In honor of Ilse Lehiste: Ilse Lehiste Pühendusteos 9783110886078, 9783110133035

Citation preview

In Honor of Ilse Lehiste

Netherlands Phonetic Archives The Netherlands Phonetic Archives (NPA) are a modestly priced series of monographs or edited volumes of papers, reporting recent advances in the field of phonetics and experimental phonology. The archives address an audience of phoneticians, phonologists and psycholinguists. Editors·. Marcel P.R. Van den Broecke University of Utrecht

Vincent J. van Heuven Univeristy of Leyden

Other books in this series: I

Nico Willems English Intonation from a Dutch Point of View

IIA

A. Cohen and M.P.R. Van den Broecke (eds.) Abstracts of the Tenth International Congress of Phonetic Sciences

IIB

M.P.R. Van den Broecke and A. Cohen (eds.) Proceedings of the Tenth International Congress of Phonetic Sciences

III

J.R. de Pijper Modelling British English Intonation

IV

Lou Boves The Phonetic Basis of Perceptual Ratings of Running Speech

V

Renee van Bezooyen Characteristics and Recognizability of Vocal Expressions of Emotion

VII

Mary E. Beckman Stress and Non-Stress Accent

Robert Channon and Linda Shockey (eds.)

In Honor of Ilse Lehiste

Ilse Lehiste Pühendusteos

1987 FORIS PUBLICATIONS Dordrecht - Holland/Providence - U.SA

Published by. Foris Publications Holland P.O.Box 509 3300 AM Dordrecht, The Netherlands Sole distributor for the U.S.A and Canada·. Foris Publications USA, Inc. P.O. Box 5904 Providence RI 02903 U.S.A. CIP-DATA In Honor of Ilse Lehiste = Ilse Lehiste Pühendusteos/Robert Channon, Linda Shockey (eds.). - Dordrecht [etc.]: Foris. 2 (Netherlands Phonetic Archives; 6) ISBN 90 6765 152 4 bound SISO 805.2 UDC 801.4 Subject heading: phonetics.

Frontcover illustration taken from F.M. Helmont [An unabbreviated representation of a true natural Hebrew alphabet, which simultaneously shows how those born deaf can be taught not only to understand others who speak but even to produce speech themselves] Pieter Rotterdam, Amsterdam 1697. ISBN 90 6765 152 4 (Bound) © 1986 By the authors No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission from the copyright owner. Printed in The Netherlands by ICG Printing, Dordrecht.

To Ilse Lehiste Few would dispute the statement that Ilse Lehiste is a remarkable person. Her very great natural talents, her dignity, and her humanity would set her apart in any case, but she also benefits f r o m what seems an unusually rich background to those of us from a single cultural heritage. She was born in Estonia and her native language is Estonian. But the young Ilse, presumably both f r o m necessity and from interest, also learned German and Russian with near-native fluency and, when circumstances carried her to the United States, added English so expertly that no one who knows her only through her writings would doubt that it was her first language. Her command of SerboCroatian can be deduced from the following: while travelling through Wales, she stopped at a pottery in Hay-on-Wye. Hearing that the potter was a Serb, she spoke to him in his own language. 'Are you a Serb?' he asked. 'No,' she replied. 'But,' he gasped, 'you have no accent!' While knowing a large number of languages is not a prerequisite for a linguist, it is by no means a hindrance and equipped her very well for her role of investigating how variation in quantity and quality over time can encode linguistic information. Ilse Lehiste also has an extraordinary degree of self discipline and dedication. She has said that when she sits down to write she often experiences a struggle between her will and her i n t e l l e c t . . . she wants to work, but her intellect refuses to focus itself narrowly on the problem at hand. Rather than musing about larger issues, she sits until her will and intellects come together to produce results. These were among the abilities which allowed her to rise to prominence very quickly when she began to study acoustic phonetics with G o r d o n Peterson at the University of Michigan. A glance at her publication list shows this to have been a very fruitful time in her career, if such can be said of a career so uniformly productive. Some of her work at this time is rightly considered among the classics of phonetic literature, and she has never ceased to make timely and vital contributions to a wide variety of fields, many of which (Germanic philology, Finno-Ugric linguistics and literature, Slavic linguistics and literature, historical linguistics, and phonetics) are represented in this volume. Ilse Lehiste is a dedicated scholar, but she is by no means totally defined in this way. She is an accomplished pianist and says (in jest, we trust) that she gave up the idea of becoming a concert pianist only when the conservatory in Tallinn burned down. She is an excellent gardener and an enviable cook. She has a playful sense of humor and is justly famed for her repertoire of Estonian anecdotes and jokes. Here house holds an impressive collection of embroidered items which she admits to having done herself. She is an enthusiastic hiker

Dedication

Vll

and lover of nature, has been known to ski to the University after a heavy snowfall, and at last report had taken up horseback riding. Ilse Lehiste excites admiration of her ability to enjoy all aspects of life. She is also a person who treasures friendship and knows how to be a loyal friend. She has been heard to remark, based on her own experience, that money and property come and go, but in the end one's wealth is in the people one trusts. We, her students and colleagues, offer this volume to her in respect and friendship to mark her 60th birthday.

Contents

DEDICATION BIOGRAPHY LIST O F PUBLICATIONS

ν Xlll xv

Part One: Phonetic Studies ROBERT AUSTERLITZ: Twelve Remarks on the Teaching of Phonetics A N T H O N Y COHEN: Juncture Revisited ARVO EEK: The Perception of Word Stress: A Comparison of Estonian and Russian KNUT FINTOFT: Toneme Patterns in Norwegian and in Swedish Dialects ELI FISCHER-JORGENSEN: Segment Duration in Danish Words: Dependency on Higher Level Phonological Units IVAN F 0 N A G Y and J U D I T F 0 N A G Y : Analysis of Complex (Integrated) Melodic Patterns VICTORIA A. FROMKIN: A Note on the Suprasegmental Representation of Prosody OSAMU F U J I M U R A : A Linear Model of Speech Timing ANTTI IIVONEN: A Set of German Stressed Monophthongs Analyzed by RTA, F F T , and LPC S H I G E R U KIRITANI and MASAYUKI SAWASHIMA: The Temporal Relationship between Articulations of Consonants and Adjacent Vowels V.A. KOZHEVNIKOV, N.E., D R O Z D O V A , and E.I. STOLJAROVA: The Application of Analog Models of Some Auditory Mechanisms for Speech Signal Processing B. LINDBLOM, J. LUBKER, T. GAY, B. LYBERG, P. B R A N D E R U D , and K. H O L G R E N : The Concept of Target and Speech Timing

7 19 33 53 75 99 109 125

139

151

161

χ

Contents

L E I G H LISKER and A R T H U R S. ABRAMSON: Phonetic Validation of Distinctive Features: A Test Case in French KATHERINE MORTON: Cognitive Phonetics - Some of the Evidence MARIA ALZIRA, NOBRE and FRANCES I N G E M A N N : Oral Vowel Reduction in Brazilian Portuguese J O H N J. O H A L A and BRIAN W. EUKEL: Explaining the Intrinsic Pitch of Vowels LINDA SHOCKEY: Rate and Reduction: Some Preliminary Evidence IMAN H. SLIS: Rules for Assimilation of Voice in Dutch KENNETH N. STEVENS, G U N N A R FANT, and S A R A H HAWKINS: Some Acoustical and Perceptual Correlates of Nasal Vowels f

183 191 195 207 217 225

241

f

TAMAS TARNOCZY: The Formation, Analysis and Perception of Hungarian Affricates M A R C E L T A T H A M : Cognitive Phonetics - Some of the Theory KULLO VENDE: The Basic Intonation Contours of the Principal Communicative Types of Utterances in English and Estonian

255 271 277

Part Two: Language Studies M O H A M E D SAMI ANWAR: Semitic Has Four Vowels HREINN BENEDIKTSSON: The First Singular Preterite Subjunctive in Germanic Z.S. BOND: Baby Talk: Adult Prediction of Child Pronunciations C A T H E R I N E A. C A L L A G H A N : An 'Indo-European' Type Copula in Plains Miwok ROBERT C H A N N O N : A Function of the Instrumental Case in Russian HERBERT G A L T O N : Unreal Hypothetical Periods in South Slavic and Greek and Related Features ULRICH GROENKE: The Minimal Pair in the Finnish Spelling Debate of the Late 18th and Early 19th Centuries, with a Glance at the First Grammatical Treatise MORRIS HALLE: The Morphophonemics of Latvian Declension ROBERT T. HARMS: What Helmholtz Knew about Neutral Vowels ROBERT K. HERBERT: Articulatory Modes and Typological Universals: The Puzzle of Bantu Ejectives and Aspirates ΜΑΤΙ HINT: Loan Words and the Estonian Grade Alternation PAVLE IVIC: The Accentuation of the Serbocroatian Dialect of Uljma WICKTOR JASSEM: Vocalic Oppositions in Monosyllabic English Words

295 307 323 331 339 357

369 375 381 401 415 433 439

Contents YAKOV MALKIEL: Sound Correspondences and Levels of Transmission: Problems in the Fluctuating Representation of Medial /kt(j)/, /pt(j')/ in Hispano-Romance KENNETH Ε. NAYLOR: The Morphophonemics of Flexion in Serbocroatian: A Comparison FELIX J. OINAS: Sämpsä Pellervoinen DAVID F. ROBINSON: Vowel Lenghthening and mitatonie rude in Lithuanian ANELYA RUGALEVA: The Grammar of Choice in Russian and English SANDRA A. THOMPSON: The Passive in English: A Discourse Perspective WILLIAM S-Y. WANG and C-C. CHENG: Middle Chinese Tones in Modern Dialects ARNOLD M. ZWICKY and ANN D. ZWICKY: Patterns First, Exceptions Later ADDRESS-LIST

xi

451 465 473 479 483 497 513 525 539

Biography of Ilse Lehiste

1922

January 31, born in Tallinn, Estonia

1948

Dr. Phil., University of Hamburg. Lecturer, University of Hamburg, 1948-49

1949

Immigrated to the United States of America

1950-51

Associate Professor of Germanic Philology, Kansas Wesleyan University

1951-56

Associate Professor of Modern Languages, Detroit Institute of Technology

1952

Naturalised a U.S. Citizen

1957-63

Research Associate in Acoustic Phonetics, Communication Sciences Laboratory, University of Michigan

1959

Ph.D., Linguistics, University of Michigan

1963-65

Associate Professor, Linguistics/Slavic Languages and Literature, The Ohio State University

1964

Presenter of Plenary Session Paper, 5th International Congress of Phonetic Sciences, Munster

1965

Professorship, The Ohio State University Chairman of Linguistics Department, 1965-71 Visiting Professor, University of Cologne

1966

Visiting Professor, U.C.L.A.

1970

Director, Linguistic Institute at The Ohio State University

1974

Visiting Professor, University of Vienna

1975

Executive Committee, Linguistic Society of America

1977

Honorary Doctor of the University, University of Essex, England

1980

Visiting Professor, University of Tokyo President, Linguistic Society of America Distinguished Research Award, The Ohio State University

xiv

Biography

1982

Honorary Doctor of Philosophy, Lund University, Sweden Presenter of Plenary Session Paper, 13th International Congress of Linguists, Tokyo

1985-87

Chairman, Department of Linguistics, The Ohio State University

Fellowships and Grants 1961-64

National Science Foundation Research Grants

1967

Guggenheim Fellowship Grant, American Council of Learned Societies

1972

Grant, American Council of Learned Societies Grant, National Science Foundation

1975

Guggenheim Fellowship Fellowship, Center for Advanced Study in the Behavioral Sciences Grant, American Council of Learned Societies

1976

Grant, National Endowment for the Humanities

1977

Grant, American Council of Learned Societies

1979

Grant, American Philosophical Society Fellowship, Japan Society for the Promotion of Science

Memberships: Pi Lambda Theta, Phi Beta Kappa, Sigma Xi, Linguistic Society of America. Acoustical Society of America (Fellow), Modern Language Association, International Society of Phonetic Sciences, Association for the Advancement of Baltic Studies, American Association of Phonetic Sciences, American Association for Southeast European Studies, American Association for the Advancement of Slavic Studies, Societe Finno-Ougrienne (corresponding member), Societas Linguistica Europaea, American Association for the Advancement of Science, Linguistic Association of Canada and the United States, European Association for Signal Processing, American Literary Translators Association, Estonian Learned Society in America, Finno-Ugric Studies Association of Canada. Editorial and Executive Activities Member of the International Bibliography Committee of the Modern Language Association; member, Board of Trustees, Center for Applied Linguistics, 1974-83 (Vice-Chairman, 1974-79); President of the Association for the Advancement of Baltic Studies, 1974-76; Co-editor, Bibliotheca Phonetica, Journal of Phonetics, and Phonetica\ Member, Visiting Committee for Linguistics; Harvard College Board of Overseers, 1979-85; Member, Council for the International Exchange of Scholars, 1977-81; Member, Permanent Council for the Organization of International Congresses of Phonetic Sciences, 1983-.

Publication List of Ilse Lehiste

MONOGRAPHS 1. 2. 3. 4. 5. 6. 7. 8. 9. 10.

Uber die Ethik der nach altnordischen Quellen geschaffenen Werke von William Morris. Dissertation H a m b u r g (1948) 1-115 (typescript). An Acoustic-Phonetic Study of Internal Open J u n c t u r e . Supplement to Phonetica 5 (1970) 1-54. (with P. Ivic). Accent in Serbocroatian: An Experimental Study. A n n A r b o r , Michigan Slavic Materials 4 (1963) 1-142. Acoustic Characteristics of Selected English C o n s o n a n t s . Publication 34 of I U R C A F L , Bloomington (1964) xii-197. Some Acoustic Characteristics of Dysarthric Speech. Bibliotheca Phonetica 2, S. Karger, Basel (1965) iv-124. C o n s o n a n t Quantity a n d Phonological Units in Estonian. Uralic and Altaic Series, Vol. 65. Indiana U., Bloomington (1966) 1-73. (Editor), Readings in Acoustic Phonetics. M I T Press, Cambridge (1967) ix-358. Suprasegmentals. M I T Press, Cambridge (1970) vii-194. (with Robert J. Jeffers). Principles a n d M e t h o d s for Historical Linguistics. M I T Press, Cambridge (1979), x-209. (with Pavle Ivic). Word a n d Sentence Prosody in Serbocroatian. M I T Press, C a m b r i d g e . (In press).

ARTICLES 1. Order of subject and predicate in scientific Russian. Mechanical Translation 4 (1957) 66-67. 2. Names of Scandinavians in the Anglo-Saxon Chronicle. P M L A 73 (1958) 6-22. 3. (with G.E. Peterson). Linguistic considerations in the study of speech intelligibility. J A S A 31 (1959) 280-286. 4. (with G.E. Peterson). Vowel amplitude and p h o n e m i c stress in American English. J A S A 31 (1959) 428-435. 5. (with G . E . Peterson). The identification of filtered vowels. Phonetica 4 (1959) 161-177. 6. Segmental and syllabic quantity in Estonian. American Studies in Uralic Linguistics. Bloomington (1960) 21-82. 7. (with G . E . Peterson). D u r a t i o n of syllable nuclei in English. J A S A 32 (1960) 693-703. 8. (with G . E . Peterson). Transitions, glides, a n d diphthongs. J A S A 33 (1961) 268-277. 9. (with G . E . Peterson). Some basic considerations in the analysis of intonation. J A S A 33 (1961) 419-425. 10. The phonemes of Slovene. International J o u r n a l of Slavic Linguistics a n d Poetics 4(1961) 48-66. 11. Some acoustic correlates of accent in Serbocroatian. Phonetica 7 (1961) 114-147.

xvi

List of

Publications

12. (with G . E . Peterson). Revised C N C lists for auditory tests. J S H D 27 (1962) 62-70. 13. Acoustic studies of b o u n d a r y signals. PICPS 4 (1962) 178-187. 14. (with P. Ivic). Prilozi ispitivanju fonetske i f o n o l o s k e prirode akcenata u savremenom srpskohrvatskom jeziku. Zbornik za filologiju i lingvistiku 6 (Novi Sad, 1962) 33-73 ( + 8 plates). 15. C o m p o u n d i n g as a phonological process. P I C L 9 (1964) 331-337. 16. Palatalization in Estonian: some acoustic observations. In V. Köressaar a n d A. Rannit, eds., Estonian Poetry a n d Language: Studies in H o n o r of Ants Oras, Stockholm (1965). 17. A poem in Halbdeutsch and some questions concerning substratum. W o r d 21 (1965) 55-69. 18. The function of quantity in the phonological systems of Finnish a n d Estonian. Language 41 (1965) 447-456. 19. Juncture. PICPS 5 (1965) 172-200. 20. (with P. Ivic). Prilozi ispitivanju fonetske i fonoloSke prirode akcenata u savremenom srpskohrvatskom jeziku, II. Zbornik za filologiju i lingvistiku 8 (Novi Sad, 1965) 75-117 ( + 11 plates). 21. (with P. Ivic). Prilozi ispitivanju fonetske i f o n o l o s k e prirode akcenata u savremenom s r p s k o h r v a t s k o m jeziku, III. Zbornik za filologiju i lingvistiku 10 (Novi Sad, 1967) 55-93 ( + 10 plates). 22. (with P. Ivic). Some problems concerning the syllable in Serbo-Croatian. Glossa 1 (1967) 126-136. 23. Vowel quantity in word a n d utterance in Estonian. Congressus Secundus Internationalis Fenno-ugristarum (1968) 293-303. 24. (with K.K. Wiik). Vowel quantity in Finnish disyllabic words. Congressus Secundus Internationalis Fenno-ugristarum (1968) 569-574. 25. Kahe uue luuletuskogu vormist ja sisust. M a n a 33 (1968) 95-101. 26. (with P. Ivic). Prilozi ispitivanju fonetske i f o n o l o s k e prirode akcenata u savremenom srpskohrvatskom jeziku, IV. Zbornik za filologiju i lingvistiku 1 2 ( N o v i S a d , 1969) 115-165 ( + 10 plates). 27. 'Being' and 'having' in Estonian. F o u n d , of Lang. 5 (1969) 324-341. 28. D i p h t h o n g s versus vowel sequences iri Estonian. P I C P S VI. (1970) 539-544. 29. Suprasegmental features, segmental features, a n d long components. P I C L X (1970) 2-8. 30. (with Konstantin Popov). Akustische Analyse bulgarischer Silbenkerne. Phonetica 21 (1970) 40-48. 31. The quest f o r phonetic reality. In A r t h u r J. Bronstein et al., eds., Essays in H o n o r of Claude M. Wise. Speech Association of America (1970) 25-35. 32. (with P. Ivic). Prilozi ispitivanju fonetske i fonoloSke prirode akcenata u savremenom s r p s k o h r v a t s k o m jeziku, V. Zbornik za filologiju i lingvistiku 13 (1970) 225-246. 33. G r a m m a t i c a l variability a n d the difference between native and non-native speakers. In G e r h a r d Nickel, ed. Papers in Contrastive Linguistics. Cambridge Univ. Press (1971) 69-74. 34. T e m p o r a l organization of spoken language. In L.L. H a m m e r i c h , et al., eds., F o r m and Substance: Phonetic a n d Linguistic Papers Presented to Eli Fischer-Jergensen. Akademisk Forlag, Copenhagen (1971) 159-169. 35. Olemisest ja omamisest eesti keeles. M a n a 38 (1971) 87-92. 36. T e m p o r a l compensation in a quantity language. P I C P S 7 (1972) 929-937. 37. (with H. H e m a m i ) . Interactive a u t o m a t i c speech segmentation. In C.F. Smith, ed., Conference Record: 1972 International Conference on Speech C o m m u n i c a t i o n a n d Processing. (1972) Paper # H5, 291-294. 38. M a n n e r of articulation, parallel processing, and the perception of duration. U. of Essex Occasional Papers 13 (1972) 1-24.

List of

Publications

xvii

39. T h e units of speech perception. C h a p t e r 6 of J. Gilbert, ed., Speech a n d Cortical Functioning. Academic Press, New York and L o n d o n (1972) 187-235. 40. Three Estonian writers and the experience of exile. Lituanus 18 (1972) 15-31. 41. T h e timing of utterances a n d linguistic boundaries. J AS A 51 (1972) 2018-2024. 42. (with L. Shockey). On the perception of coarticulation effects in English VCV syllables. J S H R 15 (1972) 500-506. 43. (with P. Ivic). Experiments with synthesized Serbocroatian tones. Phonetica 26 (1972) 1-15. 44. (with P. Ivic). Prilozi ispitivanju fonetske i f o n o l o s k e prirode akcenata u s a v r e m e n o m srpskohrvatskom jeziku, VI. Zbornik za filologiju i lingvistiku 15 (Novi Sad, 1972) 95-113 ( + 4 plates). 45. Some observations concerning the third tone in Latvian. In Albert V a l d m a n , ed., Papers in Linguistics a n d Phonetics to the Memory of Pierre Delattre. M o u t o n (1972) 209-215. 46. Presuppositsioonidest ehk semantika j a g r a m m a t i k a piirimailt. Eesti K e e l e j a Kirjanduse Instituude A a s t a r a a m a t I. Stockholm (1973) 39-49. 47. The well-formedness of an Estonian folk song line. In A. Ziedonis, Jr., et al., eds., Baltic Literature and Linguistics. A A B S (1973) 135-139. 48. T a m m s a a r e , K a n g r o a n d the Devil. In A. Ziedonis, Jr., et al., eds., Baltic Literature a n d Linguistics. A A B S (1973) 69-74. 49. Where hobgoblins spend the night. J B S 4.4 (1973) 321-326. 50. (with D. Meitzer). Vowel and speaker identification in natural a n d synthetic speech. Lang, and Speech 16.4 (1973) 356-364. 51. (with Katherine M o r t o n a n d Marcel T a t h a m ) . An instrumental study of c o n s o n a n t gemination. J o u r n a l of Phonetics 1 (1973) 131-148. 52. Phonetic disambiguation of syntactic ambiguity. Glossa 7.2 (1973) 107-122. 53. Rhythmic units a n d syntactic units in p r o d u c t i o n a n d perception. J A S A 54 (1973) 12281234. 54. (with P. Ivic). Akusticni opis a k c e n a t s k o g sistema j e d n o c c a k a v s k o g govora. N a u c n i sastanak slavista u Vukove dane 3, M e d j u n a r o d n i Slavisticki Centar SR Srbije, Beograd. (1973) 159-170. 55. (with P. Ivic). Interaction between tone and quantity in Serbo-Croatian. Phonetica 28 (1973) 182-190. 56. Duration of syllable nuclei as a function of w o r d length and stress pattern. Proceedings of the Eighth International Congress on Acoustics. L o n d o n (1974) 300. 57. T h e syllable nucleus as a unit of timing. In Luigi Heilmann, ed., P I C L 11. Muline, Bologna (1974) 929-933. 58. The role of temporal factors in the establishment of linguistic units a n d boundaries. Wolfgang U. Dressier a n d F.V. Mares, eds., Phonologica 1972, Wilhelm Fink Verlag, München (1975) 115-122. 59. Experiments with synthetic speech concerning quantity in Estonian. Congressus Tertius Intemationalis F e n n o - U g r i s t a r u m . Tallinn (1975) 254-269. 60. Some factors affecting the d u r a t i o n of syllable nuclei in English. Salzburger Beiträge zur Linguistik 1 (1975) 81-104. 61. The stressed vowels of a speaker of standard G e r m a n . Semasia 2 (1975) 187-195. 62. The attitudes of bilinguals toward their personal names. American Speech 50 (1975) 30-35. 63. The phonetic structure of p a r a g r a p h s . In A. C o h e n a n d S.G. N o o t e b o o m , eds., Structure and Process in Speech Perception. Springer-Verlag: Berlin-Heidelberg-New York (1975) 195-206. 64. Influence of f u n d a m e n t a l frequency pattern on the perception of d u r a t i o n . J o u r n a l of Phonetics 4 (1976) 113-117. 65. Suprasegmental features of speech. C h a p t e r 7 in N o r m a n J. Lass, ed., C o n t e m p o r a r y Issues in Experimental Phonetics. Academic Press, New York (1976) 225-239.

xviii

List of

Publications

66. (with P. Ivic). F o n e t s k a analiza jedne slavonske akcentuacije. N a u c n i sastanak slavista u Vukove dane 6. M e d j u n a r o d n i slavisticki centar SR Srbije, Beograd (1976) 67-83. 67. Keelekontakt- keelekonflikt. M a n a 41 (1976) 5-14. 68. (with Joseph P. Olive and Lynn A. Streeter). Role of duration in disambiguating syntactically a m b i g u o u s sentences. J A S A 60 (1976) 1199-1202. 69. C o n t r i b u t i o n of pitch to the perception of segmental quality. Proceedings of the 9th International Congress on Acoustics, Madrid (4/9-VII-1977) 522. 70. Isochrony reconsidered. J o u r n a l of Phonetics 5 (1977) 253-263. 71. (with William S-Y. Wang). Perception of sentence boundaries with a n d without semantic information. Phonologica 1976, Innsbrucker Beiträge zur Sprachwissenschaft 19 (1977) 277-283. 72. Viron sanafonologian ongelmia. Virittäjä (1977) 178-186. 73. Variability in the production of suprasegmental features. In D. Sinor, ed., Studies in Finno-Ugric Linguistics in H o n o r of Alo R a u n . Bloomington (1977) 131-139. 74. Quantity in Estonian language and poetry. J o u r n a l of Baltic Studies 8 (1977) 130-141. 75. (with D o u g l a s G. D a n f o r t h ) . Foneettisten vihjeiden hierarkia viron kvantiteetin havaitsemisessa. Virittäjä (1977) 404-411. 76. (with P. Ivic). Interrelationship between word tone a n d sentence intonation in Serbocroatian. In D . J . Napoli, ed., Elements of Tone, Stress a n d Intonation. Georgetown Univ. Press, Washington, D . C . (1978) 100-128. 77. Language barriers and the poetry of Ilmar L a a b a n . J o u r n a l of Baltic Studies 9 (1978) 305-311. 78. Einige Beobachtungen über Wortgrenzen im Deutschen. In D . C . Reichel, ed. Wege der Worte: Festschrift f u r W. Fleischhauer. Böhlau-Verlag, Köln u n d Wien (1978) 70-75. 79. T h e syllable as a structural unit in Estonian. In A. Bell and J.B. H o o p e r , eds., Syllables and Segments. N o r t h - H o l l a n d Publishing Co., A m s t e r d a m (1978) 73-83. 80. Experimental studies in the phonology of discourse. In William F. Eadie, ed., Directions in C o m m u n i c a t i o n , 1978. Ohio University Center for C o m m u n i c a t i o n Studies, Special Report # 27 (Oct. 1978), Athens, Ohio, 17-44. 81. Polytonicity in the area surrounding the Baltic Sea. In Nordic Prosody: Papers f r o m a Symposium. Travaux de l'Institut de Phonetique de Lund 13 (1978) 237-247. 82. Perception of duration in sequences of f o u r intervals. J. of Phonetics 7 (1979) 313-316. 83. (with P. Ivic). A k u s t i c k i o p i s a k c e n a t a u j e d n o m k a j k a v s k o m g o v o r u . Z b o r n i k za filologiju i lingvistiku 22 (1979) 179-192. 84. A note concerning quantity in Finland-Swedish. In Sanomia. J u h l a k i r j a Eeva KangasmaaMinnin 60-vuotispäiväksi 14.4.1979. T u r k u : Dept. of Finnish a n d General Linguistics of the Univ. of T u r k u (1979), 129-135. 85. Some temporal aspects of spoken discourse. In J . J . Wolf and D . H . Klatt, eds., Speech C o m m u n i c a t i o n Papers Presented at the 97th Meeting of the Acoustical Society of America. Acoustical Society of America, New York (1979) 191-192. 86. T e m p o r a l relations within speech units. Symposium # 5 , S u m m a r y of M o d e r a t o r ' s Introduction. In E. Fischer-Jorgensen, J. Rischel, a n d N. Thorsen, eds., 9th International Congress of Phonetic Sciences (1979), Proceedings, Vol. II, U. of C o p e n h a g e n , 241-244. 87. Perception of sentence and p a r a g r a p h boundaries. In B. Lindblom en S. Ö h m a n , eds. Frontiers of Speech Research. Academic Press, L o n d o n (1979) 191-202. 88. Sentence boundaries a n d p a r a g r a p h boundaries-perceptual evidence. In T h e Elements: A Parasession on Linguistic Units a n d Levels. Chicago Linguistic Society, Chicago (1979) 99-109. 89. Translation f r o m Russian as a source of syntactic change in c o n t e m p o r a r y Estonian. In The Elements: A Parasession on Linguistic Units and Levels. Chicago Linguistic Society, Chicago (1979) 413-419.

List of

Publications

xix

90. Phonetic characteristics of discourse. Transactions of the C o m m i t t e e on Speech, S80-4, Acoustical Society of J a p a n , T o k y o (1980) 26-38. 91. Interaction between test word duration and length of utterance. In L. W a u g h and C. Schooneveld, eds., The Melody of Language. Univ. Park Press, Baltimore (1980) 169-76. 92. Temporal relations within speech units. Symposium # 5 , M o d e r a t o r ' s Introduction. In E. Fischer-Jergensen, J. Rischel, a n d N. Thorsen, eds., 9th International Congress of Phonetic Sciences (1979), Proceedings, Vol. I l l , Copenhagen (1980) 247-254. 93. Phonetic manifestation of syntactic structure in English. A n n u a l Bulletin, Research Institute of Logopedics a n d Phoniatrics, Univ. of T o k y o , 14 (1980), 1-27. Also published in: The T o k y o J o u r n a l of Medical Sciences, 89 (1982) 3-22. 94. (with Linda Shockey). Labelling, discrimination a n d repetition of stimuli with level a n d changing f u n d a m e n t a l frequency. J o u r n a l of Phonetics 8 (1980) 469-474. 95. Estonian linguistics: state of the art. General Linguistics 20 (1980) 194-208. 96. Suprasegmental constraints on segmental representation. Seminar 6, C h a i r m a n ' s S u m m a ry a n d C o m m e n t s . In Terry Myers, J o h n Laver a n d J o h n A n d e r s o n eds., T h e Cognitive Representation of Speech. N o r t h - H o l l a n d , A m s t e r d a m (1981) 161-163. 97. The role of prosody in the internal structuring of a sentence. In Shiro H a t t o r i and K a z u k o Inoue, eds., Proceedings of the X H I t h International Congress of Linguists, T o k y o , 1982. T o k y o (1983), 220-231. 98. (with P. Ivic). The phonetic n a t u r e of the Neo-Stokavian accent shift in Serbo-Croatian. In J. Peter M a h e r , Allan R. B o m h a r d and E.F. K o n r a d Koerner, eds.: Papers f r o m the 3rd International Conference on Historical Linguistics. J o h n Benjamins, A m s t e r d a m (1982) 197-206. 99. Sentence a n d p a r a g r a p h boundaries in Estonian. In O s m o Ikola, ed., Congressus Quintus Internationalis F e n n o - U g r i s t a r u m , T u r k u 20-27 VIII. 1980. S u o m e n Kielen Seura, T u r k u (1981) 164-169. 100. Phonetics. McGraw-Hill Encyclopedia of Science and Technology, 5th edition (1982) 159-162. 101. (with Hiroya Fujisaki). Some temporal and tonal characteristics of declarative sentences in Estonian. In Shiro H a t t o r i a n d K a z u k o Inoue, eds., Proceedings of the X I H t h International Congress of Linguists, T o k y o 1982, T o k y o (1983), p. 1313. 102. (with P. Ivic). The intonation of yes-or-no questions - a new Balkanism? Balkanistica VI (1980) 45-53. 103. Some phonetic characteristics of discourse. Studia Linguistica 36 (1982) 117-130. 104. Prosodic change in progress: Evidence f r o m Estonian. In I r m e n g a r d Rauch a n d Gerald F. Carr, eds., Language Change. Indiana Univ. Press, Bloomington (1983) 10-27. 105. (with Rimvydas Silbajoris). Marie Under's poetry: Some problems of translation. Lituanus 29.3 (1983) 44-59. 106. Signalling of syntactic structure in whispered speech. Folia Linguistica 17 (1983) 239-245. 107. G e r m a n loanwords in Macedonian. Makedonski jazik X X X I (1980), 113-119. 108. The many linguistic functions of duration. C h a p t e r 6 in J a m e s E. Copeland, ed., New Directions in Linguistics and Semiotics, Rice University Studies, H o u s t o n , Texas (1984) 96-122. 109. Mötisklusi eesti keele sisemise vormi kohta. M a n a 51 (1982) 11-14. 110. Semantics, Syntax a n d Prosody. Symposium 1, C h a i r m a n ' s S u m m a r y . In M.P.R. van den Broecke a n d A. Cohen, eds., Proceedings of the Tenth International Congress of Phonetic Sciences, Utrecht, 1983. Foris Publications, D o r d r e c h t - H o l l a n d (1984) 143-150. 111. The Estonian translation of the Elder Edda: Problems of metric equivalence. J o u r n a l of Baltic Studies 14 (1983) 179-184. 112. Some remarks a b o u t the temporal structure of the D E S E T E R A C " . Z b o r n i k z a f i l o l o g i j u i lingvistiku XXVII-XVIII (1984-85), 405-410.

xx

List of

Publications

113. (with Pavle I vie). Geographical variation in the perception of Serbocroatian short accents. In K.K. Shangriladze and E.W. Townsend, eds.. Papers for the V. Congress of Southeast European Studies, Belgrade, September 1984. Slavica Publishers, Columbus, Ohio (1984) 277-283. REVIEWS 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15. 16.

17. 18. 19.

20. 21. 22.

Peter S. Green, Consonant-Vowel Transitions. A Spectrographic Study. Travaux de l'Institut de Phonetique de Lund (1959). Phonetica 6 (1961) 248-251. H.T. Lutstorf, The Stressing of Compounds in Modern English. A Study in Experimental Phonetics. Buchdruckerei Walter Fischer, Bern (1960). Phonetica 6 (1961) 246-248. K. Hadding-Koch, Acoustic Phonetic Studies in the Intonation of Southern Swedish, Lund (1961). Language 39 (1962) 352-360. A.S. Abramson, The Vowels and Tones in Standard Thai, Bloomington (1962). Phonetica 9 (1963) 135-138. G. Ungeheuer, Elemente einer akustischen Theorie der Vokalartikulation. Berlin-Göttingen- Heidelberg (1962). Language 39 (1963) 584-586. Β. Malmberg, Phonetics. Dover Publications. New York (1963). IJAL 31 (1965) 101-102. W. Shakespeare, Collected Works in Seven Volumes, Part II and IV. Tallinn (1960 and 1961). translated into Estonian by Georg Meri. Shakespeare Quarterly 15 (1964) 236-238. W. Golding, The Inheritors. Current Anthropology 6 (1965) 232. C. -Ch. Elert, Phonologic Studies of Quantity in Swedish, Stockholm (1964). Word 21 (1965) 331-334. A. Flew, ed.. Logic and Language, Doubleday (1965). J S H D 31 (1966) 97. D. Abercrombie, et al., eds.. In Honour of Daniel Jones, Longman, Green and Co., London (1964). IJAL 32 (1966) 412-414. David Abercrombie, Elements of General Phonetics, Edinburgh Univ. Press (1967). IJAL 34 (1968) 119-221. Paul Rummo, ed., Eesti luule, Kirjastus 'Eesti Raamat', Tallinn (1967). Books Abroad 43 (1978) 607. Felix J. Oinas, Basic Course in Estonian. IUPUAS 54, Bloomington (1966). Language 44 (1968) 409-411. Georges Straka, Album phonetique, Les Presses de l'Universite Laval, Quebec (1965). Romance Philology 22 (1968) 36-39. Peter Rehder, Beiträge zur Erforschung der serbokroatischen Prosodie, Slavistische Beiträge 31. Otto Sagner, München (1968). The Slavic and East European Journal 13 (1969) 406-410. Viivi Luik. Lauludemiiüja, 'Eesti Raamat', Tallinn (1968). Viivi Luik, Hääl, 'Perioodika', Tallinn (1968). Books Abroad 43 (1969) 447-448. Eva Gärding, Internal Juncture in Swedish, Travaux de l'Institut de Phonetique de Lund, VI. C.W.K. GJeerup, Lund (1967). Phonetica 21 (1970) 117-119. A.V. Isacenko, Sprachwissenschaft und Akustik, Sitzungsberichte der deutschen Akademie der Wissenschaften zu Berlin # 4. Akademie-Verlag, Berlin (1966). Foundations of Language 7 (1971) 437-438. Bobuslav Hala, et al., eds.. Proceedings of the Sixth International Congress of Phonetic Sciences, Prague (1970). JASA 49 (1971) 1123-1124. Georg Heike, Suprasegmentale Analyse, Marburger Beiträge zur Germanistik, Bd. 30, N.G. Elwert Verlag, Marburg (1969). Phonetica 24 (1971) 50-56. Leena Lehto. English Stress and Its Modification by Intonation: An Analytic and Synthetic Study of Acoustic Parameters. Suomalainen Tiedeakatemia, Helsinki (1969). Language 48 (1972) 190-195.

List of Publications 23. 24. 25. 26. 27. 28. 29.

xxi

Bernard Kangro, Joonatan, kadunud veli, Eesti Kirjanike Kooperativ, Lund, Sweden (1971). Books Abroad 46 (1972) 706-707. Peter Ladefoged, Preliminaries to Linguistic Phonetics, Univ. of Chicago Press (1971). JASA 53 (1973) 112-113. Henry Sweet, A Handbook of Phonetics. G. Oscar Russell, The Vowel. McGrath Publishing Co., reprinted 1970. IJAL 39 (1973) 123-126. Karl Ristikivi. Kahekordne mäng. Books Abroad 47 (1973) 394-395. George Kurman, Literatures in Contact: Finland and Estonia. Estonian Learned Society in America, New York (1972). Books Abroad 47 (1973) 597. S.G. Nooteboom, Production and Perception of Vowel Duration. A Study of Durational Properties of Vowels in Dutch. Utrecht (1972). Journal of Phonetics 1 (1973), 181-183. L.L. Hammerich, et al., eds., Form and Substance: Phonetic and Linguistic Papers Presented to Eli Fischer-Jergensen, Akademisk Forlag (1971). Phonetica 27 (1973) 125128.

30. 31. 32. 33. 34. 35. 36.

37. 38. 39. 40. 41. 42. 43. 44. 45.

46. 47. 48.

Klara Magdics, Studies in the Acoustic Characteristics of Hungarian Speech Sounds, IUPUAS 97, Indiana Univ., Bloomington (1969). Linguistics 29 (1974) 115-123. Hans-Heinrich Wängler, Physiologische Phonetik, N.G. Elwert Verlag, Marburg (1972). JASA 55 (1974) 377. Bernard Kangro, Öö astmes x, Lund (1973). Books Abroad 48 (1974) 399. Gerhard Neweklowsky, Slowenische Akzentsiudien, Verlag der Österreichischen Akademie der Wissenschaften, Wien (1973). Journal of Linguistics 11 (1975) 155-157. Jana Ondräckovä, The Physiological Activity of the Speech Organs, Mouton, The Hague (1973). Journal of Linguistics 11 (1975) 151-155. Odön Lavotha, Kurzgefasste estnische Grammatik. Veröffentlichungen der Societas Uralo-Altaica, Band 9, Otto Harrassowitz, Wiesbaden (1973). Die Sprache 21 (1975) 207. Edward T. Purcell, The Realization of Serbo-Croatian Accents in Sentence Environments, Hamburger phonetische Beiiräge, Band 8. Helmut Buske Verlag, Hamburg (1973). Phonetica 33 (1976) 142-145. Max Mangold, Phonetic Emphasis: A Study in Language Universals. Forum Phoneticum 10, Helmut Buske Verlag, Hamburg (1975). Phonetica 33 (1976) 467-471. Ly Seppel, Varjuring timber tule. 'Eesti Raamat' Tallinn (1974). Mana 41 (1976) 86-88. A. Eek, ed., Estonian Papers in Phonetics: EPP 1975. Academy of Sciences of the Estonian SSR, Institute of Language and Literature, Tallinn (1975). Phonetica 34 (1977) 467-471. Urve Karuks, Kodakondur, Mana, Toronto (1976). World Literature Today 52 (1978) 147-148. Sadanand Singh and Kala S. Singh, Phonetics: Principles and Practices. University Park Press, Baltimore (1976). JASA 63 (1978) 1248. Debora Vaarandi, Tuule valgel, Kirjastus 'Eesti Raamat', Tallinn (1977) World Literature Today 52 (1978) 664. Henn-Kaarel Hellat, Naiste Maailm I, II, Kirjastus 'Eesti Raamat', Tallinn (1976 and 1978). World Literature 53 (1979) 323-324. Ellen Niit, Maailma pidevus: Luuletusi aastaist 1946-1976. 'Eesti Raamat', Tallinn (1978). World Literature Today 53 (1979) 715. Masayuki Sawashima and F.S. Cooper, eds., Dynamic Aspects of Speech Production: Current Results, Emerging Problems, and New Instrumentation. U. of Tokyo Press (1977). Language 55 (1979), 957. Urve Karuks, Kodakondur. Mana, Toronto (1976) Mana 46 (1979), 65. Aarne Biin, Tema kuninglik körgus. Kirjastus 'Esti Raamat', Tallinn (1979). World Literature Today 54 (1980), 461. Aino Pervik, Kaetud Lauad. Kirjastus 'Eesti Raamat', Tallinn (1979). World Literature Today 54 (1980) 461.

xxii 49.

50. 51. 52. 53. 54. 55. 56. 57. 58. 59. 60.

61.

62. 63. 64.

List of

Publications

Estonian Papers in Phonetics. Studies on Accent, Quantity, Stress, Tone. Papers of the Symposium (Tallinn, Nov. 1978). Academy of Sciences of the Estonian SSR. Institute of Language and Literature, Phonetica 37 (1980) 278-281. Mats Traat. Valitud luuletused. Kirjastus 'Eesti Raamat', Tallinn (1979). World Literature Today 55 (1981) 152. Bernard Kangro, Kuus päeva. Eesti Kirjanike Kooperatiiv, Lund, Sweden (1980). World Literature Today 55.2 (1981), 351. Mari Saat, Laanepüü. Kirjastus 'Eesti Raamat', Tallinn (1980). World Literature Today 55 (1981) 502. G. Brown, K. Currie, and J. Kenworthy, Questions of Intonation. University Park Press, Baltimore (1980). JASA 70 (1981) 648-649. Kalju Lepik. Kogutud luuletused. Eesti Kirjanike Kooperatiiv, Lund (1980). World Literature Today 55 (1981) 703. William E. Cooper and John M. Sorensen, Fundamental Frequency in Sentence Production. Springer-Verlag, New York (1981). General Linguistics 22 (1982) 123-128. William E. Cooper and Jeanne Paccia-Cooper, Syntax and Speech. Harvard Univ. Press, Cambridge.(1980). Contemporary Psychology 27 (1982) 886-887. Jiiri Talvet, Äratused. Kirjastus 'Eesti Raamat', Tallinn (1981). World Literature Today 57 (1983) 138-139. Leelo Tungal, Mullaketraja. Kirjastus 'Eesti Raamat', Tallinn (1981). World Literature Today 57 (1983) 139. Viivi Luik. Rängast röömust. 'Eesti Raamat', Tallinn (1982). World Literature Today 57 (1983) 664. Kulturelle und sprachliche Minderheiten in Europa: Aspekte der europäischen Ethnolinguistik und Ethnopolitik. Akten des 4. Symposions über Sprachkontakt in Europa, Mannheim 1980. Herausgegeben von P. Sture Ureland. (Linguistische Arbeiten, 109). Niemeyer, Tübingen (1981). Language 60 (1984) 169-171. Eesti-inglise sönaraamat: Estonian-English Dictionary. Compiled by Paul F. Saagpakk. With an Introduction by Johannes Aavik. Yale Univ. Press, New Haven and London (1982). Language 60 (1984) 459-560. Madli Morell. Tösimäng. 'Eesti Raamat', Tallinn (1983). World Literature Today 58 (1984), 641. Mari Vallisoo. Rändlinnud körvaltoas. Kirjastus 'Eesti Raamat', Tallinn (1983). World Literature Today 59 (1985) 130. Raimond Kolk. Vööral maal kirjutatud laulud. Eesti Kirjanike Kooperatiiv, Lund, Sweden (1984). World Literature Today 59 (1985) 295.

Twelve Remarks on the Teaching of Phonetics Robert Austerlitz, Columbia University, New York, N.Y.,

U.S.A.

The following twelve points are side-products of teaching introductory phonetics for about twelve years, mostly to interested parties - students in linguistics. Some of these students develop a lasting antipathy toward the subject, which they never manage to shed. Is this because of the textbooks they are forced to read? The twelve remarks are, in part, attempts to present certain chapters in articulatory phonetics in a more reasoned way and in part suggestions for looking at some old truths in a new way. They are not intended as a flow chart for a course in phonetics. 1. Introduce the two resonance chambers, buccal (chamber 1) and pharyngo-laryngal (chamber 2), as early as possible. Insist that they are situated, purely schematically speaking, at right angles to each other. Show how the two chambers are complementary with regard to the position and hence also the bulk of the tongue: if the tongue is in the back, then there is space in front, and vice versa. ('Front' and 'back' do not refer to the traditional front and back vowels in this connection.) 2. Introduce F 2 as visually describing an abrupt descent and F1 as describing an arc under F 2 - for /, a, u, in that order, f r o m left to right. Explain that all the other vowels are accommodated within the latitudes described by i,a, and u, within both the articulatory triangle and the relationship between F l and F 2 for each vowel. Show how roundedness (labiality) is related to the non-low back vowels in their very production: the wider we open our mouth or jaw (say, for the lowest back vowel, a), the less able we are to round our lips; conversely, as we go from α to ο to u, rounding becomes easier. See point 4, below. 3. Discuss, teach, and use the IPA system as a useful set of coordinates which has been sanctioned internationally. But discredit the IPA system and any other system with pretensions of world-wide coverage of sounds - as early as possible. Complete isomorphy between script and articulatory gestures is impossible. Insist that all symbols used in phonetics are mere ad hoc tokens and have no more than momentary validity, based on a temporary contract between the writer and the reader. Encourage the use of self-made transcription systems, preferably with arbitrary symbols which have no relation to alphabetic value, e.g., a flower for a voiced velar fricative. Then

2

Robert Auster litζ

describe Alexander Melville Bell's non-arbitrary "self-interpreting physiological letters, for the writing of all languages in one alphabet." This is the latter half of the title of Bell's Visible Speech: The Science of Universal Alphabetics of 1867. Finally, return to the two chambers mentioned in 1. 4. Introduce the vowels first as products of tongue height and forwardness in chamber 1. Use the high vowels i and & (where & stands for the high, back, unrounded vowel - Turkish dotless /', unfortunately also written [ui] and thus confused with a Swedish vowel irrelevant here) first. Once the articulation of &, so exotic to the West-European ear and tongue, is absolutely clear, and only then, introduce rounding. Rounding is no more than a specific, clearly identifiable exit for / and &. It accounts for the proportion i : ü :: & : u. Therefore u is no more and no less than & with a tapered and hence optimal channel for its egress. Why optimal? Because u, with a large chamber in the front of the mouth, profits from an exit with a bottle-effect. The vowel & which does not enjoy this benefit (of a bottle-effect) is, in fact, a contradiction in terms: large chamber in the front of the mouth and large exit, also in front. That is why the IPA's classical distinction between primary and secondary vowels is useful and ought to be introduced at this point. The secondary vowels, & and ii, are less prominent in the world's languages and generally occur in addition to the primary ones (/and w) in vowel systems. The primary ones are prominent in the world's languages because they are natural. Their naturalness resides in the following: The resonance chamber for i is in the back; i therefore needs no help in exiting from chamber 1. The resonance chamber for u is in the front of chamber 1; it needs concentration (a tapered exit, rounding) in order to issue from it. Once this is clear for the high vowels, test it against the lower vowels and show why the features which distinguish i from u become less dramatic: open jaw (tongue height) and rounding are at odds with each other. Or: the more one opens one's mouth, the less able one is to round one's lips. 5. Explain that the notions of the affricate and the diphthong are related but that the terminology is lopsided: in affricates we reckon with an initial shorter segment (the stoppage), followed by a longer segment (the allegedly homotopic friction) - c is said to be a sequence of t and s. In diphthongs, the inverse is generally the case: it is the long segment a which precedes the short segment w in the diphthong aw. Thus

long portion of unit c o m e s first

comes second

consonants

st

1

ts = c

2

vowels

aw

3

wa

4

On the Teaching of Phonetics

3

The unit in cell 4 is now seldom called or considered a diphthong. The unit in cell 1 is never called an affricate. The terminological bias may come from the history of these sounds in West-Germanic, e.g., German and English aw from m, German c from t. However, the mirror images of aw and c, that is wa and st, deserve at least a name or, lacking that, discussion. - To try to distinguish "vowel plus glide" from "diphthong", finally, is both futile and impossible, unless we mix interpretation (phonemicization, structuralizing) with pure phonetic description. 6. Admit that the notion of the affricate itself contains a trace of legerdemain if we insist that the stop component and the fricative component of the typical affricate be completely homotopic (homorganic). This is more or less accurate in the peripheral consonants, in the affricates corresponding to ρ and k.-ρφ and kx (where the first of these is labiodental throughout and not pf as in German). It is not accurate in the case of pf and of those affricates which consist of an apico-alveolar or apico-dental stop component followed by hissing, hushing, or lateral friction: c, c, % (the last as in Tlingit). In these, the stoppage (impediment, occlusion) is made by one portion of the apico-laminal apparatus which is dismissed as soon as the frictional release - which is made by other parts of the tongue - sets in. Or: the channel through which the air passes during friction is not necessarily formed by those portions of the tongue which perform the initial occlusion. Test this with an ideal apicodental affricate, unprofessionally t9: the stop component is observably apical but the friction is in the post-aprical area. Furthermore, the stoppage is made against the blade of the tooth or an area close to the blade of the tooth; the friction, however, is produced further up and farther back, against the inside wall of the tooth. - Discuss the palatal affricate (IPA c) and the question whether it is an affricate or a stop, or show that the affricate/stop distinction is at its least pronounced and hence most intractable in the palatal region. - Discuss the prominence of s in the languages of the world. It is more prominent (statistically frequent) than the affricate which corresponds to it. It is also the one prominent fricative which, from the point of view of system, needs no implicative support from a stop as φ, θ, χ do from ρ, t, k (e.g., "x is only found in languages with k"). 7. Explain that giving j as the voiced pendant of the ich- Laut is wrong. The voiced fricative corresponding to voiceless (IPA) ς is the sound of the digraph ch in German ich bin as pronounced by a Russian: with voice emanating from b to its neighbor to the left, j is in a completely different class. Reserve the symbol [j] for the glide. 8. Explain that / is not the voiced pendant (companion piece) of voiceless I as in Welsh or Tlingit. This lateral voiceless fricative I can be voiced; its IPA symbol is [£]. / is not a fricative. It is in a class by itself, one which more closely resembles that of the resonance chambers, vowels. 9. Distinguish between two basically different varieties of secondary arti-

4

Robert Auster lit ζ

culation: that of the consonantal type, such as aspiration and glottalization, and that of the vocalic type, such as labiovelarization, palatalization, labiopalatalization, velarization, pharyngalization. The latter are resonance chambers - vowels - made for u, i, ü, & (see 4, above), and for a pharyngal vowel of the f - t y p e . Velarization and pharyngalization often collaborate (participation of the posterior part of chamber 1 and the upper part of chamber 2), as in the Arabic emphatics ( m u f a x x a m a ) . Pure labialization, without the accompanying support of a velar component, is unlikely to occur because of the weakness of mere lip rounding. These types of (vocalic) secondary articulation are, then, instances of setting the scene (the coarticulation of a resonance chamber) for a certain vowel during or before or after the production of a given consonant type. Likes will exclude likes, i.e., velarized velars are less likely to occur than velarized dentals. This is a kind of incest prohibition (or: no ketchup on chicken cacciatore). Voiced fricatives lend themselves much better to the description and demonstration of secondary articulations of this kind than the proverbial p,t,k. That is: use ν or ö or ζ to demonstrate velarization, palatalization, etc. - In this connection, set up two w's: one consonantal, purely labial, akin to b\ and another one equally consonantal but with a vocalic pedigree, labiovelar, akin to the vowel u. - The term "secondary articulation" is therefore overladen. It refers to a well-defined vocalic resonance chamber which accompanies the production of a given consonant on the one hand and, on the other, it refers, more loosely, to a number of consonantal articulations (glottal friction, glottal stop) which accompany other consonantal articulations. The term " a c c o m p a n y " as used here requires further specification. 10. Exploit the notion of approximant to show the intersection of the notions of vowel and consonant. Approximants, as the term is used here, refer to obstructions (consonants) in statu nascendi, articulatory gestures which never realize the obstruction aimed at. They are Platonic ideals. In Spinozan terms, they are natura naturans while pure vowels and pure consonants (stops) are natura naturata. In this way, we can envisage a vowel with an articulatory chamber almost but not completely achieving the articulation of any obstruent, e.g. Ö (and its congeners 0,d,t) All buccal articulatory configurations (chamber 1) for the production of consonants can also occur in their approximant facet. The difference resides in the fact that contact between articulator and place of articulation is aimed at but never achieved in the case of the approximants. This approach can help explain "r-color," velarization - in fact, the intersection of any two articulations. It can also explain why there is no medial friction in the normal, unguarded pronunciation of Spanish hago Ί make'. The letter g represents an approximant here. 11. Show how in chamber 2, both the voice/voicelessness distinction and the consonant/vowel distinction become less operative than in chamber 1.

On the Teaching of Phonetics

5

CHAMBER 2

CHAMBER 1

Pharyngals

Proverbials I II III IV

Ρ b

t d

Φ ß

k

-

•

Laryngals 9

I

θ

g X

tl

h

II III

ö

y

r

fi

IV

-

The sound-type represented by the symbol ? is especially vexing. Since complete closure is impossible in the pharynx ? cannot represent a stop. For this reason it is placed among the fricatives (row IV). But sustained pharyngal friction-plus-voice also requires extraordinary effort and is therefore rare. The 9 which thus results may be produced in spurts. These spurts resemble a series of stops. Therefore ? would beg to be transferred to row II, among the stops, into the spot marked *. Pharyngals and laryngals thus form a strange subsystem, consisting of one or two stops (? alone or ? and f ) and three or four fricatives (h, fi, h with or without f ). This yields more fricatives than stops - a situation unique to chamber 2 and one which characterizes it in contradistinction to chamber 1. 12. Insist that what are called the r-sounds constitute a grapic rather than an articulatory class: they are written with the symbol r in European languages. The production of these sounds involves loci ranging from the lips to the velum and beyond. In the schematic typology provided below, I and II are natural classes - places and manners of articulation. Column F is post-velar ("uvular"), etc., as indicated. 2B and 3B should accommodate Japanese hara 'abdomen', British very, and American Betty, IB the Spanish initial or strong-medial -rr-, 2D and 3D the cacuminal r's of the Indian subcontinent, 5D the syllabic and sustained r in bird, 4E and 4F the Parisian and Danish fricatives, and IF and 2F the Viennese uvular trills and flaps. 1H is a glottal tremulant, i.e., murmured voice or plain voice. 2C and 3C are too close to 2B and 3B to be distinguishable in practice but deserve to be pointed out alveolum vs. upper teeth. 1A presumably does not occur as a speech sound but it is certainly produced. 4A and 5A are surrogates or pathological ("wabbif pro rabbit, with or without bilabial friction). 4B and 4C occur as unsuccessful attempts on the part of learners, children and adults, to acquire trills, flaps or taps (1, 2, 3) in the apico-alveolar area. (This is one step away from the substitution of ζ for r; one such case is reported from Finland.) 5G, a pharyngal constriction which does duty as r is sometimes found in Europe. It also occurs as a pathological surrogate for pronouncedly velar / - among some American speakers - especially in closed syllable, where the velarity of

6

Robert Auster litζ I v. "3 >

Λ Ι-» Ο

bilabial

'α.

apico-dental

retroflex

velar

J3

A

Β

C

D

Ε

υ

j>

II

tremulants, trills

1

flaps

2

taps

3

fricatives

4

approximants, vowels

5

Λ I Ο Ö V2

falling

rising

/ Λ 2 3

5

1 1 * 3

I^

ν

1

I I I I ι '

'

'

23

1

1

1

1

ι ι

Ό ft «JÖSSOl£>' r > lr > , ol5 1D

falling 1 3

Estonians Russians

/ \

Α

νΝ' / '

11

• ο

rising

i 1 J

3

I 3

2

J

Figs.

