XXIX. LINGUISTICS"
Academic and Research Staff
R. Jakobson
M. Halle
S. Bromberger
J. A. Fodor
J. J. Katz
R. P. V. Kiparsky
E. S. Klima
G. H. Matthews
Krystyna Pomorska
A. Akmajian
S. R. Anderson
J. S. Bowers
M. K. Brame
E. W. Browne III
R. J. Carter
S. W-C. Chan
P. G. Chapin
P. W. Culicover
Janet P. Dean
R. P. G. De Rijk
Prof.
Prof.
Prof.
Prof.
Prof.
Prof.
Prof.
Prof.
Prof.
Dr. N. V. Smith
Dr. D. E. Walker
Dr. Anna Wierzbicka
Dr. Kay R. M. Williamson
C-J. N. Bailey
N. R. Cattell
B. E. Gaines
J-C. Milner
J. J. Viertel
Prof. J. R. Ross
Prof. J. H. Sledd
Dr. J. V. Canfield
Dr. Elinor K. Charney
Dr. J. B. Fraser
Dr. M. F. Garrett
Dr. J. S. Gruber
Dr. A. Schwartz
Graduate Students
R. C. Dougherty
J. E. Emonds
J. L. Fidelholtz
R. Goldfield
L. N. Gross
J. W. Harris
T. R. Hofmann
I. J. Howard
R. S. Jackendoff
R. S. Kayne
J. P. Kimball
Carol A. S. Kiparsky
RESEARCH OBJECTIVES
There have been two main traditions in the 
study of language in modern times. The
first is the tradition of "universal" or "philosophical 
grammar," which flourished in the
seventeenth and eighteenth centuries in 
intimate connection with philosophy and 
specu-
lative psychology. The second is the 
tradition of modern linguistics, a 
nineteenth and
twentieth century phenomenon that was 
also closely interwoven with the philosophy, 
psy-
chology, and anthropology of its 
day. Philosophical grammar was 
concerned with
general, universal principles of language 
structure; it attempted to ground these 
prin-
ciples in a theory of mental processes, 
and to illustrate them with detailed study 
of par-
ticular languages. By 
modern standards, the work 
lacked care and attention 
to detail,
and the conclusions that were reached, though 
often highly insightful, were deficient in
empirical support and sharpness of formulation. 
In comparison, modern nineteenth and
twentieth century linguistics has achieved 
a much higher standard of rigor, and has 
accu-
mulated linguistic data 
of an incomparably greater 
scope and variety. It has 
been limited,
however, by a much narrower interpretation 
of the purposes and goals of linguistic
science. It has eschewed theory 
construction in favor of elaboration 
of methods of
analysis, and it has not been concerned 
with linguistic universals - often, in fact, 
it has
denied that there are, in any significant 
sense, genuine and deep universal 
principles
that constrain the form and 
use of human language.
This work was supported 
principally by the U. S. 
Air Force (Electronics 
Systems
Division) under Contract AF 19(628)-2487; 
and in part by the Joint Services 
Elec-
tronics Programs (U. S. Army, 
U. S. Navy, and U. S. Air Force) 
under Contract
DA 36-039-AMC-03200(E), the National 
Science Foundation (Grant GK-835), 
the National
Institutes of Health (Grant 2 P01 
MH-04737-06), and the National 
Aeronautics and Space
Administration (Grant NsG-496).
QPR No. 84
R. Kirk
J. R. Lackner
Amy E. Myers
A. J. Naro
D. M. Perlmutter
J. T. Ritter
M. S. Snow
R. J. Stanley
R. J. Thiersch
F. J. Vandamme
Nancy Woo
261
(XXIX. LINGUISTICS)
The work in linguistics at the Massachusetts Institute of Technology represents, in a
sense, a synthesis of these two major traditions. In terms of its general goals and even
many of its specific hypotheses, this work has a very classical flavor. But in the range
and reliability of evidence and precision of formulation, this work accepts and attempts
to surpass the standards of modern structuralism.
For classical linguistics, a central property of human language is what we can call
its "creative" aspect, that is, its unboundedness and freedom from stimulus control.
Under ordinary circumstances, what a person says is not determined by the stimuli that
impinge on him or by identifiable physiological states, to any significant degree. The
unboundedness of normal language is evident from the fact that almost every linguistic
utterance produced and understood is quite new, not similar in any physically defined
sense to those that have been produced in the past experience of the language user, and
not conforming to familiar or memorized patterns, in any meaningful sense of the notion
"pattern." Nor are these utterances "generalizations" from past experience, in any
sense of "generalization" known to psychology or philosophy. Nor can language use be
described in terms of "habits" or "repertoires of responses." In recognizing these facts,
philosophical grammar was entirely correct and to the point.
To account for this creative aspect of normal language use, we must attribute to the
language user knowledge of a certain organized system of rules that establish a sound-
meaning relation for an infinite class of sentences. This knowledge is, of course, quite
unconscious, but it is nonetheless perfectly real. Thus it is quite likely that no one
reading this report has ever seen, heard, or produced the sentence
(1) What disturbed John was being disregarded by everyone.
Yet every reader will understand that the sentence may be roughly paraphrased by either
(2) or (3):
(2) Everyone was disregarding the thing that disturbed John.
(3) The fact that everyone was disregarding him disturbed John.
Thus sentence (1) is ambiguous, its possible interpretations being (2) or (3). If the word
"our" is inserted in (1), giving (4), the sentence is unambiguous.
(4) What disturbed John was our being disregarded by everyone.
The interpretation of (4) can only be along the lines of (3), with "him" replaced by "us."
Or, to choose an example from a totally different sphere of language, speakers of English
would know that the plural of the word dap is daps, whereas that of linch is linches (with
es rather than s), in spite of the fact that most of the speakers would neither know the
meanings of these words nor have heard them before.
A speaker of English has knowledge of these facts and numerous others without having
been exposed to these sentences or to any explicit "teaching." He has mastered a system
of rules that determine both the phonetic form of sentences (1)-(4) and their various
semantic interpretations. The first task of the linguist who is investigating the structure
of English is to try to determine this system of rules, the system that is called the "gen-
erative grammar of English." This generative grammar has in some manner been inter-
nalized by every speaker of English; it determines the pairing of sound and meaning for
an indefinitely large range of possible sentences. It is this internalized generative gram-
mar that makes possible the normal, "creative" use of language.
The discovery of the generative grammar of English, and other languages, is, how-
ever, only the first task that faces the linguist. To the extent that such grammars have
been developed and validated, the linguist can then turn to the question of how they are
put to use, by the speaker or hearer, in normal conversation, in literature, in internal
monologue, and so on. Furthermore, he can turn to the basic problem of classical lin-
guistics: What are the universal principles that limit the form of such generative gram-
mars ? Clearly, there must be universal principles with a very narrow and limiting
character. If this were not true, it would be impossible for the child, presented with
QPR No. 84
262
(XXIX. LINGUISTICS)
scattered samples of a language for an extremely short period, to determine for himself
the generative grammar of this language. But this is a task that normal humans accom-
plish with great facility. This indicates that they must approach the task forearmed with
highly specific advance knowledge (obviously, unconscious) of the possible form that a
generative grammar must assume. To put it loosely, although the child cannot "know"
in advance whether the language 
to which he is exposed is English, 
Chinese, and so on,
he must "know" that it is a "human language" of a highly special sort, which can only
vary in very restricted ways. The problem of "universal grammar," now, as 
in the
seventeenth century, is to determine the principles that limit the variety of human lan-
guage and make possible the acquisition of language. To the extent that such principles
can be formulated and validated, we gain insight of an unparalleled kind into the innately
determined character of human mental processes.
We feel that recent work, much of it carried out at M. I. T. , makes it possible to
formulate a fairly precise theory 
of universal grammar in this sense, 
a theory which is,
furthermore, reasonably well supported by substantial empirical evidence from a variety
of languages. The major goal of our research, then, is to sharpen and deepen 
the theory
of generative grammar, and to use it as a basis for the study of cognitive processes.
