The Formal Analysis of Natural Languages: Proceedings of the First International Conference, Paris, April 27–29, 1970 [Reprint 2018 ed.] 9783110885248, 9789027924674


185 80 31MB

English Pages 396 [400] Year 1973

Report DMCA / Copyright

DOWNLOAD PDF FILE

Table of contents :
Preface
Table Of Contents
Part One: Phonology
A New Organon. The Formalism Of Phonological Rules
Remarks On Universals In Phonology
On The Role Of Notation In Generative Phonology
The Formalization Of Exceptions In Phonology
Part Two: English Syntax
Evidence That Indirect Object Movement Is A Structure-Preserving Rule
Focus, Presupposition, And Deep Structure
A Global Constraint On Pronominalization
Slifting
Part Three: Syntax Of Other Languages
Sur Les Traits Distinctifs [± Défini], [± Spécifique] Des Groupes Nominaux: Contribution Typologique
The Turkish Copula
Restrictions De Sélection, Transformations, Et Règles De Redondance: Les Constructions Pronominales En Français
Part Four : Semantics
The Semantics Of Giving
On The Relation Between Disjunction And Existential Quantification
Sur La Typologie Des Langues Naturelles: Essai D'interprétation Psycho-Linguistique
Part Five: General Linguistics And Psycholinguistics
Remarques Sur La Méthodologie De La Grammaire Générative Transformationnelle
Morphology In Generative Grammar
Modèle Transformationnel Et Linguistique Historique
Some Remarks On Psycholinguistic Studies
Part Six: Formal Properties Of Languages
A Class Of Transformational Grammars
Généralisation de la notion d'équivalence de grammaires: une méthode topologique
On restricting deletion transformations
Index
Recommend Papers

The Formal Analysis of Natural Languages: Proceedings of the First International Conference, Paris, April 27–29, 1970 [Reprint 2018 ed.]
 9783110885248, 9789027924674

  • 0 0 0
  • Like this paper and download? You can publish your own PDF file online for free in a few minutes! Sign Up
File loading please wait...
Citation preview

JANUA LINGUARUM STUDIA M E M O R I A E N I C O L A I VAN WIJK D E D I C A T A edertda curai C. H. V A N S C H O O N E V E L D Indiana University

Series Maior,

62

THE FORMAL ANALYSIS OF NATURAL LANGUAGES PROCEEDINGS OF THE FIRST INTERNATIONAL CONFERENCE

edited by

MAURICE GROSS, University of Paris-Vincennes MORRIS HALLE, Massachusetts Institute of Technology MARCEL-PAUL SCHt)TZENBERGER, University of Paris VII and I.R.I.A.

1973

MOUTON THE H A G U E - P A R I S

© Copyright 1973 in The Netherlands. Mouton & Co. N.V., Publishers, The Hague. No part of this book may be translated or reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publishers.

LIBRARY OF CONGRESS CATALOG CARD NUMBER 72-91486

Printed in the Netherlands by N.V. drukkerij D. Reidel, Dordrecht

PREFACE

The first conference on the Formal Analysis of Natural Languages was organized at the Institut de Recherches en Informatique et Automatique (78 — Rocquencourt, France) on April 27-29, 1970, in order to bring computer scientists into contact with linguists concerned with the formalization of natural languages. In bringing these groups together it was the hope of the organizers that each would benefit from the contact. Although computers have as yet been used only sparingly in linguistic research they clearly are promising tools for research, for they make it possible to manipulate and organize large bodies of data that are much beyond the capabilities of researchers unaided by machines. For computer scientists the contact with linguistics has already born fruit. The mathematical study of context-free languages has found quite natural applications to the design and improvement of programming languages. Specialists in information retrieval have until now been perhaps less affected by the advances in syntax; these appear to have, however, quite direct application to information retrieval. While many of the papers presented at the conference1 will no doubt inspire computational applications, the majority of authors regarded their contributions as theoretical in nature. This volume, therefore, reflects the main current concerns with the theory of language. As might be expected not all areas are equally far advanced. Thus, phonology appears to have reached the point where it is possible to focus fruitfully on the abstract character of the formalism and to use it as a tool for further discoveries (see Part one). By contrast, syntactic studies seem much closer to pure observation of data. Here the formulation of the rules depends much more heavily than in phonology on the nature of the particular phenomena under study. (See Parts two, three, five). Formalized semantics as represented in Part four is a new field, still very much in flux. To all subfields of linguistics the mathematical or purely formal study of grammars is of cardinal importance. Recent developments are described in Part six. While the studies contained in these proceedings do not provide exhaustive coverage of the whole field, they are representative of the present status of research on language. 1

At the Conference papers were delivered also by R. S. Kayne, and E. S. Klima. These authors did not submit their manuscript for publication.

VI

PREFACE

They show where we are at and also provide some hints about the direction in which the various subfields are moving. We are indebted to Professor Michel Laudet, whose interest and efforts made this conference possible, and to Miles Th. Bricheteau, E. Vuillaume, and Mr. A. Delacou, who handled the practical arrangements for the conference with remarkable efficiency and kindness.

TABLE OF CONTENTS

Preface

v Part one: Phonology

Morris Halle, A new organon. The formalism of phonological rules Theodore M. Lightner, Remarks on universals in phonology James D.McCawley, On the role of notation in generative phonology . . . . Sanford A. Schane, The formalization of exceptions in phonology

1 13 51 63

Part two : English syntax Joseph Emonds, Evidence that indirect object movement is a structure-preserving rule 73 Terence Moore, Focus, presupposition, and deep structure 88 Paul M. Postal, A global constraint on pronominalization 100 John Robert Ross, Slifting 133 Part three: Syntax of other languages Wayles Browne, Sur les traits distinctifs [± défini], [± spécifique] des groupes nominaux: contribution typologique Robert B.Lees, The Turkish copula Nicolas Ruwet, Restrictions de sélection, transformations, et règles de redondance: les constructions pronominales en français

173 175 180

Part four : Semantics 205 R. M. W. Dixon, The semantics of giving Christian Rohrer, On the relation between disjunction and existential quantification 224 RenéThom, Sur la typologie des langues naturelles: essai d'interprétation psycho-linguistique 233

VIII

TABLE OF CONTENTS

Part five: General linguistics and psycholinguistics Maurice Gross, Remarques sur la méthodologie de la grammaire générative transformationnelle 251 Ferenc Kiefer, Morphology in generative grammar 265 J. Stéfanini, Modèle transformationnel et linguistique historique 281 Jacques Mehler, Some remarks on psycholinguistic studies 296 Part six: Formal properties of languages Aravind Joshi, A class of transformational grammars 329 S.-Y. Kuroda, Généralisation de la notion d'équivalence de grammaires: une méthode topologique 362 P. Stanly Peters, Jr., On restricting deletion transformations 372 Index

