Trubetzkoy's orphan: proceedings of the Montréal Roundtable ''Morphonology: Contemporary Responses'', Montréal, October 1994

TRUBETZKOY'S ORPHAN

AMSTERDAM STUDIES IN THE THEORY AND HISTORY OF LINGUISTIC SCIENCE General Editor E.F. KONRAD KOERNER (University of Ottawa) Series IV - CURRENT ISSUES IN LINGUISTIC THEORY

Advisory Editorial Board Henning Andersen (Los Angeles); Raimo Anttila (Los Angeles) Thomas V. Gamkrelidze (Tbilisi); John E. Joseph (Hong Kong) Hans-Heinrich Lieb (Berlin); Ernst Pulgram (Ann Arbor, Mich.) E. Wyn Roberts (Vancouver, B.C.); Danny Steinberg (Tokyo)

Volume 144

Rajendra Singh (ed.) Trubetzkoy's Orphan Proceedings of the Montréal Roundtable "Morphonology: Contemporary responses'

TRUBETZKOY'S ORPHAN PROCEEDINGS OF THE MONTREAL ROUNDTABLE "MORPHONOLOGY: CONTEMPORARY RESPONSES" (Montréal, September 30 - October 2, 1994) Edited by

RAJENDRA SINGH Université de Montréal With the collaboration of RICHARD DESROCHERS Université de Montréal

JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences — Permanence of Paper for Printed Library Materials, ANSI Z39.48-1984.

Library of Congress Cataloging-in-Publication Data Montréal Roundtable "Morphonology: Contemporary Responses" (1994: Montréal, Québec) Trubetzkoy's orphan : proceedings of the Montréal Roundtable "Morphonology: Contem­ porary Responses", Montréal, October 1994 / edited by Rajendra Singh : with the collabora­ tion of Richard Desrochers. p. cm. - (Amsterdam studies in the theory and history of linguistic science. Series IV, Current issues in linguistic theory, ISSN 0304-0763 ; v. 144) Includes bibliographical references and index. 1. Morphophonemics-Congresses. I. Singh, Rajendra. IL Desrochers, Richard. III. Title. IV. Series. P240.9.M66 1994 414-dc20 96-44621 ISBN 90 272 3648 8 (Eur.) / 1-55619-599-0 (US) (alk. paper) CIP © Copyright 1996 - John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O.Box 75577 · 1070 AN Amsterdam · The Netherlands John Benjamins North America · P.O.Box 27519 · Philadelphia PA 19118-0519 · USA

For Asim, Milan, and Sudesh, inlieuof what wastheirdue

Acknowledgements

The Roundtable this volume is mostly a record of took place on September 30-October 1-2, 1994 at the Université de Montréal. Its realization would not have been possible without the help of several institutions and individuals. For financial help, I am grateful to the Social Sciences and Humanities Research Council of Canada, my Vice-recteur à la recherche, my Vice-doyen à la recherche and my department. For administrative and technical help, I am grateful to Jean-François Couturier, Mohammed Seddiq Ennajih, Benoît Forest, Noureddine Kahlaoui, Diane Aubut-Moussette, Moussa Ndiaye and Param Randhawa. For moral support, I am grateful to my rector, Dr. René Simard, for finding the time to open the Roundtable, E. F. Konrad Koerner for the invitation to publish this volume in his series and for constant encouragement, and to Richard Desrochers for help, far beyond the call of duty, in the editing of this volume. Above all, I am grateful to the contributors to this volume for making a true debate on 'morphophonology' possible. Rajendra Singh

À côté de la phonologie, qui étudie le système des phonèmes considérés comme étant les idées acoustico-motrices, signifi­ catives dans une langue donnée, les plus simples, et de la morphologie, qui étudie le système des morphèmes, la gram­ maire doit comprendre encore un chapitre particulier, qui étudie I' utilisation morphologique des différences phonologi­ ques, et qui peut être appelée la morphonologie. N. S. Trubetzkoy (1929:85)

Contents Acknowledgements vii List of Contributors xiii Editor's Foreword 1 De l'autonomie de la morphophonologie: Discours d'ouverture 3 Etienne Tiffou , I. Allomorphy and Morphophonology Allomorphy or Morphophonology ? 13 Paul Kiparsky Where Does Allomorphy Begin? Comments on Kiparsky 32 Κ. P. Mohanan On the Morphology/Phonology Boundary: Comments on Kiparsky . . . 43 Douglas C. Walker Reply to Mohanan and Walker 48 Paul Kiparsky Discussion 55 II. Modularity, Morphonology, and Gradience A Functionalist Semiotic Model of Morphonology 67 Wolfgang U. Dressier Form & Content in a Functionalist Semiotic Model of Morphonology; Comments on Dressier 84 Richard Janda On A Functionalist Semiotic Model of Morphonology: Comments on Dressier 97 Douglas C. Walker Reply to Janda and Walker 102 Wolfgang U. Dressier Discussion 106 III. Linguistics without Morphophonology Quelques avantages d'une linguistique débarrassée de la morpho(pho)nologie 119 Alan Ford & Rajendra Singh Where Does Morphophonology Belong? Comments on Ford & Singh 140 K. P. Mohanan

XII

"Même après le débrouillement il peut rester de la brume": Comments on Ford & Singh Richard Janda Reply to Mohanan and Janda Alan Ford & Rajendra Singh Discussion IV. Morphoprosody Morphoprosody: Some reflections on accent and morphology Bernhard Hurch Another View of Prosody and Morphology: Comments on Hurch Glyne Piggott Reply to Piggott Bernhard Hurch Discussion V. Productivity and the Lexicon Productivity, Regularity and Fusion: How language use affects the lexicon Joan Bybee Productivity, Derivational Morphology, and Atypical Populations: Comments on Bybee Heather Goad A Reply to Goad Joan Bybee Discussion VI. Some Additional Contributions Issues in Morphophonology: A view from the floor Richard Desrochers On Morphophonology: A view from the outside Probal Dasgupta References Index

155 166 171 189 222 229 232

247 270 280 284 297 318 335 359

List of contributors Joan Bybee, University of New Mexico, Albuquerque, U.S.A. Probal Dasgupta, University of Hyderabad, Hyderabad, India. Richard Desrochers, Université de Montréal, Montréal, Canada. Wolfgang U. Dressler, Universität Wien, Vienna, Austria. Alan Ford, Université de Montréal, Montréal, Canada. Heather Goad, McGill University, Montréal, Canada. *Alexander Grosu, University of Tel-Aviv, Tel-Aviv, Israel. Bernhard Hurch, Universität Graz, Graz, Austria. Richard Janda, University of Chicago, Chicago, U.S.A. *Eva Kehayia, McGill University, Montréal, Canada. Paul Kiparsky, Stanford University, Stanford, U.S.A. *JürgenKlausenburger,University of Washington, Seattle, U.S.A. *lgorMeVcuk, Université de Montréal, Montréal, Canada. K. P. Mohanan, The National University of Singapore, Singapore. *Jean-Yves Morin, Université de Montréal, Montréal, Canada. *Yves-Charles Morin, Université de Montréal, Montréal, Canada. GlynePiggott, McGill University, Montréal, Canada. Rajendra Singh, Université de Montréal, Montréal, Canada. Etienne Tiffou, Université de Montréal, Montréal, Canada. Douglas Walker, University of Calgary, Calgary, Canada. The asterisk indicates oral contributions only

Editor's Foreword

This volume is a somewhat expanded and edited record of the Montreal Roundtable on 'morphonology' I had the pleasure of organizing in October 1994. Although it may not be entirely correct to say that Trubetzkoy invented 'morphonology' (cf. Dressler 1985a and Stankiewicz 1972), he certainly dressed it up when he put it up for adoption in 1929 under his signatures. Since then it has been something of an enfant terrible. The space it has occupied has grown or shrunk, with concomitant terminological upheavals, according to both time and place (cf. Dressler 1985a, Hooper 1976 and Martinet 1965, for example). It has always caused serious debate. It is, however, important to point out that, to mix my metaphors, the heart of the matter has always been the status of morphologically-bound alternations. As Martinet pointed out, the very second paragraph of Trubetzkoy's proclamation narrowed things down to such alternations, leaving the talk about the intersection or union of phonology and morphology as royal frills. Something of a temporary closure on the debate about 'morphonology', actually referred to as 'morphophonology' in this closure, resulted from the success of Halle (1959) and Chomsky & Halle (1968), at least in North America, and although all versions of generative phonology continue to subscribe to the Hailean view that morphonology is a part of phonology (cf. Mohanan 1995:24) the closure has not been in place since Stampe (1973), Vennemann (1974) and Hooper, and phonologists have seriously entertained all logical options, sometimes reminiscent of positions taken earlier by Bloomfield (1939), Kuryłowicz (1949) and Martinet. I thought it was a good idea to revisit the debate by inviting one advocate each for as many of the distinct contemporary positions vis-à-vis 'mor(pho)phonology' as I could see, two rapporteurs, to help us conduct that visit, for each of these advocates, and a few scholars to help us understand and interpret evidence from intimately related domains (morphoprosody, access and acquisition, and diachrony).

2

TRUBETZKOY' S ORPHAN

In the first part of the book (sections 1-3), Kiparsky articulates his current position on the phonology/morphology boundary position, Dressier sum­ marizes his well-known 'gradience' position, responds to its critics, and provides additional arguments for it, and Ford and I attempt to restore the traditional non-Pāņinian position that integrates morphologically-bound al­ ternations into morphology and provide some arguments in favour of our radically a-morphous morphology. These positions are evaluated by Walker, Mohanan and Janda, whose comments are, in turn, taken up in the replies by the advocates of these positions. In the second part of the book (sections 4-5), Hurch and Bybee look at morphoprosody and the lexicon, respectively. Their demonstrations and argu­ ments are commented upon by Goad and Piggott, responses to whose comments constitute the replies by Hurch and Bybee towards the end of theses sections. This section was also supposed to include the written version of Morin's paper on diachrony, but, unfortunately, does not because Morin decided not to publish the paper he presented. The third part of the book (section 6) contains Desrocher's "view from the floor" and Dasgupta's "view from the outside". I deeply regret Morin's decision not to publish his paper. It deprives us all of the opportunity of reading not only the written version of what he said about diachrony and the topic of the Roundtable but also several other things. It is my sincere hope that he will publish the final version of that paper somewhere, and sooner than later. There was much in it, but, given his perfectionism, we must wait to see that in what I am sure is going to be aricherversion. Each section, except the last one, is followed by a slightly edited transcript of the discussion that concluded each session. I consider that to be a very special feature of this volume. Montréal June 30 1996

Rajendra Singh

De l'autonomie de la morphophonologie: Discours d'ouverture Étienne Tiffou Université de Montréal 1. Introduction Rajendra Singh, l'organisateur de ce colloque, m'a demandé d'introduire le thème de cette rencontre, car il pensait qu'il avait une position bien définie et qu'il était préférable que cette tâche incombât à quelqu'un qui avait une vue extérieure de la question. Indo-européaniste et descripteur d'une langue non indo-européenne, je ne suis pas directement impliqué dans les débats théori­ ques qui animeront ce colloque. Beaucoup font toujours l'objet de discussions qui sont encore loin d'être épuisées. Cependant, parmi les problèmes soulevés par la morphophonologie, il en est un qui me paraît compter au nombre des plus importants: doit-on reconnaître ou non un statut autonome dans la grammaire à la morphophonologie? Pour répondre à cette question, je vais rapidement poser les données du problème. J'examinerai ensuite s'il y a un consensus sur ce point parmi les membres de ce colloque et je m'efforcerai dans un dernier temps de voir s'il est possible de reconnaître à la morphophonologie un statut en linguistique historique. J'ose espérer à tout le moins que les considérations sur ce dernier point pourront susciter quelque intérêt et qu'elles contribueront á préciser un ou deux éléments de la problématique qui fait l'objet de cette rencontre. 2. Les données du problème Il revient à N. Trubetzkoy (1957) d'avoir posé les bases de la mor­ phophonologie lors du premier colloque des slavistes tenu à Prague en 19291 Ce linguiste avait remarqué que les langues présentaient du point de vue phoni­ que des modifications combinatoires, certaines déterminées par le contexte, d'autres non, mais qui comportaient néanmoins un certain caractère systéma­ tique. La morphophonologie devait avoir pour tâche d'étudier de façon très rigoureuse ce dernier type de modification. N. Trubetzkoy entreprit donc d'é1 Le terme de 'morphophonologie' n'est pas cependant une création de Trubetzkoy. Il a été créé à la fin du XIXeme siècle par le néo-grammairien Henryk Uiaszyn (voir Singh 1986).

4

TRUBEIZKOY' S ORPHAN

crire, pour remplir ce programme, une morphophonologie du russe qui devait être suivie d'une phonologie de cette même langue rédigée par Jakobson. Or ce dernier n'a jamais réalisé ce projet et l'on peut légitimement penser que ce n'est pas un simple hasard. On peut supposer que les chevauchements entre cette nouvelle discipline et la phonologie condamnaient nécessairement à des redites et vidait cette matière d'une bonne partie de sa substance. Au reste, Martinet (1965) met en doute la légitimité de cette nouvelle discipline. Quant à lui, l'instauration d'un tel domaine en linguistique ne s'im­ pose pas. Certes, nombreux sont les cas que l'on peut relever dans une langue qui posent des problèmes dont la solution ne relève pas exclusivement de la phonologie (par exemple les cas d'umlaut en allemand, comme /u/ ~ /ü/, /o/ ~ /Ö/, /a/ « /ε/); mais par ailleurs la plupart des cas sont si variés qu'il est plus convenable de les répertorier dans la morphologie; dans ces conditions les variantes généralisables, mais peu fréquentes, ne justifient plus la création d'une branche spéciale de la linguistique. On peut comprendre d'autant mieux ces réticences que, pour Martinet, un domaine comme la morphophonologie qui s'intéressait à la fois à des phénomènes relevant de la première et de la deuxième articulation, pour employer sa terminologie, avait quelque chose d'hybride, et de ce fait créait plus de problèmes qu'il n'en résolvait2. Bloomfield (1939), qui ne se démarque pas fondamentalement de la position de Trubetzkoy, quoique la postulation de représentations donne à sa démarche un pouvoir explicatif beaucoup plus fort, est sujet aux mêmes critiques. La position de Chomsky et Halle (1968) resitue le problème en incluant la morphophonologie dans la phonologie. Les entrées sur lesquelles s'appliquent les règles phonologiques offrent toutes les informations morpho­ logiques nécessaires pour que soit assurée la bonne application de ces règles. Les règles 'phonologiques' prennent en compte les entrées comportant le marquage des frontières morphologiques; elles vont jusqu'à spécifier, quand il y a lieu, des catégories syntaxiques ou morphologiques. Ainsi, en grec ancien, pour rendre compte des jeux de l'accent, on donnera une spécification diffé­ rente aux noms et aux verbes. En effet, dans les premiers l'accent est lexical; dans les seconds, en revanche, la place de l'accent est déterminée par la longueur de la voyelle finale. La position de Chomsky et Halle est donc claire: la morphophonologie est intégrée dans la phonologie de telle sorte qu'il ne saurait y avoir, dans cette perspective, de place pour la morphophonologie. 2 Pour Luc Bouquiaux et Jacqueline Thomas (1976), qui travaillent dans l'optique de Martinet, en l'adaptant à la spécificité de leurs études, la morphophonologie intéresse le domaine des formes sous-jacentes d'un paradigme dans la mesure où il permet de dénoter la régularité de celui-ci après qu'elle a été voilée par l'application des règles phonologiques.

DE L' AUTONOMIE DELAMORPHOPHONOLOGIE

5

3. Les tendances actuelles Qu'en va-t-il des tendances actuelles? La morphophonologie peut-elle pré­ tendre, selon elles, à un statut autonome? Il est délicat de répondre à cette question, puisque, d'une part, les membres du présent colloque seront appelés à prendre position sur ce problème, soit explicitement soit implicitement. D'autre part, il est présomptueux dans un discours d'introduction nécessai­ rement succinct, de faire une analyse de la position de spécialistes qui sont euxmêmes en mesure de préciser, mieux que n'importe qui, leur pensée. Toute­ fois, après avoir pris connaissance du thème de leurs communications, il m'apparaît qu'un certain consensus peut être dégagé. Il ressort de celles-ci qu'il est difficile de considérer la morphophonologie comme un domaine totale­ ment autonome de la linguistique. Ce consensus, à vrai dire, n'implique pas une similarité de points de vue. Je distinguerai deux attitudes dominantes: il s'agit pour les uns de traiter les problèmes auxquels devait s'attaquer la morphophonologie de façon exclu­ sivement linguistique. Ainsi pour l'un des participants ce type de problème relève de la phonologie, et d'elle seule; il donne pour soutenir cette position sept raisons qu'on ne saurait prendre à la légère. Pour un autre, l'étude d'un problème ne relève pas nécessairement d'un domaine unique. L'étude de l'accentuation, par exemple, ne relève pas exclusivement de la phonologie. Selon les cas, c'est tantôt à elle qu'il faudra faire appel, tantôt à la morpho­ logie. Pour un troisième participant, il est vain de considérer les différents domaines de la linguistique comme étanches. Nous avons affaire à un conti­ nuum où ceux-ci, quand ils sont connexes, peuvent interférer. C'est à une de ces zones d'interférence que se situent les problèmes pris en charge par la morphophonologie; cependant cette zone ne constitue pas un domaine à part. Il y a lieu, dans ces conditions, de considérer la morphophonologie, non pas comme une branche autonome de la linguistique, mais comme un système de régulation qui gère divers mini-systèmes. La seconde attitude fait intervenir des facteurs cognitifs que l'on peut consi­ dérer comme extra-linguistiques. De ce fait, on ne peut plus parler de morpho­ phonologie, si celle-ci se propose de traiter les problèmes qui se posent à elle de façon uniquement linguistique. Selon l'optique de ces chercheurs, le sujet parlant finit par avoir dans son esprit des modèles auxquels il se conforme pour engendrer de nouvelles formes. Pour rendre compte des formations de mots et de leur caractère aléatoire, il convient de faire intervenir parallèlement à la grammaire acquise les capacités cognitives des sujets parlants. Certes celles-ci interviennent dans d'autres domaines de la linguistique, mais, dans celui-ci, la part qui leur revient est si importante que l'on comprend pourquoi tout effort de

6

TRUBETZKOY'S ORPHAN

théorisation se heurte à des obstacles résistants. Deux des participants justifient ce choix de manière convaincante. L'un d'entre eux remarque, en effet, qu'il n'y a pas de corrélation entre la régularité et la productivité. L'autre montre que si l'analogie proportionnelle peut être intégrée dans un processus morpho­ logique, il n'en va pas de même pour l'analogie non proportionnelle. Celle-ci suppose "l'existence de correspondances indépendantes des processus mor­ phologiques". Il est donc fondé de faire intervenir dans ce qui fait l'objet de la morpho­ phonologie des stratégies cogniüves. Les tenants de cette approche ont des po­ sitions variées et nuancées. Il ne saurait être question de les exposer en détail. Elles résident surtout dans la place et le rôle qu'ils attribuent respectivement à la grammaire acquise. Les discussions au cours du colloque permettront de les dégager mieux que je ne saurais le faire. Toutefois, il ressort, selon moi, que l'analyse gagne en adéquation à la mesure de l'importance accordée aux facultés analogiques. Le sujet parlant a dans la tête des mots qui lui servent de modèles pour en former d'autres. Le caractère aberrant de ces formations s'ex­ plique fort bien du point de vue historique. Du point de vue synchronique, elles résistent à l'analyse, ce qui ne les disqualifie pas pour servir de moule à des formes de même type. La grammaire apparaît, dès lors, moins comme un processus de formation que comme l'expression d'une tendance qui s'établit au fur et à mesure que le processus analogique valide tel ou tel type de formation. Ainsi le type de formation ornemaniste sur ornement tend à être moins fréquent que le type instrumentiste sur instrument. À vrai dire, les deux attitudes qui viennent brièvement d'être évoquées ne sont pas inconciliables et peuvent peut-être se compléter. En effet, dans le cas des formations analogiques, il y a peut-être lieu de considérer que, dans bien des cas, les contextes phonologiques eux-mêmes devraient permettre de choisir entre différents moules de formation. Il y a là toute une voie à explorer, et certains sont en train de le faire. Quoi qu'il en soit, les confusions auxquelles la morphophonologie se prêtait invitent à ne pas lui concéder un statut autonome. À cet égard, la remarque d'un des participants est éclairante. Selon celui-ci, la fusion d'une opération morphologique et d'une stratégie cognitive implique que celles-ci ne peuvent aller l'une sans l'autre. Il tire de cette observation un argument de poids pour réfuter l'existence d'une composante morphophonologique autonome. Les tendances que nous avons cru déceler dans les commu­ nications qui vont être données, même si elles ne se fondent pas sur cet ar­ gument, me paraissent aller dans ce sens. Il semble donc, dans ces conditions, préférable de renoncer à reconnaître un statut autonome à la morphophono­ logie en linguistique synchronique. Dans le meilleur des cas, la morphophono-

DE L' AUTONOMIE DELA MORPHOPHONOLOGIE

7

logie apparaît plus comme un principe de gestion qui détermine le fonctionne­ ment des mécanismes. En ce sens, le débat y gagne en clarté; la morphophono­ logie doit être considérée moins comme un domaine de la linguistique que com­ me un principe dynamique qui régule les rapports complexes entre la phonolo­ gie et la morphologie, que l'on fasse ou non appel à une stratégie cognitive de formation des mots. 4. Morphophonologie et diachronie Il faudrait donc renoncer à reconnaître un statut autonome à la morpho­ phonologie. La cause de cette discipline est-elle désespérée? Il y a une solution paradoxale à ce problème. Il est clair que Trubetzkoy a été appelé à fonder ce domaine pour répondre à des difficultés qui se posent en synchronie. En effet, la plupart des formes aberrantes peuvent trouver une justification historique, mais celle-ci ne saurait rendre compte des mécanismes de formation à un mo­ ment donné de la langue. Toutefois la morphophonologie a été prise en consi­ dération dans les études diachroniques. On connaît notamment les travaux de Bybee (1980) dans ce domaine. Mais, en dépit de leur intérêt indéniable, ces réflexions assignent, à mon avis, à la morphophonologie la même visée qu'en synchronie et l'on est, en conséquence, amené à lui reconnaître un statut simi­ laire. On peut se demander néanmoins si, en déplaçant la visée de la morpho­ phonologie, il n'y a pas moyen de lui reconnaître un statut autonome. 4.1 Les données Dans l'étude historique des langues on trouve bien des cas qui résistent à l'application des règles de la phonétique historique. La plupart de ceux-ci inté­ resse des morphèmes d'usage fréquent. Ainsi en latin, on admet communément que la marque en -ba d'imparfait remonte à un thème *bhwəı issu de la racine suffixée indo-européenne *bh(e/o)w-(e/o)oı "croître, se développer". Le mor­ phème *-bh w ə1 -am aurait dû passer à *-buam; en effet la séquence bu + [+syll] est tout à fait admissible en latin (ex.: imbuō; imbuistī etc.). Or on obtient le morphème -barn. Certaines formes du relatif-indéfini dans la même langue posent des problèmes analogues. Comment expliquer le passage de quoius qui est attesté à cujus ou de quoi, également attesté, à cui? Le cas de ubi est encore plus curieux. Les linguistes s'accordent à le faire venir du thème de relatif-indéfini, encore qu'ils s'interrogent sur le thème exact à postuler. C'est ce qui ressort de l'analyse de Monteil (1970:238-39) : Le thème de ubi < *qubi, plutôt que *k w o-, doit être un thème apparenté *kwu-, que l'on retrouve sporadiquement en grec (crétoisðrruid'où vî "là où") et, avec allongement, dans védique kứ.

8

TRUBETZKOY' S ORPHAN

Quoi qu'il en soit, la perte de l'occlusive et la vocalisation de l'appendice vélaire est difficilement explicable. Ce genre de phénomène n'est pas propre au latin. Nous emprunterons un exemple significatif au grec moderne. On sait que la marque du futur dans cette langue est le morphème θα suivi de l'infinitif. Ce morphème remonte à θελεί lvá ("vouloir que", "dans l'intention de"). L'évolu­ tion de θελεί lva à θενά n'est pas problématique (θέλει- Lva > θέλει va > θελνα > θενά), en revanche le passage de θενα á θα )θενά > θανα> θα) reste peu clair. En outre, il est des phénomènes artificiels dans la langue, comme les licences poétiques, qu'il est difficile d'expliquer par des processus réguliers. Le cas de la diectasis en donne un bon exemple. Le terme homérique κομόωντες ("chevelus") est une forme épique qui représente un compromis entre κομάωντες (forme non contracte) et κομώντες (forme contracte). La forme non contracte ne pouvait entrer dans un hexamètre puisqu'elle com­ prenait trois brèves. La distension rend la forme admissible au vers. Ce genre d'exemple est intéressant car il témoigne de formations visant à modifier la structure phonique d'un mot, et non à fournir une nouvelle information. 4.2 Pour une morphophonologie diachronique Ce type d'évolution dont il est difficile de rendre compte par des règles historiques et qui, en aucun cas, ne peut s'expliquer par l'analogie, pourrait être considéré comme relevant de la morphophonologie. Celle-ci intervient bien, en effet, dans un domaine dont la nature ne diffère pas de celle du domaine qu'on lui assignait en synchronie. Il s'agit effectivement de traite­ ments phoniques opaques; la différence réside en ceci qu'elle ne prend pas en considération des traitements productifs, mais des suites de cas particuliers. La morphophonologie devient alors une taxinomie de phénomènes phoniques non explicables par des règles historiques. Elle risque, dans cette mesure, de n'être qu'une étiquette commode pour les linguistes et de n'être qu'un écran de fumée masquant un constat d'échec : l'impossibilité de trouver une explication con­ vaincante soit faute de documents permettant de reconstituer les chaînons manquants, soit faute d'une théorie capable de résoudre ce genre de difficulté. Il peut en aller ainsi parfois, mais le plus souvent il faut bien se résigner à admettre que le traitement résiste à l'analyse. Le camouflage en faisant appel à la morphophonologie est peut-être décevant, mais il permet d'isoler un groupe déviant, ce qui permet du même coup de dégager la régularité des traitements auxquels est soumise le reste de la population. Il est loisible cependant de ne pas cantonner la morphophonologie dans un rôle aussi restreint. On remarquera que ces traitements déviants intéressent des morphèmes appelés le plus souvent à être utilisés fréquemment. Il y a là soit un

DE L'AUTONOMIE DELA MORPHOPHONOLOGIE

9

phénomène d'usure, soit une contrainte sur la longueur des morphèmes dérivationnels ouflexionnels,qui peut peser sur les mots eux-mêmes (On citera à titre d'exemple le traitement du nombre "quatre" en grec ancien où il est attesté sous la forme - dans τρνøάλεία "casque de cuir à quatre pointes" et sous la formeTdans τράπε£α 'table", "meuble à quatre pieds"). La morpho­ phonologie pourrait avoir parallèlement pour tâche de déterminer ces con­ ditions et de dégager les règles particulières auxquelles la langue recourt dans ces cas. Son statut trouverait dans cette mesure, à mon avis, un fondement im­ portant. Il reste à justifier cette position. Pourquoi l'autonomie de la composante morphophonologique pourrait-elle se justifier plus aisément en diachronie qu'en synchronie? Parce que la tâche assignée à la morphophonologie porte sur des cas uniques ou sur les déclencheurs d'une stratégie de formation de mots dans la mesure où ceux-ci résistent à l'application des règles historiques. Il va de soi que si les mots ou les morphèmes sont par la suite à la base de stratégies de formation d'autres morphèmes ou de mots, comme c'est le cas par exemple de la diectasis, à ce moment d'autres démarches interviennent. La morpho­ phonologie peut donc être une composante autonome si l'on change la visée qui lui était proposée. Il est difficile, en revanche, d'opérer ce changement en synchronie puisque celle-ci se propose de rendre compte de la dynamique de la formation des mots dans un état de langue donnée. 5.

Conclusion Au cours de cette brève présentation, il est apparu que la morphopho­ nologie considérée, lors de son institution, comme une branche autonome, ris­ quait de créer en synchronie plus de problèmes qu'elle n'en résolvait. Il est ressorti que les difficultés qu'elle prétend résoudre trouvent leur solution dans d'autres approches, et notamment dans une approche cognitive. Il faut donc considérer que le statut qu'on lui a reconnu, lors de sa création, en synchronie ne s'impose pas. Paradoxalement, alors que dans l'esprit de ses fondateurs, elle n'était pas pertinente en diachronie, c'est dans ce domaine qu'elle risque d'être vraiment justifiée à condition de déplacer la visée qui lui est assignée. La conclusion que j'avance est paradoxale, mais on peut légitimement penser qu'on a plus à gagner qu'à perdre en réajustant ainsi la place qui doit être dé­ volue à la morphophonologie. J'en laisserai l'appréciation à ceux qui ont étudié de façon approfondie ce genre de problèmes, car c'est à eux plus qu'à qui­ conque de décider comment disposer de ce que R. Singh appelle Torphelin de Trubetzkoy'.

I

Kiparsky, Mohanan, and Walker ALLOMORPHY AND MORPHOPHONOLOGY

Not only have several declensions as well as conjugations been imposed on the [Finnish] language, but the morphology has even been decorated with irregularities, anomala, and defectives, perhaps from a zeal to match other languages in these grammatical treasures, or partly from necessity due to a neglect of the study of the affinity of letters, of their mutual transitions, of consonantal fusions, in a word, of phonology in general. Elias Lönnrot (1841-42, from Wiik 1990).

Allomorphy or Morphophonology? Paul Kiparsky Stanford University 1. Locating the Morphology/Phonology Boundary According to classical grammar and traditional school grammar, words are inflected by modifying the shape of a basic form; the results are described by listing model paradigms and assigning each word to the appropriate paradigm. Modern morphology modifies this 'word-and-paradigm' model by one or both of the following ideas, both adopted in the early 19th century from Panini's grammar of Sanskrit. The first is that morphology is a combinatorial system and that words are built from minimal morphological elements (morphs, mor­ phemes). The second is to make a distinction between two sorts of contextual alternation of morphological form, morpholexical (allomorphic) and morphophonological (morphophonemic) alternation, in terms of different kinds of al­ ternation processes and/or different modules of grammar responsible for the alternation. Since the two are independent of each other, we get four theoretical positions:

(D

No morphemes Amorphous morphology No morphophonology Item-and-Arrangement Word-and-Paradigm Morphophonology

Morphemes Item-and-Process

In this paper I assume, with Item-and-Process and A-Morphous morphology, that there is a distinction between morphophonology and allomorphy, and examine the criteria by which it is drawn. Like any taxonomy, its theoretical import comes from the general principles from which they are derived, which associate each type with a set of defining properties. The tradition of Item-and-Process morphology (Pānini, Kruszewski, Baudouin de Courtenay, Sapir, Bloomfield's Algonquian studies, Jakobson 1948, and much current work, e.g., Dressler 1985a) distinguishes three types of rules: (i) morpholexical rules (also known as adjustment rules, or

14

TRUBEIZKOY'S ORPHAN

allomorphy rules), (ii) morphophonological rules (morphophonemic rules, Bloomfield's 'morphological modification'), and (iii) phonological rules (Bloomfield's 'internal combination'). In most versions, morphophonology and phonology have narrowly circumscribed properties, with morpholexical rules often serving as a 'catchall' category for everything that is not tractable within the others. In Lexical Phonology and Morphology (LPM), a related though not identical three-way distinction emerges between (i) allomorphy (negotiated in the morphology), (ii) lexical phonology, and (iii) postlexical phonology. The leading idea of this approach is that the properties of these rule types can be derived from the organization of grammar depicted in (1).