Since many of the problems of language lie in the area in which several 
disciplines
overlap, an adequate and exhaustive treatment of language demands close 
cooperation
of linguistics with other sciences. The inquiry into the structural 
principles of human
language suggests a comparison 
of these principles with those 
of other sign systems,
which, in turn, leads naturally to the elaboration of a general theory of signs, semiotics.
Here linguistics touches upon problems that have been studied by philosophy. 
Other
problems of interest to logicians - and also to mathematicians - are touched upon in the
studies devoted to the formal features of a general theory of language. The study of
language in its poetic function brings linguistics into contact with the theory and history
of literature. The social function of language cannot be properly illuminated without
the help of anthropologists and sociologists. The problems that are common to lin-
guistics and the theory of communication, the psychology of language, the acoustics and
physiology of speech, and the study of language disturbances are too well known to need
further comment here. The exploration of these interdisciplinary problems, 
a major
objective of this group, will be of benefit not only to linguistics; it is certain to pro-
vide workers in the other fields with stimulating insight and new methods of attack, as
well as to suggest to them new problems for investigation and fruitful reformulations of
questions that have been asked for a long time.
M. Halle, N. A. Chomsky
A. INITIAL CLUSTERS IN ENGLISH
In examining the range of possible consonant clusters at the beginning of a word in
English, we find not merely a lot of redundancy (as is well known), but a particular 
kind
of redundancy, a type that allows us to re-analyze these clusters as single segments.
This leads us to suspect that the structure of words may be much simpler than has yet
been suggested. Chomsky and Hallel have shown how and why the complex vowel nuclei
of syllables must be derived from simple vocalic segments and in this report we extend
this same concept, finding that many complex consonant clusters can be derived 
also
from single segments.
1. Preliminary
In English, all words begin with the following sequence: (C = a consonant, 
V = a vowel;
QPR No. 84 263
(XXIX. LINGUISTICS)
parentheses enclose optional elements and brackets 
alternate possibilities. The posi-
tions are numbered for 
ease of reference.)
(1) 
1
(s) (C)( 
) V...
12 3
Not all possibilities of the formula 
above are realized, however -
there are principled
restrictions on what may occur 
with what. For example, 1 never 
occurs with a pre-
ceding dental stop - any word 
that began *dl... would simply not 
be an English word.
Various methods have been 
proposed to schematize the 
possibilities of occurrence
and non-occurrence, ranging 
from (earliest) a simple 
list of possible initial clusters 
to
(later) what we may call "phoneme 
order charts" as exemplified 
by Harris
Z 
to (recently)
morpheme structure rules 
which effectively forbid certain 
sequences of elements.
Another method of expressing these 
restrictions is proposed below.
It is well to notice first that the y 
following a consonant is itself always 
followed by
[u] as in beauty, cute and is unlike the 
other position 3 possibilities in that 
it does occur
after voiced non-obstruents as in view, 
mute, lieu, new. (In some dialects 
this y has
dropped after dentals.) For these 
reasons, then, we may consider 
it to be part of the
[u] which necessarily follows it, 
treating the combination as a single 
vowel in the under-
1
lying representation. This treatment 
may be found in Chomsky and Halle, 
and will not
be duplicated here.
2. Position 3 Elements
We may now revise the original 
statement (1) of possibilities of word-beginnings 
to
(such statements we shall call 'canonical 
forms'):
(2)(s) (C) ( ) V ...
12 3
If we ignore the s in position 1 for 
the moment, we see that in position 
3 there are
4 possibilities; either nothing at all 
(which I shall call 0), r, 1, or w. 
A choice between
4 possibilities involves 2 bits of 
information. The maximally simple 
description would
be obtained if we could use but 
2 distinctive features to specify 
which possibility is found
in a particular word.
Arranging the possibilities as in
(3) r 1
QPR No. 84
264
(XXIX. LINGUISTICS)
we see that in the horizontal dimension, r and w oppose 1 in flatness.3 And in the ver-
tical dimension, there is a distinction in both consonantality and vocalicity between r and
1 on the one hand and w on the other. Let us use the feature vocalicity for the vertical
dimension. We then have (vertical bars enclose a conjunction of features, vocl =
vocalicity, and flat = flatness):
(4) +vocl +vocl w -vocl
+flat ' -flat ' +flat
and the last combination, -flat , is realized as$.
If we can specify that a consonant occupies position 2 by marking J+cons
I 
, then the
feature vocl, along with flat, is not used in specifying anything about that segment. It
is proposed, then, that rather than use a separate segment (position 3) to hold the 2 fea-
tures needed to specify how that segment is to be actualized, we now place these features
in the second segment and thus simplify the canonical form of words in English. A seg-
mentalization rule will be introduced to spell the features flat and vocl out of a
consonantal segment into a position immediately following it. The canonical form for
English is thus
(5) (s) (C) V ...
1 2
This segmentalization rule is placed, of course, after the stress assignment rule and
simplifies it by allowing a definition of a strong cluster as "2 or more +cons segments"
instead of "any cluster that is not a weak cluster, where a weak cluster is C( )".
The problem in stress assignment where 11 must be a strong cluster (e. g. , coste lo) in
spite of its natural inclusion in the formulation of a weak cluster is now nonexistent, as
each 1 well have to be a separate segment. We are assuming that the segmentalization
rules also account forthose medial and final clusters that are the same as permitted
initial ones. As will be seen in section 5, there is a principled restriction forbidding
initial clusters composed of any combination of r, 1, or w. In particular, there can
be no 11 initial. Hence, if 11 occurs medially, it must be 2 segments: one associated
with the preceding vowel, and the other an initial, associated with the following vowel.
3. Remarks
There are several points that should be remarked upon here. First, notice that only
by using an arrangement such as (3) can we specify both how position 3 is realized (as
r, 1, or w) and whether or not there is a position 3 type segment by using only 2 fea-
tures. Any other system must use 2 features 
to contrast the 3 possibilities r, 1,
and w and one additional feature to specify the existence of nonexistence of the
QPR No. 84 265
(XXIX. LINGUISTICS)
position 3 segment. That is, to specify whether a segment after a consonant and before
a vowel is a position 3 item or simply a vowel or a consonant, we would need one addi-
tional feature. For example, we must specify a segment to be I+cons in addition to
being +flat in order that r is output rather than u or o, whereas with the present
+vocl 
+flat- 
-
system the features +vocl are located in the position 2 segment and the extra feature
needed in the earlier system is inserted by the segmentalization rule. Of course, using
3 features to specify what 2 can specify entails some redundancy that will have to be
expressed by rule.
6
Another point is that a maximally efficient set of distinctive features can be hoped for
only if each feature divides roughly into halves the classes defined by other features.
Heretofore, this has been conspicuously lacking between the features cons and vocl, the
classes defined by one being almost duplicated by the other. 
With this change, however,
where we previously had one consonant, +cons+-vclns ,we now have, in general, 4 consonantal
+cons segments,
(6) k kr kl kw
cons + + + +
vocl - + + 
-
flat - + - +
and the class defined by I+cons I is roughly divided in half by vocl. Both of these sim-
plifications are facets of the same inner simplification.
It should also be remarked that using the feature vocl in the way we do in some sense
imputes a different definition to it. Of course, we so use 
it only on a very deep level,
where features are little more than abstract markers. If we want to put the requirement
of phonetic realism on the underlying forms, we shall either have to find another feature
to play the role of vocl in this analysis or alternatively re-define vocl so that tr can be
both vocalic and consonantal. I believe that the second alternative would be more pro-
ductive, as we would then not need an "erasure" rule to mark position 2 segments
I-vocll (after the vocl has been segmentalized, of course). This is 
not unreasonable,
as we cannot have an erasure rule to mark position 2 segments 1-flat I because a posi-
tion 2 consonant is always rounded when followed by an r. Flatness also involves itself
in the difference between the t and the ch series, which accounts in part for s becoming
sh when followed by r and for the similarity in acoustic impression between trip and
chip, etc. On the other hand, we shall need an erasure rule to mark position 2 segments
I-strid , as stridency is missing in the nasal in smile or snide.