4c and 4d. Listeners reporting stress on first syllable for varying pitch patterns of V 2 (Duration of V, = 123 msec., V 2 = 183 msec. For 4c, V, rises during 1/3 of its duration, then falls; for 4d, V, rises for 2 / 3 of its duration, then falls.)

syllable II blends with the peak of the first-syllable rise, so that, e.g., an Estonian may analyze V,=105-125 Hz and V 2 =135-l 16 Hz and analogous cases as one contour of 105-135-116 Hz. In the case of a rise + fall, the ear cannot discriminate the higher initial pitch of the fall from the lower final pitch of the rise. It also looks plausible that if such blending does take place the rise has a greater effect on stress perception than a fall with an equal pitch peak or a corresponding level tone. This might account for the significant differences in the Estonian responses observable in the left-hand part of Figures 4a and 4b. In words with a rise + rise (e.g. V,=105-125 Hz, V 2 = l 16-135 Hz, as is the case in Figure 4b, center), where the pitch peaks of the two syllables are placed farther from each other, stress again starts to be associated with the syllable which has the higher peak (syllable II is perceived as stressed even where the peaks are equal, since the longer duration of V 2 adds another point in favor of stress perception). Scores remain low, however, first because, in a

Word Stress in Estonian and Russian

29

- — pattern, a rise in syllable II is untypical of Estonian, and second because in -—syllable I sounds too high to be a proclitic. The right-hand part of Figure 4b calls attention to the fact that a contour of 2/3 rise and 1/3 fall will be understood by an Estonian listener as a rise, while 1/3 rise and 2/3 fall will be interpreted as a fall (cf. Purcell, 1976). Thus, syllable II was perceived as stressed in the case of V2—116-135-116 Hz (2/3 1/3) as often as in V 2 = l 16-135 Hz, while if V2 is 116-135-116 (1/3-2/3), syllable I is marked as stressed just as often as in the case of V 2 = 135-116 Hz. The latter case again seems to imply that the peaks of syllable I and syllable II merge into a rise-fall, as was surmised from the left hand stimuli in Figure 4b. Figures 4c and 4d also seem to support the conjecture that Estonians may well consider a 2/3 rise a rise and a 1/3 rise a fall (cf. the responses to stimuli with identical V2 F 0 characteristics in Figures 4a and 4b respectively). As for Russians, they consistently mark the longer syllable II as stressed, paying no particular heed to F 0 patterns. In stimuli with a reversed duration pattern (V!=175 msec., V 2 =124 msec., F 0 contours of and V2 the same as in Figure 4), syllable I is always marked as stressed both by Estonians and Russians (which spared us a diagram here). For Estonians was the only duration pattern where the higher-pitched syllable (here a short syllable II) was not perceived as stressed, the pattern — - being uncharacteristic of Estonian. We can therefore conclude that the perception of stress by Estonians corresponds to Hypothesis 2, while that by Russians is in accord with Hypothesis 3. In Figure 5 tone contours familiar from Figure 4 are used to study their effect on stress perception if V! and V2 are of equal duration (V,= 175 msec., V 2 =172-180 msec., the fluctuations being due to F 0 differences). From Figure 1 we may remember that stress responses elicited by V, and V2 with equal duration and similarly falling contours displayed a slight bias in favor of syllable I. If, however, the fall of V2 was in a lower region than that of V,, the duration of the vowels being more or less the same (Figure 2), the stress score of syllable I was very high indeed. In Figure 5a (Vi=125-105 Hz), syllable I also gets a high score of stressed responses if its FQ is higher, while equal falls in both syllables make the prevalence of syllable I quite small. Whereas a higher fall in syllable II will increase the perception of stress on that syllable, a rise has even more power to do so, though still failing to render the predominance of syllable II convincing. In the case of equally long V ; and V2, syllable II probably cannot be perceived as stressed unless the pitch of syllable I is notably lower (the so-called "phrase-stress pattern" -). Figure 5b serves to corroborate once more our supposition about the blending of the peak of the first syllable rise with that of the second syllable fall, as well as the fact that whether a contour is interpreted as a rise or a fall depends on the place of its peak on the time axis (cf. Figure 4b).

30

Arvo Eek V1 = 105—125Hz VI' 175 ms V2=m^180rns

V1 * 125—105Hz V1 = 175 ms V2 = 172 + 180ms

I I I I I 1 V2= G i ^ S ^ S s ^ ^ S i ü S s s ü i S i S ö S

level

falling

rising

to

ΐ>

Η es ΡΗ Ε Η

TT CS oo CS CS ON ο CS c s CS ^ r - CS Ov CS c s ^H ο CN CS c s CS m CO Ov m Ι 1 ·-Η CS ο CN CS CS CS

_

t)

ω

c s CS ON σ \ ^H ο »-H CS CS CO vo Ov '— 1 CS c s vo c s ο Ov ^H Η— ^H CN CS

ω

ro Ov Ο »-H »-H vo ON ο cs — vo ON ο CS

vo 00 •-H CO oo -H oo r-

oo oo ^H ΓΟΟ —

•—·

•—"

> i > - i s 8 d ( d ö o 3

s

IX

CS ^Η CO

CS ^Η

German Stressed

131

Monophthongs

could be argued that the F2 target should be preferred to the F t target, and that TVOL is not appropriate for the characterization of vowel quality. Fig. 3 shows the differences for all four formants at the F! and F 2 targets (left) and also at the F 2 target and the maximum volume point (right). These differences are presented numerically in Table II, where the values of the vowel types are pooled and only the means of the formants are given. Table 77. Comparison of the differences between the formant mean values (Hz) obtained at different target positions. Standard deviations are given in parentheses.

TFl«TF2 TF2 « TVOL TFl « TVOL

F,

F2

F3

F4

X

26.4 (19.2) 14.5 (12.8) 23.1 (14.7)

51.6 (28.9) 105.3 (92.7) 61.4 (86.7)

54.9 (36.1) 91.2 (60.2) 60.7 (50.4)

48.7 (38.4) 64.1 (37.9) 39.6 (38.0)

45.4 (30.7) 68.8 (50.9) 46.2 (47.5)

The common mean of the formant differences between the values at TF1 and TF2 is 45.4 Hz, that between the values at TF2 and TVOL 68.8 Hz, and that between the values at TF1 and TVOL 46.2 Hz. The qualities of [Y] and [0:] seem to be similar (cf. Fischer-Jorgensen, 1975; Sendlmeier, 1981). The very fronted position of [e:] was observed earlier by Jergensen (1969). 5. THE ANALYSIS OF THE SAME MATERIAL BY RTA A N D FFT

The same 150 vowels were also analyzed by means of RTA and F F T (cf. the technical description above). The results obtained from these measurements are presented in Table III, and a comparison of the three methods is given in Table IV, using the means of the differences. For the comparison the eigenvalues of the differences between the methods with respect to every vowel type were calculated first, and then the means and standard deviations for each formant. The methods are compared in pairs. The formant means are also indicated (column χ on the right). Because the differences in F4 values were considerably greater than those in other formants, the common mean was calculated separately for the formants F,, F 2 and F 3 (column F ^ F j ) . Table IV shows that the mean differences among the three main methods are almost equal (57.1, 57.5 and 59.5 Hz). The internal dispersion within the three F F T alternatives is smaller (25.7,35.1 and 32.1 Hz). The LPC results lie between these two groups if we consider the means obtained between TF1 and TF2 (Table II, 45.4 Hz). On the F , / F 2 plane the points obtained by RTA and F F T lie farther apart from each other than those obtained by RTA and LPC or by F F T and LPC.

132

Antti

Iivonen

German Stressed

ο ο ο Ί-

ο ο ο Ί-

ο ο

Ο Ο

ο ο

133

Monophthongs

ο ο ο ΓΊ

ο ο

ο ο ο

ο ο

ο ο ο

ο ο

ο ο ο

ο ο >Λ

ο ο

ξ.bi

134

Antti livonen

6. SUMMARY A N D DISCUSSION

The microstructure of 150 occurrences of 15 German monophthong types analyzed by means of the LPC technique showed no 'stationary' middle portions; the formants were bending in most cases during the vowel nucleus (Fig. ID). This is partly due to the 1 Hz frequency resolution of LPC and the relatively high, changing F 0 of the vowels. But it also seemed evident that the formants were regularly rising, rising-falling or falling-rising, in addition to 'level' cases. F! reached its target at the same time point as F 2 only in a few cases. The time interval between Fj and F 2 was 32.1 ms on the average (standard deviation 19.1 ms). The average intervals for different vowel types are presented in Table I. According to the F ^ F j plane the vowels form a more peripheral system at the F 2 target than at the F, target (Fig. 2). The quality seemed to be more centralized at the F, target and even more at the maximal volume point (in the latter case because of the back vowels). The differences in the formant values at the three time points are presented graphically in Fig. 3. The overall mean of the differences observed between the F t and F 2 targets was 45.4 Hz (Table II). The formant structure at the maximal volume point resembled that at the Fj target more than that at the F 2 target (differences 46.2 Hz and 68.8 Hz on the average). The differences in the values obtained by means of the different analysis techniques were greater. The differences in the F 4 values were especially noticeable (Table IV). When only the formants F l s F 2 and F 3 are considered the pairwise differences between the methods were as follows: 57.5 Hz 57.1 Hz 59.5 Hz

LPC (at F 2 target) ® RT A RTA«FFT3 FFT3«LPC

We cannot explain the causes of the differences exactly; presumably all steps of the analysis chain contribute to the variation. Note, e.g., that for practical reasons the speech material had to be fed into the three analyzing systems through different tape recorders. The RTA values are systematically slightly higher than the F F T and LPC values. The frequency resolution of RTA was greater (40 Hz) than that of F F T (16 Hz) and LPC (1 Hz), the time window was variable, etc. The internal comparison among the three F F T alternatives showed that their differences were smaller than those observed among the three main methods (see Table III). The differences were 25.7, 35.1, and 32.1 Hz. The effect of the time window (one glottal period or five periods) thus seemed to be less than that of the alternative target (cf. the LPC analysis). Note, however, that the F 1 of [e:] and [ε] was not measurable with FFT1, because of the split formant. Statistically and practically it seems to be insignificant whether formant asymmetry is taken into account or not. Although there is a difference of 32.1 Hz between the means obtained by FFT2 and FFT3, other causes of variation in other measurement types yield equal differences.

German Stressed υ Oh

Η

Monophthongs

m

et

OÄ

U 0 -

t u

Η tu Pu

< Η Di

U P-

H t u

< Η

OS

U CL, J

m

s o

r sn c i

i n f )

m rη

>n 00 sO Cl

«n Tt sO r«"i

Ο r -

m f—t

SO 00

00 f»>

P-.

< Η

ο Ο r ΓΊ

CN r -

fN fN

i n OS Ο fN

sO •5t fN

τΤ ο ΓΟ

OS so fN

OO 00 fN

Os fN

Ο 00 m

OS c s

Η Ü-, U-

< Η OÄ

_

Os «1

>-

r Os m

13

8

Ö

SO

cd

TT fN m

fN u-i

Ο

SD

i n

136

Antti Iivonen

Table IV. Comparison of the differences between the mean formant values (means and standard deviations in Hz) obtained by the different methods (RTA, FFT3, and LPC), and also for three options within the FFT analysis. The measurement of the average differences was based on the eigenvalues. LPC values are from the F2 target. FFT3 represents the FFT analysis in the above comparison, since it most resembles the other two main methods theoretically. F,

F2

F3

F4

χ

F,-F 3 (X)

65.1 (28.4) 36.5 (25.3) 52.6 (38.6)

49.0 (18.8) 56.4 (35.3) 47.3 (37.3)

57.3 (32.5) 79.5 (30.8) 78.5 (63.6)

94.5 (58.1) 145.5 (94.4) 87.9 (100.6)

66.5

57.1

79.5

57.5

66.6

59.5

FFT I, FFT2, and FFT3: 29.6 F F T 1 « FFT2 (16.8) F F T 1 « FFT3 33.0 (26.5) F F T 2 = FFT3 41.3 (16.5)

17.0 (15.2) 32.1 (30.6) 32.0 (18.8)

30.4 (24.7) 40.2 (36.8) 22.9 (21.9)

131.7 (152.7) 146.1 (167.6) 43.3 (48.2)

52.2

25.7

RTA, FFT3, and LPC: RTA « FFT3 RTA « LPC FFT3

LPC

62.9

35.1

34.9

32.1

In summary, the three main methods RTA, F F T and LPC yielded similar average formant values, although differing ca. 57-60 Hz. The conclusion is plausible that within a method the relative results are sufficiently accurate, but the direct comparison of results obtained by different methods must be done with caution. This particularly concerns studies which deal with very precise vowel qualities, such as in sociophonetics. It seems to be most adequate to measure the formant structure in monophthongs in which the vowel quality does not change over time (contrary to the glide vowels) at the F 2 target, because then a more peripheral vowel F / F j dispersion can be obtained than at the F, target. Because several variables affect the formant microstructure (e.g., the interaction between the F 0 and formant resonance, variation in the repetitions of the same word, etc.), it does not seem inappropriate to measure the formant structure either at the F[ target or at the temporal mid-point of the vowel. It also seems plausible to average the middle portion of the vowel (ca. 30 ms) and to measure the formants from the averaged spectrum. If we estimate the differences reported above from the perceptual point of view and base our assumption of the perceptually relevant difference on experiments which show that the DL of the formants is 3-4% (Nord and Sventelius, 1979), we can conclude that the formant values can be measured with better resolution than is needed perceptually for formants above 1000

German Stressed Monophthongs

137

Hz (DL > 30-40 Hz). But ifthefonnants lie below 1000Hz, an accuracy of9 12 Hz (needed at 300 Hz) or 15-20 Hz (at 500Hz) can be achieved technically only in theory. On the other hand, linguistically relevant sound patterning can hardly be based on such small differences, the main reason being that a greater security margin is needed for the auditory discrimination of sound patterns in everyday communication situations. If we evaluate the differences reported using the Bark scale asreferenceframe, they seem less crucial. For the fonnant Fl the rough difference for the differentiation of vowel qualities must be about 100 Hz(l Bark) below 500 Hz, corresponding very well to the fact *hat typological!* maximallyfivedegrees ofopening can be distinguished in the languages of the world. For F2 the relevant difference must be much greater (ca. 120-400 Hz, depending on the height of the formants to be compared). This corresponds to the logarithmic increase in the psychoacoustical response above 500 Hz (cf. Zwicker and Feldtkeller, 1967). ACKNOWLEDGEMENTS The LFC analysis for this study was carried out at the Phonetics Department of the University of Kiel in June, 1981.1 am most grateful to Prof. Dr. Klaus Kohler and his staff for their friendly help. I also thank the Deutscher Akademischer Austauschdienst for the grant which made my visit to Kiel possible. For the FFT programming I thank Assoc. Prof. Matti Kaijalainen and for the plotter programming Mr. Jouni Mikonsaari. REFERENCES Hanrrt, R_ (197®|. MmeShemisdbe numSagie amf a&msfgst&er toad pcrzepSoiixAer GrwadBmge. (Trmwmade rhaHitmt de Lm^msstiqm de Lmed.) Land: Glkcrmp. Barry» W. (1974). Penxptiam mad Pnndrnktiem öm yiMfr-pbsmrmiwkcm Boödk. (= f m&wiirinL· Arbeüem. 15.J) Tübingen: Max Miemeycr. Fistlig-JBiyiBai„ Ε. (1975). PcrcxptaJffi off Gcman amid Daraisb Vomidls width Special Reffcxmux to the German Lax Vowels. Lac G. Fault and MJLA. Tathann feds.) AmdBtory Analysis and Pemeptiom of SpeedL London: Academic Press, pp. 153-217. Hdtose, P. C1972). Spectragraphiie Analysis of Frnigfirslfa Vowels. Atamml Report of Ae Imstitaae of Pkametics of Ae Umivasity of CopemiuQcm,, ·„ 1-48. Eiwomcn, A_ |1979|L Ζακ Stndimn drantsrilirr Vokale: Farmantiiiicssgmgcn und anadüttrör SnmmDicnmg des Vofcafcqpginrinns^ In: KL Ezanra and K..H. Rcnsdh feds.) Spmdbe wmd SpnxAem: Festsckrißfmr Eberhard Zmtwrmar mm AOL Gehurssstig. Tübingen: Max Nüemefar, pp. 101-112. Brennen, A. (1982|_ SaksanTOkaaMemiaBcwHttson laadnn tmttlkiimiiitaesta koflumdllacri mannigtdmniällS. Im: A.fiwonen,S. Suhomen and P. Vktaiantta feds.) Voces amkmimm Pxsttsdhriiftt for Amtoi Sowiprvi on the occasion of bis UMe birthday 22.4.19SZ}. Memoiires de la Soäctc FiimraOngÖenne 181 _ Hritemtii:: Snitfwnmlawe^girnlaimi™ snm, pp. 12S-141. Jangemsen, H.P. {1969%. Diie gespannten mnd umgespannten Vokale in der nonlldcuitsclicn Hodhspradlic mit einer qpezifisdhen UmCTsmdhmnB der Struktur ihrer FonianttfreqiiicinEn. Phoneaka,, 19,211-245. Ka^dbmen, Μ (19M)L Design of α ißtwoproeessar-^med SystemfirSpeedk AmedysSs. F= Research Report No. 9, Tampere UmnweaMly of Technology,, Electronics Laboratory.)

138

Antti Iivonen

Kohler, K.J., Schäfer, Κ. and Thon, W. (1981). Phonetische Untersuchungen zur gesprochenen Sprache im Deutschen. Institut für Phonetik der Universität Kiel, Arbeitsbericht 16, 111-136. Lehiste, I. (1961). Transitions, Glides, and Diphthongs. Journal of the Acoustical Society of America, 33, 268-277. Lehiste, I. (1964). Acoustical Characteristics of Selected English Consonants. The Hague: Mouton. Lindblom, B. (1962). Accuracy and Limitations of Sona-Graph Measurements. In: Proceedings of the 4th International Congress of Phonetic Sciences, Helsinki 1961. The Hague: Mouton, pp. 188-202. Mangold, M. and Grebe, P. (1962). Duden Aussprachewörterbuch. Der Grosse Duden, vol. 6. Mannheim: Dudenverlag. Markel, J.D. and Gray, A.H. (1976). Linear Prediction of Speech. Berlin/Heidelberg/New York: Springer. van Nierop, D.J.P., Pols, L.C.W, and Plomp, R. (1973). Frequency Analysis of Dutch Vowels from 25 Female Speakers. Acustica, 29, 110-118. Nord, L. and Sventelius, E. (1979). Analysis and Perception of Difference Limen Data for Formant Frequencies. In: B. Lindblom and J. Lubker (eds.) Experiments in Speech Perception, Phonetics Research Seminar 1978-1979, PERILUS, Report I, 24-37. Pap9un, G. (1980). How Do Different Speakers Say the Same Vowels? Discriminant Analyses of Four Imitation Dialects. UCLA Working Papers in Phonetics, 48. Schäfer, Κ. (1981). Modulare Software zur interaktiven Bearbeitung paralleler Zeitreihen. Institut fur Phonetik der Universität zu Köln, Berichte 11, 53-67. Sendlmeier, W.F. (1981). Der Einfluss von Qualität und Quantität auf die Perzeption betonter Vokale des Deutschen. Phonetica, 38, 291-308. Strange, W. and Gottfried, T.L. (1980). Task Variables in the Study of Vowel Perception. Journal of the Acoustical Society of America, 68, 1622-25. Wiik, K. (1965). Finnish and English Vowels. Annates Universitatis Turkuensis, Series B, 94. Turku: Turun yliopisto. Zwicker, Ε. and Feldtkeller, R. (1967). Das Ohr als Nachrichtenempfänger. Stuttgart: Hirzell.

The Temporal Relationship between Articulations of Consonants and Adjacent Vowels Shigeru Kiritani, Research Institute of Logopedics and Phoniatrics, University of Tokyo, Tokyo, Japan Masayuki Sawashima, Research Institute of Logopedics and Phoniatrics, University of Tokyo, Tokyo, Japan

1. INTRODUCTION

In the production of VC and CV sequences, articulatory movements for the pre-consonantal vowel and post-consonantal vowel overlap temporally with those for the consonants. Analysis of the temporal relationship between consonant articulations and vowel articulations appears to be one of the basic problems in the understanding of the temporal organization of articulatory movements. Temporal overlap of consonant and vowel movements causes variation in the consonant articulations based on the vowel context. There have been several studies analyzing the temporal pattern of the tongue movements in VCV utterances (Öhman, 1967; Houde, 1968; Gay, 1977). However, the types of speech utterances examined in these studies were limited. Temporal characteristics of the coarticulation between consonants and vowels in the VCV sequences may vary depending on the type of segment. Different languages may exhibit different temporal characteristics due to the dissimilar patterns of temporal grouping of the sequential segments. The present study is an attempt to investigate these problems by observing movements of the tongue in the production of selected VCV sequences in Japanese and American English, using the X-ray microbeam method. The relative magnitude of the perturbations of the consonant articulations by the pre-consonatal and the post-consonantal vowels is examined both for Japanese and English, and factors (such as the type of vowel) that affect the temporal relationship between the consonant and vowel articulations are examined. 2. JAPANESE VCV A N D CVC SEQUENCES

Two sets of speech material were used. The first set consisted of meaningless

140

Shigeru Kiritani and Masayuki

Sawashima

words of the form /C^WCii/. The vowel was one of the five Japanese vowels: /i, e, a, o, u / and the Q and C2 were one of the four consonants: / m , t, k, s/. The number of test words was 80 and, for these test words, variations in vowel articulations due to the difference in C ( and C2 were examined. The second set of speech material consisted of meaningless words of the form /mViCV 2 /. The vowels and consonants were the same as in the first set and the number of test words totalled 100. For this speech material, variations in consonant articulations due to the difference in the vowel context were examined. The subject was a male adult who spoke the Tokyo dialect. The test words were uttered in a sentence which consisted of a test word and the copula 'desu'. The sentences were uttered in succession with no pause in between. The speaking rate was relatively fast. Movements of the tongue were observed by the X-ray microbeam method. In this method, several metal pellets are attached to the articulatory organs and their movements are tracked by X-ray microbeam. Deflection of the X-ray beam is controlled by an on-line computer to automatically track the moving pellets. The method has the advantage of substantially reducing radiation dosage to the subject as well as avoiding manual work in analysis and measurement of the X-ray film required in conventional cineradiography. The principle of this method has been reported elsewhere (Fujimura et al., 1973; Kiritani et al., 1975). J

ι

ι—r

makae

(a)

κ—ι lcm J

L

(b) Fig. 1. Schematic illustration of pellet locations (a) and examples of pellet trajectories for VCV utterances (b).

The Temporal Relationship between

Articulations

141

In the present experiment, four pellets were attached to the tongue, as shown in Figure 1 (a). Pellet No. 4 was located on the tongue blade, at about 1 cm back from the tongue tip. The distance between each pellet on the tongue was about 3 cm. In order to observe the movements of the jaw and the lower lip, a single pellet was attached to both the lower incisor and the lower lip. The movements of the pellets were recorded at a rate of 130 frames per second. Examples of the pellet trajectories in the production of the vowel-consonant-vowel sequence are shown in Figure 1 (b). In this figure, pellet movements for the utterances / m a k a e / and / m a k i e / are compared. In the case of the utterance / m a k a e / , pellet No. 2, for example, first moves downward and backward for vowel / a / . Then, the direction of the movement changes upward for the articulation of consonant / k / . The tongue positions considered to be representative of the first vowel / a / and the intervocalic consonant / k / are shown by connecting with solid lines the positions of the four tongue pellets at selected time moments. After reaching the highest position f o r / k / , the pellet moves downward for the second vowel / a / , and then it moves forward for the vowel/e/. When the pellet trajectories for / m a k i e / are compared with those o f / m a k a e / , it can be seen that pellets No. 1 and No. 2 start to move forward almost at the same time with the upward movement for the articulation of / k / . This forward movement is considered to be the effect of the vowel / i / which follows the consonant / k / . Thus, this utterance example demonstrates a case in which the effect of the post-consonantal vowel appears already at an early phase of the vowel-to-consonant transition. For the C]VC 2 sequences in the first set of speech material, perturbations of the vowel articulations by the consonantal environment were examined and the relative magnitude of the effect of the preceding consonant C t and the following consonant C 2 were compared. For each vowel, the standard deviation of the coordinate of pellet No. 3 was calculated for two different sets of test words. For any particular VC2 combination, there were four test words which differed in the first consonant C,. The standard deviation of the pellet coordinate over these test words was calculated and then averaged over four different C 2 . This yields a measure of perturbation of the given vowel by CV Perturbation by C 2 was calculated in a similar way. The results are shown in Table 1.

Table 1. Variations in position of pellet No. 3 for V in C, VC2 sequences (Standard deviation in mm). Vowel Variation Variation

byC, byC2

a 1.5 0.9

2.0 0.8

u 2.4 1.4

e 3.3 2.0

ο 3.4 1.5

142

Shigeru Kiritani and Masayuki Sawashima

3. RESULTS

It can be seen that the perturbation by the first consonant is about twice that by the second consonant in terms of the standard deviation of the coordinate value. That is, the left-to-right effect of the consonant on the following vowel is greater than that of the right-to-left effect of the consonant on the preceding vowel. This is true regardless of the type of vowel. For the VjCV 2 sequences in the second set of speech material, perturbations of the consonant articulations by the pre-consonantal vowel and the post-consonantal vowel were compared. For each test word, the pellet coordinates at the moment of consonant peak, which was defined as the moment of the maximal excursion of the pellet on the pertinent articulator (lip pellet for / m / , tongue pellet No. 4 f o r / t / a n d / s / , and tongue pellet No. 2 ,for / k / ) , were sampled. For a particular pair of the pre-consonantal vowel V, and the middle consonant C, there were five test words which differ in the post-consonantal vowel V2. The standard deviation of the pellet coordinate over these five words was calculated and then averaged over different V, to provide a measure of perturbation of the consonant articulation by the post-consonantal vowel. Perturbation by the pre-consonantal vowel was estimated in a similar way. The results in Table 2 show that perturbation of consonant articulation by the following vowel is greater than that by the preceding vowel. Examination of the temporal pattern of the pellet movements revealed that the relative timing of vowel and consonant articulations varied depending on the type of the vowel. Figure 2 (a) shows a superposed display of the time functions of the pellet coordinates for the two groups of words /mamVe/ and /mimVe/, where V was varied over five vowels: / a , i, u, e, o/. The curve at the top in each figure shows the vertical displacement of the lower lip. The local peak of the curve corresponds to the closing movement of the lip for / m / . The other two curves show the χ and y coordinates of pellet No. 3 on the tongue. It can be seen that, with regard to the moment of the peak of the middle / m / (which is indicated by the vertical line in each figure), transitions from vowel / a / to the following vowels in /mamVe/ start earlier than transitions from vowel / i / to the following vowels in /mimVe/. Consequently, variations in the pellet positions over five different words at the moment of peak of / m / are greater for /mamVe/ than for /mimVe/. Table 2. Variations in position of pellet No. 3 for C in V,CV 2 sequences (Standard deviation in mm). Consonant Variation Variation

by V, by V2

m 4.5 5.4

t 3.4 4.1

k 3.1 5.2

s 3.1 4.2

The Temporal Relationship between Articulations

143

Displacement

Fig. 2. Superposed displays of pellet movements for utterances mVmie, mVmae, mimVe and mamVe ( V = i , e, a, o, u). Vertical lines indicate the moments of peak displacement of the lower lip for / m / .

A similar tendency is also observed when pellet movements for the groups of utterances / m V m a e / and /mVmie/ are compared (Figure 2 (b)). In this case, transitions to the vowel / i / in / m V m i e / start earlier than transitions to the vowel / a / in / m V m a e / . Thus, when we compare the effects of vowels / i / and / a / on the tongue position at the consonantal period, the tongue gesture for the vowel / i / is more apparent than that for the vowel / a / , regardless of whether the vowel / i / precedes or follows / m / .

144

Shigeru Kiritani and Masayuki

Sawashima

4. ENGLISH CVC SYLLABLES

The speech materials examined were as follows: (1) / p V p / words uttered with a carrier phrase 'It's a again.' The word medial vowel was varied over ten vowels, /i, i, e, ε, ae, o, o, u, s, λ/. (2) Selected CVC words containing dental consonants paired with different vowels. 'It's a seat.' 'It's a sot.' 'It's a peace.' 'It's a bus.' etc. (3) Simple sentences containing a word initial / s / or word final / s / . 'Miss Pee seeds/sots it.' 'Miss Pa seeds/sots it.' 'Miss Piece eats/odds it.' 'Miss Poss eats/odds it.' In this experiment, three pellets were attached to the tongue as illustrated in Figure 3. For the / pVp/ words, the pellet coordinates at the moments of peak of the word initial / p / and the word final / p / were sampled and the standard deviations over ten different vowels were examined. The moment of the consonant peak was defined as before, i.e., as the moment of maximal vertical excursion of the pellet on the lower lip. 5. RESULTS

The results are given in Table 3. As can be seen in Table 3, the variation in the

Fig. 3. Pellet locations used in the experiment on American English.

Table 3. Variations in position of pellet No. 2 for / p / in / p V p / (Standard deviation in mm). Word initial / p / Word final / p /

1.9 8.1

The Temporal Relationship between

Articulations

145

pellet coordinate for the word final / p / is greater than that for the word initial / p / . Thus, the left-to-right effect of the word medial vowel on the word final / p / is greater than the right-to-left effect on the word initial / p / . Figure 4 (a) compares pellet movements for the utterances 'It's a peep again,' and 'It's a pip again.' The curve at the top is the vertical displacement of the lower lip, and the other two curves are the x- and y- coordinates of pellet No. 2 on the tongue. It is noted that the two utterances show a clear difference in the timing of the tongue movement for the word medial vowel relative to the lip movement for / p / . That is, with regard to the moments of the two consonant peaks, which are shown by a pair of vertical lines in the figure, maximal displacement for vowel / i / is achieved later than that for vowel / i / . Consequently, the difference in the magnitude of the left-to-right effect and the right-to-left effect of the word medial vowel is greater when the vowel is / i / than when it is / i / . Tongue movement for vowel / u / , when compared with that for vowel / u / , shows similar characteristics to that of / i / (Figure 4 (b)). Thus, these data indicate that the temporal pattern of the vowel consonant coarticulation varies depending on the type of vowel. Vowels / e / and / o / also show similar

Fig. 4. Pellet movements for tense and lax vowels in American English. Vertical lines indicate the moment of peak displacement of the lower lip for / p / .

146

Shigeru Kiritani and Masayuki

Sawashima

temporal characteristics, which may be natural if we consider that these vowels are diphthongized. The tendency that the effect of the word medial vowel on the word final consonant is greater than that on the word initial consonant is also observed in the other utterances containing the dental consonant. Figure 5 (a) shows the movements of the tongue pellets in the utterances 'It's a seat,' and 'It's a sot.' In these utterances, the pellet positions are almost stationary during the period o f / s / , and the tongue movement toward the following vowel [i] or [a] does not appear until near the release of / s / . That is, the difference in the following vowel has little effect on the tongue position d u r i n g / s / (at least for the tongue position reflected in the movement of pellet No. 2). On the other hand, for the word final / s / shown in Figure 5 (b), the effect of the word medial vowel is clearly seen at the moment of frication onset. Thus, it is not the case that the perturbation of the / s / articulation by the vowel context is generally slight. Rather, it appears that, in the case of the word initial / s / , the movement toward the following vowel is suppressed until the offset of / s / . Figure 6 compares the relative magnitude of the effects of the pre-consonantal vowel and the post-consonantal vowel on / s / . Figure 6 (a) is for the syllable initial / s / and Figure 6 (b) is for the syllable final / s / . These data show that the effect of the pre-consonantal vowel on the syllable initial / s / and the syllable final / s / are quite similar, although there is a difference

Displacement

lOOmS Fig. 5. Superposed displays of pellet movements for selected CVC words, comparing the effects of the word medial vowel on the word initial consonant and word final consonant. Vertical lines show the moments of implosion or release of the consonants.

Fig. 6. Next page. Superposed displays of pellet movements for pairs of utterances containing the consonant /%/. The effects of the pre-consonantal and post-consonantal vowels are compared for syllable initial / s / (a) and for syllable final / s / (b).

ο Ο) (V α. «Λ

148

Shigeru Kiritani and Masayuki

Sawashima

between the two cases in whether or not there is a syllable boundary between / s / and the preceding vowel. On the other hand, as for the effect of the post-consonantal vowel, there is a clear difference between the syllable initial / s / and the syllable final / s / . For the syllable initial / s / shown in Figure 6 (a), the effect of the following vowel is quite small. The result is similar to that shown in Figure 5 (a). Thus, for the syllable initial / s / , the effect of the pre-consonantal vowel is greater than that of the post-consonantal vowel. As for the syllable final / s / , it appears that the movement towards the following vowel takes place during the period of frication of / s / , and the effect of the post-consonantal vowel is greater than that for the word initial / s / . In other words, for the word final / s / , the effect of the pre-consonantal vowel and the post-consonantal vowel is nearly the same in magnitude. 6. DISCUSSION

The present study shows that in CVC sequences in Japanese, perturbation of vowel articulation by the preceding consonant is greater than that by the following consonant, while in VCV sequences, perturbation of the consonant articulation by the following vowel is greater than that by the preceding vowel. The results show that, in Japanese, a consonant and the following vowel form a cohesive unit of articulatory movement, which appears to conform to the general notion that a consonant and its following vowel constitute a syllable in this particular language. For American English, the effects of the adjacent vowels on the syllable initial and syllable final consonants were examined. The results show that the relative magnitude of the effects of the pre-consonantal vowel and the post-consonantal vowel on the consonant varies according to the position of the consonant within a syllable. For syllable initial / s / , the influence of the preceding vowel is clearly seen, while the influence of the following vowel is very small. Thus, the effect of the preceding vowel is greater than that of the following vowel. This difference between Japanese and English suggests that consonant and following vowel constitute a more cohesive unit in Japanese than in American English. In other words, the syllable initial consonant retains its identity in American English more than in Japanese with regard to the influence of the following vowel. For syllable final / s / in American English, the effect of the following vowel is more apparent than for syllable initial / s / , while the effect of the preceding vowel on the syllable initial and syllable final / s / is almost identical. These results may be interpreted to mean that, in American English, the constraint to keep a given target position is stronger for the articulation of the syllable initial consonant than for the syllable final consonant, and coarticulation with the following vowel takes place more easily for the syllable final consonant. It is well known that there are allophonic variations in the phonetic

The Temporal Relationship between

Articulations

149

characteristics of the consonants depending on the position of the consonants in an utterance, such as change in duration, intensity of aspiration or frication, etc. The present study shows that such an allophonic variation of the consonant is also observed in the temporal characteristics of the coarticulation with the following vowel. Another phenomenon found in the present study is that the pattern of the temporal overlap of the vowel and the consonant articulations varies depending on the type of vowel. In the case of American English, it is observed that tongue movements for the tense vowels are carried over into the following consonant to a greater extent than for the lax vowels. This phenomenon seems to conform to the traditional terminology 'tense' and 'lax', in that for a 'tense' vowel the tongue position is maintained longer and the movement toward the following sound takes place later (Lehiste, 1960). In the case of Japanese, there is a difference between the utterances / a m i / and / i m a / in the timing of vowel-to-vowel transition relative to the articulation of the intervocalic consonant. Namely, the timing of the / α / to / i / transition in / a m i / with regard to the lip articulation for / m / is earlier than that of the / i / to / a / transition in / i m a / . A similar timing difference is also found in other studies, such as the study on the relative timing of the pitch transition and the formant transition in Japanese (Sawashima et al., 1982). The basis of this timing difference is, however, not clear at present and further studies are needed to clarify the basis of this phenomenon. REFERENCES Fujimura, O., Kiritani, S. and Ishida, H. (1973). Computer Controlled Radiography for Observation of Movements of Articulatory and Other Human Organs. Computers Med. Biol, 3, 371-384. Gay, T.J. (1977). Articulatory Movements in VCV Sequences. Status Report Speech Research, Haskins Laboratories, SR-49, 121-147. Houde, R. A. (1968). A Study of Tongue Body Motion During Selected Speech Sounds ( = S.C.R.L. Monograph No. 2). Santa Barbara, California: Speech Communication Research Laboratory, Inc. Kiritani, S., Itoh, K. and Fujimura, O. (1975). Tongue-Pellet Tracking by a Computer-Controlled X-Ray Microbeam System, J ASA, 57, 1516-1520. Lehiste, I. (1960). Phonetic Study of Internal Open Juncture. Phonetica, suppl. to vol. 5. Öhman, S.E.G. (1967). Numerical Model of Coarticulation. J ASA, 39, 310-320. Sawashima, M., Hirose, H. Yoshioka, H. and Kiritani, S. (1982). Interaction Between Articulatory Movements and Vocal Pitch Control in Japanese Word Accent. Annual Bulletin of the Research Institute of Logopedics and Phoniatrics, Tokyo University, 16, 11-19.

The Application of Analog Models of Some Auditory Mechanisms for Speech Signal Processing fV.A. Kozhevnikov, N.E. Drozdova, Pavlov Institute of Physiology, Academy of Sciences of the USSR, Leningrad, U.S.S.R. E.I. Stoljarova, Pavlov Institute of Physiology, Academy of Sciences of the USSR, Leningrad, U.S.S.R.

1. AN AUDITORY SPECTRAL ANALYZER

This paper deals with the representation of speech signals as seen in the output of models of human auditory system mechanisms. 4 Auditory' dynamic spectrograms were obtained from the initial block of a functional model simulating auditory frequency analysis (linear version). This block was built as a multichannel analyzer. Sharpening of the spectrum-shape pattern by the nonlinear two-tone suppression model is demonstrated. A model for envelope processing in the channel simulating the 'on' and 'off responses at the higher levels of the auditory system is described, and the reaction of the model to speech material is shown. Studies in almost all fields of experimental phonetics require devices for visual presentation of the acoustic patterns of speech. Devices such as dynamic spectrographs are used for this purpose. The acoustic properties of the phonetic elements essential for human perception must be represented in the spectrum patterns registered by the device. Phoneticians analyzing dynamic spectrograms usually tend to pick out those details which are important from this point of view, omitting the less important ones. We have attempted to use an analog model of some auditory mechanisms for the analysis of speech. It was natural to assume that the perceptually important features of speech would automatically be emphasized in the output pattern, while details of little significance for human hearing would in turn not be present. The analog system built in our laboratory comprises a model of an auditory spectral analyzer, a two-tone suppression model and a model of an 'envelope processing channel.' The model of the auditory spectral analyzer developed by V.S. Supljakov (see Dolmazon et al., 1977; Goloveskin et al., 1978) is based on certain biophysical data of cochlear hydrodynamics and does not reproduce non-

152

Kozhevnikov, Drozdova, and Stoljarova

linear effects, its resulting characteristics agreeing with psychoacoustical and neurophysiological data. The input block of the model reproduces the frequency characteristics of the system of the middle ear and the basal part of the cochlea. The following electronic circuit, which models traveling wave effects in the cochlea, contains 128 series-connected frequency-selective elements of the second order, the response frequency decreasing as the number (coordinate) of the element increases. The relative bandwidth of the elements ranges up to 1.5; the elements are isolated by amplifiers with an amplifying coefficient close to 1. The frequency characteristics of the elements are multiplied in series, which causes a steep high-frequency slope and complete suppression of frequencies above the bandpass range. Additional low-pass filters of the second order at the outputs of the elements have the effect of sharpening the frequency characteristics, their relative width tending to be close to the critical bands. The output signal of a given filter represents the waveform at the corresponding point along the cochlear axis. The model reproduces frequency-to-coordinate conversion and works like an analyzer of the parallel type. The center frequencies of the 128 channels range between 30 and 9500 Hz. The relation between the center frequency and the number of the channel (coordinate) is linear below 800 Hz, with the bandwidth equal to 100 Hz for all channels. Above 800 Hz the relation of the center frequency of the channel to its number is logarithmic. The channels may be considered to be filters with an equal relative bandwidth. The amplitude/frequency characteristics are quite asymmetric: the high-frequency slope is up to 180 dB/octave, but the low-frequency slope is considerably less steep. The difference between attenuation at the resonant frequency and outside the band increases with the center frequency of the channel. The response time of the filters increases as the channel frequency decreases, and ranges from 1 msec at 2000 Hz to 5 msec at 250 Hz. The analyzer's dynamic range is 80 dB. The recording of the model's output signals is accomplished in a similar way to 'visible speech' devices (L. Chistovich et al., 1974). Since the multichannel parallel recorder is absent, the sound stimulus to be analyzed is repeatedly played back by the system by means of a rotating magnetic head and is fed into the model. An electronic switching device serves to sequence the model's outputs and to displace the optical recording head. After rectification and logarithmic compression the model's output signals modulate the intensity of the neon lamp in the optical device. The resulting image has the dimensions of time, frequency and intensity (the latter shown by the darkness of the photographic material); this picture has been termed an 'auditory spectrogram.' The analog system also reproduces spectral 'samples' (the short-term spectrum) taken at given moments in time, with the relative intensity of the frequency components expressed in decibels. To evaluate the compatibility of the model with the real human auditory

Analog Models for Speech Signal Processing

153

system, the ability of the model to reproduce just-perceptible gaps in white noise and just-perceptible spectral peaks in vowel-like stimuli was tested (Borozdin et al., 1980). Gaps with durations longer than 2 msec were clearly represented by vertical white bands in the high-frequency region of the spectrogram. The representation of short gaps in the low-frequency region was not clear; at frequencies below 500-800 Hz it started to blur when gap duration was shortened to less than 5 msec. These results seem to agree well with current psychoacoustic data on gap detection. The model was also tested for reproduction of the minimal spectral peaks essential for vowel quality perception. Synthetic-stimuli formant-amplitude thresholds specified in psychoacoustic experiments were detected well by the model. Thus the model reproduces the minimal temporal and spectral stimulus irregularities perceived by the human auditory system. By now several hundred 'auditory kHZ 6~

α

5—

J—

θ m a. 100 SEGMENT

DURATION,

150 NORMAL

200

250

CONDITIONS

300 (msec)

Fig. 4. Compensatory segment durations (Y) and corresponding normal values (X) for all stressed vowels and the adjacent preceding and following consonants. Data from all six subjects pooled.

How might the complementarity of vowel and consonant duration arise? It seems reasonable to hypothesize that the presence of a bite-block would make it natural for the speakers to undershoot labial closures. Accordingly, the effect of the bite-block would be to delay the time of closure and to release the occlusion prematurely (cf. the account of inherent vowel duration given in Lindblom, 1967). It is not the aim of the present study to test this interpretation against the movement data (cf. Lubker et al., forthcoming). Nevertheless, a preliminary indication of its realism might be obtained as follows. Since our assumption implies that the bite-block 'steals' duration from a given consonant on behalf of the vowels on either side of it and each vowel affects both its adjacent consonants, we decided to compare the (N-BB) differences in consonant duration with those in vowel duration both

The Concept of Target and Speech Timing

SEGMENT

DURATION,

173

NORMAL

CONDITIONS

(msec)

Fig. 5. Compensatory segment durations (Y) and corresponding normal values (X) f o r all unstressed segments. Those preceding the stressed syllable on the left, those following on the right.

summed over each entire word. Thus Fig. 6 shows X A C : S VS. I A V : S . Each point refers to a given word spoken by one of the subjects. The significance of data points falling in any one of the four quadrants can be stated as follows: Top left:

Vowels longer, consonants shorter in BB speech. For perfect complementarity data points must lie on the solid 135 degree line. Bottom left: Both vowels and consonants longer in BB speech. Uniform slowing down: data points on the dashed 45 degree line. Bottom right: Vowels shorter, consonants longer in BB speech. Complementarity implies negative correlation. Top right: Both vowels and consonants shorter in BB speech. Uniform speeding up: data points on the dashed 45 degree line. It is clear that, since the data points tend to cluster in the top left quadrant, there is support for the interpretation we have just proposed. We obtain a correlation coefficient of r = -0.4* and mean values for C(N-BB) and V(NBB) which are 12 and -19 ms respectively. Removing the previously mentioned cases gives r = - 0 . 6 7 * * * C(N-BB)=-15 and V(N-BB) = -15 ms.

174 υ

Φ " Ε

< CL

Lindblom, Lubker, Gay, Lyberg, Branderud, and Holmgren I Ω UJ

τ

30

(Λ Ζ

20

2

5

°




3

I

-50

/

/

/

•

•

•

/ SPEEDING UP •

COMPLEMENTARITY

s

10

S

0

Ο

υ

Ω Ζ


ο

SLOW I

-30

tu

£ ν— α: UJ UJ m

-20

/

-50

/

/

-40

DOWN

/

-30 - 2 0

-10

COMPLEMENTARITY

0

DIFFERENCE IN VOWEL Ν

AND

WHOLE

BB

X 20

10

40

50

DURATION BETWEEN

CONDITIONS

WORD

30

.

SUMMED

OVER

(msec)

Fig. 6. Each data point refers to a comparison between differences between Ν and BB durations. X-axis: vowel duration differences summed over the whole word. Y-axis: corresponding consonant duration differences summed over the whole word. Each point refers to a given talker's production of a given word.

4. CONCLUSIONS, INTERPRETATIONS A N D DISCUSSION

Summarizing, we find that: (i) All subjects achieve at least partial compensation for the bite-block in the sense that they are capable of producing lip closure for / b / in all positions. (ii) Their BB productions reproduce normal vowel and consonant durations within, on the average, ± 15 ms. (iii) The BB condition is shown to induce complementary consonantal shortenings and vowel lengthenings. (iv) We interpret this effect to be compatible with the assumption that the BB condition makes it natural for speakers to undershoot the / b / closures slightly.

The Concept of Target and Speech Timing

175

Let us pause to consider how the instructions were formulated. Recall that subjects were asked to attempt a match in terms of word rhythms or stress patterns. They were not asked to replicate segment durations - probably a more difficult instruction to interpret and carry out both for naive and trained subjects. The question thus arises as to how word rhythms and stress patterns might be coded acoustically. One possible hypothesis is that speech rhythm is signalled in terms of the sequence of 'syllable beats' and that the perception of such beats is a function of various acoustic events such as intensity discontinuities at the CV boundary of a syllable as well as characteristics of the F 0 contour near that point in time (Allen, 1972; Huggins, 1972; Rapp, 1971). Such experimental evidence seems to provide support for the view that in the case of the present /ba/-segments the syllable beats are the vowel onsets ( = the /b/-releases). Adopting vowel onset as the acoustic correlate of syllable beat we can define a given stress pattern or word rhythm in terms of its temporal pattern of vowel onsets. Table I presents the results in terms of such patterns quantified as IVOI, intervowel-onset intervals:

Table I: Intervowel-onset intervals (IVOI) in ms tabulated by subject, by test word and by experimental condition ( N and BB). The (N-BB) differences are also shown. Their mean value = 0, the standard deviation = 13 ms. Average values (n > 19 samples). IVOI Ν

220

BB

236

Ν

288

BB

273

Ν

330

BB

313

Ν

221

BB

224

Ν

219

BB

216

Ν

282

BB

285

21

-16

31

15

51

17

71

-3

81

3

91

-3

176

Lindblom, Lubker, Gay, Lyberg, Branderud, and Holmgren IVOI Ν

166

BB

163

Ν

215

BB

213

Ν

206

BB

172

Ν

153

BB

161

Ν

165

BB

173

Ν

210

BB

215

21

3

31

2

51

34

71

-8

81

-8

91

5

IVOI2

IVOIl Ν

253

21

139 17

BB

226

Ν

250

31

-13 152 143

43 BB

207

Ν

246

51

-5 148 167

24 BB

222

Ν

188

71

11 156 129

-4 BB

192

Ν

299

-12 141 157

3

81 BB

296

Ν

255

7 150 158

5

91 BB

260

ι 155

['bab:abab]

177

The Concept of Target and Speech Timing IVOIl Ν

IVOI2

158

267 4

21 BB

154

Ν

192

31

24 243 225

8 BB

184

Ν

148

51

-13 238 366

9 BB

139

Ν

130

71

-11 375 230

-12 BB

142

Ν

143

81

-25 255 282

1 BB

142

Ν

176

91

2 280 319

3 BB

Ν

-8

173

327

IVOIl

IVOI2

130

21

141 -3

BB

133

Ν

147

31

-16 157 146

9 BB

138

Ν

179

51

-22 198 201 3

10 BB

189

Ν

135

71

204 146 -16

BB

151

Ν

165

81

-12 158 170 5

-4 BB

169

Ν

180

91

165 190 5

BB

175

[ba'babrab]

-9 199

[baba'bab:]

178

Lindblom, Lubker, Gay, Lyberg, Branderud, and Holmgren

How does the complementarity of consonant shortening and vowel lengthening influence the realization of word rhythm? If the interpretation of this effect in terms of labial-closure undershoot is accepted, then the following argument would seem to hold. The syllable initiations (vowel onsets) of BB syllables are, it is true, premature in relation to their Ν counterparts. The relative timing of these events, however, would be expected to be similar for Ν and BB conditions. To evaluate the data in Table I from the viewpoint of the preceding paragraph it is first relevant to review what we know about the just-noticeable difference (JND) or difference limen (DL) values for segment duration (Lehiste, 1970; Klatt and Cooper, 1975). In her summary of the literature up to 1970 Lehiste states: . . in the range of durations of speech sounds usually from 30 to about 300 ms - the just-noticeable differences in duration are between 10 and 40 ms.' Klatt and Cooper report the results of an experiment using natural sentence materials in which the / i l / of the words deal or dealer and the / s / in fish were durationally modified. For ranges of original durations of 165-340 ms and 95-142 ms respectively they inferred JND's of 25 ms or more. Since Klatt and Cooper's / i l / durations represent IVOI values and coincide with the ranges of Table I we might apply these authors' conservative estimate of ± 25 ms to the present figures. We would then conclude that the majority of the (N-BB) differences appear to be, on the average, perceptually insignificant. Consequently, by way of interpretation, it appears justified to propose that: (v) Subjects actualized the stress patterns of the test words by controlling the relative timing of syllable (vowel) onsets; (vi) In so doing they used Ν and BB syllable timing patterns that must be regarded as closely similar, if not identical, judging by criteria of perceptual tolerance. (vii) Furthermore the present results do not preclude that the reorganization of articulatory gestures takes place with respect to segment durations only in so far as it is necessary to maintain the perceptual constancy of the stress patterns. Hence we observe selective or partial segment duration compensation. (viii) The most important assumption implicit in the preceding account is that of attributing to the speakers a 'mental plan', that is a representation of the stress pattern (rhythm, metrics) of each word. In other words, the present analysis appears to force us to postulate underlying temporal targets largely independent of the processes of motor execution. To conclude the present paper we shall make three points. We shall comment on the status of the target concept, in particular the notion of

The Concept of Target and Speech Timing

179

temporal target. We shall touch upon the question of independence of targeting and execution. We shall consider compensatory behavior in the context of motor control in general. The account given above implies a 'mentalistic' target theory of speech timing. We are aware that the application of the concept of target to the time dimension of speech was explicitly dismissed by Lisker (1972) in his theoretical discussion of time and timing in speech. The criticism leveled by Fowler (1980) at so-called extrinsic timing theories also seems relevant to the present interpretations, which refer explicitly to movement-independent temporal goals. (Extrinsic timing: 'time is not taken to inhere in, or to be essential to, the specification of . . . features, segments or syllables'.) These views certainly deserve our serious attention. There are a host of phenomena in speech timing that are likely to be intrinsic, e.g. duration of average breath group, average syllable and segment durations (Lindblom, 1983). Nevertheless, there is incontrovertible evidence making extrinsic timing models viable. One source of such evidence has to do with language-specific differences. To take only one example, it appears intuitively clear that languages can sound very different, rhythmically. Compare for instance syllable-timed and stressed-timed languages. Or take more closely related phenomena such as English and Swedish pronunciations of words that can be said to contain similar segmental targets and to be made up of a sequence of a stressed syllable followed by an unstressed one but whose timings must be said to be rather different: [' Jiliq] versus [' J"il:iq] for shilling. Here there is alanguage-dependentdifference in the quantity structure of the stressed syllable, the Swedish intervocalic consonant being obligatorily longer, the English one short. This phonological difference can be said to account for the timingdifferences. Howcouldtheyever be expected to be derivable from the inherent phonetic characteristics of such utterances? While we are convinced that great benefits will accrue from intensifying research on intrinsic timing phenomena in speech, we are not willing to totally give up the extrinsic model. For support of the present 'mentalistic' target theory applied to speech timing we should give high priority to an early demonstration by Nooteboom (1972) whose subjects were able to adjust, with great precision, vowel duration in synthetic test wordsaccordingto an 'internal criterion'. We take Nooteboom's term 'internal criterion' to be synonymous with our 'temporal target'. The rhythm hypothesis explored by Kozhevnikov and Chistovich (1965) for Russian is likewise pertinent. It was not substantiated in its strong form, that is the temporal realization of a given stress pattern was not independent of the segmental realization o f t h a t stress pattern. For instance, words containing consonant clusters were found to take longer than prosodically identical ones having single consonants. Consequently the present conclusions regarding target patterns of intersyllabic timing that are 'largely independent of the processes of motor execution' must not be taken to totally rule out temporal

180

Lindblom, Lubker, Gay, Lyberg, Brändernd, and Holmgren

interaction between an utterance plan and its execution, i.e., between extrinsic and intrinsic timing. The plastic response of the speech motor system to a large variety of goals and tasks is well documented and appears to be continuous with general biological capabilities present in the elementary locomotion, breathing and mastication mechanisms of most animals (Grillner, Lindblom, Lubker and Persson, 1982). The present results are instances of such motor adaptability. It is worth mentioning that they amplify an observation first reported by Netsell, Kent and Abbs (1978) on normal and bite-block sentence productions. For a 16 mm bite-block condition considerable increases in lip velocities were noted whose effect was that of maintaining constancy of timing at the level of the acoustic signal. As we contemplate the complexity of the brain processes that make possible the astounding reorganizational ability that we witness in speech and other motor activities, the question arises as to how deep in the nervous system the answer to many of our queries in phonetics may be hidden. At such moments it is comforting to recall the words so well chosen by Ilse Lehiste in a recent lecture* as a characterization of her own research efforts: 'Count what is countable, measure what is measurable and what is not measurable, make measurable!' ACKNOWLEDGEMENTS

The authors gratefully acknowledge the help of Liisa Karhapää who typed the manuscript and Joseph Perkell who made many valuable comments on it.

REFERENCES Allen, G.A. (1972). The Location of Rhythmic Stress Beats in English: An Experimental Study, Part I and II, Language and Speech, 15, 72-100; 179-195. Branderud, P. (1979). Blod-a Block Diagram Simulator. In: PERILUS /, University of Stockholm, 151-153. Fischer-J0rgensen, E. (1964). Sound Duration and Place of Articulation. Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung, 17, 175-207. Fowler, C.A. (1980). Coarticulation and Theories of Extrinsic Timing. J. of Phonetics, 8, 113-133. Grillner, S., Lindblom, B., Lubker, J. and Persson, A. (1982). Speech Motor Control. Pergamon Press: London. Gay, T., Lindblom, B. and Lubker, J. (1981). Production of Bite-Block Vowels: Acoustic Equivalency by Selective Compensation. J. Acoust. Soc. Am., 69, 802-810. Hixon, T. (1971). An Electromagnetic Method for Transducing Jaw Movements During Speech. J. Acoust. Soc. Am., 49, 603-606. *

SCAMP, Summer Conference on Speech Communication organized in 1981 by I.D.Α., Princeton, at Williams College, Williamstown, Mass., USA.

