Papers in Linguistics and Phonetics to the Memory of Pierre Delattre [Reprint 2015 ed.] 9783110803877, 9789027923103

Table of contents :
PREFATORY NOTE
Pierre Delattre
Publications of PIERRE DELATTRE
Tonal Experiments with Whispered Thai
Phonemics as a Discovery Procedure in Synchronic Dialectology
The Syntax of parecer
[ts] et [dz] en français canadien
Some Observations on Méchant Poète vs. Poète Méchant
Some Observations on the Pitch of Questions
Pierre Delattre, Teacher of French
De l’r gaulois à l’r français
Analyse acoustique de deux allophones du l final anglais
On Utterance-final [ł]] and [u] in Portuguese
“ptk” et “bdg” français en position intervocalique accentuée
French Listeners and the Tonic Accent
Le Système accentuel du français
Is ‘Middle French’ Necessary?
L’Origine et l’évolution du mot gnôle “eau-de-vie”: la géographie linguistique et l’homonymie
Vowel Overlap in Azerbaijani
Phonemic Texts
The Question-Phrase Fall-Rise in British English
Entry Words: Grammars and Dictionaries
Phonetic Prerequisites for a Distinctive Feature Theory
Perception phonétique au seuil d’audition
Some Observations Concerning the Third Tone in Latvian
Étude de la prononciation de e accentué chez un groupe de jeunes parisiens
In Search of the Acoustic Cues
Stop Duration and Voicing in English
New Procedures for Descriptive Phonetics
The pan-European Suffix -esco, -esque in Stratigraphic Projection
Une question de méthode et la solution d’un problème concret (esp. crecer- crezco)
La nature phonologique d’e caduc
Cinefluorographic Studies of Speech Articulation
A Comparison of Spanish Single-tap /r/ with American /t/ and /d/ in Post-stress, Intervocalic Position
Some Similarities in Old French and Modem Spanish Verb Morphology
Developmental Aspects of Auditory Discrimination
Sur la représentation par écrit des sons parlés
Le Seuil différentiel de durée
Sonocineradiography in Speech-Sound Analysis
The Loi de Position as a Pedagogical Norm
The Many Uses of Fo
Linguistique générale et théorie de la linguistique

Citation preview

JANUA LINGUARUM STUDIA MEMORIAE NICOLAI VAN WIJK DEDICATA

Series Maior,

54

Pierre Deìattre

PAPERS IN L I N G U I S T I C S A N D PHONETICS TO THE M E M O R Y OF

Ρ

edited by

ALBERT VALDMAN Indiana

University

EDITORIAL C O M M I T T E E S I M O N BELASCO Pennsylvania State University

ANDRE MALECOT University of California - Santa Barbara

F R E D E R I C K D. E D D Y University of Colorado

MICHEL MONNOT University of California - Santa Barbara

R. P I E R R E L É O N University of Toronto

C A R R O L L L. O L S E N University of Toronto

1972

MOUTON THE HAGUE · PARIS

O Copyright 1972 in The Netherlands. Mouton & Co. Ν.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: 77-190140

Printed in The Netherlands by Mouton & Co., Printers, The Hague.

PREFATORY NOTE

It had been the desire of a few former students of Pierre Delattre or linguists and phoneticians whose career he had profoundly influenced to present to him on the occasion of his sixty-fifth birthday a collection of papers in linguistics and phonetics that would reflect the breadth of his research interests and pedagogical and administrative experience. But after delays in publication and his untimely death, the efforts of the Editorial Committee were turned in profound sorrow to the preparation of this memorial volume. The broad scope of the contributions to this memorial volume bears witness to Pierre Delattre's profound influence in a large variety of areas of scholarly endeavor. While limitations of space and our attempts — alas, too successful in retrospect — to keep the preparation of the originally intended Festschrift secret to the hommagé reduced the potential number of contributors, we hope that the authors of the papers of this volume will bear testimony for all those who, touched by Pierre Delattre's magnetic energy and enthusiasm or influenced by his searching studies, would have liked to express their esteem and gratitude. The Editorial Committee

CONTENTS

PREFATORY NOTE

5

ANDRÉ MALÉCOT a i l d ALBERT VALDMAN

Pierre Delattre Publications of PIERRE

11 DELATTRE

21

ARTHUR S. ABRAMSON

Tonal Experiments with Whispered Thai

31

SIMON BELASCO

Phonemics as a Discovery Procedure in Synchronic Dialectology

45

DWIGHT BOLINGER

The Syntax of parecer

65

RENÉ CHARBONNEAU e t BENOÎT JACQUES

ft,] et [ d j en français canadien

77

PIERRE F. CINTAS

Some Observations on Méchant Poète vs. Poète Méchant

91

ANTONIE COHEN

Some Observations on the Pitch of Questions

97

FREDERICK D. EDDY

Pierre Delattre, Teacher of French

105

FRANÇOIS FALC'HUN

De IV gaulois à 1 'Γ français

109

GEORGES FAURE

Analyse acoustique de deux allophones du / final anglais

117

DAVID FELDMAN

On Utterance-final Ρ] and [u] in Portuguese

129

E l l FISCHER-JORGENSEN

"ptk" et "bdg" français en position intervocalique accentuée

143

D. Β. FRY

French Listeners and the Tonic Accent

201

ERNEST F. HADEN

Le Système accentuel du français

209

8

CONTENTS

ROBERT A. HALL, JR.

Is 'Middle French' Necessary?

215

FRANK R. HAMLIN

L'Origine et l'évolution du mot gnôle "eau-de-vie" : la géographie linguistique et l'homonymie 223 FRED HOUSEHOLDER

Vowel Overlap in Azerbaijani

229

LEE S. HULTZÉN

Phonemic Texts

231

WIKTOR JASSEM

The Question-Phrase Fall-Rise in British English

241

ALPHONSE JUILLAND

Entry Words : Grammars and Dictionaries

253

PETER LADEFOGED

Phonetic Prerequisites for a Distinctive Feature Theory

273

JEAN-CLAUDE LAFON

Perception phonétique au seuil d'audition

287

ILSE LEHISTE

Some Observations Concerning the Third Tone in Latvian

309

PIERRE R. LÉON

Étude de la prononciation de e accentué chez un groupe de jeunes parisiens . 317 ALVIN M. LIBERMAN a n d FRANKLIN S. COOPER

In Search of the Acoustic Cues

329

LEIGH LISKER

Stop Duration and Voicing in English

339

ANDRÉ MALÉCOT

New Procedures for Descriptive Phonetics

345

YAKOV MALKIEL

The pan-European Suffix -esco, -esque in Stratigraphie Projection

357

BERTIL MALMBERG

Une question de méthode et la solution d'un problème concret (esp. crecercrezco) 389 ANDRÉ MARTINET

La nature phonologique d'e caduc

393

KENNETH L. MOLL

Cinefluorographic Studies of Speech Articulation

401

MICHEL MONNOT a n d MICHEL FREEMAN

A Comparison of Spanish Single-tap /r/ with American /t/ and /d/ in Post-stress, Intervocalic Position 409 CARROLL L. OLSEN

Some Similarities in Old French and Modern Spanish Verb Morphology . . 4 1 7

CONTENTS

9

R. L. POLITZER

Developmental Aspects of Auditory Discrimination

425

A. ROSETO

Sur la représentation par écrit des sons parlés

433

MARIO ROSSI

Le Seuil différentiel de durée

435

HENRY M. TRUBY

Sonocineradiography in Speech-Sound Analysis

451

ALBERT VALDMAN

The Loi de Position as a Pedagogical Norm

473

WILLIAM S-Y. WANG

The Many Uses of F0

487

EBERHARD ZWIRNER

Linguistique générale et théorie de la linguistique

505

PIERRE DELATTRE

Pierre Delattre died on July 11, 1969 in Santa Barbara after a strenuous game of tennis with his wife and two sons, a most fitting death for this dynamic modern Renaissance man who enjoyed every moment of life to the fullest and committed himself totally, whether it be at the piano keyboard, in the concert hall, on the mountain trail, or behind the net.1 Although nearing the compulsory retirement age of the University of California system, he was deeply engaged in his Speech Synthesis Project and the organization of a graduate program that would harmonize literary and linguistic studies and train teacher-scholars of French who would follow him in delicately balancing scientific inquiry with a tolerant and rich humanism. By the lást decade of his academic and scientific career, Pierre Delattre had achieved indisputable international prominence in the field of experimental phonetics and French linguistics, recognized by numerous honors including designation as Fellow of the Acoustical Society of America — a significant honor for a non-physicist — and the award of the title of Chevalier de la Légion d'Honneur by the French government. But this intense and devoted teacher of French could not have practiced his craft for nearly forty years without leaving an indelible mark, and indeed there is scarcely any teacher of French in this country who has not been influenced, directly or indirectly, by Pierre Delattre's theory and practice of applied French phonetics as well as his ideas about foreign language (FL) pedagogy in general. Delattre was born in the town of Roanne in central France on October 21, 1903, the thirteenth child of a Huguenot minister, and the austere character of his native Calvinist milieu accounts for his capacity for long hours of hard work impervious to external distractions. Very early he demonstrated musical talent and eventually became an accomplished pianist and fair composer of short pieces. His attachment to music was to remain with him throughout his life, and it no doubt explains his strong interest in intonation and other prosodie features. After completing secondary school studies in France, he emigrated to the Detroit area where, together with his brother André, who was to become á leading specialist in eighteenth century French literature, he enrolled at the University of Michigan. While preparing his Ph.D. he 1 We should like to thank Mme Pierre Delattre, Simon Belasco, and Fred Eddy for making available to us biographical and bibliographical material.

12

PIERRE DELATTRE

taught French at Wayne State University; he also studied at the Institut de Phonétique of the Sorbonne and obtained a certificate in phonetics. In 1941 he joined the faculty of the University of Oklahoma and in 1944-45 launched an experimental course which applied some of the procedures of the Army Language School method at the University of Oklahoma. The most decisive step in his career was his acceptance of a post at the University of Pennsylvania in 1947 which placed him in propinquity to acousticians at ΜΓΓ, Bell Telephone Laboratories, and Haskins Laboratories and led to his participation in research utilizing the Pattern Playback devised by F. S. Cooper at the latter institution. At Pennsylvania Delattre also became associated with a major graduate program in Romance languages and eventually attracted several disciples to French phonetics who went on to attain scholarly and academic distinction. Except for Delattre there was no director of serious French linguistic studies in this country between the late thirties and early fifties. From 1941 until 1957 he directed the remedial phonetics program at the Middlebury College French Summer School. In 1953 he moved west to the University of Colorado, and in 1961 he established his Speech Synthesis Project which he moved to the University of California, Santa Barbara in 1964. No doubt these younger universities provided a greater outlet for his infinite creative energy than the tradition-bound Ivy League institution, but also Colorado and California gave him more opportunity to participate in sports which he enjoyed with as much exuberance and intensity as he did his research and teaching. We recall that during the last International Congress of Phonetic Sciences in Prague where he was invited to deliver one of the plenary session papers he managed to squeeze in attendance at a soccer game between two of Czechoslovakia's best teams, a feat that required finding his way without guides and using Prague's complex trolleycar system with a very rudimentary knowledge of the language no doubt enriched in advance with suitable basic sentences required for the expedition. Delattre began his career in articulatory French phonetics with a strong commitment to practical applications to the teaching of French to American speakers. Following the publication of La durée des voyelles en français: Etudes expérimentales sur la durée des E d'un Français in 1939, based in part on his Michigan doctoral dissertation, he published more than forty articles on various aspects of French articulatory phonetics and pedagogical application primarily in the French Review, the journal of the American Association of Teachers of French. In fact, his regular annual contribution, presented with great clarity and without condescension, was for nearly a decade almost the only opportunity the relatively unsophisticated audience of that journal had to gain accurate information about matters linguistic. Delattre's approach to remedial phonetics was heavily influenced by the early articulatory phoneticians : Rousselot, Passy, and Jones, and it was not until relatively late (in Comparing the Phonetic Features of English, French, German and Spanish, for instance) that he worked explicitly with the notion of the phoneme. Rather than claiming to predict very specific points of phonological interference by the comparison

PIERRE DELATTRE

13

of the phonemic inventories of the native and target language, Delattre held that the inaccurate pronunciation of the target language stemmed from interference of the native articulatory set. Differences in the phonological systems of French and English — phonemic or allophonic, segmental or prosodie — according to him had their source in contrasts between the two languages in three articulatory modes: (1) tense vs. lax; (2) fronted vs. backed; (3) opening vs. closing. Since individual features of the target language were determined by the interplay of the articulatory modes, it was necessary for the learner to acquire the latter before he could master individual features. Thus, the first stage in Delattre's method of remedial phonetics was articulatory conditioning undertaken independently of the functional use of language or the manipulation of grammatical features and lexicon. Consider the following two practice sentences from Les Difficultés phonétiques du français, which, together with its companion volume Principes de phonétique française, gave Delattre's method wide dissemination: (1) Zéro ¡pour avoir osé ôter un mégot du métro; (2) Un homme éminent n'amène aucun ennui. Sentence (1) is designed to train the learner in producing steadystate vowels by having him repeat alternately a front unrounded and a back rounded vowel in an extended series of syllables rather than, for example, contrast steady-state French vowels and glided English near-equivalents : gué/gay; l'eau/Iow. Sentence (2) provides training in open vs. close syllabification, one of the manifestations of the contrast between an opening and a closing mode of articulation, as well as the nonnasalization of vowels preceding nasal consonants which is, within Delattre's scheme, simply a consequence of open syllabification. During the fifteen year period when contrastive analysis of phonemic inventories and minimal pair or near-equivalent contrast drills derived directly from these analyses became one of the basic marks of audiolingual oriented FL teaching, Delattre's method was in relative disfavor with many leading FL teaching methodologists — but not, it must be pointed out, with rank and file practitioners. But now that the linguistic and psychological underpinnings of contrastive analysis theory are seriously questioned, it becomes more apparent that Delattre's reliance on the notion of articulatory set has two advantages. First, it deals with entities directly linked to articulatory motions rather than abstract linguistic units, and is thus compatible with a weak version of linguistic interference theory anchored in physiology, namely that articulatory habits of the native language deeply ingrained after many years of practice will inhibit the acquisition of a different set of habits. Second, since an articulatory mode subsumes a variety of target language features, the remedial phonetician needs to concentrate on intra-systemic relations rather than comparing native and target language step-by-step, which is more in harmony with the structuralist view of language as "un système où tout se tient". One of the most troublesome aspects of French for the foreign learner is the alternation in the surface form of morphemes subsumed by traditional optional liaison and elision but compounded by dialect and stylistic variation. An important task for the applied French linguist is the elaboration of pedagogical norms whereby the ragged edges that the language shows in normal use by a complex and heterogeneous lin-

14

PIERRE DELATTRE

guistic community are streamlined to make the principles that underlie the variable behavior of its speakers more readily accessible to foreign learners. A pedagogical norm cannot, however, be an ad hoc simplification of linguistic facts but must express some significant generalization about them. Delattre's research was motivated by the search for generalizations that relate various aspects of a phonological system — physiological, acoustic, linguistic, and psychological — and it comes as no surprise that without having discussed the notion explicitly he formulated powerful pedagogical norms for French pronunciation and worked out practical procedures to apply them in the classroom. French shows partial complementation between members of the mid vowel pairs [e]/[£]» [0]/[®]> and [o]/[o], such that, except for contrasts of the type gué/guet, jeûne/ jeune and paume/pomme and considerable free variation in medial syllables in certain styles of the prestige Paris dialect, the high-mid member of each pair occurs in an open syllable and the low-mid member in a checked syllable. While this partial complementation had been observed previously, notably by Passy (1887), it was Delattre who formulated it most concisely as a variable rule termed the Loi de Position ("en syllable fermée, la voyelle tend à s'ouvrir; en syllabe ouverte, la voyelle tend à se fermer") which, while it did not sacrifice observational adequacy since it was not stated categorically, provided foreign learners with a simple basis for pronunciation. "Exceptions" to the Loi de Position were progressively introduced, for example the use of high-mid [0] and [o] before [z], so that the learner's pronunciation more closely approximated that of prestige speakers. Interestingly, Delattre also viewed the Loi de Position as expressing a natural tendency of French and considered pronunciations which constituted exceptions to it "une réaction savante et contre-nature (contre les tendances phonétiques du français) imposée par les grammairiens" (1948:22). In this instance his claims were not altogether correct since the use of [ε] in final open syllables and of [0] in medial open syllables is characteristic of working class Paris adolescents (Léon, 1972), and Delattre's assertions were perhaps influenced by his own pronunciation habits in which indeed [ε] seldom occurred in final open syllables. Nonetheless, they reflect his idea that the function of the phonetician was to transcend the collection of data and to search for the small set of great principles, revealed really by exceptional insight, that underlie speech and distinguish one language from another, and in this regard his attitude joins that of present-day generative phonologists. Delattre's involvement in FL pedagogy transcended remedial phonetics and led him to undertake one of the earliest and most notable comparison experiments. During the academic year 1944-45 he decided to test the Army Language Method assertion that an initial concentration on listening and speaking while denying reference to the written word was more efficient than the traditional immediate exposure to spelling. At the University of Oklahoma, he divided a beginning class into an experimental group exposed to an innovative treatment and a control group following the traditional eclectic approach. Both groups were taught by the same instructor so that the

PIERRE DELATTRE

15

dependent variable was limited to availability or non-availability of the written word and the use of materials and procedures compatible with these two approaches. Students in the experimental group demonstrated superior proficiency in auditory comprehension and speaking and slightly better ability in writing, but the long-range importance of the experiment lies not in the demonstration that an extended prereading period leads to more efficient acquisition of audio-lingual skills since we now know that even comparison experiments with considerably more elaborate research fail to isolate as many significant variables as they control. Rather, it showed nearly two decades before programmed self-instructional courses were launched that a significant responsibility for learning can be assumed by the students directly. Students had access to a phonograph and discs prepared daily by Delattre and practiced material presented previously in class. In this way contact with the spoken language was multiplied several fold, at least for assiduous students, and much of the drill was taken over by an electro-mechanical component which could supplement and, indeed, function independently of the classroom teacher. Another innovative feature of the experimental treatment that antedated programmed instruction was a graded presentation of phonological features to reduce initial articulatory difficulty to a minimum. The first sentence students practiced, La belle demoiselle qui passe là-bas est la voisine de Jeanne, artfully combined only five of the fifteen vowels of French and avoided such difficult articulations — for English speakers — as r, front rounded vowels, and unglided final [e], [o] and [u], and thus provided a facilitative framework within which the student acquired French prosodie habits of even-stressed syllabic rhythm and sustained statement intonation. While Delattre was a precursor of programmed instruction, he never would have accepted attempts to teach the spoken language, and particularly pronunciation, by means of total self-instruction. For him, language was literally SPEECH, and he would not have accepted the extensive preliminary training in auditory comprehension contained in all current programmed audio-lingual language courses, nor for that matter would he accept any FL program that did not aim at the attainment of functional bilingualism: ... since we understand the sounds of a new language not just by ear but by direct reference to articulatory gesture of which we have already acquired the correct HABIT ... it seems that the earlier we enter the production phase the better we can discriminate among the phonemes of a target language (1966:9). Delattre's appointment to the French Department of the University of Pennsylvania was a momentous occasion, not only in his personal career, but also in the history of Acoustic Phonetics, for it brought him into contact with the principal new technological advances that were to play important roles in this new interdisciplinary science. The Bell Telephone Laboratories had produced the sound spectrograph, which for the first time performed a running Fourier analysis of the acoustical wave form and provided a coherent visual representation of speech samples. From spectrograms, hypotheses could be made as to what aspects of the total speech signal convey phonemic information. However analysis constitutes only one half of the investigative

16

PIERRE DELATTRE

processes of experimental sciences; the hypotheses must be tested by synthesis. An additional research tool was clearly indicated, namely a device that would convert spectrograms, real or hand-painted, into sound. Delattre began visiting the Bell Telephone Laboratories regularly, at his own expense, in order to familiarize himself with spectrographic analysis. It was there that he learned that F. S. Cooper, of the Haskins Laboratories in New York, had developed a successful synthesizer. Actually, Cooper, a physicist, and his colleague, A. M. Liberman, a psychologist, had been concerned with developing a reading device for the blind, and had, from that project, become interested in discovering the acoustic cues for speech. Excited by the research possibilities offered by this equipment, Delattre began going regularly to New York to visit the Haskins Laboratories. " Vini, vidi vici!" might very well have been attributed to Delattre, for that is precisely what happened, and Cooper and Liberman soon asked him to become part of their research team. Together, they made the first major breakthroughs in acoustic phonetics, methodically discovering the principal perceptual cues for phonemes, and developing a 'motor' theory of speech perception. According to the latter, the acoustic level is not the fundamental one, but rather speech is conceived as a complex of articulatory (motor) patterns, made up of units of positions and movements. These workers claimed that speech recognition involves a proprioceptive feedback in which the acoustic signals are referred back to the articulations that are their genesis. If the acoustic level were the fundamental one, that is, if the phoneme were conceived as an acoustic rather than as an articulatory event, we would expect to find a one-to-one relation between the two. On the contrary, a simpler relationship appears to exist between perception and articulation than between per* ception and the acoustic stimulus. Indications to this effect are numerous and dramatic. A good example is the role of the burst in identifying the voiceless stops : The frequency of the burst for /k/ varies with that of the second formant of the associated vowel — it is relatively high for /ki/ and low for /ku/ —, and when a burst appropriate for /ki/ is paired with /u/ the result is perceived as /tu/. Thus a number of bursts at widely separated frequencies, but paired with appropriate vowels, will be perceived as the linguistically constant /k/ — it is noted that they are normally produced in all contexts with roughly the SAME articulation —, while a given burst paired with different vowels may be heard as DIFFERENT consonants. The same lack of correspondence is observed between formant transitions and perceived phonemes, and the "Locus Theory" represents an attempt to find constant values on the acoustic level for each place of articulation. Vowelconsonant transitions do not show a constant relation to consonant class either — second formant transitions for /d/, for example, may point up, down, or straight, depending on the identity of the contiguous vowel. However, an extrapolation of their curves leads to a single point where they intersect. For dentals, it is approximately at about 1800 HE, for labials it is at about 700 H z , and for velars with all but the back vowels, it is around 3000 H z . Those points, or 'loci', can thus be considered as corresponding to articulation. Although spectrograms of real speech show

PIERRE DELATTRE

17

considerable variation in this respect, the best results obtained with synthetic speech are those which follow this principle. As a result of this and other discoveries, the Haskins Laboratories soon became an international focal point for work in acoustic phonetics, and Delattre became recognized as an authority in this field. Once the major acoustic cues had been specified, he turned his attention to comparative phonetics. When he moved to the University of Colorado, he obtained an Office of Education grant to build a laboratory, conceived on the basis of an original threeway research technique, involving spectrograph«; analysis, speech synthesis, and cineradiography, which permitted relating the elements of the acoustic signal to the articulatory phenomena that they represent. It was in the latter that Delattre looked for the broad generalizations that would explain not only systemic relations at the synchronic level but also the direction of linguistic change. He had shown an early interest in diachronic studies and had published several articles on sound changes, for example "A contribution to the history of 'R grasseyé'" (1944) and "Stages of Old French phonetic changes observed in Modern Spanish" (1946). Typical of what might be considered his mature work in experimental phonetics was his study of nasalization published in one of his last articles, "Divergences entre nasalités vocaliques et consonantiques" (1970). Spectrographs analyses reveal two types of vocalic nasalization. In the first type Delattre labelled "nasalisation par annulation" (cancellation) and characteristic of the French nasal vowels [e],[ä], [œ], [δ], there is a decrease in the amplitude of the first formant while the second formant value remains comparable to that of corresponding oral vowels. In the second type, there is a spread of formant one so that, although its amplitude is not necessarily reduced, it is spread over more harmonics. This type of vocalic nasalization, termed "nasalisation par amortissement" (damping), is found in English vowels preceding nasal consonants as in Sam and thin and in the high and high-mid nasal vowels of Portuguese. X-ray film studies show that the two types of vowel nasalization correspond to two different articulatory configurations: nasalization by cancellation is produced not only by a lowering of the velum, which also occurs in the production of nasalization by damping, but by the adjustment of the volume of the pharyngeal cavity to that of the velie cavity formed in the upper pharynx above the velum. After noting that all four French nasal vowels have pharyngeal cavities of equal volume and show identical first formants, Delattre suggested that their diachronic evolution from sequences of high and high-mid vowels followed by a nasal consonant is explained by the fact that low and low-mid vowels have, pharyngeal cavities that produce maximal perceptible nasalization. He hypothesized that, in turn, this evolution is explained by the loss of implosive consonants that changed the syllabic structure of French between the Middle Ages and the 16th century (Martinet 1965) and led to the phonemicization of nasal yowels : "une fois libérées de la tyrannie des consonnes nasales subséquentes, les voyelles nasales françaises ont eu libre cours pour évoluer vers des positions articulatoires plus favorables à la distinction orale/nasale" (1970:70). Delattre never drew a sharp distinction between pure and applied research, and

18

PIERRE DELATTRE

he did not leave the application of his use of a variety of experimental techniques in phonetics to less skilled and less knowledgeable interpreters. Indeed, much of his time and energy during his mature period was devoted to the improvement of FL teaching. In the preface to Comparing the Phonetic Features of English, French, German and Spanish he states: It occurred to us that the same techniques as, for instance, that which was used at Haskins to study the acoustic correlates of speech perception could be applied to the study of foreign accents described in terms of phonetic features which distinguish one language from another. Thus the latest electronic techniques of speech analysis and our experience in artificial speech synthesis should serve a practical purpose — better teaching of foreign languages. Pierre Delattre was a generous and affable host and displayed a charming and sparkling Gallic wit that one would hardly suspect in someone with such austere antecedents, and colleagues from the humanities who knew of his strong attachment to music and literature had difficulty imagining him as the white-smocked scientist surrounded by complex machines. But Delattre was and considered himself a humanist and he had always been enchanted by the esthetic qualities of the principal object of his study, French sounds: French ... reveals a rich and varied system of vowels and consonants in which frontal resonance dominates, open syllables that are proud of their vowels, oxytonic accentuation that emphasizes the last syllables of sense groups not by heavy marks of intensity but by restrained increases in length, an intonation capable of contrasting the lightest shade of syntactic meaning, a rhythm of equal syllables which recalls a string of pearls ... (1966b:6). André Malécot University of California Santa Barbara Albert Valdman Indiana University Bloomington

REFERENCES Delattre, Pierre 1939 La durée des voyelles en français: Etude expérimentale sur la durée des E d'un Français (Paris : d'Artrey). 1944 "A contribution to the history of 'R grasseyé", Modern Language Notes 14: 562-64. 1946a Principes de phonétique française à l'usage des étudiants anglo-américains (Middlebury, Vermont: The College Store). 1946b "Stages of Old French phonetic changes observed in Modern Spanish", PMLA 61: 7-41. 1947 "A technique of aural-oral approach, report on a University of Oklahoma experiment in teaching French", French Review 20: 238-50; 311-24. 1948a Les difficultés phonétiques du français (Middlebury, Vermont: The College Store). 1948b "Un triangle acoustique des voyelles orales du français", French Review 21: 477-84.

PIERRE DELATTRE

19

1951 "The physiological interpretation of sound spectrograms", PMLA 66: 864-75. 1955 "Les attributs acoustiques de la nasalité vocalique et consonantique", Studia Linguistica 8: 103-09. 1958 "Les indices acoustiques de la parole: premier rapport", Phonetica 2: 108-18; 226-51. 1965a "La nasalité vocalique en français et en anglais", French Review 39: 92-109. 1965b Comparing the Phonetic Features of English, German, Spanish and French (Heidelberg: Julius Groos Verlag and Philadelphia: Chilton Books). 1966 Studies in French and Comparative Phonetics (The Hague: Mouton). 1968 "La radiographie des voyelles françaises et sa corrélation acoustique", French Review 42: 48-65. 1970 "Divergences entre nasalités vocaliques et consonantiques", Word 24 : 64-72. (Linguistic studies presented to André Martinet on the occasion of his sixtieth birthday, II). Léon, Pierre, 1972 "Etude de la prononciation de e accentué chez un groupe de jeunes Parisiens", in Papers in Linguistics and Phonetics to the Memory of Pierre Delattre (The Hague: Mouton). Liberman, A.M. and F.S. Cooper, 1972 "In search of distinctive features", in Papers in Linguistics and Phonetics to the Memory of Pierre Delattre (The Hague: Mouton), 329-38. Martinet, André, 1965 "Les voyelles nasales du français", Linguistique 1:117-22.

PUBLICATIONS OF PIERRE DELATTRE

BOOKS

1. La durée des voyelles en français: Etude expérimentale sur la durée des E d'un Français (Paris: d'Artrey, 1939). 2. Principes de phonétique française à Vusage des étudiants anglo-saxons (Middlebury, Vermont: The College Store, 1946). 3. An Introduction to French Speech Habits (with a 10" L.P. record) (New York: Henry Holt and Company, Inc., 1947). 4. Les Difficultés phonétiques dufrançais (Middlebury, Vermont: The College Store, 1948). 5. Advanced Training in French Pronunciation (with a 12" L.P. record) (Middlebury, Vermont: The College Store, 1949). 6. Premier manuel: Grammaire et civilisation françaises (with a 12" L.P. record, with D.W. Alden) (New York: Appleton-Century-Crofts, 1954). 7. Phonetic Readings of Songs and Arias (with B. Coffin, R. Errolle, and W. Singer) (Boulder: Pruett Press, 1964). 8. Comparing the Phonetic Features of English^ German, Spanish and French (Heidelberg: Julius Groos Verlag or Philadelphia: Chilton Books, 1965). 9. Studies in French and Comparative Phonetics (The Hague: Mouton, 1966). 10. Word-by-Word Translations of Songs and Arias (with B. Coffin and W. Singer) (New York: The Scarecrow Press, 1966).

ARTICLES

1. "Durée consciente et durée inconsciente", French Review 12 (1938) 49-50. 2. "Investigating vowel duration in French", Papers of the Michigan Academy of Science, Arts, and Letters 23 (1938) 517-24. 3. "L'accent final en français: accent d'intensité, accent de hauteur, accent de durée", French Review 12-(1938) 3-7.

22

PUBLICATIONS OF PIERRE DELATTRE

4. "Remarques sur l'enseignement de la durée des voyelles françaises", Le Maître Phonétique 3 (1938) 66-67. 5. "Ve muet dans la coupe syllabique", Le Français Moderne 7 (1939) 154-58. 6. "Durée vocalique et consonnes subséquentes", Le Maître Phonétique 3 (1939) 41-44. 7. "Accent de mot et accent de groupe", French Review 13 (1939) 141-46. 8. "Le mot est-il une entité phonétique en français", Le Français Moderne 8 (1940) 47-56. 9. "Anticipation in the sequence: vowel and consonant-group", French Review 8 (1940) 314-20. 10. "Tendances de coupe syllabique en français", Publications of the Modern Language Association of America (PMLA) 55 (1940) 579-95. 11. "La force d'articulation consonantique en français", French Review 14 (1941) 220-32. 12. "Paul Passy, or the return to elementáis", Books Abroad 16 (1942) 241-46. 13. "Travaux d'étudiants au laboratoire de phonétique expérimentale de l'école française de Middlebury", French Review 16 (1943) 504-10. 14. "La leçon de phonétique de Maurice Chevalier", French Review 17 (1943) 99104. 15. "L'aperture et la syllabation phonétique", French Review 17 (1944) 281-85. 16. "A foreigner views basic English", Books Abroad 18 (1944) 115-20. 17. "La syllabation ouverte par la méthode compensatrice", French Review 17 (1944) 371-76. 18. "Que faut-il penser de l'anglais basique", Le Recueil 14 (1944) 1-9. 19. "International phonetics, a founder and pioneer — Paul Passy", The French Forum 7 (1944) 289-90. 20. "Vers la méthode phonétique intégrale pour débutants", French Review 18 (1944) 109-15. 21. "A contribution to the history of 'R grasseyé'", Modern Language Notes 14 (1944) 562-64. 22. "Prononciation graphique et prononciation phonétique, I. Les Consonnes", French Review 18 (1945) 219-226. 23. "Prononciation graphique et prononciation phonétique, II. Les Voyelles", French Review 18 (1945) 285-96. 24. "Spanish is a phonetic language", Hispania 28 (1945) 511-17. 25. "Tout va très bien, Madame la Marquise, transcription commentée", French Review 19 (1945) 125-29. 26. "Stages of Old French phonetic changes observed in Modern Spanish", Publications of the Modern Language Association of America (PMLA) 61 (1946) 7-41. 27. "L'enseignement de YR français", French Review 19 (1946), 428-33. 28. "Pour imiter un disque de français parlé", French Review 20 (1946) 43-48. 29. "A technique of aural-oral approach, report on a University of Oklahoma

PUBLICATIONS OF PIERRE DELATTRE

23

experiment in teaching French", French Review 20 (1947) 238-50 and 311-24. 30. "La leçon de phonétique de Charles Boyer", French Review 20 (1947) 396-98. 31. "Broad transcription for typing in American English", Le Maître Phonétique 3 (1947) 27-28. 32. "La liaison en français, tendances et classification", French Review 21 (1947) 148-57. 33. "La leçon de phonétique de Charles Trenet", French Review 21 (1948) 249-59. 34. "Phonetics in beginning language study", The Modern Language Journal 32 (1948) 373-77. 35. "Un triangle acoustique des voyelles orales du français", French Review 21 (1948) 477-84. 36. "La leçon de phonétique de Farrebique", French Review 22 (1949) 323-28. 37. "Le jeu de Y E instable de monosyllabe initial en français : I. Un seul monosyllabe", French Review 22 (1949) 455-59. 38. "Le jeu de Y E instable de monosyllabe initial en français : II. Deux monosyllabes", French Review 23 (1949) 43-47. 39. "Le surcomposé réfléchi en subordonnée temporelle", Le Français Moderne 18 (1950) 95-108. 40. "The use of the pattern playback in studies of vowel color by synthesis", Journal of the Acoustical Society of America 22 (1950) 678. 41. "Le jeu de YE instable intérieur en français", French Review 24 (1951) 341-51. 42. "The physiological interpretation of sound spectrograms", Publications of the Modern Language Association of America (PMLA) 66 (1951) 864-75. 43. "Des progrès actuels de la phonétique", French Review 25 (1951) 42-46. 44. "Voyelles synthétiques à deux formants et voyelles cardinales" (with A. M. Liberman and F. S. Cooper), Le Maître Phonétique 96 (1951) 30-36. 45. "Some suggestions for language teaching methods arising from research on the acoustic analysis and synthesis of speech" (with F. S. Cooper and A. M. Liberman), Report of the Third Annual Round Table Meeting on Linguistics and Language Teaching, Monograph Series 2 (1952) 31-47. 46. "The role of selected stimulus-variables in the perception of the unvoiced stop consonants" (with A. M. Liberman and F. S. Cooper), The American Journal of Psychology 65 (1952) 497-516. 47. "Some experiments on the perception of synthetic speech sounds" (with F. S. Cooper, A. M. Liberman, J. M. Borst, and L. J. Gerstman), The Journal of the Acoustical Society of America 24 (1952) 597-606. 48. "An experimental study of the acoustic determinants of vowel color; observations on one- and two-formant vowels synthesized from spectrographic patterns" (with A. M. Liberman, F. S. Cooper, and L. J. Gerstman), Word 8 (1952) 195210.

49. "Les modes phonétiques du français", French Review 27 (1953) 59-63. 50. "The role of consonant-vowel transitions in the perception of the stop and nasal

24

51. 52.

53. 54. 55. 56.

57. 58. 59.

60.

61. 62. 63.

64. 65.

66. 67. 68.

PUBLICATIONS OF PIERRF DELATTRE

consonants" (with A. M. Liberman, F. S. Cooper and L. J. Gerstman), Psychological Monographs 68 (1954) 1-13. "Lecture et prononciation en seconde année de français", French Review 27 (1954) 472-74. "Acoustic loci and transitional cues for consonants", (with A. M. Liberman and F. S. Cooper), The Journal of the Acoustical Society of America 27 (1955) 769-73. "Les attributs acoustiques de la nasalità vocalique et consonantique", Studia Linguistica 8 (1955) 103-109. "... Sur les origines de la prononciation française", Le Bayou 62 (1955) 316-18. "Les facteurs de la liaison facultative en français", French Review 29 (1955) 42-49. "Tempo of frequency change as a cue for distinguishing classes of speech sounds" (with A. M. Liberman, L. J. Gerstman, and F. S. Cooper), Journal of Experimental Psychology 52 (1956) 127-37. "La fréquence des liaisons facultatives en français", French Review 30 (1956) 48-54. "The Celtic element in the French language", Bulletin of the Colorado Congress of Modern Language Teachers 9 (1956) 4-6. "Speech Synthesis as a research technique" (with F. S. Cooper, A. M. Liberman, and L. J. Gerstman), Proceedings of the Vllth International Congress of Linguists (1952), London, International University Booksellers Ltd., (1956) 54561 plus 9 figures. "Acoustic cues for the perception of initial /w, j, r, 1/ in English" (with J. D. O'Connor, L.J. Gerstman, A.M. Liberman, and F. S. Cooper), Word 13 (1957) 24-43. "La question des deux Ά' en français", French Review 31 (1957) 141-48. "A propos des corrélatifs acoustiques de la distinction entre voyelle et consonne", Studia Linguistica 11 (1957) 78-83. "Effect of third-formant transitions on the perception of the voiced stop consonants" (with K. S. Harris, H. S. Hoifman, A. M. Liberman, and F. S. Cooper), The Journal of the Acoustical Society of America 30 (1958) 122-26. "Unreleased velar plosives after back-rounded vowels", The Journal of the Acoustical Society of America 30 (1958) 581-82. "Some cues for the distinction between voiced and voiceless stops in initial position" (with A. M. Liberman and F. S. Cooper), Language and Speech 1 (1958) 153-67. "Vowel color and voice quality — an acoustic and articulatory comparison", The Bulletin of the National Association of Teachers of Singing 15 (1958) 4-7. "Les indices acoustiques de la parole: premier rapport", Phonetica 2 (1958) 108-18 and 226-51. "Acoustic cues in speech", Haskins Laboratories Report (1958) 2-39.

PUBLICATIONS OF PIERRE DELATTRE

25

69. "Rapports entre la durée vocalique, le timbre et la structure syllabique en français", French Review 32 (1959) 547-52. 70. "Minimal rules for synthesizing speech" (with A. M. Liberman, F. Ingemann, L. Lisker, and F. S. Cooper), The Journal of the Acoustical Society of America 31 (1959) 1490-99. 71. "Sommes-nous préparés pour l'enseignement de la langue?" French Review 33 (1960) 483-90. 72. "Testing audio equipment by ear", Audiovisual Instruction 5 (1960) 156. 73. "Un cours d'exercices structuraux et de linguistique appliquée", French Review 33 (1960) 591-603. 74. "Testing students' progress in the language laboratory", International Journal of American Linguistics 26 (1960) 77-93. 75. "Testing students' progress in the language laboratory", Automated Teaching Bulletin 1 (1960) 21-31. 76. "Comment tester la facilité de parole dans un laboratoire de langue", Le Français dans le Monde 1 (1961) 36-38. 77. "La leçon d'intonation de Simone de Beauvoir, étude d'intonation déclarative comparée", French Review 35 (1961) 59-67. 78. "Testing the oral production of language students", The College Language Laboratory 5 (1961) 25-43. 79. "Un cours d'exercices structuraux et de linguistique appliquée", NDEA Anthology, Boston, Heath (1961) 59-72. 80. "La fréquence des liaisons facultatives en français", NDEA Anthology, Boston, Heath (1961) 90-97. 81. "A comparative study of declarative intonation in American English and Spanish" (with C. Olsen and E. Poenack), Hispania 45 (1962) 233-41. 82. "An experimental study of the effect of pitch on the intelligibility of vowels" (with J. Howie), The Bulletin of the National Association of Teachers of Singing 18 (1962) 6-9. 83. "Some factors of vowel duration and their cross-linguistic validity", The Journal of the Acoustical Society of America 34 (1962) 1141-42. 84. "Le français et les laboratoires de langue", Esprit 30 (1962) 595-604. 85. "Formant transitions and the perception of consonants", Haskins Laboratories Reports 30 (1962) 1-23. 86. "Le jeu des transitions de formants et la perception des consonnes", Proceedings of the Fourth International Congress of Phonetic Sciences, Helsinki 1961, The Hague, Mouton (1962) 407-17. 87. "Une technique 'audio-linguale' d'initiation au français", Le Français dans le Monde! (1962) 15-17. 88. "Quality in tape recording and voicing", International Journal of American Linguistics 29 (1963) 55-59. , 89. "Voyelles diphtonguées et voyelles pures", French Review 37 (1963) 64-76.

26

PUBLICATIONS OF PIERRE DELATTRE

90. "Research techniques for phonetic comparison of languages", International Review of Applied Linguistics in Language Teaching (IRAL) 1 (1963) 85-97. 91. "Comparing the prosodie features of English, German, Spanish and French", International Review of Applied Linguistics in Language Teaching (IRAL) 1 (1963) 193-210. 92. "Some experiments on the perception of synthetic speech sounds" (with F. S. Cooper, A. M. Liberman, J. M. Borst, and L. J. Gerstman), New Selected Works of Physics #28, Tokyo, Japanese Physics Association (1963) 32-41. 93. "Acoustic loci and transitional cues for consonants" (with A. M. Liberman and F. S. Cooper), New Selected Works of Physics #28, Tokyo, Japanese Physics Association (1963) 42-46. 94. "Minimal rules for synthesizing speech" (with A. M. Liberman, F. Ingemann, L. Lisker, and F. S. Cooper), New Selected Works of Physics #28, Tokyo, Japanese Physics Association (1963) 47-56. 95. "French prepositional patterns in linking a verb to an infinitive object", The Modern Language Journal 48 (1964) 29-35. 96. "Classifying speech sounds by their source", In Honour of Daniel Jones, edited by David Abercrombie, London, Longmans, Green and Company (1%4) 4652. 97. "French phonetics for the singer", Phonetic Readings of Songs and Arias, Boulder, Pruett Press (1964) 252-54. 98. "Formant transitions and loci as acoustic correlates of place of articulation in American fricatives" (with A. M. Liberman and F. S. Cooper), Studia Linguistica 16 (1964) 104-21. 99. "Comparing the vocalic features of English, German, Spanish and French", International Review of Applied Linguistics for Language Teaching (IRAL) 2 (1964) 71-97. 100. "Le son du français, impressions et réalité", Le Français dans le Monde 4 (1964) 13-16. 101. "Le jeu des prépositions dans l'enchaînement des verbes en français", French Review 38 (1964) 67-81. 102. "German phonetics between English and French", Linguistics 8 (1964) 43-55. 103. "Comparing the consonantal features of English, German, Spanish and French", International Review of Applied Linguistics for Language Teaching (IRAL) 2 (1964) 155-203. 104. "La synthèse acoustique de la parole", Bulletin de la Société des Professeurs Français en Amérique 18 (1964) 13-26. 105. "De la hiérarchie des indices acoustiques pour la perception de la parole", Proceedings of the Fifth International Congress of Phonetic Sciences, Münster 1964, edited by E. Zwirner and W. Bethge, Basel, S. Karger (1964) 244-51. 106. "Change as a correlate of the vowel-consonant distinction", Studia Linguistica 18 (1965) 12-25.

PUBLICATIONS OF PIERRE DELATTRE

27

107. "Some characteristics of German intonation for the expression of continuation and finality" (with E. Poenack and C. Olsen), Phonetica 13 (1965) 134-61. 108. "La nasalità vocalique en français et en anglais", French Review 39 (1965) 92109. 109. "Some sound changes in the light of spectrograph«; analysis and synthesis", Omagiu Lui Alexandru Rosetti, Bucharest, Editura Academiei Republicii Socialiste România (1965) 163-65. 110. "Oklahoma revisited", Advances in Teaching of Modern Languages, edited by G. Mathieu, Oxford and New York, Pergamon Press (1966) 1-10. 111. "Les attributs physiques de la parole et l'esthétique du français", Revue d'Esthétique 3-4 (1966) 250-54, 112. "La notion de structure et son utilité", Le Français dans le Monde 41 (1966) 7-11. 113. "Les dix intonations de base du français", French Review 50 (1966) 1-15. 114. "A comparison of syllable length conditioning among languages", International Review of Applied Linguistics for Language Teaching (IRAL) 4 (1966) 183-98. 115. "Acoustic or articulatory invariance?" Glossa 1 (1967) 1-25. 116. "La nuance de sens par l'intonation", French Review 41 (1967) 326-40. 117. "Language learning and linguistic interference", Dimension: Languages 2 (1967) 27-33. 118. "Principles of language instruction at the college level", Dimension: Languages 2 (1967) 120-26. 119. "Tempo of frequency change as a cue for distinguishing classes of speech sounds" (with A. M. Liberman, L. J. Gerstman, and F. S. Cooper), Readings in Acoustic Phonetics, edited by Use Lehiste, Cambridge, M.I.T. Press (1967) 159-69. 120. "Some experiments on the perception of synthetic speech sounds" (with F. S. Cooper, A. M. Liberman, J. M. Borst, and L. J. Gerstman), Readings in Acoustic Phonetics, edited by Ilse Lehiste, Cambridge, M.I.T. Press (1967) 273-82. 121. "Acoustic loci and transitional cues for consonants" (with A. M. Liberman and F. S. Cooper), Readings in Acoustic Phonetics, edited by Ilse Lehiste, Cambridge, M.I.T. Press (1967) 283-87. 122. "Minimal rules for synthesizing speech" (with A. M. Liberman, F. Ingemann, L. Lisker, and F. S. Cooper), Readings in Acoustic Phonetics, edited by Ilse Lehiste, Cambridge, M.I.T. Press (1967) 333-42. 123. "The role of duration in the identification of French nasal vowels" (with M. Monnot), International Review of Applied Linguistics in Language Teaching (IRAL) 6 (1968) 267-88. 124. "La radiographie des voyelles françaises et sa corrélation acoustique", French Review 42 (1968) 48-65. 125. "Duration as a cue to the tense/lax distinction in German unstressed vowels" (with M. Hohenberg), International Review of Applied Linguistics in Language Teaching (IRAL) 6 (1968) 367-90. 126. "From acoustic cues to distinctive features", Phonetica 18 (1968) 198-230.

28

PUBLICATIONS OF PIERRE DELATTRE

127. "A dialect study of American R's by X-ray motion picture" (with Donald C. Freeman), Linguistics 44 (1968) 29-68. 128. "Syntax and intonation : a study in disagreement", Study of Sounds 14 (1969) 21-40. 129. "Syllabic features and phonic impression in English, German, French and Spanish" (with Carroll Olsen), Lingua 22 (1969) 160-75. 130. "L'intonation par les oppositions", Le Français dans le Monde 64 (1969) 6-13. 131. "Two notes on Semitic laryngeals in East Gurage" (by Robert Hetzron with contribution by P. Delattre), Phonetica 19 (1969) 69-81. 132. "L'/R/ parisien et autres sons du pharynx", French Review 43 (1969) 5-22. 133. "Syntax and intonation: a study in disagreement", Modern Language Journal 54 (1970) 3-9. 134. "A cross-language study of the i/j distinction" (with Marie-Thérèse Delyfer), English as a Second Language — Current Issues (Robert C. Lugton, Ed.) [Language and the Teacher: A Series in Applied Linguistics # 6 ] . Philadelphia: Center for Curriculum Development, Inc., (1970) 63-87. 135. "Divergences entre nasalité vocalique et consonantique", Linguistic Studies Presented to André Martinet, Part II (Indo-European Linguistics) [Word 24 (1968) 64-72; published in 1970]. 136. "Coarticulation and the locus theory", Studia Linguistica 23 (1969) 1-26. 137. "La théorie celtique et les substrats", Romance Philology 23 (1970) 480-91. 138. "An acoustic study and articulatory study of vowel reduction in four languages", International Review of Applied Linguistics for Language Teaching (IRAL) 7 (1969) 295-325. 139. "First-formant transitions as a cue to place of articulation", Journal of the Acoustical Society of America 46 (1969) 110A. 140. "Rapports entre la physiologie et la chronologie de la nasalité distinctive", Proceedings of the X International Congress of Linguists, Bucharest, IV (1970) 221-27. 141. "Pharyngeal features in the consonants of Arabic, German, Spanish, French and American English", Phonetica 23 (1971) 129-55. 142. "Consonant gemination in four languages: an acoustic, perceptual, and radiographic study", International Review of Applied Linguisticsfor Language Teaching (IRAL) 9 (1971) 31-52. REVIEWS

1. George Gougenheim, Système grammatical de la langue française, French Review 13 (1940) 236-38. 2. Henri Malhiac, Analyse et enregistrement de la voix parlée et chantée, Language 30 (1954) 597-98. 3. F. D. Eddy, French for Children, Bulletin of the Colorado Congress of Foreign Language Teachers 9 (1958) 13-14.

PUBLICATIONS OF PIERRE DELATTRE

29

4. L. Kaiser, Manual of Phonetics, Romance Philology 13 (1959) 80-83. 5. Jean-Claude Lafon, Message et phonétique: Introduction à l'étude acoustique et physiologique du phonème, The Journal of the Acoustical Society of America 34 (1962) 995. 6. William G. Moulton, The Sounds of English and German, International Review of Applied Linguistics in Language Teaching (IRAL) 1 (1963) 142-45. 7. Diagnosticai Procedures of Voice and Hearing Dysphonia, Reports of the 12th Congress of Logopedics and Phoniatrics, 1962, The Journal of the Acoustical Society of America 35 (1963) 1069-70. 8. Alf Lombard, Le Rôle des semi-voyelles et leur concurrence avec les voyelles correspondantes dans la prononciation parisienne, Romance Philology 18 (1965) 462-64. 9. Georges Straka, Travaux de linguistique et de littérature publiés par le Centre de philologie et de littératures romanes de l'Université de Strasbourg, I, Romance Philology 19 (1965) 73-81. 10. George Scherer and Michael Wertheimer, A Psycholinguistic Experiment in Foreign-Language Teaching, Harvard Educational Review 35 (1965) 533-36. 11. Bertil Malmberg, Phonetics, Romance Philology 20 (1967) 246-50. 12. A. Moles and B. Vallancien, Phonétique et phonation, The Journal of the Acoustical Society of America 42 (1967) 1107. 13. Jean-Denis Gendron, Tendences phonétiques du français parlé au Canada, Language 44 (1968) 852-55. 14. W. Zwanenburg, Recherches sur la prosodie de la phrase française, Romance Philology 22 (1968) 209-12. 15. G. Straka, Travaux de linguistique et de littérature I : II : Linguistique et philologie, Romance Philology 23 (1970) 92-97. 16. David Abercrombie, Elements of General Phonetics, General Linguistics 9 (1969) 118-22. 17. Boudreault, Rythme et mélodie de la phrase parlée en France et au Québec, Romance Philology 25 (1971) 121-23. 18. Péla Simon, Les Consonnes françaises, Romance Philology 24 (1970) 332-37. 19. Y. Lebrun, Anatomie et physiologie de Γ appareil phonatoire, Phonetica 21 (1970) 126-28. BIBLIOGRAPHY

1. A critical bibliography of phonetics (with S. Belasco, J.-P. Vinay, and H. E. Collins) in American Speech 33 (1958), 63-69. 2. A critical bibliography of phonetics (with S. Belasco and J.-P. Vinay) in American Speech 33 (1958), 212-22. 3. A critical bibliography of phonetics (with S. Belasco and H. E. Collins) in American Speech 34 (1959), 55-63.

30

PUBLICATIONS OF PIERRE DELATTRE

4. A critical bibliography of phonetics American Speech 34 (1959), 214-21. 5. A critical bibliography of phonetics American Speech 35 (1960), 66-72. 6. A critical bibliography of phonetics American Speech 35 (1960), 226-35. 7. A critical bibliography of phonetics American Speech 36 (1961), 69-77. 8. A critical bibliography of phonetics American Speech 36 (1961), 215-23. 9. A critical bibliography of phonetics American Speech 37 (1962), 62-70. 10. A critical bibliography of phonetics American Speech 37 (1962), 222-29. 11. A critical bibliography of phonetics American Speech 38 (1963), 66-75. 12. A critical bibliography of phonetics American Speech 39 (1964), 64-74. 13. A critical bibliography of phonetics American Speech 39 (1964), 226-34. 14. A critical bibliography of phonetics American Speech 40 (1965), 68-73. 15. A critical bibliography of phonetics American Speech 40 (1965), 218-23. 16. A critical bibliography of phonetics American Speech 41 (1966), 225-34.

(with S. Belasco and H. E. Collins) in (with S. Belasco and A. Valdman) in (with S. Belasco and A. Valdman) in (with S. Belasco and A. Valdman) in (with S. Belasco and A. Valdman) in (with S. Belasco and A. Valdman) in (with S. Belasco and A. Valdman) in (with S. Belasco and A. Valdman) in (with S. Belasco and A. Valdman) in (with S. Belasco and A. Valdman) in (with S. Belasco and A. Valdman) in (with S. Belasco and A. Valdman) in (with S. Belasco and A. Valdman) in

RECORDS

1. 1-10* L.P. record for Introduction to French Speech Habits (New York: Henry Holt and Company, Inc., 1947). 2. 1-12" L.P. record of the complete material in Advanced Training in French Pronunciation (selections from Les Difficultés Phonétiques du Français) (Middlebury, Vermont: The College Store, 1949). 3. 1-12" L.P. record for Premier Manuel: Grammaire et Civilisation Françaises (New York: Appleton-Century-Crofts, 1954).

ARTHUR S. ABRAMSON

TONAL EXPERIMENTS WITH WHISPERED THAI1

In so-called tonal languages it is the general consensus among linguists and phoneticians that the feature used to distinguish most, if not all, of the tonal phonemes is pitch variation. The question arises as to what happens in whispered utterances of such a language, where there is no vocal fold vibration to produce a fundamental frequency that varies according to the levels or contours prescribed by the phonological system. In some languages, such as Swedish, tonal oppositions are restricted to certain points in the utterance and may in fact be conditioned syntactically to some extent; the question of the preservation of tonal oppositions is more intriguing in a language like Thai, in which every syllable has a tone as part of its phonemic make-up. This problem has interested a number of phoneticians working on a variety of languages. In an investigation of Mandarin Chinese, Charles Boardman Miller (1934) found that its four tonal phonemes are readily identified in whispered speech. He concluded that this was done through extensive help from the context as well as changes in energy indicating variations in pitch. Miller's experimental design, however, apparently provided no tests in which tone itself was the only variable. Panconcelli-Calzia (1955) maintains that in tonal languages context makes comprehension of longer whispered utterances possible, but he just about rules out comprehension of isolated words. Where others have claimed that whispered tones are indeed audible,4 he advances two thoughts: either they were not dealing with a genuinely voiceless whisper,8 or the semantic function of tones has been exaggerated. In a rejoinder, Giet (1956) reaffirms his stand that tonal distinctions are maintained in 1

This work was inspired by a long conversation with Pierre Delattre who was at the time looking at other aspects of whispered speech (Delattre, Liberman and Cooper 1959). Early versions were presented as oral papers at the Seventy-Third Annual meeting of the Modern Language Association of America, December, 1958 and the Fifty-Eighth Meeting of the Acoustical Society of America, October 1959 (Abramson, 1959). • E.g. Giet, 1950: p. 95. ' This is indeed something to be guarded against in such discussions. In true whisper the glottis may be somewhat narrowed but the folds do not pulsate (von Essen, 1962 : 35). In stage whisper, although the cartilaginous glottis is open to allow turbulent air through, the membranous folds will be sufficiently approximated to allow for the breathy phonation known as murmur (Zemlin, 1964: 165); this quasi-periodicity, albeit mixed with noise, can of course carry tonal information.

32

ARTHUR S. ABRAMSON

whispering and rendered perceptible, though less clearly so, through the substitution of other phonic features for pitch. The two substitutes in REAL Chinese whispering, he says, are changes in vowel color as well as increases in air flow for high tones and decreases in air flow for low tones. Addressing himself to this controversy, MeyerEppler (1957) supports Giet with spectrographs evidence of upward shifts in some vocalic formant frequencies for higher pitch, as well as an increase in intensity accompanied by noisy components in the gaps of the higher spectral regions for whispered German.4 Further work on tonal distinctions in whispered Mandarin was done by Wise and Chang (1957), who found that in tests with paired utterances minimally distinguished by tone, listeners were able to identify no more than 62% of the critical words. Kloster Jensen (1958), however, obtained somewhat higher recognition scores for Mandarin; they ranged from 73 % to 85 %, so he concluded that phonemic tones are reflected somehow in whispered speech.8 Two rather recent studies show on the one hand very little tonal information transmitted for whispered Vietnamese (Miller, 1961) and, on the other hand, considerable information on the word accents of whispered Swedish (Segerbäck, 1966). The foregoing claims and counterclaims, as well as the mixed experimental results, made it seem desirable to tackle the problem of the perception of phonemic tones in whispered speech with yet another language. The Thai language appears to be a good choice for this purpose because it has been clearly established (Abramson, 1962) that pitch movements furnish the dominant cues for the identification of the phonemic tones. Standard Thai or Siamese is the national language of Thailand and the regional dialect of the central plains including Bangkok. It is usually said to have five tones: middle, low, falling, high and rising. Spectrographic measurements of fundamental frequencies show that the mid tone starts near the middle of the speaker's voice range and remains level; if it occurs before a pause, it drops slightly at the end. The low tone starts just below the middle of the voice range, drops gradually and levels off somewhat above the bottom of the range. The falling tone starts rather high and drops rapidly to the bottom of the range. The high tone starts above the middle and rises slowly; before a pause, in certain phonetic environments, it drops slightly toward the end and shows concomitant laryngeal constriction with irregular pulsing. The rising tone starts quite low and rises rapidly to the top of the voice range.® The plan of this study was to see whether in fact Thai tones could be identified in whispered speech, and then, using equipment not available to previous investigators, to see whether such information as is transmitted is also available in normally phonated speech. After all, one might argue that the mixed results from Mandarin * Meyer-Eppler does not take German to be a tonal language but asserts that the prosodie distinction between a question and a statement is analogous. 5 Kloster Jensen's study included similar experiments with Norwegian, Slovenian and Swedish. • For more details see Abramson, 1962, especially Tables 3.2-3.3 and Figures 3.3-3.6. Hese statements are for citation forms; a full aUophonic description would have to take at least tonal environment and sentence intonation into account.