4. Leading s
Returning now to the optional s at the beginning of a word, we may notice that it con-
trasts only with its absence. This is just 1 bit of information or 1 distinctive feature -
all else is redundant. We need only 1 unused feature of its following consonant to use in
QPR No. 84 266
(XXIX. LINGUISTICS)
specifying its presence. There are several possibilities, 
but let us assume that stri-
dency (strid) is available on this deep level, i. e. , that f and s do not 
contrast with
anything by virtue of their values of stridency. (We shall soon return to 
s and 0.)
Then we may say that if a consonant is strident, an s is created before it 
by
the s-segmentalization rule. To develop s alone as an initial, 
the full form of s is
entered in the dictionary. The s-segmentalization rule will add an s before 
it and gem-
inate simplification will reduce the resultant ss to s.
The s-segmentalization rule is formalized 
as7
(7) +cons
+strid -vocl
-voicl z +strid + 1
+naslj -voic
+cont
and is followed by the stridency-erasure rule
(8)nt l ----- strid J/( +strid )
D. Perlmutter has pointed out that, by such an arrangement, the effect of a
morpheme structure rule is obtained at no extra cost. If a strident b (b) occurs in the
S
dictionary, rule (7) does 
not apply and rule (8) 
erases the stridency. 
In contrast, if ps
(a strident p) occurs, rule (7) applies to form sps , and rule (8) 
applies only to the ps by
virtue of the disjunctive ordering (roughly, if a parenthesized 
term in a rule environment
can be matched by some part of the material to which the rule is to be applied, 
then the
rule cannot be applied without so matching).
Consider the case in which positions 1, 2, 3 are all filled. While we previously
needed only 4+cons I in first segment to specify it as s, we now need 1 feature, strid on
both s and non-s-beginning words. This means that in the case of the leading s, our
present system will require more feature mentions in the dictionary, although the use
of markedness can make the 2 systems equivalent again. I am inclined to believe that
we now have the proper result. A word with a leading s carries no more and no less
information than one without s. If the number of binary distinctions needed to identify
a word is any function of the information content (or "surprise value") of that word, the
s-less and s-ful words must be specified with the same number of features.
5. Details
To meet the facts of English, we must disallow the following combinations.
(9) position 
2 position 
3
labial w
dental 1
voiced any (i. e., r, 1, or w)
continuant
QPR No. 84 267
(XXIX. LINGUISTICS)
By effecting this restriction before the segmentalization rule applies, we can handle this
with segment-structure rules that are equivalent to the morpheme structure rules that
were necessary for handling these restrictions before. The only difference between the
2 systems is that now we need not assign values for the features vocl and flat in posi-
tion 2, as we had to before. The following rules will suffice, but are not the only ones
that will do so. The chosen ones are interesting in that they approximate the acoustic
impressions experienced by a native when hearing forbidden clusters:
(10) +cons
+grav
+diff ---- -vocll
+flat
(11) +cons
+vocl
-flat - +grav
-cont
and one that does not have this property
(12) +cons 
-flat
I
+cont 
-vocl
+voic
Actually, something fundamental is happening here that is hard to capture without
introducing the concept of markedness, which is beyond the scope of this introductory
sketch. After labials, a w (a labial) is forbidden; after dentals, an 1 (a dental) is for-
bidden; and after voiced continuents, any position 3 item (all voiced continuents) is for-
bidden. Although this can be expressed with markedness, what really should be said
is that w (1) is not segmentalized out of a labial (dental) because such segmentalization
would be pure redundancy. All of the essential characteristics of the w (1) are already
expressed by the presence of the labial (dental). Such considerations are valid for the
languages with which I am familiar, but these are at best a small sample. Let us pose
this as a supposition, and if it holds water for a large number of languages, we can build
it into the theory of segmentalization. Note that the testing of this supposition rests on
deeper analyses of languages than are now generally available, and although languages
like Spanish do have pw... , the w does not result from segmentalization but derives
from an underlying vowel.
8
But recalling how we used the s-segmentalization rule with its associated erasure
rule to effect a restriction, we may ask if it is possible 
to play such a trick here,
thereby saving the cost of these segment-structure rules. It is easy to do for rule (12)
but more difficult for rules (10) and (11). The "3"-segmentalization rule will be the
following.
QPR No. 84
268
(XXIX. LINGUISTICS)
(13) +cons
-nasl a cons
fa vocl
yfric 
1 + pflat
Svoic
a vocl
p flat
The -nasll forbids position 3 creation with a nasal and the y specifications do the same
for voiced fricatives. All will have to be marked -vol ' which is motivation for a
-vocl
vocl erasure rule if this tack is taken. To disallow w after labials, we shall need as
additional features (14a), and to disallow 1 after dentals, we shall need (14b) added to
the left side of rule (13).
(14) +diff +diff
(a) +grav (b) -grav
+vocl -flat
Addition (14a) will disallow w and y from being segmentalized out of labials, and the
y will be erased anyhow to give the plain labial. Addition (14b) will disallow 1 and y
from a dental. The derivation for the plain consonants has been supposed to originate
in the underlying forms as -flat , which was spelled out as y, after which the y was
erased. We see, however, that for the labials and dentals it is easier to block the w
and 1 by also blocking
9 
the y, hence the simple labials and dentals are not marked for
flat and vocl, respectively. This is a verification of our original arrangement (3) of the
position 3 elements. There is no clear method to combine (14a) and (14b) with
(13). We can obviate the need for this by using a rule of the following form before
the "3"-segmentalization rule to cause it not to apply to any segments to which this rule
is applicable.
(15) +diff
a vocl
-a flat a- [-next 
rule]
a grav
Let us consider how 1, r, w are handled when they do not follow a consonant. The
+cons
+vocl
underlying form of 1 is -flat where is the set of features necessary to make 1
distinct from such consonants as kl, t, n, . The segmentalization rule applies giving
+cons +vocl
+vocl -flat where 
is the redundant 
set of features 
inserted in position 
3 by the seg-
-flat
+cons
mentalization rule. If the sets and I are the same, then the segmentalization
rule will change a single 1 into 11 and the geminate-simplification rule (which is exter-
nally motivated) will then change the 11 into 
1. An identical process is undergone by 
r,
except that all 1-flatl specifications are replaced by j+flat . If for both r and 1
QPR No. 84
269
(XXIX. LINGUISTICS)
(16) -gray
+diff
= -nasl
+cont
-fric
then the complete form of the "3" -segmentalization rule is
(17) a vocl
p flat
+segment flat
a cons
+cons
a vocl 
= 1 + -agrav
+diff
p flat -nasl
-nasl 
-nasl
+cont
-fric
Notice that in the more traditional analysis employing morpheme structure rules
+consl +cons and
there was a large number of consonants, -vocl but only 2 liquids, +vocl and as
a result the 2 liquids r and 1 could be specified by only 3 features apiece, while most
other consonants required more. In this analysis, the number of liquids is comparable
to the number of consonants (kl, kr, etc. are now liquids on the more abstract levels)
and the 2 "simple liquids," r and 1, must be specified to an extent that is comparable
to the specification of a simple stop.
-cons
The simple glide w, -vocl , is unaffected by the "3" -segmentalization rule because
+flat
this rule affects only I+cons I segments. It may seem surprising, at first, that the treat-
ment required of simple w by this solution is not parallel to the treatment of simple r
and 1, but this nonparallelness is also found in the data. A (position 3) w after a con-
sonant is never followed by U' ([u] or [u], usually spelled oo). If a w is in position 2
(i. e. , not following a consonant), such a vowel following 
is possible, e.g. , woo, woozy,
wood, would, wool, woof. To explain these data within the framework of 
mor-
pheme structure rules is considerably more difficult than rule (18) placed before
"3" -s egmentalization.