385

PART ONE

PHONOLOGY

MORRIS HALLE

A NEW ORGANON. THE FORMALISM OF PHONOLOGICAL RULES*

The advances that have been made in the field of linguistics during the past fifteen years have deepened our understanding of the nature of individual languages and of the nature of language. We know today a great deal more than we did fifteen years ago, not only about what it is that someone must know in order to have fluent command of English, Navaho, Japanese, or Ewe, but also about the general features of linguistic knowledge, that is, about features that knowledge of English has in common with knowledge of Ewe, Navaho or Japanese. That a better understanding of individual languages should bring with it also a better understanding of the nature of Language with a capital L is hardly surprising. The converse, however, is not quite as obvious. It is not quite self-evident that the discovery of new facts about individual languages is predicated upon particular advances in our understanding of the nature of language. Nevertheless, we may cite many instances of discoveries that have been made regarding various subtle points in the syntax and phonology of individual languages that were directly due to particular advances in our understanding of language, in general. It was only because we had learned that language, in general, exhibits a given regularity that it made sense to look for a peculiar and highly abstruse reflex of this regularity in some particular language. It is worthwhile to study in some detail one such example, namely, the regularities of stress assignment in English, because such a study is bound to teach us something not only about the methods of science, but also about one aspect of the highly abstract skill or competence that is manifested by an English speaker every time he produces an utterance. What makes our example, the placement of stress in English, especially interesting is that it shows that all of us know things about our language that we are not only quite unaware of, but that we even lack the means to express coherently. It should be noted that stress location in words differs greatly from language to language. In Finnish it falls on the first syllable of the word; in Polish it falls on the one but last syllable, whereas in French the last syllable of the word is stressed. When a speaker of any of these languages learns English, mastering the position of stress in the word is a matter of some difficulty. It is obvious that native speakers of English must also have expended some effort on learning how to stress English words, for * This work was supported in part by the National Institutes of Mental Health (Grant MH-13390-04).

2

MORRIS HALLE

knowing where to place the stress in the word is part and parcel of what we mean when we say that a person has command of English. One of the most surprising discoveries that Chomsky and I made in the course of our studies of the sound pattern of English was that the stress in a large class of English words was totally predictable from the phonetic form of the word. That is, given the sequence of consonants and vowels that compose the word, the location of the stress in the word can be determined automatically. This discovery was so surprising because textbooks had for many years specifically denied this and, moreover, because the rule that we discovered resembled not that of any Germanic language, but rather was all but identical with the stress rule of Classical Latin. In order to see how the rule operates we observe that the vowels of English fall into two disjoint classes, tense and nontense (lax). Rather than explain the phonetic difference between these two sets of vowels, we give examples of the distinction below: (1)

(a) (b)

TENSE:

pile, peel, pale, pole, pool, foul pill, fell, pulse, pull

NONTENSE:

The first thing that is to be observed about stress in English is that its location is more readily stated when the end, rather than the beginning of the word is taken as the point of departure. If viewed from this point of view, stress can go on any one of the last three syllables (cf. (2)); on the other hand, if the origin is placed at the beginning of the word there are many more positions. (2)

antepenult

penult

last

America

Arizona

Wisconsin

Tennessee

aluminum

arboretum

carborundum

supreme

original

suicidal

orchestral

parole

origin

horizon

utensil

divine

antidisestablishmentarian

developmental

Tippecanoe

endocrinological

anthropomorphous

aluminium

epistemologica! When we examine the phonetic composition of the examples it is not hard to establish the principles of stress location :

A NEW ORGANON. THE FORMALISM OF PHONOLOGICAL RULES

(3)

3

(a) The stress is on the antepenultimate vowel if the last vowel is nontense, and the penultimate vowel is nontense and followed by at most one consonant. (b) The stress is on the penultimate vowel if the last vowel is nontense and the penultimate vowel is either (i) tense, or (ii) followed by two or more consonants. (c) The stress is on the last vowel if the last vowel is tense.

These rules are so wordy that it is evidently desirable to restate them with the help of a more perspicuous formalism (4): (4)

(a)

V-[1

stress] ¡[X

C0

C 0 [ " * n S e ] C„]

(b) (i) V - [1 stress]/ [X (ii) V - [1 stress] / [X (c)

^ n s e ] C j [ _ * n s e ] C0]

C2

C0]

F-.[lstress]/[z[^=]c0]

The string to the left of the slash (/) is to be read 'place primary stress ([1 stress]) on the vowel (represented by the letter V)'. The string to the right of the slash represents the salient features of the word as they play a role in determining the location of stress in accordance with (3). The pair of bold face square brackets [ ] indicate the beginning and end of the word. C 0 and C 2 stand for sequences of zero, respectively two, or more consonants, whereas C j represents a sequence of at most one consonant. The horizontal line indicates where the vowel to be stressed is located, and X represents a sequence of any number of vowels and/or consonants within a word. As shown in (5) below, the word aluminum fits the analysis of rule (4a). (5)

al [X

u

m

i

Co [ " 7

n e

] c j

u

m

[~7Se]c0]

Up to this point the formulas in (4) are nothing but straightforward translations into an algebraic symbolism of the statements in words given in (3) above. We shall now impose a number of formal constraints on these statements; in doing this we shall force these statements into a particular mold so that they will no longer be simple translations of the English statements (3). By forcing our statements into a particular mold we are implying that there is a preferred way for recording observations about language. Clearly such an implication is justified only to the extent that we can demonstrate that the mold chosen is appropriate to the subject matter, to language.

4

MORRIS HALLE

A good way to demonstrate that the mold really fits would be by showing that with the help of statements conforming to the proposed mold we can learn new and interesting facts about language. Let us assume that the statements in (4) are actually rules which take a sequence of letters, where each letter represents (as a first approximation) a sound, and indicate the location of the stress by placing a '1' above the appropriate vowel-letter. In other words, we picture stress assignment as a process where a sequence of letters like aluminum is analyzed in accordance with the various alternatives provided by the statements in (4), the applicable analysis (cf. [5]) found, and then the sequence is modified to aluminum which, incidentally, is regarded as distinct from the former. This rather mechanical way of looking at the stress rules is not just pedantry; it is an essential step in showing what is involved in knowing a language. When one reads over the rules as stated in (3)=(4), one notices readily that they are somewhat too prolix, too 'wordy'. This prolixity can be remedied by making use of the locution 'otherwise', i.e., we can replace (3) by the less prolix (6): (6)

(a) The stress is on the antepenultimate vowel if the last vowel is nontense and the penultimate vowel is nontense and followed by at most one consonant; (b) otherwise, the stress is on the penultimate vowel if the last vowel is nontense; (c) otherwise the stress is on the last vowel.

The less 'wordy' statement will be shown to have linguistically interesting consequences. We shall therefore digress briefly and ask what formal devices would be needed so that in the statement of our rules we may make use of the locution 'otherwise'. The formal reconstruction of this concept demands that we impose the following conditions on the rules: (7) Rules apply in a linear ORDER, i.e., we test each of the rules in (4) in order to find out whether or not it applies, first (4a), next (4bi), next (4bii) and finally (4c). (8)

For certain sets of rules — and, in particular, for the set in (4) — a special condition of DISJUNCTIVENESS holds.