TONAL EXPERIMENTS WITH WHISPERED THAI

33

suggest that some speakers go through soeciai maneuvers to compensate for the missing pitch information, while^others simply supply the usual instructions to their speech production mechanisms minus fundamental frequency control. The most severe test of the phonological distinctiveness of tonal features would seem to be one in the context-free condition of isolated monosyllabic words. Before proceeding to whispered speech then, it was necessary to establish that the tones could be distinguished in phonated speech without any help from verbal context. Perception tests were prepared for each of four sets of tonally differentiated words in which each word was pronounced five times by a male native speaker of Thai. All the stimuli were presented in a random order to eleven native speakers, including the informant. Two tests gave five choices, one gave four, and the last gave three. Most of the subjects scored 100% on all the tests. There were a few scattered errors. Although these results indicate that all five tones are readily identified in isolation, R. B. Noss (1954, section 1.1.2) claims that the mid and low tones are not distinguishable in isolation but require an environment where relative pitch criteria are available. In a recent private communication, Noss has clarified this point by describing perception tests of his that do indeed strongly suggest that the tonal opposition in question is somewhat unstable in isolation. The informant used in the present study,7 he goes on to say, must have produced optimal maximally differentiated contours that the listeners had no difficulty in identifying. To reconcile the apparent conflict, it may be necessary to view the distinction between the mid and low tones as an unstable one, or at least a facultative one, in isolation. This is obviously an important observation to take into account when considering the identifiability of whispered tones in isolated words. The sets of words so well identified in phonated speech were once again recorded five times each by the same informant in a whisper. Great care was taken to insure that the speaker used true whisper. Neither auditorily nor spectrographically could any laryngeal pulses be detected. The recordings were randomized into test tapes and played to the eleven subjects used for the base line test; for two of the four tests, only eight of the subjects were available. The results are given in the form of confusion matrices in Tables 1 through 4. The words and brief glosses are given in each table. In Table 1, the words are also labelled as to tone to facilitate the reading of the tonal symbols in the rest of the tables. The convention of using double vowel symbols to represent distinctive length is followed here. The mean recognition scores of Tables 1 through 4 show a sharp drop in identifiability of whispered isolated words as compared with normally phonated words. The individual scores ranged from 73.3% for the informant himself in Table 4 down to two instances of 5 % in Table 2. Examination of the confusion matrices suggests that more information has been transmitted than indicated by the overall scores. Indeed, the pattern of responses might make one suspect the existence of a marginal tonal 7

The informant and base line tests described here are the same as those found in Abramson, 1962:

128.

34

ARTHUR S. ABRAMSON

TABLE 1 Confusion of Matrix of Tones in Words Number of subjects: 11 Percent Identification

Heard Mid Low High Falling Rising N:

Mid

Low

Whispered High

Falling

Rising

14J 9.1 27.1 5.5 43.7 55

14.5 54.5 5.5 7.3 18.2 55

23.6 14.5 40.0 9.1 12.8 55

16.4 10.9 20.0 52.7 0 55

7.3 7.3 12.8 3.6 69.0 55

Heard as intended: 46.2% /naa/ /naa/ /náa/ /nák/ /naa/

Words 'field' (Mid tone) 'custard apple' (Low tone) 'mother's younger sibling' (High tone) 'face' (Falling tone) 'thick' (Rising tone)

TABLE 2 Confusion Matrix of Tones in Words Number of subjects: 8 Percent Identification

Heard Mid Low High Falling Rising N:

Mid

Low

Whispered High

Falling

Rising

42.5 52.5 5.0 0 0 40

35.0 52.5 0 12.5 0 40

37.5 60.0 2.5 0 0 40

52.5 35.0 0 12.5 0 40

20.0 67.5 0 7.5 5.0 40

Heard as intended: 23% /k'aj/ /k'àj/ /k'áj / /k'âj/ /k'äj/

Words 'dried sweat' 'egg' 'to scoop out' 'fever' 'to unlock'

35

TONAL EXPERIMENTS WITH WHISPERED THAI

TABLE 3 Confusion Matrix of Tones in Words Number of subjects : 11 Percent Identification Whispered

Heard Mid Low High Falling N:

Mid

Low

High

Falling

12.7 61.8 9.1 16.3 55

40.0 18.2 5.5 36.4 55

31.0 9.1 12.7 47.2 55

26.4 20.8 11.6 41.5 55

Heard as intended: 21.3% Words /lom/ 'wind' /lòm/ 'mud' /16m/ 'to fall' /lôm/ 'shipwreck' TABLE 4 Confusion Matrix of Tones in Words Number of subjects: 8 Percent Identification

Heard Mid Low High N:

Mid

Whispered Low

High

57.5 37.5 5.0 40

42.5 57.5 0 40

52.5 27.5 20.0 40 Heard as intended: 45 %

Words /p'asae/ 'raft' /p'œiae/ 'to spread' /p'a¿se/ 'to be defeated' system whose categories do not coincide exactly with those of the normal tonal system. It is however hard to see a consistent breakdown into categories. Note, for example, the rather different treatment of the rising tone in Tables 1 and 2 even though in both tests there are five response choices. 8 * Chi-square tests show that the overall distribution of responses is significantly different from chance at the 1 % level of confidence in Tables 1 and 3 ; it is barely significant at the 1 % level in Table 4 and not significant in Table 2.

36

ARTHUR S. ABRAMSON

The data of Tables 1 through 4 hint that the context-free conditions of the first four tests may have been too severe to allow a by-system of whispered reflexes of tonal categories to emerge clearly, although such a system, without too much of a stretch of the imagination, may seem to be incipient in the matrices. Four more tests were prepared with sets of two and three tonally differentiated words embedded in sentence frames and randomized on magnetic tape. Because of grammatical and semantic constraints, we could not at that time think of any sentence that would accomodate five tonally differentiated words. The confusion matrices for these tests are displayed in Tables 5 through 8. The sentences, key words and glosses are given in the tables. The sentence environments seemed to induce a slight improvement in perception, although it is difficult to quantify the difference.' For three of the eight subjects there was a startling improvement. In the test underlying Table 5, one subject had 90% correct.10 All the other subjects, however, ranged from 40% to 60%, thus accounting for the poor resolution of Table 5. The somewhat better resolution of Table 6 is accounted for by the 100%, 80% and 70% achieved by three other subjects. Once we move to a three-way choice in Tables 7 and 8, we do not find such high individual scores. In Table 7 individual scores range from 33.3 % to 60%, and in Table 8 from 33.3 % to 80%, the latter achieved by the informant himself. It seems reasonable to infer from these findings that, given a sufficiently long linguistic context, some Thai speakers at least are moderately successful at using phonic features other than pitch to distinguish phonemic tones perceptually. At this point the question arose as to whether the concomitant features associated with the distinctive pitch contours were simply not as audible in whisper as in phonated speech. To test such a hypothesis it was necessary to expose the subjects to voiced speech which presumably retained the concomitant features but was neutral as to pitch. This was done by passing the sets of spoken words through an 18-channel vocoder at a constant fundamental frequency, i.e., a monotone, of 130 cps.11 Although some of the concomitant features, e.g., the abrupt amplitude drop of the falling tone, were quite detectable by ear, the four tests yielded no discrimination at all. That is, the eleven subjects either showed chance distributions of responses or assigned nearly all the stimuli to the mid tone when it was one of the response choices. It would seem that the presence of pitch, even if it was a monotone, was too great ' Chi-square tests show that the overall distribution of responses is significantly different from chance at the 1 % level of confidence in Tables 7 and 8; it is barely significant at the 1 % level in Table 6 and not significant in Table S. 10 Interestingly enough, the informant correctly identified only 50% of his own productions in this test. 11 The vocoder is a machine that processes speech, first by analyzing it and then by resynthesizing it. The analyzer separates the speech signal into information about its voiced and voiceless characteristics and its spectrum information, or roughly, into information about activity at the vocal folds versus information about articulation. The synthesizer supplies its own voiced sounds in the form of a buzz and its own voiceless sounds in the form of a hiss. These sounds are shaped by the spectrum information from the analyzer. The result is highly intelligible speech, but with a voice quality that is characteristic of the machine (Dudley, 1939).

TONAL EXPERIMENTS WITH WHISPERED THAI

37

TABLE 5 Confusion Matrix of Tones in a Séntence Number of subjects: 8 Percent Identification Whispered

Heard Low Falling N:

Low

Falling

30.0 70.0 40

32.5 67.5 40 Heard as intended: 48.8%

Sentence frame: Ip'öm ruu waà k'un c'áóp 'I know you like good Keywords: /k'aaw/ 'news* /k'aaw/ 'rice'

dii/

TABLE 6 Confusion Matrix of Tones in a Sentence Number of subjects: 8 Percent Identification Whispered

Heard High Rising N:

High

Rising

47.5 52.5 40

15.0 85.0 40 Heard as intended: 67.5%

Sentence frame: /p'öm mii hòk tua/ 'I have six Keywords: máa *horse(s)' màâ 'dogCs)'

a distraction for any kind of tonal identification based on other features. Thé experiment was repeated by resynthesizing phonated versions of the sentences used for Tables 5 through 8 on vocoder buzz to make sure, once again, that the lack of a carrier frame was not dulling the sensitivity of the subjects to such cues as were present in the key words. There was no sharpening of discrimination at all; the results were the same as for the isolated words. These findings are consistent with those of another experiment (Abramson, 1962:131-33 and Figure 3.11), in which the optimum contour for each of the five tones was artificially imposed on each member of a set of five

38

ARTHUR S. ABRAMSON

TABLE 7 Confusion Matrix of Tones in a Sentence Number of subjects: 8 Percent Identification

Heard Low Falling Rising N:

Low

Whispered Falling

Rising

17.5 5.0 77.5 40

52.5 30.0 17.5 40

5.0 10.0 85.0 40 Heard as intended: 44.1 %

Sentence frame: /pen / 'It is a .' Key words: /sia/ 'mat' /sìa/ 'upper garment* /sia/ 'tiger'

TABLE 8 Confusion Matrix of Tones in a Sentence Number of subjects: 8 Percent Identification

Heard Mid High Falling N:

Mid

Whispered High

Falling

50.0 37.5 12.5 40

27.5 52.5 20.0 40

40.0 10.0 50.0 40 Heard as intended: 50.8%

Sentence frame: /raw cà paj duu k'òbg k'un/ 'We will go look at your .' Keywords: /naa/ 'field' /ηώ/ 'mother's younger sibling' laS&l 'face'

tonally differentiated words for identification by Thai listeners. This was done with the Haskins Laboratories' Intonator, a device that enables one to pass speech through the vocoder while at the same time substituting a new fundamental frequency contour for the original one. Ten Thai subjects made nearly perfect identifications of all members of the /naa/ set of words (listed in Table 1) in terms of the synthetic tones

39-

TONAL EXPERIMENTS WITH WHISPERED THAI

rather than the tones of the original unvocoded productions.12 It is interesting to note that the experimenter himself, in taking this test, was able to recognize nearly all instances of syllables originally spoken on the falling and rising tones by listening for allophonic variations in duration and changes in the course of intensity; yet the native speakers of Thai apparently ignored these features and attended only to the fundamental frequency movements. Even while providing some interesting information in their own right, the vocoder monotone experiments did not answer the questions for which they were designed. To determine whether the concomitant features of the tones might not be more audible in phonated speech than in whisper it was apparently .necessary to strip away the voice and thus remove the distracting impression of pitch. This was done by passing both the isolated words and the sentences through the vocoder once again, but this time with the buzz generator turned off. The outputs of the analyzing channels were modulated instead upon hiss to produce vocoder 'whispering'. In these tests the subjects were told that someone else had done the whispering. The word data are displayed in Tables 9 through 12, and the sentence data 13 in Tables 13 through 15. As we look at the word data in Tables 9 through 12, we find that overall recognition is somewhat better than for the isolated whispered words in Tables 1 through 4.14 The confusion matrices show that some of the tones, in particular the rising tone, are perceived rather well on vocoder hiss. Also, as before, certain subjects did much

TABLE 9 Confusion Matrix of Tones in Words* Number of subjects: 9 Percent Identification

Heard Mid Low High Falling Rising N:

Mid

Resynthesized on Vocoder Hiss Low High Falling

13.3 0 60.0 8.9 17.8 45

20.0 66.7 0 4.4 8.9 45

0 0 75.5 2.2 22.2 45

2.2 0 26.8 71.0 0 45

Rising

0 0 0 0 100 45

Heard as intended: 71.6% * The same as in Table 1. 11 Each of the five words, of course, had the synthetic version of its own original tone imposed upon it as well as those of the other four tones making twenty-five stimuli in all. These were recorded four times and randomized into a test order of 100 items. 11 The sentences of Table 7 were not used for this part of the study. 11 Tables 9 through 12 are all significant at the 1 % level.

40

ARTHUR S. ABRAMSON

TABLE 10 Confusion Matrix of Tones in Words* Number of subjects: 9 Percent Identification

Mid

Resynthesized on Vocoder Hiss Low High Falling

17.8 26.7 6.7 8.9 40.0 45

15.6 82.3 10 2.2 0 45

Heard Mid Low High Falling Rising N:

22.1 28.9 0 13.3 35.6 45

24.4 24.4 0 51.2 0 45

Rising

6.7 11.1 0 0 82.3 45

Heard as intended: 46.7% * The same as Table 2.

TABLE 11 Confusion Matrix of Tones in Words * Number of subjects: 9 Percent Identification

Mid Heard Mid Low High Falling N:

35.6 28.9 24.5 11.1 45

Resynthesized on Vocoder Hiss Low High

28.0 72.0 0 0 43

20.0 15.6 40.0 24.5 45

Falling

15.8 15.8 9.1 59.0 44 Heard as intended: 51 %

* The same as Table 3.

better than the others. For example in Table 9, two subjects achieved identification scores of 88% and 80% respectively, while the rest ranged from 64% to 52%. In Tables 10 through 12 the individual scores ranged from 75% down to 6.6%. The sentences also show a somewhat sharper patterning of responses on vocoder hiss than do their whispered counterparts.15 Table 13 makes it dramatically evident that at least some tonal oppositions in suitable sentence contexts can be reliably distinguished. 15

Table 13 is significant at the 0.1 % level and Table 15 at the 1 % level. The distribution of responses in Table 14 is not significantly different from chance.

41

TONAL EXPERIMENTS WITH WHISPERED THAI

TABLE 12 Confusion Matrix of Tones in Words* Number of subjects: 9 Percent Identification

Heard Mid Low High N:

Mid

Resynthesized on Vocoder Hiss Low

High

33.4 4.4 62.2 45

48.8 42.3 8.9 45

27.3 9.1 63.6 44 Heard as intended: 45.9%

* The same as Table 4. TABLE 13 Confusion Matrix of Tones in a Sentence* Number of subjects: 6 Percent Identification Resynthesized on Vocoder Hiss Low Falling Heard Low Falling N:

91.7 8.3 36

2.8 97.2 36 Heard as intended: 94.4%

* The same as in Table 5. TABLE 14 Confusion Matrix of Tones in a Sentence* Number of subjects: 6 Percent Identification Resynthesized on Vocoder Hiss High Rising Heard High Rising N:

58.3 41.7 36

47.2 52.8 36 Heard as intended: 55.7%

* The same as Table 6.

42

ARTHUR S. ABRAMSON

TABLE 15 Confusion Matrix of Tones in a Sentence* Number of subjects: 6 Percent Identification Resynthesized on Vocoder Hiss Mid

High

Falling

36.2 11.0 52.8 36

30.6 36.2 33.3 36

8.3 8.3 83.4 36

Heard Mid High Falling N:

Heard as intended: 51.8% •

The same as Table 8.

Table 14, with its chance distribution of responses, is in fact less good than its whispered counterpart in Table 6. This is more apparent in the treatment of the rising tone than in the overall scores. Pending instrumental examination of the stimuli, we can only speculate that the speaker may have considerably enhanced the concomitant features in that particular rendition of whispered /máá/. Table 15 indeed shows a patterning of responses — but one that is somewhat different from that of its whispered counterpart in Table 8. It is hard to decide that there is an improvement throughout the confusion matrix, even though the falling tone in particular is now identified correctly 83.4% of the time. The results of the research presented here indicate that in the context-free setting of isolated words, whispered Thai tones cannot be well identified. As Hockett (1955:17-18) points out in discussing by-systems and marginal cases at the boundaries of language that should not be allowed to complicate one's phonologic analysis, "A whispered utterance mocks the phonologic structure of the same utterance spoken in the normal way, but almost always omits certain contrasts which are functional in normal speech." The responses to the whispered sentences, however, suggest that when the concomitant features of tonal phonemes are embedded in a sufficiently long phonic environment, at least some Thai speakers can do reasonably well at identifying certain tones. The vocoder buzz experiments, both for words and sentences, convincingly demonstrate that these concomitant features are really redundant and probably can never function distinctively in the presence of distracting pitch. This conclusion is supported by parallel work done on the tonal contours of Thai (Abramson, 1962). But the vocoder hiss experiments show that the concomitant features are indeed present in phonated speech and, once the distracting voice is stripped away, are somewhat more audible than in whispered speech. The poor transmission of tonal features in whispered speech did not seem to warrant a detailed instrumental analysis of the concomitant features for the present

ñatea speecn, me wnoie larynx is iiKeiy to nse ana iaii wiin great cnanges in hus shortening and elongating the pharyngeal tube; such changes in the length vocal tract will of course affect formant frequencies and thus sometimes, if ormant shifts exceed the psychoacoustic thresholds, cause changes in vowel (Parmenter, Trevrflo and Bevans, 1933). Auditory phonetic analysis suggests is happens in Thai, but no quantitative data are available. For this to have ect in whispered speech, it is necessary to suppose that the speaker replicates :tions' that he sends in phonated speech to the extrinsic muscles of his larynx srhaps, muscles of the tongue to make adjustments in vocal tract configuration lítate large pitch changes. 1 ' Duld seem that in Thai and no doubt other tonal languages17 whispered comition can be ambiguous in short utterances with low redundancy.

In longer

ices or in short utterances embedded in a conversation or a particular situale high contextual redundancy plus the tonal distinctions that whispering does e, combine to make whispered communication quite feasible. UNIVERSITY OF CONNECTICUT STORRS, CONNECTICUT AND HASKINS LABORATORIES NEW YORK

REFERENCES on, Arthur S., "Vocoder Output and Whispered Speech in a Tone Language: Thai", J. Acoust. Soc. Am. 31 (1568) A. The Vowels and Tones of Standard Thai: Acoustical Measurements and Experiments (= Bloomington: Indiana U. Res. Center in Anthro., Folklore and Linguistics, Pub. 20). Pierre, Alvin M. Liberman and Franklin S. Cooper, "Whispered Vowel and [h] : an Acoustic and Articulatory Comparison" (A paper presented at the Ann Arbor, Mich, meeting of the Linguistic Society of America, July 29,1959). . Hower, "Remaking Speech", J. Acoust. Soc. Am. 11:169-75. en, Otto, Allgemeine und angewandte Phonetik (Berlin Akademie-Verlag), ranz, Zur Tonität nordchinesischer Mundarten (Wien-Mödling, Missionsdruckerei St. Gabriel). "Kann man in einer Tonspracheflüstern?",Lingua 5: 372-81. t this kind of replication happens for voicing distinctions has been at least tentatively estaby transillumination of the larynx (Malécot and Peebles, 1965; Lisker, Abramson, Cooper and 1969). ί statement may not be completely true of a language in which one or more tones are characby prominent non-pitch features, such as strong glottal stop. This kind of thing may at least f explain the mixed observations of some of the sources cited.

44

ARTHUR S. ABRAMSON

Hockett, Charle» F., 1955 A Manual of Phonology (Baltimore, Waverly Press). Kloster Jensen, Martin, 1958 "Recognition of word tones in whispered speech", Word 14: 187-96. Lisker, Leigh, Arthur S. Abramson, Franklin S. Cooper and Malcolm H. Schvey, 1969 "Transillumination of the larynx in running speech", J. Acoust. Soc. Am. 45: 1544-46. Malécot, André and Κ. Peebles, 1965 "An Optical Device for Recording Glottal Adduction-Abduction During Normal Speech", Z.f. Phonetik, Sprachwissenschaft u. Kommunikationsforschung 18 : 545-50. Meyer-Eppler, E., 1957 "Realization of Prosodie Features in Whispered Speech", J. Acoust. Soc. Am. 29: 104-06. Miller, Charles Boardman, 1934 "An Experimental-Phonetic Investigation of Whispered Conversation, Considered from the Linguistic Point of View" (Ph.D. dissertation, Hamburg University). Miller, John D., 1961 "Word Tone Recognition in Vietnamese Whispered Speech", Word 17: 11-15. Noss, Richard B., 1954 "An Outline of Siamese Grammar" (Ph.D. dissertation, Yale University). Panconcelli-Calzia, G., 1955 "Das Flüstern in seiner physio-pathologischen und linguistischen Bedeutung", Lingua 4: 367-78. Parmenter, C.E., S.N. Treviño and C. A. Bevans, 1933 "The Influence of a Change in Pitch on the Articulation of a Vowel", Language 9: 72-81. Segerbäck, Borge, 1966 La réalisation tΓune opposition de tonèmes dans des syllabes chuchotées: étude de phonétique expérimentale. (= Travaux de l'Institut de Phonétique de Lund, IV) (Lund: Gleerup). Wise, C. M. and L.P.H. Chang, 1957 "Intelligibility of Whispering in a Tone Language", JSHD 22: 335-38. Zemlin, Willard R., 1964 Speech and hearing science: anatomy and physiology (Champaign, 111. : Stipes).

SIMON BELASCO

PHONEMICS AS A DISCOVERY PROCEDURE IN SYNCHRONIC DIALECTOLOGY

In recent years, the development of generative grammar has seriously brought into question the value of positing classical — sometimes called taxonomic or autonomous — phonemics as a separate level of linguistic structure. It is maintained that such a level is not incorporable into a descriptively adequate grammar, since it denies that phonetic processes are in part dependent upon syntactic and morphological structure (Chomsky, 1964:110-11; Chomsky and Halle, 1968:114-19; Postal, 1968: 240-44). Following Postal, we shall call the autonomous approach: PI, which is the view that NO information about word and morpheme boundaries, morphological and syntactic categorizations, morphophonemic alternation, etc. is relevant for the determination of phonological structure. The alternative view, where SOME reference to nonr phonetic morphophonemic and/or superficial grammatical structure is relevant will be termed the non-autonomous approach: P2. Another possibility, the inseparable approach: P3, is the view that ALL grammatical information is relevant in a phonological analysis. Since P3 has never really been maintained by anyone, it will not be mentioned any further in this study. Let us assume that P2 is the correct approach. As part of such an approach, generative phonology posits two levels: one 'systematic phonemic' — most recently termed a "phonological representation" (Chomsky and Halle: 1968:11) — and the other 'systematic phonetic' — also called a 'phonetic representation'. Roughly speaking, the output of the syntactic component (a proper labeled bracketing of a string of formatives) is connected to the phonological component by the application of a series of 'readjustment rules'. The result of this operation is the creation of the systematic phonemic level, the entire process being called a 'surface structure'. A system of phonological rules relates surface structures to phonological representations. In other words, the systematic phonetic level is derived from the systematic phonemic level by a set of fixed rules that apply in a fixed manner determined by the labeled bracketing of the surface structure. No linguistic significance is assigned to an intermediate 'classical phonemic' level between the systematic phonemic and the systematic phonetic levels.

46

SIMON BELASCO

The case against an intermediate ^autonomous' level is best exemplified by the arguments set forth by Paul Postal (1968:7-24). Briefly, they go something like this. All linguists agree that a phonological analysis should distinguish free variants from contrasting utterances. Given two utterances, such as [bin] and [pin], that may occur in two different languages, it is not possible to infer whether these utterances contrast or are in free variation in either language from their phonetic representations alone. A systematic, i.e., generative, grammar assigns two distinct phonetic representations, such as [kat] and [kaj], to the same free variation set, making them noncontrastive utterances, just in case the phonological rules derive them from "the same single input systematic phonological representation and not otherwise" (Postal, 1968: 14). Thus, omitting certain irrelevant details, Table 1 shows that the systematic TABLE 1 (After Postal, 1968) Systematic Phonemic a. #kat#

Autonomous Phonemic c. /kat/ # / # Sys. Phonemic

I

[t]

Systematic Phonetic b. [kat] b\ [kaj]

I

[1] Sys. Phonetic

phonetic representations b. [kat] and b'. [kaj] may be directly related to the systematic phonemic representation a. #kat#, i.e., they are in free variation and are non-contrastive, without the mediation of the autonomous phonemic representation c. /kat/. Table 2 illustrates an objection by Postal (1968:20-22) to Sydney Lamb's (1964:7576) claim that autonomous representations are essential in order to indicate the lack of voiced : voiceless contrast in final position in certain Russian utterances. According to Postal, items bl. [...ot...] and b2. [...ot...] are free variants because they are phonetically identical, whereas b3. [...ada...] and b4. [...ata...] are contrastive "because there is no single systematic structure to which they may be assigned" (Postal, 1968:24). In other words, the feature 'voice' is predicted as distinctive or contrastive in intervocalic (but not in final) position because the phonological rules do not derive systematic phonetic intervocalic [t] and [d] from some SINGLE systematic intervocalic phoneme. Since the facts concerning free variation and contrast can be read off directly from the two systematic levels, then autonomous representations such as c3. /...ada.../ and c4. /...ata.../ are redundant. As long as no motivation exists for reading the facts off from one level instead of two levels, there are once again "no grounds whatever for including autonomous phonemic representation in a linguistic description" (Postal, 1968:22). If circumstances turned out to be as typical and as simple as those depicted in Tables 1 and 2, then investigations in synchronic dialectology would be a relatively easy matter. In the first place, a dialectologist cannot wait until he has made an

PHONEMICS AS A DISCOVERY PROCEDURE

47

TABLE 2 (After Lamb and Postal) Systematic al. a2. a3. a4.

Phonemic

Autonomous

cl. c2. c3. c4.

...od... ...ot... ...ada... ...ata...

d

V[t]

t

Phonemic

Systematic

/...ot.../ /...ot.../ /...ada.../ /...ata.../ d t Sys. Phonemic

1 1[t]

[d]

bl. b2. b3. b4.

Phonetic

..ot...] ..ot...] [. ..ada...] ..ata...]

Sys. Phonetic

exhaustive analysis of the syntactic, if not the semantic, structure of a language before he learns the facts about free variation and contrast.1 In the second place, even if he intends to incorporate P2 in a finalized version of a total linguistic description, he is still faced with the necessity of developing an effective discovery procedure that leads from the accumulated phonetic data to the best phonemic system. To show that the circumstances are more complicated than those depicted in Tables 1 and 2, let us first compare the systematic phonetic data in Table 2 with the phonetic data in Table 3, representing an American dialect which we may call South Philadelphianese. In Table 2 the phonetic data show BOTH [t] and [d] in intervocalic position but ONLY [t] in final position. In Table 3 there is no phonetic [s] or [z] that occurs TABLE 3 (After Belasco) Phonetic Data for South Philadelphianese

(circa 1930)

Final Position

Intervocalic Position

Initial Position

because [...s] because [...z] fuss [...8] fuzz [...z] close [...s] close [...z] Japanese [...s] Japanese [...z]

partisan [...s...] partisan [...z...] citizen [...s...] citizen [...z...] Joseph L-.-s—] Joseph [...z...] resourceful [...s...] resourceful [...z...] razor [...z...] racer [...s...] cussin' [...s...] cousin [...z...] classy [...s...] jazzy [...z...]

zeppelin [ζ...] zeppelin [s...] sink [s...] sink [ζ...] sip [s...] sip [ζ...] soup [s...] soup [ζ...] sue [s...] zoo [ζ...] zip [ζ...] sit [s...]

1

Establishing MEANINGFUL communication with NON-SOLICITED informants can be a matter of life and death in the field. The possibility of eventually formulating a completed linguistic description may depend upon how rapidly such factors as contrastive and freely varying utterances are determined. For dramatic incidents suggesting what problems may be involved, see Wallis, 1960.

48

SIMON BELASCO

in one environment ONLY, i.e., phonetic [s] and [Ζ] occur in BOTH final AND intervocalic position.* Then some procedure must be established to show that phonetic [s] and [z] in the two pronunciations of utterances such as because, partisan, citizen, Joseph derive from some SINGLE systematic phoneme in each of the two environments: final and intervocalic; whereas phonetic [s] in fuss, cussin', classy only derives from systematic phonemic s in final and intervocalic position, and phonetic [z] in fuzz, cousin, and jazzy only derives from systematic phonemic ζ in these same two environments. To a dialectologist familiar with English, this may not present a major problem. There is nothing to prevent a language, however, from having two or more series of words that DO NOT constitute a predictable subdivision of the lexicon, such as [+native] or [+Romance] words, where two phones may be in free variation in the words of one series, and may contrast in the words of another series in exactly the same environments. The problem of assigning phonetic data to systematic phonemic structure is even more difficult when the dialectologist is confronted by utterances involving sandhi variation in a strange language. Some idea of the problem may be gotten from the examples in Table 4. These examples are typical of certain English dialects. TABLE

4

Phonetic Data for South Philadelphianese a. b. c. d. e.

Amina [amine] as in Amina (/ am going to) go. Whadiya [wAdija] as in Whadiya ( What do you) want? Whadya [wAdja] as in Whadya ( What did you) want? Whacha [WA£S] as in Whacha ( What do you) say, Joe? Whaja [WAJS] as in Whaja ( What did you) say, Joe?

For example, item a. [amina] is not usually identifiable out of context. Item b. [wAdija] (present tense) is distinguished from item c. [wAdja] (past tense) by the occurrence of phonetic [i] in the former. Item d. [WA£S] (present tense) is distinguished from item e. [wAja] (past tense) by the occurrence of phonetic [δ] in the former and phonetic [j] in the latter. Note that items d. and e. ([WA&] : [WAJS]) differ by only one phone and ARE contrastive. On the other hand, items c. and e. ([wAdja] : [WAJS]) differ by only one phone and ARE NOT contrastive. Note further that TWO DIFFERENT SEQUENCES pronounced 'ideally' as 'what do you' and 'what did you' may as a result of sandhi variation have phonetic representations that constitute a minimal pair [WAÒS] : [WAJS], whereas A SINGLE SEQUENCE pronounced 'ideally' as 'what do you' normally DOES NOT have sandhi representations that constitute a minimal pair [wAdija] [WAÖS].

When a dialectologist works with unfamiliar languages, he encounters morphophonemic problems of far greater magnitude. Assigning phonetic utterances to systematic phonemic structure is not impossible, but he will have to rely on some dis* To facilitate the exposition, we shall omit the phonetic data relating to [s] and [z] in initial position.

49

PHONEMICS AS A DISCOVERY PROCEDURE

covery procedure that will enable him to determine free variation and contrast between utterances, on a level lower than the still-to-be-determined systematic phonemic level, meeting the conditions of descriptive and explanatory adequacy. This is not to say that such a lower level may not be eliminated in the later stages of a completed linguistic description. When I say a lower level, I do not mean the phonemic level that is identical to the notion widely held by classical phonemicists. In 1959, I pointed out that phonemic analysis of complex distributional data may not be effectively carried out solely on the basis of complementary distribution, phonetic similarity, and pattern congruency (Belasco, 1959:269). Too often complications arise when more than one phonemic grouping is possible with phonetically similar phones in free variation or complementary distribution. A case in point (cf. Table 5) is Bernard Bloch's phonemic grouping TABLE 5 (After B. Bloch) Japanese Conservative Dialect Contrast

Free Variation

Complementary Distribution

[f] : [s]

[f] : [h] [b] : [s]

[h] : [bl [f] : tb]

[f] + Eh] + [bl = /h/ [s] = /X/

of "spirants" in one of his celebrated "Studies in Colloquial Japanese" (Bloch, 1950: 113). A short voiceless bilabial or labiodental spirant [f], a short voiceless glottal spirant [h], and a short voiceless palatalized glottal spirant [h] are all grouped in the same phoneme /h/ on the basis of phonetic similarity. The phone [f] is in free variation with the phone [h] which is in complementary distribution with the phone [h]. However, there is a short prevelar spirant [x] which is said to contrast with the phone [f ] and therefore is assigned to a different phoneme /x/ — despite the fact that phone [x] is always in free variation with phone [h]. In other words, all four phones are phonetically similar, i.e., they are spirants, and [h] is grouped with [f] and [h] because they are all "strongly pulsed voiceless onsets of a following vowel or semivowel", (Bloch, 1950:94), whereas [x] is "produced with contact between the back of the tongue and the forward part of the soft palate" (Bloch, 1950:113). But such phonemic grouping is untenable, since [x], which varies freely with [h], contrasts with [f], which varies freely with [h]. s The procedure we intend to propose will rightly group [f] and [h] as allophones of one phoneme, and [χ] and [h] as allophones of a different phoneme. * Equally untenable is the position taken by Austin (1957: 542), wherein he maintains that his unidimensional principle requires that phonetically similar phones grouped as allophones "have point of articulation or manner of articulation in common, but never both". For a discussion of this point relating to Bloch's grouping of spirants, see Belasco, 1959: 278, footnote 16.

50

SIMON BELASCO

This procedure DOES NOT make the claim that two phonetically similar phones that share no environments, i.e., that are in complementary distribution, are necessarily allophones. There is no basis for recognizing the principle of complementary distribution as scientifically motivated. It is for the most part arbitrary and without empirical justification. Nor do we claim that two phonetic representations which are free variants necessarily have identical phonemic representations. They may contain non-contrasting segments. The term 'non-contrasting' is used here in a special sense. Two NON-CONTRASTING phones are not necessarily allophonic. The term implies free variation but does not imply that two phones cannot contrast elsewhere in the same or different environments. Two segments which are allophonic are NON-CONTRASTIVE. Thus, non-contrasting phones may be either allophonic segments or freely varying non-allophonic segments. It is then possible to refer to a state of free variation between two phones without involving their phonemic status. A comparison of two utterances, such as [bin] : [pin] or [nip] : [nip'], that appear to differ by one feature, [lax] : [tense] or [non-aspirated] : [aspirated], may or may not constitute a minimal pair. Actually, any difference between two phones in a comparison may be placed in one of three mutually exclusive categories. Either the differences are (1) distinctive or (2) non-distinctive. If they are non-distinctive, they may be sub-categorized as (2a) allophonic or (2b) irrelevant. There has been much discussion in the literature recently concerning linguistic universale, both substantive and formal. It is fashionable to speak of a theory of universal phonetics that specifies the class of possible phonetic representations of sentences "by determining the universal set of phonetic features and the conditions of their possible combinations" (Chomsky and Halle, 1968 :4-5). Suppose we assume that a universal set of phonetic features does exist. It is a fact that all human societies communicate by speaking rather than — let us say — by tweaking the ears or blowing through the nose. The number of possible speech sounds that can be made are confined within the vocal apparatus to all possible movements of all possible organs involved in articulation. Since all languages past, present, and future are not available for analysis, we cannot be certain of all the possible combinations of speech sounds. Nor can we be certain that a complete inventory of the set of phonetic features has yet been made.4 What is common to all language is the fact that each sound or sequence of sounds is uttered as part of a time continuum. As such, speech is linear. Linguists have adopted the practice of dividing sequences of sounds into discrete elements for convenience of reference, and then they compare two or more sequences to determine * One only has to consider how many times distinctive feature theory and its relation to acoustic and articulatory correlates has undergone revision to appreciate the fact that the theory has yet to be stated in definitive form. (cf. footnotes and references in Postal 1968; Chomsky and Halle 1968). The recent concept of MARKEDNESS resembles a probabilistic criterion more than a universal criterion, e.g., the unmarked member of a pair of segments (1) OCCURS MORE FREQUENTLY, (2) MOST LIKELY OCCURS FIRST in language acquisition, (3) APPEARS OFTEN in neutralized environments (cf. Schane, 1968 : 714-15).

PHONEMI CS AS A DISCOVERY PROCEDURE

51

where they differ and where they are the same.6 The factors of free variation and contrast apply to selected points or positions within paired continua. We shall adopt this procedure as well as the test of segment substitutability, which involves the commutation of phonetically similar segments in common environments, of paired or phonetically similar utterances, to determine whether the original utterances undergo a change in meaning. If a change in meaning DOES occur in one of two compared utterances, then the phones are CONTRASTIVE. If no change in meaning occurs, the phones are NON-CONTRASTING. In the event that two phones share no common environment, their status may be ascertained from the relationship with some third phone with which they DO share a common environment. Despite claims to the contrary (Postal, 1968:28), we shall maintain that the discovery of phonetically minimal pairs does permit a conclusion about underlying phonological contrast. However, following Chomsky (1964:97), we shall insist that the notion of minimal pair be defined only in terms of a completed phonemic analysis. Thus the basis for allophony is inherent not only in the relationship between two phones, but between one phone and every other phone in the dialect undergoing analysis. This means that all environments have to be taken into consideration. Furthermore no segment is an allophone of a phoneme χ if it contrasts with any allophone of /x/ or if it is allophonic with any segment which contrasts with an allophone of /x/. Thus, the determining factor that allophony exists between two phones is NOT that they are phonetically similar, or in free variation, or in complementary distribution. The determining factor is ABSENCE OF CONTRAST BETWEEN PHONES. We can make this claim because a phonological system is a closed system that is limited to a set of universal relationships existing between the range of one segment and the range of every other segment in the system. We use the term RANGE here to mean the total number and kinds of environments in which a phonetic segment, i.e., a phone, occurs. The environments of phones are made up of phones. It is subsumed that phones are composed of phonetic features. When one phone in an utterance appears to be different from a phone in another utterance and the phones share a common environment, then the difference is either distinctive, non-distinctive allophonic, or non-distinctive irrelevant. Table 6A is in the form of a matrix and illustrates the universal relationships for a phonological system. The rows labeled 'identical', 'mixed y', 'mixed z\ and 'different' indicate the possible 'general' 'relationships' between the ranges of any pair of phones. The circles are called Euler circles or Venn diagrams. The letters a and b represent phones. When phones a and b occur in a single circle, it means that their ranges are exactly the same, i.e., the phones have identical environments. The 5 Pew linguists, if any, harbor the illusion that phones or phonemes may be isolated as discrete entities, either from an articulatory or an acoustic point of view. Speech-sound investigations experimenting with bursts of noise, consonant-vowel transitions, variations in formant intensity, etc., belie any such notion. For some discussion, see Joos, 1948; Schatz, 1954; Liberman et al., 1954; Delattre et al., 1955, 1967; Moulton, 1968.

52

SIMON BELASCO TABLE 6A The Universe for a Phonological System Allophones Free Variation

Sometimes Contrast Sometimes Free Variation

Contrast

(1)

(2)

(3)

Identical

Mixed y

Μ**,

Different

Non-allophonis Segments

φ (Jw s—-ν

s—v.

Θ GO Q)

Θ GO Φ ΘΘ

TABLE 6B Allophonic (1)

Non-allophonic Segments (2)

(3)

Identical Mixed y Mixed ζ Different

set of environments shared by two phones is termed the PARS COMMUNIS; the set of environments not shared make up the PARS PROPRIA.® Thus the phones a and b share a pars communis, but neither has a pars propria when their ranges are 'identical'. When their ranges are in 'mixed y' relationship, a and b share a pars communis and each phone has a pars propria. In 'mixed z' relationship, a and b share a pars communis, but only a has a pars propria. Of course the relationship is still 'mixed z' if b is assigned the pars propria instead of a, provided a and b still share a pars communis. When the ranges of a and b are 'different', they each have a pars propria and share no pars communis. The test of segment substitutability (or commutation test) can only be performed • The terms (pars) communis, propia, as well as coincident, overlapping, and relative distribution are used in Bloch 1953, but not in the way we use them.

PHONEMI CS AS A DISCOVERY PROCEDURE

53

when two phones have at least one environment in common, i.e., when the ranges of two phones are either in 'identical', 'mixed y', or 'mixed z' relationship. For the time being, let us restrict our remarks to these three relationships, omitting the relationship 'different' from the discussion. Table 6A applies to the ranges of phones in a 'complete' corpus. The numbers 1, 2, 3 (precise relationships) above the columns indicate the allophonic status of the phones AFTER the test of segment substitutability has been applied to ALL utterances in the corpus in which the phones have one or more environments in common. Thus, 3 means that for a given dialect every time phone a in one utterance has been substituted for phone b in another utterance or vice-versa, the utterances in question have changed meaning. For example, if the [s] in [fAs] is replaced by the [z] in [fvz], then the meaning of this word changes. It makes no difference what the word originally meant. It has a different meaning as a result of the substitution test. This is then a case of CONTRAST, and the phones a and b are not members of the same phoneme, i.e., are not allophones. There is an important point to be made here. The substitution of [z] for [s] could have resulted in gibberish, in which case there WOULD NOT have been contrast. For example, the [s] in [kriys] crease, when replaced by [z], would result in the word [kriyz] *creaze — a nonsense word. This would not represent a case of contrast. The word is just declared as gibberish and omitted from consideration. Nevertheless, to qualify for status 3, i.e. contrast, all the examples of the commutation test either produce contrast, i.e., produce a change in meaning, or result in gibberish but NEVER in free variation. The feature responsible for contrast in this instance, is VOICE, i.e., the distinctive difference is due to the opposition [—voice] : [+voice] IN ALL THE ENVIRONMENTS THAT THE PHONES SHARE.

For the column headed 2, the commutation test for all utterances of a given dialect must result in BOTH contrast AND free variation, and possibly gibberish. For example, a complete inventory of the utterances involving commutation of [s] and [z] in the environments shared might produce CONTRASTS such as [fvs] : [fAz] (fuss : fuzz), [reysir] : [reyzir] (racer : razor), [suw] : [zuw] (sue : zoo) AND FREE VARIATION such as [biykos] : [biykoz] (because), [sitisin] : [sitizin] (citizen), [suwp] : [zuwp] (soup), and possibly gibberish [klaesiy] : *[klaeziy], *[jaesiy] : [jaeziy]. The last two pairs of utterances will have no bearing on the allophonic status of phones [s] and [z] and will of course be discarded. When the ranges of two phones of a dialect indicate either status 2 or 3, the phones are non-allophonic and are NOT grouped in the same phoneme. Note that CONTRAST is the determining factor here, NOT FREE VARIATION. For the column headed number 1, no contrast of any sort ever obtains. Every case of commutation will produce free variation, and possibly gibberish, in every environment shared by two phones. THEN and ONLY THEN can the phones be said to be allophones of the same phoneme. Let us return to the general relationship in Table 6A, which shows the ranges of the phones a and b to be different, i.e., they never share a pars communis. In the event

54

SIMON BELASCO

that two phones do not share a single environment, their status may be ascertained by the triangle method indicated in Table 7. The solid lines in triangles A, B, and C indicate the allophonic status of phones which have already been ascertained by the commutation test. The broken line connects two phones which do not share a common environment. Another possibility exists with the broken line which we shall discuss later. Thus, for triangle A in Table 7, phones a and b have the relationship 1 TABLE 7

(B),

(A)a \

ordre.

2. EVOLUTION EN R D'UNE SPIRANTE VÉLAIRE

Dans une pièce de théâtre passée au début de 1968 à la Télévision française, un acteur, imitant une prononciation du français par un touriste suédois, faisait entendre barrage pour bagage quand il parlait des ses valises: son g intervocalique entre deux a étant spirant et non occlusif, un francophone dont Yr est dorsal ou vélaire ne pouvait interpréter ce g que comme un r, d'où le quiproquo. Des faits analogues s'observent en breton, et ont pu se manifester lors du passage du gaulois au roman. Le breton possède deux spirantes vélaires, l'une sourde et l'autre sonore, toutes deux souvent confondues sous le même signe orthographique c'h, et un r qui est vélaire en certains parlers. Dans ces parlers, locuteurs et auditeurs distinguent mal entre spirante vélaire sonore intérieure et r intérieur, entre spirante vélaire sourde et un r qui peut également s'assourdir en finale absolue ou au contact d'une consonne sourde; sans compter que la distinction entre la sonore h et la sourde χ n'est pas toujours très nette. Il en résulte des confusions possibles entre h, x, r, rh, rx, etc. ... On en trouve le témoignage dans plusieurs cartes de Y Atlas linguistique de la BasseBretagne de Pierre Le Roux. Ainsi, la carte 6, 'd'ici' note durame (point 40) à côté de do'tiarmn (point 44). Les mots contenant un groupe Ih sont particulièrement instructifs grâce à la métathèse que pratiquent certains parlers, ainsi alhouez 'clé', c. 3; gwalhi 'laver', c. 286; kelh 'cercle', c. 370. Pour alhouez, on relève [arxlqe] (pt. 63) à côté de [ahlqe] (pt. 64); pour gwalhi, [gwahlsjn] (pt. 71), [gwarlçji] (pt. 68), et [gwaxlsji] (pt. 83); pour kelh, [csrl] (pts. 99, 41) et [ckrx] (pts. 16, 17, 20). Des confusions identiques s'observent assez souvent, au Collège Littéraire Universitaire de Brest, dans les copies de thème breton d'étudiants cornouaillais (de la région comprise entre Carhaix, Quimper, et Quimperlé), dont quelques-uns ont de la peine à distinguer entre les finales de er 'air', erh ou erc'h 'neige', et neh ou nec'h 'hauteur'. Les chansons populaires sur feuilles volantes offrent de multiples témoignages de confusions semblables (Falc'hun, 1951: 50-51). Ajoutons-y l'exemple suivant: un chant du Barzaz-Breiz signalé comme recueilli en Haute-Cornouaille fait rimer 1er 'cuir' avec lec'h 'place, lieu', alors que toutes les autres rimes y sont correctes (Hersart de la Villemarque, 1867: 416). La toponymie apporte aussi son contingent d'exemples, puisés dans les Nomenclatures des hameaux, écarts et lieux-dits des départements bretons (1955, 1954, 1953, 1952). Selon celle du Finistère, on trouve en Cornouaille un hameau de Coat-anNec'h 'le bois de la colline' à Collorec, et, à une trentaine de kilomètres plus au sud, un hameau de Coataner à Scaër et un autre à Douarnenez. Il s'agit du même toponyme

DE L ' " R " GAULOIS À L ' " R " FRANÇAIS

113

qui reparaît, dans les Côtes-du-Nord, dans le nom de la Forêt de Coat-an-Hay et, avec l'article an en moins, dans le nom des hameaux de Coat-Nay et Coat-Nec'h. Le graphie Neh ne se rencontre pas isolément comme toponyme; mais elle est incluse dans le diminutif Néhan, second élément de Couesnéhan, hameau morbihannais. Le même diminutif se présente sous un travestissement dans Néant-sur-Yvel (Morb.). Si l'on compare la Haute-Bretagne francisée à la Basse-Bretagne encore bretonnante, on est amené à expliquer Pennère et Pennet comme Pen-Nec'h, Pen-Né et Pen-Nay 'sommet de la colline' (il existe un Mont-Pennay alpin, 1371 m., au nord-est de Chambéry), et Le Néard ou Le Niard (selon les cartes), village juché sur une colline de Plélanle-Grand (I.-et-V.) par le Néac'h (variante de Nec'h) des Penn-an-Néac'h finistériens. Dans Nermont, à Saint-Coulomb (I.-et-V.), mont serait la traduction de ner = nec'h. Des noms identiques ou semblables à ces toponymes bretons se rencontrent, à travers toute la France, dans un contexte géographique qui autorise leur interprétation par une variante nec -> neh -* ner ou né de la racine celtique enee, 'hauteur'. Les Néewiller du Bas-Rhin s'écrivaient aussi Nehwiller. Dans la Manche, St-Jacques-duNéhou s'accommoderait fort bien de la traduction 'Saint-Jacques-des-hauteurs', -ou étant considéré comme une désinence de pluriel celtique, comme dans le Moulin-duNeyhou de Maël-Pestivien (C.-du-N.). C'est le h des anciens Neh-willer alsaciens, des Néh-ou normands, du Neyh-ou breton, qui se retrouverait sous forme de r dans Ν ers (Gard), Nerville (S.-et-O.), Nerpol (Isère) comme dans les Coataner (= Coat-an-Nec'h) finistériens. Le bourg de Νers (Gard) s'étage au flanc d'un promontoire remarquable, et les habitants prononcent bien [ηεχ], Nerville-la-Forêt se dresse sur une hauteur en bordure de la forêt de l'Isle-Adam. Nerpol (Isère), Nerpou au XHIè siècle, qui s'expliquerait le plus régulièrement du monde comme un ancien neh-pagum, méritait en tout cas de porter un nom signifiant 'le pays des collines'. Les sommets pyrénéens du Pène Nére (2050 m., au nord de Cauterets) et du Soum de Nére (2401 m., entre Cauterets et le Pic du Midi de Bigorre), pourraient bien être des cousins gaulois, et des synonymes, des Pennère et Pen-Nec'h bretons. La racine neh semble expliquer bien d'autres noms de sommets pyrénéens, comme le Pic d'Anéou, 2179 m., ( = an Néh-ou?) à l'ouest du Somport, le Pic de Néouvielle, 3092 m., au sud du Col du Tourmalet, le Pic de Barané, 1977 m., (=Barr-an-Néh? 'sommet de la montagne') au sud-est d'Argelès-Gazost. D'autres variantes du celtique enee —> cneh ou enoe -*• enoh peuvent se cacher sous des toponymes français dont l'r continuerait un h celtique. Le moyen-breton knech est aussi représenté dans la toponymie moderne par Quénéac'h-Du, Quénarc'h-Du (Fin.) 'la colline noire', Quénah-Guen (Morb.) 'la colline blanche', Quénard (C.-du-N.), Quinard (I.-et-V.). Ou encore par Caner pour Canee'h à Brélidy (C.-du-N.). Ou aussi par les Canac'h de Laniscat et Saint-Nicolas-duPélem (C.-du-N.), auxquels correspondent les Canard(s) de Lanfains et Gouray dans la partie francisée du même département. Il n'est pas impensable que dans Conore ou Connore en Peyrilhac (Hte-V.), dans Les Conords en Vensat (P.-de-D.), l'r continue la spirante vélaire de conoch, forme

114

FRANÇOIS FALC'HUN

attestée par le Cartulaire de Redon (Courson, 1863), et variante probable du cnoch du même document. Si l'r de tous ces toponymes continue une ancienne spirante vélaire celtiaue issue de c, il est évident qu'il a d'abord figuré un r postérieur, dorsal ou vélaire. Bien d'autres toponymes français d'origine jugée obscure pourraient être mis en rapport avec cette racine celtique cnoc, enee, et le problème de la continuation possible d'une spirante vélaire gauloise par un r français. Ils ne fourniraient sans doute pas de conclusion plus assurée. Les exemples précités ne constituent pas non plus la preuve absolue d'une évolution h -> r dans le passage du gaulois au roman. Mais, appuyés sur des toponymes bretons apparemment identiques, où cette évolution h -» r ne peut être mise en doute, ils forment un faisceau de faits convergents de nature à justifier une conclusion proche de la certitude. Une telle affirmation n'étonnera pas un phonéticien expérimentateur habitué à observer dans ses enregistrements l'évolution inverse r-*h dans l'assourdissement de l'r français en position finale, ou au contact d'une consonne sourde. Il y a un quart de siècle, un grand hebdomadaire parisien rendant compte d'un spectacle de cirque, décrivait les effets comiques qu'un clown tirait de sa guitare : maguitâh ...ma guitâââhh... : à mesure que l'a s'allongeait, l'r final s'assourdissait et sonnait comme un h. A la même épique, les manuels de prononciation allemande à l'usage des Français leur conseillaient volontiers de prononcer le ch de Achtung 'attention' comme l'r de artiste. Ces faits laissent entrevoir, dans certains cas, une égalité pratique, r — h, d'où résulte une convertibilité, une évolution virtuelle en deux sens, r h et h -* r. Cette double évolution, le français actuel, ne possédant pas d'h, ne peut que l'ignorer. Mais que survienne le contact avec une langue étrangère ayant des spirantes vélaires, et la convertibilité apparaît dans les deux sens: l'r sourd du français artiste devient le ch de l'allemand Achtung, et l'A sonore (pour g) du français bagage prononcé par un Suédois devient l'r sonore du français barrage. La convertibilité semble donc pouvoir naître d'une situation où se trouvent en contact deux langues de structure phonétique différente. Tel dut être le cas en Gaule lorsque des scribes de culture uniquement latine fixèrent par écrit et latinisèrent tant soit peu des toponymes prononcés devant eux par des celtophones ou conformément à la tradition celtique. Si d'anciens Néah et Canah de Haute-Bretagne s'y écrivent aujourd'hui Néard et Canard dans un milieu francisé depuis des siècles, il est fort compréhensible que des variantes phonétiques du même toponyme, neh et conoh, soient aujourd'hui continués dans l'ancienne Gaule par Nére ou Νers et Conords ou Conore. En fin de compte, si l'on explique Trôo et Trun par un équivalent gaulois du brittonique tnou, et Νers, Nerville par un équivalent gaulois du moyen-breton (k)nech, on admet par là même que deux façons d'articuler l'r français, l'une apicale et l'autre

DE L'"R" GAULOIS À L'"R" FRANÇAIS

115

dorsale (ou vélaire), sont plus anciennes en France que la langue française elle-même, puisqu'elles seraient l'héritage d'une langue pré-romane. UNIVERSITÉ DE RFNNES

REFERENCES Arbois de Jubainville, H. d', 1890 Recherches sur l'origine de la propriété foncière et des noms de lieux habités en France (Paris, Thorin). Courson, A. de, 1863 Cartulaire de l'abbaye de Redon en Bretagne (Paris). Dauzat, A. et Rostaing, Ch., 1963 Dictionaire étymologique des noms de lieux en France (Paris, Larousse). Falc'hun, F., 1951 Le système consonantique du breton avec une étude comparative de phonétique expérimentale (Rennes, Plihon). 1966a "La doctrine de Joseph Loth sur les origines de la langue bretonne", Revue de Linguistique Romane 30: 324-43. 1966b Les noms de lieux celtiques (première série, Vallées et Plaines) (Rennes, Editions armoricaines). 1970 Les noms de lieux cettiques (deuxième série, Problèmes de doctrine et de méthode, Noms de hauteurs) (Rennes, éditions armoricaines). Hersart de la Villemarque, Th., 1867 Barzaz Breiz: chants populaires de la Bretagne (Paris, Perrin) p. 416, vers 3-4. Holder, 1896 Alt-keltischer Sprachschatz (Leipzig, Teubner) (Réédité à Graz, 1961). Le Roux, Pierre, 1924-63 Altas Linguistique de la Basse-Bretagne. Loth, J„ 1883 L'émigration bretonne en Armorique du Vè au Vllè siècle de notre ère (Paris). 1890 Chrestomathie bretonne (Paris, Bouillon). 1892 Les mots latins les langues brittoniques (Paris, Bouillon). 1907 "Les langues romane et bretonne en Armorique", Revue Celtique 28 : 374-403. Nomenclatures des hameaux, écarts et lieux-dits du Finistère, du Morbihan, des Côtes-du-Nord, de l'Ille-et-Vilaine. 8 vol., 1955, 1954,1953, 1952 (Rennes, I.N.S.E.E.).

GEORGES FAURE

ANALYSE ACOUSTIQUE DE DEUX ALLOPHONES D U I FINAL ANGLAIS

RESUMÉ L'objet de cet article est de présenter une analyse acoustique de deux allophones du / creusé final de l'anglais; le premier en position postvocalique (type: feel), le second en position postconsonantique (type:people), à l'aide du sonagraph, à partir d'enregistrements en chambre sourde sur magnétophone Nagra, à la vitesse de 38 cm/s. Cette étude nous a permis de reposer le problème de la distinction entre voyelle et consonne, tant sur le plan phonétique que sur le plan phonologique. Elle nous a également permis d'éclairer divers problèmes touchant aux effets acoustiques de la coarticulation, non seulement en ce qui concerne des fournitures vocaliques ou consonantiques en contact immédiat, mais encore entre deux fournitures vocaliques, séparées par une fourniture consonantique plus ou moins perméable. Ce travail nous a enfin, et surtout, donnée l'occasion de poser dans une perspective nouvelle les problèmes touchant à la structuration de la syllabe, en faisant intervenir les notions d'éléments intégrants, d'éléments intégrés et de hiérarchie de puissance tant au niveau génétique, qu'au niveau acoustique et au niveau perceptif.

INTRODUCTION

L'analyse acoustique de deux types de réalisation du / creusé de l'anglais — le premier en position postvocalique, le second en position postconsonantique — nous a paru de nature à éclairer, dans une certaine mesure, un ensemble de problèmes touchant à la fois à la COARTICULATION (et à ses effets acoustiques), à la DISTINCTION ENTRE LES N O T O N S DE VOYELLE ET DE CONSONNE e t à l a STRUCTURATION SYLLABIQUE. Elle nOUS a p a r u digne, à ces divers titres, d e retenir l'attention d u linguiste.

Nous avons cherché, en particulier, à déterminer dans quelle mesure ces deux types de réalisation du phonème /// doivent être considérés comme des consonnes, ou au contraire, comme des voyelles, et cela à la fois du point de vue phonétique et du point de vue phonologique. Nous avons voulu, d'autre part, déterminer dans quelle mesure ces réalisations étaient affectées par l'émission vocalique ou consonantique qui les précède. Et nous avons essayé de poser, en termes qui nous paraissent nouveaux, le problème encore mal résolu de la structuration syllabique.

118

GEORGES FAURE

1. CORPUS ET PROTOCOLE EXPÉRIMENTAL

Notre corpus, dont l'analyse expérimentale a été confiée à l'un de nos étudiants spécialisés, M. Spessa, sous le contrôle de notre collaborateur M. Rossi, a été provisoirement limité à 22 réalisations fournies par un locuteur sud-africain de langue et de culture anglaises.1 Ce corpus comprend 11 réalisations postvocaliques (type lì) à savoir: 1. feel 2. fill 3. fell 4. shall

[fi:l] [fil] [fei] [Jœl]

5. snarl 6. doll 7 .ball 8.full

[sna:l] [dol] [bo:i] [ful]

9. fool 10. dull 11. girl

[fu:l] [d/\l] [gart]

Il comprend, d'autre part, 11 réalisations postconsonantiques (type 12), à savoir: 12. 13. 14. 15.

people [pi:pl] bottle [boti] ankle [aeqkt] double [dAbt]

16. handle [hœndl] 17. angle [asrjgl] 18. final [fainl] 19. baffle [baefl]

20. travel [traevi] 21. whistle [hwisl] 22. dazzle [dœzl]

L'enregistrement a été effectué en chambre sourde sur magnétophone Nagra, à ia vitesse de 38 cm/s. Les séquences obtenues ont ensuite été analysées à l'aide du sonagraph (filtre large. Bande de 300 Hz). Sur les documents obtenus (cf. annexe) : 1 mm. en abscisse = 0,0077 seconde. 1 mm. en ordonnée = 80 Hz.

2. RÉSULTATS OBTENUS

2.1 Structuration formantique de L creusé final Un simple coup d'œil jeté sur les 22 spectrogrammes obtenus prouve que dans tous les cas, et quelle que soit la fourniture qui le précède, L final présente une structure formantique évidente. Le tableau 1 donne les valeurs de Fl, F2 et F3 de II en regard de Fl, F2 et F3 de la voyelle qui le précède immédiatement. Le tableau 2 donne les valeurs de Fl, F2 et F3 de 12 en regard de la consonne qui le précède immédiatement. Si l'on calcule la valeur moyenne de Fl, F2 et F3 pour l'ensemble des 22 réalisations on obtient les chiffres suivants: Fl = 495; F2 = 1015; F3 = 2260. Il nous paraît, dans ces conditions, évident que nous avons affaire (du point de vue strictement phonétique) à une fourniture VOCALIQUE, aussi pure que peuvent l'être les réalisations de n'importe quelle voyelle, et singulièrement de chacune des 11 1

II va de soi, que l'étude statistique détaillée que nous envisageons de poursuivre, devra porter sur un corpus nettement plus étendu, notre analyse acoustique devant, d'autre part, être complétée par une analyse parallèle au niveau perceptif et au niveau du comportement fonctionnel.