In particular, morphology shares certain principles with syntax on the one hand (including combinatorial principles governing subcategorization, feature perco­ lation, and unification), and with other lexical rules on the other (mechanisms for lexical specification of idiosyncrasy, blocking, level-ordering, cyclicity). Lexical phonology is similarly governed by the intersection of phonological principles and those which hold for lexical rules in general. According to this approach, the properties of morpholexical alternations should be consequences of their morphological status, and the properties of morphophonological rules should be consequences of their lexical phonological status. Modern 'amorphous' morphology (Matthews 1972, Zwicky 1986, An­ derson 1992) is in principle compatible with the same rule typology. It differs from the abovementioned approaches on another dimension, in that it retains — in a more sophisticated form — the classical word-and-paradigm model's assumption that words are formed not by combining morphemes but by ap­ plying morphological processes to a base. In Zwicky's formulation, each such process consists of one or more operations, which add phonological material (e.g., affixation), delete it (subtraction), or modify it (e.g., ablaut). Nevertheless, unlike the classical Word-and-Paradigm model, most

ALLOMORPHY AND MORPHOPHONOLOGY

15

versions of this theory assume that the output of morphological processes is subject to morphophonemic rules which account for general alternations of word structure. Therefore, although I will not specifically refer to amorphous morphology, the issues addressed here potentially arise in that framework as well, in ways that depend on the details of its articulation. On the other hand, my discussion is squarely incompatible with models which reject the abovementioned rule typology and treat all morpheme alternations in a uniform way, viz. pure IA on the one hand (post-Bloomfieldians, Hudson 1980), and clas­ sical WP on the other. Terminological differences aside, there are many cases where the three-way division is unproblematic or at least uncontroversial: morpholexical rules deal with suppletive, morpheme-specific alternations such as go ~ went, (box-)es ~ (ox-)en, morphophonemic rules deal (at least) with "general phonological changes which occur only in particular derivations" (Bloomfield), such as keep ~ kep-է, and (iii) phonological rules (Bloomfield's "internal combination") deal with automatic phonologically conditioned alternations such as back-[s] ~ bag-[z]. The picture is enriched by the postulation of additional rule types and more fine-grained distinctions among morphophonological rules (cyclic and word-level rules, level 1 and level 2 rules) and among phonological rules (allophonic rules, rules of phonetic implementation). The validity of the funda­ mental morphology/phonology distinction is of course not called into question by the discovery of further subdistinctions within each, and for purposes of the present discussion, it will be possible to abstract away from them. Once we look beyond the core cases, however, it is not always clear where to draw the boundaries between the three major rule types. There are important disagreements between the different versions of IP morphology, and to some extent also unclarities within some of them, about how to apportion the phe­ nomena. The distinction between morphophonological and phonological rules, and the roughly corresponding distinction between lexical and postlexical phonology, has been debated fairly extensively (see Kiparsky 1985 and Mohanan 1986 for two views in the LPM framework) and will not be addressed here. The distinction between morpholexical alternation (allomorphy) and morphophonology (or lexical phonology) has received less scrutiny, at least in the LPM literature. The topic of the present paper will be precisely this distinction. The main difficulty with drawing a principled boundary between morpho­ lexical and morphophonological phenomena is that the various criteria which have been considered relevant to the distinction, while nicely converging in the core cases, diverge in different ways at the margins. This situation is not un-

16

TRUBETZKOY' S ORPHAN

usual in linguistics. Two obvious solutions present themselves: either we select one criterion or conjunction of criteria as necessary and sufficient and deny the rest a theoretical role — as Aronoff (1976) does in positing morphological conditioning and morpheme-specific application as the criterial properties — or we retain all criteria but treat the distinction as an essentially gradient one — as Dressier (1985a) does.1 Both types of solutions pay the price of giving up a clean theoretical explanation of the convergence observed in the core cases. I will here explore a third solution, based on the observation that it is primarily one of the putative criteria for identifying allomorphy which tends to fall out of line, and that this same criterion is also theoretically unmotivated. When it is eliminated from the diagnostics, the remaining ones converge in an empirically fairly consistent way, which, I will argue, allows a principled explanation. While this obviously cannot claim to provide an automatic resolution of every problematic borderline case, it does at least reduce the indeterminacy substantially. The criterion which I propose should be eliminated is one which has often received the greatest emphasis, though for largely unexamined reasons, namely the contextual conditioning of the process. Many authors hold that morpho­ logical or lexical conditioning is a necessary condition for an alternation to be morpholexical. For example, Aronoff (1976:87) defines adjustment rules as those which "are restricted to specific morphemes and take place only in the environment of specific morphemes". Some even argue that morphological conditioning is sufficient to make an alternation morpholexical. For Ford & Singh (1983) and for Spencer (1991, section 4.4), all morphologically con­ ditioned rules are morpholexical. Anderson (1992:346) seems to be taking a similar position when he states that "once a rule is morphologized in some environment, the result is a rule whose formal properties are those of a Wordformation Rule, and not those of a purely phonological rule". I shall defend the contrary view that morphological conditioning is neither necessary nor sufficient to render an alternation morpholexical (or morpho­ logical). There are both purely phonologically conditioned morpholexical alternations, and conversely, morphologically conditioned phonological rules. Neither of these claims is new, of course. What I hope to do here is to support them with some new arguments, and to show that by taking both points seriously we can achieve a fairly tidy and theoretically principled separation

ALLOMORPHY AND MORPHOPHONOLOGY

17

between morpholexical and morphophonemic alternations. On the view I shall argue for, the essential criteria have to do with the nature of the alternation, the locality relation between the focus and the trig­ gering context, and the relationship of the process to other rules of the system. Morpholexical alternations are accounted for by the combinatorial mechanisms of the morphology, which for reasons stated below I take to be based on selection, following Lieber (1982) and Zwicky (1986). This entails (i) that morpholexical alternations are idiosyncratic (item-specific), (ii) that they may involve more than one segment, (iii) that they obey morphological locality conditions, and (iv) that they are ordered prior to all morphophonemic rules. Thus, the core properties traditionally attributed to 'allomorphy' follow trivially from assuming that 'allomorphy' is simply morphology, and rejecting the structuralist picture on which the selection of 'morpheme alternants' is distinct from the selection of the morphemes themselves. If this is the right view of the matter, these four properties should bundle together in a systematic way, whereas there is no real reason to expect them to be related to the nature of the conditioning of the alternation. If it were in fact the case that morpholexical alternations were limited to the environment of specific morphemes, this would have to be imposed as an unexplained super­ venient constraint. The truth seems to be simpler, though: just as a 'morpheme' can be restricted to a particular phonological environment, so can an 'allomorph'. In other words, morpholexical alternations can be phonologically conditioned. Morphophonemic alternations, on the other hand, are accounted for by phonological rules, applying subject to constraints on the lexical module. In consequence, they are (i) general (not item-specific), (ii) involve a single segment, (iii) observe phonological locality conditions, and (iv) follow all morpholexical processes. All these core properties of morphophonological rules follow from the assumption that they belong to (lexical) phonology. Once again the nature of their conditioning environment falls out of line. And here too there are good independent reasons to believe that the conditioning factor is not a decisive criterion: morphophonological rules can be conditioned by specific morphological elements or features, or by phonological environments, or by a combination of the two. At this point let us recall that two fundamentally different ways of accounting for morpholexical alternations have been advocated in the literature: selection and replacement. The selection method (Lieber 1982, Zwicky 1986) treats morphemes as listed sets of alternants, with one basic (default, 'else­ where') alternant, and the other alternants lexically marked as restricted to

18

TRUBETZKOY' S ORPHAN

appropriate contexts. The Elsewhere Condition or its equivalent ensures that each context gets the most specific alternant compatible with it, and the basic unmarked allomorph only appears where not blocked by any of the others. The replacement method (Aronoff 1976, Corbin 1987) posits a single underlying form for each morpheme, and allomorphy rules which substitute for it other allomorphs (or zero, in the case of 'truncation') in the appropriate contexts. On the replacement approach to morpholexical alternation, all selection takes place at the level of the morpheme, and there is a fundamental difference between the way the distribution of a morpheme is determined and the way the distribution of its allomorphs is determined. For this approach, the morpheme is thus an essential theoretical construct, and morphological elements are listed in the lexicon as morphemes. For the selectionai approach, on the other hand, lexical entries are morphs. In that approach, the 'morpheme' could be characterized derivatively as a set of morphs in a blocking relationship, though it is not so clear that there remains any theoretical need for that concept. One key prediction of the selectionai approach is that 'morphological conditioning' can be triggered by particular morphs (stems or affixes), and/or by morphological categories (feature complexes), but not by 'morphemes', in the structuralist sense of sets of '(allo)morphs' in complementary distribution. I begin with a discussion of the reasons for adopting the selectionai approach, which my subsequent argument is built on. Lieber (1982) provides evidence that allomorphs must be listed as lexical entries. Many of her arguments for the selectionai approach are however based on templatic phenomena (such as reduplication), and these arguments are superseded by the nonlinear approach to templatic morphology (McCarthy & Prince 1995). Zwicky (1986) correctly notes that the selectionai approach is to be preferred because it is more restrictive. Here I shall provide some additional arguments of an empirical nature in support of the selectionai approach. 2. Types of Morpholexical Selection In this subsection I review the factors that condition allomorphy. I argue that they can be either contextual (syntagmatic) or internal (paradigmatic). Both can involve morpholexicalorphonological conditions. Crossing these, we get four cases. 1. Contextual morpholexical selection is the most familiar case (e.g., oxen)՛,

ALLOMORPHY AND MORPHOPHONOLOGY

19

2. Contextual phonological selection is also common, and many discus­ sions of morphology explicitly recognize it (e.g., Spencer 1991:121): a. though- /Өɔ:/ before {-t} (past tense, perfect participle, nominalizing suffix), think elsewhere; b. Sanskrit Instr.pl. -ais after stems ending in short -a, -bhis otherwise, e.g., /vrkşa+ais/ vrksaih "trees", /pad+bhis/ padbhih "feet", /agni+bhis/ agnibhih "fires";  Sanskrit asthan- "bone" before vocalic case endings, asthiotherwise; d. Warlpiri Ergative -ngku after disyllabic stems, -rlu after longer stems; e. In Finnish, the 3P Possessive ending -nsa has an optional alternant -Vn after a short vowel that is part of a case ending: kirja+ssa+nsa, kirja+ssa+an "book" (Iness.sg.), but only kirja+nsa (Nom.sg., directly follows stem), kirja+a+nsa (Part.sg., long vowel precedes), vastuu+nsa "responsibility" (both conditions fail); f. Finnish Illative sg. -seen after a stem ending in an unstressed long vowel, -(h)Vn otherwise (vapaa+seen, maa+han, kirja+an); g. German ge- before a stressed syllable, ø otherwise; h. Hungarian 3sg. -ja in back harmonic contexts, -i in front harmonic contexts, e.g., var+ja "waits", kér+i "asks". 3. Internal morpholexical selection: inherent prespecification of a morpho logical feature of a stem or affix: went is intrinsically specified as [+Past], which blocks *goed, *wented, *wents, for a stem intrinsically specified as [+F] cannot receive an inflectional affix that is specified· as [-F] (because of feature conflict), nor an affix that is specified as [+F], (because of blocking/no vacuous affixation).2 4. Internal phonological selection: inherent phonological specification of a phonological property in a morph blocks it from appearing in a context which requires a contradictory phonological property to be assigned to it. The Italian paradigm of andare "go" is made up of a stressed root vad- and an unstressed root and-:3

2

This requires the assumption that inflectional affixes can't change inherent feature specifications of the stems they are added to; this follows from the assumption that they are not heads. 3 Cases like this have been considered problems for morpheme-based morphology, see in particular Carstairs (1988b).

20

TRUBETZKOY'S ORPHAN

(3)

sg. 1. vádo 2. vái 3. vá

pl. andiámo andáte vánno

We can represent this regularity by prespecifying the allomorph vad as intrinsically stressed in the lexicon, restricting it to contexts in which it can be stressed in the lexical phonology. That excludes it from appearing before the stressed endings -iamo, -ate, since words with two lexical stresses are not allowed in Italian. It seems, then, that morpholexical selection is one of the ways in which phonological constraints can be manifested.4 The four types of selection are summarized in (4):

3. Allomorphy as Morphological Selection Having summarized my case for the claim that morpholexical alternations can be conditioned phonologically as well as morphologically, I turn to the arguments for the selectionai approach to morpholexical alternations, which does away with 'allomorphy rules'. 4

An alternative might be to conversely prespecify and as inherently unstressed. We would then have to block two other potential outcomes. First, the stress rule must not simply override the lexical specification. This would follow because it is feature-filling (Kiparsky 1993). Secondly, the stress rule must not simply be blocked by the inherent specification. In the present case, this would be impossible because of the prosodic requirement that every word must have a stress. Formally, we can reconstruct the relationship between paradigmaticity and blocking through underspecification of morphological features. Both result when a language has an underspecified morph for a morphological category. In the paradigm representing that category, such a morph functions as the default form. In any cell of the paradigm, it is blocked by a more specified morph if there is one, and otherwise supplies the form. In the absence of such an underspecified morph, on the other hand, there is neither blocking, nor any guarantee of paradigmaticity.

ALLOMORPHY AND MORPHOPHONOLOGY

21

3.1 The Parallelism with Morphological Gaps The first argument for the selectionai approach as opposed to the replacement approach is that the distribution of 'allomorphs' is governed by exactly the same kinds of conditions as the distribution of 'morphemes'. If a morpholexical alternation is handled by replacement processes, this identity remains unexplained. But it follows if morpholexical alternation is handled by the apparatus responsible for generating therightmorphological combinations, since this must, on independent grounds, involve a selectionai mechanism. The evidence is that the four types of conditions governing suppletion summarized in the preceding section are exactly those that are found defining the conditions for morphological gaps. The following often-cited examples are representative: (5) a. Contextual morpholexical (category selection): e.g., adverb-forming -ly goes only on adjectives; b. Contextual phonological: noun-forming -al goes only on verbs that end in stressed syllables {remóv+al, *depósit+al). Comparative -er goes on one-foot adjectives (where -y as usual does not count as a syllable);  Internal morpholexical: people, cattle are inherently [+PLURAL], so no overt plural affix is added; d. Internal phonological: the modern Greek diminutive ending -áki must be stressed. So diminutives in -áki can't be inflected in the genitive case, where the stress would shift to the ending by the stress rules of the language: ayor+і, Gen. sg. *ayor+aki+u, Gen. pl. *ayor+aki+ón "little boy" (cf. ayóri, Gen.sg. ayori+u, Gen.pl. ayori+ón "boy").5 A similar example is the Spanish verb abolir, which has a defective paradigm defined by the condition that the stem must be unstressed. So those forms of the paradigm where the stress would have to fall on the stem are not allowed to occur: *abólol*abuelo. The selectionai approach offers an immediate explanation for this parallelism: gaps of the sort seen in (5) result from the absence o f a default morph. Note especially the parallelism between phonological suppletion of the vádol andáre type in (4) and gaps like*abólo Iabolir (5d). But if suppletive forms are treated as replacements, the conditions governing suppletion cannot 5

It is not possible to satisfy both the lexical requirement that the first syllable of -áki must be stressed and the phonological rule that requires the following syllable to be stressed because Greek prohibits stresses on adjacent syllables.

22

TRUBETZKOY' S ORPHAN

be related to the conditions governing gaps, and so we have no account of why they work the same way. 3.2 Optionality The other deviation from the ideal complementary distribution pattern, overlapping distribution, also supports the selectionai approach to some extent. In the replacement model, overlapping distribution would involve optional morpholexical replacement. In the selection model, it would involve suspen­ sion of blocking. It has been observed that such overlapping distribution of morphemes is rare in paradigmatic (inflectional) morphological categories and relatively common in derivational categories, especially unproductive ones, e.g., speciousness and speciosity (van Marie 1985, Carstairs 1988a).6 This generalization argues for the selectionai approach, because it is to be expected that blocking should be connected with the productivity and paradigmatic char­ acter of a morphological category, but there is no reason why the option-ality of allomorphy rules should have anything to do with those things. 3.3 Locality The proper locality constraints follow only on the selectionai approach. Carstairs shows that allomorphs can't be 'outwardly sensitive' that is, that the shape of  cannot be morphologically dependent on the identity of  in the configuration (6): (6)

[[A + B]C]

Note that while Bracketing Erasure can block nonlocal morphological depen­ dencies in both directions, it cannot block outward dependencies of the kind just mentioned. Bracketing Erasure must be formulated as erasure of the internal structure of embedded constituents. For example, it erases the morpho­ logical constituency of the derived stem [ A +  ] in (6) as soon as  has been added. It must be accessible up to that point because the choice of  may depend on the morphological identity of B. Even so, the choice of  may not conversely depend on the morphological identity of C. This directional asym­ metry would be mysterious on the replacement approach, but follows from the well-supported assumption that the selectionai requirements of morphemes must be met at the point at which the morpheme is introduced, which itself might be subsumed under a more fundamental morphological version of the Projection Principle. 6

Carstairs convincingly interprets it as a consequence of the 'cell-filling' imperative of paradigmatic subsystems in the morphology.

ALLOMORPHY AND MORPHOPHONOLOGY

23

(7) Morphological Projectton Principle: Selectionai requirements of morphemes must be met at all stages of the derivation. A class of systematic apparent exceptions to morphological locality, where a morpheme seems to be selected by an outer affix or by a nonadjacent inner affix, really involves dependencies between a morpheme and a morphological category. In Turkish, the Potential suffix is normally -ebil, but it is ֊e when a negative morpheme follows. In Hungarian the plural suffix is normally -k, but it is -ai when a possessive suffix follows. This is evidently not selection of one morpheme by another but dependency of a morpheme on a morphological category. Specifically, we can consider Turkish -e a 'negative polarity item', like -couth and -sufferable, and the Hungarian plural suffix -ai an obligatorily possessed element, like 'inalienable' nouns (body parts, etc.) in certain languages. Other apparent cases of nonlocal dependencies yield to reanalyses, in some cases independently motivated. For instance, the Ancient Greek optative could be seen as violating locality (Carstairs 1980): (8) a. [ [ [ tim + ā ] oi ] ēn ] b. [ [ [ lū ] oi ] mi ]

(Theme verb) "honor" (Root verb) "loose"

On the indicated morphological analysis, the choice of the person/number suffixes (-ēn, -mi) seems to depend on whether the preceding optative morpheme is in turn preceded by a theme (8a) or directly by the root (8b). This would be a nonlocal dependency, in that it crosses over the optative morpheme. But a better segmentation of the Greek verb forms (suggested to me by Michael Inman) is the following: (9) a. [[[tim + ā ] o i ē ] n ] b. [ [ [ l ū ] oi] mi] On this analysis, the optative has two allomorphs, -oi after roots, and -oiē elsewhere, and some of the person endings have different allomorphs after the two. This reduces the apparently long-distance morphological dependency to two local dependencies. In addition to vindicating morphological locality, it also simplifies the morphology by eliminating massive allomorphy of person/ number endings, as is clear from examination of the complete optative paradigm.

24

TRUBETZKOY'S ORPHAN

sg.

du. pl.

1. 2. 3. 2. 3. 1. 2. 3.

" loose" lūoimi Іūі Iūoito lūoiton lūoitēn lūoimen lūoite lūoien

"honor" tīmōiēn timōtes tīmōiē tīmōiēton tīmōiētēn tīmōiēmen tīmōiēte tīmōiēsan

(alternative paradigm) tīmōimi tīmōis tīmōi tīmōiton tīmōitēn tīmōimen tīmōite tīmōien

3.4 The Interaction of Morphology and Phonology The selectionai approach also automatically rules out unwanted interactions between morphology and phonology, namely (i) selection of allomorphs conditioned by a derived phonological property, and (ii) triggering of a phono­ logical process by a 'basic' allomorph which is then replaced by a derived allomorph. Thus, it excludes derivations like (10), where A → A' is crucially triggered by B', and  → B' is crucially triggered by A. (10)

[ A +  ]→(phonology) [ A + B'] → (allomorphy) → [ A'+ B']

The generalization is that allomorphy is always 'ordered before' phonology on a given cycle. This would have to be stipulated under the replacement approach but follows automatically under the selectionai approach. Truncation' phenomena illustrate the point particularly clearly. On the replacement approach, which treats truncation as deletion of the underlying basic allomorph, we would expect that in a structure [ [ A +  ]  ], cyclic phonological rules should apply in the first cycle to [ A +  ] before  is added on the second cycle, causing truncation of B. For instance, in Aronoff's derivation of invent+ive from /invent+ion+ive/, the t→s and s → š rules should apply on the first cycle, yielding *inven[s]ive. This sort of interaction is never observed,7 a fact correctly predicted on the selection approach, where 'truncation' is simply non-insertion of an affix, so that no phonological effect of the 'truncated' element could ever materialize. Cyclically derived properties, on the other hand, can determine morpholexical selection, as correctly predicted by LPM: (11) a. The Finnish possessive and illative allomorphy (see items (2e) and 7

"Morphemes are truncated before they can cause phonemic modifications of the preceding phonemes." (Isačenko 1973).

ALLOMORPHY AND MORPHOPHONOLOGY

25

(2f) in section 2) show this kind of interaction. The allomorphy critically depends on the phonological representations phonologically derived on the previous cycle (Kiparsky 1993); b. In Spanish, cyclically derived secondary stress is erased postlexically, hence is available in the lexical phonology to determine the choice between the two forms of the article (el ~ al), e.g., elalimitalike el álma (Harris 1989). The cyclically assigned stress on al- counts for selection of the article, word-level destressing does not;  German míss-be-hàndel-t "mistreated", míss-interpretíer-t "mis­ interpreted". The rhythmic primary stress assigned to miss before unstressed syllables at the word level does not count for purposes of the selection of ge- (see section 2 on -ge); d. Italian ándirivièni "going back and forth" (an example cited by Dressier in discussion). Rhythmic secondary stress is apparently ignored for purposes of the selection of the root form and- (for the morphology, see section 2). A range of apparent exceptions to these claims are artifacts of old-style segmental phonology and disappear under a nonlinear formulation. Thus 'overapplication' in reduplication, as in Sanskrit pari֊şa-şvaj-e, from svaj "embrace", with the ruki-rule apparently applying after a, follows because the 'copying' mechanism that spells out the reduplication morpheme is part of the phonology. The generalization is that the copied melody is always determined at the end of the cycle at which the reduplication takes place. 4. The Existence of Morphologically Conditioned Phonological Rules We are now ready to return to the 'allomorphy or phonology?' question. The classic cases on which the debate turns are rules like German Umlaut and English Trisyllabic Shortening, whose conditioning environment has become opaque through attrition by sound change, but which continue to effect phonologically characterizable alternations.8 From the present vantage point, 8

For ablaut, reduplication, and similar templatic phenomena I assume the prosodic morphology account (McCarthy & Prince 1995), according to which they are morphological elements whose phonological content is realized by principles of autosegmental phonology. Ablaut patterns are morphs consisting of vocalic melodies which are superimposed on the lexically specified vowels of the base. Reduplication morphs are morphs defined by their prosodic shape which obtain their melody from the base. Thus, ablaut and reduplication are neither morpholexical nor morphophonemic rules, and should be set aside for purposes of the present discussion.

26

TRUBEIZKOY' S ORPHAN

the answer is unequivocally that they are not morpholexical alternations, but morphophonological alternations, governed by rules of lexical phonology. Except for their conditioning environment, all their properties are clearly pho­ nological. And once we recognize that phonological rules can be morpho­ logically conditioned (just as allomorphy can be phonologically conditioned), there is no reason to question their phonological status.9 In support of the claim that umlaut in German is morpholexical it has been claimed that each morphological category involves a distinct Umlaut process (Janda 1982b, Anderson 1992:344ff.). This claim appears to be incorrect. What is true is that the comparative and superlative forms of adjectives have the peculiarity that the diphthong au is never umlauted in them: braun + er "browner" vs. bräun+lich "brownish", Bräun+e "brownness", bräun+e+n "to brown". This is the only idiosyncrasy of these suffixes with respect to umlaut. Janda claims that umlaut in the comparative and superlative does not apply across a syllable with a schwa+sonorant rhyme. But the relevant adjectives of this form either have the nucleus au, thus falling under the constraint just mentioned (sauber "clean", lauter "pure"), or they resist umlaut in any context {mager "skinny", hager "skinny"). For example, the back vowel is retained not only in the comparative mager+er, but also in the derived verb mager+n "to lose weight" (contrast änder+n "to change"). These two adjectives are then simply not umlaut-susceptible, and do not show a different type of umlaut at work.10 If there were a separate Umlaut process for each suffix, they should all have different effects on their stems, but this is not the case. All thirty-odd other umlaut-triggering suffixes actually cause the identical set of changes. The phonological nature of Umlaut is shown by the following facts: a. The target is a single segment, phonologically defined (vocalic nucleus).11 9

See Wiese (1994) for further arguments for a similar view of German Umlaut. German appears to present a similar bifurcation of comparative and superlative endings as English does, with frequent words taking a level 1 type of ending and the residue of productive cases derived with regular phonology at level 2. In English this is manifested in the well-known contrast between retention of g in a few common adjectives (lo[ηg]er, lo[ηg]+est, stro[ηg]+er, stro[ηg]+est) versus deletion in productive formations like winni[η]+est. In German there is a contrast between umlaut in common adjectives versus lack of umlaut in rarer adjectives, including in particular compounded forms of some of the same adjectives: klüg+er "smarter", alt+klug+er "more precocious", kält+est+e "coldest", eis+kalt+est+e "ice-coldest". Similarly, the consonantal alternation in näch+ste "nearest" (from nah "near") does not apply in the compound hauï+nah+st+e "nearest to the skin". 11 In the case of the diphthong au, the immediate output of fronting is subject to a rule which transfers the rounding and backness from the diphthong's second element to the first (äu —> oi). 10

ALLOMORPHY AND MORPHOPHONOLOGY

27

b. The structural change is the assignment of a phonological property ([֊Back]).

c. The triggering context must be in part phonologically defined (the trig­ gering suffix must have a nonback vowel). d. It obeys phonological locality conditions and disobeys morphological locality conditions. Umlaut never crosses a full vowel; for example, in words liko Attentat "assassination (attempt)", Europa "Europe", only the last back vowel is capable of umlauting: Attentat +er (*Ättentat+er) "assassin", Europä+isch (*Euröpa+isch) "European". In fact, if I am right in believing that there are no cases of the type Bubi→*Bübi+lein "little Bubi", Dorit→*Dörit+chen "little Dorit", we can say that German Umlaut never crosses a phonological syllable; for disyllabic stems in sonorants, such as Vater, Väter+chen "father", are phonologically mono­ syllabic, with predictable schwa-epenthesis. Since phonological rules can be either iterative or local, we would expect as another possibility that Umlaut could spread to front a sequence of back vowels. This kind of umlaut is attested in Icelandic; similarly, morphologically triggered palatal­ ization is reported to spread in iterative fashion in a class of Latvian diminutives (Ruķe-Draviņa 1959). What is excluded by phonological locality principles (but permitted by morphological locality principles) is the 'skipping' of an eligible fronting site. Any umlaut process which behaved in this way would have to be considered a morpholexical rule, in which case all the other attendant consequences should hold. The fact that umlaut disobeys morphological locality conditions is shown by cases where it crosses a morpheme, e.g., 'jumping' over the -tsuffix in past subjunctive bräuch-t-e "would need".12 e. It is fed by phonological rules (on the same or previous cycles): given the constraint that Umlaut does not cross a syllable, /jude+iš/ → jüd+isch "Jewish" shows that it is fed by schwa-truncation. f. Finally, the application of umlaut depends on the specific allomorph of the triggering context. For example, of the two plural allomorphs -s and -er, the former does not trigger Umlaut and the latter does. Thus, the Umlaut process has to 'follow' the selection of the appropriate allomorph of the plural morpheme. This is what is predicted by the claim that umlaut is a morphophonological rule. If umlaut was a morpholexical (allomorphy) 12 Zwicky (1967) and Wiese (1994) argue that cases like Fahr+er+in (*Fähr+er+in) "female driver" demonstrate that Umlaut does not cross an affix; such examples however are also excluded by the phonological constraint mentioned in the preceding item.

28

TKUBEIZKOY' S ORPHAN

process, it could not be triggered either by the plural 'morpheme' per se, or by the feature [+Plural]. A mutual selection of allomorphs would be re­ quired; it is not clear how this would be expressed in the replacement approach. Exactly the same arguments apply to Trisyllabic Shortening: a. The target is a single segment, phonologically defined (vocalic nucleus). b. The structural change is the assignment of a phonological property (short­ ening).  The triggering context must be in part phonologically defined, e.g., there is no shortening before a stressed syllable (grād+at+ion, vãc+at+ion vs. grād+u+al, e+văc+u+ate, or directly before a vowel. d. It obeys phonological locality conditions and disobeys morphological locality conditions. In Myers' analysis, it is simply the shortening of a vowel before an immediately following tautosyllabic consonant, in line with a general quantitative condition on English syllable structure. Con­ sidered as a morpholexical rule, however, it would violate the morpho­ logical locality conditions otherwise valid for such rules. This can be seen in cases like prě+sid+ent, prě+par+at+ion, de+riv+at+ion, with an obligatorily short vowel (compare prē+side,prē+pare, dē+rive, which can be pronounced with a long vowel, which to be sure can be reduced). In such words the shortening, if formulated as a morpholexical rule triggered by the suffixes -ent,-at+ion,would have to apply across the root in order to shorten the prefix vowel. e. It is fed by phonological rules (on the same or previous cycles), specifically by stress and resyllabification. The stress condition on Trisyllabic Short­ ening holds rigorously and must be worked into any formulation of the process, independently of whether Myers' morphophonological analysis is adopted or not. f. Finally, the morphological triggering of umlaut depends on specific allomorphs of the context. The lexical conditioning is by specific mor­ phemes, not morphological categories, e.g., obesity vs. obeseness. In fact, in the tightly argued and compelling analysis of Myers (1987) (pursuing a suggestion of Stampe 1973) Trisyllabic' Shortening is simply a special case of closed syllable shortening. Myers shows that much of the ap­ parent idiosyncrasy of the process is eliminated if we take into account the lexical distribution of final extrametricality, which is independently shown by stress. For example, the reason shortening normally does not apply in the penult (see (12.1)) is that final syllables in nouns and derived adjectives are

ALLOMORPHY AND MORPHOPHONOLOGY

29

normally extrametrical, as the antepenultimate stress patterns illustrated in (12.2) show: (12)

1. Length: a. glōb+al,fōc+al,pōr+ous b. arōma, arēna,ōpal,ēvil 2. Stress: a. nátion+al, devótion+al, vénom+ous, ódor+ous b. América, chólera, Timothy

And the reason derived adjectives in -ic diverge from the general pattern and normally do cause shortening (see (13a)) is that unlike other adjective-forming suffixes -ic is not extrametrical, as the stress patterns in (13b) show. (13)

-ic is usually not extrametrical: a con+ic, cycl+ic, măn+ic, stăt+ic (some exceptions: scēn+ic, bās+ic) b. acíd+ic, económ+ic, metáll+ic, Icelánd+ic, alcohól+le (rare exceptions: Árab+ic, Cáthol+ic)

The suffixes -id and -ish tend to behave in the same way. By the same logic, the consistent penultimate stress of verbs, as opposed to the antepenultimate stress of nouns and adjectives (recall (12)), shows that final extrametricality does not apply to verbs, which correctly predicts the consistently short quantity of their penult vowel: (14)

replenish, discover, inhibit, inhabit, inherit, credit, edit, deposit, ravage, develop, sever, covet, finish, wallop, follow, hover, ravish, listen, famish, shiver, quiver, cavil, menace, continue, admonish, vanish, visit

The data in (15) show that Shortening is cyclic. (15)

a. cycle cycl+ic cýcl+ic+ity b. trībe tfīb+al trīb+al+ity

The conditions which correctly distinguish (15a) and (15b), as we have just seen, hold after the first layer of affixation. The suffix -ic is non-extrametrical, so both syllables of cýcl+ic are visible, hence shortening applies. But -al is extrametrical, so the last syllable of trīb+al is invisible, so shortening is inapplicable. When the second suffix -ity is added, the stress shifts forward,



TRUBETZKOY' S ORPHAN

the stem-final consonant is syllabified to the stressed syllable, and there is no question of Shortening in either of the words. Myers' analysis further explains the phonological conditions to which the process is subject: (1) the shortened vowel must be in a stressed syllable; (2) the following syllable must be unstressed (i.e. no shortening in words like cīt+āt+ion); (3) there must be an intervening consonant (i.e. no shortening in words like riōt+ous). These are of course exactly the conditions under which resyllabification (the 'Left Capture' rule of Kahn 1976) can apply, which on Myers' proposal is what feeds Shortening. Note also that the formulation preserves the phonological locality of 'trisyllabic' Shortening, in that no three-syllable window is required. The evidence thus clearly supports the phonological status of German Umlaut and English Shortening. A similar demonstration could, I think, be given for most other classic instances of opaque rules for which the question 'allomorphy or (morpho)phonology?' has been raised, such as Finnish con­ sonant gradation, and the palatalizations of Sanskrit and Slavic. In each case, the single-segment target of the process, the phonologically defined nature of the alternation, the strict obedience to phonological locality, and the way the processes interact with incontestable cases of allomorphic and phonological alternations of the language all diagnose a rule of lexical phonology rather than an allomorphic alternation. This bundle of properties, then, seems to be sys­ tematically correlated in languages. On the other hand, the nature of the context does not seem to correlate systematically with anything else. The correlation between the properties we have taken as identifying morpholexical alternations and morphological conditioning is at most statistical. Alternations which by every other criterion are morpholexical can be phonologically conditioned, and alternations which by every other criterion are morphophonological can be morphologically conditioned. In addition, the nature of the conditioning environment has no sound theoretical basis as a criterion for the typology of rules. The version of the rule typology proposed is independently supported on the morphological side. Carstairs-McCarthy (1992) finds that predictable al­ ternations don't count for purposes of the generalizations he proposes for the universal constitution of inflectional paradigms. In particular, his principle of Paradigm Economy only works properly if English given and spoken, and Warlpiri kurdu-ngku "child-Erg" and Ngajulu-rlu "I-Erg" (see (2d)), are combined into a single paradigm each. Carstairs introduces for this purpose the concept of macroparadigm. On the proposal presented here, morphophono­ logical alternations are abstracted away from morphology and the notion of

ALLOMORPHY AND MORPHOPHONOLOGY

31

paradigm itself becomes correspondingly more abstract, in a way which I believe make it appropriate as a basis for principles of inflectional morphology such as those proposed by Carstairs and others. In particular, I conjecture that Carstairs' 'macroparadigm' is simply the morphological paradigm, minus morphophonological alternations. I conclude that the conditioning of an alternation is irrelevant to its status as morpholexical versus morphophonological, and should be eliminated from the diagnostics for this distinction. Since the remaining core criteria are theoret­ ically motivated and generally converge, the upshot is that morpholexical and morphophonological rules can then be sorted out in a principled and consistent way.

Where Does Allomorphy Begin? Comments on Kiparsky K. P. Mohanan National University of Singapore 1. Introductory Remarks There are three sets of issues that I would like to keep separate in my com­ ments on Kiparsky's paper on morphophonology: issues of (i) modularity, (ii) rule systems, and (iii) representations. Ford and Singh's paper presents ideas which are diametrically opposite to those either implicit in or explicitly articulated in Kiparksy's paper. Hence, we can understand the issues in both these papers better by juxtaposing them. Let me begin with modularity. Most debates on morphophonology involve the issue of allomorphy, morphophonology, and 'pure' phonology within their respective conceptions of morphology. Kiparsky's proposal has the effect of separating allomorphy on the one hand from morphophonology and phonol­ ogy on the other. For him, allomorphy is part of morphology, while morpho­ phonology and phonology are part of phonology. In contrast, the proposal by Ford and Singh has the effect of separating allomorphy and morphophonology on the hand from phonology on the other. For them, both allomorphy and morphophonemics are part of morphology. These proposals for dealing with the modularity of allomorphy and morphophonology crucially depend upon the conceptions of morphological representations that these authors either explicitly argue for or take for granted. Central to their discussion are two themes. One of them is that of the 'morpho­ logical compositionality of words'. The other is the concepts of 'features', 'morphemes', and 'morphs'. Kiparsky assumes that words are composed of proper subparts that correspond to morphemes or morphs. He claims that the notions of feature and morph are both needed in linguistic theory, but questions the need for the concept morpheme. Ford and Singh accept the need for features, but explicitly deny what Kiparsky takes for granted: they claim that words are not composed of word parts that correspond to the notions of morphemes or morphs. Neither morphemes nor morphs play a role in their theory. This conception of morphological representations, combined with their

A L L O M O R P H Y AND MORPHOPHONOLOGY

33

declarative bi-directional rule system, results in the modularity claim of autonomous phonology: that is, the hypothesis that phonological rules are not sensitive to morphological information. Given the centrality of the issue of representations in this debate, it seems to me that the best strategy would be to lay out the different conceptions of the representation of word structure, including those of compositionality, features, morphemes, and morphs. Begging your indulgence for talking about what we already know, I would like to begin with a characterization of features, morphemes and morphs before discussing Kiparsky's proposal and plausible alternatives in relation to these representational units. 2. Features, Morphemes and Morphs Familiar examples of the three way distinction of features, morphemes, and morphs are given in (1) and (2) below: (1) a. John has written the letter.

written [+v, -N] [+PERF] feature {write} {en} morpheme /rit/ /n/ morph1

b. The letter was written by John, written [+V,-N] [+PASS] feature {write} {en} morpheme /rit/ /n/ morph  John wrote the letter.

