SYLLABLE-BASED GENERALIZATIONS IN ENGLISH PHONOLOGY
by
DANIEL KAHN
B.S., City College of the
City University of New York
1968
SUBMITTED IN PARTIAL FULFILLMENT
OF THE REQUIREMENTS FOR THE
DEGREE OF DOCTOR OF
PHILOSOPHY
at the
MASSACHUSETTS INSTITUTE OF
TECHNOLOGY
September, 1976
Signature of Author .................................................
Department of Linguistics and Philosophy,
August 9, 1976
Certified by .- t.....................
Thesis Supervisor
Accepted by .................... ..........................
Chairman, Departmental Committee
HUM. on Graduate Students
AUG 20 1976
SYLLABLE-BASED GENERALIZATIONS IN ENGLISH PHONOLOGY
by
Daniel Kahn
Submitted to the Department of Linguistics and Philosophy
on Augurt 9, 1976 in partial fulfillment of the require-
ments for the degree of Doctor of Philosophy.
ABSTRACT
This dissertation is concerned with the role of the syllable in gen-
erative grammar. I argue that the syllable is a necessary element in
phonological descriptions by identifying aspects of phonology that seem
to call for analyses in terms of syllabic structure and demonstrating the
superiority of syllabic analyses over possible alternative solutions.
The largest part of the dissertation, Chapter II, is a study of pho-
nological rules of English which have syllable-structure conditioning.
I attempt to show, for example, that voiceless stops are aspirated when
simultaneously syllable-initial and non-syllable-final. Much emphasis
is placed on the role of syllable structure in conditioning phenomena ob-
served at word juncture which are apparently sensitive to rate of speech,
for example, the treatment of final /t/ in American English and rules of
/r/-epenthesis and deletion in various American and British dialects.
Chapter I sets the stage for the investigations of Chapter II by
discussing in general terms how it is possible to justify the use of the
notion "syllable" in linguistic desciptions, by suggesting and defending
certain basic hypotheses concerning syllabic structure, as well as by de-
ducing a set of syllable-structure assignment rules for English. An im-
portant claim of Chapter I is that one must recognize "ambisyllabic" con-
sonants on the phonetic level. The phonological implications of ambisyl-
labicity are seen in Chapter II.
The word "generalizations" in the title of the dissertation is a cov-
er term for (a) phonological rules, and (b) constraints on lexical repre-
sentations. It is often apparent that a given phenomenon is syllable-re-
lated (as I show, for example, with rega d to the unusual distribution of
[q(g)] in English) without it being clear whether what is involved is a
phonological rule (perhaps /og/ simplifies to /Q/ syllable-finally) or a
constraint on lexical representations (perhaps syllable-final /ng/ is
barred from underlying representations). In Chapter III, I approach this
problem in a general way, present evidence for "constraint" analyses of
several phenomena generally taken to be rule-based, and put forth a gen-
eral hypothesis about the ordering of syllable-sensitive rules.
Thesis Supervisor: Paul Kiparsky
Title: Professor of Linguistics
-2-
0CD
CD
4ACKNOWLEDGEMENTS
In the mid '50s, when the Brooklyn Dodgers were the dominant club
in the National League, it impressed me greatly that while other teams
had perhaps a great firstbaseman or a heroic centerfielder, the Dodgers
had a star at every position. Like the Dodgers of my boyhood, the M.I.T.
linguistics faculty is a team of heroes, and what competence I have gained
in the practice of linguistics is due in large measure to the quality of
my teachers, Noam Chomsky, Ken Hale, Morris Halle, Paul Kiparsky, David
Perlmutter, and Haj Ross.
My thesis committee consisted of Professors Kiparsky (supervisor),
Halle, and Ross. I am glad to acknowledge that whatever the quality of
this thesis, it is far better than it would have been had any one of
these three not shown the interest and provided the nelp that he did.
In my studies in phonology, I worked most closely with Professor
Kiparsky, whose depth of understanding of matters phonological is, it
seems to me, unequalled in the field. I am very grateful to him for his
guidance, his suggestions, his friendliness, and for the seriousness with
which he greeted my ideas.
In my four years at M.I.T., I spent many hours in Morris Halle's
office, discussing (mostly arguing about) the theory of phonology. I
have come away with a tremendous amount of respect for Professor Halle,
not only because, like the great men of other sciences, he was the first
to understand many of the concepts we generative phonologists now recog-
nize as basic truths, but also because of his excellence as a teacher
and his easy availability to his students.
5I regret not having spent more time discussing the topics of this
thesis with Haj Ross, for his extensive comments on a first draft of
the thesis proved extremely valuable to me. Haj may conside- himself
to be first and foremost (cf. *foremost and first) a syntactician, but
I would be proud to someday become as fine a phonologist as he is.
My fellow students provided a good deal of help, both of a theore-
tical nature and in their patience in the face of my tiresome questions
about phonemic distinctions and phonetic detail in their native dialects.
I am very appreciative of this assistance.
In short, I must acknowledge a good deal of help in the preparation
of this thesis. Nevertheless, it would be a mistake for the reader to
associate any of the blatant errors and other bits of foolishness he will
surely encounter here with anyone else but me.
TABLE OF CONTENTS
Introduction. , .*.. ...... .. .. . . .. o.a.9..9.a. v, , . ..9* o*e. * * .a. . . .. ..... 8
Chapter I - Preliminaries.........................................20
Introduction....................................................20
Section 1 Justifying Syllabic Phonology....
...
....
. . . . . . 20
Section 2 - Phonetic Descriptions and the Syllable..............27
Section 3-Ambisyllabicity....... ............
Section 4 - Phonetic Syllabication: Formalism................ ...35
Section 5 - Syllable-Structure Assignment Rules for English.....39
Section 6 - The Syllable and Phonotactic Constraints............56
Footnotes ................... ............. . . . . . . . - - . -59
Chapter II - Syllable-Structure Conditioning in Phonological Rules...68
Introduction........* ... ... * a.. .... a.0.............. ..... 68
Section 1 - The Phonological Development of Underlying /t/......68
1.1 -Aspiration....................................69
1.2 - Unreleased Stops and Glottalization...........79
1.3 - Summary and Extension Beyond Word Level.......86
1.4 - Underlying Representation of Voiceless Stops..91
1.5 - "Flapped" /t/.... ................. 91
1.6 -"Sonorization".............................98
1.7 - The Alveopalatal Allophones..................104
Section 2 - fr/-Loss........................ .......... 106
Section 3 - Pre-/r/ Vowel Quality..............................116
Section 4 - Glide Sequences............................120
7Section 5 - Raising of /5/...p...........................123
Section 6 - Simplification of/nyu/.................
Section7 -Simplification of /og/.............................130
Section 8-StressPlacement..................................134
Section 9 -Halle's Rule-Writing Convention....................142
Appendix 1 The Distinctive-Feature Description of Engl. /r/,.148
Appendix 2 A Note on the Phonetics of "Flapped" /t/..........151
Appendix 3 More on pre-/r/ Vowel Quality.....................155
Footnotes.............................................160
Chapter III - On Word-Based Generalizations with Syllabic Condi-
tioning..............................183
Section 1 - A Hypothesis................................183
Section 2 - /mr/....a......................................187
Section 3 - Short Non-Low Pre-/r/ Vowels.......................190
Section 4 - Raising of /a....................
Section 5 - The Velar Nasal..................................197
Section 6 - Summary........................................205
Footnotesgay ...........................................-.. 2C6
Bibl1iography........................................................2110 00 &0 00 00 00 00 00 00 02 1
8
INTRODUCTION
Chomsky and Halle's Sound Pattern of English is a monumental work,
not only in its establishment of a theoretical framework in which to in-
vestigate the sound systems of natural languages, but also in its appli-
cation of the theory to the analysis of a particular natural language,
English. Yet in spite of the thoroughness with which the phonology of
English is investigated, there appears to remain a class of important
phonological processes in English which are not discussed in SPE. These
processes are significant both in the major contribution they make to
the determination of the phonetic form of utterances and in the theore-
tical interest of the formalism they require.
The processes I have in mind, of which the voicing and "flapping"
of /t/ in American English in certain environments is an example, tend
to be low-level, i.e., at the phonetic end of the phonological spectrum,
productive, i.e., freely applicable to new and hypothetical forms which
happen to meet these rules' structural descriptions, and, I would claim,
sensitive to the syllabic structure of input strings. In the remainder
of this introduction I attempt to explain why such processes are of inter-
est to the generative grammarian.
Part of the foundation of the generative approach to the syntax of
natural languages consists of the simple but highly significant observa-
tion that the class of sentences on which the native speaker of a lan-
guage can make grammaticality judgements is unbounded in spite of the fact
that the learned and inherent knowledge of the speaker is surely finite.
The job of the syntactician is thus in principle a well-defined and
9achievable one: to discover the underlying system of rules and conditions
on rules that allows the generation of the unbounded class of sentences.
To the extent that the postulated system accurately reflects the
knowledge of the native speaker, we are justified in saying that it is
a correct model, that it has "psychologIceI reality." Like the models
produced by the other sciences, this linguistic competence model is
based on observed facts and subject to empirical testing. But as in the
other sciences, a major difficulty in the construction of linguistic mod-
els is the largely intuitive task of deciding which empirical data serve
to reveal the underlying system and which involve distortions of perhaps
unknown origin and can safely be excluded from immediate attention. Pro-
gress in any science is hampered under a methodology that insists that
all "facts" are equally deserving of immediate explanation. Newton's
laws of motion and law of gravitation, to take an example from physics,
have proved to be insights of great significance and generality, and yet
with regard to the simple test of letting fall pairs of objects of dif-
ferent densities, these principles account for the"facts" only in the
highly unnatural setting of the vacuum chamber. An understanding of aero-
dynamics eventually allows one to account for the way in which objects
fall in air, but one's acceptance of Newton's principles need not be
postponed until aerodynamics is understood, and in fact could not be,
because of the dependence of the latter on the former.
A good deal of current controversy among linguists is due to differ-
ing intuitions as to what facts should have first claim on our attention,
and yet in syntax there is probably universal agreement that a certain
type of observation is of very limited interest. Consider the commonly-
10
heard sentences (1)-(2) in the light of current analyses of negation and
question-formation in English:
(1) I kid you not!
(2) How goes it?
We would certainly reject these aentences as evidence on the basis of
which to modify our analyses. The reason for this is clearly the lack
of productivity of these constructions:
(1') *She understands us not
(2') *How went it in Europe last summer?
Sentences (l)-(2) may be interesting from a cultural point cf view in that
they reveal on the part of present-day speakers a certain familiarity
with an archaic form of the language, and an appreciation of anachronism.
But we realize that considering (1)-(2) will only lead us astray if we
want to produce a model of the linguistic competence of contemporary
speakers of English, and this realization stems directly from the non-
productive nature of the rules that would generate them. At best, sen-
tences like these indicate that any analysis will be left with a finite
residue of isolated sentences.
My purpose in making the above observations is to contrast the gen-
erative approach to syntax with the generative approach to phonology.
Generative phonologists have been concerned in large part with phenomena
evidenced in only finitely many forms, and phenomena for which the "pro-
ductivity"test often fails.
For example, one postulates a Tri-syllabic Laxing Rule for English
11
because for most words the generalization holds that vowels three or more
syllables from the end of the word are lax, even when the corresponding
vowel in related forms not meeting the "tri-syllabic" condition are tense
(for ex., sanity vs. sane). Nonetheless, (a) there exist glaring excep-
tions to this rule (rarity, obesity, etc.); (b) the language expands
its exicon as though the rule didn't exist (Idaho, Omaha, etc.);
(c) the exceptional forms show no tendency to regularize (one would be
vary much surprised to find a dialect in which Idaho is pronounced
[LDahow]); (d) there is no native-speaker intuition that words like
Idaho, obesity are in any way deviant; and (e) there is no native-speak-
er intuition that a hypothetical form like [owkeser], in which the Tri-
syllabic Laxing generalization has been ignored, is any less in accord-
ance with the general principles of English pronunciation than one like
[akeser], which obeys the generalization.
Is it correct for a science whose goal is to construct a model of
psychological competence to be concerned with generalizations such as
the Tri-syllabic Laxing Rule? Note that the mere fact that the generali-
zation holds for the majority of English words is insufficient reason to
assume that the rule is psychologically real. The majority of English
words are less than nineteen segments long, but this latter generaliza-
tion surely has no direct psychological representation. The origin of
the particular arrangement of the data captured in the Tri-syllabic Lax-
ing Rule is a historical rule, similar in form to the hypothesized syn-
chronic rule, which was added to the phonology at a certain point in the
history of the English language. What is at issue is whether the rule
is still operative in its original form.2
12
Without attempting to settle this issue here, let us contrast the
generalization about the tenseness of vowels three or more syllables from
the end of the word with the observation that English /p, t, k/ are as-
pirated in # -- V C0 # and unaspirated in # s -- V C #3
The latter generalization is statistically correct, as is the one
captured by the Tri-syllabic Laxing Rule. However, (a) there are simply
no exceptions to the stated generalization about aspiration, and (b) new
words coming into the language always obey the English aspiration rule,
even if they are not in accordance with it in the source language. Most
importantly, (c) there is strong native-speaker intuition that forms like
[th Ls] and [stLs] are possible, though non-occurring, words of English,
while [tLs] and [sthLs] are felt to be strongly deviant.
I hasten to point out that it would be naive to assume that the only
psychologically-real generalizations in phonology are those for which this
kind of direct evidence is forthcoming. On the other hand it is certain-
ly foolhardy to assume that every generalization that the linguist can
capture will be captured by the native speaker as well.
A survey of the literature of generative phonology makes it clear
that the questionable status of Tri-syllabic Laxing is shared by many or
most proposed phonological rules. One might trace the origin of this state
of affairs in the following way.
Work in linguistics in this century has been characterized by a
clear-cut separation of diachronic and synchronic investigations. Syn-
chronic studies involve analyses of phenomena obseved for a particular
point in the history of a language, without regard for the historical or-
igin of these phenomena or the correct analysis of them at some other
13
stage in the history of the language. But what sorts of phenomena are
of interest to the synchronic investigator? The answer depends on the
goal of the particular linguist. In pce-generative frameworks, one was
interested in an analysis of the language, not in a model of the psycho-
logical constructs accounting for the competence of a speaker of the
language. For example, Trager and Smith view as an advantage of their
analysis (1951) of the vowel system of English that it is adequate for
the description of all dialects of American English known to them. Now
surely it is irrelevant to the linguistic development of the child growing
up in Los Angeles that certain East Coast dialects have much more complex
vowel systems than the dialect of his own linguistic community, so it is
clear that Trager and Smith are analyzing an abstract entity, "American
English," rather than the competence of particular speakers of American
English. Similarly, Harris(1951) says that it makes no difference whether
in English one assigns the phonetic segments [phI and [p] to the phoneme
/p/ and [b) to /b/, or [phj to /p/ and [p] and [b] to /b/, as long as one
states accurately the appropriate allophonic rules (p. 63). Clearly Har-
ris is giving instructions for describing the language. He does not say,
"the distribution of the phonetic segments [p), [ph], and [b] in English
is such that we cannot be sure whether speakers analyze [p] as deriving
from /p/ or /b/."
Generative grammarians, on the other hand, are interested in pro-
ducing psychologically-accurate models of speaker competence. This fact
limits the range of phenomena of interest in synchronic investigations
to those "known" by the native speaker, giving rise to the following no-
tational problem. What is to be said about generalizations we have no
14
reason to believe to be part of the competence of the speaker but are
nonetheless "facts" about the current stage of the language under investi-
gation? Are they part of the synchronic description of the language?
The answer is clearly no if we take seriously the assumptions of the gen-
erative approach to linguistics. For example, what is the status of
Grimm's Law in English? Grimm's Law is certainly a diachronic rule.
But it is also a fact about modern English that it displays the Grimm's
Law alternations in many pairs of words, for example, father/paternal.
Is this spirantization process therefore also a synchronic rule? Or,
to take a different sort of example, is the fact that the English mor-
pheme has on the average n segments, where n is in principle easily de-
terminable, a synchronic fact?
There is no paradox here. The answer is that these are facts about
modern English but are not part of a synchronic generative phonological
analysis of English because they appear to play no role in the native
speaker's internal organization of his knowledge about th sound system
of his language. Now it is easy to exclude facts such as these from
consideration but one is reluctant to discard too much of what has tra-
ditionally been considered phonology. Nevertheless it is possible that
Tri-syllabic Laxing and many other of the rules discussed in the litera-
ture of generative phonology represent correct "facts'about modern Eng-
lish" but are not part of a "synchronic analysis" in the psychological
sense that generative grammarians attach to the term.
The way in which the traditional goal of sychronic investigations,
that of analyzing the "language," and the goal of the generative approach,
to produce a model of speaker competence, come together in generative
15
phonological analyses and give rise to the problem discussed above can
be seen clearly in the following pair of quotes from SPE. Consider first
this definition of grammar:
we use the term "grammar" to refer both to the system of rules
represented in the mind of the speaker-hearer, a system which
is normally acquired in early childhood and used in the production
and interpretation of utterances, and to the theory that the
linguist constructs as a hypothesis concerning the actual interna-
lized grammar of the speaker-hearer. (p. 4)
I have underlined portions of the quote to emphasize the fact that genera-
tive phonologists are indeed interested in modeling the actual psychologi-
cal constructs which account for the speaker's competence.
Next consider this passage from SPE:
We see no reason to give up rules of great generality because
they are not of even greater generality, to sacrifice generality
where it can be attained. It seems hardly necessary to stress
that if we are faced with the choice between a grammar G that
contains a general rule along with certain special rules govern-
ing exceptions and a grammar G2 that gives up the general rule
and lists everything as an exception, then we will prefer G1 .
(p. ix)
It is not clear how one can be certain that, as opposed to the linguist,
the language learner will not prefer G2. On the other hand, in a non-
psychological approach, where one is merely interested in the most com-
pact description of the facts of the language, the policy outlined in this
16
quote is clearly optimal. Thus this insistence on capturing as many gen-
eralizations as possible seems to be a hold-over from earlier approaches
to synchronic analysis and to be at odds wit!, goals expressed in the first
quote. For how can we make the very strong claim that we are modeling
the actual internalized system of rules if take as evidence for the na-
ture of the system any generalization we can find in the data? Only,
it would seem, by making the unsupported assumption that any generaliza-
tion noticed by the linguist will be noticed by the language learner as
well. ,
As pointed out above, we never make this assumption in syntax. Hy-
pothesized rules must be supported by native-speaker confirmation of pro-
ductivity. In support of the way in which generative phonology is done,
one might point out that the nature of phonological phenomena is such that
this kind of direct evidence is seldom avaliable. But surely this obser-
vation cannot be used to condone the formulation of strong hypotheses on
the basis of weak evidence. Rather, the observation, to the extent that
that it is a correct one, suggests that analytic tools more powerful than
generalization-capturing are necessary.
But one might also question the validity of the observation. There
certainly does exist a large class of phonological phenomena which can
be directly tested. In view of the questionable status of phonological
rules for which no direct evidence exists, and insofar as substantive
arguments against such rules are not forthcoming, it has seemed to me
that a productive course of action might be to limit one's attention
first to rules whose status is not in doubt, attempt to reach general con-
clusions regarding form and applicability, and only later, armed with
17
these observations, return to the study of the more questionable generali-
zations.
I have embarked upon such an investigation, with various dialects
of English as the objects of study, and report on my findings in this
dissertation. I mention here two very general results.
The first is that one need not fear that by restricting oneself to a
study of directly-confirmable phonological processes one will be hard-
pressed for material. As in syntax, the typical situation is that as one
proceeds, interesting new problems arise at a faster rate than old ones
are solved. Nor are the solutions to such problems typically transparent.
It is a mistake to use the term "low-level rule" disparagingly, as is
often done, as though the expression "low-level" described these pro-
cesses' inherent degree of linguistic interest and complexity.5
Secondly, the importance of the syllable in the description of low-
level processes quickly becomes obvious. I set out to write a disserta-
tion on productive phonological processes and soon found myself hampered
by the absence of a generative theory of syllabication.
Thus, in its final form, this dissertation represents an attempt to
outline a theory of the syllable and to apply it in the description of
certain generalizations which are clearly a part of native-speaker pho-
nological competence.
18
Footnotes to Introduction
1. rare, phonetically [rer], is from underlying /riir/, according to the
analysis of SPE. Upon suffixation and Laxing we expect [rarLDiJ, but
in fact observe [rertDi], as though Laxing has failed to apply. In some
American dialects, all /M/ in - r V has become [eJ; what is of inter-
est about the example rarity is that even in dialects in which Harry,
marry [a] are always distinct from hairy, Mary [e), rarity has [e].
See Appendix 3 of Chapter II for more discussion of dialectal variation
in pre-/r/ vowels.
2. In doubting the psychological reality of Tri-syllabic Laxing as a
phonological rule of English, one need not take the extreme position
that sane and sanity, for example, bear no more than a suppletive rela-
tionship to each other. One possible alternative is a morphological
rule which laxes vowels (or directly effects /ey/-*/ai/, /iy/-+/E/,
/ay/-i/t/) before a certain class of affixes. Evidence for this analy-
sis over the phonological-rule account is available in (a) the large
number of exceptions to the phonological generalization (Idaho, etc.)
(see also points (b)-(e) in the text, above), and (b) the existence of
monosyllabic affixes, in particular -ic, which lax preceding vowels with
the same regularity as bisyllabic -ity.
3. See Section 1.1 of Chapter 1I for a full discussion of aspiration in
English.
4. As alternative means of getting at the form of synchronic grammars,
I have in mind the evidence available from phonological change (Kiparsky,
19
1968a, 1968b), from studies of meter (Kiparsky, 1973), from "slips of the
tongue" (Fromkin, 1971; Stampe, 1972), from word games and secret lan-
guages (SPE:342-3; Hombert, 1973); for more discussion, see Zwicky (1974).
5. Low-level processes and fast-speech rules are often though of togeth-
er. While it is correct that fast-speech rules tend to be low-level pro-
cessee, the converse is not true. Essentially all of the processes I dis-
cuss in this dissertation are obligatory at normal rates of speech.
20
CHAPTER I
PRELIMINARIES
Introduction
My general goal in this chapter and the next is to provide a formal
basis for two assumptions which have been pervasive in traditional dis-
cussions of phonology but which have by and large not been adopted by
generative phonologists. These assumptions are (a) that there exists,
on the phonetic level, a well-defined unit of perception and production
larger than the segment and smaller than the word, and (b) that this unit
plays a very significant role in conditioning distributional statements,
sound changes, synchronic phonological rules, etc., i.e., that it is of
general phonological significance. The unit is of course the syllable.1
In Chapter II, I discuss a large number of phonological processes of
English, the statements of which become simpler, intuitively more meaningful,
and, most importantly, descriptively more adequate, if use is made of a
set of rules of syllable structure assignment ordered prior to the phono-
gical rules proper. This chapter provides the necessary preparation by dis-
cussing the general question of how the claim that the syllable is a phono-
logically significant construct can be justified as well as by suggesting
and defending certain general hypotheses concerning syllable structure and
proposing a set of rules for assigning syllable structure to strings of
segments in English.
Section 1 - Justifying Syllabic Phonology
1.1 It may prove helpful to approach the problem of how one might
go about justifying the intuition that the syllable has an important pho-
21
nological role to play by first considering an analogy. Let us evaluate
several possible solutions to a well-known problem of Sanskrit phonology,
that of accounting for the class of segments /r, u, k, i/ as the leftward
environment of a retroflexion rule. Assuming that this class of segments
is somewhat unnatural when expressed in terms of a given distinctive-fea-
ture framework, what sort of observation could eliminate the problematical
status of the "ruki" rule?
Suppose it were suggested that there exists some previously undis-
covered feature, [+F J, such that /r, u, k, i/ are [+F J and all other
segments [-Fxi, or that Sanskrit contained a rule inserting the hitherto
unknown boundary Q in the environment /r, u, k, i/ - , and that the
correct environment for, the ruki rule is simply [+Fx] or C - . Obvi-
ously these proposals are of absolutely no theoretical value as solutions
to the problem under consideration and represent no more than their pro-
poser's prejudice as to where a solution will eventually be found. In
the case of the new-feature "solution," no evidence has been given for the
appropriateness of the feature beyond its use in simplifying the very rule
we wish to understand, while the new-boundary "solution" eliminates a pro-
blematical retroflexion rule environment at the cost of replacing it with
an equally problematical boundary-insertion rule environment.
Suppose, however, that the proposer of either of these solutions
could go further and show (contrary to fact, as far as I know) that [F j
or as defined above could be used to condition not only the ruki rule,
but several other rules of Sanskrit phonology as well. How would we re-
evaluate the claims under consideration?
It is clear that in contrast to the earlier situation we would now
22
have to acknowledge that a discovery had been made - the discovery of a
relationship between rules not previously recognized and of a probable
need for modification of the theory, since by hypothesis /r, u, k, i/ is
an unnatural class. However the situation remains the same with regard
to the particular solutions put forward: wc have been offered no basis on
which to select one and not the other, or either rather than some third
possibility.
In order to make a truly adequate case for either of the proposed
solutions, its defender must go beyond the realm of phonological rules
and establish a link to other phenomena relevant to the type of solution
being proposed. For the new-feature solution, what would be required is
the existence of particular articulatory or acoustic/perceptual properties
corresponding to the proposed feature, a demonstration that the feature
properly distinguished between /r, u, k, i/ and all other segments of the
language, and evidence that the feature is appropriate in the phonological
descriptions of other languages. In the case of the new-boundary solution,
we would demand either the existence of some sort of special phonetic
juncture at the points where r is found or else a demonstration that the
points of Q-insertion are well-defined from the point of view of the syn-
tax of the language.
Keeping this discussion in mind, let us turn from the problem of the
Sanskrit ruki rule to a problem of the type for which I would support a
syllable-based solution.2
1.2 In the so-,called r-less dialects of English, words like car,
cart, cartel, but not rack and carry, are pronounced without phonetic [r].
The environments in which [r] is missing, from the point of view of ety-
23
mology and the r-retaining dialects, is that given in (1).
(1) U)
This environment is problematical because the class of consonants (abbre-
viated by C in (1)) and word-boundary (#) do not form a natural class -
in fact they have in common no features at all. (The theoretical frame-
work assumed here is that of Chomsky & Halle, 1968.)
Regarding this problem, the hypothetical discovery discussed in the
second part of the ruki analogy has already been made: /r/-loss is not
an isolated example of the occurrence of the strange environment of (1).
As has often been noted (for example, by Lightner, 1972:333), rules of
this type are quite common in the world's languages; I offer detailed dis-
cussions of several such rules in English, including /r/-loss, in Chapter
II. We have thus uncovered a problem for the theory of phonology.
As was the case with ruki, there are two obvious directions in which
one might look in order to understand rules with environments such as (1).
One might try to modify the distinctive feature system in such a way that
consonant and word-boundary would constitute a natural class. Or one
might look into the possibility that some sort of boundary is typically
found immediately to the left of these units. Suggestions of both of
these types have in fact been made, and in the remainder of this section
I show how such proposals have failed to meet the level of adequacy out-
lined at the end of the ruki discussion.
1.3 Among others, Lass (1971) and Lightner (1972) have suggested
that boundaries be specifiable for features normally taken to be reserved
24
for segmental units. In particular, they would expand the feature speci-
fication of word-boundary in such a way that it and the class of obstru-
ents would form a natural class. I find this proposal unacceptable for
two major reasons.
First of all, there is no hope of independently justifying the seg-
mental features imposed on word-boundary in terms of articulatory or acou-
stic considerations. This is so because word-boundary is simply not arti-
culated and is not present in the acoustic signal. Lightner (1972) tries
to associate word-boundary with silence (p. 332), but of course this cor-
relation fails in general: there is no cessation of vocal cord vibration
and sorid output at the points of sentence-internal word-boundary in the
phrase John#is#a#bum as normally spoken.
Thus segmental feature specifications would have to be assigned to
word-boundary purely on the basis of maximal simplification of phonologi-
cal rules. That this is indeed Lightner's tactic in spite of his attempt
to describe the articulation of word-boundary is made clear in the follow-
ing quote (p. 334), which is typical of several statements he makes: "In
many languages glottalized consonants lose their glottal quality before
non-glottalized consonants and word-finally. Apparently # must be speci-
fied [-glottal]." It should be clear that Lightner has observed that many
C unglottalized cons.
languages have and - (rules, not
# #
explained why they do. Similarly, Lass' (1971) only motivation for view-
ing word-boundary as an obstruent is that it allows the Pimpiification of
certain rules (pp. 24, 27).
25
It might be countered that if it were possible to specify for word-
boundary a single value of each of the distinctive features in such a way
that this set of specifications were of universal phonological usefulness,
a major theoretical contributior4 would be involved even if no explanation
external to phonological rule-systems could be given as to why word-bound-
ary takes on these particular specifications. However, the hope of deter-
mining a unique universal set of segmental feature specifications for word-
boundary seems unrealizable. In English alone, for example, there exist
both [C, #1 and [V, #] rules; an example of the latter is a rule which
tenses vowels in -- [V, #. (See also Halle, 1971.)
A second major objection to specifying word-boundary for segmental
features is that boundaries do not participate in phonological processes
the way segments do. In particular they do not occur in focus position
in rules. Having boundaries bear segmental features weakens the theory
by implying that segments and boundaries are phonologically on a par and
opening up the unrealized possibility of rules which change the segmental
feature specifications on boundaries.
Furthermore if word-boundary, which is allegedly [-syllabic], occurred
in an environment in which glides becomes vowels, would its specification
be changed to [+syllabic] and would it henceforth behave as a vowel rather
than as a non-syllabic? Presumably not, leading to the need for compli-
cation of a general vocalization rule to exclude word-boundary.
I conclude that the proposal that word-boundary is specified for
segmental features is untenable and that we must look elsewhere for an
account of the naturalnecs of rules like /r/-loss.
26
1.4 I have argued that a particular common-feature type solution to
the problem of accounting for rules involving [C, #] is inadequate and
that in general no such solution could be correct. In turning to common-
boundary solutions, in particular solutions in terms of syllable boundary,
I argue only that past treatments of this type have been inadequate, for
I believe that syllable structure is the key to understanding the [C, #]
rules as well as other phonological processes. I exemplify the basic
failing of certain analyses that have sought to simplify rules by refer-
ence to syllable boundary by briefly reviewing a paper of Hoard's.
Hoard (1971) discusses the environments for aspiration, consonantal
tenseness and certain types of feature assimilation in English and claims
that a general "syllabication" rule ordered before these segmental pro-
cesses allows the statement of them to be simplified. Hoard's syllabica-
tion rule is (2) (his (3)), where /./ represents syllable boundary and M
a* "maximal initial cluster."
(2) 0.-*/./ in env. VC0
L +stress>
Thus in a string /V1C... . C V2/, all the consonants are part of V's syl-
lable if V2 is unstressed; otherwise as many of the rightmost consonants
as is consistent with the constraints on initial clusters in English will
be syllabified with V2 , the remainder with V1 .
Although Hoard shows that the various segmental rules, as he deszribes
them, can be given simple statements in terms of this syllable boundary,
it is significant that he never discusses the syllabication that results
from (2) in terms of phonetic plausibility and conformance with traditional
27
descriptions of English syllabication. A few examples will serve to il-
lustrate how far Hoard has in fact strayed in these regards. (2) gives
ampl.i.fy, atl.as, lingu.ist, and ostr.ich. I think it is fair to
say that any syllabication scheme for English that produces syllables like
[ampl], [atl], [lingw], and [ostr] is not using the notion "syllable" in
its traditional sense, in which the term has intuitive significance and
some (admittedly elusive) phonetic reality. Since Hoard suggests no pho-
netic interpretation for his syllables, the conclusion seems forced that
they are completely abstract units, set up to simplify a certain set of
phonological rules. But then, as in the C-boundary analogy, we are being
offered an account which is essentially arbitrary.
1.5 In the remainder of this Chapter and in Chapter 11, I will try
to show that the intuitions of linguists like Hoard are basically correct:
there is ample justification for using the traditional term "syllable" to
refer to an important conditioning factor in many rules. I will attempt
to go further than Hoard has gone, however, by showing that when rules
assigning syllable structure to segment strings are correctly written,
they not only provide the needed apparatus to properly account for the
naturalness of a large class of phonological rules, but also produce cor-
rect syllabications on the phonetic level. In other words, I attempt to
demonstrate that the kind of hypothesis-justification discussed in the
final part of Section 1.1 is available with respect to the proposal that
syllable structure is behind the [C, #1 and other rules.
Section 2 - Phonetic Descriptions and the Syllable
In this section I would like to discuss the syllable's claim to a
28
place in phonetic representation. I begin the discussion with some remarks
on "acoustic correlates" of phonetic units.
Recent work in phonetics makes it clear that the view that to each
unit in a narrow phonetic transcription there corresponds a unique acou-
stic signal (i.e., a unique function relating sound pressure and time) is
hopelessly naive. For example, although the phonetician transcibes the
word Bob as [bab], with identical initial and final segments, the acoustic
signals corresponding to these segments are radically different. In the
first [b], the first and second formants (peaks in the amplitude-vs.-fre-
quency curve) start low in frequency and rise to appropriate values for
[a], while the "correlate" of the second [b] is a fall in formant frequen-
cies. Furthermore, there is no indication whatsoever in the acoustic sig-
nal corresponding to [bab] of segmentation into three units - the acoustic
parameters vary smoothly from beginning of utterance to end. In fact, if
one records the word Bob and replays an initial fraction f for a listener
who has not been told what to expect to hear, it is impossible to choose
f such that the listener can unambiguously identify the first segment but
not the second. If he hears enough of the utterance to know for sure that
the first segment is (b], he will know as well that the second is [a].
Many other such illustrations of the lack of a one-to-one correspon-
dance between phonetic segments and acoustic signals could be given. Fur-
thermore, the same phenomenon is seen on the suprasegmental level. An
obvious example involves English stress.
In phonetic transcriptions, one distinguishes between the American
pronunciations of the noun and verb forms of the word pervert by means of
a simple difference in stress placement, for segmentally both are [prvrt].
29
Yet the acoustic manifestation of phonetic stress involves a very complex
context-dependent interplay of at least three factors, fundamental frequen-
cy contour, intensity, and segmental timing (see the discussion in Sec.
4.4.2 of Lehiste, 1970, and the studies she refers to). There is nothing
in the acoustic signal which directly reflects the primary, secondary,
and reduced stresses that the phonetician perceives as the distinguishing
features of the noun and verb forms of pervert.
Should we conclude from examples of this sort that traditional pho-
netic transcriptions are hopelessly inadequate and of no value? Obviously
not. The correct conclusion is rather that phonetic representation is a
level in the overall code relating meaning and acoustic signal that admits
of a good degree of abstractness from the point of vie w of the acoustic
output.
In support of the claim that this abstract phonetic level has psycho-
logical reality and cannot be dispensed with, one might bring to bear the
following sort of argument. In spite of the obvious acoustic differences
between the two [bis of Bob, we are justified in identifying them because
we have good evidence that the language user does. In historical sound
change, such classes of distinct acoustic signals as initial and final [b]
are often grouped together as focus or conditioning environment of the
change. When speakers invent symbols for the perceived sounds of their
language, the members of classes of this type are not distinguished.