ANALYSE ACOUSTIQUE DE DEUX ALLOPHONES DU " L " FINAL ANGLAIS

119

TABLEAU 1 Mots

V

Valeur des formants de la voyelle précédant /

Valeur des formants de / final postvocalique (/l)

1. feel

[i]

Fl 400 -

F2 2000 -

F3 2560

Fl 480 -

F2 1200 -

F3 2360

2. fill

[1]

560 -

1680 -

2400

480 -

1200 -

2320

3. fell

[ε]

720 -

1600 -

2400

560 -

1120 -

2400

4. shall

Μ

800 -

1520 -

2440

560 -

1080 -

2320

S. snarl

[α]

680 -

1120 -

2320

480 -

1040 -

2240

6. doll

[0]

640 -

1160 -

2440

480 -

880 -

2200

7. ball

[o]

560 -

960 -

2360

520 -

960 -

2160

8 full

[o]

480 -

960 -

2240

480 -

960 -

2240

9 fool

tu]

400 -

920 -

2240

480 -

1040 -

2160

10. dull

[Λ]

720 -

1360 -

2400

520 -

1120 -

2160

11. girl

[3]

560 -

1280 -

2400

480 -

960 -

2160

TABLEAU 2 Mots

C

Valeur des formants de l postconsonantique (12)

12. people

[ρ]

Fl 480

-

F2 960 -

F3 2320

13. bottle

[t]

520

-

960 -

2280

14. ankle

[k]

480

-

1040 -

2120

15. double

[b]

480

-

960 -

2280

16. handle

[d]

480

-

1040 -

2320

17. angle

tg]

440

-

1000 -

2320

18. final

[n]

520

-

960 -

2400

19. baffle

[f]

480

-

880 -

2160

20. travel

[V]

480

-

1040 -

2280

21. whistle

[s]

480

-

1040 -

2360

22. dazzle

[Z]

520

-

880 -

2160

120

GEORGES FAURE DIAGRAMME 1

Diagramme des voyelles du corpus

voyelles qui précédaient LI dans notre corpus (cf. diagramme 1, où nous avons figuré "par une majuscule en caractère gras" la localisation acoustique de la 'voyelle' II). Il convient maintenant de voir (toujours au même niveau d'analyse) dans quelle mesure la structure formantique de / final peut être influencée par la fourniture (vocalique ou consonantique) qui le précède immédiatement, voire par la fourniture vocalique qui précède la consonne suivie de / final (comme par exemple dans dazzle [daezt] où il n'est pas exclu à priori que les formants de / soient influencés par ceux du [ae] qui précède [z]). Ce travail fait nous serons plus à l'aise pour aborder les deux derniers points de notre recherche, à savoir: le comportement fonctionnel (vocalique ou consonantique) de / final et l'éventuelle pesée de ce comportement sur notre conception de la syllabe. 2.2 Influence de la réalisation phonique (vocalique ou consonantique) qui précède immédiatement l final (1) Influence de la voyelle sur 11 (c'est-à-dire quand l final est postvocalique) : (a) Valeur moyenne des formants de Ll après les 11 voyelles du corpus: Fl = 500; F2 = 1050; F3 = 2250. (b) Valeur moyenne de Fl et F2 de ll en fonction du degré d'antériorité de la voyelle qui le précède: (i) après voyelles antérieures (1 à 4), Fl = 520; F2 = 1150;

ANALYSE ACOUSTIQUE DE DEUX ALLOPHONES D U "L" FINAL ANGLAIS

121

(ii) après voyelles postérieures (5 à 9), Fl = 488; F2 = 976. (c) après voyelles centrales (10-11), Fl = 500; F2 = 1040. Nous constatons à l'analyse de ces chiffres que: 1. après les voyelles antérieures: F2 de /I est supérieur à F2 moyen, de 9,5 p. 000; Fl de /I est supérieur à Fl moyen de 4 p. 000; 2. après les voyelles postérieures: F2 de /I est inférieur à F2 moyen de 7 p. 000; Fl de /I est inférieur û Fl moyen de 2,5 p. 000; 3. après les voyelles centrales: F2 de /I est pratiquement égal à F2 moyen; Fl de /I est égal à Fl moyen. Il semble donc qu'il y ait une certaine influence, non négligeable, de la voyelle sur II, en particulier au niveau de F2: APRÈS UNE VOYELLE ANTÉRIEURE LE F2 S'ÉLÈVE. APRÈS UNE VOYELLE POSTÉRIEURE LE F 2 S'ABAISSE.

Ceci peut s'expliquer par le fait que les voyelles antérieures ont un F2 beaucoup plus élevé que les voyelles postérieures (cf. Diagramme 1, Diagramme des voyelles). Il est d'ailleurs significatif de constater que la plus grande valeur du F2 de /I se trouve après [i] et [i] qui sont les deux voyelles les plus antérieures de notre corpus, donc celles dont le F2 est le plus élevé. (c) Valeur moyenne de Fl et F2 de /I en fonction du DEGRÉ D'APERTURE de la voyelle qui le précède: (i) après les voyelles fermées ou mi-fermées (1-2-7-8-9-11), 2 Fl = 486; F2 = 1053; (ii) après les voyelles ouvertes ou mi-ouvertes (3-4-5-6-10) Fl = 520; F2 = 1048. Nous constatons que: (i) après les voyelles fermées: Fl de II est inférieur à Fl moyen de 3 p. 100; F2 de il est égal à F2 moyen; (ii) après les voyelles ouvertes: Fl de II est supérieur à Fl moyen de 4 p. 100; F2 de 11 est égal à F2 moyen. H semble donc qu'il y ait une légère influence de la voyelle, selon son degré d'aperture, au niveau du Fl de /I : APRÈS UNE VOYELLE FERMÉE LE F1 A TENDANCE À S'AFFAISSER. APRÈS UNE VOYELLE OUVERTE LE F1 A TENDANCE À S'ÉLEVER. Il est, à cet égard, intéressant de constater que la plus grande valeur du Fl de /I apparaît nettement après les voyelles [ε] et [ae] qui sont les voyelles les plus ouvertes du corpus, donc celles dont le Fl est le plus élevé (cf. Diagramme 1, Diagramme des voyelles, et le tableau 1). (2) Résumons nos constatations: 1. / a une structure formantique parfaitement nette. Ses trois formants apparaissent clairement sur tous les spectrogrammes. / est donc phonétiquement une voyelle. Or, Faure (1958:70-73), Lehiste (1964:10-50), et Heffner (1960:72-80) affirment que / ressemble fort, sur le plan PERCEPTUEL, à la voyelle [υ] de l'anglais put. Qu'en est-il exactement? Si nous nous reportons au tableau 1, case 8, nous lisons full: [u]: Fl = 480 F2 = 960 F3 = 2240 9

/I : Fl = 480 F 2 = 960 F3 = 2240

Les nombres en parenthèses se rapportent aux numéros des spectogrammes.

122

GEORGES PAURE

ce qui signifie que les formants de / prolongent EXACTEMENT ceux de u. De plus (car ce phénomène pourrait n'être dû qu'à l'influence de [u] sur /l) si nous comparons les valeurs moyennes de / (Fl = 495, F2 = 1015) et de [u] (Fl = 480, F2 = 960) nous sommes obligé de constater qu'effectivement / est presque l'équivalent de [u]. Il faut préciser, d'autre part, qu'on ne voit apparaître sur aucun des 22 spectrogrammes du corpus une quelconque 'zone de bruit' qui correspondrait à la fourniture aléatoire qui est le trait acoustique spécifique de la consonne phonétique. 2. Si / est une voyelle, dans les séries voyelle: + / , tout doit se passer comme si nous émettions des diphtongues. Il doit y avoir glissement progressif d'une voyelle vers une autre voyelle. Or, nous constatons précisément que dans tous nos spectrogrammes il y a CONTINUITÉ ABSOLUE entre les formants de la voyelle et ceux de la 'voyelle /'. Dans feel Fl et F2 se rapprochent de plus en plus l'un de l'autre. Ce rapprochement s'accompagne d'un affaiblissement de l'intensité globale comme c'est toujours le cas pour les diphtongues décroissantes de l'anglais : "Le premier élément d'une diphtongue de ce type est toujours prononcé avec beaucoup plus de force que le dernier" (Faure, 1958:30). Ce phénomène est évident dans tous les spectrogrammes de 1 à 11, où le dernier élément de la diphtongue est constitué par / final. (3) Influence de la consonne sur 12 (c'est-à-dire quant 1 final est postconsonantique) 1. Valeur moyenne des formants de 12 après les 11 consonnes du corpus : Fl = 490 ; F2 = 980; F3 = 2270. Si nous comparons ces valeurs avec celles des formants de II (500-1050-2250) nous constatons que: a. Fl de L2 est très légèrement inférieur à Fl de /I (2 p. 100) b. F2 de L2 est très nettement inférieur à F2 de /I (6,6 p. 100) Cette dernière remarque permet d'expliquer la dénomination de "over-dark variety of L" donnée par Ward (1944) au / postconsonantique. Rappelons avec Malmberg (1966): 15) que si nous prononçons le série [i] [y] [u] en français, le Fl reste invariable, tandis que le F2 baisse progressivement: la voyelle [i] a ainsi un timbre clair (ou aigu) tandis que la voyelle [u] a un timbre sombre (ou grave). Or, ici, en comparant /I et 12 nous voyons que: F1 BAISSE LÉGÈREMENT, DONC QUE 12 EST UN PEU PLUS FERMÉ QUE 11.

F 2 BAISSE PLUS NETTEMENT, DONC QUE 12 EST PLUS SOMBRE (OU PLUS GRAVE)

QUE 11.

2. Valeur moyenne de Fl et de F2 de 12 en fonction du degré de fermeture de la consonne qui le précède: (a) après les six occlusives (12 à 17), Fl = 480; F2 = 990; (b) après les quatre constrictives (19 à 22), Fl = 490; F2 = 960; (c) après la nasale (18), Fl = 520; F2 = 960. Il ressort de la comparaison de ces valeurs, que: LE F2 DE 12 EST STABLE. L E F1 DE 12 CROÎT QUAND NOUS PASSONS DES OCCLUSIVES AUX CONSTRICTIVES PUIS AUX NASALES.

Cela signifie que L2 s'ouvre de plus en plus. Or le classement des consonnes d'après leur degré d'aperture est le suivant: degré 0: occlusives

ANALYSE ACOUSTIQUE DE DEUX ALLOPHONES DU " L " FINAL ANGLAIS

123

degré 1 : constrictives degré 2: nasales Il semble donc qu'il y ait une légère influence des consonnes, quant à leur mode d'articulation, sur l'aperture de 12 qui les suit, c'est-à-dire au niveau du Fl. 3. Valeur moyenne de Fl et de F2 de 12 en fonction du lieu d'articulation de la consonne qui le précède : (a) après les bilabiales (b-p), Fl = 480; F2 = 960; (b) après les ladiodentales (f-v), Fl = 480; F2 = 960; (c) après les apico-postalvéolaires (t-d), Fl = 500; F2 = 1000; (d) après les prédorso-postpalatales (s-z), Fl = 500; F2 = 960; (e) après la dorso-palatale (ri), Fl = 520; F2 = 960; (f) après les dorso-palatovélaires (k-g), Fl = 460; F2 = 1020. Il apparaît clairement que les variations des formants de 12, en fonction du lieu d'articulation des consonnes qui le précèdent, sont très faibles, et ne sont révélatrices d'aucune tendance nettement marquée. Aucun enseignement ne semble devoir être tiré de ces résultats. (4) Résumé: Nous pouvons dire en résumé, que / final est relativement stable, en particulier quand il est précédé d'une consonne. Les consonnes ne rendent / qu'un peu plus 'sombre'. En position postvocalique, le timbre de / est, par contre, plus nettement influencé par celui de la voyelle. Les différences entre 12 et 11, du point de vue quantitatif (niveau des formants), sont assez minimes. Mais du point de vue qualitatif, c'est-àdire en ce qui concerne le modelé des transitions formantielles, elles sont plus importantes : A l'examen des spectrogrammes (cf. documents), une différence fondamentale apparaît entre les séquences 'voyelle + V (1 à 11) et les séquences 'consonne + /' (12 à 22). Les premières constituent une unité homogène sans interruption dans la structure, alors que dans les secondes il y a une coupure nette, une interruption avant le /. Lehiste l'avait préssenti lorsqu'elle écrivait à propos de la séquence : "the phonemic sequence /kl/ does not seem to constitute an integral, close-knit unit: the allophone of /k/ depends on the preceding syllable nucleus, whereas the allophone of /I/ is essentially independent." (1964:24). Sur nos documents, nous pouvons vérifier sans peine qu'il en est bien ainsi, alors que pour /I se manifeste une continuité absolue du début à la fin de chaque séquence; ce qui nous invite à traiter la séquence 'V + / final' comme une diphtongue décroissante; par contre dans la plupart des séquences 'C + / final' (12-13-14-15 notamment), les transitions entre les sons précédant / sont nettes, alors que / apparaît indépendant avec ses formants plats et sans aucun lien avec ce qui le précède. Toutefois nous pouvons noter que pour 18 :final,20 : travel, 22 : dazzle, les formants de 12 viennent manifestement de la voyelle qui précède [n], [v] et [z]. Dans les trois cas, c'est surtout la transition du F2 de la voyelle qui pénètre le mieux la consonne pour venir rejoindre le F2 de /. Y a t-il ime explication à ces trois "excep-

124

GEORGES FAURE

tions" apparentes? Notons que η est une nasale tandis que ν et ζ sont des constrictives sonores. Or, Malécot (1955) nous donne un classement des consonnes d'après leur degré de force d'articulation (c'est-à-dire en fonction de la quantité d'énergie requise pour leur émission) qui s'établit comme suit (de la plus forte à la plus faible) : degré degré degré degré degré

1 : occlusives sourdes 2 : occlusives sonores 3 : constrictives sourdes 4: constrictives sonores 5 : nasales

Il est intéressant de constater que nos trois consonnes, [n], [v] et [z], sont les moins fortes du système et sont donc susceptibles d'être plus facilement pénétrées par les éléments qui les environnent, en particulier les voyelles. Ceci peut expliquer, selon nous, le fait que les formants de [ai], [ae], [ae] pénètrent aussi profondément les consonnes [n], [v], [z] pour rejoindre et influencer les formants de l qui vient en fin de séquence. On peut donc affirmer que, mis à part les trois cas que nous venons d'étudier, 12 est essentiellement indépendant de ce qui le précède. Alena Skalicková, citée par Malmberg (1961), souligne que les liens entre les sons à l'intérieur d'une syllabe doivent être plus étroits que ceux des sons de deux syllabes distinctes. Dans la plupart de nos séquences, les liens entre 12 et la consonne précédente sont inexistants. Donc, si 12 a une structure formantique qui fait de lui un élément vocalique (un peu moins intense qu'une voyelle toutefois) et si 12 est indépendant de ce qui le précède, nous pouvons en conclure que dans les séquences ' V + C + L', L constitue à lui seul une unité phonétique. C'est la raison pour laquelle 12 est généralement dénommé 7 syllabique' par la plupart des phonéticiens. Heffner (1960:114), entre autres, signale que "the liquid consonants frequently serve as syllabic elements". De même, pour toutes les séquences terminées par 12 étudiées ici, Jones (1954) donne la transcription suivante: [pi:p/l], [bot/1], [fain/1], [wis/1] ... en signalant bien la frontière syllabique entre la consonne et / final.

3.3 Le problème de la syllabe

Les constatations, que nous venons de faire, nous semblent poser en termes relativement nouveaux le problème de la syllabe. Nous ne reviendrons pas ici sur les multiples définitions déjà données de ce complexe sonore dont on s'est plu, à juste titre, à souligner la profonde cohésion. Nous ne retiendrons, à la suite de Malmberg et de divers autres chercheurs, que la déclaration (fructueuse et digne d'être creusée) selon laquelle une syllabe est un complexe sonore dont tous les éléments sont en état de 'fusion avancée'. Cette image, qui correspond à une réalité certaine, mérite d'être approfondie car elle pourrait bien se révéler d'une fécondité exceptionnelle tant sur

ANALYSE ACOUSTIQUE DE DEUX ALLOPHONES DU " L " FINAL ANGLAIS

125

le plan de la recherche fondamentale que sur celui de la linguistique appliquée (par exemple, dans l'enseignement des langues, en orthophonie, etc.). Le fait que la GRAPHIE signale à l'œil des unités distinctives (trois unités dans le mot par, deux unités dans le mot si, etc.) a pu amener certains à conclure trop hâtivement que la syllabe concrète, telle qu'elle se réalise, était formée d'éléments plus ou moins distincts qui, certes, interféreraient les uns avec les autres, s'influenceraient réciproquement, mais garderaient néanmoins un minimum d'indépendance. Or, il est bien certain que, lorsqu'on enseigne une langue vivante étrangère ou lorsqu'on cherche à redresser un défaut d'articulation, les unités minimales irréductibles dont on se propose d'assurer la réalisation sont les syllabes dont les graphies complexes ne marquent que les COLORATIONS SUCCESSIVES. C'est, en d'autres termes, à la dynamique verbale qui modèle L'UNITÉ IMPULSIONNELLE de la syllabe qu'il convient avant tous de nous attacher. Notre propos n'est pas de nous étendre sur ce point mais plutôt de nous demander, à la lumière de ce qui précède, ce qu'il faut entendre par 'fusion avancée'. Une fusion (nous parlerions volontiers d'intégration et même, pour poursuivre l'image, de 'digestion') suppose UN ÉLÉMENT INTÉGRANT ET UN OU PLUSIEURS ÉLÉMENTS INTÉGRÉS. Et l'intégration génératrice de l'unité syllabique nous semble être fonction du rapport de force entre intégrant et intégré. Il se trouve que la présence de 'vocalité' dans une fourniture phonique confère à cette fourniture une force d'intégration qui lui permet d'assimiler les fournitures non vocaliques qui lui sont contigiies. Les voyelles sont à ce titre privilégiées mais les consonnes dites syllabiques (et singulièrement L creusé final que nous venons d'étudier) sont également pourvues d'une force d'intégration, proportionnelle, nous semble-t-il, à leur pureté vocalique et qui leur permet de s'assimiler les fournitures non vocaliques (ou simplement moins purement vocaliques) qui les entourent; à moins que ces dernières soient absorbées par une fourniture vocalique plus puissante et immédiatement contiglie, laquelle peut même absorber deux unités successives de moindre puissance qu'elle. Par exemple, dans le mot français église la dynamique syllabique croissante de notre langue fait que le [i] de la deuxième syllabe absorbe à la fois les deux consonnes sonores qui la précèdent et la consonne sonore qui la suit. La voyelle de l'unique syllabe du mot français peuple ([pœpl]) absorbe à la fois les deux consonnes sourdes [p] initial et [p] postvocalique et la consonne [1] qui suit cette dernière. Par contre, dans le mot anglais people ([pi:pl]) où nous avons deux syllabes, [i] final creusé (à cause précisément de la force d'attraction que lui confère sa pureté vocalique) résiste à l'attraction de [i:] et devient du même coup noyau syllabique. Il ne s'agit là, on le voit, que d'une hypothèse de travail, dont nous poursuivons la vérification, mais qui nous paraît ouvrir des perspectives intéressantes. Il nous semble, en particulier, que si les choses se passent bien ainsi, le débat qui a opposé si longtemps et qui oppose encore certains théoriciens de la syllabe, comme Hala (1961) et Rosetti (1963), n'aurait plus d'objet. Si notre hypothèse de départ est exacte, l'intégration syllabique n'implique en effet nullement qu'il y ait présence de

126

GEORGES FAURE

vocalité dans un noyau syllabique en puissance; la 'vocalité' n'étant qu'un élément favorable, présent il est vrai (qu'il s'agisse de voyelles ou de consonnes) dans la plupart des noyaux syllabiques. Mais des syllabes commepstt\ ([pst]) ou chui\ ([Jt]) etc., s'expliqueraient parfaitement par le déséquilibre des forces entre la constrictive [s] et les occlusives [p] et [t] (dans le premier mot) ou entre la constrictive [J] et l'occlusive [t] (dans le deuxième mot). Le rapport des forces qui, à un moment donné par exemple, segmente un monosyllabe en deux syllabes est, d'autre part, très instable. Il s'agit, là encore, de marges et de seuils qu'il conviendrait d'étudier à la fois au niveau génétique, au niveau acoustique et au niveau perceptif. Nous avons procédé à cet égard à des tests sommaires mais instructifs à propos du mot anglais fuel (transcrit: [fjual] et [f)u:l] par Jones (1954)). Prononcé par un français qui ignore l'anglais, ce mot devient [fysl] avec deux syllabes à noyau vocalique très pur. Viennent ensuite la première version anglaise de Jones [fjual] où c'est la noyau vocalique de la deuxième syllabe qui se neutralise et s'affaiblit en [a] et enfin [f)u:l] où [1] creusé final forme syllabe à lui seul. Et nous atteignons un seuil au delà duquel le disyllabe se réduirait à un monosyllabe: [f)u:l] (avec un / français apical et non creusé), prononciation qu'il nous est arrivé d'entendre dans la bouche de français un peu snobs qui singent maladroitement la prononciation anglaise. On a donc, d'étape en étape: [fyel] -»· [fjual] -»· [f)u:l] [f)u:l]. La prononciation [fluì], monosyllabique, est à comparer avec [fqel], qui serait en français la réduction monosyllabique du dissyllabe [fy/εΐ]. Rien ne nous empêche d'imaginer et de réaliser plusieurs types intermédiaires entre les deux derniers de la série qui nous font basculer du dissyllabe au monosyllabe; l'inertie de l'oreille (et son manque d'aptitude aux discriminations dynamiques assez fines) fait que seule la distinction [f}u:l] / [f)u:l] facilement réalisable et facilement perceptible soit communément réalisée.

4. CONCLUSION

Si limitée soit-elle, puisqu'il ne s'agit en fait que d'une toute première étape, la présente étude nous a permis de montrer que / creusé final de l'anglais est une voyelle, très voisine de la voyelle [u]. s Cela nous confirme dans notre habitude d'enseigner aux anglicistes francophones, qui nous sont confiés, qu'ils sont beaucoup plus proches de la perfection s'ils prononcent [WEU] (pour le mot well) plutôt que [wel]. Ils ont même toutes les chances d'être pris pour des britanniques s'ils donnent au [u] final de ce mot la coloration qui est la sienne dans un mot comme look. Nous avons pu illustrer une nouvelle fois l'importance des transitions dans le continuum phonématique. Nous ne sommes pas loin de penser à cet égard, que le mot même de TRANSITION (ou de 'passage') qui implique un minimum de TENUE * Elle est, en fait, nettement plus ouverte, et donc plus pure, que notre /u/ français.

ANALYSE ACOUSTIQUE DE DEUX ALLOPHONES DU " L " FINAL ANGLAIS

127

est impropre. Nous pensons, en fait, que le continuum phonématique est aussi parfait que le continuum prosodique. La pensée dégage de ces deux continuums des unités définissables, localisables et commutables dont les réalisations impliquent des marges de tolérances beaucoup plus étroites lorsqu'il s'agit d'intonèmes que lorsqu'il s'agit de phonèmes. Nous avons par ailleurs vérifié le fait que les termes de "overdark L" et de "syllabic L" sont justifiables en termes acoustiques à partir d'un corpus limité. Nous avons essayé de dégager de nouvelles perspectives — qui nous paraissent présenter un certain intérêt — dans le domaine de l'analyse génétique, acoustique et fonctionnelle de la structuration syllabique. Nous savons qu'il reste beaucoup à faire dans ce domaine et il est bien évident que ce que nous venons de suggérer ne constitue que l'amorce d'une vaste recherche que nous n'avons pas l'outrecuidance de mener seul à son terme. INSTITUT DE PHONÉTIQUE UNIVERSITÉ DE PROVENCE (CENTRE D'AIX)

RÉFÉRENCES Faure, G., 1958 Manuel pratique d'anglais parlé, nouvelle édition (Paris, Hachette) pp. 29, 30, 70-73. Hala, Β., 1961 "La syllabe, sa nature, son origine et ses transformations", Orbis 10: 69-143. Heffner, R.-M.S., 1960 General Phonetics, troisième édition (Madison, The University of Wisconsin Press) pp. 7280, 114. Jones, D., 1954 English Pronouncing Dictionary, dixième édition (London, J. M. Dent and Sons, New York, E.P. Dutton and Co). Lehiste, I., 1964 Acoustical Characteristics of Selected English Consonants (Bloomington, Ind., Indiana University, La Haye, Mouton) pp. 10-50. Malécot, Α., 1955 "An Experimental Study of Force of Articulation", Studia Linguistica 9 : 35-44. Malmberg, Β., 1961 "A Recent Contribution to the Problem of the Syllable", Studia Linguistica 15.15: 80-88. 1966 La phonétique (Paris, P.U.F. [Que sais-je?]). Rosetti, Α., 1963 Sur la théorie de la syllabe, deuxième édition, refondue et augmentée (La Haye, Mouton). Ward, I.C., 1944 The Phonetics of English (Cambridge. Heffer).

DAVID FELDMAN

ON UTTERANCE-FINAL [1] AND [u] IN PORTUGUESE

0. INTRODUCTION

Standard descriptions of the overall pattern of contemporary spoken Brazilian Portuguese normally assign to the voiced alveolar lateral /I/ two allophones : [1] when initial or medial in syllables or when final in syllables immediately preceding a vowel in close transition; and [1] when syllable final or when utterance-final (Feldman, 1967; Hall, 1943 ; Reed and Leite, 1943). The articulatory distinction between the allophones [1] and [1] is characterized by retraction and absence or optionality of apico-alveolar occlusion in [i]. Similarly, to the syllabic phoneme /u/ four allophones are assigned: [u] when stressed, except before /l/ in the same syllable; [ij] when unstressed, except before /l/ in the same syllable; [U] when either stressed or unstressed before /l/ in the same syllable; and [u] unstressed after a vowel in the same syllable to form a diphthongal nucleus. Thus, in utterance-final position after a vowel, we would expect to encounter the [1] allophone as the realization of /I/ and the [u] allophone as the realization of /u/. There is, however, a marked tendency in the rapid, unguarded speech of speakers of Brazilian Portuguese to articulate utterance-final /l/ and /u/ after a vowel so that they are indistinguishable from each other to the average hearer. Three questions immediately arise. First, have the phonetic norms of [1] and [u] merged so that, in fact, no distinction is made or perceived? Second, if such merging has taken place, are the utterance-final shapes now described as /-(c) Vl/ and /-(c) Vu/ to be considered no longer phonemically contrastive? Third, will this imply that formerly minimal pairs such as riu:rio, in which the first member was traditionally viewed as monosyllabic and the second disyllabic, are now exclusively monosyllabic? Brazilian linguists and grammarians have long acknowledged the problem and have suggested that in rapid, unguarded speech any distinction between utterance-final β/ and /u/ is, in fact, lost: "O ([!]) final é proferido relaxado, quase velar, mas tendo-se o cuidado de näo fazê-lo igual a ú" (Bechara, 1963). Similarly: "em posiçâo pósvocálica, porém [as líquidas] tendem a cair" (Elia, 1963). In the experimental study summarized below, we sought answers to these three

130

DAVID FELDMAN

questions. Because our main concern has been to improve the detail of pedagogicallyoriented phonological descriptions of Brazilian Portuguese, this presentation is limited to a summary of the results of the experiments and does not attempt to include a full statement of all the raw data nor to analyze in detail the research methodology employed. 1. THE EXPERIMENT

Native adult informants, both male and female, from four major dialect areas of Brazil1 were first asked to aurally identify a series of minimal or analogous-environment pairs designed to reveal potential contrasts between utterance-final [1] and [u] after a vowel.2 No informant was able to identify correctly a sufficient number of ISOLATED items taken from the pairs to suggest more than chance guessing as to which of the utterance-final phones was being used. In other words, the informants were unable to decide consistently or unanimously which word they were hearing when the component words of minimal or analogous-environment pairs, such as meu/mel, céu/(pin)cel, tal/tau, viljviu, véuj{prova)vel, and sol/sou, were pronounced in isolation and in random sequence by native speakers via tape recording. This lack of uniformity in recognition was somewhat less prevalent when [I] or [u] occurred in interior syllable-final position, e.g., (farma) cêutico/céltico. This confusion was all the more interesting because, in a pair such as vil/viu, the 'ideal' realizations of the stressed vowels are [i] and [i], respectively, a clearly perceptible difference in Brazilian Portuguese. Four possible explanations of the phenomenon of aural confusion might be offered at the outset : (1) that the speakers modelling the utterances in fact made no articulatory distinction between /l/ and /u/ in the critical positions; (2) that although some articulatory differentiation was made, it was so slight that the hearers were unable to detect it systematically; (3) that utterances of this type in normal speech with /u/ as the final sound are always articulated as monosyllables ; or (4) that the common tendency of speakers to replace the phonetic norm of some vowels with slightly lower variants when those vowels are followed in the same syllable by /l/ or /u/ affected in some measure the informants' recognition of the items, e.g., the [e] of farmacéutico vs. the [ε] of céltico. A simple election of one of these possible initial explanations was further complicated by the fact that some informants were able to recognize immediately some of the items modelled by some of the speakers when the minimal and analogous-environment pairs were read as pairs, suggesting that, at least for some speakers or hearers, the pairs were contrastive and not homophonic. 1 The dialect areas represented, using the subdivisions proposed by Elia (1963: 309-310) are: (a) área cearense (informant native of Fortaleza); (b) área fluminense (informants natives of Rio de Janeiro); (c) área caipira (informants natives of Bauru and Goiânia); (d) city of Säo Paulo (informants natives of the city of Säo Paulo). * Here we use Kenneth Pike's distinction (1943: 73ff.) between minimal pairs in which only one segment is in contrast and pairs which are not identical beyond the contrasting segment.

ON UTTERANCE-FINAL [1] AND [ y ] IN PORTUGUESE

131

In this experiment, the minimal and analogous pairs used were both authentic and invented, the latter procedure being necessary because the language offers relatively few potentially minimal contrastive pairs of this type. Moreover, the primary basis of selection of pairs had to be based on orthography, since the phonetic data themselves were in question. Thus, the results obtained from the preliminary experiment could not be considered conclusive. Admittedly, not only did the authentic pairs tend statistically to be analogous rather than minimal, but it was also possible that the hearers' successful distinction between the utterances of some pairs was really based on the recognition of cues contained in the utterances elsewhere than in the supposedly contrasting segments, e.g., in an open vs. closed vowel preceding the syllable-final /1/or /u/. In order to resolve the problem, a more rigorously controlled second experiment was conducted. It was designed first to help pin down whatever acoustic and articulatory differences there may be between the two sounds in utterance-final position after a vowel; second, to examine the syllabic status of the /u/ in such utterances; and third, to suggest any possible need for revisions in existing descriptive statements regarding these articulations. For the experiment, a spectrographic study of the realization of orthographic syllable-final / and u was followed by a limited fluoroscopic study of the same articulations (see the spectrograms appended to this article). The undeniable significance of spectrographic information on certain features of the physiology of speech was made clear in 1951 by Pierre Delattre (1951). The decision to use limited fluorographic data in addition to the spectrographic information was for the purpose of gathering further information on the measurement of the oral and pharyngeal cavities. In the spectrographic study, the target items were 'hidden' in utterances of no more than 2.4 seconds duration and which always presented the target word in the final, and prosodically stressed, position.8 Other utterances presented occurrences of /u/ and β/ in other positions in order to provide a basis of comparison. The informants were unaware of the objectives of the experiment so that in modelling the utterances for the spectrograph as natural a quality of speech as possible was assured. 2. u: PRELIMINARY OBSERVATIONS

Allowing for expected variations in voice timbre among the several informants, the spectrographic realization of /u/ in stressed and unstressed (both pre- and post-tonic) positions other than syllable-final after a vowel was as expected. Formant 1 (Fl) averaged 322 and Formant 2 (F2) 740. The still-film records of the fluoroscopic study confirmed the corresponding articulatory features: 4 '

For example: Di-me o meu vs. Dê-me o mei. But because normal, unguarded speech was the desired norm for the modelling by informants, it was often difficult to pinpoint the precise moment at which the 'characteristic' organic position of the articulation was achieved. Again, because of the admittedly pedagogical focus of this summary, the figures and tabulations presented must be viewed as having a relative, rather than an absolute, validity. 4

132

DAVID FELDMAN

(1) Labialization: Sufficient constriction of the orbicularis oris occurred to reduce the lip orifice to an average of 1.6 cm. but without significant lip protrusion. (2) Tongue position:5 The dorsum was uniformly high. The apicum was characteristically lowered to the level of the base of the lower teeth, often retracted some distance behind the gums. A very slight convexity of the laminum was also noted. (3) Nasopharynx: firmly closed. (4) Epiglottis: almost vertical. (5) Pharyngeal cavity: Open, as observed by measurement of the cross-section of the pharyngeal cavity. Average width: 9.5 mm. (6) Jaw position: generally lowered with only the slightest retraction; distance between upper and lower teeth averaged 1.2 cm. (7) Larynx: lowered. (8) Duration: average .141 sec.

Fig. 1

When /u/ was articulated in unstressed utterance-final position after a consonant, no significant differences from the above were noted, save expected variations in the duration of the sound and in the microwattage of phonetic power. When /u/ was articulated in unstressed utterance-final position after a preceding vowel, however, some significant changes were noted. The spectrograms revealed an average Fl of285, with no appreciable change in F2, signifying a lowering of tongue height, but with only a slight tendency toward forward placement. The radiographs confirmed the following: (1) Labialization: slight, no protrusion. (2) Tongue position: As in [u], but with a slight lowering of the dorsum. (3) Nasopharynx: Firmly closed. (4) Epiglottis : Angled with tendency toward posterior pharyngeal wall. (5) Pharyngeal cavity: Reduction in size averaging 2.5 mm. (6) Jaw position: slightly lower than in [u], but noticeably more retracted. (7) Larynx: slightly higher than in [u]. 6

Delattre (1951) has shown that the measurement of tongue height alone is less significant in the determination of sound production than the totality of the actual back-and-up tongue retraction. This feature, as we have also noted, is intimately linked to the lowering and retracting action of the lower jaw.

ON UTTERANCE-FINAL [1] AND [y] IN PORTUGUESE

133

(8) Duration: average .117 sec. No significant departures from these general norms were noted among the informants.

Fig. 2

3. PRELIMINARY OBSERVATIONS

Portuguese /I/ in absolute initial and interior syllable-initial positions is clearly a gingival lateral, unaccompanied by labialization or nasalization. The apicum makes a medial occlusion against the upper gums. The sides of the tongue are relaxed so that the breath stream flows between the teeth near the first premolars. The spectrograph^ realization of /l/ in these positions is as expected: showing the harmonic source characteristics of vocoids and the zeros in their spectrum envelopes characteristic of contoids. Its Fl averaged 250, analogous to that of /u/. Its F2 value however, varied consistently with that of the adjacent vowel, e.g., c. 1200 when preceded by /a/, but with a high frequency spread reaching as high as 2000. (1) Labialization: insignificant. (2) Tongue position: laminum elevated with tendency toward concavity; no corresponding elevation of the dorsum. The apicum is vigorously raised to definite gingival contact. (3) Nasopharynx: closed, with moderate sustained muscular pressure on the velum. (4) Jaw position: generally lowered with slight retraction; distance between upper and lower teeth averaged .75 cm. (5) Larynx: higher than in /u/. (6) Duration: average .10 sec. No significant departures from these general norms were noted among the informants. When /I/ was articulated in utterance-final position following a vowel, however, significant changes are noted. Most notable, and consistent in all informants, is the elevation of the back of the dorsum of the tongue in the prevelar region. Informants from the city of Säo Paulo showed the following characteristics: (1) Labialization: slight; no protrusion. (2) Tongue position: humping of laminum with considerable elevation of the dorsum in the prevelaT region. Apicum is held somewhat tensely, tending toward, but not

134

DAVID FELDMAN

Fig. 3

achieving, gingival or alveolar occlusion. Occasionally, sufficient laxness of muscular tension occurs, permitting the lower surface of the apicum to contact the upper surface of the lower teeth. (3) Nasopharynx: vigorously closed, with sustained muscular pressure on the velum. (4) Jaw position : lowered and retracted. (5) Larynx: slightly lower than /l/ in syllable-initial position. (6) Duration: average .095 sec. Informants from the interior of Säo Paulo state, however, differed in the following categories: (1) Labialization: none. (2) Tongue position: humping of laminum with approximately the same degree of elevation of the dorsum in the prevelar region as in [u]; apicum held very tensely and raised to alveolar occlusion in a majority of instances tested. (3) Larynx: slightly lower than for /I/ in syllable-initial position. The informants from Fortaleza and Rio de Janeiro differed from the ones above in that the degree of elevation of the dorsum was greater and the apicum was lax, often assuming a low position behind the lower teeth.

Fig. 4

4. SYLLABIC CONSIDERATIONS

In none of the pairs used in the experiment did informants create hiatus groups composed of the vowel followed by /u/. The unanimous tendency was toward diphthong-

ON UTTERANCE-FINAL [1] AND [ y ] IN PORTUGUESE

135

ization. In a test of the duration of utterances, measurements were recorded for all pairs. In the first run-through the form ending in orthographic I was first. In the second, the form ending in orthographic u was first. In the third, the pairs were mixed in random sequence. The average duration of the forms ending in orthographic / was .397 sec., while the duration of those ending in orthographic u was .454 sec. Because we have noted that utterance-final /u/ after a vowel in utterances of the type we examined is treated as the coda of a falling diphthong rather than as a hiatic syllable, we feel justified in using the symbol [u] to represent it phonetically in the remainder of the discussion."

5. SUMMARY OF COMPARISONS OF [1] AND [u] [U]

(1) Labialization: (2) Tongue position

(3) Nasopharynx:

(4) Jaw position:

(5) Larynx: (6) Formant:

(7) Duration:

rounding with minimal protrusion. elevation of dorsum, lowering of apicum, gentle convexity of laminum. firmly closed, with sustained muscular pressure on velum. lowered (with an average of 1.5 cm. distance between upper and lower teeth); slight retraction. lowered. Ft ave. 322 F2 ave. 740 Fi ave. 1200

average .141 sec.

[1]

no rounding or protrusion. median position of dorsum, elevation and definite convexity of laminum, apicum raised to alveolar or gingival occlusion. closed, but with only moderate muscular pressure on velum. lowered (with an average of .80 distance between upper and lower teeth); definite retraction. slightly raised. Ft ave. 250 Fi consistently varies with second formant of adjacent vocoids. F3 ave. c. 1800 average .10 sec.

The lower average of Fl frequencies for [1] would indicate a generally smaller overall dimension of the oral tract then in [u]. This is confirmed by actual measurement in which the more posterior position of the dorsum and greater labialization in [u] ' This procedure is fully justified by the standard analyses of contemporary Brazilian Portuguese in which the pairs we have used are consistently listed as containing diphthongs as the final syllable. Cf. for example Pereira (1957: 26ff.)

136

DAVID FELDMAN

creates a larger oral cavity volume than we find in [1]. Because of the difficulty in achieving a second formant for [1] in adjacency to a vowel, the evidence of F2 as an indicator of back-to-front tongue placement is not precise. The lowering of F3 in [1], however, clearly reflects the apical occlusion. 6. SUMMARY OF COMPARISONS BETWEEN /l/ AND /u/ IN UTTERANCE-FINAL POSITION AFTER A VOWEL

Maximal differentiation among informants:7 M

(5) Larynx:

noticeable, but not intense, minimal protrusion moderate elevation of dorsum, lowering of apicum, gentle convexity of laminum. firmly closed approximate average of 1.0 cm. distance between upper and lower teeth; no retraction lowered

(6) Duration:

average .117 sec.

(1) Labialization:

(2) Tongue position:

(3) Nasopharynx: (4) Jaw position:

ra slight; no protrusion

moderate elevation of dorsum, concavity of laminum, tense apicum but without occlusion firmly closed approximate average of 1.0 cm. distance between upper and lower teeth; slight retraction very slightly lowered (higher than /u/) average .095 sec.

The consistently lower first formant in both sounds, as compared with [u] and [1], confirmed the generally smaller overall opening of the oral tract. The lower average of the second formant in [u] reflects the greater posterior placement of the tongue and the greater degree of labialization as compared with [1], The significantly lower third formant average of [1], however, confirms the presence of at least the vestigial tendency to raise the apicum toward occlusion. Minimal differentiation among informants:8 (1) Labialization: (2) Tongue position:

7 8

[y]

very slight; no protrusion moderate elevation of dorsum; lowering of apicum to position behind lower teeth; gentle convexity of laminum.

[i]

very slight; no protrusion moderate elevation of dorsum; lowering of apicum to position behind lower teeth; gentle convexity of laminum.

Revealed by informants from the city of Säo Paulo, the interior of Säo Paulo State, and Goiás. Revealed by informants from Fortaleza and Rio de Janeiro.

TYPE B/65 S O N A G R A M ®

KAY ELECTRIC CO.

PINE BROOK. N. J.

tttftttlftftttiillflfoJ l

ι , 4 Spectrogram 1.

T Y P E B/6S S O N A G R A M ®

[mei] range: 85-4000 cps.

KAY ELECTRIC CO.

Spectrogram 2.

PINE BROOK. N. J.

[mei] range: 85-4000 cps.

i IHllii

ü*l

TYPE B/65 SONAGRAM ®

ti « f m

ι » t »üiuMft K m

KAY ELECTRIC CO.

m m m m m

Spectrogram 3.

PINE BROOK. N. J.

ιβϋ»«ΐι i n

i

»

[rneyl range: 85-4000 cps.

TYPE B/6S SONAGRAM ®

iii¿tMAéititfttftMi

KAY ELECTRIC CO.

Ü J R i j t b t i f i i Ü Spectrogram 4.

[meu] range : 85-4000 cps.

\

TYPE B / 6 5 SONAGRAM

Spectrogram 5.

[mcu] range: 85^4000 cps. TYPE B / 6 5 SONAGRAM®

ÍiMMHÜMliéiftJiilÉ! Spectrogram 6.

KAY E L E C T R I C CO.

rliíMiyftÉtiüÉ [mei] range: 85-4000 cps.

KAY ELECTRIC CO.

PINE BROOK. N. J.

Spectrogram 7.

T Y P E B / e s SONAGRAM ®

[mei] range: 85-4000 cps.

KAY ELECTRIC CO.

PINE BROOK. N. J.

I

Spectrogram 8.

[mef] range: 85-4000 cps.

ON UTTERANCE-FINAL Ρ ] AND [ y ] IN PORTUGUESE

(3) Nasopharynx: (4) Jaw position:

(5) Larynx: (6) Duration:

[y] firmly closed approximate average of .50 cm. distance between upper and lower teeth; slight retraction lowered average of .110 sec.

137

ra firmly closed approximate average of .50 cm. distance between upper and lower teeth; slight retraction lowered average of .110 sec.

7. C O N C L U S I O N S

The articulatory and acoustic problems which underlie any attempt at a clear distinction between [1] and [u] in syllable-final and utterance-final positions are several. First, there is the basic problem of the 'duality' of the liquid (Jakobson et al., 1957: 19), bridging the vocalic and the consonantal, making [1] unique among Brazilian Portuguese phonemes. Like the Brazilian Portuguese vocoids, /l/ is produced by vocal cord and varying cavity modulation. The lateral emission of the voiced breath stream, however, finds no parallel among Brazilian Portuguese vocoids. The second problem is that Brazilian Portuguese /l/ is one of the sounds most susceptible to the effects of assimilation. Thus, for example, we have no clear definition of F2 patterning, since F2 acquires most of the characteristics of the vowel sound which precedes or follows it. The third problem is that, in comparison with the other Romance languages, Brazilian Portuguese /l/ shows certain unique characteristics, among which we may note its marked tendency toward retroflexion of the apicum without necessarily achieving occlusion, convexity of the prelaminum, a lessening of the distance between the rear surface of the root of the tongue and the rear pharyngeal wall, and the tendency toward a slight construction of the orbicularis oris. We must add to this a general process of lenition of /l/ in contemporary Brazilian Portuguese which has caused Elia to remark that: "O /l/ pós-vocálico está sofrendo um fenòmeno de deterioraçâo: ou se vocaliza, ou cai, ou se transforma na outra líquida o /r/ que ... possúi maior vitalidade" (1963: 271). In the Caipira dialect area, the substitution of /r/ for /l/ in the positions we have been discussing is attributed by many Brazilian linguists to an attempt to reach alveolar occlusion as in [1] but without lateral emission because of the speed of the articulation in the unstressed position. The result is an alveolar flap. The phenomenon has also been attributed to the effects of African substratum or influence (Mendonça, 1935: 120). In our comparison of [1] and [u], the above-mentioned characteristics of Brazilian Portuguese [1] cause it to be less clearly different from /u/ in certain positions than in, say, French or Spanish. Still, labialization, tongue position, mandibular position,

138

DAVID FELDMAN

laryngeal height, and formant frequencies (to the extent that these latter are reliable in the case of ¡if) provide us with the necessary acoustic and physiological data to predict and account for a clear differentiation between [1] and [u] from both the speaker's and hearer's points of view. In our comparison of [I] and [u], however, we note much greater dialectal divergences. In general, there is a clear tendency toward an equalization of acoustic and physiological phenomena between the two sounds even in those informants whose idiolects retain the distinction in utterance-final position after a vowel. First, even in the informants from Säo Paulo and Goiás with regard to labialization, the [u] shows a lessening of constriction of the orbicularis oris when compared to [u], while [!] shows an opposite tendency toward constriction when compared with [1]. Thus, the two articulations [1] and [u] are significantly more alike than are the articulations [1] and [u]. In these idiolects in which the distinction is in any way preserved, the [u] is articulated with a noticeable lowering of the dorsum as compared with [u], while the [t] shows a similarly noticeable raising of the dorsum as compared with [1]. The convexity of the laminum in [u], however, is distinct from the concavity of the laminum in [I]. The lax apicum in [u] is likewise distinct from the tense and raised apicum in [1]. The major feature here is the distinction between apical position in [1] and in [1]. In the case of [1], however, the apicum is tense and raised, but fails to make any occlusion with the upper surface of the oral cavity. In the case of the informants from Rio de Janeiro and Fortaleza, however, virtually all measurable distinction between utterance-final /l/ and /u/ after a vowel is lost. Thus, when compared in terms of apical position and convexity of the laminum, [1] and [u] are distinct, while [1] and [u] merge. Likewise, in the case of these particular informants, with regard to mandibular movement, [1] and [u] merge. Likewise, in the case of these particular informants, with regard to mandibular movement, [u] and [1] show a virtually equal degree of tensing and retraction. It is clear, then, that the essential distinctions between [1] and [u] are considerably reduced in [1] and [u] in the overall pattern, leaving basically only the tongue position and mandibular retraction to make the distinction. In the speech of Carioca and Ceará informants, as we have seen, no significant distinction is made. Apparently some speakers of Brazilian Portuguese tend to increase the constrictive movement of the orbicularis oris in the articulation of [1], making in fact a rounded vs. unrounded distinction in the articulation of [1] vs. [1] while the apicum remains lax and low behind the lower teeth. In our observations, the greater the degree of increased labialization, the less the degree of tenseness in the apicum. Thus, many speakers, in effect, substitute a slight labialization for apical occlusion in the articulation of [1]. Simultaneously, there is a tendency toward relaxation of apical tenseness, permitting a greater degree of mandibular retraction. Consequently, the apical position is more fully differentiated from [1] to [1]. These observations account in great measure for the inability of the informants to distinguish aurally between utterance-final [1] and [u] in minimal pair situations. But we still have not accounted for the possible effect of a preceding close or open

ON UTTERANCE-FINAL Ρ] AND [y] IN PORTUGUESE

139

vowel on such aural comprehension in analogous-environment pairs such as meu/mel ([méu]/[mél]). Both meu and mei derive from Popular Latin etyma with /ε/ : mëum and mël(lis), respectively, /mèi/ > /mèi/ is the expected development, while the raising of the /ε/ in meum to /e/ in meu was occasioned by the hiatic contiguity of a high vowel in the following syllable (Grandgent, 1908, sections 347, 385). On the other hand, if our observations of a merging of [u] and [1] are correct, then we must take into account any potential assimilative influence of [1] on the preceding vowel. The influences of final /I/ on preceding vowels have been demonstrated graphically by Potter, et al., (1966: 236ff.). The second, third and fourth formant bars of /ε/ followed by /I/, for example, are pulled sharply downward to join with the corresponding formant bars of /l/. The same occurs to the second and third formant bars of /ε/ (the fourth bar for [ε] was not realized clearly on any of our spectrograms) when followed by /1/. The influence of /u/ on a preceding /e/ or /ε/ is even more severe. Thus, to articulate a tense /e/ before [u], as in meu, without lowering the /e/ to /ε/, requires a much greater muscular effort with regard to the /e/ followed by /u/ than for /ε/ followed by /l/. This extraordinary reinforcement of the tenseness of /e/, in which the characteristic apical position is low (behind the lower teeth), would tend to impede the raising of the apicum and the tensing of its muscular structure in rapid transition. To test the hypothesis that accurate aural recognition of mei vs. meu depends upon the /e ~ ε/ contrasts and not upon an /u ~ 1/ contrast, the artificial pairs *[mél] and *[meu] were recorded. The native informants were unanimous in recognizing the first as the possessive meu and the second as the noun mei, despite the reversal of the final sounds. We may conclude that utterance final /.CVu/ and /.CVl/ are characteristically realized as single syllables. The experiment revealed that, when utterances of the /CVu/ type were read as /CV1/ (and conversely), informants consistently selected the correct utterance on the basis of the nuclear vowel or the initial consonant (or both), but not on the basis of any contrast between the final [1] or [u]. In narrow transcription, then, for the description of overall pattern the separate symbols must be retained. The acoustic and articulatory data referred to above are sufficiently distinct in [I] and [u] for the generalized pattern covering the several dialect areas represented to require the retention of separate symbolism. But for the informants from the cearense and fluminense dialect zones, the only realization of unstressed utterance-final /I/ or /u/ after a vowel is [u] and, consequently,only one symbol is required even in narrow transcription. At the level of broad transcription, however, the issue is less clear. When informants heard *[mél] or *[méu] as substitutes for mei, they invariably chose the latter as the form closest to any they had ever heard or said themselves. The former was understood, of course, but sounded 'oratorical' or, in some cases 'foreign'. 9 From the viewr " Cf. the sociological generalization by Antenor Nascentes: "O / final é pronunciado levemente pela classe culta; os pedantes exageram-no ..." (1953:48).

140

DAVID FELDMAN

point of the hearer at the phonemic level, then, a convincing case can be made for the assignment of [1] and [u] as allophones of /u/ in the overall pattern in dialectally-determined variation. The same would be true of the phonemic analysis from the speaker's viewpoint, in overall pattern. However, for those dialect areas in which the apical distinction between [1] and [u] is revealed, the articulatory features appear still to be sufficiently consistent and distinct to warrant their respective assignments to /l/ and /u/ in descriptions of the respective idiolects. The monosyllabic realization of both the /CVu/ and /CVl/ types, moreover, clearly implies the need for revision of present descriptions of Brazilian Portuguese falling diphthongs. For pedagogical purposes, the approximation of [1] and [u] can be subjected to more thorough treatment in teaching materials than has been the case. English- and Spanishspeaking students, in the transfer of phonetic habits into Portuguese, tend to hear the utterance-final / and u as the same sound. When the word is cognate with an English and Spanish form (or when the student sees the written representation of it) the learner then substitutes /u/ or /I/. Because both English and Spanish have falling diphthongs of the/-.Vu#/type, students have less trouble articulating a native-sounding [u] than [1], especially after a clear demonstration and sufficient repetition drill. But because neither the Spanish /I/ after /a/, nor the English 'dark' /I/ achieves the vocoidal characteristics of the Portuguese [1], it seems more economical to treat the syllabic structure /-.Vu#/ and /-.Ϋ1 # / as consistently ending in [u]. CALIFORNIA STATE COLLEGE AT FULLERTON

REFERENCES Bechara, Evanildo, 1963 Moderna gramática portuguêsa, Curso mèdio, 7* ed., (Säo Paulo, Companhia Editora Nacional). Carmody, F.J., 1937 "X-Ray Studies of Speech Articulations", University of California Publications in Modern Philology 20: 4 (Berkeley, University of California Press). Delattre, Pierre, 1951 "The physiological Interpretation of Sound Spectrograms", PMLA 56: 864-75. Elia, Silvio, 1963 Ensaios de filologia (Rio de Janeiro, Livraria Académica) pp. 265, 271. Feldman, David M., 1967 "Outline of a Comparison of the Segmental Phonemes of Brazilian Portuguese and American Spanish", Linguistics 29: 44-57. Grandgent, C.H., 1908 An Introduction to Vulgar Latin (Boston) sections 347, 385. Hall, Robert Α., Jr., 1943 "The Unit Phonemes of Brazilian Portuguese", Studies in Linguistics 1:1-6. Jakobson, Roman et al., 1957 Preliminaries to Speech Analysis (Cambridge, ΜΓΓ Press) p. 19. Mendonça, Renato, 1935 A influencia africana no portugués do Brasil (Sâo Paulo, Companhia Editora Nacional). Nascentes, Antenor, 1953 O linguajar carioca (Rio de Janeiro, Simöes).

ON UTTERANCE-FINAL Ρ] AND [μ] IN PORTUGUESE

141

Parmenter, C.E. and S.N. Treviño, 1931 "A Technique for Radiographing the Organs of Speech During Articulation", Zeitschrift für Experimental-Phonetik 1 : Ó3-84. 1932 "Vowel Positions as Shown by X-ray", Quarterly Journal of Speech 18: 351-69. Pereira, Eduardo, 1957 Gramática expositiva, 102nd ed. (Säo Paulo, Companhia Editora Nacional). Pike, Kenneth L., 1943 Phonemics (Ann Arbor, University of Michigan Press). Potter, Ralph et at., 1966 Visible Speech (New York, Dover Publications, Inc.). Pulgram, Ernst, 1964 Introduction to the Spectrography of Speech (The Hague, Mouton). Reed, David and Yolanda Leite, 1943 The Segmental Phonemes of Brazilian Portuguese in K.L.Pike''s Phonemics (Ann Arbor, University of Michigan Press) pp. 194-202. Russell G. Oscar, 1928 The Vowel (Columbus, Ohio State University Press).

APPENDIX Partial inventory of minimal pairs 1. [e] u vs. [ε] I meu mei •cêutico céltico o teu hotel mareo chairel vs. [ε] I 2. [ε] u cel(ta) céu (provd)vel véu 3. [a] u vs. [a] / tau tal cacau chacal

4. [o] u sou vou 5. [i] u viu 6. [i] « tio

vs. [o] / sol 'eau-de-vie [de n'importe quelle sorte]'. u

Notons que l'Atlas Linguistique de la France ne signale aucun équivalent d'hièble dans l'Allier, soit parce que cette plante n'y est pas connue, soit parce que les répondants de l'Atlas ne reconnaissaient pas son nom français. " D'après Dauzat, Dubois et Mitterand (1964), la référence de 1882 proviendrait de Lyon et non de Genève. 11 Plus à l'ouest et notamment dans une grande partie de l'Auvergne, on trouve des formes comparables mais non identiques (voir l'Atlas Linguistique de la France, carte 928, et J. Ronjat, 1930-41 : ISO) et cette ressemblance ne constitue pas une homonymie.

226

FRANK R. HAMLIN

En plus des attestations déjà citées, Esnault mentionne l'emploi de 'gnôle' 'eaude-vie* à Lille en 1905 et à Troyes en 1912. Mais ces exemples sont isolés. Bien que le dictionnaire Argot and Slang d'Albert Barrère (1906) soit assez complet, il ne donne pas le sens 'eau-de-vie' pour ce mot et il est donc évident que Barrère ne le connaissait pas14. En effet, M. Cohen observa en 1916: "il [le mot gnôle] me semble actuellement aussi général que nouveau... mais il était connu auparavant de certains corps de métier...; ce terme... est... de l'argot véritable passant par les tranchées pour se répandre à Paris". Mais Albert Dauzat fait remarquer (1917:71, note 1): "Le mot est... resté inconnu, au moins jusqu'en 1916, aux contingents du Nord auxquels des Méridionaux l'ont appris". L'emploi généralisé actuel de gnôle remonte à la première Guerre Mondiale. Avant la guerre, on le considérait comme un mot régional; après, il était devenu un mot d'argot. Qu'est-ce qui lui fit fortune? Et pourquoi réussit-il à se substituer dans l'argot à d'autres équivalents d"eau-de-vie'?16 L'usage familier employait assez généralement en France un homonyme de gnôle 'eau-de-vie', homonyme qui remonte au moins jusqu'à la fin du 17e siècle. Le dictionnaire de Furetière (1727) donne ce mot en citant Richelet: "GNIOLE, terme d'écoliers qui jouent à la toupie, c'est la marque du fer qu'on y imprime"18. Hatzfeld et Darmesteter (1890-1900) ajoutent une acception plus générale: "Coup, éraflure, qui laisse une marque sur une personne ou une chose" ; et, de même que Robert, ils font voir l'identité presque complète qui existe entre les mots gnôle et gnon™. Nous venons de voir que le dictionnaire de Barrère citait gnôle avec le sens général de 'coup.' Comme ce sens s'est généralisé à partir de la marque imprimée par une toupie, Dauzat avait sans doute raison de considérer que ce mot était en réalité une forme apocopée de torgn(i)ole (voir ci-dessus note 6). Il aurait croisé le sens de gnon (< oignon) avant de rencontrer celui de gnôle 'eau-de-vie'. Nous avons exposé tous les éléments du petit drame qui semble expliquer l'évolution sémantique de gnôle au cours de la guerre 1914-1918. Des militaires venus de différentes régions de France se sont rencontrés dans les casernes. Ceux du Lyonnais disaient qu'ils allaient 'boire une gnôle', en employant un terme régional qui leur était familier. Les non-Lyonnais croyaient au jeu de mots, parce qu'ils traduisaient instinctivement 'gnôle = coup' et l'expression 'boire un coup' leur était bien connue. Comme il arrive parfois18, ce jeu de mots a fait rage et le mot gnôle s'est rapidement popularisé avec le sens qui était jusque-là régional. Il convient de remarquer que, en se substituant à 'coup' dans l'expression 'boire un coup', et par la suite dans n'importe " Il cite gnole (p. 180) au sens de 'coup': "'slap, clout, wipe'; or, as the Americans would have it, 'biff'". 15

Notamment tac ou taco(t) et cric.

" Cette définition est suivie par Louis Chambaud (1761) qui traduit "The dint or impression made by the spike of a castle top", et par Littré (1874-83:1888). " Robert explique ainsi ce dernier mot: '"enflure provoquée par un coup'; forme apocopée d'oignon" (1953-64); de même, A. Dauzat (1917:185). 18 Des expressions ou des vers qui correspondent, à un moment donné, à la tendance des idées populaires se répandent parfois, au moyen de la transmission orale, à une vitesse extraordinaire; pour des exemples analogues à celui que nous discutons, voir I. et P. Opie (1959:5-7).