(18) +cons
-vocl -flatj U'
This rule will also enable us to explain why there is no w in two [tu] in contrast to the
w in twin, twelve, twain, between, etc. (courtesy of D. Perlmutter). Further, notice
that we have gained without cost the effect of a morpheme structure rule which forbids
all initial combinations of r, 1, and w.
The treatment of s will solve 2 problems that are lurking in the background of the
segment-structure rules mentioned above. First, s being a dental, it ought not to occur
with 1 yet it does (and is the only exception to the rule). Hence we shall enter sl in the
QPR No. 84 270
(XXIX. LINGUISTICS)
dictionary as a strident 1. The 1 is duplicated by the "3" -segmentalization rule, 
and
the leading s is inserted by the s-segmentalization rule. With geminate simplification,
this becomes sl.
The second problem was that although w does not occur in environments, CU', swU'
does, i. e. , there is swoon, swoop. Thus we write it as a strident 
w, which as we have
found, will not be affected by the "3"-segmentalization rule or by rule (18). In this case,
it is necessary to consider the s to be in the 1st position and the main consonant 
to be
w rather than an expected 2
n d 
position s consonant with a w in the 3
r d 
position
-cons +cons
-vocl -vocl
+flat +flat
( i. e. , that it is a strident 
w, +strid , not a rounded 
s, +strid
Usual formulations of English phonology employ stridency to distinguish s from 6.
This option is left open to us as long as an sO cluster is forbidden as it clearly 
is
(sthenic and sthanakvasi undoubtedly being foreign). We see that leading s occurs with
consonants p, t, k in spot, stop, and scot; with nasals in smart, snot; and with 
f as
in sphere. We have strident stops, nasals, and f, and it would seem unlikely if we did
not also have a strident 0 parallel to the f. But indeed we do, and we call it s. It is
doubled by the s-segmentalization rule and then re-singled by the geminate-simplification
rule. Thus the simple strident consonants may be charted:
(19) 
sp st sk
sm sn
sf s
In English, there is no initial sr except in Srinigar, which is clearly foreign, and
there are no palatals with 1, w, or r, except that shr.. is common (shwa, 
shluh,
schlimazel, schlemiel are foreign). These facts have prompted many to suggest that
sr obligatorily becomes shr. Of course, the more traditional phonology cannot accept
this unless Srinigar is represented with a word boundary, s#ringer, or is an exception
to the rule that converts s to sh before r. The present system captures the distinction
-vocl
nicely, shrink beginning with an s that is +flat and Srinigar beginning with an s
followed by an r, i. e. , a true cluster, which is forbidden initially in English.
6. Problem
Referring back to (14) and (15), we saw that to block w and 1 after labials and den-
tials, respectively, the position 3 y, which would have been deleted anyhow, was also
blocked. But now notice that the rule needed to remove this y will be useful only with
velars (i. e., there is no ky,
1 0 
gy). It is very tempting at this point to map these velars
+ y into the ch series (c j s), thereby explaining their presence in English and why they
are never accompanied by a position 3 element, and also removing the need of the
QPR No. 84
271
(XXIX. LINGUISTICS)
y-deletion rule. Unfortunately, if we do this, there is no source for plain k. 
If we could
make up an ad hoc feature, +3 , which would specify 
whether or not a position 3 is to
be created, k and ch could contrast on their being 1-31 and 1+3 i. 
respectively. It may
well be that this is cheaper (it depends on the relative weighting 
of feature mentions in
the dictionary and feature mentions in phonological rules); we shall 
not argue it here.
It is significant, however, that an analogous situation is found in Mandarin 
Chinese pho-
nology, that for the sake of 2 types of syllable finals, a feature 
must be introduced which
serves only to specify whether or not the final has a 
position 4 element (y, w, n, or ng).
We may ask if this is a general property of segmentalization systems, but for the 
pres-
ent we shall have to be content with hypothesizing that such an extra feature 
is sometimes
necessary to capture the full generality of the phonemic system. 10
7. Comments
A few comments are in order concerning the interpretation of these processes of
segmentalization. First, it is observed that the posited underlying segments are nigh
on impossible to pronounce; for example, a strident nasal or liquid or a 
rounded
"liquidic" (i. e. , I+vocl ) t, etc. It such underlying forms existed, there would need to
be spelling-out rules to de-focus the various components of such 
a segment into time-
wise distinct segments so that all would be pronounceable. It seems that in some 
lan-
guages (e. g. , English), there is a strong tendency to move some 
of the features on a
segment in one direction and others in another, until what was originally one 
segment
may appear to be several distinct segments in the phonetic output. In other languages
(e. g. , Kabardian 11), most of the features on a segment are realized in the output more
or less concurrently. (It is doubtful that in any language, for any segment, any 2 fea-
tures begin and end simultaneously.) But even in nonsegmentalizing languages, many
of the features on a consonant are spread out onto surrounding segments in the 
phonetic
output. For example, in East Indian languages, the retroflection 
of the retroflex series
of consonants appears recognizably only on the vowels before and after 
the consonants.
It is clear that the only reason that these languages are not analyzed as having 
pseudo
clusters like the English initial clusters is that the phonetic realization 
of the segmen-
talized features appears on both sides of the basic segment, and thus 
there is no way to
decide on which side to place them. Contrariwise, treating the pseudo clusters 
found
at the beginning of an English word as genuine clusters, e. g. , chm in Lechmere, 
is a
hang-over from the theory of classificatory phonemics.
The major difference between the use of segmentalization techniques and 
morpheme
structure restrictions is that with the latter the simpler case is the one in which 
there
is greater freedom of occurrence between elements of a cluster, 
whereas with the for-
mer strong restrictions not only are expected but required.
This segmentalization technique applies only to clusters in which the combinatorial
QPR No. 84 272
(XXIX. LINGUISTICS)
possibilities are severely restricted. Note that in the analysis above, 
the features
needed to naturally specify positions 1 and 3 were unused in the specification of the 
2
nd
segment. Requiring such a condition seems to place a natural limitation 
on the pos-
sibilities of this segmentalization technique. It is clear that without 
creating a lot
of ad hoc features, the initial consonant cluster and its following vowel could not be
spelled out of a single segment, since at least the features grave, diffuse, 
and vocalic
would necessarily have one value for the initial cluster and one value for the 
vowel, which
values would not in general be identical.
Even more natural than having the features on one segment unused in the 
other seg-
ment (as above) is the situation in which a feature that is necessary on 
one of the seg-
ments predicts the occurrence of itself or similar feature(s) on another segment 
within
the cluster. An example of this can be found in Mandarin Chinese, which I shall 
not dis-
cuss now.
This solution for initial clusters is as simple as any that has been previously 
enter-
tained. It is simpler, in fact, in terms of the number of features used in the 
rules.
Thus if this type of solution is to be rejected, the present phonological theory will have
to be modified in some non ad hoc manner, in order to exclude it automatically.
Last, consider a mutated form of English, where r, 1i, and w in position 3 are
replaced by k, t, and p. This language would be more complex, yet the list and the
phoneme-order methods of description would not show it. Another 
mutation, where r,
1i, and w in position 3 are supplemented by k, t, and p, would be simpler in terms of
morpheme structure rules (since less restrictions would have to be stated). In both of
these cases, it is only when we find that the clusters thus formed cannot be reduced to
single segments that we understand the reason why these mutations result in a more
complex language.
A forthcoming paper by J. E. Emonds, to whom I am indebted for some points brought
out in this report, will give a more general description of consonant clusters in English,
although the author disagrees with me on some points and does not go as far in some
directions as I have in the present report.
T. R. Hofmann
Footnotes and References
1. N. A. Chomsky and M. Halle, The Sound Pattern of English (to be published by
Harper and Row).
2. Z. S. Harris, Methods of Structural Linguistics (University of Chicago Press,
Chicago, Ill., 1951), p. 153.
3. This is realized as lip-rouding, or tongue retroflection, or both. Actually, we could
simply use rounding for th'is contrast, since the prevocalic r, as well as w, is
always rounded. Post vocalic r, with which we are not concerned here, is a
beast of an other nature, and is never rounded.