This kind of argumentation is of course of just the sort that can
often be used to defend the even more abstract phonological levels which
are set up to capture, for example, regularities in the way that the pho-
netic form of a stem varies as a function of its environment. Nonetheless,
30
in their most general application, such arguments show the need for one or
more phonetic/phonological levels in addition to the level of the acoustic
signal.
One can, however, go further and differentiate between the phonetic
level and the more abstract phonological levels. One distinction has to
do with considerations of universal grammar. For example, although
English does not distinguish, in final position, between the class of
acoustic signals commonly referred to as unreleased [p] and the class of
released [p), other languages do, and the English-speaking phonetician
who fails to note the difference in preparing a narrow transcription is
simply a poor phonetician. On theother hand, the rising-formants and
falling-formants varieties of [b] are not distinguished in even the most
careful of phonetic transcriptions, for these "allo-acoustic-signals" are
universally determined by phonetic context.
To summarize the main point of these remarks, even the most concrete
of the phonological levels, that of phonetic representation, is related to
the acoustic signal by an extremely complex set of context-dependent rules.
A phonetic representation does not capture the elementary physical attri-
butes of an utterance directly, but rather encodes them in a way intimate-
ly tied up with the language capabilities of the human being.
With these considerations in mind, let us turn to the notion of syl-
labic structure. The first thing to be noted is that it is not a fair
challenge to the linguist who would divide utterances into "syllables"
at the phonetic level to demand instrumental proof of "syllable boundary."
The relevant observation is rather that the notion of "syllable" as a
phonetic unit is meaningful to the naive speaker and to the phonetician
31
in the same way that the concept [b] comprising the rising- aid falling-
formants variants is. Abercrombie says (1967:34):3
One unit seems an obvious starting point for this purpose (analy-
zing speech -DK), and that is the syllable. Most people seem to
be able to say, without much difficulty, how many syllables are
contained in a word or utterance; and with perhaps somewhat more
difficulty, to say where each syllable begins and ends... The
syllable would appear to be an intuitively recognizable unit even
for primitive peoples.
We might ask why speakers and phoneticians perceive syllables. Again
it is helpful to consider this question first with respect to a less con-
troversial phonetic unit like [b]. Obviously one thing that rising- and
falling-formants [bis have in common is that they correspond to the same
articulatory gesture, bilabial closure with voicing and closed v-IIu, and
it is reasonable to assume that this is what is behind our regarding ini-
tial and final [b] as the same segment.
Phoneticians have suggested such an articulatory correlate for the
syllable. According to Pike (1947:60), phonetic syllables are "units of
one or more segments during which there is a single chest pulse and a sin-
gle peak of sonority or prominence." Similarly, for Abercrombie (1967:35)
"each contraction (of the respiratory muscles), together with the result-
ing puff of air, constitutes the basis of the syllable."
The idea that each syllable of an utterance corresponds to a pulse of
the chest musculature can be traced back to Stetson (1928), who viewed
the syllable as being basically a "motor unit."
There is some evidence that this model is not entirely adequate and
32
that the articulatory correlate of the syllable is more complex (see Aber-
crombie, op. cit.), leading Ladefoged, for example, to conclude (1971:81)
that "there is no single muscular gesture marking each syllable." But
Ladefoged warns against "being overly simplistic in our view of physiolo-
gical properties...we may still be able to define a physiological unit of
this kind (the syllable) which will account for the timing and coordina-
tion of the articulatory movements. There is evidence...that speakers
organize the sequences of complex muscular events that make up utterances
in terms of a hierarchy of units, one of which is of the size of a sylla-
ble; and it is certainly true that speakers usually know how many sylla-
bles there are in an utterance. We will therefore assume that a neurophy-
siological definition is possible, even if one cannot at the moment state
it in any way."
Also relevant to this discussion is Malmberg's (1955) claim to have
experimentally isolated an acoustic correlate of syllable division in
vowel-stop-vowel utterances in the relative timing of vowel-formant tran-
sition and stop interval.
A further point to be made in support of recognition of the syllable
involves its role in word recall. Fudge (1969)points out that some apha-
siacs apparently retain a knowledge of the syllable structure of words
none of whose segments they are able to recall, and the same is apparent-
ly true of the normal speaker with a word "on the tip of his tongue"
(Brown and McNeill, 1966).
Thus when one considers its strong intuitive meaningfulness, its con-
stant appearance in phonetic descriptions, and our partial understanding
of its acoustic and physiological basis, the syllable appears to have as
33
much clai to a role in phonetic representation as a segment like [b].
Section 3 - Ambisyllabicity
In all traditional treatments of English syllabication, a word like
atlas would consist of two syllables, [at] and [las]. Since each syllable
is well-defined, it makes sense to speak of a "syllable boundary" as oc-
curring between the [t] and [1] of atlas. Thishenomenon of well-defined
boundary is observed in a large class of cases in English, leading to the
general assumption on the part of many phonologists that it is always
possible to segment an English utterance into n well-defined syllables,
i.e., to choose (n-1) intersegmental positions as syllable-boundary loca-
tions.5
However, this conclusion is not a logical necessity. There need
not correspond to every pair of adjacent syllables a well-defined syllable
boundary. For example, as opposed to a word like atlas, where the bound-
ary between syllables is uncontroversial, it would seem completely arbi-
trary to insist that hammer contains a syllable boundary either before or
after the [m].
In the past this fact has been typically either ignored (but see be-
low), in which case one arbitrarily assigns a syllable boundary in a word
like hammer, or else taken as evidence that the concept of the syllable is
an untenable one. The position taken here is a middle one between these
two extremes: it makes sense to speak of hammer as consisting of two sylla-
bles even though there is no neat break in the segment string that will
serve to define independent first and second syllables.
Using Pike's term "sonority" (each syllable contains exactly one "peak
34
of sonority"), there appears to be a sonority trough at the [m] in hammer,
as opposed to a complete break in sonority between the [t] and (1] of
atlas. It would seem reasonable to maintain, then, that while hammer is
bisyllabic, there is no internal syllable boundary associated with the
word. As an analogy to this view of syllabic structure, one might consi-
der mountain ranges; the claim that a given range consists of, say, five
mouni;ains loses none of its validity on the basis of one's inability to
say wI'ere one mountain ends and the next begins.
The observation that polysyllabic words in English need not have well-
defined syllable boundaries has in fact been made before. Careful phone-
ticians not committed to a theory of well-defined syllabication have sug-
gested that intervocalic consonants in English may belong simultaneously
to a preceding and a following vowel's syllable.
For example, in discussing words like being, booing, Trager & Smith
(1941:233) say, "...in cases like these, the intersyllabic glide is ambi-
syllabic (i.e., forms phonetically the end of the first and the beginning
of the second syllable), so that these words exhibit a syllabic structure
exactly parallel to that of such words as bidding..."
Smalley (1968:154) points out that it is easy to identify the "crests"
of syllables but notes that "it is not always possible to determine an
exact syllable boundary. A consonant between two syllables may belong
phonetically to both." He gives the English word money as an example of
this phenomenon.
The difficulty speakers of English experience in saying, in many cases,
just where one syllable ends and the next begins, referred to by Abercrom-
bie (see quote above), is doubtless due to their uncertainty about arbi-
35
trary syllabication conventions in these ambisyllabic cases.
The only phonologists who to my knowledge try to deal formally with
the phenomenon of ambisyllabicity in English are Anderson & Jones (1974).
For them also, words like hammer, being, booing, bidding, and money would
involve ambisyllabic segments. I will have more to say about their pro-
posals below and in Chapter II.
Section 4 - Phonetic Syllabication: Formalism
In this section I will indicate what I believe to be the basic require-
ments for a theory of syllabic structure for the phonetic level.
The starting point would seem to be the binary feature [syllabic] (see
SPE p. 354; Ladefoged, 1971:81). The class of [+syllabic] segments in-
cludes the ordinary vowels as well as the so-called syllabic sonorants,
such as the last segment of the English word button in its most common
American pronunciation, [bAt9n]. As far as I know, there is no controver-
sy among phoneticians as to the syllabicity of particular segments found
in English words, although a rigorous definition of this feature has never
been achieved.
It would also seem uncontroversial to define the number of syllables
contained in a given word or utterance as equal to the number of [+syllabic]
segments in the string. Thus phoneticians and naive speakers of English
are in agreement that photo, button, and wrestle have two syllables, that
there are two easily distinguished pronunciations of wrestling, one with
two syllables, the other, perhaps less common, with three, that four con-
tains one syllable as pronounced by most Americans, two as pronounced by
many, and so on.6
36
Given the one-to-one relationship between [+syliLabic] segments and
syllables, we can regard phonetic representations as consisting of segmen-
tal strings together with an indication of syllable membership, where the
nth syllable contains the nth [+syllabic] segment as well as zero or more
left- and right-adjacent [-syllabic] segments. For a word like atlas,
which consists of two discrete syllables, and one like pony, involving an
ambisyllabic element, Anderson and Jones (op. cit.) would use the notation
indicated in (3), which allows improper bracketing:7
(3) [mtI [ lesI [po[nj
1 i2 .2 1 I X
According to this notation, the first syllable of pony is [poni, i.e., the
material in the brackets labelled 1, the second syllable [ni], the materi-
al of brackets 2. For reasons which will become clear immediately, I pre-
fer to use the graphical representation of (4):
(4) ltI a a p o n i
V \V \VVl
S1 S2 S1 S2
Given the way the term "syllable" is understood, it would seem non-
sensical to speak of discontinuous syllables. No one has ever suggested,
for example, that a language could contain a bisyllabic form like [poni]
in which [p] is a member of the second syllable. Worse yet would be the
proposal that [p] could be the final member of this syllable. Free use
of Anderson & Jones' Improper bracketing allows these possibilities to be
represented: p][on][i. In the notation of (4), such a syllable structure
cannot be represented, since even if [p] were associated with the final
syllable, it would be interpreted as the initial element of that syllable,
37
due to the more constrained nature of the graphical representation:
(5) p on1i
S S
However, even the syllable structure represented in (5) involves a discon-
tinuity. One obvious way of preventing this possibility is to adopt the
convention that the lines associating syllables and segments may not cross.
This assumption has a further desirable consequence. In Anderson & Jones'
treatment, intervocalic consonant clusters can be ambisyllabic. In fact,
they would (for phonological reasons, not phonetic ones) represent a word
like Boston as [ba[stlani. But in no sense does this word consist of a
syllable [bost] followed by a syllable [stan]. This syllabication possi-
bility, represented in line-of-association form in (6), is correctly ruled
out by the assumption that lines of association may not cross.
(6) b os t a n
S S
Since [s] appears to be an ambisyllabic element in this word, its structure
is as shown in (7):
(7) b os ta n
S S
The assumptions being made here about the syllabic structure of strings
of segments can be summarized as follows:8
38
(8) a. Each [+syllabic] segment is associated with exactly one
syllable.
b. Each [-syllabic] segment is associated with at least one
syllable.
c. Lines associating syllables and segments may not cross.
It should be kept in mind that for a given segmental string, (8) may allow
more than one syllabication; language-specific rules (and perhaps some
universal ones) will determine the correct syllabication consistent with
these general constraints.
As an illustration of the way in which (8) limits the possible sylla-
bications of a given string, consider again the word Boston. (8a) insures
that the structure indicated in (9) will be present:
(9) b O 2 tn
S1 S2
Any further association involving the vowels would violate the "one-sylla-
bic-one-syllabie" principle.
(8b) and (8c) taken together insure that [b] will be associated only
with S1 and [n] only with S2* Any other association would involve cross-
ing, violating (8c), while leaving [b] or [n] unassociated would violate
(8b).
Universal conventions (8a-c) allow several possibilities of assigning
[s] and [t] to syllables and rule out many more. One of the possibilities
permitted by (8) is (7), which I take to be the correct syllabication of
39
this English word. The claim being made in proposing the conventions of
(8), of course, is that no language could allow a segmental string
[bostan] with a syllabication other than those permitted by (8). In most
of the remainder of this chapter I will be concerned with the question of
how phonetic syllable structure is derived in English, i.e., with those
rule of English phonology which assign syllable membership to segments.
Section 5 - Syllable-Structure Assignment Rules for English
5.1 (8a) requires a one-to-one relationship between syllables and
syllabic segments. The first of the rules which produce proper syllabifi-
cations of strings of segments is thus (10):
(10) Rule 1:
With each [+syllabic] segment of the input string associate
one syllable.
Given the input Mississippi, for example, Rule I produces (11):
(11) m1i sis i p i
I I I I' I
S S S S
The remaining syllabication rules deal with associations between syllables
and non-syllabics, for as we have seen, no further associations with vowels
are possible.
5.2 As was noted in Section 4, our basic assumptions requite Ihat a
word-initial (-final) consonant be syllabified with the first (last) syl-
lable of the word. Furthermore, if we have an initial (final) consonant
cluster, the entire cluster must be contained in the first (last) syllable.
40
Thus by studying what types of cluster occur initially and finally in words
in English, we can discover at least a subset of the set of possible sylla-
ble-initial and -final clusters. We cannot be certain that this technique
will reveal all the possible syllable-marginal clusters because of the
possibility that some of the tautosyllabic clusters found word-internally
will be systematically absent in word-initial or -final position.
There are, however, two pieces of evidence suggesting that in English
all possible syllable-marginal clusters are indeed revealed in a study of
possible word-marginal clusters. First of all, were this not true, we
would expect to find medial clusters which could not be analyzed as a pos-
sible word-final cluster followed by a possible word-initial cluster.
Such a situation is, however, never observed. Given an English word
[...VC1 6..CnV...], it is always possible to find at least one i such that
[C ..,.C ] is a cluster found word-finally and [Cf1+1...C ] is a cluster
found word-initially. There could be no word like akpsa in English, cor-
responding to the fact that [kp] and [kps] are not possible word-finally,
[ps], [kps] not possible word-initially.
This evidence, it must be acknowledged, is not absolutely conclusive,
for it might be the case that a word like Medford, which could be syllabi-
fied into permissible word-marginal clusters, [med,ford], is in fact syl-
labified [medf,ord] or [me,dford]. The second important observation is
that while there are admittedly cases in English where the syllabication
of medial clusters is not immediately clear, we are never led to syllabi-
cations which involve clusters not observed at word margins. We do not
find phoneticians or naive native speakers debating, for example, whethei
album might be syllabified [a,lbum].
41
Thus I will accept as a safe working hypothesis the assumption that
the set of possible syllable-initial (-final) clusters in English is
identical to the set of possible word-initial (-final) clusters.
5.3 Even under this assumption, of course, there are many cases in
which more than one syllabication will be allowed. pony, 'for example,
might be [po,ni], [pon,i], or, as I have suggested,
p o n i
\VV/
S S
As a first step towards a determination of the actual rules which
assign syllabication of intervocalic consonants in English, let us consi-
der slow, over-precise speech, or even the type of speech one might use to
imitate a science-fiction-movie robot.9 What we find in this sort of ar-
ticulation in cases where more than one syllabication is consistent with
the general constraints is a strong tendency to syllabify in such a way
that initial clusters are of maximal length, consistent with the general
constraints on word-initial consonant clusters. Thus we find, with syl-
lable boundary optionally realized as pause, [po,ny], not [pon,yJ; [Ja,cob],
not [Jac,ob]; [ac,tress], not [act,ress]; [Bo,ston], not [Bos,ton] or
[Bost,on]. I have chosen examples with unstressed final syllables delib-
erately, for the tendency towards maximal initial clusters is if anything
even stronger when the final syllable is stressed: (De,troit], not
[Det,roit]; [a,stound], not [as,tound] or [ast,ound].
In faster speech, many of these syllabication break down. For exam-
ple, as mentioned earlier, in ordinary speech we must recognize the [n]
42
of pony as ambisyllabic.
Rather than attempt to write a single rule which will produce normal-
speech syllabication, I will first set down the syllabication rule for
slow speech, then suggest the existence of other rules which apply only
in faster, i.e., normal, speech.
There are several reasons for taking this approach. First of all,
as will be seen in Chapter II, it may be necessary to order certain phono-
logical rules between the slow-speech syllabication rules and the rules
which account for faster-speech modification of syllabic structure. Sec-
ondly, it appears to be the case that faster speech modifies syllabic struc-
ture in a well-defined way: additional lines of association are introduced.
Thus the most natural account would seem to be one in which the obligatory
asociations are set up first, followed by optional (but normally invoked)
rules which introduce additional structure. Thirdly, the ability to uIse
over-precise "slow speech" is a part of competence. A single rule produ-
cing correct syllabications for normal-rate speech would leave this aspect
of competence unaccounted for.
The slow-speech rule, which will henceforth be referred to as Rule 11,
is given in (12). (On the use of the term "permissible cluster" in this
rule, see discussion below.)
(12) Rule II
a. C ...C V-C C..CC ...C V
S n 1 ii+l. n
S S
where C +.'' Cn is a permissible initial cluster
but CC C+1...C is not.
43
b. V C1....C 4-> V C1 ... C Cj+1 ...0Cn
S S
where C ..0C is a permissible final cluster
but CV.. .0 CJ+ is not.
(x standing below a segment indicates that the segment is not associated
with any syllable.)
Several points of clarification are needed with respect to Rule II.
Civen input strings spin, *kspin, mist,*mistk, Boston, Ia produces the
associations shown by dashed lines in (13) (Rule I has already applied):
(13) s v i n k s p i n m i a t m i s t k B o s t o n
S S S S
Rule 1Ib introduces the further structure shown in dashed lines in (14):
(14) s p i n k s p i n m i s t m i s t k B o s t o n
S S S
Since, as we shall see, no following rule will acL to assign the initial
[k] of hypothetical kspin.or the final [k] of hypothetical mistk to a syl-
lable, these forms will be rejected as possible words of English, since
they violate universal constraint (8b).
Similarly, given hypothetical [...VkpsV...], mentioned earlier, Rules
I-II produce (15):
(15) ...V k p s V...
V V
S S
44
Unassociated [p] is the formal correlate of the native speaker's rejection
of [...Vkps...] as a possible word of English.
Note in (14) that JIb has applied to the post-vocalic [st) of mist
but not the post-vocalic [stj of Boston. This is a consequence of the
fact that only in mist is [st] not assigned a syllabic association by Ila.
The medial [st] of Boston has lost the opportunity to be syllable-final.
Thus the slow-speech tendency toward maximal initial clusters rather than
maximal final clusters is reflected in Rule II by the ordering (a), then
(b).
Continuing our investigation of the operation of Rule II, consider
the following problem. Given a language identical to English in all re-
spects except that it allows syllable-initial /stl/, i.e., has words like
stlit. How would Rule IIa treat al input of the form [...VpstlV...J?
Since [1] is a permissible cluster but [tl] is not, Rule Ila could act to
syllabify [lV], leaving the rest of the cluster unassociated. However,
since by hypothesis [stl] is also a permissible initial cluster while
[pstlj is not, another syllabication of the input string consistent with
IIa is [stlV]. Thus the operation of Ila would be ambiguous in such a
case.
I take it as a highly significant fact that we must turn to a hypo-
thetical language for such an example. In English, the existence of pro-
hibited initial cluster [C .. Cn] implies that the Liuster [CC..C nI]is
also non-occurring initially, for any choice of C. It follows that there
will never be more than one C as defined in Rule Ila. The same results
hold, mutatis mutandis, for final clusters and Rule IIb.
There is a further point to be made in connection with Rule 1I, hav-
45
ing to do with the expression "permissible initial (final) cluster" in
the statement of this rule. The system of rules assigning syllable struc-
ture to strings of segments, as envisioned here, does not refer back to
some general set of constraints on possible word-initial and -final clus-
ters which is pervasive throughout the phonology. It is rather in the
syllable-structure assignment rules themselves that these constraints are
found. Furthermore, the constraints are not referred to by any other
rules of the phonology. As far as I can see, no such "globality" is re-
quired.
I elucidate these points by means of two examples. Part of Rule Ila,
left unstated in (12), states that [spi but not [ksp] is an acceptable ini-
tial cluster. It is for this reason that given an input like *kspin, Rule
Ha leaves [k] unassociated, resulting in rejection of the form as a pos-
sible word of English, as we have seen. Thus it is precisely in Rule Ila
that the restriction against word-initial [ksp] is reflected formally.
There is no need of an independent restriction which Rule IIa refers back
to.10
The lack of globality of the restrictions found in the rules of syl-
lable structure assignment can be exemplified by means of the fast-speech
rule outlined in (16):1l
(16) #C 1 a X #
This rule is oblivious to the constraints of the syllable structure assign-
ment rules. In the same way that it gives rise to [klekt] from collect,
well-formed from the point of view of Ila, it produces initial [pt]
46
in fast-speech potato [pteyDel. (Note how Rules IIa and (16) work toge-
ther to account for the native speaker's intuition that [ptamij, for ex-
ample, is not a possible word of English except as a fast-speech reduction
of acceptable hypothetical [petamil.)
5.4 The result of the application of Rules I-II to money is (17a),
while the ordinary-speech structure is as shown in (17b):
(17) a. m A n i b. m A n i
V V \V\/
S1 S2 SIS2
For the reasons mentioned in the previous section, I assume the existence
of one or more rules, ordered after I-II, whose effect is to introduce
additional lines of syllabic association like the one linking S and [n)
in (17b). I investigate the conditions on one such rule in this section.
To begin, let us consider the slow speech rendering of a form like
anemic. This word has the structure shown in (18) in slow speech, in ac-
cordance with Rules I-II.
(18) a n i m L k
I V\/Vsk
S S S
The [n] in (18) is syllable-initial, as is the [n] in (17a). However in
anemic there is no obvious change in the juncture between [o] and [n] as
speech rate increases, in contrast to the situation observed for a word
like money. The syllabic break between [a] and [n] in anemic appears to
remain.
Examination of additional examples makes it clear that this junctural
47
difference between normal-rate speech money and anemic is related to the
stress contour difference between them. Thus we have appear, attack,
collide with syllable boundary between the unstressed vowel and the fol-
lowing consonant vs. happy, attic, collie with ambisyllabic consonants.12
On the basis of examples like these, there appears to exist a rule
of the form (19a) or (19b):
(19) In V1 C V2  associate C and S1 if
I V
S 1S2
.I 1 2
a) V1 is stressed
b) V2 is unstressed
Either (19a) or (19b) will give correct results in all examples thus far
considered. To distinguish between these possibilities, we must consider
examples in which VI and V2 are both stressed or both unstressed.
It seems clear that words like senile, rabbi, latex do not involve
ambisyllabic medial consonants, indicating that (19b) is correct.13
(19b) makes the prediction that the [t] of captal and the [m) of
enemy will be ambisyllabic in normal speech. This prediction seems to
me to be borne out in fact, although I find the judgement more difficult
to make in this case. It can at least be said with confidence that the
clear perception of juncture one finds between [a] and [t] in a tall man
is absent in capital; likewise for amino and enemy with respect to the
[aJ-[m] junctures.
To account for the ambisyllabic [s] of ordinary-speech Boston, we
must extend (19b) in the way shown in (20), so that the first consonant
of an initial cluster of arbitrary length is syllabified with a preceding
48
vowel.
(20) In V C C-sessJ
S1 S2
associate C and S1.
(20) gives rise to the lines of uyllabic association shown dashed in (21):
b. Haskins
hmskin z
S S
c. April
e p r a
s s
In forms like pony and April, a more accurate transcription than that
given in (21) would show the diphthongal quality of the first vowel:
[powni], [eypral]. Strictly speaking, then, (20) should not be applicable
to these forms. To overcome this difficulty, (20) must be modified to
.(2?), which will be referred to as Rule III.
(22) Rule III
In [-consj C C0 [st e 4
IN\s
S S2
associate C and S
The formulation in (22) takes advantage of the fact that glides and vowels
constitute the class of [-consonantal] segments (see Appendix 1 to Chap-
ter II for a discussion of the major class features). Note that the pre-
sence of S in (22) implies the existence of a vowel to the left of
"[-cons]" if the latter segment is not itself a vowel.14
5.5 Our investigation of the assignment of additional syllabic link-
(21) a. pony
pon i
s s
49
ages to medial consonants has been so far limited to cases in which the
leftward syllable is vowel- (or glide-) final. Turning to other cases,
consider a word like after. Due to the restriction found in Rule IIa a-
gainst initial /ft/ in English, this word has the structure shown in (23a)
after application of Rule II.
(23) a. aWf t a r b. agf t a r c. a f t a rV \V \VM
S S S S S S
If th, [-cons] restriction were removed from Rule III, it would apply to
(23a) in normal-rate speech, producing (23b). However in this case and
others like it (i.e., words with an interconsonantal syllable boundary at
the output of Rule II), this syllable structure seems simply to be wrong.
I therefore assume that Rule III should remain as it stands.
However the sharp syllabic juncture of (23a) does seem to be lost in
faster speech, with [f) the ambisyllabic element, as in (23c). That is,
in ordinary speech the syllabic structure of after seems entirely parallel
to that of Astor, Haskins (21b).
This result suggests the need for a rule like that shown in (24):
(24) Rule IV
In C C [vi associate C and S0 1 -stress] 2
1 2
Note that Rule IV, whether ordered before or after III, will not affect the
syllabication of words like hammer, Haskins, nor does its ordering with re-
spect to III affect correct operation in the case of words like after.
50
As was the case with Rule III, a restriction to the effect that S2
be unstressed must be included in Rule IV. There is no muddying of the
juncture between [f] and [t] in Haftonium.
As it stands, Rule IV appears to be too general. Words like Medford,
bodkin seem to maintain the interconsonantal juncture assigned by Rule II
in spite of the fact that they meet the structural description of Rule IV.
Now if IV applied to these words, it would give rise to syllable-initial
[df] and [dkj. Although these are in fact clusters which are prohibited
in the complete statement of Rule Ha, Rule IV cannot in general be con-
ditioned by the constraints of Ha since IV must produce, for example,
initial [ft] in after.1 5
It is possible that there exists a set of English-specific constraints
on Rule IV, independent of those found in Rule II. A study of the relevant
examples, however, suggests that IV may be subject mainly to universal con-
straints. [df] and [dk], for example, are not found as syllable-initial
clusters in any language.16 Although some of the crucial facts are debat-
able, with regard to the proper statement of both the universal phonetic
constraints and junctural properties of English clusters, Rule IV, con-
strained not to produce "impossible" clusters, seems to be required for
English.
Note, incidently, that such a restriction is not relevant to the op-
eration of Rule III, for the latter rule produces syllables closed a sin-
gle consonant, against which there are no universal constraints.
Rule IV will usually operate only in the case of intervocalic clust-
ers, since when a single consonant stands between vowels it will almost al-
ways be syllabified with the vowel to its right by Rule Ha, allowing only
51
vacuous application of Rule IV. The sole exception to this generalization
involves /9/, which does not occur initially in English. A medial /0/
will be left unassociated by Rule Ha and will be linked to the preceding
syllable by IIb, since /0/ is possible finally in English. Thus medial
/0/ will be subject to Rule IV. (25), (26), (27), (28) show by means of
dashed lines the application of Rules Ha, Ilb, III, IV, respectively, to
inputs hammer /hAmer/ and hangar /hmoe r/.
(25) h am or h aj a r
'./ ' /jIIby IIa
S S S S
(26) ha m a r hab a r
\ /Iby IIb
S S S S
(27) ha m a r h a r
V by III
S S S S
(28) ha m a r h a r
by IV
S S S S
I take the final output (28), as providing some additional evidence for
the correctness of the rules suggested, for in normal speech hammer and
hangar are indeed juncturally indistinguishable, whereas in artificially-
slow speech they differ.
5.6 Ignored so far has been the question of the domain of application of
the rules of syllable structure assignment.
Rule I, which simply sets up a syllable for each syllabic segment in
52
an input string, is not critical in this regard. Rule II, however, must
not apply beyond the domain of the word. The phrase this time, for exam-
ple, must not be syllabified [6t ,staym]. Since there is always syllable
boundary, and perhaps even pause, between words in slow speech, Rule II
must be limited to the word.
Rule III appears to be limited to the word as well. Even in normal-
rate speech, say veranda remains juncturally distinct from save Iran.1 7
On the other hand, it is not the case that normal-rate speech never
involves inter-word syllabic linkages. At the output of Rule III, the
word Hockett and the phrase hock it have the structure shown in (29):
(29) Hockett hock it
h a k i t h a k i t
\W\W\VVS S S S
But in ordinary speech Hockett and hock it are fully homophonous. There-
fore the [k] of hock must be linked to the syllable of it.
Although Rule IV would serve to establish this association, it appears
that IV must be limited to the word and another rule introduced to provide
inter-word syllabic linking.
Rule IV is both too weak and too strong to describe inter-word linking
correctly. Rule IV links the last consonant of syllable n to syllable n+l,
which is consonant-initial in most cases in which IV applies. Across word
boundary, however, a final consonant links only to a vowel-initial sylla-
ble: this array but *this parade. This restriction has long been recog-
V V
nized, and is usually illustrated in the literature on juncture in English
by pointing out that sharp juncture is obligatorily retained between the
53
[t) and [r) of night rate.
Furthermore, inter-word linking seems not to be limited to following
vowels which are unstressed. In connected speech, the [k] of the phrase
Hock all of them seems as clearly ambisyllabic as that of Hockett.
For these reasons I postulate an independent Rule V:
(30) Rule V (connected speech only)
In C V associate C and S
S
Regarding the statement of Rule V in (30), observe that it is not necessary
to show linkage between C and a previous syllable, for if C is linked only
to S, Rule V will apply vacuously and need not be made inapplicable. Nor
is it necessary to explicitly mention word-boundary in the rule, for with-
in a word a prevocalic consonant will almost always have been linked to
the following vowel by Rule IIa; when the consonant is /0/, Rule IV will
have provided the association.18 Within words, then, Rule V will always
apply vacuously.
One must draw a distinction between Rules III and IV on the one hand
and Rule V on the other with regard to conditions of applicability. It
was pointed out that in contrast to Rules I and II, Rules III and IV are
inh ibited in slow speech but are required at normal rates of articulation.
Now since even in normal speech one may pause between words, Rule V must
be limited to stretches of "connected speech." An alternative, apparently
equivalent, view, is that in normal speech Rule V is optional but usually
applies.19
A problem much discussed by the pre-generative phonemicists was that
54
of accounting for the non-homophony of pairs of phrases like a name and
an aim. The fact that both phrases contain the same segment sequence in
normal-rate speech, [aneym], together with the structuralist methodology
requiring determination of the set of phonemes of a language before in-
vestigation of morphological phenomena, led to the setting up of a phoneme
of juncture to account for minimal pairs such as these. In the generative
treatment of juncture proposed here, the difference between the two phrases
is a consequence of word-boundary placement (i.e., a morphological consi-
deration) and the fact that English has no rule which is the inverse of V,
i.e., no rule which syllabically links word-final vowels to word-initial
consonants. The phrase a name receives structure (31a) from Rules I-II,
and no further rules apply. In an aim , however, Rule V provides the
dashed line of syllabic association shown in (31b).
(31) a. a name b. an aim
a n e y m a n e y m
S S S S
As a final observation on Rule V, note that its effect is to elimi-
nate vowel-initial syllables. On the basis of cross-linguistic observa-
tions, we know that such syllable are more highly marked than consonant-
initial syllables. Within words, Rule II acts to prevent vowel-initial
syllables, if at all possible. Rule V steps in exactly where Rule II is
powerless to create a consonant-initial syllable, in the case of a vowel-
initial word. In this way we can account for the lack of symmetry men-
tioned above, the presence of Rule V in the phonology versus the absence
of a rule linking word-final vowels with word-initial consonants.
55
5.7 In (32) I summarize the rules determining English syllabic
structure that have been motivated in this section. I know of no other
rules of this type.
(32) Summary of English Syllable-Structure Assignment Rules
(Domain of application of all rules except Rule V is the word.)
Rule I
[+sylJ -- [+sylj
Rule II
a. C .... Cn V -
S
where C +0. ..C is a
initial clusters but
b. V C ...C -- +
11 in
x....
S
C ... C Ci+...C V
S
member of the set of permissible
C ci+1'''.C is not.
1 i+l6 n
S
where C ...C is a member of the set of permissible
final clusters but C ... C Ci+1 is not.
Rule III (normal-rate and faster speech only)
In [-cons] C C0  [ ] associate C and S
S1S2
56
Rule IV (normal-rate and faster speech only)
In C C -V associate C and S0 trs2
Sy S21 \ 2
Condition: CC0 must not be a member of the set of univer-
sally-prohibited clusters; certain highly marked clusters
not universally proscribed may be excluded also.
Rule V (connected speech 'aly)
In C V associate C and S
S
Section 6 - The Syllable and Phonotactic Constraints
6.1 It is usually assured that each language has a set of "norpheme
structure constraints," which place limitations on the phonological shape
of possible formatives of the language. These constraints, together with
the language's morphological and phonological rules, define the class of
possible phonetic representations for words in the language.2 0
In English, for example, there could be no word like [ff$, since /0/
is not an underlying segment (i.e., is ruled out by a morpheme-structure
constraint) nor does any phonological rule serve to introduce it. [xat],
on the other hand, is impossible in English for another reason, according
to the analysis in SPE (p. 234). Although the presence of /x/ violates no
morpheme-structure constraint, all occurrences of initial /x/ become /h/
by general rule. Finally, taking another example from the analysis of
English in SPE, English has phonetic diphthongs in spite of the constraint
57
against them in lexical representations because of the existence of the
Diphthongization Rule (p. 183).
6.2 Regarding morpheme-structure constraints, it should be noted
that the restrictions embodied in Rule II go a long way towards distin-
guishing between permissible and non-permissible morphemes. Furthermore,
in a theory with no explicit recognition of the syllable, the set of mor-
pheme-structure constraints would be considerably more complex.
For example, consider the problem of accounting for the fact that
atktin is not a possible word of English. No phonological rules are of
relevance in ruling out this form. As we have seen, the constraints a-
gainst syllable-final /tk/ and syllable-initial /kt/ contained in Rule II,
together with the universal convention on syallabic association (8), suf-
fice to exclude atktin.
Let us consider how the unacceptability of atktin would be accounted
for in a theory not making use of the syllable. We would not want to have
a constraint which simply excludes /tkt/, because the non-occurrence of
this cluster is surely related to the absence of many other clusters, for
example, /tks/. Nor could we simply rule out /tk/ or /kt/, for we must
not exclude acceptable forms like Atkins and cactus. On the other hand,
it is true that final /tk/ and initial /kt/ are not permissible. Thus we
would have to recognize constraints against the configurations in (33);
(33) *tk * kt
However, as observed in Sections 1.2, 1,3, [C, #] is an unnatural class.
As will be seen directly below and in Chapter II, it often arises when
58
syllable-conditioned phonological processes are not recognized as such.