L'ORIGINE ET L'ÉVOLUTION DU MOT "GNÔLE"

227

quel contexte, gnôle garde toujours le sens d"eau-de-vie' et ne se généralise pas au point de signifier, par exemple, 'consommation'. Observons enfin que l'expansion géographique de gnôle se rapporte directement à son extension sémantique, et que ce sont surtout les rencontres homonymiques qui ont favorisé l'un et l'autre de ces développements; l'étymologie, au sens le plus usuel de ce terme, explique la naissance d'un mot mais risque souvent d'ignorer les rapports de parenté qui l'unissent à bien des autres termes. UNIVERSITY OF BRITISH COLUMBIA

RÉFÉRENCES A. Barrère, 1906 Argot and Slang (London). Bloch, O. et W. von Wartburg, 1960 Dictionnaire étymologique de la langue française (Paris). Chambaud, L., 1761 Dictionnaire françois et anglois (Londres). Cohen, M., 1916 Bulletin de la Société de Linguistique de Paris 20. Cuny, Α., 1910-11 Mémoires de la Société de Linguistique de Paris 16 (Paris). Dauzat, Α., 1917 L'argot de la guerre (Paris). 1922 La géographie linguistique (Paris). 1929 Les argots (Paris). 1938 Dictionnaire étymologique de la langue française (Paris, Larousse). Dauzat, Α., J. Dubois et H. Mitterand, 1964 Nouveau dictionnaire étymologique (Paris, Larousse). Dottin, G., 1920 La langue gauloise (Paris). Eraout, A. et A. Meillet, 1932 Dictionnaire étymologique de la langue latine (Paris, C. Klincksieck). Esnault, G., 1965 Dictionnaire historique des argots français (Paris). Fouché, P., 19S9 Traité de prononciation française (Pai · Furetière, 1727 Dictionnaire universel (Paris). Gilliéron, J. et E. Edmont, 1900-12 Atlas linguistique de la France (Paris). Hatzfield et Darmesteter, 1890-1900 Dictionnaire général de la langue française (Paris). Holder, Α., 1896-1907 Altceltischer Sprachschatz, II (Leipzig, Β. G. Teubner). Huguet, E., 1925 Dictionnaire de la langue française au seizième siècle S (Paris, E. Champion). Jaubert, H., 1864 Glossaire du Centre de la France (Paris, N. Chaix). Littré, E., 1874-83 Dictionnaire de la langue française 2 (Paris, Hachette). Meyer-Lübke, W., 1935 Romanisches Etymologisches Wörterbuch (Heidelberg, C. Winter).

228

FRANK R. HAMLIN

Mistral, F., 1878-86 Lou Tresor dóu Felibrige 2 (Aix-en-Provence, Veuve Remondet-Aubin). Niedermann, M., 1902 Mélanges linguistiques offerts à M. Antoine (Paris). Opie, I. et P., 1959 The Lore and Language of School Children (Oxford). Orr, J., 1939 Studies in French Language and Medieval Literature presented to Professor Mildred K. Pope (Manchester). Pokorny, J., 1948 Indogermanisches Etymologisches Wörterbuch (Bern, Francke). Robert, P., 1953-64 Dictionnaire alphabétique et analogique (Casablanca). Ronjat, J., 1930-41 Grammaire historique des parlers provençaux modernes 1 (Montpellier). Seifert, E., 1923 Beihefte zur Zeitschrift für Romanische Philologie 75. Die Proparoxytone im Galloroma· nischen (Halle). Thomas, Α., 1904 Nouveaux essais de philologie française (Paris), von Wartburg, M., 1928 Französisches Etymologisches Wörterbuch (Bonn, F. Klopp).

FRED HOUSEHOLDER

VOWEL OVERLAP IN AZERBAIJANI

The possibility of achieving similar acoustic effects by different articulatory means, and the related possibility of making articulatory distinctions whose acoustic discrimination is difficult or impossible, have interested Professor Delattre. When I began the study of Azerbaijani in 1951, I often confused instances of what I later analyzed as a tense mid front rounded vowel /ö/ with those of a (lax or tense) high back unrounded vowel /ui/. Note that the contrast is a three-step one: front vs. back, rounded vs. unrounded, high vs. mid. Because of such distributional facts as (1) the absolute scarcity of /ö/, (2) the general restriction of /ö/ to initial syllables (or, in foreign words, stem-syllables), (3) the very high frequency of /m/ in suffix-syllables, and its total exclusion from absolute initial position, (4) the occurrence of palatals adjacent to /ö/, of back velars with /ui/, the functional yield of the contrast is extremely low. I can't recall finding even one genuine minimal pair (in a language where monosyllables are plentiful). The average values of Fl and F2 for 39 instances of /ö/, spoken by two male informants with rather low voices, (all these instances are in initial syllables) are: Fl = 440 Hz; F2 = 1340 Hz. A sample from the same speakers of 17 /ra/ in initial syllables (only a few being monosyllabic) shows: Fl = 440; F2 = 1330. 13 instances in medial (affixal) syllables have average values : Fl = 440; F2 = 1320. And finally, 26 instances in final syllables (some followed by one consonant, some absolutely final) average: Fl = 410; F2 = 1410. Only this last value is sufficiently different from /Ö/ to suggest easy discrimination — and this position is exactly where /ö/ never occurs. There is no difference in the range of values for either FI or F2; the medians for FI are almost identical (420 and 430), and differ only slightly for F2: about 1340 for / '

t

'

I

1

0.5

1

I

1 '

I

1

! ' I

1.0

i z ô s r e n i r i j z

ç w a j w i j ' u d p

S 3

Fig. 1

1.0

-

0.5 -

i ζ δ ε r e

η

i

r i j z i j

w a j w i

J

u

d

ρ

S3

Fig. 2

1.0 —

0.5

I w

i

1

j

'

uw Fig. 3

I

1

I '

m

m

I

' ε

I ' ι

1 ' ι

I ·

I

'

1.5

-

1.0

-

0.5

-

I

I

•—|—!—Γ-1—I

ι

'

»

I—1—I

'

1

ι—1

0.5 J

s

l

a

j

k

I

1

»

'

1.0

a

m

a

w

t

Fig. 7

1.0

J* **

0.5

• JU S Sjgff« ΐ j w ^ j y j s » · ^ * ' - *"'*

1.0

-

It

0.5

ι

ι

M^ïfa

I—J—«—ι—I—I—»—I—'—I—«—I—«—ι—1

1

0.5 d a j u w

θ

i

q

k

1

I

'

1

1

r

1-0 i

t

Fig. 9

w

i

l

b

i

d3

o

j

s

THE QUESTION-PHRASE FALL-RISE IN BRITISH ENGLISH

249

And as to the 'novelty' of the item, this is probably applicable to any 'nuclear' tune. We have compared question-phrases with a final fall-rise with those having other final 'tones' (falls, rises and rise-falls). Those other phrases have not been measured. Their number is of the same order of magnitude as that of the materials analyzed here. Whilst we would not be able to add much to the excellent description of the semantic values of such tones in general questions given by O'Connor and Arnold, we suggest that general questions ending with a FALL-RISE, at least the ones we could find in our materials, compared to similar questions with a final RISE are more FRIENDLY, FAMILIAR, INFORMAL or INTIMATE. We could hardly find a single example of our intonation in serious, classical drama. Considering that such acute observers as Palmer, D. Jones or Armstrong never mentioned it in their writings, it would also be interesting to try to find out whether this intonation was not relatively new. Only statistical treatment of very extensive materials with recorded speech of people of different ages could test this hypothesis. The greatest interval of the fall was noted in nos. 67 (26 semitones), 83 (18 semitones) and nos. 54, 62, and 88 (17 semitones), while the smallest intervals occur in nos. 24, 37, 56, 59, 87 (2 semitones) and no. 60 (1 semitone). Though, again, we cannot, for reasons of space, describe the particular situations (and even if we could, it would not be obvious how much explanation was necessary to PROVE a particular statement about the meanings carried by the intonation); it would appear that the extent of the interval is not related to any semantic (or "attitudinal") differences. The size of the interval (in our cases at least) is probably largely dependent on momentary or permanent neuro-physiological states ('emotions', etc.). The same would appear to apply to the rising part of our fall-rise. The largest intervals occur in our nos. 67, 83, 1, 41, 96, 24 and 79 (between 54 and 17 semitones), and the smallest in nos. 3, 29, 34 (three semitones) and nos. 13, 19, 25, 31, 52, 72 (2 semitones). Most descriptions of British English intonation distinguish functional units not only on the basis of the DIRECTION of the pitch movement but also according to the relation of the tone (or tones) to the lowest (and/or highest) pitch. Such pitch is, of course, relative due to individual differences between speakers. Kingdon (1958a, 1958b) distinguishes four varieties of the fall-rise: high normal, high emphatic, low normal and low emphatic. A study which would test the validity of four functionally different fall-rises (or, indeed, similar differences within other 'kinetic' or 'level' tones) would involve an analysis of considerable amounts of recorded material and/or psycho-phonetic experiments performed with natural or synthetic speech. Inspection of our materials has led us to the conjecture that the relation of the lowest tone within the fall-rise to the speaker's lowest pitch is not irrelevant semantically. To test this hypothesis rigorously would also require analytic work far beyond the scope of the present study. We would, however, like to suggest tentatively that if the speaker's lowest pitch (or lowest pitch range) is reached within the questionphrase fall-rise, this expresses more INSISTENCE, SERIOUSNESS, CONCERN or CONCENTRA-

250

WIKTOR JASSEM

TiON than if this low pitch is not reached. Again tentatively, we assume that the lowest individual pitches used in speech are somewhat like this: male (in cps.) low medium high

6 0 - 80

80-100 100-120

female (in cps.) 100-120 120-140 140-170

The above figures are based solely on the materials analyzed for the purposes of the present study, so they only apply to voices we happen to have on record. We have assumed that if the lowest pitch reached in our fall-rise is at least about three semitones above the lowest pitch range, the fall-rise may be described as 'high', otherwise it is 'low'. In our list on pp. 244-245, the low fall-rise is indicated by v , and the high fall-rise by v . Thus, phrases asking permission or making unobtrusive suggestions would have the high fall-rise, e.g., nos. 5, 6, 7, 9, 10, 31, 32, 44, 49, 50, 82 etc., whilst earnest queries would end with a low fall-rise. Thus, nos. 36,37 and 38 were spoken by an investigating police-inspector and nos. 53,54,55,56 and 61 by a doctor (apparently a psycho-analyst) when performing his duties. Other examples typically illustrating a low fall-rise are nos. 15, 21, 65 etc. The distribution of the falling-rising 'melody' over the particular syllabic tonal segments partly depends on the number of syllables within the 'nuclear tone'. If this is monosyllabic, the whole double movement is by definition contained within that syllable, as in Figure 7. In a bisyllabic 'nuclear tone' the first syllable usually carries the fall and the second carries the rise as in Figure 8. The most common distribution in a three-syllable 'nuclear tone' is: first syllable: fall, second syllable: level, third syllable: rise, as in Figure 2. In longer 'nuclei' the distribution varies. The rise may begin anywhere between the second syllable and the last stressed syllable, as shown in Figure 9. We have also measured the fundamental frequency of that part of the question phrase which precedes the fall-rise. We have found it possible to describe this part, which in British terminology is mostly referred to as the 'head', in terms of 'toues' (or 'tonal units') similar to those proposed by Kingdon, O'Connor and Arnold, and Gimson. All five types of tones have been found: level, rising, falling,rising-falling and falling-rising. A phrase-initial, unstressed, level syllable is left unmarked, otherwise we are using tone-marks similar to those used by Kingdon, with the exception that we do not distinguish a phrase-initial stressed rise from a similar unstressed rise. As can be seen from our list, in a vast majority of cases, there is an initial rise, which is often followed by a fall. The fall may begin at a stressed syllable, and is then considered to be a separate 'tone', marked (see, for instance, nos. 2, 29, 32). If the change of direction takes place within the same rhythm unit, the tone is classified as a rise-fall and marked Λ (e.g., nos. 1,28, 36, 56, 68). There are also a few cases of ' The symbol ' is reserved, in the O'Connor and Gimson systems for a falling tone reaching the low range, and we follow those authors in this respect.

THE QUESTION-PHRASE FALL-RISE IN BRITISH ENGLISH

251

high-level tones (e.g. nos. 10,13, 35, 65) and falling (or 'sliding') heads not preceded by a rise (e.g. nos. 34, 71, 76). Other tones and combinations of tones also occur, as can be seen in the list. Such differences, which may be semantically significant before other 'nuclei', are apparently in free variation before a question-phrase fall-rise. Finally, the relative pitch of the 'head' will briefly be compared with that of the fall-rise. In a great majority of cases most of the HEAD lies above the initial tone of the NUCLEAR fall. At any rate, the head mostly finishes above or at the pitch of the beginning of the fall. In a few specimens, however, the nucleus starts above the end of the head, e.g. in examples nos. 9,21,37, 39 and is then, in our tone-marking, preceded by an arrow. If this is so, the end of the head lies only about one to four semitones below the beginning of the fall. Again, this does not appear to be correlated with any clear differences in meaning.

5. SUMMARY

While terminal rises and falls have long been observed in general questions in British English, terminal fall-rises are only admitted by a few authors, and that with reservations which would imply that this is not a very usual kind of intonation in such phrases. The one hundred cases of question-phrase fall-rises collected and analyzed in this study indicate that this is an altogether common intonation. It appears to indicate friendliness, familiarity, informality or intimacy. Two varieties of the terminal fallrise have been distinguished: a high one and a low one, of which only the latter reaches the speaker's lowest pitch range. The low fall-rise indicates more insistence, seriousness, concentration or concern than the high one. There is wide dispersion in the size of the intervals both for the fall and the rise, but intervals between five and twelve semitones are typical. Diiferences in the extent of the interval do not appear to be semantically significant. The terminal question-phrase fall-rises are preceded by heads which most commonly begin with a rise and usually end higher than the beginning of the nuclear fall. POLSKA AKADEMIA NAUK

REFERENCES Candita, E. F., 1967 Present-Day English for Foreign Students, reprint (Warszawa). Gimson, A. C., 1962 An Introduction to the Pronunciation of English (London). Halliday, M. A. K., 1967 Intonation and Grammar in British English (The Hague). Jassem, W., 19S2 Intonation of Conversational English (Educated Southern British) (Wroclaw).

252

WIKTOR JASSEM

Kingdon, R., 1958a The Groundwork of English Intonation (London). 1958b English Intonation Practice (London). Lee, W. R., 1956 "English Intonation : a New Approach", Lingua 5 : 345-71. O'Connor, J. D. and G. F. Arnold, 1961 Intonation of Colloquial English (London). Palmer H. E. 1924 English Intonation with Systematic Exercises, 2nd ed. (London). Schubiger, M., 1958 English Intonation, Its Form and Function (Tübingen). Sharp, A. E., 1958 "Falling-Rising Intonation Patterns in English", Phonetica 2: 127-52.

ALPHONSE JUILLAND

ENTRY WORDS: GRAMMARS A N D DICTIONARIES

Until recently the exclusive concern of structural linguists, the formalization qf phonology and grammar, has been followed by attempts at formalizing the lexicon. The traditional idea of "viewing grammar as an integrated system, and lexicon as a miscellaneous remainder" (Gleason, 1962:96) has been challenged, and linguists have tried to determine the type of information which properly belongs in the dictionary. In addition to the topics discussed in the volume Problems in Lexicography (Householder and Saporta ed., 1962), several aspects of the problem have been examined by Garvin (1955: 1013-88), Householder (1962 : 567-76), transformational grammarians (Householder, 1962 : 572-73), and Mathiot (1967)1, to mention only a few. Essentially, they have tried to determine: (1) what linguistic information belongs in the dictionary rather than in the grammar, and (2) the way in which this information can best be formalized. Although the two questions are interrelated, the emphasis has been on the former, centering on the relationship between grammar and lexicon. In this paper, we shall focus on the latter, in an attempt to establish some general principles for consistently presenting the grammatical information usually contained in the entry-heads or entry-words of dictionaries. Needless to say, a satisfactory solution to this problem presupposes a correct understanding of the relationship between grammar and dictionary.

1. ENTRY WORDS

At least insofar as modern dictionaries are concerned, the traditional concept, which tacitly excludes formal information from dictionaries, no longer seems to apply. Dictionary entries usually consist of a heading, called the entry-form or entry-word, generally transcribed in bold-face type or in italics, and of an entry proper, which follows m regular type. Traditionally, entry-words have been conceived of as labels, their function being that of indices, pointing to the form or family of forms to which 1 This article by M. Mathiot (1967) had not yet appeared when this article was written. Page references are to a pre-publication copy.

254

ALPHONSE JUILLAND

the lexical information (concerning meaning and use) contained in the entry proper is relevant. This situation has changed in modern dictionaries, where entry-forms increasingly tend to acquire an importance of their own. Phonemic transcriptions, accentual marks, occasional morphonemic symbols, codes of morphologic class membership, syntactic abbreviations, etc., have been progressively added, altering the character of entry-forms to a point where they no longer function as pointers to information of another kind : they have become sources of information in their own right. Actually, entry-forms now tend to be considered as full-fledged conveyors of grammatical information about the unit or units whose semantic properties are dealt with in the entry proper. On this point there tends to be general agreement among specialists. For Gleason, for instance: ... the dictionary always carried a considerable mass of formal information, even if this is nothing more than the assignment of words to 'parts of speech.' But ordinarily this is not all. Hidden away in our advocacy of copious citation of usages may indeed be a demand for additional formal information, in this case largely the collocational possibilities of the forms. With function words, of course, the dictionary definitions, generally thought of as statements about meaning, are actually largely statements about use, as may be seen in any good English dictionary in the articles for the, not, etc. (1962:91). According to Hoenigswald: ... we are also used to the idea of going to the dictionary for grammatical information in the strict sense. We can never be sure just how much we will find; but there is probably no Latin dictionary worth the name that doesn't contain the principal parts of the verbs, or that fails to tell us that urbs is feminine or that utor Ί use' governs the ablative case. In other words, it is either their grouping or some added matter which — implicitly or explicitly — gives us added information, going beyond the index principle. It is as though we had here either an attempt to provide separated indexes to separate portions of the 'book' (namely, of the grammar or the semantic treatise, as the case may be); or else an attempt at not merely referring to page and line, chapter and verse, but proceeding to quote and excerpt from the 'book' right then and there. (1962:104) The question as to what information should be carried in the entry-words, and in what way, has also been considered. Here we must distinguish between the ENTRYWORD proper, generally one of the variants of the lexical family dealt with in the entry proper, and the MORPHOLEXICAL TRANSCRIPT, which sometimes follows in brackets and represents a formalized transcription of the invariant. In the absence of the latter, the entry word is selected for its 'predictability value', i.e., the ability to function as a morphological transcript: In most lexicographic practice, the problem of the morphemic shape of the entries is solved by selecting one of the several grammatical forms of the constituent word or words of the lexical unit as the base form, and entering this base form in the dictionary. In the better known languages, the basejforms of words of all major classes have long-since been fixed by tradition: thus, Latin nouns are entered in the nominative singular (with the exception, of

ENTRY WORDS: GRAMMARS AND DICTIONARIES

255

course, of pluralia tantum). Hungarian verbs are entered in the third person singular present, Slavic adjectives in the nominative singular masculine, etc. (Garvin, 1955:1017). With emphasis on American Indian languages, and especially on Kutenai, the question has been completely discussed by Garvin (1955:1017). As for the nature of the morpholexical transcript, i.e., the grammatical information to be conveyed in an entry-head, Hoenigswald believes that: A simon-pure lexical listing ought to presuppose a morphic segmentation (not necessarily unique); the items in the list should be minimal, that is, they should be single morphemes; their arrangement might be random although there can be no objection to an arrangement by morphs, that is by the phonemic shape of the allomorph or allomorphs making up the morpheme; and finally the 'reference' should be to the appropriate paragraph in the morphology and syntax, or in some imaginary, comprehensive treatment of meanings and meaning differences in the guise of a repertory of semantic fields. At this point the views would begin to diverge widely. I myself am inclined to liken the step from grammar to semantics to that from the setting up of such classifications as syllabic vs. non-syllabic on one hand to the study of the overlapping among the ranges of individual phonemes on the other. Luckily this is another question which we need not decide here. Perhaps we can agree that the privileges of occurrence for morphemes are of interest both to the grammarian and to the lexicographer. (1962:104). Malone, however, considers that: If the entry word is subject to inflection, some clue to this is usually added, as principal parts of verbs and the name and number of the declension that a noun belongs to. Irregular and anomolous forms are often entered for themselves, with cross-reference to the appropriate head-form. In the front matter one may find a list of the commoner suffixes (less often, of prefixes too) and an outline of the inflectional system of each language dealt with in the dictionary. In a good many bilingual dictionaries much is made of irregular verbs, sometimes as part of the front matter, sometimes in a special appendix. Here such verbs are listed in full, and the irregular forms of each verb are entered against the head-form (1962:112). It would seem that formalized entry-heads or morpholexical transcripts convey phonological information by means of phonemic (segmental) symbols and prosodie (suprasegmental) marks, morphophonological information by morphophonemic symbols, morphological information through codes of class membership (second declension, third conjugation, etc.) and syntactical information by parts-of-speech abbreviations ('n.' for nouns, 'v.' for verbs, etc.). It goes without saying that syntactic information is also contained in the entry proper, where examples are selected to illustrate the privileges of occurrence of lexical entities relative to one another and to the larger constructions they constitute. The question of formalized transcripts is therefore ambiguous, as both grammar and lexicon now convey grammatical information. Hence the question arises as to the proper relationship between the two : which formal features normally belong to the grammar, and which to the dictionary? Which morphological features to morphology, and which to the morphological transcript? Which syntactic features to syntax, and which to the dictionary entry?

256

ALPHONSE JUILLAND

2. GRAMMAR VS. LEXICON

In the past, grammars and dictionaries were envisaged as distinct and separate undertakings. Gleason's introduction to the discussion of "The Relation of Lexicon and Grammar", is characteristic of this outlook: In this paper I am charged with discussing the relation between lexicon and grammar. This might be interpreted as meaning either of two similar, but not identical problems : the relation of grammar (a portion of language structure) with lexicon (another portion of language structure), or the relation of a grammatical statement with a dictionary (both documents or some sort). (1962:85)

Garvin, for his part, puts it more bluntly: "Dictionaries and texts are the customary modes of presenting the corpus of a language, just as grammars — both traditional and modern analytic — are the means of presenting its structure." (1955:1013) Just as characteristic is the stand taken by Mathiot, who, in "distinguishing and separating grammar from dictionary", protests against the approach of the transformationalists, for whom a complete dictionary is scattered through the grammar. It would, of course, be possible to apply a differential marking to every member of each list and then throw all the lists together into a kind of dictionary placed, say, near the end of the grammar. (Householder, 1962:579)

For Mathiot "the above approach obscures what is by us considered a fundamental distinction, namely that between naming behavior and ordinary speech behavior". (1967:5, footnote 10) In recent years, linguists have become increasingly aware of the fact that grammar and dictionary may be interrelated, though the relationship has remained somewhat obscure. (Householder and Saporta, eds., 1962:143-50) For Bolinger, the dictionary ought to be a companion to grammar: "How would a dictionary look if it tried to satisfy the new demands of linguistic science? It would have to be what it is only in a half-hearted way now : a companion to a grammar." (1968: 289). A similar view has been expressed by Hoenigswald, who closed his discussion of the relation between lexicography and grammar on a note of applause and admiration for a development which has gained in strength, and which emphasizes the complementary nature of grammar and the lexicon. This is the grammatical companion sketch which goes with the dictionary. It has had its humble beginnings in the synopses often carried by practical dictionaries: paradigms, lists of irregularities, etc. But these beginnings are hard to recognize in some of its latest specimens. (1962:110)

In the same vein, Householder advises lexicographers "to prepare a complete syntactical grammar of their language as a necessary preliminary to writing their dictionary" (1962:576), a view whose generative verbalization sounds as follows: It appears that the sub-component of syntactic rules which enumerates underlying irarkers is divided itself into two elements, one containing phrase structure rules and the other

ENTRY WORDS: GRAMMARS AND DICTIONARIES

257

containing a lexicon or dictionary of highly structured morpheme entries which are inserted into the structures enumerated by the phrase-structure rules. (Postal, 1964: 253)' Scholars, then, who have dealt with the relationship between grammar and lexicon have insisted on the extent to which the latter presupposes the former. Insofar as we postulate that, in certain fundamental aspects, grammars and dictionaries are mutually presupposing, we go one step further, to envisage them as two sides of the same coin : not only does a dictionary presuppose a grammar, a grammar also presupposes a dictionary. To be sure, decisions made in grammatical analysis are conditioned by the requirements of a consistent, exhaustive, and economical conveying of formal information through the medium of dictionaries; and conversely, decisions involved in the compiling of formally satisfactory dictionaries rest heavily on the conclusions reached in the grammatical analysis. Indeed, insofar as the formal properties of entities entered in a lexicon must be specified with reference to the phonologic, morphologic, and syntactic classes of which they are said to be members, decisions made in the analysis of grammar inevitably affect the compiling of formally satisfactory dictionaries. This has not always been clear, because the relationship between grammar and lexicon is the most neglected among those which obtain between the main areas of language, phonology, grammar (i.e., in inflectional languages, morphology and syntax), and vocabulary. Rooted in the biological nature of man through its PHONOLOGICAL system, the grammatical core of language is linked to the sociocultural environment reflected in the LEXICAL system. While the relationship between phonology and grammar in general, between phonology and morphology in particular 3 , has been explored almost as thoroughly as the intragrammatic boundary which links morphology to syntax4, the relationship between grammar and lexicon has been neglected.8 The distinction has been tentatively rooted in several contrasts: form vs. meaning, morphemes vs. words, grammemes vs. lexemes, classes vs. members, codes vs. messages, and regular vs. irregular features. However, according to Gleason: None of these suggestions for the division of the descriptive task between the grammatical statement and the dictionary seem to provide any very adequate basis for a parallel distinction between grammar and lexicon. That is, none seem to correlate in any satisfactory way with a possible view of language structure that would seem theoretically adequate. The great problem is that most seem to point very strongly toward viewing grammar as an integrated system, and lexicon as a miscellaneous remainder. If lexicon can be given no better • As Bolinger correctly points out, "This is simply to say that words and other similar tightly bound units enter into the structure of phrases in specific ways that are among the rules of a grammar, and that a complete dictionary would have to label its entries according to the rules that apply to them". (1968:289-90) • For a critical account of the debate on grammatical criteria in phonological analysis, cf. section 8.87 in Pike, 1957. 4 For an excellent discussion of the literature bearing on the relationship between morphology and syntax, cf. de Guevara, 1955; for a critical bibliography, cf. Pike, 1957, section 11.721, "On Words and Morphology versus Syntax". • Cf., for instance, Gleason, 1962, sect. 2, "Attitudes Toward Grammar and Lexicon", and sect. 4, "Current Neglect of Lexicon and Dictionaries".

258

ALPHONSE JUILLAND

basis than this, it is clearly not to be considered as a basic segment of language structure in the way that the grammar and the phonology are to be considered (1962:96). In the following paragraphs, we shall examine the more prevalent ways of conceiving of the relation between grammar and lexicon, in order to show not only that the lexicon can be considered "a portion of language structure", on the same level as grammar, but also that grammar and lexicon are complementary notions which mutually presuppose one another.

3. FORM VS. MEANING We may begin by mentioning the traditional way of conceiving of the relationship between grammar and dictionary, which assumes that grammars deal with the formal aspect of language (expression), whereas dictionaries deal with the semantic aspect (content): "One very common implication is that the grammatical statement deals with form, the dictionary with meaning" (Gleason, 1962:90). In view of the difficulties it raises, few modern linguists have been known to defend the traditional concept. These difficulties result from a lack of co-extensiveness between units of form and units of meaning, from differences in the nature of the major lexical categories, and from the semantic requirements of grammars as well as the grammatical requirements of dictionaries. First, since the unit through which dictionaries convey information is the conventional word, they cannot handle well meanings associated with units larger or smaller than words. This explains proposals that the unit of dictionary information be the smaller morpheme (see following section), just as it accounts for suggestions that units larger that words, such as phrases or idioms, be incorporated among dictionary entries (Garvin, 1955:1017). Even in regard to words, difficulties subsist in attempting to define functional words. If the meaning of lexical words, such as nouns, verbs, adjectives, or adverbs can be handled reasonably well, that of functional words, such as prepositions or conjunctions, is difficult to pinpoint. To be sure, the meaning of function words depends on their use, hence on formal and distributional properties normally assigned to the grammar: "With function words, of course, the dictionary definitions, generally thought of as statements about meaning, are actually largely statements about use, as may be seen in any good English dictionary in the articles for the, not, etc." (Gleason, 1962:91). The question is further complicated by the semantic requirements of grammars, as not all modern linguists have followed the lead of American descriptivists, who consider that "meanings in a grammatical statement are quite incidental, and from a rigorous point of view out of place." Indeed, for many grammarians, "the giving of such meanings remains an important function of the grammatical statement itself, perhaps even its main function" (Gleason, 1962:90).

ENTRY WORDS: GRAMMARS AND DICTIONARIES

259

But the major difficulty involved in traditional attempts at rooting the grammar/ lexicon division in the distinction between form and meaning is encountered in the grammatical (formal) requirements of dictionaries, a topic which forms one of the main objects of this paper.

4. MORPHEMES VS. WORDS It has been assumed that grammars focus on the minimal morphologic constituent, the morpheme, whereas dictionaries convey information in terms of the minimal syntactic unit, the conventional word. The weakening of the word concept in modern linguistics and the increased emphasis placed by structuralism on the morpheme, has prompted suggestions that lexical information be conveyed in terms of morphemes rather than words. This ambiguity has been well summarized by Garvin : Although it has been customary in North American Indian studies to equate dictionaries with lists of morphemes of all classes, this is so clearly at variance with established lexicographic practice in the better known languages that the question of the kinds of units to be included in the dictionary (as opposed to the units properly discussed in the grammar only) deserves some serious attention. Lexicographic tradition in the better known languages has it that a dictionary properly is a collection of words. Although the word is not rigorously defined but simply taken for granted, it is delimited for purposes of lexicography to the extent that sequences with varying paradigms but containing the same, or in some ways unsubstitutable, thematic morphemes, are considered "different grammatical forms" of "the same word." One of these forms of the word is then traditionally chosen as the entry form for the dictionary. More extensive dictionaries attempt to include words of all classes; in more limited ones often full words (such as nouns, verbs, or adjectives) are better represented than particles, prepositions or the like. More recently, fixed multi-word phrases and clauses, the so-called idiomatic expressions, and other units consisting of more than one word, have found inclusion in modern dictionaries. This raises the question whether a dictionary is really but a collection of words, or — to put it more specifically — what are the proper units of lexicography. (Garvin 1955: 1015-16) Bloomfield was the first to define the lexicon in terms of morphemes: "The total stock of morphemes in a language is its lexicon" (1933:162). Gleason related this trend to a ... reluctance to view meanings as any other than improper intruders in a grammatical statement that has led many descriptive linguists to minimize this by removing much of it to the dictionary. They tend to make the dictionary not so much a list of words as of morphemes. (1962:90) The same scholar considers that Mager's Gedaged-English Dictionary is "in many respects what any dictionary ought to be", i.e., "a brief ethnographic encyclopedia combined with a glossary and linguistic definition of the morphemes of the language" (Gleason, 1955:165). A stronger stand is taken by Malone, for whom "...the main

260

ALPHONSE JUILLAND

thing is to make the morpheme, not the word, our unit in determining what items to include in the inventory" (1962:113, 117).« However, not everyone agrees that conveying lexical information in terms of morphemes rather than words would represent an improvement. Hoenigswald, for instance, thinks that: Even if we could agree that minimal, one morpheme length is ideal for dictionary lemmata, there is still the question of completeness, and one particular aspect of it which impinges on grammar. There are, in the view of many linguists, morphemes and morpheme-like entities with so-called grammatical meanings. Is it reasonable, for instance, to consider the sentence intonations of English on a par with the segmental morph-morphemes? Many of us would answer in the affirmative and would relegate the problem of whether or not to include them in a dictionary to the realm of mere expedience — and probably leave it there until we know more about the facts. What about constructions? Where would it lead us if we were to consider constructions, in the abstract, as items to be classified in the grammar but listed in the lexicon? The least difficult side of this question arises where segmental morphs form morphemes with grammatical meanings. In fact, both the theory pertaining to such forms and their actual lexical treatment are enlightening. As for the theory, it is surely trüe that the discovery of their special nature is part of the co-occurrence study to which all morphemes may be subjected. There is, therefore, no reason not to list them, along with other morphemes in the dictionary. (Hoeningswald, 1962:107) Mathiot is more direct: The decision to exclude morphemes and morpheme clusters from the dictionary was made for the following reasons: the grammatical unit best suited to the purpose of a practical dictionary is the word, and not the morpheme, since the user cannot be expected to know how to identify and look up morphemes, whereas words are traditionally set off in the orthography by spaces, a tradition which we followed in devising an orthography for Papago (1967:2). Whatever the merits of morpheme dictionaries, claims that they would be more economical insofar as there are fewer morphemes per language than words, do not stand up under closer scrutiny. First, the fact that dictionary headings are words rather than morphemes — constructions consisting of a stem and an ending instead of a single morpheme such as a bare root, an affix, or a flexive — does not necessarily mean that the lexical information provided in the entry proper is about words rather than morphemes, about morphologic constitutes rather than constituents. Indeed, word variants such as singular nominatives or infinitives, are selected as entry heads for their 'representative' character, inasmuch as their endings reveal the main distributional properties of the roots or stems to which they are attached : not only that • Malone (1962:113,117): "We come now to morphology, or the analysis of utterances into their minimal units of meaning. In the entries of the dictionary devoted to single morphemes the entry-form itself represents such an analysis, of course, but all morphemic sequences (i.e., phrases and polimorphic words) that are entered need to be analyzed into their constituent morphemes. This is done by marking the morphemic boundaries, so far as they are not already shown in the conventional writing by spacing or hyphenation. The hyphen is the obvious mark to use for this purpose. Morphological analysis enables one to determine the systems of inflection, derivation, and composition characteristics of the vocabulary of the language in question."

ENTRY WORDS: GRAMMARS AND DICTIONARIES

261

these combine with nominal rather than verbal flexives or vice-versa, but also that they are modified exclusively by nominal morphemes of the first declension or by verbal morphemes of the third conjugation. This is to say that selecting as entry-head a whole word instead of a single morpheme is often a device for conveying information about certain FORMAL properties of the constituents, the privileges of occurrence of roots relative to morphemes or paradigms of morphemes of the affixal or inflectional type, or vice-versa.' Moreover, the assumption that dictionaries of morphemes are more economical than dictionaries of words is largely illusory, notwithstanding the obvious fact that there are fewer morphemes per language than words. To be sure, the economy achieved in the lexicon by resorting to morphemes has to be compensated for in the grammar, where additional information about stem and word formation will be required. Owing perhaps to the same weakening of the word concept,8 the traditional disciplines which account for the rules governing the combination of roots with affixes into stems and of stems with flexives into words, have been largely ignored in the newer linguistics. Now, it is precisely because the information conveyed by the traditional Stamm- and Wortbildunglehre is implicitly contained in dictionaries of words, that modern grammar can afford to deal, only perfunctorily with the sub-levels of an inflectional morphology. Modern grammars restrict information of this kind to the listing of affixes, and to loose statements about their function and meaning, a superficial treatment acceptable only as long as the information about restrictions upon the privileges of occurrence of roots and affixes relative to one another and to the word (and about the functions they assume in different stem or word constructions) is conveyed implicitly in the entries of word-dictionaries. If dictionaries of morphemes were to separate endings from stems and affixes from roots, intra-word distributional information would have to be transferred to ad hoc sections of the grammar, and the economy achieved in the dictionary by resorting to morphemes instead of words would be offset by the more extensive sections grammars would have to devote to stem and word formation. While dictionaries of words can be based on morphological statements consisting essentially of an inflectional morphology (Formenlehre), dictionaries of morphemes presuppose grammars containing much more elaborate sections dealing with stem and word formation (Stamm- and Wortbildunglehre). The point is that, in one way or another, RELEVANT INFORMATION MUST BE CONVEYED: if one chooses to eliminate it from the dictionary by restricting the basic unit to the morpheme, space must be found for it in the grammar; if one eliminates it from the grammar, one must make a place for it in the dictionary by extending the basic unit to constructions (morpheme sequences) largely co-extensive ' Cf. Hoenigswald (1962:109): "In conclusion, it may be said that our dictionaries — those of the classical languages among them — carry a great deal of grammatical information and that we have every reason to be thankful for this mixed tradition. One's complaint is certainly not that they have compromised a rigid ideal demanding an abstract dichotomy; this is an ideal which it would be preposterous to uphold anyway." * On the weakening of the word concept in modern linguistics, (cf. Pike, 1957).

262

ALPHONSE JUILLAND

with traditional words. As a matter of principle, we may assume that THE NECESSARY AND SUFFICIENT INFORMATION TO BE CONVEYED BY EXHAUSTIVE DESCRIPTIONS IS GIVEN ONCE AND FOR ALL, AND THAT THE LINGUIST'S DECISIONS AFFECT ONLY ITS DISTRIBU-

More simply, a smaller dictionary presupposes a larger grammar, just as a smaller grammar presupposes a larger dictionary.

TION AMONG THE VARIOUS SECTIONS OF A FORMALIZED TREATISE.

5. LEXEMES VS. GRAMMEMES It has also been suggested that the distinction between grammar and lexicon can be established with reference to the distinction between LEXEMES and GRAMMEMES, which are categories of morphemes largely co-extensive with the roots and non-roots of traditional grammar: crudely, the former express the terms of linguistic propositions or some of their properties, while the latter express the relations which obtain among terms, or between terms and grammatical categories (Juilland, 1961: section 2).· The assumption is that lexemes belong to the lexicon and grammemes to grammar. In turn, separable grammemes, such as articles, prepositions, conjunctions, joint pronouns, auxiliaries, etc., belong to syntax (cf. chapters entitled "syntaxe des prépositions"), whereas inseparable ones, such as affixes and flexives, belong to morphology. Although the criteria for distinguishing between the two categories are sometimes ambiguous, interesting attempts have been made to distinguish between the two classes both formally and semantically. A possible way of characterizing the distinction is to say that a grammar stresses recurrences, i.e., patterns, whereas a dictionary stresses unique characteristics, i.e., items (individual agents). Linguistic patterns have been observed to be finite, thus allowing for an exhaustive listing. Linguistic items, on the other hand, have been observed to belong to two types of classes: (a) classes of unrestricted membership, such as roots and bases or nouns and verbs which do not allow exhaustive listing, and (b) classes of restricted membership, such as derivational and inflectional morphemes, or particles, which do allow exhaustive listing. We shall refer to these classes from now on as unrestricted classes and restricted classes respectively. (Mathiot, 1967:9) The semantic differences between the two classes have been defined as follows: There can be observed a relation between the distinction of restricted versus unrestricted classes on the one hand, and certain semantic characteristics on the other. The meanings of the majority of the members of restricted classes to be listed in the grammatical glossary cannot be elicited. They have to be inferred from a great variety of contexts. On the other hand, the meanings of the majority of the members of unrestricted classes to be listed in the lexicon can be obtained by elicitation from informants (with more or less ease, depending • Halle (1959) notes that "a distinction seems to be necessary between grammatical and lexical morpheme classes and that lexical morphemes must be introduced in the representation before grammatical morphemes; but he too provides that the rules which 'compromise the dictionary of the language' be incorporated within the grammar."

ENTRY WORDS: GRAMMARS AND DICTIONARIES

263

on circumstantial factors such as the informant's fluency in the investigator's language). We refer to the first type of meaning as relational meaning and to the second as content meaning. (Mathiot, 1967:12) Therefore, ... grammar and lexicon can be said to constitute two separate linguistic dimensions. Each dimension has its own set of analytic units to be defined in their own terms. While grammatical units are defined in formal terms, no definition is yet available for lexical units. However, we have ascertained that they are characterized by lexical meanings. (Mathiot, 1967:14)» But regardless of its merits, the distinction between lexemes and grammemes, content and function words, etc., does not answer the question of the relations which hold between the two categories of morphological constituents, between roots and nonroots or stems and endings. D o these relations belong to the grammar or to the dictionary? To both or neither? Notice that roots or stems may have variants whose selection depends on certain properties of the endings with which they combine, just as the occurrence of endings may depend on the properties of the roots or stems they select. Moreover, certain stems combine only with certain inflectional morphemes, e.g., members of the first declensional paradigm or of the third conjugational paradigm etc., or vice-versa, certain endings combine only with certain stems, e.g., masculine, or ending in a vowel, and not with others, e.g., feminine, or ending in a consonant. It is customary to deal with these relations in the grammar, to which good dictionaries usually refer. However, it is interesting to note that there is nothing necessary about 10

Mathiot's attempt to parallelize grammar and lexicon with "ordinary speech" and "naming" behavior deserves to be quoted in full : "We propose that grammar and lexicon correspond to two types of behavior, namely, ordinary speech behavior and naming behavior respectively. By naming behavior we mean the cultural activity which consists in the use of particular speech forms with lexical meanings as 'names' to refer to cultural phenomena (such as, for instance, 'objects', 'events', and 'qualities'). "We consider that naming behavior differs from ordinary speech behavior in terms of the particular type of awareness which characterizes it. Thus, naive native speakers reveal their actual (i.e., spontaneous, untaught, immediate and overt) awareness of their use of 'names'. This can be inferred (a) from responses such as : 'This is the way we call it', or 'We have no name for that', or Ί have forgotten its name', or Ί have the name for it on the tip of my tongue', and so on ...; (b) from the common practice of giving 'names' in one language while commenting on them in another (usually names in the informant's language and comments in the investigator's. "By contrast with the preceding, the awareness on the part of the native speakers of their use of grammatical units — such as, for instance, morphemes and words — can be characterized as follows : "(1) It is only potential, the use of these units is not accompanied by a spontaneous and overt awareness, but it can be made conscious by teaching or experience. We infer this from the observation that, for instance, the grammar of their own language can be taught to native speakers and that informants are capable of learning how to recognize certain morphemes in the course of their participation in linguistic field work. "(2) It does not apply equally to all types of grammatical units. There seems to be a higher potential level ot awareness of some grammatical units than of others. We infer this from the observation that the meaning of certain morphemes — such as, for instance, some derivational morphemes — is easier to ascertain and to teach to native speakers than that of other morphemes — such as, for instance, some inflectional morphemes." (1967:15-6)

264

ALPHONSE JUILLAND

assigning grammemes to grammars and lexemes to lexicons, i.e., about the practice of subjecting non-roots to GRAMMATICAL TREATMENT by grouping them into categories and classes (paradigms), and roots to LEXICAL TREATMENT by individually specifying their grammatical class memberships. Other considerations aside, this procedure is determined by the fact that there are many more roots than non-roots in a language, though the occurrence of non-roots is much higher in actual texts. Since the purpose of the analysis is to reduce the description without impairing its exhaustiveness, by making statements about classes instead of their individual members, it is only natural to characterize the more numerous by the less numerous, hence to group roots into a smaller number of classes with reference to the non-roots with which they combine, rather than vice-versa. It is, then, the higher ratio of roots that makes it convenient to deal with paradigms of endings rather than paradigms of stems, and to operate with INFLECTIONAL CLASSES OF STEMS rather than with THEMATIC CLASSES OF ENDINGS. Formally, grammar and lexicon can be defined with reference to the fact that GRAMMARS DEAL WITH THE CATEGORY OF MORPHOLOGICAL CONSTITUENTS WHOSE FEWER MEMBERS HAVE A CONSIDERABLY HIGHER FREQUENCY OF OCCURRENCE IN TEXTS, WHEREAS DICTIONARIES FOCUS ON THE CATEGORY OF MORPHOLOGICAL CONSTITUENTS WHOSE MORE NUMEROUS MEMBERS HAVE A CONSIDERABLY LOWER FREQUENCY OF OCCURRENCE IN TEXTS. To clarify the point, imagine the special language of some decaying civilization, preserved only in ritualistic form, a language in which a few religious concepts are conveyed through a very elaborate system of grammatical relations expressed by a variety of affixes and flexives, a language with MORE GRAMMEMES THAN LEXEMES, with more affixes than roots, and more flexives than stems. In describing such a language synchronically (rather than as the survivor of an earlier, lexically richer stage), it would be simpler to specify stem-ending relations by grouping the more numerous endings with reference to the less numerous stems, hence to establish PARADIGMS OF ROOTS instead of PARADIGMS OF ENDINGS, to characterize endings thematically instead of characterizing stems inflectionally. Consequently, the information normally assigned to the grammar would be treated in the dictionary and the materials usually assigned to the dictionary would be treated in the grammar: the description of such a language would lead the investigator to compile a GRAMMAR OF ROOTS and a DICTIONARY OF ENDINGS, whose mutual relations would be specified by determining THEMATIC CLASSES OF INFLECTIONS instead of INFLECTIONAL CLASSES OF STEMS. However extreme, this imaginary case illustrates the point that, in their formal aspect, grammars and dictionaries can be viewed as two sides of the same coin. While the constituency of inflectional constructions makes it convenient to assign stems and endings to different sections of an exhaustive description, their repartition to grammars and dictionaries is partly conditioned by the quantitative characteristics which govern their complementary distribution in systems and texts.

ENTRY WORDS: GRAMMARS AND DICTIONARIES

265

6. CLASSES VS. MEMBERS Better perhaps than in terms of the grammeme-lexeme distinction, the question of the respective domains of grammar and lexicon can be answered with reference to the distinction between CLASSES and MEMBERS. The parallelism has been suggested, but its formulation has not been clear. For Gleason, the grammatical statement... takes care of all relationships between classes. It says nothing at all about words, morphemes, or any other items, only about classes which bear a very special kind of relationship with the items. We ordinarily call the latter "members" of the classes, but we must be very clear that this means nothing like being part of the classes. The class is given a very different status than was suggested in the last section. (Gleason 1962: 94) Since establishing the RELATIONSHIP between classes is only one among four or five grammatical operations, it is difficult to see why this particular one, systematization, should distinguish grammars from dictionaries better than segmentation, identification, or classification (Juilland, 1961 : section 3). More pertinent remarks on the subject were made by Hoenigswald, who establishes a parallel between grammars and phonic SYSTEMS on the one hand, dictionaries and phonemic INVENTORIES on the other hand, to consider dictionaries as lists and grammars as systems: One fundamental aspect of dictionaries is surely the fact that they are lists — lists of 'items' above all, although perhaps also lists or 'arrangements' (the boundary is notoriously difficult to draw). On the other hand, neither grammars nor semantic treatises, whatever they may be, are essentially examples of listing per se. (Hoeningswald 1962:110) This concept, as Sledd remarks, is not in agreement with transformationalist doctrine : Mr. Hoenigswald, in turn, seems both to begin and end on a note of disagreement with the transformationalists. Contrasting grammars with dictionaries in his first paragraph, he says that "dictionaries are essentially lists"; and in his conclusion he characterizes grammar and lexicon as "complementary ways of organizing the corpus", the text which is more basic than either. For transformationalists, however, a taxonomic grammar is itself "a set of lists", lists of "phonemes, morphophonemes, morpheme classes, and morpheme class sequences"; and "organizing the corpus" is simply a way of rearranging the data which tells us nothing new, leads to no deep generalizations, and finds no explanation for the listed facts. (Sledd, 1962:143—50) These divergent views suggest the need for a tighter formulation, if the grammar/ lexicon distinction is to be fruitfully parallelized with the class/member dichotomy. Perhaps the misunderstanding results f r o m a failure to grasp the nature of mutually presupposing concepts, which do not lend themselves well to separate treatment: since classes and members cannot be grasped independently, it is difficult to distinguish dictionaries from grammars in those terms, as the two must deal with BOTH classes AND members. The difference is one of emphasis and orientation: grammars establish classes by grouping members on the basis of the formal properties they

266

ALPHONSE JUILLAND

exhibit and with reference to the relations they contract, while dictionaries determine members by specifying their formal properties with reference to the classes established in the grammar (Jutland and Macris, 1962). According to this concept, formalized entry forms are SEQUENCES OF SYMBOLS OF CLASS MEMBERSHIP, the function of each symbol being to signal the phonologic, morphologic, and syntactic classes to which the symbolized entries have been assigned in the grammar. In decreasing order of generality, i.e., reading from right to left, the symbols in a morpho-lexical transcript such as Rumanian iubi 'to love', which is {jub-ESK=i v}, must be interpreted as follows : ν signifies that the symbolized base is a member of a verbal rather than of a nominal («), adjectival (a), pronominal (/>), etc., class; =i signifies that the verb is a member of the fourth conjugational type, whose characteristic endings exhibit the vowel /i/, in contrast to a verb, such as pleca 'to leave' {pLEK«a}, whose endings exhibit /a/, etc. (Juilland and Edwards, 1971); -ESK« signifies that this verbal belongs to a suffixed sub-class whose root is extended by the variants -ésk-, éît-, and gásk* in Present 1, 2, 3, and 6, in Subjunctive 1, 2, 3, and 6, and in Imperative 3 and 6, in contrast to non-suffixed si verbs such as {vln=i}, whose stem is never extended and always consists of the bare root; the capitalized symbols of the suffix, -E-, -S-, and -K-=, signify that this verb is a member of the morphophonemic classes whose alternants differ by means of the -e- ~ -ga-, -s- ~ -s-, and -k- ~ -ë- alternances, in contrast to non-alternant =i verbs such as {sui=i v}, whose stems are not affected by alterances of the morphophonemic type;11 lack of an accentual mark signifies that the symbolized form is a member of the regular accentual class, in contrast to other verbs such as {búkur=a ν}, whose acute mark signals that the verb belongs to an accentually irregular class, whose members bear the stress on a syllable other than the last. Even the phonemic symbols can be interpreted in the same way, by assuming that a stem belongs to as many (morpho)phonemic classes as it contains (morpho)phonemic positions, e.g., cinta 'to sing' {kinTsa v} belongs to the phonemic class /k/ by its first position, to /i/ by its second, to /n/ by its third, and to the morphophonemic class /T/ by its fourth (Juilland and Edwards, 1971). In other words, a correctly symbolized form should belong to AT LEAST AS MANY FORMAL CLASSES AS THERE ARE SYMBOLS IN ITS MORPHOLEXICAL TRANSCRIPT, just as a consistent morpholexical transcript should contain NO MORE SYMBOLS THAN THE NUMBER OF CLASSES TO WHICH THE SYMBOLIZED FORM HAS BEEN ASSIGNED IN THE

In agreement with the principle of economy, which requires that the necessary and sufficient information be conveyed at minimal cost (Apostel, Mandelbrot and Morf, 1957), i.e., by a minimal number of symbols, THE SYMBOLIZED UNIT GRAMMAR.

11

Hoenigswald (1962:106), "And as for morphophonemics, it is fair to say that it represents the area of grammar (if we wish to include it there) which is most fully and most anciently treated in our dictionaries. To be sure, the way in which this treatment is executed varies greatly with various language structures and various strains of tradition. So-called automatic alternation need not of .course be treated at all. Other alternations may require cross-references from allomorph directly to the grammar. The current dictionaries of the classical languages (which I would like to use as examples) exemplify this."

ENTRY WORDS: GRAMMARS AND DICTIONARIES

267

MAY BE A MEMBER OF MORE, BUT OF NO FEWER FORMAL CLASSES THAN THE NUMBER OF

The purpose is to achieve maximal economy by conveying all relevant information by a number of symbols smaller than the number of classes to which the symbolized entity is said to belong. This can be done by exploiting the relations of implication or exclusion which obtain between classes, membership in the presupposed classes being inferred from membership in the presupposing ones, thus rendering superfluous the use of overt symbols in the morphological transcript: whenever such relationships appear to hold in the system, rules of selection can be introduced in the grammar, making dictionary specification of class membership unnecessary. Formal conditions for achieving maximal reduction in the means of symbolization are somewhat complex, as they involve the degree of generality of the criteria applied, as well as the degree of specificity of the classes yielded. A grammar implicitly determines a morpholexicon; and, vice-versa, a formally satisfactory dictionary presupposes a grammar; this relationship can be made explicit by introducing a dictionary through its grammar (Apostel, Mandelbrot and Morf, 1957) or by appending a morpholexicon to every morphology.12 We see why certain linguists insist that grammars consist of inventories which list the members of grammatical classes. The purpose of formalized dictionaries is also to list: however, they list grammatical properties or symbolize grammatical class memberships in the formalized heads of the various entries. As far as the listing of entries is concerned, the dictionary order is arbitrary (e.g., alphabetical). The same is true of grammatical listing, but only within the classes to which the entities are assigned in the final grammatical operation, CLASSIFICATION (Juilland and Edwards, 1971: section 1.31). SYMBOLS USED IN ITS TRANSCRIPT.

7. REGULAR VS. IRREGULAR FEATURES

A related concept, which views the distinction between grammar and lexicon as roughly equivalent to the distinction between REGULAR features, assumed to belong to the grammar, and IRREGULAR features, assumed to belong to the dictionary, must be briefly considered. It was explicitly formulated by Bloomfield: "The lexicon is really an appendix to a grammar, a list of basic irregularities" (1933:274). After discussing the descriptive linguist's dislike for irregularities, Gleason concludes that To a very real extent, we have made the dictionary exactly what Bloomfield suggests — a list of irregularities, the refuse of grammatical descriptions, the repository of things put aside because they would give a clear and incisive picture of a pervading structure (for some of us nothing more than a fancy synonym for irregularity) which the descriptive linguist finds aesthetically pleasing. (1962:87) " Bloch concludes his analysis of English verb morphology by a list of irregular verbal bases introduced as follows: "The following list shows the bases of irregular verbs as they might appear in the lexicon" (1957:48) (Italics ours).

268

ALPHONSE JUILLAND

He asks: "If this criterion were to be followed quite literally, I believe that it would destroy any active conception of grammar and render grammatical statements worthless." (Gleason, 1962:93) The question underlying Gleason's answer presupposes a misleading dichotomy : rules are supposed to apply to a PLURALITY OF ITEMS or to a SINGLE ITEM; differences in the scope of the rules are of the more/less rather than the either/or type: in most instances one moves from classes of very large membership, or from rules which apply to ALL items, to rules applying to MANY items, to classes of decreasing membership, or rules applying to a smaller number of items: from thousands to hundreds to half a dozen, then four or five, two or three, down to one-member classes or to one-item rules. Often, there is a GRADUAL TRANSITION from classes consisting of thousands of items to rules which apply to one single item. Now, if one-member classes belong to the lexicon, should two-member classes be transferred to the grammar? Should rules applying to three items be assigned to the lexicon, and those applying to four items to the grammar? Gleason's "puzzling" (Sledd, 1962:149-50) distinction between "loose" and "tight" structure (Gleason, 1962:91-2) seems to be a loose formulation of the situation just described, "loose" structures being those exhibited by languages in which the transition from classes with large membership to classes with small membership (or down to one-member classes) is gradual, while "tight" structures are those which offer no "irregularities", i.e., classes with small membership or one-member classes, where the irregularities are clearly marked, i.e., the transition from classes with large membership to classes with few members is abrupt rather than gradual. Strictly speaking, "irregularities" exist only in an either/or perspective, which distinguishes exclusively the one from the many; in a more/less perspective, distinguishing between one and two as well as between one and one hundred, there are no 'regular' and 'irregular' features, only classes with larger or smaller memberships, ranging in theory from all to none. Rules, then, apply to a larger or smaller number of units, up to rules which apply to all or down to rules which apply to only one. In this perspective, drawing a line between grammar and lexicon is a matter of expediency, in which statistical considerations are as much involved as structural or formal criteria.

8. CODES VS. MESSAGES

In view of the objective we assign to formalized entry heads, which is to convey a certain amount of information with a minimal number of symbols, the relationship between grammar and lexicon, here between morphological classes and morpholexical transcripts, can be viewed as a problem in communication, and conceived of in terms of information theory. The task of writing a grammar, which is to establish classes along with a means of symbolizing them, can be equated with the task of code construction; whereas the task of devising morpholexical transcripts which are to convey

ENTRY WORDS: GRAMMARS AND DICTIONARIES

269

information about the formal properties of linguistic entities in terms of symbols assigned in the grammar, can be equated with codes and morpholexical transcripts with messages, the compiler of a formalized dictionary can be compared to an encoder, and its user to a decoder. The problem confronting the linguist as a codemaker and message-sender is to devise such codes as would enable him to send minimally redundant messages. In terms of our discipline, the task of a linguist as a grammarian and lexicologist is to establish grammatical classes and to determine the relationships which obtain between them, so as to reduce to a minimum the number of symbols per entry, without any loss of relevant information. This confirms the inseparability of these fundamental aspects of language : a dictionary can no more be separated from its grammar than a set of messages can be dissociated from the underlying code.

9. CONCLUSIONS

The question of the respective domains of grammar and lexicon has been generally approached on a complementary basis, on the model of the better established and more thoroughly studied divisions of phonology vs. grammar, or, within grammar, of morphology vs. syntax: defined in terms of their respective objects, phonology deals with phonological units (phonemes), grammar with grammatical units (morphemes, logemes, tagmemes); morphology with units smaller than words (morphemes), syntax with units larger than words (phrases, clauses, sentences). In the same way, grammars and dictionaries are assumed to deal respectively with form and meaning, morphemes and words, grammemes and lexemes, regular and irregular features, classes and members, etc. Since these complementary concepts all give rise to various difficulties, a more sophisticated understanding appears desirable, conceiving of both grammar and lexicon as mutually presupposing rather than simply complementary. Indeed, both grammars and dictionaries deal with both classes and members, the difference being one of emphasis: grammars determine classes by grouping members, whereas lexicons determine members by specifying class memberships. Such an interpretation assumes that the necessary and sufficient information conveyed by exhaustive descriptions is given once and for all, the linguist's decisions affecting only the distribution of the subject matter among the sections of a formalized treatise : a smaller dictionary presupposes a larger grammar, and a smaller grammar a larger dictionary. In this perspective, formalized entry-heads are envisaged as sequences of class memberships, the function of each symbol being to indicate the classes to which the symbolized forms have been assigned in the grammar. A form should belong to at least as many classes as there are symbols used in its morpholexical transcript; a correct morpholexical transcript should contain no more symbols than there are classes to which the symbolized form has been assigned in the grammar. By exploiting

270

ALPHONSE JUILLAND

the relations of implication and/or exclusion that hold between classes, a symbolized form may be assigned to more but to no fewer classes than the number of symbols used in its transcript. In this approach, the task of the linguist as a grammarian and lexicographer appears parallel to that of the cybernetician as a code-maker and message-sender: he must devise such codes as would enable him to send exhaustively intelligible but minimally redundant messages, i.e., construct such grammars as would permit him to convey relevant formal information through the shortest possible morpholexical transcripts. STANFORD UNIVERSITY REFERENCES Apostel, L., B. Mandelbrot, and A. Morf, 1957 Logique, langage et théorie d'information (Paris). Bloch, Β., 1957 "English Verb Inflection", Readings in Linguistics, M. Joos, ed., pp. 243-52. Bloomfield, 1933 Language (New York). Bolinger, D., 1968 Aspects of Language (New York). Garvin, P. L., 1955 "Problems in American Indian Lexicography and Text Edition", Anais do XXXI Congr. Int. de Americanistas (Sâo Paulo), 1013-38. Gleason, Jr., Η. Α., 1955 Review of J. F. Mager, Gedaged-English Dictionary, Language 31:165. 1962 "The Relation of Lexicon and Grammar", in F. W. Householder and S. Saporta, eds., Problems in Lexicography (Bloomington, Indiana), 85-102. de Guevara, A. LI. M., 1955 Morfologia y syntaxis: el problema de la división de la grammatica (Granada). Halle, M., 1959 The Sound Pattern of Russian (The Hague). Hoenigswald, H. M., 1962 "Lexicography and Grammar", in F. W. Householder and S. Saporta, eds., Problems in Lexicography (Bloomington, Indiana), 103-10. Householder, F. W., 1962 "Lists in Grammars, Logic, Methodology and Philosophy of Science", Proceedings of the I960 International Congress, E. Nagel, P. Suppes, and A. Tarki, eds. (Stanford), 567-76. Householder, F. W., and S. Saporta, eds. 1962 Problems in Lexicography (Bloomington, Indiana). Juilland, Α., 1961 Outline of a General Theory of Structural Relations (The Hague). Juilland, A. and P. M. H. Edwards, 1971 The Rumanian Verb System (The Hague). Juilland, A. and J. Macris, 1962 The English Verb System (The Hague). Malone, K., 1962 "Structural Linguistics and Bilingual Dictionaries", F. W. Householder and aai saporta, eds., Problems in Lexicography (Bloomington, Indiana), 103-10. Mathiot, M. 1967 "The Place of the Dictionary in Linguistic Description: Problems and Implications", Language 43: 103-24.

