tsing Hua Journal of Chinese Studies 47(2). 231–254. https://doi.org/10.6503/THJCS.
2017.47(2).02.
Chen, Fulang. 2023. Obscured universality in Mandarin. MIT dissertation.
Chen, Tingchun. 2010. Restructuring in Mayrinax Atayal. McGill University BA Thesis.
Bibliography 357
Chen, Victoria. 2017a. A Reexamination of the Philippine-Type Voice System and its Impli-
cations for Austronesian Primary-Level Subgrouping. University of Hawai‘i at Mānoa
dissertation. http://hdl.handle.net/10125/62502.
Chen, Victoria. 2017b. Philippine-type voice affixes as A′-agreement morphology: evi-
dence from causatives. Proceedings of AFLA 23. 35–49.
Chen, Victoria. 2023. Voice morphology as the spell-out of parallel chains: Insights from
Austronesian and beyond. Ms., Victoria University of Wellington.
Chen, Victoria. 2025. The syntax of Philippine-type alignment: Insights from case-
marking. Natural Language & Linguistic Theory. 1–60. https://doi.org/10.1007/s11049-
025-09655-7.
Chen, Victoria & BradleyMcDonnell. 2019.Western Austronesian Voice.Annual Review of
Linguistics 5(1). 173–195. https://doi.org/10.1146/annurev-linguistics-011718-011731.
Chierchia, Gennaro. 2025. Movement and crossover in three languages. Natural Language
& Linguistic Theory. 1–48. https://doi.org/10.1007/s11049-024-09641-5.
Choi, Jaehoon & Heidi Harley. 2019. Locality domains and morphological rules. Natural
Language & Linguistic Theory 37(4). 1319–1365. https://doi.org/10.1007/s11049-018-
09438-3.
Chomsky, Noam. 1965. Aspects of a theory of syntax. Cambridge, MA: MIT Press.
Chomsky, Noam. 1977. On Wh-movement. In Peter Culicover, Thomas Wasow & Adrian
Akmajian (eds.), Formal syntax, 71–132. New York: Academic Press.
Chomsky, Noam. 1981. Lectures on Government and Binding. Dordrecht: Foris.
Chomsky, Noam. 1993. A Minimalist program for linguistic theory. In Ken Hale & Samuel
Jay Keyser (eds.), The View from Building 20: Essays in linguistics in honor of Sylvain
Bromberger, 1–52. Cambridge, MA: MIT Press.
Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press.
Chomsky, Noam. 2000. Minimalist inquiries: the framework. In Martin Roger, David
Michaels & Juan Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in Honor
of Howard Lasnik, 89–155. Cambridge, MA: MIT Press.
Chomsky, Noam. 2001. Derivation by phase. In Michael Kenstowicz (ed.), Ken Hale: A life
in language, 1–52. Cambridge, MA: MIT Press.
Chung, Sandra. 1994. Wh-Agreement and ”Referentiality” in Chamorro. Linguistic Inquiry
25(1). 1–44. http://www.jstor.org/stable/4178847.
Cinque, Guglielmo. 1999. Adverbs and functional heads: A cross-linguistic perspective. Ox-
ford: Oxford University Press.
Cinque, Guglielmo. 2020. The Syntax of Relative Clauses: A Unified Analysis. Cambridge:
Cambridge University Press. https://doi.org/10.1017/9781108856195.
Clem, Emily. 2019. Amahuaca ergative as agreement with multiple heads. Natural Lan-
guage & Linguistic Theory 37(3). 785–823. https://doi.org/10.1007/s11049-018-9431-2.
Clem, Emily. 2022. Accounting for parallels between inverse marking and the PCC. Pro-
ceedings of NELS(52). 149–162.
Clem, Emily. 2023. Cyclic Expansion in Agree: Maximal Projections as Probes. Linguistic
Inquiry 54(1). 39–78. https://doi.org/10.1162/ling_a_00432.
358 Bibliography
Clem, Emily & Amy Rose Deal. 2024. Dependent Case by Agree: Ergative in Shawi. Lin-
guistic Inquiry. 1–47. https://doi.org/10.1162/ling_a_00529.
Codrington, Robert H. 1885. The Melanesian Languages. Oxford: Oxford University Press.
Colley, Justin & Dmitry Privoznov. 2020. On the topic of subjects: composite Probes in
Khanty. Proceedings of NELS 50. 111–124.
Collins, Chris. 1996. Local economy. Cambridge, MA: MIT Press.
Collins, James N. 2019. Definiteness determined by syntax. Natural Language & Linguistic
Theory 37(4). 1367–1420. https://doi.org/10.1007/s11049-018-9436-x.
Compton, Richard. 2016. Mutually conditioned mood and object agreement in Inuit. Pro-
ceedings of NELS 46. 241–250.
Comrie, Bernard. 1986. Markedness, Grammar, People, and the World. In Fred A. Eckman,
Edith A Moravcsik & Jessica R. Wirth (eds.), Markedness, 85–106. New York: Plenum
Press.
Conrod, Kirby. 2019. Pronouns Raising and Emerging. University of Washington disserta-
tion.
Coon, Jessica. 2019. Building verbs in Chuj: Consequences for the nature of roots. Journal
of Linguistics 55(1). 35–81. https://doi.org/10.1017/S0022226718000087.
Coon, Jessica, Nico Baier & Theodore Levin. 2021. Mayan agent focus and the ergative
extraction constraint: Facts and fictions revisited. Language 97(2). 269–332. https : / /
muse.jhu.edu/pub/24/article/794875.
Coon, Jessica & Alan Bale. 2014. The interaction of person and number in Mi’gmaq. Nord-
lyd 41(1). 85–101.
Coon, Jessica, Pedro Mateo Pedro & Omer Preminger. 2014. The role of case in A-bar ex-
traction asymmetries: Evidence fromMayan. Linguistic Variation 14(2). 179–242. https:
//doi.org/10.1075/lv.14.2.01coo.
Crysmann, Berthold & Ana R. Luís. 2024. Morphology of Clitic Pronouns in the Romance
Languages. In Oxford Research Encyclopedia of Linguistics. https://doi.org/10.1093/
acrefore/9780199384655.013.694.
Cysouw, Michael. 2003. The paradigmatic structure of person marking. Oxford: Oxford Uni-
versity Press.
D’Alessandro, Roberta. 2020. Agreement across the board: Topic agreement in Ripano. In
Peter W. Smith, Johannes Mursell & Katharina Hartmann (eds.), Agree to Agree: Agree-
ment in the Minimalist Programme, 235–270. Berlin: Language Science Press. https :
//doi.org/10.5281/ZENODO.3541757.
Dawson, Virginia. 2024. Inclusive plural in a ‘general number’ language. Proceedings of
TripleA 10. 33–46.
De Guzman, Videa. 1988. Ergative analysis for Philippine languages: an analysis. In
Richard McGinn (ed.), Studies in Austronesian Linguistics, 323–345. Athens, Ohio: Ohio
University.
Deal, Amy Rose. 2015a. A note on Nez Perce verb agreement, with sample paradigms.
Proceedings of the International Conference on Salish and Neighbouring Languages 50.
389–413.
Bibliography 359
Deal, Amy Rose. 2015b. Interaction and satisfaction in 𝜑-agreement. Proceedings of NELS
45(1). 179–192.
Deal, Amy Rose. 2016a. Cyclicity and Connectivity in Nez Perce Relative Clauses. Linguis-
tic Inquiry 47(3). 427–470. https://doi.org/10.1162/LING_a_00218.
Deal, Amy Rose. 2016b. Syntactic Ergativity: Analysis and Identification.Annual Review of
Linguistics 2(1). 165–185. https://doi.org/10.1146/annurev-linguistics-011415-040642.
Deal, Amy Rose. 2017. Syntactic Ergativity as Case Discrimination. Proceedings of WCCFL
34. 141–150.
Deal, Amy Rose. 2022. Negative concord as Downward Agree. Proceedings of NELS 52(1).
235–244.
Deal, Amy Rose. 2024a. Interaction, Satisfaction, and the PCC. Linguistic Inquiry 55(1).
39–94. https://doi.org/10.1162/ling_a_00455.
Deal, Amy Rose. 2024b. Probe-specific locality. Talk given at the Berkeley Syntax & Se-
mantics Circle, March 15, 2024.
Deal, Amy Rose. 2025. Case sensitivity reflects case structure: agreement, extraction, and
clitics.
Deal, Amy Rose. to appear. Current models of Agree. In James Crippen, Rose-Marie
Déchaine & Hermann Keupdjio (eds.), Move and Agree: towards a formal typology.
Amsterdam: John Benjamins.
Deal, Amy Rose, Line Mikkelsen & Ellen Thrane. 2024. Syntactic ergativity without inver-
sion in Kalaallisut. Talk presented at NELS 54 (MIT), January 26, 2024. https://nels54.
mit.edu/sites/default/files/documents/deal-mikkelsen-thrane-handout.pdf.
Deal, Amy Rose & Justin Royer. 2025. Mayan animacy hierarchy effects and the dynamics
of Agree. Natural Language & Linguistic Theory 43(3). 1607–1663. https://doi.org/10.