The Concept of Target and Speech Timing

181

Huggins, A.W.F. (1972). On the Perception of Temporal Phenomena in Speech. J. Acoust. Soc. Am., 51, 1279-1290. Kozhevnikov, V.A. and Chistovich, L.A. (1965). Speech and Articulation. Moscow-Leningrad. (English Translation: J.P.R.S., Washington D.C., JPRS 30543). Klatt, D.H. and Cooper, W.E. (1975). Perception of Segment Duration in Sentence Contexts. In: A. Cohen and S. Nooteboom, (eds.): Structure and Process in Speech Perception. Springer Verlag: Heidelberg. Lehiste, I. (1970). Suprasegmentals. M.I.T. Press: Cambridge. Lehiste, I. (1972). The Timing of Utterances and Linguistic Boundaries. J. Acoust. Soc. Am., 51, 2018-2024. Lehiste, I. (1975). Some Factors Affecting the Duration of Syllabic Nuclei in English. In: G. Drachman (ed.): Proceedings of the First Salzburg Conference on Linguistics. Verlag Gunter Narr, 81-104. Liberman, M.Y. (1977). Further Work on Duration Modelling in Reiterant Speech. 94th Meeting of Acoustical Society of America, paper V9. Lindblom, B. (1967). Vowel Duration and a Model of Lip Mandible Coordination. Speech Transmission Laboratory Quarterly Progress Report, 4, 1-29. Lindblom, B. and Sundberg, J. (1971). Acoustical Consequences of Lip, Tongue, Jaw and Larynx Movement. J. Acoust. Soc. Am., 50, 1166-1179. Lindblom, B., Lyberg, B. and Holmgren, K. (1981). Durational Patterns of Swedish Phonology. Indiana University Linguistics Club. Lindblom, B., Lubker, J. and Gay, T. (1979). Formant Frequencies of Some Fixed-Mandible Vowels and a Model of Speech Motor Programming by Predictive Simulation. J. of Phonetics, 7, 147-161. Lindblom, B. (1983). Economy of Speech Gestures. In: P.F. MacNeilage (ed.): Speech Production. Springer Verlag: Heidelberg. Lisker, L. (1972). On Time and Timing in Speech. In:T. Sebeok,(ed.): Current Trends in Linguistics XII. The Hague: Mouton. Lyberg, B. (1981). Temporal Properties of Spoken Swedish. MILUS 6, Dissertation, Stockholm University. Netsell, R., Kent, R. and Abbs, J. (1978). Adjustments of the Tongue and Lips to Fixed Jaw Positions during Speech: A Preliminary Report. Conference on Speech Motor Control, Madison, Wisconsin. Nooteboom, S. (1972). Production and Perception of Vowel Duration. Dissertation, University of Utrecht. Perkell, J.S. and Nelson, W.L. (1982). Articulatory Targets and Speech Motor Control: A Study of Vowel Production. In: Grillner et al. (1982), see above. Rapp, K. (1971). On Syllable Timing. PILUS, Stockholm University.

Phonetic Validation of Distinctive Features: A Test Case in French Leigh Lisker, Haskins Laboratories New Haven, CT., U.S.A. and University of Pennsylvania, Philadelphia, PA., U.S.A. Arthur S. Abramson, Haskins Laboratories New Haven, CT., U.S.A. University of Connecticut, Storrs, CT., U.S.A.

and

Much of the phonological literature shows little concern for recent phonetic data. Even in a provocative overview of Jakobsonian phonology (Jakobson and Waugh, 1979) that does give much attention to recent phonetic research, the latter is not exploited very convincingly in defining certain distinctive features. A case in point is the notorious French chestnut embodied in vous la jetez vs. vous I'achetez, a pair of expressions traditionally said to be distinguished by a voicing feature in the palatal fricatives, which appear here as initial elements in consonant clusters with / t / . It is reported, however, that the /•$/ of jetez is devoiced through assimilation to the following / t / , and it is argued that a feature of 'fortisness' or 'tensity' is therefore needed. We have tested two hypotheses: (1) Such pairs are likely to be distinguished in production and perception. (2) When they are distinguished, the phonetic basis is glottal adduction vs. abduction. Readings by native speakers of standard French of written sentences terminating in la jeter and I'acheter were collected and those tokens in which the terminal items were pronounced as disyllables were presented to French listeners for identification. Their responses suggest instability of the distinction, with a perceptual bias toward / J / , thus largely negating the first hypothesis. Insofar as the distinction is maintained, spectrographic analysis and perceptual tests involving the manipulation of / $ / and /J7 noise segments do not argue against a hypothesis of laryngeal control. 1. INTRODUCTION

The once intimate connection between phonetics and phonology has in recent decades been stretched to the point where many linguists practice in one area to the neglect of developments in the other. For some linguists the neglect may well be 'benign,' but for many it would seem to have become a matter of principle. Ilse Lehiste, whom we honor in this volume, is among the very few linguists who refuse to divorce phonological from phonetic concerns, and to make of phonology a purely formal exercise in the elaboration of ostensibly phonetic statements whose scientific respectability derives more from traditional opinion than from rigorous testing. If phonology is to

184

Leigh Lisker and Arthur S. Abramson

be taken seriously as more than an elaborate spelling exercise - in other words, if the assertions of phonetic fact are not just objects to be manipulated but rather statements whose truth values are thought relevant to linguistic description - then they deserve the respect implied by careful and appropriate testing. Terms such as 'voiced' and 'fricative' have physical meanings that are generally recognized. Provided that the linguist who says that a given utterance type involves a voiced fricative grants physical meaning to those terms, the statement may be checked against physical observation. Linguists may not want to test their phonetic judgments, even though ostensibly they are making claims about the physical nature of speech signals. Quite frankly we find such an attitude deplorable, even if we acknowledge that beliefs about the nature of the world are also facts worth studying. Some kinds of phonetic judgments are, moreover, not easily translated into terms that allow ready testing. An outstanding example is the claim that two utterance types are distinguished by a difference in force of articulation, where the so-called 'fortis-lenis' distinction is attributed to particular segments. It might be argued that if the fortisness of a particular segment is a matter of belief that is widely shared, then it may not be dismissed as groundless just because laboratory phoneticians have failed to find an appropriate measure. But there is a difference between taking such a belief seriously and regarding it as sacrosanct. We prefer to take it seriously, and that means to view it critically. The claim that a phonological distinction is based on a fortis-lenis difference is not easily tested for another reason, namely because most often a non-controversial difference is present, one that is physically interpretable. Only rarely is an alleged fortis-lenis difference unaccompanied. One of these cases seems to be in French, a language that distinguishes two sets of obstruents, one usually voiced and the other voiceless. A number of linguists (e.g., Armstrong, 1932; Delattre, 1941; Malmberg, 1943), most recently Jakobson and Waugh (1979), have said that the palatal fricatives / $ / and / J / , usually voiced and voiceless respectively, are lenis and fortis as well. They claim, moreover, that in the phrase Vous la jetez 'You throw it' a common pronunciation omits the schwa that in a more deliberate style separates the / $ / and the / t / , and also devoices the fricative. The resulting form, it is further said, is distinguishable from the semantically different expression Vous I'achetez 'You buy it,' despite the alleged absence of any voicing difference. The aim of the exercises to be reported here was to test the proposition that the distinction just described cannot be attributed to a difference in laryngeal action, and that we must look for something else that can plausibly be regarded as a consequence of a difference in articulatory force. The strongest acoustic evidence for a difference in laryngeal management would be the presence of glottal pulses during the fricative noise of /•$/, and the absence of such pulses during the /]"/ noise. The acoustic indices of articulatory force that are commonly proposed are duration and intensity

Phonetic Validation of Distinctive

Features

185

level, in this case the relative durations and intensities of the / $ / and / J / noises. (It must be pointed out that, on the one hand, the absence of glottal pulses during the /•$/ noise does not conclusively demonstrate that the laryngeal action is the same for /•$/ and / J / , while a difference in either noise duration or intensity may as plausibly be attributed to a difference in laryngeal management as to one of articulatory force.) 2. THE TESTS

Three tests were run: first, native speakers of French recorded a set of sentences read from a written list, and the recordings were played back to French listeners for identification of the intended target forms; second, selected sentence tokens were edited so that fricative intervals from well-identified jeter and acheter were interchanged; finally, the intensities of the fricative intervals were varied to determine whether this would affect listeners' identifications of the sentences. The first test was run just to make sure that sentences meant to differ only as to whether they contained jeter or acheter could be distinguished if pronounced with fricative-stop clusters. Three speakers of standard French were recorded in readings of the following sentences. The sentences were listed in a random order. II faut la jeter. U faut l'acheter. II ne faut pas la jeter. II ne faut pas Pacheter. II devrait la jeter. II devrait l'acheter. On a fini par la jeter. On a fini par l'acheter. Elle a fini par la jeter. Elle a fini par l'acheter. J'ai decide de la jeter. J'ai decide de l'acheter. Elle ne pouvait pas la jeter. Elle ne pouvait pas l'acheter. Est-ce que vous voulez la jeter? Est-ce que vous voulez l'acheter? On dit que vous voulez la jeter. On dit que vous voulez l'acheter. Moi, j'ai peur de la jeter. Moi, j'ai peur de l'acheter. Moi, je ne veux pas la jeter. Moi, je ne veux pas l'acheter.

186

Leigh Lisker and Arthur S. Abramson

Est-ce que vous ne voulez pas la jeter? Est-ce que vous ne voulez pas l'acheter? One speaker read all the sentences containing jeter with this word pronounced as a disyllable. Since her productions could not be used to test our hypothesis, they were discarded. A second speaker always pronounced jeter as a monosyllable, while the third nearly always did so. Randomizations of the sentences recorded by these latter two speakers were played back to native listeners, both the speakers and others. The listeners' judgments as to the identity of the final words (if you like, their judgments as to the speakers' intentions) are presented in Table 1. Speaker G.P., who pronounced all his tokens of jeter as monosyllables, very clearly produced sentences that were ambiguous; roughly two thirds of both intended jeter and acheter were judged to be the latter by the three listeners who rendered a total of 280 responses. In the case of D.E.'s readings, although intended acheter were more often reported as acheter than were intended jeter, it can hardly be said that the 704 responses by four listeners provide strong evidence that the / 3 / - / J / distinction can survive deletion of the schwa oijeter. D.E.'s intended jeter were so identified just at chance; her acheter tokens, reported 60% as acheter, were perhaps more often produced with fully voiceless fricative-stop clusters, combinations that might predispose listeners to report acheter. Chi-square tests of individual listeners' responses revealed only a single case in which a speaker's intended forms were correctly identified at better than chance: D.E. as listener was able to identify her own recorded sentences at a level better than ρ < .001. Table 1. Labeling of Original Recordings Speaker: G.P. 3 listeners 280 responses Intended

Reported

jeter acheter

jeter 34% 30%

acheter 66% 70%

Speaker: D.E. 4 listeners 704 responses Intended

Reported

jeter acheter

jeter 51% 40%

acheter 49% 60%

Phonetic Validation of Distinctive Features

187

The data of our first test suggest that there is little basis, at least for these speakers and listeners, for the claim made as to the robustness of the / 3 / - / J / contrast in the context under study. The fortis-lenis difference, so hard for the laboratory phonetician to lay hands on, seems to be no less elusive for our French speakers and listeners. Of course, while our test subjects are certifiably native speakers of French, and the claim is about French, somewhere there may be whole communities of speakers who behave as the claim we are testing says speakers of French do generally. But at the moment we do not know whether or where they are to be found. At this point we might have dropped the whole matter. We were persuaded to continue, however, by the following consideration. If we could find any sentence tokens with intended jeter that were so identified, and that we could say were produced in accord with the schwa-deletion rule, and if we also found other tokens regularly judged to contain acheter, then we might still pose the original question: does a difference in labeling responses require us to recognize a phonetic basis other than laryngeal? Of the more than 40 sentences that D.E. recorded containing intended jeter, just three were reported, at 90% or better, as ending with jeter. Of an equal number of tokens with intended acheter there were six that were as often so reported. Our data do not compel the conclusion that these particular tokens reflect real auditory/phonetic differences, since purely random labeling behavior might have yielded the results obtained. On the other hand, we cannot absolutely reject the possibility that these jeter and acheter tokens differ acoustically in a way that can explain why listeners reported them differently. We proceeded therefore to examine spectrographically all the unambiguously labeled sentence tokens, looking for differences that might consistently distinguish members of the two sets, and, if such were to be found, determining whetherthey were of laryngeal or extra-laryngeal origin. Figure 1 reproduces narrow-band spectrograms of two sentence tokens with well-identified jeter and acheter. The short vertical lines at the base of each spectrogram mark off the fricative noise intervals. The two intervals differ very little in duration (perhaps 5%), but they do differ in two other aspects. The amplitude profile for the fricative of acheter has a higher peak value, and this is as proponents of a fortis-lenis distinction would predict, although it is also consistent with the higher airflow that should result from the abduction of the vocal folds that occurs in voiceless fricatives. The other difference is in the extent to which the harmonic pattern that characterizes both signals just before the fricative intervals persists past the onset of the noise. In the upper spectrogram of Figure 1 the harmonics fill well over half the fricative interval; in the lower one they damp out much earlier. The spectrograms do not tell us whether amplitude or voicing is perceptually significant, but they suggest that perhaps one or both of them may play some role.

188

Leigh Lisker and Arthur S. Abramson

II n · faut p a · la J· tar.

Η faut raehatar. Fig. 1. Narrow-band spectrograms of sentences with well-identified tokens of jeter and acheter. The short vertical lines mark off the fricative noise intervals.

In order to see whether the category assignments of the items differently labeled can be ascribed to the fricative segments, we selected four sentence tokens, two for each reported word, for further testing. For each token the fricative segment was first excised with the help of a waveform editing program, and then each of the four segments was in turn introduced into the gaps left in each of the sentences. The 16 acoustically different signals were then presented in random order to three of our French listeners. Their responses are represented in Table 2. Each number in the table represents the averaged responses to four stimuli. For example, the four combinations of the two / $ / noise segments and the two contexts that originally included those segments elicit an average of 77% jeter identifications. The four combi-

Phonetic Validation of Distinctive Features

189

nations of those same contextual signals with / J / noises elicited, on the average, only 35% jeter judgments. Combinations of / J / noises with their proper contexts were reported 75% as containing acheter. The same contexts with /·$/ noise yielded stimuli that were quite ambiguous. When the responses of each listener were submitted to a simple Chi-square test of significance, only one was found to distinguish reliably between the two classes of stimuli (p < .001). Possibly it is significant that this listener was the speaker D.E. The fact that two of our three listeners failed to distinguish two categories makes still more doubtful the proposition that jeter and acheter maintain phonetic distinctiveness in contexts of the kind tested, in the absence of the schwa that elsewhere marks jeter, even if there seem to be differences in the extent to which voicing accompanies frication. The fact that the percentage 'correct' scores obtained were lower than the 90% obtained for the test tokens in the initial labeling test is not readily explained, but it can be pointed out that three of the four stimuli on which each of the values given in Table 2 is based were 'unnatural' combinations of frication noises and sentence contexts, and the process of cutting and recombining may well have introduced incongruities of intensity, duration and fundamental frequency that could contribute to listener uncertainty. Our last test involved no commutation of segments. Instead, the four noise intervals were presented in their native contexts, but at two intensity levels. In the acheter sentences the fricative segments were played back at their original levels and also with 10 dB attenuation. The corresponding segments in the jeter sentences were also replayed at their original intensities, and at intensities 10 dB higher. As Table 3 shows, the effects of modifying the intensities of these segments are not spectacular; acheter responses decreased little more than 10% with decreased noise intensity, while jeter responses actually increased with increased intensity, possibly reflecting the effect of the increased salience of the voicing harmonics. Chi-square tests of the responses of the four listeners who underwent this test showed that varying the noise intensities had no statistically significant effect on labeling behavior. Table 2. Responses to Cross-Matched Fricative Noises Speaker: D.E. 3 listeners 192 responses Fricative noise from acheter

Fricative noise from jeter Context from

jeter acheter

Reported jeter 77% 50%

acheter 23% 50%

Context from

Reported

jeter acheter

jeter 35% 25%

acheter 65% 75%

190

Leigh Lisker and Arthur S. Abramson

Table 3. Responses to Fricative Noises at Two Intensity Levels Speaker: D.E. 3 listeners 160 responses Intended

Reported jeter

jeter

0 dB 75%

acheter

-10 dB 73%

+ 10 dB 80% acheter 0 dB 85%

To conclude, we have little reason, on the basis of the data gathered in the course of this study, to believe that speakers of standard French reliably maintain the contrast between a sentence pair vous la jetez and vous I'achetez in the absence of differences of vocalization and voicing. Thus the alleged basis for an independent fortis-lenis contrast in French seems to us to be very possibly entirely illusory. However, even //sporadically we find well-identified fricative-stop clusters that hint at a contrast, we find no compelling evidence to reject an explanation in terms of a difference in laryngeal behavior. ACKNOWLEDGEMENTS

This is an elaborated version of a paper presented at the 56th Annual Meeting of the Linguistic Society of America, New York City, December 1981. The work was supported by NICHD Grant HD01994 and BRS Grant RR05996 to Haskins Laboratories. REFERENCES Armstrong, L.E. (1932). The phonetics of French. London: Bell and Sons. Delattre, P.C. (1941). La force d'articulation consonantique en fran?ais. The French Review, 14, 220-232. Jakobson, R. and Waugh, L. (1979). The sound shape of language. Bloomington: Indiana University Press. Malmberg, B. (1943). Le systeme consonantique du frangais moderne. Lund: Gleerup.

Cognitive Phonetics - Some of the Evidence Katherine Morton, University of Essex, Colchester, U.K.

Cognitive phonetics has been proposed (Tatham, 1983, 1986) to try to deal with the theoretical problem concerning the link between phonology and phonetics. The source of the problem is that two areas of linguistics have developed as different types of model. Phonology has been concerned with hypothesizing mental processes in dealing with establishing the sound pattern and phonetics with the neuromuscular events related to physical realization of the output of these phonological processes. The difficulty lies in trying to account for the physical implementation of mentalistic units, or looked at the reverse way: how can a mentalistic unit be programmed into the neuromuscular system? Cognitive phonetics is about the mental processes involved in manipulating the vocal tract mechanism. It is concerned with the control of the speech mechanism after the phonology has established the sound pattern of the language. Cognitive phonetics differs from phonology in that phonology selects from an inventory of sounds provided by phonetics; cognitive phonetics then makes decisions about how to realize the requirements of the phonology. I should like to try to substantiate the view that cognitive phonetics is not the same as a low-level area of phonology by looking at particular types of data. The speech mechanism is configured in various ways to produce sound. Within this range of sound produced, the hearer can identify different groups of sounds and differentiate between them. The human perceptual system has limits of discrimination which constrain the size of the inventory of sounds usable in the language. This set of sounds is further filtered by the need to be able to reproduce these sounds in the same manner in a relatively unvarying fashion. They must be repeatable. Thus the set of physical sounds which a phonology may employ as mental representations can only have members which are (a) able to be perceived as distinct from each other, and (b) able to be produced reliably such that they can be perceived unvaryingly by the hearer. Sounds that are capable of being perceived (a) and those that can be produced for perception (b) are variables dependent on the overall inventory selected by the phonology. Two sounds, SI and S2, might not satisfy condi-

192

Katherine Morton

tions (a) and (b) in a particular language. They may not be opposed in any encoding or decoding operation, but S1 may be used in one language and S2 in another. So SI and S2 are measurably different in the laboratory, are available for selection for a phonological inventory, but cannot both be selected for the same language. An example of this is [s] in English and French. In general, [s] is articulated in French with a narrower constriction and greater intraoral air-pressure than in English. This gives rise to greater aerodynamic post-constriction turbulence and a higher frequency and amplitude. The mental representation of both sounds at the phonological level is identical, in that [s] is different from [5] and [f], etc. The phonological decision for both languages is to have the single abstract object we symbolize as / s / ; the phonetic decision for French is that this should be actualized with narrower constriction than for English. This cognitive process is phonetic, not phonological, since it has nothing to do with the sound pattern. It is also language-dependent but not a property of the vocal tract mechanism itself, since a single individual can produce either sound at will, given a bit of practice. The constriction, which means width of gap between tongue-front and palate, is a physical variable which is under cognitive control. It is limited by the mechanics of the vocal apparatus. The apparatus is able to produce this kind of constriction; the phonology can abstractly employ it for sound pattern encoding; the phonetics can choose a degree of precision within the mechanical limits for production. What cannot happen is that the phonology selects an English type and a French type sound in the same language, (1) because the degree of phonetic control available cannot guarantee a precise enough acoustic event to enable reliable perceptual discrimination, or (2) because perceptual discrimination is not fine enough even if the phonetic control is there. Another example: phonetic realization of the phonological feature [voice] in stop consonants produces the articulatory effect of delay of vocal cord vibration onset for vowels immediately following these consonants. Many researchers have shown that the actual average delay varies from language to language even in those languages where only two segments are being realized and differ between [+voice] and [-voice]. The delay, called VOT (voice onset time), which is phonologically dominated by the same [voice] feature in two different languages, may be systematically different. This difference, which is under voluntary control but which is systematic, is, controlled cognitively. As another example, take the different tongue heights involved in actually pronouncing the vowel / i / in French and English. French is consistently higher than English, but by the same reasoning as before, no single language may have both [i] of the French type and [i] of the English type. The phonological mental representation is identical; the phonetic representation is different. The precision with which these sounds are produced in each

Cognitive Phonetics - Some of the Evidence

193

language is related to the contrasts required by the phonology of that language. As a further example, consider / s / as realized phonetically in standard European Spanish. Phonetically this approaches [δ] in English, having a wider constriction, less turbulence and lower acoustic amplitude and frequency than English [s]. Probably also there is wider variation in its articulation. The reason for this is that Spanish has chosen not to have, phonologically, contrasting / s / and / s / , leaving open, for phonetic realization of / s / ; a greater acoustic range than could be allowed in English without causing perceptual confusion and possible problems of accuracy of motor control. These examples involve decisions by the speaker about selection from a range of possibilities. When a particular realization of a phonological requirement is possible, then a process of selection occurs. It is clearly not left up to the vocal mechanism itself to freely realize abstract phonological requirements. It is suggested that a cognitive phonetic capacity controls this selection for the purposes of the phonology. There is another aspect to the role cognitive phonetics plays. Realizations of intended phonological units which are repeated exhibit variations. The output of the vocal mechanism is not in a one-to-one relationship with the underlying representation of the utterance in the phonology. In the phonology segments are either the same or different, which means unambiguously able to function distinctly for encoding purposes. However, in deciding the inventory for this functioning, control of variability is a major consideration. Too much closeness or overlap in realization and the distinction between units cannot be made. Cognitive phonetics controls the amount of separation necessary for the hearer to make a necessary phonological distinction and dictates the amount of variability acceptable. In addition, we can observe that variation in articulation of segments or features of segments is often expressed differently from language to language or within one language between phonological context. Thus a segment might show on repeated articulation a wider variation, or less precision, than the same segment in another language. The existence of variation is inherent in the mechanical system, but since the variation is systematically different in different languages or in different contexts then the degree of the variation is under control. That control is cognitive since it involves a computation based on the sound pattern of the language. This also results in an increase in the size of the potential phonological inventory. If one thinks of phonological segments as occupying a perceptual space, the range of variation determined by the mechanical system takes up a particular area. A phonology would be unable to use segments whose area overlapped. So intrinsic variability, that variability determined by the mechanical system, constrains phonological candidature for a segment.

194

Katherine Morton

However, if it were possible to reduce these areas by special control so that there would be no overlap and this control could be consistently precisely maintained, adjacent segments could be used in their abstract versions in the phonology. It seems to be the case, for example, that where a language has vowels 'bunched' in a particular part of the vowel space, such narrowing of variation takes place. Or another example, the front/back range of variability in palatal consonants is narrower the more of these consonants a language has. It's thought to be unlikely that a mature speaker sets up a phonological inventory each time he speaks. But there is one phenomenon that has been observed which suggests some sort of active ongoing cognitive process. Precision, or narrowness of area of variation, tends to vary itself during utterances for the articulation of the same segment in different contexts. This may mean that degree of precision of realization of the segment is not set once and for all by some phonological consideration but can vary as and when necessary. It is suggested that such precision is cognitively dominated. It involves a knowledge of the perceptual confusion effects of certain phonological contexts, and involves knowing if and to what extent certain phonetic contexts are going to suddenly change the variability of this segment's realization. It also involves a computation based on these factors which can assess their relative weighting of importance. In considering a theory of cognitive phonetics, one needs to examine speech with a view to determining the nature of intervention in the actual speech production mechanisms. It seems clear that there can be such intervention. A theory which describes the output of a phonology and the intrinsic properties of the vocal mechanisms cannot on its own explain the data observed in the laboratory. Although there is of course a great deal that remains for phonology to describe and we don't know everything about motor control and aerodynamics, it is possible to see systematic phenomena which cannot be explained by either. These phenomena do not appear to be generated automatically, they seem to have mental representation, yet not be strictly phonological. These are the phenomena which constitute the evidence for a cognitive phonetics. REFERENCES

Tatham, M.A.A. (1983). Cognitive Phonetics. In: S.K. Ghosh (ed.) Human Language: Biological Perspectives. Tatham, M.A.A. (1986), Cognitive Phonetics - Some of the Theory. In this volume.

Oral Vowel Reduction in Brazilian Portuguese Maria Alzira Nobre and Frances Ingemann, University of Kansas, Lawrence, KA., U.S.A.

1. INTRODUCTION

Reduction of vowels in unstressed syllables occurs in most, if not all, languages to some degree, although the amount and type varies from language to language. Factors which have been found to differentiate stressed from reduced vowels have included vowel quality and duration (Lehiste 1970, Gay 1978λ, b). Brazilian Portuguese is normally analyzed as having three different phonemic oral vowel sets depending on location of stress: /i e ε a ο ο u / in stressed syllables, / i 3 ο u / before stressed syllables, and /i s u / in unstressed word-final position (Mattoso Camara 1972). For convenience, vowels in syllables preceding stressed syllables will be called by the traditional term pretonic. Few acoustic studies of Portuguese have been published and most of these deal with Continental Portuguese (see Godinez 19786 for a bibliography of phonetic studies of Portuguese). Of those on Brazilian Portuguese, only two are relevant to this study. Godinez (1978a) measured formant frequencies of stressed vowels in Brazilian Portuguese and compared them with those reported by Delgado Martins (1973) for Continental Portuguese. Major (1979, 1981) measured syllable length as a byproduct of his study of interstress intervals in Brazilian Portuguese. At the time we began our research, to our knowledge there had been no acoustic study of vowel reduction in Brazilian Portuguese. Our study was designed to determine the extent of vowel formant frequency change as the contrastive sets went from seven phonemes in stressed position, to five in pretonic position, and three in word-final position. We were also interested in what durational differences might accompany the qualitative differences. 2. PROCEDURES

Words for this study (Table 1) were selected so that vowels in the three stress conditions appeared in so far as possible in similar phonetic environments. The principal corpus (hereafter referred to as Set A) consisted of 21 words: 9

196

Maria Alzira Νobre and Frances Ingemann

Table 1. Words Used for Analysis Set A. Words elicited in the Frame 'Diga 1. 2. 3. 4. 5. 6. 7. 8. 9.

chique seca cheque saca choque soco suco Chica saco

['Jiki] ['seks] ['feki] I'saka] ['Joki] ['soku] f'suku] ['Jlka] 1'saku]

'chic' 'drought' 'check' 'sack' 'shock' 'blow with the fist' 'juice' a nickname 'bag'

mais uma vez, ta?' 10. 11. 12. 13. 14. 15. 16. 17. 18. 19. 20. 21.

chiqueza secura checado sacola chocado soquete sucoso sicario sequestro sacrario socorros sucata

[Jl'keza] [se'kurs] [Je'kadu] [sa'kola] [Jo'kadu] [soket y i] [su'kozu] [si'karyu] [se'kwestru] [sa'kraryu] [so'koxus] [su'kata]

'elegance' 'dryness' 'checked' 'handbag' 'shocked' 'ramrod' 'juicy' 'hired assassin' 'kidnapping' 'sanctuary' 'aids' 'scrap iron'

Set B. Phrases with Final Vowels Between Voiced Consonants 1. tese grada ['tezj/grada] 'scholarly thesis' 2. gole grande ['goli/gräd y i] 'big gulp' 3. Lisa Gomes [liza 'gomis] a female name 4. gola grande ['gab 'gräd y i] 'big collar' 5. jogo duro ['30gu 'duru] 'hard game' 6. siso duro ['sizu 'duru] 'hard wisdom tooth' Underlined vowels were used for formant frequency comparisons.

disyllabic words with stress on the first syllable and 12 trisyllabic words with stress on the second syllable. The disyllabic words were selected to provide at least one example of the 7 phonemic stressed vowels and 3 examples each of the 3 phonemic unstressed word-final vowels. In the disyllabic words, all stressed vowels occurred in the environment / s _ k / or / J _ k / ; all unstressed vowels in the environment / k _ # m / . The 12 trisyllabic words were stressed on the second syllable. Seven were words derived from 7 of the disyllabic words representing the seven stressed vowels, providing examples of how each of these underlying vowels is realized as one of the 5 phonemic pretonic vowels. To these were added five unrelated words containing each of the 5 phonemic unstressed pretonic vowels in the environment / s _ k / or / J _ k / . Because there were no consistent differences between pretonic vowels derived from stressed vowels and other pretonic vowels, all pretonic vowels have been grouped together in the analysis. Words from Set A were placed in the carrier sentence 'Diga mais uma vez, 'ta?' 'Say one more time, OK?' The tag question was added to induce the speakers to place sentence emphasis on mais or vez. The speakers were asked to read the sentences naturally and were allowed to record the sentences as many times as they wanted or needed.

Oral Vowel Reduction in Brazilian Portuguese

197

Because unstressed final vowels after / k / often turned out to be partially or totally devoiced, accurate measurement of their formant frequencies was difficult. To provide additional data on final vowels, a supplementary set of 6 phrases (Table 1, Set B) with unstressed word-final vowels in voiced environments was added. The sentences and phrases were read by four adult male speakers of three different dialects of Brazilian Portuguese: two Carioca, one Gaucho and one Mineiro. 1 Spectrograms were made on a Kay Sonagraph Model 6061-B. Formant frequencies were normally measured at a point after transitions where both F l and F2 seemed to have levelled off and become stable. Measurements were made of the F l and F2 frequencies of: 1) stressed vowels in the 9 two-syllable words of Set A 2) pretonic vowels in the 12 three-syllable words of Set A 3) final unstressed vowels in the original 9 two-syllable words of Set A and in the six additional words with voiced consonants of Set B. Since each word was read by 4 speakers a total of 144 vowels were analyzed. Durations of Set A utterances, test words, and syllables, vowels, and consonants within the test words were also measured. No intensity measurements were made but visual inspection of the spectrograms shows, as expected, that unstressed vowels appear to be lower in intensity than stressed ones, especially in final position. 3. RESULTS

Formant frequency Mean formant frequencies for vowels in Set A are given in Table 2. Comparison of Stressed and Pretonic Vowels Mean formant values of stressed and pretonic vowels are plotted in Figure 1. One continuous line connects stressed vowels and another continuous line connects the pretonic vowels. Broken lines show the acoustic displacement of the unstressed vowels with respect to the stressed ones. Mean Fl frequencies of pretonic vowels are at a point halfway between those of adjacent vowels, which corresponds to a lowering of the higher vowels [i, e, o3 u j and a raising of the lower vowels |X> ßj °1· Mean F2 frequencies of the pretonic vowels [e, a, 8, u] all indicate centralizing tendencies, with the back vowels exhibiting more centralization than [e]. However, the F2 of [T| has nearly the same frequency as its stressed counterpart. Since F2-F1 correlates well with backing (Lindau 1978), there appears to be little if any backing for [T]. The articulatory pattern of vowel reduction from stressed to pretonic vowels in Portuguese can be summarized as follows: 1) There is less displacement of the high vowels on the vertical axis than of

198

Maria Alzira Nobre and Frances Ingemann




3

C/2

es

£

t s Tt

ο Μ

in m

ο tN

ε ο β **

^ —

ON m

Ό c cd

Ο C 2 - g S

fifi s vi

~ (Λ

ίί t

3 Ο υ

$ Ο

Μ Ο Η

Rate and Reduction

221

explained partially above. On the average, there was a 27% reduction in sentence duration at each stage slow — normal — fast. Tables 2a,b display the behaviour of the three consonant clusters under observation. They show the durations in milliseconds of the consonants in the clusters, if clusters there be, the number of tokens showing clusters, the durations of the fricative consonants remaining when the stop closure is deleted, and the number of tokens of this second type. 4. REDUCTION CHARACTERISTICS

Subject 1, in the 'natural' condition, shows no reduction in the slow rate. In the normal rate, one finds no reduction in the -st# clusters, but only six of the clusters are realized. At the fast rate, one finds reduction in all three clusters, five reductions before a vowel, three before a consonant, and nine in the sts clusters. In the 'metronome' condition, Subject 1 shows no reduction at the slow rate. At normal rate, she shows two reductions in the st#C cluster and eight in the sts cluster. At fast rate, there are seven reductions in the st#C cluster and all ten of the sts clusters are reduced. In the 'fast and careful' condition, the reduction is similar to that at the normal rate, though the consonants are shorter. Subject 2 also shows no reduction in the natural, slow condition. At the normal rate, he begins to reduce: twice for st#V, five times for st#C, and eight times for sts#. At the fast rate, only one sf#Fcluster remains, and there are no s i # C or sts clusters. In the 'metronome' condition, we see no reduction for st# clusters at the slow rate, but five of the sts clusters reduce. At the normal rate, three st#V and two st#C clusters reduce and, again, half of the sts sequences. At the fast rate, nine sf#Fclusters reduce, five st#C, and nine sts clusters. Unfortunately, there was a recording error for Subject 2 in the 'fast and careful' set of sentences, and they were not measurable. As a summary of the reduction results, there is consistently greater reduction at a faster rate. In general, the most reduction appears in clusters, next most in st#C sequences, and the least in st#V sequences. (There is only one condition where the last two are reversed.) At a normal rate, there were 39 reductions out of a total of 120 tokens, i.e., overall about 33% of the tokens reduced, 13% of the st#V type (all contributed by one speaker); 23% of the st#C type, and 63% of the sts# type. At a fast rate, there were 107 reductions out of 160 tokens - about 67% of the tokens reduced overall. Of the st#V type, 53% reduced, of the st#C type, 58%, and of the sts# type, 92%. Slightly less reduction is found in the fast and careful data for Subject 1 than in the fast data.

222

Linda

Shockey

5. TEMPORAL CHARACTERISTICS

In general, all segments became shorter in duration as rate increased. There were only two counterexamples. This is not surprising when it is considered that at the fastest rate there is an average (for both speakers) of about 60 msec, available per segment as opposed to 119 msec, at the slowest rate. Deletion of the closure in these clusters does seem to save time, but not strikingly. Out of 21 cases where it is possible to make a comparison between the reduced and unreduced clusters, the following distribution is found: saving in time of: over 40% 30-39% 20-29% 10-19% 1-9% no saving or increase

number of cases: 1 2 3 7 3 5

The average reduction in duration with deletion of closure is just 20 msec, or 15%. Whether this can be significant is a matter for speculation. When the closure is missed out in sts sequences, the resulting fricative is longer at all rates than the corresponding s reduced from st clusters - as if it is, in fact, a cluster temporally. Subject 2's sts clusters in the metronome condition are of special interest in this respect. The t in the st#C sequence is consistently shorter than that in the st#V sequence, no doubt indicating that the stC sequence is integrated into a consonant cluster. In fact, the duration of the t in the [st0] cluster is very similar to that of the t in the sts cluster. (The rather striking exception to this generalization is Subject 2's t in st#C clusters at the slow rate: he consistently used the voiceless stops as 'holds' or 'lengtheners' when speaking unnaturally slowly.) This shortness may give some clue as to why the t deletes more frequently in these environments (see Neu, 1980 for a survey of work on this phenomenon). 6. DISCUSSION

Let us examine the question of how well the data substantiate a theory that suggests that phonological reduction is directly linked to increase in speech rate. We might set up two alternate forms of such a theory, the first claiming that speakers articulate faster and faster, progressively shortening segments until

Rate and Reduction

223

they speak so fast that they have to leave out or modify some of the segments in order to articulate the message in the short time allotted. The present data could not be said to provide evidence for such a view because, even though there were some circumstances under which no closures were achieved, e.g., Subject 2's sts clusters in the natural, fast condition, these are always balanced by other circumstances where the speaker was articulating at the same rate but did manage at least one closure (see, e.g., the same cluster in the metronome, fast condition for Subject 2). The second version would be that as speech rate increases, it becomes harder to articulate certain segments, so that phonological reductions occur unless extra effort is exerted, i.e., reductions are related to speech rate, but also reflect the degree of effort the subject is willing to make. Fast rate is a sufficient cause, but not a necessary one. The current data provide some evidence for this point of view as when, for example, Subject 1 is able to produce more t closures in her fast and careful metronome speech than in the fast metronome speech. In the former condition, she does not achieve the number of closures she demonstrated in her careful speech, and we are still left with the question of whether this result is because of inability to reach closure due to mechanical inertia or to other factors. However, the fact that she was able to produce a t closure for underlying st#C clusters six out of ten times at her fastest rate strongly suggests to me that she could reach closures consistently if there were some reason to do so. I find more promise in some approximation to the second version of the theory, since I think that phonetic realization works not on a principle of least effort, but on a principle of diminishing returns: human beings probably do not, in fact, ever speak so fast that it is impossible for them to realize all the phonetic distinctions which would be present in a slow version of any given utterance. (Please note that I am not saying that it is impossible to speak so fast that distinctions must be omitted... only that the option to do so is not chosen.) But, weighing the effort that would be needed to realize all these distinctions against loss of intelligibility, the phonology allows many reductions to occur. Clearly, more investigations are called for, especially with respect to the degree of precision possible at faster rates. Arriving at an algorithm for reduction as a function of intelligibility would be much more difficult, as the procedure is context-sensitive. Neither of the above hypotheses could explain the reduction of sts clusters which Subject 2 shows in his slow, metronome speech. But phonological reduction in slow speech is another topic, and one beyond the scope of this paper.

224

Linda Shockey

REFERENCES Harris, J. (1969). Spanish Phonology. (= Research Monographs, 54.) Cambridge, Mass. and London: MIT Press. Neu, H. (1980). Ranking of Constraints on / t , d / Deletion in American English. In: W. Labov (ed.) Locating Language in Time and Space. New York: Academic Press, pp. 37-54. Shockey, L. (1973). Phonetic and Phonological Properties of Connected Speech. (= Ohio State Working Papers in Linguistics, 17.) Columbus, Ohio: Ohio State University, Department of Linguistics. Zwicky, Α. (1972). On Casual Speech. In: Papers from the Eighth Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, pp. 607-615.

Rules for Assimilation of Voice in Dutch Iman H. Slis, Institute of Phonetics Catholic University, Nijmegen, The Netherlands

1. INTRODUCTION

A tremendous amount of research has been done on single voiced and voiceless consonants. Combinations of these consonants in clusters are less thoroughly studied. Previously, however, in the 'pre-instrumental times' of phonetic research, assimilation of voice was a topic among Dutch phoneticians. The views on the subject varied considerably and were sometimes even conflicting. For instance, Van Ginneken (1935) stated that assimilation of voice was largely regressive because of foreign influences; this regressive assimilation had so much in common with preslavic regressive assimilation that this could not be accidental. On the other hand, Pee (1948) concluded on the basis of a large amount of material that progressive assimilation is nearly always the rule in Dutch, although not carried through equally in all combinations. Demeulemeester (1962) was of the opinion that assimilation was not regular; he observed that ' . . . with radio broadcasters a specific system cannot be found for the direction their assimilations take . . ( 1 9 6 2 : 3 1 , my translation). 2. RULES FOR THE ASSIMILATION OF VOICE

Despite unpredictable assimilations, as observed by Demeulemeester, and differences in opinion, for instance Van Ginneken versus Pee, most investigators agreed on some general tendencies. Two main rules can be formulated for clusters consisting of plosives and fricatives: (1) Regressive assimilation is observed in the majority of two-consonant clusters if the second consonant is a voiced plosive; (2) Progressive assimilation is generally found if the second consonant is a voiced fricative. A third rule that is relevant states that: (3) All syllable final plosives and fricatives are devoiced in Dutch. Therefore, the cases of assimilation of voice under investigation consist of a voiceless plosive or fricative followed by a voiced plosive or fricative. Practically all investigators in the field observed a number of deviations from the rules; many of these deviations could be attributed to the dialectical backgrounds of the investigators and their subjects. Moreover, a great number of other possible influences were assumed. A survey of all the factors that were suggested are listed below in random order:

226

Iman Η. Slis

1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14.

Phonological compositions of the utterance (e.g. Cohen et al. 1971) Frequency of occurrence of the speech sounds used (Meinsma 1958) Spoonerisms (Meinsma 1958) Etymology (e.g. Van Ginneken 1935, Leenen 1954) Orthography (e.g. Eijkman 1933) Word accent (e.g. Eijkman 1933, Van Haeringen 1955) Semantic value of words (Van Haeringen 1955) Word boundaries (e.g. Meinsma 1958) Type of text (e.g. colloquial, official, . . . ) (Leenen 1954) Emotion (Van Ginneken 1935, Kaiser 1942, Meinsma 1958) Speech rate (e.g. Kaiser 1942 and 1958, Meinsma 1958) Experimental situation (e.g. Meinsma 1958, Pee 1948) Sex of the speaker (Demeulemeester 1962, Kaiser 1958) Dialect (e.g. Demeulemeester 1962, Eijkman 1933, Van Ginneken 1935, Kaiser 1958, Meinsma 1958) 15. Perceptual errors of the investigators (Meinsma 1958)

3. RULES BASED ON TWO FEATURES

The large number of suggested influences illustrates the complexity of the problem. Therefore it is not surprising that in generative phonology complex rules are used to describe assimilation of voice. In addition to the distinctive feature "voice", a second feature is introduced. In this context we find the features "foreign" (Mey 1968), "pause" (Tops 1974 referring to Brink 1970), and "tense" (e.g. Hubers and Kooij 1973). Among phoneticians also we find supporters of two independent features to describe the voiced-voiceless distinction (e.g. Debrock 1977, Kim 1965). In addition to the voicing feature they assume a feature incorporating articulatory effort, e.g. "tense" or "fortis". Others prefer one underlying feature which can be reflected in a number of articulatory and acoustic parameters (e.g. Slis and Cohen 1969, Slis 1970). Among Dutch phonologists a two-feature description seems to be most common. In the grammar proposed by Hubers and Kooij (1973) a final devoicing rule is hypothesized, which operates in three steps: (a)

[Obs] - [+ Tn^ /

(b)

[Ob| -

[-

(c)

[Ob| -

[ - a Vcej / [a

#

Vcej / [ + Tns| Tn|

These rules change all final obstruents into voiceless, tense obstruents. This corresponds to the general tendency we mentioned under (3) above. A second set of three operations concerns assimilation:

Assimilation of Voice in Dutch

(d)

227 Obs - Cnt Vce.

[Obs] - [ + Vcej / .

This operation voices all obstruents before voiced plosives and is, in fact, regressive assimilation of voice before voiced plosives (general tendency (1), above). (e)

[Obsj — [ -

Vce) /

Obsl - Vce

J

This operation devoice§ all obstruents after voiceless obstruents. Since in operation (d) all obstruents before plosives assume [ + Voice], only obstruents before fricatives remain voiceless. Consequently, step (e) describes progressive assimilation before fricatives (general tendency (2)). (f)

[Ob§ — [ + Tn§ /

Obs + Tns

#

In this operation all obstruents after [ + Tense] obstruents become [ + Tense]. Since in the two preceding assimilation rules the [ + Tense] character of the first consonant is not changed, all second consonants become [ + Tense]. This results in a cluster of two [ + Tense, - Voice] consonants if the second consonant is a fricative. These can be regarded as two normal voiceless consonants and therefore progressive assimilation takes place. If the second consonant is a plosive, however, the rules yield two [ + Tense, + Voice] consonants. Therefore regressive assimilation takes place with respect to voicing and progressive assimilation with respect to tenseness. The feature [ + Voice] promotes a voiced realization of the cluster and the feature [ + Tense] a voiceless realization. In this way an explanation is provided for the observed inconsistencies in assimilation before stops. A shortcoming of these rules is that in fact they generate question marks in the latter Case since they do not predict under what conditions the ambivalent [ + Tense, + Voice] result tends towards a voiced and when towards a voiceless realization. The literature indicated that deviations are to a certain extent predictable. F o r instance, Leenen (1954) observed that a group o f function words beginning with a / d / led to progressive assimilation instead of regressive assimilation. It has been suggested that for exceptions like this special rules have to be added. One may wonder whether for each of the possible influences that are suggested in the literature special rules ought to be introduced. This would lead to a very complex system. Moreover, a grammar based on binary features and unequivocal rules will never lead to inconsistencies as observed by Demeulemeester (1962) or to different grades

228

Iman Η. Slis

of partial assimilation as observed by e.g. Van Rijnbach and Kramer (1939). We think that a better description can be obtained by introducing gradual features such as proposed by Ladefoged (1975) and variable rules of the kind proposed by Labov (1972) for sociolinguistic problems. We will illustrate this by means of the results of an experiment in which we investigate the influence of speech rate on assimilation of voice. 4. EXPERIMENT: ASSIMILATION AS FUNCTION OF SPEECH RATE

The original aim of our experiment was to investigate whether the subjective data we found in the literature could be confirmed by objective measurement data. For this purpose we needed objective criteria which will be described below. In addition, we wanted to examine to what extent some of the factors mentioned play a role in changing the direction of assimilation. In the present experiment we first looked at speech rate. In addition, we examined influences of the phonological composition of the clusters and of the sex and region of origin of our speakers. Ten different sentences were used, containing thirteen obstruent-obstruent clusters in a pre-stress position. Four of the clusters contained two stops (/pb, tb, pd, t d / ) / , four a fricative followed by a stop ( 2 x / f b / , / s b , fd/), three a stop followed by a fricative ( 2 x / t v / , / p z / ) and two contained two fricatives (/sv, fz). Twenty-five subjects were instructed to read these ten sentences aloud first slowly, then at a normal speed and finally as fast as they could. The 25 subjects (students in the linguistics departments of Utrecht and Nijmegen) were not informed about the aims of the experiment. All subjects spoke standard Dutch although traces of their local pronunciation were observable. Among the subjects, eight men and eight women were from the same regional background; their speech was used to study the influence of sex on assimilation. Besides we could compare a group of four men and four women from Limburg (south-east) with a similar group from the middle-east region of the country, and a group of seven men from Holland (west) with eight men from 'other regions' (middle and south-east). 5. OBJECTIVE CRITERIA

We based our criteria on measurements on single voiced and voiceless stops. In intervocalic stops voice activity continued till about 10-20 ms (s.d. 10 ms) after the moment of oral closure (Slis 1968 and 1970). We assumed that a continuation of voicing, which we called "voice tail", between zero and 50 ms is normal in voiceless obstruents. If the voice tail exceeded 50 ms, we took it that the post-vocal obstruent became voiced by assimilation. In initial prevocalic voiced stops we found a voice onset before the

Assimilation of Voice in Dutch

229

moment of oral release, in other words a negative voice onset time (V.O.T.). With initial and intervocalic voiceless stops the voice onset occurred after oral release. Klatt (1975) observed for English that in clusters with a voiceless fricative followed by a voiceless stop, V.O.T. is advanced to the moment of opening. Although we did not examine this for Dutch, we observed (unpublished data) that the same holds for Dutch. We therefore defined the second obstruent voiced if the V.O.T. is negative and voiceless if the V.O.T. is zero or positive. One additional remark has to be made. We observed clusters in which voicing activity started during the closed interval after a preceding voiceless obstruent and in which voicing was interrupted a second time before oral release. We hypothesized that this second voiceless interval was due to aerodynamic conditions and that this second interruption was meant to be voiced. We defined the first time the voicing started during the closed interval to be the V.O.T. and ignored the second interruption. A similar reasoning held for clusters in which voicing was interrupted after a voice tail longer than 50 ms; in these cases we assumed that voicing was intended to continue and we regarded these as cases of regressive assimilation. Summarizing, we arrived at the following definitions for three categories of assimilation (see also fig. 1): 1. No assimilation: The first consonant is voiceless and the second voiced. The voice tail is shorter than 50 ms and the V.O.T. is negative. Voicing may be interrupted after the V.O.T.; this interruption is ignored. 2. Regressive assimilation: Both consonants are voiced. In most cases voicing continues during the closed interval. Voicing may be interrupted after a voice tail longer than 50 ms; this interruption is regarded as being due to aerodynamic conditions and will be ignored. 3. Progressive assimilation: Both consonants are voiceless. The voice tail is shorter than 50 ms and V.O.T. is zero or positive. Although slightly different conditions for our definitions can be thought of, the ones chosen proved to be satisfactory since they are easy to use; only in a few cases there was uncertainty about the right categorization. Also from the perceptual point of view objections can be raised. It is possible that a number of listeners will perceive a voiced consonant if V.O.T. is zero. However, a V.O.T. of zero is relatively easy to measure and, moreover, it is impossible to decide on a positive value for V.O.T. that meets all requirements. 6. MEASUREMENTS

For the sake of the determination of voice tails and V.O.T.'s it was necessary to measure the moments of oral closing and opening and of the end and beginning of voice activity.

230

Iman Η. Slis no

ass imitation

J \ A A A / i tail
ui at ec hugdizt nu hoggua noreg i hond J)er' (OH 1922:6613); 'Ec Jjottvmz nv Noreg i havnd t>er havova' (OH 1941:48413). The 1st sing, hegga, 3rd sing, hegge, etc., no doubt represent the earliest pret. subj. formation of this verb. Only one example of the corresponding pret. subj. of bua 'to live, dwell', viz. 3rd sing, begge, etc., appears to be on record: 3rd sing. 'beyGi' (or 'beyggi') in two late, mutually independent transcripts of (probably) an early 14th-century vellum manuscript of Hrafns saga Sveinbjarnarsonar (the Α-version) which is now lost, viz. the 17th-century AM 155 fol. (by Jon Erlendsson) and the 18th-century Thott 983 fol. (by Oddur Jonsson). The spelling 'ey', corresponding to Olcel. 0, no doubt derives from the lost vellum and might even go back all the way to the original of the saga itself, probably from the middle of the first half of the 13th century (MRD 1960:22323; see also pp. lxi-lxiv, lxxii-lxxiv, and lxxxvi-lxxxvii). The pret. subj. forms hegge, begge, etc., correspond to the pret. ind. (3rd plur.) hjoggo, bjoggo, etc. In these forms there is some evidence to show that the root vowel (in the combination jo < e by breaking) was mid ο rather than low 2 (see H. Benediktsson, 1963:429-30, 1979:81-2); this is true especially of Old Norwegian, while in the earliest Old Icelandic (in the period when Q a n d # were still distinct) the scarcity of examples makes it impossible to determine the quality of the vowel, the uniform spelling 'io' being ambiguous. In Old Icelandic of the 13th and early 14th centuries, on the other hand, there is clear evidence for the pret. ind. hjöggu, bjöggu, pret. subj. hjöggi, bjöggi,etc., e.g. in the 13th-century Gks 1157 fol. of Grägäs, Gks 2365 4° of the Poetic Edda, and Sthm Perg. 4° No. 2 of Olafs saga helga, and in the mid-14th-century Kroksfjaröarbok (AM 122 a fol.) of Sturlunga saga, whereas there is no record of such forms in Old Norwegian. After the merging of ρ a n d ^ t o ö, the

316

Hreinn

Benediktsson

pret. subj. was clearly remodeled on the basis of the plural of the pret. ind., which would be the regular descendant of earlier hjgggo, bjqggo, if these forms actually existed. The pret. ind. hjuggu, bjuggu, etc., pret. subj. hyggi, byggi, etc., were apparently already current in Old Norwegian in the 13th century, though the actually occurring examples - especially of the pret. subj. - are very few in number; in Old Icelandic these forms became the rule only in the course of the 14th century, though there are sporadic examples earlier, viz. in the early 13th-century AM 645 A 4° and the late 13th-century Morkinskinna (Gks 1009 fol.). In the 15th century the pret. subj. of höggva is regularly (viz. in all the six examples on record) spelled with 'iu' (or 'iv') (3rd sing./plur. 'hiuggi'), while there is only one example of this spelling in the pret. subj. of bua (2nd plur. 'biuggit' in Flateyjarbok, from the end of the 14th century); conversely, there is one example of the pret. ind. 'buggu' in Gisla saga Sürssonar in AM 556 a 4° from the 15th century (ed. 1929:43). The problem is whether 'iu' is to be assumed to stand for y (see H. Benediktsson, 1977:38) or for the combination ju. In the latter case, the pret. subj. hjuggi would have to be regarded as the result of a remodeling on the basis of the pret. ind. hjuggu, with no /-umlaut alternation, possibly on the analogy of the earlier inflection ind. hjöggu, subj. hjöggi. The later pret. subj. hjyggi (or, with the delabialization of y, hjiggi), the earliest example of which is probably the 1st sing, 'hiygga' in AM 309 4° (from 1498) mentioned above, could then be explained as being due to contamination of hyggi and hjuggi. In support of this the later difference between the two verbs höggva and bua is perhaps worth observing: while hjiggi (now regularly spelled hjyggi) is the predominant form later on, biggi (byggi) is the normal pret. subj. of bua\ bjiggi (bjyggi) occurs only sporadically, as does also the pret. ind. buggu. 5. THE PROTO-NORDIC DEVELOPMENT

As far as the 1st singular preterite subjunctive is concerned, the evidence of the Old Norse forms listed in §3 all points in the same direction: no -j- is present before the word-final -a even in those forms in which, if it was present in Proto-Germanic, it should have been retained according to regular phonological development. 10 Two different explanations seem to be possible, the choice between them depending, at least in part, on the assumed prehistory of this form (see §3). On the one hand,, if the Proto-Germanic formative was *-jau (or *-jöu) (replacing PIE *-ye-m), one would have to assume that the *-j- was lost by analogy in such forms as bfya, teka(Heusler, 1932:106).11 Since, for instance, the -j of the infinitive ofy'a-verbs, weak or strong, was not affected by any such analogy (cf., e.g., telja, bflpja), the determining factor for the analogical

1st Sing. Pret. Subj. in Germanic

317

remodeling of the 1st singular preterite subjunctive may well have been the coalescence of the present and preterite subjunctive formatives in all other persons, e.g. (taka), taker, take, etc. and (*tekja), teker, teke, etc.; of course, one would then have to assume that the analogical loss of -j- was extended to the preterite-present verbs (vita, skyla, etc.), even though the relationship of the present to the preterite subjunctive was quite different in these verbs, while the fact that the -j- was not lost by analogy in the 1st singular present subjunctive of ja-vtrbs (e.g. telja, teler, tele, etc., bfyja, bfyer, bi)pe, etc.) could well have been due to the influence of the infinitive and other present-stem forms of these verbs. On the other hand, if the Proto-Germanic formative was *-T(n), by generalization of the zero-grade optative morpheme variant, the Old Norse 1st singular in -a and the Gothic form in -jau might well represent two independent secondary formations, on the analogy of the present subjunctive; that is, ON bi\>a, teka, skyla might go back to *biöö, *tökö, *skulö< * -au and not to *-jau as in Gothic (Boer, 1920:195).12 The /-umlaut in the 1st singular is no obstacle: through its morphologization the /-umlaut, where applicable, became an integral and inseparable part of the preterite subjunctive derivation, so that even an early Proto-Nordic form like 1st sing. *tökö, beside 2nd sing. *tökTR [ t ö k l R ] , etc., would be expected to have undergone /-umlaut at a later stage, in the wake of the phonologization and subsequent morphologization of the /-umlaut. The development would thus have been parallel to that of the present singular indicative of strong verbs, with, e.g., 1st sing, bryt, instead of *brjot (from brjota 'to break'), beside 2nd and 3rd sing, brytr, with phonologically regular /-umlaut. 13 To the extent that the basic problem has been noticed at all (§2), the first view has probably been more widely accepted. But it is by no means obviously correct, and the second explanation indeed seems to deserve serious consideration, although it is of course true that a definitive choice between these two alternatives is unlikely to be possible without new data; unfortunately, no 1st singular preterite subjunctive form is recorded from the ProtoNordic period. NOTES 1,

2.