ENTRY WORDS: GRAMMARS AND DICTIONARIES

271

Pike, K. L., 1957 Language: In Relation to a Theory of the Structure of Human Behavior (Glendale). Postal, P. M., 1964 "Underlying and Superficial Linguistic Structure", Harvard Educational Review 34: 247-66. Sledd, J., 1962 "Comments", in F. W. Householder and S. Saporta, eds., Problems in Lexicography (Bloomington, Indiana), 143-50.

PETER LADEFOGED

PHONETIC PREREQUISITES FOR A DISTINCTIVE FEATURE THEORY

The sounds of one language are often difficult to describe in terms of the categories used for describing another; and the more one looks at the languages of the world, the more one seems to have to increase the number of phonetic categories required for making adequate descriptions. Whether this is so or not depends in part on what one means by making an adequate phonetic description. We will assume that an adequate phonetic description must be able to specify both the linguistic oppositions (such as the phonemes) which occur within a language, and also the characteristics of that language as opposed to other languages. Given this we may then ask whether we can set a limit on the number of phonetic categories required for describing the languages of the world. It seems that we can do so provided that we restrain our setting up of categories in several ways. The most important is that categories should be regarded primarily as ways of classifying only the linguistic contrasts within a language. Thus, to take an example, the English b in bet is different from the French b in bête, in that in French the vibrations of the vocal cords occur throughout the stop closure, whereas in English they often do not. But we do not need to specify these two b sounds in terms of different categories as long as we are limiting ourselves to specifying only the oppositions within each language. This is the essential insight behind Jakobson's concept of distinctive features. In a long series of publications (Jakobson, 1962, Jakobson, Fant and Halle, 1951, Jakobson and Halle, 1956) he and his colleagues developed definitions of about 14 binary oppositions which they thought sufficient to characterize all possible phonemic contrasts. Although their categories may not be altogether appropriate, the concept of a distinctive feature specification has been shown to be remarkably useful (e.g. by Halle, 1964, and by Chomsky and Halle, 1968). Provided that we can find a rigorous way of deciding when an opposition in one language (such as p-b in English) can be equated with a similar opposition in another language (such as p-b in French) then a definition of a limited set of features seems not only practical but also desirable as part of a general theory of language. Nearly all theories of linguistic description have assumed that it is possible to describe languages in terms of a set of general phonetic categories. But we should note

274

PETER LADEFOGED

that in the works of the older great phoneticians such as Bell (1867) and in the prephonemic era of the development of the IPA, there was often no clear distinction between categories necessary for specifying the oppositions within a language, and the additional theoretical apparatus required for characterizing that language as opposed to all others. Even at a much later date, Joos (1957) and other linguists seem to maintain that there should be no such distinction and to advocate ad hoc description for each language; Hockett (1955) has said that "it is impossible to supply any general classificatory frame or reference from which terms can be drawn in a completely consistent way for the discussion of every individual language." But, as has been frequently pointed out (Jakobson and Halle, 1956; Chomsky and Miller, 1963), all linguists (including those cited above) need phonetic descriptions which imply some absolute frame of reference. Unless we are able to use categories such as alveolar and bilabial, which refer to observable phenomena, we cannot describe the first and last sounds in bib as being in some way the same, and also know that the first sound in bib is not to be identified with the last sound in did (cf. Fischer-J0rgensen, 1952). Furthermore, when we compare languages at the phonological level we must have some common frame of reference. Otherwise we could do no more than count oppositions, without even knowing what proportion of the oppositions were between vowels and what between consonants in each language. Of course, it is possible to take the view that each language has to be described as an entity in itself, and that we can never compare languages. I would rather hold the view that the framework required for comparing languages may be very large and very complicated; but our job as linguists requires us to define it. The position taken here is that the frame of reference is much simpler when we separate out the task of specifying the oppositions within a language from that of characterizing phonetic differences between languages. An explicit way of making this distinction has been proposed by Halle (1964). He suggests that it is possible to use the Jakobsonian distinctive features (which he assisted in developing) as a set of binary categories for classifying the oppositions within a language; and then, at a later stage in a series of rules, give a more detailed phonetic account of the characteristics of the language under description by replacing the binary categories with a quantitative specification of the relative values for each feature. In the case of the differences between b in English and French, we would therefore begin the description of each language by using categories which distinguish the oppositions between each of these sounds and all other contrasting items within each language, before going on to consider what it is that characterizes English b as opposed to French b. The first step might involve stating whether each sound was, or was not, more voiced, bilabial (or grave), and interrupted (like a stop) than the contrasting sounds; and the second might involve stating the degree of voicing (and other concomitant features) present in each case. It is useful to classify the sounds of a language in terms of such general categories at an early stage in the set of ordered rules which form the description; the classificatory categories may then be used for dividing the linguistic oppositions within each

PHONETIC PREREQUISITES FOR A DISTINCTIVE FEATURE THEORY

275

language into a number of natural classes, the members of which operate in the same way from the point of view of the grammar of the language. Thus, in English, we need a class of voiced sounds as opposed to voiceless sounds so that we can make simple statements about the form of the plural suffix in nouns. But if our description of English is to be complete, we must, at a later stage, specify how much voicing there is in each sound in each circumstance, and thus characterize precisely what it is that makes an Englishman sound like an Englishman even when we are paying no attention to the meaning of what he is saying. As Firth (1957) has said "it is part of the meaning of an American to sound like one." This is an unconventional use of the word meaning, which has not always been understood by Firth's detractors (Langendoen, 1964; but see Langendoen, 1968). It would probably be more acceptable if we rephrased the same thought in current terms, and said that a linguistic theory should be able to characterize both the oppositions within a language (the differences between the members of the set of all possible sentences) and the contrasts between languages (all and only the features which mark the sounds of the language as being different from the sounds of other languages). One of the objects of my current work (Ladefoged, 1971) is to assess the extent to which it is possible to construct a phonological theory which will achieve this goal. Others have suggested that phonetics should play a different part within linguistic theory. Thus, for the traditional American linguists, phonetic analysis of a language had to precede its description in linguistic terms ; and it had to be made completely independently of any knowledge of the oppositions which occurred. A satisfactory way of making phonetic descriptions for this purpose may be theoretically conceivable, but it would be very complicated. The best attempts are those of Pike (1943) and Peterson and Shoup (1966). However, Chomsky (1964) has shown that it is not possible to describe a language adequately by starting from descriptions of sounds without reference to their linguistic function. So we do not need a theory of phonetics which will include a procedure for classifying sounds irrespective of their function within a language. Much to the discomfort of some phoneticians (and some linguists), phonetics is not an independent science but an integral part of linguistics. All that is necessary for linguistic purposes is a theory of phonetics which will allow us to account first for the oppositions within a language, and then for the relations between languages. But if it is to be interesting, the description of each language must also be testable; and the possibility of making a sufficient test must be inherent in the underlying theory. The difficulty is that languages are abstractions. What we call a language such as English or French is a code that enables individual Englishmen and Frenchmen to communicate with one another. But we cannot test descriptions of a code without reference to its manifestations; the only data we have for checking our descriptions of a language are the utterances of individual speakers. It seems, then, that there are three stages which a phonological theory must be capable of handling. First, it must permit the oppositions within each language to be specified; this is what Chomsky

276

PETER LADEFOGED

(1964) calls systematic phonemics. Secondly, it must provide a way of accounting for the particular characteristics of each language; this might be systematic phonetics. Thirdly, it must lead to the specification of actual utterances by individual speakers of each language; this is physical phonetics. Linguistic descriptions which do not meet all three of these requirements are apt to be trivial. In practice the first step involves allocating sounds to contrasting categories, the second to designating relative values of each category, and the third to interpreting these values in terms of measurable units. Thus, to continue our example, both English and French may need the categories voiced and voiceless; but English initial voiced stops may be said to have 15 degrees of voicing, whereas French may have 60; and a particular English speaker may make a given word with a voiced initial stop in which the voicing lasts 15 ± 2 msecs., whereas a particular French speaker may, in his circumstances, have an initial stop in which the voicing lasts 60 ± 2 msecs. We may now consider the general form of the kind of phonetic description that is being proposed here. It must, like other parts of the description of a language, be capable of being expressed completely in a set of explicit statements or rules, so that we can be sure that no intuitive (possibly fallacious) concepts are required for its interpretation. Ultimately it would be convenient if the rules produced a set of signals which could control a speech synthesizer. Then we could be certain that the entire account of a language was contained in the rules and the theory (which would have to include a specification of the speech synthesizer). Such a description could, in a very literal sense, be part of a generative grammar; and the grammar would be very powerful in that it would contain rules which were not merely possible (specifying correct but not necessarily all the phonetic correlates) but necessary and sufficient (containing all and only the information required to generate speech). One of the major difficulties in achieving this kind of description is in relating the essentially continuous nature of speech with the essentially discontinuous nature of a linguistic description. All linguistic descriptions involve segmentation of some kind, since they all distinguish between an infinity of possible sentences by specifying different arrangements of a small number of discrete units. Virtually the same problem arises irrespective of whether we are attempting to account for what happens in the juxtaposition of segments of the size of a phoneme or a syllable or any other unit. One solution is that we should specify the ideal form or target for each unit (or, more precisely, for each category within each unit), and then provide a rule or a set of rules which will specify the extent to which this target is missed because of the influence of the adjacent items. Our description of a language would then include a table of values specifying the targets in this language, and rules, which we will call interpretation conventions, accounting for the partial overlap or way of getting from one target to another. The table of values might be expressed in terms of numbers representing relative values of categories, or parameters for synthesizing speech, such as formant frequencies and durations; or conceivably it might be in terms of numerical values of Jakobsonian distinctive features, in which case it would occupy a place in

PHONETIC PREREQUISITES FOR A DISTINCTIVE FEATURE THEORY

277

the description somewhat analogous to Halle's descriptive phonetic matrix. Whatever the form of the categories or distinctive features used in the specification of the targets, the values given will be largely in relative terms, as long as the description is of a language, and not of an individual. In the process of synthesis, the values will be transformed by a constant (perhaps multiplication by unity) and then considered to have measurable physical properties which may be associated with an individual (cf. Kim, 1967). In this way we will have a motivated principle for stopping at a given point in the rules which produce a more and more detailed phonetic description. The systematic phonetic level of description may be said to be that level which specifies all the targets necessary for the description of a particular language as opposed to all other languages, but contains no information of the kind that is used simply to specify one speaker of that language as opposed to other speakers. Halle has not (to my knowledge) considered the form of the conventions required for turning phonetic segments into continuously varying parameters; but a number of other investigators (Lindblom, 1963; Öhman, 1964 and 1967; Holmes, Mattingly, and Shearme, 1965; Lindblom and Studdert-Kennedy, 1967) have suggested parts of possible schemes. The basic principles which seem most appropriate for the theory of phonology being propounded here have been derived mainly from the work of Lindblom and öhman. In particular, öhman (1967) has shown that a theory involving target positions and interpretation conventions (which may be given precise mathematical formulation) can be used to account for cineradiographic data on consonant assimilations in phrases comparable with a key and a car. It seems probable that in some such way, with the aid of a table of values for the ideal form of each segment, we should be able to go from a description of speech in terms of discrete segments to a specification in terms of continuously varying parameters. This approach forces us to consider the circumstances in which we will consider a speech sound, at the systematic phonetic level, to consist of a single segment or target, as opposed to those which require us to specify it as a sequence of two targets. It is at this point that we must remember two of the requirements of a theory of phonetics. It must provide a set of categories for use in descriptions of the phonology of a language; and it must also provide for the interpretation of these categories in non-linguistic terms. There is a conflict between these two requirements when we come to consider items such as affricates. On the one hand these items clearly consist of a sequence of two items, stop and fricative, which may be independently motivated targets. But on the other hand, affricates often have to be considered as single phonological units. In languages such as Hindi and other Indo-Aryan languages, and Makua and other Bantu languages, there are contrasts between aspirated and unaspirated affricates. If we regard affricates as sequences at the phonemic level, then we have to have extra statements (phonological rules) which allow for the possibility of fricatives being aspirated only when they are preceded by a stop within the same syllable. Similarly, there are many languages (e.g. Quechua) where there is a series of ejective stops which includes an affricate; and we do not want to have a special rule saying that

278

PETER LADEFOGED

fricatives can be ejective only when following a stop which is accompanied by an upward movement of the closed glottis. Clearly, we have to provide for the possibility of a fricative release being part of the same segment as the stop. Further examples of this kind of problem are not hard to find. In English, laterally released stops (as at the ends of the words riddle and little) are sequences of stop followed by lateral; but in Navaho and other languages, where tl' forms part of the series of ejective stops, it is more convenient to regard the lateral release as part of the same segment as the stop. In Tiv there is a series of prenasalized stops contrasting with other stops and nasals. This phenomenon, which is common in Bantu languages, leads us to consider stops with and without nasal onsets as single units at the phonemic level, although this is not the simplest possible description at the phonetic level. Similarly insistence on simultaneity in secondary articulations may seem ill-advised on phonological grounds. Many languages (e.g. Twi) have sets of consonants which differ in that one set has a slightly subsequent but considerably overlapping feature of labialization or palatalization. The rules accounting for the phonological patterns within these languages will be considerably complicated by a specification of the underlying form in terms of sequential items, although this may be the best account of the surface forms. It seems that a theory of phonology will have to permit some sounds to be categorized as single segments at the phonemic level, and yet to regard them as sequences of targets at the systematic phonetic level. Any description of linguistic forms must ultimately be interpreted (by a human or a machine) in terms of a sequence of items referable to a table of values corresponding to the targets and durations of each item, and interpretation conventions specifying the extent to which each target is missed because of the influence of the adjacent items; and it will be much easier to operate such a system if we maintain a clear distinction between those items which involve simultaneous targets, and those which involve sequential targets. But it would seem preferable to let the requirement of making an adequate phonological description of a language constrain the theory we must develop so that it will be able to account for the cases we have described above by having more categories or possible combinations of categories than are necessary to satisfy the requirement of providing an adequate interpretation of linguistic items in nonlinguistic terms. The latter requirement may, however, result in the theory having some categories which are necessary at the level of systematic phonetics, but which are not required at the phonemic level. Thus it seems probable that at the phonetic level we may need categories distinguishing between syllabic and nonsyllabic sounds ; but with the possible exception of Japanese (McCawley, personal communication), this distinction is not needed at other phonological levels. Thus in English, the surface contrasts between hungry and Hungary or between coddling and codling, do not occur in the underlying forms. In whatever way it is arranged, the table of values which is involved at the systematic phonetic level will have to contain sufficient information to mark all the linguistic features of the language being described. The most well-known of these involve

PHONETIC PREREQUISITES FOR A DISTINCTIVE FEATURE THEORY

279

phonemic oppositions. There is no need to discuss here the concept of a phoneme (or morphoneme — I follow Chomsky, 1964, in finding no necessity for two separate levels, and normally say phoneme and phonemic where older American linguists might have said morpho(pho)-neme and morpho(pho)nemic). Similarly the contrastive nature of suprasegmental units such as pitch and stress is sufficiently well known to require no further elaboration. But there are other contrasts which cannot be specified simply by taking into account the -segmental phonemes, and suprasegmentale such as pitch and stress. Consider the example (suggested by Bloch in a personal communication) I'm going to get my lamb prepared as opposed to I'm going to get my lamp repaired, in which the sequences of segmental phonemes and stresses are identical. At some point in the description of English these two will be distinguished by a juncture mark, or word boundary, or similar device. But later this distinction will have to be expressed in terms of categories characterizing the allophones of the particular segmental phonemes, since only these units (and the supra-segmentals which are irrelevant here) can be interpreted in physical terms. Allophones such as final ρ and initial phwill have to be composed of distinctive features or categories with separate listings in the table of values for English. Their parametric specifications are different; and, although the differences are linguistically predictable in terms of juncture, they cannot be regarded as the product of interpretation conventions or conjoining rules. Junctures and all other boundary marks cannot be listed in a table of values as having certain relative identifiable properties such as formant frequencies and durations; they can be taken into account only by noting their effect on other units. Allophones which are generated in this way (or through the effect of any other higher level units such as stress or vowel harmony marks) may be called extrinsic allophones, in contrast with those which are due to the partial overlapping of the articulations of adjacent phonemes, which will be called intrinsic allophones. These two kinds of allophones have to be distinguished because features characterizing extrinsic allophones have to be given individual listings in a table of values, but those for intrinsic allophones do not. Some further examples may help in clarifying the differences between these two types of allophones. We have already mentioned the well-known fact that in the English words key and car the initial consonants differ in that the stop in the first word has a more forward articulation than that in the second. These two stops are intrinsic allophones, since it is possible to postulate a single target position for English initial k and a rule which enables one to calculate the actual articulatory position by taking into account the overlap with the articulations required for the neighboring vowels. Similarly the difference between the voiced r in dry and the largely voiceless r in try can be predicted by a rule specifying the nature of the overlap between the states of the glottis in the adjacent sounds. None of these allophones has to be specified separately in a table of values. But the difference between the t in top and the t in mountain (which in many forms of American English may be accompanied by a glottal stop)

280

PETER LADEFOGED

cannot be ascribed simply to the overlapping articulations of neighboring sounds; nor can the difference between the r in American English reed and that in deer·, nor the differences such as those in timing between the two η sounds in nun or the two k sounds in kick. These extrinsic allophones must be characterized by features with different listings in the table of values. Both the targets and the interpretation conventions may be language dependent. There are many linguistic universals; but, for example, the effect of neighboring vowels on the articulation of velar stops is not one of them. This may be seen by comparing English and French. In both languages the initial stops vary in much the same way in pairs such as English key-car and French qui-car; but there is a much greater difference between the final stops in French pique-Pâques than there is between those in English peak-pock. Some of this difference may be due to the differences between the vowels in the two languages. But an explanation of this kind is not sufficient. It seems that in English coarticulation consists mainly of anticipation of the following item, but in French preceding vowels have as much effect on a consonant as following ones. In other words, the conjoining rules for English and French have to be different. Further evidence which may be interpreted in this way is available in the literature. Thus Jones (1956) noted with regard to palatograms of English two and tea in comparison with palatograms of French tout and type ... these palatograms corroborate a curious point previously ascertained by direct observation that while the English t is articulated further back when followed by sounds of the o type than when followed by sounds of the i type yet in French the opposite is the case. Sounds which can be correlated with phonemic oppositions or extrinsic allophones within a single language clearly need to be specified in different ways. But before we can formalize a set of distinctive features we must state a principle for classifying similar sounds which never contrast within a single language, such as Hindi dfi and English ö. We can obviously reduce the number of categories required for describing the languages of the world by classifying similar phenomena in different languages as variants belonging to the same category. But unless we are careful this is going to lead us into the game-playing arbitrariness of some adherents of Jakobsonian distinctive feature theory. There are two principles which will help us here. In the first place we should try to account for linguistic oppositions in terms of distinctive features which fit into a scheme of speech synthesis and will account for all the linguistic aspects of speech. We should not be satisfied with distinctive features that are merely possible in that they characterize some and not all of the distinctive aspects of the linguistic oppositions (which is the best that could be claimed for the Jakobsonian distinctive features). In the second place we should describe sounds in terms of the same distinctive features only if they differ in degree and not in kind. This second principle can be put more formally (and in a more widely applicable form) by saying that TWO DIFFERENT PHENOMENA SHOULD BE DESCRIBED IN TERMS OF THE SAME DISTINCTIVE FEATURE IF AND ONLY IF THE FEATURE HAS MEASURABLE PROPERTIES AND

PHONETIC PREREQUISITES FOR A DISTINCTIVE FEATURE THEORY

281

THE PHENOMENA CAN BE DIFFERENTIATED SIMPLY IN TERMS OF NUMBERS SPECIFYING THE DEGREES IN WHICH THEY POSSESS THESE PROPERTIES.

We will find it useful that this second principle applies just as much to the description of two different sounds within a language as to sounds in different languages. It makes it immediately obvious that, despite the phonetics of Trager and Smith (1951), at the systematic phonetic stage of the description the first sound in English hot and the last sound in English law can no more be in the same categories than the first and last sounds in English hang. As there is no gain in grouping any of these items together at an earlier stage in the rules (except a pseudo-economy which later has to be resolved), we never have any motivation for this part of the Bloch-TragerSmith analysis of English. Our knowledge of phonetics and speech synthesis by rule is still too limited for us to be sure of the details, but I would guess that a principle of the kind emphasized above should enable us to evaluate many other disputed points in the phonology of English and other languages. So far, no explicit statement has been made about the relations between segmental targets, and the distinctive features which characterize the segments. Earlier investigators (Holmes, Mattingly and Shearme, 1965) envisaged a system in which segments such as phonemes or allophones were assigned values by finding the entry corresponding to each segment in a look-up table. Recently, Kim (1967) has proposed an interesting way of making generative rules for going from the systematic phonetic level to the target specifications of physical phonetics. He was concerned with interpreting systematic phonetic features in terms of acoustic parameters; but the underlying ideas are equally applicable to physiological specifications of speech. His principle (slightly reinterpreted) is to begin by a rule assigning values which would specify a neutral position of the speech mechanism. He then interprets each group of features, such as those required for vowel height, by means of another rule giving a value for a degree of movement and a set of rules such as: (1) if high up two degrees; (2) if low down two degrees; (3) if tense to a certain degree (of tenseness) then up to that same degree (but of movement). Kim's rules are expressed in a more elegant way, using a notation with variables which is partially of his own devising. But even the formulation given above enables us to appreciate a number of the advantages of his system. Firstly it clearly distinguishes between the properties of the language and those of the individual speaker. The rules for assigning values to specify the neutral position, and the rules that give the magnitude of a degree, correspond to properties of individual speakers. Changing them will be equivalent to changing the personal quality of the speaker being synthesized. But the rules concerning the interpretation of features, such as those exemplified above, are properties of the language. A second advantage is that it has often been shown (cf. Halle, 1964; Chomsky and Halle, 1965) that phonological rules are expressed most appropriately in terms of features rather than segments. It would be disingenuous, however, to gloss over the difficulties in the phonological theory proposed here. It must be admitted that at the level of interpretation conventions the notion of distinctive features seems to be an added complication rather than

282

PETER LADEFOGED

an explanatory device. The fact is that the difference between, say, English ρ and b in a given environment (say, utterance final) is not simply that one is voiced and the other voiceless. They differ in length, the shape of the vocal tract at the moment of lip closure, and probably in other ways. Similarly the difference between initial s and t is not just a manner of articulation. These sounds also differ in length; the action of the glottis has to be specified in a different way for each of them; and they have very different target positions and rates of transition. The velar stops k and g differ not only in the voicing feature but also in the rate of transition (rate of movement of the tongue from the stop position) which is faster for g then for k. A host of facts such as these forces us into making extremely context-restricted statements for each distinctive feature. 'Voiceless' has to be interpreted as having certain properties when it co-occurs with 'stop' as opposed to when it co-occurs with 'fricative'; 'alveolar' (or whatever other place feature is used) also has one value when it is associated with 'stop', and another with 'fricative'. It appears that the simplest form of conversion to physical phonetics may not rely on distinctive features. The interpretation conventions would have to include a large number of context-restricted rules which would not be needed if the conversion were directly from segments (or larger units) to physical phonetics. It is perfectly possible to organize the controls for a speech synthesizer in terms of distinctive features; an appropriate program is now being developed at UCLA using a small digital computer. A system of language description of this kind has a great deal of intuitive appeal, and is very elegant in that it uses the smallest possible number of primitive categories, and the greatest number of formal rules. But human beings are not necessarily elegant, and do not work like computers. The indications from neurophysiology and psychology are that, instead of storing a small number of primitives and organizing them in terms of a large number of rules, we store a large number of complex items which we manipulate with comparatively simple operations. The central nervous system is like a special kind of computer which has rapid access to the items in a very large memory, but comparatively little ability to process these items when they have been taken out of memory. There is a great deal of evidence that muscular movements are organized in terms of complex, unalterable chunks of at least a quarter of a second in duration (and often much longer) and nothing to indicate organization in terms of short simultaneous segments which require processing with context-restricted rules. We are then left with the problem of the reality of units such as distinctive features. It has been pointed out that they have explanatory power in that they permit the construction of natural classes which are required in phonological rules. But we may still ask whether they are simply necessary items in a linguistic analysis, subject only to the constraints imposed by the theory (including the constraints concerning their relation with physical phenomena which we have been discussing) but with nothing further that can be said about them; or whether they have in addition simple behavioral correlates. A possible answer is that they may play a more important part in the perception of speech. Speech production and speech perception are sometimes pre-

PHONETIC PREREQUISITES FOR A DISTINCTIVE FEATURE THEORY

283

sumed to be reverse processes; but this is a naive, untenable, view. The response to any stimulus depends in part on the set of the observer; and in the case of the perception of speech his set may be largely equated with his linguistic competence, which may include statements structured in terms of distinctive features. Language is a very complex process, with many facets, some of which require superficially incompatible analyses. Eventually our linguistic theories must account for all possible data. But in the present state of our knowledge we should not expect to find a single simple form of description which has equal explanatory power in all areas. It follows from this that the relation between a speaker's competence and his performance may be very complex. It is also possible that a speaker's competence should be related to his performance both as a listener and as a speaker. Studies of the perception of speech are of little assistance in this respect, since almost the only relevant data available to the experimenter are the responses of subjects. But these responses are discrete items which reflect an encoding of the sensations corresponding to the incoming stimuli. Subjects can report only the results of their perception; they cannot make explicit anything about the process of encoding sensations. It is possible that this process involves something similar to the activity reflected in the rules for going from systematic phonetics to physical phonetics but in reverse. At the moment this hypothesis is uninteresting because it is untestable. Meanwhile these interpretation rules are in any case an essential part of the description of a language, since without them the description has no substance and is itself untestable. It is tempting to imagine that there might be some universal perceptual categories so that we could test a grammar without going to the length of requiring the interpretation of our description in terms of distinctive features to include all the information necessary for synthesizing natural speech. It would be so much simpler if the interpretation had to indicate only some of the perceptual data which a hearer might be assumed to use in decoding utterances; and it would seem that a legitimate aim for a linguistic description might well be to characterize simply the important perceptual features in terms of a set of universal perceptual categories. There are two objections to this approach. In the first place unless we generate all the information for synthesizing speech, we cannot be sure that we have in fact synthesized the most important perceptual features. Speech is enormously redundant, and subjects in experiments can base their responses on cues which they might never use in other circumstances. Secondly, it is difficult to know what could be meant by universal perceptual categories, if perception is taken to include the encoding of sensations. Perception in any other sense is unknowable; and in this sense is dependent on the listener's previous linguistic experience. Subjects in experiments can usually categorize speech sounds only in terms of sounds which they can make themselves. Trained phoneticians may do better, but if we use them as subjects, so that we partially remove the influence of the subject's linguistic background, then we might just as well remove all such influences and devise distinctive features based on the physical and not the perceptual

284

PETER LADEFOGED

properties of speech sounds. If a particular perceptual category has any universal validity, it is almost certainly because of some property of the physical stimulus. Distinctive features which simply reflect the linguistic competence of listeners cannot be used in a theory of phonetics without destroying the whole concept of phonetic specifiability.1 UNIVERSITY OF CALIFORNIA LOS ANGELES

REFERENCES Bell, A. M„ 1867 Visible Speech (London). Chomsky, A. N., 1964 "Current issues in linguistic theory", in The Structure of Language, J. Fodor and J. Katz (eds.) (Englewood Cliffs: Prentice-Hall), 50-118. Chomsky, A. N. and Halle, M., 1965 "Some controversial questions in phonological theory", Journal of Linguistics 1.2, 97-138. 1968 The Sound Pattern of English (New York : Harper and Row). Chomsky, A. N. and Miller, G., 1963 "Introduction to the formal analysis of natural languages", in Handbook of Mathematical Psychology, R. D. Luce, R. R. Bush, and E. Galanter (eds.) (New York: John Wiley and Sons, Inc.), 3. 269-323. Firth, J. R., 1957 Papers in Linguistics, 1934-1951, (London: OUP). Fischer-jOTgensen, Eli, 1952 "The phonetic basis for identification of phonemic elements", Journal of the Acoustical Society of America 24, 611-17. Halle, M., 1964 "Phonology and generative grammar", in The Structure of Language, J. Katz and J. Fodor (eds.) (Englewood Cliffs: Prentice-Hall), 334-52. Holmes, J. N., Mattingly, I. G. and Shearme, J. N., 1964 "Speech synthesis by rule", Language and Speech 7.3, 127. Hockett, C. F., 1955 A Manual ofPhonology = International Journal of American Linguistics 1.24, pt 1 (Bloomington: Indiana University). Jakobson, R., 1962 Selected Writings I: Phonological Studies (The Hague: Mouton). Jakobson, R., Fant, G. M. and Halle, M., 1951 Preliminaries to Speech Analysis (= Tech. Report - No. 13, Acoustics Laboratories, M.I.T.) (Cambridge, Mass. : M.I.T. Press). Jakobson, R. and Halle, M., 1956 Fundamentals of Language (The Hague: Mouton). Jones, D., 1956 An Outline of English Phonetics (8th edition) (Cambridge, England: Heffer). Joos, M. (ed.), 1957 Readings in Linguistics (Washington: American Council of Learned Societies). 1

Since 1967, when this article was written, The Sound Pattern of English (Chomsky and Halle, 1968) has appeared. The references in this article have been updated, but some of the arguments should also have been revised. Some of the requisite additional constraints on a feature theory are discussed in Ladefoged (1971).

PHONETIC PREREQUISITES FOR A DISTINCTIVE FEATURE THEORY

285

Kim, C.-W., 1967 The Linguistic Specification of Speech (= Working Papers in Phonetics, No. 5) (UCLA). Ladefoged, P., 1971 Preliminaries to Linguistic Phonetics (Chicago: University of Chicago Press). Langendoen, D. T., 1964 Review of "Studies in linguistic analysis", Language 40.2, 305-21. 1968 The London School of Linguistics: A Study of The Linguistic Theories of B. Malinowski and J. R. Firth (Cambridge, Mass.: M.I.T. Press). Lindblom, B., 1963 "Spectrograph«: study of vowel reduction", Journal of the Acoustical Society of America 35, 1773-81. Lindblom, B. and Studdert-Kennedy, M., 1967 "On the role of formant transitions in vowel recognition", Journal of the Acoustical Society of America 42.4, 830-43. öhman, S. E. G., 1964 "Numerical model for co-articulation, using a computer simulated vocal tract", Journal of the Acoustical Society of America 36, 1038. 1967 "Numerical model of co-articulation", Journal of the Acoustical Society of America 41.2, 310-20. Peterson, G. and Shoup, J., 1966 "A physiological theory of phonetics", Journal of Speech and Hearing Research 9.1, 6-67. Pike, K. L., 1943 Phonetics (Ann Arbor: University of Michigan Press). Trager, G. L. and Smith, H. L., 1951 An Outline of English Structure, (= Studies in Linguistics, Occasional Papers, 3). Reprinted 1957 (Washington, D.C.).

JEAN-CLAUDE LAFON

PERCEPTION PHONÉTIQUE AU SEUIL D'AUDITION

La mesure de l'intelligibilité de la parole, c'est-à-dire celle de la qualité d'identification des unités acoustiques verbales, est passée ces cinquante dernières années par plusieurs stades. Une première époque va des travaux en France de l'Abbé Rousselot (1924) à la publication en 1929 aux U.S.A. de l'ouvrage de Fletcher (1929). Ce dernier a étudié la formulation de la qualité d'un message verbal à travers un système transmetteur. De 1930 à 1948, les problèmes ont été étudiés sur un plan différent: définition psychophysiologique de l'audition, méthodes de mesures et études d'appareils de codage du type du vocoder (Campanella, 1958). Actuellement ces études ont été renouvellées par la construction du sonagraph (Potter, Kopp et Green, 1947) et grâce aux machines électroniques d'analyse et synthèse de la parole. De nombreux auteurs se sont intéressés à la réception de la parole et à son identification, en particulier dans les pays anglo-saxons. L'étude présentée se situe dans l'orbite des ces travaux, elle concerne la perception de la parole à ses limites, c'est -à-dire entre le niveau où le sujet est incapable d'identifier une structure phonétique et celui où la reconnaissance est pratiquement toujours correcte. Les sujets soumis à l'examen sont normaux tant sur le plan de l'audition que sur celui des facultés perceptives. Nous ne considérerons que la parole normale. D'autres études ont porté sur la parole codée (Campanella, 1958; Lafon, 1961; Pimonow, 1962) ou sur les altérations pathologiques (BorelMaisonny, 1961 ; Mounier-Kuhn et Lafon, 1961). Les structures acoustiques de la parole sont très hétérogènes : impulsions laryngées marquées par la résonance des cavités bucco-naso-pharyngées pour les voyelles et les consonnes semi-voyelles, impulsions laryngées et bruit d'explosion ou d'écoulement pour les consonnes sonores, bruit marqué d'une résonance spécifique pour les consonnes sourdes. Chaque élément positif du spectre représente un formant s'il peut être utilisé pour la discrimination du son de parole. Mais l'étude de la perception ne peut se réduire à celle des spectres acoustiques, les variations d'intensité, les mélodies fournies par le son laryngé ont également un rôle dans l'identification. De plus, l'oreille reçoit de façon continue ces signaux évoluant dans le temps; ce sont à des ensembles que l'auditeur attribue une valeur de signe, de message constitué de symboles, unités discrètes, les phonèmes. L'empreinte de la reconnaissance est mar-

288

JEAN-CLAUDE LAFON

quée par le souvenir de l'émission verbale; on identifie par le souvenir du mouvement à effectuer pour une émission acoustique correspondante: l'identification est liée à l'articulation, on reconnaît dans le son le mouvement des organes phonateurs qui est imagé pour préciser le contour des formes acoustiques perçues. L'importance de ces phénomènes psychophysiologiques est telle que même dans la parole synthétique, c'est un auditeur qui juge de la qualité ou ce qui revient au même les réponses statistiques d'une population déterminée. Etant donné la complexité de ces formes verbales, on ne peut en étudier la perception à l'aide des données acoustiques élémentaires, ou des seuils différentiels de ces unités (intensité, fréquence, temps), en leur comparant la structure acoustique de la parole étudiée : on ne peut se passer du niveau verbal, c'est-à-dire du langage du sujet examiné. Nous avons donc dans ces études deux données: celle qui correspond à des critères acoustiques et celle des caractères psychophysiologiques de la perception. Ces deux données ont une importance variable suivant la taille des unités verbales: phonème, mot monosyllabe, mot plurisyllabique ou phrase. Les liaisons répondent à des règles qui en devenant plus nombreuses avec la taille de l'unité en augmente d'autant la probabilité d'une identification exacte. Howes (1957) a montré que plus le mot est long, plus le bruit toléré est grand; la valeur critique du rapport signal-bruit varie de 15 pour les mots de 3 lettres à - 10 pour les mots de 21 lettres à fréquence d'occurrence très faible donc à probabilité d'erreur réduite. La forme de l'erreur se rapproche d'autant plus d'une confusion acoustique que le mot est court et l'influence d'une structure linguistique est progressivement croissante du phonème au morphème, du morphème au mot, du mot à la phrase (Fry, 1964). La position du phonème dans le mot n'est pas indifférente mais les avis sur la place du phonème sont divergents: position finale plus difficile (Oyer et Dowdna, 1959), position initiale (Black, 1939), position indifférente (Lafon, 1966). Pour les études d'intelligibilité on utilise des logatomes syllabiques (Fletcher, 1929; Kryter et Whitman, 1965), des mots phonétiquement équilibrés (Black et Haagen, 1963) en notant un tout ou rien pour l'étude quantitative. Borel-Maisonny (1961), Moser, Fotheringham et Hender (1961), emploient des phonèmes isolés. L'étude phonétique a été faite par différents auteurs (Miller et Nicely (1955), Lafon (1956), Oyer et Dowdna (1959), Tobias (1965)) qui en donnent des tableaux de confusion. Il devrait être plus facile d'étudier la perception de la parole à partir de mots ou de syllabes sans signification dont les phonèmes ont été assemblés au hasard. Le lien de probabilité d'un phonème à l'autre peut théoriquement être considéré comme nul. En fait, seuls des sujets entraînés à une telle perception donnent des résultats cohérents. Nous avons préféré nous adresser, pour les études de la perception phonétique au seuil de l'audition, à des structures de la langue considérée, donc à des mots existants réellement. 1. LE PHONÈME CARACTÉRISTIQUE

Dans un mot tous les phonèmes n'ont pas la même valeur informative, certains sont

PERCEPTION PHONÉTIQUE AU SEUIL D'AUDITION

289

utiles, d'autres peuvent éventuellement être mal perçus sans que la reconnaissance du mot qui les contient en soit perturbée. Lorsque le phonème est indispensable à la reconnaissance du mot on dit qu'il est un PHONÈME CARACTÉRISTIQUE du mot. Ce qui veut simplement dire que la probabilité de retrouver le phonème mal évoqué ou déformé est suffisamment faible pour qu'une erreur ait quelque chance d'apparaître. Dans un tel mot, il est ainsi possible de connaître les modalités de perception du phonème caractéristique, donc à travers l'étude de la reconnaissance d'un mot de noter celle du son de parole évoquant le phonème considéré. Sur ce principe nous avons bâti un test publié à partir de 1956 et appelé "test phonétique" dont la traduction anglaise a paru récemment (Lafon, 1966).

2. LE CALCUL D E L'INTELLIGIBILITÉ

Nous avons l'habitude sous l'influence des ingénieurs des téléphones de parler d'intelligibilité, c'est-à-dire de reconnaissance de mots. En fait, il est beaucoup plus intéressant de compter en erreurs, c'est-à-dire en unités mal reproduites tout particulièrement lorsqu'on s'adresse aux sons de parole. Les calculs statistiques, ceux d'une probabilité en sont beaucoup facilités. Et ceux qui nous intéressent le plus sont certainement les mots ou les phonèmes sont mal identifiés ; ils nous apprennent plus que ceux qui le sont correctement. Tous les graphiques et les calculs de cette étude portent donc soit sur un pourcentage soit sur une probabilité d'erreur. Sous l'influence des psycho-acousticiens nous avons l'habitude de calculer un seuil comme étant la valeur à 50 % des sujets de la population étudiée. Si pour la perception tonale, le seuil de perception des sons purs, cette valeur est assez proche des limites physiologiques, il n'en est pas de même de la parole. Nous savons (Fletcher, 1929) que la marge entre la non-perception d'une forme verbale et la complète identification peut atteindre et dépasser 40 dB. Prendre une valeur moyenne rend mal compte de ce résultat. Il nous paraît donc préférable d'utiliser une unité plus physique, le décibel que nous employons correspond à une référence de 2 X 10" 5 Pascal, celle-ci étant mesurée pour la fréquence 1.000 Hz, la courbe de réponse des amplificateurs étant linéaire pour l'ensemble des fréquences utilisées dans les sons de parole (de 30 à 10.000 Hz).

3. LE TEST PHONÉTIQUE

Le test mesurant la reconnaissance des sons de parole que nous avons construit, utilisant la notion de phonème caractéristique, est compté en erreurs à partir d'un niveau en décibels référence physique. Les listes-mots du test contiennent uniquement des mots de la langue considérée. Les mots ont la même longueur phonétique, et nous avons choisis, pour des raisons

290

JEAN-CLAUDE LAFON

pratiques, des mots de trois phonèmes. Trois phonèmes ne demandent aucune effort de mémoire au sujet testé, ils peuvent être répétés facilement dans leur succession, même si le mot n'est pas compris; l'ensemble de la langue contient suffisamment de mots de trois phonèmes ayant des unités caractéristiques pour construire un nombre de listes suffisant, ce qui n'est pas le cas de mots de deux, de quatre ou de cinq phonèmes. Pratiquement nous avons relevé les mots qui ne diffèrent entre eux que par un seul phonème, par exemple: /mar/ - /mor/ - /mar/ - ou bien /mar/ - /nar/ - /lar/ ou bien /mal/ - /man/ - /mas/, et avons noté pour chaque mot quels sont les phonèmes qui peuvent être confondus avec d'autres phonèmes; autrement dit, nous avons noté la probabilité d'erreur de chaque phonème dans l'identification du mot par le biais de la nécessité d'un choix entre différents mots si ce phonème est mal perçu. Nous avons ainsi pu disposer pour la langue française de plus de 450 mots dont nous avons étudié les distorsions chez 150 sujets sourds, ce qui nous a permis d'attribuer en plus de la probabilité d'erreur de chaque mot un coefficient de distorsions correspondant à la difficulté d'identification dans la surdité. Nous avons ainsi des listes de 17 mots correspondant à 51 phonèmes (assimilés à 50) ayant comme caractéristique de présenter tous les types de phonèmes de la langue, d'avoir un éventail de consonnes occlusives, constrictives, de voyelles nasales, de consonnes sourdes, etc.... identiques d'une liste à l'autre de telle sorte que l'échantillonnage acoustique ainsi que l'échantillonnage phonétique soient équilibrés. Enfin, nous nous sommes efforcés, de plus, d'avoir un équilibre de distorsions d'une liste à l'autre. Ainsi nous avons pu construire 20 listes de mots, la qualité étant testée par des analyses de variance. Ces mots sont présentés au sujet, soit enregistrés, soit prononcés au micro, contrôlés au vu-mètre, et l'on note chaque phonème mal répétés que l'on coche en précisant le phonème de remplacement utilisé. Il est ainsi possible d'avoir une étude quantitative et qualitative de l'audition. Ces listes on été appelées "listes cochléaires" parce que leur étude à différents niveaux d'intensité indiquent, lorsque la discrimination auditive est normale, les erreurs provoquées par les distorsions acoustiques de la cochlée. Elles renseignent donc sur la valeur de l'oreille dans la perception de la parole. Il existe dans le test phonétique une deuxième méthode d'exploration de l'audition, appelée TEST PHONÉTIQUE D'INTÉGRATION. Pour ce test nous avons choisi des mots où tous les phonèmes sont très caractéristiques, c'est-à-dire où les possibilités de choix portent sur un tellement grand nombre de phonèmes de la langue que la probabilité d'une réponse exacte est très faible. De plus les listes groupent tous les mots différents qui auraient pu être évoqués à l'occasion d'une erreur sur le phonème testé. Tous les mots font partie de la même liste, c'est ainsi que dans une liste d'intégration où il y a le mot /mar/ apparaissent également les mots /lar/ - /nar/, etc.... les mots /mar/ - /mer/ etc.... Donc la probabilité de voir apparaître une erreur est grande si l'identification est mauvaise. Comme tous les mots probables existent dans la liste, il ne peut y avoir de mémorisation des mots-tests, pourvu que l'ordre de présentation ne soit pas le même. Il est alors possible

PERCEPTION PHONÉTIQUE AU SEUIL θ'AUDITION

291

de répéter indéfiniment ces listes, sans que le souvenir des mots entendus permette de compenser une distorsion ou une déformation de la perception phonétique. Présenté dans deux conditions acoustiques différentes, nous aurons les mêmes mots, les mêmes structures phonétiques, les mêmes structures acoustiques, la même composition linguistique avec simplement des niveaux de difficultés de perception : on peut émettre la liste à l'intensité optimale d'identification, puis avec des facteurs venant modifier les possibilités d'identification comme la variation d'intensité faible et forte ou bien encore l'adjonction de bruits. A l'intensité optimale, les erreurs sont dues à la connaissance de la langue par le sujet, à des difficultés expressives, difficultés articulatoires, éventuellement difficultés acoustiques d'ordre cochléaire. Les erreurs commises en plus à l'adjonction de bruit ne pourront être que celle qui dépendent de la discrimination de la parole dans son fond sonore. La différence d'erreurs dans les deux mesures permet de connaître de façon précise les qualités de discrimination du sujet testé. Et si l'on définit statistiquement chez le sujet normal l'intervalle de confiance de cette identification, on peut connaître les niveaux qui sont pathologiques.

4. EXPÉRIMENTATION

Nous avons fait deux séries d'expérimentations. La première concerne la perception de la parole au seuil d'audition, c'est-à-dire entre le moment où le sujet commence à reconnaître des éléments phonétiques jusqu'au moment où le sujet perçoit convenablement la parole. La deuxième étude a été faite en utilisant un rapport signal-bruit variable, avec une intensité de parole toujours identique. Ces deux études sont faites en bi-auriculaire. 4.1. Etude de la perception au seuil d'audition Pour cette mesure nous avons utilisé les listes du test phonétique cochléaire dans un ensemble audiométrique comportant un amplificateur gradué en décibels à partir du niveau de référence physique, une émission par haut-parleur dans une cabine sourde communiquant avec la cabine du testeur par un interphone. L'interphone sert uniquement au testeur à connaître les réponses du sujet. Toutes les erreurs sont notées phonème par phonème de façon à avoir pour chaque liste un rapport de 50 phonèmes testés. Nous avons remarqué en explorant le champ des intensités qu'en dessous de 20 dB aucune mesure n'est valable étant donné la variabilité des réponses d'un sujet à l'autre, et pour un même sujet, d'une liste à l'autre ou d'un moment à l'autre. La dispersion des erreurs est beaucoup trop grande pour pouvoir définir une norme de perception et il suffit de variations extrêmement faibles de l'attention pour modifier considérablement l'identification phonétique. Le simple fait de déglutir, de respirer ou de bouger entraîne une impossibilité d'identifier et même de percevoir les sons de la parole.

292

JEAN-CLAUDE LAFON

D'autre part, nous nous sommes aperçus qu'au-delà de 45 dB, la perception phonétique est relativement normale et les erreurs sont sporadiques. L'étude a donc été entreprise à 20,25,30,35 dB où les résultats sont suffisamment cohérents pour pouvoir être dépouillés statistiquement. A chaque intensité nous avons choisi d'émettre cinq listes, de façon à pouvoir faire une analyse de variance pour l'ensemble des sujets. Chez 15 sujets entendant normalement, nous avons étudié les erreurs commises dans l'identification des phonèmes près du seuil. Nous retiendrons de ces mesures les perturbations phonétiques notées en bi-auriculaire, la moyenne générale étant calculée à partir des chiffres moyens donnés pour chaque sujet à chaque intensité. Ces résultats peuvent être portés sur un graphique. Lorsqu'on considère le graphique obtenu, on s'aperçoit que l'échelonnement de 5 en 5 dB, donne à la courbe la forme d'une progression géométrique. Π est alors possible de calculer les valeurs constantes de cette progression et de la définir mathématiquement. Le dépouillement des résultats expérimentaux nous a montré que pour des intensités dm, progressant de façon arithmétique à raison de Q dB la probabilité d'erreur suit une progression géométrique de raison r, nous avons donc les relations suivantes:

et d'autre part: 4n(l + 1) = dt + iQ on remarque que a = η lorsque = ¿i si on tire (i) de la deuxième relation ont trouve: . _ O

es en

60,7

ci

VO

23,3

t·^

00 *— 'shy, evasive' > 'surly, churlish, wicked, vicious'); the Lusism conceivably struck root in Castilian through the late medieval love lyric. 1 * Aren-isco holds a strategic position in the lexical edifice because it seems to mediate between the adjectival -isco series here under consideration and the 'meteorological' group of nouns in -isco, -isca (nev-isca 'light snowfall, flurry, sleet', vent-isco, -isca 'blizzard, drift of snow'), which in turn tie in with a larger contingent of formations in -asco, -asea and with a few in -usco, -usca. Neither the problems raised by these scattered nouns, nor those posed by the 'diminutive' adjectives in -isco, -izco, -usco, -uzeo against the broader background of -azo, -izo, -uzo, nor indeed the separate rôle levant-isco. In the first half of the 16th century French absorbed a limited number of Hispanisms, none among them ending either in -isque or in -esque (see Gerard J. Brault's material) ; Rabelais referred to berne(s) à la moresque 'sorte de manteau', clearly in Italianate disguise (see RPh. 12.233). 16 Aebischer has extracted (1951: 14-15), from Vol. 1 of J. Rius' cartulary of the monastery of Sant Cugat del Vallès, noteworthy examples of proto-Catalan "in ipsa calciata Francischa" (A.D. 978 and 979), a record which parallels R. Lapesa's mention of proto-Leonese franc-isca, used in reference to the manufacture of swords (see fn. 13, above), and which harmonizes with Olt.franc-esco 'French', again as a qualifier of a route. The subsequent choice of the Gallicism franc-és for 'French' has afforded speakers of Spanish and Portuguese the chance of a sharper differentiation of ethnonym and saint's name—a polarization quite differently achieved in modern French ( franç-ais ~ Franç-ois). 16 The classic derivation, advocated by — among others — F. A. Coelho, R. Menéndez Pidal, A. d. R. Gonçâlvez Viana, and A. Castro (add to this chorus the voice of R. de Sá Nogueira), has of late been subjected to searching reappraisal. J. Corominas, as early as 1949-50 (and for subsequent reactions cf. the entries in DCE and in BDE) identified the hypothesis' inherent chronological and semantic flaws and critically assessed some equally weak alternatives (from ARÄRE 'to plough' : 'husbandman' > 'uncouth chap'? from Goth. *FRISKS! from *ARIS-T-ULA, *-CLA 'small fishbone'?). On the positive side of the ledger we may place Corominas' analysis of Sant. jar-isco as a blend of ar-isco with josco, the local representative of Sp. hosco, fusco 'dark, sullen, gloomy' — an interpretation obviating the need for V. Garcia de Diego's conjectural derivation of the regionalism from FER US 'wild'. Participants in the latest discussion include J. da Silveira (1949-50)* ARID-ISCU, quite unconvincing; but the author, gratifyingly, adduces Trasm. brav-isco, Gil Vicente's presumably jocular nonce grav-isco, the 16th-century nickname Ruv-isco 'strawberry-blond', and an otherwise unidentified 15th-century item roman-isco. The pervasive emphasis on 'wildness, aloofness', likewise detectable in Sp. dial, mont-isco (see RPh. 4.33 and fn. 104), may be traced to mor-isco, evocative of a forcibly converted Moor shunning any contact with the close-knit group of "rancid" (i.e., inveterate) Christians. For further repercussions, see W. Giese (1952), and B. Pottier in Rom. 73.278. To the old record of arisco add P. Ferruz, Cane, de Baena, No. 305. Aren-isco 'sandy' is in full bloom ; thus, Azorin in his short story "Los niños en la playa" speaks of la piedra arenisca.

THE PAN-EUROPEAN SUFFIX "-ESCO, -ESQUE" IN STRATIGRAPHIC PROJECTION

371

played b y n o n - d e c o m p o s a b l e Hellenisms a n d Latinisms in -isco ( m a n y o f them o f n o n - I E stock) can here be scrutinized — however challenging these side-issues. 1 7 O n e reason, incidentally, f o r the blocking o f adj. -esco in H i s p a n o - R o m a n c e (and f o r the confinement of -isco t o a very narrow groove) was the fairly strong developm e n t o f -engo, which served astonishingly similar purposes : Contrast Sp.

abol-engo

'patrimony', abad-engo

'Jewish'

( = O S p . jud-iego)

'abbacy, abbatial' with OVrov.frair-esc;

with O P r o v . juvez-esc,

Vtg. jud-engo

etc.

T h e record o f Portuguese discloses little, except for lexical details, that is n o t ascertainable f r o m Spanish; but Catalan, lying astride the frontier between Galloand H i s p a n o - R o m a n c e , again demonstrates its transitional character. O n the whole, it gravitates rather toward Provence than toward A r a g o n and Castile: Hisp.-Lat. argento

span-escho

capell (colteli) recall grez-esch grez-esca, spanesco

( A . D . 986), asino espan-esco

cathalan-esch

( A . D . 1027), R a m o n Muntaner's

beside the g l o t t o n y m s (en) gres-esch,

sarrahin-esch

in Boeci ( w . 205, 207), as well as the g r o u p espan-esca,

sarrazin-esca

in Sainte

jointly

franc-esca,

Foy ( A . D . 1060); cf. also the phrase cum

in a Provençal testament dated 1005. (The local reflexes of Hispaniscus

freno are

the m o r e remarkable as that Late Latin adjective is s u p p o s e d t o have crystallized, o f all places, in the area o f Barcelona.) But the inventory o f the personal belongings o f K i n g A l f o n s o V (ed. L. G o n z á l e z Hurtebise) contains turch-esch(a) 17

beside

mor-

To start with the learned layer, note that the underlying Latin words, insofar as they are not straight Hellenisms (like ASTER-ISCUS, BASIL-ISCUS — already favored by Pliny —, OBELISCOS, TROCH-ISCUS; note BAL[N]ISCUS 'small bath' in Petronius 42.1), are, as a rule, assigned by experts to a "Mediterranean" rather than to the IE stock; cf. Ernout and Meillet's comments on LENT-ISCUS 'mastic tree' (secondarily associated with LENTUS 'slow' on account of the unhurried flow of its resinous sap?), M ALU A U-ISCUS 'marsh mallow', TAMAR-ISCUS 'tamarisk' (Palladius, beside -IX-ÎCIS in Columella and Lucan, -ICIUMm. Itala, -INDA in Mulomedicina Chironis, etc.; cf. the hydronym TAMARIS and the ethnonym TAMARlCl), and TURB-ISCUS 'spurge flax', which, aside from Sp. torvisco, Ptg. trovisco, B. forbisco, may have left reflexes in Sard, tru-iscu, -vuzu, see Wagner (1943 : 544-46), Bertoldi (1939: 81-82, with further bibliography), and Ernout-Meillet with a reference to J. André. Both the last-mentioned word, obviously semilearned and subject to all manner of associative interferences, and some of the more strictly erudite items have given rise to new derivatives in Spanish, e.g. torvisc-al 'field of spurge flax', trocisc-ar 'to make into losenges', or have generated new meanings: obelisco '(typ.) dagger', or else have affected phraseology: estar hecho un basilisco 'to be in a rage'. On Jud.-Sp. basal-isco see R. D. Abraham's brief comment (1961); basil-isco as the name of an old-style cannon is discussed by Kahane-Tietze, 1958: §81. As regards segmentable Sp. mord-isc-ar 'to nibble, gnaw at' (mord-isco 'bite, nibble'), Ptg. lambisc-ar 'to pick at (food)', lamb-isqu-etro or -ador 'fond of nibbling dainties', one is tempted to start from well-authenticated iterative-diminutive LAMB-ISC-ÄRE 'to lick, lap' (through repeated movements of the tongue) beside standard LAMB-ERE. From here an easy road leads to Ptg. chov-isc-ar 'to drizzle' beside chov-er 'to rain' and establishes a link to nev-isca, etc. (undeniably, that far-flung group, esp. in its -asco branch, may have been nourished by additional roots). Worthy of special attention is the fact that the emphasis throughout is not on 'wind', 'snow', 'rain' viewed as meteorological phenomena, but on such concomitants as 'snowflakes', 'drops of rain', 'whirl of falling, or piles of fallen, leaves' (hojar-asca), an overtone particularly perceptible in the phrase a barr-isco (or berr-isco) 'sweepingly', abundantly documented by Gillet (1951: 216). For preliminary exemplification I refer to Hanssen (1913: §359), whose parenthetic bare mention of barbasco and lambrusca seems unhelpful; the cited phytonym verb-asco 'great mullein' is an outgrowth of VERB ASC UM, worthy of note because even scholars as cautious as Ernout and André seem disposed to credit it to Ligurian.

372

YAKOV MALKIEL

isca, while the dialectal Sardinian deposits of Catalan to this day preserve mor-çsku alongside -isku. It is the key-word mor-isco, then, that Catalan shares with HispanoRomance proper.18 3.6. Old French Since the situation in Old French was recently subjected to a searching review in the article devoted to the gab-ois, ir-ois, jargon-ois type, the presentation of this facet of the problem can be appropriately succinct. Ever since W. Förster's classic note on OFr. franc-eis, -esche as the forerunner of franc-eis, -eise, it has been suspected that the -eis(e) > -ois(e) suffix contains a sizable admixture of -iscu formations assimilated to the better-represented progeny of -Ë(N)SE. In most cases the extant records are not old enough to enable us to discriminate between these two ingredients on the basis of direct evidence; at this point the comparative method alone can help us close the gap. Since It. ted-esco, Olt. franc-esco dovetail smoothly with the archaic -eis, -esche type in Old French, we can, with a measure of confidence, exploit the Italian (and Sardinian) facts symptomatically. Thus, OFr. anten-ois 'year-old (lamb)', a curious distortion of ANNÖTINUS (itself erratically linked to ANNU 'year'), calls to mind Tusc. (Sen. Pist.) cord-esco 'agnello di seconda figliatura' and the like. OFr. de-manois 'at once' may point in the direction of It. man-esco; for a closer semantic parallel note Engl, out of hand. OFr. espan-ois, as P. Aebischer (1948) observed in his meticulous survey of all variants culled from medieval texts (21-2), jibes, both along the temporal and the spatial axes, too closely with Hispan-iscus to be genetically divorced from it. What matters most for our purpose is the fact that between the latest manifestations in Northern France of an autonomous, neatly identifiable organic descendant of -iscus and the infiltration of the earliest Italianisms in -esque there intervened a period of over three centuries in which the suffix at issue was not at all recognizably represented in French. Therein lies the major difference between French and HispanoRomance, because in Spanish and Portuguese the (far earlier) influx of Italian adjectives in -esco encountered a bedrock of preëxistent indigenous -esco (as in parent-esco) and -isco (as in mor-isco).

4. THE WAVES OF ITALIANISMS

4.1.

The General Perspective

The virtually pan-European suffix -esc(o), -esque, as adumbrated in the opening pages of this paper, is only obliquely connected with the primary, paleo-Romance formatives whose separate, though often parallel, trajectories we have been laboriously piecing " See Hoepffner and Alfaric, 1926: 256; Aebischer, 1948: 19-34; Chaytor, 1945: 126; Wagner, 1952: §136.