Whereas in general more than one rule may apply to a given word — and we shall see below examples of this — no more than one of the rules in a disjunctive set may apply to a given word, and, moreover, the rule that applies is the one ordered earliest in the set. To see what this means, assume that the rules in (6) are disjunctively ordered. Clearly (6a), which is identical with (4a), applies to the word aluminum as shown in (5) above. Since (6) is a disjunctive set no further rules can apply to this word. This is important since as formulated above, the last vowel of the word aluminum is nontense so that (6b) would also be applicable, but this would lead to two stressed syllables in

A NEW ORGANON. THE FORMALISM OF PHONOLOGICAL RULES

5

the word, which is clearly an incorrect consequence. The reason that (6b) does not apply, in fact, is the word 'otherwise', which introduces it and is the way everyday language expresses the formal constraints of a disjunctively ordered set of rules. We spoke above of the needless prolixity of the rule (3)=(4) and noted that by use of the concept 'otherwise' or its formal equivalent 'disjunctively ordered set of rules' a less prolix statement can be obtained. To see the improvement we translate (6) into the symbolic notation developed above: (9)

(a) F - [ l stress]/[X

C0 [ "

(b) V -» [1 stress]/[X

C0

(c) V

C0]

[1 stress]/[X

c

l ["

C„]

* n S e ] C0]

This translation brings out a very interesting formal property, namely, in (9) we can obtain the later rules by deleting portions of the earlier rules. Thus we can obtain y

C\

y

I, and we obtain

"-tense! ^ . This property of (j J later rules being — as it were — contained within earliier rules is a unique feature of disjunctively ordered rules. Only and all sets of disjunctively ordered rules possess it. To bring out this property more clearly we shall abbreviate (9) by enclosing the substrings to be deleted in parentheses as shown in (10): (9b) from (9a) by deleting in the latter the substring

(10)

V

[I stress]/[X

(Coif"^]

c

o)

C

°]

We shall make the assumption that (10) is the form in which the stress rule appears in a true grammar of English. We achieve this by establishing yet a further condition: (11)

The prolixity of a rule or set of rules is measured by the number of symbols (C, V, -tense, etc.) appearing in the statement of the rule. All other things being equal, less prolix formulations are always to be chosen over more prolix formulations. We must now demonstrate that these formal conditions are indeed of some value to us in deepening our understanding of the nature of language, in general, and of English, in particular. Observe first that rule (3b)=(4b) does not tell us how to stress bisyllabic words such as those in (12) where the first syllable ends with a nontense vowel followed by no more than one consonant: (12)

venom

mucus

Paris medal

missile villa

Rule (6b)=(9b), on the other hand, does handle these words properly; for it supplies stress to the penultimate syllable of any word to which (6a)=(9a) does not apply and

6

MORRIS HALLE

which, moreover, has a last vowel that is nontense. Notice that there is no necessity that this be the case; English bi-syllabic words of this type might very well have been stressed on the last syllable. Nor were bi-syllabic words of this type taken into consideration when the stress rules were formulated above (cf. (2)). In other words, what has happened here is that when we tried to describe the facts for one set of words we automatically also described the facts for a totally different set of words. Extrapolations of this sort are of great importance, for they testify to the correctness of the proposed formal constraints which provided for these extrapolations in the first place. Two further sets of facts can be extrapolated from (9) which were not implied by (3). First, monosyllabic words with nontense vowels are not handled by (3), for (3a) applies to tri-syllabic and longer words, (3b) to bi-syllabic and longer words, and while (3c) applies also to monosyllables, it requires that their vowel be tense. Hence words such as those given in (lb) above are outside of (3), nor were they considered when the rules were formulated. The formulation (9), however, implies that such monosyllabic words will receive stress. Again there is nothing automatic about this fact and monosyllabic words with nontense vowels could logically very well have been stressless. The second class of cases is provided by the examples in (13): (13)

Suez

Madrid

Berlin

Vermont

Japan

Saigon

All these words would normally have received initial stress by (9b). In fact, the town Berlin, New Hampshire, pronounces its name with initial stress. The name of the German capital is thus clearly an exception; i.e., it is marked in some way so that (9b) does not apply to it. But once it has been marked as an exception to (9b), its stress is automatic — it goes on to the last syllable — because (9c) as stated will supply final stress to all words that have not been operated on by (9a) or (9b). Thus, the formal features of the rules we have adopted imply that certain other things should also be true of language, and as we have just seen, these things are indeed true. Up to this point we have dealt with matters that can readily be ascertained by the layman and that may, therefore, not strike the reader as being very surprising. The examples to be discussed directly are rather less self-evident but share with the ones discussed previously the interesting feature that they are predicted by the theory from certain other more readily observed facts. In most examples above there has been only one stressed vowel which we have designated with the numeral '1' in a purely conventional manner. The English language forms compound words by juxtaposing two simple words; e.g.: (14)

1 arthritis cobra

2 symptom venom

1 aluminum 1

parole

2 utensil 2

violation

The thing to observe about these compound words is that the main stress in each of the two words remains on the same place as when the words are pronounced in isola-

A NEW ORGANON. THE FORMALISM OF PHONOLOGICAL RULES

7

tion, but the main stress on the second word is lower than the main stress on the first word. We have indicated this fact by writing a '2' above the vowel carrying the main stress in the second word. What we observe here, then, is an instance of stress subordination: the main stress of the second words becomes subordinated to that of the first, and we need some formal mechanism that would allow us to express this. The mechanism that we propose is somewhat indirect and consists of two interdependent parts. Assume that the individual words of a compound first receive stress by the Main Stress Rule (9), so that the Compound Stress Rule will apply to sequences such as i i radio station where there are at least two words with primary stress. Note incidentally that here we have an instance of nondisjunctive rule order. A word can be subject to both the Main Stress Rule and the Compound Stress Rule. We shall say that the effect of the Compound Stress Rule is to assign primary stress to the vowel bearing main stress in the first word. This may, at first sight, appear totally useless as the vowel to which primary stress is being assigned has already been assigned by the Main Stress Rule. Our move becomes understandable when we postulate the general principle (15) to which all the operation of all stress rules is subject: (15)

When [1 stress] is assigned to some vowel all previously assigned stresses within the domain to which the rule applies are lowered by one.