(2) a. John has shattered the vase.

wrote [+V, -N] {write} /rout/

[+PAST] feature {ed} morpheme morph

shattered [+V, -N] [+PERF] feature {shatter} {en} morpheme /šætor/ /d/ morph

b. The vase was shattered by John. shattered [+V, ֊N] [+PASS] feature {shatter} {en} morpheme /šætər/ /d/ morph 1

Whether the morph of {write} in written is /rit/, /rait/, or /riit/ does not matter for the purposes of this illustration.

34

TRUBETZKOY'S ORPHAN

. John shattered the vase.

shattered [+v, -N] [+PAST] {shatter} {ed} /šætər/ /d/

feature morpheme morph

Features like PERF, PASS, PAST, PLUR, MASC, like distinctive features in phonology, are part of a universal inventory. Both morphemes and morphs are part of the language particular inventory in a given language. These assumptions are stated in (3): (3) a. Features: part of a universal inventory in Universal Grammar: e.g., PERF, PASS, PAST, PLUR, MASC, NOM, ACC ... b. Morphemes: part of the language particular inventory in an indi­ vidual grammar: e.g., In English, the morpheme {en} bears the features PASS and PERF; In Malayalam, the һ {appet} bears the feature PASS (but not PERF).  Morphs:. part of the language particular inventory in an individual grammar: e.g., In English, the һ {} is realized as /n/, /d/, etc. We must assume that features are part of UG because we need to make crosslinguistic assertions about, say, how TENSE features (PAST, PRESENT, FUTURE) have cross-linguistic consequences for the overt manifestation of grammatical subjects in many languages. If {en} is  һ and /n/ and /d/ are һs, then one may say that rpһms are language particular entities associated with semantic, syntactic, and phonological properties. Thus, even though for mnemonic urposes І have labeled the morphemes associated with PERFECTIVE and PASSIVE as {en}, and the һ associated with PASSIVE and PAST as {ed}, this is only for the sake of cross-reference. The representations of morphemes would look somewhat like those illustrated in (4): SYNTAX a. morpheme {28}: b. һ {94} c. һ {45} d. һ {72} e. morpheme {104} Morphemes

PERF, PASS PASS, PAST VERB, INTRANS VERB, TRANS ADJ

(4) The Representation of Morphemes

SEMANTICS

 MOVE AWAY [x ACT UPON y], &.. MENTALSTATE

PHONOLOGY /n/, /d/,ablaut, etc. /ď, ablaut, etc. /gou/, /went/ /distroi/, /distrukt/ /sæd/ Morphs

ALLOMORPHY AND MORPHOPHONOLOGY

35

In the light of diagram (4), we may expand the statements in (3) as those in (5): (5) a. A morpheme is a language particular entity. b. A morpheme is associated with syntactic, semantic, and phono­ logical representations.  A morpheme can be associated with more than one syntactic re­ presentation, semantic representation, or phonological represen­ tation. d. One or more of these three dimensions of representation may be absent in a given occurrence of a morpheme. e. Linguistic representations (syntactic, semantic, and phonological) are made out of a universal inventory of atomic elements (some of them are 'features')· f. The phonological representation of a morpheme (= underlying representation) is called morph (= formative in SPE). g. When a given occurrence of a morpheme is not associated with a phonological representation, we call it zero morph. h. When a morpheme is associated with more than one morph, we call it allomorphy. 3. Status of Allomorphy 3.1 Three Positions on Allomorphy Hypothesis (5c) permits a morpheme to be paired with more than one underlying phonological representation. Kiparsky refers to this as 'allomorphy as selection'. The choice between two alternative phonological representations, that is, two different morphs, is made in terms of a distributional statement that tells us which representation to use where. This view is implicit in (4) and (5). In contrast, if we view 'allomorphy as replacement', we are allowed only one phonological representation for each morpheme. The other morphs are derived through a rule that effects a change in the phonological representation of the morpheme, resulting in either partial or total replacement of the phonological material. This position is indicated in (6):

36

TRUBETZKOY'S ORPHAN SYNTAX

. mrpһm {28}: b.mrpһm{94} .mrpһm{45} d.mrpһm{72} e. morpheme {104} Morphemes

SEMANTICS

PHONOLOGY

/n/

PERF, PASS PASS, PAST

/d/

VERB, INTRANS x MOVE AWAY VERB, TRANS [x ACT UPON y], &.. ADJ MENTAL STATE

/gou/ /distrai/ /sæd/ Morphs

(6) Allomorphy as Replacement: Change in the phonological representation of the morpheme.

The crucial difference between (4) and (6) is that (4) allows more than one morph or underlying representation for a given morpheme, while (6) allows only one underlying representation. In advocating һ as selection, Kiparsky rejects (6). However, in addition to advocating һ as se­ lection, he also claims that the notion of morphemes as a primitive unit is unnecessary in morphological theory, as distinct from the universal inventory of features and the language particular listing of morphs. If I understand his position correctly, it can be pictured as (7): SYNTAX

SEMANTICS

PASS, PAST VERB, INTRANS

 MOVE AWAY

VERB, TRANS

[ ACT UPON ], &... MENTALSTATE

ADJ

PHONOLOGY

/n/, /ď, ablaut, etc. /d/, ablaut, etc.

PERF, PASS

LMoıphemes

/gou/, /went/ /distrai/, /distrukt/ /sæd/

Morphs

J

(7) Kiparsky's Position: һ as a pairing between features and formatives.

There are three related issues we are facing here: (8) a. Issue I: What are the primitive units of representation? b. Issue II: How do we state the regularities in representations?  Issue III: What are the modules of linguistic organization? Each of these questions has different possible answers, as indicated in (9): (9) Primitive Units of Representation a. The primitive units are morpheme, feature and morph. b. The primitive units are feature and morph. Morpheme is a derivative notion.

ALLOMORPHY AND MORPHOPHONOLOGY

37

(10) Statement of Regularities in Representations a. (i) A given morpheme can have more than one phonological repre sen tation (= morph). (ii) Allomorphy is a relation between a morpheme/feature and its phonological representations, (iii) Allomorphy is expressed as a declarative statement (of selec tion). b. (i) A given morpheme can have only one phonological represen­ tation (morph). (ii) Allomorphy is a relation between two phonological representa­ tions. (iii) Allomorphy is expressed as a procedural statement (of replace­ ment). (11) Modularity a. Allomorphy is part of the module of morphology. b. Allomorphy is part of the module of phonology. In Kiparsky's argumentation, the different combinations of these answers are conflated into a binary opposition of 'selection' and 'replacement'. In order to evaluate his position, it is important to unpack the different assumptions that each package contains. 3.2 Arguments Against Allomorphy as Replacement The bulk of Kiparsky's argumentation is based on the observation that allomorphy does not have the properties characteristically associated with phonological patterning, namely, parallelism with gaps, optionality, locality, and phonology-morphology interaction. This would be an argument for treating allomorphy as part of the module of morphology rather than phonology, that is, for (11a) as opposed to (11b). However, I fail to see how this can be an argument for the rule formalism of selection, that is, for (10a), as opposed to (10b). In principle, one can combine (10b) and (11a) as indicated in (12): MORPHOLOGY

/distroi/ + ion/ → distrΛkt + ion allomorphic replacement rule

PHONOLOGY

distrΛkšən phonological rule

(12) Allomorphy as Replacement in the Module of Morphology: EXAMPLE: derivation of destruction [distrΛkšən]

38

TRUBEIZKOY'S ORPHAN

All the crucial properties that Kiparsky points out would follow from the replacement model in (12). Thus, what is at stake here is not the rule formal­ ism, but the issue of modularity. As far as I can see, the choice between selection and replacement is simply a matter of the choice of rule formalism. It does not have any empirical consequence, and hence has no theoretical significance.2 3.3 Is Allomorphy a Pairing between Features andMorphs? If we allow the construct morpheme in linguistic theory, allomorphy is a relation between two language particular entities, namely, morpheme and morph. On the other hand, if we eliminate the construct morpheme and adopt Kiparsky's treatment of allomorphy, then allomorphy is a relation between a universal entity and a language particular entity, namely, feature and morph. The latter solution runs into problems of duplication in the statement of allo­ morphy. Take, for example, the statement of the allomorphy of the perfective and the passive in English. In the selection model in (4) which incorporates the notion morpheme, allomorphic selection will be stated only once: (13) Allomorphy as a pairing between morphemes and morphs: Morpheme {28} : a. morph = (i) /n/ in write, break, hide, speak... (ii) ablaut in sing, ... 2

Halle & Mohanan (1985) treat allomorphic alternations such as those in sit/sat in English as part of the phonological module. Central in Kiparsky's proposal is the rejection of the modularity in such an analysis. Now, the alternation in sit/sat in English could be stated either as a morphological replacement rule as in (i), yielding the derivation in (ii):

or as a morphological selection rule as in (iii), yielding the structure in (iv), with the effect of replacing the specification [+high] with the specification [+low]: (iii) Select the superimposed morph /æ/ for PAST in words of class X. (iv)

These seem to me to be technical alternatives without empirical consequences. Either way, the device that Kiparsky abandons is also one that Ford & Singh object to, namely, the device (found in Halle & Mohanan 1985, Kiparsky 1985 and others) of a phonological rale whose environment is specified as a zero morph that precedes or follows the stem morph. If we accept Kiparsky's current proposal, the noun-verb alternation in houseNlhousey or measureN/ measureV does not require the postulation of a zero morph.

ALLOMORPHY AND MORPHOPHONOLOGY

39

(iii) . . .

(x) otherwise /d/ b. syntax = (i) PERF (ii) PASS In this model, allomorphy is a pairing between language particular morphemes and language particular morphs. If we eliminate the notion 'morpheme' as Kiparsky suggests, and state allomorphy as a pairing between universal features and language particular morphs, the statements of the allomorphy of passive and perfective will involve a duplication: (14)

Allomorphy as a pairing between features and morphs: a. PERF: morph = (i) /n/ in write, break, hide, speak, ... (ii) ablaut in sing,... (iii) ... b. PASS: morph =

(x) (i) (ii) (iii)

otherwise /d/ /n/ in write, break, hide, speak, ... ablaut in sing,... ...

(x)

otherwise /d/

Needless to say, if we use the intermediate 'feature' PAST PARTICIPLE as a diacritic for the combination of perfective and passive in English, and state the allomorphy in terms of this feature as in (15), the problem of duplication can be eliminated. (15) Allomorphy as Pairing between Diacritic Features and Morphs PAST PARTICIPLE: a. morph = (i) /n/ in write, break, hide, speak,... (ii) ablaut in sing,... (Üİ) . . .

(x) otherwise /d/ b. syntax = (i) PERF (ii) PASS The problem that we now have to deal with will be the universal status of the feature PAST PARTICIPLE such that it can participate in statements of allomorphy. If this feature has no universal characterization, but exists only in

40

TRUBETZKOY'S ORPHAN

the grammar of English, it is simply a flag for the 'morpheme' {en}. We must then acknowledge that allomorphy is a relation between two language particular entities, not a relation between a universal entity and a language particular entity. The nature of the diacritic solution in (15) becomes more transparent when we consider the 'past tense' allomorphy in Malayalam, illustrated in (16): (16) "run" a. -um b. c.

"open" -um -

"hear" -um kee -

future form present form past form

What I have glossed as 'future form', 'past form' and 'past form' are diacritic labels which can be interpreted only within the grammar of Malayalam. They are markers of tense when they occur as the last verb in a finite clause. But they also occur as non-finite forms, before the last verb, as in (17a), or in obligatory control clauses as in (17b): (17) a. child-NOM run-'past' walk- 'future' "The child will walk in a 'running' manner." b. door-NOM open- 'past' child-NOM out "After opening the door, the child went out."

go- 'future'

As far as I know, there is no syntactic theory which would analyze the 'past' form in (17a) and (17b) as instances of the syntactic feature PAST. The syntactic TENSE of (17a) is future, not past. The morphological past form is simply a requirement on the non-final forms of the serial verb construction. Though in the 'past form', the verb ooti "run" here is syntactically non-finite. The verb in (17b) is also non-finite. As far as I know, all current syntactic theories assume that obligatory control is found only in non-finite clauses. If we analyze the form as being in the PAST TENSE, we will not be able to explain why the embedded clause in (17b) is a construction of obligatory control. If we allow morpheme as a primitive unit, the allomorphy of the verbal forms in Malayalam can be described as (18).

41

AIXOMORPHY AND MORPHOPHONOLOGY

(18)

Allomorphy as a pairing between morphemes and morphs Morpheme {732}: a. morph = (i) "open", ... (ii) "hear"... (iii)... () otherwise, /і/ b.

syntax =

PAST, NON-FINITE

If we eliminate the construct morpheme, there does not seem to be a way of avoiding the duplication in (19): (19)

Allomorphy as pairing between features and morphs a. PAST: morph = (i) "open", ... (ii) "hear"... (iii)... () otherwise, /і/ b. NON-FINITE: morph = (і) (ii) (iii)...

"open",... "hear"...

() otherwise, /i/ Finally, the notion of allomorphy as a pairing between universal features and language particular morphs makes no sense when we come to stem allomorphy, as in destroy and destruction. The entity paired with the morphs /distroi/ and /distrυkt/ can be viewed as a language particular unit. I don't see what kind of universal feature these morphs can be paired with. 4.

Conclusions In sum, the evidence I have discussed in these remarks points to the following conclusions, most of which are not dramatically different from traditional views: (20) a. The smallest units of words are morphemes, some of which cannot be words by themselves (affixes, bound roots). b. A morpheme is associated with phonological, syntactic, and semantic representations.  A morpheme can be associated with more than one syntactic representation, semantic representation, or phonological representation.

42

TRUBETZKOY'S ORPHAN

d. One or more of these three dimensions of representation may be absent in a given occurrence of a morpheme. e. Linguistic representations (syntactic, semantic, and phonological) are made out of a universal inventory of atomic elements (some of them are 'features'). f. Allomorphy is part of the module of morphology.

On the Morphology/Phonology Boundary: Comments on Kiparsky Douglas C. Walker University of Calgary In the context of studies of rule taxonomy, it is clear that most work has focused on the fundamental distinction between morphophonemic and allophonic (or lexical and post-lexical) rules and on internal divisions within these categories.1 Less attention has been paid to the way which (morpho)phonology interacts with morphology and the lexicon; that is, to the allomorphic/ morphophonemic interface. It is therefore a pleasure to be able to benefit from Paul Kiparsky's observations concerning the morpholexical/morphophonology distinction.2 Before turning to certain detailed implications of Kiparsky's proposals, I will tabulate what seem to me to be some of his central claims: 1. There is a core of clear cases where a coordinated set of criteria render the three-way distinction allomorphy/morphophonology/phonology "unproblematic or at least uncontroversial." 2. Beyond this core, the allomorphy/morphophonology distinction may break down. In such cases we may choose one or more criteria as definitive, or we may deny a discrete division between rule types. Either solution gives up the "clean theoretical explanation of the convergence observed in the core cases." 1

Or, in the case of e.g., Halle (1959), Chomsky (1964) or SPE, the absence of such a distinction. 2 The terminology in this domain is far from established. Kiparsky appears to use the following terms in essentially synonymous ways: 'suppletive', 'morpheme-specific', 'mor­ pholexical alternation' and 'allomorphy'. Thus, "morpholexical rules deal with suppletive, morpheme-specific alternations...morphophonemic rules deal (at least) with 'general phono­ logical changes which occur only in particular derivations' (Bloomfield)" and phonological rules "deal with automatic phonologically conditioned alternations" (cf. p. 24). It is not clear, however, that there are no significant distinctions to be made within the set of terms involving allomorphy. For example, see Meľčuk (1982:110-118) for one indication of the complexities of suppletion.

44

TRUBETZKOY'S ORPHAN

3. The putative factor responsible for the lack of a clear distinction between allomorphy and morphophonology is reliance on morphological or lexical conditioning as a necessary and sufficient criterion for identifying allomorphic rules. 4. This criterion must be abandoned since there exist morpholexical or allomorphic rules without such conditioning and morphophonemic rules with such conditioning. Once we eliminate this criterion, a clear allomorphy/morphophonology distinction based on a concurrent set of prop­ erties returns. 5. As a result, the set of properties which combines to distinguish allomorphy from morphophonology is the following:3 Allomorphy: - item-specific - may involve more than one segment - obeys morphological locality conditions - takes effect before all morphophonological rules - morphologically or phonologically conditioned Morphophonology (or Lexical Phonology): - not item-specific - involves single segments only - obeys phonological locality conditions - takes effect after all allomorphy rules - morphologically or phonologically conditioned Against this background, we may return to a discussion of the allomorphy/ morphophonology distinction. My strategy will be to consider individually a number of the criteria in (5), with a view to showing that they may be violated independently and that the conjunction of these criteria is therefore un­ warranted. This tactic parallels what has happened to the separation of lexical from post-lexical rules in Lexical Phonology, where an analogous breakdown of a formerly homogeneous set of diagnostics appears to be taking place. Thus, Kaisse & Hargus (1993a: 16) acknowledge, with respect to the set of properties putatively distinguishing lexical from post-lexical rules,4 that "we 3

Linked to this division is the proposal that allomorphy is accounted for by selection from among a list of allomorphs, one of which is the elsewhere case, rather than by replacement of a single basic representative by substitution of other allomorphs in various contexts. Lexical entries are then sets of morphs in a blocking relationship, and allomorphy functions just like morphology in this respect. I view Kiparsky's arguments concerning this matter as highly attractive, and will not deal with them further. 4 E.g., cyclicity, word-boundedness, access to word-internal structure assigned at same level, order with respect to post-lexical rules, presence of exceptions, application in derived environments and so on. Cf. Kaisse & Hargus (1993a: 16-17) for details.

ALLOMORPHY AND MORPHOPHONOLOGY

45

now know that many of these characteristics cannot be considered diagnostic of the lexical or postlexical status of a rule." Partly as a consequence, the number of rule types is now multiplying to yield (at least) cyclic and post-cyclic lexical rules and types P1 and P2 of post-lexical rules, where P1 rules share properties with lexical rules (e.g., cyclicity). In other words, the discrete division between lexical and post-lexical rule types, formerly assured because a set of criteria functioned in unison, is dissolving. It is possible to foresee a similar fate for the allomorphy versus morphophonology partition. Thus, I do not agree with Kiparsky that "the picture is enriched by the postulation of additional rule types and more fine-grained distinctions among morphophonological rules...and among phonological rules" (p. 15). Rather, the boundaries between rule types are dissolving because certain diagnostic properties straddle these boundaries. We appear instead, if the same lack of correlation between properties emerges in the domain of allomorphy, to be headed toward the gradient scales characteristic of Natural Phonology and Morphology. Which properties enumerated by Kiparsky as distinguishing allomorphy from morphophonology appear fragile? I will deal with three. To begin with, it is not clear that suppletion or allomorphy is necessarily item-specific. There may well be cases where we are not dealing with a general alternation, but where a small set of items is involved. Numerous cases illustrated by problems with the learned vocabulary in the Romance languages come to mind (French mère-maternel, père-paternel, frère-fraternel, for example). More generally, restricted lexical domains (learned words, kinship terminology, number or pronominal systems, proper nouns and so on) appear to lend themselves to just this type of 'suppletion'.5 Thus, allomorphy need not be item specific. We need a transitional stage between the 'true' suppletion of the je vais-nous allons-nousirons type applicable to single morphemes and the (potentially limited) generality of morphophonemic rules. Secondly, the restriction of morphophonological rules to apply to single segments only would appear to be contradicted by examples of productive reduplication, metathesis, contraction, mutual assimilation or exchange rules, examples of which have been extensively presented in the literature and not all of which are reducible to single targets. (I assume it is not legitimate simply to break a complex rule down into its component parts for the sake of preserving the 'single segment' claim.) Thirdly, the rigid ordering relationships between allomorphic and morphophonological rules (all allomorphy » all morpho­ phonology) would appear to be contradicted by the various marked but I am indebted to Professor Kiparsky himself for discussion of this point.

46

TRUBETZKOY'S ORPHAN

nonetheless widely occurring rule interactions of the type illustrated in Anderson 1975, among others. In particular, those cases where morpholexical selection depends on derived phonological (or even allophonic6 ) properties are problematic for a taxonomy based on a conjunction of the properties in (5). As a second type of general comment concerning "On the morphology/ phonology boundary", it is perhaps worth noting, particularly in the context of this conference, its 'static' character and the absence of arguments imported from the so-called external domains. This absence is all the more striking in a paper by the person who introduced into the generative grammatical lexicon the use of arguments from historical change as a justification for synchronic structures (cf. Kiparsky 1968). An abrupt shift from morphophonemic to allomorphic (i.e. to suppletive) rules, for example, does not appear to lend itself to an insightful characterization of the way languages change, particularly if all the properties of (5) are to change in unison. To the extent that a synchronic theory must permit a proper characterization of its own internal dynamic, its evolutionary patterns and possibilities, the tripartite model sketched here ap­ pears too restrictive. A gradual transition from one rule sub-type to another, incorporating in particular a lexical dimension, would be more in keeping with our historical expectations. One final general observation may be relevant in the light of much recent work in phonological and morphological theory. Sets of properties such as those in (5) lead one immediately to think of the ranked constraints char­ acteristic of Optimality Theory. One productive avenue of research, here as well as in the morphophonemic versus allophonic or lexical versus post-lexical domain, might involve the reinterpretation of these criteria as constraints on rule types or rule interactions, followed by a determination of the appropriate rankings of these or related properties. This would presumably allow for a much more gradient approach to rule taxonomy and for the language-specific variation that results from the differential rankings of such constraints. There appears to be no reason, at least at this stage of our knowledge of rule taxonomy and rule interaction, to rule out an optimality theoretic approach to the ranking of rule properties (or any other parameters, for that matter). Thus, with respect to (1) above, while there may be a set of 'clear cases' where the properties of (5) unite to reinforce a discrete partition into allomorphic and morphophonemic rules, there also appears to be reason to doubt the stability claimed in (4). Despite their initial attractiveness, we must question 6

Anderson (1975:56) cites the case of Rotuman, where the morpholexical rule determining the form of the incomplete phase of stems must follow an allophonic rule determining vowel quality.

ALLOMORPHYANDMORPHOPHONOLOGY

47

claims as to the ease with which linguistic rules or processes sub-classify into a limited set of types and interact in a hermetic or disjoint fashion. While we must be grateful to Paul Kiparsky for drawing our attention to the necessity of examining anew the factors which organize different types of rule behaviour, past and recent studies of rule taxonomy lead me to believe that the situation is more chaotic than his present paper proposes.

Reply to Mohanan and Walker Paul Kiparsky The main thesis of my paper is that the status of an alternation depends on its form and on its locality properties, and not, as has often been claimed, on how it is conditioned. Mohanan and Walker do not dispute this directly, but they do question some of the claims and assumptions that I make along the way. Mohanan doubts whether I have really shown that allomorphy involves selection rather than replacement, and that the morpheme is a derivative notion rather than primitive notion of the theory. On the first point, he claims that my evidence can also be accounted for by entirely separating morphology from phonology, as diagrammed in (12) of his response. But that model is a nonstarter because it cannot account for the fact that affixal allomorphy can be sensitive to derived phonological properties of the base (as Walker indeed notes in his comments).1 Accounting for this kind of morphology/phonology interaction has been one of the basic motivations for Lexical Phonology and Morphology, the theory presupposed in my paper. My paper proposes a constraint on this interaction. The empirical generalization is that allomorphy is sensitive only to the locally accessible phonological representation, viz. the output of the phonological rules of the previous cycle. The theoretical account of this generalization is in turn the basis of one of my arguments for how to handle allomorphy: the selectionai approach allows the interaction of phonology and morphology to be intrinsically constrained by the independently given organization of the grammar so that the right interactions fall out, whereas the replacement approach requires additional extraneous constraints to achieve it. Mohanan's first point is therefore simply mistaken. Contrary to what he claims, my proposal has empirical consequences which are supported by the evidence. This said, there remains a more basic disagreement between us concerning the positivist philosophy of science which he assumes throughout, and which I See Kiparsky (1993) for examples.

ALLOMORPHY AND MORPHOPHONOLOGY

49

reject. In particular, I think that a formal difference may well have 'theoretical significance' even if it appears to have no empirical consequences for the moment. It happens often enough in science that divergent theoretical con­ ceptions seem empirically indistinguishable when they are first formulated, yet fundamental differences between them become apparent as their latent implications are explored more deeply. Had Mohanan been at Ghent in 1938 when Jakobson introduced his conception of features as the ultimate components of phonemes, he would no doubt have joined the chorus of complaints that it was only a 'technical alternative', empirically indistinguish­ able from the view of features as 'cover terms' for classes of phonemes. Of course, the empirical evidence soon began to emerge, and turned out to support Jakobson's theory, beginning with his own findings on acquisition, aphasia, and typology (Jakobson 1971[1941]), continuing with Halle's insight that the evaluation measure requires it (Halle 1962), and culminating with the results of autosegmental phonology (Goldsmith 1976). Mohanan then turns to the status of the morpheme, which was briefly alluded to in my paper, though it hardly figured in the way I proposed to draw the distinction between allomorphy and morphophonology. Contrary to Mohanan, the view of morphology I sketch out in my paper in no way commits me to treating allomorphy as a pairing between universal features and language particular morphs. Nor is the assumption that 'past participle' is a morphological feature tantamount to recognizing a 'morpheme' {-en}. On the view I suggested in my paper, the past participle can very well be a morpho­ logical category (in line with both traditional grammar and recent morpho­ logical studies, e.g., Fabri & Wunderlich 1995). Specifically, English verb stems can be organized in terms of two cross-classifying morphological features, PAST and FINITE, where the 'past participle' is the [+PAST, -FINITE] form. Mohanan's duplication argument thus evaporates. Yet the different forms marking it (-ed, -en, ablaut) need not be listed together in the lexicon as so many 'allomorphs' of some superordinate entity. They are related derivatively as a class of elements in a blocking relationship. The 'morpheme' as it were constitutes itself through the Elsewhere Condition. A closer look at Mohanan's own example reveals that the duplication argument actually speaks in favor of my proposal. The distribution of -ed versus the various ablaut forms and irregular stems is largely the same in the past finite (past tense) as in the past nonfinite (participle). In the distribution of the shared formatives /d/, /t/, and most of the ablaut forms, the two categories are actually identical. The principal divergence between them is that forms in -en are strictly restricted to the participial function. But, since the past tense and

50

TRUBEIZKOY'S ORPHAN

past participle are different 'morphemes', a Mohanan-style traditional descrip­ tion ends up stating the shared part of the distribution twice. Thus, (la)-(2a), (lb) = (2b), etc., and only (li)-(lk) lack a counterpart in (2): ( 1 ) 1 . PAST, NONFINITE:

(a) /t/ in burn, feel, deal, kneel, dream, lose, weep, breed, shoot2 ... (b) /t/ and stem suppletion in bring, seek, catch, fight.. (c) /d/ and shortening in say,flee,shoe... (d) /d/ and stem suppletion in make (e) ablaut /o/ in win, shine... (f) ablaut /u/ in sing, ring, hang,fling,stick, dig, sneak... (g) ablaut/e/ in hold (h) ablaut /a/ in sit (i) /n/'mfall, show, grow, take, beat, eat... (j) /n/ and ablaut /o/ in get, tread, swear, weave, break... (k) /n/ and shortening in write, smite, ride, drive, do... etc. (1) /d/ otherwise

(2) 2. PAST, FINITE: (a) /t/ in burn, feel, deal, kneel, dream, lose, weep, breed, shoot3 ... (b) /t/ and stem suppletion in bring, seek, catch, fight... (c) /d/ and shortening in say,flee,shoe... (d) /d/ and stem suppletion in make (e) ablaut /o/ in win, shine... (f) ablaut/u/ infling,stick, dig, sneak... (g) ablaut /e/ in fall, hold (h) ablaut /a/ in sit, sing, ring, hang... (i) ablaut/o/ in rise, get, trad, swear, weave, break... etc. (j) /ál otherwise The appropriate generalizations can be expressed by treating the shared inflectional forms (a)-(h) and (11) = (2j) as homonymous between past tense and past participle — in other words, as general [+PAST] forms — and stating their distributional conditions in a single statement. We can then collapse the bulk of (1) with (2) (along with many other ablaut cases and irregularities 2 I assume that this l-t/ triggers shortening of the stem vowel and degeminates with a preceding t-, d-. 3 With precisely the same morphophonemic effects as in the past participle forms.

ALLOMORPHY AND MORPHOPHONOLOGY

51

omitted here). The -n ending will then block these general-purpose past forms in the past participle forms that select it. Thus, depending on its morphological features, a special form either blocks a general form entirely (e.g., the special [+PAST] form flung blocks the 'elsewhere' form of the [+PAST] *flinged), or it blocks it in a particular function (e.g., the specific past participle forms sawn, flown restrict the [+PAST] forms sawed,flewto the past tense function). The more complex such a blocking network, the more generalizations will be lost by segregating the forms into 'morphemes' and stating the distribution of the allomorphs separately for each morpheme. Thus, the considerations of economy that Mohanan cites against me turn out to cut the other way. Walker claims that there is in fact no sharp distinction between morphophonology and allomorphy to be made. To show that the distinction is gradient, he presents three arguments that even the remaining traditional criteria for allomorphy which I propose to retain do not always coincide. One of these criteria for allomorphic alternations is idiosyncrasy, or itemspecificity, by which I mean that the alternants are listed for particular individual morphemes or lexemes rather than being introduced by rules that apply to classes of items defined in terms of general structural properties. The expectation is then that general alternations are due either to the application of (morpho)phonological processes or to the addition of morphological elements, subject respectively to phonological and morphological locality principles; morphological elements may be melodically unspecified prosodic templates, or prosodically unspecified 'floating' melodies, which combine with the stem in non-linear fashion, as in reduplication, infixation, and other kinds of prosodic morphology, again governed by phonological locality conditions (McCarthy & Prince 1995). On this approach, ablaut forms and similar alternants (at least in so far as they reflect general patterns) are not listed as allomorphs but derived by superimposingfloatingvocalic melodies on the inherent melody of the root. Walker's first argument is that allomorphic alternations can be general rather than item-specific. He cites the French suppletion pattern seen in the three words père/paternel, mère/maternel,frère/fraternel.4These alternations are rather marginal and unproductive, so that it might not be unreasonable to simply list the allomorphs for each stem; on that analysis the alternation would in fact be item-specific. Walker disregards this alternative, presumably on the grounds that a generalization (however tenuous) is thereby lost. So let us accept for the sake of the argument that there is indeed a general morphological alternation here, and assume that it applies to kinship terms, and that the disyllabic /-atєr/ alternants are selected by 'learned' suffixes. We can factor out Similar examples in English would be pope ~ papal, nose ~ nasal.

52

TRUBETZKOY'S ORPHAN

the /ε/ ~ /a/ alternation because it is triggered by the so-called 'learned back­ ing' rule (e.g., père ~ parricide). Suppose now that the normal 'learned' forms of these stems are formed by adding an infix t after the root vowel, e.g., /par/→ /patr/, and a stem-forming suffix, such as -, -і in patronymique, patrologie, patriarcal. If the stem-forming suffix is consonantal, as in /patr-n-/, the cluster is broken up by insertion of  /ә/, which in this context surfaces as /є/ by general rules (cf. genre ~ génér-ique). In sum, Walker's first example is inconclusive, since it is not clear that the alternation is really general rather than item-specific. However, if it is con­ sidered general, a morphophonological analysis is available which is consistent with the theory proposed in my paper. One of the criterial properties of morphophonological rules is that they, like phonological rules, can apply to just one segment at a time. This is simply a corollary of the assumption that morphophonological rules have the properties of phonological rules and that phonological rules can apply to just one segment at a time. If there is a type of phonological rule that can apply to more than one segment, then exactly such morphophonological rules should also be able to apply to more than one segment. Consider in this light Walker's second ar­ gument that no clear boundary can be drawn between allomorphy on the one hand and morphophonology/phonology on the other. It is based on the claim that there are morphophonological rules that target more than one segment. As examples he cites productive reduplication, metathesis, contraction, mutual assimilation, and exchange rules, in each case assuming pre-autosegmental analyses of them, which are now superseded. As for reduplication, it has been clear since Marantz (1982) that there are no morphophonological rules of reduplication. Rather, reduplications are affixes in their own right, consisting of prosodic templates which borrow their phonemic shape from their stems according to principles of autosegmental phonology. Therefore, reduplication processes actually confirm my position. The other processes mentioned by Walker are a heterogeneous group, but in so far as they are morphophonological rather than simply morphological/ allomorphic, they likewise confirm rather than refute my claim that morpho­ phonological processes and phonological processes are subject to the same formal constraints. Qua processes, metathesis, contraction, and mutual assimi­ lation function in exactly the same way whether they are morphophonological or phonological in character. Where apparent metathesis is a synchronic process, it is always a templatic effect, in the cases I am familiar with. It results from the accommodation of a morpheme's phonemic content to different syllabic structure patterns, which may be imposed either by the language's

ALLOMORPHY AND MORPHOPHONOLOGY

53

phonology or as specific morphological templates, or in the formation of the Rotuman incomplete phase mentioned by Walker. Contraction also involves accommodation to syllabic templates. Any derivational/ processual relation to the uncontracted form in such cases involves the deletion of a segmental slot in the skeleton, with subsequent reassociation of the remaining slots of which the segments of the melody. In these cases as well, the contraction process itself is formally identical whether the skeletal deletion that triggers it is phonologically or morphophonologically determined. As for morphophonological exchange rules, there is very little evidence that they are required as either phonological or morphophonological processes at all. Thus the class of processes mentioned by Walker confirms the formal parallelism between morphophonological and pure phonological operations which was one of the key points of my argument. Whether any of them provides good evidence that morphophono­ logical rules can target several segments at a time is a side issue in this context, but it is any case doubtful. For the reasons just stated, contraction and metathesis are not cases in point. As for bidilectional assimilation, it sometimes must be decomposed into its unidirectional component operations (e.g., Hayes 1986), and I am not aware of any cases where it cannot be so decomposed. Walker's third argument is that morpholexical selection can depend on derived phonological properties. Far from being overlooked in my paper, let alone being a difficulty for the argument put forward in it, this point is actually part of the evidence supporting my position. The fact that allomorphy can depend on derived phonological properties is what the architecture of Lexical Phonology is designed to account for through the interleaving of phonological and morphological constraints. The point made in section 3.4 of my paper is that this dependence seems to be limited in a particular way; the generalization is that allomorphy is always ordered before the phonology on any given cycle, hence allomorphy is selection rather than replacement. However, allomorphy can of course be sensitive to the derived phonology of the previous cycle. While such phonology/morphology interactions do refute the view that phono­ logy and morphology are entirely separated (as advocated e.g., in Mohanan's reply), they corroborate the position based on LPM that I defend in my paper. Finally, I disagree with Walker's view that recent research has somehow been 'dissolving' the boundaries between phonological rule types. I see no basis for his prognosis that we are 'headed toward gradient scales', much less that the situation is 'chaotic'. On the contrary, the distinction between lexical and postlexical rules, for example, has been confirmed time and again by numerous studies on a variety of languages. I would be hard put to cite a single in-depth analysis of a language's phonology of the last ten years which has not

54

TRUBETZKOY'S ORPHAN

empirically confirmed the reality of this fundamental distinction. In general, the more these questions have been studied, the more clearly articulated the rule typology and the organization of the phonology has become.