QPR No. 84
273
(XXIX. LINGUISTICS)
4. M. Halle (in class lectures) has treated kw as a single segment (thereby allowing
a weak cluster "formula," C({})), but was unwilling to treat tw and dw similarly.
Our direction here requires us to go on and accept not 
only tw but also kr, kl, tr,
pl, and pr as single segments. Recently G. Bedell
5 
has proposed such an analysis
for the 'final' part of the syllable in Mandarin Chinese. 
The only thing new here,
as with Bedell's analysis, is the use to which this technique is put, that of creating
canonical positions. But even this is not novel, we may view the canonical position
of the vowel off-glides in English as being created and filled by the influence of the
preceding vowel.
5. G. D. Bedell IV, "Syllable Finals in Chinese Phonology, Quarterly Progress Report
No. 81, Research Laboratory of Electronics, M. I. T. , April 15, 1966, pp. 186-190.
6. Although one part of this redundancy is perhaps universal, the impossibility of
-cons
-flat in position 3 will have to be mentioned, which will have the effect of, and
-vocl
will cost the same as, the rule in our new system which deletes such a segment.
7. We shall assume that f is I-stridI at this level of abstraction, and that it is stri-
dentized later.
8. J. A. Foley, Jr. , "Spanish Morphology," Ph. D. Thesis, Department of Modern
Languages and Linguistics, M. I. T. , June 1965 (unpublished).
9. Here, to block means "to cause not to appear," rather than the more usual "to mark
as deviant."
10. Notice the similarity in saying that k is -31 and that k is an exception to the
"3" -segmentalization rule.
11. A. Kuipers, Phoneme and Morpheme in Kabardian (Mouton and 
Company, The Hague,
1960).
QPR No. 84
274
(XXIX. LINGUISTICS)
B. ESQUISSE A PROPOS D'UNE CLASSE LIMITEE D'ADJECTIFS
EN FRANgAIS MODERNE
Ii existe en Frangais moderne une variation souvent observe, qui transforme le sens
de l'adjectif, suivant qu'il suit ou prdcede le nom auquel il se rapporte:
(1) un simple soldat / un soldat simple
une simple robe / une robe simple
(2) un ancien roi / un roi ancien
une ancienne armoire (=un meuble qui a 6t6 une armoire, et sert a
present a d'autres usages) / une armoire ancienne
(3) une 6pop6e vraie / une vraie 6pop6e
(4) une vague id6e (=quelque chose qui m6rite a peine le nom d'idde) / une idee vague
(=une idde impr6cise)
(5) une apparente folie / une folie apparente
D'ordinaire, on rambne ce phenomone a celui, plus g6ndral, de la place de l'adjcctif
6pithbte en Frangais, et de sa variation. Ainsi les oppositions mentionn6es sont
rapprochdes de l'opposition entre homme grand et grand homme, qui change le sens, ou
entre verts paturages et paturages verts,courageux soldat et soldat courageux etc., qui
change la nuance. C'est meme dans ce dernier type de variation qu'est trouv6e la clef
de l'ensemble: avant le nom, verts, dira-t-on, n'est pas un adjectif descriptif, mais
d6signe un attribut permanent du nom; au contraire, aprbs le nom, verts est descriptif
et d6signe un attribut accidentel. Ainsi, dira-t-on, il existe une opposition g6ndrale
gouvernant la place et le sens de l'adjectif, entre par exemple le propre (apr6s le nom)
et le figur6 (avant le nom), comme on peut le voir, semble-t-il, dans les couples (I) a
(5) ou dans le couple homme grand / grand homme.
2
Notre hypothbse est au contraire que la s6rie des variations (I) a (5) est a s6parer
des autres ph6nomhnes apparemment analogues (y compris l'alternance pr6sent6e par
grand), et que les adjectifs qu'elle affecte forment une classe limit6e, reconnaissable
a certains caractbres.
Par exemple, ils n'admettent pas les degr6s de comparaison: on ne dit pas
(6) *une moins simple simple robe, une robe moins simple
*une plus simple robe, une robe plus simple
(7) *un moins ancien professeur, un professeur moins ancien
*un plus ancien professeur, un professeur plus ancien
du moins au sens de une simple robe, un ancien professeur.
Il est ill6gitime de les coordonner avec des adjectifs ordinaires, ce qui, suivant les
r6gles de la coordination, suffit a prouver la diff6rence des statuts. On ne dit pas *une
QPR No. 84 275
(XXIX. LINGUISTICS)
simple et petite porte, "un vieil et ancien 
professeur.
Mais de facon plus radicale encore, il n'est 
que de rechercher leur origine profonde
pour manifester une singularit6 essentielle 
de ces adjectifs: on ne peut les employer 
en
attributs aprbs la copule. Une robe est simple 
ne peut avoir que le sens de une robe
simple, un professeur est ancien, le sens de un professeur 
ancien. Cela suffit '
interdire de d6river ces adjectifs de relatives 
pr6dicatives: (8) ne peut venir de (9).
(8) je porte une simple robe
(9) je porte une robe
[une robe est simple]
En meme temps qu'il les distingue, ce caractbre dbfinit le veritable problbme syntaxique
que posent ces adjectifs: s'il est impossible 
de les d6river des relatives, quelle est leur
origine profonde ?
On peut remarquer ici que l'impossibilit6 de pr6diquer 
est l'aspect le plus important
d'un ph6nomone plus g6n6ral, l'impossibilit6 de s6parer 
les 616ments du syntagme, par
exemple en supprimant l'16ment nominal dans une anaphore. Ainsi (10) est possible,
(11) ne l'est pas.
(10) Quel crayon voulez-vous ? Le petit.
(11) *Quelle robe voulez-vous ? La simple.
De meme (10a) C'est le petit que je veux.
(Ila) *C'est la simple que je veux.
Tout se passe donc comme si le groupe forme par l'adjectif et le nom 4tait inseparable,
et provenait en bloc de tours pr6dicatifs tels 
que
(12) Louis est un ancien 
professeur
3
(13) Ceci est une simple robe.
S'il en est bien ainsi, ce seront ces groupes unitaires 
que l'analyse devra expliquer a
partir de la pr6dication qui les constitue, et d'abord 
il convient d'en examiner la sig-
nification exacte.
En effet, l'on peut admettre sans difficult6 que l'6nonc6 Ceci 
est une belle robe
exprime que l'objet appartient a l'ensemble des robes, au sous-ensemble 
des belles
robes, une belle robe 6tant une robe caract6ris6e par sa beaut6. En revanche, 
par
l'6nonc6 Ceci est une simple robe, on exprime que l'objet appartient 
a l'ensemble des
robes, mais il n'y a pas de sous-ensemble des simples 
robes; une simple robe en effet
n'est pas une robe caractdrise par sa simplicit6, mais 
une robe tout simplement. Ce
qui se trouve ainsi d6fini, ce n'est pas l'appartenance 
a un sous-ensemble, mais une
modalit6 de l'appartenance a l'ensemble.
4
Or il est remarquable que tout naturellement, 
pour paraphraser l'6nonc6 a
QPR No. 84
276
(XXIX. LINGUISTICS)
interpr6ter, nous soyons parvenu plus haut ' un tour adverbial: une simple robe est une
robe tout simplement. Il ne s'agit pas 1l d'un hasard; il apparait en effet que l'on peut
relier les adjectifs que nous traitons a des adverbes synonymes:
(14) a. Jean 6tait anciennement un professeur
b. Ceci est une robe tout simplement
c. Ceci est apparemment une folie
d. Jean est faussement un pretre
e. Jean est vraiment un h6ros
(15) a. Jean est un ancien professeur
b. Ceci est une simple robe
c. Ceci est une apparente folie
d. Jean est un faux pretre
e. Jean est un vrai h6ros
Bien que les phrases (14) soient inigalement acceptables, il est clair que toutes peuvent
recevoir une interpr6tation, dont relbvera tout aussi bien la phrase (15) correspondante.