Or at least in those of the 1st and 3rd classes, e.g. tefya from telja, vek\>a from vaka 'to be awake', while the 2nd-class verbs, e.g. kallapa from kalla 'to call', without /-umlaut, are in any case different. Thus, in Gothic, only five strong verbs (runnjau; qemjau, -nemjau\ qe\)jau, wesjau) and one reduplicating verb (-staistaldjau), as well as ten weak verbs in -dedjau and two preterite-present ones (wissedjau, -kurifyedjau) occur in the 1st singular preterite subjunctive; in addition to the pres. wiljau (see footnote 12), five preterite-present verbs occur in the 1st singular present subjunctive with the same formative (witjau:; -daursjau, kunnjau·, skuljaw, magjau) (Tollenaere and Jones, 1976:287).

318

Hreinn

Benediktsson

3.

It should be mentioned that Cleasby and Vigfusson (1874) listed the 1st sing. pret. subj. 'fengja', 'flygja', 'hengja' (s. vv.fa,fljuga, hanga), without, however, citing any examples to substantiate these forms. It should therefore be noted (see §4.3) that the form 'hjöggja' cited by Jon (wrkelsson (1888-94:256) is from the 1864 edition of Eyrbyggja saga, which was prepared by Guöbrandr Vigfusson. It should also be observed that Munch and Unger (1847:32) cited 'toekja' as a by-form of 'toeka', though without a text reference, while Rask (1818:164) has only forms like 'laega', 'feinga'. 4. The most extensive collection of examples is to be found in Jon £»rkelsson, 1887,1888-94, 1895; see also the current dictionaries, especially Larsson, 1891 and 1956, and Holtsmark, 1955, as well as Cleasby and Vigfusson, 1874 (1957), Fritzner, 1886-96 (1954) and 1972, Gering, 1903,and F. Jonsson, 1931. Finally, there are the collections of the Old Norwegian Dictionary project at the University of Oslo and of the Arnamagnean Old Norse Dictionary project at the University of Copenhagen; unfortunately, an exhaustive examination of these collections has not been possible, but numerous spot checks (of typical verbs) in both collections indicate that they contain at most very few additional examples. 4a. Long after the present article was completed, another occurrence of the 1st sing. pret. subj. bypa (in the compound fyrirbypa from fyrirbjofya 'to forbid') came to light (and is included in the list in §3). The example is in Guömundar saga biskups, the D-edition (BS 1858-78:2.96'), and is present in both of the two principal manuscripts, Sthm Per. Fol. No. 5 (fol. 25va33) and A M 396 4° (fol. 24rl 1), both from the second half of the 14th century. While the chapter in question is not in the earliest edition (A = AM 399 4°, probably from the period 1330-50), the corresponding passage in the B-edition (in AM 657 c 4°, probably from the period 1340-90) has a different, though synonymous, verb in the pres. subj.: banna (fol. 38rl5; BS 1858-78:1.575'), and (according to information from Mr. Stefan Karlsson of the Arnamagnean Institute, Reykjavik) the same verb is used in the C-redaction (which is preserved only in 17th-century transcripts). The youngest redaction (D), usually ascribed to Brother Arngrimr Brandsson, was not compiled until towards the middle of the 14th century, i.e., well after the emergence of the 1st singular ending -i instead of -a. Yet the author, choosing a verb and a grammatical form of his own, different from that of the preceding redactions, retains the old ending -a, which accodingly, had most likely become an archaism by this time. In other words, the hypothetical conclusion mentioned above would not necessarily be affected by this additional example. 5.

6.

7. 8.

This form, spelled 'kiera', occurs in Gudbrandsbiblia (1584); see Bandle, 1956:397. For the retention, at least in the written language, of the ending -a alongside -i down to the 18th century, see Bandle, 1956:383-4. The form 'saeia' in Guöbrandsbiblia (Bandle, 1956:384), apparently with a -j-, is of course of no significance in this connection, in that it postdates both the conditioned loss of intervocalic gj and j and the diphthongization of ae > [ai], the result of which was that such Old Icelandic forms as sga, hleja, and lggja 'to lower' began to rhyme (in [-aia]). I am indebted to Mr. Forrest Scott, who is preparing an edition of Eyrbyggja saga, for additional comments on the manuscript relationship. Contrary to what at first glance might be taken for granted, the form 'hvoGva' for höggva is probably not a misspelling. There are five otherexamples of this oranequivalentspelling(viz. 'hu-')on record: they occur only in the infinitiveand the neuterofthe past participle,andallsix date from the late 14th or early 15th century (assuming that the present example was derived from *V). There can, in other words, be no doubt that these examples represent a genuine by-form inf. hvöggva, neut. past part, hvöggviö (beside höggva, höggvfö), whatever the explanation of this form may be. There is also one example of the inf. 'hueggua' from the late 15th century (DI VI No. 492 from 1485); this is probably nothing but a hypercorrect form for hvöggva, based on the alternation between ve and vo, e.g. in kveldand fcvoWevening', which had arisen by this time.

1st Sing. Pret. Subj. in 9.

10. 11. 12.

13.

Germanic

319

All three copies (written by Äsgeir Jonsson) in fact have the usual Icelandic 'ae' in this form, though the editor, in his critical apparatus, represents this by the ligature 'oe', which may of course have been in the Norwegian original. Even the late and secondary hlypa (for earlier *Μόρα) from hlaupa 'to run', occurring in Guöbrandsbiblia(Bandle, 1956:407), agrees with this. Kieckers( 1928:197)ascribedthelossof*-y-eveninthepret. subj. ft>w/a'bound'toanalogy;but thisis of course an unnecessary assumption. The ON 1st sing. pres. vilja 'will, wish', which occursafewtimesasamain-clauseverb(withthe function, apparently, of an indicative form) in early poetry (beside the usual viT), has been considered to be the historical equivalent of Goth, wiljau and thus to be a remnant of the athematic subjunctive paradigmofthisverb which isstillpreservedinitsentiretyinGothic(see e.g.Noreen, 1923:361). However, under thesecondhypothesis, thisis virtuallyinconceivable, since the expected Proto-Nordic form would be *wilö, not *wiljö. But under the first hypothesis, the fact that the *-_/- was not lost by analogy, as in bfya, teka, skyla, could easily be explained by the relative isolation of the paradigm of vilja. Yet,inanycase,it seemsmorelikely that the few occurrences of lstsing. v/#asimplyrepresenttheregularpresentsubjunctiveofthe ya-stem of this verb (inf. vilja, etc.); see e.g. F. Jonsson, 1931:616. In the generative phonological analysis of Old Icelandic the preterite subjunctive is bound to play a central role in determining the form and nature of the /-umlaut rule synchronically operative at the Old Icelandic stage; as is well known, the main question is whether this is tobe regarded as a strictly morphological rule or as having remained a phonological one. A form like the 1st sing. pret. subj. teka, in particular, without -j- before the -a, strongly suggests, indeed presupposes, a morphological rule: if, as described above, theoriginalform was *tökjö ( a n d the *-_/-was lost by analogy,though theformretaineditsj'-umlaut.aprerequisite forsuchananalogicalremodelingisthemorphologizationofthei-umlautrule;similarly,ifthe original form was * tökö(l i az u tjax ('Maybe there would be a time for me, too, to go to their place'). These are not hypothetical periods, and they have a minimum of apparatus - really only the ubiquitous conjunction da. Otidex is a pv. Ipf. and is by itself untranslatable into any language I know, the other form is Plup., but occurs here obviously in a modal, not a temporal function. All North Slavic languages would simply use an infinitive here, or a conditional with by\ S.-Cr. is somewhat apart. Admittedly, such formations are infrequent in Big., but are perfectly alive and immediately understandable without having to be taught at school. The Ipf. leaves the outcome of the deliberation in limbo, as it were, and Aor. cannot be thus used. It is in my opinion not at all necessary to derive the so-called 'iterativehabitual' function of the future in the past, for which an ordinary Ipf. can likewise be employed, from a full-fledged conditional one; the two probably developed alongside each other. In an example like Kovacevski (1976:39) Taa bese patriarxalno vospitana, so onoj so kogo ce se zasakase se sakase za brak, za deca (Mac.) ('She had been raised in a patriarchal manner, she would love him with whom she would fall in love, for marriage, for children'), the forms used are halfway between a conditional and an iterative meaning, about equidistant from both. Basically, the temporal meaning still shines through, as an event in the future envisaged from a standpoint in the past; the habitual meaning is not far to seek, because a number of repeated events can be assumed before a successful upshot may result. Iterative habitual and conditional (not unreal) meaning have an equal share in Kovacevski (1976:102): Ako napolnev voda od cesmata ... vujce ke ja poznaese vodata, ke gi isturese gumovite ('If I filled the vessels from the well, uncle would recognize the water and overturn them') (note that English would can occur in a similar function). The iterative function of Ipf. surely goes back to Indo-European and has been illustrated above from Greek. It is doubled by the aspect, however, in

366

Herbert Galton

the Southeast of the Slavic territory (not in Greek, although it does possess verbal aspect). Thus, in the perfective aspect, the imperfect is especially apt to bring out the pointlike emergence of an event along a series of repeated events in the past, without the future particle Re (Mac.) if a conjunction is employed, as in Fotev (1974:218) Vo manastirot Sveta Petka, vo oltarot na edna crkva imase edno izvorce, cija voda lekuvase sekakvi bolesti: ako se napiese covek od taa voda ('In the monastery of S.P., at the altar of a church, there was a spring whose water cured all sorts of ailments: if a person would (!) drink of that water'). The transition to a hypothetical clause, arrested, one might say, by Fotev's colon, could not be much clearer. And if there is no direct syntactic dependence, ke serves to provide the stage for pv. Ipf., resulting in the formation under discussion here, as in Janevski et al. (1962:94): Sedev vnatre cetiri dena i cetiri nok/i. Po malku ke izlezev koga ne pukaa ( Ί sat (i.e., in a beech tree) for four days and four nights. After a while I would go out when they did not shoot (i.e., repeatedly)'). The I.-E. Ipf. has undergone rather a sea-change here, but still denotes repetition, as among other things it always did, in a somewhat modified manner indicating, by means of the future particle, the repeated emergence of an event from the vantage point of a continuing past. The aspect is, of course, a Slavic innovation. It is very doubtful that the addition of the particles ke, ste to help in the constitution of such a complex meaning properly qualifies the conjugational system as developing along the lines of increasing analytism, as the Bulgarians, following Russian precedent in the treatment of their own language, seem to think (cf. Gennadieva Mutafcieva, 1970:209). Surely a sentence like Da bese pone po-xubava, stjax da ja obikna ('If she had been at least more beautiful, I should have fallen in love with her') can hardly be compressed any further. The conjunction da which introduces the whole hypothetical period is nothing like specialized in this function; beginning its career as an exhortative particle in Slavic (it accompanies the imperative in the Kiev Missal and still does so in Slovene), it has spread out of bounds as compared with its usage in O.C.S. (cf. Mircev, 1963:243). Finally, let me mention a curious parallel arriving at the same result (i.e., a future in the past with a predominantly hypothetical function) from the opposite end, and from the other end of the I.-E. linguistic territory. Whereas the Big. type stjax da kaza ( Ί was going to say — would have said') proceeds from Ipf. of the modal verb that came to be appropriated for the morphological expression of the future and is based on a future in the past (imperfect), it is equally possible to reach the same temporal level by proceeding from a future tense in the first place and dropping this down, as it were, to the imperfect. In such a highly complex procedure, no split into separate indices of the future and ' the imperfect is needed, as confronts us most clearly in Big. ste xodex, Mac. ke odev. This piece of ingenuity was left for the (East) Indians to develop. Proceeding from the future stem marked by the (apparently I. E.)

Unreal Hypothetical Periods in South Slavic and Greek

367

suffix *-sy~, precisely the endings of the imperfect tense can be added to yield a conditional, typified by dbharisyat ('he was going to carry off—would have carried o f f ) . This formation was almost unknown in the Vedic Samhitäs, but is much more frequent in the ancient Indian epics and dramas (Macdonell, 1958:178; 1927:129), so that it is fair to say that the Indians hit independently upon this idea. Its inflection is that of an imperfect of the first conjugation, as has already been pointed out by Whitney (1913:200,334). The reason for this different approach with practically the same result was probably the greater vitality of a separate future tense as against that of an Ipf. with a clearly defined function in Sanskrit, as compared with Late Latin or older Slavic for that matter. The meaning of an irrealis is fully developed in the Upanisads, for which I will quote just one single example from Chändogya Upanisad V, 3, 5 (Radhakrishnan, 1953:427): Yady aham imän avedisyam katham tenavaksyam iti ('If I had known them, how should I not have told them to you?'). It might be mentioned, in conclusion, that in formations like those dealt with in this study, any simple binary approach proceeding from some past/ present (or 'non-past') dichotomy breaks down; in the Sanskrit formation it was a very definite future tense which sprouted a past, more particularly an imperfect. Mutatis mutandis, no binary system, let alone one on privative marked/unmarked lines, holds out any hope of much advancing our understanding of either the Old Church Slavic or the modern Big. and Mac. verbal systems with their plethora of tenses. In Sanskrit as well as in modern East South Slavic, the future tense is very much on a par with the other tenses, albeit with very different morphological markings, and no amount of philosophical speculation on the uncertainty of the future can belittle this fact. If one must philosophize, then one may admire the ingenuity of the ancient Indians who were able to visualize the past from the standpoint of the future, which requires a greater mental effort than the other way round, where the future in the past has in all probability by now become the past, or else impossible, unreal. SOURCES OF LITERARY Q U O T A T I O N S

Arsovski, T. (1969). Cekor do esenta. Skopje: Misla. Cornford, F.M. (1957). Plato's Cosmology. New York: Liberal Arts Press. Fotev, M. (1974). Potomcite na kat. Skopje: Kultura. Janevski, S. et al. (eds.) (1962). Temni kazuvanja. Skopje: Koco Racin. Kovacevski, Z. (1976). Semejni fre ski. Skopje: Kultura. Njegos, P.P. (1953). Gorski Vijenac. Belgrade: Cjelokupna Djela. Radhakrishnan, S. (1953). The Principal Upanisads. London: Allen and Unwin.

368

Herbert

Galton

REFERENCES Andrejein, L. (1978). Osnovna bilgarska gramatika. Posthumous edition. Sofia: Nauka i izkustvo. Bräuer, H. (1957). Untersuchungen zum Konjunktiv im Altkirchenslavischen und Altrussischen (= Veröffentlichungen des Slavischen Seminars, 11). Wiesbaden: Otto Harrassowitz. Camara, J.M. (1964). Une categorie verbale: le futur du passe. In: H. Lunt (ed.) Proceedings of the Ninth International Congress of Linguists. The Hague: Mouton, pp. 547-551. Delbrück, Β. (1897). Vergleichende Syntax der indogermanischen Sprachen, II. Teil. Strassburg: Trübner. Gebauer, J. (1929). Historickä mluvnice jazyka ceskeho, Dil IV: Skladba. F. Trävnicek (ed.) Prague: Ceskä akademie ved a umeni. Gennadieva-Mutafcieva, Z. (1970). Podlinitelnijat s~bjuz DA ν sbvremennija bblgarski ezik. Sofia: Bblgarskaakademija na naukite. Gotgb, Z. (1964). Conditionalis typu baikanskiego w jezykach pohidniowoslowianskich ze szczegölnym uwzglqdnieniem macedonskiego. Wroclaw-Kraköw-Warszawa: Polska akademia nauk. Koneski, B. (1967). Gramatika na makedonskiot literaturen jazik, Del I i II. Skopje: Kultura. Kühner-Gert, R. (1904). Ausführliche Grammatik der griechischen Sprache, II. Teil: Satzlehre. 3rd ed. Hannover: Hahn. Lambertz, M. (1914/15). Die hypothetische Periode im Albanischen. Indogermanische Forschungen, 34, 44-208. Macdonell, A. (1927). Α Sanskrit Grammar for Students. London: Oxford University Press. Macdonell, A. (1958). A Vedic Grammar for Students. Reprint. London: Oxford University Press. Minova-Gurkova, L. (1967). Svrznikot ako vo makedonskiot jazik. Makedonski jazik, 18, 111-140. Mirambel, A. (1959). La langue grecque moderne. Paris: C. Klincksieck. Mir£ev, K. (1963). Istoriöeska gramatika nabblgarskija ezik. 2nd ed. Sofia: Nauka i izkustvo. Panzer, B. (1967). Der slavische Konditional (= Forum slavicum, 14) Munich: Wilhelm Fink. Potebnja, A.A. (1958). Iz zapisokpo russkojgrammatike, Tom I—II. Moscow: Gosudarstvennoe ucebno-pedagogiceSkoe izdatel'stvo. Stankov, V. (1967). Modalna upotreba na glagolnite vremena ν si>vremennija bblgarski knizoven ezik. Izvestija na Institute za bblgarski ezik, 15,346. Stevanovic, M. (1969). Savremeni srpskohrvatski jezik, II: Sintaksa. Belgrade: Naucna knjiga. Trost, K. (1972). Perfekt und Konditional im Altkirchenslavischen. Wiesbaden: Otto Harrassowitz. Veenker, W. (1967). Die Frage des finnougrischen Substrats in der russischen Sprache (= Uralic and Altaic Series, 82). Bloomington: Indiana University. Whitney, W.D. (1913). Α Sanskrit Grammar. 4th ed. Leipzig-Boston.

The Minimal Pair in the Finnish Spelling Debate of the Late 18th and Early 19th Centuries, with a Glance at the First Grammatical Treatise Ulrich Groenke, Cologne University, Cologne, West

Germany

The reading of Finnish texts of older provenience - say, of the Agricola era, the 17th and 18th centuries - is hampered by an orthography that betrays the writer's ( a n d / o r printer's) lack of insight into the system of the language on the phonemic and morphophonemic planes. It bespeaks the writer's more or less futile struggle with the multitude of phones that must be identified with a lesser number of superordinate phonic units in order to be reduced to writing with the inherited Latin alphabet. The problem of establishing a phonemicgraphemic equation, reduced to a highly efficient, unambiguous, economical orthography (by which modern Finnish script is distinguished) was one that they could not solve satisfactorily. Doubtless, writing traditions taken over uncritically f r o m Swedish, Latin and German blinded the eye and deafened the ear of those early writers. It would be tempting - though futile - to ponder what early Finnish script might have looked like if its creation had been promoted not by the Reformation, but by some missionary activity of the Eastern Church, necessitating an adaptation of Church Slavonic Cyrillic, or perhaps its transformation into a Latin-letter script, resulting in an original Finnish azbuka or alphabet. We would do better to abandon such speculative ideas, however, and make peace with the fact that older orthographies, and even modern orthographies bound up in centuries-old traditions, demonstrate the idiosyncrasies of writing systems - systemoids, I'd rather say - of the era before linguistic science. Indeed, instead of negative criticism we might show how well the old classics did after all, considering the shortcomings in linguistic analysis of their time, and pay our tribute to the pioneering achievement of Mikael Agricola. His teacher Martin Luther was certainly making too harsh a judgement of his disciple's orthographic attempts when he flung the inkwell at him. Yet, almost eight centuries before modern linguistics, four centuries before Agricola, an Icelandic scholar demonstrated how by c o m m o n sense and rather trivial linguistic operations one could arrive at a workable phonemicgraphemic analysis. The work of this anonymous Icelandic scholar has gone down in the history of science as the First Grammatical Treatise (FGT). 1 The

370

Ulrich Groenke

linguist will remember that the author, the First Grammarian (FG), applies the criterion of the commutation test with minimal pairs, that is, he shows that the substitution of phonic segments in constant phonic surroundings on the level of expression produces symmetrical changes on the level of content. Going even further than the modern linguist, he incorporates his minimal pairs into meaningful sentences such as: sar veitti madr mir eitt, sgr mgrg veitta ek honum 'one wound (sär) a man inflicted on me, many wounds(i ) I inflicted on him'; siir eru augu syr, slik duga betr en spryngiyr 'sour (siir) are the sow's (syr) eyes, such are better than if they popped out'. Establishing an inventory of 9 basic vowels for 12th century Icelandic, the FG proceeds by the same method to establish the oral-nasal correlation, and finally, with 18 minimal pairs and exemplifying sentences, the quantity correlation. This, now, is of particular interest to us. The F G - very fortunately for the Nordist - observed Icelandic before the quantity revolution and the radical restructuring of the entire vowel system. At the time of the FG, vowel quantity was distinctive in Icelandic, and, like in Finnish, long vowels could precede long - or geminated - consonants, as demonstrated by minimal pairs like minna vs. minna or skjot vs. skjott in the FGT. Now, as is well known, vowel quantity bears a maximal functional load in the construction of linguistic signs in Finnish. Modern textbook authors have no difficulty in presenting any number of minimal pairs, and, if they resort to word forms involving two vowels and a consonant, they can organize whole networks of interdependent minimal pairs. 2 Actually, the minimal pair should have been invented by the Finns, and it is amazing that Agricola and his followers should have failed to see the value of this testing device for the maker of orthographic norms. The fame of having realized the demonstrative value of the minimal pair, though, goes to our FG. Interestingly enough, it was as late as 1942 that linguists discovered they had a forerunner to their pit-pet-patput-pot gimmick in Iceland in the 12th century (Bergsveinsson, 1942). The first to subject Finnish orthography to linguistic insights, applying minimal pairs to underpin his argumentation, was Bishop Daniel Juslenius (1676-1752), known to the Fennist as the creator in 1745 of the first workable and sizeable dictionary of the Finnish language (Juslenius, 1968). Up to Juslenius' time, and far into the 19th century, the graphic representation of long vowels and consonants by double letters was practiced irregularly and inconsistently. Writing habits markedly improved at times, only to worsen again. Agricola's Schriftbild already shows fluctuations in quality; thus, for example, his Almanac of 1544 'looks better' than his Preface to the New Testament of 1548. In the Praefatio to his Finnish Lexicon of 1745 Juslenius draws attention to the importance of the quantity correlation in Finnish, and emphasizes the necessity of marking quantity in script. His illustrative examples are minimal pairs, or, rather, minimal triplets - double pairs of the type A vs. Β vs. C:

The Minimal Pair in the Finnish Spelling Debate tuli puki

'ignis' vs. 'induit' vs.

tuuli pucki

'ventus' 'caper'

vs. (vs.

371 tulli puhki

'telonium' 'per')

He does not fail, either, to pinpoint another spelling weakness of early Finnish writers, i.e., the distinguishing of the mid and low front vowels in script. His demonstrative minimal pair is: että = 'ut, qvo, qvia, qvoniam' vs. ette = 'non vos' Juslenius' argumentation and presentation does not have the technical stringency of the medieval Icelandic genius, to be sure, but at least, like the FG, he underlines the importance of his remarks with strong rhetoric: 'quae nisi observaveris, toto coelo errabis, futurus loquens audienti barbarus, et loquentem barbarum habiturus' (1968: Praefatio par. 10). This is reminiscent of one of the FG's polemic outbursts: 'But, if anyone is so insistent or obstinate as to contradict so many sensible men who have said that they themselves pronounce this word as it is here written . . . , then I shall apply Cato's favorite advice, which he gave his son in these verses: Contra verbosos noli contendere verbis; sermo datur cunctis, animi sapientia paucis' (Haugen, 1972: 22). Juslenius' pioneering introduction of the minimal pair seems to have impressed, if not Finnish writers, at least Finnish grammarians, though somewhat belatedly in the 19th century. Carl Hellenius (1838) uses pairs like häätä vs. hätä, kuulunut vs. kulunut, and vaara vs. vara in his argumentation; Kustaa Renvall (1840) presents Juslenius' old standby tuuli vs. tuli and demonstrates the distinctive grammatical function of quantity, which he finds disregarded in the latest Bible redaction of 1776, with pairs like taloon vs. talon and tulee vs. tule. As a matter of fact, his last two examples demonstrate cases of fusion or innere Beugung in Finnish and show that the graphic visualization of the quantity opposition is indispensible in the grammatical decoding of a written Finnish text. We may recall here the FG's (quality) pair quoted above, sdr (sg.) vs. sor (pi.). Renvall demands that in all instances, whether in lexemes or in grammatical morphemes, vowel quantity must be strictly observed in writing, using single letters for short and double letters for long units. Criticizing the 1776 Bible redaction, he denounces the inconsistent and irregular marking of vowel quantity as a form of 'carelessness (värdslöshet) that creates confusion and quibbling in recitation' (Rapola, 1965:75).3 Hellenius (as cited in Rapola, 1965) takes a half-hearted attitude to the problem. He demands the spellings aan, ään, een, iin, oon, yyn, öön in word forms in the illative case, and points out (with the minimal pairs quoted above) instances of possible confusion

372

Ulrich Groenke

that should be avoided by using single vs. double letters, but in general he relegates the whole matter to the intuition of the native speaker, 'since the Finn knows anyway when a vowel is long or short in his language. He who does not know should learn it through grammar or language usage, just as in any other language' (Rapola, 1965:75-76).4 This calls to mind the argument of the FG's imagined heckler - a rhetorical device of our Icelander - who claims, Ί can read the Danish language perfectly well, even though it be written with the regular Latin letters. I can make out how to pronounce it . . . I don't care whether you write your g or a, I or e,yor if (Haugen, 1972:15). Of course, the FG then refutes the argument: To that I say, 'It is not the virtue of the letters that enables you to read and to make out the pronunciation where the letters are unclear. That is, rather, your virtue, and it is not to be expected that I, too, or anyone like me, if such there be, shall be able to read well and to make out which path to take where more than one course is possible because it is written one way but not clearly determined, and one then has to guess, as you claim you can do so well. But even if everyone could make something out of it, it is practically certain that not everyone will arrive at the same result when the meaning is changed in this way, particularly in the laws' (Haugen, 1972:15). As early as 1818 it was suggested that long vowels be represented by single letters with diacritic marks instead of writing double letters, e.g., ό instead of oo, δ instead of öö, etc., as was and is practiced in Hungarian orthography. The Hungarian model was suggested by Rasmus Rask, who discussed Finnish orthography with Renvall and Jaakko Juteini in Turku. The latter agreed with Rask, uncovering an orthographic problem that could be solved easily ä la hongroise: 'Writing the long vowels undoubled, and with only a prolongation mark over the same, would be - quite apart from the brevity of writing - of great importance for pronunciation, in particular for those who are not versed in the basic words and the words' basic syllables, and who could easily assume two separate vowels to be one long vowel, such as in kalaaika or koota, which otherwise ought to be hyphenated, which I do not favor' (Rapola, 1965:74-75).5 As a matter of fact, modern Finnish orthography, which adheres to writing long vowels with two letters, is hampered in this respect and, indeed, occasionally has to resort to hyphenating, e.g., ala-arvoinen. Juteini's casual remark that using the Hungarian model would also make writing shorter is, of course, more amusing than practical. Still, in word forms like pää, kuu and the like one could save one third of the writing space, word forms like myydään could be written with five instead of seven letters, and publishers could save somewhat less than ten per cent of the paper used if Rask and Juteini had had their way. In a paper-producing country like

The Minimal Pair in the Finnish Spelling Debate

373

Finland this sort of economizing is likely to carry little weight. To our FG, though, a man of the quill and parchment era, devices for abbreviating were, it seems, of real economic importance: in his treatise he suggests writing geminated consonants with small capital letters, for which he introduces linguistically relevant spelling names, e.g., Ρ = 'epp' as opposed top = 'pe'. It is, obviously, not without pride that he comments on his small capital letter device with these well chosen words: Ί let this one letter represent as much as if two of the same kind were written, so that the writing may be less and quicker and the parchment last the longer' (Haugen, 1972:25). NOTES 1.

2.

The chief works on the F G T are Benediktsson (1972), Haugen (1972) and Leoni (1975). Quotations in English are taken from Benediktsson and Haugen, slightly altered by the author. For example, the following set is taken from Karlsson (1976). Tule tänne come here Ulkona ei tuule the wind does not blow Ulkona ei tuulle the wind is unlikely to blow Ulkona tuulee the wind is blowing Pekka tulee Pekka is coming Pekka ei tulle Pekka is unlikely to come Pekka tullee Pekka is likely to come Ulkona tuullee the wind is likely to blow This could be presented as a network of oppositions: r—t u 1

3. 4. 5.

e

Slightly abbreviated and translated by the author. Slightly abbreviated and translated by the author. Slightly abbreviated and translated by the author.

REFERENCES Benediktsson, H. (1972). The First Grammatical Treatise: Introduction, Text, Notes, Translation, Vocabulary, Facsimiles. (= University of Iceland Publications in Linguistics, \.) Reykjavik: Institute of Nordic Linguistics. Bergsveinsson, S. (1942). Wie alt ist die phonologische Opposition in sprachwissenschaftlicher Anwendung? Archiv für vergleichende Phonetik, 6, 59-64. Haugen, E. (1972). First Grammatical Treatise: The Earliest Germanic Phonology. An Edition, Translation and Commentary. 2nd, revised edition. London: Longman.

374

Ulrich Groenke

Hellenius, C. (1838). [Citations taken from Rapola (1965).] Juslenius, D. (1968). Suomalaisen Sana - Lugun Coetus. Näköispainos. (Facsimile). (= Suomalaisen Kirjallisuuden Seuran Toimituksia, 288.) Helsinki: Suomalaisen Kirjallisuuden Seura. Karlsson, F. (1976). Finskans struktur. Lund: Liber Läromedel. Leoni, F.A. (1975). II primo trattato grammaticale islandese. 5 Studi linguistici e semiologici. Bologna: Societä editrice II Mulino. Rapola, M. (1965). Suomen kirjakielen historia. I. (Reprint of 1933). (= Suomalaisen Kirjallisuuden Seuran Toimituksia, 197.) Helsinki: Suomalaisen Kirjallisuuden Seura. Renvall, K. (1840). [Citations taken from Rapola (1965).]

The Morphophonemics of Latvian Declension Morris Halle, Research Laboratory of Electronics, Massachusetts of Technology, Cambridge, MA., U.S.A.

Institute

Latvian declension presents an instructive example of the need to assume underlying representations that in some cases are strikingly different from their surface realizations. In Latvian the interaction of rules of elision with a rule of metathesis conspires to render the relationship between surface and underlying representations particularly obscure. Since each of the rules is independently motivated, the discrepancies between surface and underlying representations present no serious problem to the language learner. The facts below are offered as a contribution to the ongoing debate concerning the abstractness of underlying representations. As shown in Tables (1) and (3), Latvian noun declension distinguishes five cases, two numbers, and two genders. Masculine Gender

(1)

Sg Ν G D A L PI

Ν G D A L

(a) gulb-i-s 'swan' -j-a -i-m -i -i

(b) zaen - -s - -a -a -m -u -a

(c) tirg-u-s 'market' -u-s -u-m -u -u

gulb-j-i -j-u -j-iaem -j-us -j-uos

zaen- -i - -u - -iaem - -us - -uos

tirg-

-i -u -iaem -us -uos

Each form is composed of a stem followed by a thematic vowel to which may be added a case ending. We see this most directly in the forms of the i-stem noun (cf. gulbis 'swan'). The thematic vowel / i / becomes / ) / before case endings beginning with a vowel, and it is lengthened in the loc. sg. The composition of the forms is less obvious in the two other masculine paradigms - where the thematic vowels are / a / and / u / , respectively - because the thematic vowel is deleted before case endings beginning with a vowel, and,

376

Morris Halle

therefore, does not surface in many instances, e.g., in any of the plural forms. To account for the facts just noted, we postulate rule (2) which desyllabifies / i / and deletes all, other vowels in prevocalic position. (2)

/

[-syl]

Γ +highl L -back]

[+syl]

[+syl] /

[-

-]

In addition to rule (2) we require a word formation rule that deletes certain theme vowels before the case ending / s / : / a / in the masc. sg. and / i / in the fem. sg. (see below, (3)). (We shall not state this rule formally.) Before accounting for the remaining forms in (1) - i.e., the acc. and loc. singular - we give the feminine paradigms illustrated in (3): Feminine Gender

(3)

Sg Ν G D A L PI

(a) ziv- -s 'fish' - -s -i-j -i -i

N / A ziv-i-s G -j-u D -7-m L -is

(b) mäs-a 'sister' -a-s -a-j -u -a

(c) zem-e 'earth' -e-s -e-j -i -e

mas-a-s - -u -a-m -a-s

zem-e-s -j-u -e-m -e-s

We note at once that just as in the masculine nouns the thematic vowel is long in the loc. sg., and it is long also in the dat. pi. and loc. pi. We capture this fact with a second word formation rule: (4a)

V — [+long]

/

stem + —

in

loc. sg. loc. pi. fem. dat. pi. fem.

Finally, we observe that in the acc. sg. the thematic vowel is high (and round if back). This fact is captured by a third word formation rule: (4b)

+high I around J

7

aback

m

acc. sg. fem (gen.pl.

Morphophonemics of Latvian Declension

377

If we assume that the word formation rules precede all other phonological rules, and in particular rule (2), then we can account for the gen. pi. forms of the feminine declension by the simple expedient of letting rule (4b) apply to these forms as well. As shown in (5a) the declension of indefinite adjectives is identical to that of the -a-stem nouns (cf. (lb) and (3b)). (5a)

Masculine

Feminine

Ν G D A L

lab- -s 'good' - -a -a-m -u -a

lab-a -a-s -a-j -u -a

Ν G D A L

lab-

lab-a-s - -u -a-m -a-s -ä-s

-i -u -iaem -us -uos

The definite forms given in (5b) are considerably more complex. (5b)

Masculine

Feminine

Sg

Ν G D A L

lab-aj- -s -a -aj-a-m -u -aj-ä

lab-a -äs -aj-a-j -u -aj-ä

PI

η G D A L

lab-iae

lab-as

-U3

-uo

-aj-iaem -uos -aj-uos

-aj-a-m -as -aj-a-s

We observe at once that some forms surface with the suffix -aj- whereas others do not have this suffix. The forms with the suffix -aj- are otherwise completely straightforward: the rules developed above readily account for all of their peculiarities. We shall assume therefore that all definite forms are generated with the special suffix -aj- to which are then added the thematic

378

Morris Halle

vowel / a / and the regular case endings, all of which are subject to the three rules of word formation and the phonological rule (2) given above. This will account for the nine forms in (5b) where the suffix -aj- appears intact on the surface. In the remaining eleven forms the underlying / j / of the suffix does not appear on the surface and must be assumed to have been deleted. We note that in these eleven forms (but not in the nine above) the suffix -aj- is followed either by a word-final short vowel or by a short vowel plus a word-final obstruent. It is in this environment, then, that a rule of j-deletion (6) is postulated. (6)

j —-«Γ/

V ([-son]) #

We order j-deletion after rule (2) because the / a / of the definite suffix is not deleted, as shown, e.g., by the fact that in the gen. sg. forms the ending surfaces with a long rather than a short vowel.1 The effects of (2) and (6) on the underlying representation are shown in (7): Feminine

Masculine

(7) Sg

G A

lab aj-ii-a lab-a/-u

PI

Ν G A

lab-a/-jl-i lab-a^-ji-u lab-a/-ji-us

Ν G A

lab-a/-a lab-a/-a-s lab-aj-u

N/A G

lab-a/-a-s lab-a^-u

To obtain the correct output for the rest of the forms in (7) we must add a rule of metathesis (8a) followed by a rule of round/back assimilation (8b). (8a)

(8b)

in certain environments

V! V2 - V2 V,

[

aback 1 around]

/

+high aback around

As pointed out by Halle and Zeps (1966), rules (8a) and (8b) are independently motivated. They are needed to account for the quite regular alternations between / e j / and /iae/ in verbs such as / s e j u / Ί bind' /siaet/ 'to bind'; or / l e j u / Ί pour' /liaet/ 'to pour', as well as for the otherwise quite isolated alternation / d e w - u / 4I gave' / d u o t / 'to give'. In verbs metathesis occurs before endings beginning with a consonant, whereas in definite adjectives metathesis takes place in word final position.

379

Morphophonemics of Latvian Declension

Finally, if we make the natural assumption that a sequence of identical vowels surfaces as a single long vowel, nothing further needs to be said about the gen. sg. masc. or about the nom. sg., gen. sg. and nom./acc. pi. fern, forms. To conclude we list below the rules that have been developed in this discussion in the order in which they must be applied: (9)

Word formation rules a)

Raise thematic vowel in acc. sg., etc. (4b)

b)

Lengthen thematic vowel in loc. sg., etc. (4a)

c)

Delete thematic vowel

d) e) f) g)

i -* j and V — 0 before V (2) j — 0 (6) metathesis (8a) assimilation (8b)

before / s / , etc.

NOTE 1. It was observed by Zeps (1969) that in the meter of Latvian folk songs it is necessary to distinguish a special class of syllables that may occupy two adjacent positions in the metrical pattern. The case endings before which j-deletion does not take place are among the syllables that may count metrically as equivalent to two syllables.

REFERENCES Halle, M. and Zeps, V.J. (1966). Survey of Latvian Morphophonemics. MIT RLE Quarterly Progress Report, 83, 105-113. Zeps, V.J. (1969). The Meter of the Latvian Folk Dactyl. Celi-Rakstu Kräjums Lund: Ramave.

What Helmholtz Knew about Neutral Vowels Robert T. Harms, University of Texas, Austin, TX., U.S.A.

0. INTRODUCTION

The essence of vowel harmony is a constraint on the vowels within a larger domain, grammatical or prosodic, requiring that all the vowels agree in a given phonological feature (e.g., backness, rounding, tongue root advancement). For example, in Diola Fogny all vowels in a word must agree in tongue-root position (Sapir, 1965, where this feature is called 'tense'). That very few languages in fact exhibit this ideal harmony type has remained a source of challenge to phonology, both in the descriptive treatment of specific languages and in the revision of formal theory to account for the problems which arise. Three basic difficulties may be noted. First, there is the problem of ad-hoc domains, i.e., where the domain of harmony does not correspond to any prosodic or grammatical structure of the language. For example, in Turkish, rounding agrees over the domain: any vowel followed by subsequent high vowels (ignoring intervening consonants). Or in Votic, with backness harmony, ο is opaque and requires subsequent vowels to be back, e.g., maenko-za 'in the game' (cf. maenciizae 'in playing' < maenki-A-zA, slciiza 'in the straws' < slkE-i-zA). Second, there is the problem of ad-hoc features, i.e., where the harmonizing vowel classes are not naturally characterized by established feature parameters. For example, the dominant/recessive harmony alternations of Nez Perce (ae > a, u > o, but i is either dominant or recessive), viewed by Kiparsky (1968) as 'attraction toward a' of the dominant set, allow a rule which reflects that generalization only at the cost of treating dominant /'s as underlying abstract a, i.e., [-l-backj-high]. 1 Finally, there is the problem of neutral vowels, i.e., where not all vowels are subject to the harmony constraints. For example, in the tongue-root-advancement harmony systems of West Africa, the low vowel a is frequently neutral (Ladefoged, 1968:36-38). Similarly, among the many tonality harmony systems of Eurasia, it is not uncommon for i to be neutral. Although one can readily imagine both articulatory and acoustic reasons why a low vowel might not participate in a tongue-root-advancement opposition system, the neutrality of the front vowel with highest tonality within an otherwise productive front/back harmony remains without an obvious explanation under current theories.

382

Robert Τ. Harms

It is this last problem, the neutral vowels within tonality harmony systems, for which I hope to provide a new perspective in this paper. I shall first examine the vowel systems of three Uralic languages with productive tonality harmony - Votic, Finnish and Hungarian. In an attempt to shed light on the phonological role of the neutral vowels in these languages I shall propose a new set of phonological vowel features based upon functional interpretation of vowel spectra. Finally, I shall suggest that the solution to the problem of perceptual vowel normalization lies in the geometry of the energies of the lower harmonics rather than in a calculation based upon the frequencies of the fourth and higher formants. 1. VOTIC (BASED ON ARISTE, 1948)

The harmony status of the Votic vowels is given in (1). (1)

Harmonic: Front ü e ö ae

Neutral: i

Back 4- u s ο a

Alternating pairs: ae

a : joomma 'we drink' - veemmae 'we take' jsimma 'we drank' - veimmae 'we took' e a : juvvsza 'in drinking' - vidydyezae 'in taking' naizalaa 'to the woman' - mehelee 'to the man' ü ® u : joonnu 'drunk' - veennü 'taken' All vowels occur both short (ρ alto 'field') and long (vaaras 'guest'). Although 4 is lexically limited to Russian loans, these appear to be well assimilated to Votic phonological and morphological patterns, e.g., kit-rüba 'whale' (cf. Rus. kit 'whale', ryba 'fish', although Russian has no such compound), rinkoilla 'at the market' ( < r4nko-i-llA\ cf. Rus. rynok 'market'). Neutral i occurs freely with both harmony sets: sika 'pig': ikae 'each', talvi 'winter': maeci 'hill', issuizimma 'we would sit': ciisiiizimmae 'we would ask'. A stem with all z's takes front suffix variants: pilli ssae 'from the musical instrument', irt-tae 'log (part.)'. (Note, however, mi-lta 'from me', a suppletive contraction from minulta.) ö occurs beyond the initial syllable only in limited forms (such as Finnish loans like tüttö 'girl'), leaving ο essentially outside the set of alternating vowels, other than a few cases such as laehtö 'departure' ( < laehte'leave') alongside the regular maenko 'game' ( < maenki- 'play'), ο always

What Helmholtz Knew about Neutral Vowels

383

requires back harmony of subsequent syllables; thus, leipoita 'loaves of bread (part.)' < leipae-i-tA; cf. leipaeae 'bread (part.)' < leipae-tA, ssbrsita 'friends (part.)' < ssbrai-tA, psltoiza-pzltaiza 'in the fields' < p s l t o - i - t A . Apart from the neutral vowel i, the words with ο in non-initial syllables and a small list of affixes which are always back (e.g., -kaa '2 pi. imperative': söökaa 'eat'), the normal domain of harmony is the word. 2. FINNISH

The vowels of Finnish are given in (2). (2)

Harmonic and alternating: Front Back y (=ü) u ο ο ä (=ae) a

Neutral: i, e

Vowels occur both short {tuli 'fire') and long (tuuli 'wind'). Except for the two neutral vowels, words are restricted to vowels agreeing in backness, e.g., talossansako 'in his house?' - kynässänsäkö 'in his pen?' (-j-ϊΛ 'in', -nsA '3 pers. possessive', -kO 'interrogative'). The neutrality of i,e is illustrated in (3). (3)

a. sika 'pig', laki 'law', melu 'noise', sade 'rain' ikä 'age', mäki 'hill', pesä 'nest', säde 'ray' itikka 'bug', sovitellessa 'in adjusting' helinä 'tinkling', käsitellessä 'in handling' b. itu 'sprout (noun)' < itä 'to sprout' veto 'pull (noun)' < vetä- 'to pull' (cf. lähtö 'departure' < lähte- 'to leave') selvyys 'clarity' < selvä 'clear' herruus 'supremacy' < herra 'master' veljeys 'brotherhood' < veli, velje- 'brother' c. miehuus 'manliness' < miehe- 'man' miehyys'masculinity' heinikko 'hayfield' < heinä 'hay' metsikkö < metsä 'forest' mesikka 'arctic bramble' < mesi/mete- 'honey' mesikkä 'sweet clover'

384

Robert Τ. Harms d. itku 'crying' < itke- 'to cry' (cf. itkenyt 'cried') meno 'going' < men- 'to go' (cf. menkö '3 pers. imperative [negative]') seistä 'to stand' (remaining forms of the paradigm are based on seiso-) seisoa e. nominative essive partitive

'sea' meri merenä merta

'water' vesi vetenä vettä

Closer inspection of the morphological formations in (3b-e) gives some notion of the extent to which the neutral vowels complicate the phonology of modern Finnish. The forms of (3b) reveal problems in the rule of stem vowel truncation and its ordering relative to vowel harmony. The apparent inconsistencies among the derivations for the Standard Finnish words in (3c) point up difficulties in determining those rules (and exceptions) which underlie the competence of speakers of modern Finnish. Although the usage of some of these forms is not widespread, native speakers accept all as well formed. The recent formation miehyss and the botanical term mesikkä might suggest a shift toward the rule for productive inflectional harmony, by which stems containing only neutral vowels take front-vowel endings (cf. 'water' in (3e)). On the basis of forms of the type given in (3d) (cf. also miehuus, itu, veto) Kiparsky (1973) has proposed a different harmony role for monosyllabic neutral vowel-roots when followed by a vowel-initial affix: for the sequence C0(i/e)C0 + Vi•..., Vj (if harmonizing) will be back, whereas for C0(i/e)C0 + CVj . j will be front. The fact that formations are restricted to a small set of nonproductive derivatives led Harms (1966) to posit a difference in boundaries as well as underlying backness for the V; suffixes. The lack of harmony throughout an inflectional paradigm is exhibited by only two roots, mer- 'sea and ver- 'blood', as in (3e), with partitive -ta for the expected -tä. Here the tonality-lowering effect of the er rhyme noted by Wiik (1975) accounts not only for these forms, but also for the absence of C(i/e)rCästems in general. The impact of the neutral vowels upon Finnish phonology, as illustrated in (3), is significant. The various synchronic rules required to handle the harmony relationships must necessarily establish formal means to oppose two different vowel classes: (a) neutral vowels as distinct from other front vowels (especially i,e vs. a), and (b) harmonizing vowels as distinct from neutral vowels. Five different strategies for Finnish vowel harmony may be seen in McCawley (1964 - abstract central vowels+, a and a rule of absolute neutrali-

What Helmholtz Knew about Neutral Vowels

385

zation), Harms (1966 - an ad hoc feature parameter 'acute' is used in addition to the harmonizing graveness feature), Kiparsky (1973 - only underlying back vowels are subject to harmony, with secondary restorative backing in the veto, meno types), Wiik (1975 - two distinct harmony rules: fronting of all vowels posterior to the first harmony vowel and a backing harmony agreement rule) and Ringen (1980 - left-to-right iterative rule requiring specification of both the neutral and harmony classes of vowels). It is not my intention here to add yet another set of rules for describing the complexities of Finnish vowel harmony. I wish merely to establish the central role the neutral vowels must play in any attempt to achieve an adequate phonology of Finnish. 3. HUNGARIAN

Hungarian presents yet a somewhat different set of harmony relationships as outlined in (4) (with long vowels indicated by acute accents as in Hungarian orthography: ιϊ = [ü]). (4)

Harmonic: Front ü, u ö, ο e (=[e,ae])2

Neutral: i, i, e (=[e]) Back u, ύ ο, ό a, ä

Alternating sets: u » ύ« ό o^ a,a

ü hat-ul 'behind' - fel-ül 'above' ΰ X-haj-ύ 'X-haired' - X-szem u 'X-eyed' ό häz-bol 'from a house' - fej-böl 'from a head' ö ^ e var-tok 'you wait' - jöt-tök 'you come' - men-tek 'you go' e,e haz-a 'his house' - fej-e 'his head' häz-ä-ban 'in his house' - fej-e-ben 'in his head' häz-ä-näl 'his house (adessive)' - fej-e-nel 'his head (adess.)'

Neutral vowels occur with both back and front harmonizing vowels (e.g., bika 'bull', pelda 'example'). Unlike Finnish, roots with only neutral vowels must to some extent be keyed lexically with regard to harmony type, cf. viz-nel 'water (adess.)' - hid-ηάΐ 'bridge (adess.)', kez-nel 'hand (adess.)' - eel ηάΐ 'goal (adess.)', videk-e 'her region' - ferfi-ja 'her husband'. A count of lexical association of the various neutral vowels with back harmony gives the order ί - i - έ, with only four items, all nouns, having e. In leäny 'girl' and a few words of foreign origin even short e appears marginally neutral, e.g., Agnesnak « Agnesnek 'to Agnes'. Suffixes with i, i, e are strictly neutral in the sense

386

Robert Τ. Harms

that they do not influence the harmony class of word structures, e.g.,piszk-itni-uk 'for them to make dirty' (cf.piszok 'dirt'; contrast also Tibi, diminutive form of Tibor, Tibinek 'to Tibi') - ker-ni-ük 'for them to ask' (cf. ker-ek Ί ask'). For four different formal treatments of Hungarian vowel harmony, see the recent series of views expressed by Jensen, Phelps, Ringen and Vago in Linguistic Inquiry, 9 (1978). 4. T H E H I G H TONALITY AS N E U T R A L TONALITY P A R A D O X

In summary, tonality harmony continues to play a productive role in the phonologies of the various Finno-Ugric languages surveyed here, although it is significantly complicated by the presence of neutral vowels. On the one hand, the tendency toward a regular association of the neutral vowels with front vowel harmony seems perfectly natural in view of their phonetic frontness. On the other hand, strange as it may seem, the neutral vowels are just those with the highest tonality (i.e., F 2 /F 3 frequency values): the non-low unrounded front vowels. Indeed, it is possible to recognize an implicational hierarchy i-e - z, ranked by decreasing tonality values, by which the lowest occurring neutral vowel entails the neutrality of all higher vowels in the series (Figure l). 3 F 2 /F 3 i e ε

Votic Hungarian Finnish

ΰ ae ö Fig. 1. Tonality scale for front vowels based on F 2 / F 3 values.

In my opinion this apparent contradiction of the front/back harmony principle by the fact that the neutral vowels are the most front presents a rather serious obstacle to attempts to provide a plausible nonlinear (autosegmental or metrical) accounting for such systems. Attempts to explain the problems associated with the neutral vowels as historical accident, the consequence of contact with European languages, or with vague notions of markedness do not completely succeed. Early Hungarian is generally assumed to have had consistent front/back harmony, with

What Helmholtz Knew about Neutral Vowels

387

neutral vowels arising through loss of a central *+or monophthongization of back vowel diphthongs, but not all back-harmonizing neutral vowels are the result of secondary fronting, e.g., the German loan eel 'goal' (cf. cel-nak 'to the purpose'). Clearly the neutral vowel category has taken on a purely synchronic role. In the case of the more closely related Baltic-Finnic Finnish and Votic, the currently prevailing opinion is that the neutrality of i and e reflects the original Proto-Baltic-Finnic system, if not that of Finno-Ugric, with unrounded central vowels (as in Votic and Estonian) as secondary developments (Hakulinen, 1961:21). Although I accept instead the opposing view that Proto-Baltic-Finnic had complete tonality harmony, with central unrounded vowels (cf., e.g., Kettunen, 1929:30), I do not consider the presence of the neutral vowels in Finnish in itself to be an argument for an earlier merger of central and front vowels (as suggested by Aoki, 1968:144 45). And appeals to markedness (Aoki, 1968:145) do not allow us to account for the merger of */':*+in Votic and Estonian alongside the retention of the supposedly even more highly marked *e:*s opposition - and this in an area with evidence of contact influence (e.g., consonant palatalization, voicing of obstruents) from Russian, which maintains a primary [4] vowel target type. The unquestionable continuing vitality of vowel harmony among the widely dispersed western Finno-Ugric languages in the face of the neutrality of the front unrounded vowels must be accounted for on purely synchronic grounds. I shall argue below that a fresh approach to the feature parameters for vowels - one based on psychoacoustic and functional considerations leads to a view of the neutrality of non-low front unrounded vowels as perfectly neutral. 5. VOWEL FEATURES

The set of vowel features I shall introduce here represents a highly speculative revision on substantive, psychoacoustic and functional grounds. In large part it was inspired by the mandate of Liljencrants and Lindblom (1972:859): to derive linguistic form as a consequence of various substantive-based principles pertaining to the use of spoken language and its biological, sociological, and communicative aspects. We would then proceed by asking the following questions: (a) What are the mechanisms available for human speech communication? (b) How can the use of these mechanisms be constrained and optimized with respect to various psychological conditions and communicative efficiency?

The possible implications of these features for phonology, including their application to the problems of vowel harmony, were considered later in testing the system against feature related aspects of phonological systems, e.g., natural class specifications, the typology of vowel systems, etc.

388

Robert Τ. Harms

My preliminary analysis was based on the spectra for Swedish vowels presented by Fant (1973:40).4 Thus a wide range of vowel types, 19 in all, within a rich system of phonemic contrasts and the most advanced experimental data were assured. Fant's spectra for u, ii, i, a and ae are given in Figure 2.

dB

-

. . 1..I . -20.