Allomorphy and Morphophonology Discussion KIPARSKY: In response to Walker I should say first that these further distinctions among lexical rules — let's say those between the word level and cyclic rules — in no way weaken the evidence for a superordinate category of lexical rule. They do not in any way call into question the basic distinction between a lexical and a post-lexical rule. As far as I know it is not even controversial: almost everyone has something like this distinction. What has happened is that the theory has different ways of elaborating or specifying this distinction. The second point I wanted to make is that the criteria in the original list of properties distinguishing lexical rules from post-lexical rules that turned out to be wrong were exactly those which were not based on any principles; they were just listed as properties. The proposal I made here does not have this character, I'm not making a laundry list of properties of allo­ morphy and another laundry list of properties of morphophonology. I am saying that if you make some fundamental assumptions about morphological selection versus phonological interpretation, then a whole range of properties follows. So these are principled in a way in which the original distinction between lexical and post-lexical rules was not. As for the question of allomorphy produced by phonology, which was raised as another potential problem, I think this problem is purely termi­ nological because the structuralists called anything that involved an al­ ternation in the shape of a morpheme 'allomorph', whether it was produced by a morphophonemic or an allomorphy/morpholexical rule, and that's precisely because they didn't make a fundamental distinction between these two types of rules. If you want to use 'allomorphy' as a useful term then I think it should be restricted to the alternation between different morphs, that is different things listed in the lexicon, essentially suppletion if you like, phonological alternations could be called 'allomorphy' if you wished to retain structuralist terminology, but I don't see any point in that. The issue of mutual conditioning seems to me actually to be an argument for the selectionai approach to allomorphy. It makes sense on the assumption

56

TRUBETZKOY'S ORPHAN

that what allomorphy really is is morphs being selected in contexts, so the shoe is on the other foot in that case. As for Learned Backing, I just wanted to throw this question back at you. I didn't understand in what sense Learned Backing in French could be in any way a problem for the approach I propose. Is your suggestion that it really should be an allomorphy process? WALKER: I don't have the answer to that at the moment. My recollection of Learned Backing is that the suffixes which condition Learned Backing can follow suffixes which themselves do not and the segments undergoing backing do not form a natural class phonologically, so that from the point of view of this criteria they do not appear to be morphophonemic rules. They appear instead to be morpholexical rules, and if so, they conflict with other properties of morpholexical rules, at least well-documented rules of that type. DRESSLER: I think that in all those cases where you criticize certain of my approaches I can wait for tomorrow for my paper. I have a few other remarks to my handout: the basic thing is 'condition'. The criterion affects a single segment', has implications for vowel harmony, for fusion processes, where several segments are affected, and for the German umlaut au to äu. Another point is locality where there is the problem of morphological in­ sertion rules like interfixation where clearly there is dependence in the wrong direction, that is where what you inserted depends on what follows linearly. Another point is forms like [invenšiv] which do, although morphophonological rules are involved, occur rarely in speech errors. KIPARSKY: If I may I'll just take up at least a few of these points. First of all, about single segments. Of course harmony should be treated as a rule which applies to a single segment but iterates. This is the standard analysis. It is in no way is intended to be excluded here. On the contrary, precisely because morphophonemic rules are phonological they can apply in the two ways that are familiar from all phonological rules, either once or iteratively. So actually this again confirms the point. Fusion is typically analyzed as the elimination of the vocalic or consonantal slot, so that two melodies are fused into a single slot. As for umlaut, the umlaut of au to äu has been analyzed by Wiese and others fronting followed by unrounding (the article by Wiese has been published in the Düsseldorf Working Papers). The question of infixation once again, if McCarthy and Prince and many other people are right, which I think they are, then infixation is affixation at an edge. DRESSLER: Sorry, interfixation. KIPARSKY: Interfixation. What particularly do you have in mind? DRESSLER: When for example you have [pueblo] [puebleθito], where you have the interfix [es] or [θ] according to dialect.

ALLOMORPHY AND MORPHOPHONOLOGY

57

KIPARSKY: There is an analysis of Jim Harris' that is consistent with what ľ ve said here which appeared in Linguistic Inquiry a few years back, the one that deals with stem-forming affixes in Spanish, so in that particular case I don't think it poses a problem; it may be that there are others. As for speech errors, I would imagine that [invenšiv] arises as the result of abstracting the verb [invenš] from [invenšan] ad hoc and then forming [invenšiv] from that verb. That would mean that one could also have speech errors such as [invenš]. In umlaut, the nucleus ai of -lein functions as a front vowel in German phonology in other ways as well. For example, it never umlauts itself, which it would if it were a back vowel, so it systematically patterns with other front vowels as well. This is not all that abstract an analysis. For example, Würzel has it in his book on German phonology, in 1970. DRESSLER: In 1970, quite outdated. KlPARSKY: In 1970, and it has never been refuted. In cases like Väter, Gepäck, where there is no overt vowel, there is evidence for a schwa there which drops. So for example something like Väter e with a schwa is im­ possible. In exactly those declensions where one expects a schwa, that schwa is systematically missing precisely after an unstressed sonorant stem. So this umlaut is triggered by the same schwa which is in Baum/Baüme. In the cases of [la] (I don't know these dialects myself), I would suspect that this is a realization of schwa in those dialects, that is that in some German dialects merge in schwa and [a]. In any case the striking thing is that no analogical extension of umlaut was ever triggered by any of the overtly back suffixes like -tum whereas it spread to the non-back suffixes by analogy. JANDA: I always thought someone should write a paper on Spanish and German data. I am happy to hear about stem allomorphy. I think it makes it clear what you are talking about, that you are listing some stems to be used for various purposes. I think that convergence is productive. Depending on the theory, if the affixes are processes, the difference comes up. I just wanted to make three points, and the first one, after this agreement, is that amorphic morphology in some earlier instantiations, did have an empirical side. Based on Anderson's work, in my dissertation I took up some conditions that Stephen proposed in order to differentiate rule-types, rules being divided not as lexical or post-lexical but as morphological, purely phonological and lexical correspondence. The claim there is that there are various types of conditions that would treat the rules differently, that the Elsewhere condition would apply to phonology, without morphology, that lexically free rules, exchange rules could be morphological but not phonological and so forth. There were actually six of those. Now you might

58

TRUBEIZKOY'S ORPHAN

disagree with the way that it came out, but there was a real empirical side to that. The way people name their theories is really hard to disagree with — theirs is natural and everything else is unnatural. So, amorphous morphology is one kind of process morphology that's word and paradigm. There's both the right hand side and the left hand side, the right hand side of your tax­ onomy runs together more than word and paradigm, more than the classical theory, at least not too many people do any more. This also has an empirical part and its processual focus per se. The second point has to do with context, so we will hear about that tomorrow, we heard some from Walker today. I feel a fast one is being pulled when I see the [t] of thought as a phonological complex, because [t] itself then is annotated down the line as only occurring after thought and bought and certain other verbs and if you could imagine an experiment where new affixes were created in English beginning with [t], I have the feeling they wouldn't provoke various selection allomorphs like [ba] and [θο]. I would naturally go to an analysis where thought is listed in a redundancy rule which is purely a redundancy rule, it's never used productively to form new past tenses in forms with [t]. So, this would be a listed form parsed by a redundancy rule. I would then equally naturally ask you your view of re­ dundancy rules, because I think strangely enough this is the most telling question you can ask a linguist. If you're talking about life, you ask do you believe in the eternal soul? If you're talking to a linguist, you can ask do you believe in redundancy rules? And if you do, let's divide suppletion in two ways. I don't know if it was Hudson, but someone divided partial suppletion and suppletion per se, so went/'go, that's full suppletion, but if you listed both think/thought and sing/sang, you have only a small difference, do you have redundancy rules somewhere to deal with these relations? Those are my three questions. KIPARSKY: I will try to be short but these are interesting questions. Yes, I agree with Anderson about being empirical. Well, what is my world view on redundancy rules? I think they are rules that assign default values. So in fact in that sense a lot of phonology is done by redundancy rules. So, there is a another sense in which that word 'redundancy rule' is sometimes used, let's say by Jackendoff, in which they are generalizations about fully specified listed items. To me one of the goals of grammar is precisely to eliminate redundant information. FORD: What I wanted to say was that the crucial factor about the distinction between morpholexical and morphophonemic rules was that it seems to

ALLOMORPHY AND MORPHOPHONOLOGY

59

eliminate the possibility of there being another strategy for word formation so that a word like shoot that has a past form shot will in children and many other speakers of the language have a past form in shooted whereas a word like boot generally has a past form in booted but can often have the form bot I don't see how if you're making the distinction you are making you can take account of this sort of rivalry. KIPARSKY: The analysis of shoot/shot, bite/bit etc. that I would advocate capitalizes on the fact that English contains vowel shift and a degemination rule. This is a case of shortening before a geminate, basically the same thing as shortening before a cluster in keep/kept. That means that what the children who are saying shooted for shot are doing is that they are assuming that it's a Level  or regular past tense instead of having learned an irregular one, and the children who are saying bot instead of booted are overgeneralizing the irregular paradigm to a new verb, so it's essentially the same phenomenon as the child saying brang for bringed. The phenomenon is that first they say brought, then they say bringed and then they say brang. They say bringed again so a new generalization that is learned gets overgeneralized; the correct irregular form previously learned by rote memory. FORD: So, there is some kind of hierarchy, then? KIPARSKY: There is a hierarchy of irregularity, of course, so brought is the most idiosyncratic, it is just listed as an allomorphy rule. Shot is formed in the phonology by shortening. BYBEE: It seems that vowel harmony rules do not use morphological conditions and they also seem to have parallel rules applying quite often postlexically. So it seems that they are more phonological and less morpho­ logical. KIPARSKY: What you say is true and goes along with the very interesting fact that lexical harmony processes are almost always structure-preserving. There is a morphological aspect about them, which is connected with the phe­ nomenon of lexical rules applying to prosodic domains. In some languages nouns have a stem cycle and verbs don't. Nouns then have affixed vocalism, while verbs undergo vowel harmony triggered by their suffixes. The reason is that the defaut value of the harmonizing feature is assigned cyclically. So morphology plays a big role mostly in the sense of defining the domain. I just want to say that they are not rules. I should have been more careful and said 'alternations', because I ended up arguing that the very difference between morpholexical and morphophonemic alternations is that one is not a rule but rather a selection and the other is. PIGGOTT: A short question. Why is the final [t] in went not treated like the final

60

TRUBETZKOY' S ORPHAN

[t] in bought? KIPARSKY: You mean as derived from wend? PIGGOTT: I don't have a claim one way or the other. I don't understand the criteria. KIPARSKY: I suppose that is a possible analysis. It is not something that would be completely impossible but I don't see what would be gained from it. PIGGOTT: Would it work? KIPARSKY: It would not be something you could do in all cases, let's say bit and bought and so forth. As soon as you get to these marginal cases there are always alternatives. The less structure, the more alternatives there will be. This is not claimed to give you a magic wand that will automatically sort everthing into morpholexical versus morphophonemic. There are going to be borderline cases. JANDA: Could we continue that question? I'll go ahead and ask, I'll be the obvious strawperson. What do we gain by having the [t] in thought as separate? You might think there is as much reason to have the [t] in thought as the [t] in went, but you turn it around and say there might be as little reason to have the [t] in thought as the [t] in went and this goes for Doug Walker's comments because it is true that there is a small class of verbs, thought, fought, etc. Is that a big enough generalization to warrant being captured, and if so why should that be excluded? KIPARSKY: Yes, it's there. I thought I would like to have an English example and this was the closest I could get. Strike that one if you don't like it. JANDA: There's the [d] in sold for example. KIPARSKY: Pick any analysis of thought or fought as lists or whatever, I don't much care. I think the very next example, for example nobody will want to argue with. There is a little bit of generality in that there are several suffixes: there's past tense, there's the participle, there's the nominalizing -t- suffix, all of which happen to have the same allomorph. The interesting general point behind all this is that allomorphy arises in completely different ways historically. One way it arises is by distinct lexical items — distinct morphemes — joining into a paradigm, for example go/went. And another way it arises is by phonological processes becoming no longer transparent, for example, think/thought. There are masses of examples of both types, and they have exactly the same properties. In spite of these completely different sources you get the same principles of construction. JANDA: I want to ask about that because it leads to the umlaut questions. We have locality conditions jumping in over things. Then it doesn't really matter whether the [t] is a separate morpheme or not. Leuchte we should bracket

ALLOMORPHY AND MORPHOPHONOLOGY

61

because that's just one exceptional form of a verb that has a regular past tense but has umlaut in the past subjunctive and you can see that as an analogy. Nowhere else is there a regular past tense that is in umlaut, it's an irregular past tense that gets it, but then that leads to say bringe/brachte/brächte, dachte/dächte and those things are actually listed as wholes and those have umlaut, so whether you segment the [t] in certain places actually makes a difference for claims that locality is being violated. So I think that's important but you circumvented the problem of having umlaut, because what defines it is often morphological conditions and you said that that's not that relevant distinction so the fact that comparatives and superlatives can't umlaut au and can't jump over a syllable, that's just a morphological condition so that you've already taken care of that, that's not a criterion for distinction. There might be types or something but it's not whether it's morphological or not. I think that's relevant but I'll go back to you again. KIPARSKY: The fact that comparatives and superlatives from au can't umlaut is to the point. It's true and I have no explanation for it. It goes along with a whole bunch of other strange properties of the comparatives and superlatives. One in of them is that unlike any other morphological category, it fails to apply in derived compounds e.g., nah, näher "near", hautnah, hautnaher "close to the skin". It's been shown in psycholinguistics ex­ periments the comparatives and superlatives stick out as forms in which the umlaut is not adopted. There is also some relationship to irregular allomorphy seen in the suffix itself, gross, grösste vs. blass, blasseste. So there is a correlation between the form of the affix and whether it triggers umlaut here and I suspect what might be going on. JANDA: Would you say one has a front vowel and one has a back vowel? KIPARSKY: I have taken no position on exactly what is the triggering of this, but I am attracted to the proposal by Wiese and Lieber and several other people that the stem receives its specification for the back feature from the affix. BYBEE: I'd like to hear more about how you think about these marginal or difficult cases. Presumably you believe in rules, and that rules are organized into components or groups, and I assumed you thought these types are discrete but you mentioned marginal phenomena a couple of times. Does that mean there are just difficult areas or are there actual gradations between rule types in your way of thinking? KIPARSKY: This is partly a philosophical question and partly an empirical one, I think. The theory doesn't have gradations. In that way it is different from some other theories in which the categories themselves have things belonging

62

TRUBEIZKOY' S ORPHAN

to them to a greater or lesser extent. This is not the case here. However, the evidence for assigning a phenomenon to this or that category may be more or less solid and I wouldn't be in the least surprised if there were phenomena where the evidence is too small to permit a clear decision one way or another. What do we do in such cases? Well, one way would be to say that there is a preference for assigning those phenomena to one or the other category, for example morphology in favor of phonology. Many people have suggested such a preference and argued for it on the basis of the phenomenon of morphologization. So you can of course have the default assumption that unless something can be shown to be phonological, it is morphology. That's entirely plausible, but it wouldn't mean the theory itself permits gradations in assigning phenomena to categories. BYBEE: TO get even more philosophical than before, I just wanted to be sure you agreed with my articulation. WALKER: You've got small sets of words that are phonologically and semantically close and behave in a parallel way. That makes them look morphophonological though that is really suppletion and you can multiply examples. There are many many things which in some sense seem to straddle morphophonemics and allomorphy but you would like the theory to have a neat way of resolving that gradience. Lots of languages seem to me to have this mixture or gradience and I just want to know why this also is not a neat way of solving the problem. KIPARSKY: I thought French could be analyzed along the lines proposed by Selkirk and Dell. They propose a feature assigned to lexical elements and to affixes, and a mechanism by which these features percolate up. That is by which bequeaths its features to the resulting stem an affix. Some constraint must ensure you can't have words which are made up of elements which are plus and at the same time minus (i.e. there is unification of the features). We have in English something like this to make sure that we don't add Romance suffixes to Germanic words and vice versa. In English perhaps not so much in French) there are words which join both vocabularies, long. That suggests a way of dealing with these phenomena on a purely morphological basis. What you raise is a very interesting question which deals with these small subsystems of words, kinship terms, where you get submorphemic re­ semblances. Numbers are another example of a lexical field where this happens, as are dimensional adjectives. There you have submorphemic but systemic patterns which are not entirely covered by either phonological rules or by any kind of morphological principles. This is to me a mysterious phenomenon which deserves more study.

ALLOMORPHY AND MORPHOPHONOLOGY

63

DRESSLER: These are cases of complete suppletion, where these regularities are clearly of a diachronic nature in the sense that suppletions, to be able to survive, need some support in the stage of acquisition so if there is such a minisystem where members, so to say, support each other by semantic and formal properties then there is what Stemberger has called the 'Gang effect' which makes them survive. Now this is a question of diachrony and acqui­ sition, rather and it's very difficult to have it projected into a principle of synchronic grammar. SINGH: Just one clarification. If I put your proposal in the context of earlier proposals that try to make a distinction between rules of allomorphy and morphophonemic rules, such as Matthews, for example, and Sommerstein, it is not clear to me for example whether you are proposing that these criteria, — assuming that they make the distinctions that you want them to make, which is not what I am debating at the moment — if you want their joint satisfaction, if you want, in other words, all of them to be jointly satisfied at the same time or if you consider the possibility that they may be graded in the sense that some may have precedence over some others. KIPARSKY: Yes SINGH: Yes, some have precedence?

KIPARSKY: Yes, I would consider that possibility. SINGH: Let me be less polite. Are you proposing that they must all be satisfied at the same time, is there any order in them, or no order? KIPARSKY: Normally, they're not.

II Dressler, Janda and Walker MODULARITY, MORPHONOLOGY, AND GRADIENCE

Describing a language, like mapping a country, is an affair of cutting and smoothing. The problem is where to cut, and where to smooth. C.E. Bazeli (1953:93)

A Functionalist Semiotic Model of Morphonology Wolfgang U. Dressier Universität Wien 1. Introduction The model I am going to present here is essentially identical with that of my monograph Morphonology (Dressier 1985a) which consists of chapters which lay the ground (1-4), an ample descriptive and classificatory part (5, 8), two case studies (6, 7), a chapter on language acquisition (9), and a long explanatory chapter (10). I am going to refer constantly to these chapters. In this paper I am going to outline functionalist analysis and explanation (2), then apply it to morphonology (3), present a semiotic model of morphonology by concentrating on problems of universal markedness or preference theory (4), explicate and exemplify main consequences for diachronic change (5). After brief remarks on morphonology in language acquisition (6) and the level of system adequacy (7), I will deal in the conclusions (8) with critiques of Dressier (1985a). 2. Functional Analysis Although my stance is decidedly functionalist, I oppose mutual exclusion of functional and formal explanation (cf. Hall 1992; Newmeyer 1992, 1994). I consider formal explanation rather as part of functional explanation. My main reason is (cf. Dressler & Merlini Baraberesi 1994:1.2-1.6) that I belong to those who see human language as a specific type of human interaction so that (text) pragmatics is to be seen as the all-encompassing highest level of which grammar is just a subordinate subpart. Now, for explanation in pragmatics, discourse and sociolinguístics, where speaker's intentions are crucial, only functional explanation is adequate. Particularly for diachronic change at least some elements of functional explanation appear inevitable, since change always involves pragmatic and sociolinguístic aspects and function in discourse. In other publications I have dealt with the rationale of functional explanation in linguistics in general and in various areas including diachrony and language acquisition (Dressier 1995, 1985a:261ff.; Dressier & Dziubalska

68

TRUBETZKOY'S ORPHAN

Kołaczyk 1994; Dressler & Karpf 1995). Here I can only briefly indicate: 1) that I do not equate function with communicative content or meaning; 2) that functional explanation is also applicable to linguistic structure and structural change, where we may abstract from speakers' intentions; 3) that functional explanation is not reducible to causal explanation (in the sense of HempelOppenheim-type deductive-nomological explanation); 4) that apparent weak points of functional explanation (such as goal conflicts, multiple strategies, plurifunctionality, competition of alternative explanations) are the necessary reflections of the complexity and multicausality of linguistic phenomena, particularly in diachronic change and language acquisition; 5) that the identification of functional analysis with a caricature of teleologicai explanation is confused; 6) that the important methodological concept of asterisks in generative grammar has a correspondent in functional analysis, namely the notion of dysfunction or functional deficiency and of dysfunctional change; 7) functional explanation is more ambitious than deductive-nomological ex­ planation but, as we have seen, also more difficult to disconfirm with 'internal evidence', therefore the use of 'external' or 'substantial evidence' (particularly from diachrony and language acquisition) is highly important. Functional explanation is applicable to all three subtheories (A-) of nat­ uralness theory which I espouse (cf. Dressler 1985a:261ff., 291ff.; Dressler et al 1987): A) the subtheory of universal preferences, where the semiotically-based parameters of transparency and biuniqueness and most iconicity parameters generally disfavour morphonology, whereas indexicality parameters gen­ erally favour morphonology. B) the subtheory of typological adequacy, where morphonology is typeadequate in inflecting, introflecting and incorporating languages, but not so in isolating languages, and in agglutinating languages only vowel harmony is type-adequate. C) the subtheory of language-specific system adequacy, where, for example in Polish, certain morphonological rules (MPRs) may be adequate in declen­ sion, others in conjugation. Functional explanation of speaker intentions applies to MPRs (as to other rules of grammar) in reference to speaker evaluations of variants during language acquisition and in phases of diffusion (spread) of innovations in ongoing language change.1 In other words, I claim a general correspondence 1 Sociolinguístic investigations (e.g., Allony-Feinberg 1977) have shown that also structural properties, referring to each of the three naturalness subtheories, are evaluated by speakers and are thus corresponsible for adoption by speakers in diffusion processes, In this way language

MODULARITY, MORPHONOLOGY, AND GRADIENCE

69

between my model of morphonology and criteria of speaker evaluations in the processes of language acquisition and language change. 3. Functionalism and Morphonology The units of morphonology are morphonological rules (MPRs), the most important subclass of which roughly corresponds to the lexical phonological rules of Kiparsky's Lexical Phonology.2 I assume rules not alternations, 1) for reasons that Classical Generative Phonology has given against struc­ turalist morphophonemics (cf. also Bromberger & Halle 1989); 2) because of similarities of MPRs to phonological processes (henceforth PRs), which have to be described in terms of processes (cf. Stampe 1969), particularly phonostylistic processes of casual/fast speech;3 3) because I conceive, within my dynamic model, of rules, processes and strategies as operations which serve specific functions.4 Whereas PRs serve the function of making language pronounceable and perceptible (or, even more, facilitating pronounciation and perception, as particularly in the case of phonostylistic processes), morphological rules (henceforth MRs) serve the following functions: 1) the function of motivating morphologically complex forms semantically, formally (tactically), and prag­ matically;5 2a) inflectional rules have the syntactic function of providing syntax with inflectional word forms, 2b) word-formation rules have the function of lexical enrichment. Do MPRs share (any of) these functions? As to morphological functions, MPRs do not have, on their own, functions of 2a) and 2b), and as to motivation (function 1), they complicate it insofar as they contribute to morphotactic opacity. As to the two functions of phonology, many MPRs clearly do not have them. Let us contrast phonological and morphonological palatalization in Polish (more in Dressier 1984; 1985a: 182ff.): the PR of acquisition and diachronic change are linked to the synchronic distribution of MPRs in a way which allows functional explanation (cf. Hall's 1992 notion of linkage). 2 Cf. Booij (1993). Here I will not discuss the class of 'prelexicaľ MPRs (cf. Dressler 1985a:219ff). They correspond to those morpheme structure constraints of other models which can be considered to be of a morphonological nature. 3 Notice also the incapability of Harris (1990) to account for all process types of Natural Phonology within his static format. 4 In my notation of MPRs I use the format of classical generative phonology, with the exception that I will usually write phonemes instead of distinctive features or their equivalents, because — as understood already by Kuryłowicz — they often apply differentially to single phonemes of natural classes of phonemes. 5 This leads to the differentiation of the dimensions of morphosemantics, morphotactics, and morphopragmatics (cf. Dressler & Merlini Barberesi 1994).

70

TRUBETZKOY'S ORPHAN

surface palatalization palatalizes consonants before the most palatal segments /i, j / , and /k, g/ before some morphs in -e (see below). This assimilation process clearly facilitates pronounceability (cf. Bhat 1978; Lindner 1975); moreover, certain consonant-vowel combinations are unpronounceable in Polish without surface palatalization, for example, *[ki, gi] vs. palatalized [k'i, g'i]. The MPRs and AMRs (allomorphic morphological rules) of postalveolar formation (historical 1st palatalization), however, transform /k, g, x, h/ into cz, di, sz, i [č, dž, š, ž] before a variety of suffixes which start with many different segments, e.g., before the fem, suffix -anka and 'agent' nouns in -і, fem. -nica, as in ziemniak "potato"→ziemniacz-anka "potato soup", rok "year" → rocz-nica "anniversary". This rule clearly does not facilitate pronounceability nor does it make the sequences /k+a, k+n/ pronounceable, because [ka, kn] are legal sequences in Polish, which often even occur with the same suffixes, as in wy-krzyk-nąc/wy-krzycz-ęć "to exclaim"→wy-krzyk-nik "exclamation point". For the few cases where MPRs resemble PRs so much that they too seem to have a phonological function, we will have to decide whether this is only an appearance or whether there is a vestigial phonological function (see below).6 But MPRs do have a function of their own, the function of cosignalling MRs (cf. Dressler 1985a:268f.). For example, the application of the Pol. MPR /k/ → cz in ziemniacz-anka is an advance signal that one of a specific set of suffixes is to follow. In terms of Marslen-Wilson's (1984) cohort model, a Polish hearer may reach earlier the point of recognition of the word being produced. Or in the terminology of cybernetics and communication theory (cf. Klaus 1968), MPRs are redundant, but in terms of useful redundancy, not of dead redundancy, i.e., within always unreliable communication they make signs more reliable. Clearly, if an MPR applies in the linear sequence after the triggering MR, i.e., if the morphonological cosignai occurs after the main morphological signal, then the cosignalling function of the MPR is less important although it still works as safeguard in communication. If the motto of functionalism 'Form follows function to some extent' is true, then we would expect that MPRs should be more frequent (in type and token frequency) and diachronically more stable, if their effects precede the morphological trigger than if they follow it. This is an empirically testable claim. In fact, among MPRs of 6 The functional distinctions discussed so far represent the explanatory basis for the descriptive criteria of phonetic plausibility, process matching and pronounceability/ perceptibility as discussed in Dressier (1985a:59ff., 68ff.) and used for differentiating MPRs from PRs and AMRs.

MODULARITY, MORPHONOLOGY, AND GRADIENCE

71

palatalization, regressive (anticipatory) palatalizations have been preserved in Polish (as PR or morphologized), progressive palatalization has been lost. Thus the rise of progressive palatalization in Alb. Imp. piq from jek "I cook" (cf. Dressler 1985a:82) seems to be rather exceptional. This cosignalling function, however, is also shared by many AMRs, for example, by those German umlaut rules that cosignai plural formation in addition to suffixation of a plural marker as in Wolf "wolf ' → Wölf-e, Fuchs "fox" → Füchs-e, or by English plural fricative voicing, as in wife→wiv-es. And the cosignalling function is even shared by some PRs that occur predominantly in morphological alternations and are easiest to notice among PRs, i.e., obligatory neutralizing phonemic PRs, especially if they are of the anticipatory assimilation type, e.g., voice assimilation of obstruents, as in the Latin past participles iunc-tus, tac-tus, tec-tus from iungo "I join", tango "I touch", tego "I cover". But whereas the cosignalling function is typical for MPRs, it is less so for allomorphic MRs and rather occasional for PRs. Maiden (1991:95) has stressed the possibility that the morphonological cosignai may become the primary signal. This I have always assumed in the diachronic change of, e.g., plural formation in Austro-Bavarian dialects, as in Tag "day" → pl. Täg-e→Täg. Here final schwa-deletion was first optional (casual speech PR), thus diminishing the reliability of the main, morphological signal of suffixation, then successive generalization of the phonostylistic PR to an obligatory process further diminished and finally canceled the relevance of the suffix. The non-existence of a separate function which holds only for MPRs fits in well with my assumption (held since Dressier 1977) that morphonology is not a grammatical component of its own or, more precisely, a module or submodule of grammar. Morphonology is rather an area of interaction between morphology and phonology with fuzzy boundaries. This has to be demon­ strated with the units of morphonology, viz. MPRs.7 But, before, I want to add a precision: in any language, morphonology is an area of interaction between the modules of morphology and phonology, i.e., between grammatical morphology and grammatical phonology. This precision has several consequences, notably the following two: 1) Morphonology, as far as I can see, is not involved in extragrammatical morphology and phonology,8 e.g., in the make-up of interjections, 7 Maiden (1991) insists on the conjunctive or cooperative interaction of morphology and phonology. I have never denied this possibility, but judging from the examples I know, I still think that conflictual interaction is more frequent. 8 This concept (cf. Dressler & Merlini Barberesi 1994:1.9) overlaps with Zwicky &

72

TRUBETZKOY'S ORPHAN

submorphemes (as in E. gl-eam, gl-ow, etc.), echo-word-formation, as. in E. tick-tock,Turk.kitab-mitab "books and other such things", cf. E. variables-shmariables (for French, cf. Mayerthaler 1977, for German, Wiese 1990), etc. 2) As to type-adequacy (cf. Dressler 1985a:337ff.), we can safely predict that a thoroughly isolating language which has nearly no grammatical morphology (e.g., Semai), cannot have morphonology either. Extragrammatical morphology, however, may abound in such languages (such as total reduplication in Semai). Thus, whereas the distinction of grammatical and extragrammatical morphology is relevant for morphonology, the distinction between prototypical and non-prototypical morphology or inflection or derivation (cf. Dressler 1989b) is not. Thus MPRs are as likely to occur in non-prototypical plural formation as in prototypical case formation. The concept of prototypicality, however, will be important for MPRs themselves (see below). 4. A Semiotic Model of Morphonology: Level of universal preference theory My choice of semiotics as a metalevel has had several aims (cf. Dressler 1985a:279ff.; 1989a), such as: a) to arrive at broader and deeper generaliza­ tions; b) to base axioms of a linguistic model on a promising more general model, thus avoiding ad hoc assumptions, and also in order to restrict the gamut of conceivable axioms; c) to connect the level of MPRs with other lin­ guistic levels in a principled way. Peircean semiotics (cf. Peirce 1965; Hookway 1985) has appeared to be the most adequate semiotic model. Within such a dynamic semiotics, where signs are understood as actions of semiosis (Parret 1983:30), rules and processes as well can be understood as signs. Thus, an MPR, MR or PR connects, in the mind of its producer, receiver or (more general) its interpreter, its input as its signatum with its output as its signans. Such a semiotic model allows the deduction of a consistent characteri­ zation of MPRs, and results in a continuum of gradual transitions between PRs and MPRs on the one hand, and between MPRs and AMRs on the other. A first semiotic parameter is diagrammaticity; a sign is the more natural the more diagrammatic the relationship between its signatum and its signans.9 And the more natural a sign, the higher its (type and token) frequency {ceteris paribus). Thus, metathesis is antidiagrammatic insofar as the linear sequence Pullum5 s (1987) notion of expressive morphology (and its phonological counterpart). Diagrams are the subtype of icons that is most important for grammar. All linguistic signs are conventional (symbols), but in addition they may be also iconic and/or indexical.

9

MODULARITY, MORPHONOLOGY, ANDGRADIENCE

73

of (part of) the input is reversed in the output (cf. Dressier 1985a:307f). As a consequence, synchronic metathesis rules/processes are extremely rare. Phonological diagrammaticity can be measured by the number of distinctive feature values of the input which are retained in the output. On average, intrinsic allophonic processes are phonologically most diagrammatic, extrinsic allophonic processes less, phonemic neutralizing processes and MPRs still less; allomorphic MRs are the least diagrammatic. This semiotic criterion is at the basis of the frequently used criterion of phonetic distance or similarity between input and output (cf. Dressler 1985a:64ff, 120ff). For example, attested Celtic lenitions started out as allophonic processes of either only voicing or only fricativization. When they changed to MPRs and MRs, always more feature changes were added. Morphological diagrammaticity is, so to say, the reverse of phonological diagrammaticity. Morphological diagrammaticity means constructional diagrammaticity, where additive rules are highest valued, modificatory rules less, and subtractive rules least. This hierarchy holds best for MRs, i.e., for their frequency distributions (additive MRs are preponderant, genuinely subtractive rules of grammatical morphology extremely rare), whereas the set of MPRs seems to be less additive and more modificatory and subtractive, PRs (particularly casual/fast speech processes) much more often subtractive (deletion processes). The importance of constructional diagrammaticity lies in its relevance for word and morpheme recognition and in its function of morphotactic motivation (cf. 3), and here a cline from MRs to PRs via MPRs is self-evident (for semiotic, psycho- and sociolinguístic reasons). Also the criterion of regularity (cf. Dressler 1985a:65ff.), in terms of uniformity of change from input to output, is related to phonological and morphological diagrammaticity, insofar as the structural change of PRs can be described with a distinctive feature change which is identical for all inputs, whereas such a description is often non-sufficient for MPRs. Morphological uniformity (one of Mayerthaler's 1981 basic principles of Natural Morphol­ ogy) is highest with non-allomorphic MRs. Since Naturalness Theory claims that the relative degree of naturalness results in relative frequency, we may expect that MPRs are both less frequent than PRs and than MRs, provided that we assume that phonological natural­ ness is irrelevant for MRs, and morphological naturalness for PRs, whereas they are both relevant for MPRs, with the effect that MPRs represent, so to say, the turning point or valley between morphology and phonology. This is an empirically testable claim, as are many others deductively derived in this chapter.