La relation predicative donc parait modifi6e (modalisde) de manibre identique par
l'adjectif et l'adverbe.
Sans les analyser pour eux-memes, nous pouvons cependant reconnaitre a ces
adverbes certaines particularit6s qui les opposent aux adverbes de manilre avec lesquels
on les confond ais6ment.
1. Il est impossible de les coordonner avec ces derniers. On peut par exemple dire:
(16) Il conduit rapidement et bien
mais non pas
(17) *Ii conduit vraiment (apparemment) et bien
2. De fagon plus r6v6latrice encore, il est impossible d'employer ces adverbes pour
r6pondre a une interrogation Comment?, De quelle manibre?.
(18) De quelle mani6re conduit-il?
a. Prudemment
b. Rapidement
c. *Apparemment
d. *Anciennement
e. *Vraiment
En revanche, (18) c, d, e r6pondent a la question Conduit-il?.
3. Pas plus que leurs correspondants adjectifs, ils n'admettent les degr6s de comparaison:
QPR No. 84
277
(XXIX. LINGUISTICS)
(19) *Jean est plus (moins) vraiment (apparemment) 
malade que Louis
4. Du point de vue de leur origine profonde, 
le sens meme interdit que ces adverbes
proviennent d'un tour Pr6position 
+ Nom. En effet, dans les phrases 
que nous
consid6rons, simplement ne peut signifier 
'avec simplicit6', ni anciennement 
'avec
anciennet', ni vraiment 'avec v6rit6', 
ni apparemment 'avec apparence', 
La seule
paraphrase possible serait ici compos6e de l'adjectif qualifiant un nom tel que mani6re,
fagon etc., soit de manibre simple, 
ancienne, vraie, apparente, fausse, 
rdelle etc.
5
Seule correcte pour le sens, cette paraphrase 
fournit aussi la base de l'analyse
syntaxique. II suffit de reconnaitre 
qu'en v6rite, le nom n'est dou6 ici d'aucune
autonomie s6mantique: tel la copule dans 
l'nonc6 pr6dicatif, il n'est la que pour
effectuer la compl6tude et l'autonomie syntaxique 
du syntagme, qui rdduit au seul
adjectif ne pourrait en Fran9 ais obtenir 
de statut plein.
6 
C'est en fait le repr6sentant
superficiel de ce qui doit 'tre dans la 
structure profonds un "dummy element", un
l16ment non sp6cifiable, sinon par sa cat6gorie 
et sa fonction.
Nous parvenons dbs lors a la repr6sentation suivante:
MOD 
NP Adjectif 
[+blDqueD]]
oii D repr6sente un N non 
sp6cifi6,
MOD d6signe la cat6gorie Modalitd, que nous introduisons 
dans la grammaire afin
de nommer la structure sous-jacente aux 
adverbes dont nous avons reconnu la
sp6cificit6.
La figuration [+Oblique] tient compte du fait que, vraisemblablement, 
de et a ne sont
pas en Frangais des prdpositions ordinaires et jouent 
plutot comme des traits confer6s
a un '16ment nominal par sa place dans le P. 
M. , c'est-a-dire assign6s ' certaines
fonctions (donec analogues 
a des cas).7
Devant la synonymie des phrases (14) et (15), 
nous avangons l'hypothbse que
l'adjectif dans les phrases (15) est un adjectif modal, d6riv6 du syntagme de modalit6,
sous-jacent a l'adverbe dans les phrases (14). En 
rigueur, l'adverbe en -ment 6tant
lui-meme d~riv6 du syntagme de modalit6, le processus 
d6rivant l'adjectif ne passera
pas par l'adverbe lui-meme, mais deplacera 
directement l'adjectif (issu d'une relative
r6duite) hors de la structure de modalit6, et l'adjoindra 
par une transformation que
nous ne tenterons pas ici de mieux d6finir au NP Attribut de la phrase predicative. Ainsi,
adjectif 6pithbte ordinaire dans le syntagme [NpAdjectif D], celui-ci 
devient adjectif
7b
modal (issu d'un tour adverbial) 
dans le syntagme attributif d6riv. 
7b
Nous n'avons pas ' d6cider ici de la place hi6rarchique 
de la Modalit6; l'hypothbse
la moins couteuse, sinon la plus vraisemblable, est de la placer au 
meme niveau que VP.
Nous admettrons donc que pour d6river Jean est un ancien 
professeur, nous devrons
QPR No. 84
278
(XXIX. LINGUISTICS)
partir de la description suivante:
S
NP Predic
VP
Copule Attribut
N NP
Jean est 
un prol
ate - Phrase
Moda I ite'
Fesseur
NP
Adjectif N
I I
ancien D
E + Oblique]
Diagram XXIX-1.
La d6rivation proc6dera comme suit:
1. Adjonction de l'Adjectif domino par la categorie Modalit6, au 
NP domin6 par la
cat6gorie Attribut.
2. Plutt qu'une d6ltion particulibre, le "dummy element" D suivra une rbgle g6ndrale
conf6rant a tous les "dummy" non explicitement remplac6s une interpr6tation 
phon6tique
zdro. D6s lors on pourra conserver le noeud Modalit6 dans le PM g6n6ralis6, 
et l'on
parlera de phrases modales, caractdrisees par sa pr6sence. D'ohi
NP Predicate - Phrase
VP Moda I ite'
Copule Attribut
NP
Adjectif NP 
D
Jean est ancien un professeur H
Diagram XXIX-2.
La transformation se d6crit ainsi: (# denote l'Adjonction)
+ X [PredP [VP 
Copule NP] Y 
[MOD [NP Adj 
D]] Z +
1 2
1 2
3 4 5
36#4 5
OPT
6 7 89
0 7 89
QPR No. 84 279
(XXIX. LINGUISTICS)
Remarques: 1. Si la transformation sur la Modalit6 n'est pas appliqu6e en ce sens
prend place la transformation adverbiale: [simple D] devient simplement. Entre l'une
et l'autre transformations, il y a choix optionnel, mais il est obligatoire de transformer.
On peut alors parler d'un couple obligatoire de transformations optionnelles sur la
Modalit6.
2. La transformation adjective est cyclique (d'oti la pr6sence des + marquant les
"phrase boundaries"). Temoin les phrases
(19) Jean est un simple ancien professeur
(20) Jean est un parfait simple soldat
3. La place de l'adjectif modal est fixe: il prc6dde tous les 616ments du NP auquel
il est adjoint, non pas 6videmment l'article, toujours en tete (une rbgle sp6ciale y
pourvoira), mais tout adjectif non-modal, ou tout adjectif modal issu d'une d6rivation
antdrieure. Ainsi s'expliquent les apparentes infractions aux rigles d'ordre rythmiques
(pourtant si contraignantes), suivant lesquelles un adjectif polysyllabe suit un nom
monosyllabe (cf. pour cette rbgle Gr6visse, Le bon Usage, ? 398). On dira donc Un
ancien roi, un parfait sot, un v6ritable don. Si le nom est d6j* pr6c6d6 
d'un adjectif,
on dira par exemple: Une simple petite porte, Un faux grand homme et non *Une petite
simple porte, *Un grand faux homme. Ii faut reconnaitre du reste que la rencontre de
deux adjectifs devant le nom est d'ordinaire 6vit6e grace au tour adverbial; plutot que
Un vrai bel appartement, on dira Un appartement vraiment beau etc.
Ainsi se trouve d6velopp6e l'origine des tours pr6dicatifs du type Ceci est une simple
robe, et de meme que l'adjectif 6pithhte ordinaire, ce sera de ces tours que l'adjectif
modal sera d6riv6 dans ses emplois non-pr6dicatifs
(21) a. Je porte une simple robe
b. une simple robe suffira
(22) a. Je vois un ancien professeur
b. Un ancien professeur arrive
Comme pour l'adjectif ordinaire, il faut partir d'une relative, celle-ci ne qualifiant pas
un nom sp6cifi6 cependant, mais ici encore un N 
non specifi6, analogue a quelqu'un,
quelque chose. En effet nous avons vu, c'6tait notre point de d6part, que les groupes
simple robe, ancien professeur 6taient ins6parables, ils devront donc provenir ensemble
de la relative et dans la principale ne subsiste plus qu'un z6ro phondtique, repr6sentant
un "dummy" sous -jacent.