1

-40-

dB 0- Τ - 1 0 -

1

u

+I

I"

dB 0-

dB 0-10-

-40-

-10-

lljli.t..ril, 1ll|lUI mp' nt' nk'. The process of ejectivization is rare, occurring only within the Nguni group (Zulu, Ndebele, Xhosa, Swazi, etc.). No simple articulatorily based model such as that proposed for aspiration presents itself here. Several preliminary points require mention. First, it will not suffice merely to refer to similarities between ejectives and aspirates, since the Nguni case involves both the introduction of post-nasal ejectivization and the loss of aspiration. That is, Bantu *p t k~> ph th kh except under nasal influence when they become p't' k'. There is good reason to suspect that the development of aspiration in simple voiceless stops is older than post-nasal ejectivization, since the former involved all Southeast Bantu languages, whereas the latter is restricted to the Nguni group. Second, both aspiration and ejectivization affect voiceless consonants exclusively. Further, both developments serve to 'heighten' the voicelessness

406

Robert Κ. Herbert

of these consonants in the sense that they 'check' other processes such as voicing, which is the most common process affecting post-nasal stops. Thus, it might be suggested that both ejectivization and aspiration are perceptuallymotivated processes. While there is no evidence to counter this proposal, there is no direct evidence to support it, either; the fact that one 'heightened' type (aspirates) is replaced by another (ejectives) is vexing, although it might be possible to argue that voiceless-aspirated-ejective can be regarded as a sliding scale of complexity. Phonetically, the ejectives in Zulu are only weakly ejected. Doke (1926) reported weak or absent ejection for many informants except in careful, emphatic speech. The same facts are described by Selmer (1933). In modern Zulu, especially urban varieties, it is possible to find speakers who seem to lack ejectives entirely. These sounds thus differ from the typical Amerindian or Caucasian ejectives, where the glottal release is rather prominent. Also of probable importance in explaining the ejective pattern is the role of aspirates, ejectives, and so-called implosives as areal features in Southern Bantu. While a process of unrestricted aspiration of Bantu stops can be reconstructed for the Southeastern languages, aspiration is in general typologically uncommon in Bantu. 5 Similarly, Bantu implosives are largely confined to this area, and the click consonants occur in no other Bantu languages. It is well established that clicks entered the Bantu languages through contact with Khoisan languages, probably Hottentot. An essential difference between Bantu clicks on the one hand and ejectives, aspirates, and implosives on the other is that clicks occur mainly in borrowed words, whereas the others are regular developments from Proto-Bantu consonants. Lanham (1964) and Westphal (1963) have described the linguistic and cultural prehistory of southern Africa. It is possible that, in addition to the enormous influence on the Bantu phonemic systems and lexicon, the contact situation might have resulted in the development of a Southern Bantu 'articulatory mode'. The nature of such a mode would involve a predilection for glottalic consonants, clicks, aspirates, etc.6 Some support for the notion of ejection as an articulatory mode or 'family universal' (cf. Lass, 1975) derives from the fact that in several of the languages ejection is characteristic of all plain voiceless obstruents. For example, Bailey (1976) reports that in Copi, which has a rich consonant inventory, ejection is optional with plain voiceless consonants, which are opposed to voiceless aspirates, implosives, and 'voiced aspirates'. In Phuti, post-nasal voiceless consonants are rare since Ν -» 0 / C, but they are ejective when they do occur in Sotho borrowings; the plain voiceless stops are 'slightly ejective' and are opposed to aspirates, voiced stops, and implosives (Mzamane, 1949). Bantu *mp nt nk> b dg>ρ t k>p't' k' in Sotho,although,as Tucker (1929) noted, ejection is an optional feature of these consonants. 7 Returning to the puzzle of post-nasal ejective consonants in Zulu and other

The Puzzle of Bantu Ejectives and Aspirates

407

Nguni languages, it is suggested that this process can be understood by reference to two basic typological facts. The first is the universal tendency for voiceless obstruents in contact with preceding nasals to be unaspirated (although this preference is counteracted in the Southern Bantu language groups). The second typological fact has to do with the articulatory mode of Southeastern Bantu in which weak ejection is characteristic of plain voiceless stops and affricates. Thus, in place of the simple formula *mp nt nk > mp' nt' nk' or the phonological rule [+aspirate] — [+ejective]/ Ν , one needs to understand a two-stage process: (1) deaspiration (2) ejectivization of plain voiceless obstruents Such an analysis has the advantage of bringing Nguni within a single typological class insofar as aspiration is concerned; 8 as mentioned earlier, aspirated clicks unambiguously lose their aspiration in this group. In Swazi, another Nguni language, post-nasal stops are deaspirated, but only nt is subject to ejectivization: *p t k> ph th kh\ *mp nt nk > mp nt'/nt nk (Ziervogel, 1952). It may be significant that in Swazi, with its non-ejective mode, there is a tendency to voice post-nasal stops, e.g., / m p / tends to be [mb]. For some speakers, Bantu *mp nt nk and *mb nd ng are thus not distinguished in some styles of speech. We have thus far briefly considered the data concerning the post-nasal ejectivization reported for Nguni languages. It has been proposed that, instead of such a process, which runs counter to the evolutionary preference for (ND) sequences, the data exhibit the combined effects of deaspiration and an 'articulatory mode' wherein all nonaspirated voiceless obstruents are subject to ejectivization. This ejection is most clearly marked in post-nasal consonants, but is, in addition, optional throughout the system in most languages. The stronger ejection following nasal consonants may be related to timing considerations, or the ejective quality of these consonants may function to preserve their voicelessness as suggested by the Swazi data. 3. ARTICULATORY MODES

At this point the question of the exact nature of an articulatory mode arises. The concept is cited with some frequency in the literature, especially by authors who have at some time been associated with the University of Edinburgh. There is some variation in its use, but the basic notion seems to be that material for language acquisition involves the articulatory mode/set for a particular language/dialect and forms part of the material which must be acquired in learning it. One may say that 'articulatory mode' refers to general phonetic principles which govern the articulation of a particular language or language variant. These modes are frequently associated with voice qualities, and anyone who has heard much spoken Zulu cannot fail to have been

408

Robert Κ. Herbert

impressed by the 'breathy voice' of many speakers. There is no reason why the concept of articulatory mode might not be extended to include, for example, the ejective tendency of many Southern Bantu languages. Such an approach has much in common with the characterization as a 'low level phonetic realization rule', although it is difficult to specify what these two different views of ejection entail. Assuming the validity of an ejective articulatory mode in Southern Bantu, its origins are apparently buried in prehistory. It is tempting to speculate, however, that its genesis might lie in language contact. There is excellent evidence for substantial Bantu-Hottentot contact, which included widespread replacement of Hottentot languages by Bantu languages (Westphal, 1963). It is not unreasonable to hypothesize, on the basis of the extant Hottentot languages, that the contact language was characterized by considerable complexity in phonation types; in particular, the presence of clicks seems to imply the presence of glottalic consonants. Further evidence which suggests a contact basis for the ejective mode is the fact that the Hottentot words for 'cow', 'sheep', and 'milk' are borrowed into all Southern Bantu languages except Shona, which has reflexes of Bantu terms for these items. This opposition between Shona and the other Southern languages is the same as that found for the ejective mode, i.e., the opposition between voiceless consonant types in Southern Bantu tends to be between aspirates and ejectives, except in Shona where neither type occurs. There are other features distinguishing the two groups of languages, and some of these may also derive from contact. Evidence for a relatively late origin of the ejective mode comes from the fact that the Proto-Bantu voiceless stops are not ejectives in any of these languages; they were aspirated in early times (pre-contact) and thus not subject to ejectivization. 9 The aspirated stops were liable to further development, but the aspiration typically reappears under nasal influence, e.g., Venda *p t k> φ r h, and synchronically / Ν + φ r h / [p h t h k h ]. 10 Had the ejective mode been present in earlier forms of Southern Bantu, ejection would have inhibited aspiration, since the two phenomena are mutually exclusive. In a stimulating article Lass (1975) proposed the notion of'family universal,' which has obvious affinities with the concept of articulatory mode. Briefly, Lass took issue with the concept of universal marking constraints in generative phonology, which asserted that certain features and sounds will be less common among the world's languages, will tend to be lost over time, will be acquired later by children, etc. He considered the cases of front rounded vowels in Germanic and retroflex stops in Dravidian, both marked sound types, and he proposed that these sounds are neither marked nor unmarked in their respective language families: they are simply part of the inventory. Germanic languages such as Swedish work as well as Germanic languages without front rounded vowels, and Lass suggested that markedness theory

The Puzzle of Bantu Ejectives and Aspirates

409

could then be reduced to a weak statement of statistical probability with no inherent content. Herbert (1978) examined the Lassian notion of 'family universal' and, citing data involving clicks and geminate consonants, argued that these segments exist only in opposition to less marked types. That is, clicks and geminates occur and are stable in certain languages, but if the simple stops undergo spirantization, we expect that their geminate counterparts will be 'reinterpreted' as simple stops. The set of predictions provided by markedness theory is borne out in such cases. Marked segment types may develop within a language (e.g., as the result of introducing an umlaut process) or may increase in number (e.g., if new forms are subject to an existing umlaut process), but the development of marked types cannot be context-free. In such a situation markedness predictions would be falsified. For example, the shift *i — y cannot occur as an unconditioned sound change unless i.n The reverse process, *y — /, should, by contrast, be common both in historical development and in acquisition. Applying these principles to ejective consonants, the prediction is that they should preserve this feature only in the presence of non-ejective counterparts. The unconditioned process C' — C should be common, whereas C C' should be uncommon, unless other shifts occur within the consonantal system resulting in replacement of the plain consonants. In the absence of plain voiceless consonants, it is expected that ejectives will be reinterpreted in that way. Unfortunately, it does not seem possible to reconcile the Southern Bantu data with this approach. Consider, for example, the stop inventory of Venda(Doke, 1954): P' Ph b mb

r th d nd A

Α

Λ

t' th d nd

k' kh g ng

The same four distinctions are present in the affricate series. Such an inventory seems 'ripe' for reinterpretation, yet there is every indication that it is stable and has been so for generations. This sort of situation is typical of Southern Bantu languages. Before abandoning markedness the points under consideration here should be summarized. First, although ejection has been a stable feature over time, it is phonetically optional; this is usually stated explicitly in grammatical descriptions, or it is reported that ejection is 'not very prominent.' (My impression is that ejection varies with speech style, being most noticeable in careful speech.) Second, there are very few studies of the acquisition of a Southern Bantu language, and these deal only with morphology (Kunene, 1979; Suzman, 1980); however, informal observation suggests that children do not noticeably mark ejection either. Third, it has been noted above that

410

Robert Κ. Herbert

within Southern Bantu ejection exists in opposition to aspiration. Both are generally unknown in Shona, although there are Shona dialects where both occur or where aspiration alone occurs; no dialect, it seems, presents ejectives in the absence of aspirates. This intriguing fact may point to a special status enjoyed by ejection in Southern Bantu. Fourth, ejection in Southern Bantu seems to differ phonetically from ejection elsewhere in that glottalization, when present, is not very marked: the degree of supraglottal pressure created is not very great and a separate glottal release is seldom heard. Fifth, the 'least marked' stop type, plain voiceless consonants, does not typically occur in these languages. 12 It is tempting, on the basis of the above, to propose that Southern Bantu ejectives are not 'typical ejectives.' What ejection means for Southern Bantu is that in the articulation of unaspirated voiceless obstruents the vocal folds are brought rather closer together than is typical for plain voiceless consonants. The opposition aspirated:unaspirated is interpreted as relatively large glottal opening:relatively restricted/closed glottis. Ejection, in this sense, differs from its usual meaning. If so, such a difference may reflect the difference between an articulatory mode and a phonetic realization rule. This attempt at a solution to the problem of ejection in Southern Bantu, viewing it as a phonetic mode, is not any more satisfying than describing a language with the vowel inventory / y e u o a / a s having an articulatory mode for roundness in vowels. No such vowel inventory exists, and markedness theory, with bases in articulatory and perceptual phonetics, claims it cannot exist. However, as Roger Lass once informally observed: one doesn't disprove the egg-laying nature of the duck-billed platypus by producing a picture of one - or even a dozen - not laying eggs. The lack of content in its predictive value is what led Lass to reject the markedness theory of generative phonology. The theory is, he claimed, devoid of intrinsic content to the extent that it cannot reconcile meta-theoretical considerations of simplicity with language-internal economy of phonological systems. There is no apparent way to reconcile the phonetic mode of Southern Bantu with universal considerations of markedness in consonant types. One approach might be to analyze the ejective consonants as underlying plain consonants that are realized with optional surface ejection. Thus, there would be no contradiction of markedness predictions at the underlying level. However, this is a disguise rather than a solution to the problem. There seems to be good reason to believe that the underlying contrast is between ejective and aspirated types and that the realization rule effects weakening of the ejective consonants. Obviously, a single example such as the above does not invalidate markedness theory. Rather, this example points to the need for more precise definition of the term articulatory/phonetic mode and the status of such a concept, if any, within phonological theory. In the absence of any explana-

The Puzzle of Bantu Ejectives and Aspirates

411

tion for the ejective mode of Southern Bantu, all we can do at present is to note, using Lass' metaphor, that Southern Bantu is 'that kind of beast 1 and that ejectives are 'like gizzards in chickens,' i.e., simply part of the inventory. ACKNOWLEDGEMENTS

This research was in part supported by an ACLS postdoctoral fellowship and SUNY Faculty Research Grant to the author. I am grateful to the Department of Phonetics and Linguistics at the University of the Witwatersrand for extending me research facilities in 1981. A version of this paper was presented at the Sixth International Conference on Historical Linguistics, Poznan, 1983, and will appear in the proceedings. NOTES 1.

2.

3. 4.

5.

6. 7.

8.

9.

10.

There remains the larger question of whether true explanation is possible within linguistics, a question which hinges on the definition of'explanation.' In the realm of historical studies actuation remains the most vexing problem; see Jeffers (1974) for a discussion oiexplanation in historical linguistics. The term 'ejectivization' is recognizably deficient since it suggests that the ejective quality is a secondary articulation and that this process operates upon plain consonants, whereas in fact the term refers to the replacement of a plain consonant by its ejective counterpart. I am grateful to Erhard Voeltz for pointing out that Panconcelli-Calzia (1915/16) provided an antecedent model that closely approximates that of Hinnebusch in broad outline. There are six groups of languages recognized within Southern Bantu (Guthrie's Zone S). Five of them are represented here by Shona (Manyika) (S13), Venda (S21), Sotho (S33), Tswa (S51), and Copi (S61); the Nguni group (S40) is represented below by data from Zulu (S42), Xhosa (S41), and Swazi (S43). In former classifications, Shona was assigned to a separate Zone T, and the remaining five to the Southeastern Zone. There are good reasons which motivate such a distinction, some of which are mentioned below. It occasionally occurs that a specific environment will induce aspiration of voiceless stops; the most common such environment will induce aspiration of voiceless stops; the most common such environment in Bantu seems to be the position preceding one of the Proto Bantu 'super close' vowels, e.g., Makua (P31) *p > ph/ j, but elsewhere *ρ > v. This proposal has much in common with Westphal's (1963) notion of 'phonation areas'. The same variation is characteristic of Southern Bantu implosives: / 6 / { / d / rarely occurs) varies from a voiced bilabial fricative to a lenis stop to a true (weakly) imploded stop. Deaspiration of post-nasal consonants is noted as a Nguni innovation, since other Southern languages show the opposite pattern. However, it is perhaps significant that Nguni is the only group in which *p t k> ph t* k* without further development, while in the other groups they are weakened, e.g., Venda φ r h. The dating of this contact is uncertain, since there is some disagreement among the archaeological, linguistic, and folkloric evidence, but a time depth of about 500 years can be assumed. This process is now morphophonological; the nasal appears only in restricted environments, e.g., before monosyllabic stems and when it represents the first person singular object marker.

412 11. 12.

Robert Κ. Herbert This view obviously has much in common with Martinet's concept of push- and dragchains. There are a few Southern Bantu languages which appear to present simple voiceless stops, e.g., / k / in Zulu, but in fact these are 'semi-voiced'.

REFERENCES Bailey, R.A. (1976). Copi Phonology and Morphotonology. Unpublished B.A. Honors Thesis, University of the Witwatersrand. Catford, J.C. (1968). The Articulatory Possibilities of Man. In: B. Malmberg (ed.) Manual of Phonetics. Amsterdam: North-Holland Publishing Co., pp. 303-333. Catford, J.C. (1977). Fundamental Problems in Phonetics. Edinburgh: Edinburgh University Press. Devine, (1974). Aspiration, Universale and the Study of Dead Languages. Working Papers in Language Universals, 15, 1-24. Doke, C.M. (1926). The Phonetics of the Zulu Language. Bantu Studies, Special Number. Doke, C.M. (1954). The Southern Bantu Languages. London: Oxford University Press. Givon, T. (1974). Rule Un-ordering: Generalization and De-generalization in Phonology. In: A. Bruck, R.A. Fox and M. W. La Galy (eds.) Papers from the Parasession on Natural Phonology. Chicago: Chicago Linguistic Society, pp. 103-115. Greenberg, J.H. (1966). Synchronic and Diachronic Universals in Phonology. Language, 42, 508-17. Greenberg, J . H . (1970a). Language Universals. In: T. Sebeok (ed ,)Current Trends in Linguistics, vol. 3. The Hague: Mouton, pp. 61-112. Greenberg, J.H. (1970fc). The Role of Typology in the Development of a Scientific Linguistics. In: L. Dezsö and P. Hajdü (eds.) Theoretical Problems of Typology and the Northern Eurasian Languages. Budapest: Akademiai Kiado, pp. 11-24. Greenberg, J.H. (1970c). Some Generalizations Concerning Glottalic Consonants, Especially Implosives. International Journal of American Linguistics, 36, 123-145. Greenberg, J.H. (1978). Diachrony, Synchrony, and Language Universals. In: Greenberg, Ferguson and Moravcsik, (eds.) (1978), vol. 1, 61-91. Greenberg, J.H., Ferguson, C.A. and Moravcsik, E.A., (eds.) (1978). Universals of Human Language, 4 vols. Stanford: Stanford University Press. Guthrie, M. (1967-70). Comparative Bantu, 4 vols. Farnborough: Gregg International Publishers. Herbert, R.K. (1978). Another Look at Meta-rules and 'Family Universals'. Studies in African Linguistics, 9, 143-165. Herbert, R.K. (1979), Typological Universals, Aspiration, and Post-nasal Stops. In: Proceedings of the Ninth International Congress of Phonetic Sciences, II. Copenhagen: University of Copenhagen, pp. 19-26. Hinnebusch, T. (1975). A Reconstructed Chronology of Loss: Swahili Class 9/10. In: R. Herbert (ed.) Proceedings of the Sixth Conference on African Linguistics. Columbus: Ohio State University Working Papers in Linguistics, pp. 32-41. Jakobson, R. (1941). Kindersprache, Aphasie und Allgemeine Lautgesetze. Uppsala: Spräkvetenskapliga Sällskapets i Uppsala Förhandligar. Jakobson, R. (1958). What Can Typological Studies Contribute to Historical Comparative Linguistics? In: Proceedings of the Eighth International Congress of Linguists. Oslo: Oslo University Press, pp. 17-35. Jeffers, R.J. (1974). On the Notion 'Explanation' in Historical Linguistics. In: J.M. Anderson and C. Jones (eds.) Historical Linguistics II. Amsterdam: North-Holland Publishing Co., pp. 231-255.

The Puzzle of Bantu Ejectives

and

Aspirates

413

Kunene, E. (1979). The Acquisition of Siswati as a First Language. Unpublished Ph.D. Dissertation, U.C.L. A. Ladefoged, P. (1971). Preliminaries to Linguistic Phonetics. Chicago: University of Chicago Press. Lanham, L.W. (1964). The Proliferation and Extension of Bantu Phonemic Systems Influenced by Bushman and Hottentot. In: Proceedings of the Ninth International Congress of Linguists. The Hague: Mouton, pp. 383-391. Lass, R. (1975). How Intrinsic is Content? Markedness, Sound Change, and 'Family Universals'. D. Goyvaerts and G. Pullum (eds.) Essays on the Sound Pattern of English. Ghent: Story-Scientia, pp. 475-504. Lisker, L. and Abramson, A. (1964). A Cross-language Study of Voicing in Initial Stops. Word, 20, 384-422. Mzamane, G.I.M. (1949). A Concise Treatise on Phuti with Special Reference to Its Relationship with Nguni and Sotho. Fort Hare Papers 1(4). Ohala, J. (1980). Introduction, Symposium on Phonetic Universals in Phonological Systems and Their Explanations. In: Proceedings of the Ninth International Congress of Phonetic Sciences, III. Copenhagen: University of Copenhagen, pp. 181-185. Panconcelli-Calzia, G. (1915/16). Objektive Untersuchungen über die stimmlosen Nasale im Ndonga. Zeitschrift für Kolonialsprachen, 6, 257-263. Selmer, E. (1933). Experimentelle Beiträge zur Zulu Phonetik. Avhandlinger utgitt av Det Norske Videnskaps-Akademi i Oslo, II, No. 1. Suzman, S.M. (1980). Acquisition of the Noun Class System of Zulu. Papers and Reports on Child Language Development, 19, 45-52. Tucker, A.N. (1929). The Comparative Phonetics of the Suto-Chuana Group of Bantu Languages. London: Longmans, Green and Co. Westphal, E.O.J. (1963). The Linguistic Prehistory of Southern Africa: Bush, Kwadi, Hottentot, and Bantu Linguistic Relationships. Africa, 33, 237-265. Ziervogel, D. (1952). A Grammar of Swazi. Johannesburg: Witwatersrand University Press.

Loan Words and the Estonian Grade Alternation Mati Hint, Tallinn, Estonia,

U.S.S.R.

1. Baltic Finnic languages, including Estonian, have during their long history borrowed words from their various neighbors: from Baltic (in the 2nd millenium BC), Proto-Germanic (in the first millenium BC) and Slavic (in the first millenium AD), and subsequently from Low and High German, Russian, Swedish and other languages. Phonetic/phonological integration of loan words into the system of Estonian has been treated in many studies and the thematic vowel of loan words (as an indicator of loan source) has been studied as well. The morphological and morphophonological integration of loan words, on the other hand, has not attracted the attention which it deserves. The largest and the most significant part of Estonian morphophonemics is its grade alternation system. The type of grade alternation in loan words (or the lack of alternation) has come up in previous studies only when the inflectional type of the loan word is mentioned (e.g., Hinderling, 1981:58ff.), or when several inflectional forms of one loan word are used as examples of borrowing (e.g., Liin, 1968; Ariste, 1981). But grade alternation in Estonian is of such linguistic importance that it requires more attention, especially because the integration of loan words into the different grade alternation types is not free of significant problems. This paper constitutes only a preliminary treatment of some questions in this area. In the following an attempt is made to clarify whether and how loan words help to characterize the productivity of the various grade alternation types at the time of borrowing. This question has several different aspects. First we will take a closer look at what types of grade alternation have been applied with consistency to loan words and what types have not. The consistent absence of grade alternation or the presence of inconsistent alternation in loan words having certain phonetic structures makes it possible to ask whether at the time of borrowing a given alternation type was no longer productive or had not yet been fully generalized to all the words with suitable phonetic structure. For example, the Low German loan wordspriigi (MLG brugge 'rubbish') and nagi (MLG knagge 'coat-rack') are inflected without grade alternation (nom. sg. priigi, nagi; gen. sg. priigi, nagi', cf.

416

Mati Hint

alternating inflection such as inpügi :p'öe 'shearing, nom.: gen.' or lagi: Γae 'ceiling, nom. : gen.'). 1 ' 2 We may then ask 1) whether grade alternation (loss or spirantization of the stop at the onset of the second syllable) in words with a short open first syllable was no longer so widespread at the time of borrowing that it would automatically have been generalized to all words having such a structure, or 2) whether grade alternation had not yet been generalized and there existed, side by side, two major inflectional types: one with grade alternation and the other without it. These questions are as essential for the understanding of the integration of loan words into Estonian morphology as they are for the investigation of Estonian grade alternation itself (the generalization, loss of productivity, and ultimately partial elimination of various types of grade alternation). 2.

There are three different types of grade alternation in Estonian, the history and the productivity of which in the contemporary language are quite different. In the case of consonant mutation (qualitative alternation) there is a stop or s at the onset of the second syllable in the strong grade form (S), which in the weak grade form (W) is either lost, assimilated to a preceding sonorant or replaced by a semivowel, e.g. rege (S, part, sg.): fee (W, gen. sg.) 'sledge', kulda (S, part, sg.): kulla (W, gen. sg.) 'gold', sada (S, nom. and part, sg.): saja (W, gen. sg.) 'hundred', töbe (S, part, sg.): töve (W, gen. sg.) 'disease', etc. Alternation of geminate stops takes place after a long first syllable: the strong grade has a geminated stop, and the weak grade has a single (phonetically lenis) one, e.g., nurka : nurga (=[nurkka]: [nurGa]) 'corner, part, and gen.', tafpu : taibu (=[taippu]: [ta'iBu]) 'understanding, part, and gen.', etc. In the case of late prosodic quantity alternation the long first syllable is pronounced in the strong grade with extra intensity, greater length and sometimes (in the case of voiced sounds) with a particular, marked tone contour. This pronunciation is traditionally called the third degree of quantity (Q3) or overlength. In the weak grade a long first syllable is pronounced without these special features, and this pronunciation is called the second degree of quantity (Q2), e.g.: Q2, gen. sg. Q3, part. sg.

'forest'

'mercy'

'town'

'song'

'salt'

metsa metsa

armu a'rmu

linna li'nna

laulu Γ aulu

soola s'oola

(=[söla]) (=[söla])

Today these grade alternation types have merged to a considerable extent, but the historically very early qualitative alternation and the alternation in the geminate stops, on the one hand, and the historically late (not earlier than

417

Loan Words and the Estonian Grade

the 15th century AD, but presumably much later) prosodic Q3 : Q2 alternation, on the other hand, have retained their main point of difference: in the first two alternation types the segmental composition of the strong and weak grades is different (in terms of segmental phonemes), while in the case of late quantity alternation the difference between the strong and weak grades is exclusively prosodic. There are intermediate alternation types as well (e.g., alternation of geminate stops after a short vowel as in se~ppa: sepa (=[seppa]: [seppa]) 'smith, part, and gen.', etc.), the identification of which is not relevant for purposes of the present paper. The historically early alternation in the geminate stops and the historically late quantity alternation (Q3: Q2) are entirely productive morphophonological processes in the modern language. Almost all words having a suitable phonetic structure are inflected according to these alternation types (including proper nouns and foreign place names such as New Yorki : New Yorgi, Stuttgafti : Stuttgardi 'illat. vs. gen.', and even foreign personal names such as Shakespeare or Bergmann, which have regularly alternating last syllables in Estonian). Qualitative consonant mutation is no longer productive in the modern language. Recent loan words and international terms are not inflected according to the old pattern of consonant mutation, although inherited lexical items and early loan words retain this morphophonemic pattern (especially words with a high frequency of occurrence in everyday speech). Inherited lexical items and early loan words: consonant mutation

S: W:

Recent loan words and international terms: prosodic quantity alternation

'bread' 'gold'

'ox'

'watchman'

Γ eiba leiva

harga härja

vahti vahi

ku Ida kulla

S(Q3): W(Q2):

'washer'

'guild'

'erg'

'shaft'

s'eibi seibi

gi'ldi gildi

ergi ergi

ia'hti iahti

In these examples the strong grade is illustrated by part. sg. and the weak grade by gen. sg. 3. Loan words cannot supply substantial information about the history of late prosodic quantity alternation and alternation in geminate stops for a few reasons: 1) since both these types of alternation are fully productive in the contemporary language, the inherited lexical items, the early borrowings and the later borrowed vocabulary cannot be divided into contrasting groups on the basis of alternation or lack of it; 2) the most numerous group of borrowings in Estonian consists of Low German loan words (and subsequently High German loan words), but the orthography of Estonian at the time of these borrowings was too inconsistent to reflect the lack of alternation in

418

Mali Hint

geminate stops reliably. (Late prosodic alternation is not indicated even in modern Estonian orthography; the grave accent in Q3 syllables is used only in normative dictionaries and in phonetic transcription.) There are sufficient grounds to assume that alternation in geminate stops was a very productive morphophonemic process at the time of the borrowings from Low German and that all loan words with the appropriate phonetic structure were quickly integrated into this alternation system (e.g., trukkima : trüki MLG drücken 'to print, inf. and imp.', märki: märgi MLG merk 'mark, part, and g e n . \ p i n k i : pingi MLG benk 'bench, part, and gen.', etc.). But late prosodic quantity alternation was presumably only in the embryonic stage of development at the time of the borrowings from Low German and the contrast Q3 : Q2 was not yet distinctly established. 4. The situation of Low and High German loan words with respect to consonant mutation patterns presents some problems of a more intriguing nature, though. First, in the case of consonant mutation the difference between strong and weak-grade stem-forms is clearly expressed even in the inconsistent orthography of the 16-17th centuries. It should be kept in mind that even the best and most influential authors - such as H. Stahl - represented the difference between the strong and weak grades very inconsistently. Therefore, the overuse of the strong grade forms need not mean that the word had no alternation in the stem. It is rather an indication of the inability or unwillingness of grammarians and writers to understand the system of stem alternations. For example, Stahl uses the strong grade in such forms as nachkast (for nahast) 'skin, elat.', Kesckust (for käsust) 'order, elat.', Techtest (for tähest) 'star, elat.', lindust (for linnust) 'bird, elat.', etc. In spite of these "malformed" elatives, there is almost no doubt that at that time the weak grade was, at least in these alternation types, identical to the modern form in all essential features. Systematic orthographic distinction between strongand weak-grade stem variants is, of course, sufficient evidence for regular grade alternation, e.g.,jalgk: jallast(ja lg: jalast 'foot, nom. and elat.'), selgk : seljast (selg : seljast 'back, nom. and elat.'), kohus : kochtust (kohus : ko'htust 'court of law, nom. and elat.'), techt : tehhe : techte (taht : tähe : tahte 'star, nom., gen., part.'), etc. Second, Low German loan words with phonetic structure appropriate for grade alternation (a stop at the onset of the second syllable) in the contemporary language belong to two groups with regard to grade alternation: 1) + grade alternation (+GA), e.g., vahti: vahi MLG wachte 'watchman, part, and gen.', and 2) -grade alternation (-GA), e.g., priigi: prügi: prügi MLG brugge 'rubbish, nom., gen., part.'. This difference is both central and problematical. The spelling of Estonian in the 17th century offers some insights for constructing an explanation of this difference.

Loan Words and the Estonian Grade

419

5. Early Estonian sources enable us to affirm that at the time of the borrowings from Low German the consonant mutation system was already established and productive in the majority of word types where the mutating stop (or 5) occurred after a long first syllable and the weak-grade stem was disyllabic. In the contemporary language the majority of the inherited lexical items having this phonetic structure, as well as Baltic, Proto-Germanic, Slavic, Low German and earlier High German loan words, have morphophonemic grade alternation. (A few exceptions are the result of later degeneration of the consonant mutation system and generalization of one of the stem variants - S or W; there are some Low German loan words which present problems in this connection, see below.) The situation is quite different with regard to the occurrence of grade alternation after a short first syllable (after a short main stressed vowel): a large part of the inherited Baltic Finnic words and of the earlier Baltic and Proto-Germanic loan words belong to the mutating type, while a small part of them (including all onomatopoeic words) and all Low German loan words are inflected without grade alternation. This situation enables us to conclude that at the time of the borrowings from Low German (13th-17th centuries) grade alternation either was no longer productive in this phonetic type or had not yet been generalized into a fully productive morphophonemic device. PRINCIPAL ALTERNATION TYPES IN LOAN WORDS Alternation examples in earlier loan-words: Baltic (B), Germanic (G), and Slavic (S)

Alternations in Low and High German loan-words

1) Loss of intervocalic b, d, g, s and automatic insertion of a semivowel, e.g., Γauda: laua (=[laüwa]), feist reied (=[reije]): Γ auda: laua (B) 'table' k'uube: kuue M L G schube 'coat' Γuuda\ luua (B) ' b r o o m ' r'eisi.reied (B) 'thigh, pi.' r'auda: raua (G) 'iron' Γ iuda: liua (S) 'dish' and others 2) Loss of the voiceless fortis stops t and k after the spirants h a n d s: luhta: luha (B) 'water-meadow' ja'hti: jahi and ja'htima: jahi tohtu: tohu (B) 'birch-bark' M L G jacht 'hunting, to h u n t ' sa'hka: saha (S) 'plough' va'hti: vahi and va'htima: vahi vahki: vähi (B) 'crayfish' M L G wachte and wachten 'watchand others man, to watch' pfhti: pihi and pfhtima: pihi M L G bichte, bichten 'confession, to confess' ri'htima: rihi H G richten 'to get straight'

420

Mati Hint

ta hti: tahi MLG dacht 'wick' lii'hti: liihi MLG luchte 'lantern' la'htima: lahi MLG slachten 'to slaughter' vi hti: vihi HG Gewicht 'weight' ki'hti: kihi HG Schicht 'layer' prahti: prahi MLG vracht 'rubbish' ναΓ ski: val'si and völ"skma: völsi lai'ska: laisa (B) 'lazy' (dial.) MLG valsck, valsch, MG falroo ska: roosa (S) 'whip' schen 'false, to falsify' tu ska: tusa (S) 'grief kriiskama: kriisata MLG krischen, and others HG kreischen 'to screech' hei'skama: heisata (HG heischen ?, cf. Hinderling 1981: 58, fn. 95) 'to hoist' 3) Assimilation of the voiceless lenis stops b, d to preceding sonorants: si'lda: silla (B) 'bridge' kilda: killa MLG gilde 'gang' kulda: kulla (G) 'gold' mo'l'di: mol'li MLG molde 'trough' ra'nda: ranna (G) 'strand' punda: punna MLG punt '20 pounds' suAdima: surini (S) tundi: tunni MLG stunde 'hour' to Ida: tölla MLG telde 'coach' 'to force' mel'dima: mel'li MLG melden parda: parra (?S, ?G) 'to report' 'beard' and some others and others

The strong grade in these examples is represented by part. sg. and -ma infinitive forms; the weak grade is represented by gen. sg. and 2 sg. imp. or -ta infinitive forms. There are some additional alternation types which are not commonly found in loan words (e.g., varbama : värvata HG werben 'to recruit'), but which fit into the general patterns of grade alternation after a long first syllable exhibited by native words. These consonant mutation patterns seem to have been relatively automatic. Early Estonian sources confirm that these alternations occurred even in occasional loan words in the 17th century (e.g., landsknecht: landsknehit 'nom. sg. and pi.', (*richte): rihhit(= [rihit]), cf. MLG richte 'dish', (*wunde): wunnith 'nom. pi.' MLG wunde 'wound', and others). But there are some signs that these mutation patterns began to lose their universal productivity just at the time of later Low German borrowings, although there do occur some later loan words from High German which show quite regular alternation. 6. In the period of Low German borrowings several phenomena are worthy of attention. First, there are loan words which have avoided the alternating inflection type: ta^sku MLG taske, tasche (? Swedish loan, cf. Ariste, 1981:139, 153; Hinderling, 1981) 'pocket', lask(i) MLG vlasche 'bottle',

Loan Words and the Estonian Grade

421

me*ski MLG mesch 'mash', tfsk(i) MLG disch, disk 'dish', karske MLG karsch 'chaste', prfske MLG vrisch 'fresh', värske MLG versch, wersck 'fresh'. But almost all these exceptions are regular with regard to grade alternation in some dialect area, e.g., värske and karske in the Western dialect and in the Island dialect belong to the /-stems, and in the Western dialect area the mutating type of inflection is found ( v ä f s k : värsi, karsk: karsi); tasku has mutation in Western Estonia and on the island of Saaremaa (ta^sk : tasu), and occasionally in the South Estonian Tartu dialect (ta^sk: tasi)\ me*ski is typically found with mutation in the Western dialect (mö^sk: mösi). Second, the consonant mutation processes did not operate without exception in the production of otherwise very regular verbal derivatives during the period of Low German borrowings, e.g., vanderdama M L G wanderen 'to wander' in Estonian should be vannerdama (because of the closed second syllable) but such a form is not attested. High German verb stem borrowings always have a phonetic shape characteristic of the strong grade in such derivatives: rehkendama H G rechnen 'to calculate', mehkeldama HG einschmeicheln 'to ingratiate', sahkerdama H G schachern 'to huckster', and others. Third, various sound changes had by this time broken the phonetic conditioning of the strong- vs. weak-grade alternation. This is reflected both in the aforementioned cases and in polysyllabic borrowed nouns, where it is difficult to take account of all the modifications which occur, e.g., vammuiMLG wambds 'doublet' has a weakened stem (-mb- has assimilated to -mm- in grade alternation position), but olderman has preserved the original consonant cluster with a stop as its second member. Consonant mutation seems to have lost its productivity just at the time of the later Low German loans; when gilde was borrowed for the second time (in the meaning 'craft, guild'), it had no consonant mutation, though it does show late prosodic quantity alternation: gild: gildi: gildi (contrast of Q3 : Q2 : Q3 in nom., gen. and part, sg., respectively; for its treatment when borrowed earlier see the chart of Principal Alternation Types in Loan Words above). 7. Despite the evidence of irregularity, the consonant mutation types dealt with up to now may be considered relatively productive and general at the time of the borrowings from Low German. However, the situation for words with a short first syllable (a single stop at the onset of the second syllable after a short vowel) is quite different. In the modern language the majority of such words belong to the inflection type without grade alternation and no Low or High German loan word of this phonetic type has an alternating stem.

422

Mati Hint

Inherited and early borrowed words 2) without grade 1) with grade alternation alternation S ( = nom.):W ( = gen.)

Low and High German loan words (no grade alternation)

lagt Γ ae 'ceiling' pügi: p"öe 'shearing' higt h'ii 'sweat' (dialectal inflection) töbi: töve 'disease' vagu: vao (B) ' f u r r o w ' tuba: foa (G) 'room* hobu: Κ oo 'horse' (archaic inflection) and many others

nagi M L G knagge 'coat rack' pigi M L G pek, pik 'cobbler's wax' ribi M L G ribbe 'rib' priigi M L G brugge 'rubbish' nibu M L G snibbe, Swed. snibb 'nipple' plagu M L G flagge, Swed. flaggo 'flag' logima HG loggen 'to keep a

vaba (S) 'free' modu (?G, ?B) 'mead' tädi 'aunt' and many others Descriptive and pejorative words: kröbi, köbi, logu 'wreck' tobu 'silly' libu 'whore' and many others

It should be stressed that among inherited vocabulary and early loan words of this phonetic type a considerable subgroup has presumably never had grade alternation (onomatopoeic words). Nevertheless is it remarkable that all Low German (and, of course, all later High German) loan words of this type are inflected without grade alternation. According to F.J. Wiedemann's Estonian-German dictionary (1869) two Low German loan words included here - nagi and nibu - may be inflected either with or without grade alternation: nagi (nom. and part, sg.): nae or nagi (gen. sg.), and nibu : neu (neo) or nibu. Inflection with grade alternation was structurally possible, of course, but Wiedemann has given no hints concerning the dialectal background of the forms which show alternation. The data from early literature are too scanty to answer these detailed questions exhaustively. The other Low German loan words of this type - prügi, ribi, pigi, plagu - have only non-alternating inflection even in Wiedemann's dictionary, and there are no forms with mutation in the files of Estonian dialect vocabulary of the Institute of Language and Literature of the Academy of Sciences of the Estonian SSR. Early Estonian literature (17th century) offers interesting material with regard to grade alternation in words with a short first syllable, especially when compared with the types described previously. It is not unusual that in 17th century literature (and later, too) the wrong grade forms were used by non-native writers, so that the weak-grade stem may appear instead of the strong grade or, more frequently, vice versa. In the case of word types where the alternating stop occurs after a long first syllable, the spellings of the weak grade forms reflect very clearly that the weak grade of the stem was already established, e.g., lauwa (laua 'table, gen.', cf. S lauda), hauwa (haua 'grave, tomb, gen.', cf. S hauda), reyat {reiad 'gash, nom. pi.', cf. S reiga), noya (nöia 'witch, gen.', cf. S nöida), Rye (riie 'cloth, nom.', cf. S hide), oyendanut

Loan Words and the Estonian Grade

423

{öiendanud 'to settle, p. act. part.', cf. S öige), kuwe (kuiie 'coat, gen.', cf. S kuube), kullane ('golden', cf. S kulda), tahab ('want, 3sg. pres.', cf. S tahta), jallakeüya (jalakäija 'walker, nom. and gen. sg.', cf. S jalga), vche and iiche (ühe 'one, gen.', cf. S iihte), and many others. But, in contrast to the situation with a long first syllable, the spellings of the alternating stop after a short first syllable do not show that the weak grade was already established. In this type very often the strong-grade stem is found instead of the weak grade: piddab ipidab for peab 'must, 3sg. pres.'), eb pidda (ep pida for pea 'must, pres. neg.'), tobbe kaas (tdbega for tövega 'disease, comit.'), soddamees (södamees for söjamees 'soldier'), keddest, keddel (kädest, kädel for käest, käel 'hand, elat. and adess.'), wedde (vede for vee 'water, gen.'), and many other examples. Some spellings enable us to conclude that the weak-grade stem was considered to have two syllables, but nothing very definite can be said about the pronunciation of such weak-grade forms as kehdet and kehjel (käed and käel 'hand, nom. pi. and adess. sg.'), wehje (vee 'water, gen.'), errakawub (ära kaob 'to get lost, 3sg. pres.', cf. S kaduma), etc. Such spellings are so numerous in 17th century texts, e.g., those by G. Müller (in Saareste and Cederberg, 1925-31) and Stahl (1637) that it is impossible to regard them as a reflection of insufficient knowledge of the language. (In the word types where the alternating stop occurs after a long first syllable this explanation would be quite possible, since the cases of grade mistakes there are relatively rare and unsystematic.) 8.

For these reasons it is appropriate to suppose that at the beginning of the 17th century grade alternation and the weak grade were not yet fully established in the word type where the alternating stop occurs after a short first syllable. Potentially alternating words probably were used occasionally without grade alternation, or, at least, there probably were pronunciations with a syllable boundary in the weak-grade stems (in place of an alternating stop or s), so that the weak-grade stems were perceived as disyllabic forms. Under this interpretation it is understandable that there were such spellings as wehje and waee (both W väe 'force, gen.', cf. S vägi), wehje and wedde (both W vee 'water, gen.', cf. S vesi) and the very frequent forms kehje and kedde (both W käe 'hand, gen.', cf. S käsi) which occur side by side with rare monosyllabic weak-grade forms such as kehs, kehst, kehl (käes, käest, käel 'hand, iness., elat., adess.'), waee (W väe 'force, gen.', cf. S vägi), and others. The files of early Estonian vocabulary and word usage of the Department of Estonian Language at Tartu State University show more than 80 kehje-forms (W käe), numerous kehdet-forms (W nom. pi.), a dozen or so kedde-forms (W gen.), and some isolated monosyllabic weak-grade spellings, such as kehl, kehs, kehst, kehdta {käel, käes, käest, käeta 'hand, adess., iness., elat., abess.'), and nom. pi. forms kehdt (käed).

424

Mati Hint

It is not surprising that in vesi: vee ('water, nom. and gen.') the monosyllabic weak-grade stem established itself more easily and earlier than in käsi: Κ äe, although both these words belong to the same archaic morphophonemic and morphological subtype. In vesi : ν ee the loss of the alternating s resulted in the formation of a long vowel in the weak-grade stem, while in käsi : kx äe the weak-grade stem was characterized by a so-called late diphthong äe, whose pronunciation as a syllable nucleus was unfamiliar. The prevalence of the monosyllabic weak-grade forms weh, wehs, wehst (vee, vees, veest - gen., iness., elat.) compared to the disyllabic forms wehje and wedde (both genitives) is evident from the early Estonian vocabulary material, although the forms of this word are not as frequent as käsi: k"äe. The spelling of postpositions in the 17th century constitutes the most convincing evidence for the assertion that at that time the weak grade was not fully developed to its modern form in the word type with a short first syllable. In the case of postpositions there is no paradigm which could influence the choice of strong- or weak-grade variants. It is very striking that the monosyllabic (weak) postposition forms ees 'in front o f , eest 'from in front o f , eel (spelled eel and eht) 'before' are used by Müller only 22 times according to the early Estonian vocabulary data, while the strong-grade (or, at least, disyllabic) stem variants eddes and eddest appear about 300 times. It is true, though, that the (historical) comparative eemale (or eemalle) < *edem(m)alle 'into the distance' is represented only by the weak stem form (25 times), which shows once again that the weak grade established itself in different words and different forms at differing rates. Approximately the same picture can be seen when observing the weak- and strong-grade stem variants of the postpositions sees (W), siddes (S) 'in, inside' and seest (W), siddest (S) 'from'. In modern Estonian the strong-grade forms of these postpositions have been lost completely, and remnants can be found only in the illative-form postpositions e^tte 'forward, ahead' and sfsse 'into'. In this context it is appropriate to recall the observations made by some historians of Estonian who have dealt with the evolution of grade alternation. Kettunen (1962:60-61) presents only those examples of irregular strong-grade spellings in the 17th century where the alternating stop occurs after a short first syllable: wägest (väest 'force, elat.'), Häteggust (hääteost 'good deed, elat.'), näghete (näete 'you see'), mäghest (mäest 'hill, elat.'). Kask (1970:37ff.) explains the unstable spelling of weak-grade forms as the influence of older church literature (where the strong grade was often used instead of the weak) and as interference from the most conservative dialectal usage (in the Northern Coastal dialect the strong grade occurs more widely than in other dialects). It is clear that there can be varying opinions about the time of the ultimate formation of the weak grade. Hinderling (1981:131, 133) considers that the long vowel or diphthong became established in weak-grade forms in the 15th

Loan Words and the Estonian Grade

425

century. Kettunen (1962:61) writes that in some dialects the weak grade had not yet taken its contemporary form even as late as the beginning of the 17th century. The present analysis confirms the appropriateness of the weak-grade spellings in the 17th century and the participation of Low and High German loan words in grade alternation. Alternating stop or s after a long first syllable

Alternating stop or 5 after a short first syllable

I. The weak grade in most types has the contemporary form in early sources.

I. In early sources the weak grade is very often spelled as a disyllabic stem or the strong grade is used instead of the weak grade. II. Low and High G e r m a n loan words have no grade alternation in the modern language (the 17th century data are very scanty).

II. Low G e r m a n and early High G e r m a n loan words mostly have grade alternation both in 17th century literature and in the modern language.

(vahti: vahi, va!" ski: val'si, mol'di: mol'li, k'uube: kuue, etc.). 9. It seems very likely that grade alternation (consonant mutation) was generalized in various phonetic environments at different rates. Descriptive or pejorative words with a short first syllable apparently avoided the alternating inflection type. In addition, there are serious phonotactic problems connected with the formation of weak-grade stem variants in this phonetic type: the loss of the alternating stop in many cases brought about the juxtaposition of two vowels which did not fit into the previous system of diphthongs. As a result, the whole diphthong system was reorganized by adding a new morphophonemic rule: when the stop is lost (in grade alternation), the [+high] vowels adjacent to the lost stop become [-high] in the weak grade {tigu : feo 'snail, slug',/?wg/:p*öe 'shearing')· Although these vowel alternations in some dialects may have a less general form or may be later developments, they nevertheless indicate that there were serious complications connected with the establishment of the weak grade in this phonetic type. Viitso (1962:59) hypothesizes that grade alternation began in the word type where the stop occurred after a long first syllable, and spread thereafter to single stops after a short first syllable. Low German and early High German loan words allow us to assert that this tendency toward generalization . was not fully realized: before the goal was reached grade alternation began to be lost in the word type with a short first syllable. Today this alternation type is losing more and more words as some archaisms go out of use and some words are transferred to an inflectional type without grade alternation (e.g., kubu : k"oo 'armful' — kubu : kubu, and many others, cf. Hint, 1981:256ff.). In the

426

Mati Hint

contemporary language the majority of words with the syllable structure (C)VKV (where Κ is a stop) belong to the nonalternating type and this predominance can only increase, although grade alternation will without any doubt maintain its position in frequently-used words. 10.

Another Low German loan word type whose relation to grade alternation (consonant mutation) is problematical has the structure ( C X C y V ^ K i , where Wx\x represents a long vowel, Κ is an alternating stop and -i is the stem vowel. In early Estonian literature Low German loan words of this type are for the most part inflected without consonant mutation: ρ f aadi MLG brade 'roasted meat', pr aadima MLG braden 'to roast', vaagi M L G wage 'weigh', ν aagima MLG wegen 'to weigh', Γ aadi MLG (büssen)läde 'gunstock', Γ aadima G laden 'to load', Γ oodi MLG lode, lot '[unit of measure]', Γ öödi MLG lode 'weigh, bullet', s'aagi MLG sage 'saw (n.)', s aagima MLG sagen 'to saw', r aadi MLG rät 'town council', p^oodi MLG bode 'shop', m^oodi G mode 'fashion, way', Iceedi MLG kede, keden 'necklace', s~iidi MLG side, siden 'silk', viigi MLG vige 'fig'. Strong-grade forms like pohdi (gen.) mees (= poemees) 'shopkeeper', kehdit (= keed) 'necklace, nom. pi.', saagi (gen.)purro (= saepuru) 'sawdust', wahgki-kodda (= vaekoda) 'weigh hous e\pradi-warras (= praevarras) 'roasting spit', etc., are very common in the 17th century from Müller and Stahl to Göseken (1660). The occurrence of relatively few weakgrade forms shows that this type was intermediate between the two types treated above with regard to grade alternation. In this type grade alternation was not automatic, but was nevertheless productive and it may have spread only slowly in these words, as the weak-stem spellings confirm: Rae-Issand (cf. S f aadi) 'member of the town council', vllewayb (= üle vaeb, cf. S ν aagima) 'to weigh, 3sg. pres.' erra saijn (— ära saen, cf. S s aagima) 'to saw, lsg pres.'. This loan-word group is divided into two inflectional types in the contemporary literary language: 1) in the standard literary language, and in most of Northern Estonia, words with a [-high] long vowel in the first syllable are inflected with consonant mutation (s^aagi: s^ae,p'oodi : p* oe, pr aadima : prae, Γ aadima : Γ ae, etc.), and 2) words with a [+high] long vowel in the first syllable are inflected without consonant mutation, but with late quantity alternation, e.g., s"iid{nom. sg. in Q3): siidi(gtn. sg. in Q2): s"iidi(part. sg. in Q3). In one word of the first group the weak-grade form is generalized and the word is inflected without stem alternation: free (nom.): k*ee (gen.), instead of k"eed : k*ee. In the South Estonian dialect area both subtypes are inflected without consonant mutation (but with alternation of geminate stops or prosodic alternation). The question here is to determine when the first subtype acquired conso-

Loan Words and the Estonian Grade

427

nant mutation in Northern Estonia and what the motivation for this change was. There were only a few inherited words or early borrowings with mutation in this phonetic type, with -u or -a, and not -i, predominating as the thematic vowel (e.g., roogu : roo 'reed, nom. and gen.' or rooga: roa 'food, nom. and gen.'). Moreover, the generation of the weak-grade stem required a rather complicated set of ordered rules: 1) loss of the alternating stop: p'oodi — ρ ooi, s aagi —• s"aav, 2) lowering of the stem vowel -/" to -e\ p" ooe, s*aae\ 3) formation of a diphthong: p*oe, s"ae. Although all these rules already existed in the grammar, they were not used in exactly this combination in other inflectional types. Grade alternation in this type was in a sense a generalization of the mutating type of inflection, which had to overcome several obstacles, but nevertheless spread to these words, at least for a time in the main Northern Estonian dialect area. The weak-grade stems sae, poe, prae, etc., were probably perceived as disyllabic for a long time, and traces of such syllable counting in Estonian poetry can be found in the case of late diphthongs even up to the present time. In addition to these factors the phonetic structure of the loan words cited above is not conducive to consonant mutation in another special respect - the stem vowel -i makes it impossible to generate the usual type of weak-grade stem produced by substituting the semivowel [j] for the lost stop (since the sequence *-ji is phonotactically restricted). Normally the semivowels [j] and [w] occur in the environments

and

respectively. While insertion of a semivowel at the onset of the second syllable is very common in gradation (e.g., fiidu: riiu (= [riju]) 'quarrel', Γ uuda: luua (= [luwa]) 'broom'), in the case of the weak-grade forms from s" aagi,pr aadi, p"oodi, rnoodi, etc., this process is doubly blocked: the [-high] long vowel in the first syllable and the stem vowel -i in the second syllable do not constitute a suitable environment for [-j-]. In the case of s'iidi, viigi, reidi G Reede 'roads' [-j-] is blocked only by the -i in the second syllable. 11.

When did the phonotactic restriction on the sequence *-ji arise in Estonian? There is no clear answer to this question. Most likely the restriction is of later origin than the establishment of grade alternation (consonant mutation) in the loan words under study. Such a chronology would explain the generaliza-

428

Mali Hint

tion of the weak grade in the Low German loan word fceedi: k"ee 'necklace' — fc'ee : Icee (without grade alternation) and the same phenomenon in some inherited words {s"üüdi : s^üü-* s*üü : s*üü 'guilt', rüüdi: rüü — rüü : rüü 'cloth', Γeesi: Γ ee — Γ ee : ΐ ee 'hearth'). This explanation assumes the initial generalization of consonant mutation at least after long high- and mid-front vowels: Iteedi: *[keji] and s^üüdi: *[süji], like modern fiidu : riiu (= [riju]) 'quarrel' or riie ( = [rije]) : riide 'cloth, nom. and gen.'). When *-ji became phonotactically impossible, the sequence was simply dropped and the weak grade genitive stem was generalized in order to eliminate the irregular consonant mutation type k'eed : k*ee : k"eedi (nom., gen., part, sg.) which would otherwise have resulted. In this way kee, lee, rüü, süü were transferred to the nongrading inflection type; only the idioms s~üüdi olema 'to be guilty' and s'üüdi möistma 'to condemn' preserve the strong-grade stem. The other way of avoiding a phonotactically impermissible sound sequence in the derived weak-grade forms would have been, of course, the generalization of the strong grade stem: the old Germanic loan word faud(i) 'disease' and the inherited word haug(i) 'pike' (a substratum word, according to Ariste, 1981:16) have lost consonant mutation in order to avoid the disallowed sequence *-wi in the regular weak-grade forms * tauwi and *hauwi. Both these words have generalized the strong-grade stem and have moved over to the inflection type with quantity alternation, e.g., faud (nom.sg. in Q3) : taudi (gen. sg. in Q2): f audi (part. sg. in Q3), in the major dialects. It then comes as no surprise that the South Estonian e-stem variant h"auge : [hauwe] has retained consonant mutation in this word. Wiedemann's dictionary (1869), alongside phonotactically irreproachable genitives for these words, offers alternate weak-grade forms with mutation: tauu (= [taüwu]), taui (= [tauwi]) and haui ( = [haüwi]), which all violate the late phonotactic restrictions prohibiting *-wu and *-wi. In this century the weak-grade genitive forms taui, tavvi, tavi are attested only in the extremely conservative Northern Coastal dialect. It is likely that in some central dialects later Low German loan words with long -/- or - Vi- in the first syllable never underwent these processes (acquisition of consonant mutation and then its subsequent loss for phonotactic reasons), and either were always inflected without consonant mutation or switched to the non-mutating type very soon after borrowing, as the restriction on *-ji became established. There are but isolated cases of weak-grade forms like wijipuhst 'fig tree, elat.', Wiji-MarjaPuhd'fig-berry trees, nom. pi.', Wii Puud 'fig trees' in 17th century ecclesiastical literature (mostly at the end of the century in literature from southern Estonia, e.g., in the New Testament of 1686). The dialectal vocabulary files have no such forms. There are hardly any weak-grade mutational forms like siiji or sii (W from s"iidi) in early sources, and only the most peripheral Estonian dialects have lent support to the theoretical possibility of these stem variants, e.g., si in Hargla Möniste

Loan Words and the Estonian Grade

429

(Southeast Estonia), zii in Leivu (Northern Latvia), elat. si, ist in Kadrina Vöhma (Northern Coastal region). Wiedemann's dictionary (1869) gives the weak-grade genitive variants sii and vli or vlji, but marks viji as South Estonian. These data allow us to conclude that the restriction on *-ji began in Central Estonia and had not yet spread over the whole language area before s^iid(i) and viig(i) were borrowed from Low German. In the modern language, too, this restriction is observed most strictly in Central Estonia and in the literary language. 12.