In other words we account for the stress pattern in a compound word such as radio station by assuming that the stress rules (9) first apply to the individual words and assign primary stresses. The compound rule applies to the sequence i radio

i station

and assigns primary stress to the first word, i.e., to the vowel with primary stress in the first word. Simultaneously the stress lowering convention applies and lowers the only other previously assigned [1 stress] to [2 stress], yielding thus the correct stress contour 2 1 radio station We now state the compound rule more formally: (16)

V-* [1 stress]/[X Y# # Z ] [1 stress] Condition: Z does not contain # # Y contains no [1 stress]

The symbol # # represents the boundary between words. For obvious reasons of visibility we cannot employ here the space that is utilized for this purpose in writing and

8

MORRIS HALLE

printing. We recall that X represents — as in (9) — an arbitrary sequence of consonants and vowels, and we now specify further that these sequences may, but need not include word boundaries, unless expressly forbidden to contain such boundaries. Precisely the same is true of Y and Z. They also represent arbitrary sequences of vowels, consonants and boundaries except that Z is expressly forbidden to contain such boundaries. In view of this the compound word radio station is analyzed as shown below: 1 station a dio 1 1 Y ## When primary stress is assigned to the first vowel in radio the stress lowering convention applies automatically and lowers the primary stress in station to secondary. Given only such compound words as those in (14) we cannot justify the conditions that we have appended to the compound stress rule, but we shall see directly that they are needed. It is a well known fact that there are numerous compound words in English which do not have the falling stress contour as those in (14), but instead have a rising stress contour: 1 2 Madison Street (17) Madison Avenue 2 1 Union Station Union member

1



Christmas oratorio

Christmas story

What is going on in these words is quite similar to what is going on in the words in (14) except that in (14) the main stress of the second word was subordinated to that of the first word; here it is the main stress of the first word that is subordinated to that of the second. We can capture the fact just noted by a second compound stress rule: (18)

F - * [ 1 stress]j[X]_ Y] [1 stress] Condition: Y may not contain p

s

^essj

The compound Madison Avenue would be analysed by (18) as shown in (19): (19)

Madison

venue

The rule would then assign primary stress to the initial vowel in Avenue and this in turn would automatically result in the lowering of the [1 stress] in Madison to [2 stress]. Comparison of (16) and (18) immediately reveals their near identity. We can obtain

A NEW ORGANON. THE FORMALISM OF PHONOLOGICAL RULES

9

(18) from (16) by deleting the subsequence # # Z . This means that (16) and (18) must be abbreviated into the single rule (20) and that they are disjunctive. (20)

V -> [1 stress]/[X

Y(##Z)] [1 stress]

Conditions: Y contains no

[

1 stress! V

J

Z does not contain # # Observe that both (16) and (18) represent actual word forming processes in present day English; i.e., both rules are part of the native speakers' knowledge and must, therefore, be given. However, a given word is subject to either (16) or (18), never to both. In other words, these two rules are, in fact, disjunctive. The language thus confirms to us what we have already discovered by a purely formal argument. The compound words in (17) are entered with a special marker which indicates that they are not subject to rule (16). By the conventions of our description this means that they are subject to (18). The fact that there are stress gradations in compound words quite naturally leads us to investigate stress gradations in simple (noncompound) words. It is not hard to see that words such as those in (21) have a subordinate stress on the last syllable in addition to a primary stress on some earlier syllable: (21)

anecdote brigantine hurricane compensate expedite interpolate personify solidify

anticipate

All these words end with a tense vowel and should, therefore, receive stress on the last syllable by (9c) of the Main Stress Rule. But we know that their primary stress is not on the last syllable but rather on the antepenult. We need, therefore, a special stress retraction rule which will apply after the Main Stress Rule has done its work and assign primary stress to the antepenultimate vowel in words that have primary stress on the last vowel: (22)

V->[1 stress]/[X

C0VC0 ^

S CSS

£

] C0]

Words such as Tippecanoe, Tennessee, etc. would have to be marked as exceptions to (22); whereas bi-syllabic words cited in (2) above are not subject to (22) because it applies only to words three syllables and longer. We illustrate the operation of this rule to the word personify. (23)

person [x

if

y

-c0FCo[lst;ess]c0i

Because of the stress lowering convention (15), the assignment of primary stress to the o of personify automatically brings with it the lowering of the primary stress on the

10

MORRIS HALLE

last vowel to secondary. Rules (9) and (22) yield therefore the stress contour 2 1 personify But now recall that we also have rules (16) and (18)=(20) in the language and we need to inquire whether and how these rules affect the proposed account. Clearly (16) is irrelevant since it requires that there be at least two words in the domain of the rule. Note, however, that (18) is not similarly restricted since while the strings represented by X, Y may contain internal word boundaries they are not required to do so. That is, in determining the stress contour of words such as personify rule (18) will apply in addition to (9) and (22), and, moreover, the three rules apply in the order Main Stress Rule (9) Stress Retraction Rule (22) Rule (18). We show this in (24): (24)

personify 1 1 2 1 3 1 3 personify

input Main Stress Rule (9c) Stress Retraction Rule (22) Rule (18) output

The rules we have imply, therefore, that the subordinate stress in a simple word such 1 3 1 as personify is lower than the subordinate stress in a compound word such as radio 2

station. This is so small a distinction that speakers may have a bit of difficulty in convincing themselves that they, in fact, make this distinction; the distinction, for all that, is nonetheless real, even if rather subtle, and phoneticians have noted it a long time ago. The interesting fact from our point of view is that this very subtle distinction which was not taken into consideration when we formulated the stress rules of English turns out to be an automatic consequence of these rules. We have here an empirical validation of the theoretical apparatus that we have employed in formulating the rules, for it is by virtue of the theoretical apparatus that we were led to discover — or rediscover — extremely subtle facts about English pronunciation. The question may now be raised what is the broader, extra-linguistic significance of the various constraints that we have imposed on the form of the language rules which have been discussed above. We have shown that if we impose these conditions on statements dealing with a particular body of data we can deduce from these statements that certain other facts are true about the language. Consider now this

A NEW ORGANON. THE FORMALISM OF PHONOLOGICAL RULES

11

observation from the point of view of a person learning a language. Clearly he must have some way of committing to memory the facts that he learns; e.g., that there are 1 2 two phonetically distinct classes of compound nouns in English — i.e., Madison Street 2 1 versus Madison Avenue — and that there is a stress retraction rule in English. We have shown that if he commits these facts to memory in a particular form, he will also be able to deduce the fact that in noncompound English words the subordinate stress is at most tertiary. In other words, if he commits to memory the facts in the form proposed he gets further facts about English for free, as it were. From the point of view of a person learning a language this is not a bad bargain. Consider further that when a child learns his mother tongue he is exposed to the language in the most unsystematic fashion, so that enormous portions of the language are presented to him only in the most fragmentary fashion or not at all; yet in spite of this highly unsatisfactory apprenticeship, a child learns the language without difficulty and with astonishing rapidity. The only way in which this achievement becomes at all comprehensible is if we can assume that the child has available some sort of device which leads from a very small amount of data to the correct inferences about a large additional set of facts. In other words, the conditions under which children learn suggest rather forcefully that children must get a lot for free as it were. And the formal constraints I have sketched above are a possible mechanism whereby this extra information is made available to children at no cost. Massachusetts Institute of Technology