74

TRUBETZKOY' S ORPHAN

Another semiotic parameter, connected with the property of the reliability of signs, is represented by the ordered set of (context-sensitive!) 'biuniqueness-uniqueness-simple ambiguity-multiple ambiguity'. Evidently only allophonic PRs and non-allomorphic MRs may be biunique, whereas phonemic neutralizations, MPRs and allomorphic MRs are not. A weaker form of biuniqueness, the non-dependence of rule application on lexical factors, is at the basis of the frequently used criteria of generality, obligatoriness, productiv­ ity and recoverability/inferability (cf. Dressler 1985a:82ff., 89ff., 92ff., 129ff.). Consequences in external evidence are the relative frequency of loan­ ing and of speech errors. Both loaning of MPRs (cf. Dressler 1985a: 103ff.; 1992) and occurrence of MPRs in speech errors (cf. Dressler 1985a:105ff.) is less frequent than of PRs and MRs. A third semiotic parameter is indexicality. An index is a sign which exhibits a direct connection between indexicai signans and signatum. All contextsensitive rules are indexicai, because they point to their relevant context of application. For PRs, the phonological context is much more important than the morphological, in other words: distribution is more important than alterna­ tion in phonology (quite in contrast to usual claims in generative phonology). For MRs the morphological context is all-important, a phonological context relatively seldom relevant, unless in cases of dissimilatory allomorphy, as in the substitution of the Hungarian 2nd sg. /s/-suffix by an /l/-suffix, for example in autóz-ol"youdrive a car" (cf. also Carstairs 1990). For MPRs, both phonological and morphological indexicality count, as in the often cited E. electric-ity class, where velar softening applies before a small set of suffixes which have a palatalization-favoring initial segment. Pre­ dictably, morphological indexicality counts more. For there is a semiotic priority of morphology over phonology.10 This interplay of morphological and phonological indexicality is at the basis of the frequently used criteria of phonological vs. morphological conditioning or domains of rule application (cf. Dressler 1985a:71-82, 126ff.). The semiotic priority of morphology over phonology is also at the basis of the rule order: first MRs, then MPRs, finally PRs (cf. Dressler 1985a: 111ff. and the studies in Lexical and Cyclic Phonology). The essence of this part 4 has been so far to show that the basic descriptive criteria and results of differentiating MPRs from PRs and MRs can be derived from very few basic semiotic principles, in addition to the basic functions of phonology and morphology (3). If these principles are interconnected (and 10

Here I will not discuss the consequences of semiotic priority of words over morphemes and MRs, which explains lexicalisation.

MODULARITY, MORPHONOLOGY, AND GRADIENCE

75

very few minor principles added), the explanatory power of the model is still increased (cf. Dressler 1985a, chapter 10). In other words, from the universal preferences we can derive testable claims. Since many intervening variables interfere, testability boils down to statistical significance, similar to the situation in natural and social sciences. What remains to be shown is the fuzziness of the 'upper' and 'lower boundaries' of morphonology against morphology and phonology. This fuzziness has already been predicted in 3 because of functional considerations: Morphonology has no function strictly limited to all and only MPRs. The same is predicted by our semiotic model here in 4: most differences between MPRs and either PRs or AMRs cited are only gradual or the borderlines have to be drawn outside morphonology, i.e., within phonology or within morphology. Moreover, as I have evidenced over and over again in Dressier (1985a), there is not a single positive property that holds for morphonology alone, i.e., for all and only MPRs. Let me exemplify the gradual transition between PRs and MPRs with German and Russian final devoicing, two often cited phonemic neutralizing PRs. 11 However, in contrast to a prototypical PR, they allow certain exceptions to obligatoriness and productivity, particularly in extragrammatical morphology: a) in abbreviations, their application can be blocked: thus many German speakers clearly distinguish Log and Lok, abbreviations of Loga­ rithmus "logarithm" and Lokomotive "train engine", respectively.12 Similarly the pronounciation of the abbreviation of R. ministerstvo inostrannyx del "ministry of foreign affairs" fluctuates between [m'id] and [m'it] (Comrie & Stone 1978:79). b) In subtractive extragrammatical Russian vocatives, final devoicing does not apply (cf. Dressler 1985a:93): Lida→Lid, Leva → Lev, Ugo → Ug. In all other respects, however, both German and Russian final devoicing seem to operate like normal phonemic neutralizing PRs. Problems for well distinguishing MPRs and AMRs appear, e.g., with Polish affricate spirantization of dz → z, which follows Polish allomorphic postalveolar formation g→di and is dissimilatorily blocked by preceding і (which may be due to shibilant assimilation), as in the diminutives bóg 11 They neutralize final obstruents, irrespective of the question whether the preceding vowel may be longer before an input voiced obstruent than before an input voiceless one (cf. Dinnsen 1985). 12 This behavior is quite different from 'distinctive non-application' of final devoicing in technical terms (i.e., in language for special purposes as opposed to everyday language where abbreviations such as Lok and Log are common). An example would be the distinction made by many speakers between Sulfit "sulphite" and Sulfid "sulphide" (with final [d], pronounced without final devoicing).

76

TRUBETZKOY'S ORPHAN

"god" → DIM boi-, but mózg "brain" → móżdż-ek, drobiazg "trifle" → drobiaidi-ek. Thus there is an (indirect) morphological conditioning and one (negative) phonological condition. Since other criteria of its application do not allow a distinction between MPR and AMR, it is questionable whether this one phonological condition is enough to classify Polish affricate spirantization as a MPR. 5. Diachronic Morphonological Change The type of MPRs discussed in this contribution originates from PRs, more precisely, there is a unidirectional change of morphologization a) from (phonemic neutralizing) PRs to MPRs (or, in terms of Lexical Phonology, of postlexical to lexical rules) and b) from MPRs to MRs (unless either PRs or MPRs are eliminated). In my view, the main mechanism of this change is a shift from phonological to morphological indexicality, triggered again by the semiotic precedence of morphology over phonology (cf. 4). Let me illustrate this again with Polish palatalizations (see 3, 4 above). The PR of surface palatalization palatalizes consonants before the most palatal segments /i, j / and it palatalizes /k, g/ before some morphs in -e (see below), in conformity with the universal phonological hierarchy of palatalization. The historical 1st palatalization (or synchronic postalveolar formation), however, transforms /k, g, x, h/→ into cz, di, sz, i [č, dž, š, ž] before a few declensional suffixes (e.g., recessive vocative kozacz-e from kozak "cossack"), many conjugational and derivational suffixes, largely irrespective of their initial segment. Thus its context is primarily morphological and only marginally phonological. In the imperative płacz! of płak-ać "to weep" there is even no phonological context (i.e., due to rule split, this imperative formation rule has become an AMR). This represents a shift from primarily phonological indexicality (PR) to primarily morphological indexicality (MPR), and finally to morphological indexicality alone (MR). This shift has even initiated with surface palatalization in the phonologically least favorable context, i.e., before /e/, where it applies only before a few suffixes, e.g., the instrumental masc. sg. suffix /em/, as in Polak "Pole" → Polaki-em, Maks Plank → Maks(-em) Planki-em. And there exist minimal pairs distinguished by the contrast [k] vs. [k'] before /e/, as in Acc. sg. Polsk- [polske] "Poland" vs. Nom./Acc. pl. Adj. polski-ę [polsk'e] "Polish". A shift from primarily phonological indexicality to primarily morphological indexicality follows from the semiotic priority of morphology over phonology. If we look for the locus of change (or the scenario of change), then we can argue that the realization of this priority and the resulting morphologization of

MODULARITY, MORPHONOLOGY, AND GRADIENCE

77

neutralizing PRs presuppose the acquisition of a fair amount of morphology and the awareness of the system adequacy of morphonological alternations. Therefore young children cannot be candidates for being the perpetrators of morphologization. Moreover, many children acquire some MPRs of their language only when in elementary school (cf. Dressler 1985a:246ff.), i.e., they reproduce MPRs by analyzing alternations fairly late in the acquisition of both phonology and morphology. Finally, young children have been observed to reanalyze very general MPRs as PRs (cf. Gósy 1989); thus, later on, they have to re-re-analyze them as MPRs. All this points to older children as potential sources of morphologization of PRs. I am not aware of any com­ parable observations with adolescents and adults. Last not least, we must envisage the two following alternatives (a and b): a) if recoding a PR as an MPR is a discrete change, then this means that a PR is given up. And since the acquisition of PRs in terms of natural processes is considered to be finished with childhood, i.e., when the articulatory basis is fixed, the originators of morphologization must be children, i.e., older children, because of the previous arguments. b) The alternative is to consider the morphologization of PRs as a gradual change, in line with the view that the distinction between PRs and MPRs is not discrete but gradual (see above). This leads us to the next question within the theory of diachronic change: when does phonological indexicality cease to exist? Per definitionem, this should occur, in general, in the transition from an MPR to an AMR. But does phonological indexicality still play a role in MPRs? Or does it belong to dead redundancy? This would be in line with Newmeyer's (1994:67) observation: "it is frequently the case that the original motivating factors have ceased to exist". Reasons for my assumption that phonological indexicality is at least vestigial in MPRs is the productivity or at least apparent activity of phono­ logical contexts/domains/conditions of MPRs, particularly the observation that phonological hierarchies still play a role in the application of MPRs. For example, there is still a marked difference in the productivity of the Italian palatalizationk→č before -i and before -e (cf. Dressler 1985a: 168ff.), as in ami[k]o "(male) friend" → pl. ami[6]-i vs. ami[k]a "female friend" → pi. ami[k]-e; gre[k]o "Greek"→gre[č]-izzare "to hellenize" vs. gre[k]-eggiare "to use Greek expressions"; musi[k]a "music"→musi[č]-ista "musician" vs. musi[k]-essa "female musician" (suffix -essa < late Lat. -).13 13

Maiden (1991:105) has objected that fem. PL -e derives from the Latin diphthong -ae; this objection does not hold for the other cases though, and [kae] has changed to [č] in many words anyway, such as in Lat. Caesar(em) > It. Cesare, toponym Caere > Cer(veteri), caelus

78

TRUBETZKOY' S ORPHAN

In Dressler (1985a:295), I have interpreted such phenomena as quantitative change of distinguishing criteria which must accumulate in order to effect qualitative change from a PR to an MPR or from an MPR to an MR. At least, such gradual transitions speak against those models which predict a si­ multaneous change of all relevant criteria when a PR turns into an MPR or an MPR into an MR. So far I have not seen any serious counter-example violating the constraint of morphologization being a unidirectional type of change, i.e., one which has no counterpart in 'phonologization' of MRs or MPRs (cf. van Coetsem & McCormick 1990; Mendez Dosuna & Pensado 1986). Of course, replacement of a PR by a different MPR does not contradict the claim: a case in point, is the replacement of older Romance phonological total assimilation by Italian 'radoppiamento fonosintattico', i.e., by initial gemination triggered by prosod­ ic, morphological and lexical factors, as shown in detail by Loporcaro (1994): an example is a lui "to him" → [almi], where the preposition a comes from Lat. ad. The most serious counterexample to the claim of unidirectionality is Morin, Langlois & Varin's (1990) interesting and solid analysis of tensing of wordfinal -o in 19th and 20th century French. The nucleus of their painstaking analysis is the claim that this change presents a phonologization of the rule of plural tensing in, e.g., le gigot "leg of mutton" vs. les gigots which differed only in the word-final lax vs. tense /o/, which would mean a change from an AMR directly to a PR. Since there were (and partially still are) some lexical exceptions, such as Pierrot and trop "too" (Morin et al. 1990:518ff.), there must have been a process of lexical diffusion. They exclude that this may present the diffusion of a phonological process, because if it were "it should have applied uniformly across the lexicon" (p.519). There are, however, good examples where lexical diffusion of a PR respects lexical and morphological categories (cf. Dressler 1978:152) and this may be the case here as well. Semiotic priority of morphology over phonology and ensuing change from phonological to morphological indexicality explain the unidirectionality of change from PRs to MRs and thus represent the general motivation or cause of morphologization. This general cause must be supplemented, in the sense of ubiquitous multicausality of diachronic change, by specific causes, such as opacification by other PRs, as in the cases of final devoicing in Yiddish and several German dialects (cf. Shannon 1987), and vowel change in Andalusian Spanish (cf. Manaster Ramer 1987). Let me illustrate this unidirectional change of morphologization with a 'heaven' > c(i)elo, etc.

MODULARITY, MORPHONOLOGY, AND GRADIFNCE

79

change from an Ancient Greek PR to a Modern Greek MPR, a change whose most important steps took place in late Ancient Greek koiné (period of late Hellenism and Roman Empire): Classical Greek had a PR of assimilation of manner: (1) as shown by distribution (cf. Lupas 1972) as in /thm, tm, dm, khm, km, gm, *phm, *pm, *bm/) and alternation as in blépō"\ look", kléptō"\ steal", krúptō "I hide", koptō "I hit, cut", tribō "I rub", gráphō "I write" vs. passive perfect participles beblemménos, keklemménos, kekrumménos, kekomménos, tetrimménos, gegramménos. The glide /w/ did not undergo this PR, as in páwō "I cease", passive perfect participle pepawménos, cf. pnéwma "spirit". In the koiné, the following sound changes took place: (2)

b, w > ß > v p h >f

Much later consonant degemination occurred. In Modern Greek we have the corresponding distribution: vléma "look" vs. pnévma and the alternations vlépo - vleménos, klépto - kleménos, krivo kriménos, kóvo - koménos, trívo - triménos, grafo - graménos vs. pávo pavménos, filévo "I have a guest" - filevménos, jirévo "I seek" - jirevménos (cf. Koutsoudas 1962:9f.; Joseph & Philippaki Warburton 1987:236-240). Thus an AGk. assimilation PR changed into an MGk. MPR, whereas degemination is still a PR. Although the MPR still maintains the phonological characteristics of an assimilation rule, its application is restricted to certain morphological categories with lexical exceptions; the importance of triggering factors is thus: morphological > lexical > phonological. This illustrates the general cause of shift of indexicality, and the specific cause of opacification. The claim of unidirectionality of successive change from PRs to MRs via MPRs is strengthened by the generalization that also allophonic PRs may change unilaterally into phonemic neutralizing PRs, but not vice versa (cf. also Labov 1994 on apparent merger). A further argument can be adduced from the parallel process of morphologization of syntax, i.e., various types of grammaticalization (cf. Hopper & Traugott 1993; Compes, Kutscher & Rudorf 1993), which also represent a unidirectional change towards morphology. Within morphology, however, inflectional MRs may change into derivational MRs and vice versa, allomorphic MRs may become non-allomorphic and vice versa. This speaks, again, against including MPRs and morphonology within morphology.

80

TRUBETZKOY'S ORPHAN

MPRs, since they originate from PRs, are "old rules" in Wurzel's (1994) terms.14 MPRs share with 'old' MRs the property of being diachronically recessive. There is, however, the problem that some AMRs (which are also 'old', insofar as they have originated from MPRs) may expand, whereas MPRs do not.15 Examples of such expansive allomorphic MRs are German umlaut plurals of the class Mops "pug", General "general" → pl. Möps-e, Generäl-e (cf. Dressler 1985a:92) and Portuguese plural formation of the type coraçaõ "heart" → corações. This, again, is an argument for differentiating MPRs from allomorphic MPRs. 6. Morphonology in Language Acquisition Finally, supplementing the above arguments on diachronic change through language acquisition, I want to indicate briefly how first language acquisition fits into my model of morphonology. In Dressier (1985a:246ff.) I have accumulated evidence that MPRs are acquired after phonology and much of morphology is acquired. Both more recent publications of whatever persuasion16 and the results of an ongoing Viennese project on the acquisition of morphology support this view. 7. Remarks on the System Adequacy of Morphonology As already anticipated in 3, a MPR may be more adequate within one part of morphology than within another, for example Polish palatalization →č rather in verbal than in nominal inflection, Polish palatalization →  the reverse. More general, the analogical leveling of MPRs may follow morphological subclassifications, such as the leveling of Italian monophthongization (a rule inversion of an older diphthongization), as in suono "I sound", Inf. sonare to suono, suonare in the first macroclass rather than in the second, as in siedo "I sit", Inf. sedere (cf. Dressler & Thornton 1991). Or as Maiden (1991) has shown, morphonological umlaut is system-adequate in some Italian dialects rather than in others (and in the standard). Or, diachronically, MPRs may split according to morphological categories, as it happened with umlaut in German (cf. Roberge 1980; Wurzel 1984b; Janda 1987).17 14

Where "old" is understood in developmental and not in chronological terms. Exceptional cases of expansion of Italian dialectal MPRs of metaphony and hypermetaphony have been adduced by Maiden (1991). The lack of further material and in my knowledge of Italian dialectology have hindered me so far to check the evidence. 16 E.g., Jones 1991; Plunkett & Marchman 1993; Prasada & Pinker 1993; Marcus et al. 1992. 17 Many more examples in Dressier (1985a:352ff., 82, 90, 179f.). 15

MODULARITY, MORPHONOLOGY, ANDGRADffiNCE

81

Now we can also attempt the characterization of what a prototypical MPR is: its indexicality is primarily morphological, secondarily phonological, values on the scales of semiotic and descriptive criteria for the differentiation of MPRs, PRs, and MRs are medium (for details see 4); it is likely to be acquired neither by pre-schoolers (as vowel harmony in Turkish and Hungarian) nor by adolescents or adults (as some rare MPRs in rather sophisticated loan-words, such as velar softening in G. Rubrik "rubrique"→verb rubriz-ieren), but by older children (depending on the degree of marginality of MPRs and of the richness in morphology, it may be earlier, as in Polish or Russian, or later, as in English); diachronically a prototypical MPR comes from an earlier PR; its probable location depends on system-adequacy. Claims inherent in this definition of a prototype can be empirically tested (although testing is harder than with a discrete definition): prototypical PRs should be, crosslinguistically, more frequent and diachronically more stable than non-prototypical ones. In addition, the parts of the definition must be mutually coherent. 8. Conclusion Of all the reviews of Dressier (1985a), the most penetrating one is Singh's (1986) review article. Thus a brief reply to at least some of his arguments is in order: The first critique I want to reply to, is his statement (p.352) that I underrated intramodular dialectics. To this I can answer that 1) Marxist dialectics (what is commonly understood as dialectics and also marginally used in my tenth chapter) a) has the basic problem of reifying HegeUan dialectics of ideas, b) most of the time it is not restrictive enough, too rich in explanation (although I still find F. Engels' concept of the leap from quantity to quality a useful one), c) conflict theory still strikes me as more important for morphonology than a cooperation model (à la Maiden 1991). 2) (cf. p.353): intramodular means that morphonology should be taken as part of morphology, to which I will reply below. Second, Singh (1986:354) thinks that "[h]e [= Dressier] seems to take too much for granted in morphology". This refers A) to the concept of morphology-less languages: but this concept holds only: a), for the ideal (in the sense of a construct, as envisaged by V. Skalička) of an isolating language; b) it holds only for morphological grammar (as distinguished from extragrammatical operations, cf. my section 3 above); c) this concept gets its full impact with the assumption that only little of modularity is innate (cf. Dressler & Karpf 1995), with the effect that morphology as an innate module or grammatical submodule of its own becomes dispensable.

82

TRUBEIZKOY' S ORPHAN

) Singh has no directional morphology, i.e., MRs: but a) notions of basic vs. derived morphological structures and derivation paths are best captured by rules and rule ordering (as argued in classical generative phonology); b) one observes performance of stepwise output derivations, which serves as an analogue to a rule model; c) bidirectional morphology cannot properly distinguish backformation from MRs. In classical generative grammar, it has often been claimed that a rule model of competence does not imply that such rules are actually used in performance, a perfectly legitimate view (although I do not share it), the assumption of a 'static' model of morphology is obliged to the more problematic opposite view: the model contains only static alternations or relations between different representations, whereas there is evidence from performance for the dynamic derivation of derived representations from (more) basic ones. Such evidence comes, for example, from fieldwork. For example, in field-work on Sepecides Romani, my student Petra Cech found out that native speakers invariably form infrequent or non-existing but potential inchoatives in two steps: first they form the plural of the noun presented to them, and then they add the verbal suffixes of the derived inchoative verb — and this corresponds exactly to the grammatical derivation of an inchoative form from a noun via its nominative plural. Similar evidence comes from morphological production of aphasics (cf., e.g., Dressler & Dziubalska-Kołaczyk 1994) and from the way nonceforms (either neologisms or occasionalisms) are formed, particularly by children. Finally, d) a dynamic model of morphology allows a closer interaction between diachrony and synchrony, a standard argument of rule-models, which becomes more critical with the concept of scenario of diachronic change (cf. Dressler 1994). Third, Singh (1986:353) criticizes my assumption of a 'quasi-module of morphonology', whereas he considers morphonology to be a part of morphology, a) His interpretation of my model as containing a quasi-module of morphonology is not my own, in fact I have claimed that morphonology represents the area of interaction of (the modules of) morphology and phonology. If I may briefly add now, I understand this as meaning that morphonology corresponds to a non-modularized part of the grammar module, with a continuum of substantial criteria which gradually distinguish MPRs from PRs and AMRs. A similar (albeit less radical) view is expressed by Booij (1993) in regard to what I call non-prototypical inflectional morphology: he differentiates between internal inflection (roughly corresponding to my nonprototypical inflection as of Dressier 1989b) and contextual inflection (roughly

MODULARITY, MORPHONOLOGY, AND GRADIENCE

83

corresponding to my prototypical inflection), but he considers these two areas not as two discrete components or submodules, but as groups of rules which differ by a varying number of substantial properties. b) It has to be established that MPRs are not simply an integral part of MRs. The same problem holds, still more critical, for AMRs. Thus let me give diachronic external evidence that even an AMR is not an integral part of an MR. My case 1) an AMR. of umlaut being triggered by a MR of diminutive suffixation, and 2) influence of standard Dutch on the Dutch dialect of Twente [on the German border], as studied by van Bree (1994). Diminutives are formed in Standard Dutch by the suffixes -je, -tje, -etje without umlaut, but in the Twente dialect by the umlaut-triggering suffixes ֊(e)ke, ֊she. Now the crucial question arises: do suffixes and umlaut, in the process of adaptation to the standard, change together? Do they change independently? i.e., is umlaut an independent rule or simply part of the MR of diminutive formation? Consider the following common base forms and the diminutive forms of Standard Dutch (St.), in the Twente dialect (Tw.) and the transition forms of the dialect under influence of the Standard (Tr.): (3) boom "tree": St. boompje, Tw. b , Tr. boomke, *beumpje kan "can": St. kannetje, Tw. , Tr. kenneke, kanneket *kentje, *kennetje stok "stick": St. stokje, Tw. stökske, stökke, Tr. stokske, stokke, *stökje. Thus transition forms with independent umlaut do not occur spontaneously, and such forms very rarely occur in the tests administered by van Bree (1994). This is evidence for a separate AMR of umlaut (which comes from an earlier MPR) triggered by diminutive suffixation. And if this counts for the separate status of an AMR, we may assume, a fortiori, the same for MPRs of umlaut formation. c) For the reasons given in my section 5, the unidirectionality of change from MPRs to MRs but the bidirectionality of changes within morphology speak for the differentiation of morphonology and morphology. d) the vestigial phonologicity of MPRs points in the same direction. With this presentation I hope to have been able to clarify, and elaborate on, my model of Dressier (1985a), to have bolstered up argumentation against the view that morphonology is a part of morphology (as in older literature and in Darden 1989; Singh 1986, 1990a:111), and to have addressed Carstairs's (1986) critique in stressing empirically testable predictions the model is able to make.

Form & Content in a Functionalist Semiotic Model of Morphonology: Comments on Dressier* Richard Janda University of Chicago Dressler's paper not only "constantly refer[s] to the... [eleven] chapters" of his long 1985 book Morphonology: The dynamics of derivation, but even declares the model which it argues for to be 'essentially identical' to the one in that monograph (cf. §1, "Introduction"). Given both this close but asymmetrical relationship of article to book and the fact that Dressier (1985a) refers occasionally to Albanian, I may perhaps be forgiven for comparing the assignment of critiquing "A Functionalist Semiotic Model of Morphonology" to the task which faced any country dealing with the Albanian/MainlandChinese bloc during the period ca. 1961-1977. That is, after President Enver Hoxa had broken with the USSR and made the People's Socialist Republic of Albania a client state of the People's Republic of China, his propagandists occasionally referred to Albania and China as "together a billion strong",1 but it was always clear that the security of the PSRA (with a population of under three million) depended overwhelmingly more on the forces of the PRC than on its own. Similarly, Dressler's current paper represents only a tip compared to the iceberg of his earlier monograph, and the challenge of discussing a model associated with a work of such titanic proportions (nearly 400 pages of text and 60 pages of references, plus 930 footnotes, with examples drawn from more than 200 languages) obviously cannot be met with anything approaching completeness by a short critique such as this.2 For helpful comments and other aid in the preparation of this paper, I would like to express my gratitude to all the participants at the Table Ronde sur la Morphophonologie (Réponses contemporaines) held at the Université de Montréal from September 30 to October 2, 1994 — especially its organizer, the more-than-patient R. Singh, and W.U. Dressier — as well as to L. Dobrin, J. Goldsmith, J. Sadock, H. Suzuki, and B. Vance. 1 This account admittedly sounds apocryphal, but I have heard it confirmed both by National Public Radio correspondent Daniel Schorr and by my Albanologist colleague Victor Friedman. 2 A comprehensive critique of Dressier (1985a) is, however, available in the form of the

MODULARITY, MORPHONOLOGY, AND GRADIENCE

85

Instead, I will focus on a small complex of largely related issues, directing most of my attention to the elaboration of probing questions intended to elicit clarification of key points. The reason for this is that, despite Dressier's extensive writings on morphonology3 as a sphere of interaction between Natural Phonology and Natural Morphology, his use of functional and semiotic criteria to a degree unfamiliar in most other theories has sometimes met with non-understanding or even misunderstanding of his model on the part of formally-oriented phonologists and morphologists. In this regard, however, I will be concerned more with the grammatical (especially morphonological) consequences of the appeal to functionalism and semiotics in his paper than with specific invocations of these orientations themselves (for these, see his §2, "Functional Analysis", or §7, "Remarks on the System Adequacy of Morphonology"). Yet, before leaving the latter topics, it is appropriate to note that, in already advocating a hierarchy of variably rankable and hence violable constraints, Dressier (1985a) showed himself to be a practitioner of Optimality Theory before the fact (cf., e.g., Prince & Smolensky 1993, as well as the related but distinct proposals of Sadock 1995b). Still, it is hard to avoid the conclusion that the particular combination of a morphonological focus with a semiotic perspective has rendered the overall Natural Phonology/Natural Morphology approach less accessible to formal linguists who display an antifunctional bias. Let me state at the outset, though, that Dressier's latest exposition of his views on morphonology not only already provides a large and welcome measure of additional clarification for, but also elaborates on, the 1985 version of his model — most notably, by emphasizing factual predictions in such spheres as diachronic morphological reanalysis. Since the author himself review article by Singh (1986), which provides a balanced evaluation of that work from a general perspective similar to my own in most major respects; cf. also Singh (1996) on 'phono(norpho)logy'. 3 Dressler (1985a:l-3) follows Trubetzkoy (1929, 1931, 1957) in employing only a haplological variant — in English, morphonology (derived by subtraction from morphophonology) — and thereby distinguishes his own approach to the subject from related but distinct frameworks which refer to morphophonology or morphophonemics. Since the present discussion focuses centrally on Dressier's model (and contrasts it most closely with a framework which posits no significant domain of morpho(pho)nology), I will likewise make (nearly) exclusive use of the term morphonology — even though the haplology here creates an unfortunate ambiguity, due to the homophony of English more and mor-. The possible confusion at issue is illustrated by a surely apocryphal anecdote of David Stampe's. It is alleged that someone once remarked to Chomsky in 1957, "I like Syntactic Structures very much; the only problem is that it has no morphology"; the author is then said to have replied, "No more what?. Actually, though, there are just as many pages in Chomsky 1957 on which the term morphophonemic(s) appears as ones on which it does not.

86

TRUBETZKOY' S ORPHAN

mentions such elaboration and testability as explicit objectives of his article (cf. §8, "Conclusion"), it must be qualified as a success on both these counts. As for the remaining goal — that of "bolster[ing]... argumentation against the view that morphonology is a part of morphology" — I must confess that the paper did not succeed in changing my mind on this matter. It is here, I believe, that even further clarification is needed within the Functionalist Semiotic model of grammar, both in order to demonstrate (at least to skeptics like Ford & Singh 1996 and myself) that this is indeed an empirical issue, rather than just a terminological one, and in order to establish the precise meaning of the crucial notions involved in the debate, so as to make future exchanges more productive. Some idea of the confusion which still reigns on this score can be gained from the passage in Dressler's current §8 "Conclusion" where he emphasizes that the entity targeted by Singh's (1986) criticisms of an assumed 'quasimodule of morphonology' is not, after all, to be found in Dressier (1985a). As a matter of fact, the term in question occurs only in Zwicky's "Preface" to the latter work. Yet Zwicky's definition of a "quasi-module... [as] a set of rules that is distinguishable from other sets on various criteria but nevertheless lacks characteristic properties of its own" (p. vii) appears to match closely several of Dressler's latest statements made toward the end of his paper's §4, "A Semiotic Model of Morphonology...". There, we read: "Morphonology has no function strictly limited to all and only MPRs",4 and "most differences between MPRs and PRs or allomorphic MRs... are only gradual...[,] or the borderlines have to be drawn outside morphonology, i.e., within phonology or morphol­ ogy".5 "Moreover,... as evidenced... in Dressier (1985a), there is not a single property that holds for morphonology alone, i.e., for all and only MPRs".6 Nor is clarity enhanced by the contradictory-sounding assertion of the §8 4 Cf. also the end of §3, "Functionalism and Morphonology", for a similar statement: "The non-existence of a separate function which holds only for MPRs fits in well with...[the] assumption (held since Dressier 1977) that morphonology is not a grammatical component of its own or, more precisely, a module or submodule of grammar". Following Dressler's practice in both his 1985 monograph and the most recent paper, I use the following abbreviations: MPR for morphono logic al rule, PR for phonological rule, and (A)MR for (allomorphic) morphological rule. 5 This statement is difficult to interpret: i.e., the borderline between two domains can certainly be outside both of them (e.g., on an edge between them), but how can such a boundary be outside one domain and within another? 6 This conclusion does not, however, square exactly with the statement in §3, "Functionalism and Morphonology", that "MPRs do have a function of their own,... [that] of cosignaling MRs (cf. Dressler 1985a:268f.)". Still, whereas "...[t]his cosignaling function... is also shared by many allomorphic MRs...[(AMRs), it] is typical for MPRs, [but]...less so for allomorphic MRs and rather occasional for PRs".

MODULARITY, MORPHONOLOGY, AND GRADIENCE

87

"Conclusion" that "morphonology corresponds to a non-modularized part of the grammar module". And immediately preceding is a difficult-to-interpret statement that "morphonology represents the area of interaction of (the modules of) morphology and phonology", echoing the statement in Dressier (1985a:4) that "...[m]orphonology belongs neither to morphology nor to phonology; it mediates between both components without being itself a basic component". While the proposals of "A Functionalist Semiotic Model of Morphonology" along these lines have suggestive plausibility and are rich in implications, we simply cannot subject them to a full and proper evaluation until we know exactly what is meant by such concepts as 'module' (plus 'grammar module' and 'modularized'), 'submodule', 'area of interaction', 'borderline' or 'bound­ ary', 'component' (plus 'basic component'), and "mediation]... between... components". In reading Dressier (1985a), one does have the feeling that "it's all in there somewhere",7 but, as a practical matter, even the most recent paper on the subject leaves us wishing for a single place (like a short chapter or summary of a few pages' length, or even just a bare-bones table or schematic diagram) where the essentials of the Natural Phonology/ Natural Morphology model of morphonology could all be brought together and set forth explicitly and concisely. Until this need is met, it will remain understandable why even specialists in the field have been tempted to pay perhaps undue attention to Zwicky's "Preface", which in the space of less than a page manages to provide convenient brief characterizations not only of notions like 'quasi-module' but also of the three rule-types PR, MPR, and AMR (on the abbreviations in question, cf. fn. 4). At the same time, it should be stressed that both "A Functionalist Semiotic Model of Morphonology" and Dressier's (1985a) monograph make unmis­ takably clear the essential core-notion of the morphonological theory which they espouse. Namely, each module of sound-structure consists of a set of rules, and boundaries can be established between the various modules because each rule-set possesses a unique range of values for a number of distinguishing properties.8 Since the latter criteria are intended to evaluate not the absolute but the relative extent to which a rule does or does not fulfill particular functions 7

One must sympathize with Dressler's chagrin, expressed in footnote 1 of his response, that the 1985 monograph's "usability... has been reduced by the refusal of the publisher to print a subject index". 8 It sometimes appears, however, that a module can partly be defined, not by the degree to which all rules in it possess a certain functional property, but by the percentage of its rules which possess a high degree of that property. Cf., e.g., what is said about the typicality of the cosignaling function exhibited by MPRs in the remark from Dressler's paper quoted above.