THE PAN-EUROPEAN SUFFIX "-ESCO, -ESQUE" IN STRATIGRAPHIC PROJECTION

373

together in Section 3. Rather can it be defined as a suffix that diffused in the Renaissance (and that has kept spreading ever since) as a key ingredient o f the general radiation o f Italian culture in an arch stretching f r o m the Iberian peninsula, via France and England, all the w a y — with increasingly weaker results — into Central and Eastern Europe. W h a t is more, -esco has become, the world over, an Italianate derivational suffix par excellence; latives in -issimo,

it has, i n fact, n o rivals i n that category, since the absolute super-

w h i c h inevitably c o m e t o m i n d next, may, at least in part, b e attrib-

uted t o the equally powerful, n o t to say endemic, Latinizing influence. As usually happens under such circumstances, certain leader words have acted as wedges, opening a path through the foreign vocabularies susceptible t o Italian penetration. O n e such word, particularly well-studied because o f its artistic piquancy — with equal reference t o fine literature and to representational arts — has been grotesque,

from It.

grottesco.1β 19

Of sustained appeal to the general reader is W. Kayser's broad-gauged book Das Groteske (1957; the English translation, issued six years later, is far better indexed). From Kayser (50-58) one can glean such entertaining bits of information as the practically free variation of the nouns Groteske, Arabeske, and Moreske in Goethe (1789) and in Germany's tone-setting writers of the following years, down to 1800 (F. and A. W. Schlegel, Tieck, Jean Paul, and a few minor luminaries). As regards the origin of It. grott-esco (adj.) and la grott-esca, Kayser ably summarizes (20-25) the findings of earlier art historians: The word, a derivative from grotta 'grotto', owes its peculiar meaning to the discovery, toward the end of the 15th century, of a long-forgotten style of ornamental painting through excavations — hence the allusion to 'caves' — made in Rome and elsewhere. Giorgio Vasari's learning enabled him to identify the newly-unearthed specimens with an ornamental fashion expressly condemned as bizarre, monstrous, and incongruous in Vitruvius' realistically slanted treatise De architectura, traceable to the Augustan Age. The ornamental grotesque, juxtaposing and interweaving stylized blossoms, leaves, and animals, owes its revival and newly-won respectability to Raphael's playful, light-winged adornment of papal loggias (1515). From these initially harmless combinations the path led, via all manner of nightmarish distortions ("sogni dei pittori", Dürer's Traumwerk), to the fantastic world of H. Bosch and of the members of the Breughel family, and to the "Knorpelgroteske" characteristic of 17th-century Germany. As regards the word itself, note that Fischart (1575) spoke of Grubengrotteschisch, while Lukas Kilian issued three instructional booklets titled Grotesken für die Wand (1607), Neues Grodesko Büchlein, and Grodisko-Buch (1632), each displaying a different variant of the key word. A record traceable to the year 1612 in Frankfurt a. M. speaks of Krodischkenwerk, with a change gr- > crpresumably suggested by contemporary French crot-esque (an Italianism adulterated through contamination with Fr. crot, adj. croté ?) and with a regression from e to i conceivably stimulated either by neo-Latin models or by the vogue of Sp. mor-isco — against the background of Spain's heavy involvement in Central Europe and of the impending amalgam of the two ornamental styles, the Hispano-Oriental and the Italian-Renaissance. For a fuller array of excerpts from older texts see P. Knaak's Greifswald dissertation (1913); for peculiarities of Renaissance usage, P. Scheuermeier, 1920: 71-83; for late-Romantic revival, G. Matoré's article (1946) on some of Th. Gautier's lexical proclivities. For a thorough, critical assessment of Kayser's book see L. Spitzer's review, identified in fn. 28 below. Aside from setting straight the record of certain literary figures (e.g., of Montaigne), Spitzer misses in Kayser's treatment an adequate appreciation of the grotesque streak in medieval art and literature. (Several months after the completion of this article there appeared a lavishly illustrated 200-page monograph by Nicole Dacos: La découverte de la Domus Aurea et la formation ¿les grotesques à la Renaissance, Studies of the Wartburg Institute 31 [London and Leiden, 1968]. From the publisher's prospectus one gathers that this study sheds new light on such matters as the influence of the ancient paintings on Ghirlandaio, Pinturicchio, Perugino, Filippino Lippi, Signorelli, Raphael and his School. There is no evidence of any major discovery with a direct bearing on our linguistic problem.)

374

YAKOV MALKIEL

The point to remember is that, in Italian, -esco refers both to nouns endowed with, and to those utterly devoid of, humorous effects, overtones of artistic refinement, and the like. On the other hand, the Italianisme (-esco words included) borrowed by neighboring languages at the height of the Renaissance — and later imitated — for the most part evoked only features distinctive of Italian culture. Thus, there was no cogent reason for absorbing "neutral" bambin-esco or fanciull-esco, since children behave alike nearly everywhere (conversely, for a while donn-esco proved to be a tempting morsel); it is items relating to the artistic bohème, to regal pomp, and to the modish styles of writing, painting, and living that have more and more dominated the new phase of Sp. -esco as well as of Fr. and Engl, -esque. 4.2.

Golden-Age and Modern Spanish

Over the years the Italianate suffix in Spanish has increasingly tended, along the semantic axis, in the directions of 'exquisiteness', 'delicacy', 'daintiness', 'exoticism', 'artfulness', on the one hand, and of 'oddity', 'bizarreness', 'grotesquerie', on the other. The coinage of relatively neutral words, such as fland-esco 'Flemish' and navaresco 'nautical' — the latter figures in the title of an (unpublished) anonymous late16th-century dictionary made accessible through S. Gili Gaya's Tesoro — would be inconceivable today.20 Also, the substantival series suggestive of 'community, life, manners, behavior of ...': ladron-esca 'gang of thieves', rufian-esca 'gang or conduct of scoundrels', soldad-esca 'soldiery, undisciplined (mercenary) troops, soldiership', must be regarded as closed. The sole truly productive subgroup is that of adjectives, whose primitives can be roughly categorized as follows: (1) The World of Make-Believe (Art and Literature): (a) Names of artists: churriguer-esco, goy-esco, miguel-angel-esco; rembran-esco; (b) Names of writers: boccacc-esco, cervant-esco (beside -ino), dant-esco, gracianesco, guevar-esco, lop-esco (beside -ista), lucian-esco, naharr-esco (in reference to B. de Torres Naharro), petrarqu-esco, queved-esco, unamun-esco ; (c) Names of famous protagonists (and of works of fiction named after them): celestin-esco, donjuan-esco, quijot-esco·, (d) Names of literary genres or styles: caricatur-esco, diecioch-esco, folletin-esco, grot-esco, guiñol-esco, luis-catorc-esco, miniatur-esco, novel-esco, romanc-esco, zejelesco. 10

See Hanssen, 1913: §359 (rub-esco seems to be a ghost word — confusion with RUB-ESCÖ Ί am turning red'?) and Alemany, 1918: §81, who erroneously conjoins turqu-esco and Turqu-ia, rather than turco. Terlingen's Index contains the following hard core of Italianisms absorbed before 1700 : arabesco, brut- beside grut- and grot-esco (also grot-isco in Cristóbal Pacheco), dueñ-esco (three times in Don Quijote, apparently an ephemeral adaptation of It. donn-esco due to a private whim of Cervantes), a la forfant-esca 'roguishly', mariner-esco, merca(da)nt-esco, soldad-esca. The derivatives brut-, grut-esc-ante 'designer of grotesque ornaments' apparently mark a Hispanic innovation. One may safely assume obs. barbar-esco 'barbarous' (now replaced by -icó) to have been another temporary transplant from Italian. See the Supplement for selective illustrations.

THE PAN-EUROPEAN SUFFIX "-ESCO, -ESQUE" IN STRATIGRAPHIC PROJECTION

375

(2) The Real World (measured by the old standards): (a) Exotic countries, viewed in "stylized" projection: chin-esco, tud-esco, turqu-esco; (b) Colorful ethnic (or social) groups: gauch-esco, gitan-esco; (c) Artistic professions: pintor-esco, plater-esco, trobador-esco; (d) Aristocratic environment and festive events suggestive of pomp, pageantry, and refinement: carnaval-esco, princip-esco, versall-esco·, (e) The domain of chanceries, monasteries, and schools — viewed by a jaundiced eye: canciller-esco, capellan-esco, frail-esco, oficin-esco, pedant-esco; (f) The gaudy realms of underworld, demi-monde, adventure, entertainment: bufonesco, canall-esco, chalan-esco ('horse-trading, sharp-dealing'), charlatan-esco, chul-esco, fuller-esco, goliard-esco,juglar-esco, labrador-esco, ladron-esco, maton-esco, pelon-esco (pelón 'poor, penniless'), picar-esco, Ptg. popular-esco ('vernacular, vulgar'), rufianesco, soldad-esco; (g) Sprites and magicians: bruj-esco, culti-diabl-esco, hechicer-esco; (h) Names of animals: (i) generic: animal-esco, brut-esco; (ii) specific: ardill-esco, gat-esco, hormigu-esco, lob-esco, raton-esco, simi-esco. There remains a relatively small residue of formations difficult to pigeonhole by semantic criteria; as a rule, some suggestive formal feature may have acted as a catalyst. Thus, talon-esco 'pertaining to the heel' is queer if viewed in isolation; but given the affinity of -n- to -esco in such highly motivated cases as truhan-esco 'knavish, clownish', villan-esco 'rude, boorish', also Lope's humorous soplon-esco from soplón 'tattletale' (a pattern of behavior for once portrayed as a profession), talon-esco loses some of its oddity. Sayal-esco 'sackcloth' and gregu-esco (frequently misspelled gregüescó) beside grigu(i)-esco 'Grecian wide breeches' stand apart as petrifacts, reflecting a substantival masculine use of -esco no longer pressed into service.21 4.3.

Modern French

There exists a sizable corpus of significant publications on Fr. -esque, and a brief survey of this literature, chronologically slanted, is a suitable avenue of approach 11

A mere handful of formations are troublesome or call for special comment. Faron-esco seems to have been idiosyncratic with Lope. (S. Montoto, 1958:135 cites the passage : "Gente pedante, faronesca y dura/ de su opinión ...)". One is tempted to link it to It. (Jari)farone 'swaggerer, braggart' which, in penetrating into Spanish, was eventually influenced by the native suffix-chain -arr-όη, yielding fanfarrón. Rubén Darío's montjuich-esco (from Montjuich, the name of Barcelona's citadel) owes its genesis, it would seem, toa craving for lexical piquancy rather than to slavish imitation of Barcelona usage; Catalan favors in comparable derivatives from oro- and topo-nyms the suffixes -ench, -és or -I. In the Conclusion to his novel Misericordia Β. Pérez Galdós launched zapaí-esco as a mockserious derivative from the family name (Juárez y Zapatas) of his protagonist; this item is to be distinguished from CI.Sp. zapat-er-esco 'shoemaker-like', which M. Alonso, Enciclopedia del idioma, p. 4236, cites from Garcia, La desordenada codicia (1619), p. 85. The great masters of Golden Age literature — Cervantes, Lope, Quevedo — used -esco freely and imaginatively. From W. Krauss' collection I have excerpted andant-esco, azot-esco, cristian-esco, gat-esco, gobernador-esco, hechiceresco, labrador-esco, matrimonial-esco, tobos-esco as highly characteristic of Cervantes' preference in Don Quixote.

376

YAKOV MALKIEL

to the problems involved. Darmesteter opened the discussion in 1877 by identifying a kernel of formations as Italian by descent (arab-esque, barbar-esque, burl-esque, charlatan-esque, chevaler-esque, etc.), by recognizing the new impetus the Romantic vogue gave a few of them (e.g., grot-esque), and by specifying some as being of recent coinage, despite the continued referential emphasis on the Italian scene: cardinalesque, carnival-esque, dant-esque, michel-angel-esque, raphaél-esque (1877:207-8). Plattner, twelve years later, was concerned only with derivatives from proper names; he supplied copiously documented examples of the anthroponymic series (aristophan-, arnal-, bernin-, charlemán-, chaíeaubrian-, donjuán-, figar-, ingres-, maréchal-, moliér-, raphaél-esque, pitted in rivalry with adjectives harnessing other suffixes), but erred in labeling vatican-esque as toponymie in background; it is the atmosphere of the papal court rather than its location that the suffix served to evoke. In 1894, Meyer-Liibke was too preoccupied with the remote provenience (Greek vs. Germanic) of primary -esc(o) to devote more than a couple of lines to the French Italianisms. It is to Nyrop (1908:§§371-3) that we must turn for the next break-through: Apart from furnishing a much-enriched inventory and from identifying the creators of a few neologisms, he was the first who neatly segregated from the bulk of the material those items which posed formal problems, e.g., charleman-esque, chateaubrian-esque, hugo-l-esque, rembran-esque, vlan-t-esque, zola-t-esque, proposing plausible models for all deviations from the norm (gigant-esque, soldat-esque). He was likewise the first to point out the slight admixture of words borrowed from Spanish (picar-esque) or evocative of Spain's romantic past (alhambr-esque). Reverting to the problem, in 1920, MeyerLübke toyed with the felicitous idea of ascribing the formations' frequently comic effect to the occasional clash between vernacular radicals and the exotic suifix, as in livr-esque. Had he pursued the matter further, he might have made it clear why -esque failed to become acclimatized in French, as it did in Spanish (the two reasons were its structure, with the heavy consonant cluster in its center, and the chronological break). Eloquent, let me further add on my own, is the proximity of Fr. pittor-esque to its Italian model, as against the secondary adjustments undergone by Sp. pintor-esco (rapprochement with native pint-ar, -or) and by Engl, pictur-esque (Latinization of the radical). The more recent analyses, to the extent that they are not focused on special literary usages (as is, e.g., G. Matoré's, 1946), tend to disregard the patterns of sources and of stratigraphy and to ascertain in descriptive fashion the suffix's 'central' meaning. J. Giraud (1957) seems to despair of attaining this goal, seeing a kaleidoscopic profusion of irreconcilable pejorative, neutral, and laudatory nuances, but for his critic J. Marouzeau (1958) this confusing motley effect characterizes only the surface, not the deeper reaches, of the pertinent slice of material. For these underlying reasons a fairly simple, non-contradictory formula can be devised: "Son rôle est essentiellement de désigner une anormalità ou du moins une originalité dans le bon comme dans le mauvais sens". Only one basic ingredient is missing from the otherwise flawless formula: the flamboyance of the writer's or speaker's style is as irreducibly vital a

THE PAN-EUROPEAN SUFFIX "-ESCO, -ESQUE" IN STRATIGRAPHIC PROJECTION

377

condition as is the aberrant character of the object envisaged. For this reason many formations pertain chiefly or even exclusively to certain peculiar styles that fit this description: pungent art criticism, spicy feuilletons, frothy travelogues flavored with all kinds of lexical bric-à-brac, etc. Small wonder that the stabler nucleus of the lexicon, ministering to the speaker's basic needs in everyday routines, has — as J. Dubois correctly observes (1962: 106) — been but minimally affected by these spells of ebullience and fluctuation. The chief mistake of present-day descriptivists has been to neglect, in their search for a unitary tag, the heterogeneous character of -esque and its continued dependence on the sources feeding it. Thus, for the study of the cinematic world, which lies at the heart of Giraud's inquiries, the 'Hollywood-ej^we' bridge to the English-speaking countries demands heightened attention. In a more academic ambiance, the preference of such contemporary French Hispanists as M. Bataillon, P. Hourcade, and P. Le Gentil for words charged with 'Luso-Iberian' local color, like jongIer-esque,pessoesque, pinzon-esque, zéjel-esque, presupposes the earlier use of juglar-, pinzón-, zejelesco in Spanish and of pesso-esco (in reference to Fernando Pessoa) in Portuguese. Sp. pinzon-esco, grafted onto the name of a companion of Columbus (Martín Alonso Pinzón), is in turn best explained through the previously noted affinity of -esco to -an-, -in-, -on-, -un- in Hispano-Romance (cf. capellan-esco, chalan-esco, donjuan-esco, gitan-esco, lucian-esco, rufian-esco, truhan-esco, villan-esco ; celestin-esco, chin-esco, oficin-esco ; bufon-esco, ladron-esco, maton-esco, raton-esco, soplon-esco·, unamun-esco). The threads are, then, far more involved than appears at first glance; and if France loses, in the process of such impartial reappraisals, part of the credit for originality, it stands to recoup with a vengeance, because Sp. diecioch-esco, {grati) guiñol-esco, luis-catorc-esco, miguel-angel-esco and similarly daring derivatives from entire phrases — known as favorites of Rubén Darío's — prove to be deft and tasteful imitations of French prototypes.22 From the myriad individual lexical developments into which, upon close inspection, the growth of Fr. -esque dissolves one may single out for specific mention roman" In his brilliant essay on Dario's language, R. Lida sagaciously includes the "orgies" of derivatives in -esco (1967:357) among the manifestations of the poet's verbal inventiveness stimulated by domestic and foreign models. As regards combinations of phrase plus suffix, I may add on my own that (despite Lope's facetious cidti-diabl-esco) the impact of French was extra-heavy; on this point see Pichon's perceptive and well-documented analysis (1942: 59-67) of "Attache d'un suffixe à un complexe". Pichón (1942) also offers incidentally a profusion of other pertinent data and comments: on rembranesque's affinity to printan-ier and ortieman-iste (26) ; on the -I- of hugo-1-ûtre, -ien (30), which reappears in hugo-l-esque ; on zola-t-esque (55) ; on (casually overheard) baldesquatzar -esque, from Bal des Qua?ζ* Arts (64); on Verlaine's silvio-pelliqu-este, involving for once the suffix of agr-este, cil-este chosen to avoid cacophony, and on saint-papoul-esque (65). Rather characteristically, the French call petit Mauresque a kind of lingua franca or sabir used in Algeria. See Harvey, Jones, and Whinnom's recent note in RLC. Even where a word in -esque is not, strictly speaking, Hispanic, its history may have a trans-Pyrenean admixture, cf. the joint statement by A. Dauzat, J. Dubois, and H. Mitterand (1964) on the late-19th-century neologism rocambolesque 'grotesque': "De Rocambole, nom d'un personnage aux aventures extraordinaires créé par Ponson du Terrail ... d'après le mot vieilli rocambole, 1680, Richelet, s.f. 'ail d'Espagne', d'où 'chose piquante', de l'ail, Rockenbolle".

378

YAKOV MALKIEL

esque, both because it happens to be currently at the center of lively esthetic discussion 2 3 and because, placed in a broader, pan-European context, it exemplifies certain complexities of h o m o n y m y and polysemy. As the result of a historical accident to which G. Paris' classic article alerted scholars nearly a century ago, a single radical, namely Rom-, has been called u p o n — throughout the West — t o satisfy a profusion of competing, equally pressing demands o n the referential side: It serves t o suggest '(ancient and modern) Roman', 'Romance', 'R(o)umanian\ 'Romanesque' (as against Byzantine or Gothic, chiefly in architecture, more sparingly in sculpture, wall painting, polyphony, lyric, and binding), 'novel of adventure', 'ballad', 'attuned t o a m o o d suggestive of, or suggested by, novels with romantic overtones'. For the most part the meanings here tentatively isolated and contrasted can be brought out separately through the clever exploitation of a set of nearly synonymous derivational suffixes; within this gamut -esc(o), -esque almost everywhere plays a major rôle, but, paradoxically, in each language it is charged with individuating a different component of the radical's complex meaning. French separates roumain 'Rumanian' from romain 'Roman', and both of these from roman, which to laymen and literary scholars means any 'novel' (not only medieval 'romance'), to philologists 'Romance', and to art historians 'Romanesque', while roman-esque has the above-mentioned meanings of 'romantic, adventurous, extravagant, enchanted, far-fetched'. Other languages distribute the congeners of the same suffixes in different patterns. 24 23 Fr. roman-esque has inspired a significant corpus of writings, some of which, placing the word in the modern perspectives of alienation and appealing to mediation, chiefly concern the literary critic; cf. Girard's book, in its original form (1961) and in the English translation (1965), as well as the finely nuanced spectrum of reactions it has provoked; also Champigny, 1963. More directly relevant to our problem are W. von Wartburg's statements (1962) s.w. RÖMÄNICE and RÖMÄNUS. Aside from noting the neutral use in Italy, including that country's strain of Medieval Latin (Città di Castello, A.D. 1538: lana roman-esca), Wartburg reports the adoption of roman-esque by Du Bellay, that enthusiastic admirer of things Italian, then traces substantivized roman-esque (fem.) 'sort of song' to 1572 and 1634 and 'kind of lively dance* to 1703, treating separately, as a homonym, the words extracted from roman 'novel': roman-esque 'merveilleux, fabuleux comme les aventures du roman' (1628, 1634, 1666; subst. 1689), 'exalté comme les personnages de roman, comme les sentiments qu'on leur prête' (1628); -esquement (1676), -esquerie 'tendance à être romanesque, exalté, chimérique'. His sharp condemnation (p. 456α) of W. Krauss' sensitive article (1937-38) is groundless; the assumption of a measure of continuity from roman-esquet (rooted in RÖMÄNUS) to romanesque2 (extracted from roman) seems defensible. 14 Thus, It. roman-esco refers to modern Rome, including its lower-class speech, as against romeno 'Rumanian*, rom-ano '(ancient) Roman' or 'Romanesque', and romanz-esco, semantically the exact replica of Fr. romanesque. Rumanian uses -esc to refer to the native culture alone: (ara románeascä 'Wallachia', limba romän-eascä 'Rumanian language'. Spanish has at its disposal self-explanatory rum-ano vs. rom-ano-, it endows romance with the meanings 'Romance' (in rivalry with román-ico) and 'ballad', assigning to romanc-esco the meaning 'ballad-like', while novela denotes the narrative prose genre. English suffers from a plethora of spellings fo Rou-, Ru-, Ro-man-ian; it has allowed romance, orig. 'medieval novel of adventure', to develop quite far in the direction of '(typically, Platonic) love affair' ; -esque is here confined to the style preceding Gothic, while Roman-ic is an optional philological tag. German discriminates between rumänisch, römisch, romanisch, romanhaft; it needs no adjective from Romanze (more musical than folkloristic) and dispenses altogether with -esk, but boasts an extra slot for roman-ist-isch 'devoted to Romance philology*. On Ptg. roman-isco see fn. 16, above.

THE PAN-EUROPEAN SUFFIX "-ESC», -ESQUE" IN STRATOGRAPHIC PROJECTION

379

4.4. English As regards social matrix, semantic ambit, and stylistic flavor, Engl, -esque, enormously successful in certain quarters at the present moment, differs but little from its French and Spanish cognates, except perhaps for a somewhat heavier concentration on proper names. Among these the names of actors, motion-picture directors, dancers, painters, and all manner of intellectuals, especially those known for a dosage of sensationalism and even of exhibitionism, occupy a place of honor: Bergman-esque2S, Boudin-esque (but Picasso-ish, Sisley-ish), Chaplin-esque, Duncan-esque, Giott-esque, Hepburn-esque, Hitler-esque (as a jocular variation upon -iari), Kinsey-esque, Menckenesque, Persian-esque, Robinson-esque, Tristan-esque are cases in point, outnumbering by a wide margin such innovations as girl-esque (despite the support — in a music-hall environment — which the latter word draws from burl-esque). What is uniquely peculiar to English is the care with which the final vowel of the primitive — its inalienable core-constituent in this language — is preserved (as it is in Jingo-ism, Faro-ese, and the like): witness bikini-esque, Dali-esque, Hardy-esque, Jtmo-esque, Kafka-esque, Rousseau-esque, Stravinsky-esque (but only Prokofief-style !), Tito-esque. True, there are on record Giott-esque beside Giotto (see above), Dant-esque beside Dante, and Goy-esque beside Goya — but all three are directly modeled on foreign usage. The southern Romance speech communities follow a different path: They unhesitatingly lop off -a, -e, -o (less consistently -i and -«), which lexically represent little more than an outer shell or wrapping, before any derivational suffix ushered in by a vowel. Similarly, French willingly sacrifices its -e, while introducing an intercalary consonant to salvage other endangered final vowels, cf. hugo-l-esque, zola-t-esque. Where the hazard of an inopportune encounter and subsequent conflation of -e and -esque begins to loom in English, the risk is averted through the use of differently-shaped rival suffixes, e.g. Rilke-an (not unlike Proust-ian, Shakespearian) in preference to *Rilke-esque. Among the pan-European words shared by English burl-esque invites special attention. It. burla 'practical joke, prank, harmless trick' (cf. -etta 'vaudeville') invaded, along with burl-esco 'jesting, farcical', many neighboring languages; thus, French b(p)urle 'plaisanterie' was tolerated from Brantôme to the 17th century, and bourrel· esque (Satire Ménippée, 1594), ultimately transmuted into burl-esque, survived, even The side-dividends of Krauss' paper include a short list (301) of -esque formations demonstrably in use before 1600 — a list heavily studded, as one would expect, with almost unadulterated Italianisms: arab-esque, baragouin-esque, barbar-esque, battier-esque, bougr-esque, bourdontt-esque, brigandesque, calvin-esque, camibal-esque, courtisan-esque (as against Chapelain's undisguised cortigianesque), crot-esque,forfant-esque, livr-esque, porch-esque, priap-esque, ripon-esque, turqu-esque. One deviant Italianism propelled through a non-literary channel of transmission is (bozza alla) barb-ar-esca 'ring stopper' (from barba 'beard', possibly influenced by barbar-esco 'barbarous'): It yielded Prov. barbar-asso -*• Fr. -asse, recorded in 1792. See Kahane and Tietze, 1958 : §70. 15 The intrinsic connotation of 'artfulness, sophistication' makes it unmistakably clear that the reference is to the experimentally-minded director Irtgmar, not to the actress Ingrid celebrated for the candor of her portrayals.

380

YAKOV MALKIEL

though derivationally it had cut loose from its moorings. English followed suit, using burl-esque for 'derisively imitative or exaggerated (performance)' ; at that stage burlesque and grot-esque, semantically akin and both straddling the domains of adjectives and substantives, supported each other. But the literature of burlesque now belongs to the past; what remains intact is the reference to 'theatrical entertainment consisting of coarsely comic skits and provocative strip-tease acts', with marked overtones of the tasteless and erotically risqué.26 Note that picar-esque 'roguish' rides the crest of a vogue today, while no attempt has been made to adopt the underlying Sp. picaro. 4.5.

German and Russian

German uses its version of our suffix very sparingly and only in a recherché style. Burl-esk is distinctly rarer than drollig, schnurrig, spasshaft; grot-esk, except in reference to art forms, trails behind auffällig, seltsam, wunderlich ; pittor-esk is overshadowed by malerisch. Dant-esk is marginally tolerated (K. Wais, a deft stylist, accepts it); but Sp. goy-esco, lop-esco are usually translated 'auf Goya, Lope bezüglich'. When Carl Zuckmayer speaks of chaplinesker Gang, he evokes not only the inimitable gait of the comedian, but also the tone of American motion-picture criticism (note that the German writer, during his exile, was a long-term resident of Hollywood). One finds in German a greater predisposition in favor of the substantivized forms. Ordinarily, the designation of a specific work of art exhibiting a conspicuous style appears in the feminine, as in (die) Arab-eske, Grot-eske, Humor-eske, whereas the preferences of Spanish are divided: el arab-esco, grot-esco, but Las Goy-escas (title of E. Granados' suite).2' In addition, German esthetics favors a philosophical neuter alluding to an immanent principle, as when Justus Moser spoke of the GroteskKomische and K. F. Flögel, in 1788, published a full-blown Geschichte des GroteskKomischen (subsequently brought up to date by F. W. Ebeling & M. Bauer), a concept which Nicolai Hartmann, in his posthumous Ästhetik (1953), was later to separate from the Feinkomische. The preference for substantivized forms is even more sharply pronounced in Russian, where the generative force of the suffix dwindles almost to the zero point. Speakers are undecided as to which is more appropriate in the singular, arab-esk (m.) or -eska (f.), though in the plural, which refers not nearly so often to the architectural ornament as to a miscellany of literary or musical pieces, the masculine is strictly " The quasi-obligatory element of piquancy has made it difficult for English-language art critics to rely on pictur-esque in translating G. maler-isch, which is of course innocuously romantic. Hence the tentative coinage of painterly, a not entirely successful word reminiscent of Russ. zivopisnyj and of Pol. malowniczy. " Limitations of space preclude an excursus on the suggestive use of "Romantic" -esque in titles of literary, musical, and visual-arts compositions. Suffice it to recall Gogol's Arabeski, Poe's Tales of the Grotesque and the Arabesque, Chekhov's Jumoreski, Ravel's Grotesques, and Granados* Goyescas, a six-part suite evoking the gaudy, sensual world of the 18th-century artist's paintings and tapestries (whereas Goya himself was content to use caprichos).

THE PAN-EUROPEAN SUFFIX "-ESCO, -ESQUE" IN STRATIGRAPHIC PROJECTION

381

required. The dictionaries list an adjective grot-esk-nyj (as a qualifier of 'style' or 'performance, spectacle'), but the sequence -sktt- is so woefully unattractive as to prompt speakers and, above all, writers, to circumvent the difficulty by saying evasively stiV grot-esk, thus avoiding the use of a normally developed adjective; as a substantive, grot-esk is fully acceptable, contrasting, as regards gender, with jtimoreska. As is well known, phonological constraints have made it difficult for speakers of Russian to adopt so international a word as modernise) — *modern-nyj would be infelicitous in the extreme —, while modern-ism causes no trouble at all.28

5. CONCLUSION

Each language considered at any length in our bird's-eye survey shows some distinctive peculiarity either in the range of its version of the pan-European suffix -esco or in the specific semantic hue of that polychromatic formative. In Rumanian, -esc(u) is characterized by its simplicity and unpretentiousness, a feature that invites its frequent transfer to Slavic stems; it also abounds as a marker of native family names. Sard, -iscu has preserved discernibly more than elsewhere the diminutive value of its Greek ancestor; it has also generated a minor vocalic gamut supported by the consonantal pillar -SK-, in part through absorption of Tusc. -escu, in part through spontaneous production of a rival branch of somewhat differently colored derivatives in -uscu. In Italian, the highest evolutionary potential of -esco has been fully realized; quite often it serves as a purely relational suffix, without conspicuously humorous implications, and even enters into family names; then again, it may suggest artistic individuality and originality, or any kind of striking shape, attitude, or performance, with occasional overtones of (the onlooker's) surprise or amusement. In Old Provençal, adjectival -esc was weakly developed, but, as if in compensation, substantivized -esc and -esca (typically confined to abstracts and mass-nouns) reached their peak, of all places and periods, in medieval Southern France. Old Spanish and Old Portuguese showed no trace whatever of vernacular -esco in " Long after completing this article I discovered Leo Spitzer's lengthy analysis of -esco, which extends over the two concluding pages of his sparkling review of Kayser's Das Groteske (Spitzer, 1958). Taking issue with Kayser's rather amateurish treatment of the linguistic facet of the problem, Spitzer correctly recognized the fundamental difference between (a) It. -esco, which is neutral ("einfach Herkunft und Artung bezeichnet"): cagnesco, (canti) carnascialeschi, (gatto) lupesco, (cavallo) poltresco, and acquires facetious overtones only when the primitive communicates them, as in pazzesco, and (b) Fr. -esque, which suggests exoticism as a result of the slant of its leader words, borrowed from a certain section of the Italian lexicon: (a) arabesque, mauresque-, (β) bourrelesque (later burlesque)·, (if) grotesque. From here the trajectory leads to caricaturesque, funambulesque, etc. An excellent connoisseur of Rabelais, Spitzer identified a passage (the obscene litany in Bk. 3, Ch. 26 ; pubi, in 1546) testifying to this amalgam: "couillon de crotesque, couillon arabesque ... couillon troussé à la levresque" (elsewhere, in Bk. 4, Ch. 12, one finds the equally Italianate accoustré à la tigresque).

382

YAKOV MALKIEL

the ranks of adjectives; they did tolerateparent-esco 'kinship' (reminiscent of Provençal usage) and a fairly sizable series of -isco adjectives, in which one detects the separate strain of Latin-oriented chanceries (adaptation of -iscus) and of Hispano-Arabic 'Mischkultur' (assimilation of Semitic -i). Starting with the Renaissance, the Hispanic languages became accessible to Italian influence, adopting merely the characteristically, entertainingly authentic elements of It. -esco. This new and narrow semantic and stylistic content has carried over into the later and even the most recent periods: It. dugent-, trecent-esco, etc., constitute an open series producing such adjectives as qualify any kind of temporally delimited human activity; Sp. diecioch-esco is restricted to a single highly sophisticated and artful style (of living, architecture, etc.) and does not readily lend itself to proliferation. Along this path Spanish has occasionally overtaken Italian ; thus, the latter is content with cin-ese, while the former, beside extending the ambit of chino (cuentos chinos 'extravagantly mendacious tales'), for a while tolerated a facetious derivative -esco, witness J. P. Forner's Los gramáticos: historia chinesca and Jovellanos' "ideas ... un poco chinescas" 'paradoxical, bizarre' (ca. 1780). Native French -eis, -esche fell prey to an early homonymie collision and then exerted no influence on -esque, subsequently imported from Italy. This foreign-sounding -esque, evocative of exotic art, has had a major vogue — during the Renaissance and, independently, after 1800 —, appearing in derivatives from proper and common nouns alike. In English the piquancy and esotericism of -esque reach their apogee; the suffix is used, typically among connoisseurs and art critics, in derivatives of (preferably, foreign) names, often with full preservation of the primitives' final vowels. In German -esk is a marginal, optional suffix, which has failed to win a place in the core vocabulary and may thus be altogether avoided. Russian shows little more than vestigial evidence of -esk(a), in a sprinkling of borrowed substantival formations. The one almost pervasive characteristic that one can distill from all these analyses is the overwhelmingly human ingredient in the spectrum of referential values of -esco : Whether directly absorbed from Late Latin or refracted through the prism of Italian, the suffix relates, with negligible exceptions, to the realm of man and of his immediate concerns, whether these involve an élite's lofty devotion to arts and letters or the poor man's humble assortment of domestic.animals. UNIVERSITY OF CALIFORNIA, BERKELEY

SUPPLEMENT: A SHORT SELECTIVE GUIDE TO HISPANIC -esco FORMATIONS ANDANTESCO: Blanca de los Ríos Lampérez, Obras completas, V (1912), 12 ("nuestra españolería andantesca"). ANIMALESCO : A. Alonso, De la pronunciación medieval a la moderna ..., I (Madrid, 1955), 119 ("las letras ammalescas"); D. Maçàs, Os animais na lingua portuguesa (Lisbon, 1950-51),

THE PAN-EUROPEAN SUFFIX "-ESCO, -ESQUE" IN STRATIGRAPHIC PROJECTION

383

p. 290 ("para a contribuïçào animalesca nas religiöes"). ARDILLESCO: J. Zorrilla, Prólogo a El cantar del romero, in Obras, II (1943), 287 ("el ardillesco movimiento y bárbaro tecnicismo del chulo"). BARBARESCO: see Alemany Bolufer, Tratado, §81. BARRIOLATINESCO: R.Darío ("noche barriolatinesca"); cf. R. Lida, Rev. iber., XXXIII (1967), 357. BOCCACCESCO : M. R. Lida de Malkiel, La originalidad artística de "La Celestina" (Buenos Aires, 1962), p. 47. BRUJESCO: see W. Meyer-Lübke, GRS, II, §520. BRUTESCO: Cervantes, Don Quijote, Part I, Ch. 1 = ed. Bonilla and Schevill, II, 371 ("a lo brutesco"). BUFONESCO: Β. Pérez Galdós, Misericordia (1897), Ch. 32 ("una inmensa falange gloriosa y bufonesca")', Lida de Malkiel, La Celestina, pp. 140 ("un bufonesco contrapunto"), 616 ("un dúo bufonesco"), 627 ("actitud bufonesca"). BURLESCO: See F. Hanssen, Gramática, §359. CABALLERESCO (cf. Hanssen, Gramática, §359, and Alemany Bolufer, Tratado, §81, who is ill-advised in labeling the word obsolete): Pérez Galdós, Misericordia, Ch. 37 ("sus habilidades caballerescas")·, Menéndez Pidal, title of a pamphlet (1948): Cervantes y el ideal caballeresco·, Lida de Malkiel, La Celestina, p. 564 ("novela caballeresca")·, note that -esco here relates to -osa as Engl, chivalr-ic to -ous. CANALLESCO: Lida de Malkiel, La Celestina, pp. 160 ("gula canallesca"), 713 ("a lo canallesco"). CANCILLERESCO: i. Slaby, Sp.-D. Wörterbuch (estilo cancilleresco 'Kanzleistil'). CAPELLANESCO: Pérez Galdós; see Η. Ν. Bershas, RPh, IX (1955-56), 26, who suspects the influence of Quevedo ("esta vida capellanesca, escolástica y cardenalicia"). CARICATURESCO 'caricatural'. CARNAVALESCO: see Alemany Bolufer, Tratado, %81. CELESTINESCO·. Lida de Malkiel, La Celestina, p. 578 ("género celestinesco"). CERVANTESCO : rather than -ino (but only gongorino). CULTIDIABLESCO: Lope de Vega, see S. Montoto, BRAE, XXVI (1948), 456. CHALANESCO'. Pérez Galdós, Misericordia, Ch. 16 ("este oficio chalanesco"). CHARLATANESCO: rare, like pedantesco, because charlatán and pedante lend themselves to adjectival use. CHINESCO: Cervantes, Don Quijote, II, Dedic. Ill ("en lengua chinesca")', J. Cadalso, Cartas marruecas, p. 52 ("cartas chinescas"). CHULESCO: C. Clavería, Estudio sobre los gitanismos del español (Madrid, 1951), ρ. 31n. ("estilo chulesco"). CHURRIGUERESCO: relating to the inordinately ornate architecture of José Churriguera, Ribera, and their 18th-century followers, flanked by churriguer-ismo 'overloading with unsuitable décor' and -ista 'architect adopting the extravagances of this style'. DANTESCO: see Alemany Bolufer, Tratado, §&1. DIECIOCHESCO: L. Lázaro, RFE, XXXVII (1953), 221 ("la frondosa literatura dieciochesca"). DONJUANESCO: Lida de Malkiel, La Celestina, p. 396 ("su donjuanesco amigo"). FARONESCO: Lope, cf. Montoto, BRAE, XXVIII (1948), 135 ("gente pedante, faronesca ..."). FOLLETINESCO: Lida de Malkiel, La Celestina, p. 466. FLANDESCO: Lope, see Montoto, ΒRAE, XXVII (1947), 137, beside Beir. FRANDESCA 'kitchen knife', cf. D. Maçâs, BF, XII: 2 (1951), 205. FRAILESCO: see Hanssen, Gramática, §359; a shade less offensive than frailuno (RPh, IV [1950-51], 32), but poles apart from reverential cardenalicio, papal, etc. (for references to A. de Cáceres Sotomayor, Lope de Vega, Fray H. F. Paravicino, Pragmática de tasas [1680] and for delimitation vis-à-vis competitors in -engo and -iego see t/CPX, IV: 3 [1951 J, 146). FULLERESCO : see Alemany Bolufer, Tratado,% 81. GATESCO: Cervantes, cf. RPh, IV (1950-51), 37 (the only passages C. Fernández Gómez's dictionary [1962] cites are from Don Quijote, III, 3: "las gatescas heridas" and "aquella canalla gatesca"). GAUCHESCO: the lone American Spanish derivative. GITANESCO: Pérez Galdós, Misericordia, Chs. 22 ("facciones algo gitanescas"), 27 ("su educación gitanesca")·, Lida de Malkiel, La Celestina, p.5 58 ("halago gitanesco"). GOLIARDESCO : R. Lapesa, De la Edad Media a nuestros días (Madrid, 1967), p. 74. GOYESCO: as against goyano, from Goya (Argentine top.) and Goyaz (name of Brazilian state). GRACIANESCO: see M. Alonso's Enciclopedia del idioma for a quotation from F. Maldonado, Salvaciones (1953), p. 95. GREGUESCO 'Grecian (wide breeches)': Quevedo, Buscón, ed. 1911, p. 186 ( < Louis.-Fr. grègue 'drip coffee pot'? Cf. W. A. Read, "Some Words from French Louisiana", RPh, VII [1953-54], 183) beside grigu(i)esco, favored by Lope (Montoto, BRAE, XXVII [1947], 142). GROTESCO: J. Benavente, Prólogo a Los intereses creados ("son las mismas grotescas máscaras de aquella comedia"). GUEVARESCO: Lida de Malkiel, La Celestina, p. 576 ("estilo guevaresco"). G UIÑOLESCO : Benavente, Prólogo a Los intereses creados ("es una farsa guiñolesca"). HECHICERESCO: see Alemany Bolufer, Tratado, §81. HORMIGUESCO: Golden Age Spanish, see NRFH, ΧΠΙ (1959), 247. JUGLARESCO: cf. the title of R. Menéndez Pidal's book (1924), Poesía juglaresca y juglares, preceded by the article (1923), "Caracteres de la poesia juglaresca" (Revista de Occidente, II [1923], 171-200); patterned after It. giullaresco? LABRADORESCO: Cervantes, Don Quijote, Part II, Ch. 47 = ed. Bonilla and Schevill, IV, 98 ("las bodas labradorescas"). LADRONESCO: see Alemany Bolufer, Tratado, §81. LOBESCO: Lope, see Montoto, BRAE,

384

YAKOV MALKIEL

XXVni (1948), 309. LOPESCO: dominant, but E. Asensio, RFE, XXXVI (1952), 332, prefers: "teatropreiopistd". LUCIANESCO: Lida de Malkiel, La Celestina, p. 626 ("comedia lucianesca"). L UISCATORCESCO, cited from R. Darío ("tipos luiscatorcescos") by R. Lida, Rev. Iber., XXXIII, 357. MARINESCO·, see Alemany Bolufer, Tratado, §81. Matonesco: in a passage from A. Bonilla y San Martín ("tipo matonescó"), adduced by Lida de Malkiel, La Celestina, p. 701. MIGUELANGELESCO: Pérez Galdós, Misericordia, ch. 33 ("mascarón escultórico, miguel-angelesco"). MINIA TURESCO : Lida de Malkiel, La Celestina, p. 40 ("el enfoque miniaturesco"). MONTJUICHESCO: in a passage from R. Dario ("militarismo montjuichescó"), cited by R. Lida, Rev. Iber., XXXIII, 357. NAHARRESCO (in reference to the playwright Β. de Torres Naharro): E. Asensio, RFE, XXXVI (1952), 332 ("estudios naharrescos"). NAVARESCO 'nautical, maritime, navigational': cf. the anonymous Vocabulario navaresco, a late-16th-century MS put to use by S. Gilí Gaya in his Tesoro. NOVELESCO: see Alemany Bolufer, Tratado, §81. OFICINESCO: cf. the fragment cited in Tarr-Centeno's Grammar, §187 ("... interrumpió su oficinesca labor"). PEDANTESCO, currently supplanted by pedante: see Alemany Bolufer, Tratado, §81. PELONESCO: Lope, see Montoto, BRAE, XXVIII (1948), 471. PETRARQUESCO: F. Weber de Kurlat, in RPh, XXII (1968-69), 248 ("la vena petrarquesca"). PICARESCO : used by Juan Valera in a letter dated Dec. 1, 1896; cf. D.Alonso, BRAE, XXXIII (1953), 352. PINTORESCO: see Alemany Bolufer, Tratado, §81, and note the tertiary derivative pintoresquismo in Clavería, Estudio sobre los gitanismos del español, p. 27. PLATERESCO: see Alemany Bolufer, Tratado, §81. POPOLARESCO: S. da Silva Neto, História do latim vulgar, p. 49 (cf. Engl, popularist, G. volkstümlich). PRINCIPESCO: residencia principesca (Slaby); but califal, condal (or, ironically, condesil), ducal. QUEVEDESCO: obligatory. QUIJOTESCO: see Alemany Bolufer, Tratado, §81 (as against E. quixot-ic). RATONESCO: Golden Age Spanish, also in F. M. Samaniego, cf. RPh, IV (1950-51), 37 and NRFH, XIII (1959), 247. REMBRANESCO : recorded in the Espasa-Calpe encyclopedia, from Rembrandt)·, but prerrafaelista, to match E. pre-Raphaelite, and velasqueño, to avert cacophony. ROMANCESCO 'relating to balladry': favored by M. Alvar, RF, LXIII (1951), 254; by P. Salinas in a celebrated essay; and by Lida de Malkiel, La Celestina, p. 42; cf. Alemany Bolufer, Tratado, §81. RUFIANESCO: a favorite with Valera (cf. his letter, dated Dec. 1,1896, quoted by D. Alonso, BRAE, XXXIII, 352); with J. López Silva, La musa del arroyo (Madrid, 1911), p. I l l ("en tu rufianesco argot"), cf. Clavería, Estudio sobre los gitanismos, p. 33; and with Bonilla y San Martin ("tipo rufianesco"·, see Lida de Malkiel, La Celestina, p. 701), cf. Alemany Bolufer, Tratado, §81. SIMIESCO: translated by J. Slaby as 'affenartig', beside cara simiesca 'Affengesicht'. SOLDADESCO: Β. de Torres Naharro's play Comedia soldadesca has been likened to G. Β. Della Porta's La fantesca·, Lope, Peribáñez, 11.2405 f., uses the derivative to refer to a level of speech ("... hoy os hablara / en lenguaje soldadesco"). SOPLONESCO : Lope, cf. Montoto, BRAE, XXIX (1949), 143. TALONESCO : see Alemany Bolufer, Tratado, § 81. TROBADORESCO: T. Navarro, Métrica española (Syracuse, 1956): "canción trovadoresca" — where, for once, Italian prefers trovadorico. TRUHANESCO: J. F. Montesinos, BH, LIV (1952), 402. TUDESCO, obsolescent in the literary language (Alemany Bolufer, Tratado, §81; Hanssen, Gramática, §359), has seeped into rural dialect speech, cf. C.-Ast. (Cabranes) tudescu 'persona muy desabrida, muy seca de palabras' (M. J. Canellada, El habla de Cabranes [Madrid, 1946], s.v.). TURQUESCO (not from Turquía, with loss of i [Alemany Bolufer, Tratado, §81], but through elaboration on turco): in Duque de Osuna's letter to Philip III, dated 1626 and quoted by A. Castro in the Preface (p. viii) to his ed. (1927) of Quevedo's Buscón ("aderezos turquescos de diferentes piedras y valores"). UNAMUNESCO: A. Sánchez, RFE, XXXVI (1952), 328 ("español y unamunesco"). VERSALLESCO : clearly paralleling dieciochesco. VILLANESCO : a word sanctioned by Cervantes (Rinconete y Cortadillo, ed. F. Rodríguez Marín, p. 353) and, of late, resuscitated as a qualifier of a level of speech: E. Asensio, RFE, XXXVI (1952), 335 ("jerga villanesca"); R. Menéndez Pidal, Toponimia prerrománica hispánica (Madrid, 1952), p. 157 ("el lenguaje villanesco"). ZAPATERESCO 'relating to the cobbler': a mock-refined formation cited by M. Alonso, Enciclopedia del idioma, s.v., from Garcia, La desordenada codicia (1619), p. 85. ZAPATESCO 'relating to Juárez y Zapatas'': Pérez Galdós, Misericordia, Epilogue, p. 393 ("las efemérides ['diaries'] zapatescas"). ZEJELESCO: Hispano-Arabic hybrid launched ca. 1950 through the joint efforts of D.Alonso, R. Menéndez Pidal, and E. García Gómez.

THE PAN-EUROPEAN SUFFIX "-ESCO, -ESQUE" IN STRATIGRAPHIC PROJECTION

385

REFERENCES Abraham, R.D., 1961 RPh 14 : 242, footnote 28. Adams, E.L., 1913 Word-Formation in Provençal (New York and London, Macmillan). Aebischer, P., 1948 Estudios de toponimia y lexicografía románica (Barcelona, Escuela de Filologia), pp. 1348 ("El étnico español: un provenzalismo en castellano"). 1950 "La literie et l'histoire du matelas d'après des matériaux médiévaux romans", ZRPh: 66.303-337. 1951 "Une possibilité nouvelle concernant l'origine du fr. 'chaussée', esp. 'calzada'", RFE 35: 8-28.

Alemany (Bolufer), J., 1918 "Tratado de la derivación y composición de las palabras en la lengua castellana", BRAE, vol. 5. Alessio, G., 1950 "Saggio di etimologie francesi", RLiR 17: 158-207. Alonso, D. 1953 "Esp[añol] lata, latazo", BRAE 33: 351-388. Baldelli, I., 1955 RPh 9: 43. Bertoldi, V., 1939 Questioni di metodo nella linguistica storica (Napoli, Stabilimento tipografico). Branca, V., 1959 RPh 13: 356. Brault, J.G., 1950 RPh 15: 129-38. Butler, Jonathan L., 1969 "The Latin Derivational Suffixes ima, ina, ¡nus, and ineus", unpublished dissertation (University of California, Berkeley). Catalán Menéndez-Pidal, D., 1953 Poema de Alfonso XI: fuentes, dialecto, estilo (Madrid, Gredos). Chambers, F. M., 1945 MLN 60: 475-77. Champigny, R., 1963 Le genre romanesque: essai (Monte-Carlo). Chaytor, H.J., 1945 From Script to Print: An Introduction to Medieval Literature (Cambridge, at the University Press). Contra?, E., and M. Popescu-Marin, 1967 "La suffixation dans le roumain des XVI'-XVIII· siècles", RRL 12 : 397-416. Coraminas, J., 1949 RPF 3: 35-39, 51. Darmesteter, Α., 1877 De la création actuelle de mots nouveaux dans la langue française et des lois qui la régissent (Paris, F. Vieweg). Dauzat, Α., J. Dubois, and H. Mitterand, 1964 Nouveau dictionnaire étymologique et historique (Paris). Devoto, Giacomo, 1966 Avviamento alla etimologia italiana. Dizionario etimologico (Firenze, Le Monnier) ("Dizionari Le Monnier"). Dubois, J., 1962 Étude sur la dérivation suffixale en français moderne et contemporain (Paris, Larousse). Garcia de Diego, V., 1950 "El castellano como complejo dialectal y sus dialectos internos", RFE 34:107-124.

386

YAKOV MALKÏEL

Giesc, W., 1952 Ζ RPh 68: 173-74. Gillet, Joseph E., 1951 "Propalladia" and Other Works of Bartolomé de Torres Naharro, Vol. 3 (Bryn Mawr, Pa.), Notes. Girard, René, 1961 Mensonge romantique et vérité romanesque (Paris, Grasset). (Note the following year's discussions by A. Hoog, RR 53: 316-18, and by F. P. Bowman, MLR 57: 454-55.) Translated in 1965 by Yvonne Freccerò, under the title Deceit, Desire and the Novel (Baltimore, Johns Hopkins Press); reviewed in 1967 by K. D. Uitti in HR 36: 189-194. Giraud, J., 1957 "Quelques néologismes dans la langue du cinéma", FM 25:212-16. 1962 "Sur les suffixes '-esque' et '-ien"\ FM 30: 115-19. Hanssen, Federico, 1913 Gramática histórica de ta tangua castellana (Halle, Niemeyer). Hoepffner, E., and P. Alfaric, eds., 1926 La Chanson de Sainte Foy, vol. 1 (Paris, Les Belles Lettres). Kahane, Henry and Renée, and Andreas Tietze, 1958 The Lingua Franca in the Levant. Turkish Nautical Terms of Italian and Greek Origin (Urbana, 111., University of Illinois Press). Kayser, Wolfgang, 1957 Das Groteske, seine Gestaltung in Malerei und Dichtung (Oldenburg and Hamburg, G. Stallin). (Translated: The Grotesque in Art and Literature [Bloomington, Indiana University Press, 1963].) Knaak, P., 1913 Über den Gebrauch des Wortes "grotesque" (Diss. Greifswald). Kraus, W„ 1937 "Zur Bedeutungsgeschichte von Romanesque im 17. Jahrhundert", ZFSL 61: 297-320. Lapa, M. Rodrigues, 1930 LP 1:308. Lapesa, R., 1959 Historia de la lengua española, 4th ed. (Madrid, Escelicer). Lida, Raimundo, 1967 "Notas al casticismo de Rubén [Darío]", Rev. Ib. 33: 333-58. Malkiel, Y., 1950 "The Latin Background of the Spanish Suffix '-uno'", RPh 4: 17-45. 1951 "Graeco-Latin 'Iûdaeus' and 'Iüdaicus' in the Romance Languages, with Special Reference to Ibero-Romance", UCPSPh 11: 327-338 (Festschrift William Popper). 1957 "The Semantic Link between Latin 'bi(s)-' and Romance *bes-\ 'bis-'", Studies Presented to Joshua Whatmough on his Sixtieth Birthday (The Hague, Mouton). 1958 "Old Spanish 'judezno', 'morezno', 'pecadezno' ", PhQ 37: 95-99. 1959 "Nuevas aportaciones para el estudio del sufijo '-uno'", NRFH 13: 241-90. Malkiel, Y., and Karl D. Uitti, 1968 "L'ancien français 'gab-ois', 'ir-ois', 'jargon-ois' et leurs contreparties dans l'anglais d'Amérique", RLiR 32: 126-74. Marouzeau, J., 1958 "Note sur la valeur du suffixe '-esque"', FM 26: 1-2. Matoré, G., 1946 "En marge de Th. Gautier", Études romanes dédiées à Mario Roques par ses amis, collègues et élèves de France (Paris, Droz), pp. 217-39. (Cf. the assessment by J. Jud, in VR 11: 257 [1950]). Meyer-Lübke, W., 1894 Grammatik der romanischen Sprachen, vol. 2 ( = Romanische Formenlehre) (Leipzig, Fues). Migliorini, Bruno, 1939 "Note sugli aggettivi derivati da sostantivi", Mélanges de linguistique offerts à Charles Bally (Genève, Georg et Cie), pp. 251-62.

THE PAN-EUROPEAN SUFFIX "-ESCO, -ESQUE" IN STRATIGRAPHIC PROJECTION

387

1943 "Sulla tendenza a evitare il cumulo dei suffissi nella formazione degli aggettivi", Sache, Ort und Wort: Jakob Jud zum sechzigsten Geburtstag (Genève, Droz), pp. 442-52. 1950 "Convergences linguistiques en Europe", Synthèses (April), pp. 185-91. 1956 "Le lingue classiche, serbatoio lessicale delle lingue europee moderne", Lingua Nostra 17:33-38. 1963 Saggi sulla lingua del Novecento, revised 3rd ed. ( = Biblioteca di "Lingua Nostra", vol. 1) (originai ed., 1941) (Firenze, Sansoni). Montoto, S., 1948 BRAE 28: 135. Nyrop, K., 1908 Grammaire historique de la langue française, vol. 3: Formation des mots (Copenhague, E. Bojesen). Pascu, Giorge, 1916 Sufixele românesti (Bucureçti, Socec Ampers and Co.). Pérez Galdós, Benito, 1897 Misericordia Pichón, É., 1942 Les principes de la suffixation en français; l'enrichissement lexical dans lefrançais d'aujourd'hui (Paris, d'Artrey). Piel, J.M., 1966 Revision of W. Meyer-Lübke, Historische Grammatik der französischen Sprache, vol. 2: Wortbildungslehre (Heidelberg, C. Winter). Plattner, Ph., 1889 "Personal- und Gentilderivate im Französischen", ZFSL 11: 105-66. Puçcariu, S., 1905 Etymologisches Wörterbuch der rumänischen Sprache, I : Lateinisches Element (Heidelberg, C. Winter). Rohlfs, Gerhard, 1954 Historische Grammatik der italienischen Sprache und ihrer Mundarten, vol. 3 : Syntax und Wortbildung (Bern, Francke). S[ánchez] C[antón], [F.J.] 1942 "Para la historia de 'arabesco' ", Correo erudito 3:55. Sá Nogueira, R. de, LP 1:295. Scheuermeier, P., 1920 "Einige Bezeichnungen für den Begriff 'Höhle' in den romanischen Alpendialekten", ZRPh, Beiheft 69. Silveira, J. da, 1949 RPF 3:46-51. Spitzer, Leo, 1958 Review of W. Kayser, Das Groteske, in Göttingische Gelehrte Anzeigen, CCXII: 95-110. Stern, Charlotte W., 1968 RPh 22:127. Terlingen, J.H., 1943 Los itaÜanismos del español desde la formación del idioma hasta principios del siglo XVII (Amsterdam, North Holland). Tiktin, H., 1905 Rumänisches Elementarbuch (Heidelberg, C. Winter). Wagner, Max L., 1943 "Iberoromanische Wortmiszellen", Sache, Ort und Wort: Jakob Jud zum sechzigsten Geburtstag (Genève, Droz), pp. 544-62. 1957-64 Dizionario etimologico sardo, 3 vols. (Heidelberg, C. Winter). Wartburg, W. von, 1962 FEW 10:454,457-58. Wilkins, Ernest Hatch, and Thomas Goddard Bergin, 1965 A Concordance to the Divine Comedy of Dante Alighieri (Cambridge, Mass., Harvard University Press).

BERTIL MALMBERO

U N E QUESTION DE MÉTHODE ET LA SOLUTION D'UN PROBLÈME CONCRET (ESP.

CRECER-CREZCO)

Plusieurs linguistes ont attiré récemment l'attention sur un parallélisme frappant entre des solutions 'génératives' et des faits diachroniques.1 La linguistique traditionnelle avait trop souvent confondu les deux points de vue, en utilisant par exemple le même terme en parlant de l'alternance morphologique du type all. Sohn - Söhne, suéd. son - söner, angl. man - men ou mouse - mice et de l'évolution /o/ + pal. > ¡0¡ ou /a/ + pal. > /e/, etc. et en y voyant, plus ou moins inconsciemment et sans le dire en termes explicites, le même phénomène. En réalité, le reflet des mutations phonétiques dans les faits d'alternance synchroniques est naturel et ne pose pas de problèmes autres que ceux qui sont dus à la maladresse du descriptiviste. D'autre part, il est évident que les deux séries de faits sont en principe entièrement différentes et qu'il serait vain de chercher une correspondance parfaite entre l'une et l'autre. J'ai attiré récemment l'attention sur une solution 'générative' proposée par Sol Saporta (article cité) de la formation de la première personne du singulier et de tout le présent du subjonctif des verbes espagnols en -acer, -ecer et -ocer (aussi -ucir) dans les régions dites du seseo2. En espagnol castillan, les formes en question peuvent être engendrées à l'aide de la règle suivante: ont /-ko/, /-ka/ etc. les verbes dont le radical se termine en /Θ/ (ex. crecer - crezco, crezca)·, tous les autres ont /o/, /a/ (ex. coser coso, etc.). Une liste d'exceptions d'une longueur raisonnable est à ajouter {cocer cuezo, hacer - hago). Comment arriver à trouver une règle qui permette d'engendrer les formes en cause dans les régions qui ne connaissent pas l'opposition /s/ ~ /Θ/? Sol Saporta se tire d'affaire en supposant aussi pour les régions du seseo un phonème /Θ/ non manifesté (la règle

-*• [s] l'indique), donc un système consonantique qui est identique sur ce

point à celui du castillan. La description 'générative' fait donc revivre une distinction perdue et un système plus archaïque.3 La seule justification de l'introduction du I Voir par exemple Bengt Sigurd, 1966, Nils Erik Enkvist, 1966, et Sol Saporta, 1965. ' En principe l'Espagne du Sud et toute l'Amérique latine. Voir Bertil Malmberg, 1950: 172-77. II n'y pas de différence, au point de vue qui nous intéresse ici, entre 'seseo' et 'zezeo'. ' Je fais abstraction ici du fait que, historiquement, l'opposition moderne /Θ/ — /s/ reflète une opposition plus ancienne /ts/ ~ /s/ (et aussi /dz/ ~ /z/) et que. dans les régions du seseo, ça a dû être cette

390

BERTIL MALMBERG

phonème /θ/ dans les dialectes du seseo est-le désir de pouvoir donner une règle valables des formes en cause, une règle que l'auteur préfère à une longue liste d'exceptions. Je souligne que, pour un Andalous ou un Sudaméricain, des verbes comme coser, crecer se terminent par les mêmes phonèmes : /-ser/. J'ai discuté cette solution de Sol Saporta dans un article à paraître dans les mélanges von Wartburg à propos d'une discussion du problème général de savoir si, oui ou non, il peut être légitime de recourir à des faits morphologiques pour la solution de questions de phonologie. Même si je suis en principe en faveur d'une analyse linguistique qui tient séparés les deux niveaux du contenu et de l'expression (principe saussurien développé par la glossématique) et qui ne tient compte du rapport entre les deux que dans la commutation, j'ai utilisé des faits d'alternance morphologique pour interpréter même les cas les plus extrêmes de modification du -î implosif espagnol (en Andalousie et en Amérique latine), pas seulement le passage très général à une spirante laryngale [h] ou [ h ] mais aussi des modifications de timbre et des allongements, comme des réalisations du phonème /s/ (ex. [-0] = /os/, [a:] = /as/). Les cas soulevés par Sol Saporta sont différents. J'ai souligné ailleurs4 que je regarde comme inadmissible toute solution phonologique qui implique l'introduction dans le système d'une unité fonctionnelle qui ne connaît pas d'autre justification que celle créée par l'interprétation. C'est le cas ici. La seule raison d'être du phonème /Θ/ en espagnol américain est le désir de donner une règle à la place d'une liste pour les verbes qui insèrent un /k/ dans certaines formes. En réalité cette ambition du transformationaliste peut être satisfaite d'une façon beaucoup plus simple et qui cadre mieux avec des faits vérifiables si l'on suppose pour les verbes orthographiés -ecer un radical /-esk/ (donc */kresk-er/, etc.) à partir duquel il sera possible d'engendrer toutes les formes à l'aide de la règle que -sk-/V pal. -* -s- (donc/kre'ser/, /'krese(s)/, /kre'sia/, /kre'sido/, etc. mais /'kresko/, /'kreska/). Il est facile d'entrevoir à traver cette règle des faits évolutifs connus. La vélaire du radical latin /kresk-/ s'est conservée devant /a/ et /o/ mais s'est palatalisée devant /e/ et /i/, devenant en ancien espagnol /ts/, d'où en castillan, à partir du XVIIe siècle, /Θ/. Les dialectes du seseo ont eu /s/ partout. Dans un dictionnaire phonologique il faudrait donc mettre comme radical /kresk-/ ou /kreGk-/ respectivement, tandis que pour coser le radical sera /kos-/. Le parallélisme entre la description transformative et la description diachronique est frappant. Il n'est pas impossible que ce soit à partir de la conscience d'un radical /kresk-/ que le sujet parlant, même sans l'appui du /Θ/ castillan et de l'orthographe, arrive à engendrer correctement les formes en /-sk-/. opposition qui s'est perdue directement sans passer jamais par l'étape /Θ/ ~ /s/. Ceci n'influe pas sur la manière de résoudre notre problème, aussi peu, du reste, que le fait connu que le /Θ/ de crezco, crezca est analogique d'après crecer, etc. et qu'une forme étymologiquement 'correcte' aurai un /s/ (on connaît l'hésitation dans l'ancienne langue entre conosco et conozco, à laquelle a dû contribuer aussi la faiblesse de l'opposition en fin de syllable; voir V. Garcia de Diego, 1957:139, 203. « Voir par exemple Bertil Malmberg, 1967: 84-85.