DISCUSSION

McCawley: Your treatment of Berlin, etc. is to mark them as exceptions to parts (a) and (b) of your stress rule, so that they undergo part (c) and receive final stress. However, there is great asymmetry as regards which parts of the rule a word may be an exception to: while there are literally hundreds of words which would have to be treated as exceptions to (a) and (b), there are very few words (e.g. allegro, composite) which would have to be treated as exempt from (a) but undergoing (b). This provides support for Ross's contention that there are two separate rules in English: one of non-final stress (encompassing what you take in under (a) and (b)) and one of final stress, that words ending in certain consonants (which are not the same for nouns as for verbs or for adjectives) undergo the one rule or the other, as an idiosyncrasy of the word, and all other words only undergo the rule of final stress. Your implicit treatment of Madison Avenue requires a different principle of rule government than the one given in Sound Pattern of English, according to which a rule whose structural description is met will apply or fail to apply on the basis of whether the morpheme containing the affected segment is marked as an exception to the rule or not. The anomalous stress on

12

MORRIS HALLE

Madison Avenue is due to Avenue, which does not contain the segment which the first part of the rule would mark as [1 stress]. However, the basic idea of your rule government principle could be saved if the principle were interpreted as exempting the exceptional morpheme from undergoing any effect of the rule, e.g., the stress reduction which results from putting a stress on Madison by the ordinary compoundrule.

THEODORE M. LIGHTNER

REMARKS ON UNIVERSALS IN PHONOLOGY*

In the historical development of many languages, a long nasalized P appears in place of an earlier sequence VN.1 This P may subsequently become denasalized, leaving oral V in place of the original VN sequence. Even after denasalization, however, the historical process may leave an effect on the synchronic description of a language. Thus, to give an example treated in greater detail in § 4 below, in Lithuanian one finds third person past brento~ infinitive brgsti (phonetically, the alternation is [en~ie]) from, roughly, /brent+o/ and /brent+ti/. The process of vowel nasalization raises at least two interesting questions. First, we must ask why the process should occur independently in many different languages. Second, we must decide what to do about the synchronic description of languages which have no phonetic nasal vowels; for Lithuanian, for example, we must decide whether the synchronic description should include a rule of vowel nasalization (followed by a rule which denasalizes vowels) or whether the synchronic description should not mention nasalized vowels at all. One approach to these problems is to argue that there is a set of universal phonological rules, one of which nasalizes vowels in some environment. Such an approach is attractive for a number of reasons. First of all, it permits an immediate explanation for why vowel nasalization (and other common processes, such as truncation of vowels before vowels, the shift of high vowels to glides before vowels, word final obstruent devoicing, and so on) should occur independently in many unrelated languages. Second, since the proposed rule for vowel nasalization is universal, we can assume that the synchronic description of a language like Lithuanian will include this rule. And third, we will have a principled reason for formulating 'natural' phonological rules in synchronic descriptions (as opposed to 'crazy' rules; cf. Bach and Harms [forthcoming]). In this paper I shall examine some possibilities for formulating a universal rule to account for the process of vowel nasalization. This investigation will lead us to consider several other well-known rules in various languages and also to comment on a few of the distinctive features used in phonology. * I am indebted to Emmon Bach, Bob Freund, Eugene Grace, Maurice Gross, Bob King, S-Y. Kuroda, and Jerrold Sadock for their help in preparing this paper. 1 The symbol Kis used for any vowel, Pfor any long vowel, Kfor any short vowel, TV for any nasal consonant, C for any non-vowel (except, sometimes, /), Nasalized vowels are indicated either with a tilde (3, e, V etc.) or with a nasal hook (q, § etc.).

14

THEODORE M. LIGHTNER

1. COMPENSATORY LENGTHENING One might be tempted to begin describing the process of vowel nasalization by proposing that in certain environments a vowel before a nasal is nasalized, and that the nasal is then dropped with 'compensatory lengthening' of the preceding nasalized vowel. Before discussing the nasalization of vowels, therefore, let us consider briefly the notion of compensatory lengthening. Take, for example, the case of Lachmann's Law: a vowel is lengthened when followed by VOICED CONSONANT+VOICELESS CONSONANT. Thus Latin fació¡factus, but ago/actus, video/visus etc. It is clear that as stated, the 'law' is impossible. The most interesting phonological treatment I have seen is Foley's; he suggests that in going from ag+t+us to phonetic [àxtus], the 'weakening' of g to [x] requires a corresponding 'strengthening' elsewhere, manifested here as length of the preceding vowel. But if the shift of g to [x] causes lengthening of the preceding vowel, why doesn't the shift of k to [x] in [faxtus] also cause lengthening? 2 Moreover, there are forms like seded/sessum, stringo/strictus, findó/fissus scindo/scissus etc., where lengthening does not occur at all (stringò, findò, scindo have nasal infix, but this is irrelevant; cf. tangó/tàctus).3 2

Foley (n.d.) has the strength scale k>g> x> y. Given this scale, the shift of k to x involves even greater 'weakening' than the shift of g to x and ought, following this reasoning, to cause even greater length of the preceding vowel. Moreover, it is not overwhelmingly clear that Latin ct is realized [xt] and not Pet] ; if the realization of ct is [kt], then Foley's argument reduces to the old argument mentioned in note 4 below. 8 In view of the highly problematic nature of whatever process is involved here, it is surprising to find that Lachmann's Law is the example chosen to demonstrate that synchronic ordering need not reflect chronological acquisition (Kiparsky [1965], Postal [1968]). Compare Watkin's analysis (forthcoming) in which he claims to give "irrefutable evidence that no phonetic conditioning whatever is involved in Lachmann's Law". But Watkins goes too far in the opposite direction (i.e., away from a 'phonological' analysis): one fact that EVERYONE agrees on is that lengthening can occur ONLY if the vowel is followed by a voiced consonant. But if Lachmann's Law is phonologically conditioned, then the rule which accounts for the process will have to a 'crazy' rule because vowels simply do not lengthen before a consonant cluster; in case after case, from Germanic languages to Sierra Miwok languages, vowels SHORTEN before consonant clusters. It seems, then, that Lachmann's Law must be handled by a minor rule, a rule which mentions both the voiced segment and the morphological category; those roots which undergo the minor rule are so marked in the lexicon (discussion of minor rules, see Lakoff [1965], and Lightner [1967b]). As regards synchronic ordering and chronological acquisition, there are not very many straightforward examples which justify the claim that Kiparsky (1965) and Postal (1968) try to substantiate, but an interesting example has recently come to my attention : in the 'alfalfa' secret dialect of English, the rules are that after each vowel, If is inserted, followed by a copy of the preceding vowel. Thus, not-+nolfot, suffix-*sulfufilfix, and so on. But in English, vowels are laxed before two consonants if the consonants are not both dentals (cf. Chomsky and Halle [1968], 171-172, 175-176, et passim); since / i s not a dental, the question arises as to how tense vowels before If are pronounced. It turns out that vowels are lax here. Thus by [baj] is not pronounced *[bàjlfàj] but [balfaj], no [now] is pronounced [nslfdw], gate [gsjt] is pronounced [gslfEjt], and so on. Clearly, since the secret language rules are acquired later than the regular rules of English, the synchronic ordering of the 'alfalfa' rules (which must apply not only before vowel taxing but before several other phonological rules as well) does not mirror the chronological acquisition.