1007/s11049-024-09648-y.
Déchaine, Rose-Marie & Martina Wiltschko. 2002. Decomposing Pronouns. Linguistic In-
quiry 33(3). 409–442. https://doi.org/10.1162/002438902760168554.
den Besten, Hans. 1981. On the interaction of root transformations and lexical deletive
rules. Groninger Arbeiten zur Germanistischen Linguistik 1–3. 1–78.
den Besten, Hans. 1983. On the interaction of root transformations and lexical deletive
rules. In Werner Abraham (ed.), On the Formal Syntax of the Westgermania, 47–131.
Amsterdam: John Benjamins.
dos Santos, Wesley Nascimento. 2023. Wh-agreement in Kawahíva relativization. Ms., UC
Berkeley.
Drummond, Emily N. 2023. Clause structure and ergativity in Nukuoro. UC Berkeley dis-
sertation. https://escholarship.org/uc/item/1xv905sb.
Embick, David. 2007. Linearization and Local Dislocations, Derivational Mechanics and
Interactions. Linguistic Analysis 33(3-4). 303–336.
Embick, David. 2010. Localism versus Globalism in Morphology and Phonology. Cambridge,
MA: MIT press.
Embick, David & Roumyana Izvorski. 1997. Participle-auxiliary word orders in Slavic. In
Natasha Kondrashova, Wayles Browne, Ewa Dornisch & Draga Zec (eds.), Formal Ap-
360 Bibliography
proaches to Slavic Linguistics (FASL): The Cornell Meeting 1995, vol. 4, 210–239. Ann
Arbor: Michigan Slavic Publications.
Embick, David & Alec Marantz. 2008. Architecture and Blocking. Linguistic Inquiry 39(1).
1–53. https://doi.org/10.1162/ling.2008.39.1.1.
Embick, David & Rolf Noyer. 2001. Movement Operations after Syntax. Linguistic Inquiry
32(4). 555–595. https://doi.org/10.1162/002438901753373005.
Erlewine, Michael Yoshitaka. 2018. Extraction and Licensing in Toba Batak. Language
94(3). 662–697.
Erlewine, Michael Yoshitaka. 2020. Patterns of relativization in Austronesian and Tibetan.
Ms., National University of Singapore. https://mitcho.com/research/handout-tibetan-
austronesian.pdf.
Erlewine, Michael Yoshitaka, Theodore Levin & Coppe van Urk. 2017. Ergativity and
Austronesian-type voice systems. In Jessica Coon, Diane Massam & Lisa Travis (eds.),
The Oxford Handbook of Ergativity, 373–396. Oxford: Oxford University Press.
Erlewine, Michael Yoshitaka & Cheryl Lim. 2023. Bikol clefts and topics and the Aus-
tronesian extraction restriction. Natural Language & Linguistic Theory 41(3). 911–960.
https://doi.org/10.1007/s11049-022-09555-0.
Erlewine, Michael Yoshitaka & Joey Lim. 2023. Deriving scrambling with word order freez-
ing. Glossa: a journal of general linguistics 8(1). https://doi.org/10.16995/glossa.8364.
Erlewine, Michael Yoshitaka & Carly J. Sommerlot. 2023a. The Malayic verbal phase and
Cyclic Linearization. Talk presented at AFLA 30.
Erlewine, Michael Yoshitaka & Carly J. Sommerlot. 2023b. Voice and extraction in Malayic.
Ms., National University of Singapore.
Ershova, Ksenia. 2019. Syntactic ergativity in West Circassian. University of Chicago dis-
sertation.
Ershova, Ksenia. 2022. Unexpected Ā-movement in West Circassian: Theoretical implica-
tions for syntactic ergativity.
Ershova, Ksenia. 2024. Ergativity in the syntax without syntactic ergativity. Ms., UCL.
Evans, Bethwyn. 2003. A study of valency-changing devices in Proto Oceanic. Canberra:
Pacific Linguistics.
Farkas, Donka & Henriëtte de Swart. 2010. The semantics and pragmatics of plurals. Se-
mantics and Pragmatics 3. 6:1–54. https://doi.org/10.3765/sp.3.6.
Feldcamp, Zachary Satoshi. 2025. Precedence-sensitive locality in English locative inver-
sion. Ms., MIT.
Ferrell, Raleigh. 1979. Construction markers and subgrouping of Formosan languages.
South-East Asian Linguistics Studies 3. 199–212.
Fitzpatrick, Justin Michael. 2006. The syntactic and semantic roots of floating quantification.
Massachusetts Institute of Technology Thesis. https://dspace.mit.edu/handle/1721.1/
37420.
Foley, Steven. 2013. The Syntax of Georgian Relative Clauses. NYU Unpublished BAHonors
Thesis. https://docslib.org/doc/2498918/the-syntax-of-georgian-relative-clauses.
Bibliography 361
Foley, William A. 2008. The place of Philippine languages in a typology of voice systems.
In Peter K. Austin & Simon Musgrave (eds.), Voice and grammatical relations in Aus-
tronesian languages, 22–44. CSLI Publication: Stanford.
Fong, Suzana. 2019. Proper movement through Spec-CP: An argument from hyperraising
in Mongolian. Glossa: a journal of general linguistics 4(1). https://doi.org/10.5334/gjgl.
667.
Fox, Danny. 2002. Antecedent-Contained Deletion and the Copy Theory of Movement.
Linguistic Inquiry 33(1). 63–96. https://doi.org/10.1162/002438902317382189.
Fox, Danny & Jon Nissenbaum. 1999. Extraposition and Scope: A case for overt QR. Pro-
ceedings of WCCFL 18. 132–144.
Frostad, Benedicte Haraldstad. 2006. Syntactic and semantic aspects of some verbs of motion
and location in Äiwoo. University of Oslo MA thesis.
Ganenkov, Dmitry. 2020. Missing Elsewhere: Domain Extension in Contextual Allomor-
phy. Linguistic Inquiry 51(4). 785–798. https://doi.org/10.1162/ling_a_00358.
Georgi, Doreen. 2012. A Local Derivation of Global Case Splits. In Artemis Alexiadou,
Tibor Kiss & Gereon Müller (eds.), Local modeling of non-local dependencies in syntax,
305–336. Berlin: De Gruyter Mouton.
Gerdts, Donna B. 1988. Antipassives and causatives in Ilokano: evidence for an ergative
analysis. In Richard McGinn (ed.), Studies in Austronesian Linguistics, 295–321. Athens,
Ohio: Ohio University.
Gong, Zhiyu Mia. 2023. A/Ā-Operations at the Mongolian Clausal Periphery. Journal of
East Asian Linguistics 32(4). 413–457. https://doi.org/10.1007/s10831-023-09268-4.
Grewendorf, Günther & Joachim Sabel. 1999. Scrambling in German and Japanese: Ad-
junction Versus Multiple Specifiers. Natural Language & Linguistic Theory 17(1). 1–65.
https://doi.org/10.1023/A:1006068326583.
Grishin, Peter. 2023. Omnivorous third person agreement in Algonquian. Glossa: a journal
of general linguistics 8(1). https://doi.org/10.16995/glossa.8874.
Grishin, Peter. 2025. Impersonal impersonals and personal third persons: An argument for
binary [±PART]. Proceedings of NELS 55. 1–10.
Guilfoyle, Eithne, Henrietta Hung & Lisa Travis. 1992. SPEC of IP and SPEC of VP: Two
Subjects in Austronesian Languages.Natural Language & Linguistic Theory 10(3). 375–
414. http://www.jstor.org/stable/4047773.
Halle, Morris. 1997. Distributed Morphology: Impoverishment and Fission. MIT Working
Papers in Linguistics 30. 425–449.
Halle, Morris & Alec Marantz. 1993. Distributed Morphology and the Pieces of Inflection.
In KenHale & Samuel Jay Keyser (eds.), The View from Building 20: Essays in Linguistics
in Honor of Sylvain Bromberger, 111–176. Cambridge, MA: MIT Press.
Halpert, Claire. 2019. Raising, unphased. Natural Language & Linguistic Theory 37(1). 123–
165. https://doi.org/10.1007/s11049-018-9407-2.
Hammerly, Christopher. to appear. The relativized EPP: Evidence from agreement and
word order in Border Lakes Ojibwe. Canadian Journal of Linguistics/Revue canadienne
de linguistique.
362 Bibliography
Hammerly, Christopher & Eric Mathieu. to appear. Linking agreement and movement: A
case study of long distance agreement in Border Lakes Ojibwe. Linguistic Inquiry.
Han, Chung-hye, Jeffrey Lidz & Julien Musolino. 2007. V-Raising and Grammar Competi-
tion in Korean: Evidence from Negation and Quantifier Scope. Linguistic Inquiry 38(1).
1–47. https://doi.org/10.1162/ling.2007.38.1.1.
Han, Chung-hye, Julien Musolino & Jeffrey Lidz. 2016. Endogenous sources of variation in
language acquisition. Proceedings of the National Academy of Sciences 113(4). 942–947.
https://doi.org/10.1073/pnas.1517094113.