88

TRUBETZKOY' S ORPHAN

(especially ones of signaling, or semiosis), such properties must be viewed as η-ary, scalar features, rather than binary, privative ones. From this, it should follow in principle that — to the extent that values of a given property have the characteristic of density or betweenness (such that, for any two arbitrarily chosen values, it is always possible to find a third, intermediate value) — the transitions between rule sets and hence modules must be continuous, and indeed the §8 "Conclusion" of Dressier's latest paper mentions "a continuum of substantial criteria which gradually distinguish MPRs from PRs and AMRs". Nevertheless, Dressier (1985a) stresses at several points that, in practice, "such a continuum cannot be represented by a steady slope but by a curve", because "...[t]he transition interval... is rather abrupt" (cf., e.g., p. 21), and Zwicky's "Preface" gives the characterization that "the transitions between (quasi-)modules, though not sharp, are steep; most rules clearly belong in one module or the other". Still, both authors refer to blurring between modules; cf. Dressler's (1985a:3) statement that "there are proto­ typical... MPRs... with fuzzy boundaries to... PRs... and... AMRs". Here, it is crucial that, due to the scalar nature of the properties which distinguish the rule types across the various modules (and assign differing degrees of prototypicality to particular rules within a single module), it is possible for the rules of a given module to share with those of another module the characteristic of having at least a certain value for some property while still having values so low that they fail to count (to reach critical mass, as it were) for some evaluative purpose. This is what leads to the otherwise paradoxicalseeming situation that, in the Natural Phonology/Natural Morphology framework, morphonology is claimed to share significant properties with both mor­ phology and phonology while at the same time being neither a part of one of those modules nor a module of its own. On the one hand, this combination of what Zwicky's "Preface" calls 'homogenist' and 'modularist' assumptions endows the Functionalist Semiotic approach with greatflexibilityand the potential for making and testing a wide spectrum of empirical predictions. But, on the other hand, the fact that the model strives to posit multiple sets of scalar properties in order to differentiate across (and within) rule sets also results in significant complexity which makes it difficult to verify such predictions in practice. While many of the properties are assumed to correlate, their largely continuous nature tends to yield a situation where the theory is hard-pressed to constrain the varied individual property-values and value-combinations which are applicable to rules. More importantly, the fact that relative and not absolute values of properties are criterial for defining modules (and non-modules) allows and perhaps even

MODULARITY, MORPHONOLOGY, AND GRADIENCE

89

guarantees a certain amount of arbitrariness in determining where the cut-off values between modules are. In the worst case, this could threaten to color one prominent portion of the overall enterprise with a distinctly taxonomic tinge that would allow morphonology to be defined and grammatically situated at least partly according to individual taste. This conclusion can best be illustrated with reference to a graphic representation of the overall Natural Morphology/Natural Phonology approach to morphonology. Accordingly, consider the diagram below, which I have constructed on the basis of the information in both Dressier (1985a) — especially the graphs for the continuum from rule-governed alternations to suppletion on p. 21 — and "A Functionalist Semiotic Model..." — particularly the remark in §3, "A Semiotic Model...", that "MPRs represent... the turning point or valley between morphology and phonology": (1) One interpretation of the Natural Phonology / Natural Morphology model of sound-structure:

This construct is to be viewed as primarily a horizontal scale along which both specific exemplars and general types of sound-structural rules can be ordered, from left to right, according to a simultaneous set of functional and/or semiotic defining-properties (and evaluative criteria) which the individual rules and rulesets possess to varying degrees, shown by a vertical scale from low to high. For simplicity's sake, a single line descending from the upper left toward the lower right combines all those properties whose functions relate in some way to phonological naturalness (e.g., cf. Dressler's current footnote 6 and 1985a:59ff., 68ff. on 'phonetic plausibility', 'process matching', and 'pronounceability/perceptibility'). Similarly, one line ascending from a point near the lower left up to the upper right combines all properties that are in any way functionally associated with the promotion of morphological naturalness (e.g.,

90

TRUBETZKOY'S ORPHAN

§3 of Dressler's current paper mentions "lexical enrichment", "providing syntax with inflectional word forms", and the avoidance of "morphotactic opacity"). The gray columns represent the abovementioned 'fuzzy boundaries' at the outside borders of phonology and its PRs (on/from the left) and of morphology and its (A)MRs (on/from the right); they are thus aligned with those parts of the phonological and morphological naturalness-curves which show an abrupt increase or decrease in value. The arrows and ellipses at the inside edges of phonology and morphology are intended to suggest that these domains extend at least into the gray areas (and perhaps beyond — a point which I take up again further below). Morphonology and its MPRs are situated in the middle, between the two gray areas (into which the arrows and ellipses likewise show an extension), centering on the cross-over point of intersection between the decreasing vs. increasing scales of phonological vs. morphological naturalness, respectively. To the right, within morphology, an internal division has been indicated between allomorphic rules (Dressler's 1985a: 12 "AMRs... that govern morphologically and/or lexically conditioned allomorphy"), which are aligned toward the left, vs. non-allomorphic rules (cf. Dressler's 1985a:22 "spell-out MR[s]"), located at the far right. To avoid clutter, the leftward-oriented arrow placed before the label "Allomorphy" has not been extended across the diagram through the two gray areas and beyond, as actually should have been the case, but its triple arrowheads and their directionality should be understood as suggesting this, since MPRs as well as many PRs are likewise responsible for creating allomorphy. Before this graphic representation of the Natural Phonology/Natural Morphology model is used to illustrate a number of questions raised by such an approach to grammatical organization, two caveats are in order — one concerned with idealization and the other involving artificially imposed determinacy as a solution to problems due to my insufficient or faulty un­ derstanding. First, it is worth emphasizing again that Dressler's Functionalist Semiotic framework takes pains to include a maximal number of evaluative criteria (properties) that bear on phonological and morphological naturalness. A less abstract diagram would therefore presumably show a large quantity of non-identical but largely parallel curves descending from upper left to lower right and from upper right to lower left (plus ones which might follow these trajectories for only part of their course), with the two sets of lines tending to cluster like a bundle of dialectological isoglosses. It is possible that even related properties might differ in where their curves bend abruptly (as marks of a transition between rule types); this phenomenon may in fact be what defines

MODULARITY, MORPHONOLOGY, ANDGRADffiNCE

91

the abovementioned gray areas and is thus responsible for the fuzziness of boundaries between rule sets. In addition, a more realistic diagram would likewise surely not present two naturalness-curves which are completely symmetrical mirror-images of each other (as is the case above), and so their intersection might well lie closer to one of the borderline transition-areas marked here by gray columns. Along the same lines, it is further likely that their transition points would not occur at the same 'height' on the vertical scale — if, indeed, it is even valid to imagine that numerous related properties can all be calibrated in comparable fashion on the same (sort of) scale, let alone two sets of disparate criteria. Second (and relatedly), the present placement of the lowest portions on the phonological and morphological naturalness-curves represents nothing more than a compromise guess motivated by Dressler's remark in his §4 "...Semiotic Model..." that "phonological naturalness is irrelevant for MRs, and morphological naturalness for PRs, whereas they are both relevant for MPRs". After all, this implies that some residue from each kind of naturalness is present for at least some rules in what can be considered the (polar) opposite module (i.e., that phonological naturalness is present to a small degree in a few morphological rules, and vice versa). Otherwise, of course, it would have been possible to align the two curves so that they both reach zero somewhere in the transitional gray areas. There remains another possibility, however: it is imag­ inable that phonological and/or morphological naturalness could in fact be asymptotic, so that one or both curves might stretch all the way from corner to corner, rather than petering out halfway through the opposite module. Not being able to find conclusive evidence in Dressler's most recent paper or 1985 monograph for either of the latter two module-marginal endpoints, though, I have instead opted for a middle-ground compromise-solution which seems intuitively plausible but which I would not insist on. With the above diagram portraying the Natural Phonology/Natural Morphology model of sound-structure available as at least a schematic basis for reference, it is now possible to address some of the previously mentioned concerns regarding both the strong points and the weak points of the Func­ tionalist Semiotic approach. Let us start with the former, and note that the model has solid empirical underpinnings. In particular, it makes rather so­ phisticated predictions concerning the correlations which should obtain between and among the values that a rule can exhibit for a wide range of functional (and/or semiotic) properties — including not only direct correlations (within a bundle of related phonological or morphological properties) but also inverse correlations (between phonological vs. morphological properties).

92

TRUBETZKOY'S ORPHAN

Precisely because these functional properties and their correlations are both the essence and the bulk of the theory, though, additional issues — like whether morphonology is its own module, an independent non-module, or part of another module — turn out to have much less intrinsic interest or factual content. For one thing, the two principal transition-points — between PRs vs. MPRs, and between MPRs vs. (A)MRs — exist independently of the specific names or modular statuses which are attributed to the three (or four) rule-sets that they separate. Moreover, once one makes the homogenist assumption that two rule-sets may be said to 'share a property' even if they possess that feature to noticeably different degrees, it begins to seem arbitrary that only the rule set which exhibits the property to the highest degree, or with a value above a certain cut-off point, can be said to have it as a defining characteristic. But this is exactly what occurs in the Functionalist Semiotic approach when MPRs are said to "possess no properties of their own". Still, what synchronic evidence is there to stop us from defining the possession of, say, morphological nat­ uralness in terms of having any value above the point where the curve for this property(-set) shows a first abrupt transition upward (across the left-side gray column)? If this step is taken, then MPRs and AMRs (and MRs) come to share a major property, to the exlusion of PRs, and one can just as well say that morphonology and morphology belong to the same module — or that mor­ phonology is part of morphology. Yet this is where taxonomic arbitrariness threatens to enter the picture, since one could also define phonological naturalness in terms of having any value above the point where the curve for this property(-set) shows a second abrupt transition downward (across the right-side gray column) — which would allow us to say that phonology and morphonology belong to the same module, or that morphonology is part of phonology. One could even concoct a new functional property defined as the degree to which a rule has intermediate values for the other two characteristics (phonological and morphological natu­ ralness), in the manner of the Jakobsonian acoustic feature [compact] (cf. Jakobson, Fant & Halle 1952:27ff. on, e.g., the formant values of [a]). This composite feature would show a line whose curve rises from PRs to prototypical MPRs and then falls again to AMRs and other MRs, thereby allowing a claim that morphonology is its own module because it uniquely possesses high values for this property.9 Given such a situation allowing this 9

There is actually no need to invent a composite property in order to come up with a characteristic whose peak-values are borne by MPRs; recall from footnote 6 that Dressier's §3, "Functionalism and Morphonology", already gives essentially this kind of description for

MODULARITY, MORPHONOLOGY, AND GRADIENCE

93

kind of latitude for variant interpretations of sound-structural organization, it seems no less arbitrary to follow the alternative practice of anti-morpho(pho)nologists like Ford & Singh (1996) and Janda (1987), who use a single absolute (privative) binary criterion to locate the boundary of morphology visà-vis other domains. Namely, any sound-structural rule that makes reference to one or more morphological features is a morphological rule, period — so that the boundary between phonology and morphology is absolutely sharp and clear, even if MPRs and AMRs may differ in phonological naturalness. And morpho(pho)nology can be given a similarly non-fuzzy definition: it is the set of all morphological rules that make use of operations where anything other than simple addition is involved (hence subtraction, replacement, permutation, and combinations thereof). Within the Functionalist Semiotic model, however, perhaps the most intriguing alternative view of morphonology is one that actually sounds much more like Dressler's informal characterization of morphonology as being (or arising from) the interaction of phonology and morphology. One could define phonology as extending from the far left all the way to the second abrupt downward transition of the phonological-naturalness curve (at the right-side gray column), and morphology as correspondingly extending from the far right all the way to the second abrupt downward transition of the morphologicalnaturalness curve (at the left-side gray column). This would make morpho­ nology literally the area of overlap between phonology and morphology, thereby explaining why it shares characteristics with both of those modules and so is not an independent (non-overlapping) module of its own; such an analysis could be labeled 'Non-Linear Morphonology'. On such a view, the two gray areas would represent the blurring of boundaries, not only between PRs vs. MPRs and MPRs vs. AMRs, but also between phonology and morphology, in two places. This would suggest that the core of the above diagram — the bottom horizontal scale along which rules and rule types are arranged — is indeed a single continuum of sound-structure extending from prototypical PRs via prototypical MPRs and AMRs to prototypical MRs (as Dressier implies in various places), with those internal divisions being dis­ tinguishable by reference to the two areas showing abrupt transitions. But the view just adumbrated goes in the direction of suggesting that there is actually only a single continuous module of sound-structure, rather than two (or two the relation between MPRs and the cosignaling function. It is not clear why this does not suffice to make morphonology a distinct module, but perhaps module status is itself a gradient notion which depends on the relative number of peak properties that a rule type possesses, and one is simply insufficient.

94

TRUBETZKOY' S ORPHAN

and a half) dichotomous ones. To summarize, then, what I have suggested is that such a conclusion regarding the variety of possible interpretations which a Functionalist Semiotic model leaves open for the organization of phonology, morphology, and morphonology is a result of the fact that, once one has accepted an arrangement with two overlapping and inversely-related property-curves, each marked by a pair of abrupt internal transitions, there is not really much more to say. At least as regards synchrony, virtually everything having to do with the graded horizontal ranking of rules falls out automatically from their values on the two naturalness-scales. Due to the availability of two abrupt transition-points as anchors for internal divisions, the rest is mainly taxonomy and terminology, such as the question of whether or not morphonology is independent and/or modular.10 This then makes diachrony the major place to look for possible empirical evidence bearing on the latter two issues. And that aspect of the Natural Phonology/Natural Morphology model is precisely what is focused on by both Dressier (1985a) and his current paper's §5, "Diachronic Morphonological Change". Such a focus unfortunately entails that references to his earlier monograph must substitute for the length which one might otherwise have hoped to find in §6, "Morphonology in Language Acquisition" — and the same holds true for aphasia, which is omitted entirely, despite the volume of Dressler's previous (and continuing) work on that topic (cf., e.g., Dressier 1978). In any case, the key diachronic claim of "A Functionalist Semiotic Model of Morphology" is that there is "a unidirectional change of morphologization... from... PRs to MPRs... and... from MPRs to [A]MRs...". Here, we may note that the very description of the phenomenon as 'morphologization' could again be taken to suggest that the major watershed in sound-structure is found between PRs and everything else (although another interpretation is that both PRs and MPRs act as if they held in common an ultimate goal of becoming (A)MRs). Dressler's latest paper adduces only one piece of historical evidence against combining MPRs and AMRs within the same (morphological) module, and this is that, within morphology, "allomorphic MRs may become nonallomorphic and vice versa", while (A)MRs never become MPRs (again). Nonetheless, the latter assertion seems to be in need of continued empirical testing. This is because the main criterion for AMR status is apparently that a rule must apply in at least some cases where it is not triggered by an 10

Another way of phrasing this point is to say that it is not clear what else in grammar follows from a decision to assign morphonology to any particular grammatical module(s) or non-module.

MODULARITY, MORPHONOLOGY, AND GRADIENCE

95

accompanying affix (cf. Dressler 1985a:61 on Modern Standard High German umlaut as an AMR in, e.g., Bruder/Brüder "brother"/"brothers"). Yet it does not appear impossible that such unaccompanied alternations could be leveled out while cosignaling ones would remain — in which case the rule at issue would have (re)gained MPR status. Indeed, something very like this phenomenon may have occurred in Standard Dutch. Here, the only remnant of an umlaut MPR occurs in the pair stadlsted-en "city"/"cities", but the earlier situation is reflected in non-standard regional dialects that also have words where umlaut may be unaccompanied by affixation — and so seems to be an AMR. Furthermore, even if there were no cases of the change from AMR > MPR to match historical conversions of MPR > AMR, it cannot be the true that bidirectionality of reanalysis between two adjacent rule-types is a necessary concomitant of their belonging to the same module in a Natural Phonology/ Natural Morphology framework. After all, Dressler's §5 discussion of diachrony explicitly states that "allophonic PRs may change unilaterally into phonemic neutralizing PRs, but not vice versa". Yet the larger problem here concerns unidirectionality in general, since at least two claims regarding 'impossible' paths of rule-type mutation which are made by "A Functionalist Semiotic Model of Morphonology" are contradicted by existing studies. First, the assertion that there is "no... 'phonologization' of... MPRs" is contravened by the fact that exactly such an alteration affected a rule tensing final open  in the earlier history of Modern French (cf. Morin et al. 1990 and Morin 1994). Second (though less importantly for morphonology), the claim that the diachronic relationship between syntax and morphology also involves "uni­ directional change towards morphology" is falsified by a rather large collection of examples which show affixes becoming clitics (as with English - 's/-s'; cf. Janda 1981) or even non-clitic words (as with Enontekiö Northern Saame taga; cf. Nevis 1986); their implications are discussed by Joseph & Janda (1988). What such instances show is that there exists no diachronic conveyor-belt ineluctably transporting phonology and syntax off into morphology. Given the right (admittedly rare) circumstances, younger speakers may jump to the conclusion that some phenomenon is a syntactic or phonological one even though it receives a morphological analysis in the grammars of other, older speakers. This way of describing matters reminds us that the transmission of language is discontinuous (as emphasized by both Halle 1962 and Andersen 1973), and so extreme caution must be exercised in comparing the varying status of 'the same rule' across different grammars. Consequently, the

96

TRUBETZKOY'S ORPHAN

historical evidence discussed by Dressier in the 1985 monograph and his most recent paper must be evaluated from such a perspective. My own conclusion in this regard is that diachrony turns out to be a much less revealing indicator of modular relations than it at first appears to be. Similarly, since all linguistic data necessarily owe their origin either to reanalysis or to inheritance from humankind's original language(s), history also provides an alternative to Dressler's assumption of a "naturalness theory [which] claims that... relative degree of naturalness results in relative frequency". I have argued elsewhere (cf. Janda 1984) that the relative rarity of morphological metathesis-rules is due, not to some to inherent markedness of such a grammatical process, but rather to the scarcity of the two situations which give rise to it via reanalysis: (i) phonological metathesis and (ii) morphological stress-alternations associated with different loci of deletion in strings where successive syllables have identical vowels (thus, /...éCe.../vs. /...eCé/> /...éC.../ vs. /...Cé/). Still, mentioning the issue of frequency and numbers should remind us that there cannot be many linguists who have as thoughtfully and insightfully considered as many facets of as many languages as has Dressier — certainly not in the study of how phonology, morphology, and morphonology interact, or of what their optimal arrangement in grammar is. In examples ranging from Albanian to Chinese, he has forced his colleagues to devote greater attention to both functional and formal considerations, and, even if I have here found myself disagreeing with his most recent paper on several issues, the agenda has largely been set by topics whose discussion Dressier initiated in his 1977 and 1985 books. Admittedly, there are no Mandarin data in his current paper (though cf. Dressler 1985a: 109-110). Yet, by grappling with a host of functional principles within a polycentristic theory (cf. Dressler 1978), all his work moves us toward fulfilling Mao Tse-Tung's 1957 exhortation concerning the best way to "promote the flourishing of the arts and the progress of science": "let... a hundred flowers blossom and a hundred schools of thought contend". One might wonder if there could ever be a hundred aspects of thought on morphonology, but why not? We already know of at least one Round Table's worth!

On A Functionalist Semiotic Model of Mor phonology: Comments on Dressier Douglas C. Walker* University of Calgary For well over a decade, Wolfgang Dressier and his colleagues have been developing what they have come to call a 'natural' approach to phonology, morphology and related questions. Among the most salient properties of this approach, we may list at least the following: (i) the admixture of both formal and functional propositions in the de­ scription and explanation of linguistic phenomena, with the latter propositions subsuming the former; (ii) recourse to a (Peircean) theory of semiotics as the appropriate meta­ language in which to formulate explanations of such phenomena; (iii) special interest in external evidence (from e.g., language acquisition, va­ riation, pathology, evolution); (iv) incorporation of results from typological studies as a way of accounting for structured variation within and between languages; (v) proposal of specific semiotically-oriented criteria which illuminate the interrelationships among linguistic phenomena: iconicity, diagrammaticity, indexicality, the semiotic priority of morphology over phonology, measures for the reliability of signs, etc.; (vi) development of scales, based on the criteria in (v) (and many others), which allow for rules to be classified as reflecting, to a greater or lesser degree, a prototypical rule type: morphological (MRs), allomorphic mor­ phological (AMRs), morphonological (MPRs), phonological (PRs), and so on; (vii) insistence on the multicausality and multifunctionality of linguistic pro­ cesses, and on the conflict between competing tendencies, as a way of accounting for observed diversity. Much of this material is set out in detail in Dressier's chef ď oeuvre, * Work on this paper has been funded by a grant from the Social Sciences and Humanities Research Council of Canada, whose support is gratefully acknowledged.

98

TRUBETZKOY'S ORPHAN

Morphonology: The dynamics of derivation (Dressier 1985a), and his paper in this volume is, in a sense, a compact summary of and good introduction to that work, as well as a reference to many supporting theoretical and empirical studies. Natural Morphology has also attracted significant international atten­ tion (via, for example, inclusion in standard handbooks of morphology such as Bauer 1988 or Carstairs-McCarthy 1992, or via critical review in the scholarly literature, e.g., Carstairs 1986, Singh 1986). It is thus unfortunate that, for reasons of time, Dressier was here able to offer us neither his detailed response to the reactions of others nor his observations regarding, competing or complementary approaches.1 Nonetheless, there is sufficient material in the present paper to convey the substance of the Natural Phonology and Morpho­ logy (henceforth NPM) approach to morphophonology, and to allow a number of questions to be raised (albeit by one who is strongly sympathetic to the approach in question). Paradoxically, one can almost conclude from Dressler's work that morphophonology does not exist, in the sense that, unlike PRs and MRs, there is no clearly delimited component or module containing all and only morphophonological rules. Nor can we define a set of properties or functions neces­ sary and sufficient to identify the latter (although their cosignalling of MRs comes close). Instead, morphonology functions as a zone of mediation be­ tween the competing requirements of morphology and phonology. Dressler's approach is therefore clearly a gradient one (although no doubt in the middle of the continuum from phonological to morphological rules, we will encounter, in many languages, language-specific exempla of the typical MPR). In such a context, the interaction of the attributes listed in (v-vii) is crucial, and much of Dressler's work is devoted to the elaboration of scales illustrating the operation of these descriptive criteria. The existence of such gradualness, and particularly the role played by multicausality/multifunctionality and by goal conflicts, lead one to ask what state of affairs could not be described by Dressler's theory, what situation would be incompatible with its conclusions. Dressler's answer, reasonably enough, is that we require large scale, broadly-based empirical studies, inasmuch as most of the predictions made by NPM are statistical in nature. Variation (or stability) in space and time, or longitudinal studies of acquisition, 1

Dressier does conclude with a number of brief comments addressing Singh's detailed review of his 1985 monograph. Of particular interest are Dressler's remarks concerning the directionality of morphophonological operations and the (non-)existence of a morphophonological module. Probing as these reactions may be, they are too brief to be compelling (as Dressier himself indicates), and these key questions regarding the essence of morphophonology remain (in my opinion) open.

MODULARITY, MORPHONOLOGY, AND GRADIENCE

99

or cross-linguistic typological comparisons, to name obvious possibilities, will reflect the patterns implicitly or explicitly inherent in the theoretical claims. Although Dressier refers to a number of such studies, including a major on­ going project in Vienna on language acquisition, I believe it is fair to say that the empirical evidence in support of (or not in conflict with) Natural Morpho­ logy should now be brought more to the forefront as a means of theoretical proselytization, if nothing else. For Dressier, "the units of morphonology are morphonological rules (MPRs)," written using "the format of classical generative phonology, with the exception that I will usually write phonemes instead of distinctive features or their equivalents, because — as understood already by Kuryłowicz — they often apply differentially to single phonemes of natural classes of phonemes". (1996:fn.4). To say the least, this set of proposals requires extensive discussion, as Kiparsky's paper in this collection makes clear with respect to the rule versus alternation (i.e., replacement versus selection) question. But this Natural Morphonology approach also sidesteps a vast number of questions that have dominated much of the phonological literature of the past decade: autosegmental, metrical and prosodic phonology (and morphology); feature geometry; underspecification theory, to say nothing of the now ubiquitous Optimality Theory. Given the profound implications of this range of work for our understanding of morphophonology, it would seem appropriate for NPM to address it in detail. It will no longer do, for example, to circumscribe mor­ phophonology to "segmental morphonology, the usual understanding of morphonology" (Dressier 1985a:38) and to give non-linear approaches the short shrift they appear to have received up to now (e.g., the scant Prosodic Phonology and Morphology chapter of Dressier 1985a occupies pp. 33-40). One theoretical cornerstone of NPM involves the semiotic priority of morphology over phonology. From this, Dressier derives a number of important organizing principles, including that involving synchronic rule order ("first MRs, then MPRs, finally PRs", p. 10) and a related one constraining the directionality of historical change ("there is a unidirectional change of morphologization a) from (phonemic neutralizing) PRs to MPRs... and b) from MPRs to MRs" p. 12). If these principles are absolute, the first appears contradicted by the use of derived phonological (or even allophonic) infor­ mation in conditioning allomorphy, as evidenced by Anderson (1975), for example. There also exist proposals, such as that of Morin, Langlois & Varin (1990), purporting to show the incorrectness of the second. Dressler's framework might even appear to make provision for such cases to arise, since he cites Gósy (1989) as showing that young children can reanalyze very

100

TRUBETZKOY'S ORPHAN

general MPRs as PRs, subsequently having to reanalyze them as MPRs (p. 13). It appears, then, that both synchronic derivations and diachronic change may impose a greaterflexibilityon the organization of grammars than Dressier is willing to recognize, although the rarity (not to say abnormality) of such phenomena permits the natural synchronic ordering MR » MPR » PR (or its diachronic reverse, PR » MPR » MR, both of which emphasize the marginal position of MPRs) to be preserved. Let me base a final comment on Dressler's article and on NPM in general on a particularly appropriate (albeit lengthy) quotation originating in the domain of Optimality Theory (Prince & Smolensky 1993:198): Recourse to functional explanations, couched in gradient terms, is often accompanied by severe loss of precision, so that one cannot tell how the purported explanation is supposed to play out over specific cases. A kind of informal terminological distinction is sometimes observed in the literature: a 'law' is some sort of functional principle, hard to evaluate specifically, which grammars should generally accord with, in some way or other, to some degree or other; a 'rule' is a precise formulation whose extension we understand completely. Thus, a 'law' might hold that 'syllables should have onsets,' where a 'rule' would be 'adjoin  to V'. 'Laws' typically distinguish better from worse, marked from unmarked; while 'rules' construct or deform. Linguistic theory cannot be built on 'laws' of this sort, because they are too slippery, because they contend obscurely with partly contradictory counter-'laws', because the consequences of violating them cannot be assessed with any degree of precision. With this in mind, one might feel compelled to view a grammar as a more-or-less arbitrary assortment of formal rules, where the principles that the rules subserve (the 'laws') are placed entirely outside grammar, beyond the purview of formal or theoretical analysis, inert but admired. It is not unheard of to conduct phonology in this fashion. We urge a reassessment of this essentially formalist position. If phonology is separated from the principles of well-formedness (the 'laws') that drive it, the resulting loss of constraint and theoretical depth will mark a major defeat for the enterprise. The danger, therefore, lies in the other direction: clinging to a conception of Universal Grammar as little more than a loose organizing framework for grammars. A much stronger stance, in close accord with the thrust of recent work, is available. When the scalar and the gradient are recognized and brought within the purview of theory, Universal Grammar can supply the very substance from which grammars are built: a set of highly general constraints which, through ranking, interact to produce the elaborate particularity of individual languages.

Applied to approaches such as Lexical Phonology and Morphology, these reflections encourage the hope that functional considerations will assume greater importance. In the NPM context, to which they seem remarkably congenial, they should incite Dressier and his colleagues to examine in more depth certain formal questions, particularly those involved in matters of phonological and morphological representation. This being said, NPM should

MODULARITY, MORPHONOLOGY, AND GRADIENCE

101

find solace that such an analysis arises within Optimality Theory: NPM is explicitly both formal and functional; scalar properties and ranking are theoretically central; explanatory bridges to a number of external evidential domains have been established. There is every reason to believe, therefore, that NPM stands as a rich and productive counterexample to Anderson's 'isolationist' view (Anderson 1981) of the relationship between phonological grammars and their users.

Reply to Janda and Walker Wolfgang U. Dressier Adequate answers to the interesting questions, suggestions and critiques brought forth by the two discussants of my paper in this volume (numbers will refer to sections) would require a new book. Thus here I can reply only sum­ marily, relying on either references to other publications or contenting myself with a very sketchy format. First of all I am going to deal with explicit or implicit requests for clarification by the two discussants. The first such issue raised by Janda is my stand on 'morphonology and modularity'. It is clear enough (also to the discussants) that one of the main points of my 1985 book has been that morphonology is not a module or submodule of its own. Since I take modularity in a Fodorian sense (cf. Dressler & Karpf 1995), the notion of a 'quasi-module' is a contradiction in terms, but one may describe it as a non-modularized domain (in the sense of Karmiloff-Smith 1992). More complex is the question of differentiation of PRs (phonological processes) vs. MPRs (morphonological rules) vs. AMRs (allomorphic rules). Janda draws a diagram (with which I basically agree, provided that the grad­ ually of the differentiation has to be reduced to a two-dimensional diagram). Then he proposes different variants and discusses which one would be the best fit. In my view, instead of having two lines or parallel curves, one should rather allow branching lines (how the lines should be drawn in detail, is a matter of empirical research). More important, however, are Janda's parallel suggestions a) that AMRs may have some residual phonological naturalness, and b) that PRs may have some residual morphological naturalness. Now b) is excluded by definition, and in regard to a), what may look like phonological naturalness of AMRs is phonologically totally irrelevant, comparable to dead redundancy (Dressier 1985a), as opposed to useful redundancy. Still more important is the fact that any two-dimensional diagram is necessarily reductionist, because, in addition to the two opposed, polar participants phonology and morphology, there is also the lexicon, which often puts

MODULARITY, MORPHONOLOGY, AND GRADIENCE

103

restrictions on the generality of MPRs (cf. Dressler 1985a, again). An important request for clarification raised by Walker is: "what state of affairs could not be described by Dressler's theory". I limit my short reply to five domains of evidence: I) synchrony, II) diachrony, III) first language acquisition, IV) second language acquisition, V) loans (contact). I) In synchronic accounts, my model notably excludes the coexistence of incompatible properties (including incompatible degrees or scores of naturalness on different scales), i.e., interscalar incompatibility (as ac­ knowledged by Janda). Since many of these incompatibilites are fully spelt out in the conclusion of chapter V of Dressier (1985a: 146-149), I will not repeat them here, but just cite one example: if a rule of palatalization (as discussed in chapters VI, VII) attains score 2 on the scale 'aid to pronounceability', it may not attain scores 2-4 on the scale of process matching or be restricted to certain morphological or lexical categories. II) In diachrony, my model excludes the change (A)MR > MPR > PR. Janda and Walker maintain that there are counterexamples, referring notably to Morin, Langlois & Varin (1990). Maybe they have not seen the final written version of my paper, where, in section 5 I have stated why I do not think that Morin, Langlois & Varin's (1990) case of French word-final tensing is completely convincing. And this has been the best candidate of a counterexample I have ever seen.  ) In first language acquisition, my model excludes the emergence of an MPR prior to that of all obligatory PRs (with the only exception that the degree of generality of the PR may still change). Variable or optional phonostylistic PRs, however, may still emerge later, because children start to acquire the whole gamut of adult (stylistic, pragmatic, sociolinguistic) variation later than the first MPRs (at least in morphology-rich languages such as Russian, where the acquisition of both MRs and then MPRs starts much earlier than in English or German). IV) In second language acquisition, interference of MPRs but not of PRs and of MRs triggering such MPRs is excluded (cf. Dressler 1985a:101-103; 1992). V) The same holds for loaning of processes/rules (cf. Dressler 1985a: 103105; 1992). A final request I want to address is Janda's wish to have a summary of the essentials of Dressier (1985a). For this I may refer the reader to the

104

TRUBETZKOY'S ORPHAN

conclusions (146-150, 180, 376-379).1 After these clarifications I want to resolve a misunderstanding by Janda. He incorrectly seems to attribute to me the view that "the main criterion of AMR status is apparently that a rule must apply in at least some cases where it is not triggered by an accompanying affix", as in G. umlaut Bruder/Brüder "brother/ brothers" vs. Land/Länd-er "land/land-s". However, I have never implied such a view. On the contrary, I have repeated ad nauseam that there are many different German umlaut rules, some of them AMRs, some MPRs, a syn­ chronic situation which is diachronically due to rule split (as also abundantly shown in Janda 1987). In a similar vein, Janda (note 2, p.99) appears to infer from my statements that a high degree of cosignaling function is restricted to MPRs. But also AMRs may cosignai to the same degree, such as the AMR of umlaut in Land → Land-er.2 Next I want to argue against three other critical remarks. Although I fully agree with Janda's insistence on the importance of diachronic explanations for morphonological phenomena, I disagree with a general reduction of explanation to diachronic explanation and with his specific example of metathesis. He explains metathesis with loss of adjacent unstressed vowels in stress alternations. There exist, however, other types of metathesis where this explanation is not applicable, such as in AGk. tí-kt-ō < /tí-tk-ā/ "I give birth to" vs. aorist é-tek-on. Walker refers to my citation of Gósy's (1989) material which shows that small Hungarian children may reanalyse the Hungarian MPR of vowel har­ mony as a PR, and takes this as evidence that diachronically an MPR may turn into a PR. However, not all changes that children introduce into their early grammars have a chance of success, in the sense that they may persist until they lead to language change (more in Dressier 1994). Janda argues that the fact that only AMRs and MPRs share important properties of morphological naturalness, is evidence for both types of rules belonging to morphology. But he interprets the parallel fact that only PRs and MPRs share important properties of phonological naturalness as being ev­ idence for MPRs forming a subpart of morphology. This asymmetry in argumentation remains without further explication or explanation. Interest1 Which may have easily escaped the attention of readers due to the lamentable lack of a detailed table of contents and subject index, which the publisher of Dressier (1985a) had refused to include. 2 Here I must concede an inconsistency in Dressier (1985a), noticed by Janda (note 6): not always have I insisted that the cosignaling function only typically belongs to MPRs (but is not excluded from AMRs and even from some PRs; cf. Janda's note 8).

MODULARITY, MORPHONOLOGY, ANDGRADIENCE

105

ingly, the opposite argumentation is typical within generative paradigms: phonologicity makes what we call morphonology a part of phonology, where­ as morphologicity (e.g., morphological conditions) makes it a proper subdomain of phonology (e.g., the stratum or strata of lexical or cyclical rules). Finally I have to refer to future work planned in the following two domains: Janda rejects my brief remark on unidirectional change from syntax towards morphology with the example of suffixes turning into clitics. Cases cited in the literature are, however, due to highly theory-dependent inter­ pretations. Walker and Janda suggest a critical comparison with, and discussion of Optimality Theory. There is no space here for this, but two general remarks may be in order: 1) Proponents and practitioners of Optimality Theory do not discuss naturalist preference theories either (beyond occasional mentions of rather old publications in (mostly) Natural Morphology). 2) From the point of view of Naturalness Theory, there is still much arbitrariness in the basic claims and options of Optimality Theory. There is still a rather conspicuous absence of acknowledgement and discussion of the extralinguistic bases of grammatical preferences, a lack of semiotic metatheory and of the epistemology of func­ tional explanation (cf. Dressler 1985a:261ff.; Dressler & Dziubalska-Koîaczyk 1994; Dressler 1995). At the moment it seems that Optimality Theory will just have to offer more in one of these respects. When its practitioners discuss naturalist preference theory seriously, it could provide an interesting challenge.