REMARKS ON UNIVERSALS IN PHONOLOGY

15

In the example of Lachmann's Law, then, the notion of compensatory lengthening has no explanatory value whatever.4 To take another example, consider short i and u in Japanese. Under various obscure conditions, these vowels may be pronounced voiceless;5 voiceless y and / may then be dropped with 'compensatory lengthening' of the preceding consonant. How is one to capture the process of dropping with compensatory lengthening? We could have two ordered rules: (DROP-opt)

{/, w} -* 0, where 0 represents 'zero'

(LENGTHEN)

C->[+long]/

( # ) C6

But these rules hardly capture the notion of 'compensatory lengthening'. Another set of ordered rules might be: (LENGTHEN)'

C->[+long]/ {ji, /}, if and only if environmental {«, (} undergo the following rule:

(DROP-opt)'

{&/}-» 0-

Since the applicability of the first of these rules is dependent on the application or non-application of a rule which follows, it is far more than likely that the first rule is not well-formed. This set of rules, however, comes closer than the first set to capturing the notion of 'compensatory lengthening' because the two rules are linked together by the condition that one applies if and only if the other also applies. Both sets of rules suffer from what might turn out to be a different defect: it might be the case that unconditioned deletion is not a possible phonological process. If it is the case that all rules of the type X-* 0 must be environmentally conditioned, then neither set of rules is possible because both involve a rule of unconditioned deletion. The statement that voiceless y, / may be dropped, and that if they are dropped, there is compensatory lengthening of the preceding consonant implies that a single, optional process is at work: (K~C) 4

a voiceless vowel may optionally become a copy of the preceding consonant.

It is hardly necessary to observe that Foley (n.d.) was not the first to attempt an explanation in terms of compensatory lengthening. Thus Buck (1933), 94, points out that the lengthening of the root vowel "is regarded by many scholars as due to a lengthening which attended the change of the voiced consonant to a voiceless", and correctly goes on to note that such an explanation is not acceptable. 5 The traditional explanation has it that short i, u may be pronounced voiceless between voiceless consonants provided the i, u are not highpitched; if a sequence of syllables, all containing either i or u, occurs, then every other i, u in the string may be pronounced voiceless. Such a description runs afoul of minimal pairs like the following (raised + indicates high pitch): kftd 'North' and kfta 'came'; sfku 'four times nine' and sjku 'to spread'. One of my informants has tisjk[zm 'intellectual' with voiceless/'s in two successive syllables. Needless to say, there is a lot of dialectal variation; in some dialects, voiceless /, v are permitted only if the following syllable has a low vowel, for example (see Muraki [1970]). 6 Japanese words all have CVCV structure; consonant clusters and word final consonants arise ONLY through the application of (DROP).

THEODORE M. LIGHTNER

16 But if

C ) is the correct rule, then the n o t i o n o f dropping a v o w e l with c o m p e n -

satory lengthening o f the preceding c o n s o n a n t is meaningless: n o v o w e l is dropped. A g a i n , consider the case o f m o n o p h t h o n g i z a t i o n o f eu/ou t o u: (ejo ~ w)

{e, o} -* uj

u

Surely n o o n e w o u l d say that in this case e a n d o are dropped with c o m p e n s a t o r y lengthening o f the f o l l o w i n g vowel. A n d yet this case is fundamentally the same as the process in Japanese. Or consider Latin scrlptus,

octo>Italian

scritto,

otto. N o o n e w o u l d say here that

c o n s o n a n t s are dropped with compensatory lengthening o f the f o l l o w i n g stop. W h a t is involved in all these cases (and in numerous others; a f e w further examples are given in the text below) is merely complete assimilation o f o n e segment to a c o n t i g u o u s segment. D a t a such as these indicate that it w o u l d be inadvisable t o begin a study o f v o w e l nasalization with the h o p e o f using, s o m e h o w , the apparently ill-conceived n o t i o n o f "compensatory lengthening". Let us turn n o w to a n examination o f s o m e data o n v o w e l nasalization. 2. OLD CHURCH SLAVONIC (OCS) 7 In O C S , and presumably also in Proto-Slavic, V N before w o r d boundary or any c o n sonant except j results in P; thus o n e finds alternations like klçtl, klçlu~klinq dgtl~dumq

'blow', but first person singular lomljg 'I break' f r o m lom

'curse',

+j+â+m.s

7 There are two letters used to represent the long, nasalized vowels in Slavic, g and g. These vowels were non-high even though they may have their source in high oral vowels, as indicated by the examples given in the text below. There is a general tendency in language to lower high nasal vowels. Thus, besides Slavic, compare French finir, parfumer ~ fin, parfum, i.e., [i, y ~ g, œ]. Shevelov (1965), 325, writes that in Kashubian, ç shifted to j but that this j was soon denasalized; moreover, before hard dentals and finally, ç shifted down to q. Compare also Gordon's comments (§ 3 below) on the lowering of nasalized i and u. The reasons for preferring lower nasal vowels to higher ones are at least two. First, a high tongue position and a low velum form a constricted passage, which is undesirable in vowels (it is no accident that the lax HIGH (=constricted) vowels drop in Japanese, Slavic, Armenian, Iranian, Uzbek, and no doubt countless other languages). Second, the nasality of nasal vowels is optimally achieved by reducing the first formant of the vowel, a trick that can be done by equating the volume of the pharyngeal cavity with that of the nasal cavity. Delattre (n.d.) writes, "Pour /i/, /y/, et /u/ nasalisés la cavité pharyngale est très grande, ce qui explique en partie un premier formant très bas. Pour /ë/, /dé/, /5/, cette cavité pharyngale est relativement petite, ce qui explique en partie un premier formant plus élevé. Pour /a/ nasalisé, la cavité pharyngale est très petite, ce qui explique en partie un premier formant très haut. Pour /â/ la cavité est plus grande que pour /a/ nasalisé mais elle est encore relativement petite. De fait les quatre voyelles nasales modernes ont à peu près la même dimension de cavité pharyngale. Comme elles ont également la même fréquence de premier formant, il ne faut pas voir là une coïncidence et il faut en chercher la raison. La voici, nous semble-t-il. Pour que les quatre nasales aient abouti à la même dimension pharyngale il faut que cette dimension soit la plus FAVORABLE à la nasalité distinctive. Or on observe que cette 'dimension favorable' est semblable à celle de la cavité rhino-pharyngale qui la surplombe. Et la théorie acoustique veut que lorsque deux cavités ont même volume comme