UNE QUESTION DE MÉTHODE

391

De toute façon, notre interprétation a le grand avantage, par rapport à celle de Sol Saporta, d'être basée sur des faits concrètement vérifiés. La sienne implique un procédé qui à mon avis est inadmissible.® Mon objection contre la réintroduction du phonème /λ/ dans les systèmes qui l'ont perdu ('yeísmo') est en principe la même. Il me paraît inadmissible de supposer un phonème /λ/ seulement pour pouvoir rendre compte des alternances aquel - aquellos (IV - /j/)> ley - leyes (/j/ - /j/; je fais abstraction des différences de réalisation phonétique qui existent entre les différentes régions et qui sont parfois considérables (cf. Malmberg, 1967:104-12; 150-56) et papel - papeles. Cette règle /λ/ [1] / - # reflète une règle plus générale, celle qui interdit les palatales (autres que /jf) à la finale en espagnol (et qui est aussi responsable des formes syncopées telles que val, cal de valle, calle). Le pronom aquel y est parfaitement parallèle et n'implique aucune difficulté en castillan. On part dans le lexique d'un radical /a'keX-/ d'où aquel est engendré à l'aide de la règle ci-dessus. Le problème pour Sol Saporta est de savoir si (dans les régions du 'yeísmo') un radical en /j-/ (se. /a'kej-/ et /lej-/ respectivement) doit garder [j] à la finale absolue ou le transformer un [1] (d'après le modèle /a'kej-/ -* /a'kel/ ou selon /lej-/ -* /lej/?). C'est cette hésitation qui, dans les régions du seseo, explique la création sporadique des formes avec /-j/ à la place de /-!/, telles que ey pour él dans les textes populaires argentins; selon l'équation /'lejes/ - /lej/ = /'eja/ - x, où χ = /j/ (orth. -y). Voir Malmberg, 1950:155.

RÉFÉRENCES de Diego, V.G., 1957 Gramática histórica española (Madrid) pp. 139, 203. Enkvist, N.E., 1966 "Tre modeller för Ijudhistorisk forkning", Societas Scientiarum Fennica 44B. 4. Malmberg, B., 1950 Études sur la phonétique de l'espagnol parlé en Argentine (Lund) pp. 172-77. 1967 Structural Linguistics and Human Communications, 2e ed., pp. 84-85. Saporta, S., 1965 "Ordered Rules, Dialect Differences, and Historical Processes", Language 41. 2. Sigurd, B., 1966 "Generative Grammar and Historical Linguistics", Acta Linguistica Hafiiiensia 10. 1.

5

La règle peut être utilisée aussi pour engendrer les formes écrites des verbes: sont orthographiées '-C-' (crecer, crece, etc.) celles qui sont derivées de la base /-sk-/; les autres ont '-s-' (coser, cose, etc.). Ma remarque ne concerne pas les autres interprétations 'génératives' données par Sol Saporta dans l'article cité. Je ne vois pas pour le moment comment ont pourra formuler une règle qui permette d'engendrer dans les régions du seseo et sans tenir compte de l'orthographe, des pluriels comme ¡ápices 'crayons' /'lapises/ de lapiz en face de lunes 'lundis' de lunes. Pourtant, le dernier type n'est pas très fréquent, et aucune règle ne pourra tenir compte de tous les cas qui existent.

ANDRÉ MARTINET

LA NATURE PHONOLOGIQUE D'E CADUC

Le 3 mars 1968, vers 8 heures 10, Robert Villard, annonceur à Europe no. 1, présente sur les ondes le nouveau produit Jexnet [3eksnst]. Il éprouve, de toute évidence, une certaine difficulté à prononcer ce complexe; il se reprend, marquant un temps d'arrêt après [3sks], et s'excuse en arguant de la difficulté que présente la prononciation d'un tel vocable. Cet épisode rappelle opportunément que la fameuse 'loi des trois consonnes' n'a pas cessé d'être valable, même pour un annonceur de radiodiffusion que sa profession amène à prononcer, à longueur d'émission, les noms anglais, ou de consonance anglaise, des produits les plus divers. Ce même annonceur aurait sans doute articulé sans encombre des groupes consonantiques plus 'lourds' [-kspl-], [-kstr-] dans des mots connus comme expliquer ou fox-trot. Mais la rencontre avec une forme inattendue revèle qu'un groupe consonantique comme [-ksn-] peut poser un problème pour n'importe quel francophone. On a beaucoup écrit au sujet de la 'loi des trois consonnes', surtout d'ailleurs pour signaler les exceptions à la règle.1 On a pu relever que tel groupe, [-str-] par exemple, ne faisait difficulté pour personne. Mais ceci veut simplement dire que la 'loi' reste valable à condition de rendre sa formulation plus explicite. On a relevé qu'elle ne s'appliquait pas nécessairement au passage d'un mot à un autre, voire d'un premier élément de composé à un second (gard'-malade), ce qui s'explique soit par la possibilité effective de marquer une légère pause, soit par le sentiment confus d'une virtualité de pause qui justifie toutes les audaces : notre annonceur n'aurait pas marqué la moindre hésitation, le moindre arrêt, s'il s'était agi de dire: Jex nettoie mieux, avec le même groupe [-ksn-] que dans Jexnet. On a par ailleurs relevé que ce qui valait pour le synthème garde-malade [gaxdmalad], avec son deuxième élément disyllabique, ne valait pas pour garde-chasse [gairdajas].2 Mais tout ceci ne fait que préciser les conditions d'existence et d'apparition d'e caduc, en fonction du contexte phonique

1 Voir, par exemple, P. R. Léon, 1966: 118. * Ce que nous notons [s] n'est pas nécessairement une voyelle centrale sans arrondissement ni rétraction, des lèvres, mais toute réalisation d'e caduc qui ne se confond pas totalement avec les réalisations normales, dans la position considérée, d'un phonème vocalique antérieur arrondi.

394

ANDRÉ MARTINET

et des pauses réalisables ou réalisées. Cette masse d'observations très utiles ne repose pas le problème du statut phonologique de cet e caduc. L'essentiel, à cet égard, est ce qui a été énoncé, dès 1933, par Josef Vachek (1933): Ve caduc n'a pas de fonction phonologique. Son apparition est prévisible puisque déterminée par la 'loi des trois consonnes'; il ne correspond donc pas à un choix du locuteur; un mot français comme revenir se transcrira donc /rvnir/, la prononciation [κ3νηίκ] étant celle qui s'impose après consonne finale de l'élément précédent, dans pour revenir, par exemple, [κνοηϊκ] apparaissant automatiquement après voyelle, comme dans à revenir. La formulation peut-être la plus suggestive est celle selon laquelle l'e caduc caractérise une variante de chacun des phonèmes consonantiques du français: soit le phonème /d/; il se prononce [d] devant un phonème vocalique, dans dans, dors, dîne-, il se prononce également [d] après voyelle et devant consonne unique, comme dans là-dedans [laddô]; mais il se réalise comme [da] entre consonnes, dans par devant [paifdavà] par exemple. Gilè Vaudelin, qui nous a donné une belle description de la phonologie du français en 1700, n'avait pas à sa disposition les concepts nécessaires pour présenter les choses dans les termes où nous le faisons ici. Mais notre formulation résume exactement ses constatations.® A l'appui de la conception de Vaudelin, on rappellera qu'aucun mot français ne commence par un e muet. Une prononciation [alagar)], pour élégant, qu'on prête aux gens du Midi à la scène et à l'écran, est, en fait un [œlœgaq], avec, à l'initiale,, le phonème vocalique antérieur-arrondi-d'aperture moyenne des usages méridionaux du français. Lorsqu'on examine les conditions d'emploi des variantes vocalisées des consonnes françaises, on arrive à la conclusion, celle qui résume la 'loi des trois consonnes', que la fonction de Ve caduc est celle d'un lubrifiant phonique. L'inexistence phonologique d'e caduc est bien ce qu'on doit enseigner à tout étranger pour lui faire comprendre le comportement le plus naturel des usagers parisiens du français.4 Ceci ne veut pas dire qu'on le mettra nécessairement, dès l'abord, devant des transcriptions phonologiques comme /rvnir/ ou /3mldmäd/. La transcription utilisée dans l'enseignement des langues doit toujours représenter un compromis entre la phonologie de la langue enseignée et les habitudes articulatoires, phonématiques et graphiques de l'étudiant. Mais il faudra bien faire sentir à ce dernier que Ve caduc n'est pas une voyelle comme les autres, que son rôle est de soutien et non d'information, qu'il pourra, dans le corps des mots, en risquer quelques-uns qu'on n'entendrait guère à Paris (dans un débit lent, [puinravanÎK] est parfaitement acceptable), mais qu'il ne doit guère compter sur cet e pour identifier ce qui lui est dit. Une fois acquis le caractère généralement non distinctif de Ve caduc, il reste à voir '

Cf. Marcel Cohen, 1946, et André Martinet, 1946. Les Parisiens ne sont pas, en l'occurrence, les seuls en cause. Mais nous ne saurions, ici, chercher à déterminer quelle extension les faits décrits ont dans l'espace et la société. Par ailleurs, il y a à Paris bien des gens, d'origines diverses, dont le système phonologique n'est pas conforme à celui qui fait l'objet de notre examen. 4

LA NATURE PHONOLOGIQUE D'E CADUC

395

quelles sont les circonstances qui s'opposent à une élimination intégrale et définitive de l'e caduc du système phonologique du français contemporain. Parmi ces circonstances, on trouve, en tout premier lieu, la graphie traditionnelle et la diction poétique classique qui, pour l'essentiel, se calque sur elle. Il est vrai qu'un usager du français qui n'est pas sur ses gardes insérera un [a] entre le [k] et le [b] du groupe [-Kkb-] et dira arcque-boutant pour arc-boutant ; mais alors que tout le monde prononce arquebuse avec un e caduc, il y a bien des gens qui prononcent arcboutant avec trois consonnes successives sans effort apparent. Si l'on rapproche arc-boutant de la prononciation gard'malade citée ci-dessus, on pourra considérer la prononciation sans [a] de ce mot comme résultant du sentiment de l'existence d'une pause virtuelle entre les deux éléments du composé, alors qu'il ne saurait y avoir rien de tel dans arquebuse. Mais, il n'est pas niable que des sujets qui, en toute innocence, prononçaient [aifkabutô], sont parvenus à dire [airkbutä] dès qu'on leur a rappelé l'absence d'e dans la graphie du mot. Il est certain que le sentiment latent de différences dans le degré de cohésion des monèmes en présence joue ici un grand rôle: l'auteur de ces lignes prononce naturellement [uKsablâ] pour ours blanc, équivalent de l'anglais polar bear ; mais il désignerait certainement comme un [UKS bla] un animal en peluche blanche ayant l'aspect d'un ours de nos climats. Il n'en est pas moins vrai que les e de la graphie ne sont pas sans influence sur la prononciation la plus naturelle de bien des gens. Il y a donc, chez ceux-ci, la possibilité de distinguer entre deux unités de sens du fait de la présence ou de l'absence de [a] en un point de la chaîne phonique. On hésitera, toutefois, à attacher trop d'importance à des oppositions qui, en tout état de cause, ne vaudraient pas pour l'ensemble des francophones parisiens. Il est une situation particulière où une distinction traditionnelle pourrait très naturellement être interprétée comme résultant d'une opposition d'e caduc à son absence. C'est celle où des 1ère et 2ème personnes du pluriel de conditionnels de verbes du premier groupe s'opposent aux mêmes formes de verbes à l'infinitif en -re. Soit, par exemple, nous fonderions et nous fondrions. L'auteur de ces lignes, et il ne doit pas être seul dans son cas, prononce [födsifjö] d'une part, [födirijö] d'autre part. On pourrait, certes, transcrire phonologiquement /födriö/ et /födrijö/ et voir dans le /-dri-/ de la première forme une succession de trois articulations non syllabiques allégées par l'insertion d'un [a] entre /d/ et /r/ (ou, si l'on veut, par l'utilisation dans ce cas de la variante [da] de /d/). Toutefois, comme le groupe [-dri-] suivi de voyelle développe automatiquement un [j] de liaison, on pourrait préférer transcrire le second terme de l'opposition /födriö/ ce qui nous contraindrait à représenter le premier comme /födariö/. 5 Nous aurions donc ici une paire de quasi-homonymes établissant la pertinence de l'opposition d'e caduc à son absence. Mais, de nouveau, la distinction des deux formes est loin d'être universelle. Dans la bouche de deux professeurs de français à la Sorbonne, parlant publiquement dans une commission ministérielle, on a * Dans notre interprétation de la phonologie du français, [j] et [i] sont, entre consonne et voyelle, des variantes d'une même unité; /j/ ne s'oppose à /i/ qu'après voyelle (voir Martinet, 1946).

396

ANDRÉ MARTINET

relevé à plusieurs reprises demanderions réalisé constamment comme [dmädirijö] avec la syllabation qu'on attendrait dans vendrions ; ceci laisse croire que l'opposition entre [födoirjö] et [fôdirijô] est loin d'être générale, et témoigne du fait que les traces d'opposition entre e caduc et son absence sont très généralement instables. Le statut phonologique d'e caduc à la finale ne pose guère de problème : à Paris, coq et coque sont de parfaits homonymes; peu importe ce qui se passe à la finale de peuple ou de chambre, puisqu'il n'y a pas, dans ce cas, possibilité d'utiliser à des fins distinctives deux types différents de vocalisation.® Aucun problème non plus à l'intérieur du mot entre deux syllabes à voyelle stable: à l'initiale près, calepin se prononce comme alpin, et âpreté comportera nécessairement le lubrifiant entre [-pr-] et [-t-]. Mais la situation n'est pas aussi claire là où l'e caduc est, s'il se prononce, la première voyelle du mot. Soit le mot meringue. On aura régulièrement [lamirœg] sans [a], [ynmoiràëg] avec [a] ; à supposer que le mot apparaisse à l'initiale absolue, dans une énumération par exemple, on entendrait probablement [moirâîg] ; mais comme il n'existe pas, en français, de mot commençant constamment par [mir...] qui pourrait entrer en opposition avec [msKàëg] à l'initiale absolue, une transcription phonologique /mrëg/ est parfaitement correcte. Ce qui est dit ici du groupe initial /mr-/ vaut généralement pour les groupes consonantiques initiaux résultant de la disparition, dans la prononciation, d'un e de la graphie devant /r/, dans cerise ([sk]) par exemple. Ces groupes ne semblent que rarement comporter de première consonne qui se combine traditionnellement avec un /r/ suivant: on a donc [m(o)K-], [s(s)k-], etc. d'une part, [bK-], [tK-] d'autre part; cf. toutefois querelle. Le cas de vrai (cf. angl. very, d'un lat. uërâcum) suggère que l'absence de [b(o)K-], [ί(ο)κ-] résulte de réductions très anciennes sanctionnées par la graphie traditionnelle. Il n'y a donc ici guère d'oppositions entre les groupes où [a] peut apparaître entre les deux consonnes et ceux où ceci n'est pas le cas : /mr-/ sera [mK-] ou [maK-] selon les contextes, /br-/ sera toujours [bK-], jamais [boK-].

Le problème se pose autrement avec les groupes à second élément [1]. A côté des combinaisons nouvelles résultant de la chute de e, dans cela [sia] par exemple, d'emprunts comme chelem ou de formations argotiques du type schlass, on trouve ici des groupes instables correspondant à des graphies avec e entre les consonnes, dans pelote par exemple, en face des groupes traditionnels stables de plat, blanc, claie, qui présentent les mêmes consonnes. Dans le vocabulaire le plus ordinaire, on ne constate guère de distinction entre les deux types lorsqu'il s'agit de formes qui ont assez de corps: peloton, peloter ne se prononcent pas autrement que [piotò], [piote]; ce sont des mots employés généralement dans le discours bien avant qu'on les trouve dans les textes écrits, et leur graphie en pel- étonne un peu à la première rencontre. Toutefois, dans des mots plus courts comme pelote et peluche, il semblerait que Y e n'ait pas disparu de toutes les prononciations. Il est vraisemblable qu'on ne confond • Cf. André Martinet, 1945: 56.

LA NATURE PHONOLOGIQTJE D'E CADUC

397

pas totalement en pelant et en plan. Quant aux noms propres Blin et Belin, qui au départ sont deux variantes d'une même forme, on les distingue sans difficulté aujourd' hui. D'autres mots français présentent, dans la graphie, un e après consonne finale et devant consonne unique autre que /l/ ou /r/ ou devant groupe traditionnel consonne plus /!/ ou /r/, par exemple semelle, secrétaire, penaud, guenon, jeton, cheval, retard. A l'initiale absolue, la première consonne est normalement séparée de la seconde par [a]; en principe, des groupes de consonnes comme [-gn-], [-JV-], [-Kt-] ne sont prononçables que si la première consonne s'appuie sur une voyelle précédente; dans je me dis fôamdi] est plus normal et fréquent que fômadi]. Ailleurs, on s'attend à ce que joue la "loi des trois consonnes", donc [fc>3tô] pour faux-jeton, [ajval] pour à cheval, [airtaïf] pour en retard. Mais certaines combinaisons sont généralement évitées : tout penaud se prononce [tupano] et la guenon [Iaganô], peut-être en réaction contre une assimilation du type [lagnò] qui rendrait le mot trop différent du [ganô] d'une guenon. Rien de tout ceci n'affecterait le statut de lubrifiant d'e caduc, si les combinaisons de consonnes en cause résultaient toujours de son élimination. Mais des emprunts à diverses langues ont introduit ces mêmes combinaisons sous une forme stable ou que la graphie suppose telle. On a ainsi gnome, gneiss avec [gn-] en face de guenon, guenille où l'on entend normalement [gan-], et pneumatique, pneumonie avec [pn-] qui s'oppose au [pan-] de penaud. On notera toutefois que les mots empruntés font souvent difficulté: pour pneu, par exemple, on entend souvent [pano]. En résumé, on peut dire que, n'était la pression permanente exercée par la graphie, les locuteurs parisiens, dans les contextes signalés jusqu'ici, n'emploieraient jamais l'e caduc à des fins distinctives. L'action combinée de la graphie et de la diction poétique traditionnelle n'est arrivée à fixer définitivement un emploi distinctif de l'e caduc que dans des cas très particuliers comme celui des noms propres Blin et Belin mentionnés ci-dessus. Il est cependant une situation particulière que nous n'avons pas envisagée jusqu'ici, où l'opposition de [a] à son absence semble se maintenir assez généralement chez les francophones adultes. Il s'agit de ce qui se passe devant ce qu'on appelle l'A aspiré. Chez les locuteurs parisiens, l'A dit aspiré n'a aucune existence phonique. Il ne correspond nullement à un coup de glotte, comme on a pu parfois le lire. Phonologiquement, il ne se manifeste par aucun trait particulier lorsque l'élément 'commençant par h aspiré' est immédiatement précédé par un élément qui se termine par une voyelle stable: dans la hache, la succession /...aa.../ ne diffère en rien de celle qu'on trouve dans à Avignon·, dans un hêtre la succession [...œe...] est celle qu'on entend dans Γ un est... ; le traitement [la] de l'article féminin, au lieu de [1], dans l'hirondelle par exemple, est un fait de morphologie pure et simple. Là où se pose un problème particulier, c'est lorsque 1"A aspiré' est précédé d'une consonne, le /l/ de l'article défini masculin ou le /n/ de l'article indéfini féminin une. Dans ce cas on serait tenté de dire que la consonne se réalise sous la forme de la variante à e caduc. Toutefois cette formulation ne serait licite que si le Ά aspiré' correspondait à une réalité phonologi-

398

ANDRÉ ΜΑΚΉΝΕΤ

que. C'est ce que je n'avais pas hésité à postuler dans le premier article consacré à la phonologie du français (Martinet, 1933). Cette attribution d'une réalité phonologique à une entité phoniquement inexistante est à la source de la théorie de la latence qu'a développée plus tard Louis Hjelmslev (1953:59-60). Je ne pense plus qu'elle soit admissible en phonologie. Il nous faut donc reconnaître que les oppositions le hêtre ~ l'être /laetr/ ~ /letr/, une hache ~ une ache /ynaaj"/ ~ /ynaJ/ se fondent sur celle de [a] à son absence, ce qui veut dire que, dans certaines conditions, hêtre se distingue de être et hache de ache par un phonème initial supplémentaire /a/. On peut naturellement noter ce phonème au moyen de /h/, mais il ne devra être marqué que là où il existe effectivement, donc, par exemple, dans /ynhaj/ une hache, mais non dans /laajy la hache. On comprend qu'en pratique, on préfère parler à"h aspiré', ce qui permet de grouper sous une même rubrique l'absence d'élision dans la hache, l'absence de liaison dans les haches, faits morphologiques, et l'apparition de /a/, fait phonologique, dans une hache. Mais une analyse sérieuse se doit de mettre les deux traits dans deux chapitres différents et de poser ici une opposition /a/ ~ zéro, même si l'on ajoute immédiatement qu'elle est normalement neutralisée presque partout ailleurs. Les difficultés qu'éprouvent les enfants et certains adultes à réaliser comme [daoïr] et non [doir], [deoK], [dejoir] le monème unique dehors, marquent bien les limites normales du phénomène. Restent les cas très particuliers où e caduc se trouve en position prépausale, c'est-àdire, comme on le dit parfois, 'sous l'accent'. Il s'agit de la prononciation du pronom le comme objet d'un impératif, dans bois-le, arrêtez-le par exemple, de celle de ce dans le désinvolte sur ce ou de la lettre e, de ce qui se passe lorsqu'on traite, en 'métalangue', d'un mot 'atone' comme je, me, te, se, ou qu'on hésite sur la conjonction que. Pour les locuteurs, et ils sont nombreux, qui emploient dans tous ces cas la voyelle de feu, le problème est résolu puisqu'il ne s'agit plus d'e caduc, mais d'un phonème /»/ bien caractérisé. Pour ceux qui n'identifient, dans ce cas, la voyelle à aucun des phonèmes qu'ils utilisent à la finale absolue, on pourrait, bien entendu, poser la règle qu'en contexte prépausal, un phonème consonantique qui est le constituant unique d'un monème particulier se réalise avec son support vocalique. On transcrirait dans ce cas bois-le comme /bua 1/ en marquant par un espace la frontière entre les deux éléments signifiants. Il faudrait cependant prendre garde de ne pas identifier, dans les transcriptions, le /I/ prononcé [la] de bois-le tout /bua 1 tu/ et le /l/ pronconé [1] de bois le tout, et donc transcrire ce dernier /bua ltu/ ce qui serait, il faut le dire, passablement artificiel. Doit-on cependant poser qu'il y a ici pertinence de l'opposition /a/ ~ zéro et transcrire /buala tu/ d'une part /bua 1 tu/ d'autre part? Il faudrait, avant de trancher, être sûr que nos sujets distinguent effectivement entre bois-le tout et bois le tout indépendamment de tout emploi de la pause. Il semble, en fait, que l'un et l'autre syntagmes se confondent en [bwaltu] en prononciation naturelle, ce qui est d'autant moins gênant qu'ils ne diffèrent guère par le sens. Pratiquement, on peut, je crois, recommander une transcription phonologique où l'on ne noterait pas le produit de la neutralisation de l'opposition /a/ ~ zéro, où donc

LA NATURE PHONOLOGIQUE D'E CADUC

399

on ne trouverait de /a/ qu'à l'initiale des mots à 'h aspiré' et, de façon générale, dans les contextes où Ve caduc, première voyelle d'un mot polysyllabique, semble pouvoir s'opposer à son absence; ceci permettrait de distinguer entre les initiales de /pano/ penaud et /pn0/ pneu, /pal ä/ pelant et /plä/ plan, /bale/ Belin et /blë/ Βlin1. Ailleurs, tout emploi de /a/ ne peut que donner une idée fausse de la nature réelle du fonctionnement de la langue: le futur d'accoupler se forme régulièrement en ajoutant /-ra/ au radical /kupl-/, même si l'on entend en fait [kuplaira]. On ne devra pas hésiter devant des graphies sans [a] comme /3mldmäd/, /mrëg/, /Jval/, /rtar/, /arkbutä/ et /ursblä/ qui peut correspondre à la variante [uKsablä], en face d'/urs blä/ toujours sans [a]. Dégager le système phonologique d'un parler n'est jamais chose simple. Mais lorsqu'on a affaire à une langue parlée par des millions de sujets appartenant à des classes, à des provinces, voire à des nations différentes, une langue qui est écrite depuis des siècles, dont l'orthographe n'a que des rapports capricieux avec la prononciation contemporaine et qui est maniée par des gens qui, dans leur grande majorité, savent lire et tendent à modeler leur prononciation sur la graphie, il est presque impossible d'arriver à des conclusions tranchées même si l'on a pris la précaution de se limiter à l'usage d'un secteur limité de la communauté linguistique, voire à un idiolecte. A ne présenter qu'une liste de phonèmes sous la forme de symboles graphiques, on donnera nécessairement une idée inexacte du comportement phonologique réel des usagers. A côté d'un noyau central d'unités parfaitement discrètes, il y a partout un nombre plus ou moins considérable d'éléments marginaux qu'il ne convient ni d'écarter dédaigneusement ni d'identifier aux traits réellement constants et proprement constitutifs de l'idiome. SORBONNE

RÉFÉRENCES Cohen, M., 1946 Le français parlé en 1700 d'après le témoignage de Gilè Vaudelin (Paris). Christensen, B.W., 1967 "Phonèmes et graphèmes en français", Acta Linguistica Hafniensia 10: 217-40. Hjelmslev, L., 1963 Prolegomena to a Theory of Language (Baltimore, Linguistic Society of America). Léon, P. R., 1966 "Apparition, maintien et chute du e caduc", La Linguistique 2: 111-22. Martinet, Α., 1933 "Remarques sur le système phonologique du français", BSLP 34: 191-202. 1945 La prononciation du français contemporain (Paris; 2ème éd. Genève, 1971). 1946 "Note sur la phonologie du français vers 1700", BSLP 43: 13-23. Vachek, J„ 1933 Über die phonologische Interpretation der Diphthonge (Prague). ' Bjarne Westring Christensen (1967 : 223) a raison de noter /pala/ pour pelant ·, mais le /a/ qu'il emploie dans /davä/ ou /premie/ ne s'oppose jamais à son absence.

KENNETH L. MOLL

CINEFLUOROGRAPHIC STUDIES OF SPEECH ARTICULATION

For the past 15 years, the technique of cineradiography has been utilized for the study of speech articulation.1 Early users of this procedure emphasized its potential for providing direct information concerning movements of the articulatory structures during connected speech (Cooper and Hofmann, 1955; Bjork, 1961; Moll, 1960). Undoubtedly, cineradiography has contributed importantly to our knowledge of speech production by providing data on velar activity (Bjork, 1961; Moll, 1962), articulation of consonants and vowels (Truby, 1959; Lindblom, 1964; Perkell, 1965; Heinz and Stevens, 1965), articulatory differences between various languages (Delattre, 1965), and coarticulation phenomena (öhman, 1966; MacNeilage and DeClerk, 1967). Although considerable data have been obtained, it appears to this writer that the use of cineradiography has not been as extensive as might have been predicted and that the potential for utilization of this procedure in speech research is largely untapped. If this viewpoint is valid, it probably relates directly to certain limitations of cineradiography which have yet to be overcome. Although these limitations have been described previously (Subtelny et al., 1957; Moll, 1960, 1965), I would like here to evaluate their present status and to discuss possibilities for their solution. In addition, a discussion will be presented concerning some of the basic issues involved in the design of speech articulation research.

1. PROCEDURAL ASPECTS

The most obvious limitation of cineradiography is that observations are restricted, for the most part, to a lateral, two-dimensional view of the speech articulatory structures. Attempts have been made to obtain frontal views (Astley, 1958) and to use rotational techniques (Massengill, et al., 1966); however, such procedures have not received extensive use. 1

It should be recognized that this paper was written in March, 1968 and does not reflect more recent research in this area.

402

KENNETH L. MOLL

Even if one accepts a lateral, two-dimensional view as being adequate for the investigation of a great number of speech phenomena, problems still remain in achieving adequate definition of the articulatory structures for accurate visualization. The quality of structural definition is dependent, of course, on such factors as film type, film processing and printing procedures, and radiographic technique. These factors have been discussed previously (Moll, 1960) and will not be considered in detail here. The enhancement of structural definition by radiopaque marking procedures does warrant some consideration. Most cineradiographic speech research has involved the marking of structural outlines with a radiopaque paste, usually a barium-sulfate solution. A swab generally is used to coat the midline of the tongue and the solution may be inserted through the nose to mark the velar and pharyngeal structures. Such procedures present certain problems. The marking substances tend to disperse over the entire structural surface and spread to adjacent structures. This means that one cannot be certain that the marking he sees is at the tongue midline or even that it is not on some other structure. For example, in Astley's (1958) study of lateral pharyngeal wall movements, there is no way to ascertain whether the radiopaque marking he visualized actually was on the lateral walls or whether he may have been observing movement of the faucial pillars. In recent years, various methods have been used to mark specific points on the articulatory structures, rather than marking entire structural outlines. Small pieces of radiopaque metal are affixed to the structures at selected points. The most commonly used material for affixing the metal has been Eastman 910 Adhesive (Lifschiz, 1963); however, it appears that the possible side-effects of the intraoral use of this material have not been fully explored. This author has attempted to use an experimental adhesive bandage (Orahesive bandage produced by the Squibb Company) for marking purposes. Although partial success is often attained if care is taken in removing moisture from the structure, this bandage does not adhere well to such organs as the tongue which do not have a firm skeletal base. Houde (1967) utilized small gold cups which were pressed into the tongue surface. These markings remained in place well on the dorsum of the tongue; however, they could not be used on the tongue blade and tip where the tongue structure is less 'spongy'. Since the marking of specific anatomic points has a number of advantages in relation to data reduction and interpretation, further efforts in finding safe, effective adhesives are needed. A procedural issue which has been discussed by numerous authors (Subtelny, et al. 1957; Moll, 1960; Shelton, et al., 1963) involves the filming speed necessary for studies of speech articulation. Obviously, there is no one answer to this issue, since the needed sampling rate depends on the purposes of the individual investigation. Shelton, et al. (1963) concluded that a speed of 24 frames per second was adequate to detect openings of the velopharyngeal port, while Subtelny (1964) estimated, from data on lip movements, that 240 frames per second would be necessary to study small articulatory movements. The procedure used by Daniloff (1967) for assessing required camera speed appears to have advantages. On the basis of a test film, he determined the frame

CINEFLUOROGRAPHIC STUDIES OF SPEECH ARTICULATION

403

rate needed so that the fastest-moving structure studied (the tongue) would show less than three mm movement between adjacent film frames. In his study, this rate was found to be 150 frames per second. Such a speed involves a time uncertainty of seven msec., which is relatively short compared to the duration of most phones. It would appear reasonable for each investigator to apply similar empirical tests to determine necessary film speed, based upon criteria consistent with the phenomenon being studied and with the length of exposure time needed. Although the use of different camera speeds may present some problem in obtaining a closely synchronized acoustic record, this topic will not be discussed here. The reader is referred to papers by DeClerck, et al. (1965) and Cooper and Hofmann (1963) for excellent solutions to this problem. Undoubtedly, the major problems limiting cineradiographic studies of speech lie in the area of data reduction. It should be recognized first that the slow-motion viewing of short film loops, possibly with the speech on the sound track expanded (Cooper, et al., 1963), is an excellent and relatively simple procedure for determining the general articulatory characteristics of speech samples. Eventually, however, most investigators wish to reduce the structural position data to numerical form so that comparisons between various speech samples can be made on a quantitative basis. The basic procedure utilized to achieve this purpose has been the tracing of individual film frames and the extraction of certain measurements from the tracings. The time and effort required for such analysis procedures have presented formidable obstacles to cineradiographic research and have made it necessary to limit the number of speech samples and subjects used. The ideal solution to the film analysis problem undoubtedly would involve some type of optical scanning device which would automatically extract structural outline information directly from the film frames. As yet, devices which will accurately accomplish this task have not been developed. The problem lies in the fact that the signal-to-noise ratio (structural contrast) of cineradiographic films is not great enough to permit accurate scanning. In addition, the film density differences to be detected would need to be varied depending on which articulatory structure was being observed. At this stage, it appears that a human observer is required to identify the structural outlines on the film frames. Advances have been made, however, in development of more automatic methods of recording such identifications and processing the obtained data. The basic procedures involve the expression of structural outlines or specific anatomic points in terms of X-Y coordinates. One method (DeClerk, et al., 1965) involves the aligning of cross-hairs on a point, which is then coded in relation to the coordinate system. The movement of the cross-hairs appears to place severe limits on the speed of analysis and on the freedom of the operator to accurately trace continuous outlines. A device which does not seem to suffer from these disadvantages is one described by Heinz and öhman (1966). In this system, the X-Y position of a tracing stylus is reflected by voltage changes across two potentiometers to which the stylus is mechanically coupled. The voltages are sampled at selected points along the

404

KENNETH L. MOLL

curve. Shaft-angle encoders also can be used to replace the potentiometers in such a system. A method described by Ramsey (1964) for analyzing laryngeal films also is applicable to cineradiographic research. In this method the stylus is attached to a light-pen so that the traced outline also is traced onto the face of a computer-coupled oscilloscope. Although all of these methods generally have been used on-line with digital computers, this does not appear to be necessary. If the sampling logic is built into the X-Y reduction device, the data can be stored on magnetic or paper tape for later off-line computer processing. The expression of structural outlines or anatomic points in terms of X-Y coordinates results in data which can be processed in various ways on a digital computer. For example, distances between structures can be computed and plotted as a function of time and coordinate axes can be rotated as desired. With appropriate output equipment, tracings of the individual frames can be obtained. The effects of such analysis and processing procedures and of anticipated future improvements in the procedures undoubtedly will reduce the time and effort presently required. Eventually, such a reduction should allow the use of longer speech samples and of greater numbers of subjects. Another often-discussed issue in cineradiographic speech research concerns the types of measurements which should be made from the films. Again, the answer to this question depends on the purpose of the particular investigator. The measurement systems which have been utilized can be differentiated into two general classes : those designed primarily to reveal the effect of articulatory movements on the shape of the acoustic transmission tube (Heinz and Stevens, 1965) and those which attempt to express movements of structures or anatomic points in space (Houde, 1967). Obviously these are not independent systems, since the first also provides information about structural movement and the latter can be used to predict the effects of movements on tube size and shape. The differentiation is made here, however, to point up some subtle differences between the two systems. It should be recognized that measures related to changes in the tube configuration may not accurately reflect the direction or the nature of the articulatory movements involved. For example, observed reduction of the distance between the tongue and hard palate at some particular point along the tube does not necessarily reflect an upward movement of the tongue. It could result from tongue movement in an anteroposterior dimension so that a higher point on the tongue is moved to the position at which the measurement of cross-section is being made. The observation also could reflect upward movement of the jaw rather than independent tongue movement. Thus, caution must be exercised in interpreting changes in tract configuration in terms of movements of specific structures. The fact that changes in measures made at some point along the vocal tract do not usually yield information on movement of a specific anatomic point has important implications. Houde (1967) points out that the fitting of mechanical-system response functions to cineradiographic measures may be difficult if the measures involve dif-

CINEFLUOROGRAPfflC STUDIES OF SPEECH ARTICULATION

405

ferent anatomic points at different times. It also is difficult to make inferences concerning underlying muscle activity unless the extent and direction of structural movements are accurately defined. The above discussion is not meant to ignore the fact that measurements related to tube configuration are of primary significance in determining the acoustic changes resulting from articulatory movements. It does not necessarily follow, however, that information not directly related to tract configuration is irrelevant to the study of articulation, öhman (1966) suggests that a model of speech articulation which is expressed in terms of positions and movements of structures may have advantages over those based on vocal tract shape functions. For example, Coker's (1966) model specifies articulatory characteristics of a given speech segment in terms of structural positions. The acoustic effects of changes in positions are then calculated as a second step toward speech synthesis. The primary point to be made concerning cineradiographic measurement systems are that systems designed to describe movements of structures may provide data which are most relevant to theoretical models of speech articulation. This conclusion emphasizes the importance of developing radiopaque marking procedures which allow accurate observations of the movement of specific anatomic points in time. 2. SPEECH ARTICULATION RESEARCH

The procedural problems discussed undoubtedly have limited the use of cineradiography for speech articulation research. This is unfortunate, since investigation of the articulatory processes appears to be of utmost importance. Although considerable research has been carried out recently in this area, available information on articulatory functioning is rather meager. There are little empirical data on the articulatory characteristics of speech sounds and even less on how these characteristics vary as a function of phonetic context, stress, or other factors. Most articulatory phonetic descriptions are based on introspective observations which may be inaccurate. Only a beginning has been made in the study of speech sound coarticulation and of articulatory similarities and differences between languages. More complete data on speech articulation obviously are needed for the physiological description of speech. In addition, such data are essential to an understanding of the manner in which speech production is programmed and controlled. Since any proposed model of speech production must account for observed variations and overlapping of articulatory characteristics, the availability of sufficient empirical data to test such models is important. There is little doubt that observations of coarticulatory phenomena (Kozhevnikov and Chistovich, 1965; öhman, 1966; Daniloff, 1967) have provided impetus to a re-evaluation of traditional concepts concerning the behavioral units in which speech is learned and produced. Yet, even in this instance, the available data are not adequate to provide critical tests of syllable-input or phoneme-input hypotheses.

406

KENNETH L. MOLL

In relation to model-testing, it should be noted that there has been a recent increase in studies of muscle activity associated with speech production. The argument might be made that study at this level provides information closer to the neural control system than do observations of articulatory movements. It would seem, however, that we eventually must describe articulation at all levels of the process and must define the interrelationships which link one level to another. Thus, the study of articulatory movements by cineradiography or other techniques is an essential aspect of a speech research program. There are two primary cautions related to the interpretation of articulatory data which should be emphasized. First, it should not be assumed that all movements observed are due to active muscle contraction or to muscular forces acting directly on the structure moved. Houde's (1967) demonstration that small perturbations in tongue movement are related to the intraoral pressure build-up on stop consonants suggests that some structural movements may be passively induced. The observation of Kozhevnikov and Chistovich (1965) that velar elevation is faster and greater on voiceless stops than on voiced stops may not justify the conclusion that velar muscle activity is different. Since the pharyngeal, velar and laryngeal structures are interconnected, this observation could be due to adjustments in the laryngeal structures on voiced sounds which place a mechanical drag on velar movement. Such examples suggest the need for exercising caution in concluding that articulatory differences reflect a change in the control of that structure at higher physiological levels. The importance of considering other factors, such as mechanical interconnections of structures and air pressures and flows, suggests that simultaneous observations of various phenomena eventually will be required to explain articulatory observations. The second aspect of data interpretation to be discussed involves segmentation of the speech string. Most investigators have used a synchronized audio record to segment utterances and thus to determine the time intervals during which measurements relevant to a given vowel or consonant sound should be made. As a general rule, the consonant-vowel and vowel-consonant transitions have been included as part of the vowel. From an articulatory point of view, however, it might be argued that a phone begins when the prime articulator first begins appropriate movements associated with the phone. For example, it has been demonstrated (Daniloff, 1967) that in a /VCCV/ sequence in which the second vowel is rounded, lip rounding begins simultaneously with the start of articulator movement from the first vowel to the first consonant in the string; that is, with the beginning of the vowel-consonant transition. Whether one considers the transition as part of the vowel or part of the consonant thus becomes important in reaching conclusions concerning the extent of coarticulation in such a string. At higher physiological levels, segmentation procedures may be even more critical. For example, there must be some delay between the beginning of electromyographic activity and the resultant structural movement and acoustic output. As a result, it is doubtful that one should describe the timing of muscle activity on the basis of acoustic segmentation of the speech string. The activity may occur during

CINEFLUOROGRAPHIC STUDIES OF SPEECH ARTICULATION

407

the acoustic segment assigned to a given phone; however, it could be causally related to structural movements associated with a later phone or phones in the sequence. It appears that if one wishes to describe the physiological characteristics of given units, the time intervals assigned to each unit can most validly be determined by observations on the same physiological level at which the descriptions are made.

3. SUMMARY

An attempt has been made to review the status of cineradiography as a technique for the study of speech articulation. Although definite procedural advances have been made in the past ten years, this technique continues to suffer from several severe limitations, primarily related to data reduction and processing. Various other procedures for observation of articulatory movements have been developed. For example, Kozhevnikov and Chistovich (1965) utilized an 'electropalatographic' method which detects tongue-palate contacts. The use of ultrasonic techniques for this purpose also has been attempted. It is not clear at this point, however, that these techniques can provide adequate and accurate information on positions and movements of the articulatory structures. Thus, cineradiography is likely to remain an important tool for the study of speech and to be used more extensively if solutions are found for some of the present procedural limitations. UNIVERSITY OF IOWA

REFERENCES Astley, R., 1958 "The Movements of the Lateral Vails of the Naso-Pharynx: a Cineradiographic Study", J. Laryngology and Otology 72: 325-28. Bjork, L., 1961 "Velopharyngeal Function in Connected Speech, Studies Using Tomography and Cineradiography Synchronized with Speech Spectrography", Acta Radiologica Suppl. 202:1-94. Coker, C.H., 1966 "Model for Specification of the Vocal-Tract Area Function", JASA 40: 1271. Cooper, F.S., D. Zeichner, D.M. Speaker, A.S. Abramson, 1963 "Slow Speech", Paper presented at 15th Annual Convention of the Audio Engineering Society (New York City). Cooper, H.K., and F.A. Hofmann, 1955 "The Application of Cinefluorography with Image Intensification in Field of Plastic Surgery, Dentistry, and Speech", Plast. Reconstr. Surg. 16:135-37. Daniloff, R.G., 1967 "A Cinefluorographic Study of Selected Aspects of Coarticulation of Speech Sounds", Ph.D. thesis (University of Iowa). DeClerk, J.L., L.S. Landa, D.L. Phyfe and S.I. Silverman, 1965 "Cinefluorography of the Vocal tract", Se Cong. Int. d'Acoustique, Paper A26. Delattre, P., 1965 Comparing the phonetic features of English, French, German and Spanish: An Interim Report (Philadelphia, Chilton Books).

408

KENNETH L. MOLL

Heinz, J.M., and S. öhman, 1966 "Models for Articulation", Speech Transmission Lab. Quart. Prog, and Status Rep. 1: 2-13. Heinz, J.M., and K.N. Stevens, 1965 "On the Relations Between Lateral Cineradiographs, Area Functions, and Acoustic Spectra of Speech", 5e Cong. Int. d'Acoustique, Paper A44. Houde, R.A., 1967 "A Study of Tongue Body Motion During Selected Speech Sounds", Ph.D. thesis (University of Michigan). Kozhevnikov, V.A., and L.A. Chistovich, 1965 Rech' Artikuliatsia i Vospriiatie (Moscow-Leningrad) (Transi. "Speech: Articulation and Perception", Joint Publications Research Service (JPRS) : 30543 (Washington, D.C.). Lifschiz, J. M., 1963 "Tongue Markers for Cinefluorographic Analysis", J. Speech Hearing Dis. 28: 393-96. Lindblom, B., 1964 "Articulatory Activity in Vowels", Speech Transmission Lab. Quart. Prog, and Status Rep. 2:1-5. MacNeilage, P. F., and J.L. DeClerk, 1967 "On the Motor Control of Coarticulation in CVC Monosyllables", Conf. Reprints, 1967 Conf. on Speech Communication and Processing (Air Force Cambridge Research Laboratories): 157-63. Massengill, R., Jr., G. Quinn, W.F. Barry, Jr., and K. Pickrell, 1966 "The Development of Rotational Cinefluorography and its Application to Speech Research", J. Speech Hearing Res. 9: 259-65. Moll, K.L., 1960 "Cinefluorographic Techniques in Speech Research", J. Speech Hearing Res. 3: 227-41. 1962 "Velopharyngeal Closure on Vowels", J. Speech Hearing Res. 5: 30-37. 1965 "Photographic and Radiographic Procedures in Speech Research", Am. Speech Hearing Assoc. Reports 1: 129-39. Öhman, S. E. G., 1966 "Coarticulation in VCV Utterances: Spectrograph«; Measurements", J. Acoust. Soc. Amer. 39: 151-68. Perkell, J. S., 1965 Cineradiographic Studies of Speech: Implications of a Detailed Analysis of Certain Articulatory Movements", 5e Cong. bit. d'Acoustique, Paper A32. Ramsey, J.L., 1964 "Logical Techniques for Glottal Source Measurements", Instrumentation Papers #48 (Air Force Cambridge Res. Labs. Hanscom Field, Mass.) : 1-10. Shelton, R.L., Jr., A.R. Brooks, K.A. Youngstrom, W.M. Diedrich, and R.S. Brooks, 1963 "Filming Speech in Cinefluorographic Speech Study", J. Speech Hearing Res. 6: 19-26. Subtelny, J.D., 1964 Physio-Acoustic Considerations in the Radiographic Study of Speech", Cleft Palate J. 1:402-10. Subtelny, J.D., S. Pruzansky, and J. Subtelny, 1957 "The Application of Roentgenography in the Study of Speech" in Manual of Phonetics, L. Kaiser, ed. (Amsterdam: North Holland Publishing Co.). Truby, H.M., 1959 "Acoustico-Cineradiographic Analysis Consideration with Especial Reference to Certain Consonantal Complexes", Acta Radiologica Suppl. 182.

MICHEL MONNOT and MICHEL FREEMAN

A COMPARISON OF SPANISH SINGLE-TAP /r/ WITH AMERICAN /t/ A N D /d/ IN POST-STRESS INTERVOCALIC POSITION

1. PURPOSE

In teaching a foreign language it is not always necessary to accustom the students to a totally new phonetic system for, in most cases, the target language has phonemes similar to those existing in the native tongue. For instance, an American student learning French will not have to pay much attention to the fricative sounds: he has all of them, plus an additional two in his native English. Conversely, a French student learning English will have to hear and then produce the difference between /i/ and /i/, a difference for which he has no reference in his native French. The problem becomes more delicate when two languages have a similar sound which pertains, however, to different phonemes in each language. For instance, a frequently used technique in teaching the single-tap Spanish /r/ to American students is to point out the auditory similarity existing between that sound and the sound of an American intervocalic, post-stress /t/ or /d/. In this study, we shall examine by means of spectrography and cineradiography, whether such an assumption of similarity between these two sounds is purely subjective or whether it has some objective basis. A perception test will be used to verify the findings.

2. HISTORY

In English, the question of the occurrence of [r] as an allophone of/t/ or /d/ in intervocalic position has been raised on several occasions. As far back as 1938, Haugen (1938:630) notes that "voiced t is physically more like d or a trilled r than a t. It is a voiced, unaspirated alveolar fricative." Soon after Kenyon (1940:232) remarks: "Voiced t is often described as a single tap r. To American ears, the two are quite distinct." Almost at the same time, Bloch (1941) perceives the problem in this manner: "In the speech of many Americans, the /t/ phoneme includes as one of its constituent sounds or allophones an alveolar flap, which occurs

410

MICHEL MONNOT AND MICHEL FREEMAN

intervocalically after a stressed vowel, as in butter, betting, kitty." And Oswald (1943) sums up the dilemma: Whatever variants of the sound occur, whether they be well articulated, or trilled, or tapped, or merely flipped, they are all voiced sounds of some sort produced by a more or less rapid contact between the tongue-tip and the alveolar ridge. Now there is nothing about such a sound phonetically speaking, which would lead an objective observer to conclude that it is a kind of a [t]. Halliday, Mcintosh and Strevens (1964:68) also freely acknowledge the presence of [r] in a "narrow transcription" of English. For the sentence "Yes but I don't see why", they transcribe: "[jiss bar ae daent si wae]". And lastly Chomsky (1964) has given great consideration to the linguistic value of this sound. From a pedagogical point of view, however, it is Martinet and Stockwell who best handle the problem. Referring to the Spanish single-tap /r/, Martinet (1956:28,36) describes it as "une vibrante apicale à un seul battement, une sorte de [d] de tenue extrêmement brève"; he then draws the parallel with American English where "la vibrante [r] n'est qu'une variante fréquente de /d/ (et même dans certains usages de /t/) entre voyelles", to conclude by the following advice: "Pour obtenir d'un Américain une prononciation acceptable du nom espagnol Pérez, on peut lui présenter une forme écrite Peddith." Similarly, Stockwell and Bowen (1965:51) recommend that "as a point of departure, the English speaker may do well to associate the Spanish /r/ between vowels with English Iti or /d/ between vowels (if the stress is on the first of the two syllables)."

3. DATA

The previous considerations could be considered as rather subjective. Therefore, we submitted this problematic sound, which we shall henceforth designate by [r] for both English and Spanish, to objective verification. In American English, all the instances of this [r], allophone of /t/ and /d/, occur after a stressed vowel. Words such as attention, addiction certainly do not contain this feature, but 1addict could, because of the poststress position of the /d/. In relaxed speech, we find these instances, either within the word as in Betty, water, daddy, Fido, or at the juncture of two words as in about it, he beat him, dead end. We recorded on tape three native speakers uttering the following list of words three times at normal speed: Betty, water, butter, daddy, dead end, it's about it, Fido, it's a pot of tea, berry. Berry was inserted to give us a comparison with the normal American /r/. Using three native Spaniards, the same procedure was followed for Spanish where we used words with different vocalic environments. The list was also read three times by each informant, and was made up of : ire, oro, aro, bura, muri, lira, beri, es para ti.

A COMPARISON

411

Furthermore, one informant of each language was X-rayed while reading his list to which was added he beat him, he bit him for English and la península Ibérica for Spanish.

4. ANALYSIS

4.1. Spectrographic It must first be said that not all utterances of a post-stress, intervocalic /d/ or /t/ will automatically be perceived as [r]. It is a matter of degree which depends upon two closely linked factors: articulation and duration. These factors can alter the perception of the phone from a voiceless stop to an alveolar fricative. The spectrograms of Figure 1 show two such degrees. Spectrogram A, on the left, represents an utterance of butter clearly understood as [bAta·] while spectrogram B, on the right, reproduces the utterance of the same word by the same speaker, but perceived this time as [bAra·]. It must be noted that these two examples were not posed, that they naturally occurred within the list of words spoken three times by the informant. However, while this person was a little rigid at the first reading (spectrogram A), she felt more confident and relaxed at the second reading (spectrogram B). There are three main differences between the two spectrograms. The stressed vowel [ Λ ] is shorter in A than in Β and this is naturally linked to another factor: the hold of the consonant. We notice in A that the duration is the same for the hold of the consonant as for the surrounding vowels. In Β on the contrary, the place of the consonant is marked only by a short interruption in the vowel formants. In terms of duration, the consonantal hold of A measures 8 cs., while the consonantal break of Β measures 3 cs., for a ratio of 2.6 to 1. But this is not a statistical study and these measurements are given only for reference. The third difference between A and Β can be seen in the lower part of the spectrograms : in A, the fundamental is clearly interrupted and the break in the second harmonic is already as wide as in the higher frequencies; in B, the fundamental and the second harmonic, except for a slight loss of intensity, remain practically unaffected. It must be emphasized that this [r] has nothing to do with the American /r/ (described in detail by P. Delattre and D. Freeman, 1968) which, for clarity, we represent in Figure 2 by the symbol R. Spectrograms A and Β of Figure 2 offer a comparison between these two sounds. Spectrogram A is the name Betty uttered with a very short alveolar hold, perceived as [r]. As in Figure 1, spectrogram B, we notice that the fundamental and first few harmonics are only slightly interrupted. Moreover, the three formants do not present any noticeable transitional changes either before or after the consonant. In spectrogram B, which is the representation of berry, we find the typical post-stress intervocalic American /r/, with its rising first formant and its second and thfrd formant lowering respectively to 1150 and 1450 cps. These two

412

MICHEL MONNOT AND MICHEL FREEMAN

formants can reach much lower but are hindered here by the vocalic environment peculiar to /ε/ and /I/. In contrast to A, spectrogram C shows the word beri as uttered by a Spanish informant. As in A, we see the first few harmonics practically unaffected by the consonant and the second formant-transition dipping slightly. The duration of the hold is very similar in both cases, and if it were not for the formant frequencies of /ε/, 550 and 1750 cps in A and 400 and 1950 cps. in C, it would be easy to mistake one for the other. As shown in Figure 2, there seems to be no criterion to differentiate between a Spanish single-tap /r/ and an American post-stress intervocalic /t/ or /d/ perceived as [r]. 4.2. Cineradiographic If from the acoustic point of view the Spanish and American phones present practically no divergencies, what is their articulatory nature? Do they belong to the tap or the flap categories? Let us first define both terms. While many phoneticians including Pike, Jones and Heffner fail to distinguish clearly between the two, Ladefoged (1964:30) states that a tap involves a rapid movement of the tip of the tongue up to tap (and occasionally to trill) against the forward part of the alveolar ridge. The retroflex flap is made by drawing the tongue tip up and back, and allowing it toflapagainst the posterior part of the alveolar ridge as it comes down. With this definition as a criterion for distinction, let us examine the X-ray drawings of Figure 3. Column A (English) represents three consecutive frames of the word water, (1) just preceding [r], (2) [r] itself, (3) just following [r]. Column Β (Spanish) represents three consecutive frames of the word Ibérica, (1) just preceding [r], (2) [r] itself, (3) just following [r]. (The X-ray movie was taken at the speed of 24 frames per second.) From this figure, two conclusions can be drawn. First, in terms of definition both sounds can be classified as taps for neither one shows the retroflex and falling action of the tongue against the alveolar ridge, and both show a rapid surge of the tip of the tongue to the "forward part of the alveolar ridge". Second, in terms of contrast between the Spanish [r] and the English [r], only minor distinctions can be made, as in the acoustical analysis. In frame 1, the tip of the tongue in English is already prepared to go to the alveolar ridge while the tip of the tongue in Spanish shows no consonantal anticipation. This is due to the closed syllabification characteristic of English as opposed to the open syllabification tendency of Spanish. In frame 2, in English, the tongue tip seems to include a wider part of tne dorsum. In contrast to this, the tip alone seems to be involved in Spanish, while the rest of the tongue, probably due to the influence of the vocalic anticipation, prepares itself for

b

λ

b

t

λ

r

su

Fig. 1

b

e

r

i

b

c

R Fig. 2

I

b

e

r

i

413

A COMPARISON

Β

Fig. 3

the upcoming [i]. In English therefore, there is slightly more contact between the tip of the tongue and the alveolar ridge. In frame 3, in both English and Spanish, the tip of the tongue has been completely withdrawn from the alveolar ridge and the tongue is preparing for the articulation of the following vowel, [ar] in English, [i] in Spanish. From this analysis it can furthermore be confirmed that the duration of the tap

414

MICHEL MONNOT AND MICHEL FREEMAN

has been completed, in both languages, in 1/24 of a second, roughly equal to 4 cs. This is approximately the same duration that was obtained in spectrogram Β of Figure 1. This coincides also with Atkinson's oscillomink study on the duration of taps and flaps in several languages (1967). She found the average length of a tap to be 3.9 cs. Thus, for all practical purposes we can state that from an articulatory viewpoint there is no critical difference between the English [r] and the Spanish /r/.