A Functionalist Semiotic Model of Morphonology Discussion DRESSLER: Les diverses valeurs de ces échelles de diagrammaticité découlent de la définition de la diagrammaticité et de la dimension phonologique où ça s'applique. Comme vous l'avez dit, l'effacement est très peu naturel dans cette échelle de diagrammaticité phonologique, mais 'très peu naturel' n'équi­ vaut pas à 'exclu', cela veut dire que l'effacement est très difficile selon des prédictions pour l'apprentissage, pour le traitement, pour la stabilité, avec des conséquences pour la distribution et la fréquence des distributions, soit dans type, soit dans token. Par conséquent, il est très difficile d'avoir une véritable règle phonologique obligatoire stable d'effacement, tandis que les règles phonologiques d'effacement facultatives ne sont pas exclues du tout, parce que la diagrammaticité sert à la facilité de la perception. On sait bien que dans le débit rapide et négligé, la perception acoustique n'est pas impor­ tante. On peut raccourcir comme on veut, et plus la situation de communi­ cation est intime, familière, plus importants sont les facteurs extra-linguis­ tiques de la compréhension. L'effacement phonologique n'a guère d'incon­ vénients dans ces situations, ainsi pour [ministr], [minist], [minis], les auditeurs n'ont pas de problème du tout. MORIN, J.Y.: Laurent [Santerre] pourra peut-être me corriger, mais ce n'est pas spécifique à un registre. C'est familier, c'est du français québécois, mais je pourrais très bien dire ici [minis] sans problème, c'est naturel, mais ce n'est pas particulièrement relâché DRESSLER: Ça dépend aussi de la définition de la situation. Au Québec, les situations de communication sont souvent plus relâchées qu'à Paris. So now let's come to morphonology, again. I quite agree that a module must be functionnally transparent and that therefore morphonology is not a module. What it is, is a domain, and this is not the same as a module, as, for example, Annette Karmiloff-Smith has shown quite well in her last book. Now, the question whether it is a domain or to a certain extent a submodule depends of course on the definition of a module, but if one takes the usual definition, let's say of the Fodorian type, then, according to the empirical

MODULARITY, MORPHONOLOGY, AND GRADIENCE

107

evidence at my disposal, it's only a domain, it's not a module. But this is open, of course, to empirical disconfirmation. Now, I also agree that it is not an interface, in the sense of autolexical syntax, but it is a domain where there is an interface, or to put it metaphorically, since baseball is a sport which is very popular on this continent, it is so to say a playground where two dif­ ferent teams are in competition and play, and thus phonology and morpho­ logy play so to say on the battle-ground of morphonology. There are these rules which follow properties of both morphological and phonological rules, although due to this competition, not to full extent; and, there is in the long run always a winning team, this is morphology, but, what morphonology looks like is fairly language-specific according to, as I have tried to show, certain principles of both typological adequacy and system adequacy. And I do not find any of these positive definitions of morphonology, and this is one of the reasons why I cannot subscribe to classifying morphonological rules as representing a module or a submodule of its own. MEL'ČUK: Before we start, I would like to make a very important point. Whether something is something or not, whether an x is a word always depends on the definition of a word. So before we discuss whether it is a module or not, we have to have a definition of a module. And I believe, if I did not misinterpret you, that you've given a brilliant definition of a module. A module is a being with private parts. Is it OK? Now I am completely satisfied, and know what a module is and that's OK with me. MORIN, J.Y.: Something that was not clear in your 1985 monograph for me, is whether you are making the claim that all morphophonological rules come from phonological rules or if this is the case most of the time. I agree with that, of course. DRESSLER: I am not speaking about rules corresponding to morpheme structure constraints. I also distinguish between phonological, morphono­ logical, morphological ones, and it is often very difficult to see where they come from. There, I cannot vouch for any unidirectionality. But the claim that morphonological rules of the type we are discussing here come from phonological rules or are replacements of phonological rules. SINGH: I just have a brief question. It's not clear to me what your principled explanation might be of the facts that I understand you as having supported: (i) that the change of morphonology is unidirectional; (ii) that part of it is separable, as you try to demonstrate towards the end of the paper. Do you see some semiotic or other sort of principle that might combine these two parts which otherwise might plausibly seem to be contradictory? DRESSLER: I don't see where they are contradictory. As I said there is this

108

TRUBETZKOY' S ORPHAN

semiotic principle of the priority of morphology over phonology because words are primary signs, only secondarily one analyzes words into mor­ phemes, and then one analyzes morphemes into what are their phonological parts. So on the one hand there may be a doubt in language acquisition about these critical points or stages where a reanalysis may occur. As to these critical stages then, the innovators, in doubt, may decide for morphology being still more important than previous speakers, such that starting from very small importance of morphological indexicality, at the end there is only morphological indexicality left. On the other hand, it's quite clear that a morphonological rule or an allomorphic rule of this type, as I was discussing here, can be only a co-signal of a morphological rule and not vice-versa, since again there is this precedence of morphology over phonology and therefore also over morphonological facts. So I don't see any contradiction there. SINGH: I want to understand clearly the idea behind form following function. Is it possible that in the case of co-signaling it is really perhaps the opposite, meaning the co-signaling is there because there are two formal indications? Because there are two signals. The two signals aren't there because there is co-signaling, but the co-signaling is there because there are two signals. DRESSLER: Well, this might be an induction on the part of a language learner, this is, of course, possible. But, if the language learner will then identify what is the main signal and due to this precedence of morphology, he will, as in the case of Umlaut, identify the morphological suffixing as the main signal. And the other one is the co-signal. Diachronically this may change but just in case that there are phonological rules which weaken or finally delete the main signal. Therefore we have then these developments like Austro-Bavarian sg. {ta:g}, pl. {te:gə} > {te:g}, i.e., in these dialects the final schwa was the signal and the umlaut the co-signal, but if then, a phonological rule which started in casual speech and then became more formal until it became obligatory, deleted the final signal such that at a certain point of course, there must have been a cognitive switch in the perception of what is the main signal. MEL'CUK: This also happened in Yiddish. KIPARSKY: About the role of co-signaling and the role it played in your signaling explanation, for the loss of the palatalizations in Slavic, you argued that it's better to have the co-signal come before the main signal than after, and that's the reason why you lose the progressive palatalization and keep the regressive one. Now, I wanted to know why it is more important to have them come in that order if you are going to have a main signal and then an

MODULARITY, MORPHONOLOGY, AND GRADIENCE

109

auxiliary signal; why does the order matter, is there some psycholinguistic reason? DRESSLER: This is a good question, I haven't done tests and have not seen any test. This is of course a lacuna in my semiotic reasoning. KIPARSKY: Actually, there was a second part to the question. Given what we know about co-articulation of consonants and vowels, is there such a tem­ poral difference in the first place that could play a role in perception? The vowel and the consonant when they are adjacent are practically super­ imposed. Can one really argue that one in some interesting perceptual sense comes as a prior signal that can be utilized before the other, a millisecond apart? DRESSLER: OK, the first part again, if I may say so, unfortunately, I do not know of any tests, but there is, in communication theory (and with the semiotician Morris), this idea that what is important for a co-signal is that it is preparatory. Preparatory signals are important to stimulate, so to say, or determine the attitude and the expectancy of the hearer. This is a concept which I've taken from communication theory. MEL'CUK NOW, in Russian, there is a very strong tendency to change the vowel as a function of the preceding consonant. If it is not vocalized, the vowel doesn't change. If it is palatalized and the vowel is not under stress, it changes so much that very often the foreigners, even other Slavs, don't recognize it at all. A typical Russian speaker does not see any difference, but it's really huge. KIPARSKY: That suggests a generalization which is completely different from the one you proposed, which covers both the Russian case and the Polish case: that tautosyllabic assimilations are kept. That's the consonant and vowel in Russian and the progressive palatalization in Polish. DRESSLER: There is one thing, speaking about -articulation as well, which has to do with what I consider phonology which is very phonetically based, and where this question of co-articulation, of syllable division is important whereas what I'm concerned with is morphonology and allomorphy and there the question is somehow independent of these phonetic frames. Because it doesn't have to do with phonetic planning. KLAUSENBERGER: I'd like to ask about the proposed morpho-centricity that you are suggesting and, of course, the change from phonology to morpho­ logy, I would go along with as you explain it by indexicality of morphology. I think Yves Morin is not necessarily in agreement with that. But you also say that from syntax you go into morphology, so how would you explain that in the history of languages there is always a rise of new syntax after

110

TRUBETZKOY' S ORPHAN

morphology gets fossilized? Not always, but that's a common development. So you say morphology wins out but then it seems to be a replacement also further on. DRESSLER: It's similar in phonology: new phonological processes become visible. As to syntax, the question is, what is the natural basis of syntax, and this is another question. Clearly one can make the argument also that syn­ tactic rules are, so to say, more natural than morpho-syntactic ones, in a similar way that one can say that phonological rules are more natural in several parameters than morphonological rules. What we have then in this diachronic convergence in morphology is simply conventionalization. As Peirce put it, symbols, that is conventional signs, come into being especially from icons, that means, more iconic signs become, due to convention­ alization, less iconic signs in diachrony, and this happens in morphologization both of syntax and of phonology. BYBEE: Following up on that question, I wanted to ask you about the role of frequency in naturalness. I think you mean frequency in the languages of the world and frequency within particular languages. And among the several things that define naturalness, one is frequency and the other is relative stability in diachrony. I'm interested in what the mechanism is for the creation of the frequency. Now what I argued in my paper was that the more transparent constructions are more frequent because of the way morphology develops. When grammaticization takes place, transparent constructions are created and it takes a long time for them to become highly fused and opaque in the morphotactic sense. For me, that explains why transparent con­ structions are more frequent. Actually I don't have any particular quarrel with the idea that they might be more stable in some way, but I'm just wondering what your mechanism is. Is the only mechanism for the frequency of transparent constructions their stability or is there some other relation I'm not seeing in your conception? DRESSLER: This is a very long question. First of all, higher frequency of one of two comparable constructions is a result of one being in specific pa­ rameters more natural than the other one. For, according to what we think of how languages are acquired and are used, it's easier to acquire and use more natural constructions than less natural constructions. This is a general answer. Of course there is a backfeeding process, in the sense that more frequent constructions in acquisition become more salient for being acquired. But in the general explanation, frequency is something we predict and not something which is our reason why we inductively generalize and say "oh, this must be more natural because it is more frequent." No, this is not the

MODULARITY, MORPHONOLOGY, AND GRADIENCE

111

way we work. We try to deduce what should be, in a given constellation of facts, more natural, and as compared to something else, which is com­ parable, meaning, in the same class of objects we want to compare, then predict higher frequency for the more natural solutions, and then we look whether this prediction is true, whether it is more frequent or more stable. This is the research strategy. And also the main explanation and of course the locus where more naturalness effects more higher frequency is language acquisition and language use. BYBEE: Could I follow up on that? Why would a construction become opaque if it's really better for it to be transparent? Let me give your answer then refute it also, if you don't mind. For example, why would a suffix with a palatal vowel be replaced by a stem change (that's umlaut), and you might say that it's just a phonological process which has a competing motivation. The way I review that process of replacement of the suffix with a stem change is a gradual fusion of the suffix with the stem, so that the phonological properties of the suffix end up on the stem vowel and now there is phonological motivation, but you end up with a highly fused element and a form that is not transparent anymore. It seems to me that if there is a real push for transparency, that sort of thing would be resisted and it wouldn't happen. DRESSLER: This is again a very complex question. We view language as any other human activity, as a domain of conflicts, of antagonistic processes. If one looks at them in a very general global way, then there is no predictability at all, seemingly. And it seems that you can never disconfirm anything that is claimed, because immediately you point to an antagonistic change, etc. So what is the way to get around this? The way is to split the problem into different constellations, like in a natural experiment, i.e., this is an experiment that nature makes for us, and this is the reason why I think external evidence, substantial evidence, is essential, for example diachronic change. Again, diachronic change cannot be dealt with as a global whole, because again there are these antagonistic forces. Therefore what is necessary, and this is something I'm currently engaged in, is to construct various explicit scenarios of change. Thus for each scenario one has to identify and weigh the factors that play a role and as in what I tried to do it in a nutshell, the scenario of morphologization. Morphologization occurs with children when they have detected the great importance of morphology in comparison with phonology. So something which is phonologically more diagrammatic, as I've tried to show is at the same time morphologically less diagrammatic, and therefore, you have to have these parameters interact in a

112

TRUBETZKOY' S ORPHAN

specific way and the important thing for explanation is then what I call various scenarios of diachronic change where I try to establish under what conditions, in which domains, in which locus, in which linkage, (cf. Christopher Hall), a change occurs. BYBEE: I wanted to make a comment on co-signaling, and then ask a separate question. In co-signaling you can see a nice explanation for the fact that the minor co-signal precedes the suffix in suffixing languages, but in prefixing languages, because of the tautosyllabic nature of assimilation for example, your explanation doesn't seem quite as easily available. It might be an interesting research strategy to look at what happens in co-signaling in prefixing languages, because I expect there may be some differences there. DRESSLER: If one studies Christopher Hall's book Morphology and Mind, where he tried to account in a better way than others did previously for the suffixing preference, then it becomes quite clear why it is the case that the combination of prefixes with following stems is much more transparent than between stems and suffixes, and it has to do with the answer to your question because a co-signal that comes after the prefix is of little use, compared to a co-signal which precedes the suffix, therefore you get more fusion then finally as a result in diachrony with the sequence stem suffix, than with the sequence prefix stem. WALKER: I'm not sure that Yves-Charles is going to cite himself, so I'll do it, with respect to the unidirectionality of change from phonological to morphonological to morphological rules. There is at least one good example in the literature of a morphonological rule becoming a phonological rule and that's the one Yves has discussed. An implication of this unidirectionality seems to me, again, that you never get morphonological rules coming into the language without first having evolved from phonological rule, and there are other examples of that that also take place, and you provide in a sense in your paper a mechanism whereby that might arise, where you say that children have been observed to reanalyse very general morphonological rules as phonological rules, and that in fact may be a mechanism for a bidirectional possibility for change. Morphonological rules under those circumstances become phonological rules. So I wondered if you could comment on why it is that that mechanism isn't exploited somewhat more than it appears to be. DRESSLER: First of all, here I must confess to a lapse of memory. I read of course your paper, and at that time I have concluded that this is not a counter-example, but unfortunately I've forgotten all the details (but cf. the written version of my paper). I can be much more positive about the Spanish examples that have been brought. Certain people have looked at certain

MODULARITY, MORPHONOLOGY, AND GRADIENCE

113

Spanish monophthongizations that Malkiel had studied as a case. Moreover there has never been a synchronic phonological rule that has originated from a previous morphonological rule, but only a replacement of input. A com­ pletely different question is that young children have been observed both in Hungarian and Turkish to reanalyze the morphonological rules of vowel harmony as truly phonological ones and to apply them as exceptionless rules, at a very early age. Now this very early age, of course, has extremely little chance to be the locus of language change. This is similar to Bybee's conclusions in one of her recent papers, To put it in a very blunt way, if certain infants produce only papa and mama, there is little chance that this will result in a language which consists only in the two words papa and mama. JANDA: Two more questions about diachrony. I wonder if you can confirm that you still have the view that you had in your book, which I think deserves more attention, about quasi-phonemes and marginal phonemes. I think, based on the natural phonological focus that you have, that if the phonetic distance is too great between two alternants, you would not see that as a phonological rule, it would be pushed to be morphophonological, as with [s] for velar softening. Because of [1] and [r] and [1] and [n] in various languages. I wouldn't say that I would rely on my judgements of what's phonetically distant, but for [k] and [s], I would really have trouble thinking that speakers would not notice the huge difference or would't consider them as co-allophones. They would be viewed as morphophonological, I think. DRESSLER: Yes, this is morphonological. JANDA: So, you are still viewing then morphologization of phonological rules as something that can take place while the phones are still in complementary distribution. This was the idea of the quasi-phoneme. But then Twaddell had it backwards. When you lose the conditioning environment, the implication is, then it becomes phonologized. Now that's the term by which it must have happened, the terminus. So that phonologization can take place before you have a reduction in the environment, a loss in the environment. So I wanted to confirm that, that basically velar softening couldn't be a phonological rule. DRESSLER: It cannot be. JANDA: This leads to the Twente case. Sometimes with language change arguments, the problem is that that was then and this is now. Do we, in other words, always know what the previous data was, based on what we have later? Notice that the argument was that the earlier stage in the non-standard dialect had umlaut triggered by the suffix, but obviously they don't have that analysis later because they have the suffix and not the umlaut. Could they not

114

TRUBEIZKOY' S ORPHAN

have said, for example, "well what I notice about Standard Dutch is that they don't have umlaut and I'm going to focus on an umlauted root or stem, and I'm going to take that Dutch diminutive, that's not this nice little thing that we have to stick on the ends of words, so I'll use my diminutive and their stem." It's just a different analysis. It doesn't really imply that at the earlier stage, the umlaut and the suffix were not co-signals, rather like in the model of Singh and Ford. It seems to me difficult, because the thing has changed, to claim to know how it was beforehand, simply because now it's different. DRESSLER: About the question of phonemes, one of my main examples has been the German ich- and h-laut, where I claim that the ach-laut is synchronically derived from the ich-laut but that the ach-laut sticks out already that much, is that much salient that it is so to say, a quasi-phoneme. And this view has been deepened by an interesting analysis of a Rhenanian regiolect by Elke Ronneberger-Sibold, who has shown that, due to a change of /r/ to a back vowel, where the ich-laut stayed there is new phonemic contrast. As to the Twente dialect situation, I follow the Dutch dialectologist Lor Van Bree. First of all, what seems to be garanteed is the sociolinguistic frame, which is that this is a dialect which, although on the German border, far away from the centre of the Netherlands, is under heavy Standard Dutch influence. In tests, Van Bree gave native speakers all sort of conceivable combinations, there he got minimal percentages of wrong transitional forms (i.e., with an important statistical difference) such that they must have been due to some artefact of the test. Now the question is, what is here the explanation? Of course, you can make a different story and say, well in accommodating to the standard, young speakers make certain things because they develop a strategy of accommodating in a certain way rather than in another one. But as long as you say it in this way, then it's of course only stipulated. You must explain why they are doing that in a certain way, and it seems that there is no way that somehow they try to accomodate more dialect vowels than dialect consonants for example. JANDA: Can I just ask you to clarify? Is your argument that because the umlaut disappears in the accommodated form, it must have been dissociable in the dialect once? DRESSLER: Right. The question is what does my model predict if I compare it, under the same pre-conditions and the same framework, with a Singh-Ford model, let's say. My model would predict that it's not possible that they change the suffix but leave the umlaut behind, whereas this is predicted as possible within the Singh and Ford model. Therefore the Twente case is compatible with my model. This type of approach would be a perfectly

MODULARITY, MORPHONOLOGY, AND GRADIENCE

115

legitimate type of formal explanation, but this explanation is couched in a functional explanation of the type that young speakers are accommodating in the sociolinguistic situation to the standard and they're doing it in taking over elements of the standard. Now there comes the formal explanation again, a relatively autonomous element is the choice of the suffix whereas the choice of the umlaut is a secondary element triggered by the suffix. SINGH: Just a very brief, comment. Suspending the possibility of motivating the sort of reanalysis that I would predict, notice that if that sort of analysis, the one that Professor Janda just mentioned, could be motivated, then the alleged non-generalizeability of morphophonology which both you and Ï subscribe to seems to become a necessary consequence of separability. One may not have to make a separate statement to that effect, if it is sustainable. DRESSLER: Yeah but separability in a certain sense. SINGH: In the specific sense in which I used it in, let's say, Singh (1996). DRESSLER: But the two separate objects are not autonomous; rather one object is triggered by the other one. SINGH: That's right, the alleged phonological part of the morphophonological, so-called morphophonological rule cannot be taken out of it and put somewhere either in a module or a submodule or whatever. DRESSLER: Let me again precise here that according to my analysis, the case at hand is an allomorphic umlaut, it's not a morphonological umlaut. The subdivision in other theories is different of course, and I took this, first because it was a nice example I found, but, second, also because an allomorphic co-signal should be still less separable from a morphological suffix in a rule than the morphonological one. Therefore, a fortiori, the reasoning is it should then be impossible for a morphonological umlaut as well.

III Ford & Singh, Mohanan and Janda LINGUISTICS WITHOUT MORPHOPHONOLOGY

Rien de tout cela ne justifie réellement l'existence d'une morphonologie qui aurait un statut analogue à ceux de la phonologie et de la morphologie. La morphonologie serait, tout au plus, une annexe de cette dernière et l'on aurait évité bien des confusions si l'on s'était abstenu de suggérer des rapports intimes entre cette discipline et la phonologie par le terme même qui la désignait. A. Martinet 1965:25

Quelques avantages d'une linguistique débarrassée de la morpho(pho)nologie Alan Ford & Rajendra Singh Université de Montréal La réintégration des procédés dits morpho(pho)nologiques au sein des opérations morphologiques permet de rendre compte, de façon naturelle, de la concurrence qui semble toujours caractériser les rapports entre stratégies de formation de mot et opérations dites morphonologiques, en particulier là où ces dernières visent à rendre compte de la totalité de la différence formelle entre deux mots. Par exemple, si au lieu de postuler, comme le fait Bybee (1980), qu'un procédé morphophonémique intervient pour assurer l'alternance vocalique qu'on observe dans la première syllabe des deux mots espagnols cierro "je ferme" et cerramos "nous fermons", on intègre cette alternance à une opération morphologique globale de la forme de (1) : (1)

[XjéRo]v1

→

[XeRámos]v1p

qui est directement concurrencielle à l'opération (2): (2)

[Xo]v1

→

[Xámos]v1p

on réussit à rendre compte tout naturellement de l'existence de deux dialectes ou idiolectes où l'on produit soit cerramos soit cierramos comme forme de la première personne du pluriel, soit cierro soit cerro comme forme de la première personne du singulier au présent de l'indicatif. Cette analyse prédit correctement l'existence de trois dialectes qui utilisent respectivement cérro et cerràmos, cierro et cierramos et cierro et cerramos, et explique la nonoccurrence d'un dialecte avec cerro et derramos. Notre morphologie est une composante autonome d'une grammaire générative, clairement identifiable par la forme distinctive et l'ordre de ses opérations d'une part, et par la directionnalité de ses rapports avec les autres composantes de l'autre1. Nous répondons ainsi concrètement à une objection théorique de Sadock pour qui une telle

120

TRUBEIZKOY'S ORPHAN

La composante morphologique prend la forme d'un ensemble de stratégies de formation de mot dont la sortie alimente la composante syntaxique et, à travers elle, la phonologie de la grammaire. Les corpuscules2 dont elle assure la génération se branchent sur de l'information qui, dans l'esprit de la physique quantique du regretté David Bohm (Bohm & Hiley 1993), prend une forme ondulatoire, celle des scénarios et des métaphores de l'ethnie qui se sert de la langue. Le mot que définit la morphologie devient ainsi un quantum d'information, une pièce dans le jeu du locuteur. Observés du point de vue d'une taxonomie aristotélienne, les procédés morphologiques ont pu paraître comme ceux de (3) qui ont opéré une évolution à travers deux visions différentes du langage, celle d'Aristote lui-même, à qui nous pouvons attribuer rétrospectivement le schéma (3) :

et celle d'un modernisme enraciné dans la phonétique, si chère aux structuralistes, qui réduit le schéma (3) en (4) :

grammaire n'aurait pas sa raison d'être (cf. aussi Mohanan 1995:44-45). Nous pensons que la question de "which component feeds which component" (Sadock 1991:10) n'est pas aussi périphérique que Sadock le prétend. 2 Ces corpuscules sont ici bien sûr des mots. Si nous nous passons de ce terme, c'est parce qu'il connaît plusieurs emplois en linguistique dont tous ne sont pas évoqués ici. En particulier, nous voudrions éviter de faire référence à un fait qu'on trouve dans beaucoup de langues et qui constitue une opération syntaxique de formation de mots prenant souvent la forme de l'ajout d'une particule enclitique à un mot, et qu'il est important de ne pas confondre avec une opération morphologique qui est clairement pré-syntaxique, comme l'intériorité des affixes par rapport aux enclitiques en témoigne.

LINGUISTICS WITHOUT MORPHOPHONOLOGY

121

(4)

Cette réduction de la gamme des procédés morphologiques s'est effectuée essentiellement aux frais de la phonologie qui s'est vue attribuée l'ensemble des catégories non-affixales et mixtes, bien qu'une autre réduction soit effec­ tuée en assimilant la catégorie des similaires à celle des différents moyennant l'introduction d'un élément différenciateur factice appelé 'zéro'. Mais l'un des coûts les plus onéreux pour notre science prend la forme de la naissance de la morphonologie qui constitue un effort supplémentaire de ranger, entre autres, le débris que laisse traîner l'assimilation partielle de la partie dite phonologique de la catégorie mixte à la phonologie. Nos propres recherches nous ont amenés plus loin sur le chemin de la réduction de la gamme des procédés morphologiques, mais dans notre cas cette réduction s'est effectuée en évitant soigneusement qu'elle alourdisse la phono­ logie. Selon notre aphorisme de 1991 les procédés morphologiques se limitent à un seul, que nous formulons comme (5a) : (5) a. La théorie morphologique exposée ici repose sur l'hypothèse suivante : le rapport morphologique entre deux mots, morphologi­ quement apparentés et de la même langue, s'exprime par une règle générative ayant la forme générale (b) ci-après (ce que nous appe­ lons dorénavant une 'stratégie morphologique') : b.

[X]A↔[X']B

où (i) [X]A et [X']B sont des mots, (ii) A et  sont des catégories, (iii) '↔' indique un rapport d'équivalence, c'est-à-dire une im­ plication bi-directionelle, (iv) X et X' sont sémantiquement apparentés, (v) la prime ['] symbolise une différence formelle entre les deux éléments du rapport, appellée désormais 'l'opération' ou la 'constante' morphologique, (vi) la prime ['] peut être nulle si A≠. (Ford & Singh 1991) Les objections à cette vision du langage et, en particulier, au rôle qu'y joue la morphologie, peuvent paraître nombreuses actuellement en linguistique généra­ tive où l'on réclame, viva voce, l'identification des opérations morphologiques avec celles de la syntaxe d'une part, ou avec celles de la phonologie ou d'une

122

TRUBETZKOY' S ORPHAN

prétendue morphonologie, de l'autre. Voilà pourquoi, dans cet article, nous répondons à quelques objections à l'existence même, partielle ou totale, de la morphologie ou du moins, d'une morphologie telle que nous la prévoyons. Sadock (1980 et 1991, entre autres) revendique une place dans la grammaire pour une syntaxe prélexicale, une place où loger un procédé qu'il identifie sous la rubrique de 'noun incorporation' (incorporation nominale) qui serait indispensable pour rendre compte de la structure et du comportement de certains verbes de l'inuttitut, du tiwa du sud et sans doute, dans l'esprit de sa vision, de phénomènes analogues qui caractériseraient beaucoup d'autres langues. Même si cette classe de verbes résulte, de son propre aveu, de l'ajout de suffixes, employés pour les 'dériver' de substantifs, Sadock ne veut pas appeler ces opérations 'morphologiques'. La raison en est que les verbes qui en résultent, dont quelques exemples sont listés sous (6), accompagnés des noms appartenant à des structures3 d'où, selon Sadock, ils dérivent, ont plusieurs propriétés que l'on arrive à expliquer 'naturellement' seulement en faisant appel à une opération syntaxique d'incorporation nominale. (6) 18)4 19) 20)

Nom qimmiq "chien" ABS. sapangaq "perle" ABS. nirriviq "table" ABS.

Verbe qimmiqarpuq "Il a un chien" IND.3sg. sapangarsivuq "Elle achète des perles" IND.3sg. nirriviliurpuq "Il a mis la table" IND.3sg. (1980:306)

Ces propriétés, qui identifient les mots comme qimmiqarpuq, sapangarsivuq et nirriviliurpuq comme structures syntaxiques, sont, toujours selon Sadock, les suivantes : A. La modification; B. La possession;  La référence. A. La modification : un modificateur de verbe de cette classe prend une forme nominale au même cas et nombre que prendrait le complément du verbe dans une paraphrase de cette même structure, munie d'un verbe et d'un complément explicites et distincts. Par exemple, 26) serait une telle paraphrase de 27) : 3 Pour Sadock, le radical (stem) qui, selon sa grammaire, sous-tend un mot comme qimmiqarpuq proviendrait d'une structure arborescente de la forme : V[V[-arpuq] N[qimmiq]] où [-arpuq] serait un verbe 'abstrait'. 4 Pour tous les exemples du gronlandais cités ici, nous conservons la numérotation des exemples des articles de Sadock d'où ils sont tirés. Dans le texte de ces exemples, nous remplaçons l'orthographe employée par Sadock par une transcription phonologique, dont nous avons besoin pour formuler nos stratégies morphologiques. Ainsi les voyelles e et  des exemples de Sadock figurent respectivement comme i et  dans les exemples cités.

LINGUISTICS WITHOUT MORPHOPHONOLOGY

123

26) sapanngamik kusanatumik pisivuq perle-lNST. be lle-lNST. Il acquiert quelque chose-INST.IND.3sg. " s'est acheté une belle perle." 27) kusanatumik sapangarsivuq belle-lNST. Il acquiert une perle-INST.IND.3sg. "Il s'est acheté une belle perle." où sapanngamik, complément du verbe pisivuq, est singulier et au cas instrumental. Kusanatumik, modificateur de sapanngamik, s'accorde avec ce dernier en nombre et en cas. En 27) le complément du verbe kusanatumik est également singulier et instrumental. Selon Sadock, ce ne peut être que pour la même raison qu'en 26), à savoir, parce qu'il s'est accordé avec le complément d'objet du verbe -vuq, qui s'y est incorporé, c'est-à-dire que même si, lors de l'opération d'incorporation, sapanngamik se serait transformé en sapangaq qui serait, selon Sadock, la forme suffixale de sapanngamik, il conserve quand même certaines de ses propriétés, en particulier son nombre et son cas. Pour rendre compte de la différence formelle entre les deux mots, Sadock opte pour une explication qui fait appel au niveau d'application de l'opération : si les marques formelles de nombre et de cas ne sont pas présentes, c'est parce que le mot qui a pour rôle de porter ces marques ne se réalise que lorsqu' arrive le moment approprié dans la génération. Il faut donc à la grammaire de Sadock deux niveaux d'opérations syntaxiques, l'un qui comprend l'incorporation et qui est prélexical (donc, selon son emploi de ce terme, 'pré-morphologique'), et l'autre post-lexical et capable d'hériter des marques de cas et de nombre, qui ne sont pas, dans l'esprit de Sadock, engendrées morphologiquement mais plutôt syntaxiquement. Curieusement, ni le redoublement de la syntaxe, ni la fragmentation de la morphologie en morphologies pré- et post-lexicale ne semblent lui créer le moindre ennui malgré son souci certain d' 'économie'. En résumé, sa génération de kusanatumik sapangarsivuq "Il/elle s'est acheté(e) une belle perle" est la suivante :

124

TRUBETZKOY'S ORPHAN

(7) Dérivation de kusanatumik sapangarsivuq "Il/elle s'est acheté(e) une belle perle." d'après Sadock.

LINGUISTICS WITHOUT MORPHOPHONOLOGY

125

Lexicalisation (morphologie 2)

De conclure Sadock : If 27) were derived either from a structure very much like 26) or from a somewhat more abstract structure into which the empty stem had not yet been inserted, then the case for the modifier would automatically be assigned by independently-needed rules. However, the generalisation that is so obvious between 26) and 27) would be obscured if object incorporating verbs had to appear fully formed in deep structure. (Sadock 1980:307)

À son avis, dans ce cas, des redondances seraient créées au niveau de la grammaire : d'une part une règle syntaxique 'ad hoc' pour assurer l'attribution du cas, de l'autre une règle sémantique, également 'ad hoc', pour interpréter le complément explicite comme modificateur du complément incorporé. Selon la solution que nous proposons, ni l'une ni l'autre de ces redondan­ ces ne s'avère nécessaire non plus, mais nous nous passons de deux compo­ santes syntaxiques en nous contentant d'une morphologique unique et unifiée. En inuttitut le lien morphologique entre les deux mots sapangaq et sapangarsivuq s'exprime à travers la stratégie (8) : 8) [X]abs.s ↔ [Xsivuq]intl.3s La différence de forme entre sapangaq+sivuq, qui est la forme de sortie morphologique, et sapangarsivuq, sa forme phonétique, résulte d'une condi­ tion de bonne formation phonologique *qs propre à l'inuttitut. Une telle situation se rétablit phonétiquement en réalisant /q/ comme [r]. Au niveau de la syntaxe, pisivuq et sapangarsivuq sont deux verbes qui prennent un

126

TRUBETZKOY'S ORPHAN

complément facultatif à l'instrumental. Sur la grammaire de ces formes, il nous semble qu'il n'y a rien à ajouter, si ce n'est que la morphologie du cas qui permet d'identifier sapangamiq comme la forme instrumentale au singulier correspondant à la forme absolutive sapangaq du mot qui désigne une perle décorative. Elle est donnée, entre autres, par la stratégie (9), dont nous vous épargnons la lecture ici : (9) [Xq]abs.s.↔ [Xmiq]1inst.s. B. La possession : un verbe de cette classe peut prendre un complément nominal explicite au cas relatif, qui s'interprète comme le 'possesseur' du complément sous-entendu du verbe, comme dans l'exemple (33) où tuttup, au cas relatif (cas du possesseur en inuttitut), est le possesseur d'une niqaanik "viande" sous-entendue : 32) tuttup caribou-REL. 33) tuttup caribou-REL.

niqaanik nirivunga viande-INST.3sg. manger-IND.lsg. niqiturpunga en manger de la viande-IND. lsg.