REMARKS ON UNIVERSALS IN PHONOLOGY

17

les cavités pharyngales et rhino-pharyngales, leurs ondes de résonance vibrent en phase inverse et s'annulent mutuellement, ce qui annule, ou tout au moins réduit fortement, les harmoniques du premier formant. C'est donc en ajustant le volume de la cavité pharyngale à celui de la cavité fixe rhinopharyngale qu'on réduit le plus l'intensité du premier formant et qu'on produit le mieux l'impression de nasalité. Le simple abaissement du voile du palais (sans ajustement de la cavité pharyngale au 'volume favorable') ne produit qu'un amortissement général du premier formant, amortissement qui affaiblit déjà, en effect, l'intensité subjective (loudness) du premier formant (puisqu'elle est moins concentrée) mais d'une manière moins efficace que ne le ferait l'annulation des harmoniques." In spite of the general tendency to lower high nasalized vowels, there are, of course, many languages with such phonetic vowels. So, for example, in English {hint, punt etc). In Portuguese, there are not only high nasal vowels but high nasal diphthongs: comum [komu] 'common', Jim [fï] 'end', muito [muPtu] 'much', leôes [ljôïs] 'lions' (cf. sg. leào [ljtû]). Some languages, such as the Athapaskan language spoken by the Hare Indians in northwestern Canada, have only high nasalized vowels; Hoijer (1966), 507 writes that ' 'Hare j/ and Hare i( followed by -n have four sources : they may derive from PA *u followed by -n, PA *u followed by -n, PA *a followed by -n, or PA *A followed by -n. Hare q, it should be noted, is extremely rare; it occurs only in three or four forms, none of which can so far be etymologized." It is not clear to me how this tendency to lower high nasalized vowels should be handled. Possibly some marking convention could be invoked. I think it important to note that there seems to be a tendency to nasalize first non-high vowels and only then — after the introduction of phonetic nonhigh nasalized vowels into the system — to nasalize the high vowels, and also a tendency to keep the height of the underlying vowel intact and only then — after the introduction of the whole set of phonetic nasalized vowels — to lower the high ones. This, for example, seems to be the case in the development of French, about which Delattre writes " O n sait qu'en ancien français les voyelles suivies de consonnes nasales se sont nasalisées au point de jouer linguistiquement un rôle distinctif (annu > an [â]), à condition toutefois que la consonne nasale ne passe pas à la syllabe suivante à l'époque (tardive) où les syllabes tendent à s'ouvrir: annata> année [â'ne]> [a'ne], antiu>alné [ë'ne] > [e'ne]. Cette évolution de la voyelle nasalisée a procédé avec un remarquable lenteur puisqu'elle a vraisemblablement commencé la première fois qu'un Gaulois a essayé de prononcer un mot latin. On sait d'autre part que cette nasalisation n'a pas abouti à la nasalité distinctive à la même époque pour toutes les voyelles. Si l'on se fie aux rimes et aux assonances des textes littéraires, entre les nasalités distinctives les plus précoces et les plus tardives il y a un écart d'au moins trois siècles et peut-être de six. Dans la Chanson de Roland, par exemple, les /a/ nasalisés n'assonaient plus aves les /a/ oraux, tandis que les /i/, les /y/, les /o/ nasalisés assonaient encore avec les /i/, les /y/, les /o/ oraux. La chronologie de la nasalité est approximativement la suivante. D'après M. K. Pope, les /a/ nasalisés ont atteint le degré distinctif dès le dixième siècle, les /e/ nasalisés au onzième ou plus tard, les /o/ nasalisés au douzième ou plus tard. Quant aux /i/, aux /y/, et aux /u/, ils l'ont atteint au plus tôt à la fin du treizième siècle, mais peut-être seulement au seizième siècle quand leur modification de /i/, /y/, /u/ nasalisés en /ë/, /&/, /ô/ a atteint son timbre définitif." In Slavic, the evidence is less clear-cut. But we may note that the palatalization of velars first observed by Baudouin de Courtenay (1894) occurs only after high front t (with very complicated additional requirements): OCS ovicâ 'sheep' (cf. Lith. avikè 'small lamb'), OCS dvi3àtl 'to move (cf. OCS dvigngti 'to move'), and so on. The palatalization also occurs after g < in, as in OCS kunç3i from Germanic kmingaz. If we assume that nasalization in Slavic retained the original height of the oral vowel and then later lowered the high nasal vowels, the environment of the BdC palatalization can be I X, where ¿"represents any high front vowel whatever ( = nasalized or not), and X represents whatever other conditions must be imposed on the proper formulation of this rule. But if in passed immediately to non-high ?, then the environment for the Baudouin de Courtenay palatalization rule would have to be formulated with an optional nasal, t(n) X. 8 Discussion of OCS conjugation, see Lightner (1966). The first person singular forms present a real problem in analysis, not with vowel nasalization, but with the prediction of transitive softening (—shifts like m > mlj ; b > blj ; / > si ; s > £). It is not entirely clear what the correct underlying representation of the 1 sg. ending is; in writing â + m, I follow Meillet, (1965), although this may not be correct. The problems mentioned in this footnote have nothing to do with vowel nasalization, and I shall therefore not continue this discussion (see Kayne, 1968, Lightner, 1972, for further details).

THEODORE M. LIGHTNER

18

The Slavic case is thus a paradigm example of what might be interpreted as vowel nasalization followed by loss of N with compensatory lengthening of the preceding vowel. But if the compensatory lengthening treatment is not allowed (cf. § 1 above), there still remains an elegant analysis. First, long nasalized vowels are derived ( # in the rule below and throughout this paper represents word boundary): Cv~P)

v -*P i

Arj^J

All that remains now is to drop the nasal. But in Slavic, it is the case (1) that no words end in a consonant and (2) that j, w drop before a consonant (1 sg znajq ~ inf znatl, 1 sg zlwg~ inf zltl). Thus — regardless of how nasalization is treated — the grammar must include the following two truncation rules: (1)

C->0/

(2)

{ j , w}->0/

# C

|, If rule (2) is written not with i v o c a ' l to the left of the arrow but with | [-cons J L-obstrJ this rule will drop not only j, w, but also n, m: (2)'

{j, w,«,m}->0/

C

Thus the entire process of vowel nasalization in Slavic can be accounted for with rules ( V ~ P ) , (1), and (2)', where the last two rules are independently motivated. This was, in fact, the analysis proposed in Lightner (1966). Consideration of the nasalization process in other languages, however, will show that if one attempts to broaden the scope of phonological rules — if one attempts to account not only for relations found within a single language, but also for language-independent relations — this seemingly elegant analysis is not correct. 3. OLD NORSE