Harbour, Daniel. 2016. Impossible Persons. Cambridge, MA: MIT Press.
Harðarson, Gísli Rúnar. 2018. Forming a Compound and Spelling it Out. UPenn Working
Papers in Linguistics 24. 87–96. https://opinvisindi.is/handle/20.500.11815/1934.
Harizanov, Boris. 2014. Clitic doubling at the syntax-morphophonology interface. Natural
Language & Linguistic Theory 32(4). 1033–1088. https://doi.org/10.1007/s11049-014-
9249-5.
Harizanov, Boris. 2019. Head movement to specifier positions. Glossa: a journal of general
linguistics 4(1). https://doi.org/10.5334/gjgl.871.
Harizanov, Boris & Vera Gribanova. 2019. Whither head movement? Natural Language &
Linguistic Theory 37(2). 461–522. https://doi.org/10.1007/s11049-018-9420-5.
Harley, Heidi. 1995. Subjects, events and licensing. MIT dissertation.
Harley, Heidi. 2013. Diagnosing head movement. In Lisa Lai-Shen Cheng & Norbert
Corver (eds.), Diagnosing Syntax, 112–119. Oxford: Oxford University Press.
Harley, Heidi & Rolf Noyer. 1999. Distributed morphology. Glot international 4(4). 3–9.
Harley, Heidi & Elizabeth Ritter. 2002. Person and Number in Pronouns: A Feature-
Geometric Analysis. Language 78(3). 482–526. https://muse.jhu.edu/pub/24/article/
19419.
Hedding, Andrew & Michelle Yuan. 2023. Phase unlocking and the derivation of verb-
initiality in San Martín Peras Mixtec. Proceedings of WCCFL 41. 1–10.
Herbeck, Peter. 2015. Overt PRO in Romance: Towards a unification of PRO and pro. In
Rachel Klassen, Juana M. Liceras & Elena Valenzuela (eds.), Hispanic Linguistics at the
Crossroads: Theoretical linguistics, language acquisition and language contact. Proceed-
ings of the Hispanic Linguistics Symposium 2013, 25–48. John Benjamins Publishing
Company. https://www.degruyterbrill.com/document/doi/10.1075/ihll.4.02her/html.
Heycock, Caroline, Antonella Sorace & Zakaris Svabo Hansen. 2010. V-to-I and V2 in
subordinate clauses: an investigation of Faroese in relation to Icelandic and Danish.
The Journal of Comparative Germanic Linguistics 13(1). 61–97. https://doi.org/10.1007/
s10828-010-9035-7.
Himmelmann, Nikolaus P. 2002. Voice in western Austronesian: an update. In FayWouk &
Malcolm Ross (eds.), The History and Typology of Western Austronesian Voice Systems,
7–16. Canberra: Pacific Linguistics.
Himmelmann, Nikolaus P. 2005. The Austronesian Languages of Asia and Madagascar:
Typological Characteristics. In Alexander Adelaar & Nikolaus P. Himmelmann (eds.),
The Austronesian Languages of Asia and Madagascar, 110–181. London: Routledge.
Bibliography 363
Himmelmann, Nikolaus P. & Sonja Riesberg. 2013. Symmetrical voice and applicative al-
ternations: Evidence from Totoli. Oceanic Linguistics 52(2). 396–422.
Hoji, Hajime. 1998. Null Object and Sloppy Identity in Japanese. Linguistic Inquiry 29(1).
127–152. https://doi.org/10.1162/002438998553680.
Holmberg, Anders. 1986.Word Order and Syntactic Features in the Scandinavian Languages
and English. University of Stockholm dissertation.
Holmberg, Anders. 1999. Remarks on Holmberg’s Generalization. Studia Linguistica 53(1).
1–39. https://doi.org/10.1111/1467-9582.00038.
Holmen, Sindre Sperre. 2020. Voice Choice in Äiwoo: A Discourse-Functional Study of the
Symmetrical Voice System of Äiwoo. University of Oslo MA thesis.
Holmen, Sindre Sperre & Åshild Næss. 2025. Voice choice in Äiwoo and the symmetrical
voice-to-transitivity shift. Oceanic Linguistics 64(1). 1–45.
Hsieh, Henrison. 2019. Distinguishing nouns and verbs. Natural Language & Linguistic
Theory 37(2). 523–569. https://doi.org/10.1007/s11049-018-9422-3.
Hsieh, Henrison. 2020. Beyond nominative: A broader view of A′-dependencies in Tagalog.
McGill University dissertation.
Hsieh, Henrison. 2023. Observations on Tagalog Genitive Inversion. Oceanic Linguistics
62(2). 242–266. https://muse.jhu.edu/pub/5/article/913560.
Hsieh, Henrison. 2025. Locality in Exceptional Tagalog Ā-Extraction. Linguistic Inquiry
56(3). 519–560. https://doi.org/10.1162/ling_a_00505.
Hulsey, Sarah & Uli Sauerland. 2006. Sorting Out Relative Clauses. Natural Language Se-
mantics 14(2). 111–137. http://www.jstor.org/stable/23749618.
Jackendoff, Ray. 1977. X̄-Syntax. Cambridge, MA: MIT Press.
Jaeggli, Oswaldo. 1982. Topics in Romance syntax. Dordrecht: Foris.
Jaeggli, Oswaldo. 1986. Three Issues in the Theory of Clitics: Case, Doubled NPs, and
Extraction. In Hagit Borer (ed.), The Syntax of Pronominal Clitics, 15–42. New York:
Academic Press.
Jarvis, Rebecca. 2023. Subject movement in Atchan relative clauses: A two-probe, mixed
A/A’-account. Proceedings of WCCFL 41. 1–10.
Jarvis, Rebecca. 2025. Raising and head-external relatives in Atchan. Natural Language &
Linguistic Theory. 1–41. https://doi.org/10.1007/s11049-024-09651-3.
Jenks, Peter. 2023. Topic and focus in Tira. Talk given at the Berkeley Syntax & Semantics
Circle.
Jouitteau, Mélanie &Milan Rezac. 2006. Deriving the Complementarity Effect: Relativized
minimality in Breton agreement. Lingua 116(11). 1915–1945. https://doi.org/10.1016/
j.lingua.2005.03.012.
Julien, Marit. 2002. Syntactic heads and word formation. Oxford: Oxford University Press.
Julien, Marit. 2007a. Embedded V2 in Norwegian and Swedish. Working Papers in Scandi-
navian Syntax 80. 103–161.
Julien, Marit. 2007b. On the relation between morphology and syntax. In Gillian Ram-
chand & Charles Reiss (eds.), The Oxford Handbook of Linguistic Interfaces, 209–238.
Oxford: Oxford University Press.
364 Bibliography
Julien, Marit. 2015. The force of V2 revisited. The Journal of Comparative Germanic Lin-
guistics 18(2). 139–181. https://doi.org/10.1007/s10828-015-9073-2.
Kayne, Richard S. 1975. French syntax: The transformational cycle. Cambridge, MA: MIT
Press.
Kayne, Richard S. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press.
Keenan, Edward L. & Bernard Comrie. 1977. Noun Phrase Accessibility and Universal
Grammar. Linguistic Inquiry 8(1). 63–99. http://www.jstor.org/stable/4177973.
Keine, Stefan. 2010. Case and agreement from fringe to core: A minimalist approach. Berlin:
Mouton de Gruyter.
Keine, Stefan & Rajesh Bhatt. to appear. Crossover asymmetries. STAR: Syntactic Theory
and Research.
Keine, Stefan & Hedde Zeijlstra. 2025. Clause-internal successive cyclicity: Phasality or
DP intervention? Natural Language & Linguistic Theory. 1–64. https : / / doi . org / 10 .
1007/s11049-024-09627-3.
Kinyalolo, Kasangati K. W. 1991. Syntactic dependencies and the Spec-head agreement hy-
pothesis in Kilega. UCLA dissertation.
Ko, Heejeong. 2014. Edges in Syntax: Scrambling and Cyclic Linearization. Oxford: Oxford
University Press.
Koopman, Hilda. 1999. The Internal and External Distribution of Pronominal DPs. In Kyle
Johnson & Ian Roberts (eds.), Beyond Principles and Parameters: Essays in Memory of
Osvaldo Jaeggli, 91–132. Dordrecht: Springer Netherlands. https://doi.org/10.1007/
978-94-011-4822-1_4.
Krajinović, Ana, Jozina Vander Klok & Kilu von Prince. 2023. When (not) to establish a
new category. The case of perfect, ‘already’, and iamitives. Ms., Tilburg University &
Humboldt University of Berlin & Heinrich-Heine-Universität in Düsseldorf.
Kramer, Ruth. 2014. Clitic doubling or object agreement: the view from Amharic. Natural
Language & Linguistic Theory 32(2). 593–634.
Krapova, Iliyana. 2010. Bulgarian relative and factive clauses with an invariant comple-
mentizer. Lingua 120(5). 1240–1272. https://doi.org/10.1016/j.lingua.2009.08.002.