There is a problem concerning (33) as hypothetical morpheme-structure
constraints beyona che appearance of [C, #]. For certain choices of C,
/tkC/ and /Ckt/ must not be excluded. Examining the case /tkC/, we find
that unlike atktln, hypothetical atklin and atquin are possible words of
English. The non-syllabics which may follow /tk/ are in fact just those
which may follow /k/ in initial clusters. In a system a morpheme-structure
constraints not recognizing the syllable, this fact will be represented as
accidental. But this generalization (and many others like it) is perfect-
ly predictable. As mentioned in Section 5.2, medial clusters in English
simply obey the constraint that they must be analyzable into a possible
final cluster followed by a possible initial cluster. Thus any attempt
to write constraints on possible medial clusters will miss important gen-
eralizations. As we have seen, it will also leave unexplained why word-
boundary is found co-occurring with classes of consonantal segments in
morpheme-structure constraints. Thus the general problem of accounting
ror possible morpheme shapes provides additional evidence for the correct-
ness of introducing the syllable into phonological descriptions.
However, it must be acknowledged that even in a syllabic treatment of
morpheme-structure conditions certain well-known problems arising from the
presence in the grammar of morphological processes like affixation do not
find an immediate solution.21 For this reason I make no attempt to out-
line a general theory of phonotactic constraints. It should be clear how-
ever that any such theory must countenance the syllable as a significant
phonological unit.
59
Footnotes to Chapter I
1. Some phonologists, of course, have tried to incorporate the syllable
into a generative framework; cf. Vennemann (1974, 1972), Hooper (1975,
1972), Stampe (1972), Hoard (1971), and many papers, published and un-
published, by C.-J. Bailey, for example (1968, 1975). Although I have
benefited from the work of these authors, I find myself in disagreement
with them on some major points, as will become clear.
I am not aware of any serious attempt in the literature to show that
the syllable and generative phonology are mutually exclusive. Leben (1973)
is sometimes cited as an "anti-syllable" work, but Leben's arguments, if
valid, do not bear on the claims I will make here. His main points regard-
ing the syllable are that (a) the syllable is not appropriate as a unit on
which to express certain phonological features, and (b) analyses have ap-
peared in the literature in which reference to syllables is used merely as
a notational trick to simplify rule statements, without providing any ad-
ditional insights into the workings of phonology.
2. For serious discussion of the Sanskrit ruki rule, see Zwicky (1970)
and the references cited there.
3. I take the inclusion of the word "even" in the last sentence of the
quote not as a slur on "primitive peoples" buc rather as an admonition
that intuitions about syllable structure cannot be attributed to wide-
spread literacy.
4. For example, those cited in the first paragraph of footnote 1; also,
Abercrombie (1967), O'Connor & Trim (1953), and Fujimura (1975).
60
5. One reason for this assumption on the part of non-phonologists is the
desire for a well-defined set of conventions for end-of-line word-breaking
in written texts. In spite of the irrelevance of such conventions to the
question of phonetic representations, they have influenced people's think-
ing on phonetic syllabication.
6. One should not be surprised to find a minority of cases in which even
trained phoneticians would be unsure as to the syllabicity of a given pro-
duction of some sonorant. Such instances of uncertainty should not be
allowed to cast doubt on the basic correctness of [syllabic] as a binary
feature. It must be kept in mind that this sort of problem is pervasive
in phonetics.
In my transcriptions I use a tick mark [XI to indicate the [+syllabic]
equivalent of a normally [-syllabic] segment and a bow (X] to indicate the
[-syllabic] equivalent of a normally [+syllabic] segment. For example,
monosyllabic pronunciations of four include [for] and [foe], bisyllabic
pronunciations [foe] and [for].
7. In indicating syllabications I use broad transcriptions or even stan-
dard orthography where phonetic detail is not important.
8. There is an important relationship between the proposals I make here
and the "autosegmental" theory of John Goldsmith, and in fact my own work
has benefited from Goldsmith's insights.
A central notion of Goldsmith's theory is that the traditional view
of phonological segmentation is wrong in that it assumes that a unique
analysis of a stretch of phonological material into segments is appropri-
ate for the specification of all the phonological "distinctive features;"
61
in his treatment, the various features, in particular the so-called
"suprasegmental" ones, induce analyses potentially distinct from the
traditional segmentation but related to it by means of well-formedness
conditions, which may be said to preserve linearity.
In spite of the fact that in this dissertation I am not concerned
with the possibility that there exist more appropriate domains of defini-
tion for certain phonological features than the traditional segment, in
a broader sense I am working within the "autosegmental" framework. This
is so because all theories of the syllable, including my own, are "auto-
segmental" in that they involve parallel analyses of phonological materi-
al into (traditional) segments and syllables.
Traditional treatments of the syllable differ from mine in that they
impose a well-formedness condition which in my opinion must be relaxed,
the requirement that each segment be associated with no more than one syl-
lable. I retain, however, the remaining well-formedness conditions of
traditional syllabication; these are stated In (8), in a form influenced
by the work of Goldsmith.
It should be clear why I consider (8a) a traditional assumption re-
garding syllabication. (8b), together with the condition mentioned above
which I reject, gives rise to the traditional view that each [-syllabic]
segment is associated with exactly one syllable. (8c) corresponds to the
implicit traditional assumption that a form like [ponil may not consist
of the syllables [pi] and [on], or have [po] as the second syllable, [nil
as the first.
9. Other situations in which this type of speech may be heard include
dialog with infants, with foreigners, over noisy telephone circuits, etc.
62
10. No English word begins with //, and, correspondingly, a word with
medial /0/, like hangar, has the slow-speech syllabication [hM ,er].
Thus a restriction against initial /o/ must be embodied in Rule ia.
(On the normal-speech syllabication of hangar, see the discussion of
Section 5.5; on the question of whether [9] is underlyingly /ng/, see
Section 5 of Chapter III.)
Now note that although there are no native words with initial /i/,
the slow-speech syllabication of pleasure appears to be [ple,aerJ, not
[plez,ar], just as pressure is [pre,ser], not [prei,er]. Thus there can
be no restriction against syllable-initial /1/. Although it might be sup-
posed that there exist a constraint against word-initial /0/, independent
of the syllable-structure constraints, it seems more likely that a simple
gap is involved, i.e., that it is accidental that there are no words with
initial /E/. In support of this analysis, note that the ordinary speaker
of English pronounces the initial // of the Russian name Zenya with no
apparent difficulty, while initial [01 is accessible only to the student
of phonetics.
11. There are additional restrictions on this rule that are of no rele-
vance here. See Zwicky (1969).
12. I offer some additional support for these claims. Even those phone-
ticians who speak of ambisyllabic medial consonants in the case of words
like happy (see Section 3) do not do so in the case of words like appear.
Dictionaries (for example, AHD, RHD) show a junctural difference between
happy, etc., on the one hand, and appear, etc., on the other. This dis-
tinction is also recognized by phonologists who do not recognize ambisyl-
63
labicity, for example, Stampe (1972) and Hoard (1971).
13. Since the doubly stressed examples chosen have contour 1,3, the re-
quirement for introduction of the additional line of association could
not be "if stress on V1 greater than stress on V2 ." This possibility
might be expected on the basis of the original examples (money vs. inemic,
etc.), together with the observation that the medial consonant of 3,1
words does not become ambisyllabic: tcoon, Kowloon.
14. In the notation I will use to write rules conditioned by syllable
structure, the absence of a line of syllabic association in a rule state-
ment does not imply that the rule's applicability is dependent on the ac-
tual absence of the line. In (22), for example, S1 can have any number
of additional leftward lines of association.
Where it is necessary to specifically exclude certain linkages, such
exclusions are shown explicitly. In (i), for example,
(i) t V
x*
S
/t/ may or may not be initial in S, must not be associated with the pre-
vious syllable, if any.
15. This is another example of the non-globality of the constraints on
clusters found in Rule II.
16. /ft/, on the other hand, though not found initially in English, vio-
lates no universal restriction on possible initial clusters: Modern Greek
hinds, 'inexpensive,' Russian vtoroy [ftaroy], 'second.'
It may however be necessary to constrain Rule IV from producing cer-
64
tain highly-marked, though not universally proscribed, clusters. Further
discussion of this point is found in Chapter II.
17. Because of Rule V, to be discussed below, this pair of phrases would
be homophonous to the third syllable if Rule III applied to the first of
the phrases.
18. Rule IV provides linking to following unstressed vowels only.
/...rV.../, then, would seem to be a word-internal input to (30). However,
/0/ does not occur before stressed vowels in English word-internally, so
the generalization in the text holds.
19. While the likelihood of applying Rule V may be greater in (i) than
in (ii),
(i) Hock it!
(ii) After she decided which jewels to hock, it became clear...
it is certain that syntactic conditions are not relevant in the statement
of Rule V, for it quite normally applies across the strongest of bounda-
ries, as in (ii). Non-application of V in (ii) is more natural than in
(i) simply because pause is more natural in (ii).
20. This position is taken explicitly by Postal (1968).
Shibatani (1973) argues that there must in addition exist "surface
phonetic constraints." At least two of Shibatani's arguments for the re-
cognition of SPCs are fallacious. Consider a language all of whose mor-
phemes have the form CVC and all of whose suffixes have /t/ as the initial
segment. Then all consonant clusters within words in this language will
65
have /t/ as the second element. According to Shibatani (p. 92), the lat-
ter generalization cannot be expressed without SPCs. Actually, SPCs are
needed only to state the generalization directly. Even without SPCs the
generalization is immediately deducible from the morpheme-structure con-
ditions of the language and the rules of morphology (in particular,
word - stem + suffix).
Shibatani alst argues, p. 95-6, that although in the standard analy-
sis of German one can account for the non-occurrence of final voiced ob-
struents by means of the final-obstruent devoicing rule, there must exist
a SPC ruling out forms like *[bund], for German speakers will presumably
immediately reject them as possible words of the language. According to
Shibatani's line of reasoning, unless such a conotraint existed, speakers
could never reject hypothetical forms like [bund] since while it is true
that /bund/ as an underlying form could not give rise to [bund], some oth-
er lexical representation might, and a speaker would have to run all pos-
sible underlying configurations through the rules of the phonology before
being able to state with certainty that [bund] is impossible.
It seems to me that this sort of reasoning is psychologically naive.
Placed in a twenty-foot-square room I immediately know that there is no
way I can touch two opposite walls simultaneously, presumably without un-
consciously considering all the possible ways I could orient my body,
stretch my arms and legs, etc. Or to take an example from syntax, how do
we immediately reject *John climbed the mountain the cliff? There is no
constraint against surface NP V NP NP and there is an infinite number of
deep structures and derivations to consider as possible sources.
21. If phonotactic constraints are literally morpheme-structure con-
66
straints, certain serious problems arise. Hooper, who shares the view
that the syllable is directly relevant to the statement of phonotactic
constraints, discusses these difficulties in Hooper (1975). She points
out that under the standard morphological analysis of Spanish, a word
like comprando 'buying' consists of the morphemes /kompr+a+ndo/, in which
one morpheme contains the final cluster /mpr/, another the initial cluster
/nd/. A study of Spanish words makes it clear that the presence of /mpr/
as a final cluster is directly conditioned by the presence of a following
vowel-initial morpheme, cf /nd/ as an initial cluster by the presence of
a preceding vowel-final morpheme. Intramorphemic constraints cannot ac-
count for the observed facts.
Another difficulty surrounding morpheme-structure constaints is that
some morphemes have no independent phonological form. It is hard to ima-
gine, for example, what sort of phonotactic constraints the past tense
morpheme in brought is subject to.
On the basis of problems like this, Hooper concludes that phonotactic
constraints must be stated at the level of the word, with no regard to
morphemic boundaries. Spanish allows comprando because /koN/, /praN/ and
/do/ are possible syllables; English allows brought because it is a per-
missible syllable.
Hooper notes, and in my opinion too rapidly dismisses, a problem
arising under this hypothesis. In English the sequence voiced-obstruent/
voiced-obstruent is generally absent in final position unless morpheme
boundary breaks the sequence: robbed, dogs. This is a serious problem
for her theory, for it suggests that some phonotactic constraints are
stated at the morphemic level after all.
She attempts to circumvent the problem by noting that there do exist
67
some monomorphemic words with final voiced-obstruent/voiced-obstruent clus-
ters, for example, adze, Ides, i.e., by suggesting that the constraint is
not a real one, and argues that at one time the restriction was operative
and word-based - rubbed was then bisyllabic and hypothetical adze ex-
cluded - and that now it is merely a historical accident that there exist
so few examples of this configuration.
However while exceptions in /-dz/ like those cited do exist, there
are simply no morphemes in /-bd/, /-Jd/, /-gz/, etc., although these clus-
ters are common when split by morpheme boundary. Thus Hooper has at best
shown that English has no morpheme-based general constraint against all
voiced-obstruent pairs, i.e., has argued against an artificially-strong
hypothesis. Also to be noted is the fact that there are many more exam-
ples of clusters permitted across morpheme boundary but never within mor-
phemes, for example, the well-known /ksOs/ of sixths.
Further direct confirmation of the morphemic status of many sequen-
tial constraints is available. For example, hypothetical forms like
[nAbdi, [negz] are interpretable only as preterites and plurals, not as
monomorphemic words.
Thus both morpheme- and word-based systems of phonotactic constraints
face serious problems, problems for which I have no general solutions to
offer. If however one sets oneself the more restricted problem of account-
ing for the possible phonetic forms of monomorphemic words in English, the
syllable seems indispensable, as I have tried to argue in the text. It is
hard to see how any solution of the more general problem could obviate
this need.
68
CHAPTER II
SYLLABLE-STRUCTURE CONDITIONING IN PHONOLOGICAL RULES
Introduction
In this chapter, I show how rules I-V, established to provide proper
syllabic structure for phonetic strings, can be used to condition many
phonological rules of English in a simple and natural way. Some of the
rules discussed are clearly contained in a synchronic phonology of English,
while others, though obviously actual phonological rules at some point in
the history of the language, may have been lost from the phonological
component.
Section 1 - The Phonological Development of Underlying /t/
A very interesting example of the use of syllabic structure to condi-
tion phonological rules is provided by the problem of accounting for the
phonetic forms which correspond to underlying /t/. Where relevant, I will
also comment on the allophones of the other voiceless stops.
There exist at least four distinct allophones of /t/ in all American
dialects (la-d), and two more in many (la', c').
(1) a. voiceless aspirated alveolar stop [thj: creativity
b. unreleased voiceless alveolar/glottal stop (i.e., closure
is made simultaneously at alveolar ridge and glottis) [t9 ]:
create
c. released voiceless unaspirated alveolar stop [t: stem, list
d. voiced1 alveolar tap [D]: creating
69
1 2 h
a'. voicelss aspirated alveopalatal stop [chJ: train
c.' released voiceless unaspirated alveopalatal stop [c:
strong
In the following subsections, the distribution of each of these allophones
is studied, and, where appropriate, additional remarks on phonetic detail
are made.
1.1 - Aspiration
A. The voiceless stops /p, t, k/ are all aspirated in the same en-
vironment. The usual description is that given in (2):
(2) /p, t, k/ are aspirated if and only if they
a. stand before a stressed vowel, and
b. do not follow /s/
In conformance with this description, the /t/s in after, stem, and Boston
are unaspirated, for they fail to meet conditions (a), (b), and both (a)
and (b), respectively, while ten and attempt have [th], since they satisfy
both conditions.
Closer examination of the facts, however, shows that (2a) is a spur-
ious generalization. (Thus the lack of aspiration in after must be trace-
able to a different sourL2.) Let us consider briefly the phonetics of
aspiration.
In words of the form #TVX# (T - voiceless stop, i.e., /p/, /t/ or /k/
X = any string of segments), first bilabial, alveolar, or velar closure is
made and pressure is built up behind the obstruction. At a certain point
in time, to, the closure is suddenly released and the tongue is quickly
70
brought into appropriate posizion for articulation of the following vowel.
It is not, however, until an appreciable timt later, t,, that voicing of
the vowel begins. The interval of time (t1 -t0), the "voice onset time,"
is a period of voiceless escape of breath, or aspiration. A crucial fact
to be noted is that voice onset time varies within quite wide limits in
words of the type under consideration.
There is an obvious correlation between the length of the aspiration
interval in the articulacion of an aspirated stop and the amount of stress
on the following vowel. Emphatically stressed words are very strongly
aspirated: I said two, not ten. Unemphatic primary-stressed syllables
have less aspiration, secondary-stressed syllables still less, other things
being equal (compare the stops in c0ptoe).
Now if close attention is paid to the manner of articulation of the
initial stops of words like tomorrow, Pacific, collide, where the stops
otand before completely stressless vowels, it will be seen that there is
still a noticeable period of aspiration, although it is shorter than be-
fore stressed vowels. The manner of articulation is quite distinct from
that of the true unaspirated cases like spin, stem, skin, in which the
onset of voicing is essentially simultaneous with release of the stop (see
Lisker & Abramson, 1964). This aspiration of initial voiceless stops be-
fore unstressed vowels is typically overlooked in the literature, but see,
for ex., Trager and Bloch (1941:225).
Thus although there may seem to exist a continuum in voice onset time
when one considers in turn ten, ten, temperamental, t6morrow, stem, there
appears to exist a categorical distinction between the first four cases
and the last.
71
In terms of Halle & Stevens (1971) proposals on the laryngeal feat-
ures, it appears that the first four stops are [+opread glottis] while
the fifth is [-spread glottis], with all five [-constricted glottis] and
+stiff vocal cords
-slack vocal cords f
There are several techniques one many .se to make oneself aware of
this categorical distinction. First of all, if words like tomorrow,
Pacific, and collide are uttered slowly and over-distinctly (but not neces-
sarily in such a way as to introduce stress onto the first vowel), the
aspiration will become even clearer. On the other hand, in my speech at
least, no amount of emphasis will introduce aspiration in the post-/s/
h&
cases: [sthaptt] is not possible for stop it'
A very direct confirmation of the claimed difference in articulation
of the stops in tomorrow, Pacific, and collide on the one hand and stem,
spin, and skin on the other can be made by speakers of English who have
good control over the pronunciation of the voiceless unaspirated stops of
languages like French, Spanish, and Italian. The English post-/s/ stops
are just like the Romance stops, while the stops in # - are quite
different. Furthermore, if one pronounces tomorrow, Pacific, collide with
the Romance stops, the result is decidedly unnatural for English, in just
the sane way that it is unnatural to pronounce ten, pen, Ken with the Ro-
mance stops.
Finally, note that there is no contrast in English between voiceless
unaspirated stops and voiced stops when not preceded by a sonorant. Thus
there could be no sdem, Bosden, akder contrasting with stem, Boston, actor.
When English-speaking students of phonetics are first made aware of the
72
fact that the /t/ of stick is quite different from the /t/ of tick, they
often identify the /t/ of stick with the /d/ of Dick. In fact, it has
often been observed that in words with initial /b, d, g/, true voiced
stops and voiceless unaspirated stops are in free variation for many spea-
kers of English.3 On the other hand, the aspirates are always kept dis-
tinct from /b, d, g/ in English.
Thus if the /p, t, k/ of tomorrow, Pacific, collide are unaspirated,
as usually assumed, /b, d, g/ should be freely substitutable, while if
they are aspirated, as claimed here, the same perception of phoneme re-
placement should be noted upon substitution of /b, d, g/ that one observes
when /b, d, g/ are substituted for /p, t, k/ in pen, ten, Ken. In fact it
is the latter situation that is observed: introduction of initial /d, b, g/
in tomorrow, Pacific, collide results in the perception of different (non-
existent) words, even if /b, d, g/ are not pre-voiced. Compare also
Pacific with Basilica, tomorrow with the hypothetical name DeMorrow.
One must conclude, then, that (2a) is not a correct condition for
aspirated (i.e., [+spread glottis]) articulation of the English voiceless
stops.
B. The next question to consider is why the single segment /s/ ap-
pears in (2b). The answer would seem to be that /sf is the only consonant
that forms syllable-initial clusters with following voiceless stops. If
this is true, the aspiration rule should not have to specify /s/. Par-
tial confiration of this hypothesis is available in the pronunciation of
the show-business term of Yiddish origin shtick, the /t/ of which must be
unaspirated.
Thus at least in monosyllabic words, voicelss stops that precede the
73
vowel in their syllable are aspirated if and only if they are initial in
their syllable. These observations suggest rule (3):
(3) /p, t, k/ are aspirated if and only if they are syllable-
initial
Notice that such a rule accounts for the aspiration of the intial stops of
tomorrow, Pacific, collide, since these stops must be syllable-initial.
We will see that a slight modification of (3) is required.
C. Let us consider next words in which /p, t, k/ have a vowel some-
where to the left and somewhere to the right of them, i.e., words for which
we must turn to the rules of Chapter I to determine the syliabic membership
of the stop. Relevant examples, together with the syllable structure as-
signed by Rules I-V of Chapter I, are given In (4):
(4) a. support b. capon c. happy
a 5 p 6 r t k p"an h m'pI
V \V/ V \V \4V
S S S S S S
d. aspen e. asparagus
aDsp e n a a p h r a g a s
S S S S S S
Of the five cases illustrated, only in the first two is aspiration ob-
served. Here the /p/ is syllable-initial, which is in accordance with
the generalization of (3). In (4d, e), the /p/ is not syllable-initial,
nor is it aspirated, again as predicted by (3). (4c), however, presents
a problem for (3) as ft is stated, since the /p/ of happy is syllable-
74
initial, yet unaspirated. The key observation is that in this case /p/ is
simultaneously syllable-final, a condition which does not hold for the
aspirated cases. Thus the rule for aspiration of voiceless stops appears
to be (5):
(5) /p, t, k/ are aspirated if and only if they are both
syllable-initial and non-syllable-final.
Or, in the notation introduced in Chapter I, and in terms of distinctive
features,
(6) Aspiration
continuan 
.[+spread glottis]
S
In the notation of (6), the presence of the lower x serves to insure
that /p, t, k/is syllable-initial, the upper x that /p, t, k/ is not syl-
lable final..
Note that although all aspirated stops have a following vowel, the
vowel need not be mentioned in rule (6) since conventions (8a-c) of Chap-
ter I insure that a syllable-initial consonant must have a vowel somewhere
to the right in the segmental string. Furthermore, it is not necessary
that the vowel follow immediately. The non-syllabic segments that can
follow the voiceless stops in intial clusters are /1, r, w, y/ (except
/tl/, and in many dialects /ty/) and they do not impede aspiration: play
pray, puerile, pew [ph].
In summary, voiceless stops are aspirated after juncture, i.e., when
75
their only syllable membership is initial. More evidence for the correct-
ness of this claim will become apparent as we proceed.
D. Note that in a word like happy, which is initially syllabified
by Rules I-II in the way shown in (7),
(7) 
h m P
\/V
SIS2
the voiceless stop would be aspirated were it not for Rule III, which in-
troduces a line of syllabic association between S1 and /p/, thereby remov-
ing it from the domain of the aspiration rule. (Recall that Rule III was
introduced not for this purpose but rather to account for the fact that in
words like happy, hobby, hammer, the medial consonant is phonetically ambi-
syllabic.)
Observe further that in unnaturally slow speech, in which Rule III
does not apply, i.e., in which the syllabication of (7) is correct, /p/
is aspirated. Thus the aspixation rule (6) is perfectly general and needs
no further conditioning in terms of style of speech, etc. In addition,
an example such as this illustrates that an aspiration rule stated purely
in terms of segmental environment cannot be adequate, for it would fail
to capture the fact that /p/ in slow-speech happy is aspirated and that
it is aspirated for just the same reason that /p/ in p4k is aspirated.
E. Aspiration rule (6) straightforwardly produces correct results in
a wide range of cases. In addition to the five types of examples illus-
trated in (4), (6) correctly predicts the presence or absence of aspira-
tion in words like pin, spin, Pacific, clap, clasp, claps, clasps. There
76
is a small set of examples, however, in which the lack of precision in
the statement of Rule IV leads to possible difficulties. In some of these
cases, aspiration is actually optional. I discuss these cases for complete-
ness; I do not believe they challenge the correctness of the aspiration
rule given in (6).
I examine first a case in which Rule IV, though potentially applica-
ble to a voiceless stop, is not relevant to the operation of the aspira-
tion rule. In words of the form /...VTCV.../, the output of Rules I-III
(III is actually not applicable) is as exemplified in (8):
(8) Catford lapsing
kwm t falr d ps SL
S1 S2 S1 S2
Rule IV, if it applied, would introduce a line of syllabic association
between the voiceless stop and S2 , making the stop syllable-initial. How-
ever, since the stop remains syllable-final in any case, (6) should not be
applicable, and, iin fact, no aspiration is observed. (It is my impression
that the [p1 of lapsing tends to be ambisyllabic but not the [t) of
Catford, implying that /ps/ but not /tf/ is a member of the set of per-
missible initials of Rule IV.)
Next consider words of the form /...VCTV.../ in which C is such that
/CT/ is not a permissible initial cluster. Rules I-III produce (again III
is not applicable):
(9) V C T V
V V
SI S 2
77
Here the operation of Rule IV is crucial, since /T/ meets the structural
description of the aspiration rule only if IV does not act to associate
C with S2'
Words with a double medial stop before an unstressed vowel are of
the type under consideration. In words like Abco (company name), bodkin,
I assume Rule IV is inapplicable due to the universal restriction against
initial clusters like /bk/ and /dk/; cf. Section 5.5 of Chapter I. This
result is unproblematical, since there is no intuition of ambisyllabic
/b/ or /d/, and /k/ must be aspirated.5
A more interesting situation arises in words in which both stops are
voiceless, for example, napkin, Atkins, the pre-IV structure of which is
shown in (10):
(10) n w p k L n
S1 S2
It is the second stop that is of interest, for as we saw in the discussion
of words like lapsing, the initial stop will not meet the structural de-
scription of the aspiration rule whether IV applies or not. Initial
double voiceless stops are found in many languages (for ex., Russian,
fast-speech English - see rule (16) of Chapter I) and thus cannot be ruled
out on universal grounds.
In slow speech, Rule IV is not applicable, and the 1k/s of napkin,
Atkins are aspirated, as predicted by (6). In faster speech, I find as-
piration optional, with a preference for unaspirated /k/. Rule IV, then
may be optional in cases like these. However, in the absence of a clear
78
intuition that the first stop is ambisyllabic just when the second is un-
aspirated, I will refrain from claiming the behavior of /k/ in napkin,
Atkins as evidence of the correctness of the postulated system of rules.6
Turning to two final cases, consider the /t/ of words like after and
Washington. As mentioned earlier in connection with after in the discus-
sion of Rule IV in Chapter I, there is a strong intuition of ambisyllabi-
city with respect to /f/ in normal-rate speech (i.e., after is syllabical-
ly equivalent to aspen), indicating that Rule IV is applicable in this
case, which is consistent with the observation that /ft/ is not a univer-
sally proscribed initial cluster. The line of association introduced by
Rule IV removes after from the input of the aspiration rule, and in fact
the /t/ of after is not aspirated except in very slow speech.
In Washington, on the other hand, /t/ must be aspirated, a fact which
can be accounted for under the assumption that Rule IV fails to apply in
this case since it would give rise to initial /t/. Consistent with this
account is the fact that there ib no intuition of ambisyllabic /o/ - the
contrast between after and Washington in this regard is quite striking.
F. In summary, it may be necessary to acknowledge that there is an
idiosyncratic component to the operation of Rule IU (unexpected exclu-
sions from the set of permissible clusters, unpredictable optionality
of operation). Howe-er one should keep in mind that the major hypothesis
being examined in this chapter is that phonetically correct syllabications,
however obtained, are a major conditioning factor in phonological rules.
There are no serious challenges to the aspiration rule of (6) as evidence
for this hypothesis.
79
1.2 Unreleased Stops and Glottalization
A. "Release" of a stop refers to the reopening of the vocal tract
after complete blockage. Now obviously all stops are eventually released,
so it is important to understand the way in which the terms "released"
and "unreleased" stop are used.
In my most natural pronunciation of a word of the form #XVt#, like
mat, the vowel is terminated by cutting off the airstream by what appears
to be simultaneous closure of the glottis and placement of the tip of the
tongue on the alveolar ridge. Due to the glottal closure, no significant
amount of pressure is built up behind the oral cavity occlusion. Thus
when the tongue-tip is dropped, which may occur almost immediately, or,
if /t/ is utterance-final, may be after a rather long interval of time,
there is no sound produced. Such a stop is unreleased in the sense that
the eventual release is irrelevant to the acoustic form of the phone.
There is another way in which a stop may be unreleased. There may
6 be no glottal closure and hence a positive pressure behind the occlusion,
but if the stop is utterance-final, the pressure may remain unreleased
until the velum is lowered to resume normal breathing. The nasal release
is silent and such a stop is considered unreleased. This sort of articu-
lation seems to be common in the case of final /p/, although /p/ also may
involve glottalization.
Returning to the pronunciation of mat described above, if there is
glottal closure and little or no pressure build-up behind the tongue in
this articulation of /t/, is it distinct from a pronunciation involving
glottal closure and no raising of the tongue, [m?]? The answer is yes,
significantly so. As the tongue-tip is raised to alveolar position, the
80
oral resonances, which-had corresponded to the tongue position for [M],
begin to change, and have changed significantly by the time the oral cav-
ity oscillations have died out due to the glottal closure. In [ma'], on
the other hand, the oral resonances remain essentially constant, since
there is no movement of the tongue. I use the symbol [t?] to indicate
simultaneous alveolar and glottal closure.8
In contrast to the /t/ of mat, in a word like stem there is no glot-
tal stricture; pressure is built up behind the alveolar obstruction, and
is suddenly released as voicing for the vowel begins.
Returning to the phonetic nature of unreleased /t/, the following
quote from Heffner (1950:13'6) is of interest:
If the breath stream is wholly occluded by the vocal bands wh..le
it is also stopped at any one of the positions possible for the
tongue or lips, the result is a glottalized stop. The air con-
fined between the glottis and the lingual or labial occlusion may
by (a) compressed, (b) rarefied, or (c) merely confined.
Heffner goes on to describe the compressed glottalized stops, or ejectives
/p', t', k'/, produced by a raising of the larynx, and the the rarefied
glottalized stops, or suction stops /'p, 't, 'k/, produced by a lowering
of the larynx, but has nothing further to say about his third category,
in which there is closure of the glottis but no raising or lowering of the
larynx. It would seem that the unreleased /t/ I have described fits into
this category.
I would suggest, again making use of Halle & Steven's (1971) laryn-
geal features, that both of the unaspirated stop types described above,
the released and the unreleased, are [-spread glottis], but that the re-
81
leased stops are [-constricted glottis] while the unreleased stops are
[+constricted glottis]. The released and unreleased unaspirated voiceless
stops, as well as the aspirated stops, would all be [+stiff vocal cords,
-slack vocal cords].
Let us now proceed tc investigate in a systematic way the environments
in which the two variants of unaspirated stops are found.
B. I consider first /p, t, k/ in utterance-final (i.e., pre-pausal)
position, and discuss two subcases, with preceding vowel and with preced-
ing consonant.
It seems to me that there exists a clear distinction in the behavior
of the three stops in the first subcase, V - Pause, exemplified by
words like mat, map, Mac. I find that /t/ must be unreleased, with a re-
leased /t/ sounding over-precise and artificial. /p/ also tends to be
unreleased, but release is not as unnatural as in the cas4 of /t/. /k/
on the other hand definitely tends to be released. I have no explanation
for this state of affairs, but it is important for what follows to state
the facts explicitly.
The second subcase involves the behavior of the voiceless stops in
the environment C - Pause. Where C is an obstruent, release seems to
be required: apt, list, clasp, ask. When C is a sonorant, though, unre-
leased /p, t, k/ seems to be the norm in many cases. However we must pay
careful attention to phonetic detail here if we are not to be led astray.
Consider first cases where nasal consonants appear to precede the
stops: c , can't, hank. The facts here appear to be just identical to
the V - Pause case: /t/ and /p/ strongly tend to be unreleased, /k/ to
be released. Now in words of this type, the vowel is clearly nasalized
82
and it is difficult to tell whether a nasal consonant is actually present.
Malecot, who studied this problem instrumentally, reports (1960) that in
pairs like hit/hint, back/bank, _cap/cam, the distinction resides almost
entirely in the nasality of the vowel: [httJ vs. [htt], etc. Thus the
case in question, which appears to be of the form C - Pause, reduces to
the V - Pause case, where V is here a nasalized vowel, and we have a
potential explanation for the fact that pre-pausal stops behave identically
after vowels and after what appear to be nasal consonants.
After /1/, on the other hand, I find it possible, though not obliga-
tory, to release all the voiceless stops, even /t/: alp, belt, milk. I
would relate this to the release of /p, t, k/ after the obstruent conso-
nants (on the noted possibility of non-release after /1/, see below), and
generalize that voiceless stops are released in C - Pause.
The remaining sonorant is /r/, which works just like the vowels: re-
lease in heart is unnatural, in harp somewhat less unnatural but clearly
not preferred, and in hark greatly preferred.9
At this point it is once again necessary to consider questions of
phonetic detail, this time with respect to the properties of English /r/.
To avoid a lengthy discussion I have included this material, wE '.ch will
be needed elsewhere in this dissertation as well, as Appendix I to this
Chapter. The principal result is that the English /r/ of rain, heart,
etc., is a glide, not a sonorant consonant. The features that distinguish
/r/ from the other glides involve specification of tongue shape only.
- Once /r/ is recognized as a glide, the observed pre-pausal allophones
of /p, t, k/ in harp, heart, hark can be seen to follow from a more gener-
al fact. Observe that it is not only when following a pure vowel, as in
83 .
mat, map, Mac, that /p, t, k/ take the allophones described above, but
also when following a diphthong: tight, type, tyke. Thus the leftward
environment must include vowels and glides, that is, must be [-consonantal]
(see Appendix 1). Words like heart, harp, hark, then, come under this
environment, and the fact that one observes the same allophones of
/p, t, k/ here as in tight, etc., can be taken as additional evidence for
the claim of Appendix 1 that English /r/ is [-consonantal].
Let us summarize now the results of our investigation of the pre-
pausal allophones of /p, t, k/. Following a [-consonantal] segment, for
example the oral vowel /w/ of mat, the nasal vowel // of can't, the
glides /y, r/ of height, heart, /t/ is unreleased, /p/ is preferably un-
released, /k/ is preferably released. Following a [+consonantalj segment,
for example the stop /p/ of apt, the fricative /sf of list, the sonorant
/1/ of belt, all three stops are released.10 From this point on I will
limit my attention to /t/. As mentioned earlier, I have no explanation
to offer as to why the three stops behave differently in the pre-pausal
environment.
Assuming that the underlying form of /t/ is the allophone that is not
glottally stopped, we could state the result of the above observations
as rule (11):
(11) [-cons] t Pause
[+C.G.J (C. G. = constricted glottis)
This is only a tentative statement of the rule, for additional results will
make it possible to derive (11) as a special case of a more general rule,
84
as well as to eliminate the imprecise term "pause."
C. I turn now to the behavior of /t/ in the environment - [-syl].
In words like cats and atlas, but not mattress and matriculation, there
appears to be glottalization of the pre-consonantal /t/, as was the case
with pre-pausal /t/.
Pre-consonantal glottalization har the same left environment condi-
tion as pre-pausal glottalization. As opposed to cats, where glottal
stricture accompanies the ore.1 articulation of the /t/, in a word like
casts the glottis is completely inactive during the articulation of
/sts/.11 Furthermore, it is only [+consonantal] non-syllabic segments
that prevent glottalization. Compare casts [...sts] with heights [...yts]
and hearts [...rt'sJ.