According to Gordon (1957) 267: "As late as the last half of the twelfth century the Iceland vowels and dipththongs also occurred nasalized, when immediately preceded or followed by a nasal consonant, or if followed immediately by a nasal consonant in Prim. Norse or even in Germanic, which has been lost. Thus syna, mer 'to me', / (PrN. in), far (Ger. *fanh-) 'takes', had nasalized vowels. The nasal quality was lost earliest in unaccented syllables, and earlier in a vowel following a long syllable than in one following a short syllable". On page 275: "When p, t, or k followed a nasal consonant, the nasal was assimilated to the following consonant; a preceding (nasalized) i was then lowered to e, and a preceding (nasalized) u to o, as in sakkva (from *sekkwa), pp. sokkinn; cf. OGut. sinqua, pp. sunken... . The same lowering took place when a following nasal was lost, as in porr, from *punraR (OE punor)." On page 276: "Long vowels were shortened before double consonants (except tt from

REMARKS ON UNIVERSALS IN PHONOLOGY

19

ht), as in gott, neut. of godr; minn 'my', cf. fem. mln." And, finally, on pages 282-283: "A nasal consonant was assimilated to a following p, t, k: mp became pp: kappi (cf. OE cempa). nt became tt: batt, pa. t. of binda, mitt, neut. of minn; vetr ( = O E winter). Forms such as vant, seint have n restored by analogy, nk became kk: drekka ( = O E drincan), gekk [in PrN, final g was devoiced; hence *geng>genk, ultimately gekk TML], ykkr ( = O E incer)." I am not sure how reliable these data are; I include them for two reasons: first, one does not expect vowel nasalization caused merely by the presence of a nasal on either side of the vowel, although note Portuguese muito [muintu] 'much', with nasalization due, apparently to the preceding m, and, to cite a more regular case, the nasalization of vowels in the Amoy dialect of Chinese after word initial nasal (cf. Chu, Yen); and second, it is clear that in forms like drekk- from drink-, the nasal has NOT dropped with compensatory lengthening of the preceding vowel. In the absence of supporting evidence, the development drink- -> dr§k- -* dr§k- -* drekk- is highly unlikely. I assume that the nasal responsible for nasalizing the preceding vowel subsequently assimilated completely to a following stop (as Gordon suggests): drink- -*dr§nk- -* dr§kk- -* drekk-. The Old Norse nasalization rule cannot derive a long nasal vowel from a short oral vowel; therefore, if there is to be a universal rule for vowel nasalization, the rule given above in § 2 for Slavic cannot be correct because it would lead to incorrect results in Old Norse. Consideration of the Lithuanian and the Polish data in §§ 4-5 below will confirm this conclusion. 4. LITHUANIAN

Lithuanian has a phonetic contrast between long and short vowels. Short vowels may be stressed, in which case they have high pitch. Long vowels and diphthongs may be stressed, in which case they have either a rising contour (indicated by the diacritic ~ in orthography, e.g.,plaukti 'to swim') or a falling contour (indicated by the diacritic' in orthography, e.g., augti 'to grow'). So-called 'mixed diphthongs' are formed from a vowel followed by a nasal or liquid; these diphthongs, when accented, show either a falling contour (antis 'duck') or a rising contour (antis 'breast'). An underlying VN sequence is realized as a long oral K when followed by { j , v, I, r, m, n, s, z, s, £}, a class of segments I will temporarily refer to with the arbitrary cover symbol Z. 9 Thus third person past brehto but infinitive brgsti 'rot' 1 0 (see page 20). 9

Kenstowicz (1969), 15, gives the following examples with the prefix san- 'con-': simburis assembly BUT: s£junga union sdmpilas store, stock s£voka idea sdndora covenant sjiskambis harmony sintaka confluence s££lavos sweepings sdnkaba coupling s£2ine conscience (e — long e) s£lytis contact (y = long i) sjirasas register s£mokslas conspiracy s£narys joint

For more discussion of Z, see page 32.

20

THEODORE M. LIGHTNER

One reasonable way of capturing the suprasegmental phenomena just described is to consider long vowels as sequences of identical short vowels and to mark vowels for a binary accent feature which is interpreted phonetically as

Thus in the string CVC, the vowel is short, stressed, and high-pitched. In the string CVVC, the drop from a high-pitched first vowel to a non-high-pitched second vowel automatically results in a falling contour. A rising contour is indicated CVVC (plaUkti), CVLC (kaltas 'guilty') or CVNC (trumpas 'short'). Now consider the third person past and infinitive pairs brento, brgsti and brendo, bresti 'ripen'. The underlying representations (after the dental assimilation mentioned in note 10) are brent+0, brens+ti and brSnd+d, bres+ti, respectively. The past forms are immediately interpreted as brento and brendo. But in the infinitive forms, vowel nasalization takes place because en is followed by a member of Z. Suppose the rules are as follows: (i)

V-* [+nasal]/

(ii)

N-+Q/V

NZ

Brensti is now incorrectly realized as *br$sti, with a short nasal vowel and loss of all suprasegmental information. Similarly, brSnsti is incorrectly realized as *br§sti, with a short high-pitched nasal vowel. If rule (i) is changed to derive a long nasal vowel, then underlying brinsti is realized incorrectly as * bresti, with a long high-pitched nasal vowel instead of a long nasal vowel with falling contour. The problem in the two inf. forms is to have the derivations start with ¿n, en and end with §§, §§, respectively i.e., with long nasal vowels having a falling and a rising contour, respectively. This can be done by means of the following two ordered rules: (V ~ V)

V

[ + nasal] /

(N ~ V)

N

[«features] /

NZ V [afeatures]

Rule (N ~ V) states that a nasal assimilates all the features of a preceding nasal vowel: 10 In Lithuanian, a dental stop shifts to a continuant before a dental stop; this is presumably an inherited IE assibilation rule, found also in Iranian, Slavic, and Greek. Meillet (1964), 131, claims that the Latin reflex of geminated dental stops is ss, and, indeed, we find sessum, e.g., from sed +1 + um. But it is possible that in Latin, the reflex of dental stop before dental stop is st as in Iranian, Baltic etc. (note forms like rodo 'I gnaw', rostrum 'beak'); if this were correct, then ss in forms like sessum would be from st with assibilation in intervocalic position. De Saussure (1877), however, has argued convincingly against such a development, noting, e.g., that original st does not shift to ss (thus ustus, gustare, pistor etc.). I am inclined to agree with de Saussure's conjecture that the intermediary stage in the Latin development is ts (cf. esse < *etse, *edse). Thus both developments of d/t followed by t are similar: in one case the first stop spirantizes (W > st), in the other case the second stop spirantizes (tt>ts). 11 The question of how a vowel gets marked for accent in the first place is irrelevant to the present topic; discussion of this important problem, see Heeschen (1966), Darden (1970), Kenstowicz (1969).

REMARKS ON UNIVERSALE IN PHONOLOGY

21

¿n->gn-*gg and en^>n-*§§. Notice that in neither of these derivations can the accent feature be assimilated; if accent were assimilated, then *££ would be derived from f?«, and *§§ from