Krapova, Iliyana & Guglielmo Cinque. 2008. Clitic reduplication constructions in Bulgar-
ian. In Dalina Kallulli & Liliane Tasmowski (eds.), Clitic Doubling in the Balkan Lan-
guages, 257–287. John Benjamins Publishing Company. https://doi.org/10.1075/la.130.
15kra.
Kroeger, Paul R. 1993. Phrase structure and grammatical relations in Tagalog. Stanford, CA:
CSLI Publications.
Kroeger, Paul R. & Sonja Riesberg. 2024. Voice and transitivity. In Alexander Adelaar &
Antoinette Schapper (eds.), The Oxford Guide to the Malayo-Polynesian Languages of
Southeast Asia, 795–824. Oxford: Oxford University Press.
Lackey, William James & Brenda H. Boerger. 2021. Reexamining the Phonological History
of Oceanic’s Temotu Subgroup. Oceanic Linguistics 60(2). 367–411. https://muse.jhu.
edu/pub/5/article/836178.
Bibliography 365
Landau, Idan. 2006. Chain Resolution In Hebrew V(P)-fronting. Syntax 9(1). 32–66. https:
//doi.org/10.1111/j.1467-9612.2006.00084.x.
Lasnik, Howard. 1981. Restricting the theory of transformations: A case study. In Norbert
Hornstein & David Lightfoot (eds.), Explanation in linguistics: The logical problem of
language acquisition, 152–173. London: Longman.
Lasnik, Howard. 1995. Verbal morphology: Syntactic Structures meets the Minimalist Pro-
gram. In Hector Campos & Paula Kempchinsky (eds.), Evolution and revolution in lin-
guistic theory: Essays in honor of Carlos Otero, 251–275. Washington: Georgetown Uni-
versity Press.
Latrouite, Anja. 2011. Voice and Case in Tagalog: The coding of prominence and orientation.
Heinrich-Heine Universität in Düsseldorf dissertation.
Lebeaux, David. 1988. Language acquisition and the form of grammar. UMass Amherst
dissertation.
Lebeaux, David. 1991. Where Does the Binding Theory Apply? Tech. rep. NEC Research
Institute Technical Report.
Lee, Tommy Tsz-Ming. 2022. Towards the unity of movement: implications from verb move-
ment in Cantonese. University of Southern California dissertation.
Lee, Tommy Tsz-Ming. 2024. The Unity of Movement: Evidence from verb movement in
Cantonese. John Benjamins. https://doi.org/10.1075/la.283.
Legate, Julie Anne. 1999. The morphosyntax of Irish agreement. MIT Working Papers in
Linguistics 33. 219–240.
Legate, Julie Anne. 2012. Subjects in Acehnese and the nature of the passive. Language
88(3). 495–525. https://muse.jhu.edu/pub/24/article/485162.
Legate, Julie Anne. 2014. Voice and v: Lessons from Acehnese. Cambridge, MA: MIT Press.
Levin, Ted. 2015. Unmarked case is unvalued case: Default Voice in Formosan restructur-
ing. Ms., MIT.
Levin, Theodore & Maria Polinsky. 2019. Morphology in Austronesian Languages. In
Oxford Research Encyclopedia of Linguistics. https : / / doi . org / 10 . 1093 / acrefore /
9780199384655.013.527.
Lewis, M. Paul & Gary F. Simons. 2010. Assessing endangerment: expanding fishman’s
GIDS. Revue roumaine de linguistique 55(2). 103–120.
Liao, Hsiu-Chuan. 2004. Transitivity and ergativity in Formosan and Philippine languages.
University of Hawai‘i at Mānoa dissertation.
Lincoln, Peter C. 1978. Reef–Santa Cruz as Austronesian. In Lois Carrington & Stephen
AdolpheWurm (eds.), Second International Conference on Austronesian Linguistics: Pro-
ceedings, 929–967. Canberra: Pacific Linguistics.
Liu, Kang Franco. 2025. Transitive subject relativization restriction in Northern Tujia and
beyond. Ms., UC Berkeley.
Lohninger, Magdalena. 2025. Patterns in Chaos: Composite A’/A Probes. University of Vi-
enna dissertation.
Lynch, John, Malcolm Ross & Terry Crowley. 2002. The Oceanic Languages. Richmond:
Curzon.
366 Bibliography
Mahajan, Anoop. 1990. The A/A-bar distinction and movement theory. MIT dissertation.
Marantz, Alec. 1988. Clitics, morphological merger, and the mapping to phonological
structure. In Michael Hammond & Marie Noonan (eds.), Theoretical morphology, 253–
270. San Diego: Academic Press.
Marantz, Alec. 1991. Case and licensing. In German Westphal, Benjamin Ao & Hee-Rahk
Chae (eds.), Proceedings of the 8th Eastern States Conference on Linguistics (ESCOL 8),
234–253. Ithaca, NY: CLC Publications.
Marantz, Alec. 1995. “Cat” as a phrasal idiom: consequences of late insertion in Distributed
Morphology. Ms., MIT.
Marantz, Alec. 1997. No Escape from Syntax: Don’t Try Morphological Analysis in the Pri-
vacy of Your Own Lexicon. In Alexis Dimitriadis, Laura Siegel, Clarissa Surek-Clark &
Alexander Williams (eds.), Proceedings of the 21st Annual Penn Linguistics Colloquium,
UPennWorking Papers in Linguistics Vol. 4.2, 201–225. Philadelphia: University of Penn-
sylvania.
Matushansky, Ora. 2006. Head movement in linguistic theory. Linguistic Inquiry 37(1). 69–
109.
McCloskey, James & Kenneth Hale. 1984. On the syntax of person-number inflection in
modern Irish. Natural Language & Linguistic Theory 1(4). 487–533. https://doi.org/10.
1007/BF00417057.
McGinn, Richard. 1988. Government and Case in Tagalog. In RichardMcGinn (ed.), Studies
in Austronesian Linguistics, 275–293. Athens, Ohio: Ohio University.
McGinnis, Martha. 1998. Locality in A-movement. MIT dissertation.
McGinnis, Martha. 2005. On markedness asymmetries in person and number. Language
81(3). 699–718.
McKay, Graham R. 1978. Pronominal person and number categories in Rembarrnga and
Djeebbana. Oceanic Linguistics 17(1). 27–37.
Merchant, Jason. 2015. How Much Context Is Enough? Two Cases of Span-Conditioned
Stem Allomorphy. Linguistic Inquiry 46(2). 273–303. https://doi.org/10.1162/LING_a_
00182.
Miller, George A &NoamChomsky. 1963. Finitary models of language users. In R. Duncan
Luce, Robert R Bush & Eugene Galanter (eds.), Handbook of Mathematical Psychology,
vol. 2, 419–491. New York, NY: John Wiley.
Mithun, Marianne. 1994. The Implications of Ergativity for a Philippine Voice System. In
Barbara A. Fox & Paul J. Hopper (eds.), Voice: Form and Function, 247–277. Amsterdam:
John Benjamins. https://doi.org/10.1075/tsl.27.11mit.
Miyagawa, Shigeru. 1997. Against Optional Scrambling. Linguistic Inquiry 28(1). 1–25.
http://www.jstor.org/stable/4178963.
Moskal, Beata. 2015. Limits on Allomorphy: A Case Study in Nominal Suppletion. Linguis-
tic Inquiry 46(2). 363–376. https://doi.org/10.1162/LING_a_00185.
Moskal, Beata & Peter W. Smith. 2016. Towards a theory without adjacency: hyper-
contextual VI-rules. Morphology 26(3). 295–312. https : / / doi . org / 10 . 1007 / s11525 -
015-9275-y.
Bibliography 367
Müller, Gereon. 1997. Optional Movement and the Interaction of Economy Constraints. In
Chris Wilder, Hans-Martin Gärtner & Manfred Bierwisch (eds.), The Role of Economy
Principles in Linguistic Theory, 115–145. Akademie Verlag. https://www.degruyterbrill.
com/document/doi/10.1515/9783050072173-006/html.
Müller, Gereon. 1998. Incomplete Category Fronting. Amsterdam: Springer.
Myler, Neil. 2017. Cliticization feeds agreement: a view from Quechua. Natural Language
& Linguistic Theory 35(3). 751–800. https://doi.org/10.1007/s11049-016-9351-y.
Næss, Åshild. 2000. Possessive Marking in Pileni. Sprachtypologie und Universalien-
forschung 53(3-4). 308–318. https://doi.org/10.1524/stuf.2000.53.34.308.
Næss, Åshild. 2002. Transitivity and accusativity in Pileni. Rongorongo Studies 12(2). 66–
67.
Næss, Åshild. 2004a. Serial verbs and complex constructions in Pileni (Solomon Islands).
In Isabelle Bril & Françoise Ozanne-Rivierre (eds.), Complex predicates in oceanic lan-
guages: Studies in the dynamics of binding and boundedness, 225–250. Berlin: Mouton
de Gruyter.
Næss, Åshild. 2004b. Spatial deixis in Pileni. In Gunter Senft (ed.), Deixis and Demonstra-
tives in Oceanic Languages, 81–97. Canberra: Pacific Linguistics.