Considerations of syllabic structure provide a ready explanation for
the observed difference between cats/atlas and mattress/matriculation.
Only in the latter case will /t/ be syllable-initial by Rule II. The ap-
plication of Rule III to mattress establishes a syllabic link between /t/
and the first syllable, but of course the original association as first
element of the second syllable remains:
(12) k a t s m t 1 aB m a t r s S m a t r Lk...
\V/ V \V VV V NV/
S S S S S S S
Thus the crucial observation is that the pre--consonantal /t/s which glot--
talize are non-syllable-initial:
(13) [-cons] ] [-syl]
[+C.G. I
85
(13), like (11), is a tentative statement, to be generalized when addition-
al facts are taken into account.
D. The final environment to be considered is - [+syllabic]. In
this environment, glottalized stops are never observed: ten, stem, attic,
Bostun, attack. Now Rile II will always assign a pre-vocalic /t/ to the
tollowing vowel's syllable, and once again, whether or not Rule III applies,
/t/ retains its [+syllable-initial] structural descr4otion. Thus here as
before syllable-initial /t/ fails to glottelize. Although the glottali-
zation rule could be written in such a way as to exclude the environment
-- [+syllabic], since pre-vocaltc /t/ never glottalizes, doing so would
obscure the simple generalization that before segments of all types, post-
non-consonantal /t/ is glottalized when nol. syllable-initial. Additional
strong evidence that this generalization is correct will be presented in
Section 1.3.
E. (14) is a summary of the environments in which we have discovered
glottalized /t/:
(14) a. [-cons ] - Pa:Ee
b. [-cons] [-syl]
As we have seen, the [-syllabic] specification can be removed from (14b)
since /t/ will never be [-S.I.] when it is followed by a [+syllabic]
segment:
(14b') [-cons]
86
But now observe that (14a) is just a special case of (14b') and can thus
be dropped from a general glottalization rule. (15), then, accounts for
all the occurrences of glottalized /t/ that have been mentioned (and all
known to me):
(15) [-cons]
[+C.G.]
1.3 Summary and Extension Beyond Word Level
A. We have now isolated two environments, the statements of which are
exceedingly simple in tarms of syllabic structure, where particular allo-
pnones of /t/ are found. Repeated here, they are12
(16) Aspiration
t -- [+spread glottis]
x
x
S
(17) Glottalization
[-cons] t -* [+constricted glottial
x
S
As has been pointed out, the other voiceless stops, /p, k/, follow the
same aspiration rule as /t/, while the glottalization rule seems to me to
be fully accurate only for /t/, although it also predicts /p/'s behavior
corrictly in most cases.
B. We have thus far limited our investigation of the distribution of
87
aspirated and glottalized /t/ to strings which do not exceed the domain of
the word. When this limitation is dropped, additional strong evidence in
favor of the syllabic approach becomes apparent.
We saw in Chapter I that there is a need for a rule which associates
a final consonant with the initial syllable of a following vowel-initial
word. We might ask whether this rule, V, interacts with either the as-
piration rule or the glottalization rule.
C. Since Rule V does not apply in the case of consonant-initial
words, words with initial /p, t, k/ will remain syllable-initial and non-
syllable-final in connected speech, so that aspiration of initial voice-
less stops is predicted by (16) for all speech-rates. This is a correct
prediction.
When /p, t, k/ is final, Rule V will associate /p, t, k/ to a follow-
ing word-initial vowel in connected speech. However, whether or not Rule
V applies to make /p, t, k/ syllable-initial, its original syllable-final
structure must remain, excluding it from the domain of the aspiration rule
(16). This is also a correct prediction, for whether tap Ann, hit Ann,
shake Ann are pronounced with or without pause, /p, t, k/ must not be as-
pirated.
Finally, word-internal /p, t, k/ is unaffected by the operation of
Rule V and thus displays the same allophone whether in an isolated word
or in a connected phrase.
Thus (16), derived by means of a consideration of words in isolation,
holds on all domains. We have now discovered four major advantages for
the syllabic analysis (16) of English aspiration over the traditional seg-
mental one (2): a) more accurate description of the facts (recall discussion
88
of t6morrow, etc.); b) the availability of a single, simple, potentially
explanatory rule covering all cases; c) the availability of a simple
rule, not explicitly conditioned by speech-rate, which automatically
accounts for the alternation between slow-speech aspiration vs. norm3l-
speech non-aspiration of medial stops like that of happy; d) the lack
of any need to limit the rule to the domain of the word: if this limita-
tion is not imposed on (2) it will predict aspiration in tap Ann, for
example.
Of these four difficulties surrounding (2), note that by means of an
improved segmental version one can hope to overcome only the first. In
regard to (c), observe that although any need to condition the syllabic
aspiration rule with respect to speech rate is eliminated by introducing
this conditioning elsewhere in the grammar, viz., in Rule III, the latter
rule is independently needed, so that (c) does indeed count as a major in-
dication of the correctness of the syllabic approach to aspiration.
D. Let us now examine potential inputs to the glottalization rule
which go beyond the domain of the word.
As we have seen, word-initial /t/ is always syllable-initial and thus
should never be glottalized, even in connected phrases, if (17) is correct
as it stands. Initial /t/ is in fact never glottalized.
Final /t/ is always non-syllable-initial in isolation, and thus is
glottalized if preceded by a non-consonantal segment. In connected phrases,
however, Rule V will introduce a line of association to the initial sylla-
ble of a vowel-initial word. We thus predict - and in fact observe -
a) glottalization of the /t/ of hit Ann if and only if spoken with pause;
b) glottalization of the /t/ of hit Bill, hit Ron whether or not spoken
89
with pause.
Thus (17), like (16) is a single, simple rule which describes the
behavior of /t/ both internally and at word-boundary, both in isolated
words and in connected speech.
Recall that in subsection 1.2D I observed that glottalized /t/ is
never found before vowels (only words in isolation were under considera-
tion), but suggested that the glottalization rule, rather than be written
to explicitly exclude this environment, make use of the fact that /t/ is
never [-S.I.] before vowels in order to subsume the [-syllabic] and
- [+syllabic] environments under a ,single generalization. We now have
additional evidence for the correctness of this choice. Were the glotta-
lization rule written in such a way as to predict no glottalization in
the environment - [+syllabic], it would have to be limited to words in
isolation, for the pre-vocalic /t/ of hit Ann is glottalized if the phrase
is spoken slowly and distinctly (i.e., in such a way that Rule V does not
apply). (17), on the other hand, is correct as it stands for all domains
and all speech styles.
One further observation on (17) might be made. In the discussion of
pre-consonantal /t/ in 1.2C, it was noted that cats and atlas, but not
mattress and matriculation, have glottalized /t/. This observation was
related to the differing syllabic structure of the two pairs of words,
not to the type of non-syllabic following /t/. Here also an advantage
beyond formal simplification of the isolated-word rule is apparent. Were
the non-glottalization of the /t/ of mattress accounted for by reference
to the following /r/, the behavior of /t/ in both mattress and hit Ron
could not have been accounted for by means of a single rule.
90
E. It must be realized that the presence vs. absence of word-bound-
aries cannot be used to condition the glottalization rule and thus account
for the unglottalized /t/ in mattress vs. the glottalized /t/ in hit Ron
without reference to syllabic structure. This is so because the /t/ in
a phrase like hit Ann exhibits these same two allophones, yet is syntac-
tically the same whether spoken over-distinctly (i.e., i such a way that
Rule V does not apply) or normally (i.e., with application of V). 1 3
In order to see more clearly the nature of the difficulty encountered
by an analysis of glottalization in terms of word-boundary, consider some
comments on "glottal reinforcement" (the process whereby /t/ becomes /t9/
or /9/14) from Selkirk (1972). On p. 196, she states that "glottal rein-
forcement occurs when a voiceless stop is preceded by a vowel, liquid, or
nasal, and followed by another consonant" and15 that it also takes place
"in pausa, i.e., - ##j." But of course "in pausa" and "-- ##" are not
equivalent. In a phrase like hit Ann, the /t/ will be glottalized if and
only if there is a pause between hit and Ann, while, as noted above, hit
and Ann are separated by ## whether spoken over-distinctly or normally
(cf. also fn. 13). Note that even in the sentence of (18), where there
is a maximally strong syntactic boundary between hit and Ann,
(18) Just as he was looking for someone to hit, Ann walked in
the same facts obtain, glottalization if and only if juncture. In fact,
one can get the non-glottalized allophone of /t/ when a vowel follows /t/
even in a syntactically independent sentence:
(19) Joe is liable to get hit. Ann doesn't stand for any nonsense.
91
Rule V, which is oblivious to syntactic boundaries, is as free to apply
here as in hit a ball, although there may well exist extra-grammatical
considerations which lead to a higher incidence of its operation within
sentences.
E. In summary, it should be clear that the type of structure which
conditions the aspiration and glottalization rules is independent of syn-
tactic structure, and thus that correctly written rules for these phenom-
ena will apply on any syntactic domain.1 6
1.4 Underlying Representation of the Voiceless Stops
When not aspirated or glottalized by Rules.(16)-(17), /p, t, k/ sur-
face, with one important ex;eption, as released unaspirated unglottalized
voiceless stops [p, t, k]. In view of the wide variety of environments in
which such stops are found (for example in spin, happy, clasp, clasps,
ta/ Ann, I will asume that /p, t, k/ are underlying, that is, that in
lexical representations the voiceless stops carry the feature specifica-
tions shown in (20):
(20) -continuant
-spread glottis
-constricted glottis
+stiff vocal cords
L-slack vocal cords
The important exception referred to above concerns /t/ and is the
subject of the next section.
Li "Flapped" /t/
A. A historical change which is unusual in having apparently affected
all American dialects and no current British ones involves the addition of
92
a rule which turns /t/ into an alveolar flap or tap in certain positions.
In this section I arrive at a general statement of the environment in
which this process occurs and show the relevance of the rule to the theory
of syllabic phonology.
B. "Flapped" /t/, for which I will use the symbol [D, is heard in
the American pronuciation of words like butter, city. "Flap" refers to
the characteristic motion of the tongue-tip in articulating [D], which Is
quite different from normal stop articulation.
The literature on the phenomenon of "flapped" /t/ reveals a large
degree of conf-sion as to the phonetic nature of this segment and its
dialectal variation. In Appendix 2 to this chapter, I try to summarize
what can safely be said about the phonetics of "flapped" /t/ and provide
an adequate description for the needs of this section. I turn now to a
discussion of the distribution of [D].
/t/ is usually described as flapping in the environment V - V.
The requirement that the second vowel be unstressed can be exemplified
by the pair liter [DI vs. latex [thi. The failure of /t/ to flap in
latex further shows that it is not only a following main stress that pre-
vent flapping, but in fact any [+stress] vowel. It -also shows that the
absence of flapping in a word like attack is due not to the [-stress]
specification of the first vowel but rather to the [+stress] specifica-
tion of the second. In fact, the [+stress] requirement on the first vowel
of the environment given above appears to be simply in error, the result
of looking at too small a sample. For example, the /t/ in words such as
capital, obesity is flapped.1 7
We are thus led to the environment V -- [srJ. The next point to[ Vt
93
be considered concerns the possibility of an optional non-syllabic segment
between the focus and either of the vowels.
To the right, such an additional segment, even a glide, is out of the
question: [D] is impossible in words like Atkins, atlas, mattress, amateur
(even when not [...c...).
However, the occurrence of [D] in such words as loiter [loyDar],
shouting [SawDLQ], shows that at least a glide is permissible to the left
of focus. Furthermore, words such as forty and winter have flapped /t/.
Selkirk (1972) discusses the flapping of /t/ and /d/ and states the
environment for this process as (21) (p. 197)
(21) )- 
-stess
Apparently her omission of /r/ from the left environment is a mere over-
sight, for in "/r/-retaining" as well as in "/r/-dropping" dialects, un-
flapped /t/ in parting is as unnatural as it is in potting.18 On the other
hand, one of the environments she does include is far from universal. Many
speakers who never fail to flap in all the environments so far discussed
have unflapped /t/ in Vl - V (malted, shelter).
It seems clear that a fully adequate restatement of (21) requires
reference to the observations of Section 1.2B, where it was noted that
although glottalization of /t/ appears possible between sonorant and pause
in many cases, all such examples reduce to the environment [-cons] -
Pause, when phenetically accurate observations are made.
As argued in Appendix 1 and elsewhere in this Chapter, English /r/
must be viewed as a glide. Thus the examples better, loiter, shouting,
94
parting are all covered by the rule in (22):
(22) t--PD / [-cons] - +syllabic
-stress
(Note that the lack of specification for the feature [syllabic] in the
leftward environment allows both glides and vowels in this position. Of
course, when the [-cons] position is filled by a glide, a vowel is required
to its left, but this requirement need not be stated in the rule, for the
sequence [C, #]-glide-/t/ does not occur.)
Now in words like winter, we appear to have a [+cons] segment, /n/,
before flap. But as with words like hint, this /n/ nasalizes the preceding
vowel and then may be deleted. In the flapped pronunciation of winter /n/
has in fact been elided: [wtDer]. In fact, many Americans have two pro-
nunciations of words like winter, [wtnter] in careful speech, and ['wDar]
otherwise. It is important to note that this variation cannot be attributed
to any optionality of the flap rule, for where no /n/ precedes the candidate
for flap, as in better, unflapped /t/ is unnatural even in very careful
speech. Thus it appears that nasal-consonant loss is an optional (but
preferred) rule, and that upon deletion of the /n/, /t/ finds itself in
the environment for (obligatory) flapping.19
Since nasalized vowels are certainly [-consonantal], Rule (22) covers
the flap in winter. Remaining to be discussed is the possibility of flapping
after /l/,as in shelter. If /1/ is consonantal, i.e., if the tip of the
tongue contacts the roof of the mouth in its articulation, flap seems to
me to be simply impossible; i.e., the result is a sequence not heard in
any American dialect. I would relate the tendency of many, including my-
95
self, not to flap /t/ after /1/ to our tendency to maintain a consonantal
pronunciation for /1/. When I do flap the /t/ of words like shelter, I
must produce a non-consonantal /1/, and I believe that this is what is
universally done by those speakers who habitually have flapped /t/ in this
position. Cf. footnote 10.
Thus (22) seems to be an accurate account of the cases discussed so
far. Let us now consider initial and final /t/, respectively preceded
by vowel- (or glide-) final words and followed by vowel-initial words, to
determine whether flap occurs across word boundary.
C. Where /t/ is initial, it does not flap even if the requirements
of (22) are otherwise met: buy tomatoes cannot be pronounced [bayDa...].
However, word juncture following /t/ does not impede flap. The nor-
mal pronunciation of get#a horse and get##Annette home on time is [geDe...],
suggesting that (22) must be extended to (23):
( [U) +syl lab ic
(23) t --9D / [-cons] --- ( ) strL-stress j
This treatment leaves as an arbitrary fact the presence of optional word-
boundaries on only one side of focus. However there is an even bigger my-
stery concerning the flapping of word-final /t/.
I noted an absolute restriction against flap of pre-stress /t/ when
only words in isolation were under consideration; attack and latex, tor
example, cannot be pronounced with [D]. Yet across word-boundary, /t/ can
be flapped if pre-vocalic even if the following vowel is stressed:
get##Ann home on time also has [D]. In addition, note that for dialects
in which latter - ladder, the phonetically transcribed portion of the
phrase we [s&Dhsker] on the radio can be interpreted as set Oscar or
96
said "Oscar". In other words, the [-stress] condition in (23) can be re-
laxed if and only if the word-boundary option is taken:
(24) t-D / [-cons] -(#(#)> +syllabic
t-stress>
condition: ~a=* b
Although the standard notation for phonological rules allows expres-
sion of the facts as we have observed them, (24) is a highly complex and
arbitrary description of what is after all a natural phonetic process.
Actually, (24) must be further complicated if it is to achieve de-
scriptive adequacy. Recall from Section 1.3D that in a phrase like
hit Ann, the glottalized allophone of /t/ is observed if the words of the
phrase are not allowed to flow together. This is as predicted by Rule
(17), since in such speech Rule V does not apply to link /t/ to the syl-
lable of Ann and thus add the structural condition [+S.I.] to /t/. It
is only when Rule V does apply (i.e., the normal case) that [D is heard
in hit Ann. Furthermore, i the overprecise type of articulation described
in connection with the motivation of Rule Il in Chapter I, flap is not ob-
served word-internally: better [be,ther]. Incorporating these additional
conditions into (24) leads to (25):
(25) t--+ D / [-cons] 
-- #(#)>[(stress)
Conditions: 1. -a =*b _
2. if a, within connected phrases only
3. does not apply in artificially-slow speech
Note that conditions 2 and 3 are not general ones on phonological rules
97
and indeed have to be stated explicitly in (25): compare the SPE rules of
Tri-syllabic Laxing, Velar Softening, etc., which cannot be "inhibited,"
I have looked at the phenomenon of flapped /t/ in detail because
it provides additional evidence for the correctness of the hypothesis
that a class of phonological rules cannot be properly condition' I without
reference to syllabic structure. When two crucial observations are made,
most of the complexity and arbitrariness of (25) evaporates.
First of all, for a given occurrence of /t/ to be flapped, it must
be tautosyllabic with a following [+syllabic] segment. Thus we observe
[D] in hit Ann only if Rule V has applied, associating /t/ with the sylla-
ble of Ann. Note that this tautosyllabicity condition is met in all in-
stances of word-internal [DI as well.
Secondly, only ft/s which have not been aspirated are candidates for
flap. Word-initial /t/ is always syllable-initial and never syllable-final
and is thus always aspirated by (16). On the other hand, word-final /t/
is never aspirated, as we have seen. In this way one can account for the
lack of symmetry mentioned in connection with (23). Similarly, in artifi-
cially slow speech, in which Rule III does not apply, the /t/ of better
will be left [+S.I., -S.F.] and will be subject to (16), while in normal
speech, III bleeds Aspiration. Finally, within words, /t/ following a
[-consonantal] segment will be aspirated in pre-stress position in all
cases (cf. Rules Ila, III, and (16)), while across word-juncture, /t/
is never aspirated, even if a stressed vowel follows (cf. Rules IIb, (16)).
Thus if the input segment to the flap rule is taken to be a simple
unaspirated /t/, and if fusthermore this rule ia ordered after Aspiration
(16), (25) reduces to (26):
98
(26) t--> D / [-cons] - [+syl]
S
In summary, if the various environments in which /t/ cannot be flapped
are examined, it will be found that they fall into two categories, those in
which /t/ will have been aspirated by Rule (16) (tack, attack, latex,
buy tacks, buy tomatoes), and those in which the structural description of
(26) is not met (tack, stem, mats, mast, Boston, Atkins, hit//Ann). I
would claim then that (26), ordered after Aspiration (16) and applicable
on all syntactic domains, is a correct account of /t/-"flapping" in Ameri-
can English, and that the complexity of (25) is a direct result of the
failure to recognize the phonological significance of the syllable.
1.6 "Sonorization"
A. I summarize now the results so far achieved in our study of the
allophonic development of /t/.
(27) Aspiration
continuant -+[+spreadglottis]
+stiff V. cl
x
S
Syllable-initial non-syllable-final voiceless stops are
aspirated.
99
(28) Glottalization[-continuant
[-cons] +stiff v.c. j-- [+constricted glottis]
+coronal
S
Non-syllable-initial /t/ is glottalized following a non-
consonant.
(29) Flap (ordered after (27))
-continuant
[-cons] +stiff v.c. [+syllabic]
L+coronal J
-stiff v.c.
L+sonorant S
/t/ is flapped between a non-consonant and a tautosyllabic
vowel.
I am not sure how to characterize fully the flap [D] in distinctive
features. I represent it in (29) as [-stiff v. c.] to indicate that it
is voiced. The sonorant specification is included in order to capture
the tapped articulation, basea on the general classification of trilled
/r/s as sonorants. Further specification may be necessary, for example
to represent the fact that [D] is a single-tap trill.
B. The fact that Rules (28) and (29) each have [-consonantal] in
the leftward environment suggests that a further generalization might be
captured. The following remarks represent speculation on my part as to
what is involved.
Observe that flapping will always and only occur in the configuration
100
of (30):
(30) [-cons] t [+syll
S S
To see that whenever /t/ is flapped it is ambisyllabic, note that if line-
of association a were not present, /t/ would be [+S.I., -S.F.] and there-
fore would be aspirated, while if b were not present, the environment for
(28) would be met and /t/ would be glottalized. Although (30) holds in
exactly those cases in which /t/ is flapped, the question of interest is
how much of the specification of (30) is actually contained in the flapping
rule.
Next note that what Glottalization and Flap have in common is the
configuration of (31):
(31) [-cons] t
S
Beyond this distributional similarity, flapped and glottalized /t/s have
something in common phonetically.
As mentioned above, taps and trills are always described as sonorants,
sounds whose articulation does not result in any significant pressure
build-up in the vocal tract, and this characterization seems to be an ac-
curate one. [D] then should be classified as [+sonorant]
Now as pointed out in Section 1.2A, the glottalization of /t/ in words
like mat deprives the alveolar obstruction of the opportunity of exercising
its normal function of cutting off the airstream and holding back the re-
sultant pressure build-up. Whether or not it is the glottalization in
101
[t] that is the primary cause of the absence of pressure build-up, the
fact remains that if we follow our definition of sonorant strictly, the
[C0] of mat is [+sonorant]. 2 0
Since glottalized and flapped /t/, but none of /t/'s other allophones,
are phonetically sonorants, it follows that /t/ surfaces as [+sonorant]
exactly when it stands in the configuration (31). This observation sug-
gests a rule of "sonorization," preceding the rules of Glottalization and
Flap:
(32) Sonorization2 1
-continuant
[-cons] +coronal -+.[+sonorant]
[+stiff v.c.
S
Any output of (32) will be subject to either the Glottalization rule,
which adds the feature specification [+constricted glottis], or the Flap
rule, which specifies [-stiff v.c.] (and is thus more properly a voicing
rule). The requirement for "Voicing" is simply that a vowel follow, in
the same syllable, while in the inverse of this environment, the glottali-
zation rule applies. Thus Voicing and Glottalization are as stated in (33):
(33) Voicing/Glottalization
[-stiff v.c. / - [+syl]
t t (a)
[+son]j/
[+constricted glot.] otherwise (b)
Rule (33) produces a glottalized /t/ in all those cases in which /t/ has
102
been made [+sonorant] by (32) except where a vowel follows in the same
syllable. There are thus several ways in which glottalization can occur:
a) in utterance-final position: I saw the cat; b) pre-consonantally:
cats, the cat sipped; c) before a word-initial vowel precede by pause:
the cat // ate.
If a following vowel is in the same word, /t/ will never be glottal-
ized, since Rule II insures that /t/ will be syllabified with the vowel,
thus triggering Voicing. Across word-juncture, syllabic linking in ordi-
nary speech is established by Rule V, with the consequence that we get (D
not [t?] in this context, unless a pause blocks Rule V.
C. It should be understood that if this analysis in terms of Sonori-
zation is correct, the lack of pressure build-up behind the [t?] of mat
has as its direct cause the altered type of tongue-tip articulation intro-
duced by the sonorization rule, not the simultaneous constricting of the
glottis. Note that we know independently that coronal articulation is
consistent with the feature specification [+sonorant] as well as with
[-sonorant]. Coronal sonorant consonants can arise in ways which are not
of direct interest to us here, as in the case of /n/, in which the lowered
velum prevents pressure build-up. Directly relevant, though, are Spanish
/r/ and /rr/ (single and multiple-tap trills, resp.), whose sonorant (no
pressure build-up) nature can only be the result of the way in which the
tongue-tip is controlled. /D/, which is very similar to Spanish /r/ (cf.
Appendix 2) clearly also has this special type of coronal Articulation as
the cause of its [+sonorant] specification. Entailed by the analysis in
(32)-(33) is the claim that this sonorant type of coronal articulation is
imposed on /t/ in American English whenever it stands in the configuration
103
[-cons) -
S
and that /t/ would surface as a voiceless /r/ when not in the Voicing en-
viroiment were it not for the following glottalization rule, which makes
Bernoulli oscillation, even to the extent of one tap, impossible. Although
admittedly speculative, I find this analysis appealing not only for the
formal simplification it allows but also because as far as supra-glottal
articulation is concerned, the /t/s of mat and matter seem very similar;
assigning them to different major classes, obstruent vs. sonorant, simply
seems wrong.
With the proviso that not all the feature specifications may be cor-
rect as they stand, I offer the following analysis of the allophonic develop-
ment of /t/ as an alternative to that given in (27) - (29):
(34) a. relevant feature specifications of the segment underlying
[t], [t h],I [t*], [D]:
-sonorant
-continuant
+coronal
-spread glottis
-constricted glottis
+stiff vocal cords
-slack vocal cords
b. Aspiration (as in (27)
c. Sonorization (as in (32))
d. Voicing (as in (33a))
e. Glottalization (as in (33b))
104
f. Rule Ordering:
b/c then d then e (but cf. fn. 21)
Note that the /t/s of stem and apt will be affected by none of the
rules (34b-e) and thus will surface with the feature specifications of
(34a), i.e., as [t].
1.7 The Alveopalatal Allophones
In presenting an analysis of the allophonic development of /t/,
I have left (what appears to be) the 'easiest task for last. I refer to
the investigation of the alveopalatal variant [c], [ch] of /t/, found in
words such as train, strong in many dialects.
The most obvious observation is that [c, chi occur only before /r/.
Furthermore, the aspirated variant occurs just where other aspirated stops
are found, in [+S.I., -S.F.] position. Examples are train, tremendous,
attract, nitrate. When this condition on syllable structure is not met,
/c/ shows up unaspirated: strong, mattress. Thus it is clear that the
Aspiration Rule (27) is involved in the derivation of [chI.
Alveopalatalization does not occur across word boundary: night rate
[t1. There are two ways of accounting for this fact. In the standard
view of phonological rules, word-boundary is not "transparent" and must
be mentioned in a structural description if it is to be allowed to appear
in input strings. Thus by simply writing the rule as in (35), one cor-
rectly predicts that [cr], [chr] will be found only within words:
(35) t r
c
105
On the other hand, word-boundaries seem not to be relevant in many
low-level allophonic-type rules, as we have seen. Furthermore, since
syllable-structure conditioning appears to be needed in many such rules,
one must also consider the following formulation of the alveopalatalization
rule:
(36) c
t
t r
V
S
To see that (36) gives the desired results, note that within words /tr/
must always be followed by a vowel and that Rule II will always syllabify
/trV/ together. In /...t##r.../ on the other hand, none of the syllabi-
cation rules is applicable, for II through IV apply only within words and
V links final consonants only to initial vowels.
/c/ never glottalizes or flaps. Expressed as (28) and (29), Glotta-
lization and Flap need no special restriction to preventapplication to /c/,
since /c/ will never meet the structural description of these rules.23 In
the reanalysis of (34) we can assume that Sonorization applies only to
anterior coronals (i.e., to /t/ but not to /c/), since otherwise an input
like that of (37) would trigger Sonorization:
(37) mattress
m to e r a s
S S
This would seem to be an entirely natural assumption since coronal taps
106
and trills are usually (always?) alveolar, not alveopalatal.
In summary, the alveopalatal variants of /t/ are accounted for by
supplementing analysis (34) with the alveopalatalization rule (36) (or
(35)), ordered before Sonorization, or by adding (36) to analysis (27)-
(29), with no explicit ordering statements required.
Section 2 - /r/-Loss
2.1 Two of the most discussed phenomena in works on English dialecto-
logy are the widespread loss of "post-vocalic" /r/ and the complex dialectal
treatment of pre-/r/ vowel quality. The variable retention of "post-vocal-
ic" /r/ is the subject of this section, phenomena concerned with vowel-
quality in - /r/ of the next
As is well known, a word like card is pronounced without phonetic /r/
in most of England and Australia and in parts of the eastern and southeast-
ern United States. /r/ is for the most part retained in all phonological
environments in Scotland, Ireland, Canada, the mid-western and western
United States, as well as much of the eastern U. S.
The so-called "r-less" dialects are of course not completely /r/-
less, for no dialect deletes the initial /r/ of rain, for example. The
usual statement is that /r/ is deleted in post-vocalic position, but even
this description is inaccurate, for the /r/ of Mary is not deleted except
perhaps in fast speech in certain dialects. What seems to be behind ref-
erences to "post-vocalic" /r/ is an implicit assumption of syllable struc-
ture: /r/ which follows a vowel and stands fully within that vowel's syl-
lable is subject to deletion: car, card. I believe this view is basically
correct and will incorporate it into my treatment of /r/-loss.
107
While historical pre-consonantal /r/, for example in card, is irre-
trievably lost in the r-less dialects, the traditional description has it
that deleted final /r/ re-appears before vowel-initial words: Mister! [0],
Mister Smith [0], vs. Mister Adams [r] is observed for most r-less dia-
lects. However, before the generative phonologist concludes that the
r-less dialects differ from the r-ful ones only in containing a rule which
deletes /r/ before consonant and pause, he must consider the phenomenon
of "intrusive /r/." An unetymological /r/ is often observed between a
vowel-final and a vowel-initial word in r-less dialects: law and order
[loranoDe]. In fact, it is typically the case in /r/-less dialects that
a pair of words like tuner and tuna are identical to each other in all
environments, for example being pronounced with final [a] in
This - seems all right and with [ar] in This - isn't any good. This
fact suggests a general lexical loss of /r/ in the environment - [C, #]
and an /r/-epenthesis rule in V - V.
It turns out that if one studies carefully the various dialects in
which card is pronounced without /r/, several different patterns emerge,
leading to different synchronic solutions for different dialects. The
one synchronic grammar that seems not to exist is one sometimes taken to
be the normal case in r-less dialects, a grammar which has final /r/s cor-
rectly placed in the lexicon from the point of view -of etymology, and only
one phonological rule involving /r/, a deletion rule before consonant and
pause. That is, there seems to be no naturally-occurring dialect (some
British speakers strive towards this norm) that consistently has [a] in
tuner-Pause and tuner seems and [ar] in tuner is that does not also have
[ar] in tuna is. 2 4
108
2.2 I will discuss below what appear to be the underlying forms and
relevant phonological rules of some common r-less dialects. Let us con-
sider first the historical change which gave rise to these dialects.
There can be little doubt that /r/-loss first came into the language
through the addition of a rule which had the effect of deleting /r/ (or
reducing it to [9J) before consonant and pause. This assumption is uncon-
troversial (see for ex., Kurath 1964:76), and is the only one that allows
an adequate explanation of the patterns of /r/-loss and -epenthesis in
the current dialects.
Thus /r/ was lost in essentially the environment in which /t/ is
glottalized (cf. Section 1). It will be seen that the same sort of facts
which force a syllabic analysis of glottalization make a purely segmental
account of the original loss of /r/ impossible.
Examples of the environments in which /r/ was retained and lost are
found in (41):
(41) a. /r/ retained
red cry Mary correct bar is
r ed k r a y m e rf ke''r e k t b a r L z
\ \\V/ \WV V \V/ NVV
S S S S S S S S
b. /r/ lost
bar seems bar//is bark barker
b a r s i m z b a r t z b a r k b a r k a r
S S S S S
Stated in terms of syllable structure, a very simple rule accounts for
109
this pattern:
(42) [+syllabic] r -+ 0
x
S
According to (42), post-vocalic /r/ is lost unless it also serves as a
syllable-initial segment. The inverse environment is also extremely sim-
ple:
(43) /r/ is retained only in [+syllabic]
S
That is, /r/ is retained only when syllabically linked to a following vow-
el. 25
No statement of /r/-loss stated entirely in terms of segmental and
syntactic-boundary environment could be adequate, for in phrases like
those in (44), /r/'s environment is fixed, yet retention is conditional
on whether the words of the phrase (or discourse) are pronounced in a
flowing or staccato manner:
(44) a. the bar#is
b. the bar##allows
c. [Face the bar] [It....]
Furthermore, within words pre-vocalic /r/ is always retained, so that a
non-syllabic account could not be patched up by adding a condition such
as "I/r/ is retained before vowel except in slow speech." From the point
of view of syllabic structure there is no mystery connected with these
facts. Rule II, which applies in all speech styles, assigns the structu-
110
ral description [+syllable initial] to pre-vocalic /r/ in words, removing
it from the input of Rule (42), while in the phrases of (44), indepen-
dently-motivated Rule V causes /r/ to become initial in the following
syllable in connected speech, bleeding such inputs from (42).
In summary, (42) is a formally simple rule which correctly describes
the behavior of /r/ in the prototype /r/-dropping dialect both word-in-
ternally and -finally and on all syntactic domains.
2.3 I turn now to a survey of some well-known "r-less" dialects and
offer synchronic analyses where sufficient data is available to me.
A. As far as I can see, there is no evidence of lexical retention
of /r/ in southern British English ("R. P.") in those environments in
which it was deleted by (42). The example concerning tuna and tuner
given in Section 2.1 holds for this dialect: both words have [a] before
consonant and pause, [ar] before vowel-initial words in connected speech.
Thus the analysis that seems required is a syllable-structure constraint
(stated in Rule IIb) against post-syllabic /r/, alone or in clusters, to-
gether with a rule of epenthesis, a first approximation to which is sta-
ted in (45):
(45): 0-r / (a, } --- V
Rule (45) introduces [r between words like tuna/tuner [a], fear [-iel,
fair [-ea], and a following vowel-initial word.
Epenthesis quite often takes place after the low vowels [a:] (far,
Shah an [o- (sre,26Shah) and [o:] (sore, saw) in this dialect. However, in spite of this
observation, the fact that in producing final syllabic nuclei in English
ill
that are not followed by [w] or [yJ the tongue tends to glide into a mid-
central position27 suggests that all cases ox /r/-epenthesis may actually
be occurring in the environment a - V. The only potential difficulty
such an analysis would encounter involves words like tuna/tuner, which
would normally be described as ending in [a], rather than [ea]. However
one would not not expect a rule applying in V a - to fail to apply
in a - . I will assume then that if the distinction between [a] and
[at] is more than an artifice of the notation, tuna/tuner ends in [a],
i.e., is [tuwnee], and that aside from a modification to be introduced
directly below, (46) is an accurate description of "linking" and "intru-
sive" /r/: 2 8
(46) 0-+r / a - V
Now /r/-epenthesis is like the rules of Section 1 in that it is sen-
sitive to junctural phenomena other than syntactic boundaries. In the
same way that the /t/ of hit Ann is flapped only if the words are pronounced
as a connected phrase, so epenthetic /r/ is observed in beer#1is, where
#I1 stands for a syntactic boundary of arbitrary strength, only if there is
no phonetic "break" between the words. I propose to account for this fact
in a way analogous to the treatment of sandhi phenomena in Section 1, that
is by placing a syllable-structure condition on the epenthesis rule which
will be met only if Rule V of Chapter I has applied:
(47) /r/-Epenthesis
a V
S
r
112
Note that in addition to the imposition of a syllable-structure condition
in (47), the latter rule differs from (46) in taking [] to fr] rather
than inserting [r] after [a]. (47) seems phonetically accurate, producing
[tuwnertz] and [birLz], and is also more attractive phonologically: as
outlined in Appendix 1, /a/ and /r/ are quite similar, apparently differ-
ing only in that the latter segment alone is [+high] and [+coronal]; an
epenthesis rule would have to introduce all the featuresnecessary to spe-
cify /r/, while the structural change of (47) is merely [+high, +coronal],
B. The /r/-less dialects of Eastern New England display /r/-epen-
thesis and appear to require an analysis essentially identical to that of
standard British English. Epenthesis after the low vowels, [a:] (car, spa)
and [o:] (for, saw) is regularly observed (in fact, for it or saw it,
car is or spa is without [r] would be highly unusual in connected speech).