5. PERCEPTION TEST

Since the spectrographs and cineradiographic analyses revealed no pertinent difference between these two sounds, it was thought that a perception test should verify these results. We had an American informant record the sentences he bit him, he beat him, in a very relaxed manner, as he would utter them in a non-affected everyday conversation. This, of course, produced the desired [r] in both sentences. It also resulted in the loss of all four /h/'s. Hence, devoid of any phonetically recognizable context, these sentences could easily appear nonsensical to a naive listener. We presented these two sentences twice to 23 persons who were chance visitors to our laboratory. Each one indicated his native tongue, his father's native tongue, his mother's native tongue, his own linguistic history and background, and wrote phonetically or in any manner that which he had just heard. We were, naturally, chiefly interested in the critical consonant. Eight listeners were women, 15 were men. Among those, 11 were native speakers of American English, 7 of German, 2 of Spanish (1 Mexican, 1 Castillian), 2 of French and 1 of Arabic (Lebanese). The test elicited three types of responses: 10 listeners heard [t] or [d] in both cases, 11 heard [r] in both cases and 2 heard [r] for one sentence and [d] for the other. A study of the linguistic background of these persons is very revealing. Of the ten who heard [t] or [d] in both sentences : (1) Six were native speakers of American English: two electronic engineers who had no foreign language training, two undergraduate English majors with limited knowledge of French, a doctoral candidate in French, and a chemistry professor with a scant knowledge of Spanish. All six transcribed the sentences he bit him, he beat him without any hesitation, re-establishing the /h/'s. (2) Three were native speakers of German who had neither an apical [r] in their dialect nor any training in a language with such an [r]. (3) One was a native speaker of French whose only foreign language training was in English: he had been in the United States for sixteen years. Among those who heard [r] in both cases were: (1) Four native speakers of American English: one spoke Norwegian fluently, another had Russian speaking parents; a third, although not speaking Spanish, had been

Λ COMPARISON

415

brought up in a Mexican-American neighborhood and the fourth had had four years of High School Spanish and she pronounced that language very well. (2) Three native speakers of German, two of whom had an apical [r] in their dialect, the third having had two years of High School Spanish and a mother with an East Prussian dialect (apical [r]). (3) Two native speakers of Spanish. (4) One native speaker of French, who spoke Spanish and Italian fluently. (5) One native speaker of Arabic, a language with an apical [r] in its phonetic system. It is interesting to note that all of them (including the four Americans) transcribed the sentences with an r, without any h and always wrote e for bit and i for beat. The third group of responses came from two persons, one American and one German who transcribed respectively ibering, he beat him and ibering, ibiding. The American had no history of apical [r], the German had an apical [r] in his native dialect. It can be gathered from these results that the persons who had never been or not sufficiently exposed to an apical [r] could not recognize such a sound except in terms of a post-stress intervocalic /t/ or /d/, and were compelled to write out the sentences either with the correct English orthography or with something approximating it: ibeding, ebiding. On the contrary, those for whom this sound did not relate to such allophonu; realization of ft/ or ¡d/, or who either had an apical [r] in the phonemic system of their native tongue, or had enough interference carried from another language with an apical [r] in its system, readily understood and transcribed an [r]. Here, the relaxed manner in which the recording was made proved useful, for if the /h/'s had been clearly present, their occurrence would have given a more precise phonetic context, their influence probably would have overridden that of [r] and at least the four Americans would have been expected to transcribe the sentences correctly. Consequently this test would have been of no value. In this manner, however, it demonstrated that the [r] is similar enough in both languages to be interpreted both ways, either as a Spanish /r/ or as an American intervocalic /t/ or /d/, depending on the prevailing phonetic habits of the listener.

6. CONCLUSION

Can it be safely assumed by instructors who teach Spanish to American English speaking students that the single-tap Spanish [r] has an equivalent phone in English, the allophone of /t/ or /d/ in post-stress, intervocalic position; and that, according to Martinet's advice, the way to obtain from an American an acceptable pronunciation of the Spanish name Pérez is to present him with the written from Peddith? We do not wish to argue here the validity of giving written forms of one language to improve pronunciation habits in another. However:

416

MICHEL MONNOT AND MICHEL FREEMAN

(1) from the study of spectrograms which show virtually no difference in the acoustic rendition of this [r] sound in English and Spanish, (2) from the study of motion X-ray which also tends to prove the articulatory similarity of that consonant in both languages, (3) from a perception test which shows that the differentiation between the two depends mainly upon the linguistic background of the listener, and not upon the acoustical and articulatory characteristics of the sound, it can be stated that this [r] phone, which has been described as a tap, is the same in English as in Spanish and that, from a pedagogical viewpoint, lists of word oppositions such as Betty, beri, and such as those presented in Stockwell and Bowen (1965) are not only valid in the teaching of Spanish to American students, but are also useful and objectively correct. UNIVERSITY OF CALIFORNIA AT SANTA BARBARA

REFERENCES Atkinson, K., 1967 "Definition and Differentiation of Alveolar Stops, Flaps and Taps", unpublished article. Bloch, Β., 1941 "Phonemic Overlapping", American Speech 16: 278-84. Chomsky, N., 1964 "Current Issues in Linguistic Theory" in The Structure of Language, J. A. Fodor and J.J. Katz, eds. (New York), 50-118. Delattre, P., and D. Freeman, 1968 "A Dialect Study of American R's by X-Ray Motion Picture", Linguistics 44: 29-74. Halliday, M.A.K., A. Mcintosh and P. Strevens, 1964 The Linguistic Science and Language Teaching (Bloomington, Indiana). Haugen, E., 1938 "Notes on voiced t in American English", Dialect Notes 6/16-17: 627-34. Kenyon, J.S. 1940 American Pronunciation (Ann Arbor, Michigan). Ladefoged, P., 1964 A Phonetic Study of West-African Languages (Cambridge). Martinet, Α., 1956 La description phonologique avec application au parler franco-provençal d'Hauteville (Savoie) (Genève). Oswald, V. Α., 1943 "Voiced t — A Misnomer", American Speech 18: 18-25. Stockwell, R., and J.D. Bowen, 1965 The Sounds of English and Spanish (Chicago).

CARROLL L. OLSEN

SOME SIMILARITIES IN OLD FRENCH A N D MODERN SPANISH VERB MORPHOLOGY

Two genetically related languages such as French and Spanish may have similarities on any level of analysis, from phonetic to syntactic. The student who studies both languages notices that there are phonetic similarities such as lack of aspiration after voiceless stops in French la capitale and Spanish la capital and morphological similarities such as the construction of the future and conditional tenses by using an infinitive plus inflectional endings — French chanterai and chanterais compare with Spanish cantaré and cantaría. The linguist, in addition to observing and carefully recording these similarities, is often interested in defining the time at which they appear, the extent to which they exist, and the reasons for their occurrence. On the phonetic level, Pierre Delattre made a pioneering comparison of Old French and Modern Spanish (1946). Through a perceptive analysis, he observed thirty-one similarities between the two languages which he related to degree of articulatory laxness. The success of this study leads the reader to wonder if, on another level of analysis, the comparison would be as fruitful. In the following pages, therefore, an attempt is made to discover similarities on the morphological level and to determine to what extent they are present. Thefieldof investigation is limited to verb morphology in the two languages. Only those similarities are considered which are proper to both Old French and Modern Spanish and which have, for the most part, disappeared in Modern French.1 Each point of comparison is discussed briefly under a separate heading.

1. THE STRONG PERFECT

The perfect tense in Old French is especially inviting to the linguist because it contains a rich variety of conjugational types which can be described in relatively precise terms. 1

The term 'Old French' is considered in its broadest sense as expressed by F.B. Luquiens (19S1 : 11) as "... the transitional stage between Latin and modern French". Examples for Modern Spanish have been verified in the Gramática de la Lengua Española (1962).

418

CARROLL L. OLSEN

Structurally, the verbs of this tense fall into two major categories which are identified by position of stress and are usually called WEAK and STRONG. 2 The weak perfects, those stressed throughout on the suffix, comprise four subcategories: α-type: chantai, chantas, etc., /-type: dormi, dormis, etc., «-type: valut, valus, etc., and dedi-type: perdi, perdis, etc. The first two categories are, for the most part, descendants of the Classical Latin verbs in -Ävl and -Ivi, while the latter two are new formations. The origin of the w-type is uncertain, but according to M. K. Pope (1961:370), it is probably related to the rapid extension of a new weak past participle in -ÜTUM. 3 The new dedi-type supposes an intervening form based on the perfect of DARE as in PERDIDÏ > *perdedi > perdi. As a free form DEDI was lost, but as a bound form in this re-formation it gained a new and invigorating breath of life. The strong perfects, those containing forms with stress on the radical, also comprise four sub-categories: /-type: vin, venis, etc., s-type: dis, desis, etc., w-type: dui, deus, etc., and the perfect of the verb estre : fui, fus, etc. The maverick perfect of estre is considered separately because it came to have stress throughout on the radical in Old French. The three other types of strong perfects, mostly descendants of similar types in Classical Latin, contain a fluctuation of stress between the radical and the suffix. It is precisely these three types in -i, -s, and -u which interest us here because the shifting of stress between the radical and the suffix is characteristic of a large number of verbs in Old French, was later lost in Modern French, but is nicely preserved in Modern Spanish. In Old French the strong perfects were not evenly distributed among these three types. The language shows certain definite preferences. Perfects in -s, sigmatic perfects, were by far the most frequent, followed by those in -u, and finally by those in -i. Nyrop gives 49 examples for the sigmatic type, 24 for the -u type, and three for the /-type (1903:138-46). W. Meyer-Liibke gives a similar list with corresponding forms in several Romance languages where again the sigmatic type is by far the most frequent (1895:357-76). It is from this preferred type that the following example is taken. The perfect of DÏCERE is shown with conjugated forms in Latin, French, and Spanish; accent marks are added to show at a glance the position of stress. The salient feature which relates this verb to others within the three types of strong perfects is distribution of stress. In Old French the tonic accent falls on the radical

* The terms 'weak' and 'strong' were first used in Germanic studies where strong verbs are those which show inflection through apophony (sing, sang, sung), and weak verbs are those which show inflection through addition of a dental suffix (work, worked, worked). Friedrich Diez (1874: 116-19) applied these terms to the Romance Languages where they are identified with position of stress. Strong forms have stress on the radical; weak forms have stress on the suffix. It should be noted that most of the perfects in Spanish and Old French which are referred to as 'strong' contain two or at the most three strong forms. The only verb in Old French which has strong forms throughout in the perfect is estre (fui, fus, etc.). ' "The rapid extension of the new weak type of past participle in -UTUM, and the tendency to secure conformity between perfect and past participle led to the creation of a new perfect type in -(/ stressed, and this type received much extension among verbs with radicals in /, m, n, r."

SOME SIMILARITIES IN OLD FRENCH AND MODERN SPANISH VERB MORPHOLOGY 419

in the first person singular and the third persons singular and plural but on the suffix in the other three forms. In other words, Old French is quite faithful to the stress pattern of Classical Latin. The only discrepancy is in the first person plural where Classical Latin stressed the radical and Old French stresses the suifix. This change had already taken place, however, in Vulgar Latin where, by analogy, the first person plural had been brought into conformity with the second person plural. Classical Latin

Old French

Modern French

Modern Spanish

DÍxi

dis

dis

dije

DÍXÍSTÍ DÍXIT

desís díst

dis dit

dijiste dijo

DÍXIMUS

desímes

dimes

dijimos

οίχίβΉ$

desistes

dites

dijisteis

DÍXERUNT

dístrent

dirent

DÏXÉRUNT

dijéron

The third person plural is of special interest here because Classical Latin presented two possibilities for position of stress, one on the radical and another on the suffix. The two Romance languages took divergent routes at this point: French chose the Latin pattern with stress on the radical while Spanish chose the pattern with stress on the suffix. Changes from Old to Modern French severely affected the strong perfects. Only a small number of them were permitted to survive. Some were replaced by weak forms, many were completely lost, and those that managed to weather the storm were mutilated to such an extent that fluctuation of stress became impossible. Through vowel reduction in weakly stressed syllables, the disyllabic strong perfects were reduced to monosyllables where the only syllable to remain was the one which carried the tonic accent. The Modern French perfect of the verb dire in the preceding table is representative of this monosyllabic shape. Spanish, on the other hand, is much more conservative with respect to the strong perfects. The modern language has preserved several verbs with fluctuation of stress in the perfect. The Spanish Academy lists 14 representatives of this type, a few of which can enter into additional compounds. 4 These verbs, like decir in the table above, have a stress pattern which is identical to that of Old French with the one exception which has already been mentioned. In the third person plural, Spanish has stress on the suffix and Old French has stress on the radical. The striking fact which comes to light through this comparison is that a structural pattern which was prevalent in Old French and which was lost in Modern French, a pattern of stress fluctuation in the strong perfect, is nearly the same in Modern Spanish. 4 Gramática de la Lengua Española (1962: IOS). The verbs with strong forms in the preterit are listed as caber, decir, haber, hacer,poder, poner, querer, saber, tener, traer, venir, andar, estar, conducir.

420

CARROLL L. OLSEN

2. VOCALIC ALTERNATION IN VERB RADICALS

Another phenomenon which is often related to the position of stress is vocalic alternation. In accordance with phonetic laws, the vowels of the radical of many Old French verbs developed differently as the tonic accent fell on the radical or on the suffix. The following conjugation of the Old French verb plorer in the present indicative testifies to the regularity of this occurrence and to its dependence upon stress: pleur pleures pleurent)

plorons plorez pleurent

Strong stress falls on the radical in all forms except the first and second plural where the tonic accent falls on the suffix. The radical alternation is eu ~ o. This type of alternation, known as apophony, is quite frequent in monosyllabic verb radicals of Old French.5 We have only to glance at an early text to find such pairs as: lief ~ levons, espeir ~ esperons, demeur ~ demorons. As a matter of fact, W. Meyer-Lübke (1895:253) states that Old French surpasses all other Romance languages in the diversity of its vocalic formulae, and M. K. Pope (1961:350) lists as many as 14 different alternations for the twelfth century. How rich in alternating vocalic tones the language must have been! Regretfully, as French developed into modern times, vocalic alternation was slowly eroded by the destructive force of analogy. One of the two radical forms was usually generalized, most frequently the one with the atonic vowel: lef ~ lavons > lave ~ lavons; truef ~ trovons > trouve ~ trouvons. In a few cases, however, the tonic vowel was generalized as in aim ~ amons > aime ~ aimons. This fad of reconstructing the radical by eliminating apophony was so contagious that, at the present time, only a few remnants of the forms with vowel alternation exist. As apophony played an important role in the structure of Old French verbs, so it does in Spanish at the present time. The Modern Spanish verb system is characterized by vocalic alternations in such verbs as: contar, perder, sentir, morir, and pedir. The alternation o ~ ue appears in contar ~ cuento·, e ~ ie in perder ~ pierdo·, e ~ ie ~ ι in sentir ~ sienta ~ sintamos·, o ~ ue ~ u in morir ~ muera ~ muramos; and e ~ i in pedir ~ pido. Certain of these alternations clearly depend on the breaking of Latin short e and o into ie and ue under strong stress. There are, however, other factors which are at play. For those interested in a full discussion of this subject, Yakov Malkiel has presented a masterful study in a recent issue of Language (1966). With respect to the number of different formulae, Spanish cannot rival Old French. ® The more frequent alternation in Old French verbs with disyllabic radicals is between presence and absence of a vocalic element as in parol-parlons. This alternation was realized through the loss of the counterfinal vowel and was later eliminated by generalization of one of the two radicals. In those cases where the counterfinal vowel was not completely lost, it was reduced to the neutral vowel which, in Modern French, is unstable and alternates with open [ε] in a number of verbs: acheter ~ achète; appeler ~ appelle, etc.

SOME SIMILARITIES IN OLD FRENCH AND MODERN SPANISH VERB MORPHOLOGY

421

Some five alternations in Modern Spanish compare with approximately 14 in Old French. But in the frequency of their occurrence, the two languages are certainly more comparable. A. Bello and R. J. Cuervo (1941:139-49) give revealing statistics for the various alternations in Spanish. In their counts only the simple verbs are listed if the same alternations are reflected in the compounds. They list 64 verbs with the alternation e ~ ie; 54 with the alternation ο ~ ue; and 23 with the alternation e ~ i. The alternation e ~ ie ~ i is found in infinitives ending in -ferir, -gerir, -vertir, and the verbs arrepentirse, herir, hervir, mentir, requerir, sentir, and their compounds. The only simple verbs which contain the alternation o ~ ne ~ u are dormir and morir. These statistics reveal that apophony is indeed quite extensive in Modern Spanish as it is in Old French. Therefore, Old French shows again a much greater similarity to Modern Spanish than to Modern French. Even though the individual verbs with apophony in Modern French are often of high frequency, they are, as a group, much fewer in number than those in Old French and Modern Spanish.

3. IMPERFECT INDICATIVE SUFFIXES

The development of the Latin imperfect suffixes -ÄBAM, -ËBAM, and -IËBAM into the Romance languages poses interesting problems, not the least of which is the transformation of intervocalic -B-. Since some of the Romance languages show forms both with and without this labial element, much of the discussion has centered around either its irregular loss (Bourciez, 1956; Meyer-Lübke, 1895; Thurneysen, 1882) or its irregular retention (Rebecca Posner, 1961). Because our interest here is more with the result than with the details of the transformation, it will suffice to mention a recent article by K. Togeby (1964) where several of the theories are presented and a new explanation is offered. Several degrees of labial loss can be identified in the Romance area. In Italian the intervocalic labial is retained throughout the conjugations while in Rumanian it is invariably lost. But in Old French divergent forms appear showing both its presence and its absence. The following forms with different inflectional suffixes frequently appear in Old French texts : chanteve chanteves chantevet chantevent

chantoue chantoues chantouet chantouent

vendeie vendeies vendeie vendeient

The three series of inflectional suffixes are not, of course, interchangeable. Both -eve and -oue are used with verbs of the first conjugation and are developments of Classical Latin -ÄBAM, while -eie is used with verbs of the other conjugations and probably finds its origin in Classical Latin -ËBAM. In addition, -eve and -oue are rival

422

CARROLL L. OLSEN

suffixes — the former is characteristic of the East and the latter of the West. In a given area, therefore, there were usually two series of inflections developing at the same time, -eie and either -eve or -oue. Of the two parallel forms developing in both areas, an important differentiating factor is labial constriction. The labial element is visible in -eve where -v- is the regular development of Latin intervocalic -B- (FABA > fève) and in -oue where M. K. Pope (1961:184) states that Latin -B- opened and velarized to -«-. The labial element is clearly absent in -eie. Therefore, the two series of inflectional suflixes existing side by side can be characterized by presence and absence of a labial element. In a similar fashion, Modern Spanish also provides the speaker with two sets of inflections for the imperfect tense — one with a labial, -aba {cantaba), and one without, -ía (comía, dormía). Spanish, then, is parallel to Old French. Modern French, however, is less conservative than either Old French or Modern Spanish. The Old French rival forms -eve and -oue were supplanted by -eie by the end of the thirteenth century, so that Modern French is characterized by only one series of inflections which are the regular development of the forms in -eie. Again Modern French stands alone, and Old French proves to be closely allied with Modern Spanish.

4. O L D F R E N C H ESTRE

VS. SPANISH ESTAR

AND

SER

A final point which does not have the scope of the previous three comparisons but which is certainly worth mentioning concerns the Old French verb estre. The imperfect indicative of this verb contains a series of forms which are derived from the Classical Latin verb ESSE: ERAM > iere, ere; ERAS > ieres, eres; etc. These Old French forms co-exist with another series of the type esteie, esteies, etc. The latter set was either formed by analogy to the infinitive stem est- plus imperfect inflectional endings or was borrowed from the Old French verb ester, a descendant of Latin STARE. The latter hypothesis is not to be rejected lightly because the verb estre underwent other influences of the verb ester. The present and past participles STANTEM and STATUM of STARE give estant and este which were used in the conjugation of estre. In many of the Old French texts, the two series of forms for the imperfect exist side by side as in the following lines from the Chanson de Roland: 719: "Sunjat qu'il eret as greignurs porz de Sizer" 726 : "Qu'il ert en France a sa cápele ad Ais" 2860: "A Eis estete a une feste anoel" In addition to their co-existence, the two series seem to be at least partially synonymous as is seen in verses 726 and 2860. The forms derived from Latin ERAM, however, were supplanted by those of the type esteie. It is from this latter type of longer forms that the Modern French conjugation developed. The shape of the two Old French series certainly calls to mind similar forms in

SOME SIMILARITIES IN OLD FRENCH AND MODERN SPANISH VERB MORPHOLOGY

423

Modern Spanish within the imperfect tense of estar and ser. Estaba is similar to Old French esteie, and era is certainly related to Old French iere and ere. A full statement of this relationship proves to be quite delicate. For the moment it must suffice to say that in both Old French and Modern Spanish two forms exist which are similar in shape, a similarity which is probably due to their origin in the same Classical Latin verbs STARE and ESSE. The two series of forms seem to converge semantically in Old French, and the one series was lost in the modern language. Therefore, with respect to the number and shape of forms in the imperfect indicative of estre, Old French reveals a rapport with Modern Spanish which it does not hold with Modern French.

5. CONCLUSION

This brief comparison of Old French and Modern Spanish on the morphological level has revealed four points of similarity which are related to (1) the strong perfects, (2) vowel alternation in verb radicals, (3) imperfect suffixes, and (4) the imperfect tense of Old French estre and Modern Spanish estar and ser. Even though this investigation is limited to the verb system alone, it reveals that such a comparison on the morphological level of analysis is possible and even quite productive. A consideration of the complete morphology of the two languages should be rich in possibilities. The similarities found here, however, are far less numerous than those identified by Pierre Delattre. Our four compare with his thirty-one. The morphological similarities are, nevertheless, usually of a larger scope than the phonetic ones. In the phonetic comparison, individual sounds are described. Here, focus is centered on patterns which cover several or many phonetic segments. It might be said, then, that what the morphological analysis lacks in number of points covered, it compensates for in breadth. As a final note, a word should be said about the reasons for resemblance. Delattre related the phonetic similarities between Old French and Modern Spanish to articulatory laxness, a phenomenon which developed in both languages. The morphological similarities differ by being at least partially due to a common structure which both languages inherited from Latin. The presence of strong perfects, a labial element in the imperfect inflections, and two forms for the imperfect in Old French estre as compared with Spanish estar and ser reflect a structure which already existed in Classical Latin. On the other hand, the development of vowel alternation in verb radicals and the weakening of the intervocalic labial in the imperfect suffixes can probably be related to articulatory laxness. UNIVERSITY OF CALIFORNIA AT BERKELEY

424

CARROLL L. OLSON

REFERENCES Bello, A. and R.J. Cuervo, 1941 Gramática ele la lengua castellana (Buenos Aires, Libreria Perlado). Bourciez, E., 1956 Eléments de linguistique romane, 4e ed. (Paris, Klincksieck). Delattre, P., 1946 "Stages of Old French Phonetic Changes Observed in Modern Spanish", PMLA 61: 7-41. Diez, F., 1874 Grammaire des langues romanes, vol. 2, translated by A. Morel-Fatio and G. Paris (Paris, Librairie A. Franck). Luquiens, F.B., 1951 An Introduction to Old French Phonology and Morphology (New Haven, Yale University Press). Malkiel, Y., 1966 "Diphthongization, Monophthongization, Metaphony: Studies in Their Interaction in the Paradigm of the Old Spanish -ir Verbs", Language 42, 2: 430-72. Meyer-Lübke, W., 1895 Grammaire des langues romanes, Vol. 2, translated by A. Doutrepont and G. Doutrepont (Paris, H. Welter). Nyrop, K., 1903 Grammaire historique de la langue française, Vol. 2 (Paris, Alphonse Pickard). Pope, M. K., 1961 From Latin to Modem French, 2nd ed. (Manchester, The University Press). Posner, R., 1961 "The Imperfect Endings in Romance", Transactions of the Philological Society·. 17-55. Real Academia Española, 1962 Gramática de la lengua española, 9th ed. (Madrid, Espasa-Calpe, S.A.). Thurneysen, R., 1882 Das Verbum 'être' und die französische Conjugation (Halle). Togeby, K., 1964 "Les désinences de l'imparfait (et du parfait) dans les langues romanes", Studia Neophilo· logica 36, 1: 3-8. Whitehead, F., ed., 1965 La Chanson de Roland, 2nd ed. (Oxford, Basii Blackwell).

R. L. POLITZER

DEVELOPMENTAL ASPECTS OF AUDITORY DISCRIMINATION*

That auditory discrimination between foreign speech sounds is influenced by the phonemic pattern of the native language is a well known fact. 1 It is comparatively easy for us to perceive sound différences which have counterparts in phonemic distinctions made in our native language — and it is difficult to perceive differences which do not correspond to contrasts which are utilized for phonemic purposes in our native language. Several years ago, I undertook a study concerning the auditory discrimination of French vowels by speakers of English (Politzer, 1961). The study showed that French sound contrasts involving nasals and/or rounded front vowels are difficult to perceive by English speakers and that in addition contrasts like i/e or u/o may also present problems — the latter evidently because a very high pronunciation of [e] or [u] in French may move these sounds into the range of English [i] (bit) and [ij] (good).2 In addition, my study dealt with auditory discrimination between French vowels and their diphthongal English counterparts. It showed that even after several years of exposure to French, speakers of American English have some difficulty discriminating between such pairs as English do / French doux, English sea / French si, etc. (Politzer, 1961:38). In pedagogical discussion it is generally assumed that there is an obvious connection between auditory discrimination and pronunciation. Correct pronunciation depends on the individual's ability to monitor his own responses. If he cannot distinguish the phonemes of the target language from each other, he cannot monitor the intelligibility of his response. If he cannot distinguish the phonemes of the target language from those of his native language, he cannot tell whether or not he is substituting his native speech sounds for those of the target language. Auditory discrimination training * The data on which this study is based were gathered as part of a research project on auditory discrimination and echo response sponsored by the Proctor and Gamble Fund, School of Education, Stanford University. The author is indebted to Mr. Louis Weiss (Stanford Center for Research and Development in Teaching) for his help in gathering data and to the Palo Alto Unified School District for the permission to conduct research in the Palo Alto schools. 1 For studies showing the influence of the native system on auditory discrimination see Polivanov, 1931 and Sapon and Carroll, 1958. • α . Politzer, 1961: 37; Delattre, 1946: 24-48.

426

R. L. POLITZER

has thus been made an integral part of many language courses and has become the subject of various experiments concerning pronunciation, and auditory discrimination tests are often accepted as substitutes for pronunciation tests.3 It was particularly because of this tie between auditory discrimination and pronunciation, that I thought it of interest to test the auditory discrimination ability of boys and girls at different age levels. The test used in the investigation was based on items used in the study mentioned above. However, the 1961 study used the technique of asking subjects to recognize one different word in a group of four. The study reported in this article was based entirely on the subject responding with 'same' or 'different' to the presentation of pairs. The auditory discrimination test consisted of 48 items: (1) ten French monosyllabic minimal pairs, (2) ten disyllabic minimal pairs, (3) ten identical French monosyllabic pairs, (4) ten identical French disyllabic pairs, (5) eight English/French pairs differentiated primarily by the contrast between the French and English vowel. The test was recorded by a near bilingual speaker with expertise in phonetics. Identical pairs were recorded through actual rerecording of the first word. The items used in the test are listed below. (In the actual administration of the test the items were, of course, scrambled and did not follow each other in the order given below.) (1) 1. basse/bosse 2. doux/du 3. peur/père 4. lin/l'un 5. pend/pas 6. les/lit 7. deux/du 8. beau/bout 9. rue/ri 10. l'on/l'un

(2) (3) 1. cirer/serrer 1. les/les 2. le pire/le pure 2. du/du 3. l'on dit/lundi 3. beau/beau 4. empreinte/emprunte 4. rue/rue 5. ma serre/ma soeur 5. l'on/l'on 6. eunuque/une nuque 6. pend/pend 7. le dos/le doux 7. basse/basse 8. pâté/poté 8. doux/doux 9. buter/bouter 9. peur/peur 10. tenter/tâter 10. l'un/l'un

(4) 1. une nuque/une nuque 2. le doux/le doux 3. pâté/pâté 4. ma serre/ma serre 5. empreinte/empreinte 6. buter/buter 7. santé/santé 8. cirer/cirer 9. le pire/le pire 10. lundi/lundi

(5) 1. sea/si 2. do/doux 3. doe/dos 4. fay/fee 5. key/qui 6. shoe/chou 7. bow/beau 8. lay/les

The auditory discrimination test was administered to first, third, fifth, seventh and ninth grade classes. All pupils in the classes took the tests with the exception of a few who had some training in/or knowledge of French. Also excluded from the data were the scores of a few first graders who had evidently and most obviously not understood the test instructions. Starting with the fifth graders, the subjects taking the test had had some training in Spanish. This training was comparatively light at the elementary school level, but was more intensive starting with the Junior Highschool (seventh grade) level. The results of the auditory discrimination test are summarized in Tables 1, 2, 3, 4, and 5. The tables indicate the following results: (1) The relative difficulty of the categories of items is (5) (14.6% errors), (2) (8.6%), (3) (8.0%), (1) (4.9%), (4) (4.5%). In other words, the contrasts between English * For a brief discussion of discrimination training and pronunciation see Henning, (1966). For 'pronunciation tests' based on auditory discrimination see Lado, 1961: 46-77.

427

DEVELOPMENTAL ASPECTS OF AUDITORY DISCRIMINATION TABLE 1

1

2

4 10 14 6 11 17 2

3 4 7 1

Test Items girls boys total third grade girls boys total girls fifth grade boys total seventh grade girls boys total ninth grade girls boys total Total girls Total boys Total first grade

16 23 39 30 21 51 22 35 57 37 23 60 28 22 50 133 124 257

—

—

1 — —

2 1 1 2

— — — —

—

—

—

—

13 22 35

4 4 8

3

4

5

3 4 7 2 5 7 1 5 6 — 3 — 8 — . 11 3 — 1 — 4 4 12 11 23 15 35

1 3 4 1

—

6 6 2 3 5 2 2 4

1

6

7

8

9

2 2 —

1 2 3 2

1 4 5 1

2 4 6 4

—

2

1

4

—

1 1 —

—

2 4 6

% Errors 10.0% 16.9% 14.1% 6.3% 9.1% 7.4% 2.3% 2.9% 2.6% 1.1% 4.7% 2.5% 1.1%" 0.5% 0.8% 3.5% 6.5% 4.9%

3 3

3 2 5

2 4 6

6 4 10

1 3 4

7

8

9

10

Total Errors

% Errors

2 2

1 1 2 1

27 58 85 40 33 73 11 27 38 6 11 17 4 5 9 88 134 222

16.9% 25.2% 21.8% 13.3% 15.7% 14.3% 5.0% 7.7 % 6.6% 1-7% 4.8% 3.2% 1.4% 2.3% 1.8% 6.6% 10.8% 8.6%

1 1

—

1

Total Errors 16 39 55 19 19 38 5 10 15 4 11 15 3 1 4 47 80 127

—

—

10

—

—

—

—

—

— —

—

1 —

—

2 2

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

1 1

TABLE 2

Test Items

1

girls 16 boys 23 Total 39 third grade girls 30 boys 21 Total 51 fifth grade girls 22 boys 35 Total 57 seventh grade girls 37 boys 23 Total 60 ninth grade girls 28 boys 22 Total 50 Total girls 133 124 Total boys Total 257

1 6 7 1

first grade

2

3 7 10 2 — 1 1 3 —

—

1 1

1 1 1

— —

—

1

— —

—

—

—

—

—

2 7 9

6 9 15

3

4

5

6

11 5 16 5 27 2 16 — 18 2 34 1 5 — 15 1 20 — 5 — 6 — 11 — 2 — 2 — 4 3 39 5 57 8 96

3 6 9 2 3 5

3 4 7 2 2 4

—

—

—

1 1 1

—

2 2

— — —

—

—

—

1 —

1 1

—

—

5 8 13 12 — 6 1 18 — — 4 — 1 7 — 1 11 —

—

3 3 2 2 4

—

1 1

— —

—

—

—

—

—

—

—

—

—

—

—

—

5 9 14

5 9 14

1 3 4

—

1 1 —

1 1 2 2 2 8 4 10

2 2 2 2 4 23 25 48

and French words caused the largest number of errors. Perceiving identical French disyllabics as different caused the least number of errors. The most difficult item was (5) 6. (shoe/chou). It was followed by (2) 4. {empreinte!emprunté), (5) 2. (do/doux),

R. L. POLITZER

428

TABLE 3

1

3

4

5

6

7

8

9

10

4 4 6 5 8 8 9 12 14 2 4 3 1 4 5 3 8 8 2 2 1 1 3 1 3 5 2 1 3 5 2 1 1 3 4 6 1 6 2 6 — 2 1 8 8 11 14 21 15 16 17 26 30 38

7 7 14 6 2 8 3 4 7 3 1 4

1 2 3

3 2 5 1 2 3 1 1 2 1

1 3 4 1 1 2

2 4 6 2

4 5 9 2 2 4 1 2 3 4 2 6

1 3 4 3 2 5 2 1 3 3

Test Items girls 16 boys 23 total 39 girls 30 third grade boys 21 total 51 girls 22 fifth grade boys 35 total 57 37 seventh grade girls boys 23 total 60 ninth grade girls 28 boys 22 total 50 Total girls 133 Total boys 124 Total 257 first grade

2

3 3 19 16 35

— — —

1 —

1 — —

— —

2 3

—

1 1 1 1 2

—

1

—

—

—

3 —

1 1

—

3

1 1 — — 1 1 — — 2 6 3 7 11 9 2 5 7 6 11 6 4 11 10 13 22 15 — —

Total Errors 33 47 80 24 19 43 16 14 30 21 9 30 9 13 22 105 101 206

% Errors 20.6% 20.4% 20.5% 8.0% 9.0% 8.4% 7.3% 4.0% 5.3% 5.7%* 3.9% 5.0% 3.2% 5.9% 4.4% 7.9% 8.2% 8.0%

TABLE 4

Test Items

1

2

3

4

5

6

girls 16 boys 23 total 39 third grade girls 30 boys 21 total 51 girls fifth grade 22 boys 35 total 57 seventh grade girls 37 boys 23 total 60 ninth grade girls 28 boys 22 total 50 Total girls 133 Total boys 124 Total 257

1 6 7 2 1 3

2 2 4 5

2 1 3 1

1 1 2 2 1 3

1 4 5 1

3 3 6 2 2 4

first grade

—

—

5 2

— —

—

2 2 4 6

— — — — — — —

3 7 10

—

1 1 11 7 18

1 2 1 3 2 1 3

—

1 2 4 6

—

2 2 1 —

—

—

— — —

2

—

1 1 —

1 1

1 — — 1 1 1 — 1 1 1 1 2 7 5 5 5 4 5 9 10 11 10 14 15

1 1 3 —

3 5 2 7

9

10

1 1 2

3 3 6 5

2 2 4 3

2 2

2

3 3

8

—

—

—

—

1 1 1

7

—

—

5

3 1 1 2

—

2 2 — 1 — 1 — 2 — 1 1 1 1 2 1 10 4 7 5 17 — —

— — — — — —

6 3 9

Total Errors

% Errors

16 23 39 23 6 29 7 14 21 6 9 15 6 6 12 58 58 116

10.0% 10.0% 10.0% 7.7%' 2.8% 5.7% 3.2% 5.0% 3.7% 1.6% 3.9% 2.5% 2.2% 2.7% 2.4% 4.4% 4.7% 4.5%

(2) 10. {tenter/tâter), (3) 3. (beau/beau), (3) 4. (rue/rue), (1) 1. (basse/bosse), (1) 4. (lin\Vun). Thus the difficulty of individual items also confirms that the problem areas are evidently the discrimination of English/French vowels (especially English [ o ] /

429

DEVELOPMENTAL ASPECTS OF AUDITORY DISCRIMINATION TABLE 5

1

2

3

4

5

6

7

8

5 5 10 3 6 9

8 14 22 12 12 24 5 4 9

2 4 6 2 1 3 2 2 4

1 2 3 5 5 10

3 11 14 3 1 4 1 2 3

14 17 31 13 15 28 14 22 36 8 3 11 2 2 4 51 59 110

3 11 14 3 3 6 4 3 7

6 9 15 5 8 13 3 3 6 2

Test Items girls 16 boys 23 39 total 30 third grade girls boys 21 total 51 fifth grade girls 22 boys 35 total 57 seventh grade girls 37 boys 23 total 60 ninth grade girls 28 boys 22 50 total Total girls 133 Total boys 124 257 Total first grade

—

1 1 —

—

2 2

— —

—

2 2

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

—

8 12 20

25 32 57

6 7 13

6 9 15

7 14 21

—

2 2 1 —

—

2 — —

1 11 19 30

—

11 25 36

Total Errors

% Errors

42 73 115 46 51 97 29 39 68 10 7 17 3 2 5 125 177 302

32.8% 39.7% 36.9% 19.2% 30.4% 23.5% 17.9%" 13.9% 14.9% 3.4% 3.8% 3.5% 1-3% 2.5% 1.3% 12.5% 17.9% 14.6%

French [o]) and the discrimination between French nasals (especially [è]/[œ]) or between French nasal and oral vowels fla]/[a]). (2) In the total of all categories girls made fewer errors than boys. Only in a few instances do boys score better than girls. (Girls' scores lower than boys' scores are marked by an asterisk. (3): fifth, seventh and ninth grade; (4): fifth grade. (5): fifth grade.) In general, the slight superiority of girls over boys thus follows a pattern which is not unusual for various types of skills and which seems also in accord with the findings of Carroll and Sapon (1959) concerning the slight superiority of girls over boys on language aptitude tests. (3) Auditory discrimination ability increases very rapidly from grade level to grade level. In the Graph, which indicates the percentage of errors made in each category at each of the grade levels at which the test was given, we see also how the percentage of errors in each category drops at successive grade levels. As is indicated by Table 6, TABLE 6 Improvement in % Error by Grade Level Category 1 2 3 4 5

Grade 1 — — — — —

Grade 3 6.7% 7.5% 11.1% 4.3% 13.4%

Grade 5 4.8 % 7.7% 3.1% 2.0% 8.6%

Grade 7 0.1% 3.4% 0.3% 0.2% 11.4%

Grade 9 1.7% 1.4% 0.6% 0.1% 2.2%

430

R. L. POLITZER

Graph Showing % of Errors in each Category by Grade Level

40 .

35 .

®\

\

\

\

\\ \

(2) \

\

1 3 5 7 9 the greatest improvement in each category occurs between the first and third grades and there is a general tendency toward leveling off after the fifth grade. The only notable exception to this pattern is category (5) : the discrimination between English diphthongal and French non-diphthongal vowels. In that particular category, seventh grade shows a sharp improvement over fifth grade. The conclusion suggests itself that the training in Spanish may be responsible for this increase in ability to distinguish English diphthongal vowels from non-diphthongal ones. The main conclusion of the study is obvious: AUDITORY DISCRIMINATION ABILITY QUITE EVIDENTLY INCREASES WITH MATURATION. Except for the case of discrimination of English vowels vs. non-English vowels, where a special training factor has probably influenced the results, auditory discrimination seems more or less fully developed by the time the child reaches the fifth grade. Improvement after fifth grade seems comparatively slight. The results of the study thus face us with a paradox: (1) Auditory discrimination is connected with the ability to pronounce;

DEVELOPMENTAL ASPECTS OF AUDITORY DISCRIMINATION

431

(2) Auditory discrimination ability is not reached until fifth grade. How do these statements square with the well-known fact that small children can acquire perfect pronunciation much better than grown-ups? Of course, we need further investigation of the ability of small children to acquire correct pronunciation before we can give a definite answer to this question. However, we can suggest that the paradox is apparent rather than real: The auditory discrimination ability tested in our study involved the decision of labeling sounds as 'same' or 'different'. This decision is indeed affected by the phonemic system of the native language. As we acquire the phonemic system of our language we learn to ascribe perceived bits of phonic substance to the phonemes of the native language. The acquisition of language may be pictured as a process in which — to use Harlan Lane's terminology (Lane, 1966:216:22) — bits of phonic substance and corresponding events are mapped into categories: phonemes and sememes — categories of sound and categories of meaning. If we hear the sounds of a foreign language, we identify them according to the system of the native language. A foreign language sound may be identified as the same as a native sound if it falls within the range normally associated with the latter's phonic substance or at least shares some of its phonetic features. Identification of foreign speech sounds is indeed as J. C. Catford called it, "phonological translation" and the "basis for translation equivalence in phonological translation is the relationship of SL (subject language) and TL (target language) phonological units to the 'same' phonic substance" (Catford, 1965:56). The ability to discriminate which was tested in our investigation may thus be described as the ability to 'translate' phonic substance of a foreign language into the categories of the native language. It is this ability which increases rapidly with maturation — and which undoubtedly accounts for the fact that the ability to learn foreign languages without accent decreases with age. The younger the child, the less the decision of 'same' or 'different' can be made in terms of the native language. Evidently, in learning to assign phonic substance to a phonemic category, the smaller the child, the more it is likely to 'start from scratch'. As I have stated elsewhere, "the child's resistance to interference or negative transfer may be — in part at least — the result of a less developed ability to transfer under any circumstance, positive or negative" (Politzer, 1965:28). This relative inability to transfer phonemic discrimination from the native language puts the small child at a definite disadvantage in an auditory discrimination test in another language. It is this same 'inability' to transfer which evidently also accounts for the fact that the small child is ultimately able to acquire the foreign language without 'interference' and to learn a foreign language as a 'coordinate' system rather than a system that is 'compounded' with the native language.4 The rapid increase in auditory discrimination ability during the primary grades, thus corresponds in fact to a decrease in the ability to learn a foreign language as a completely separate system without native language interference. STANFORD UNIVERSITY * For the distinction of 'coordinate' and compound systems, see Brooks 1964, SO.

432

R. L. POLITZER

REFERENCES Brooks, Nelson, 1964 Language and Language Learning. Theory and Practice, 2nd ed. (New York, Harcourt, Brace and World). Carroll, J.B. and S.M. Sapon, 1959 Manual, Modern Language Aptitude Test (New York, Psychological Corporation). Catford, J.C., 1965 A Linguistic Theory of Translation (London, Oxford University Press). Delattre, Pierre, 1946 "Les Indices acoustiques de la parole: premier rapport", Phonetica 2: 242-48. Henning, William Α., 1966 "Discrimination Training and Self-Evaluation in the Teaching of Pronunciation", International Review of Applied Linguistics in Language Teaching 4: 7-17. Lado, Robert, 1961 Language Testing (London, Longmans, Green and Co. Ltd). Lane, Harlan, 1966 "Identification, Discrimination, Translation: The Effects of Mapping Ranges of Physical Continua onto Phoneme and Semene Categories", International Review of Applied Linguistics in Language Teaching 4: 215-26. Politzer, R.L., 1961 "Auditory Discrimination of French Vowels by English Speakers", Journal of the Canadian Linguistic Association 7: 32-44. 1965 Teaching French, An Introduction to Applied Linguistics, 2nd ed. (New York, Blaisdell Publishing Company). Polivanov, S., 1931 "La Perception des sons d'une langue étrangère", Travaux du Cercle Linguistique de Prague 4: 79-96. Sapon, S.M. and J.B. Carroll, 1958 "Discriminative Perception of Speech Sounds as a Function of Native Language", General Linguistics 3: 62-73.

A. ROSETTI

SUR LA REPRÉSENTATION PAR ÉCRIT DES SONS PARLÉS

Nous nous proposons d'apporter ici quelques éclaircissements sur le problème de la représentation par écrit des sons parlés. L'écriture est, comme on le sait, une approximation: les signes à l'aide desquels nous notons les sons parlés nous sont imposés par l'orthographe de chaque langue.1 Ils s'appliquent à des types phoniques déterminés, les SONS-TYPES. Le son-type que nous notons par écrit à l'aide d'une lettre ou d'un signe donné représente le son en général, tel que le conçoit chaque sujet parlant (Rosetti, 1965: 144). Ainsi, dans l'alphabet latin, e, par exemple, note les diverses variantes de cette voyelle existantes dans la langue parlée, qui diffèrent l'une de l'autre d'une manière quelconque: è (ouvert), é (fermé), accentué, non accentué, etc. (Janakjev, 1964). L'écriture retient donc, pour les noter, seulement les caractères distinctifs de chaque son-type, la communication par le langage étant fondée sur les différences phoniques à valeur sémantique entre les mots. Le mécanisme de la représentation par écrit des sons parlés a été décrit par Troubetzkoy de la manière suivante: man schreibt nicht das, was man wirklich auspricht, sondern das, was man zu sprechen meint oder zu sprechen beabsichtigt ... Man muss sich immer daran erinnern, das die Schrift nicht das phonetische, sondern immer nur das phonologjsche System der Sprache wiedergibt, und dass das phonologische System sich nicht mit dem phonetischen deckt (1933:111).

Par rapport au son-type, ou au phonème, le graphème est indivisible: la lettre a, par exemple. Le graphème n'est autre chose que la représentation par écrit du sontype, qui réunit dans une seule unité, comme nous l'avons indiqué ci-dessus, les diverses variantes du son parlé. A la différence du son-type, le phonème ne fait pas état du psychologisme: c'est un invariant placé à un autre échelon des faits de langage (Rosetti, 1965; Avram, 1962:11). Dans le procès de la notation par écrit des sons du langage humain, il arrive que 1

Voir notre exposé (Rosetti, 1947: 10-11), ainsi que Janakjev, 1964: 61-62, et Benveniste, 1966: 22. Ce dernier déclare (1966): "[le locuteur] en entendant des sons ... identifie des phonèmes (22) ... [Dans l'alphabet latin] chaque lettre correspond toujours et seulement à un phonème (24)."

434

A. ROSETTI

celui qui rédige le message épèle, tout en écrivant, les sons qui forment le mot, et qu'il accompagne l'épellation de l'émission vocale du son, à voix chuchotée ou même à haute voix (Rosetti, 1965:94). Cette constation nous aide à interpréter certaines graphies qui, au premier abord, peuvent sembler bizarres: les sons superflus, notés dans ces cas, sont provoqués par l'épellation et n'appartiennent pas à tel ou tel mot. Si le scripteur note, en roumain, sikeris pour scris 'écrit', c'est que cette notation est dûe à l'épellation du mot: sî-kî-ris. La phonologie apporte donc une contribution essentielle à l'interprétation des sons représentés par écrit. UNIVERSITÉ DE BUCAREST

RÉFÉRENCES Avram, Α., 1962 "Sur quelques particularités des systèmes graphématiques", Cahiers de linguistique théorique et appliquée 1. Benveniste, E., 1966 Problèmes de linguistique générale (Paris). Janakjev, M., 1964 "Teorija orfografii i reèi", Voprosy jazykoznanija 13. Rosetti, Α., 1947 Mélanges de linguistique et de philologie (Copenhague-Bucarest). 1965 Linguistica (La Haye). Troubetzkoy, N.S., 1933 "Les systèmes phonologiques considérés en eux-mêmes et dans leurs rapports avec la structure générale de la langue", Deuxième Congrès international de linguistes (Paris), pp. 120125.

MARIO ROSSI

LE SEUIL DIFFÉRENTIEL DE DURÉE

Le problème si important de la durée vocalique a depuis longtemps suscité l'intérêt de nombreux phonéticiens. P. Delattre lui-même, avec la rigueur scientifique qu'on lui connaît, y a consacré de nombreuses études (voir la bibliographie). Tout récemment encore, P. Delattre, pionnier de la vérification par la synthèse, a été amené à contrôler sur le plan perceptif, le rôle de la durée vocalique comme indice acoustique du mode d'articulation consonantique (Delattre, 1967). Il écrit: "L'analyse, est le point de départ des recherches. Elle permet de faire des hypothèses qui seront vérifiées dans les expériences par la synthèse ... L'analyse donc ne peut que mettre sur la voie des indices. Il faut ensuite vérifier par la synthèse. Les exemples de conclusions erronées faites sur la base de l'analyse seule abondent" (Delattre, 1966:252). La vérification par la synthèse, d'autre part, a, comme condition préalable, l'étude des seuils de perception. C'est là également une idée chère à P. Delattre. Or, curieusement, tandis que les recherches sur le seuil différentiel d'intensité et de hauteur abondent 1 , celles qui concernent le seuil différentiel de durée sont plutôt rares. Le travail le plus récent qui présente un intérêt phonétique est celui de Creelman (1962).2 Creelman cependant se borne à définir le seuil différentiel de durée pour les sons purs. Pour ce qui est des sons de la parole, nous pouvons peut-être tirer indirectement quelques indications des travaux de Fry sur l'accent (Fry, 1955) et de ceux de Deneä sur le rôle de la durée vocalique dans la perception de la sonorité consonantique (Deneäs, 1955). Creelman présente au hasard à des sujets surentraînés des stimuli accompagnés de bruit blanc (seuil masqué); il impose un choix forcé aux sujets qui doivent dire si le signal le plus long apparaît le premier ou le deuxième. Les durées-étalons se situent entre 0,02 s et 0,32 s; l'écart entre les valeurs de la variable est de 0,01 s. 1

Du moins les recherches sur le seuil différentiel d'intensité et de hauteur pour les sons purs. Outre les travaux de Creelman sur la perception de la durée, on peut citer ceux de Chjtovich, 19S9, de Clausen, 1950, de Henry, 1948, de Small and Campbell, 1962 et de Stott, 1935. 2

436

MARIO ROSSI

Lorsque l'écart est égal à 0,04 s, Creelman obtient 95 % de réponses correctes, pour une durée-étalon de 0,12 s, 90% pour 0,18 s et 85% pour 0, 265. La loi de Weber semble donc être en gros vérifiée. En effet, le rapport AT/Τ (rapport du seuil différentiel à la durée-étalon) calculé pour 75 % de réponses correctes est à peu près constant pour les valeurs moyennes situées entre 0,10 s et 0,30 s et s'établit aux environs de 20% de la durée-étalon. Des études de Fry et de DeneS qui ne sont pas des travaux systématiques sur le seuil différentiel de durée, nous ne pouvons tirer que des renseignements fragmentaires et approximatifs, renseignements intéressants tout de même car Fry et Denes procèdent à une vérification par la synthèse. Les résultats obtenus par Fry montrent qu'une augmentation de 50% du rapport de la première voyelle à la deuxième, dans des mots comme 'object', entraîne un changement significatif dans le comportement des sujets (Fry, 1955:767).® Doit-on en conclure que AT est moins fin pour la parole que pour les sons purs? Nous ne le pensons pas. C'est un faisceau d'indices acoustiques qui rend compte de la proéminence accentuelle. Si on isole une composante, la durée par exemple, il faudra faire varier sa valeur d'un pourcentage bien supérieur à AT, pour obtenir un changement significatif dans le comportement des sujets. L'étude de Deneä, quant à elle, prouve que la durée vocalique joue un rôle dans la perception du trait de sonorité de la consonne subséquente: lorsqu'on allonge la durée de la voyelle d'une valeur égale à 5 c/s, le pourcentage de sujets qui perçoivent une consonne sonore après cette voyelle augmente de façon significative.4 Malheureusement, Denes utilise, dans la progression des durées vocaliques, une valeur absolue et non un rapport constant: ainsi, lorsque la durée de la voyelle passe de 5 c/s à 10 c/s, AT/T = 1, de 10 c/s à 15 c/s, AT/T = 0,50, et de 15 à 20 c/s, AT/Τ = 0,33. Dans le premier cas, le pourcentage de réponses de sonorité augmente de 10%, dans le deuxième cas de 20%, dans le troisième de 10%. Il est difficile dans ces conditions de se faire une idée précise de la valeur du seuil différentiel relatif de durée pour les sons de la parole. Pour tenter de combler cette lacune, nous avons entrepris une série d'études systématiques dont nous donnons ici les premiers résultats. Avant de définir la valeur AT pour les sons en contexte, nous avons estimé qu'il était utile et éclairant de travailler sur des voyelles isolées. Dans ces quelques pages, nous essaierons de préciser la valeur du seuil différentiel relatif de durée pour la voyelle [a].

* Le pourcentage de sujets qui perçoivent l'accent sur la première syllabe passe alors de 38 à 55 %. 4 5 c/s = 0.05 s. Au cours de cet article, l'unité utilisée pour exprimer la durée est le centième de seconde.

LE SEUIL DIFFÉRENTIEL DE DURÉE

437

1. PROTOCOLE EXPÉRIMENTAL

Une voyelle [a] tenue a été enregistrée dans une chambre anéchoïque. La fréquence fondamentale et l'intensité de cette voyelle sont constantes5; le spectre et les caractéristiques de la voyelle sont représentés dans les figures la, lb, le et ld. C'est à partir de ce modèle que nous avons obtenu, par segmentation et repiquage, tous les stimuli dont nous avons besoin. Nous avons éliminé l'attaque vocalique; la segmentation a été faite pour chaque stimulus en un point situé à 3 c/s après le début de [a]. La fréquence fondamentale en effet est plus basse à l'attaque que pendant la tenue (—10 Hz): pour les durées très brèves de l'ordre de 6 c/s nous introduisions ainsi une variable supplémentaire qui risquait de fausser les résultats. Les précautions que nous avons prises nous ont permis de présenter aux sujets des stimuli qui ne diffèrent que par la durée; le spectre, la fréquence fondamentale et l'intensité étant par ailleurs identiques.® Chaque sujet doit comparer un stimulus-étalon à un stimulus variable.7 Nous avons calculé AT pour les trois valeurs étalon: 12 c/s, 18 c/s et 26 c/s, de façon à couvrir tout le champ des durées effectivement utilisées dans la parole normale. L'écart entre les valeurs de la variable est fixé à 2 c/s. Les durées comparées respectivement à 12 c/s, 18 c/s, et 26 c/s sont les suivantes: (12) — 6 c/s 8 10 12 14 16 18 20 22

(18) — 10 c/s 12 14 16 18 20 22 24 26

(26) — 18 c/s 20 22 24 26 28 30 32 34 36

Pour chaque durée-étalon, 11 séries de stimuli sont présentées au sujet (la première, dont on ne tient pas compte dans le calcul, permet au sujet de s'adapter). Dans chaque série, l'ordre de présentation des paires 'étalon-variable' et l'ordre de présentation à l'intérieur de chaque paire sont aléatoires (figure 2). • Micro Philips électrostatique LBB 9060 (position omnidirectionnelle. Magnétophone Nagra ΠΙ B.) Fréquence fondamentale de la voyelle: 140 hz; intensité RMS: 70 db. RE., à 30 cm du micro (calibration en niveau faite avec un micro B. et K., capsule 4133, Κ = + 1 3 . 6 db.) * Intensité RMS mesurée au niveau de l'oreille des sujets, dans une chambre anéchoique : 70 db. R.E. Nous précisons cette valeur car, comme le fait remarquer Creelman (1962:585), AT augmente rapidement pour les trop fortes ou les trop faibles intensités; la discrimination est meilleure pour des intensités de l'ordre de 70 db. (Source sonore linéaire de 40 à 20.000 H J . 7 Au sujet de la méthode de comparaison utilisée ici voir Clausen (1950:756). Pour ce qui est de la méthode suivie dans la mesure du seuil voir Noizet et Gary 1966:55-73.

438

MARIO ROSSI DURÉE-ÉTALON 12 c/s 1

2

3

4

5

6

7

8

9

10

1

8-12

12-18

22-12

12-10

16-12

6-12

12-12

12-10

12-8

12-16

2

12-12

16-12

12-14

8-12

12-18

16-12

12-6

14-12

12-14

8-12

3

12-14

12-14

18-12

18-12

14-12

18-12

12-22

20-12

12-12

12-12

4

12-10

12-22

8-12

12-20

12-10

12-10

16-12

8-12

12-10

12-20

5

12-18

12-8

12-10

12-12

12-22

22-12

12-8

18-12

18-12

12-22

6

12-6

12-6

6-12

22-12

12-20

12-20

20-12

12-16

22-12

12-18

7

22-12

12-12

12-20

16-12

12-12

12-14

12-10

12-22

16-12

10-12

8

12-16

20-12

12-12

12-14

6-12

12-12

18-12

6-12

12-20

12-14

9

12-20

12-10

16-12

6-12

8-12

14-12

12-12

12-6

6-12

a-a

12-8

Fig. 2 Tableau des séries prises en compte dans le calcul du seuil pour le stimulus étalon 12 c/s. Verticalement les 10 séries contenant chacune 9 paires de stimuli placés dans un ordre aléatoire.

Dans une paire, une durée de 80 c/s sépare le stimulus-étalon du stimulus variable; entre chaque paire on laisse un silence de 3 s. Le sujet, placé dans une chambre anéchoïque, a le choix entre 3 réponses : -+- si le deuxième stimulus est jugé plus long que le premier, — si le deuxième stimulus est jugé plus bref que le premier, ? si le sujet est incertain ou juge que les deux stimuli sont égaux. Les calculs sont établis à partir des réponses de 23 sujets non entraînés, de 25 à 40 ans, pris au hasard parmi les étudiants et le personnel technique de la Faculté des Lettres d'Aix-en-Provence. Il s'agit de locuteurs méridionaux dont le système linguistique ne comporte pas de corrélation de longueur.

2. EXPLOITATION DES RÉSULTATS

Sur la feuille de résultats de chaque sujet sont indiquées, en colonne, les valeurs du stimulus variable, en ligne, les séries de présentation. On dénombre dans chaque colonne les réponses de même catégorie en attribuant: la valeur 0 à la réponse — la valeur 0,5 à la réponse ? la valeur 1 à la réponse +

LE SEUIL DIFFÉRENTIEL DE DURÉE

439

DURÉE-ÉTALON 18 c/s 10

12

14

16

18

20

22

24

26

1

-

-

7

-

7

?

+

+

+

2

-

-

-

-

?

+

7

+

+

3

-

-

-

?

?

7

+

+

+

4

-

-

-

-

+

+

+

7

+

5

-

-

-

?

7

+

?

+

+

6

-

-

-

7

?

7

?

+

+

7

-

-

?

-

7

+

7

+

+

8

-

-

-

7

7

?

?

+

+

9

-

-

?

-

7

+

+

+

+

-

7

7

+

+

+ 100

10 XlO Conversion en Ζ

-

0

0

15

20

55

75

75

95

-1.04

-0.84

+0.13

+0.67

+0.67

+ 1.64

Fig. 3 Tableau d'exploitation des réponses du sujet L.R. En colonne les valeurs du stimulus variable (de 10 à 26 c/s) comparées au stimulus étalon. En ligne, les 10 séries. Pour l'interprétation des résultats les valeurs du stimulus variable sont classées dans l'ordre croissant. Le signe donné dans la réponse du sujet est converti en fonction de cet ordre : ex. à l'audition de la paire ordonnée 10-18 le sujet doit inscrire + sur sa feuille. Nous supposons momentanément que la durée-étalon précède dans tous les cas la variable. Par conséquent, si la réponse est juste, nous inscrirons le signe — dans la colonne des valeurs inférieures à l'étalon et le signe + pour les valeurs supérieures à l'étalon. Le signe + pour les premières et le signe — pour les secondes indiquent des réponses erronées. Les réponses de doute sont inchangées.

La somme des valeurs de chaque colonne est ensuite convertie en pourcentages (figure 3). Nous avons ensuite estimé le seuil, pour chaque sujet, à partir de la droite de Henri. On porte en abscisse les valeurs du stimulus variable et en ordonnée les valeurs de Ζ positives et négatives qui correspondent aux pourcentages obtenus plus haut. La droite de Henri est construite à partir de deux de ses points; les coordonnées de ces points sont déterminées de la façon suivante: (1) on partage les valeurs du stimulus variable en deux groupes égaux: la médiane de chacun des groupes constitue l'abscisse de chaque point. (2) La moyenne des valeurs de Ζ pour chaque groupe détermine l'ordonnée de chaque point (figure 4). Nous avons pris comme estimation du seuil différentiel la demi-différence entre les

440

MARIO ROSSI

Fig. 4 Droite de Henri d'après les résultats obtenus avec le sujet L.R. (Durée étalon : 18 c/s). En abscisse les valeurs de la variable, en ordonnée celles de z. La droite de Henri coupe les droites d'ordonnée + 1 et — 1 aux points d'abcisse 14.25 et 22.25. C'est la demi-différence entre ces deux durées qui détermine le seuil.

abscisses des points d'intersection de la droite avec les parallèles à l'axe des abscisses d'ordonnées: Ζ = + 1 et Ζ = — 1. La valeur ainsi trouvée correspond à la détermination graphique de l'écart-type (68% de réponses correctes). Le point d'intersection de la droite de Henri et de l'axe des abscisses précise la valeur du point d'égalisation perceptive (PEP); cette valeur représente la perception que le sujet a de l'étalon. Le calcul de t par la formule relative à la comparaison de deux moyennes dans le

90

dX

80

70

t 60

t

50

63 65

üg •ii^

1

70

• «·

r c/s

jtl

J*