Næss, Åshild. 2006. Bound Nominal Elements in Äiwoo (Reefs): A Reappraisal of the ”Mul-
tiple Noun Class Systems”. Oceanic Linguistics 45(2). 269–296. http://www.jstor.org/
stable/4499965.
Næss, Åshild. 2011a. Case on the margins: Pragmatics and argument marking in Vaeakau-
Taumako and beyond. In Case, Animacy and Semantic Roles, chap. Case, Animacy
and Semantic Roles, 305–328. John Benjamins Publishing Company. https : / /www.
degruyterbrill.com/document/doi/10.1075/tsl.99.11nae/html.
Næss, Åshild. 2011b. Directional Verbs in Vaeakau-Taumako. Oceanic Linguistics 50(1).
120–139. https://muse.jhu.edu/pub/5/article/447598.
Næss, Åshild. 2012a. Cutting and breaking in Äiwoo: Event integration and the complexity
of lexical expressions. Cognitive Linguistics 23(2). 395–420. https://doi.org/10.1515/
cog-2012-0012.
Næss, Åshild. 2012b. Structural Parallels between Vaeakau-Taumako and the Vanuatu
Outliers: Capell Revisited. Oceanic Linguistics 51(2). 567–588. https://muse.jhu.edu/
pub/5/article/493457.
Næss, Åshild. 2013. From Austronesian Voice to Oceanic Transitivity: Äiwoo as the ”Miss-
ing Link”. Oceanic Linguistics 52(1). 106–124. http://www.jstor.org/stable/43286763.
Næss, Åshild. 2015a. The Äiwoo verb phrase: Syntactic ergativity without pivots. Journal
of Linguistics 51(1). 75–106. https://doi.org/10.1017/S0022226714000048.
Næss, Åshild. 2015b. Voice at the Crossroads: Symmetrical Clause Alternations in Äiwoo,
Reef Islands, Solomon Islands. Oceanic Linguistics 54(1). 270–307. https://muse.jhu.
edu/pub/5/article/586715.
Næss, Åshild. 2017a. A short dictionary of Äiwoo. Canberra: Asia-Pacific Linguistics. https:
//openresearch-repository.anu.edu.au/handle/1885/112469.
368 Bibliography
Næss, Åshild. 2017b. Beyond roots and affixes: Äiwoo deverbal nominals and the typology
of bound lexical morphemes. Studies in Language 41(4). 914–955. https://doi.org/10.
1075/sl.16087.nae.
Næss, Åshild. 2018a. Plural-Marking Strategies in Äiwoo. Oceanic Linguistics 57(1). 31–62.
http://www.jstor.org/stable/26779820.
Næss, Åshild. 2018b. Standing up to the canoe: Competing cognitive biases in the encoding
of stative spatial relations in a language with a single spatial preposition. Cognitive
Linguistics 29(4). 807–841. https://doi.org/10.1515/cog-2017-0096.
Næss, Åshild. 2021a. Äiwoo wâtu and the typology of comparatives. Studies in Language
45(4). 938–967. https://doi.org/10.1075/sl.20022.nae.
Næss, Åshild. 2021b. Voice and Valency Morphology in Äiwoo. Oceanic Linguistics 60(1).
160–198. https://doi.org/10.1353/ol.2021.0005.
Næss, Åshild. 2023a. Binominals in Äiwoo: Compounds, possessive constructions, and
transitional cases. In Binominals in Äiwoo: Compounds, possessive constructions, and
transitional cases, 181–212. De Gruyter Mouton. https : / / doi . org / 10 . 1515 /
9783110673494-006.
Næss, Åshild. 2023b. Voice and Pluractionality in Äiwoo. Oceanic Linguistics 62(1). 47–80.
https://muse.jhu.edu/pub/5/article/885181.
Næss, Åshild. 2024. Voice and transitivity in Äiwoo and Engdewu: A case study in align-
ment change. Studies in Language. 1–36. https://doi.org/10.1075/sl.23034.nae.
Næss, Åshild. in prep. A grammar of Äiwoo.
Næss, Åshild. in prep. Äiwoo phonology and morphophonology. Ms., University of Oslo.
Næss, Åshild & Brenda H. Boerger. 2008. Reefs-Santa Cruz as Oceanic: Evidence from
the Verb Complex. Oceanic Linguistics 47(1). 185–212. http://www.jstor.org/stable/
20172343.
Næss, Åshild & Even Hovdhaugen. 2007. The History of Polynesian Settlement in the Reef
and Duff Islands: The Linguistic Evidence. Journal of the Polynesian Society 116(4).
433–449.
Næss, Åshild & Even Hovdhaugen. 2011. A Grammar of Vaeakau-Taumako. Berlin: De
Gruyter Mouton. https : / / www . degruyterbrill . com / document / doi / 10 . 1515 /
9783110238273/html.
Næss, Åshild & Mathias Jenny. 2011. Who changes language? Bilingualism and structural
change in Burma and the Reef Islands. Journal of Language Contact 4. 217–249. https:
//doi.org/10.1163/187740911X589253.
Nakamura, Masanori. 1996. Economy of chain formation. McGill University dissertation.
Neu, Eva. 2023. Against a low subject analysis of causatives of unergatives. Proceedings of
the LSA 8(1). 5527–5527. https://doi.org/10.3765/plsa.v8i1.5527.
Neu, Eva. to appear. Direct causatives of unergatives: A view from variable unaccusativity.
Natural Language & Linguistic Theory. 1–33.
Nevins, Andrew. 2004. Derivations without the activity condition. In Proceedings of the
EPP/Phase Workshop. Cambridge, MA: MITWPL.
Bibliography 369
Nevins, Andrew. 2007. The representation of third person and its consequences for person-
case effects. Natural Language & Linguistic Theory 25(2). 273–313. https://doi.org/10.
1007/s11049-006-9017-2.
Nevins, Andrew. 2011. Multiple agree with clitics: person complementarity vs. omnivo-
rous number. Natural Language & Linguistic Theory 29(4). 939–971. https://doi.org/
10.1007/s11049-011-9150-4.
Nie, Yining & Henrison Hsieh. 2024. Tagalog Valency Morphology and its Neutralization.
Proceedings of AFLA 30. 1–15.
Noyer, Rolf. 1992. Features, positions and affixes in autonomous morphological structure.
MIT dissertation.
Nunes, Jairo. 2004. Linearization of Chains and Sidewards Movement. Cambridge, MA: MIT
Press.
O’Herin, Brian. 1993. Wh-agreement in Abaza. In Eric Potsdam & Geoffrey K. Pullum
(eds.), Syntax at Santa Cruz, vol. 2, 25–56. Linguistics Research Center: University of
California Santa Cruz.
O’Herin, Brian. 2002. Case and agreement in Abaza. Dallas, Texas: SIL International. https:
//www.sil.org/resources/publications/entry/8421.
Ostrove, Jason. 2018. When phi-Agreement Targets Topics: The view from San Martín Peras
Mixtec. UC Santa Cruz dissertation. https://escholarship.org/uc/item/55b8d7jw.
Otsuka, Yuko. 2006. Syntactic ergativity in Tongan. In Alana Johns, Diane Massam & Ju-
venal Ndayiragije (eds.), Ergativity: Emerging issues, 79–107. Springer.
Otsuka, Yuko. 2010. DP ellipsis in Tongan: is syntactic ergativity real? Natural Language
& Linguistic Theory 28(2). 315–342. http://www.jstor.org/stable/40856167.
Oxford, Will. 2024. Probe Specification and Agreement Variation: Evidence from the Al-
gonquian Inverse. Linguistic Inquiry 55(3). 489–535. https://doi.org/10.1162/ling_a_
00478.
Paparounas, Lefteris. 2023. Visibility and Intervention in Allomorphy: Lessons from Mod-
ern Greek. Linguistic Inquiry. 1–41. https://doi.org/10.1162/ling_a_00479.
Paparounas, Lefteris & Martin Salzmann. 2024. First conjunct clitic doubling in Modern
Greek: Evidence for Agree-based approaches to clitic doubling. Natural Language &
Linguistic Theory 42(1). 323–382. https://doi.org/10.1007/s11049-023-09585-2.
Partee, Barbara H. 1975. Montague Grammar and Transformational Grammar. Linguistic
Inquiry 6(2). 203–300.
Payne, Thomas E. 1982. Role and Reference Related Subject Properties and Ergativity in
Yup’ik Eskimo and Tagalog. Studies in Language 6(1). 75–106. https://doi.org/10.1075/
sl.6.1.05pay.
Pearson, Matthew. 2001. The clause structure of Malagasy: A minimalist approach. UCLA
dissertation.
Pearson, Matthew. 2005. The Malagasy Subject/Topic as an A′-Element. Natural Language
& Linguistic Theory 23(2). 381–457. https://doi.org/10.1007/s11049-004-1582-7.
370 Bibliography
Perry, J. Joseph. 2025. On the Irish agreement pattern and its parallels: Asymmetric
chains and defective goals. Journal of Linguistics. 1–43. https : / / doi . org / 10 . 1017 /
S0022226725000088.