Here, as in R.P., I know of no evidence supporting retention of post-
vocalic /r/ in lexical entries.
C. /r/-less speech is very widespread in the southeastern part of
the United States, especially in coastal regions. A feature of this dia-
lect often remarked upon is the general absence of "linking" and "intru-
sive" /r/ (see, for ex., Bronstein, 1960:118). Thus not only tuna is but
also tuner is is pronounced without [r] even in connected speech, in this
dialect.
At first glance it appears that Southern speech29 has the simplest
grammar of all with respect to post-vocalic /r/, full lexical loss and no
rule of epenthesis or deletion. Both tuna and tuner would be underlyingly
/tune/ and would always appear as such on the surface.
The situation is complicated, however, by the necessity of accounting
113
for [e]/[r] alternations within words, at morpheme boundary: stare [stee] or
[stee], staring [stertr]; structure [...e8j, structural [...EerelJ is ob-
served in many dialects.3 0  Two possible analyses suggest themselves: (a)
underlying final /r/ in words that alternate, /a/ otherwise, and an /r/-
deletion rule; (b) underlying final /a/ in all words of the relevant type
and an /r/-epenthesis rule applicable before vowel-initial suffixes. Nei-
ther of these solutions is fully safisfactory, as far as I can see.
There is a technical problem in the case of (a), involving the precise
environment of the deletion rule. Suppose the rule is written as
r ---. / --- #
By the standard assumptions, as input to the phonological component staring
is /ster#ing/, and this rule would incorrectly lead to loss of /r/. On the
other hand, if the rule is modified to require - ##, /r/ would be improp-
erly left undeleted in those cases in which the boundary conventions3 1
provide only a single word-boundary between words. For example, due to
the fact that at and of are not members of major lexical categories, only
one word-boundary would separate them from /r/-final words in phrases like
stare at them, structure of the sentence, and a - ## rule would be unable
to reduce /r/ to /a/.
Possibility (b) could be directly tested if there existed word of
the relevant type normally heard only uninflected (thus always with [eJ)
which were nevertheless possible inputs to productive suffixation rules
involving vowel-initial suffixes. If (b) is correct, we would expect ex-
crescent [r) in such words when suffixed. Unfortunately there is an ex-
treme paucity of crucial examples, due to a general absence of vowel-vowel
114
sequences within words in English on the phonetic level. Possible test
cases include rumba+ing, subpoena+ing, Selma+ite. As far as I can deter-
mine, excrescent /r/ is not to be expected in these cases. Possibilities
(a) and (b) thus both meet serious problems, and in my ignorance of addi-
tional relevant facts concerning the dialect, I will have to let the
matter stand unresolved.
D. An interesting situation exists in New York City, which is loca-
ted between traditionally r-less and r-retaining regions. r-less pronun-
ciations of words with historical /r/ (card [ka:dJ), as well as "intru-
sive" /r/ (I saw it [aysorLt]) are well attested here, but the overall
pattern is quite different from that of England and New England.
The investigations of Labov (1966, 1972) are of great value in this
regard. Labov is interested in the correlation between certain linguis-
tic and sociological variables. He discovered in New York City a consis-
tent connection between the retention of post-vocalic /r/, fricative
pronunciation of th, etc., and considerations such as socioeconomic class
of the speaker, his class aspirations, formality of speech, and so on.
In spite of the complexity and inherent interest of these findings, only
one rather simple point ii of relevance to the aims of this section. In
words like tuner (when noc before a vowel-initial word) and card, the per-
centage of retention of /r/ varies from a few per cent to near 100%, the
actual figure depending on the factors mentioned above. When /r/ is not
pronounced, tuner and card become homophonous with tuna and cod. Now the
crucial fact is that whatever the percentage of /r/ in tuner and card,
the naive32 New York speaker has exactly 0% occurrence of /r/ in tuna
(when not before vowel) and cod. This fact forces the conclusion that in
115
spite of their frequent homophony, card and cod, tuner and tuna are under-
lyingly distinct. The obvious choice of distinguishing feature is inclu-
sion of /r/ in the lexical representation of tuner and card but not tuna
and cod.33 Thus we must conclude that the synchronic grammar of this dia-
lect contains an optional /r/-drop rule. In common with other "fast-"
or "casual-speech" rules, it has the properties of being invoked more often
in colloquial or hurried speech, of being avoided by speakers who strive
to speak clearly and "correctly," and so on. The form of the rule is just
that of the historical rule (42).
What is surprising about the situation in New York is that in addition
to this rule of /r/-deletion there must exist an /r/-epenthesis rule, for
as mentioned earlier, "intrusive /r/" is a common phenomenon of this dia-
lect: tuna seems, always [a], vs. tuna is, often [ar]. (47) appears to
be fully adequate for New York. 3 4
I would like to make a brief remark on the theoretical significance
of discovering a rule and its inverse (r-+o0, 0--+r) in a single dialect.
It seems to me that this is neither a problem for the theory nor particu-
larly difficult to understand. Consider a native speaker of an /r/-re-
taining dialect who (consciously or unconsciously) wishes to imitate the
speech pattern of England or New England. His attention would quickly be
drawn to two types of unfamiliar patterns, words like card being pronounced
without [r] and phrases like tuna is with intrusive [r. These peculiari-
ties could rapidly be incorporated into his own speech by addition of (42)
and (47) as late rules. We know in fact that dialect imitation typically
takes place in just this way. If these rules are retained in optional
form, we have the New York City pattern. In view of the fact that current
116
New York City speech is the result (among many other influences) of the
imposition of an r-less norm of early-19th-century British origin on an
r-retaining base (see for ex., Kurath, 1964:76), this account seems en-
tirely reasonable.
Section 3 - Pre-/r/ Vowel Quality
3.1 A comparison of English orthography and pronunciation gives a
hint as to the quality changes /r/ has induced in preceding vowels. To
give just one example, the vowel of orthographic C aC 1 is [a} except when
the initial consonant of C is r, in which case it is [a.] (Ij, mat, Mac,
mass, but mar, mark).
Although the sound change responsible for this particular spelling
inconsistency has affected nearly all dialects of English, other pre-/r/
vowel mutations took place only in some dialects; furthermore there are
many cases of rules with identical structural descriptions in various di-
alects but different structural changes.
Most of the important changes of the type under consideration had
already taken place before the wave of post-vocalic /r/-dropping of the
early 19th century, so that current r-less dialects display a wide range
of altered vowels before lost /r/. Even dialects in which /r/ is not
nounced in cart have a vowel in this word which is more back than the
vowel in cat, for example.
It is beyond the scope of this chapter to attempt anything approa-
ching a general survey of the dialectal treatment of pre-/r/ vowels. What
is of direct interest is that most of the historic and synchronic rules
involve syllable-structure conditioning, a claim which I will attempt to
117
establish as true by means of a few examples involving easily confirmable
facts.35
3.2 In the mid-18th century, words like cat, carry, car, cart had
the same vowel ([a] in America and in the major British dialects). Some-
time thereafter, [a] shifted to [a] or [0.] when followed by final or pre-
consonantal /r/ (car, cart), but remained otherwise (cat, carry36) (Kurath,
1966:177-9). Expressed in standard form, the phonological rule accounting
for the result of this sound change is (48):37
(48) fc J[ (a)
a r # (b)
[+bkJ
The appearance of the unnatural class in (48) suggests that this
rule is actually syllable-conditioned. Noting that according to the sylla-
ble-structure assignment rules of Chapter I, only car and cart have a fol-
lowing /r/ wholly within /a/'s syllable, we might relate cases (a) and (b)
of (48) by re-writing (48) as (49):
(49) r
[+bk
where [-S.I.] is an abbreviation for "not the initial segment of any sylla-
ble;" cf. fn. 4.38
Use of syllable-structure conditioning in (49) allows elimination of
the unnatural [C, #] in (48). However, it should be noted that the /ar/
-- + /on/ rule differs in two important respects from rules discussed
earlier in this chapter which at first glance appear quite similar, for
118
example, /r/-loss.
The first difference is that the unnatural environment in the non-
syllabic account of /m/-backing is [C, #], of /r/-loss [C, Pause]. To
see the second difference, which is intimately related to the first, re-
call that once syllable structure has been assigned by Rules I-V, rules
like /r/-loss apply freely without regard to boundaries assigned by the
syntax. Now observe that (49) cannot be allowed to apply in a similar
manner, for the /r/ of car, which is [-S.I.] when car is spoken in isola-
tion or stands before a consonant-Initial word, becomes [+5.I.] by Rule
V when spoken in a connected manner before a vowel-initial word, yet the
the [a] of car does not revert to [m] under these circumstances in contem-
porary dialects.
To see how the challenge to the correctness of (49) entailed by these
observations might be overcome, consider the structure of four key inputs
car, cart, carfis, and carry at various points in the syllable structure
de t4.ation:
(50) Syllable structure of
car [a] cart [a] car"is [a.] carry [w]
a. after Rules I-II
k a r k art k arL z kma r i
\W \V/\W V V V
S S S S S S
b. after Rules III-IV (IV is not applicable)
k r k ar t k awrtL z k a r i
\7 \V/ \VV \V
S S S S S S
119
c. after Rule V
k wr k w r t kamr tLz k xri
S S S S S S
It is clear that Rule (49), if it is to produce correct results, must be
ordered before Rule V, since after applicatica of V, car is and carry are
syllabically indistinguishable, yet one has [a], the other [W], and since
[a] is found in car is whether or not spoken in such a way that V applies.
On the other hand, (49) must follow Rule II, which distinguishes between
the /r/s of car/cart and carry.
The differences between rules like /r/-loss on the one hand and /a/-
backing on the other are theoretically significant, and will be discussed
further in Chapter III, where the ordering solution suggested here will
be re-examined. As we proceed in this chapter, we will encounter other
rules which are like (49) with regard to the properties just discussed.
3,3 In all major contemporary dialects of English, curse, first,
and herd have the same vowel, [r or [3]. The appearance of u, i, e in
the spelling is an accurate representation of the Middle English vowels
in these words; the rule which served to neutralize this historical dis-
tinction is the subject of this section.
It is only before /r/ that /u, i, e/ merged, but the merger failed
to take place before /r/ followed by a vowel: hurry [A], mirror [L],
merry [e).39 Thus we have the following historical rule of assimilation:
4 0
(51) +syl +high
-cons -round r
-low +b ack [-S . I.
-long +coronal
120
(51) of course is a rule operating in just the same environment as (49),
and the remarks made in connection with (49) regarding the desirable elim-
ination of [C, #], as well as the need for pre-Rule V ordering, are rele-
vant here. It is significant, however, that these rules represent inde-
pendent sound changes, (51) having taken place before the time of Shakes-
peare (Kurath, 1964:122), (49) around 1800.
3.4 In Sections 3.2 and 3.3 I discussed rules affecting /i, e, u, u/
before /r/. The remaining short vowel of Middle English, /o/, was also
influenced by a following /r/.
In many dialects, the same sort of conditioning observed in rules
(49) and (51) is observed in connection with /3/. We find in most dia-
lects of the eastern and southern United States, as well as in standard
British English (RP), odd, orange [a.] ([o in RP) contrasting with or, orb,
orbit [o] or [o], all five words corresponding to earlier short /o/.
Thus although the dialectal development of M.E. /o/ is extremely complex,
it is clear that the history of the dialects just mentioned involves a
rule sensitive to the environment [r .]a41
Section 4 - Glide Sequences
I have argued in Appendix 1 and elsewhere in this chapter that Eng-
lish /r/ should bear the feature specification [-consonantal], i.e., that
it is a glide like Iw/ and /y/, not a sonorant consonant like /n/ and (in
most dialects) /1/. An examination of the pattern of glides associated
with the reflexes of Middle English long vowels and diphthougs provides
more evidence for this claim, as well as an additional illustration of
121
the way in which syllable structure conditions phonological processes.
Although the vowels of both code and core derive from M.E. /3/, un-
derwent vowel shift together, and currently have mid back rounded sylla-
bic nuclei, many or most current speakers of English do not perceive them
as the same vowel. The most obvious phonetic difference between them is
that only the vowel of code is followed by a /w/-glide: [kowd] vs. [korj.42
This observation is part of a more general fact about modern English:
tense vowels are not diphthongized before the glide /r/ in monosyllables:
(55) feed [fiyd], fear [fir]
moon [muwn], moor [mur]
cage [keyj], care [ker
roam [rowm], roar [rorn
Where the syllabic nucleus is a high vowel, following homorganic glides
are hard to hear, but I believe the first two pairs of transcriptions are
generally correct for American English. The third and fourth pairs of
examples should be uncontroversial.
(55) illustrates the reflexes of the Middle English non-high long
vowels in pre-/r/ and non-pre-/r/ position. In the case of the high
vowels, the facts are different. Cf. (56):
(56) A B C D
tide [tayd]; tire [tayr], [ta:r], ?[tayr]
out [awt]; hour [awr], [a:r], ?[awr]
In the most widespread American dialects, the glides are retained even
before /r/, following the reflex of a M.E. high vowel, but the /r/ is made
122
syllabic (column B); in some dialects the bisyllabicity of words like tire
is extremely obvious: [tay-ye(r)]. A minority of speakers drop the glides
(column C), as is the general rule in the case of the other vowels (cf.
(55)). I am not familiar with pronunciations in which the glide is re-
tained and /r/ is kept truly non-syllabic (column D), lIke the /1/ of tile.
Thus there appears to be a general principle operating to prevent the
sequence [VGr] in monosyllables (G = glide). But this observation is sure-
ly related to the fact that no sequence of two glides is ever found in
English monosyllables: *[wy, ywJ. If, as I have argued, English /r/ is a
glide, the absence of [VGr] falls out from the general restriction on GG
sequences. In fact, none of the possible combinations of [r and glide
occur in monosyllables: *[rw, ry, wr, yr].
Furthermore, no other non-syllabic shares this property of not form-
ing sequences with glides. Examining the most obvious possibility, we
find /1/ following glides in tail [teyl], tile [tayl], toll [towli, toil
[toyl], and, accepting the hypothesis under discussion, in Carl [karl].
I take these observations as additional evidence that /r, w, y/ form a
natural. class among the English non-syllabics.4 4
Extending the domain of the investigation beyond monosyllables, we
find that GG sequences do occur, provided the glides are not tautosyllabic.
For example, [yr] is found in tyrant, [wr] in Lowry, [yr] in wayward, [rwJ
in Harwell. Thus Rule II of Chapter I apparently disallows GG sequences as
parts of initial and final clusters, while a syllable-final glide is free
to abut a syllable-initial glide of a following syllable.
Furthermore, if the synchronic Vowel Shift/Diphthongization analysis
of SPE is correct, there must be a rule of glide deletion in the environ-
123
ment (57a) ((57b) in a theory not recognizing the syllable),
(57) a. --- r b. -[r
S
or a restriction on the general diphthongization rule preventing appli-
cation in this environment.
Section 5 - Raising of /m/
5.1 Most Americans make a very obvious distinction in the vowels of
pairs of words like mat/man, in spite of the fact that both of these words
show the same vowel in pronouncing dictionaries. There are at least four
differences in the vowels exemplified by mat and man:
(58) a. the vowel of man is longer than the vowel of mat
b. the vowel of man is nasalized; the vowel of mat is not
c. the vowel of man is articulated with a higher tongue po-
sition than the vowel of mat
d. the vowel of man is tense, of mat, lax
Distinction (58a) is simply a special case of the well-known generalization
whereby vowels in English are considerably shorter before voiceless segments
than elsewhere (louse and Fairbanks, 1953; Peterson and Lehiste, 1960), and
will not be of interest in this section. Likewise the nasalization of
vowels before nasal consonants is a general process in English, so that
(58b) does'not represent a special property of /ai/. met and men, for ex.,
differ in the way described in (58b).
On the other hand, in the most common dialects of American English,
only /is/ is tensed and raised (58c, d) before nasals.45 In many of these
124
dialects, /3/ is tensed and raised before all voiced segments (mat 0 mad)
and before fricatives even if unvoiced (mat # mass);46 however, I limit
the discussion below to tensing/raising before nasals.
5.2 The phonetics of tensed/raised /3/ (for which I will use the
symbol /E/) presents a problem for a distinctive-feature framework that
countenances only three tongue-height levels, and I comment on this matter
now before turning to the distribution of /E/.
Although phoneticians often point out t.at the two "mid front" vowels
of English, the vowels of met and mate, differ in tongue height, it has
been argued (see, for ex., Halle, 1973b:930)) that the basic distinction
between these two vowels, /e/ and /e/, is captured by the feature [tense]
(/e/ is [-tense], /e/ [+tense]), and that the height difference is merely
a secondary effect of the tensing.
The relevance of /E/ to this hypothesis is that /E/ has approximately
the height of /e/, not /e/, yet unlike /E/, is tense. Thus even among
[+tense] vowels there appear to be two heights distinguished within the
"mid" range.
Representing /E/ as a tense counterpart of low /3/ does not seem an
acceptable way out of this problem. While two vowels might differ slight-
ly in height as a secondary result of some other feature difference, to
claim that a vowel with the specification [+low] (/E/ under the hypothe-
sis being examined) and one with the specification [-low] (/E/) have the
same height seems an abandonment of the usual assumption regarding the
phonetic significance of distinctive-feature specifications. Secondly, it
may be necessary to make a three-way distinction between /a/, a tense low
125
vowel, and a tense vowel having the height of /e/ (- /E/): see the dis-
cussion of the variants heard for the vowel of bad in New York City speech
in Labov (1966) as well as Trager's (1930) discussion of "/AI."
It is beyond the scope of this section to attempt a resolution of
the question of how many vowel heights a correct distinctive-feature sys-
tem should allow (see Kiparsky, 1968b; Ladefoged, 1971; Halle, 1973b;
Kiparsky, 1975). I have included these remarks here because I would like
to see future discussions of the problem take into account American /E/
as well as to make explicit the nature of the vowel-quality distinction
referred to in this section.
5.3 In some dialects a following /m/ or /n/ triggers /u/-+/E/ under
all circumstances, and the simple rule of (59) is an adequate description:
(59) w {m, n)
I
E
(/m/ never occurs morpheme-finally, so there are no boundary phenomena
to consider).
Also common, though, are dialects in which man and plan have [E] while
manage and planet have [m].48 The crucial difference between man and manage
is not syllable count, however, for in these dialects mandible and pLanta-
tion have [E]. If syllable structure is not taken into account, the rule
accounting for the distribution of [E] must be written as in (60) (which
also correctly predicts to [E] of words like plant):
(60) m (m)J[ )
E
126
Once again we encounter the unnatural grouping [C, #] in a rule environ-
ment. Here, as before, if one thinks in terms of syllable structure, the
conditioning becomes quite straightforward: to effect //--/E/, the nasal
consonant must reside entirely in the syllable; if it simultaneously serves
as initial consonant of the following syllable (manage) or is not in the
syllable at all, it does not affect /aB/:
(61) [Mn]
E
We have of course seen this sort of conditioning several times already
in this chapter. That (61) correctly accounts for the distribution of
[E] can be seen by considering the syllabic structure assigned by the rules
of Chapter I to the following key words:
(62) plan plant planet plantation
p 1 m n p 1laBn t plI m n L t p1 w n t e A a n
NV/ \Vl \\//VJV
S S S S S S S
Only in the case of planet is /n/ syllable-initial and only here does /w/
remain.
As was the case with the /wr/-+ /ar/ rule discussed in Section 3.2,
(61) is accurate in words only. The phrase plan it, as normally spoken,
has the same syllabic structure as planet (see (62)), yet the phrase has
[E], the word [w], in the dialects under discussion. See Chapter III for
further investigation of this phenomenon.
127
Section 6 - The Sipplification of /nyu/
The simplification of /yu/ to /u/ after coronals, widely observed in
current dialects of English, provides another illustration of the need for
phonological descriptions in terms of syllable structure.
It is generally assumed that at an earlier point in the history of
English,words like new, dew were universally pronounced with prevocalic
/y/ which has been lost in many contemporary dialects. (See, for ex.,
Kurath 1964:77-8.) The change is usually described as loss of /y/ after
coronals, for in all dialects the /y/ of pew and cue remains.
The facts, however, are not quite so simple. While it is true that
/yu/ never simplifies after non-coronals, it sometimes also fails to re-
duce after coronals. In what follows, I limit my attention to the coron-
al /n/.
Even in dialects in which new is never pronounced with [y] (hence-
forth the [nu] dialects), the glide is found in annual, continue, and
sometimes in avenue. The intuition of speakers of such dialects is that
even though [nyu] represents a strange clustering of segments, annual
[anyuel] is unexceptional since it consists of the acceptable sequence
[an] followed by the acceptable sequence [yu].
Suppose we assume that the following rule was added to the phonology
of the [nu] dialects:
(63) n
x/A
As in the case of several other rules discussed in this chapter, the re-
quirement is that the conditioning element be entirely in the syllable of
128
the segment affected.
Now at the point at which (63) enters the language, new is [nyu],
i.e., [ny] is a possible initial cluster. Thus the rules of Chapter I
provide the following syllable structure for annual,
(64) a n y u a 1
V\/ V
S S S
while in new the /n/ is exclusively syllable-initial. (63) operating on
new and annual, then, produces the desired result.4 9
While (63) seems to correctly characterize the rule that caused the
change of /nyu/ to /nu/, it is not likely that it exists in the synchron-
ic grammar of speakers of the [nu) dialect. Consider a child learning
English in an environment in which the adult speakers have (63) as an
obligatory rule. Not hearing any words like [nyu], the child's language-
learning mechanism wi'.l assume that /ny/ is not a permissible initial
cluster, the underlying form of new will be /nu/, and (63) is not needed.5 0
Thus I would propose (63) as a historical rule of the current [nu]
dialects and as the innovation added to the phonology by [nyu] speakers
who come under the influence of the (apparently expanding) [nu] dialect.
The explanation for the retention of /y/ in continue is the same as
that given for annual: Rule (63) does not apply in the environment of
[+S.F.] /n/. It is interesting to speculate on why the /y/ of avenue may
have been lost in the subset of dialects in which this word has [nul
while continue has [nyu].
Trisyllabic words with initial main stress and no strtss on the penult
fall into two classes, those with secondary-stressed final vowel (lberate),
129
which according to SPE are derived by means of the final-stress clause
of the Main Stress Rule and the Alternating Stress Rule, and those with
unstressed final vowel (MAdison), which have been subject only to the ante-
penult case of the M. S. R. In the case of words with a tense vowel in
word-final position, it is often not clear whether the tenseness is under-
lying, in which case the word belongs to the first class, or derived by
means of the final-vowel tensing rule of SPE from an underlying lax vowel,
in which case the word is a member of the second class. Buffalo is an
example of such a word. The difference between the stress patterns XXX
and XXX is difficult to hear without the aid of the reduced/tense vowel
distinction which occurs only before consonants and in the case of low
vowels.
Now in those words with [ny] in which (63) fails to apply, the fail-
ure is due to /n/'s being [+S.F.], which in turn, in the cases of interest
here, is the result of Rule III, which associates the initial consonant of
S with S if the vowel of S is unstressed. Thus depending on whether
n+l n n+l
avenue is a member of the first or second class of trisyllabic words de-
scribed above, it will have syllabic structure (65a) or (65b),
(65) a. Dv 06n y u b. a v n y u
W \VV/\
S S S S S S
with the additional line of association in (65b) produced by Rule III due
to the lack of stress on the final vowel. Since (65a) but not (65b) meets
the structural description of Rule (63), we have a way of relating the
variability in preservation of [y] in avenue to an ambiguity in stress
contour through a general convention of syllable structure assignment.51
130
Section 7 - Simplification of /rg/
There are certain facts which lead Chomsky & Halle (1968) to propose
a special analysis of phonetic [0] in English, in spite of the fact that
triplets like those in (68),
(68) a. Kim [m]
b. simmer [m]
c. ramp [mp]
d. Ander(son) [nd]
kin [n]
sinner [n]
rant [nt]
amber [mb]
king [ol
Singer [0]
rank [ 0k]
anger [g]
show that the three occurring nasals of English have at
similar distributions.
The facts concerning /q/ that motivate the special
are summarized in (69): 52
least partial'ly
analysis of SPE
(69) a. [m] and [n] but not [] occur in # - V
b. [m] and [n] occur freely in V - V in morphemes, while
there is only a limited number of examples of [0] in
this environment
c. only [n] occurs with a following homorganic voiced stop
in - #. (Note that [m) patterns with [0] in this
regard.
d. [m] and [n] but not [0) occur after the reflexes of under-
lying tense vowels
e. there are [q]/[Qg] alternations but no [n]/[nd] or [mu/
[mb] alternations53
Most of these facts can be accounted for by means of the analysis in
131
(70), which is essentially the treatment proposed in SPE (cf. pp. 85, 369,
419). 54
(70) a. /Q / does not occur in lexical representations; all sur-
face occurrencesof this segment derive from /n/ in
the environment - {k, g)
b. Nasal Assimilation Rule (approx. statement):
n-+0 / - k, g)
c. /g/-Deletion Rule (ordered after (70b)):
g-+0 / -#
d. morpheme-structure constraint: *#[+nasal][+consJ...
While there are serious problems with this analysis, some of which will
be discussed below, others in Section 5 of Chapter III, it is quite an
attractive account of the facts in (69).
The absence of initial [01 (69a) follows directly from (70a) and
(70d). Fact (69b) is also a direct consequence of (70), (70a) prohibiting
underlying /V0V/ and (70c) producing free [] only before #. Similarly,
the obligatory character of (70c) prevents the occurrence of [...rgl.
(The absence of [...mb] is not covered by (70).)
Fact (69d) is accounted for by (70a), together with a general rule
which laxes vowels before non-coronal clusters (SPE: 172, 241), ordered
before (70c). Finally, since (70c) deletes /g/ only before #, [r]/[gg]
alternations are to be expected in the case of suffixal derivation from
words with final /-ng/ underlyingly (69e).
I turn now to a major difficulty faced by (70). There is an addit-
ional fact, not recognized in SPE, which shows immediately that the analy-
132
sis of (70) is not fully adequate to describe the distribution of [r].
In words like those in (71),
(71) angma 'velar nasal consonant,' angstrom, anxiety
tungsten, Yngve
we find [0] not followed by either [g] or #
Assuming that (70a) is correct, the domain of /g/-Deletion (70c) must
be expanded, for we find /Og/ simplifying to /o/ not only before # but
also before consonants, as illustrated in (71). To achieve descriptive
adequacy, (70c) must be generalized to (72):
(72) 0 g (C, #}
0
Once again we encounter the configuration {C, #}l, and here as before there
is good evidence that the absence of syllable-structure conditioning is
behind the appearance of this unnatural class.
Before turning to specific evidence in favor of a syllabic account
of /g/-deletion, let us establish that the extension of (70c) to apply
before consonants is required in general. As mentioned in (69b), there
are some occurrences of [0] before vowels as well (for ex., hangar), sug-
gesting that the presence of free [01 in the words of (71) might simply
be exceptional.
In fact, however, the strongest kind of evidence is available to the
effect that the words of (71) represent a true generalization about Eng-
lish. Not only are there no occurring words like those in (71) but with
[0g], there could exist none. Hypothetical [ieogma] is fully as deviant as
133
non-occurring [#wog#], which SPE's analysis correctly excludes. (In con-
trast, *[haoger] is an accidental gap: cf. anger.)5 5
If (70a) is correct, the facts so far discussed call for the /g/-
deletion rule stated in (73),
(73) n g
0
as can be seen from the syllabications of three key examples by the rules
of Chapter I:56
(74) hang [f angma [r) anger (0g]
h a r0g 0g ma a g a r
\V/ \T V V \
S S S S S
Next note that (73) predicts a class of exceptions to the /gC/ ex-
pansion of (72). Where C is such that /gC/ is a permissible syllable-ini-
tial cluster, and C is followed by a vowel, the output of Rule II will
be as shown in (75) (contrast angma in (74), with non-permissible /gm/):
(75) V 0 g C V
VtV
S S
Here /g/ is [+S.I.], so that if (73) is correct, /g/ should not be deleted.
That is, words of the form [...VqgCV...] should be possible just when /gC/
is a permissible syllable-intial cluster. This prediction is borne out:
the non-syllabics which can form initial clusters with /g/ are /1, r, w, y/
134
(glow, grow, Gwen, gules), and these are just the non-syllabics which
tolerate preceding [0g): anglicize, angry, anguish, angular.57 1 con-
clude that if all surface [0] is from /0g/, (73), and not (70c), is
required in the phonology of English.5 8
Section 8 - Stress Placement
8.1 One of the most obvious applications of the syllable to problems
of English phonology is in accounting for a class of exceptions to the
generalization, derivable from the Main Stress Rule of SPE, that vowels
in the environment of (76) are stressed.
(76) 
-- C 2 V C0 IN
As has often been pointed out (for ex., by Kahn, 1972; McCawley, 1974;
Anderson & Jones, 1974), SPE's weak clusters, like /br/, which apparently
do not meet the C2 part of the structural description in (76), are possi-
ble syllable-initial clusters, while most of SPE's strong clusters, like
/nd/, are not possible syllable-initial clusters.
I have avoided using the phenomenon of the weak cluster in stress
assignment as evidence for the correctness of the view that phonological
rules may be conditioned by syllable structure because there is a basic
weakness in such an argument. The stress rules of English are first of
all riddled with exceptions and secondly fail the simplest attempts to
confirm them directly, in many cases. As examples of the first point,
one might mentioq with respect to environment (76) words like vanilla,
Manischewitz on the one hand, and ttllsmn, character59 on the other.
As for the second point, consider a word like Popponesset, in which there
135
is no question of either of the last two vowels being tense. Persons un-
familiar with this Massachusetts town name do not insist on antepenult
stress, as might be expected if the correct model of their psychological
competence included the Main Stress Rule of SPE; in fact, penult stress
strikes me as more natural (I have never heard the word pronounced), in
direct contradiction to the M. S. R.
In contrast, most of the rules I have discussed to this point cannot
be challenged in this way and thus are good examples to use in trying to
reach conclusions of a general nature about the form of phonological
rules. Nevertheless, I have made certain observations concerning the
relationship between accentual phenomena and syllabic structure, and I
would like to mention them here in the hope that they will add to our
general understanding of stress placement in English.
8.2 The basic claim of those who would use syllable structure to
condition the Main Stress Rule is that the C2 condition in (76), which
serves to distinguish, for example, Canada from veranda, represents an
incorrect generalization. In this alternate view, the crucial considera-
tion is whether or not the penult syllable is closed by one or more con-
sonants; closed penults receive stress, while open penults allow stress
to fall on the antepenult, if present. Under the general assumptions made
by those who would assign discrete syllabications to words, initial clus-
ters are as large as possible consistent with certain general constraints.
Thus for the two words in question we have Ca,na,da and ve,ran,da; only
the second has a closed penult.
If the open/closed penult analysis is correct, then the general oc-
136
currence of antepenult stress in (77),
0 0 ''
(77) [# X V C 0 C0 C0 #
is due to the fact that a single intervocalic consonant is always sylla-
ble-initial, from which it follows that there is no chance of a closed
penult; when C1 is replaced by C2 there is the possibility of syllable02
boundary internal to the cluster, which would result in a closed penult.
Thus the crucial- cases for deciding between the two hypotheses would
seem to be words which meet the structural description in (76) but whose
penult cluster is such that it is a possible syllable-initial. Viewed
in this way, the ante-penult stress of words like algebra, vertebra counts
in favor of the open/closed penult analysis.
Of course, this latter hypothesis makes the prediction that (76) will
fail not only when C2 is a possible initial -cluster of the type called
"weak" by SPE (viz., clusters consisting of C(r, w}) but when C2 is any
cluster which is possible syllable-initially.
Although I have not made the kind of statistical study necessary to
reach a final judgement in this matter, there are certainly many cases of
this type that work in the way predicted by the open/closed penult princi-
60'# 61lPple. Examples are Alistair, Arista, minister, orchestra, sacristan.
This principle obviates the need for any special treatment of a large
class of words in final orthographic -y: industry, tapestry, chemistry,
amnesty.
On the other hand, there are certainly examples in which the open/
closed penult principle appears to fail: Alaska, NebrAska, asbestos. In
order to properly evaluate the significance of these examples, one must
137
take cognizance of the large number of exceptions to the rule of antepenult
stress assignment, whether formulated as in SPE or restated in terms of
syllable structure. Relevant to the examples Alaska, Nebraska is the fact
that observed penult stress where antepenult stress is expected is the
rule in U. S. state names rather than the exception: Alabama, ColorAdo,
Kentutcky, Mississippi, etc. Even beyond the domain of place names, the
occurrence of penult stress in words with a penult of the form VCO
where both principles predict antepenult stress, is so great as to lead
to the conclusion that both types of stress placement must be allowed for
the structure in (77). Penult stress in these cases may be the result
of abstractness in underlying representations, as suggested in SPE:146-52,
or of stress marking in the lexicon. In either case the factor that re-
sults in the differential stress of Mississippi and America is ad-hoc and
lexical. 62
For this reason, one should not expect to be able to decide between
the principles being compared here on the basis of words with penult stress.
On the other hand, when words with antepenult stress are examined, it is
found that the majority of them can be accounted for by either the Main
Stress Rule of SPE unbuttressed by the special weak cluster provision, or
the open/closed penult principle. Most of the residue is in accordance
with only the latter principle (algebra, Arista), although many words of
this class come under the SPE weak-cluster analysis (algebra).63 On ba-
lance, then, if stress is to be assigned by phonological rule, syllable-
structure conditioning seems desirable since it allows doing away with the
special case of the weak cluster definition (for algebra)and subsuming ad-
ditional cases under a general rule (Arista). It should be noted, however,
138
that most of the additional cases involve the cluster /st/, rather than
other possible syllable-initial clusters, such as /sp/.
Before turning to other aspects of stress assignment, let us deter-
mine how the open/closed penult principle might be incorporated into a
phonology containing the system of syllable-structure assignment rules
of Chapter I.
Observe that if main-stress assignment is to be conditioned by con-
siderations of syllable structure, the stress assignment rule must be or-
dered between syllable-structure assignment rules II and III. It must
follow II in order to make use of information concerning possible sylla-
ble-initial clusters; it must precede III since III makes use of stress
information.
This result is not problematical, for the additional lines of sylla-
bic association assigned by Rules III and IV do not appear to be relevant
in any way to stress assignment. (Recall that this was not the case with
regard to many of the rules discussed earlier in this chapter.) Rules
I-II produce the following syllabications of the relevant examples of
this section:
(78) C a n a d a a geb A r i a t a
V V V V V WV I V N/
S S S S S S S S S
miViipi ve r an d a
V V V V V \V V
S S S S S S S
Only in the case of veranda is the penult syllable closed, and as noted
above, it is only when this condition is met that we can predict stress
placement with confidence in words whose last two vowels are lax.