When the restriction on *-ji became established, words with -ji- in the second syllable of the weak-grade stem had to change their morphophonemic and morphological type, while new borrowings went to the inflectional type without consonant mutation (and thus without *-ji in the second syllable). The other way of satisfying this restriction would have been to change the stem vowel, but for one or another reason this alternative was not employed. There is a certain systematicity in this restructuring. The majority of words with a long high- or mid-front vowel or with a Vi diphthong in the first syllable generalized the weak-grade form, e.g., riiiidi : rüü — rüü : rüü, s'üüdi: s*üü — s'üü: s'üü, Γeesi: Tee — Tee: Tee, fceedi: Wee (from MLG)-~ kxee : k'ee, fäidima : fäi (from *täiji) — fäima : fäi 'not begrudge, inf. and imp.', and some less familiar words. The variants faigliha and failiha 'lean meat' exist in almost the entire Estonian language area. The stem taig- is used in the strong grade with two different stem vowels - -a and -i (part. sg. faiga or faigi)\ the weak-grade variant from faigi is regularly tai- (< *taiji). The old Germanic loan word faigen, 'dough', which has a variant tainas (gen. faina) without -g- in the stem, seems to be a special case of *-_/'/; this form is probably a reflection of an /-stem weak grade (cf. Finnish taikina). In this first subtype only s'iidfi), ν iig(i) and reid(i) have avoided this regular development (together with later loan words, e.g., s*eibi 'washer'). Words with a long low- or mid-back vowel in the first syllable, on the other hand, preserved grade alternation by losing the semivowel at the onset of the second syllable and lowering the stem vowel -i to -e in the weak-grade forms. The majority of Low German loan words was adapted in this way: Γaadi: Γae, Γaadima : Tae,T oodi: Γ oe, moodi: moe,poodi: p~oe,praadi: prae, praadima : prae, f aadi : f ae, s^aagi : sae, s'aagima : s ae, vaagi: v'ae, vaagima : vae\ as indicated above, the weak-grade stems may have retained disyllabic pronunciation for a long time. One part of this second subtype has lost consonant mutation in the literary language rather recently: Γ aadi, Γ oodi, Γ aadima (in one meaning) are no longer inflected according to the mutating type; as mentioned earlier, in some dialects the majority of these words has always been non-mutating. It is interesting to note that these two

430

Mati Hint

methods of dealing with the restriction on *-ji are entirely complementary in terms of the vocalic nuclei of the first syllable. The development of consonant gradation in the Low German loan word type s'aagi: s ae,pr aadima : prae,p'oodi: p*oe,etc., is in a way paradoxical. The avoidance of the weak grade would be much more expected, since the hypothetical pronunciations *[säji] (W from s^aagi), *[pöji] (W fromp'oodi), *[präjin] (W f r o m p r a a d i m a ) should have been to some extent contrary to the phonotactic innovation which restricted *-ji soon after the borrowing of these words. Of course, it is possible that the restriction on *-ji began in the narrower environment /ii — / and /Vi — / (as in s"iidi, viigi, reidi) and was later generalized. And it is also possible to suppose that the [j] was never clearly pronounced in these forms at the onset of the second syllable after long back vowels. The Low German dialectal pronunciations krage (= [kräje]) and sage ( = [säje]), as well as fige ( = [fije]), yielded Estonian kfae 'collar' and sae (dialectically without consonant mutation). It is difficult to imagine how these pseudo-weak-grade pronunciations could have given rise to the mutational morphophonemic pattern s aagi : [säje], and how this pattern could have been generalized into the numerous words with -dipfaadi: prae, etc.). The early literary language does not give clear answers to these questions. Hinderling (1981:113, 141) has stressed that the spellings Krdje and Saje by Gutslaff (1648) (as cited by Hinderling, 1981) and Kraije, Kraij, Saij by Göseken (1660) mark disyllabic pronunciation. Nevertheless, it is not certain that these words were really pronounced with -ji or -je in the second syllable in Estonian; they were more likely pronounced without -j- at the onset of the second syllbale, on the order of [säi] «=» [säe] and [kräi] [kräe]. Stahl (1637) uses the spellings sai, saist (elat.) and kraj, kräjst (elat.); it would be rather strained to interpret these spellings as referring to a pronunciation with ji in the second syllable. But if these and the other forms discussed above really were pronounced with ji in the second syllable, then the restriction on *-ji was without any doubt an important factor in pushing the weak-grade forms into monosyllabicity (säji -* säe —· s"ae, and, in the same way,p"oe,prae, etc.). The morphophonemic pattern of Low German loan words presents several problems which are central both to modern Estonian and to its history. It is clear that some consonant mutation patterns were no longer productive during the period of borrowing from Low German; for example, there is no attested case of mutation of s in these borrowings, cf., on the one hand, kruusi MLG krüs 'cup', reisi MLG reise 'journey' and many others with non-mutating -s-, and, on the other hand, the old Baltic loan word reis: reie ( = [reije]) 'thigh' or the native uus : uue ( = [üwe]) 'new'). The development of consonant mutation in some loan word types is quite obscure. A more thorough investigation of the older Estonian literary language (17th-19th centuries) may give fuller answers to the problems raised in this paper. But

Loan Words and the Estonian

Grade

431

some problems may still remain as insoluble as the definitive analysis of a substratum language on the basis of the superstratum (cf. Lehiste, 1965). ACKNOWLEDGEMENTS

The author wishes to express his thanks to Professor Paul Ariste for his kindness in looking over this paper and discussing some problems which relate to his long-standing field of interest. NOTES The author wishes to express his thanks to Professor Paul Ariste for his kindness in looking over this paper and discussing some problems which relate to his long-standing field of interest. 1. A note on transcription Where greater precision is not needed ordinary orthography or the orthography of Estonian standard normative dictionaries is used; in the latter case an acute accent on a consonant denotes palatalization and a grave accent marks overlength of the following consonant, vowel or diphthong. Square brackets denote standard Finno-Ugric phonetic transcription (in this transcription " denotes a short s o u n d , ' on a consonant denotes overlength, long vowels are transcribed as V, and overlong vowels as V). The older orthographic J in the examples f r o m early sources and in G e r m a n words is replaced by s. 2. Loan words have been drawn f r o m the studies by Ariste (1963; 1981), Hinderling (1981), Liin (1964; 1968) and Uesson (1970). The forms and spellings of early Estonian literature are cited as given in the studies by Ariste and Liin, in the reader by Saareste and Cederberg (1925-1931), and, principally, in the Tartu State University files of early Estonian vocabulary. REFERENCES Ariste, P. (1963). Saksa laensönad Heinrich Stahli eesti keeles. Emakeele Seltsi Aastaraamat, 9, 85-119. Ariste, P. (1981). Keelekontaktid. Eesti keele kontakte leiste keeltega. (= Eesti NSV Teaduste Akadeemia Emakeele Seltsi Toimetised, 14). Tallinn: Valgus. Göseken, Η. (1660). Manuductio ad Linguam Oesthonicam: Anführung zur Ohstnischen Sprache. Reval. Gedruckt und verlegt von Adolph Simon. Photoreprint: Fenno-Ugrica, Band 3. Hamburg: Helmut Buske Verlag. 1977. Gutslaff, J . (1648). Observationes Grammaticae circa Linguam Esthonicam. D o r p a t . Exudebat Johannes Vogel, Acad. Typogr. Photoreprint: Fenno-Ugrica, Band 2. Hamburg: Helmut Buske Verlag. 1976. Hinderling, R. (1981). Die deutsch-estnischen Lehnwortbeziehungen im Rahmen einer europäischen Lehnwortgeographie. Wiesbaden: Otto Harrassowitz. Hint, Μ. (1971). The Semi-Vowels [j] and [w] in the Phonological and Morphophonological Systems of Estonian, I and II. Soviet Finno-Ugric Studies, 7, 73-85 and 169-186. Hint, M. (1981). Neodnorodnost' sistem ceredovanija stupenej estonskogo jazyka. Summary: Non-homogeneity of the Estonian Grade Alternation Systems. Soviet Finno-Ugric Studies, 17, 247-265. Kask, A. (1970). Eesti kirjakeele ajaloost, I. Tartu: T a r t u Riiklik Ülikool.

432

Mati Hint

Kettunen, L. (1962). Eestin kielen äännehistoria. Kolmas painos. (= Suomalaisen Kirjallisuuden Seuran Toimituksia, 156.) Helsinki. Lehiste, I. (1965). A Poem in Halbdeutsch and Some Questions Concerning Substratum. Word, 21, 55-69. Liin, H. (1964). Alamsaksa laensönad eesti vanimas kirjakeeles. Tartu Riikliku Ülikooli Toimetised, 162, 32-74. Liin, H. (1968). Alamsaksa laensönad 16. ja 17. sajandi eesti kirjakeeles. Unpublished dissertation. Manuscript in the Estonian Dialect Sector, Institute of Language and Literature of the Academy of Sciences of the Estonian SSR. Saareste, A. and Cederberg, A.R. (1925-31). Valik eesti kirjakeele vanemaid mälestisi a. 1524-1739. (= Akadeemilise Emakeele Seltsi Kirjastus, 16) Tartu. Stahl, Η. [= Stahlen, Henrico] (1637). Anführung zu der Estnischen Sprache. Revall. (Photoreprint, Maarjamaa, Rome, 1974). Uesson, A-M. (1970). On Linguistic Affinity. The Indo-Uralic Problem. Malmö: Eesti Post. Wiedemann, F.J. (1869, 1893). Ehstnisch-deutsches Wörterbuch. St. Petersburg. (Photoreprint, Tallinn, 1973: Valgus). Viitso, T-R. (1962). Tüvelisest astmevaheldusest (eriti eesti keeles). Emakeele Seltsi Aastaraamat, 8, 44-62.

The Accentuation of the Serbocroatian Dialect of Uljma Pavle Ivic, Academy of Sciences, Belgrade,

Yugoslavia

In 1972 and 1973 I investigated the local dialect of Uljma in Southeastern Banat for the General Slavic Linguistic Atlas (OLA). 1 The prosodic pattern of the dialect proved to be very peculiar. It comprises the typical four Stokavian "accents" 2 ("long falling", written", "short falling" "long rising"' and "short rising" "), but the distribution of those entities is different from that in any other dialect known so far. It is remarkable that this is the only Serbocroatian prosodic pattern which combines a full inventory of four ν

Stokavian accents with a complete lack of quantity contrasts in unaccented syllables. I shall try here to give an outline of the genesis and the characteristics of that pattern. The dialect of Uljma belongs to those peripheral Stokavian dialects in which the Neostokavian accent retraction took place only partially. 1. A short accent was shifted from the ultima to the penultima. The result of retraction was a long rising accent in all cases, regardless of the original quantity of the formerly pretonic vowel: (a) od bica < od bicä, boba G s g , j i c m a , krsta Gsg, vipru, volu, sela Gsg, tele, da ska, deca, kosa, meda, sestru,po zemni, dobra,jeno < jednö, moji, one pers. pro., svega, zovi, doslo, prosli, rekla; reseta, brzinu, cistoca, dubina, gospoda, krtina, ladnoca, lob0da,po[äne, putdna, vrucina, sarine, zeleni,pecino, izaslo, nazderi se\ dovedena, povedeni; bostan, bozic, cekrk, covek, istok, jecam, kdzan, kiipus, momak, o%ak, sirket 'vinegar', secer, tocak, iiskrs, poslom, brkat, diibok, visok; arpd%ik, kukuruz, dugäcak. (b) gnizdo (< gnezdo ),jme, krilo, mleko, svince, usta, cerku, jila, ρύζα f., räna 'food', sfne, ζύήα, slano, jemo, nisu, bile·, imäne, bu^ova 'Sambucus nigra', kericu, cupämo, idemo, stojite, zapisi, poräslv, krmacice, [ubicica, sangarepa', Karasovanr, güsak, mäcak, pisak; ma^ärac 'a dance', Pancevac, sumärak. In a small group of examples belonging to type (a) the short accent still appears on a closed ultima: danäs, onomäd, gotöv, devedez [gödina]. The explanations seem to be clear: danäs was influenced by the parallel f o r m danäske, onomäd is a shortened form of onomädne, and the final accent in gotöv is supported by the paradigm (gotöva, gotövo, etc.); the cardinal numbers in des (-deζ) are shortened alternates, and the full varieties in -deset still exist in the dialect.

434

Pavle Ivic

2. Retraction also affected all other examples where an accent, short or long, was preceded by a long vowel: (c) grideti (< gredefi), kosovi, piitovi, kdjsije, lenstina, kdkoce, birali, cipala, lizali, sädile\ rdskrsnica, sonicama·, ozdribila, podavila, ροάέΐΐο (cf. also obrno, posrädo, raspäro < obrnuo, posträdao, raspärao)·, (d) iz gläve (< iz glave), väsi Gpl, be züba, päse,3 rästu, znedu; dugova, meseci, sinova, zecova, gutddu, najidu se, preglidi, udäju, zijädw, (e) övcom (< ovcöm), celog, rästem, sicem, vucem; ( f ) pütova (< putöva), bräzctene, gledite, plevimo, vucete', sa zalivanom\ izorävane. 2a. Corresponding retraction to a preceding short syllable did not take place: (g) vinögrad, miseve, oräse, s ovnövi, zublma, kopine « kupine), kopita f., siklra, strnika, decäma, od oviju, starii, ublse, posejo, doveo, izeo, nosili, opleo, sedeo, voleo, oräces, prepelica, rastövina; jareblca, majstorica, bogacii compar.; pozelenelo·, (h) ajgir, bu$ak, petlic, repüv, rukom, ovim Dsgm .Jedän, doim, idem, sedim, zoves\ razumem\ (i) opänce Apl, bagrene, sibärje 'bushes', devöjka, igränke, kokoske, nosila, izäite, ujeda, uftdrnik; otidnemo, vospitano; saraniva; ponedelnika, otkravlivamo, preselävadu. 3. There is one exception to the rule in 2a. The accent was shifted even to a short penultima, where a short rising accent appeared, if the final vowel was long: (j) kose Gsg ( < kose), iz vdde, mene, sebe (cf. mene, sebe in various dialects), cega, kakve, moje Gsg, täko, bole 3 sg. presl, cveta, cita, doi, drzu 3 pi. pres., plete, stöi, stoju, zove, peske\ onäko, oväko, marvena, donese, dovede,prenocu. Cf. also smeju-se but sijä-se, tako-je, all with enclitics, and the typeposo < posö < posäo, doso < dosö < dosao, pöso, prbso, otiso, contrasting with imäo, lagäo, ustäo, which allows us to conclude that the sequence -do was contracted to -δ only in forms where ä was the reflex of a strong jer, i.e., where it did not have support from the other forms in the paradigm (cf. posao, posla, poslu etc., dosäo, dosla, doslo etc., vs. imäo, imala, imalo etc.). This particular retraction of the accent from a long final vowel is the most striking peculiarity of the dialect of Uljma, especially because it created a short rising accent, which otherwise does not appear in the accentual pattern. However, this phenomenon cannot be separated from the shortening of a final long vowel in open monosyllables: (k) vo ( < vo),jä, fi, od ne, o(d) te zene, dvä, dve, tri, pre. The change under (k) must also have affected a long accent on the open ultima of polysyllabic words (kose > kose), thus paving the way for the retraction mentioned in 1 (kose > kose). The relative chronology was undoubtedly: 1) (a) and (b), 2) (c)-(f), 3) (k), and also in polysyllabic words, 4) (j)·

Accentuation in a Serbocroatian

435

It might seem tempting to put (k) before (a) and (b), reducing the number of processes (the change under 4 would then become a subtype of 1). However, such a view is unacceptable for two reasons: i) The result of the change kose > kose > kose is a short rising accent, whereas in type (a) we have kosä > kosa, with a long rising. If the retraction of the final accent in kosä and kose had taken place simultaneously, the results would be identical (thus either kösa, kose or kosa, kose). ii) Changes of types (a) and (b) are extremely widespread in Serbocroatian dialects, while the shortening of a long accent on word final vowels has a much more restricted domain. In Stokavian it embraces only several dialects closely related to that of Uljma, such as those of Alibunar in Southeastern Banat (Ivic, 1957:36), Landol (field notes), Kusadak(field notes), and Nemenikuce (Remetic, 1981:490) in the neighboring part of Northern Serbia, as well as the dialect of the so-called Gallipoli Serbians (Ivic, 1957:33-34), who emigrated from Northern Serbia to Turkish Thracia about four centuries ago. In a number of the dialects involved the process is not yet complete, e.g., in Landol we find zove, zne, ot kose, trci, razumu, but also zove,juce, gubi, etc., and in Kusadak dve, zivine, do zemne, ovolikl alongside dve, ide, icupä, etc. Thus the shortening of a long accent on a final vowel is to be considered a relatively recent process, and the Uljma retraction of the short accent generated by that process must be even more recent. Nevertheless, the changes under (a), (b), and (j) are basically similar. Their common impetus is the tendency to eliminate short accents from the ultima, which is characteristic of so many South Slavic dialects - Slovenian, Kajkavian, Cakavian, peripheral Stokavian, Macedonian, and even Bulgarian. The old short accent on the ultima was eliminated comparatively early (kosä > kosa), and a secondary short vowel in an open ultima also underwent a shift, but without lengthening of the penultima (kose > kose). Thus the distributional rule about the inadmissibility of a final short accent has been restored. An additional argument for a recent date of the shift in type (j) is furnished by the fact that Turkish loanwords such as bostan, cikrk, käzan, sirket or secer have a long rising accent, which suggests that they were borrowed in (or even before) the period of the kosä > kosa retraction rather than in the period of the shift found in (kose >) kose > kose. However, we cannot exclude the possibility of influence from the accentuation of Slavic words such as potok, Gsg potoka upon the accent of borrowings such as bostan, Gsg bostäna. A more complicated problem of relative chronology is the one concerned with the changes which have affected our types (d)-(f)· Like all other dialects of the Smederevo-Vrsac and Kosovo-Resava areas, the dialect of Uljma has abolished quantity contrasts in postaccentual position: (1) kräcä4 (< kräcä), bega, Hitnö, rzö, veliki, läde Gsg, usäkä, star ι pi m., saramvä; kürjäk, rädös, sä mnöm, po stärlm; gölübi, küpimo, üdäta, precvetälo, üzoräli, molded.

436

Pavle Ivic

In these examples shortening occurred in a position which was postaccentual throughout the history of Serbocroatian. However, in our material under (d)-(f) an unaccented short vowel now stands in the place of an originally accented long one. Three relative chronologies of the events leading to this situation can be imagined: (A) The accent shift took place first ([glave > gldve) and later the final long vowel was shortened more or less simultaneously with the process under (1), thus gldve > gldve as well as läde> lade\ (Β) The retraction {gläve > glave) occurred at a time when the final long vowel in type (1) had already been shortened (lä lääe), so that quantity contrasts in postaccentual vowels were reintroduced into the system; only later were they abolished again (gldve > gldve)', (C) The shortening took place while still under the accent {glave > glave, pütöva > pütö να, etc.) as a result of a tendency to avoid sequences of two long syllables; the retraction followed the shortening {gläve > gldve, pütöva > pütöva). None of these possibilities can be excluded. Perhaps the least probable is (A). The elimination of postaccentual length in examples of type (1) embraces a huge area stretching from the Albanian to the Rumanian border. In none of the dialects concerned can we find either exceptions to the rule or traces of an earlier situation. This speaks in favor of an early date for the shortening. As for the accent shift to a preceding long vowel, which occurs in various places in the area mentioned above, it seems to be a more recent phenomenon, in many cases only inconsistently realized {gldve alongside with older gläve, gride[i alongside with grede[i, etc.). Finally, the possibility described under (C) appears to be the most probable. In other dialects of Southeastern Banat, such as those of Kruscica and Vracev Gaj, a long falling accent has been shortened when preceded by a long vowel: Kc. Bogojäv\ene > Bogojävfene, pitäne >pitäne, etc., YG. prices ci väne >pricesciväne, etc. (Ivic, 1958:328). In the examples given, which belong to our type (f), the first phase of the development supposed in hypothesis (C) has taken place, while the second phase did not occur. However, in examples such as Kc. sinova < sinövä, VG. sa Vdsom < sa Väsom (our types (d) and (e)), we find the same situation as in Uljma. It is very likely that here the chain of events was sinövä > sinövä > sinova, which corresponds exactly to hypothesis (C). Notice that in these dialects, too, an original short accent has been shifted from the ultima to the preceding syllable: mlekö > mliko, svetäk > svetak. The present state of the prosodic pattern of the dialect of Uljma can be described as follows: I. Three features are distinctive: the place of the accent, quantity and tone. II. The accent can appear in any position except on the open ultima of polysyllabic words.

Accentuation in a Serbocroatian

437

III.

Quantity is distinctive only in accented vowels (but it is neutralized in open monosyllables, which always have short vowels, and the distinctive role of quantity is small in the closed ultima of polysyllabic words, since shortness is exceptional in that position). IV. Tone, too, is contrastive only under accent. Tone contrasts do not occur in the ultima; this precludes their occurrence in monosyllabic words. In other positions tone is distinctive under a long accent, and in the penultima also under a short accent. Once the quantity of the syllables in an accentogenic word is known, the place of the rising tone (and accent) is predictable. If the word does not contain a falling accent, a rising tone appears on the long syllable; and in words consisting only of short syllables it falls on the penultima. The distribution of the four accents is as follows: • The long falling accent occurs in all positions except on a vowel in final position in the word (i.e., it is impossible on an open ultima or on open monosyllables); • The short falling accent occurs in all positions with the exception of the open ultima of polysyllabic words; it appears only exceptionally on aclosedultimain such words; • The long rising accent can stand on non-final syllables of polysyllabic words; • The occurrence of the short risingaccent is limited to the penultima. It is obvious that the domain of the falling accents is broader than that of the rising accents, and that the long accents can stand in more positions than the short ones. We may conclude that the prosodic pattern of the dialect of Uljma is one of the most complex both within the Serbocroatian language area and in the Slavic linguistic world in general. NOTES

1. 2.

3.

4.

A short sketch of the phonological pattern of the dialect of Uljma is given in Ivic (1981). The term "accent" rather than "stress" is used throughout this paper since investigations have shown that tone and duration play a more prominent role than intensity as physical correlates of Serbocroatian "accent," cf., e.g., Lehiste and Ivic (1963). Traditional terms such as "long rising" or "short falling" will be used here as labels for the associated phenomena, although it has been shown that the physical nature of these "accents" does not always coincide with the description implicit in the terms. In various Serbocroatian dialects the thematic vowel e in examples such as pase, rästem or vucete is short. However, it is highly probable that the shortness of that vowel in the dialect of Uljma is secondary. In all the dialects belonging to the same group the vowel in such examples is long unless regular phonetic changes have shortened all long vowels in the given position. In this paragraph the sign ~ is used to denote a short vowel that has replaced an earlier long one.

438

Pavle

Ivic

REFERENCES Ivic, P. (1957). Ο govoru Galipoljskih Srba ( = Srpski dijalektoloski zbornik, XII). Beograd: Srpska akademija nonka i umetnosti. Ivic, P. (1958). Mesto banatskog erskog govora medu srpskim dijalektima. In: Banatske Ere. Novi Sad: Matica srpska, pp. 326-353. Ivic, P. (1981). Uljma. In: Ivic et al. (eds.) (1981), pp. 491-496. Ivic, P. et al. (eds.). (1981). Fonoloiki opisi srpskohrvatskih/hrvatskosrpskih, slovenackih i makedonskih govora obuhvacenih OpSteslovenskim lingvistiikim atlasom. Sarajevo: Akademija nauka i umjetnosti Bosne i Hercegovine. Lehiste, I. and Ivic, P. (1963). Accent in Serbocroatian. Ann Arbor: University of Michigan Press. Remetic, S. (1981). Nemenikuce. In: Ivic et al. (eds.) (1981), pp. 485-490.

Vocalic Oppositions in Monosyllabic English Words Wiktor Jassem, Acoustic Phonetics Research Unit, Polish Academy of Sciences, Poznan, Poland

1. INTRODUCTION

Although there are many definitions of the phoneme and various approaches to the problem of phonological distinctiveness, almost all phonologists make use, at some stage, of minimal pairs, i.e., pairs of isolated words, in their citation forms, such that the members of a pair differ by no more than one linear element (phone, segment, etc.). Quite often, multiple oppositions are quoted. In this case, three or more minimally different words are cited. These may be considered as forming a 'commutation series' (cf., e.g., Hockett, 1964; Gimson, 1980). One kind of linguistic redundancy consists in the fact that the (semi-) open set of lexical items in the given language never exhausts all the phonotactic possibilities so that there are always considerably more potential words, i.e., strings of segmental elements that are phonotactically admissible, than there are actual lexical items, for any given length of the string. English has often been described as a language tending towards monosyllabicity, and it is common knowledge that indeed a substantial majority of the lexemes with the highest text frequency of occurrence are monosyllabic. Thus, the proportion of actual monosyllabic English words related to the total number of potential words, i.e., the phonotactically admissible strings of sequential units, is considerably higher than that of actual polysyllabic words. Commutation series including more than two minimally different words of two or more syllables (such as picket, packet, pocket) are rare whereas multiple oppositions of monosyllables are, as will be shown, quite numerous. A numerical description of both vocalic and consonantal oppositions is called for, but consonants have to be treated differently because, as syllable margins, they can form clusters, which the vowels, as monophthongal or diphthongal syllable peaks cannot. Minimal pairs or commutation series are often used for didactic purposes. Many practical handbooks of the pronunciation of English (or any other language) contain them. Segmental oppositions, then, are of both theoretical and practical interest.

440

Wiktor Jassem

It is sometimes pointed out that oppositions may have unequal load. Some of them, that is, may be illustrated by more lexical items than others. But although this is inherently a quantitative problem, it has not been elaborated in numerical terms, at least not for the phonemes of Received Pronunciation (RP). 2. T H E MATERIAL

We are here interested in one particular record of what is regarded as standard British English pronunciation, viz. the 13th edition of the English Pronouncing Dictionary (EPD) (1975) and with the particular phonological interpretation of the English vowel system which underlies the transcription used in that record. Of the several possible interpretations, this one posits the maximum number of vowel phonemes in Present-Day RP. As we are dealing with phonological forms, homophones are collapsed so that, e.g., right, rite, wright and write are entered only once, under / r a i t / . Conversely, variant pronunciations that are recorded as phonologically different are treated as separate items, not only in /maes/ - / m a : s / , mass, /lost/ - /lo:st/, lost, etc., but also, e.g., / k j u o / - /kjo:/ - /kjoo/, cure. A phonological form like /Jo:/ may represent several lexical items:pshaw, shore, shaw, and sure, whilst being one of several phonologically distinct variants of one particular lexical entity: sure, which is recorded as / J u s / , /Jo:/, /Jos/, and /Jo:/. An individual speaker may only use one of the variants, but we are here interested in the phonology of a speech community and are dealing with lexical units irrespective of semantic identities or differences. This is why we treat free morphophonological variants in the same way as distinct lexemes. Our inventory of vowel phonemes consists of the following units: /v., i, e, ae, a:, o, o:, u, u:, λ, ο:, ei, ou, ai, au, oi, ui, io, εο, oo, uo, eo/. This vowel system includes, besides the traditional 21 monophthongs and diphthongs tabulated in the successive editions of Jones (e.g., 1976:61), the items / u i / and / e o / (as in one pronunciation of ruin and payer) because they are treated in our data source as single syllable nuclei and must therefore, by implication, be considered within that particular system as phonemic units just like, e.g., / o i / and / ε ο / . Indeed, our ensemble of diphthongs is that given in the Introduction to EPD (1975:xxxv). The consonants are as follows: / p , b, t, d, k, g, tj, d3, m, η, η, 1, f, ν, Θ, 6, s, z, r, J, 3, h, w, j / . As Jones' position with respect to / t r / and / d r / is not entirely firm (Jones, 1976:165-167), we have treated them as successions of two consonant phonemes, this interpretation being preferable on distributional grounds (Gimson, 1980:166-168). When a phonological commutation series is used ad hoc for didactic or experimental purposes, it does not matter what stock of words the items are drawn from. Both inflected and uninflected forms may be included and

Vocalic Oppositions in Monosyllabic English

Words

441

proper names may appear alongside common lexical units. For a quantitative analysis, however, it becomes necessary to restrict the material according to some clearly definitive criteria. We have limited the scope of our words by applying the following restrictions: (1) Entries in E P D representing only proper names were excluded. The material was thus made more homogeneous. Proper names form a set which is more open still than ordinary lexical items, so that there is of necessity more arbitrariness in their inclusion in phonetic and phonological data. (2) Regularly inflected forms of verbs and nouns were excluded. If a language has very few inflections, it becomes doubtful whether inflected forms should appear in sets of minimal pairs or in commutation series because they create an imbalance by inordinately favoring certain structures. In our case, the number of items with final / s / , / z / , / t / , / d / would be out of proportion, and treating, e.g., / v e t s / vets exactly like / v e k s / vex seems exceptionable on structural grounds. (3) While it was necessary to include such variant pronunciations as differ by the alternation of vowels so as to exhaust the supply of possible interrelations between vowels, little purpose would have been served by taking into account variations with respect to consonants, the more so as these are more predictable. Thus, forms with / \ / and / 3 / alternating with / t j / and /&·§/ (e.g., in lunch, hinge) were not included, nor were other forms with elided middle consonants (e.g., / g l i m s / glimpse or contextually elided final consonants (e.g., / n e k s / next). Of the two variants with / w / - / h w / ( / λ λ / ) only the former was included. (4) Unstressable ('weak') forms were excluded (e.g., / a z / cj) because they should be treated as a separate morphophonemic category, and so were forms with sounds not representing any of the assumed regular phonemes of RP, as enumerated above. 3. ANALYSIS AND RESULTS

For the purposes of this study, a 'phonological word' will be a definite string of segmental elements irrespective not only of the spelling but also of its morphonological and semantic relations with other strings, whether it represents one or more lexemes, as explained above. We will refer to such strings simply as 'words.' The words were transliterated from the phonetic notation for computer-processing, and arranged into classes and subclasses. Each class represents a distinct C / V syllabic structure, and each subclass contains words differing only by the vowels, or one word if there is no vocalic opposition. The word /aekt/, act, for instance, forms a single item subclass. In Table 1, the leftmost column indicates all the syllabic structures found in our set of words. The column heads refer to the number of words in

442

Wiktor Jassem - Ν η η

ο Η

^ η r^ m m mΝ - Tt η

—

ο\ »s

ΟΟΟ r - N m o c i » i N m « m m η o o m f i T r o \ t — — > v o w - i w > 0 \ o o r - ' t f N Ν Ν

a

f

© o o v i -

5 vo

>

o r - i O m t —

ο ο τ

—

m

»

-

r—

CN v o

f i

ο

Ο vi

^

in

-

^

^

c ο

Ο ο. α. ο

cd

o ω X) E 3 ζ

-Q r a

(N Γ-

«

v o m r —

o

w

^

O

Ptg./e/to, Sp. hecho 'done, made'; LXCTE > Ptg. leite, Sp. leche 'milk'; ( 2 ) PÄCTU > Ptg. (with change of gender and ending)peita 'bribe' (ci.peitar 'to bribe', peiteiro 'briber'), Sp. pecho 'tax, tribute' (cf. pechero 'taxable', 'taxpayer, commoner'); 3 (3) LECTU > Ptg. leito 'bedstead', 'couch, berth, cot, bunk', Sp. lecho fig. 'bed' (as in lecho de rio 'river bed', lecho de roca 'bedrock', also lecho nuptial)·, PECTUS > Ptg. peito, Sp. pecho (orig. pechos) 'breast, chest'; (4) COLLECTA > Ptg. colheita 'harvest(-time), crop', Sp. cosecha, obs. and dial. cogecha\ DE-, DT-RECTU > Ptg. direito, Sp. derecho 'right-hand, straight', 'law'; (5) STRICTU > Ptg. estreito, Sp. estrecho 'narrow, tight, close'. But in addition to such very tidy correspondences, and not a few others like them, one also runs into a radically different situation, involving, as expected, / j t / in Portuguese, but - obviously learned - / k t / in at least the standard variety of Spanish, witness these examples: LECTORE > Ptg. leitor, Sp. lector 'reader' beside LECTÜRA > Ptg. leitura, Sp. lectura 'reading'; PERFECTU > Ptg. perfeito, Sp. perfecto·, RESPECTU > Ptg. respeito 'esteem', 'consideration', 'relationship', Sp. respecto 'reference, relation'. Here the vernacular transmission in one language clashes with the erudite adoption in its congener. As if this were insufficient, there arises yet another situation where vernacular Ptg. -eit-, clearly ar'rived at by word-of-mouth transmission, is opposed to 'semilearned' Sp. -t-, involving a mixed record of late adoption, then farreaching adaptation, thus: RESPECTU > Ptg. respeito, Sp. respeto 'esteem'; SUBIECTU > Ptg. sujeito, Sp. sujeto. There are still more variations on record: Lat. -ECT- may have yielded -et- in the West of the Peninsula, but -ect- in the Center, as is true of TRAIECTU > Ptg. trajeto, Sp. trayecto 'passage, course, route, way'; or may have produced -t- in both languages, cf. O B I E C T U > Ptg.

Sound Correspondences

and Levels of

Transmission

453

objeto, Sp. objeto, in which case the contrast has been neutralized, as it indeed has where -CT- has everywhere been taken over intact, as holds for aspecto. At the opposite end of the scale of possibilities, the two languages share an -eitsequence traceable to -ECT- only where a two-pronged Provencalism has asserted itself, as applies to deleite 'delight', deleitar 'to enrapture', deleitoso (rare in present-day Spanish) 'delightful, delicious', and also to more complexly structured (with prefix change) Ptg. enfeite 'ornament', enfeitar 'to bedeck' beside Sp. afeite 'make-up', afeitar 'to shave' (orig. 'to trim, embellish'), which in the last analysis go back to AFFECTU. (Ptg. azeite, Sp. aceite 'olive oil' hardly qualify as a parallel, involving as they do an Arabism.) Completely isolated is the case of TECTU > Ptg. teto, Sp. techo 'ceiling, roof, shelter'. Then again, there emerge such unusual splits as Ptg. fato as against Sp. hecho (from older /echo) 'fact, event' < FACTU. In certain instances there simply exists no dual representation, either under pressure f r o m morphological shifts, as when ELECTU 'chosen' has given rise to Ptg. eleito, while yielding ground to the innovative 'weak' participle elegido in Spanish, or, conversely, in cases like Ptg. ceitil 'farthing', fig. 'trifle, mite', which - whatever its background - lacks in Spanish a relevant lexical counterpart that might invite comparison. 4 At this point, it becomes clear that the characteristic correspondences mentioned above - add to them Ptg. asseito, Sp. acecho (from older assecho) Ί a m b u s h ' < ASSECTOR (*-o) and Ptg. proveito, Sp. provecho 'advantage, benefit' < PROFECTU - to be sure, represent the nucleus of the problem, but do not begin to exhaust the range of possible relationships between cognate forms. II

It would thus be arbitrary to limit the process under observation here to such clear-cut oppositions as Ptg. barbeito, Sp. barbecho 'land left fallow', at one extreme of a scale, and, at the other, Ptg. efeito, Sp. efecto < EFFECTU, 5 Ptg. prefeito, Sp. prefecto 'prefect, mayor, governor' < PRAEFECTU, with the unavoidable afterthought that Ptg. trecho 'space, distance, interval', in lieu of expected *treito, is best categorized as an adoption and, ultimately, an adaptation of Sp. trecho 'stretch, while' < TRACTU. 6 (A similar suspicion lingers on in the case of Sp. Ptg. bochorno 'hot, sultry air or wind' < VULTÖRNU [for the source of the -ch- see below]; in Ptg. flecha/frecha, Sp./7echa 'arrow' the ultimate source was extra-Peninsular. 7 ) There exist, to begin with, very close parallels to the course followed by PECTUS, VERVACTUS, etc., but, for a change, exhibiting, within the critical segment, a stressed vowel other than e - e.g., an ο - cf. Ptg. noite, Sp. noche 'night, evening' < NOCTE and Ptg. oito, Sp. ocho 'eight' < ÖCTÖ. (The mutual relation of Ptg. muito, Sp. mucho 'very, much' seems similar at first glance, except that here - uniquely after υ - / I t / rather than / k t / is involved, cf. the

454

Yakov Malkiel

outcome of VÜLTÜRNU, above.) But for 'eighth' the ideal bicuspid pattern ceases to work, since Ptg. oitavo and Sp. octavo < OCTAVU are in an opposition reminiscent of the conflicts of efeito, perfeito with efecto, perfecto; Old Spanish still happened to favor ochavo. The situation is further complicated by the familiar free alternation oi ~ ou widely tolerated by Portuguese since the Middle Ages (oiro ~ ouro 'gold' < AURU). Hence the rise of the by-form noute for 'night'; hence the even stronger predilection for outubro 'October', lit. 'eighth [month]', as against Sp. octubre (but OSp. ochubre), conceivably under pressure from outono 'fall' < AUTUMNU, whose ou diphthong was far more solidly moored; and hence, finally, the selection of doutor 'doctor' < DOCTÖRE, possibly with some help from doito ~ douto = OSp. du(e)cho 'agile' ( < DOCTU X DUCTU?), which, in turn, may have had its share of contacts with fouto ~ (a)foito 'bold, daring' (from FAUTUS lit. 'favored' > 'overconfident'). The / w t / variants, in all likelihood, reflect a far-reaching confusion in the Romance vernacular of Lat. -CT- and -PT-, independently suggested by such perhaps semilearned - developments as Ptg. ( > Sp.) auto 'public ceremony, kind of short play' < XCTU, Sp. cautivo 'prisoner' < CAPTTVU (over against atar 'to tie' < APTÄRE 'to fit', OSp. Ptg. cativo echoing It. cattivo - conceivably in a more rustic speech register), and, above all, independently suggested by Rum. lapte 'milk', noapte 'night', opt 'eight', piept 'chest', and the like. There is no free interchange of ei and eu in modern Portuguese on a scale even remotely comparable to oi ~ ou, but certain leaps from one camp to the other have left readily detectable traces, e.g., aceito Ί accept' < ACCEPTO, the iterative-intensive of AcciPiö, or conceito 'concept, idea', from CONCEPTU, in sharp contrast to Sp. acepto, concepto. Less deep penetration into folk speech is displayed by Ptg. exceto 'save, except' < EXCEPTU; at least the Portuguese do not emulate their neighbors' punctilio in attempting to pronounce the ρ of excepto. To return to MULTU: it has bifurcated into heavily-stressed mucho and pretonic muy, originally muyt / m u j t / - which, appearing before a consonant as it predominantly did, shows a thwarted development of / j t / , reminiscent of that of buitre, dial, butre ~ bueytre, Ptg. avutre 'vulture' < VULTURE/VOLTURE and of Nav.-Arag. cuitre 'plowshare' < CULTRU, * - R E , over against cuchillo, OSp. co-, cu-chiello 'knife' (lit. 'small plowshare') < CÜLTELLU. Since, in contrast to the stability - indeed vogue - of ei in older Portuguese, ui before long began to exhibit sporadic leaps to u (witness dial, and OPtg. chuiva 'rain' > mod. chuva), it causes no surprise to find cutelo 'cutlass, cleaver' (for the ordinary 'knife'/aca is used in Portuguese). Ptg. and Sp.fruta 'fruit' look alike at present, but for the historian the reduction o f f r u i t a to fruta along the Atlantic Coast and the replacement of medieval frucha by fruta on the Peninsula's central plateau represent irreconcilably different processes. OPtg. fruita >fruta may involve a tendential sound change; OSp. frucha > fruta seems to have been induced by the introduction of semilearned

Sound Correspondences and Levels of Transmission

455

fruto fig. 'fruit' < FRUCTU. In the lexicon far removed from the rural scene the use of -uct- in full-fledged Spanish cultismos (e.g., seductor 'tempter, tempting') is as much to be foreseen as the predictable use of -ut- in their Portuguese counterparts (sedutor). The opposition Ptg. -it- vs. Sp. -ich- is rather exceptional and excludes ancestral -ICT- as the starting point, because that sequence produced / i j t / and, through eventual absorption of / j / by the preceding homorganic vowel, / i t / everywhere, witness frito 'fried' < FRICTU. One accounts for Ptg. Sp. escrito 'written' < SCRTPTU either by bracketing it with atar 'to tie' < APTARE 'to fit' (see above), or by appealing to confusion of -CT- and -PT-. Ptg. dito, Sp. dicho 'said' < DYCTU (but bendito 'blessed', maldito 'cursed') stand apart, because the phonologically regular development should have produced *deito, *decho (witness Tusc. detto). However, in contrast to FILICTU > Ptg. feleito, Sp. helecho and to STRTCTU > Ptg. estreito, Sp. estrecho, which followed the normal course of events, the product of DTCTU, at a certain juncture, fell under the influence of the 'strong' preterit DTXT'I said', etc. ( > Ptg. disse, OSp. dixe / d i s e / > mod. dije /dixe/), so far as the stressed vowel was concerned. By then, Proto-Sp. / j t / must have been well on its way to compression into a unit phoneme, namely / l · / } We have so far .consistently disregarded the segment -ach- because, in striking contrast to all our experiences so far, it is ordinarily shared by the two languages under scrutiny. This deviation from -oit-/-och-, etc., is due to the fact that -ach-, typically, goes back to a prototype other than A + CT, as when Sp. Ptg. macho 'male (animal), virile' reflects MASC(U)LU,9 or as when the suffix -acho perpetuates ancestral -ACEU (presumably transmitted through a Mozarabic conduit). There are also borrowings from a common source; thus, Ptg. Sp. fachada 'facade', fig. 'appearance' are transparent Italianisms (as is Sp. facha 'comic face' in its own right); Ptg. Sp. empachar 'to hinder, encumber' is a Gallicism, as is its semantic opposite, despachar 'to expedite, dispatch.' 10 Related and resemblant to -CT- and -PT-, but entitled to separate consideration, have been the segments / k t j / and / p t j / in Folk Latin. Where these appeared in a very rustic layer of the lexicon, e.g., in the newly-derived verbs in *-CTIXRE and *-PTIARE (on the order of *CAPTIXRE 'to catch, seize' > Ptg. OSp. cagar 'to chase, hunt'; *RUPTIARE 'to break [ground]' > Ptg. OSp. rocar 'to grub, stub, clear [land], nibble [grass]'), / k / and / p / were dropped through assimilation to the emerging affricate, without leaving the slightest trace. The abstracts in -CTI5NE and -PTIÖNE, however, by definition culturally more refined, took a somewhat different course. Here Ptg. eleigäo 'choice' (almost entirely vernacular), with / j s / , conflicts with Sp. eleccion (almost entirely learned), with / k 0 / in most of the Peninsula (including Madrid) and with / k s / elsewhere; both words represent ELECTIONE. Similarly, Ptg.perfei-

456

Yakov Malkiel

?äo alongside Sp. perfection 'excellence' echo PERFECTIÖNE. Ptg. Ιϊςαο 'lesson' has veered away more radically from LECTIÖNE lit. 'reading' than has Sp. lection1, metaphonic raising of -eic- to -ic- may have been suggested here by the example of profissäo 'calling, career, vocation, occupation' < PROFESSIONE, as against Sp. profesion, uneventfully imported. Upon occasion, however, it is Spanish that overtakes Portuguese in their race toward -if-, easier to pronounce in a pretonic syllable, cf. Sp. aficion 'fondness, zeal' vs. Ptg. afeigäo, both from AFFECTI5NE. In this subfield, predictability is close to zero. Yet in the vast majority of word pairs reflecting parental -CTIÖNE, it is customary to find in one language -ςαο, with / s / , and in the other -ccion, with / k 9 / or / k s / , respectively, even though in Portuguese the spelling may tend to lag behind the actual pronunciation. Hence Ptg. α(ο)ςάο 'action', dire(c)gäo 'direction', fra(c)gäo 'fraction, fracture', se(c)gäo 'section', vs. Sp. action, direction, fraction, and section, against the background of ACTIONE, DTRECTIÖNE, FRÄCTTONE, and SECTIÖNE, so that here, typologically, the familiar pattern efeito/efecto, perfeito/perfecto - as if to soothe the bilinguals' bewilderment - reappears. Where the Latin prototype contains the sequence /iktj/, as in DICTIÖNÄRIUM, one is not surprised to find Ptg. diciondrio pitted against Sp. diccionario. If the word were less erudite, one might argue that -it- constitutes a compression of */ijt/, an analysis entailing a certain parallelism between, say, eleigäo and diciondrio·, but given the learned 'cut' of the word as a whole, it seems preferable to collocate diciondrio on the same level as αςάο. Also placeable on that same plateau is construgäo, in contradistinction to Sp. construction, with / k 9 / or / k s / . But the symmetry is dented, or breached, in the case of destruifäo, which, once a perfect parallel to eleifäo, was in the end reinterpreted as destru'ifäo, given the extreme rarity of the diphthong ui in pretonic position. 11 Not so in the ranks of the progeny of / p t j / words, where the near-equidistance of the modern congeners from the common starting point was distinctly better maintained; witness Ptg. acepgäo, concepgäo, recepgäo, and the like (with the fringe benefit, into the bargain, of neat distinction between concepςαο 'conception' and concessäo 'concession'). But on the plateau of archaisms - e.g., in ecclesiastic terminology, conservative by definition - one is hardly surprised to encounter individual instances of / p t j / having sailed in the wake o f / k t j / , as was indeed the case with Conceigäo '[immaculate] Conception', as a New Testament episode. Conceigäo, then, parallels aceitar 'to accept' in depth of erosion. But this is not all: Between -eigäo and -epgäo the Portuguese have managed to establish an intermediate, or transitional, level, as in exceqäo < EXCEPTIÖNE, with the further complication that in the - necessarily less well integrated - derivatives from this word the -p- may optionally be introduced; hence exce(p)cion-al, -ar, also, exce(p)tivo.

Sound Correspondences and Levels of Transmission

457

III If we now attempt to reexamine the triangular situation in a historical perspective and start out with a sketch of the state of affairs in Old Portuguese,12 we immediately run into one difficulty to which we come unprepared from earlier experience: the heavily Latinizing orthography randomly favored by medieval scribes. Because, again and again, we find spellings like doutor, fruyto, and (per)feyto alongside less realistic douctor, fructo, and (per)fecto, we conclude, in a probabilistic vein (and in most instances no doubt correctly), that the 'etymological' c, in such contexts, was a mere embellishment; but a narrow margin of error in such extrapolations is possible. Ρ before ς and t was treated in similar fashion, and occasional scribal confusion of these two letters in such attempts at restoration ('false corrections,' 'hypercorrections') has traditionally been interpreted as additional proof that c and ρ so placed merely illustrate the given scriveners' pretentiousness and cannot serve as clues to actual pronunciation. As the case of fruyto shows, the tendency to reduce ui to u had not yet asserted itself at that time, so that one finds cuitel(o) for modern cutelo, much like entruido < INTROITU for modern entrudo 'carnival (sport), revelry'. 13 And just as, typologically, OPtg. cuitelo is one notch closer to OSp. cuchiello than are their respective modern descendants, so OPtg. as-, es-cuitar 'to listen' < A(U)SCULTARE was closer to OSp. as-, es-cuchar than are their present-day respective counterparts escutar and escuchar to each other, as a result of the fact that Ptg. cutelo and escutar have moved one step more away from the ancestral bases than their Spanish cognates (and, moreover, by coincidence look and sound as if they were semilearned). The basic direction of the movement across the centuries has been from wholly vernacular through semilearned to thoroughly learned representatives of the Latin lexical items salvaged from attrition. Thus, for 'exception', a juridical term in the Middle Ages, the older texts would favor eixegom - better still eiceifon - < EXCEPTIÖNE, where the modern language favors exce(p)fäo; the probability that the ρ will be pronounced increases in such less frequently used and more 'academic' derivatives as (adj.) exce(p)cion-al, (v.) exce(p)cionar, as well as exce(p)tivo (see above). For 'blessed', the old texts offer beeito < BENEDICTU where today's speakers make do with less characteristic bendito or abenfoado·, cf. the family name Beneite < (gen.) BENEDICT? in Western Spain. But this principal evolutionary line is intersected, as it were, by all sorts of semi-autonu mous side developments, as where the presence of -ut- / w t / (chiefly from parental -PT-, but by extension occasionally also from -CT-) made itself far more strongly felt in the Middle Ages than in later periods. Thus, au(c)to would stand for 'act, action, deed' (i.e., for modern ato < ÄCTU) and not just for the narrower range of meanings corresponding to modern auto·, for 'christening', the modern form is batizar (or, among those particu-

458

Yakov Malkiel

larly finicky about orthoepically accurate spelling, bätizar) where medieval texts exhibited bautizar, reminiscent of Spanish; and, for 'to treat', one encounters traces of trautar. The confusion between -ΐςάο and -εϊςαο assumed chaotic proportions at an early date; one stumbles across ίηΐϊςοηι where today's usage pits eleigäo against Ιϊςάο', conversely, descrefom preceded discrigäo 'discretion', a variant possibly favored in an effort to reduce the annoying homonymic overlap with descrigäo 'description'. (Descregom, incidentally, descending as it does from discretione, parallels devogom 'piety' < devötiöne; neither abstract involves any reverberation of ancestral / k t j / or / p t j / , and they are cited here simply to round out the picture.) As the examples of ba(u)tizar and cuitelo have shown, the two languages under scrutiny displayed actual or merely typological resemblances in the lexical sector at issue which have since become blurred; one might add OPtg. outubre as being in one peripheral respect - namely in its ending - closer to OSp. ochubre, modern octubre than is its own modern outgrowth outubro, with -o on the model ofJaneiro, etc.; by the same token, except that this time two prefixoids are involved, OPtg. dereito 'right-hand, etc.' resembled (O)Sp. derecho more than does today's direito. But the Atlantic Coast was at all times far richer in words containing an -eit- segment than was Spanish in the matching -ech- words, either because of lacunae in the lexical inventory (thus, Sp. lechon and Ptg. leitäo 'suckling', from leche/leite 'milk', are perfect congeners, but only Old Portuguese boasted, in addition, leitiga with the same meaning, reminiscent in its suffixal segment of rapariga 'girl'), or because the straight vernacular channel of transmission was, at the critical stage, open for Portuguese but blocked for Spanish. Thus, the Atlantic Coast's despeito and the Central meseta's despecho 'spite, resentment, pique' both perpetuate despectu, but no such form as *aspecho has, at any phase, been even remotely conceivable in Spanish, while OPtg. [mal\ aspeitado 'viewed with an unfriendly mien' is on record. The West's greater openness to -eit- words (and, on a more modest scale, to -oit- words as well) may interlock with the longer survival of certain 'strong' -ctu past participles (cf. OPtg. encolheito 'timorous' = modern encolhido) and with their more vernacular 'cut' (contrast OPtg. esleito with modern eleito 'elected [bishop]'). Even the tributary of Gallicisms and Provensalisms that fed Old Portuguese enriched its inventory of -eit- words (witness espreitar 'to ambush', linkable to OFr. espleitier 'to exploit') more than did the parallel stream emptying into Old Spanish in regard to -ech- words. Finally, if the actual incidence of words is to count in our weighting, medieval geito and geitar deserve far greater prominence than what remains of this word family (actually, it was deitar that chiefly ousted geitar). The situation in older Spanish can be thus adumbrated. There existed, to be sure, individual cases of prevalence of 'regular' -ch- from ancestral -ct- in

Sound Correspondences and Levels of Transmission

459

which, at a distinctly later date, a semilearned, or wholly learned, or abnormal vernacular form lacking / £ / - e.g., given the local phonotactic unavailability of the sequence * / c r / - eventually became superimposed on the medieval variant; here my earlier comments on ochavo and ochubre need not be repeated. Note the following instance of crass asymmetry: fechoria 'misdeed, villainy', from older malfechoria, still exists, given away by its/- as a legal term (cf. fecha 'date' rn lieu of hecha, lit. 'issued'); but, despite the existence in Old Spanish of bienfecho (adj.) 'shapely, handsome', (n.) 'benefit', hence also of bienfechor 'benefactor, donor' and bienfechoria, -uria 'status of a donor', only the variant behetria < be(n)fe(y)tria, endowed with a newly-acquired facetious meaning ('confusion, disorder, pandemonium'), has of late eked out a precarious existence. However, the truly remarkable feature, which - on the level of the literary idiom - solidly brackets the Middle Ages to the entire Golden Age period and even slightly beyond, concerns the development tagged at present as semilearned, but which until 1700 was learned par excellence. That is, / k t / , and, in addition, / k 0 / , / p t / , / ρ θ / , remained as 'unpronounceable' as, say, the / m n / of columna or the -gn- oidigno, for which, as we know for sure from rhymes, even the highly educated of that long era substituted coluna and dino as the currently closest attainable approximations to Classical Latin usage. Thus, in lieu oiefecto and perfecto, peculiar to the most refined variety of 20th-century Spanish, efeto and perfeto were in use, being fully acceptable in rhyme with completo (from COMPLETU), objeto, and sujeto\ thus, respeto cumulated the semantic ranges of modern respeto and respecto\ thus also, ator, dotor, condutor, otimo for centuries ranked as the standard forms, even though such pronunciation was apt, here and there, to be camouflaged by a deceptively more erudite spelling. On the other hand, the rare sequences -eyt- and -oyt-, unless, of course, supported by a following r (as in empeytrar 'to press [against the chest]', coytral 'plowshare', dial, bueytre 'vulture'), would point to a word's foreign provenience or transmission, with the further complication that -oit- would tend to alternate with -uit-.14 IV

While many side developments traced here have .turned out to be thoughtprovoking, and while disregarding or telescoping them might have been tantamount to distorting the record, the one feature that really stands out as being of potential relevance to students of general glottodiachrony is the uneven distribution, in terms of lexical range, of Sp. / c / , especially in the word-medial segment -ech-, and of Ptg. / j t / , especially as a component of -eit-, to the extent that there is mutual correspondence between them and preferably where a common ancestral prototype (namely -CT- / k t / preceded

460

Yakov Malkiel

by one of five front or central vowels) was involved. We have run into this conspicuously uneven distribution synchronically, in our contrastive examination of the lexical inventory of the two languages, and diachronically as well. While asseito and acecho, barbeito and barbecho, deitar/geitar and echar, estreito and estrecho,feleito and helecho, peito andpecho,proveito and provecho are some of the 'perfect' or 'near-perfect' illustrations of the longago recognized isogloss capsulating two parallel sound developments, -eithas an unexplained noteworthy margin over -ech-, as follows from the confrontations of efeit ο and efecto (in lieu of *efecho), leitor and lector, leitura and lectura,perfeito and perfecto,prefeito a n d p r e f e c t o , with the proviso that the Portuguese -eit- variants are valid for all earlier periods, from the dawn of vernacular records, whereas for the Spanish Middle Ages and Golden Age such intermediate forms as efeto, letor, etc., are found instead; an occasional form so shaped tias, as a matter of fact, continued to be used (sujeto, beside Ptg. sujeito), and there has occurred a semantic split in the case oirespeto/respecto (over against Ptg. respeito, marked by a broader scope), as mentioned above. The forms with -ct- must be understood as learned, those with -t- as semilearned. Words involving vowels other than e or such medial consonant clusters at the ancestral stage as / k t j / , / p t / or / p t j / exhibit similarly discrepant evolutions, but less abundantly represented and/or less sharply characterized ones. Since there is general agreement on the learned or semilearned (or else retarded) transmission of parental / k t / as -(c)t- in contradistinction to either its Western-style transmutation into / j t / or its Central-style change into -ch- / c / , what, to rephrase the issue, could possibly have prompted the speakers of Spanish to indulge the use of these cultismos on such a distinctly more liberal scale than their opposite numbers in Portugal? To begin with, the category considered is by no means isolated. As earlier work has disclosed (Malkiel, 1944; Malkiel, 1945), the derivational suffix -ENTiA has been copiously represented in Portuguese, starting with the Middle Ages, by the reflexes (vernacular) -εηςα and (learned) -encia·, Spanish, almost from the start, has preferred -encia to the virtual exclusion of vernacular -ienga, except in a handful of petrifacts like semienfa 'seed'. 15 From Menendez Pidal (1950: par. 98-9) we can confidently borrow the idea that / c / especially as a local substitute for earlier / j t / preserved in neighboring territories - was widely felt as one of those three or four phonic peculiarities of Proto- and Old Castilian speech that constituted downright vulgarismos when judged by the contemporary surrounding standards of socially more highly esteemed vernacular speech. We must next introduce into our analysis the notion that primitive Romance speech, just as it was evolving from local varieties of provincial Imperial Latin, undoubtedly had several social registers, an idiosyncrasy conducive to the incubation of doublets and triplets. By combining these two presuppositions, we can simply argue that, being not unaware of the rusticity, perhaps despicability of / c / , in their native dialect's

Sound Correspondences

and Levels of Transmission

461

phonology, and, in its morphology, of the crudeness of /jentsa/, the organizers of the first specimens of the Castilian vernacular were more consistently inclined to palliate what they must have considered as its shortcomings by an extra generous compensatory injection of more or less learned variants on a far greater scale than was deemed necessary in nearby Portugal. If that is so, it becomes arguable that, under certain conditions of societal structure and pressure, an excess of vulgarismo and of cultismo, far from being mutually contradictory, in sober fact can rank as supporting and even conditioning each other. 16 NOTES 1.

2.

3.

4.

5.

6.

I have left the tag 'semilearned' surrounded by quotes because of its transparent inadequacy: no suggestion of a 50:50 ratio, or even of an approximation to one, has, to my knowledge, ever been intended by those who have, somewhat rashly, used it. The wisdom of replacing it by some more suitable substitute ('partly learned', 'retarded', 'hybrid'?) cannot be discussed in the present context. Meillet's thinking (1931:113-4) had been stimulated by Wahlgren's (1930) provocative analysis of the idiosyncratic development Lat. -D- > OFr. -r-, apparently confined to a special layer of the lexicon, as in MEDICU > mire 'doctor'. Pecho < PÄCTU and peckos < PECTUS were initially near-homonyms in Old Spanish; when, through hypercharacterization of the singular, the -s was eventually dropped in the latter word, the former, visibly endangered, began to yield ground to pago, paga 'pay'. In Portuguese a measure of differentiation was achieved through (a) gender switch and (b) semantic deterioration ('illicit payment') of the product of PÄCTU. For details see a lengthy article of mine, with a focus on homonymy (Malkiel, 1952). This thought raises the question as to how far minor deviations from a pattern of perfect comparability can be safely disregarded. For our purposes it seems legitimate to treat Sp. echar 'to throw, pour, (refl.) to lie down' and Ptg. deitar as mutual opposites, even though, strictly, the former echoes (E)IECTARE and the latter (DE)IECTARE. Also, there are syntactic, semantic, and stylistic discrepancies between the two: in high-level discourse Spanish uses acostarse in lieu of echarse; Ptg. deitar atipara . . . 'to last' is unparalleled in Spanish; echar menos and, ultimately, echar de menos 'to miss' involves the fanciful reinterpretation of borrowed achar menos, lit. 'to find less' (cf. Class, hallar menos). The OPtg. reflex of IACTARE, namely geitar, has become obsolete, as I have tried to show elsewhere (Malkiel, 1949), but cf. its trace injeito 'way, manner'. Note finally that for 'rejecting, discarding' DISSpanish favors desechar, an elaboration on DEIECTARE, with des- < DIS- antihiatically substituted for de-, while Portuguese has opted for rejeitar < REIECTARE. The polarization within the second group is enhanced through use of divergent prepositions in the respective current adverbial phrases for 'in fact': Ptg. comefeito can be pitted against Sp. en efecto. The deeper rooting of trecho in Spanish follows from: (a) the availability in that language exclusively of the offshoot (p\.)pertrechos 'supplies, tools', which Corominas (1967:454a) gratuitously connects with the p.part, PROTRACTU rather than with the family of PERTRACTARE well known to Ernout and Meillet (1959-60:699a), who give this definition of the infinitive: 'manier longuement ou avec soin, examiner en detail', with mention of two satellite formations; (b) the existence of stereotyped phrases: a trechos 'at intervals', (muy) detrechoen trecho 'only once inawhile',apparentlyabsentfrom,orlesscommonin, Portuguese. Note that older Portuguese distinguished between a / ! / , spelled x, and a / δ / , spelled cAandpreservedin

462

7.