QPR No. 84
280
(XXIX. LINGUISTICS)
On obtient alors la description suivante:
NP VP
NP S
N QUI S
NP VP
Copule Attribut
NP
Adjectif NP
D ( =Ce) qui D est une simple robe suffira
Diagram XXIX-3.
Ii est ais6 de voir que la transformation 
appropri6e sera ici la r6duction des relatives,
dont la description structurale sera obtenue a partir de celle valable pour l'Anglais, soit
X [NP NP [S QUI Copule Y]] Z OPT
1 2 3 4 5
1 2 0 4 5
Cependant, pour des raisons en partie ind6pendantes de notre pr6sent objet, la pr6sence
de D doit etre sp6cifi6e dans la description structurale, et entraine certaines
modifications. Il convient en effet de pr6ciser que la r6duction n'est pas possible si
l'ant6c6dent du relatif est cet 616ment non-sp6cifi6 D, et l'attribut un adjectif: 
Tout ce
qui est bon ne se r6duit pas a "Tout bon.
En revanche, si l'attribut est un nom pourvu ou non d'un adjectif modal 
ou non, la
r6duction est possible: Tout ce qui est une 
belle robe se r6duit a Toute belle robe,
Toute ce qui est une simple robe a Toute simple robe, Tout ce qui 
est une robe ' Toute
robe.
Une version particulibre de la r6duction des relatives apparait ainsi necessaire:
QPR No. 84
231
(XXIX. LINGUISTICS)
X [NP D [S QUI Copule [ATT [NP 
Y 
NP]]]] Z OPT
1 2 3 4 5 6
1 2 0 4 5 6
Dans cette formulation, la sp6cification 
de l'ancienne variable 4 est lide 
a la
sp6cification de l'616ment 2. Il est 
clair que ce n'est pas la une rbgle ad hoc, 
formul6e
pour rendre compte seulement des tours modaux 
non-pr6dicatifs, mais une regle
concernant l'ensemble des rdductions 
de relatives attach6es a un 616ment non-sp6cifi6
nominal: en fait une regle mineure (minor 
rule). Il est vrai aussi cependant que 
cette
formulation pr6cisbe permet seule d'obtenir 
la d6rivation que nous souhitons. Grace
a elle, nous pouvons d6river de D [D 
est une simple robe] suffira, Une simple 
robe
suffira.
D recevra une interpretation phonetique z6ro, 
mais l'on peut admettre comme 6nonc
6
d6rive D l'ancien professeur, D une simple 
robe.
8 
Une simple robe, l'ancien
professeur se trouvent 
ainsi engendr6s comme 
des caractdrisations 
appos6es d'un
l66ment non-sp6cifi6, et de ce fait meme g6n6rique, ce qui 
correspond bien au
sentiment intuitif que ces 6nonc6s comportent une 
r6f6rence vague a un ensemble plus
vaste; on entendra par exemple (En fait de v tements) une 
simple robe-...; on peut
observer du reste que le tour est bien plus fr6quent 
avec un article ind6fini.
Au terme de cette double d6rivation qui 
nous a procur6 d'abord les phrases modales
pr6dicatives, puis par une rdduction particulibre, les non-prddicatives, 
il apparait que
les alternances dont nous 6tions parti sont illusoires: ce n'est pas le meme adjectif qui,
changeant de place, change de sens, mais il y a deux 
adjectifs de forme identique et
d'origine diff6rente, dont l'un, l'adjectif modal, a une place 
fixe, devant le nom, et
en exclut l'adjectif non-modal homonyme.
Cependant cette homonymie n'est pas un fait de hasard, 
puisque l'adjectif modal est
bien issu de l'adjectif non-modal, par l'intermddiaire 
de l'adverbe de modalit6, soit
le processus
Adjectif non-modal - Adverbe de modalit6 -
Adjectif modal
Simple
(1)  
> Simplement > Simple
( 2 )
Du meme coup s'expliquent et la parent6 et la difference 
des sens entre les deux
adjectifs homonymes: parent6 du fait d'une commune origine 
lexicale, difference dans
les processus de d6rivation.
On peut affirmer sans grave erreur que les alternances ainsi 
rduites formaient
l'616ment principal des variations de place affectant le 
sens de l'adjectif; si l'on
QPR No. 84
282
(XXIX. LINGUISTICS)
d6montre d'autre part que parmi les alternances non-r6duites, 
les plus apparentes sont
6galement illusoires, pour d'autres raisons, on aura 6cart6 
l'ensemble des variations
de place entrainant d'importantes polarisations s6mantiques. Or bien 6videmment 
une
grammaire transformationnelle pouvait difficilement traiter ces ph6nom6nes, tels qu'ils
apparaissaient ' premiere vue; deux solutions se presentaient ici (elles ont 6t6
effectivement proposdes par G. Lakoff et J. R. Ross): ou bien attribuer 
le d6placement
a une rigle mineure (minor rule), mais alors une transformation se trouvait affecter 
le
sens,10 ou bien conserver deux entrees lexicales, ce qui ne rendait pas compte de la
parent6 des sens. C'est donc simplifier et g6nAraliser la grammaire du Frangais que
d'y admettre la nouvelle transformation et la nouvelle classe deriv6e des adjectifs
modaux.
Il faut prendre garde que par la ne se trouve nullement r6solu le probl6me g6ndral
de la place de l'6pithbte. Bien au contraire, alors que les grammaires 
traditionnelles
apercevaient dans les variations semantiques polarisdes le ph6nombne le 
plus
caract6ristique et la clef g6n'rale du systhme adjectif, notre hypothbse 
entend les mettre
a part et ne pr6juger en rien de ce que doit etre l'analyse des adjectifs 
ordinaires. Tout
au plus, dissipant une apparence dont celle-ci prenait jusqu'a pr6sent 
son d6part, peut
elle pr6tendre en former une 6tape pr6liminaire.
J. C. Milner
Notes
1. De ce fait, des ph6nombnes diff6rents se trouvent m6lang6s dans les listes donndes
par les grammaires et les descriptions (cf. e. g. , Brunot, La Pens6e et la 
Langue,
pp. 638-643, Damourette et Pichon, Des Mots a la Pensde, II, par. 512-516,
Grevisse, Le bon Usage, par. 398). Les releves les plus complets de ces variations
apparaissent chez les grammairiens du XVIIIe sibcle, qui les consid6raient comme
essentielles ' la pr6cision et l'exactitude du style (Des r6f6rences sont donndes par
Damourette et Pichon, ibid.; y ajouter pour une expression caract6ristique du point
de vue g6n6ral Rivarol, Prospectus d'un nouveau dictionnaire de la langue frangaise.).
2. Si les exemples (1) a (5) relbvent tous de la classe que nous entendons d6crire, ils
n'en 6puisent pas l'extension. D'autres adjectifs apparaitront occasionnellement
dans la suite, mais nous ne tenterons pas ici de dresser une liste complEte.
De plus, dans l'examen du comportement particulier des adjectifs concernds,
la bribvet6 nous impose de faire valoir un ou deux cas (en g6n6ral, simple et ancien)
pour l'ensemble. Nous ne pr6tendons pas que certains ne pr6sentent pas parfois
certaines idiosyncrasies, mais nous les n6gligerons ici.
3. Notons encore un trait particulier des adjectifs consid6r6s: de meme que l'on peut
avoir, sans article, Jean est soldat, Jean est professeur (noms de fonction), on aura
Jean est simple soldat, simple professeur. En revanche, un adjectif ordinaire
entraine automatiquement Particle: les 6nonc6s *Jean est pauvre soldat, Jean est
vieux professeur sont exclus.