6 4
139
It appears then that one part of a basically correct phonological
generalization for English is (79):65
(79) Closed penult syllables are stressed
(ordered between syllable-structure ass4gnment rules II and
III)
8.3 I outline in this section another application of syllable struc-
ture to a problem of stress assignment.
It is usually possible to predict whether an initial syllable which
stands before a stressed syllable will itself be stressed. A first approx-
imation to a correct generalization is that the vowel of such a syllable
will be stressed if and only if it is tense or is followed by two or more
consonants: Ohio, Alberta, but Dakota. This observation was incorporated
into the system of stress rules of SPE by means of "Auxiliary Reduction
Rule II." In Halle's modified system of stress rules (1973a), two rules,
the Initial Stress Rule and the Destressing Rule, work to produce the same
effect, as well as to allow accounting for generalizations regarding the
stress patterns of bisyllabic words observed by Oehrle (1972). The ISR
blindly stresses vowels in initial syllables, while the DR removes stress
from the first vowel in a word provided it is lax and followed by a sin-
66
gle consonant.
Although more often than not producing correct results, the rules
of Halle (1973a) regarding stress on initial syllables admit of a large
number of exceptions, many of which can be eliminated by means of a syl-
lable-sensitive reformulation, as I will show below. For completeness,
as an aid to those who would study these problems further, and to avoid
140
giving the incorrect impression that attention to syllable structure pro-
vides a cure-all for the ailments of English stress-assignment, I mention
as well types of exception which remain unexplained even under a syllabic
treatment.
There are ficst of all cases in which it appears that the Destressing
Rule simply fails to apply. Some examples are given in (82): 67
(82) a. alloy, Kellogg, monarch, reggae, (Walter) Matthau,
sapphire, satire
b. baboon, ballet, cafe, chateau, Hanoi, hello, raccoon,
tattoo
At best these exceptions require lexical marking; at worst they challenge
the basic generalization.
There are also many exceptions of the opposite type, in which Destres-
sing appears to have applied in spite of the fact that its structural de-
scription is not met (alternatively, the ISR has failed to apply in these
cases). I am aware of four classes of exceptions of this type.
First of all, there are certain prefixes of Romance origin which are
destressed (in verbs only) in spite of a consonant cluster which follows
the vowel: c6mbAt, permit. (Suggestions for building this observation
into the formalism are contained in SPE and in Leben, 1974.)
Secondly, there are vowels in initial syllables which should retain
stress by virtue of their apparent underlying tenseness, but lose both
tenseness and stress. Examples are binoculars (cf. binary), miijority
(cf. mlor), minority (cf. minor), penultimate (cf. p nult), ps chia-
trist (cf. psyche, psychologist), stability (cf. stable), trannical
141
(cf. tyrant), vdcabulary (cf. v5cal). (In all examples except majority
and stability, stress- and tenseness-loss are optional.) 6 8
Thirdly, stress is not infrequently lost in #C0 [tas] a C X#
when C is a sonorant: Kentucky, Minhattan (especially as pronounced
in New York), Lmbardo, BUllgaria (opt.). When the sonorant is /r/, re-
duction is found in a very large number of examples: Bermuda, Bernoulli,
berserk, Vermont, vernacular, Virgtsia, etc. Regarding this generaliza-
tion, note that if, as argued in this dissertation, /r/ is to be grouped
with the glides rather than with the consonants, we should expect it to
occur in syllabic form underlying, as the glidewdo (syllabic fw, y/ -
/u, i/). Now if Bermuda, for example, is underlyingly /brm.../, this
expectation is borne out and moreover words like Brmuda become well-be-
haved with respect to the initial-stress generalization. Words like
Brlin, on the other hand, which are fairly rare, would be exceptions to
Destressing, like Hnoi. (Cf. Appendix 1, where phonetic syllabic /r/
is discussed.)
Fourthly and finally, when in #C V[tnsC V2V X # cluster C2 is of
the type permissible syllable-initially, loss of initial stress appears
to be the rule: Astair, %str 6 nomy, D6tr6 it, Dibrvnik, MAdrid, msaguito,
suprime. I studied words of the form [# C Vs t V X #1 fairly systema-
tically and found very few exceptions to this generalization (one is
gstition). Far more common are examples like iistnish, custdian, dis-
tfnguish, gistalt, nbstAlgia. Even very uncommon wordsoften follow this
generalization: Astyanax.6 9
Thus it appears that the proper environment for initial destressing
is open syllables, this generalization encompassing examples like
142
Alberta and D&kota, as well as examples like Madrid.
I conclude that Halle's (1973a) treatment of initial stress in Eng-
lish can be improved by recognizing /r/ as an underlying simple vowel
and by destressing initially in open syllables, rather than in #C0 [-tnsj
C V X#.
Section 9 - Halle's Rule-Writing Convention
9.1 Halle has suggested a convention for writing phonological rules
which would eliminate the need for the appearance of the unnatural class
[C, #]. In view of the fact that the hypothesis that phonological rules
may make reference to syllabic structure draws some of its support from
its ability to perform this same desirable function, a look at Halle's
proposal would seem to be in order.
Halle suggests that rules that apply within word boundaries be writ-
ten to indicate this fact, and that variables be used where necessary to
complete a structural description. Using the /m/-raising rule of Seztion
5 as an example, instead of writing the environment as shown in (83),
(83) -- N {} a)
one would express case (a) of the rule as (84),
(84) # X-- N C Y #
where X and Y are variables. Then case (b) can be incorporated into (84)
by means of the introduction of cost-free parentheses:
(85) # X -N ( C Y ) 1
143
Thus one could argue that the appearance of [C, #] in rules is merely an
artifact of improper rule-writing conventions.
In what follows, I will argue that the availability of the rule for-
mat illustrated in (85) does not diminish the need for syllable-structure
conditioning of phonological processes. I will try to demonstrate that
Halle's proposal does not really represent an improvement over a theory
in which rules like (83) can be written (Sec. 9.2); that it allows the
writing of rules for which there is no demonstrated need (9.2, 9.3); and
that it does not provide the apparatus for representing certain observed
natural-language processes closely related to those for which it was pro-
posed (9.4).
9.2 (83)/(85) also represents the environment for vowel nasalization
in French (subsequent rules delete the nasal consonant and effect vowel
quality changes to produce the correct phonetic output; see Schane, 1968).
While it is not immediately clear why a nasal consonant should induce
tensing and raising of a preceding vowel, as it does in the case of the
English rule of Section 5, the French rule is simply one of feature assim-
ilation.
According to the syllable-structure account of French Nasalization,
a vowel assimilates to a following nasal only if the nasal is tautosylla-
bic,70 a condition which is surely to be favored by a proper rule-evalua-
tion scheme. On the other hand, I would argue, (85) and Halle's conven-
tion are as far from a formal account of the naturalness of French Nasal-
ization as (83).
To see this, note first that in a non-syllabic theory, in order to
144
avoid writing French Nasalization as (83) (i.e., in oder to avoid repre-
senting the process as consisting of two unrelatable parts), one must ei-
ther (a) try to show that C and # form a natural class, a possibility
which Halle (1971) and I (Chapter I) agree must be rejected, or (b) try
to show that once the rule has been written for one of the two environ-
ments in (83), the other environment comes "for free." It is approach
(b) which is taken in Halle (1971).
The first question to be asked in comparing this approach with the
syllabic account is what formal consideration serves to value (84) more
highly that (86), which involves an equivalent "feature counts"
(86) # X-- N V Y #
Cross-linguistic considerations make it clear that (84) is indeed more
natural than (36), where a vowel-nasalization context is at issue.
Let us assume that this question can be answered in a satisfactory
way, although no answer has yet been suggested, and consider the general-
ization of (84) to (85). Note that the convention that parentheses do
not add to the cost of a structural description provides no formal expla-
nation for the fact that (85) is more natural than (84). Even more impor-
tantly, since parentheses are "free," why isn't (87) as highly valued as
(85)??7l
(87) #X-( N C Y) #
The only possible answer to this question is "because the second expansion
of (87) is (88),
145
(88) # X-- #
and (88) is not a natural nasalization context." In accordance with such
reasoning, as opposed to (87), (85) is a natural generalization of (84)
because its second expansion is the highly valued (89):
(89) # X - N #
That is, the logic of this argument is that (85) is highly valued because
it expands to (84) and (89), each of which we know from cross-linguistic
considerations to be natural nasalization contexts. But (84) and (89),
judged independently, are exactly equivalent to (83). Since there is no
other way of justifying (85) as the proper genralization of (84), it
must be concluded that (85) is a notational variant of (83), at least
insofar as the [C, #1 configuration is concerned.
9.3 Most of the rules in which [C, #] appears to the right of focus
have an environment ot the form (90) if the focus segment is [-syllabic],
(91) if [+syllabic]:
(90) 
- C
#
(91) - C' C'''
where C', C'', C''' represent various classes of [-syllabic] segments.
/t/-glottalization and /r/-loss are examples of the first type, /r/ --V
/o.r/ and French Nasalization of the second.
As far as I can see, this fact is accidental from the point of view
146
of Halle's suggested convention. There is no formal distinction between
(92a) and (92b), for example.
(92) a. C-+[ ] / ( C X ) #
b. VC-L ] / -- ( CX ) #
On the other hand, when rules with right-of-focus [C, #1 are translated
into syllabic form, the reason behind this correlation becomes clear.
Rules of the form (90) describe the behavior of a syllable-final consonant,
(91) the behavior of a vowel in a closed syllable. Were the focus [+sylla-
bic] in (90) or [-syllabic] in (91), structural descriptions which are
completely undistinguished from the point of view of syllabic structure
would be involved. That is, in the latter cases syllabic considerations
offer no formal simplification while the former cases are found to reduce
to (93) and (94),
(93) +sF.] 
.or [C
(94) [ 1]
lt t
V C or V C
S Sx
when the classes of consonants involved and general initial/final cluster
restrictions are taken into account.
9.4 The demonstration that Halle's (1971) rule-writing convention
does not eliminate all the undesirable configurations of segment and
boundary that the syllable-structure theory does is quite straightforward.
147
Recall the rules discussed earlier in this chapter in which pause func-
tions on a par with a class of [-syllabic] segments. As we saw, it is
indeed pause or syllabic juncture and not word-boundary that is involved.
The theory of Halle (1971) has nothing to say about these rules, which
find such natural expression in terms of a syllabic theory.
Finally, note that as pointed out in Chapter I, in the realm of
phonotactic constraints statements like that in (95) are required,
(95) *tk
where C' is the class of non-syllabics prohibited from forming initial
clusters with a preceding /k/. As argued in Chapter I, such constraints
must be viewed as restrictions on possible syllable structure:
(96) *V t k
S
Thus (95) represents another type of occurrence of [C, #] in phonology
which is not covered by Halle's proposed convention, for re-writing (95)
as ...(C'X)# does not invalidate the criticisms of (95) made in the next-
to-the-last paragraph of Section 6 in Chapter I.
148
Appendix 1 - The Distinctive-Feature Description of English /r/
In the distinctive feature system I use in this dissertation, the
major class features are [syllabic], [consonantal] and [sonorant]. The
definitions I give below of the latter two are based on M. I. T. class
lectures by Morris Halle. The first, however, [syllabic], is a feature
that is easy to use but extremely hard to define, as discussed in Chap-
ter I, Section 2. I follow standard practice in taking vowels and "syl-
labic sonorants" to be [+syllabic], all other segments to be [-syllabic].
It is with [-syllabic] segments that I will be concerned in this appendix.
A segment is [+consonantal] if there exists an obstruction in the
central passage of the oral part of the vocal tract. "Obstruction" is
taken to mean that a pair of articulators is brought at least close enough
together to produce turbulence in a stream of air moving past them. Thus
stops, fricatives, nasal consonants, etc., are [+consonantal], while
glides are [-consonantal].
The class of [+sonorant] ([-obstruent]) segments is defined as the
class of segments whose articulation does not result in a pressure build-
up in the vocal tract. The nasal consonants, for example, fit this des-
cription, since in spite of the outward airstream and oral closure associ-
ated with these segments, the open nasal passage insures that no pressure
can build up in the oral cavity. Similarly, /1/ is a sonorant, since in
spite of tongue to roof-of-mouth contact, the wide lateral openings pre-
vent pressure build-up.
Since [-consonantal] segments involve no obstruction at all, there
will a fortiori exist no pressure build-up and such segments will be
[+sonorant]. We thus have the three major classes of [-syllabic] segments
149
shown below:
a. [oIsno obstruction: the glides
b.[+cons] obstruction not causing pressure build-up: the
+son nasals and liquids
c. +cons obstruction caus.ng pressure build-up: the obstru-
L-son J]ents (stops, fricatives, affricates)
Where does /r/ fit into this taxonomy? The "corresponding" sound
in other languages is classified with the liquids, but English /r/ in
its most common American and British variants is extremely rare among the
world's languages and is quite different from the more common "/r/ sounds."
The most striking characteristic of English /r/ is that at no time during
its articulation is there any obstruction, in the sense defined above.7 2
Thus on the phonetic level at least, English [r] must be a [-consonantal]
sonorant, i.e., a glide.
I would like to make some suggestions as to how English /r/ should
be distinguished from other glides. In producing /r/, the tongue tip is
retroflexed and the sides of the tongue are tensed and curled up, a con-
figuration which is quite similar to that involved in the production of
retroflex consonants in the languages of India and elsewhere, with the ob-
vious exception that in the case of English /r/ the tongue-tip remains
sufficiently far from the roof of the mouth that a [+consonantal] articu-
lation is impossible (compare the diagrams on pp. 168, 180 of Smalley,
1968).
On purely articulatory grounds, the feature specifications [+high],
[+backj, and [+coronal] seem justified in describing English [r]. In
150
M. I. T. class lectures (Fall 1975), Paul Kiparsky presented a good deal
of evidence of a phonological nature that the feature complex [+high, +back]
is in general appropriate for the retroflex sounds. I will therefore take
English /r/ to include the feature specifications shown in (d):73
d. -syllabic
-consonantal
+sonorant
+high
+back
+coronal
In the literature on the dialectal treatment of "post-vocalic /r/'
(cf. Section 2), the distinction between the "/r/-ful" and "/r/-less"
pronunciations of a word like course, [kors] and [kogs], respectively,
is often described by saying that /r/ is consonantal in the first pronun-
ciation but has reduced to a glide in the second. (One also sees refer-
ences to the "vocalization" of /r/; this misleading term is used to mean
that /r/ has become /e/.) The claim I wish to make is that both pronun-
ciations involve glides between the [o] and the [s], and that what is lost
in the "/r/-less" pronunciation are certain feature specifications rele-
vant to the shape of the tongue, [+high] and [+coronal] if the suggestions
of the preceding paragraph are correct, /r/ and /a/ otherwise having fea-
ture compositions which are nearly or exactly identical; in particular,
/r/ and /a/ are both [-syllabic] and [-consonantal], i.e., glides.
Additional justification for regarding English /r/ as a glide will be
found elsewhere in this dissertation. I mention here two patterning ar-
guments.
The other English glides, /y/ and /w/, may be viewed as occurring in
151
three forms phonetically. They are pre-vocalic glides in yjt, wet, stressed
syllabics (i.e., [i, u]) in beat, boot, and form the second element of a
diphthong in toy, toe. /r/ and no other segment in English shows these
three variants: red, burn, far. (Note that burn must be transcribed
[brn], not [bVrn], where V is some non-retroflex vowel, for there is no
portion of the syllabic nucleus that is not retroflex in standard Ameri-
can speech (see, for ex., Heffner, 1950:108). The conventional symbol
for the syllabic equivalent of the glide [r] is [1] when stressed and
[0] when not.)
In addition, note that in monosyllables the glides [y, w] are found
after the tense vowels, say, so [sey], [sow], but not after the lax vowels
*[sey, sew; swy, sww; etc.] in the standard dialects. /r/ is the only
other non-syllabic following this pattern: [ker, kor] but *[ker, kar, etc.].
In summary, there is good reason for viewing English /r/ as a
[-consonantal] segment in the feature system outlined above, and for re-
garding /r, w, y/ as a natural class of English non-syllabics.
Appendix 2 - A Note on the Phonetics of "Flapped" /t/
In studying the properties of "flapped" /t/, a good starting point
is the observation that /d/ is subject to flapping as well. If one com-
pares the pronunciations of ebbing, heading, and egging with each other
and with the pronunciation of /b, d, g/ in other contexts, one finds a
distinct manner of articulation in the case of the /d/ of heading. The
tip of the tongue is raised toward the alveolar ridge, but rather than
being placed in position, held, and released, as is normal with stops,
the tenseness of the tip is adjusted so that the egressive airstream will
152
set the tip of the tongue into oscillation by the Bernoulli effect, as
in the case of a trill. However, for "flapped" /t, d/, the tongue drops
away after the first tap of the trill. Thus a better description of the
/d/ of heading would be "one-tap trill," reserving the term "flap" for
the ballistic movement of the tongue against the roof of the mouth de-
scribed for many languages of India (see Abercrombie, 1967:49-50).
The crucial role of the airstream in producing the tap of heading
can be appreciated b: attempting to mouth silently the words heading,
head, and din with no airstream passing through the vocal tract. (This
manner of articulation differs from a whisper in that the latter involves
air flow.) The absence of airstream in no way impedes production of the
/d/ of head and din, for these articulations are achieved by use of the
tongue muscles alone, whereas under these conditions it becomes impossi-
ble to articulate the /d/ of heading in the way it is normally done.
As pronounced by most Americans, heading is probably exactly equi-
valent to the pronunciation of herring by Englishmen who have a "trilled"
/r/ in V - . "Flapped" /d/ is also very similar to the single-tap /r/
of Spanish miro, although the latter may involve some additional shaping
of the tongue. Certainly the American who attempts to approach the Span-
ish pronunciation of miro by concentrating on saying "meedo" will be com-
ing very close.
One result of the way in which /d/ is articulated in heading is that
the consonant is very short. In some informal instrumental measurements
I made, the /d/ of heading had less than one third the duration of the
/b, g/ of ebbing, egging.
In view of the above observations, the /t/ of a word like heating,
153
which is homophonous with heeding in my speech, cannot be viewed as merely
a voiced [t], i.e., a [d]. Although tapped /t, d/ is voiced, its oral
articulation is quite different from that of [d], as we have seen.
The description of tapped /t/ is additionally complicated by the
fact that while many Americans pronounce pairs like heating/heeding iden-
tically, others claim to maintain a distinction in the intervocalic conso-
nant. Identity has been reported for a (geographically) wide range of
dialects, and I would claim that the following pronouncement of Kenyon's
(1966:126) is simply inaccurate*
In American English [t] is often voiced between voiced sounds,
as in better [bet], battle. Yet voiced [t) is not the same as
[d], and does not belong to the [d] phoneme, since Americans
do not confuse such words at latter [ltd] -- ladder [ladat], or
putting [potL] -- pudding [podLQJ.
In Kenyon's speech,
the chief difference between voiced [t) and [d] is that [t] is
less than half as long as [d] in a given utterance and the tongue
contact for [t] is much less firm than for [d. (p. 127)
Judging by this description, it may be the case that Kenyon simply does
not flap /d/, while his [t] is equivalent to my tapped /t, d/.
It is probably significant that although the above quotes are from
the 1966 edition of Kenyon's American Pronunciation, the book was written
in 1924. Although Kenyon's description is undoubtedly correct for his
speecn, and while the distinction he describes may be quite widespread
even in 1975, it is certainly not universal in current American speech.7 4
It is somewhat difficult to design an experiment to determine whether
154
tapped /t/ and /d/ are identical in a given dialect. This is so because
vowels are shorter before voiceless consonants, and in some cases have
different quality, and thus may prevent homophony in pairs of words in
which /t/ and /d/ are pronounced identically. (Cf. Chomsky, 1964.) For
example, although I am convinced that my flapped /t/ and /d/ are identi-
cal, plotter and plodder are not homophonous, the latter word having a
distinctly longer vowel, perhaps with slighly different quality. Recog-
nizing that the longer vowel of plodder is a direct consequence of the
longer vowel in plod vis-a-vis plot, one might try to eliminate this prob-
lem by limiting one's attention to pairs of underived words like latter/
ladder, where the length difference is not observed. Bu, exactly because
these words are underived, we cannot be sure that they have different con-
sonants underlyingly. Thus the discovery that they are homophonous would
be inconclusive.
The literature contains a few very limited attempts at experimental
determination of the homophony of tapped /t/ and /d/. Oswald (1943)
found listeners in general unable to distinguish between American pronun-
ciations of bleeting/bleeding and similar pairs. Sharf's (1960) results
are ambiguous. (One of his two speakers producd voiceless unflapped /t/s
in words with orthographic t, like matter, probably under the influence of
the microphone and the experimental setting.) I am sure that a careful
study, involving speakers of many different backgrounds, would show that
for many American dialects /t/ and /d/ undergo merger in the tapping en-
vironments, while for others they dq not.
In the text I describe my own speech, in which tapped /t/ and /d/
are identical. I believe that the claims made regarding the environment
155
of /t/-tapping hold as well for speakers who never merge /t/ and /d/.
Comments on the feature specification of tapped /t/ are to be found
in the text. It is not at all clear, however, how Kenyon's latter, his
ladder, Spanish miro, and Spanish perro (multiple-tap trill) are to be
distinguished in terms of distinctive features.
Appendix 3 - More on pre-/r/ Vowel Quality
A. Sections 3.2, 3.3, and 3.4 were concerned with the phonetic
changes induced in the reflexes of the Middle English (M.E.) short vowels
by a following fully-tautosyllabic /r/. In the absence of further changes,
one would expect to find, in current dialects, identical vowels in the
environments - rV and - , where F - segment other than /r/. While
this situation is observed for some dialects and some vowels, further
changes have complicated the pattern in marny cases, giving rise to vowel-
quality distinctions in the environments just mentioned. Furthermore,
mergers have taken place between the reflexes of M.E. long and short
vowels in - r. I mention here some of the major patterns observed in
the United States.
B. As a result of Rule (49), which took /m/ to /a./ in - r {C, #1,
[mr] occurred only in -- V, as in carry. In some American. dialects, typ-
ically found in the West and Midwest, a much more recent rule has raised
the im/ of carry to [e], thereby giving rise to a merger of M.E. /W/ and
/h/ in - rV. (M.E. /h/ became modern /e/ in all environments by means
of the Great Vowel Shift.) A well-known example of the homophony result-
ing from these changes is that of marry, Mary (M.E. /mari/, /miri/, re-
spectively.)
156
The raising rule affecting carry, marry can be stated as (e):
(e) 
-low I/
[+tense
Although (e) always applies in the environment rV, the rule need not
be made conditional on the presence of V, since all far is in -V,
as we have seen.
It might also be mentioned at this point that the dialects which un-
derwent (e) are a subset of a set of dialects which show a loss of the
tense/lax distinction in certain pre-/r/ vowels:
(f) V
-bk -+---[atensei / - r
.- low
where 0, = + in some dialects, - in others. For example, while more con-
servative dialects have mirror [t] (<M.E. /mir.../) contrasting with
hearer [i] (< M.E. /hir.../ via Vowel Shift), in other dialects the words
rhyme. Similarly, merry (conservative [e] < M.E. /meri/) and Mary (con-
servative [e] < N.E. /miri/ by Vowel Shift) fall together. As was the
case with (e), (f) had the opportunity of affecting vowels only in - rV,
the hypothetical input U, e} r {C, #} no longer existing because of the
rule discussed in Section 3.3.
Note that in dialects affected by both (e) and (f), triplets like
merry/Mary/marry will be indistinguishable. (The varying homophony across
dialects of the latter set of words is too well-known for it to serve as
a reliable shibboleth in investigations of Rules (e)-(f); some speakers
have apparently adopted non-systematic pronunciations for one or more of
157
these words. Better test triplets include parry, peary 'like a pear,'
Perry; Harry, hairy, herring; bury is a good test item because its
spelling disguises its class membership: bury - berry, not Barry or
beary 'like a bear,' in the conservative dialects.)
C. There is a similarity in the development of front and back vowels
in pre-/r/ position. Two facts, however, prevent full parallelism in the
front and back series. First of all, at the time the pre-/r/ vowel-muta-
tion rules came into the language, English had two front non-high short
vowels, but only one back non-high short vowel, /i5, e/ vs. /5/. Thus
there is no possibility of a merger, among the back vowels, corresponding
to the one that gave rise to homophony between marry and merry in some
dialects. In this respect, the back-vowel developments are somewhat sim-
pler. A second difference, however, increases the complexity of the back-
vosel facts. The front low short vowel, //, was backed under the influ-
ence of a following fully tautosyllabic /r/ (mar(k) [a]), preventing //
in this environment from merging with any other front vowel. In the back
series, on the other hand, the low short vowel, /5/, remains back in all
environments, even before tautosyllabic /r/, and in fact merges with the
reflex of earlier /5/ in certain dialects.
In the most conservative American dialects, spoken in various parts
of the Eastern United States, we find the reflexes of earlier /5r/ and
/5r/ distinct in all environments: for [o] or [D] (</S/) vs. fore, four
[o] (</3/ via Vowel Shift); forest [a.] or [D] (< /5/) vs. forum to]
(< /5/ via Vowel Shift). I will refer to dialectashowing thes3 contrasts
as E dialects.
Other Eastern U. S. dialects (E2 dialects), as well as standard Bri-
158
tish English, retain the distinction between /Sr/ and /Sr/ only when /r/
is followed by 4 vowel, i.e., is not fully tautosyllabic with the vowel
in question. In such dialects then, for and four are homophonous [o]75
while forest, [o.] or [v], and forum [o]75 remain distinct.
The rule accounting for the merger of /5/ and /5/ in E2 , referred
to in Section 3.4, apparently applied in the environment-
and in this respect parallels the front-vowel rule of Sec 3.2, (49).
Dialects EI and E2 agree in distinguishing the words in (g) from
those in (h):
(g) [(.] or [i] < /SrV/
Doris horrible
foreign Morris
forest orange
historic Oregon
(h) [o] < /SrV/
boron orient
forum story
glory torus
There are other American dialects ("W dialects") in which all the words
of (g)-(h) have a mid vowel, approximately [o].76 The W dialects are spo-
ken west "of a continuous line on the map from central Vermont to western
Texas" (Thomas, 1958:200). The rule that gave rise to the W dialects
can be written as (i), added as an innovation to an E2 dialect:
159
(i) 9 -low /--r
+tenseJr
This rule parallels Rule (e), the innovation that merged marry and Mary,
in several respects: (a) both are conditioned by a following /r/; (b) both
apply only in - rV but need not have V stated explicitly, for other in-
puts (mar(k), for(k)) have been bled by previous rules applying before
non-syllable-initial /r/; (c) both rules raise low vowels to mid position;
(d) both rules applied in approximately the same geographic area, west of
the line mentioned above. It seems clear, then, that (e) and (i) are
special cases of a general pre-/r/ raising rule which affected Midwestern
and Western speech.78
160
Footnotes to ChapterII
1. occasionally voiceless; see Appendix 2
2. These allophones are sometimes described as being the affricate /E/.
This may be an inaccurate description traceable to the fact that // is
the closest freely-occurring phone in English to /c/, or it may actually
occur in some dialects.
3. In the experiments described in Lisker and Abramson (1964), most oc-
currences of initial /b, d, g/ in English were articulated as voiceless
unaspirates.
4. 1 will also, where convenient, treat the conditions "syllable-initial"
and "syllable-final" as though they were features on a segment; thus (6)
might be rewritten as
-continuant
+stiff v.c. --- +spread glottis)
-S.F.
It should be clear, however, that [S.I.] and [S.F.] are not distinctive
features, but rather structural conditions that may be imposed on segment
classes in rules.
On the use of the feature specifications [-continuant, +stiff v.c.]
to capture the set of English voiceless stops, see Halle & Stevens (1971).
5. Words like bodkin thus provide another type of example illustrating
the incorrectness of (2a).
6. There is another possible source for the unaspirated articulation of
161
/k/ in these words.
Consider first how a simple intervocalic stop is articulated. Stop-
page of the vocal tract is made at some point, held for a short interval,
and then released from that same position. Thus Abercrombie (1967:140)
refers to the "three phases of a stop," the shutting phase," "the closure
phase," and "the opening phase." Now when there is a two stop intervocal-
ic cluster, one does not observe two-times-three or six phases in ordinary
speech, but again only three. First there is the bringing together of
the articulators at the point of articulation of the first stop, followed
by a closure phase during which there is no escape of air or sound, fol-
lowed by an opening at the point of articulation of the second stop. Dur-
ing the closure phase, the point of stricture moves silently from that
corresponding to the first stop to that corresponding to the second.
Acoustically, a medial stop cluster looks like a single stop, except that
its silence interval is somewhat longer and its opening and closing for-
mant transitions do not correspond to the same point of stricture.
Now if for the purposes of establishing syllabic structure the se-
quence closure-holding-release is interpreted as a single segment, whether
or not closure and release involve the same point cf articulation, the
structure of napkin would be,
n a pk t n
S S
with the original lines of association between /p/ and S1 and /k/ and S2
joining to form ambisyllabic /pk/ and giving napkin a syllabic structure
162
identical to that of happy. At present I know of no further evidence for
or against this proposal.
7. I believe the description in this paragraph in fact holds for most
American speech, but I have not investigated the matter.
8. In many English and Scottish dialects, /t/ is replaced with a pure
glottal stop not involving alveolar closure. Such stops are perceptually
very different from the glottalized/t/ described here, presumably for the
reasons outlined in the text.
In addition, it should be mentioned that there are several percep-
tually distinct sounds for which one might use the symbol [t?], differing
perhaps in the forcefulness of the glottal stoppage and the relative tim-
ing of the oral and laryngeal closures. In the /t/ described in the text,
the glottalization is not particularly prominent.
9. As is well known, in many dialects orthographic r is not pronounced,
or is only optionally pronounced, in this position (cf. Section 2). The
observation being made here, of course, is that even where /r/ is pro-
nounced, /p, t, k/ take the V - Pause allophone.
10. In some dialects, words like belt seem to be consistently pronounced
with [tJ, and, as implied in my description, such a pronunciation is an
option for me. I have not tried to pass over this fact as possible coun-
terevidence to my claim that /r/ is [-consonantal] in English while /1/
is [+consonantal]. On the contrary, it is my own very clear intuition
that release of /t/ is far more acceptable in colt than in court, for ex-
ample (as well as the differential behavior of /1/ and /r/ with respect
163
to "flapping" of a following /t/; see Section 1.4) that led me to inves-
tigate the articulation of /1/ and /r/ and observe that only /r/ is oblig-
atorily a glide in the way that this term is usually understood (Appendix
1).
I would relate the variability in release of /t/ following /1/ to
the observation that in many dialects (optionally in most?) contact is not
established between tongue-tip and palate in articulating /1/ in post-
vocalic position (see, for ex., Bronstein, 1960:128). In extreme form,
this tendency gives rise to pronunciations like [mLwk] for milk, quite
common in some parts of the United States. However, it is possible to
attain "/1/ quality" (even as judged by speakers who tend towards conso-
nantal /1/) with no contact between tongue and roaf of mouth, as a simple
experiment clearly illustrates. Now once contact is lost, we have a
[-consonantal] articulation, as discussed in Appendix 1, and we expect
the unreleased pronunciation of /t/ and /p/. It thus seems to me that
the suggested analysis fits the observed facts quite satisfactorily.
11. In cases like casts and asks, a fast-speech option for many speakers,
including myself, is elision of the stop. However the resulting geminate
[ss] does not seem to simplify, except perhaps in extremely hurried speech.
Thus casts # Cass, asKs ass.
12. Regarding the statement of (17), which is equivalent to (15), note
that S may link segments to the right of /t/ and of course segments to
the left of the [-consonantal] segment. It need not be stated explicitly
that S's vowel is to the left of /t/, since were it to the right, the
[-consonantal] segment would necessarily be [-syllabic], giving rise to
164
an initial cluster of the shape C1-glide-/t/, which is of course non-oc-
curring.
13. It is sometime suggested that "fast-speech" processes be accounted
for in terms of a rule which weakens or deletes boundaries at above-normal
speech rates. There are two indications that such an account, if ever
correct, is not behind the phenomena described here.
First of all, the usually-described "fast-speech" phenomena come into
play only in greatly accelerated rates of articulation. If boundary-
weakening is to account for these phenomena, it must occur only in truly
hurried speech, with normal speech retaining the boundaries assigned by
the 'syntax (subject to the conditions outlined in SPE and in the Introduc-
tion to Selkirk, 1972). But the distinction between glottalized and un-
glottalized (actually "flapped") /t/ in hit Ann corresponds to artificial-
ly-distinct speech and normal speech, not normal speech and fast speech;
by hypothesis normal speech has maximal boundaries, so differences between
slow speech and normal speech cannot be related to boundary differences.
Secondly, if it were assumed that the words of the phrase hit Ann
are separated by ## only in slow speech and by # or by no boundary at all
in normal speech (an assumption that would give rise to problems else-
where), we could not account for the glottalized /t/ of slow-speech
hit a ball, in which /t/ is followed by single #, or of atlas, which in-
volves no boundary at all.
14. In the dialect I describe, /t/ is never fully replaced by [9]. In
comparing Selkirk's statements regarding the general process of glottali-
zation and my own, one should keep in mind that her sources of phonetic
165
data on glottalization are largely British, whereas I attempt to describe
my observations of American speech.
15. Recall that I have attempted to demonstrate that the "liquids and
nasals" after which glottalization occurs are illusory. On the other hand,
Selkirk omits glides from the enumeration of possible leftward environ-
ments, presumably an oversight.
16. There is however a point of contact between syntactic structure and
syllabic structure, Rules II-IV, which apply to words and to which certain
generalizations concerning words and aspiration/glottalization can be
traced (for example, the fact that word-initial /p, t, k/ is always aspi-
rated.) Such generalizations must not be allowed to obscure the fact
that considerations of syllabic structure, not syntactic structure, con-
dition proper statements of the aspiration and glottalization rules.
17. In some words which appear to be entirely on a par structurally with
words like capital, failure to flap is not quite so serious an affront to
the American ear as the absence of flap usually is. Compare better,
capital with marital. Even in the case of the latter word, however, /D/
is preferred greatly.
18. Another inaccuracy in Selkirk's rule is found in her attempt to col-
lapse the environments for the flapping of /t/ and /d/. Although I agree
with her that latter and ladder are homophonous for most Americans, center
and sender are not, at least in normal-rate speech. That is, a preceding
/n/ hinders /d/-flapping but not /t/-flapping, a fact which is surely re-
lated to the elision of nasal consonants (after vowel-nasalization) before
166
voiceless stops only (cf. Malecot). See discussion below.
19. For me, there is a third option in the pronunciation of words like
winter, possible only in decidedly hurried speech, [winer] - winner. Some
speakers appear to have homophony in winter and winner even in normal
speech. I cannot relate this loss of /t/ to the other rules discussed in
this chapter in an illuminating way.