8.

9. 10.

11.

12.

13.

14.

15.

16.

Yakov Malkiel northern dialects to this day, a circumstance which must have facilitated the borrowing process. Namely, Germanic and diffused via Old French, cf. It.freccia; see Meyer-Lübke (193035 :par.9424a), who allowed himself to be guided entirely by Gamillscheg. Add to the contingent of such words of exotic descent containing the -ech- segment: Ptg. Sp. mecha 'wick', apparently Greek in the last analysis. With Spanish emerging, again and again, as the language more prone, of the two, to shelter cultismos, one striking detail worth noting is that in the adj .fixo / f i k s u / 'fixed' and the corresponding verb fixar Portuguese, for once, favors the learned interpretations of the grapheme, very much in contrast to Spanish, which equates it with / x / , orig. / s / : an isolated instance of spelling pronunciation? Ptg. macho 'mule' goes back to older muacho, from mu < MULU; Sp. macho 'sledgehammer' continues MARTULU. Also, not a few words displaying -ach- are restricted to one of the languages, such as Ptg. bolacha 'cracker, biscuit, cookie,' although the presence of some other vowel, we recall, does not preclude such an awkward situation (add Ptg. bochecha 'cheek' to the record). In this context, the clash between Sp. muchedumbre 'crowd' < MÜLTITÜDINE and hybrid Ptg. multidäo, in preference to *muitidäo, acquires a new dimension. The vogue of -uifäo led to the analogical coinage of Ptg. constitu'ifäo, distribwgäo, etc., as pendants of Sp. constitution, distribution, etc., from parental -UTIONE. Most of my information comes from such, in general, trustworthy sources as the glossaries appended to Leite de Vasconcelos (1922) and Rodrigues Lapa (1970). However, Leite allowed a bad misprint to slip through: the correct gloss for auctos should have been 'autos' rather than 'autor' (1922:89, 170). The temporary rise of baffling (a)ventuira '(ad)venture', desventuirado 'hapless' could then be explained away as a case of false regression, were it not for the alternative possibility of recognizing in it a lame attempt to imitate OFr. / ü / in these three words closely associated with medieval courtly French culture. Suffice it to refer cursorily to OSp. coita/cuita 'trouble, worry, sorrow', borrowed from Old Provencal or from Old Galician-Portuguese and traceable to COACTA, "COCTA (as distinct from OPtg. coito, OSp. cocho 'cooked, baked' < C O C T U ) , to (em)pleita 'plaited strand of esparto grass', from Gr.plexte passed on by Mozarabic, and to OPtg. pleito ~ preito, Sp. pleito 'lawsuit' < PLAC(Y)TU via OFr. plait. For brief explanatory statements to this effect see Corominas (1967: s. vv.). The information on the conflicting records of Sp. -encia and Ptg. -εηςα will be found in Malkiel (1944) and its companion piece Malkiel (1945). I have later on several occasions reverted to this theme, e.g., in the context of the divergent transmissions of pretium 'price' and a propos OPtg. ga(a)nfa vs. Sp. ganancia 'gain'. Limitations on space have prevented me from going into several important side issues e.g., certain relevant contrasts between Lisbonese and Brazilian pronunciations of medial clusters, vowel quality in the pretonic syllables of certain Portuguese abstracts, and the reaction of speakers of substandard Spanish to the newly-posed challenge of pronouncing the heavy clusters / k t / and / p t / .

REFERENCES Corominas, J. (1967). Breve diccionario etimologico de la lenguacastellana. Rev. 2nded. Madrid: Gredos, S.A. Ernout, Α., and Meillet, A. (1959[-60]). Dictionnaire itymologique de la langue latine. Rev. 4th ed. Paris: C. Klincksieck. Leite de Vasconcelos, J. (1922). Textos Arcaicos. Rev. 3rd ed. Lisböa: Teixeira.

Sound Correspondences

and Levels of Transmission

463

Malkiel, Y. (1944). Three Spanish-Portuguese Etymologies: pendencia, primencia, fimencia. Romanic Review, 35, 307-323. Malkiel, Y. (1945). The Development of the Latin Suffixes -antia and -entia in the Romance Languages. University of California Publications in Linguistics, 1:4, 41-186. Malkiel, Y. (1949). Latin iactare, deiectare, and eiectare in Ibero-Romance. Boletim deFilologia, 10, 201-214. ( = Miscelänea ... ά Memoria de Francisco Adolf ο Coelho). Malkiel, Y. (1952). Studies in Hispano-Latin Homonymics: pessulus, pactus, pectus, despectus, suspectus, fistula in Ibero-Romance. Language, 28, 299-338. Meillet, A. (1931). Review of Wahlgren (1930). Bulletin de la Societe de Linguistique de Paris, 31:2, 113-114. Menendez Pidal, R. (1950). Origenes del espanol. Estado lingüistico de la Peninsula iberica hasta el siglo XI. Rev. 3rd ed. Madrid: Espasa-Calpe. Meyer-Lübke, W. (19[30-] 35). Romanisches Etymologisches Wörterbuch. Rev. 3rd ed. Heidelberg: Carl Winter. Rodrigues Lapa, M. (ed.) (1970). Cantigas d'escarnho e de mal dizer dos cancioneiros medievais galego-portugueses. Rev. 2nd ed. Editorial Galaxia. (Bound with a separately paginated Vocabulärio Galego-Portugues.) Wahlgren, E.G. (1930). Un probleme de phonetique romane: Le developpement d> r. Skrifter utgivna av Kungliga Humanistiska Vetenskapssamfundet, 26:4. Uppsala and Leipzig.

The Morphophonemics of Flexion in Serbocroatian: A Comparison Kenneth E. Naylor, Ohio State University, Columbus, OH., U.S.A.

1. INTRODUCTION

It is my intention in this paper to summarize the material which I have gathered and published over the past decade on the morphophonemics of Serbocroation flexion (Naylor, 1969; 1975; 1981)1 and to compare and contrast the patterns found in the various parts of speech to determine if there are any which are common to all (i.e., which might be considered 'common core') as well as to establish if there are patterns whose distribution might be seen as having a functional role in distinguishing one part of speech from the others. 2 In this way, it is hoped that we can speak more confidently when comparing the morphophonemic patterns of flexion to those of derivation. Unlike my earlier studies, which focused on the patterns of a specific part of speech, I shall consider the distribution of the alternation among the parts of speech. (A composite picture of the distribution of the alternation is given in Figure I.) 2. ACCENTUAL ALTERNATIONS

2.1. Quantity It can be seen that alternations of quantity are found primarily in the declension of substantives and adjectives. Even in those cases where it is found, the alternation is largely automatic and can be subsumed under one of the following headings: (a) lengthening of a short vowel before a sonorant in closed syllable; (b) shortening of an inherently long vowel before another vowel, specifically / o / which alternates with / l / in closed syllable, (c) shortening of a long vowel in a stressless stem before a disyllabic desinence (found in substantives only). There are two interesting instances where, however, the quantitative alternation must be interpreted as being morphological; when it is in the flexion of pronouns and the verbs, specific conditions must be stated. In the first instance, sequences of the s h a p e / o j e / may optionally b e c o m e / o / in the gen. sg. and dat.-loc. sing, of nonfeminine adjectives. (This is incidentally only optional and as such deserves special mention as all other alternations are obligatory.) The second instance occurs in verbs of the I conjugation

466

Kenneth Ε. Naylor Nouns

Verbs

Adjectives

+a +b

+ c+r

(+y -H

+ +

(+) f

+»

+ +

+ + +

Pronouns

Accent β«α

_m _m

Quantity (-)

f

-H

-

+n

Consonantal

κ«δ K«C K«C Jotovanje Others Vocalic #«a fe^e/o/a

-

(+) -

-

+h

V

-

-

-

k

c

-

+k

+ +1

-

+

+

-

Fig. I.. Occurrence of Morphophonemic Alternations in Serbocroatian by Morphophological Class. a b

c

d

e f

8 h

1

1

k

1

Alternation found in I (masculines and neuters), II, and III declensions. Masculine and neuter I declension and II declension, if nom. sg. taken as base form for II declension. In any event, alternation is limited to a small number of stems for masculines (I declension) and II declension stems. Found only in III declension before < - j u > of instr. sing.; this ending is in free variation with < - i > under certain circumstances. The alternation < # > e s / ä / before the gen. pi. desinence < - 5 > is considered automatic because only long vowels may occur before this ending. Limited; found only in the imperative of a few verbs. Alternations of this type are found but are better analyzed, from a morphological point of view as sequences o f / a j e / , / e j e / or / a j ä / , / e j ä / > / £ / and / e / , with a special rule which applies only to the verbal system. We do find examples in the verbal system of V»»V, e.g., krasti, krädem or musti, mutem where the alternation is found in the stem but these are very limited in number and the examples suggest that the alternation may be connected with the alternation in the stem which takes place in the infinitive. Including changes of where there is no / j / involved. Includes alternations of the type T»»/s/ before t, truncation of / j / before consonantal desinence and T-truncation before / l / . Unlike the alternation < # > = / a / , found elsewhere, < φ > must be in the a 2 syllable of the stem, and the results are not predictable. The alternation is found between positive and comparative, specifically between the indefinite of the positive, as well as between indefinite and definite. This alternation is found only between the positive and comparative; but there would be no evidence of such an alternation in the adjectival system if we considered the formation of the comparative derivation, rather than flexion, as many do. Some stems undergo shortening from positive to comparative; all of them have a / k / in the positive which is lost before forming the comparative, e.g., visok, gladak, sladak, tezak. This shortening, however, is not automatic, witness tanak, comparative tanli.

The Morphophonemics of Flexion in Serbocroatian

467

which have stems ending in / a j / and / e j / ; here there is an alternation o f / a , e / with / a , e / when the vowel following / j / is nonhigh and long; ( / j / is also lost in this environment). It is interesting to note that there are limitations on this alternation in terms of both the type and length of the following vowel for this to take place. Another instance which might well be treated morphologically, but which seems to me to be automatic, is the general alternation between the indefinite and comparative of adjectives where long vowels shorten. This contrasts strikingly with the restricted alternation of the same type found between the definite and indefinite. Finally, the lengthening of short vowel (or < # > ) , before the gen. pi. desinence < - a > , might well be viewed as morphological because it occurs under specific morphological conditions, i.e., before the genitive plural, but I shall consider it as being automatic because it is always found (when this suffix is used). 2.2. Stress Shifts3 Stress shifts are found in all parts of speech in Serbocroatian, except pronouns. In part, it is easy to understand why pronouns lack them, in view of their stem-shape, which is basically mono-or asyllabic. Although the majority of attention here will be given to stems which have nonautomatic stress shifts, i.e., those which are not due to phonotactics and the number of syllables in the stem/word, it would be wrong to allow this to obscure the fact that the majority of nonpronominal stems have fixed stress. (Included in this class are the verbal stems which have an automatic retraction of stress on a syllable in an open stem when the stem vowel is truncated before a vocalic desinence.) 2.3. Progressive Shifts (a ^ ßj Examination of Figure I shows that alternations of the progressive type, from

m

Alternations are largely absent due to the shape of pronominal stems; many of which are either asyllabic, or if syllabic, show a suppletion to another stem in flexion. n Limited alternation between / o j e / and / δ / in the nonfeminine genitive singular oblique. This alternation is stylistic rather than systematic and need not occur. (The fact that the alternation does not occur with feminine adjectives which contain a similar sequence suggests that it may be conditioned by the fact that the nonfeminines are potentially trisyllabic or that sequences of / o j e / are not affected.) ° Truncation of / j / is found in the deictic pronouns, i.e., ovaj, taj, onaj, which can also be analyzed as having a zero. Suppletion is found in a few other pronouns, i.e., it- /c- οη-/ή-, m-/n-, ja/men-. p There are a few limited instances of vocalic alternations, specifically, ! = = ! /β/ (men-mn-) and / e / ~ / o / , / t o b - , s o b - / / t e b - , seb-/.

468

Kenneth Ε. Naylor

stem to desinence, are more widespread than the regressive shifts (from desinence to stem). Examples of this alternation are found in substantives, verbs and adjectives, although in the latter instance, it may be viewed as somewhat marginal. It is interesting to note that the substantival stems which show the alternation belong to all three declension types - 1 , II, and III - and stems of all genders - masculine, neuter and feminine - are found. In the verbal system, the progressive shifts are the most common and productive and have the widest distribution. On the other hand, we must consider the alternation somewhat marginal for adjectives; it is found as an alternate to fixed stress - in the formation of definites from indefinites and comparatives from indefinite positives. We might note, however, that this pattern seems to be gaining ground among adjectives, which is not surprising given its productivity in substantives and verbs. 2.4. Regressive Shifts fß ~ a) Unlike the progressive shifts just discussed, alternations from desinence to stem are much more restricted in their distribution, frequency of occurrence and productivity. In the verbal system, they are extremely limited and found only in the imperative of a few verbs. Among the substantives, the pattern is found only with a few stems of the I declension, mostly neuters. 5 In adjectives, it is found, like the progressive shift, both between the indefinite and definite and the positive and comparative, but it is productive rather than marginal, especially with stems which use the formant < - j - > in making the comparative. The restrictions on its occurrence are themselves noteworthy when compared to the far more common progressive shifts. 3. CONSONANTAL ALTERNATIONS

As Figure I shows, we can distinguish three types of consonantal alternaν

'

tions, viz., those involving velars, viz., Κ « C, Κ « C, Κ « C, (b) those involving / j / , traditionally called ''jotovanje', which involve substitutive softenings, and (c) 'others.' 6 This last group is most difficult to characterize in that it includes truncation, i.e., the loss of / j / before consonants, / t , d / before A / , the change of / t , d / to / s / before / t / and the alternation of/s,z,k,g,h/ to /s,z,c,z,s/ where there is no phonological or morphological / j / . In all instances we have ignored the automatic voicing assimilations which take place with the loss of an / a / (which alternates with < # > ) , and the alternations of / o / with / l / , both of which are phonologically predictable. In terms of distribution, it is once again interesting that the pronouns lack alternations except for a few instances of truncation of / j / , and some suppletions. Given the generally static nature of the pronominal stems, which we have already noted, this should not be surprising. What is surprising, how-

The Morphophonemics of Flexion in Serbocroatian

469

ever, is the fact that we find that verbal stems exhibit consonantal alternations of all five types which we have outlined and that in fact, some are found only in the verbal system. The nominal system shows far less in the way of consonantal alternations; for substantives only K ^ C , and Κ » C are found regularly. Even here, Κ » C is limited to the voc. sg. of I declension masculine stems, which means that it is extremely restricted in its frequency of occurrence. In the same way K ^ C exhibits a number of limits on the alternation with II declension stems, with a number of specific restrictions, e.g., it is not found with proper names which have stem final clusters or certain derivatives. (The fact should also be noted that / k / tends to keep the alternation better than either / g / or / h / . ) The fact that the desinence which is needed to produce 'jotovanje' in substantives (specifically of the instr. sg. of the III declension) is in free variation with the ending / i / , forces us to consider the alternation as marginal. In the flexion of adjectives two things must be noted; first, and most important, the alternation occurs only between the positive and comparative, which might be considered derivation rather than flexion, and second, given the fact that we can posit that a morpheme which forms the comparative contains / j / , it is possible that we are only dealing here with ' j o t o v a n j e a l t h o u g h some cases would appear to be examples of Κ « C. It is important to point out that many alternations, K » C , and those which have been assigned to the class O t h e r ' , are found only in the verbal system. 4. VOCALIC ALTERNATIONS

The basic vocalic alternation in literary Serbocroatian is that of v o w e l « zero. It is necessary, however, to distinguish two kinds of 'zero,' i.e., < # > which alternates with / a / , and < φ > which is realized as / e , o, a / . Vocalic alternations are found in all parts of speech, although it should be noted that the pronouns show only three instances, all of which are examples of < φ > rather than < # > , with a full vowel. In terms of their distribution, all classes show examples of < # > » / a / while the alternations of < φ > are restricted to the verbal system, except as noted above. In the first case, it is necessary to mention that the / ä / found in the genitive plural before the ending < - ! > is interpreted as a regular alternation with the lengthening found being ascribed to the rule mentioned above which permits only long vowels before that suffix. On the other hand, it should be noted that / a / ' s which alternate with < # > before sonorants in the nom. sg., masc. indef. of adjectives do not lengthen as we might expect. Two features which distinguish the second zero, which has been noted with the symbol < φ > , are (1) the fact that it occurs only in the penultimate syllable of the verbal (or marginally pronominal) stems and (2) the fact that it is realized as a full vowel when the vowel in the following syllable is truncated.

470

Kenneth Ε. Naylor

Also its realization as a full vowel is unpredictable, i.e., / e , o , a / may be found (cf. brati, berem, zvati, zovem, trnati, tarnem). 5. OBSERVATIONS

Based on the survey which has just been presented, it is clear that a number of generalizations can be made about flexional morphophonemics of Serbocroatian. (1) The lack of morphophonemic alternations in the pronouns is striking and suggests that this class is marked morphologically by the absence of alternations. (2) The predominance of consonantal alternations, both in terms of distribution and variety, in the verbal system compared to their limitations in the substantival and adjectival systems, would seem to be functional rather than purely accidental. This is also striking given the very different patterns when we compare the patterns of vocalic alternations between the two classes. (3) Accentual alternations, both of quantity and of place, are more similar in the nominal systems, i.e., between substantives and adjectives, than between the nominal and the verbal systems, although there is an overlap between the two major classes. Although it would be difficult to find a set of distinctive features which mark the morphophonemic system of any of the parts of speech definitively, it is possible to point out that using all four alternation types examined here, a fairly clear picture can be drawn. As an examination of Figure I will show, sufficient distinction is found to suggest that a hierarchy can be established, viz., (a) pronouns tend not to have morphophonemic alternations, (b) verbal stems differ from nominal stems in the variety and number of their consonantal alternations, (c) adjectives are distinguished from substantives by differences in their patterns of accentual and consonantal alternations, and (d) the existence of two types of vowel/zero alternations sets verbal (and marginally pronominal) stems apart from nominal stems.

NOTES 1.

2.

In this paper, as in the other articles which I have done dealing with the flexional morphophonemics of Serbocroatian, I have chosen to use a morphophonemic - rather than a phonemic (or phonetic) analysis of stress. A description of this interpretation can be found in Naylor (1969:79-80). Jakobson (1948:157) provides an excellent example of this for Russian where / i / always causes palatalization when it is part of a verbal or pronominal ending but it does not do so when it is part of a nominal ending. Clearly this difference must be seen as morphological in the modern system as we do not expect the phonology of a language to be selective in the ways phonemes work and we know that the situation in Common Slavic was not selective.

The Morphophonemics 3.

4.

5.

6.

of Flexion in

Serbocroatian

471

In view of the regularity of the shift and the fact that it serves to keep the stress on the stem, it is fair to say that this is an automatic rather than a morphological alternation. The fact that it is generally viewed as 'stem internal', i.e., α 2 = α ι , when a 2 is truncated before a vocalic desinence supports the automaticity of the alternation. In Naylor (1969) these stems were analyzed as having a progressive shift, i.e., α « β , because the accusative, rather than the nominative, singular was used as the base. If the nominative singular is taken, however, as the base the alternation would be of the regressive type, i.e., β « « . Both analyses are valid but the net result would be to change slightly the distribution of accentual types here. For the purpose of this discussion, I have ignored the marginal regressive alternation found in masculines (see Naylor, 1969:86) because we find nothing similar in the other parts of speech. Mark (1982), which I discovered after having finished this paper, has an alternative way of describing consonantal alternations in Serbocroatian but it should be noted that he deals with all alternations - derivational as well as flexional - so that the results and his conclusions will differ from those which I have offered here.

REFERENCES Jakobson, R. (1948). Russian Conjugation. Word, 4, 155-166. Merk, Η. (1982). Sustav konsonantskih alternacija u srpskohrvatskom knjiievnom jeziku. Aarhus: Aarhus Universitet, Slavisk Institut. Naylor, K.E. (1969). Morphophonemics of the Serbocroatian Declension. International Journal of Slavic Linguistics and Poetics, 12, 79-89. Naylor, K.E. (1975). Morphophonemics of the Serbocroatian Conjugation. International Journal of Slavic Linguistics and Poetics, 18, 83-92. Naylor, K.E. (1981). Morphophonemics of Serbocroatian Adjectives and Pronouns. Wiener slawistischer Almanack, 8, 307-317.

Sämpsä Pellervoinen Felix J. Oinas, Indiana University, Bloomington, IN., U.S.A.

Sämpsä (Sämsä, Sampsa) Pellervoinen's song, a song of framework repetition, was once used to accompany the performance of an annual rite to a fertility deity (Kuusi, Bosley, and Branch, 1977:544). There are two versions of this song, both of them rare. One of these, recorded in defective form in North Karelia, tells that Sämsä fled to the North (Pohjola), after he lay with his sister or stepmother. Ahti offered money to anybody willing to go and fetch Sämsä back, in order to sweeten fields and 'produce a hardy crop, make a better ear.' The wolf goes, and on its second trip proves successful. (Comparative evidence shows that the actual inviters of Sämsä are the 'summer-boy' and the 'winter-boy.') 'Sämsä the Pellervö b o y / took up a few seeds/ a mixture of seeds/ black specks from an ermine's tail/ from a swan's footprints.' Upon arrival he sowed the fields and forests (Kuusi et al., 1977:256-258). The other version was recorded from Ingria. It begins with a picture of disaster: no grain or grass is growing, since Sämpsä is asleep in bed. The winter-boy takes a wind stallion, goes to Sämpsä and asks him to get up. Sämpsä refuses, because the winter-boy has caused so much damage to nature, blowing the trees leafless and the maids out of their bloom. When the summer-boy goes in his turn to invite Sämpsä, he is willing to come, since the summer-boy has reversed the ill effects of the winter-boy (Kuusi et al., 1977:259-261). The song of Sämpsä Pellervoinen thus has two basic themes: (1) Sämpsä's disappearance, which causes the grain to cease growing, and (2) the invitations to return. In addition, the first version has an introduction describing Sämpsä's incest as the reason for his departure. Finnish scholars have unanimously pointed to the international background of the song. Kaarle Krohn (1903:763-769) saw in the motif of Sämpsä's illicit love the influence of the Scandinavian Freyr cult. Uno Harva (1948:177ff.) suggested that the invitation of Sämpsä reflects a type of game known in Europe, and even in Asia, which enacts the victory of summer over winter. Martti Haavio (1952:250-251) thought an international myth to be the basis for summer's victory. In a special monograph (Haavio, 1961) he compared the Sämpsä song with the Greek Dionysios cult, pointing out numerous common traits between them, which - according to his cautiously

474

Felix J. Oinas

expressed conclusion - could testify to the influence of the Greek tradition on the Finnic. Scholars have also pointed to the similarities of Sämpsä to the Roman Bacchus, the Latvian Jumis, the Estonian Peko, and the Russian Orthodox saint Samson (Kuusi et al., 1963:351-352; Kuusi et al. 1977:545; Kirkinen, 1967; Turunen, 1979:295). It cannot be denied that there is a general similarity between the Sämpsä song and the international tradition quoted, but there is not a one-to-one correspondence. This holds true especially for the parallels given by Haavio from the Dionysian myths; hardly any of the numerous parallels listed would in themselves constitute a strong case for establishing the influence of the Greek Dionysian tradition on the Finnic. Despite considerable work, no scholar has yet successfully tackled the essence of the Sämpsä song - the departure of the fertility divinity and his return - with all the consequences. In Propp's terms, this is the 'lack - lack liquidated' theme. It is the purpose of the following lines to suggest a new solution. Among Hittite texts there is a story, termed as 'outstanding,' which deals with Telipinus, 'the god who disappears and takes the whole of life with him, who is sought for, brought back, and appeased by ritual' (Burkert, 1979:123). We do not know the introduction to the story, since the upper third of the tablet is broken off. The narration begins with a description of the all-embracing catastrophe caused by Telipinus' desertion: fires in houses go out, grain and spelt no longer thrive, trees shrivel, cattle do not breed, etc. The Storm-god, Telipinus' father, sends the eagle to look for him, but the bird cannot find the deserter. Then the Storm-god goes himself, but without result. Finally the mother of the gods sends out the bee, who discovers Telipinus lying asleep in a meadow. It stings him and thus awakens him. The enraged god returns only after a special purification ritual has been performed. Now the good life is re-established. A pole is erected before Telipinus; from it the fleece of a sheep is suspended, which signifies (contains?) sheep fat, kernels of corn, wine, etc. (Pritchard, 1955:126-129; Burkert, 1979:123-124). As Burkert has shown, this story is related to a Greek myth of Black Demeter. According to the Arcadian version, Demeter, angered after being raped by Poseidon or because her daughter Persephone had disappeared, hid herself in a cave. As a consequence, nothing grew any more and people were dying. None of the gods knew where she was. Finally Pan found her in the cave, and the Moirai brought her back. When returning, she brought the earth's fruit with her. She also prescribed a special form of sacrifice for herself: fruits, grapes, honey-combs, etc. The disappearance of the fertility goddess symbolizes famine. On the occasion of a disastrous famine, such as occurred in the fifth century B.C., people felt the need to renew the cult of Black Demeter in the cave. This meant, as Burkert suggests, 'to 'find out' the god offended, the cult neglected.' The cave in a gorge was rediscovered and the cult installed (Burkert, 1979:125-126).

Sämpsä Pellervoinen

475

In another version, the Homeric hymn to Demeter, the offended goddess retreats to her temple at Eleusis. The earth ceases to bring forth any crops. Finally, the mother of the gods (as in the case of Telipinus) brings her back, and new 'crops grow again the fields, leaves and flowers sprout, life is restored' (Burkert, 1979:128).' It has been pointed out that the temple motif in the Homeric version is secondary; since the goddess is known to be in her sanctuary, there is no need for a search, only an invitation for her to return. It could be mentioned that Demeter's priestesses are called melissai, 'bees,' - an additional trait connecting this version with the Telipinus story (Burkert, 1979:128-129). The Telipinus-Demeter myth strongly resembles the song of Sämpsä Pellervoinen. In all of them the god or goddess of fertility leaves, and, as a result, growth stops. The deserter is sought, found, and brought back, after which life continues normally. The Near Eastern (Anatolian) myth was created as the rationalization of famines and droughts: the catastrophe in nature is attributed to the fertility god's anger, aroused by the non-observance of ritual. The renewal of the ritual brings the divinity back. We have every reason to believe that the Finnish Sämpsä song was created following the Near Eastern myth. In addition to the identity of the basic pattern, there are a few similarities in detail. The introductions of the Demeter and Sämpsä myths contain, among other things, an unnatural sexual incident (rape or incest), which serves as the cause for the divinity's departure. The intensity of Telipinus' rage makes us believe that his desertion, too, must have been caused by an uncommonly serious offense. In the Telipinus myth, the Storm-god exhibits special eagerness in finding his son, first sending out the eagle for the search and then going himself. In the Sämpsä song, too, the wind has an important role. The winter- and summer-boy go in search of the fugitive with the wind stallion and by blowing cause a great change in nature. In the Near Eastern myth, sacrifices are offered to the returned divinity. The fleece of a sheep, fat, kernels of corn, etc., are set up before Telipinus, and fruits, grapes, etc., before Demeter. Can they not be compared with Sämpsä taking a few seeds and black specks from an ermine tail with him when returning? In a version thought to refer to Sämpsä six seeds are put in the summer squirrel's hide or in the black part of the ermine's tail (Harva, 1948:174). It is of paramount importance in establishing the prehistory of the song of Sämpsä Pellervoinen that the song appears on the eastern periphery of the Baltic Finns - Finnish and Russian Karelia and Ingria. It can be assumed that the Anatolian myth of the temporary disappearance of the fertility divinity migrated to the Greeks and, through several intermediaries, to the north, and reached the Finns via the Russians. So far no Russian correspondent has surfaced.

476

Felix J. Oinas

ADDENDUM

I have subsequently come across a Mordvin song and a Russian one that represent the link between the Hittite-Greek versions and the Karelian Finnish 'The Sower.' These songs are about the gods' feast. A Mordvin variant {Obrazcy, 1882:64ff.;Harva, 1952:203) goes as follows: The Gods' Feast There is a large field. In the field is a hill, on the hill a valley, in the valley is an apple tree. The roots of the apple tree spread throughout the earth, its branches spread throughout the sky. At the ends of the branches are buds, from the ends of the buds drips honey, on the leaves lies wax, on the tips of the leaves burn candles. Vere-pas [upper god, heaven god] prepared food, beer and drinks, Vere-pas made a barrel of beer; he invited all the gods, called together all the Niske [creators, heaven gods], only Norov-pas [grain god] was missing, only Norov-ava [grain mother] was not among them. 'Whom will we send for Norov-pas, whom will we send for Norovava? Let us send the quick messenger, the gracious Mikila. - Mikila, go fast!' - Where is Norov-ava staying, where is Suron-pas [grain god] staying? - Norov is staying in the wheat field, in a wheat field, on a straight boundary strip. Mansikka (1941:108) gives the following list of motifs of this song, compiled from all the available variants: In a large field grows a miraculous tree (apple tree, birch). A god (Vere-pas, Skaj [creator], Niske-pas) has arranged a feast near it. All the gods (saints) have, by invitation, gathered to eat and drink. Only one, Norov-ava (Norov-pas) has not arrived. God asks whom He should send to get the missing guest, and He decides to send Mikola (Purgine [thunder], Ilja the Great). Norov-ava is reported to be staying on the boundary strip between the cornfields. Norov-ava refuses to come, since she herself has a feast (the sowing feast) the same day, and she receives food and drink in great quantity from the people. Purgine-Ilja brings the missing guest along with him. The Mordvin song of the gods' feast is thematically closely related to the Hittite-Greek song. It tells of a feast to which the fertility god has not come. She is in the field. Quick messengers are sent to invite her. She usually refuses, and only rarely does she come. The role of the fertility god is sometimes transferred to Mikula (St. Nicholas), whom the Mordvins consider the promoter of the growth of grain (Mansikka, 1941:107).

Sämpsä Pellervoinen

All

Some variants of the gods' feast contain motifs that bring them closer to the Hittite-Greek song, e.g., the Moksa Mordvin variant, in which it is Mikola who has not been invited to the feast (Mansikka, 1941:108-109).2 The supreme God sends Christmas and Easter to call Mikola, but he does not come. Then God decides to go Himself. His question is indicative: 'Oh, Mikola, why are you angry?' In another variant Mikola goes by horse as a messenger to invite the missing Norov-ava (Harva, 1952:202).3 A third variant from Moksa enumerates the offerings Norov-ava gets from the people: wine, beer, chicken, bread and salt (Harva, 1952:202-203, after Snesnickij, 1870:702). The Mordvin song is no doubt of Slavic origin. Belorussians have the ν corresponding koljada, used for wishing good luck at Easter (Sejn, 1874:77ff.; Bezsonov, 1871:1-2; Mansikka, 1941:109). St. Mikola or some other saint does not come to the feast; he is walking along the boundary strip. Sts. Peter and Ilja refuse to go to invite him, but St. George goes. He finds Mikola walking through the field and taking care of the growth of the grain. The fact that unmistakable traces of the Anatolian and Greek myths are found in Russia makes our conclusion concerning the ultimate origin of the Karelian Finnish 'The Sower' in these myths more plausible. NOTES *

I am indebted to Professor Jaan Puhvel, University of California, Los Angeles, for his suggestions. 1. Greek mythology also tells of the disappearance of Hera, which is interpreted as the equivalent of a drought. The bringing back of her daidalon (wooden image) causes Hera to return (Burkert, 1979:132-133). 2. Cited by Mansikka after a manuscript of Paasonen. 3. Cited by Harva after a manuscript of Solnikov.

REFERENCES Bezsonov, P. (1871 ).Belorusskiepesni, I. Moscow: Obs£estvoljubitelej rossijskoj slovesnosti. Burkert, W. (1979). Structure and History in Greek Mythology. Berkeley: University of California Press. Haavio, M. (1952). Kirjokansi. Porvoo-Helsinki: Werner Söderström. Haavio, M. (1961). Kuolematonten lehdot. Porvoo-Helsinki: Werner Söderström. Harva, U. (1948). Suomalaisten muinaisusko. Porvoo-Helsinki: Werner Söderström. Harva, U. (1952). Die religiösen Vorstellungen der Mordwinen (= Folklore Fellows Communications, 142). Helsinki: Suomalainen Tiedeakatemia. Kirkinen, H. (1967). Oliko Sampsa historiallinen henkilö? Kalevalaseuran Vuosikirja, 47, 103-132. Krohn, K. (1903). Kalevalan runojen historia. Helsinki: Suomalaisen Kirjallisuuden Seura. Kuusi, M. (1963). Suomen kirjallisuus I. Helsinki: Suomalaisen Kirjallisuuden Seura and Otava. Kuusi, M., Bosley, K. and Branch, Μ. (1977). Finnish Folk Poetry: Epic. Helsinki: Finnish Literature Society. Mansikka, V.J. (1941). P. Nikolaus mordvalaisilla. Virittäjä, 1941.

478

Felix J. Oinas

Obrazcy. (1882) Obrazcy mordovskoj narodnoj slovesnosti, I. Kazan': Tipografija gubernskogo pravlenija. Pritchard, J.B. (ed.) (1955). Ancient Near Eastern Texts Relating to Old Testament. Princeton, N.J.: Princeton University Press. Sejn, P. (1874). Belorusskie narodnye pesni. St. Petersburg: Tipografija Majkova. Snesnickij. (1870). Penzenskie epiarxial'nye vedomosti. Turunen, A. (1979). Kalevalan sanat ja niiden taustat. Lappeenranta: Karjalaisen Kulttuurin Edistämissäätiö.

Vowel Lengthening and metatonie rude in Lithuanian David F. Robinson, Ohio State University, Columbus, OH., U.S.A.

1. INTRODUCTION

The expression metatonie rude, apparently coined by Ferdinand de Saussure, refers to the replacement of non-acute intonation (in Lithuanian, this involves circumflex ['long rising'] intonation) by acute intonation ('long falling' intonation), regardless of the origin of such metatony. Stang (1966: 111-112) correctly observes that Lithuanian exhibits many more cases of acute intonation than one would predict from Proto-Indo-European. While there is undoubtedly such a group of Lithuanian words whose intonation is inherited from PIE, a second group of cases may be of Proto-Balto-Slavic origin, although the latter may sometimes reflect changes in length and only secondarily differences in tone. A third group of instances of apparent mitatonie rude seems to be purely Lithuanian in origin, involving acute intonation in a stressed originally non-acute syllable where the following syllable is (or has historically been) lost, e.g. gäl from gäli 'can (3 pers.)', or gelz{ from gelezj 'iron (acc.sg.)'. In a wide-ranging article on Proto-Indo-European accent Paul Kiparsky develops a theory of 'accent spread' which, combined with a mora analysis of Lithuanian phonology, seems to account for the metatony exhibited in the third group above. Such metatony, in the terms of his analysis, becomes a 'pseudo-problem' (1973:836). Kiparsky handles this purely as a matter of vowel lengthening, with intonation derived automatically from the place of stress within a two or three-mora syllable. Without entering into a discussion of the merits of the Baltic portions of Kiparsky's article or of the advantages and disadvantages of a mora analysis of Lithuanian, we can certainly agree that in this case Kiparsky is right in showing that the intonation of Lithuanian words of the third group above is purely a matter of lengthening of stressed low vowels, a rule required elsewhere in the grammar of Lithuanian. Earlier in his article, Stang recognizes that intonation and length are intimately connected in Lithuanian, particularly in diphthongs. He states, with reference to the second (Balto-Slavic) group of cases of mitatonie rude, that such metatony 'is to be found in formations which almost certainly contain a lengthened first diphthong component (yrddhi)' (1966:112). In the same place, Stang states that the third (purely Lithuanian) group of cases of

480

David F. Robinson

metatonie rude 'can be explained in Lithuanian . . . on a purely phonetic basis . . . in the final syllable the combination of stressed short [sic-DR] vowel + sonant becomes an acute diphthong if a vowel disappears after the sonant . . . ' The difference of opinion between Stang and Kiparsky on this point, then, seems to revolve principally on the nature of the Lithuanian acute tone in a stressed diphthong, i.e., whether it is reducible to and, in fact, equivalent to a sequence of long vowel plus high vowel, liquid or nasal, or whether, as Stang seems to suggest, it is actually an additional prosodic feature automatically added to such a sequence. As we have suggested earlier (Robinson 1968:208,210), Lithuanian 'tone,' at least in the standard dialect, is now no more than a sequence of phonemes as discussed just above. A few Lithuanian scholars seem to support this view, especially Kazlauskas(1968:8)andPakerys (1967:14-15), although they would probably have some reservations about putting the matter so flatly. Klimas (1970: 101) and others have stated that most Lithuanian speakers do not make even this distinction inthediphthongs/eand uo,sothatnowthereisnodifference of any kind between certain 'acute' and 'circumflex' syllables. We cannot hazard a guess as to when Lithuanian tone began to d r o p out of the standard language, but we can observe that the process has surely begun and can suppose that it will probably continue.

2. LITHUANIAN ACUTE INTONATION

The nature of Lithuanian acute intonation as merely an alternate name for a sequence of long vowel plus high vowel, liquid or nasal in the same syllable, gains some support f r o m a consideration of the substantival dative plural ending in Lithuanian. This ending alone (together with the dative singular masculine ending in the pronominal system) exhibits acute intonation when stressed. We earlier proposed (Robinson 1979:463) that the dative plural ending exhibits acute intonation by a rule that automatically assigns such intonation to stressed vowels before resonants in the same syllable, and simultaneously lengthens such vowels. It would be preferable, it seems, to restate this rule as part of the general rule that stressed low vowels (a and e) are regularly lengthened in Lithuanian. The language seems to have changed since c. 1650 in this respect, such that at some time before 1650, the lengthening rule must have been: 1. Lengthen stressed non-final low vowels in stems and disyllabic

suffixes

This will account for the Old Lithuanian endings -'amus [-a:mus], -'omus, etc., which all show length on the penultimate syllable when the ending is stressed. (Note that the originally short a of -amus is affected by this rule; the

Vowel Lengthening in Lithuanian

481

rule applies vacuously to the inherently long ο of -'omus.) The question of why this ending should show penultimate as opposed to absolute final stress remains unanswered. After c. 1650, by a separate process, the dative plural ending is shortened to -dms, -oms, etc. In more recent times, whenever new contractions take place, whether in allegro forms or otherwise, vowel lengthening does not take place, e.g., vienäm 'one (loc. sg. masc.)' from viename, cf. tdm 'this (dat. sg. masc.)' from older tamui. The first example shows retraction from a deleted final stressed syllable; in this case, the newly-stressed syllable -äm is not lengthened. In other words, after c. 1650, the above lengthening rule ceases to operate and is replaced by the following: 2. Lengthen stressed low vowels in stems of loanwords This rule applies to all new words in the language, e.g., advokätas 'lawyer', pdrkas 'park'. Evidence for rule 2 is provided also by the large Lithuanian speech area where stress is retracted from final short stressed syllables, as in gälva 'head', cf. standard dialect galvä (Senn 1966:96). These dialects show acute intonation in gälvq (acc. sg. of preceding), just as the standard dialect does; since, however, the retraction shown in the nominative singular occurred after 1650, but in a native word, lengthening of the stressed stem vowel does not take place. This brings about a paradigm in which one and the same syllable exhibits sometimes acute intonation, reflecting the operation of rule 1 before 1650, and sometimes circumflex, reflecting stress placement after that. In this century, or perhaps even earlier (and noted in East Lithuanian before entering the standard dialect), the rule has been modified and broadened further, so that it now reads: 3. Lengthen stressed vowels in stems The operation of this rule in the standard dialect is extremely controversial, but see Kazlauskas (1968:8) for at least one authoritative hint at this sound change in progress. It appears that, for probably a majority of speakers of the standard dialect, stress and length are now inextricably connected. How far this change will spread, or whether it will be resisted, we will be in a position to judge in the coming decades.

482

David F. Robinson

REFERENCES Kazlauskas, J. (1968). Lietuviu kalbos istorine gramatika [Historical Grammar of the Lithuanian Language]. Vilnius: Mintis. Kiparksy, P. (1973). The Inflectional Accent in Indo-European. Language, 49, 794-849. Klimas, A. (1970). Some Attempts to Inventory Lithuanian Phonemes. In: Th.F. Magner and W.R. Schmalstieg (eds.) Baltic Linguistics. University Park and London: Pennsylvania State University Press, pp. 93-102. Pakerys, A. (1967). Dabartines lietuviu literaturines kalbos dvibalsiu pagrindinio tono (F 0 ) moduliacija [Modulation of Fundamental Frequency in Contemporary Standard Lithuanian Diphthongs}. Kalbotyra, 17, 5-15. Robinson, D.F. (1968). Some Acoustic Correlates of Tone in Standard Lithuanian. Slavic and East European Journal, 12, 206-212. Robinson, D.F. (1979). On Dating Sound Changes in Lithuanian. Papers from the Conference on non-Slavic languages of the USSR. Chicago: Chicago Linguistic Society, 462-464. Senn, A. (1966). Handbuch der litauischen Sprache, I. Grammatik. Heidelberg: Carl Winter. Stang, C.S. (1966). 'Metatonie Douce' in Baltic. International Journal of Slavic Linguistics and Poetics, 10, 111-119.

The Grammar of Choice in Russian and English Anelya Rugaleva, Ohio State University, Columbus, OH., U.S.A.

0. INTRODUCTION

Alongside the traditional grammar of forms and structures, a natural language can also be described in terms of a so-called 'grammar of choice'. The latter predicts which words, forms or structures are likely to be used rather than which words or structures can be used in a given language. The rules of the grammar of choice differ from what is usually understood by the term 'rule of grammar' in that they are formulated as tendencies or probabilities, rather than as deterministic laws. Jakobson (1959:235-236) writes, 'As Boas neatly observed, the grammatical pattern of a language (as opposed to its lexical stock) determines those aspects of each experience that must be expressed in the given language.' By 'grammatical pattern' Jakobson, and Boas, understand the grammar of forms and traditional grammatical categories, such as aspect, number, etc.; they explicitly oppose grammar to the lexical resources of a language. The 'grammar of choice' approach adopted here allows for a broader interpretation of 'those aspects of each experience that must be expressed in the given language.' This study will consider tendencies and preferences which influence the selection of lexical items in Russian and English for describing the same extra-linguistic situation. The main objective is to demonstrate that on the lexical level, too, to use Jakobson's words again, 'languages differ essentially in what they must convey and not in what they may convey' (1959:236). Conclusions will be based on the interpretation of data from Russian and English original texts and their published translations. In order to eliminate, or at least to reduce, the element of the author's or translator's personal stylistic preference, examples have been drawn from works by different authors and translators. 1 The study was done within the framework of semantic syntagmatics, which investigates how components of meaning (semes) combine in linear speech sequences, how a given language designates a given situation, what information about that situation the language selects as relevant, and how these portions of information are encoded in linguistic forms. One of the basic concepts of semantic syntagmatics is the law of semantic agreement. Semantic agreement, like any other agreement, is based on iteration. For a sentence

484

Anelya

Rugaleva

to be well-formed semantically, a certain amount of information must be repeated in more than one linguistic item. In other words, a well-formed sentence of a natural language must contain redundancy of information. The law of semantic agreement has different manifestations in Russian and in English. Anticipating the results, I wish to claim that in Russian the requirement of informational redundancy is more evident in the surface lexical pleonasm than it is in English. The Russian and English phrases designating the same situation will be compared in the following aspects: what types of lexical items the two languages typically choose to denote an object, a process or a property; what degree of statement of detail is normally accepted as sufficient by each language; how much denotative (referential, logical) information is explicitly conveyed by the words in Russian and English sentences; and to what degree the connotative (expressive, emotive and stylistic) information is made overt in both languages. 1. GENERAL A N D SPECIFIC

When man perceives reality and expresses his perception with language, he always does so against certain probabilities: something is always known to the speaker and to the listener about the situation and the objects to be described. This background information, or presupposition, affects the choice of one or another linguistic sign. If some elements of the extra-linguistic situation are included in the presupposition, they may remain latent and not be expressed lexically without infringing on the amount of relevant information included in the message. Presupposition on which the speaker relies can be of two types: a) knowledge shared by all humans or by all members of the given civilization; b) knowledge related only to the given situation. The former is best illustrated by the so-called dispositional predicates (a term used in logic), which express the object's intrinsic 'disposition' to display or manifest itself. Thus in Russian, in the situation 'object in motion', man is usually described by the verb idti, a dog is described by the predicate bezaf, a bird by letet', a snake by polzti. Existence and location in space, which certainly may be expressed by the verbs byt' or naxodit'sja in Russian, are usually rendered by specific verbs containing the same semes that are also included in the noun: the manner of existence of plants is usually denoted by the verb rasti, of animals by the verb vodit'sja, of fire by the verb goret', of lightning by sverkaf, of buildings by stojaf, of rivers by tec', etc. In both languages either a specific or a general term may be used: There stood/was a church at the far end of the street - V konce ulicy stojala/naxodilas' cerkov'. The choice of a variant does not affect the amount of information about the situation, and the lexical expression of the predicate is selected according to a

The Grammar of Choice in Russian and English

485

certain tendency characteristic of the given language. For a wide array of dispositional predicates English usually prefers a general term, while Russian tends to employ a more specific one and thus to hypostatize each object not only in the nominal lexeme, but also in the verbal lexeme. (1)

There was a slight noise behind her . . . (Mark Twain. The Adventures of Tom Sawyer) pozadi poslysalsja legkij sorox . . . (Translation by K. Cukovskij) . . . the edifice was a small, plain affair, with a sort of pine-board tree-box on top of it for a steeple (Ibid.) . . . zdanie malen'koe, nekazistoe, i na kryse torcalo necto vrode uzkogo jascika iz dosok - kolokol'nja. 'To hell with everything. That's really what you'd like.' This was unexpected. (Gore Vidal. Washington, D.C.) 'K certu vce. Vot cego ty na samom dele xoces'.' Eto prozvucalo neozidanno. (Translation by V. Smirnov and A. Fajngar) •

(2)

(3)

t

(

In (1), the noise,is simply brought into existence by the most abstract and general verb be in English, while in Russian a specific term of auditory perception poslysat' sja is used (aside from this, the nature of the noise, namely sorox, is specified, too). In (3), the deictic this and eto both refer to the spoken words in the preceding sentences, and again, the modus vivendi for the sound is the general was in English and the specific prozvucalo in Russian. In example (2), English compresses the expression of existence into a prepositional phrase (with a sort of...), completely eliminating the verb, whereas Russian resorts to a specific verb torcat' in order to describe the particular geometric position of the steeple 'sticking out' on the top of the church. Redundancy or economy of expression can also depend on a concrete situation. Here, too, Russian employs more specific words and describes a greater number of aspects and elements of the situation. English uses general terms which are applicable to other situations as well, in fact, to a wide range of situations. Such general terms acquire their concrete and specified interpretation only by virtue of the specificity of the situation itself. (4) The old lady pulled her spectacles down and looked over them about the room; then she put them up and looked out under them. (Mark Twain. The Adventures of Tom Sawyer) Staruska spustila ocki na koncik nosa i ogljadela komnatu poverx ockov; potom vzdernula ocki na lob i gljanula iz-pod nix. (Translation by K. Cukovskij) In a real situation, indeed, one pulls one's spectacles down 'onto the tip of one's nose', and pulls them up 'onto one's forehead', however, English omits

486

Anelya Rugaleva

the localizers, while Russian finds it necessary to mention them. The English verbs pull down and push up can be applied to many other objects in various situations, but in this particular situation, this particular object, namely spectacles, can only be pulled down and pushed up in a quite specific manner. Thus, both Russian and English provide the same information about this situation, but in Russian certain portions of this information are provided several times: first they are introduced as the description of the situation as a whole (an old lady manipulating her spectacles), second, they are encoded in linguistic signs. (5)

About half past ten the cracked bell of the small church began to ring, and presently the people began to gather for the morning sermon. The Sunday-school children distributed themselves about the house and occupied pews with their parents, so as to be under supervision. (Mark Twain. The Adventures of Tom Sawyer) Okolo poloviny odinnadcatogo zazvonil nadtresnutyj kolokol malen'koj cerkvi, i prixozane stali sobirat'sja k utrennej propovedi. Uceniki voskresnoj skoly razbrelis' ν raznye storony po cerkovnomu zdaniju, usazivajas' na te ze skam'i, gde sideli ix roditeli, ctoby vse vremja byt' pod nadzorom starsix. (Translation by K. Cukovskij)

Even without any preceding context, the mention of the small church and the morning sermon sets up a situation specific enough to narrow down the meaning of the nouns people, children (in Sunday-school children), house. Nevertheless, the Russian translation again and again reminds the reader about the locale and the occasion of the situation being described through the words prixozane, i.e., people who belong to the church, uceniki, i.e., children who are taught in Sunday school, po cerkovnomu zdaniju. It is true that the word house in the lexicon has, as one of its meanings, the meaning 'church, temple': it is also true, however, that this meaning can be realized only in an appropriate context, and the situation described in (5) provides such a context. A recipient of the above message can easily picture how children distribute themselves about the church, but the Russian word, which can only be applied to humans, makes this motion doubly specific. The physical position of both the children and their parents on the church benches is quite obvious even if a very general verb occupied is used to describe it, but Russian characteristically is lexically redundant in usazivajas' and sideli·, furthermore in a situation involving children and adults where children are said to be under supervision, it is clear who the supervisors are, and in this respect, too, the Russian addition of starsix {pod nadzorom starsix) is redundant. More examples of this type are given below.

487

The Grammar of Choice in Russian and English (6)

She had said that she was in bed and ill. (Thackeray. Vanity Fair) Bekki pisala, cto ona bol'na i lezit ν posteli. (Translation by M. D'jakonov)

(7)

Can we meet for a smoke in the Teacher's Lounge between classes? (Kaufman. Up The Down Staircase) Nel'zja li nam ν peremenu vykurit'po sigaretke ν ucitel'skoj? (Translation by E. Ivanova and S. Sajkevic)

(8)

Then Henry and the Packard arrived and Burden got in without a word. (Gore Vidal. Washington, D.C.) V etu minutu pod"exal Genri na 'pakkarde', i Berden sei ν masinu, ne skazav ni slova. (Translation by V. Smirnov and A. Fajngar)

(9)

She waved to him. (Ibid.) Ona pomaxala emu rukoj.

(10)

He was not in the chair but beside it, talking to a group of Senators. (Ibid.) On ne sidelna svoem predsedatel'skom meste, a stojalrjadom, beseduja s gruppoj senatorov.

Example (10) is particularly characteristic. Evidently, being in the chair, one is supposed to sit, and being beside it, to stand. Such an interpretation seems equally natural for both the English and the Russian observer of situation (10). However, a literal translation into Russian: On by I ne na stule, a rjadom s nim would produce a somewhat comic effect on the Russian reader, or, at the very least, would puzzle him. The tendency towards the specific and particular in Russian affects the syntagmatics of Russian speech: two or more words are selected in such a way that they share a greater number of specific semes than is the case in English. That is why such syntagms as subject - predicate, verb - object (or complement), verb - adverb (or adverbial phrase) or attribute - noun display greater mutual 'adjustment' and selectivity of lexemes. (11)

The first thing the beetle did was to take him by the finger. . . . the hurt finger went into the boy's mouth. The beetle lay there working its helpless legs unable to turn. (Mark Twain. The Adventures of Tom Sawyer) Zuk pervym dolgom vpilsja emu ν palec . . . ukusennyj palec Tom totcas ze sunul ν rot. Zuk upal na spinu i bespomoscno baraxtalsja ne umeja perevernut'sja. (Translation by K. Cukovskij) ν

ν

488

Anelya Rugaleva

(12)

Verx avtobusa byl zasteklen kakim-to neobyknovenno golubym steklom. (Iskander. Sozvezdie Kozlotura) The top of the bus was covered with a tinted-blue glass ... (Translation by Helen Burlinghame)

(13)

The conductor came around for old Mrs. Morrow's ticket. (Salinger. The Catcher in the Rye) Vosel konduktor proverjat' bilet u missis Morrou. (Translation by R. Rajt-Kovaleva)

Syntagmatic semantic agreement and informational redundancy often come to the fore when Russian designates those actions, processes or states that are expressed in English by the so-called verbs with wide semantics: go, come, make, take (as in 11), get and some others. (14)

She made a face. (Gore Vidal. Washington, D.C.) Ona skorcila grimasu. (Translation by V. Smirnov and A. Fajngar)

(15)

Na Ivana ραχηύΐο vlaznym teplom. (Bulgakov. Master i Margarita) A wave of moist heat came at Ivan. (Translation by Mirra Ginsburg)

(16)

. . . that makes him come up to the top. (Mark Twain. The Adventures of Tom Sawyer) Ot etogo utoplenniki vsplyvajut naverx. (Translation by K. Cukovskij)

In (14), the mutual adjustment goes both ways in the Russian syntagm skorcila grimasu: it would be vacuous to try to determine which of the two elements 'comes first'. An interesting consequence of the above examples would be a re-evaluation of the semantic and syntactic role of the verb in sentence structure. The supremacy of the verb which is claimed by case grammars can at least be questioned in view of the Russian data. Indeed, the verb vsplyvajut in (16) does not predict the choice of the subject, since all kinds of things submerged in water can 'float up'. On the other hand, the subject utoplenniki 'drowned bodies' predetermines the selection of the verb. 2. DENOTATION A N D CONNOTATION

Each language has in its lexicon so-called 'expressive' or 'emotive' words which, besides being the signs for extra-linguistic referents, also express the speaker's attitude towards these referents. As a rule, an emotive word has a non-emotive, or 'neutral' counterpart whose meaning is exclusively, or almost exclusively, referential. In many cases, the referential (denotational) component of the meaning of an expressive lexeme is roughly the same as that

The Grammar of Choice in Russian and English

489

of its non-expressive synonym, which means that they both can be used to denote one and the same object, process, etc., cf. in Russian zacem on sjuda xodit and zacem on sjuda sljaetsja. Very often Russian makes the attitude of the speaker (or subject of the discourse) explicit while English leaves it implicitly included in the situation described. (17)

Then came the measles. During two long weeks Tom lay a prisoner dead to the world and its happenings. (Mark Twain. The Adventures of Tom Sawyer) Potom prisla kor'. V tecenie dvux dolgix nedel' Tom provaljalsja ν posteli, kak uznik, umersij dlja mira i dlja vsex celoveceskix del. (Translation by K. Cukovskij)

Here the translator prefers the expressive provaljalsja to the neutral prolezal in order to make explicit the sick boy's attitude: note also the redundant reference to the place where the patient lay ν posteli 'in bed'. (18)

Da, on, vidimo, znal cenu etogo utrennego starikovskogo casa s cajkom i tabackom. (Tvardovskij. Pecniki) Obviously he knew the value of his old man's morning hour for tea and a smoke. (Translation by R. Dixon)

Derivational morphology as one of the sources of expressive vocabulary is richly developed in Russian. Diminutive and, to a lesser extent, augmentative suffixes are widespread and productive. Such suffixes may either express objective quantitative characteristics: nosik - a small nose, tixonecko, tixoxon'ko - very quitely, domisce - an enormous building; or they can have a purely attitudinal or emotive role: caek, xlebusek, golovuska, rozica\ or they may combine both: domina - not just a very large building, but also ponderous, plat'ice - a little and/or pretty dress, rucka, nozka, liciko - any imaginable combination of diminutive denotation and endearing connotation. If the Russian suffixed word is indeed diminutive or augmentative, the translation usually contains the necessary quantitative seme also, at least at the first entry of the object. (19)

Iz komnaty vysel Alik, zametno podrossij, blednoe vesnuscatoe sozdanie let odinnadcati, ν ockax na tonkom nosike. (Trifonov. Obmen) Alec came out of the room, noticeably taller, a pale freckled creature of about eleven, with glasses on his thin little nose. (Translation by E. Proffer)

The emotive connotations, however, are most often left implicit in English

490

Anelya Rugaleva

and are not encoded in lexemes or morphemes; in Russian they are not only made explicit, but may be reiterated several times. In translating from Russian to English expressive components are usually omitted; in translating from English to Russian they are added. (20)

. . . odin znakomyj jakoby upal ν prorub', probaraxtalsja podo l'dom dvenadcat' minut i vylez ottuda celexonek, zivexonek i zdorovexonek . . . it appears one fell into an ice hole, floundered under the ice for twelve minutes and crawled out safe and sound. (Translation by R. Dixon)

(21)

Ja i sam by sejcas na balkoncike cajku popil. (Bulgakov. Master i Margarita) I would much rather be sipping tea on a balcony myself. (Translation by Mirra Ginsburg)

(22)

The young lady teachers 'showed off . . . lifting pretty warning fingers at bad little boys, and patting good ones lovingly. (Mark Twain. The Adventures of Tom Sawyer) Moloden'kie ücitefnicy 'kozyrjali' po-svoemu . . . s ulybkoj grozja xorosen'kim palcikom neposlusnym i laskovo gladjapo golovke poslusnyx.(Translation by K. Cukovskij)

(23)

. . . there was not a dry spot to sleep on, anywhere around. (Ibid.) . . . krugom ne ostalos' ni odnogo suxogo mestecka.