4. La meme analyse pourrait 'tre faite de Jean est un ancien professeur, Ceci est une
apparente folie, mais la relation n'est pas simple, un ancien professeur n'tant plus
professeur,une apparente folie n'6tant pas une folie. Cependant 
ici encore l'adjectif ne
modifie pas le nom en le caract6risant, mais la relation elle-m@me 
en la modalisant.
QPR No. 84 283
(XXIX. LINGUISTICS)
On peut ainsi comprendre que tout adjectif probatif (excellent, parfait, 
mauvais,
vrai, faux, r6el etc.) peut entrer 
dans le systhme des adjectifs consid6r6s,
puisqu'ils sont predisposes 
' exprimer que les conditions 
d6terminant
l'appartenance a une classe (definie en intension, 
non en extension, soit un concept
freg6en) sont ou non remplies.
5. On pourrait objecter que 
tous les adverbes de maniore 
en -ment admettent
6galement cette interpretation. 
Mais il est clair que, pour 
le sens, il est
indifferent de deriver par exemple 
rapidement de avec rapidit6 ou de 
de manibre
rapide. Le trait discriminant n'est pas 
que l'analyse proposee soit possible, mais
qu'elle soit dans certains cas pr6cis 
la seule possible.
6. De meme que dans bien des 
langues, en certains cas, la copule 
peut etre
repr6sent6e par un zero phondtique, 
de m@me le "dummy" auquel se rapporte
l'adjectif dans les tours sous-jacents 
a l'adverbe. Comparer e.g., certains
adverbes de manibre en latin: crebro, cito, 
falso.
Le fait que le frangais, s'opposant 
au latin de la meme manibre dans 
les deux
cas, soit oblig6 d'employer un mot de statut 
grammatical plein et de s6mantisme
vide pour soutenir l'adjectif, qu'il s'agisse de la predication ou du tour adverbial,
pourrait manifester une correlation 
typologique importante.
7. L'hypothse a et' avanc6e et 
solidement fond6e par E. Benveniste, 
notre
presentation n'en est qu'une interpr6tation 
plus formelle.
Le trait [+Oblique] recevra l'interpr6tation 
phon6tique de ou a suivant les con-
textes syntaxiques (ici de). On peut meme penser que par, introduisant 1'agent,
est la r66criture de ce meme trait dans 
le contexte de la transformation passive.
7b. Cette adjonction n'est pas sans 
rapport avec la rbgle reportant 
la n6gation du
verbe subordonn6 au verbe recteur dans une phrase telle 
que:
I didn't believe that he would leave until tomorrow
Cela malgr6 des differences 6videntes: dans cette rbgle, il s'agit d'un trait [+Neg],
non d'un constituant; de plus le deplacement de l'616ment 
se fait du syntagme
domind au syntagme dominant, non pas 
entre syntagmes de meme niveau
hidrarchique.
8. Il est possible que ce soit l1 l'explication 
de la place de tout dans les 6nonc6s 
tels
que Toute la maison, tout le corps, si l'on 
admet qu'ils soient a d6river de Tout
ce qui est la maison, tout ce qui est le 
corps, c'est-a-dire Tout [D qui est] la
maison, Tout [D qui est] le corps.
9. Ainsi par exemple la c6l6bre alternance 
homme grand / grand homme. En effet
1. cette alternance ne vaut que pour le mot 
homme; partout ailleurs, grand a une
place fixe, avant le 
nom,
2. malgr6 l'apparence grand conserve le meme sens 
dans tous ses emplois:
la diff6rence entre les 6nonc6s un grand tapis / un 
grand pianiste rdside seulement
dans la classe semantique des 616ments nominaux. 
Neanmoins certains traits
lexicaux du nom peuvent affecter l'adjectif; ainsi 
grand sera ou non adjectif de
mesure suivant que le nom sera ou non marqu6 
du trait lexical [+Mesure]. On
posera donc une rbgle d'assimilation, 
comparable a une rigle phonologique:
Grand . [a Mesure] /- N [a Mesure]
3. s'il y a une apparente variation du sens 
de grand entre homme grand et grand
homme, elle ne peut etre due qu'a une dualit6 
du nom, et en effet le mot homme a
deux s6ries d'emplois: dans l'une, il est syntaxiquement 
analogue aux noms de
fonction; on peut alors le pr6diquer sans article 
(Je suis homme, comme Je suis
pianiste, fonctionnaire, professeur etc.) et 
il faut l'affecter du trait [-Mesure].
Dans ce cas, grand est d sa place regulibre 
et n'est pas adjectif de mesure.
QPR No. 84
284
(XXIX. LINGUISTICS)
Dans l'autre emploi, homme est un terme 
"biologique", comparable aux noms
d'animaux, et il est affect6 du trait [+Mesure] 
(cf. un homme de 1, 80 m), le con-
f6rant du meme coup a grand. C'est l'homonymie 
avec le premier tour, ph6nomhne
purement idiosyncratique, qui exclut 
alors grand de la place precedant le nom.
A titre de contre-6preuve, on observera 
que femme, n'ayant que le sens
"biologique", ne prdsente pas la m'me 
alternance; non seulement Une femme grande
aurait le mime sens que Une grande femme, mais 
le premier -nonc serait bien
infirieur au second, et a peine acceptable 
(Noter l'opposition nominale grande 
femme
[+Mesure] / grande dame [-Mesure]).
10. Ajoutons que les variations de sens joueraient alors de manibre d6concertante; 
si
l'on admet en effet, comme il est vraisemblable, que la place rdgulibre de l'adjectif
est devant le nom et la place d6riv6e, aprbs le nom, comment expliquer 
que le sens
"propre" apparaisse ' la place d6riv6e, et le sens figur6 (qu'on imaginerait plutot
issu de la transformation) a la place r6guli6re ?
11. Si toutes les variations s6mantiques se trouvent r6duites, seules subsistent des
variations de nuance (cf. e.g., verts paturages / paturages verts,courageux
soldats / soldats courageux). II est possible, dans ces conditions, 
que le
deplacement de l'adjectif 6pithbte ordinaire rel6ve ' nouveau d'une 
rogle mineure
(minor rule), et que les variations de nuance soient attribu6es ' des ph6nom6nes
de discours (essentiellement d'intonation); peut-etre faut-il aussi tenir compte 
de
la difference entre relatives restrictives et non-restrictives.
Note ajoutde a la relecture: Nous avons pris connaissance 
trop tard pour en faire
6tat ici de l'article d'E. 
Bach "Nouns and Noun 
Phrases in English" (non 
publi6),
oti pp. 12 et 13, des remarques et des hypotheses analogues 
aux n^tres se trouvaient
deja enoncees.
Nous ne discuterons pas les suggestions de Bach et 
remarquerons seulement
la difference qui s'tablit n6cessairement 
entre la grammaire de l'Anglais et celle
du Frangais; dans les deux langues, la sp6cificit6 
des adjectifs modaux est claire
et s'atteste par le meme 
trait: l'absence de predication 
(cf. pour l'Anglais, former,
alleged, real, mere cit6s par Bach.); mais le 
probl6me qui se pose alors en Anglais
est seulement de rendre compte de la 
d6rivation syntaxique de cette classe. En
Frangais, a une hypoth6se de cet ordre, s'en ajoute 
encore une autre touchant la
place de ces adjectifs; nous avons avanc6 ainsi 
deux hypotheses distinctes et, il
faut le remarquer, ind6pendantes:
1] les adjectifs modaux proviennent de tours adverbiaux
2] ils ont une place fixe
Ii n'y a aucune relation d'implication entre ces 
deux hypoth6ses: la premiere
seule pourrait pr6tendre etre 6tendue ' l'Anglais 
et ainsi s'opposed ' celle de Bach;
la seconde est propre au Frangais et, en tout 
6tat de cause, n'est nullement affect6e
par le sort de la premiere.
QPR No. 84
285