20. Assuming of course that the expression "pressure build-up in the vo-
cal tract" mentioned in the definition of sonorant is taken to mean "pres-
sure build-up in the vocal tract prcper, i.e., not including the region
below the glottis." I take this to be the way in which the definition
should be understood. Note for example that SPE (p. 354) classifies glot-
tal stop as a sonorant.
21. Note that although in all cases in which Sonorization applies there
will exist the syllabic linking shown in (32), this structure can be elimi-
nated from the rule by ordering the rule after Aspiration and including
[-spread glottis], i.e., unaspirated, in the structural description of
Sonorization. For then inputs in which [-cons] and /t/ are not tautosyl-
labic will have been aspirated by the Aspiration rule. I don't know of
any fact which chooses between this analysis and (32).
22. The appearaze of underlying /t/ as an alveopalatal before /r/ is
surely an assimilation. /c/ differs from /t/ in being [-anterior) (cf.
SPE:304), a specification which /r/, as a retroflex segment, can be taken
to share. /t/ does not take on /r/'s [+back] specification in words like
train, although it appears to be [+high] under the influence of /r/ (cf.
167
Appendix 1), and may in this respect differ from alveopalatal stops found
in other languages.
Thus the structural change of the /t/ -+ D/c/ rule, to be given
below, can be taken as [-anterior, +high].
For its part, /r/ is affected by a preceding /c/, becoming conso-
nantal and fricative.
23. Glottalization (28) requires a [-.I.] stop preceded by a [-cons]
segment. As we have seen, /c/ is either [+S.I.] (in {V, #1 - rV) or
else preceded by a [+cons] segment, namely /8/. As for Flap (29), it has
the rightward environment [+syllabic], while /c/ is always found before /r/.
24. In additon to my own observations, I offer the following comment from
Kenyon (1964:164-5):
...intrusive /r/ is the use of an /r/ sound that is not spelled
andwas originally not sounded. It is a very common practice
among cultivated speakers in England and eastern America...The
evidence in these regions is so overwhelming that it is mere
ignorance of the facts of cultivated usage to deny it.
Gimson (1962:204), discussing R. P. (standard British English) describes
"intrusive /r/" as being pervasive.
25. The discussion in the text would be incomplete without reference to
the fact that even within /r/-retaining dialects, the structural descrip-
tiot of rule (42) is significant. In most of these dialects, post-vocalic
/r/ is articulated differently from /r/ in environments in which it is
universally retained. (See Kurath and McDavid, 1961:115, for a descrip-
168
tion of the difference in some American dialects; see also fn. 73.) Thus
it is possible that the correct historical solution is a universal rule
of the form /r1 /-+'/r2/, followed by the context-free innovation /r2/~0
in the "r-less" dialects. Cf. Gimson (1962:205).
26. Gimson (1962:204) on R. P.: "...the comparative rarity of the cases
of such possible intrusive /r/s [after low vowels not followed by ortho-
graphic r -DK] tends to make speakers more aware of the "correct" form;
thus I saw it [...r...], drawing [...r... j, are generally disapproved of,
though it is likely that many RP speakers have to make a conscious effort
to avoid the use of such forms."
27. This fact quickly becomes obvious when an English speaker attempts
to produce sonograms illustrating pure vowels, i.e., vowels with steady
formants.
28. If (46) is correct., the intrusive /r/ following the word eh, 'lax mid
front vowel,' which I observed on the part of a Boston-bred linguist
("the vowels [e] and [L]," phonetically [...erane]) is to be traced to
the fact that English speakers, even with phonetic training, produce [ea]
when attempting to articulate [e] in isolation.
29. I have observed, but will not discuss here, Southern dialects in
which post-vocalic pre-consonantal /r/ tendr to be retained, final /r/
dropped: force the Senato(...
30. Dialects without [r] in staring and structural probably also occur
in the southern United States.
Note that in the British and New England dialects discussed above,
169
where such alternations are also observed, epenthesis rule (45), which
applies on all domains, would introduce an /r/ in the inflected forms.
31. See SPE and Selkirk (1972Introduction).
32. I include this qualificatian to exclude possible hypercorrection
effects due to conscious or unconscious imitation of other dialects on
the part of speakers who have close contact with natives of other regions.
33. This sort of phonological situation is not at all uncommon. Again
in New York speech, then is sometimes [6], sometimes [d], while den is
always [d], never [6]. Or to give a more universal example, in spite of
the fact that several is almost always pronounced [...vr...], the fact
that it is occasionally pronounced [...ver...j implies that it is distinct
underlyingly from Chevron, which is never [...var...].
34. In New York speech at least, derived final /a/ does not allow "epen-
thetic" /r/: in the phrases a lotta-apples, you shoulda-eaten,
I'm gonna-ask 'im, /r/ in the positions indicated by dots is not observed
and is felt to be severely deviant, in spite of the fact that the final
segments of tuna and of lotta, etc. are phonetically identical.
A possible way of accounting for this phenomenon is to order rule (47)
before the various rules which generate final /a/, of-+"/e/, to--/tO/,
etc. (Note that these rules must be distinguished from the general SPE
rule which takes unstressed lax vowels to /a/: /algebra-r-is.)
There is at least one indication that this solution is not correct,
however. When the rightward environment of the epenthesis rule is derived,
/r/ appears normally: give Linda-r-istorical information,
170
let's subpoena-r-'em. Thus all rules (the sample is admittedly small)
that create a rightward environment for epenthesis would precede the lat-
ter rule, while lefrwrd-environment producing rules would follow it.
These facts suggest that words with "lexical" /a/, like Linda, tuna,
(and, contra SPE, algebra) are somehow marked as alternating with /Ir/,
while lotta, shoulda, being derived, lack this specification. Note that
this marking could not be the inclusion of a final /r/ which is deleted
before consonant and pause by (42), for such deletion would have to be
absolutely obligatory (recall that tuna is never [-r] before consonant
or pause), while (42) is normally optional in New York: Peter, tuner are
sometimes [-r], sometimes [-a] before consonant and pause; cf. discussion
in the text. I have no satisfactory solution to offer.
35. In addition to the specific references found in this section, the
following works served as general sources of information: Bloomfield
(1935), Bronstein (1960), Kenyon (1966), Kenyon & Knott (1953), Kurath
(1964), Kurath & McDavid (1961), Thomas (1958), Trager & Bloch (1941),
Trager & Smith (1951).
36. In some common American dialects a much more recent rule has raised
the unaffected /i/ of carry to /e/; see Appendix 3 for further discussion.
37. A historically more accurate account might have to break (48) down
into a lengthening rule in the environment of (48) and a context-free
/i/--/&/ rule. (See Kenyon, 1966:178 fn 94.)
38. Certain other non-syllabics, /f, 0, s, m, n/, induced the /w/-.P/a/
change, but only in a limited set of words (about 150); in contrast, the
171
backing of // was exceptionless before non-syllable-initial /r/). [a]
is found in place of [m] in these words in England and, in an even more
restricted set of words, in Eastern New England (where the vowel tends
to be [a:]), but not in the rest of the United States. Examples for both
England and New England include bath and can't. Although there is no
general rule accounting for which words with /m/ + /f, 0, s, m, n/ under-
went the backing rule, it is significant that in all cases the condition-
ing consonant is [-S.I.]: no word like asset had its vowel backed.
39. In some American dialects, an independent (chronologically much la-
ter) rule took /A/ to /F/ in words like hurry. On the fate of mirror
and merry in the U. S., see Appendix 3.
40. If there really is a difference between /frrst/ and /frst/, the non-
syllabic /r/ may have to be deleted.
Rule (51)'s structural description is such that all short non-low
vowels are affected. Due to an asymmety in the M. E. short vowel system,
two front vowels but only one back vowel undergo the rule.
41. See Appendix 3 for further discussion of the development of /or/.
42. With regard to these transcriptions and those in (55), I do not
mean to imply that the syllabic segments of pairs like core and code,
care and cage are identical, but merely that they are both typically mid.
43. Words like what, whether (where contrasting with watt, weather) might
be viewed as requiring the weakening of the claim in the text to "sequences
of voiced glides are not found in monosyllables," in view of the standard
transcription [hw...] for such words. However [hw] for wh is simply in-
172
accurate, at least for the types of American speech I have come in contact
with. what is not pronounced with [h] followed by (voiced) [w], but
rather with a voiceless [w]-glide, [gat], or with a voiceless fricative
(velar?) with labialization [xWot]. I can see no justification for re-
garding wh as consisting of a sequence of segments.
44. Haj Ross has pointed out to me that the analysis of this section
draws additional support from two further observations: (a) even those
dialectswhich retain pre-[u) [y] after [1] (lure, lute) fail to show
[y] after [r]: ruin *[ryuwn], a consequence of the restriction against
GG sequences and the major-class distinction between /r/ and /1/; (b) as
a syllable terminus in English, we find /rl/ (Carl) but not /lr/; this is
seen to be related to the hypothesis that /r/ is a glide while /1/ is
a sonorant when it is observed that it is true in general in English
that syllables may end in glide+sonorant but not sonorant+glide.
45. Often, /A/ is raised before /m/ and /n/ but not before /Q/, or
else is raised to /E/ (see below) before /m, n/ and to /e/ or /e/ before
/0/.
46. See, for example, Trager (1930), who claims that this situation holds
for most American speech. I have observed tensed and raised /a3/ in words
like mad, mass in the speech of many Easterners, Southerners, and Midwest-
erners.
47. Although the distinction seems to be mainly one of tenseness, [E
tends to be longer and end in a more pronounced centralizing glide. One
can see that these latter differencesare secondary in nature by conscious-
173
ly lengthening and diphthongizing the vowel of men and noting that it
still maintains a quality distinct from that of man.
48. This situation obtains in New York City and elsewhere in the U. S.
49. The crucial difference between new and annual is their respective
syllable structures, not the fact that [nyu] is the first syllable of
new and the second of annual: compare Agnew, which is like annual in
this regard but in which the /n/ cannot be associated with S due to
the presence of the /g/; in Agnew, /y/ is deleted.
50. Of course, the reason why the rule can be lost and replaced with a
syllable structure constraint is that adoption of therule did not give
rise to large-scale alternations between [n] and [ny]. If this were the
case, /y/ and Rule (63) might have been preserved.
51. In trisyllabic words with main stress on the penult, secondary
stress on the final syllable is not the usual case. Thus continue
is universally interpreted as having contour 010, /n/ is ambisyllabic,
and Rule (63) is blocked.
52. In this section, I follow SPE in describing the distribution of [0]
as observed only in the most common American dialects. In areas of both
the United States and Great Britain, one finds retention or re-introduc-
tion of historical [g) in words like hang, hanger.
Furthermore, even in the standard dialects, the facts are not always
as stated here and in pronouncing dictionaries. While [g] never replaces
[0], in rapid speech words which are cited as having [g] are often ob-
174
served with [01 alone. The ease with which this simplification can take
place shows a certain variability, governed by factors I have not studied.
For example, although Kenyon & Knott, AHD, etc., are agreed that English,
longer, and prolongation have [g], I find simplification most natural in
the case of English and least natural for prolongation, with longer fall-
ing somewhere in between.
In the discussion that follows in this section and in Chapter III,
the statement that a word has [og] is to be taken to mean that [g] is
permissible in the word, while the claim that a word has [0] corresponds
to a judgement that [g] is not possible in the word even in careful speech.
53. bomb/bombard is the only possible exception to this claim that I am
aware of, and counts as an exception only if the morphemic analysis /bomb+
ard/ rather than /bom+bard/ can be justified. The only evidence for /ard/
as a morpheme comes from the words drunkard and dullard, but here the
suffix would function to form a noun from an adjective as opposed to a
verb from a noun as required in bombard.
As will be pointed out in Chapter III, there are actually very few
[0]~[rg] alternations.
54. Cf. also Sapir (1925), who considered the "mental" representation
of all occurrences of [0) in English to include /g/.
55. The claim about native-speaker competence being made here can be est-
ablished in several ways. Speakers of the dialect under discussion who
have some phonetic training simply report that, for example, [htogo] is
a possible word of English while [IhLtgso] is not. A more naive speaker
175
might be asked to repeat words spoken to him; his responding correctly
to [hLogo] while distorting [htogsoj to [hLoso] (or, if the [g] catches
his attention, to (hLogaso]) or reporting that he is unable to repeat
[hLogso] exactly as it was spoken to him, would support the hypothesis.
So would a subject's reading hingo (presented as a made-up word) with [g]
but hingso without.
56. Note that /g/-Deletion is another rule that must be ordered before
Rule V: V would cause the /g/ of hen it to become [+S.I.] but the /g/
must be deleted.
57. Of course, /1, r, w, y/ is not an unnatural class (= [-syllabic,
+sonorant, -nasal] and it might be countered that one should modify (72)
so as to apply before #, nasals, and non-sonorants. (73) is preferable
because rather than representing the complimentarity of /Qg/-simplifica-
tion environments and environments in which syllable-initial /g/ is pos-
sible as accidental, it relates them in a principled way. Furthermore,
(73) allows the elimination of a conjunction of # and classes of segmen-
tal units. Finally, since the /g/ of anglicize, etc., is [+S.I.] phone-
tically, (73) is merely making use of independently motivated phonological
machinery.
58. As ib well known, at an earlier stage in the history of English, [0]
occurred only as the pre-velar allophone of /n/; there were no words like
[sLo], and sing was pronounced [sLog]. I would attribute the appearance
of free [0] in words like sing, Yngve to the addition of Rule (73) as an
(Early Modern English) phonological innovation.
176
59. By the standard argumentation, character cannot be bisyllabic
/ksrVktr/: cf. characteristic *[karektrLstLk].
60. pronounced [Iliste] by Alistair Cooke.
61. cf. ministerial: minister is underlyingly trisyllabic.
62. For example, SPE (p. 148) argues that underlying geminates and a de-
gemination rule are required for English. If this hypothesis is correct,
Mississippi could have /ip/ or /ipp/ as its pre-final cluster, as far as
segmental considerations are concerned. The choice of /ipp/ is motivated
entirely by the desire to help reduce the stress rules to regularity. On
the other hand, simply marking Mississippi for penult stress is equally
(though more obviously) ad-hoc.
(Regarding gemination as an account of anomalous penult stress assign-
ment, note that since voiced-obstruent clusters are exceptional in English,
one should not expect voiced-obstruent geminates in lexical representa-
tions. This prediction is noted in SPE, where it is claimed that in fact
the geminate artifice is required only in the case of voiceless obstru-
ents and sonorants. In this regard, consider Aladdin, Armagejddon,
Chappaquidick, elAven, Manischewitz, Nebuchadnzzar, pjlyer.)
The largely idiosyncratic nature of English stress placement is
more explicitly recognized in more recent formulations of the stress
rules. For example, in Halle (1973a), it is acknowledged that the con-
trast between a pair of nouns like Ivan, icon is a matter for ad-hoc
lexical marking (p. 454-5).
177
63. In the small class of words where even the closed-penult principle
fails, an observation made elsewhere in this Chapter is relevant. If
words like Jespersen, Jef&rson, nergy, Beverly, whose penult vowel is
phonetically [r], are taken to have underlying /r/, then the penult is
open, and the observed antepenult stress is unparadoxical. See also
Section 8.3, as well as Section 3 of Chapter III.
64. I mentioned in Chapter I that the only other formal phonological
analysis known to me which incorporates the notion of ambisyllabicity
is that of Anderson & Jones (1974), and I indicated there why I felt
their general approach is wrong.
A&J apply their syllabic formalism to just one phonological problem,
that of English stress assignment. It turns out, however, that their
treatment, rather than being an improvement over the SPE stress rules,
actually tails to work in one of the most straightforward cases.
In A&J's theory, all syllables have maximal initial and final clus-
ters, regardless of the amount of overlapping entailed. Boston, for ex-
ample, consists of the syllables /bost/ and /ston/.
Like SPE, A&J attempt to define "weak cluster" in such a way that
the part of the MSR that assigns non-final stress reads in essence, "as-
sign antepenult stress if the penult is weak, penult stress otherwise."
They claim, however, that through the use of syllable structure one can
give a simple definition of weak cluster not involving the mention of
special consonant classes, as the SPE definition does.
They note that under their syllabication conventions, the penults
of cinema, algebra, veranda are /nem/, /geb/, /raud/, respectively. Since
178
only veranda has penult stress, and only it is closed by more than one
consonant, their first approximation to a definition of "weak cluster" is
V C0 ], where brackets enclose syllables.
The above analysis is based on English nouns. In verbs the weak/
strong cluster principle finds applicability one syllable further to the
right, choosing between final and penult stress. Here also A&J's defini-
tion seems workable: edit, final syllable weak /dit/, vs. collapse,, final
syllable strong /laps/.
A&J then note that a noun like pentathlon has a weak penult /ta8/
by their definition, yet penult stress is observed and seems generally
required. On the basis of this sort of example, they modify their weak-
cluster definition. To produce correct results in pentathlon, a conven-
ient definition of weak cluster would be Vj, a definition which is also
satisfactory for correct assignment of stress in words like veranda.
However in verbs the original 1 C) is required. Taking advantage of
the fact that for verbs the pivotal syllable is final and thus followed
by #, they conclude that two subcases are needed to define "weak cluster,"
V] and V C0 #1, or in abbreviated form, V ( C0 # ) # . This is their final
formulation.
What A&J seem to have overlooked is that while the noun expansion
V] gives correct results for pentathlon and veranda, it fails for algebra
(penult = /geb/), and more importantly for the whole class of unproblema-
tical antepenult cases, cinema (penult = /nem/), for example.
More than inadvertancy is involved here. From a more general perspec-
tive, A&J have chosen a poor example to illustrate the possibility of non-
discrete syllabication. As we have seen in this chapter, English stress
179
assignment differs from many other phonological rules in applying at a
level at which ambisyllabicity, necessary for accurate phonetic represen-
tations and relevant to the conditioning of certain low-level rules, is
not yet appropriate. Note that Rule II-produced penults for antepenult-
stressed cinema and algebra are /ne/ and /ge/; for penult-stressed veranda
and pentathlon, /ran/ and /taO/.
65. Penults containing tense vowels are always stressed in English.
It would be desirable to be able to discover a unifying principle linking
tense vowels and vowels followed by two or more consonants in their
stress-attracting properties. The fact that tense-vowel syllables are
always closed by a post-vocalic glide suggests that Rule (79) may repre-
sent such a link. However this generalization is not capturable without
a major reformulation of the SPE analysis of the deep phonology of Eng-
lish, for in this analysis Diphthongization/Vowel Shift must follow the
M.S.R.
66. In Halle's most recent analysis of English stress (class lectures,
M. I. T.), the need for an independent rule assigning initial stress is
obviated by a reformulation of the Main Stress Rule as a rule schema ex-
panding to assign the feature [+stress] indefinitely many times in indef-
initely long words. Since in particular initial stress will always be
assigned, the critique of Halle's published formulation of the facts of
initial stress I give in this section is not invalidated by his reanalysis.
67. All the words of (82) are exceptions to Destressing. Those in (b)
are in addition exceptions to Halle's (1973a) Detail Rule, which chooses
one among several [+stress] vowels as the position of main stress.
180
68. These examples should perhaps be viewed as a challenge to the assump-
tion that "related" forms must have continuous strings of identical und-
erlying segments rather than to the system of stress assignment rules.
In the same way that the word majority has semantic properties beyond
those predicted by its components major and -ity, it has phonological
properties not predictable from those of major combined with general pho-
nological considerations. Cf. Section 2.2 of Chapter III.
69. Cf. Fidelholtz (nd). Fidelholtz shows that destressing/reduction
in #C --- C2 V correlates with frequency of usage of the word under con-
sideration. The fact that an uncommon word like gustAtion fails to under-
go destressing cannot be allowed to cast doubt on the correctness of a
syllable-based account vis-a-vis Halle's (1973a) treatment however, since
even words with a single consonant following the first vowel follow Fidel-
holtz' generalization (he limits his attention to the C2 case) rare Negev,
chateau vs. common terrific, Xmerica.
70. I assume that syllabication is discrete in French, as it !sually ta-
ken to be. If not, read "only if the nasal is fully within the vowel's
syllable." The latter condition is appropriate for the Erglish /m/-
raising rule.
71. If the convention suggested in the final paragraph of Halle (1971)
is taken at face value, it is in fact (87) and not (85) which is predic-
ted as a proper generalization of (84).
72. There are two types of exception to this statement. Some dialects in
Scotland, for example, have trilled and uvular-fricatve /r/s. Secondly,
within the more widespread dialects, there is an allophone of /r/ involv-
181
ing frication: compare train with frication vs. rain without.
73. As opposed to "post-vocalic" /r/, syllable-initial /r/ may involve
some labialization: cf. bar and rob in most American /r/-retaining dia-
lects.
74. Among U. S. radio and television announcers, who are undoubtedly
taught that it is wrong to pronounce words such as latter and ladder
alike, a quite different way of distinguishing them is often heard. Un-
like Kenyon, who voices both /t/ and /d/ but articulates them differently,
these speakers "flap" both but keep flapped /t/ voiceless.
75. I transcribe the vowel in these words as [o] since it is mid. The
vowel is not as high, however, as the vowel of four in E1 .
76. Thus it is possible to distinguish three major dialect areas in the
U. S., each represented by tens of millions of speakers:
E1E2 W
for - four? no yes yes
for(est) - for(um)? no no yes
The fourth possibility, forest - forum but for * four, is not found
anywhere in the English-speaking world, as far as I can determine.
The "r-less" vs. "r-ful" split is 1 dependent of the classification
above. Of the six possible combinations, at least five occur:
182
r-less E
r-ful E :
r-less E2
r-ful E2:
r-less W;
r-ful W:
77. There are
St. Louis, Mo.
E dialect.
eastern New England
Inland southern U. S., Scotland
standard British English
middle Atlantic coast
midwest and west U. S.
some exceptions to this geographical generalization. The
area, well to the west of Thomas' isogloss, displays an
78. Most of those regions of the U. S. which have undergone rules (e)
and (i) in addition have [r] as the first vowel in words like courage,
hurry, while the E1-E2 areas show [A]. This third dialectal split should
quite probably be related to the first two, though it differs in showing
an assimilation La addition to a raising (and a raising to high rather
chan mid position, if the remarks of Appendix I are correct in general.)
183
CHAPTER III
ON WORD-BASED GENERALIZATIONS WITH SYLLABIC CONDITIONING
Section 1 - A Hypothesis
Three types of phonological generalization requiring a syllabic treat-
ment were isolated in Chapters I and II:
(1) a. processes like /t/-Glottalization (Chapter II (28)), which
are ordered after Rule V (Chapter I (32)) and apply on all
domains without regard to syntactic boundaries
b. processes like /ar/--*/r/ (Chapter II (49)), which are
ordered before Rule V and thus in effect apply only within
the word
c. generalizations like the absence of syllable-final /tk/
(Chapter II (96)), which correspond to simple constraints
on possible syllable structure
The reasor.s for regarding processes like /t/-Glottalization as being
ordered after Rule V and as being free of syntactic conditioning are fully
discussed in Chapter II. I turn now to a closer look at processes of the
types (lb) and (lc).
Recall first the discussion of Section 6.1 of Chapter I, where it
was pointed out that under the standard assumptions, there are two sources
of general phonetic gaps *[A] / P - Q: (a) /PAQ/ is absent from lexi-
cal representations and ungenerable from allowable sequences by means of
rules of the phonology; (b) /PAQ/ is permissible underlyingly but subject
184
to an obligatory phonological rule R which destroys the underlying config-
uration, and [PAQ] is not otherwise generable.
It is generally assumed that analysis (b) is justified as an account
of the absence of phonetic [PAQ] only when there is independent evidence
for underlying /PAQ/ or for rule R (preferably both). In the case of the
absence of [...tk{C, #1],1 for example, there is nothing in the phonology
suggesting that we are dealing with a configuration that needs to be al-
lowed underlyingly, nor is there any independently-motivated rule of
English phonology which would prevent the sequence from reaching the sur-
face, were it present in lexical representations. Thus we must view this
gap as resulting from a constraint rather than from a rule (a syllable-
structure constraint, if the arguments of Chapter I, Section 6 are valid).
Now note that rule (2) (= (49) of Chapter II),
(2) r
[+bk]
ordered before Rule V, effectively prevents the occurrence of phonetic
[...ar{C, #)], giving rise to a phonetic gap similar tc that produced by
the underlying constraint which excludes [...tk{C, #}]. Implicit in the
discussion of /wr/ in Chapter II was the assumption that there is justi-
fication for allowing /arC, #}/ in lexical representations and including
an /wr/-.'/or/ rule, (2), in the phonology; hence the mention of (C) in
(lb). This assumption must be examined. If justification is not forth-
coming, we must be prepared to view the absence of [...ar(C, #}1 as being
on a par with the absence of [...tk{C, #}], the result of a constraint on
185
possible syllable structure. Such a constraint would be found in Rule
IIb and would have the form shown in (3):2
(3) * V r
S
As noted in Chapter II, the [...aer{C, #}1] gap and several others
like it arose through the introduction into the phonology of a syllable-
conditioned rule. A hypothesis that I would like to put forth is that
(a) as innovations, syllable-conditioned rules always enter the phonology
ordered after Rules I-V, and (b) such rules either remain in this posi-
tion or else are re-interpreted as syllable-structure constraints,3 but
(c) rules cannot be re-ordered so as to follow Rule II but precede Rule
V, the reordering mentioned in Chapter II as a possible way of accounting
for the [a.] of car is vs. the [] of carry, the [E] of plan it vs. the
[m] of planet, etc.
In support of the first part of my hypothesis, I would point out that
to my knowledge, it is never necessary to postulate the introduction of a
syllable-conditioned rule at a point other than somewhere after Rule V,
and that sometimes direct evidence of initial post-V ordering is available
in the case of generalizations which are now constraints of type (lc), or,
if part (c) of my hypothjsis is wrong, perhaps rules of type (lb).
For example, the absence of [0] followed by [-S.I.] [g], dibcussed
in Section 7 of Chapter II, is a syllable-based generalization of type
(lb) or (lc), for were the generalization due to a post-Rule V /g/-dele-
tion rule, the distinction between fungus [0g] and hung us [0] could not
186
be accounted for
(4) a. f A Q g a s b. h A 0 g a s
\VAV
S S S S
Rule V would introduce the line of association shown dashed in (4b), de-
stroying the distinction between fungus and hung us. Thus either the
/g/-deletion rule is ordered before Rule V, or has been lost from the
grammar and replaced with a syllable-structure constraint of the form
shown in (5):
(5) *Vj0 g
S
However we know that at the time (17th century) that /og/-simplifi-
cation came into the language, the rule was ordered after Rule V, i.e.,
that although hung and sint had [r)] in isolation and before consonants,
phrases like hung us and sing aloud had [0g] when spoken without pause.
(See Jespersen, 1909:1:217 for documentation.)
As a second example of this type, recall the discussion of /r/-loss
Chapter II, Section 2), which arose as a post-Rule V rule, but survives
in standard British English as a constraint against post-vocalic /r/ in
syllables.
The part of the hypothesis outlined above that I want to devote most
attention to is the suggestion that ordering of rules between syllable-
structure assignment rules IT and V should not be allowed.
In the sections that follow, I investigate several phonological
187
generalizations of Chapter II that might be taken, synchronically, to be
due to a syllable-structure-conditioned rule of type (lb) and attempt to
show for each that there is little or no evidence for the existence of a
rule, and, where possible, that there exist facts which argue specifically
for a solution in terms of a constraint. Corresponding to each generali-
zation I assume a syllable-conditioned phonotogical statement (rather
than a boundary/segment conditioned one) on the basis of the observations
of Chapter II, and limit my attention to the issue of whether a constraint
or a rule is involved.
Section 2 - /mr/
In this section I look into some of the pros and cons of deriving
words like car, cart [Ar] from underlying /ar/ by means of Rule (2).
2.1 There is a strong intuition that hypothetical forms like [kar],
[kaort] could not be words of English. While the rule analysis (2) adequate-
ly accounts for this intuition, so does the constraint analysis (3).
Thus there is no reason to postulate an /wr/---*/Or/ rule on these grounds,
just as one could not justify a /t/ --- * /s/ / - k rule on the basis of
the undeniably valid intuition that *retk is not possible, while *resk is.
2.2 The strongest argument for the inclusion of a postulated rule In
the phonology is the demonstration that it is needed to account for a gen-
eral phonological alternation. In the case of (2), no such evidence is
available.
SPE, which derives [ar] from [ar, offers (p. 216) four pairs of re-
lated words showing an [ar]/[ar] alternation: bar/barrier, bar/barrister
188
car/carriage, par/parity. These altzrnations can hardly be considered
convincing (and are not claimed to be in SPE). First of all, there exists
only a handful of examples. Secondly, with the possible exception of par/
parity, the alternants are not related by general morphological rule.
Thus it is not necessary that the initial underlying segments of barrier,
for example, be those of bar, since barrier is not the result of the suf-
fixation of bar. (In contrast, consider bat [tJ/batting [D]; unless
[0] and [D] are derived from a single underlying segment in these words,
it will not be possible to generate batting by means of the general rule
Participle = [ I....] ingi.)
VV
Thirdly, when a wider range of facts is considered, the cited exam-
ples of alternation appear to be accidental. pair is as closely related
to par as parity is, yet under the SPE analysis pair [per] must be from
/pir/, while par is /par/ underlyingly. That is, allomrphy in weakly re-
laced forms must in general be tolerated.4
Compare also bare (< /bir/) vs. barren (< /bwrVn/), which are at
least as close semantically as any of the SPE alternants. If barrier
must be from /BAR+ier/, where /BAR/ is the lexical representation of bar,
why need barren not be representable as /BARE+Vn/?
Fourthly, [or] from derived /ar [C, #]/ fails to be observed where
expected. From scarce (< /skirs/) + ity we expect [skarstDi] via Trisyl-
labic Laxing and /wr/--*/ur/, yet scarcity is [skersLDi].
Similarly, consider bear [ber] vs. birth. Since bear must be /bir/
underlyingly, we expect /bir+O/ for birth, becoming /barO/ by pre-cluster
laxing (SPE:172) and finally *[barO].
As another example of this sort, note that when Peter [piDer] is
189
abbreviated to Pete, the shortened form is subject to the rules of the
phonology: [pit?], yet Lar', short for Larry, is [lmr],5 not [lor].
Finally, note that the optional rule that accounts for the alternants
[jenaral]/[jEnrel] for generl produces, with Marilyn as input, [marlLn],
not [marltn].
In any case in which the vowel of derived /ar [C, #]/ fails to become
[a] it might be argued that jar/-*/ar/ is ordered before the particular
rule creating /er [C, #]/. While this account is certainly possible, there
some reasons for rejecting it. First of all, such "counter-feeding" order
am-ng phonological rules appears to be universally marked (cf. Kiparsky,
1968b). Secondly, the rule shortening Larry to Lar' is presumably a mor-
phological rule; phonological rules would be expected to follow such a
rule (cf. Peter/Pete; see Section 4.4 for some additional discussion of
this point). But most importantly, note that under an analysis in which
there is simply a constraint against tautosyliabisl/ar/, there is an im-
mediate explanation to our observation: derived /mr/ remains because there
is simply no /ar/--+ /ar rule. More generally, the fact that no rule
appears to precede putative /ar/ -- /ar/ falls out as a direct consequence
of the constraint analysis.
In summary the evidence from alternations does not support a syn-
chronic /ar/--+/or/ rule, and in fact may be viewed as favoring a con-
straint analysis.
2.3 The assumption that symmetry is a desirable property of under-
lying segmental systems might be used to justify an analysis in which
[or] is underlyingly /mr/, for unless /#Xair [C, #]/ is allowed in lexical
190
representations /r/ will provide an exception to the generalization that
/aBC#/ is possible underlyingly for all C. However, in the absence of evi-
dence that language learners seek to set up phonological systems with this
sort of symmetry, argumentation of this type cannot be considered persua-
sive. Furthermore, there are other gaps in /Xw - #1; /6/, for example,
does not occur in this environment.
In addition, as will be seen in Section 3, there must exist gaps of
the form /V r [C, #]/, where V is a non-low vowel. This fact is not only
an indication that the kind of asymmetry that would result from the ex-
clusion of tautosyllabic /wr/ is tolerable, but also that the absence of
the latter configuration may be part of a wider generalization.
2.4 In view of the above considerations, it must be concluded that
there are no reasons at all compelling for assuming that a word like cart
is underlyingly /kart/ and that there exists an /ar/ -- w /cr/ rule in the
synchronic phonology of English. If these assumptions are dropped, a
rule not supported by alternations can be omitted from the phonology,
the disparity between phonetic and phonological representations can be
lessened, and the fact that all phonological rules appear to follow hypo-
thetical /r/-P./r/ can receive a simple explanation.
Section 3 - Short Non-low Pre-/r/ Vowels
Recall the historical rule, discussed in Section 3.3 of Chapter II,
which merged /i, e, u/ in - r [C, f]. For ecsentially all words with
phonetic [r] as a result of this rule (for ex., first, herd, curse),
there is no synchronic evidence as to which of the three pre-merger
191
vowels is the precursor of modern [ri. Thus a single vowel, perhaps /r/,
perhaps /i/, /e/, or /u/, must underly all words of this type. We must
countenance, then, at least two systematic gaps of the form "/VC#/ not
possible when C = /r/." (Cf. discussion in Section 2.3.)
Let us consider whether there is any evidence that tautosyllabic
Or/ must be allowed underlyingly at all, where V - /i/, /e/, or /u/,
and that the historical rule is also a synchronic one.
It might be argued that heard [hrd] is synchronically from /h-er#d/
(hear + past morpeme /d/) via the special #-deletion rule postulated for
use where inflected forms unexpectedly undergo pre-cluster laxing (cf.
kept; see discussion in SPE, p. 369), pre-cluster laxing, and a synchron-
ic /er/- /r/ rule. There are several reasons for rejecting this analy-
sis.
One is the existence of dialects, like my own, in which erred is
only [Erd],6 while heard is [hrdj, not [herd]. That is, underlying /er/
does not reduce to [r] synchronically.
Secondly, observe that since all non-low short vowels merged before
tautosyllabic /r/, four of the six underlying "vowel shift" vowels postu-
laced by SPE (i.e., all except /i, 5/) should surface as [r] when short-
ened before tautosyllabic /r/, if the neutralization rule is a part of
the synchronic phonology. The absence of additional examples, then,
does not speak well for synchronic merger. Furthermore, there is at least
one case of [r] "derived" from the pre-cluster laxing of one of the two
vowels which should not feed the neuralization rule: birth [brO], from
bear [bern, lexical /bir/. This sort of example makes it clear that to
cite isolated heard as evidence for synchronic /er/--.'/r/ is totally
unjustified. Cf. fn. 4. 192
There are some reasons for choosing an analysis with underlying /r/
over one with a /[i, e, u] r/-- n/r/ rule beyond the lack of demonstrated
need for the latter rule. Recall the class of words mentioned in fn. 63
of Chapter II (Bevgrly, Jesp6rsen, etc.) whose antepenult stress is anom-
alous if the penult is taken to be /VrC/. Underlying /rC/ would reduce
these examples to regularity from the point of view of the stress rules
and would capture the phonetic representation directly.