As was the case with denotative redundancy discussed earlier, Russian displays a tendency to iterate a large number of emotive semantic components as well; and such expressive redundancy may also be either situational or syntagmatic. English tends to use a neutral word and leave the expressivity to the situation itself, while Russian chooses a specialized expressive lexeme. (24)

Sadly he allowed the vision of Thuvia and the Tharks to break into bits, along with the highball glass that his father had apparently thrown at either his mother or the floor. (Gore Vidal. Washington, D.C.) Videnie Tulii i Tarkov razletelos' na kusocki, kak tot vysokij stakan, kotoryj otec, verojatno, svyrnul ν mat' ili na pol. (Translation by V. Smirnov and A. Fajngar)

(25)

. . . his coat, when he wore one, hung nearly to his heels . . . (Mark Twain. The Adventures of Tom Sawyer)

The Grammar of Choice in Russian and English

491

. . . pidzak, ν te redkie dni, kogda Gek napjalival ego na sebja, doxodil emu cut' ne do pjat . . . (Translation by K. Cukovskij) (26)

V nacinajuscixsja sumerkax Berlioz otcetlivo razgljadel, cto usiski u nego, kak kurinye per'ja, glazki malen'kie, ironiceskie i polup'janye, a brjucki kletcatye . . . (Bulgakov. Master i Margarita) In the falling twilight Berlioz could clearly see that his moustache looked like chicken feathers, his eyes were small, ironic and halfdrunk, and his tight trousers were of a checkered m a t e r i a l . . . (Translation by Mirra Ginsburg)

Even greater redundancy is evident in numerous cases of syntagmatic semantic agreement, when expressive components are pleonastic and iterated in several (usually two) words. The rules of mutual adjustment for emotive lexemes are the same as those operating on purely referential words. (27)

No dlja etogo razresite esce odin kroxotnyj nomerokl (Ibid.) But in order to do that may I present just one more tiny item?

(28)

This superintendent was a slim creature of thirty-five . . . (Mark Twain. The Adventures of Tom Sawyer) Direktor byl pljugavyj celovecek let tridcati p j a t i . . . (Translation by K. Cukovskij)

The diminutive nomerok is obligatory with adjective kroxotnyj: kroxotnyj nomer would sound awkward, almost ill-formed. The adjective pljugavyj chosen as the translation of slim conveys directly the author's (and the character's) attitude, and the diminutive form celovecek is also derogatory here. Note that besides having syntagmatic redundancy the Russian phrase pljugavyj celovecek is situationally redundant too; the English slim creature may lack the derogatory connotation in a different context, e.g., as a description of a romantic heroine it would acquire quite positive, even poetic connotations (cf. Russian izjascnoe/vozdusnoe sozdan'e). 3. REGISTER A N D STYLE

The choice of a lexeme may depend on the circumstances of the act of communication (official, formal, friendly, intimate), on the social status of the participants, on their age, etc. A certain stylistic homogeneity characterizes any text. The vocabulary of each language has special words for different speech registers or styles.2 Barxudarov (1975:111) discusses the following five registers: 1) intimate, 2) casual, 3) neutral, 4) formal, 5) elevated. 3 The majority of words in the vocabulary stock belong to the neutral register;

492

Anelya Rugaleva

however, they can be used in any other register, from familiar to elevated. Different registers are perceived as such only in contradistinction to the neutral register; and, obviously, there must be words in different registers having the same referent; cf. Russian dryxnuf (intimate) - spat' (casual and neutral) - otdyxaf (formal) - pocivaf (elevated). The choice of a specialized word stylistically marked for the given register results in stylistic agreement. Stylistic agreement differs from semantic agreement in that it relates to the text as a whole, rather than to a sentence or a syntagm. Cukovskij (1964) gives some comments on the Russian translation of Dickens' The Life and Adventures of Martin Chuzzlewit made by N. Daruzes. In one episode, a rude and despotic husband shouts at his defenseless wife: 'Why do you show your pale face?', which can be translated into Russian 'literally' (i.e., using the neutral dictionary equivalents) as 'Pocemu ty pokazyvaes' svoe blednoe lico?' The translator, however, renders this ν

ν

phrase as: 'Cego ty sues'sja so svoej postnoj rozej?' According to Cukovskij, this translation is 'much closer to the style of the original . . . Postnaja rozathis is precisely what the original says, judging from the context' (1964:94) [trans, mine - A.R.]. I would like to emphasize the words 'closer to the style of the original' and 'judging from the context'. What the English recipient gathers from the overall stylistic coloring of the situation the Russian recipient must get from lexical expression. Another example of the same type: (29)

Then he noticed Peter in the window. 'Go away Γ he shouted like a farmer to a bird that has suddenly landed on a new-sown field. (Gore Vidal. Washington, D.C.) Tut on uvidel ν okne Pitera. - Provalivaj otsjuda! - zakrical on, kak fermer, uvidevsij pticu, sadjascujusja na tol'ko cto zasejannoe pole. (Translation by V. Smirnov and A. Fajngar)

In the English original the stylistic interpretation follows from the immediate context: " . . . like a farmer to a bird . . . " This context provides the necessary stylistic coloring to the exclamation "Go away!" In Russian, the same context triggers the choice of the appropriate register of speech for the exclamation: "Provalivaj otsjuda/" rather than the neutral"Uxodi/". It is in this sense that we are speaking of stylistic agreement and stylistic redundancy in Russian. When translating from English to Russian, one often has to highlight stylistic overtones lexically, i.e., to "lower" or "elevate" the register (29, 30 and 31 are examples of lowering; 32 and 33, of elevating); in translations from Russian to English, on the other hand, we observe a certain lexical "levelling" of style (as in 34 and 35). (30)

I am here because it amuses me to stir things up. (Ibid.) Znaete, pocemu ja zdes'? Menja zabavljaet vozmoznost' balamutif eto boloto.

The Grammar of Choice in Russian and English

493

(31)

He made a speech that lasted about ten hours. (Salinger. The Catcher in the Rye) On otgroxal rec' casov na desjat'. (Translation by R. Rajt-Kovaleva)

(32)

. . . the ladies would lift up their hands and let them fall helplessly in their laps. (Mark Twain. The Adventures of Tom Sawyer) . . . damy vozdevali ruki k nebu i totcas ze bespomoscno ronjali ix na koleni. (Translation by K. Cukovskij)

(33)

. . . and then a solemn hush fell upon the church. (Ibid.) . . . i zatem na cerkov' snizosla torzestvennaja tisina.

(34)

Odin xudoznik mnit sebja tvorcom nezavisimogo duxovnogo mira, i vzvalivaet na svoi pleci akt tvorenija etogo mira, naselenija ego, ob"emljuscej otvetstvennosti za nego - no podlamyvaetsja, ibo nagruzki takoj ne sposoben vyderzat' smertnyj genij . . . (Solzenicyn. Nobelevskaja lekcija). One kind of artist imagines himself the creator of an independent spiritual world and shoulders the act of creating that world and the people in it assuming total responsibility for it - but he collapses for no mortal genius is able to hold up under such a load . . . (Translation by F.D. Reeve)

(35)

Elena. Cto ν vas est' xorosego? Servinskij. Da vy vsmotrites'. Elena. Nu pobrjakuski ad"jutantskie, smazliv, kak xeruvim. I golos! I borse nicego! (Bulgakov. Dni Turbinyx)

ν

Elena. What's good about you? Shervinsky. Look closely. Elena. Well, your aide's insignia, and you are as handsome as a cherub. And your voice. And that's all. (Translation by F. Reeve) 4. CONCLUSION

Our study demonstrates that in Russian the law of semantic agreement has wider application than in English, and that the selection of lexical items, both within the sentence and in supra-sentential sequences, is governed by the requirement to make explicit and to iterate a larger portion of the information. In this connection the following questions arise: Do Russian translations - with their tendency to make things concrete, specific and explicit distort the meaning of the English original? Is the additional information

494

Anelya Rugaleva

included in the Russian sentence a mere otsebjatina on the part of the translator? And, on the other hand, is the English translation - with its omissions and generalization - deprived of necessary information that was intended in the Russian original? Diderot (as cited in Gak, 1977:108) once noted that an expression is weak or false if it creates uncertainty about the feeling expressed. Apparently both the recipient of the English text and the recipient of the Russian text have the same impression about the situation, objects, emotions, etc. In order to produce equivalent impressions, and due to different grammars of choice on the lexical level, the two languages utilize different means. In Russian, specific terms and emotive or stylistically marked vocabulary items must be used in order to render the meaning adequately: failure to employ a specific or emotive word would create an "uncertainty about the feeling". Lexical explicitness is a characteristic trait of a Russian text. In English, a considerable portion of the information remains outside the linguistic expression proper. English is characterized by greater implicitness and tends to avoid conveying that which is clear from the situation: the recipient retains the context in his memory, and this allows him to narrow the interpretation of a general term and make it specific. Thus English words appear to be more independent of context in their use, or encoding (since the same word can fit into various contexts), and more dependent on context for their interpretation, or decoding. Russian words, on the other hand, are easier to decode (since most characteristics and connotations are made explicit in the word itself), but more difficult to encode, since the encoding involves greater selectivity, i.e., a longer search for an appropriate lexical item. NOTES 1.

2.

I deliberately avoid any literary evaluation of the translations, and in that I share the view expressed by the Russian linguist V.G. Gak (1979) on the different approaches to translated material of a "translatologist" and a contrastive linguist. The former explicitly evaluates all aspects of the translated passage: its adequacy in meaning, style, pragmatics, poetics, etc; he also points out the translator's failures and errors. A contrastivist evaluates the translation, but implicitly - by the fact of his choosing to use it as a source of linguistic data: if he finds a given translation inadequate, he simply does not use it. Besides,"... a contrastive linguist, while analyzing a certain phenomenon, has the right to limit his analysis to this particular phenomenon, and to disregard the merits or drawbacks of the rest of the translated passage, even to omit it if it is irrelevant to the interpretation of the phenomenon under investigation" (Gak, 1979:12-13) [trans, mine - A.R.]. In this paper the terms "style" and "register" are used loosely and synonymously since there was no need for greater rigor in differentiating language varieties vis-ä-vis the immediate situation of the utterance for this particular study. In other works different terms are used to describe language varieties along various parameters. Thus Catford (1965:85) employs three terms which he defines as follows. "Register: variety related to the

The Grammar of Choice in Russian and English

3.

495

wider social role being played by the performer at the moment of utterance: e.g., 'scientific', 'religious', 'civil service', etc. Style: variety related to the number and nature of addressees and the performer's relation to them: e.g., 'formal', 'colloquial', 'intimate'. Mode: variety related to the medium in which the performer is operating: 'spoken', 'written'." Barxudarov correlates his five "registers" (his Russian terms are, respectively: 1) famil'jarnyj, 2) neprinuidennyj, 3) nejtral'nyj, 4) formal'nyj, 5) vozvySennyj with the five "styles", viz. intimate, casual, consultative, formal and frozen in Joos (1961:1 Iff.).

REFERENCES Barxudarov, L.S. (1975). Jazyk i perevod. Moscow: Meidunarodnye otnoSenija. Catford, J.C. (1965). A Linguistic Theory of Translation. London: Oxford University Press. t u k o v s k i j , Kornej. (1964). Vysokoe iskusstvo. Moscow: Iskusstvo. Gak, V.G. (1977). Sopostavitel'naja leksikologija. Moscow: Meidunarodnye otnoSenija. Gak, V.G. (1979). Sopostavitel'nye issledovanija i perevodieskij analiz. Tetradiperevodiika, 16, 11-21. Jakobson, R. (1959). On Linguistic Aspects of Translation. In: R. Brower(ed.) On Translation. Cambridge, Mass.: Harvard University Press, pp. 232-239. Joos, M. (1961). The Five Clocks. New York: Harcourt, Brace and World.

The Passive in English: A Discourse Perspective Sandra A. Thompson, University of California, Los Angeles, CA.,

U.S.A.

1. INTRODUCTION

It will be my purpose in this paper to propose an answer to the question of what the parameters are for English which determine the choice between an active and a passive clause at a given point in a discourse. I will suggest that the decision to use a passive clause is based on a consideration of two different pragmatic discourse factors, one of which predicts an agentless passive, and the other of which predicts an agented passive. Jespersen (1965:167-168), though he did not cite any discourse data and did not actually quantify his claim, also noticed that the parameters determining this choice were different for agentless and agented passives. Given a two-participant situation in which one of the participants is an agent, the strategy for choosing a passive over an active coding can be stated in terms of three ordered statements: (1)

A. B.

If the agent is not to be mentioned, use the passive. If the agent is to be mentioned, then use the passive only when the non-agent is more closely related than the agent either B.l. to the 'theme' of the 'paragraph,' or B.2. to a participant in the immediately preceding clause.

Justification of this three-part strategy for the use of the passive will form the body of this paper. 2. THE DATA

I used data, from various styles of English for this study. The formal written English data base consisted of articles in The Christian Science Monitor, the UCLA Daily Bruin, and passages from the following books: Severin's Explorers of the Mississippi, a detailed historical survey of the exploration of the great river from the point of view of the explorers themselves, Woodcock's Asia, Gods and Cities, which is a travelogue, and Tannahill's Food in History, a historical study of how people's eating habits have influenced the structure of human societies and culture.

498

Sandra Α. Thompson

The radio provided somewhat less formal (spoken) English. News broadcasts are written to be read, and are thus intermediate between highly formal written and conversational spoken English. I taped 1'/2 hours of news broadcasts from National Public Radio during January and February, 1982. Interviews are less formal and less planned. I taped one hour of radio interviews from National Public Radio's 'All Things Considered' and 'As It Happens' during the same time period, with topics ranging from international and national political issues to matters of artistic or literary concern. Also quite informal, though edited and printed, are the monologues which I took from Terkel's Working, a collection of people's often lengthy responses to an invitation to talk about their work. 1 Finally, for highly informal conversational English, excerpts were taken from transcripts in the files of E. Schegloff, 2 and from Carterette and Jones (1974). Anticipating a question as to what counts as a 'passive' in this study, I note that I am interested here only in those passives (called 'agentive' by Svartvik (1966)) which could occur with agents. Those which do (as in (2)) I am, following Svartvik, calling 'agented'; those which do not (as in (3)) I am calling 'agentless.' (2) (3)

The cavalcade had been delayed by the police. The city was renamed Leningrad.

I am thus not considering such uses of the 'passive participle' as these, in which expression of an agent is impossible: (4)

(5) (6)

'statal passive': Peering inside, we saw that the walls were hung with medieval tapestries (*by the monks). attributive 'passive adjectives': She collects abandoned (*by neighbor kids) motorcycles. 'lexical passives': She was pleased (*by Irwin) with the way the office looked. Jamie will be heartbroken (*by the coach) when he finds out he can't play.

3. AGENTLESS PASSIVES

There is widespread agreement that agentless passives far outnumber agented passives in English. Svartvik (1966) found that 80 percent of the passives in his extensive written corpus were agentless. The percentage for my radio news broadcasts was only slightly higher: out of 185 passives, 82 percent were agentless. Givon (1979:58-60) reported a similar percentage of agentless passives: the average for three pieces of fiction was 83 percent. Most English

The Passive in English: A Discourse Perspective

499

passives, then, are used when there is sufficient reason to forego mention of the agent. Jespersen proposed the following two circumstances under which the agent would not be indicated (1965:167): (a) 'The active subject is unknown or cannot easily be stated' (b) 'The active subject is self-evident from the context' Givon suggests that the missing agent is always 'recoverable': either the identity of the missing agent can be determined by inference from the immediately preceding or following discourse, or, based on generally available pragmatic information, the type of agent that is most likely to have been involved can be established (1979:59-61). Here is an example of Givon's first type from Working. A person describing the end of his job as a proofreader in a law firm says: (7)

I left under circumstances of considerable honor. I was given a farewell luncheon by half the staff of the law firm, meanining the lawyers themselves. I was asked to make a speech and I was much applauded, (p. 667)

Here it is obvious from the remarks preceding the italicized passives that it was the lawyers hosting the luncheon who asked the narrator to make a speech and who applauded. An example of the second kind, where it is the type of agent that is inferred, comes from another monologue in Working. A 'memorial counselor' whose domain is burials talks about upkeep: (8)

In the olden days, the maintenance of the cemetery was left to the individual family. One family would pay and the others didn't. You would have weeds in one area and someplace else cared for. Today, in a modern cemetery, you have trust funds. Whenever a family purchases, a part o f t h a t money is put into a trust. This trust is inviolate. In this state it's held by a third party, a bank. You know that cemetery is gonna be cared for. (p. 656)

In (8), the addressee is not able to infer the exact agent, but rather can infer the type of agent: the trust fund pays for groundskeepers to care for the cemetery. For agentless passives, then, there is relatively little controversy over the claim that the agent which is not expressed is either inferrable from the context or has a referent whose exact identity is not important. Before leaving the topic of agentless passives, I want to mention another option available in informal English for situations in which the identity of the agent is inferrable or unimportant: the use of non anaphoric they. This option

500

Sandra Α.

Thompson

is restricted in my data to informal conversational discourse; it is consistently used, although the passive would also be possible, where the agent is known only to be the people who are typically responsible for events such as that named by the verb phrase (as suggested in Halliday and Hasan (1976:53)). Here is an example from one of the Schegloff transcripts, a portion of a phone call to the New Orleans Civil Defense Headquarters during the 1964 hurricane: (9)

Because they're building eh they - when they bil - they're building this new subdivision?

No reference is made elsewhere in the conversation to the builders of the subdivision; the pronoun is thus clearly not anaphoric. It refers vaguely to whoever the builders of the subdivision are. Since the exact identity of the agent is not important, the passive, according to our rule in (1), could equally well have been used. In fact, although the numbers are too small for us to be confident of this as a prediction, it does appear from my data that in informal English, a non-anaphoric they in the active is favored over the passive whenever the inferrable but unimportant agent is 'those typically in charge of such activities.' Here are two more examples f r o m another telephone conversation, this time between two friends: (10)

A: B: A: B: A: B:

(11)

B:

Oh my mother wanduh know how's yer grandmother *hhh Uh::, (0.3) / d o n ' k n o w I guess she's aw- she's awright she went to the uh:: hhospital again tihda:y, Mm-hm? *hh t? *hh A:n:: I guess t'day wz d'day she's supposetuh find out if she goes in ner n o t . = = O h . Oh::. Becuz they're gonna do the operation on the teeuh duct. -ffi:rst. Before they c'n do the cataracts. N o I have my early class tihday et four thi:rdy. Tch! Except that cla- That class is suh:: yihknow, this is the indian class 'n- *ohhh They stuck us in this cra:zy building thet they juh - they're not even finished with it.

The unmentioned agents in the passives in the informal English data, on the other hand, are never of this type; here are three illustrative examples: (12) (13) (14)

Becuz the to:p wz ripped off ye iz car (Schegloff transcript) G o t - few shingles blown off of it (Schegloff transcript) I've been told that I was going to be an opthamologist an eye doctor like my daddy (Carterette and Jones)

The Passive in English: A Discourse Perspective

501

In such examples as these, the identity of the agent is clearly either inferrable or unimportant, as Givon suggests, rather than 'those typically in charge of such activities.' The discourse function of informal non-anaphoric they can thus be seen to overlap partially with that of agentless passives. Agentless passives, then, are essentially obligatory under certain conditions, these conditions having less to do with the structure of the discourse than with the encoder's intentions and assessment of the decoder's inferential capabilities. Passives with agents, on the other hand, are of greater interest from the point of view of discourse organization; we consider those next. 4. AGENTED PASSIVES

The second part of the strategy given in (1) for choosing a passive coding over an active coding deals with cases where the agent is mentioned. Here the issue of choice arises, since for an agented passive the same general message could have been expressed with an active. Under what conditions, then, does the language user choose the more 'marked' passive coding? Before turning to the proposed answer to this question, I would like to note again how infrequent the agented passive is: in formal written English roughly only 20 percent of all passives are agented, as mentioned above. The radio news and radio interviews, slightly less formal, yield 17 percent. In informal English the percentage is even lower. In the informal monologues in Working, only eight out of 56 passives, or 14 percent, had agents. Shintani (1979) found that out of 57 passives in the Carterette and Jones (1974) adult conversation transcripts none had agents, and I found that none of the five passives in the Schegloff transcripts was agented. Part Β of the strategy for using passives in English given in (1) consists of two parts; it is repeated here for convenience: If the agent is to be mentioned, then use the passive only when the non-agent is more closely related than the agent either B.l. to the 'theme' of the 'paragraph,' or B.2. to a participant in the immediately preceding sentence. Each of these factors provides a discourse motivation for the passive as a device which favors the non-agent over the agent for the role of grammatical subject. In other words, it appears that users of English are content to code the agent as subject unless broadly thematic or more local cross-clausal considerations require an alternative coding. These two sub-strategies are, of course, not unrelated: they are illustrations of the same cohesive principle of thematic unity, with B.l. simply operating across a larger domain than B.2. In other words, discourses are more cohesive (and presumably easier to process) the more their sentences have topics (in

502

Sandra Α. Thompson

English, subjects) which relate to the overall theme of the paragraph or to material in the immediately preceding clause. When a non-agent relates in one of these two ways more successfully than the agent does, the passive will be used. Each of these two factors accounts for about half of the 97 agented passives in my data. 3 Turning first to the question of the 'theme' of the 'paragraph,' 4 1 will make concrete the claim that the referent of the non-agent subject unequivocally either is, or is intimately related to, the 'theme' of the paragraph, or what the paragraph is 'about,' while the agent is more incidental to this 'theme.' As a first example, consider an article from the Christian Science Monitor entitled 'Estimates of Illegal Aliens Tumble: Has Threat of a 'Silent Invasion' Been Overstated?' (January 28,1982, p. 1). As the title suggests, the theme of the article is estimating the number of illegal aliens already in the United States and the number entering each year. In the sixth paragraph, the fact that the U.S. Immigration and Naturalization Service (INS) released an estimate that there were up to twelve million illegal aliens in the U.S. is introduced. The noun phrase the old INS figures appears again in the thirteenth paragraph; then paragraph seventeen opens with: (15)

The old INS estimate of 12 million illegals is still being used, however, by some in calling for legislation to curb illegal entries.

The subject of the passive clause, the old INS estimate . . . , is clearly part of the theme of the article, while the some who are calling for legislation is incidental to that theme. As another illustration of this point from news reporting, consider the article entitled 'Reagan's MX Plan - Budget Beats Bullets' (Christian Science Monitor, October 5, 1981, p. 1). If we take the headline of a news article as a label of the theme of that article, we find our hypothesis confirmed here as well. Of the twelve passives in this article, only one has an agent; that one perfectly supports the hypothesis: (16)

Reagan's decision to build 100 B-l bombers . . . is being viewed by some presidential watchers here as mainly a political move . . .

The opening sentence in the story had identified the 'MX plan' as including both an MX and a Β 1 component; the subject of the clause in (16), which opens the fourth paragraph, is, then, part of the MX plan, but the referents of some presidential watchers are not related to the plan, play no role in the overall goal of communicating the contents of the plan, and are not mentioned again. Working provides further examples. An occupational therapist, after talking at some length about what her job entails, says:

The Passive in English: A Discourse Perspective (17)

503

The kind of thing we do can be done by anybody in a general hospital, (p. 645)

The theme of the monologue is what it is that an occupational therapist in a hospital does. The subject of (17) is simply an overt reference to this theme. But the agent, anybody, is incidental to the theme; it is in fact non-referential, so there is no referent to appear again elsewhere in the discourse. The Working monologues are, naturally, overwhelmingly first person in theme; one thirty-nine-year-old man, in talking about the succession of positions he has held, uses almost exclusively sentences with first person subjects. The following example is another good illustration of the non-agent being more thematic than the agent: (18)

I was a young Columbia man while I worked in a cafeteria from 6:30 A.M. to 3:00 P.M. I was much respected by the management, even though I drove the people I worked with insane, because I had standards they couldn't cope with. (p. 666)

Finally, here are two examples from the radio news; the first opens a paragraph about the shooting death of an American military attache in Paris: (19)

Police in France are worried that this morning's shooting death of an American military attache is not an isolated incident, but part of an organized terrorist effort. Lieutenant Colonel Charles Ray was shot to death in front of his Paris apartment by a lone gunman, who fled the scene.

The motivations behind this murder and why it was Ray who was chosen are the themes of the remainder of the paragraph; the identity of the hit man is clearly incidental. The second radio news example occurs in this story about an airline crash: First the newscaster reports: (20)

Nearly 100 people were killed today when an Air Florida 737 jetliner crashed on take-off from Washington D.C.'s National Airport. The aircraft hit a highway bridge that spans the Potomac River, broke into two pieces, and fell into the water. The bridge was crowded with commuters on their way home in a heavy snowstorm. Six people on the bridge were killed, and it appears that most of the 73 aboard the plane also died.

Then the newscaster switches to the reporter on the scene:

504

Sandra Α. Thompson The Potomac River is lit by floodlights, but the search for bodies has been suspended for the night. The Potomac is covered with broken ice, which hampered the rescue, and is making the recovery tougher.

The scene of the crash, the Potomac River, is clearly more closely related to the theme of the story, namely the accident itself, than are the floodlights; hence the passive. The final example in this medium is a lengthy paragraph whose theme is articulated by the opening sentence: (21)

A mass of arctic air froze most of the nation again today with sub-zero temperatures recorded even in the deep South.

Several cities are named as subjects of sentences describing their cold conditions. The next paragraph continues in this mode, the general locale this time being Europe. Again, several places are singled out for mention; in the italicized agented passive, it can be seen that the subject, villages in Wales and northern France, is a part of the theme of places in Europe suffering from the cold weather, but snow drifts is non-thematic: (22)

Europe is going through its worst winter weather in 18 years. Thousands of miles of roads in Great Britain are impassable. Villages in Wales and northern France are cut off by snow drifts. British farmers have dumped thousands of gallons of milk because trucks cannot get through to collect it. In Salzburg, Austria, 28 inches fell within 24 hours. Over the past 4 days at least 23 deaths in Europe have been blamed on the weather.

In other words, if snow drifts were the subject of its clause and villages the object, the clause would violate both the structural parallelism and the listener's expectation that the thematic material, i.e., the places in Europe used as examples, will be in subject position. These examples serve, then, to support the claim that one of the two conditions under which an agented passive coding is chosen over an active one is the higher thematicity of the non-agent, as determined by the context of the paragraph in which the clause plays a role. As pointed out above, this condition accounts for about 50 percent of the agented passives in the data. Tomlin (1985), in discussing the relationship between information structure and clausal grammar in ice hockey sportscasts, has made a similar point. He observes that, given the goal of a play-by-play description of a game, the following 'thematic hierarchy' can be established: (23)

player with the puck > puck > player without the puck

The Passive in English: A Discourse Perspective

505

What this means for the grammar of the sportscast is that (24)

in each clause in which there are competing NPs for the subject relation, the NP which wins is the NP whose referent is highest on the thematic hierarchy . . . One repercussion of this for the grammar is the presence of a passive transformation to guarantee that the most thematic NP gets into subject. (1985:71)

One example of the use of the passive when the non-agent is thematic can be found in the following fragment: (25)

. . . Quick pass ahead to Errol Thompson, trying to work past Lapointe, it came back to Polonich. Polonich never got the shot away, checked by Lapointe. (1985:70)

Tomlin says: . . . initially the puck comes to Errol Thompson . . . While trying to skate past Lapointe, Thompson loses control of the puck . . . The next clause has a reference to the puck (//) and a reference to a player without the puck. The puck is the subject here, as predicted by the hierarchy. The videotape reveals that Polonich now has control of the puck. Note that the next clause contains a reference to both the puck {the shot) and a player with the puck. According to the hierarchy, the player with the puck should be the subject, and it is. The next clause describes what Lapointe, player without the puck, does to Polonich, player with the puck. Again, the NP highest on the hierarchy is in the subject, and in this case there is a passive clause. (1985:70-71) Tomlin found that the generalization in (24) accounts for all the passive clauses in his data; 5 his findings thus strongly confirm the first part of the strategy for using agented passives given above (B.I.): since the theme of the entire sportscast is to describe how the puck is manipulated by the players in such a way that one team is victorious, the passive is used when the non-agent is more closely related to this theme than the agent is.6 Sometimes, however, this condition is not fulfilled but the passive is nevertheless preferred. Here is an example from the radio news: (26)

As the first group of about 1000 Salvadoran soldiers began training at Fort Bragg, North Carolina, protesters outside the gates shouted slogans and carried signs. The soldiers are being trained by the Green Berets as part of a military assistance agreement between the U.S. and El Salvador. At a news conference today, protesters compared the

506

Sandra Α. Thompson plan to U.S. involvement in Vietnam, and said the U.S. is sitting on the edge of a volcano in El Salvador.

In this paragraph, which constitutes one complete item of news, it is simply not the case that soldiers is more thematic than Green Berets: the paragraph is about the protest. Why, then, is the passive chosen at this point in the discourse? The answer, as suggested in B.2. of (1) above, is that the non-agent, the soldiers, is more closely related (identical, in fact) to a participant in the immediately preceding sentence than is the agent. We might call this the 'interclausal anaphora' factor as a condition for the use of agented passives. Underlying this condition is the central principle of discourse construction: cohesive coding for easier processing. The greater the number of connections we can establish between identical or related participants, the more cohesive and continuous the discourse is, and, presumably, the easier it is to process. Coding non-agents as subjects in English allows a transition between two successive clauses without the potentially distracting and disruptive introduction of the agent. Let us consider some further examples of this function. Here are three from The Explorers of the Mississippi. In the first, following a paragraph about Henry de Tonti's father, Lorenzo, a new paragraph opens with: (27)

Lorenzo arrived in Paris as a down-at-heel political refugee without friends or money; luckily for him France at that time was ruled by an Italian, Cardinal Mazarin, in the minority of the twelve-year-old Louis XIV. (p. 145)

Clearly France is not thematic here: the paragraph is not about France, but about Lorenzo. What France does relate to, of course, is Paris, the focus of the preceding clause. A bit later, Severin describes a skirmish between the Illinois Indians, at the head of whom was Henry de Tonti, and the Senecas. Here is the relevant portion of the story: (28)

A young Seneca warrior, eager for glory, sprang forward and stabbed Tonti in the body. Luckily for the Italian it was a glancing thrust and the knife blade was deflected by a rib. (p. 153)

Here, the knife blade itself is not directly related to the theme of the paragraph, but it is obviously inferentially related to a glancing thrust in the preceding clause. The rib, needless to say, plays no significant role with respect to either the theme or the surrounding discourse. As a final example from the explorers' story, consider Severin's description of the retreat of Tonti and the Illinois from an Iroquois attack to the Jesuit mission of Green Bay:

The Passive in English: A Discourse Perspective (29)

507

Every few miles the travellers had to go ashore to repair the battered hull of their canoe, stitching the seams, caulking with pine gum or sticking on patches of cloth and bark. On one of these halts, the seventy-year-old Father Gabriel wandered off into the woods to pray and while he was thus absorbed, the gentle old man was surprised by a gang of marauding Kickapoos. (pp. 155-56)

Father Gabriel, who has not been mentioned for five pages, is incidental to the flight to Green Bay, which is the theme of the paragraph, but the cohesion provided by the coreferentiality of the gentle old man and he, the only nominal in the preceding clause, is obvious. One of the most striking ways in which the subject of the passive clause relates to a participant in the immediately preceding clause is in a reduced relative clause. Here is a paragraph with three such examples from a radio news item about the discovery of the Zimbabwe Patriotic Front's arms cache that led to the dismissal of Joshua Nkomo from his ministerial post: (30)

Some people suggest that Mr. Nkomo and three other Patriotic Front ministers may lose their portfolios, or that at least Mr. Nkomo will be sacked, and that other valued Patriotic Front ministers will be invited to stay in the cabinet if they join Mr. Mugabe's party. At the very least the government is expected to order the police to seize the thirty businesses and hotels, and more than twenty-five farms, owned by the Mitterand Company, a company set up by the Patriotic Front ostensibly to help rehabilitate former guerrillas from Mr. Nkomo's ZIPRA force. It was on farms in the Bulawayo area, owned by Mitterand, which had Mr. Nkomo as a co-director, that the national army and police found enough arms, ammunition, and military vehicles to equip 20,000 men, according to one source, but certainly at least a brigade of soldiers.

Each of the indicated passive participial clauses is dependent on the clause preceding it for the interpretation of its missing subject. Each of these clauses, in other words, is a reduced relative clause which is coded in the passive in order to render it a 'subject relative,' i.e., a relative clause in which the noun identical to the head noun is the subject. Subject relatives are highly favored over 'object relatives' in texts - and are, in fact, categorically favored in some languages (see Keenan and Comrie, 1972; and Givon, 1979:152) - because they allow for greater text cohesion in connected discourse when it is the non-agent rather than the agent which is identical to the head noun; passive coding is one obvious way to create subject relatives in these instances. In written English the cohesion can be enhanced by the economy of omitting the redundant subject in the relative clause, resulting in reduced relatives of the type brought out in (30) above.

508

Sandra Α. Thompson

The fact that subject relatives are preferred to object relatives is demonstrated for written English by a text count carried out by Barbara Fox. In 32 pages of Woodcock's Asia, Gods and Cities, she found 150 subject relatives but only 22 object relatives. But even more striking for the point I am making, when a patient is relativized it is twice as likely (46 to 22) to occur as the subject of a passive than as the object of an active clause; further, of these 46 passive relative clauses, the great majority (83 percent) are found in reduced clauses of the type illustrated in (30). It is clear, then, that reduced passive relative clauses function to ease memory and processing burdens by providing topic continuity between clauses in which the non-agent in the second clause is identical to a participant in the first. Such clauses are canonical examples of the second part of the strategy for choosing agented passive coding (B.2.), according to which the referential identity between the non-agent and a noun in the preceding sentence is a factor affecting this choice. An Associated Press news brief entitled 'Cory Out of Senate, Gubernatorial Races,' carried in the UCLA Daily Bruin March 4, 1982, page 2, provides another example. (31)

State Controller Ken Cory pulled out of the race for the Democratic nomination for governor Wednesday, announcing he will seek a third term as controller instead. Cory made his announcement unexpectedly in an address to a special joint session to the state Senate and Assembly on California's budget crisis. Cory had talked about challenging Los Angeles Mayor Tom Bradley for the gubernatorial nomination since late 1980, and in recent weeks he had been rumored as a possible challenger instead for the U.S. Senate seat being sought by Gov. Jerry Brown.

The missing subject of the reduced passive clause is the U.S. Senate seat, which is not the theme of the paragraph, but which is the focus of the preceding clause. The agent, Gov. Jerry Brown, by contrast, is not mentioned at all in the preceding discourse. Earlier I suggested that, of the two factors motivating the use of agented passives - thematic cohesion and interclausal anaphora - each accounted for roughly half of the 97 agented passives on which this study is based. In fact, as might be expected, there are a few examples (5 out of the 97) which could be interpreted as being governed by either factor. Here is one such case, from the travelogue Asia, Gods and Cities: (32)

The palace where King Phumipon was entertaining King Beaudoin

The Passive in English: A Discourse Perspective

509

lies on the slopes above Wat Sutep, and we drove from Chieng Mai to the foot of the mountain along an avenue hung with hundreds of fluttering flags. The mountain road, built thirty years ago by the voluntary labour of devout Buddhists, was a charming highway, serpentining sharply through a high jungle interlaced by cascading streams and full of the scent of tropical flowers, which, whenever we slowed for a turn, came drifting in through the car windows, (p. 138) The theme of this paragraph is the trip from Chieng Mai to the temple of Wat Sutep in northern Thailand; the (missing) subject mountain road of the reduced passive clause is clearly more closely related to that theme than is the agent the voluntary labour of devout Buddhists. This passive, then, counts as an instance of the thematic factor named in strategy (1) B.l. But it equally well illustrates the interclausal anaphora strategy in (1) B.2., in that it is a reduced relative whose subject is elided on identity with a participant in the immediately preceding clause. Thus the two strategies, as we can see, sometimes conspire in producing a passive clause. The pair of strategies we have been discussing sheds light on an observation made by Givon (1979:60, 63); he notes that the majority of passive agents in English are indefinite (90% according to his count), which indicates that the agent is part of the 'new information' imparted by the sentence. If 'new' is taken in Chafe's sense of'not in the consciousness of the addressee at the time of utterance' (Chafe, 1976:30), then examples such as (25) would seem to cast doubt on the validity of this inference: the passive agent Lapointe in that example, having been mentioned in the preceding sentence, cannot be far from the addressee's consciousness. Nor does it appear that we can take 'new' to suggest introducing a participant into the discourse which will continue to play a role in the subsequent discourse: numerous examples such as (28), where the passive agent is never mentioned again, would invalidate this interpretation. When these agented passives are viewed in the context of the surrounding discourse, though, it becomes clear that the tendency for passive agents to be indefinite follows from their relative lack of significance in the discourse (either in broad thematic terms or in local cohesion terms) rather than from their role in introducing 'new information'. The findings I have presented here support the intuitive observations on agented passives made by Jespersen (1965:168), who also suggests the following two related but independent uses for the English passive: a. b.

'the passive turn [i.e., voice - S.A.T.] is preferred if one takes naturally a greater interest in the passive than in the active subject' 'the passive turn may facilitate the connexion of one sentence with another.'

510

Sandra Α. Thompson

This study of passives in their natural contexts provides empirical substance to Jespersen's notions of 'taking naturally a greater interest in' the passive subject (cf. B.l.) and 'facilitating the connexion between' two sentences (cf. B.2.). 5. C O N C L U S I O N S

We have seen that an understanding of the English passive depends on recognizing that different communication needs underlie each of the two types: for agentless passives (A) the pragmatic consideration of the identity or significance of the agent is the crucial factor determining passive coding, while for agented passives (B.l., B.2.) the decision seems to be based on the discourse structuring principles of thematic and inter-clausal continuity. The passive in English has long been interpreted as a device for allowing a non-agent to be the subject of a clause (e.g., Fillmore, 1968:36-39; Halliday, 1967:215-218). By demonstrating that three empirically testable communicative principles stand behind the use of the passive, I hope to have shown precisely under what circumstances we may be called upon to opt for coding clauses in this way. ACKNOWLEDGEMENTS

It is a particular pleasure to contribute this paper in honor of Ilse Lehiste; I offer it as a small token of the esteem in which I hold her as my former teacher and mentor. I appreciate the valuable discussion of these ideas I have had with Wallace Chafe, Barbara Fox, Nikolaus Himmelmann, Paul Hopper, Robert Kirsner, Leland McCleary, William Mann, James Martin, Christian Matthiessen, Ellen Prince, and R. McMillan Thompson. The responsibility for form and content, however, rests entirely with me. NOTES

1.

2. 3.

Unfortunately, the editors of Working do not respond to my inquiries regarding the extent to which these responses have been edited. It is clear that pauses, false starts, and repairs have been edited out, but I have assumed that gross coding patterns such as active versus passive have not been tampered with. I am grateful to Professor Schegloff for the use of these transcripts; the original transcription has been preserved. There may well be cases in which the conditions for the use of the passive - as I have outlined them - are met, and yet an active is chosen. William Mann and Christian Matthiessen have suggested to me that one such case might be the following memo: (i) Ellen McConnell has cancelled her appointment with us on Wednesday morning. I have two other applicants coming in on Wednesday - if they look interesting I will try to schedule them to see you at that time.

The Passive in English: A Discourse

4.

5.

6.

Perspective

511

Strategy B.l. suggests that the italicized clause ought perhaps to be a passive, since the theme of the paragraph could be claimed to be the allocation of Wednesday morning, to which her appointment is more closely related than Ellen McConnell. Yet the passive was not used, and, in fact, our intuitions are that it would not have been particularly felicitous. However, it is important to note that this memo is interpretable only in the context of the larger enterprise of hiring a new secretary, which was assumed by both the sender and the receiver of the memo. It seems reasonable to propose that Ellen McConnell is in fact more closely related to this larger theme underlying the memo than is her appointment. By 'paragraph' here, following Longacre (1979:118), I mean simply the coherent stretch of discourse with thematic unity of which the sentence in question is a part. The term is intended to be neutral between written and spoken discourse. Interestingly, neither passive nor active is preferred when an offense man is shooting at the goal; Tomlin argues convincingly that this reflects the fact that in such a play, unlike plays involving bringing the puck down the ice, the relative thematicity of the goalie is about equal to that of the puck. As we will see, the passive in (25) can also be explained by B.2. of the strategy given in (1); it would be interesting to examine Tomlin's data to see to what extent our accounts would converge.

SOURCES O F DATA Carterette, E. and Jones, M. (1974). Informal Speech. Berkeley and Los Angeles: University of California Press. Severin, Τ. (1967). Explorers of the Mississippi. London: Routledge and Kegan Paul. Tannahill, R. (1974). Food in History. New York: Stein and Day. Terkel, S. (1972). Working. New York: Avon Books. Woodcock, G. (1966). Asia, Gods and Cities. London: Faber and Faber, Ltd. REFERENCES Chafe, W. (1976). Givenness, Contrastiveness, Definiteness, Subjects, Topics, and Point of View. In: Charles N. Li (ed.) Subject and Topic. New York: Academic Press, pp. 27-55. Fillmore, C. (1968). The case for Case. In: E. Bach and R. Harms (eds.) Universals in Linguistic Theory. New York: Holt, Rinehart and Winston, pp. 1-88. Givön, T. (1979). On Understanding Grammar. New York: Academic Press. Halliday, M.A.K.H. (1967). Notes on Transitivity and Theme in English, Part 2. Journal of Linguistics, 3.2, 199-244. Halliday, M.A.K.H. and Hasan, R. (1976). Cohesion in English. London: Longmans. Jespersen, Ο. (1965 [1924]). The Philosophy of Grammar. New York: W.W. Norton and Company, Inc. Keenan, E. and Comrie, B. (1977). Noun Phrase Accessibility and Universal Grammar. Linguistic Inquiry, 8.1, 63-99. Longacre, R. (1979). The Paragraph as a Grammatical Unit. In: T. Givon (ed.) Discourse and Syntax (= Syntax and Semantics 12). New York: Academic Press, pp. 115-134. Shintani, M. (1979). The Frequency and Usage of the English Passive. Unpublished Ph.D. dissertation, University of California, Los Angeles. Svartvik, J. (1966). On Voice in the English Verb. The Hague: Mouton. Tomlin, R, (1985). On the Integration of Syntactic Subject, Thematic Information, and Agent in English. In: J. Wirth (ed.) Beyond the Sentence: Discourse and Sentential Form. Ann Arbor: Karoma Publishers, pp. 61-80.

Middle Chinese Tones in Modern Dialects William, S-Y. Wang, University of California, Berkeley, CA., U.S.A. Chin Chuan Cheng, University of Illinois, IL., U.S.A.

1. INTRODUCTION

We are pleased to join in this occasion to honor our dear friend and esteemed colleague, Ilse Lehiste, whose contributions have enriched many distinct areas of linguistics. Our little offering here is in suprasegmental phonology, an area in which she has done pioneering work. Our aim is to discuss some aspects of lexical tones in Chinese from a general point of view, in the hope that these remarks will be of interest to linguists at large, rather than exclusively to the specialist in East Asian languages. Explicit discussion of lexical tones did not appear in the Chinese literature until Shen Yue (441-513 A.D.). The tradition started with Shen referring to the four historical tone categories as PING, SHANG, QU, RU, or 'level,' 'rising,' 'departing,' and 'entering' respectively. Through the centuries, philologists have attempted to provide various descriptions for these labels, such as 'hasty,' 'sad,' 'distant,' etc. It is clear that these terms, both the category names and their descriptors, cannot be taken literally. One reason is that systematic investigations into the phonetic properties of tones did not begin until this century. So the early philologists really had virtually no precise knowledge of the phenomenon they were discussing, and on which they were constructing principles of poetry and rime. Their labels, though useful heuristically, had no more physical basis than the singing teacher's 'chest register,' or 'pear shaped tone.' Another reason the traditional labels cannot be taken literally is that the values of the tones have changed a great deal over these 15 centuries - in different ways according to the dialect. Even though two dialects may have preserved intact a historical CATEGORY of tone, its modern VALUE may be quite different; e.g., it may be rising in one dialect while falling in the other dialect. We may take as an example the tones of Beijing and Jinan as shown below in Table I. Note that, although the two dialects have retained the identical set of historical categories, the tone values are quite different.

514

William, S-Y. Wang and Chin Chuan Cheng

Table I: Values of tones in t w o Chinese dialects

Iu Iv II III

Beijing level rising dipping falling

Jinan dipping high falling level low falling

For these reasons, it is preferable to use a set of semantically neutral labels for these historical categories. We will use the numerals: I, II, III and IV. Furthermore, the tones are intimately linked in their development with the initial consonants. We will refer to the voicing in these consonants with the letters u and v, for unvoiced and voiced respectively. Lastly, the u consonants are divided into η and a, for nonaspirated and aspirated respectively; and the ν consonants are divided into ο and s, for obstruent and sonorant respectiveiy· The traditional terms for the consonants are QING, meaning 'clear,' and ZHUO, meaning 'murky.' These are further divided into QUAN QING, CI QING, QUAN ZHUO, CI ZHUO, where QUAN means 'completely' and CI means 'secondarily.' These four divisions correspond to our un, ua, vo and vs respectively. As we will see, these consonantal features of voicing, aspiration and sonorancy have played an important role in tone development over the past 15 centuries. 2. H I S T O R Y O F T O N E S

It is generally believed that lexical tones emerged considerably earlier than Shen Yue's time, though it is difficult to date their emergence precisely. Based on a variety of comparative and philological evidence (including loan words from and to neighboring languages), some scholars have argued that tones II, III and IV were all marked by obstruent endings in early times. Haudricourt suggested in 1954 that tone III words ended in the obstruent s, which was a morphological suffix. Pulleyblank (1973) further hypothesized a glottal stop ending for tone II. According to Mei Tsu-Lin (1970), this glottal stop can still be heard in some contemporary Min and Wu dialects. Presumably, these endings were responsible for the emergence of tone in some way yet to be understood. Scholars are not agreed on the nature of the evidence that has been adduced for this view of tone emergence, nor on the dates for the emergence; see Ting (1981). In the modern dialects, however, the s and the glottal stop have all but completely disappeared. Only tone IV has preserved its obstruent endings to varying degrees in the southeastern dialects. Even these endings, notablyp, t, and k, are rapidly becoming lost, often through an intermediate value of glottal stop.

Middle Chinese Tones in Modern Dialects

515

Fig. 1. The 9 tones of Cantonese

3. CANTONESE TONES

Figure 1 illustrates the system in Cantonese, with a rich inventory of 9 tones. These pitch contours were extracted from the voice of an adult male from Hong Kong, and photographed from the screen of a Line 8 computer. Each tone is marked with a gloss. Furthermore, measurements are indicated at the position of the cursor for time (from beginning of frame), fundamental frequency, and amplitude (in arbitrary units). The 3 tones in the left column are Iu, IIu and IIIu, i.e., the initial consonants were unvoiced in Middle Chinese (ca. 600 A.D.). The 3 tones in the middle column are Iv, IIv, and IIIv, i.e., the initial consonants here were voiced in Middle Chinese, though they have merged into their unvoiced counterparts since that time. These tones all occur on the segmental sequence si. In other words, si pronounced with a high level tone means 'poetry,' with a rising tone means 'history,' with a falling tone means 'time,' and so on. The 3 tones in the right column derive from Middle Chinese tone IV, which has retained the obstruent endings in Cantonese. The segmental sequence here is sik. Tone IVu has 2 reflexes in Cantonese, illustrated by the tones for 'know,' and 'Wuxi.' Note that the tone IV syllables, being checked, are considerably shorter than the other syllables. Also, the fundamental frequency of the ν tone is always lower than that of the corresponding u tone in Cantonese. This is just another instance of the well-observed phenomenon that voiced consonants have lower pitches; see Wang (1967), Möhr (1971) and Cheng (1973).

516

William, S-Y. Wang and Chin Chuan Cheng

I

O ON £

00

3

I

3 rf

in '—'

ro

—

0 ON > rn Π1 —

Λ On 3 00 n->

—

ΪΛ Tf > ON CT* CN

I

I

m

I

I —

I ON ON

"

I CN COn

I oo — ON

Μ Ο —> CN

i n « ts — Ov

NO CN ON

Ο no > r— CN

cn r ON

η

cd ο 3 CN

NO m ON

a Tt

On

οΗ

I

rs —< CN i n ON

η

y S

I —• — r On

ι

Μ

-

I

Ο

I CN

CI ON

I OO ή ON

ΤΟΝ

I — On OO OO

r » cn On

OO

3 VO — Ο

I

I Ο

C CM 3 CN (N CN

m ON

I I — CN ON

Ο

Ό

«"> CN

On

vo - · ON

—

vo

Ο

I

I C I ON

r-

ι

ι —ι

I NO U — « Ν

ri

-s SP S ^ s χι >-ι — —

m

τι-

00

Ov

ON

C

ι

I

On

"a

I

es oo —I 00

VO CN cn c i

On

I

I OO OO 00

t-» ο

ι

— CN NO OO

VO CN — CN ON

1

Ο On

I

I —• Γ— ON

ι -

I — ON

I CN OO On

"—>

Ο Ο

I

I

I

Λ f 3 VO (N

ο

I cn

I Γ-

I —ι

I

( N Nf ON

I Ο ON —ι OO

§ s CN —ι

v> On

On

On

ι

a Η

ι

I

—

Ov —' oo 00

> ΓΊ crt -η

·£ c ο

ι CN

517

ι

-

I

ι

ι

ι

I

I

I

I

ι

g J2 ν 3 c « x i

ov OO

I

I

I

—

—·

I — VO Ο

I

vo rn

I

CN — vo

I

I

I

ι

rOv

tt

ι

ι

tN

ι

-«• ο

I

c>

I

m

w —

m t-

—

— 00

t

Μ ov

—

e n —ι

•«fr tn

Ο

4

00

CN

I

I

I

I —' —

η 3

vo in

I

I

I

I

I

c 3 Tf

— ν-> —

—

ι

I

I

—

ςη c Ο ο —

> θ\ m —

ο

ο Ον

Ι Ο

ι

> c d . O a l . ß c e X ) c d . O

^

—

η

ο

I

I

r oo in γί

I

τ* vo

Ο

S

ι

I

C CN 3 CN CS CN

η 3

3 •5 β Ο n

«Λ —i

m γ*ί Ον

Ν

r»-> —

I

—

I

Tf —«

VO ov

I

ι

r-1

(S -

η

-

η η ON

ι

" d

ι

I

Τ

m — Ov

00 es OO

ι

fA © > OO a \o 3 Ό

I

I

C -H 3 V> TJ- -H

—

> rr> ΓΛ -Η

— —

Ο

ON

Λ 3

ON OO

—1

> PI ΓΛ —
fN r -

I

I — v>

I

I

ο

y—I

—

ON

ι OO

On

—Η

Tf 00

I OO -Η ON

™

(Ν Ό Ο — oo

C fN 3 fN fN fN

fN

ON

I

ι VI fN ΟΝ

Μ fN > Ο — fN

fN I ON

Ο NO > r~ — fN

— tt — m — rfN

Λ

r— ι — - • — - • ι ON

NO — — — ON

oo Β cU υ c Λ -, Ζ

I t >o OS

I

I

—

I

I — ΙΟΝ

Ν

Nt i n ο ON

ΓΊ

->T

ON

OO

-

I

OO

Ο

NO

oo

Ν t

I

^

I

v i fN 00 NO 00

—

NO

-

—

I

Ι -

NO

- NO VI

I Ό Η

—. (-.

ι

ι

ON

I «

I

I

I fN fN NO VO f«Ί

I

I Ο fN 00 t— fN

1

ι

I fN

I

I fN

—

I

I

ι

ι

—> ι fN

I

Ο

I fN Ο ΓΊ

I

-

I Ifl Nf ON

Ο

I

I

ι

I

fN

I

I

I

ι

ι

ι

ι

V)

Ο

I

I

I

I

I

I

—

| τ* — V) 0C fN NO

I

I

I

I On fN Ό OO

I

I

I

I

η

ι

I

I

I

—

I

I

I

I

I

I

I

I

I

I

I

I -Η

I

I

I

I

Γ - —1 — Ο —I ON

I

I

I

I

I

«

I

—ι

ι

— r~ ON

_

ON

I

Ι ΟΝ

VI CI ON

I

ι r-

ON

Ο Νί η ON

• ON

Ο fN

M

I

I — —

fN

3

I

ON

I

I

3

ο J3 Ν 00 cΛ

c 2 'S — fN

ca χι ΓΊ Tt

S

.2

£ fN

3 cd ja O ca

-

ΙΛ

> Ti-

a 3

I

OS oo

Ο oo

ts

so v~t

ο m „ι

ο > ΓΊ

os ro —

rt 3 ΓΊ

os oo

I

Ο

Ο

Ο

00 on

ο

η

«

t

ιη on

«

ι

Π

Ν

«

00

«

Ή

Π

m

ο OS

c —

3 f l

so fN

> fN

OS

I

I

I

ρ»

m

fN in

ττ

ι/ι m

t

t

ρ

νο

^

I

I

I

oo oo

m

I

I

in

«ι oo

π

ι

τ}·

m

I

ο

fN

Ο

—

I

I

00

fN

Ρ-

00

I

I

fN

I

SO OO

Ό

«Λ

I

I

fN

ι

sf Os

m

fN Os

CO

Ο

fN OS

—

m

Λ 3 fN

Tt NO

C 3 fN

fN fN fN

—

>

fN Ο fN

Ο > —

so PfN

-tf-

cd 3

Ο fN

Ο Os

t

C 3 —

·

I

m

m On

—

ι

rn

ι

—c

Ν

ι

ι

so — — OO —1

Ο

I

—

•