Pescarini, Diego. 2023. Clitic Cluster Restrictions. In The Wiley Blackwell Companion to
Morphology, 1–39. John Wiley & Sons, Ltd. https://doi.org/10.1002/9781119693604.
morphcom018.
Pesetsky, David. 2013. Russian Case Morphology and the Syntactic Categories. Cambridge,
MA: MIT Press.
Pesetsky, David & Esther Torrego. 2001. T-to-C movement: Causes and consequences. In
Michael Kenstowicz (ed.), Ken Hale: A life in language, 355–426. Cambridge, MA: MIT
Press.
Pesetsky, David & Esther Torrego. 2007. The syntax of valuation and the interpretability
of features. In Simin Karimi, Vida Samiian & Wendy K Wilkins (eds.), Phrasal and
clausal architecture: Syntactic derivation and interpretation, 262–294. Amsterdam: John
Benjamins.
Pizarro-Guevara, Jed Sam & Rowena Garcia. 2024. Philippine Psycholinguistics. Annual
Review of Linguistics 10. 1–23. https://doi.org/10.1146/annurev- linguistics-031522-
102844.
Pizarro-Guevara, Jed Sam & Matthew Wagers. 2018. Agent extraction under patient
voice in Tagalog is acceptable: Evidence from acceptability ratings in various A’-
dependencies. Talk presented at AFLA 25.
Pizarro-Guevara, Jed Sam&MatthewWagers. 2020. The Predictive Value of Tagalog Voice
Morphology in Filler-Gap Dependency Formation. Frontiers in Psychology 11. https :
//www.frontiersin.org/articles/10.3389/fpsyg.2020.00517.
Pizarro-Guevara, Jed Sam & Matthew Wagers. 2021. (A)Symmetries in Tagalog Relative
Clause Processing. Proceedings of AFLA 27. 113–128.
Pizarro-Guevara, Jed Sam & Matthew Wagers. 2024. A tale of two Tagalogs. Glossa: a
journal of general linguistics 9(1). https://doi.org/10.16995/glossa.11032.
Polinsky,Maria. 2016.Deconstructing Ergativity: Two Types of Ergative Languages and Their
Features. Oxford, New York: Oxford University Press.
Polinsky, Maria & Eric Potsdam. 2001. Long-Distance Agreement And Topic In Tsez. Nat-
ural Language & Linguistic Theory 19(3). 583–646. https : / / doi . org / 10 . 1023 / A :
1010757806504.
Postal, Paul. 1971. Cross-Over Phenomena. New York, NY: Holt, Rinehart and Winston.
Rackowski, Andrea. 2002. The Structure of Tagalog: Specificity, Voice, and the Distribution
of Argument. MIT dissertation.
Rackowski, Andrea & Norvin Richards. 2005. Phase Edge and Extraction: A Tagalog Case
Study. Linguistic Inquiry 36(4). 565–599. https://doi.org/10.1162/002438905774464368.
Ramchand, Gillian & Peter Svenonius. 2014. Deriving the functional hierarchy. Language
Sciences 46. 152–174. https://doi.org/10.1016/j.langsci.2014.06.013.
Bibliography 371
Remijsen, Bert & Otto Gwado Ayoker. 2018. Forms and functions of the base paradigm of
Shilluk transitive verbs. In Bert Remijsen & Otto Gwado Ayoker (eds.), A descriptive
grammar of Shilluk, 1–80. Honolulu: University of Hawai’i Press.
Reuland, Eric & Tanya Reinhart. 1995. Pronouns, Anaphors and Case. In Studies in Com-
parative Germanic Syntax, 241–268. Springer, Dordrecht. https://doi.org/10.1007/978-
94-015-8416-6_11.
Rezac, Milan. 2008. Phi-agree and theta-related case. In Harbour, David Adger & Susana
Béjar (eds.), Phi theory: Phi-features across modules and interfaces, 83–129. Oxford: Ox-
ford University Press.
Richards, Norvin. 1997. What Moves Where When in Which Language? MIT dissertation.
Richards, Norvin. 2000. Another look at Tagalog subjects. In Ileana Paul, Vivianne
Phillips & Lisa Travis (eds.), Formal Issues in Austronesian Linguistics, 105–116. Dor-
drecht: Kluwer.
Richards, Norvin. 2013. Lardil “Case Stacking” and the Timing of Case Assignment. Syntax
16(1). 42–76. https://doi.org/10.1111/j.1467-9612.2012.00169.x.
Richards, Norvin. 2014. A-bar movement. In Andrew Carnie, Dan Siddiqi & Yosuke Sato
(eds.), The Routledge Handbook of Syntax, 167–191. London: Routledge.
Riesberg, Sonja. 2014. Symmetrical voice and linking in western austronesian languages.
Berlin: Mouton de Gruyter.
Rizzi, Luigi. 1986. On the Status of Subject Clitics in Romance. In Oswaldo Jaeggli & Car-
men Silva-Corvalàn (eds.), Studies in Romance Linguistics 24, 391–420. Berlin: Mouton
de Gruyter. https://doi.org/10.1515/9783110878516-025.
Rizzi, Luigi. 1990. Relativized minimality. Cambridge, MA: MIT Press.
Rizzi, Luigi. 1997. The Fine Structure of the Left Periphery. In Liliane Haegeman (ed.),
Elements of Grammar: Handbook in Generative Syntax, 281–337. Dordrecht: Springer
Netherlands. https://doi.org/10.1007/978-94-011-5420-8_7.
Rizzi, Luigi. 2004a. Locality and left periphery. In Adriana Belletti (ed.), The cartography of
syntactic structures. Structures and beyond. 223–251. Oxford: Oxford University Press.
Rizzi, Luigi. 2004b. The cartography of syntactic structures. The Structure of CP and IP.Vol. 2.
Oxford: Oxford University Press.
Roberts, Ian. 1994. Two types of headmovement in Romance. InDavid Lightfoot &Norbert
Hornstein (eds.), Verb movement, 207–242. Cambridge: Cambridge University Press.
Roberts, Ian. 2010. Agreement and Head Movement: Clitics, Incorporation, and Defective
Goals. Cambridge, MA: MIT Press. https://doi.org/10.7551/mitpress/9780262014304.
001.0001.
Roberts, Ian & Ur Shlonsky. 1996. Pronominal enclisis in VSO languages. In Ian Roberts &
Robert D. Borsley (eds.), The Syntax of the Celtic Languages: A Comparative Perspec-
tive, 171–199. Cambridge: Cambridge University Press. https : / / doi . org / 10 . 1017 /
CBO9780511586279.006.
Rodrigues, Cilene. 2004. Impoverished morphology and A-movement out of case domains.
University of Maryland dissertation.
Ross, John. 1967. Constraints on Variables in Syntax. Boston: MIT dissertation.
372 Bibliography
Ross, Malcolm. 2004a. Demonstratives, local nouns and directionals in Oceanic languages.
In Gunter Senft (ed.), Deixis and demonstratives in Oceanic languages, 175–204. Can-
berra: Pacific Linguistics.
Ross, Malcolm. 2004b. The grammaticization of directional verbs in Oceanic languages. In
Complex Predicates in Oceanic Languages, chap. Complex Predicates in Oceanic Lan-
guages, 297–330. De Gruyter Mouton. https://www.degruyterbrill .com/document/
doi/10.1515/9783110913286.297/html.
Ross, Malcolm & Åshild Næss. 2007. An Oceanic Origin for Äiwoo, the Language of
the Reef Islands? Oceanic Linguistics 46(2). 456–498. http : / /www. jstor .org / stable /
20172324.
Ross, Malcolm, Andrew Pawley &Meredith Osmond. 2016. Introduction. In Malcolm Ross,
Andrew Pawley & Meredith Osmond (eds.), The Lexicon of Proto Oceanic. The culture
and environment of ancestral Oceanic society. Volume 5 — People: body and mind, 1–36.
Canberra: Asia-Pacific Linguistics.
Roversi, Giovanni. 2019. The morphosyntactic structure of the Äiwoo verb. University of
Oslo MA thesis. http://urn.nb.no/URN:NBN:no-72417.
Roversi, Giovanni. 2020. How to satisfy probes: person/number hierarchy effects in Äi-
woo. Proceedings of NELS 50(3). 99–112.
Roversi, Giovanni. 2024a. Condition C and Anti-Cataphora in Äiwoo. Talk presented at
GLOW in Asia 14 (Chinese University of Hong Kong); March 8, 2024.
Roversi, Giovanni. 2024b. Exceptional Ā-extraction in Austronesian informs theories of
voice systems. Talk given at CLS 60 (University of Chicago), April 27, 2024.
Roversi, Giovanni. 2025a. ”Non-local A-movement” is predicted to exist, and it does. Talk
presented at WCCFL43; University of Washington, Seattle.
Roversi, Giovanni. 2025b. Possession and syntactic categories: An argument from Äiwoo.
Natural Language & Linguistic Theory 43(1). 385–420. https://doi.org/10.1007/s11049-
024-09623-7.