Similarly, it was noted in Section 8.3 of Chapter II that words whose
spelling suggests a lexical representation of the form #C0VrCVX# (Bermuda,
Vermont, etc.) are almost always exceptional with respect to the generali-
zation that strong initial syllablee retain stress in pre-stress position.
Here also the hypothesis that /r/ appears in underlying representations
simplifies the stress facts and lessens the disparity between lexical and
surface representations.
Finally, as an observation paralleling that of Section 2.1, note that
allowing underlying /r/ and prohibiting tautosyllabic /ir/, /ur/, and in
many dialects, /er/, is as descriptively adequate an account of the in-
tuition that words with [{L, A, (e)} r (C, #}] are deviant as a rule serv-
ing to destroy this configuration.
Summarizing, the lack of evidence for a synchronic /ir, ur, er/ ---.
/r/ rule, together with the desirability of phonemic /r/, argue against
retaining the historical rule as a synchronic rule of type (lb).
Section 4 - Tensing and Raising of Pre-nasal /o/
In this section I discuss the possibility that the /a3/ -- /E rule
of Chapter II, Section 5,
193
(6) ( Chapter II (61))
E
which accounts for the contrast mat, manage [m) vs. man, mandible [E,
has been lost from synchronic grammars and replaced by a syllable-
structure constraint of the form7
*m im, n}
V
S
4.1 As in the case of the generalizations discussed in Sections
2 and 3, strong intuitions about I/m and /E/ are observed, but do not
call for a rule analysis rather than a constraint analysis: either is
fully capable of accounting for the fact hypothetical *zan, *zanda, have
/E/, not /m/, *zana /a/, not /E/.
4.2 Nor can one argue for the existence of an /w/--/E/ rule on
the basis of [m]/[E] alternations, of which there is a general absence.
Even where alternations are expected, they are not observed. tanner
'more tan' has [E], just as tan does. Jan', short for Janice, has the
full name's [w] (cf. lexical Jan, with [E]).
4.3 In the case of the /m/--+/E/ rule a further demonstration
that lexicalization and rule-loss have taken place Is possible. I noted
above that an occurrence of /m/ which is in the environment of the puta-
194
tive tensing/raising rule by virtue of what appears to be morphological
derivation (Janice -o Jan') remains low and lax, suggesting that the pho-
nological component, acting on the output of the morphological rules, con-
tains no /a/-*/E/ rule.
Another process, even more clearly morphological, serves to create in-
stances of /k/ followed by tautosyllabic /m, n/, and here too /a/ remains.
The process is the rule which phonologically interprets the past-tense
morpheme, when attached to r member of a certain class of verbs, as ab-
laut of a [+high] stem vowel to /m/, as in sitsat. The crucial examples,
began, ran, swam, have fml, not /E/,8 suggesting that there is no // --
/E/ rule.
4.4 It is instructive to consider the sequence of grammar changes
which served to create the contrast between, for example, Japan [E) and
began [ai].
We can assume that Rule (6) was added to the end of the phonological
component at a certain point in the history of the dialects under consid-
eration, given the prevalence of this sort of sound change (cf. King,
1969, 1973). Such a change would give rise to [E] in both Japan and
began. We must then account for the "re-appearance" of [w] in began.
We cannot hypothesize that Rule (6) was simply lost from the grammar
at some subsequent point in time, for we would then be unable to account
for the retention of /E/ in Japan, and more importantly, for the intuition
that a hypothetical form like zan has /E/.
The problem under discussion has the appearance of one in which rule-
reordering would offer a solution. In fact if one postulates that the
195
tensing/raising rule was reordered so as to precede the ablaut rule, the
Japan/began contrast can be accounted for:
(7) /wj pwa n/ /b a g i n/FAST
E --- -* E
-aAblaut
[E] [M] Output
However, in spite of the fact that the derivation in (7) gives observa-
tionally-correct results, there are two general theoretical considerations
which suggest that the hypothesis that Ablaut and Tensing were reordered
must be rejected.
First of all, in the original order, Tensing is fed by Ablaut and
is a perfectly transparent rule, for there are no occurrences of [M]
followed by [-S.I.] [m, n], the structural description of Tensing, on
the phonetic surface, or of phonetic [E] from other sources.9 One would
not expect a new ordering to arise in which the two rules are in a counter-
feeding relationship and Tensing has become opaque, as is the case in the
ordering illustrated in (7).
Secondly, the ordering of (7) represents an interspersing of morpho-
logical and phonological rules. It is in general theoretically preferable
to have two rule-components, each with strong restrictions on rule-types,
rather than a single component with restrictiors weak enough to accomodate
rules of both types (cf. Chomsky, 1972:Sec. 2.2), and in fact a segre-
gation of morphological and phonological rules nearly always seems possi-
ble (see Aronoff, 1974). On these grounds too, then, the reordering analy-
sis of (7) seems poorly motivated.
196
On the other hand, suppose a constraint analysis of the [m/L[E] facts is
correct. As suggested in Section 4.3, such an account offers an explana-
tion of the [w] of began, etc., since these preterites, as derived forms
rather than lexical items, would not necessarily be in conformance with
constraints on underlying representations.
Under the constraint analysis, the generation of language learners
following the one in which the // --. /E/ rule was added as an innovation
lexicalizes /E/ wherever possible, for example in Japan, and sets up a
syllable-structure constraint to account for the distribution of /a/ and
/E/.
The [E] of began, observed by the language-learners in the speech
of their parents' generation, will be a special case, since began is the
result of a morphological operation on the lexical item begin rather than
a lexical item in its own right and its [E) cannot be "lexicalized."
Since the Ablaut rule which produces /4/ is needed for a class of verbs
in which /E/ is not involved (sit, etc.) it will continue to operate in
the grammar; if begin is taken to belong to this class of verbs by the
new generation,10 Ablaut will produce /w/ in begin + Past.
The reversion of /E/ co /w/ might occur at just this point, as a re-
sult of imperfect learning, i.e., the result of the toleration, for the
sake of overall simplicity, of [w] in began where the older generation
has [E].
However, we need not make the assumption that the replacement of the
rule by a constraint will have as an immediate consequence the reversion
of [El to [L] in began. Suppose the /x/ to /E/ rule, though not used in
the derivation of Japan and thousands of similar words, is preserved to
197
take the // of preterites like began to /E/ so that the grammar of the
new generation can produce the parents' output:
(8) /j ap E n/ /b a g i n/
a- Ablaut
---- E w -+ .E
[E] [E] Output
Now in the analysis summarized in (8), the /a/-/E/ rule, though not
requiring any special restrictions, is a very minor rule in that it is
used in the derivation of only a handful of forms. The loss from the
grammar of such a rule would be a natural type of phonological change.1
With the rule gone, the ablaut-generated /w/ of began will surface. I
would suggest that such a loss has in fact taken place in the dialects
in which Japan has [E], began [m].
If this analysis, which seems to be the most reasonable account of
the exceptional [a] in words like began,12 is correct, the distributional
facts and native-speaker intuitions discussed above are due to a syllable-
structure constraint against tautosyllabic /mm/ and /mn/, and there is no
synchronic /a/-+/E/ rule.
Section 5 - The Velar Nasal
As a final example of a phonological generalization which might be
taken to be due to a rule of type (lb) but which upon closer examination
appears to be better described as a constraint of type (lc), I discuss
in this section the absence of [r] in position before non-syllable-initial
[g].
198
5.1 Any analysis of the velar nasal in English must account for
the fact that hypothetical words like [hmog], [mogme] are not observed,
and, more importantly, for the strong intuition that there could exist
no such words. As detailed in Chapter II, Section 7, the SPE analysis,
buttressed by a syllable-structure condition on the /g/-deletion rule,
accounts for these facts. On the other hand, if there is no /g/-deletion
rule, the distributional and intuitional facts just mentioned are due to
a constraint against syllables having the shape /C0 VQg/. There is thus
no reason for preferring a rule analysis to a constraint analysis on
these grounds.
5.2 The rule analysis, which derives all [0 from /ng/ in -[C, #1,
has no way of accounting for phonetic [VV] in those cases in which it is
not possible to motivate # between /o/ and the following vowel. As it
stands, the constraint analysis, which allows syllable final /0/ (hang
would be from /hmo/, for example), predicts that morpheme-internal [VV]
should be possible. As is well known, there do exist occurrences of this
sequence, as in hangar [hmoer]. In what follows I argue (a) that the
general system of rules and constraints it the phonology should not exclude
all instances of [VOV] and (b) that the recessity of allowing [VV] weakens
the rule analysis vis-a-vis the constraint analysis.
The words in (9) display the sequence [VOVI:13
(9) dingaling 'fool'1 4  orangutan
dinghy Ringling (Bros.)
hangar Schlesinger15
dingus Singapore
humdinger Singer (Sewing Machine Co.)
199
One might challenge the significance of these examples by pointing out,
correctly, that they are limited in number. It must be kept in mind,
however, that the status of this observo:ion is quite different from the
status of the fact that there are no words with syllable final [Lg]. The
conclusion that there are very few words with [VQV] is the result of the
study of a large number of English words, a statistical fact not directly
testable for psychological significance.16 That the absence of words
with syllable-final [0g], on the other hand, is a reflection of native
speaker competence is directly confirmable.
In support of the claim that there is no direct evidence that the
paucity of [VrV] words is more than accidental, note that words like
hangar are in no way sensed as deviant or as somehow different from the
vast majority of other common English words.17 Nor is it clear whether
a hypothetical word like zanger should be pronounced with [g]18 (Of course,
in some intervocalic environments, [0g] seems decidedly better than [0J:
consider, for example, hypothetical Zingdma. I do not mean to imply that
[0 is freely occurring in V1 -- V2 for all VS, V2 and all stress patterns;
certain specific constraints against the occurrence of [V0V] undoubtedly
have demonstable psychological reality. I am arguing in this section
against a phonological analysis in which all instances of [V0V] are ex-
ceptional.)
Consider next a somewhat indirect piece of evidence regarding the
status of the sequence [VQV], an observation which again suggests that
the rarity of this string is, at least in certain environments, acciden-
tal rather than systematic. The word hangar is a direct borrowing from
French (first attested for English in 1852, in the meaning 'large shed')
200
and in French the j is pronounced. For some time the . was pronounced
in English as well. The 1933 edition of 0. E. D. gives only the hybrid
pronunciation [hagar].19 [heogar] was at one time standard in the U. S.,
with [haearI becoming the norm in this country only in the'last genera-
tion or two (cf. Thomas 1958:82; Bronstein 1960:222). Thus unless one
accepts [V0V] as a phonetically well-formed sequence of English, one must
claim that the borrowed word hangar drifted from a natural (cf. anger)
to an unnatural pronuciation from the point of view of the phonology of
the borrowing language, an unacceptable result. This observation sup-
ports the suggestion that in a theoretical framework that aims at psycho-
logical accuracy, the generalization that words with [V0V] are rare may
be better left uncaptured.
Modern lexicography apparently also recognizes that there is no
longer a systematic gap in English with respect to [V0VJ in certain en-
vironments. For example, in the entry for Engels, RHD gives [eOalzi as
the pronuciation of the name of the German socialist leader and [e~gelz]
as the pronunciation of the name of a Russian city, corresponding to the
absence and presence, respectively, of [g] in the German and Russian pro-
nunciations of this word. These transcriptions do not, however, repre-
sent an attempt to describe the pronunciation of Engels in these foreign
languages - non-English pronuciations are always noted as such (cf.
"Friedrich [fridrLk], Ger. [fRidRLx]"). Note also the [ergelzj, the pro-
nunciation given foi the Russian city, incorporates an anglicization:
there is no [1 in Russian, even before velars, but (engols), the actual
Russian pronunciation, would be unnatural for English. Thus we may con-
clude that these entries have been modified as much as necessary to make
201
them conform to the general principles of English phonology, as perceived
by the panel of native speakers who prepared RHD's transcriptions, and
that in the case of [eralz], no anglicization was felt to be necessary.20
In summary, if words like hangar are not deviant from the point of
view of English phonology, then even a rule analysis of
the deviance of tautosyllabic [rg] must allow underlying /o/. But if
/Q/ is a phoneme, there is no barrier to its use in the lexical repre-
sentation of a word like hang and in the vast majority of words with pho-
netic [9].
5.3 A remaining obstacle to the dropping of the /g/-deletion rule
from the grammar might be the desire to account for [0]/[Qg] alternations
in a way not involving allomorphy. In this regard it is essential to note
that here, as in the case of the putative rules discussed in Sections 2, 3
and 4, the evidence from alternations is almost vanishingly small.
I am aware of only three words with [01 that alternate with forms
with [0g]: long, strong, and young. ([ag] is found in the comparative
and superlative of each of these, as well as in elongate (-ion) and
prolongate (-ion); length has an optional [k] which can be taked to be
from /g/ by assimilation to /0/.) Thus whatever the source of [g] in these
few examples, they can hardly be viewed as evidence for analyzing all oc-
currences of free [0)] as surface manifestations of underlying /ng/.
Furthermore, it is not clear how an alternation such as long/longer
would be handled even in an analysis with a /g/-deletion rule. It is
suggested in SPE (p. 369-70) that a special rule deletes the internal #
in comparatives and superlatives, which have the form [#[#...#]er/est#,
A A
202
at the input to the phonological component, allowing final underlying /g/
in longer to surface, since /g/ is deleted only before #. However the
applicability of the #-deletion rule cannot be stated in general terms.
It could not be conditioned by the morphemes long, strong, young, since
it does not apply in strongly, longish, etc., *[0g]. Jor could the more
likely possibility, that the comparative and superlative suffixes are
general conditioning factors for #-deletion, be true. First of all, the
comparatives of adjectives which are not normally compared, for example,
hung, wrong, as well as actually occurring forms like lovingest,
swingingest, must retain # since they have no [g]. Secondly, retention
of # in forms like soberer, soberest21 is required under the standard
assumptions, either to allow syllabication of the sonorant /r/ (if the
underlying form of sober is /s5br/), or to prevent Tri-syllabic Laxing
of the initial vowel (if /sobVr/). In summary, then, neither of the some-
what general rules in (10) can be correct.
(10) a. long
strong # Affix
young
b. Adj. # comparativel
f Luperlativei
0
Rather, the ad-hoc (11) would be required:
203
(11) long comparative I4strongt # superlativej
young Js 
l
0
In view of this result, which is equivalent to marking these three
lexical items for insertion of /g/ in the comparative and superlative,
it must be conceded that the evidence from alternations does not support
a synchronic analysis in which all free [0] is from underlying /ng/.
5.4 Finally, let us consider whether the fact that tense vowels
appear to be absent before /Q/ (cf. (69d) of Chapter II) in any way mil-
itates against a constraint analysis.
I point out first that the attempt to identify-the class of vowels
occurring before /o/ with the class of lax vowels fails in both directions.
First of all, in many American dialects, words like hang are pro-
nounced [her]. There is no evidence that this [e] is derived from a lax V.
Secondly, one of the vowels that universally appears before // pho-
netically is [o] (long, song). This vowel is described as phonetically
tense in SPE and furthermore corresponds in general to underlying tense
[i] in monosyllables in the SPE analysis (cf. SPE Section 4.3.7). Thus
in the most obvious analyses, the generalization that tense vowels are
not found before /0/ is falsified.
There are three obvious possibilities for dealing with the occur-
rence of [5d]. One could conclude (a) that [0] does not correspond to
an underlying cluster, or (b) that pre-cluster laxing does not hold in
general, or (c) that [0] derives from some lax vowel in the case of the
environment --- ). In SPE (p. 209) the derivation of [0] from /ng/ and
204
the correctness of the laxing rule are taken to be beyond question.2 2
Thus SPE concludes that a solution to the problem must be sought in (c),
and the suggestion is made that long has underlying lax /o/. But ob-
viously if one presupposes that no tense vowels occur before /)/ and
uses this assumption to argue that some lax vowel must occur underlyingly
in this position, even when general considerations suggest that a tense
vowel is required, it becomes impossible to argue that [0] must corres-
pond to an underlying cluster on the basis of the fact that only lax
vowels occur before it underlyingly.
Furthermore, there are several lax vowels that fail to occur before
[rl, strongly suggesting that deriving [0l from /ng/ is not the right way
of dealing with //'s distributional peculiarities. First of all, there
23 '
are no words of the form [Xee]. Secondly, the sequence [uQ] is complete-
ly absent in English. Finally, note that unstressed vowels before [0] are
always [L]; [a] is prohibited in this position, even in dialects which
have (or allow) [a] in all other unstressed positions.2 4
In addition, it must be noted that the pre-cluster laxing rule is
itself suspect as a living phonological process: note coax, Colgate, James,
traipse, etc. Without such a rule, the absence of tense vowels before /0/,
to the extent that it constitutes a correct generalization, is irrelevant
to the hypothesis that all [0] is from /ug/.25
5.5 In summary, there seems to be no valid evidence favoring a
rule-based analysis of the facts of /b/-distribution over the more
straightforward constraint analysis.
205
Section 6 - Summary
The generalizations discussed in this chapter are alike in that
(a) for each generalization a rule-based analysis seems possible
and has been suggested
(b) each generalization is syllable-based, as demonstrated in
Chapter II
(c) the rule that would account for each generalization under a
rule analysis would have to be ordered prior to the final
syllable-structure assignment rule, Rule V
I have attempted to show that in the case of each generalization a sylla-
ble-structure constraint provides at least as satisfactory an account as
a phonological rule. In view of this result and the fact that the need
for syllable-conditioned rules ordered after Rule V (for ex., /t/-Glotta-
lization) and for syllable-based constraints (for ex., the one against
tautosyllabic /Vtk/) seems beyond question, the hypothesis of Section 1,
repeated here as (12), seems justified in the interest of formulating
the strongest possible theory.
(12) Syllable-conditioned rules enter the phonology ordered after
the block of syllable-structure assignment rules I-V, and
retain this post-V ordering unless replaced with syllable-
structure constraints.
206
Footnotes to Chapter III
1. where C is such that /kC/ is not a permissible initial cluster
2. For example, since the configuration of (3) would be found in /kar/,
/kart/, such inputs to Rule II would be excluded. (3), if correct, is
likely to be part of a more general constraint; see Section 3.
3. It is beyond the scope of this dissertation to attempt to set down
the conditions under which re-interpretation rather than rule-retention
takes place, a problem of great theoretical interest.
4. In fact, pairs of words which are even more obviously from the same
stem than the ones discussed in the text but which require allomorphy
(or, equivalently, ad-hoc phonological rules) are extremely common in
English, a fact often overlooked. As a result, it is possible to find
a few examples of alternations supporting virtually any phonological
analysis. Some illustrations:
a) giant/gigantic "evidence" for a /g/-deletion (or -epenthesis)
rule
b) e2o/egotism: /t/-deletion
c) grammar/grammatical: /r/ --*/t/
d) Peru/Peruvian: /v/-deletion
e) ghost/ghastly: /5/--/M/
f) will/won't: /1/ -D/5/; /1/-deletion
g) money/monetary: /u/ -- /o/ and /t/-deletion
207
h) appear/apparent: // - /e/
i) dynamite (cf. dynamic) /dyn-o-mite: /'/ -. /5/
5. in /r-retaining dialects which have not undergone rule (e) of Chapt-
er II, Appendix 3.
6. err [er] contrasts with air [ern, and our [ar] ([orn and [awr also
possible for our) in my dialect.
[sr], while not a universal pronunciation of err ([r] is probably
more widespread), is not uncommon and is cited in many current dictionaries
(AHD nd Webster, for ex., give [eri as first choice).
7. On the basis of my own observations and those of William Labov, I can
state with certainty that the particular phenomena discussed in this sec-
tion are commonly observed in the speech of New York City and Philadelpia;
I beleive they are more widespread.
8. ran may also occur with /E/
9. For general discussion of feeding and counter-feeding rule order,
rule opacity, ard the significance of these concepts to the theory of
language change, see Kiparsky (1968b, 1971).
10. Not an unlikely choice, since /a/ and /E/ are phonetically close and
since there may be more conservative speakers among the older generation
who have [w] in began, Japan, etc.
11. Note that rule loss, as opposed to re-ordering, is a type of change
which results in a formal simplification of the grammar. For additional
208
discussion of this point, see Kiparsky (1974).
12. There is another often-observed exception to the general appearance
of [E] before tautosyllabic [m, n]: can 'be able' [kan) (cf. can
'container' and the related verb, [kEn]). To be covered by the above
account of the appearance of [m] in began, etc., can, though at one
time [kEn], would have to have failed to be relexicalized as such, so
that [kanJ could resurface upon rule loss. As a somewhat remote pos-
sibility as to why can failed to relexicalize, note that only if can
retains /a/ lexically can its past-tense form could be generated by the
rule that takes fellow modal shall to should. However it is quite like-
ly that some other mechanism is behind the [w] of modal can.
13. In Ringling, [0] is followed not by a vowel but by a non-syllabic
which follows /g/ in initial clusters. Since under the bypothesis being
examined only tautosyllabic /Og/ simplifies and since /0g/, if present
underlyingly, would not be tautosyllabic in this word, it is as relevant
as the other words in (9) to the investigation of this section. Cf.
tingling (ogli.
14. The relevance of this example must not be underestimated. The fact
that dingaling is not part of the formal vocabulary, with an "official"
pronunciation, is if anything an advantage in an investigation of this
sort.
V
15. and similar names, when not [...Jar].
16. The paucity of words with [VV] is of course historically due to the
209
fact that there is no source of such words in earlier scages of English.
What is at issug is whether synchronic phonologies contain a formal
restriction against words of this type, as implied by the SPE analysis.
17. It should be kept in mind that there do exist common words which are
immediately sensed to be outside the normal range of English phonology,
for example, uh-huh 'yes' [NhZ], uh-uh 'no' [AA]. Cf. also Bach, commonly
pronounced [botx].
18. Unfamiliar names in -nger tend to be pronounced without [g]. Note
also that a speaker unfamiliar with a name like Plomber or Abercrombie
would probably not be sure whether to pronounce the b. Just as both [m]
and [mb] are permissible in V - V and both can be spelled mb, so [0
and [0g] seem both permissible in V - V in many cases, both correspond-
ing to the Lpelling p&.
19. 1 have translated all dictionary transcriptions into IPA notation.
20. The fact that some speakers of English have [g] in (Friedrich) Engels
is irrelevant to the argument made in the text.
RHD's pronunciation possibilities for Engels can be compared to
the two common pronunciations of the name Wagner. Speakers of English
are free to use initial [w] when referring to former New York City mayor
Robert Wagner and [v] in the case of composer Richard Wagner since [w]
and [v] are both phonetically permissible in initial position. Wagner
with final uvular /r/ in English or with initial Iw/ in German, on the
other hand, is possible only as a conscious and obvious imitation of a-
nother language's sound pattern. (cont'd)
210
In addition to Engels, note the near-minimal pairs hangar/anger,
Singer/finger, dingus/cunnilingus.
21. Although more, most sober may also be acceptable, there is little
doubt as to the acceptability of the cited forms. AHD, for example,
gives only soberer, -est. Cf. also "...the soberest and most respected
[emphasis mine -DK) of astronomers..." (Pirsig 1974:261).
22. "The other alternative [for phonetic [51 in long, soft, etc.] is
underlying /i/, as in lawn, fraud. This is ruled out in such words as
long and soft, however, since tense vowels do not occur before such clus-
ters..." (SPE:209)
23. In some American dialects, words like hang, anger are pronounced with
[M], in which case /w/ rather than /e/ provides an example of a lax vowel
not found before /o/. In still other forms of American speech, hang and
and anger have [e], as mentioned above; for these dialects we have two
lax-vowel gaps, in addition to the ones to be mentioned below.
The preferred pronunciation (Kenyon & Knott, AHD, RHD) of words like
penguin, Bengali is [ng], although [Og] is also cited.
24. Compare: Rosa's [a], roses [i] or [a]l, dozin' [L] or [a], vs.
dozing [L,], *[8].
25. Since [...iyo], [...uwo], as well as [...u0], etc. seem clearly de-
viant, we presumably need constraints against these configurations.
211
Bibliography
Abercrombie, D. (1967) Elements of General Phonetics, Edinburgh U. Press
AHD - American Heritage Dictionary of the English Language, Houghton
Mifflin Co., Boston, 1969
Anderson, J. & C. Jones (1974) "Three Theses Concerning Phonological
Representations," J. of Linguistics 10:1-26
Aronoff, M. (1974) Word-Structures, M.I.T. dissertation
Bach, E. and R. T. Harms (1968) Universals in Linguistic Theory, Holt,
Rinehart and Winston, New York
Bailey, C.-J. (1968) "Dialectal Differences in the Syllabification of
Non-Nasal Sonorants in American English," General Linguistics
8:2:79-91
Bailey, C. J. (1975) "Gradience in English Syllabization and a Revised
Concept of Unmarked Syllabization," unpublished ms.
Bloomfield, L. (1935) "The Stressed Vowels of American English." Language
11:97-116
Bronstein, A. J. (1960) The Pronunciation of American English, Appleton-
Century-Croft&, Inc., New York
Bronstein, A. J. (1962) "Let's Take Another Look at New York City Speech,"
American Speech 37:13-26
Brown, R. W. & D. McNeill (1966) "The 'Tip of the Tongue' Phenomenon,"
J. of Verbal Learning and Verbal Behavior 5:325-37
Bruck, A. (ed.) (1974) Papers from the Parasession on Natural Phonology,
Chicago Linguistic Society
212
Chomeky, N.
Chomsky, N.
Chomsky, N.
Chomsky, N.
(1964) Current IssuesinLinguisticiThor, in Fodor and
Katz (1964)
(1965) Aspects of the Theory of Syntax, M.I.T. Press, Cam-
bridge, Mass.
(1974) "Some Empirical Issues," in Chomsky, Studies in Seman-
tics in Generative Grammar, Mouton, The Hague, 1972
& M. Halle (1968) The Sound Pattern of English, Harper &
Row, New York
Dingwall, W. 0. (1971) A Survey of Linguistic Science, Univ. of Mary-
land Press
Fidelholtz (n.d.) "Word Frequency and Vowel Reduction in English" Univ.
of Maryland ms.
Fodor, J. A. & J. J. Katz (1964) The Structure of Language, Prentice-
Hall, Englewood Cliffs, N. J.
Fromkin, V. (1971) "The Non-Anomalous Nature of Anomalous Utterances,"
Language 47:27-52
Fudge, E. C. (1969) "Syllables," J. of Linguistics 5:253-86
Fuj imura,
Fuj imura,
0. (ed.) (1973) Three Dimensions in Linguistic Theory, TEC
Press, Tokyo
0. (1975) "Syllable as a Unit of Speech Recognition," IEEE
Transactions on Acoustics, Speech, and Signal Processing,
Vol. ASSP-23, No. 1, Feb. 1975
Gimson, A. C. (1962) Introduction to the Pronunciation of English, St.
Martin's Press, New York
213
Goldsmith, J. (1974) "English as a Tone Language," M.I.T. ms.
Goldsmith, J. (1975) "An Overview of Autosegmental Phonology," M.I.T.
lecture and ms.
Halle, M. (1962) "Phonology in Generative Grammar," in Fodor & Katz (1964)
Halle, M. (1971) "Word Boundaries as Environments in Rules," Linguistic
Inquiry 11:4:540
Halle, M. (1973a) "Stress Rules in English: A New Version," Linguistic
Inquiry IV:4:451
Halle, M. (1973b) Review of Ladefoged (1971), Language 49:926-33
Halle, M. & K. Stevens (1971) " A Note on Laryngeal Features," in Q.P.R.
No. 101, Research Lab. of Electronics, M.I.T.
Harris, Z. (1951) Structural Lingusitics, U. of Chicago Press
Heffner, R.-M. S. (1950) General Phonetics, U. of Wisconsin Pres'
Hoard, J. E. (1971) "Aspiration, Tenseness, and Syllabication in Eng-
lish," in Language 47:133-140
Hombert, J.-M. (1973) "Word Games: Some Implications for Analysis of Tones
and other Phonological Processes," ms. to appear in U.C.L.A.
Working Papers in Phonetics
Hooper, J. B. (1972) "The Syllable in Phonological Theory," Language
48:525-40
Hooper, J. B. (1975) "The Archi-segment in Natural Generative Phonology,"
Language 51:536-60
214
House, A. & G. Fairbanks (1953) "The Influence of Consonant Environment
upon the Secondary Acoustical Characteristics of Vowels,"
J. of the Acoustical Society of Arerica 25:105-13
Jespersen, Otto (1909) A Modem English Grammar on Historical Principles
Joos, M. (1957) Readings in Linguistics, American Council of Learned
Societies, Washington
Kahn, D. (1972) "Syllable Boundaries in Phonological Rules," M.I.T. ms.
Kahn, D. (1973) "On the Relationship Between Vowel Length and Vowel
Quality in American English /w/ and /a/," M.I.T. ms.
Kenyon, J. S. (1966) American Pronunciation (10th ed.), George Wahr,
Ann Arbor
Kenyon, J. S. & T. A. Knott (1953) A Pronouncing Dictionary of American
English, Merriam Co., Springfield, Mass.
King, R. D. (1969) Historical Linguistics and Generative Grammar, Prentice
Hall, Englewood Cliffs, N. J.
King, R. D. (1973) "Rule Insertion," Language 49:551-78
Kiparsky, P. (1968a) "How Abstract is Phonology?" in Fujimura (1973)
Kiparsky, P. (1968b) "Linguistic Universals and Linguistic Change," in
Bach & Harms (1968)
Kiparsky, P. (1971) "Historical Linguistics," in Dingwall (1971)
Kiparsky, P. (1972) "Explanation in Phonology," in Peters (1972)
'A
215
Kiparsky, P. (1973) "'Elsewhere' in Phonology," in Anderson, S. R. &
P. Kiparsky, A Festschrift for Morris Halle, Holt, Rinehart
& Winston, New York
Kiparsky, P. (1974) "On the Evaluation Mesure," in Bruck (ed.)
Kiparsky, P. (1975) " A Note on the Vowel Features," in New England
Linguistic Society Papers V: 162-7
Kurath, H. (1964) A Phonology and Prosody of Modern English, Carl Win-
ter, Universitaetsverlag, Heidelberg
Kurath, H. & R. McDavid (1961) Pronunciation of English in the Atlantic
States, U. of Michigan Press
Labov, W. (1966) The Social Stratification of English in New York City
Washington, D. C.: Center for Applied Linguistics
Labov, W. (1972) Sociolinguistic Patterns, U. of Penna Press, Philadelphia
Ladefoged, P. (1971) Preliminaries to Linguistic Phonetics, U. of Chicago
Press
Lass, R. (1971) "Boundaries as Obstruents," J. of Linguistics 7:15-30
Leben, W. (1973) Suprasegmental Phonology, M.I.T. Dissertation
Leben, W. (1974) "English Stress: Some Phonological and Morphological
Regularities," Stanford U. ms.
Lehiste, I. (1970) Suprasegmentals, M.I.T. Press, Cambridge, Mass.
Lightner, T. M. (1972) Problems in the Theory of Phonology, Vol. 1,
Linguistic Research Inc., Edmonton
A
216
Lisker, L. & A. Abramson (1964) "A Cross-language Study of Voicing in
Initial Stops: Acoustical Measurements," Word 20:384-422
Malecot, A. (1960) "Vowel Nasality as a Distinctive Feature in American
English," Language 36:222-9
Malmberg, B. (1955) "The Phonetic Basis for Syllable Division," in Le-
histe, I., ed., Readings in Acoustic Phonetics, M.I.T. Press
Cambridge, Mass.
McCawley, J. D. (1974) Review of Chomsky & Halle (1968), in IJAL 40:50-88
O'Connor, J. D. & J. L. M. Trim (1953) "Vowel, Consonant, and Syllable,
a Phonological Definition," Word 9:103-22
Oehrle, R. (1972) "The Stress Pattern of B.isyllabic Words in English,"
M.I.T. ms.
Oswald, V. A. (1943) "'Voiced T' - A Misnomer," American Speech 18:20
O.E.D. - Oxford English Dictionary, Oxford U. Press, London, 1933
Peters, S. (1972) Goals of Linguistic Theory, Prentice-Hall, Englewood
Cliffs, N. J.
Peterson, G. & I. Lehiste (1960) "Duration of Syllabic Nuclei in English,"
JASA 32:693-703
Pike, K. L. (1947) Phonemics U. of Mich. Press, Ann Arbor
Pirsig, R. M. (1974) Zen and the Art of Motorcycle Maintenance, WM.
Morrow & Co., New York
Postal, P. (1968) Aspects of Phonological Theory, Harper & Row, New York
A
217
R. H. D. = Random House Dictionary of the American Language, 1966
Sapir, E. (1925) "Sound Patterns in Language," Language 1:37-51, also
in Joos (1957)
Schane, S. A. (1968) French Morphology and Phonology
Selkirk, E. 0. (1972) The Phrase Phonology of English and French, M.I.T.
dissertation
Sharf, D. J. (1960) "Distinctiveness of 'Voiced-T' Words," American
Speech 35:105-9
Shibatani, M. (1973) "The Role of Surface Phonetic Constraints in Gen-
erative Phonology," Language 49:87-106
Smalley, W. A. (1968) Manual of Articulatory Phonetics, Cushing-Malloy,
Ann Arbor
SPE = Chomsky &-Halle (1968)
Stampe, D. (1972) How I Spent My Summer Vacation, Ohio State d.
Stetson, R. H. (1928) Motor Phonetics, Archives neerlandaises de phon-
tique exprimentale III 1-216
Thomas, C. K. (1958) An Introduction to the Phonetics of American Eng-
lish, Ronald Press, New York
Trager, G. L. (1930) "The Pronunciation of "Short A" in American Standard
English," American Speech 5:396-400
Trager, G. L. & B. Bloch (1941) "The Syllabic Phonemes of English," in
Language 17:223-246
A
218
Trager, G.
Vennemann,
L. & H. L. Smith (1951) An Outline of English Structure,
Battenburg Press, Norman, Okla.
T. (1972) "On the Theory of Syllabic Phonology," Linguisti-
sche Berichte 18:1-8
Vennemann, T. (1972a) "Rule Inversion," Lingua 29.209
Vennemann,
Vennemann,
T. (1972b) "Phonological Concreteness in Natural Generative
Grammar," in Shuy and Bailey (eds.) Towards Tomorrow's
Linguistics
T. (1974) "Words and Syllables in Natural Generative Grammar,"
in Bruck (1974)
Vennemann, T. (1974a) "Restructuring," Lingua 33:137-156
Webster = Webster's Third New International Dictionary of the English
Language, G. & C. Merriam Co., Springfield, Mass., 1966
Zwicky, A.
Zwicky, A.
Zwicky, A.
Zwicky, A.
M. (1969) "Note on a Phonological Hierarchy in English," in
Stockwell, R. P. & R. K. S. Macaulay, eds., Linguistic Change
and Generative Theory, Indiana U. Press, 1972.
M. (1970) "Auxiliary Reduction in English," Linguistic In-
quiry 1:3:323-36
(1970) "Greek-Letter Variables and the Sanskrit ruki Class,"
Linguistic Inquiry 1:4:549-54
M, (1974) "The Strategy of Generative Phonology," in Dressler
& Mares (eds.) Phonologica 1972 Wilhelm Fink, MUnchen, 1974.
'A