Roversi, Giovanni & Åshild Næss. 2019. Jespersen in the Reef Islands: Single versus Bipar-
tite Negation in Äiwoo. Oceanic Linguistics 58(2). 324–352. https://muse.jhu.edu/pub/
5/article/751107.
Royer, Justin. 2022. Elements of (in)definiteness and binding: A Mayan perspective. McGill
University dissertation.
Royer, Justin. 2025. Binding and Anticataphora in Mayan. Linguistic Inquiry 56(2). 247–
310. https://doi.org/10.1162/ling_a_00498.
Rullmann, Hotze & Aili You. 2006. General Number and the Semantics and Pragmatics of
Indefinite Bare Nouns in Mandarin Chinese. In Klaus von Heusinger & Ken Turner
(eds.), Where semantics meets pragmatics, chap. Where Semantics meets Pragmatics,
175–196. Leiden: Brill. https://doi.org/10.1163/9780080462608_012.
Saab, Andrés. 2024. Neither agreement nor pronouns: Clitic doubling andWeak Crossover
in Rioplatense Spanish. Isogloss. Open Journal of Romance Linguistics 10(5). 1–31. https:
//doi.org/10.5565/rev/isogloss.379.
Bibliography 373
Sabbagh, Joseph. 2016. Specificity and objecthood in Tagalog. Journal of Linguistics 52(3).
639–688. http://www.jstor.org/stable/44076669.
Saito, Mamoru. 1992. Long distance scrambling in Japanese. Journal of East Asian Linguis-
tics 1(1). 69–118. https://doi.org/10.1007/BF00129574.
Saito, Mamoru. 2007. Notes on East Asian Argument Ellipsis. Language Research 43. 203–
227.
Sauerland, Uli. 1998. On the making and meaning of chains. MIT dissertation.
Sauerland, Uli. 2000. Two structures for English restrictive relative clauses. Proceedings of
GLOW in Asia 2. 351–366.
Sauerland, Uli. 2003. Unpronounced heads in relative clauses. In Kerstin Schwabe & Su-
sanne Winkler (eds.), The Interfaces: Deriving and interpreting omitted structures, 205–
226. Amsterdam: John Benjamins. https://doi.org/10.1075/la.61.10sau.
Sauerland, Uli, Jan Anderssen & Kazuko Yatsushiro. 2005. The Plural is Semantically Un-
marked. In Stephan Kepser & Marga Reis (eds.), Linguistic Evidence: Empirical, The-
oretical and Computational Perspectives, 413–434. Berlin: Mouton de Gruyter. https :
//doi.org/10.1515/9783110197549.413.
Sauppe, Sebastian, Åshild Næss, Giovanni Roversi, Martin Meyer, Ina Bornkessel-
Schlesewsky & Balthasar Bickel. 2023. An Agent-First Preference in a Patient-First
Language During Sentence Comprehension. Cognitive Science 47(9). 1–36. https://doi.
org/10.1111/cogs.13340.
Schachter, Paul. 1973. Focus and Relativization. Language 49(1). 19–46. https://doi.org/10.
2307/412101.
Schachter, Paul. 1976. The subject in Philippine languages: topic, actor, actor-topic, or
none of the above? In Charles Li (ed.), Subject and Topic, 491–518. New York: Academic.
Schachter, Paul. 1996. The subject in Tagalog: Still none of the above. UCLA Occasional
Papers in Linguistics 15. 1–61.
Schachter, Paul & Fe Otanes. 1972. Tagalog Reference Grammar. Berkeley, CA: University
of California Press.
Scott, Tessa. 2021. Formalizing two types of mixed A/A’ movement. Ms., UC Berkeley.
Scott, Tessa. 2023. Pronouns and agreement in San Juan Atitán Mam. UC Berkeley disser-
tation. https://escholarship.org/uc/item/9vw0v4x5.
Sichel, Ivy. 2001. Studies in the Syntax of Pronouns and Features with Particular Reference
to Hebrew. City University of New York dissertation.
Sichel, Ivy. 2002. Pronoun movement and attraction. Proceedings of the Israeli Association
of Theoretical Linguistics (ITAL) 18.
Sichel, Ivy. 2018. Anatomy of a Counterexample: Extraction from Relative Clauses. Lin-
guistic Inquiry 49(2). 335–378. https://doi.org/10.1162/LING_a_00275.
Sichel, Ivy & Maziar Toosarvandani. 2020. Pronouns and Attraction in Sierra Zapotec. In
Andrew Hedding & Morwenna Hoeks (eds.), Syntax & Semantics At Santa Cruz, vol. 4,
97–116. Santa Cruz, CA: University of California, Santa Cruz.
Sichel, Ivy & Maziar Toosarvandani. 2024. The Featural Life of Nominals. Linguistic In-
quiry. 1–56. https://doi.org/10.1162/ling_a_00517.
374 Bibliography
Sigurðsson, Halldór Ármann. 1991. Icelandic Case-Marked Pro and the Licensing of Lexi-
cal Arguments. Natural Language & Linguistic Theory 9(2). 327–363. http://www.jstor.
org/stable/4047761.
Sigurðsson, Halldór Ármann. 1992. The case of quirky subjects. Working Papers in Scan-
dinavian Syntax 49. 1–26.
Solomon Islands National Statistics Office. 2023. 2019 Population and Housing Census Na-
tional Report. Tech. rep. Honiara, Solomon Islands: Ministry of Finance and Treasury.
Spriggs, Matthew. 1997. The Island Melanesians. Oxford: Blackwell.
Stump, Gregory T. 1984. Agreement vs. Incorporation in Breton. Natural Language & Lin-
guistic Theory 2(3). 289–348. https://doi.org/10.1007/BF00133790.
Sumbatova, Nina R. 2011. Person hierarchies and the problem of person marker origin
in Dargwa: facts and diachronic problems. In Gilles Authier & Timur Maisak (eds.),
Tense, aspect, modality, and finiteness in East Caucasian languages, 131–160. Bochum:
Universitätsverlag Dr. N. Brockmeyer.
Sun, Zhouyi & Peter Grishin. 2025. The ordering of postsyntactic opeerations: Evidence
from Algonquian. Poster at CLS 61, University of Chicago.
Suñer, Margarita. 1988. The role of agreement in clitic-doubled constructions. Natural
Language & Linguistic Theory 6(3). 391–434. https://doi.org/10.1007/BF00133904.
Svenonius, Peter. 2012. Spanning. Ms., University of Tromsø.
Syed, Saurov & Andrew Simpson. 2024. An Argument for Non-Agree-Driven Movement.
Studia Linguistica 78(2). 229–253. https://doi.org/10.1111/stul.12227.
Szabolcsi, Anna. 2009. Overt Nominative Subjects in Infinitival Complements Cross-
Linguistically: Data, Diagnostics, and Preliminary Analyses. In Patricia Irwin & Vi-
oleta Vázquez-Rojas (eds.), NYU Working Papers in Linguistics Volume 2: Papers in Syn-
tax, 1–55. https://philarchive.org/rec/SZAONS-2.
Taghipour, Sahar & Arsalan Kahnemuyipour. 2021. Agreement with Deficient Pronouns
in Laki: A Syntactic Repair to a Clitic Cluster Restriction. Proceedings of WCCFL 38.
417–426.
Takahashi, Daiko. 2008a. Noun phrase ellipsis. In Shigeru Miyagawa & Mamoru Saito
(eds.), The Oxford Handbook of Japanese Linguistics, 394–422. Oxford: Oxford Univer-
sity Press.
Takahashi, Daiko. 2008b. Quantificational Null Objects and Argument Ellipsis. Linguistic
Inquiry 39(2). 307–326. https://doi.org/10.1162/ling.2008.39.2.307.
Takahashi, Shoichi. 2006. Decompositionality and identity. MIT dissertation.
Takahashi, Shoichi & Sarah Hulsey. 2009. Wholesale Late Merger: Beyond the A/Ā Dis-
tinction. Linguistic Inquiry 40(3). 387–426. https://doi.org/10.1162/ling.2009.40.3.387.
Tanaka, Nozomi, William O’Grady, Kamil Deen, Chae-Eun Kim, Ryoko Hattori, Ivan Paul
M Bondoc & Jennifer U Soriano. 2016. An agent advantage in Tagalog relative clause
comprehension. Proceedings of AFLA 22. 191–201.
Thomas, David D. 1955. Three analyses of the Ilocano pronoun system. Word 11. 204–208.
Thráinsonn, Höskuldur. 1979. On complementation in Icelandic. New York: Garland.
Bibliography 375
Tomioka, Satoshi. 2003. The semantics of Japanese null pronouns and its cross-linguistic
implications. In Kerstin Schwabe & Susanne Winkler (eds.), The Interfaces: Deriving
and interpreting omitted structures, 321–339. John Benjamins Publishing Company.
https://doi.org/10.1075/la.61.16tom.
Tomioka, Satoshi. 2014. Remarks on missing arguments in Japanese. Proceedings of FAJL
7. 251–264.
Tomioka, Satoshi. 2021. Japanese -tati and generalized associative plurals. In Patricia
Cabredo Hofherr & Jenny Doetjes (eds.), The Oxford Handbook of Grammatical Num-
ber, 463–484. Oxford: Oxford University Press.
Torrego, Esther. 1995. On the Nature of Clitic Doubling. In Hector Campos & Paula Kem-
pchinsky (eds.), Evolution and Revolution in Linguistic Theory, 251–275. Washington,
D.C.: Georgetown University Press.
Travis, Lisa. 1984. Parameters and effects of word order variation. MIT dissertation.
Travis, Lisa. 2010. Inner Aspect. Dordrecht: Springer Netherlands. https://doi.org/10.1007/
978-90-481-8550-4.
Tryon, Darrel T & Brian D Hackman. 1983. Solomon Islands languages: an internal clas-
sification. Canberra: Department of Linguistics, Research School of Pacific Studies,
Australian National University.
Tyler, Matthew & Itamar Kastner. 2022. Serial verb constructions and the syntax-prosody
interface. Natural Language & Linguistic Theory 40(1). 285–306. https : / /doi .org/10.
1007/s11049-021-09507-0.
Uriagereka, Juan. 1995. Aspects of the Syntax of Clitic Placement in Western Romance.
Linguistic Inquiry 26(1). 79–123. https : / /www. jstor . org . libproxy .mit . edu / stable /
4178889.
Vaa, Anders. 2006. Some Aspects of Referent Mention and Discourse Organisation in Äiwoo.
University of Oslo MA thesis.
Vaa, Anders. 2013. A grammar of Engdewu. Oslo: Universitetet i Oslo dissertation.
van den Berg, René & Brenda H. Boerger. 2011. A Proto-Oceanic Passive? Evidence from
Bola and Natügu. Oceanic Linguistics 50(1). 221–246. https : / /muse. jhu.edu/pub/5/
article/447601.
van Gelderen, Elly. 2000. Bound pronouns and non-local anaphors: The case of Earlier
English. In Zygmunt Frajzyngier & Traci Walker (eds.), Reflexives: Forms and functions.
Volume 1, 187–226. John Benjamins Publishing Company. https://www.degruyterbrill.
com/document/doi/10.1075/tsl.40.08gel/html.
van Urk, Coppe. 2015. A uniform syntax for phrasal movement: A case study for Dinka Bor.
MIT dissertation.
van Urk, Coppe. 2018. Pronoun copying in Dinka Bor and the Copy Theory of Movement.
Natural Language & Linguistic Theory 36(3). 937–990. https://doi.org/10.1007/s11049-
017-9384-x.
van Urk, Coppe. 2020. Successive Cyclicity and the Syntax of Long-Distance Dependen-
cies. Annual Review of Linguistics 6. 111–130.
376 Bibliography
van Urk, Coppe & Norvin Richards. 2015. Two Components of Long-Distance Extraction:
Successive Cyclicity in Dinka. Linguistic Inquiry 46(1). 113–155. https://doi .org/10.
1162/LING_a_00177.
Vander Klok, Jozina. 2022. Revised modal questionnaire for cross-linguistic use. http: / /
www.eva.mpg.de/lingua/tools-at-lingboard/questionnaires.php.
Vergnaud, Jean-Roger. 1974. French Relative Clauses. MIT dissertation.
Vergnaud, Jean-Roger. 1977/2008. Letter to NoamChomsky and Howard Lasnik on ’Filters
and control’. In Robert Freidin, Carlos Peregrin Otero &Maria Luisa Zubizarreta (eds.),
Foundational issues in linguistic theory: Essays in honor of Jean-Roger Vergnaud, 3–15.
Cambridge, MA: MIT Press.
Viken, Christopher. 2024. Clause Structure and Intonation in Äiwoo: The Thetic-Categorical
Distinction in an Oceanic Language. University of Oslo MA thesis. https://www.duo.
uio.no/handle/10852/116476.
Vikner, Sten. 2017. Object Shift in Scandinavian. In The Wiley Blackwell Companion to
Syntax, Second Edition, 1–60. John Wiley & Sons, Ltd. https : / / doi . org / 10 . 1002 /
9781118358733.wbsyncom114.
Webelhuth, Gert. 1989. Syntactic saturation phenomena and the modern Germanic lan-
guages. UMass Amherst dissertation.
Weisser, Philipp. 2014. Mismatch verbs: A unified account of unaccusatives and depo-
nents. In Franz Rainer, Francesco Gardani, Hans Christian Luschützky & Wolfgang
U. Dressler (eds.), Morphology and Meaning: Selected papers from the 15th Interna-
tional Morphology Meeting, Vienna, February 2012, 315–330. John Benjamins Publish-
ing Company. https://www.degruyterbrill.com/document/doi/10.1075/cilt.327.22wei/
html.
Woolford, Ellen. 2008. Active-stative agreement in Lakota: Person and number alignment
and portmanteau formation. Ms., UMass Amherst.
Wu, Yvette Yi-Chi. 2024. Affix ordering in Seediq and other Formosan languages. Ms.,
Harvard University.
Wu, Yvette Yi-Chi, Tamisha L. Tan & Giovanni Roversi. 2023. Syntactic vs. morphologi-
cal verbal concord across Austronesian. In Vera Hohaus, Jens Hopperdietzel & Siena
Weingartz (eds.), Proceedings of TripleAFLA: 9th TripleA workshop for semantic field-
workers; 29th annual meeting of the Austronesian Formal Linguistics Association. 104–
118.
Wurm, Stephen Adolphe. 1972. Notes on the indication of possession with nouns in Reef
and Santa Cruz Islands languages. In Papers in linguistics of melanesia 3 (series a-35),
85–113. Canberra: Pacific Linguistics.
Wurm, Stephen Adolphe. 1978. Reefs–Santa cruz: Austronesian, but ... ! In Lois Carring-
ton & Stephen Adolphe Wurm (eds.), Second International Conference on Austronesian
Linguistics: Proceedings, 969–1010. Canberra: Pacific Linguistics.
Wurm, Stephen Adolphe. 1981a. Notes on nominal classification systems in Äiwo, Reef
Islands, Solomon Islands. In Andrew Gonzalez & David D Thomas (eds.), Linguistics
Bibliography 377
across continents: Studies in honor of richard s. Pittman, 123–142. Manila: Summer In-
stitute of Linguistics and Linguistic Society of the Philippines.
Wurm, Stephen Adolphe. 1981b. The possessive class systems in Äiwo, Reef Islands,
Solomon Islands. Papers in New Guinea linguistics 21. 181–209.
Wurm, Stephen Adolphe. 1982. The Papuan languages of Oceania. Tübingen: Günter Narr.
Wurm, Stephen Adolphe. 1991. Language decay and revivalism: The Äiŷwoo language
of the Reef Islands, Santa Cruz archipelago, Solomon Islands. In Robert Blust (ed.),
Currents in Pacific linguistics: Papers on Austronesian languages and ethnolinguistics in
honour of George W. Grace (Series C-119), 551–560. Canberra: Pacific Linguistics.
Wurmbrand, Susanne. 2016. Complex Predicate Formation via Voice Incorporation. In Léa
Nash & Pollet Samvellian (eds.), 48–290. Leiden: Brill. https://brill-com.libproxy.mit.
edu/display/book/edcoll/9789004307094/B9789004307094_010.xml.
Wurmbrand, Susi & Koji Shimamura. 2017. The features of the voice domain: actives,
passives, and restructuring. In Roberta D’Alessandro, Irene Franco & Ángel J. Gallego
(eds.), The Verbal Domain, 179–204. Oxford: Oxford University Press. https://doi.org/
10.1093/oso/9780198767886.003.0008.
Yip, Moira, Joan Maling & Ray Jackendoff. 1987. Case in tiers. Language 63. 217–250.
Yuan, Michelle. 2018. Dimensions of ergativity in Inuit: theory and microvariation. MIT
Thesis. https://dspace.mit.edu/handle/1721.1/122054.
Yuan, Michelle. 2020. Dependent case and clitic dissimilation in Yimas. Natural Language
& Linguistic Theory 38(3). 937–985. http://www.jstor.org/stable/48741768.
Yuan, Michelle. 2021. Diagnosing Object Agreement vs. Clitic Doubling: An Inuit Case
Study. Linguistic Inquiry 52(1). 153–179. https://doi.org/10.1162/ling_a_00366.
Yuan, Michelle. 2022. Ergativity and object movement across Inuit. Language 98(3). 510–
551. https://muse.jhu.edu/pub/24/article/864634.
Yuan, Michelle. 2024. More on ’Lexical DP Blocking’ effects in the PCC: Evidence from
Mixtec. Proceedings of NELS 54. 1–10.
Yuan, Michelle. 2025. Morphological conditions on movement-chain resolution. Natural
Language & Linguistic Theory 43(1). 509–562. https://doi .org/10.1007/s11049- 024-
09620-w.
Zwicky, Arnold M. & Geoffrey K. Pullum. 1983. Cliticization vs Inflection: English N’T.
Language 59(3).
Zyman, Erik. 2023. Raising out of Finite Clauses (Hyperraising). Annual Review of Linguis-
tics 9. 29–48. https://doi.org/10.1146/annurev-linguistics-022421-070658.