Selon Sadock le verbe niqiturpunga de la phrase 33) serait du même ordre que sapangarsivuq de 27). C'est à dire que son radical {stem) résulterait d'une opération d'incorporation du nom niqaanik, explicite en 32) où il se trouve à la forme du singulier du cas instrumental en tant que possession de tuttup, son possesseur au cas relatif. En 33) tuttup figure également au cas relatif en tant que possesseur de la forme incorporée de niqaanik qui serait le niqi- du verbe niqiturpunga en 33). Selon Sadock, la position de tuttup, qui précède obli­ gatoirement le verbe en 33), serait aussi significative car "possessor nouns are in the relative case and stand before possessed nouns" tandis que, comme il le signale à plusieurs reprises, l'ordre des constituants syntaxiques du groënlandais est généralement libre. Il en déduit que "[t]he fact that incorporated nouns can be possessed is exceedingly important evidence for a syntactic incorpo­ rating rule, since sentences like 33) could not otherwise exist." Il conclut : Ex. 32) shows an independent verb and an object NP. The object consists of a possessor in the relative case, followed by the possessed — whose inflection indicates the person and number of the possessor, as well as the case of the entire NP. In 33), however, we find a denominal verb; but there is still a possessor, and the incorporated noun is understood as possessed. Note in particular that the case of the possessor is relative, just as it would be in an overt possessed-possessor construction. Obviously, if (33) is derived from a structure very much like (32), the case of the possessor, as well as the semantics of the sentence, is accounted for directly. (1980:309)

LINGUISTICS WITHOUT MORPHOPHONOLOGY

127

Selon notre analyse qui ramène à la morphologie ce que Sadock récupère pour la syntaxe, un verbe comme niqiturpug "en manger de la viande" se lie mor­ phologiquement au nom niqi "viande" par la stratégie (10) : (10)

[X]abs.s ↔ [Xturpuq]ind.+rel.3s

où l'ajout de la propriété de prendre un complément facultatif au cas relatif est partie intégrante de l'opération de formation de mot. Bien sûr, notre analyse n'apporte aucune réponse, comme Sadock peut prétendre le faire, à la question de la raison de cet ajout, mais nous représentons le phénomène tel qu'il se déroule et ce qui s'avère l'essentiel du contexte linguistique où il se déroule. La réponse à cette question, de notre point de vue, se situe ailleurs, plus pré­ cisément au niveau du vécu du locuteur et de son interlocuteur : si tuttu pniqiturpunga, alors il y a dans mon univers un lien entre ce queJemange et le caribou. Ce serait étonnant que ce ne soit pas le lien qu'on qualifie le plus souvent, du moins que Sadock qualifie de lien de 'possession'. Un peu de la même façon que de l'expression je mange on déduit quo je mange de la nour­ riture. Il nous semble que l'Inuktitut, quand il entend tuttup niqiturpunga, déduit que la viande que je mange est de la chair de caribou plutôt que d'interpréter que cet animal entretient une autre sorte de relation avec l'événement. Il doit du moins se demander ce que fait le caribou dans ce contexte si ce n'est pas là son rôle. Si sa maison est en aval de la mienne, il y a certainement un cours d'eau quelque part dans le décor, même si ce n'est pas explicite dans la phrase. C'est ainsi que, pour répondre à la question troublante que Sadock semble se poser, nous faisons appel à un principe naturel d'éco­ nomie du discours, un peu à la Grice : ce qui se présume ne s'énonce jamais explicitement (ou bien si c'est là, cela signifie tout à fait autre chose). C. La référence : En s'inspirant de la notion d'îlot anaphorique qui est une propriété que Postal (1969) attribue au mot, Sadock (1980) affirme que "We do not expect pieces of words to have independent referential or discourse properties." Pourtant il constate que "Incorporated nominals in Greenlandic have the same kind of semantic/pragmatic status as independent nominals would be expected to have." Dans Sadock (1991), sa conviction paraît moin­ dre et il affirme que "While it is frequently the case that noun incorporation is accompanied by lack of semantic/pragmatic autonomy of the incorporated nominal (particularly where there is no indication of the syntactic independence of the nominal), it is not always the case" (op.cit., p. 86). À titre d'illustration, on nous fournit l'exemple 23) :

128

TRUBETZKOY' S ORPHAN

23) kisiannimi usi nassataqarpunga katersuriarlugit ingerlaannarlunga Mais/en effet/j'en ai des bagages/les ramasser 3p/j'ai juste à aller-lND.ls. "Mais, en effet, j'en ai des bagages, je n'ai qu'à aller les ramasser." (Sadock 1991:87) où le 'nom', dit incorporé du verbe, est coréférenciel avec le complément 3p explicite dans la morphologie du verbe katersuriarlugit. Pour assurer la coréférence entre cette marque morphologique et le 'complément incorporé' du verbe nassataqarpunga il faudrait, nous dit Sadock, que le complément de ce verbe prenne la forme d'un N" explicite en structure D. Pour résumer la propriété que Sadock identifie sous la rubrique de modification : il pense qu'un verbe intransitif comme sapangarsivuq prend un complément implicite au cas instrumental parce qu'il existe un calque transitif de ce verbe qui prend un complément au cas instrumental. Pourtant, il existe d'autres verbes en inuttitut qui prennent un complément au cas instrumental, et qui n'ont pas de calque correspondant à un verbe ayant un complément ex­ plicite à l'instrumental; des exemples en sont fournis sous (11) : (11)

tingummik alallijunnijaruk du foie-INST. s'il te plaît apporte le lui "Apporte-lui du foie s'il te plait" aalisakkanik ilinnut nuwisitsiiqquwai du poisson-INST. "Il te demande de mettre du poisson dans son ventre" inungnguwanik piingasungnguwanik takungngilatit petites personnes-INST. "N'as-tu pas vu trois petits?"

Pourquoi cette propriété de prendre l'instrumental, que symbolise le '1' qui figure à droite de la flèche de notre stratégie (9), ne résulterait-elle pas, le cas échéant, de l'opération morphologique qui crée le verbe? Quant à ce que Sadock appelle la 'référence', le problème de son analyse est qu'il n'y a rien découlant des faits signalés qui impose l'analyse incor­ porative qu'il en donne. Sapangarsivuq est un verbe morphologiquement in­ transitif qui, syntaxiquement, prend un complément au cas instrumental. Cette représentation des faits se justifie par le fait qu'il existe déjà beaucoup de verbes de cette classe en inuttitut, comme nous l'avons indiqué en (9). Ces formes résultent en partie au moins, selon Bergsland (1956), d'une opération morphologique qui les lie à des verbes transitifs sans qu'aucune opération d'incorporation ne soit évoquée. Ce dernier en dit ceci :

LINGUISTICS WITHOUT MORPHOPHONOLOGY

129

An instrumental form may have the force of a more or less indefinite object or a remoter object, in combination with certain intransitive or intransitively used verbs, especially verbs with an intransitivizing (medializing) derivational suffix, e.g., niqinik mayuussillutik 'to bring up the meat' inst. (op.cit., p. 76) Most transitive verbs, including verbs transitivized by the suffixes mentioned in 63-64 and 66 may have their object turned into an instrumental term [cf. exemple ci-haut—F & S] — and, consequently, their dependent subject turned into an annexed one — through a medializing suffix, (op.cit., p. 108-109)5

Nous en déduisons que parmi les opérations concomitantes à un changement de catégorie lors d'une opération morphologique, dans notre sens, il se peut qu'il se produise une attribution ou un changement de propriétés syntaxiques telle que, par exemple, le cas que prend le verbe, sans pour autant qu'il soit question d'une opération syntaxique, un peu de la même manière qu'une opération de suffixation pour créer un nom à partir d'un verbe, par exemple (12), attribue automatiquement le genre masculin au nom en français : (12)

[X]v3sg. ↔ [Xma]Nmasc.sg.

par exemple dans le cas de mots comme abonnement, bouleversement, changement, débrouillement, etc., l'opération (13) crée en inuttitut un verbe morphologiquement intransitif qui prend l'instrumental à partir d'un verbe transitif qui prend l'absolutif. (13)

[Xpaa]tra.3s

↔

[Xpuq]int.3s

Même si, lors de l'opération d'incorporation, sapanngamiq s'était transformé en sapangaq qui serait, selon Sadock, la forme suffixale de sapanngamiq, il conserve quand même certaines de ses propriétés, en particulier son nombre et son cas. Ce n'est pas parce qu'une opération manipule des traits catégoriels syntaxiques qu'elle est, ipso facto, de nature syntaxique elle-même; une opé­ ration morphologique a l'habitude d'en faire autant. D'autres générativistes ont aussi remis en question l'autonomie de la morphologie à base d'analyses syntaxiques de phénomènes analogues à l'in­ corporation nominale dans d'autres langues, notamment Baker (1988) pour le mohawk et Lieber (1992) pour une variété de langues. Nous les passons sous silence ici mais nous y reviendrons le cas échéant. Entretemps, répondons à l'autre assaut que subit la morphologie, qui lui vient d'un autre angle. De la même manière que la syntaxe s'est permise de s'accaparer d'une partie du domaine légitime de la morphologie, la phonologie, elle aussi, s'est 5 Les chiffres dans la citation de Bergsland se réfèrent à des exemples que nous reproduisons ici sous (9)

130

TRUBETZKOY' S ORPHAN

donnée comme tâche l'appropriation d'un terrain naturellement morpholo­ gique. Un aspect important de notre théorie des rapports entre morphologie et phonologie est qu'elle exclut toute possibilité de l'intervention d'information morphologique au niveau de la phonologie proprement dite, ce qui veut dire concrètement toute mention de structure morphologique au niveau des condi­ tions de bonne formation phonotactiques. Cette hypothèse est directement mise en cause à l'heure actuelle par la notion de morpheme structure condition (désormais 'Condition de Structure Morphologique'), ardemment défendue par plusieurs linguistes, en particulier récemment par Mohanan (1996), qui conclut : The construction types I have demonstrated to be relevant for phonology include the following types of information: morpheme (formatives and features), word, stem, affix, types of affixation, head, modifier and complement. (p 45) 6

À l'exception du concept de 'mot', qui est aussi fondamentalement un élément de la structure phonologique qu'il en est un de la structure morphologique, dans ce sens que toute CBF phonologique spécifie sa portée, soit syllabe, soit mot, dans son formalisme, les éléments morphologiques qu'évoque Mohanan n'ont pas leur place en phonologie. Pour nous en convaincre davantage, exa­ minons soigneusement les arguments mis de l'avant par Mohanan en faveur de la position qu'il défend. En fait, ces arguments se réduisent à un ensemble d'illustrations de descriptions de structures phonologiques exigeant chacune un CSM. Le pre­ mier phénomène concerne l'existence en anglais de mots de la forme de (14) : (14)

act, lift et risk

et l'absence d'autres de la forme de (15) : (15)

*zbin, *rizg et lizb

ceci couplé avec le fait que certains mots comme twelve, qui ont une consonne sonore en finale absolue, désonorisent cette consonne lors de l'ajout d'une constante morphologique qui débute en consonne sourde, par exemple dans twelfth, tandis que d'autres qui finissent en consonne sourde, comme cat, 6 En fait, il va plus loin que ceci quand il affirme que "[t]he evidence I have reviewed above suggests that phonological principles need to refer directly to morphosyntactic constructs such as the morpheme, head, complement and modifier..." (p. 45-46), mais comme nous défendons la position qu'il n'y a pas plus de morpho-syntaxe qu'il n'y a de morphonologie, nous ne prendrons pas ici la place nécessaire pour explorer toutes les implications de ces remarques.

LINGUISTICS WITHOUT MORPHOPHONOLOGY

131

désonorisent la consonne initiale sourde d'un suffixe qui s'y ajoute, par exemple cats [kats], comme prononciation de la structure morphologique du pluriel, qui est /katz/, cat plus le suffixe /z/. La solution de Mohanan à cet ensemble de faits est de postuler l'existence de la CSM(16): (16) 'Adjacent obstruents within a syllable are voiceless. (Absolute morpheme internally, weak across morphemes).' Le problème de cette solution est, d'abord, qu'elle exclut de la langue un très grand nombre des mots qui se trouvent chez de nombreux locuteurs. Nous pensons, par exemple, à des noms de famille comme Briggs, Dodds, Higgs, Hobbs, Mabbs, Tibbs et Woods ainsi que des toponymes très connus comme Hyves et Agde, lieux de villégiature très populaires chez les Britanniques, et des expressions fréquemment employées dans certains dialectes comme a diggs pour désigner un logement, typiques des parlers de Londres et sans doute d'ailleurs, et his nibs, expression courante chez Dickens et habituelle encore dans plusieurs dialectes pour désigner le diable et, par extension, chez certains, une personne dont on parle mais dont on veut éviter de prononcer le nom, généralement pour qu'elle ou une autre ne s'aperçoive pas du fait qu'on parle d'elle. Ensuite, (16) ne prévoit pas la création de néologismes tel qu'il s'est produit il n'y a pas si longtemps dans le cas d'un mot comme AIDS. Cet exemple de CSM de Mohanan nous fournit l'occasion idéale d'illustrer comment notre approche permet de résoudre le problème sans aucun des inconvénients que présente sa solution, et surtout sans l'emploi d'aucune CSM. La CBF phonologique de l'anglais qui est pertinente ici est bien (17), et sa violation est réparée par une assimilation progressive, tel que l'illustre un exemple comme cats. (17)

Un mot comme twelfth, par contre, résulte de l'application d'une stratégie morphologique qui, tout en suffixant le morphème /-0/, désonorise la con­ sonne précédente. Remarquez que cette transformation de la consonne n'est pas d'origine phonologique, comme on pourrait bien se l'imaginer, car elle ne dépend d'aucune CBF, la séquence [vø] étant bien attestée dans des mots comme givth, livth, troisième personne singulier, archaïques ou dialectales, du présent des verbes give et live. Comme l'illustrent nos propres exemples, il n'y a aucune restriction sur

132

TRUBETZKOY'S ORPHAN

des codas de syllabes de la forme [bz], [dz], [gz], et la morphologie des exemples de Mohanan prend la forme des deux stratégies (18) et (19)7 : (18) (19)

[X]N.sg.↔

[Xz]N.pl.

cats dogs, etc.

a. [X/+voix/]N.sg. ↔ [X/-voix/θ]Adj. twelfth b. [XC]N.sg.↔ [XCθ]Adj. fourth,fifth,sixth, etc.

Dans ce qui suit, nous présentons des arguments contre les autres CSM de Mohanan. Son deuxième exemple prend la forme de (20a) : (20) a. Morpheme internally in English, a coda can have at most three consonants. Le phénomène que vise à expliquer cette condition est le fait que, dans le cas de mots qui résultent d'une suffixation, il se trouve des codas de syllabe ayant plus de trois consonnes, par exemple dans le cas d'un mot comme texts, que pourrait engendrer notre stratégie (16). Nos recherches ne nous ont pas permis de dénicher de contre-exemple à la condition (18a), ce qui ne veut pas dire pour autant qu'elle reste valable, car ce n'est pas notre intuition qu'un locuteur de l'anglais ne serait pas capable de prononcer des mots de la forme proscrite. II n'y en a pas, tout simplement, du moins à notre connaissance. Imaginez, par exemple, qu'il se crée une compagnie au nom enregistré de American Continental Shipping and Transport Services. Le locuteur serait-il incapable de parler de an ACSTS truck ou de dire ACSTS shares rising at the stock market?Les locuteurs avec lesquels nous avons vérifié cet exemple, ainsi que d'autres du type : Universal Turbine Throttle Cleaning Service (an UTThCS van) n'ont pas l'impression que ces mots échappent à la prononciation, bien qu'ils ne soient pas facilement réalisables tout de suite. Un troisième exemple de CSM de Mohanan, qui concerne aussi l'anglais, est (20b) : (20) b. In English, a consonantal segment can be syllabic only at the end of a morpheme/stem. Pour que cette condition soit respectée il faudrait qu'un mot comme 7 Bien que le mot twelfth semble illustrer le phénomène de hapax legomena auquel nous nous référions plus tôt.

LINGUISTICS WITHOUT MORPHOPHONOLOGY

133

Middlesborough ou Edinborough soit structuré comme dans (21), où 7/' (le tiret double) représente une frontière de morphème et 7' (le tiret simple) une frontière de syllabe : (21)

Middlesborough /mid/lz/brə/

Edinborough /ε/dIn//brə/

Pour que ceci soit le cas, il faudrait qu'il existe une stratégie de formation de mot de la forme de (22): (22)

[X]? ↔ [Xbrə ]N. pl. Middlesborough, Edinborough, etc.

Le problème est que, selon notre théorie, pour que (22) soit une stratégie bien formée il faudrait que /midlz/ et /εdIn/, les éléments auquels on ajoute le suffixe /brə/, soient des mots, ce qui implique qu'ils soient des éléments bien identi­ fiables comme tels, et appartenant à une même catégorie morphologique. Ceci n'étant pas le cas, la stratégie (22) devient douteuse et ne fait probablement pas partie de la morphologie de l'anglais, étant donné que la catégorie signalée en (22) par un point d'interrogation reste indéterminable. De notre point de vue donc, ces mots constituent des contre-exemples à la CSM (20b), bien que Mohanan, ne partageant pas notre vision de la morphologie, ne trouvera peutêtre pas cette objection valable. Le quatrième exemple de CSM de Mohanan concerne aussi l'anglais : (20) c. Morpheme internally, in English, dental fricatives cannot occur after an obstruent in a coda.  nous semble qu'un mot comme plimpth constitue un contre-exemple à cette condition, et il y en a sans doute d'autres. D'autres reproches concernent des points techniques de la forme de la stratégie morphologique elle-même, de l'information qu'elle actualise et de son organisation au sein de la composante morphologique. Certains nous repro­ chent l'absence d'une place explicite pour ce qu'on appelle depuis longtemps, sans toutefois vouloir toujours désigner un même phénomène, la 'produc­ tivité'. Sur ce point, nous profitons de l'occasion pour identifier toute forme de 'productivité' comme une construction statistique qui ne saurait réfléter quoi que ce soit à propos des procédés morphologiques et de leur fonctionnement, tout comme elle ne saurait rien refléter non plus du fonctionnement de la phonologie. Toutes nos stratégies sont également productives dans la mesure où elles s'appliquent là où elles peuvent. La nature de leur concurrence est partiellement déterminée par la forme de la stratégie qui définit sa portée, mais en cas de véritable concurrence directe entre deux stratégies ayant des portées

134

TRUBEIZKOY'S ORPHAN

qui se recoupent, les facteurs déterminants se doivent d'être extralinguistiques : il n'y a rien dans la grammaire qui nous oblige à dire deux chevaux plutôt que deux chevals, he strove plutôt que he strived, chupa-ajîs plutôt que chupaajises, mais il peut y avoir bien des facteurs extralinguistiques qui y contribuent. Cette concurrence des stratégies morphologiques non seulement constitue leur trait distinctif par rapport aux stratégies de réparation phonologiques, les­ quelles, par contraste, sont toujours absolues, mais porte aussi un coup mortel à la morphonologie. Si l'alternance formelle entre goose et geese s'explique par une opération dite morphonologique et celle entre duck et ducks par une stratégie morphologique, comment rendre compte de l'existence des formes gooses et geeses qui caractérisent le parler de la plupart des locuteurs de l'anglais à un moment donné de leur apprentissage? Par contre, si les deux alternances sont purement morphologiques, la variation paraît naturelle. Un aspect de notre morphologie qui ne semble pas avoir été bien compris est le fait que celle-ci ne paraît pas s'accommoder des mots dits 'composés', c'est-à-dire des mots qui, dans d'autres analyses, proviendraient de l'asso­ ciation de deux mots. Vu sous cet angle, il est vrai que notre morphologie n'engendre aucun mot de cette nature, et ceci tout simplement parce qu'il n'y en a dans aucune langue. Les mots du type ouvre-boîte, noeud papillon ou mainlevée, auxquels on a attribué ce qualificatif en français, ne se composent pas du tout de deux autres mots mais plutôt, conformément à notre stratégie morphologique, d'une constante qui, dans ces exemples, prend une forme qui ressemble à celle d'un véritable mot qui, en tant que tel, pourrait bien se trouver dans un autre contexte (c'est ce qui correspond dans nos trois exem­ ples aux formes ouvre-, -papillon et main-), et d'une variable qui, en tant que membre d'une catégorie morphologique, est nécessairement et par définition un mot. Parmi les stratégies morphologiques du français se trouvent donc les trois stratégies de formation de mot (23), (24) et (25), qui engendrent respectivement et entre autres les mots ouvre-boîte, noeud-papillon ou mainlevée : (23)

[X] N

↔

[uvrX]N masc.

(24)

[X] N

↔

[Xpapij]N

(25)

[X]Adj.fém.

↔

[mX]N fém.

Une autre objection à cette vision unifiée du procédé morphologique concerne

LINGUISTICS WITHOUT MORPHOPHONOLOGY

135

la bidirectionnalité des stratégies. Nous avons longuement justifié ailleurs8 l'existence de cette propriété, mais ces arguments peuvent être renforcés en soulignant un aspect de la question sur lequel nous n'avons pas insisté dans notre article, à savoir le fait que la bidirectionnalité des stratégies nous permet de nous débarasser complètement de la notion de 'zéro' comme marque morphologique, et nous permet en même temps de bien caractériser le rapport de similarité qui constitue toujours une stratégie concurrentielle au même titre que tout autre. C'est ainsi que nous rétablissons l'ordre aristotélien que renverse le structuralisme de la façon que nous avons signalée au début de cet article. Signalons, à côté de cet avantage qu'apporte notre morphologie, d'autres avantages qui aident également à résoudre certains problèmes qui ont long­ temps hanté de nombreux linguistes. Le premier concerne la question de l'ac­ centuation et sa place dans la grammaire. Il est clair que, pour les langues qui connaissent une position variable d'accent, cette variation dépend, de notre point de vue, soit de facteurs phonologiques, soit de facteurs morphologiques. Lors d'une description dans un cadre théorique qui comprend une morphonologie — les exemples classiques sont ceux de l'anglais ou du latin — l'accent est soit morphonologique, soit phonologique. On dirait que la base supposément phonétique de ces deux prétendus sous-domaines de la gram­ maire a continuellement sollicité la recherche d'une théorie unifiée de l'accen­ tuation. En réunifiant la morphonologie à la morphologie nous pouvons entretenir l'espoir d'étouffer cette tentation.  est clair que dans ce genre de langue le déplacement de l'accent figure fréquemment comme partie intégrante d'une opération morphologique dans le sens où nous l'entendons, laquelle constitue parfois à elle seule la différence formelle entre deux mots : pego "je frappe" et pegó "il frappa" en espagnol ou protestv et protestN de l'anglais sont des exemples classiques. En séparant ces exemples clairs d'accentuation morphologique des cas d'accentuation phono­ logique, par exemple la condition de bonne formation qui limite à un seul le nombre d'accents dans un mot espagnol ou celle qui, dans la même langue, empêche toute accentuation à plus de trois syllabes de la fin du mot, nous espérons encourager certains à abandonner la recherche d'une théorie unifiée de l'accentuation dans ces langues et à concentrer leurs énergies sur des objectifs plus réalistes, puisque réalisables. Remarquons finalement comment la fusion de la stratégie de formation de mot en un seul procédé avec l'opération dite morphonologique que, dans une analyse qui maintient la morphonologie, elle est supposée déclencher, permet Ford et Singh (1985).

136

TRUBETZKOY'S ORPHAN

d'expliquer pourquoi on n'observe jamais l'application de l'une sans celle de l'autre, ce qui constitue à notre sens un argument de poids contre l'identifi­ cation d'une composante morphonologique autonome. Un facteur important en faveur de cette hypothèse provient de la diachronie, qui ne semble fournir aucun appui au fait que la partie 'phonologique' d'une opération dite morphophonologique peut s'en abstraire et se généraliser. Quant aux contre-exemples putatifs au fait que la morphonologie ne se généralise pas en diachronie, étant donné le fait que Malkiel (1982:248) a retiré son exemple espagnol de 1976 (cf.Mendez Dosuna & Pensado 1986, aussi), il n'en reste pas beaucoup, et nous devons peut-être nous limiter au cas récent et presque convaincant présenté par Morin, Langlois & Varin (1990), que nous avons traité dans Ford & Singh (1991). Ces auteurs prétendent que ɔ → o, l'alternance indiscutablement 'morphonologique', de leur point de vue, du français du 17 ième siècle, s'est généralisée pour devenir ce que nous estimons tous être un procédé global, purement phonologique, par exemple le fait que l'alternance [so] "stupide"Sg. [so] "stupide"p1. se soit neutralisée en [so] aux deux personnes. Au sujet de ce phénomène, quatre points sont clairs: (i) il était 'morphophonologique' dans le sens de Morin, Langlois & Varin, au 17 ième et après; (ii) il est actuellement phonologique dans le sens où nous l'entendons tous; (iii) il ne s'est jamais généralisé, dans ce sens qu'il n'en est jamais venu à signaler une nouvelle opposition morphologique, ni ne s'est jamais incorporé au sein d'une opération de formation de mot avec laquelle il a déjà été associé, et (iv) il a connu une diffusion lexicale, d'abord à l'intérieur de sa catégorie d'origine et, plus tard, de façon intercatégorielle en tant que 'règle de redondance lexi­ cale'. L'impression qu'il donne de s'être transformé en 'règle phonologique' est en grande partie un résultat du fait que l'autre catégorie majeure, le verbe, ne connaissait pas de forme en  final (Morin, Langlois & Varin 1990:516). La généralisation d'une 'règle morphophonologique' comme 'règle phono­ logique' doit, nous semble-t-il, impliquer l'appropriation d'autres caté-gories morphologiques, ce qui n'est pas le cas lors de la tension de o, car cette tension passe du pluriel vers une 'règle de redondance' de plus en plus trans­ catégorielle, pour terminer comme condition de bonne formation, sans jamais être déclenchée par un autre procédé morphologique. Quoique Morin, Langlois et Varin aient encore raison dans le sens où cette opération ou, plus précisé­ ment, la neutralisation dont elle rend compte, soit phonologisée, leur exemple semble être un cas de phonologisation sans généralisation du genre que nous estimons un critère du phénomène que nous abordons. Il semble s'agir d'une décomplexification globale plutôt que de la généralisation d'une 'règle

LINGUISTICS WITHOUT MORPHOPHONOLOGY

137

morphophonologique' comme 'règle phonologique' Le phénomène n'a pas pénétré d'autre procédés morphologiques, mais a quand même atteint le statut d'une alternance automatique, du moins pour un locuteur Imaginatif. Quoiqu'il s'agisse d'une des études les plus sérieusement documentées de ce que les auteurs estiment un dépouillement progressif des conditions morpho-logiques et lexicales d'une 'règle moiphophonologique', il n'est pas besoin, heureuse­ ment, de l'interpréter comme un cas clair de 'règle morphophono-logique' qui viole nos restrictions de 1983.9 Le cas présenté par Morin et al. vise à remettre en question Hooper (1976:91), qui exclut le dépouillement des conditions sur une 'règle morphophonologique' comme une évolution possible. Les conditions nécessaires pour que le dépouillement mène à la 'phonologisation' d'une alternance que Hooper qualifie de 'règle MP' demande une considération étroite. Car même le genre d'exemple évoqué par Morin et al est rare. Peut-être la véritable importance de l'exemple sous considération se trouve dans le fait que nous faisons l'erreur d'évoquer dans un même contexte la synchronie, la capacité communicative d'extraction et la possibilité de générali­ sation sociale diachronique. Le locuteur n'extrait rien; on est plutôt en présence de facteurs subtils qui produisent une situation qui ne saurait se distinguer empiriquement de celle que Hooper cherche à exclure. La rareté d'un tel développement n'est ici qu'une conséquence du fait que la combinaison heureuse requise ne se produit pas fréquemment. La généralisation d'une 'règle morphophonologique' dans le sens interdit signalé reste quand même inattestée. Peut-être faudrait-il mettre en évidence le point crucial qui est que la partie soi-disant 'phonologique' d'une règle complexe qui incorpore à la fois un affixe et une alternance non automatique ne saurait s'extraire ni s'abstraire de façon à devenir disponible pour s'associer à un autre affixe avec lequel elle 9 Les restrictions en question ont été formulées comme suit dans Ford & Singh (1983:67) : "Fourthly, morphophonological alternations do not act independently in the historical evolution of a language. We do not wish to affirm by this that one cannot witness the extension of a morphophonological process from generation to generation as the opposite is easy to attest. Neither do we wish to affirm that a morphophonological alternation could not be produced spontaneously during the evolution of a language. What we wish to affirm is that a morphophonological alternation is not dynamic in the sense that it can wander from one morphological context to another. For example, we'd like to claim that the mark of the imperfect or the morpheme -los could not trigger umlaut in German and that diminutive -ito /-ita could not trigger a monophthongization in the Spanish noun. We think, until proof to the contrary is offered, that historical changes of this form are impossible and that a morphophonological process cannot be mobile in this sense. Fifthly, a morphophonological alternation can spread without there being present the phonological conditions that were present at its appearance."

138

TRUBETZKOY'S ORPHAN

ignorait auparavant toute association. Voilà ce que nous voulions signaler quand nous avons indiqué que -ito ou -ita ne pouvait jamais déclencher une monophiongaison en espagnol. Il est peut-être possible, là où aucune ex­ traction de ce genre n'est requise, que cette partie de la morphonologie se généralise au moyen de la règle courante de concurrence entre stratégies morphologiques. Le point crucial est que la soi-disant 'phonologie' d'une opération morphologique complexe en est une partie intégrante et ne saurait ni s'isoler, ni se séparer d'elle. La diachronie confirme notre hypothèse de nonséparabilité. Voilà le sens de notre remarque voulant que la 'morphophono­ logie' n'est pas mobile, et nous ne voyons pas pourquoi nous devrions nous écarter de cette prise de position. De façon analogue, notre morphologie permet de rendre compte du comportement des locuteurs en situation de langues en contact ou lors de l'acquisition d'une langue seconde. On sait que dans les contextes d'acquisition de langue seconde et de langues en contact en général, ce qu'on a appelé la 'morphonologie', et la mor­ phologie, telle que nous la concevons, se comportent de la même manière : aucune des deux ne semble créer ce qu'on appelle souvent de  'interférence'. Dans le contexte de l'acquisition d'une langue seconde, les 'erreurs' que l'apprenant est sensé produire proviennent de la phonologie et de la morpho­ logie de sa première langue (cf. Singh 1991, Singh & Ford 1987, Singh & Martohardjono 1988, et Singh & Parkinson, in press). Dans le contexte de ce qu'on appelle le plus souvent langues en contact, l'adaptation phonologique et morphologique se renforce sous l'influence de la phonologie et la morphologie de la langue qui emprunte. Dans les deux contextes, la morphonologie se comporte comme les bound morphemes de Moravcsik (1978:110); autrement dit, elle ne se déplace pas sans se faire accompagner des mots auxquels elle appartient. C'est précisément à cause de ce mode de déplacement que la mor­ phonologie d'un ensemble de mots emprunté à un nombre indéterminé de langues se comporte morphonologiquement d'une façon qui se rapproche beaucoup de son comportement dans sa langue d'origine. De notre point de vue, un ensemble d'emprunts, phonologiquement et sémantiquement apparen­ tés, donne lieu à une stratégie de formation de mot qui ne saurait varier d'une langue source à une autre. La limite de leur variation est celle que détermine leur phonologie. Il importe peu que la différence phonologique soit au niveau des traits, des segments, de la prosodie ou des affixes. Une fois la stratégie disponible, elle peut servir à produire des mots qui pour certains locuteurs doivent toujours mettre en évidence leur complexité morphologique de manière différente. Ill-formity en anglais est au moins aussi possible que priesthood, et

LINGUISTICS WITHOUT MORPHOPHONOLOGY

139

mongeese aussi possible que mongooses. Peu importe le statut du principe universel #2 de Moravscik, il ne semble pas s'imposer de distinction entre la morphonologie et la morphologie en ce qui concerne l'acquisition de la langue seconde ni le contexte de langues en contact. En prime, il s'ensuit que la dissociation de la morphonologie de la phono­ logie, tout en nous permettant de rendre compte des faits empiriques, donne lieu à une phonologie beaucoup plus restreinte que celle qui résulte du maintien d'un lien entre les deux. Pour conclure, nous estimons que les objections formulées à l'égard d'une composante morphologique destinée à alimenter directement la phonologie de la grammaire, sans aucune intervention intermédiaire, ne s'avèrent dans aucun cas valables et qu'il est souhaitable de poursuivre sans crainte de perte impor­ tante d'information l'emploi de ce modèle au niveau de la description lin­ guistique.

Where Does Morphophonology Belong? Comments on Ford & Singh

. . Mohanan National University of Singapore I'd like to begin by reminding the reader that the issues of morphophonology spelt out in my comments on Kiparky's paper are equally relevant to Ford and Singh's paper. The reader may therefore find it useful to begin with the first two sections of my comments on Kiparsky before going on to my comments on Ford and Singh. 1. Compositionality of Words Most approaches to morphology assume that the syntactic representation of words can specify word internal constituency relations. I will refer to the class of morphological theories that subscribe to this position as 'compositional mor­ phology'. To illustrate, the word internal syntactic structure of the words in Those boys saddened Julia within compositional morphology will be as exemplified in (1): (1) Compositional Morphology:

LINGUISTICS WITHOUT MORPHOPHONOLOGY

141

I am using the terms sentence syntax and word syntax in Selkirk's sense here. Thus, (2a) is part of sentence syntax, while (2b) is part of word syntax: (2) a. The NP those boys consists of the DET those and the N boys. (sentence syntax) b. The V sadden consists of the ADJ sad and the suffix en. (word syntax) Central to statements such as those in (2) is the claim of word internal compositionality, which Ford & Singh trace back to Pānini, stated in (3): (3)

Word Internal Syntactic Compositionality (WISC) Words have internal syntactic constituency relations.

As remarked above, Word Internal Syntactic Compositionality is taken for granted (often without argumentation) in most approaches to morphology. The paper by Ford and Singh challenges this position, and claims that the syntactic representation of words cannot specify word internal constituency relations. I will refer to this hypothesis as 'non-compositional morphology'. Non-compo­ sitional morphology poses a bold and serious threat to the central dogma in morphological theory. Being the more restrictive hypothesis, in the absence of evidence against it, non-compositional morphology is the one that we must adopt. As far as I know, the challenge posed by this theory has.not been ad­ dressed satisfactorily in the literature. The theory that Ford and Singh propose, they call 'Projection Morpholo­ gy'. Another theory of non-compositional morphology is Anderson's 'Amor­ phous Morphology'. Both these theories accept the legitimacy of statements like "The NP those boys consists of the DET those and the N boys" in (2a), but reject the legitimacy of statements like "The V sadden consists of the ADJ sad and the suffix .'' in (2b). Their version of the structure of Those boys saddened Julia is given in (4): (4) Non-compositional Morphology (Ford & Singh, Anderson)

TRUBEIZKOY' S ORPHAN

142

Thus, the notion morpheme, and word internal syntactic constituency, has no place in this view of morphology. 2. Relatedness of Words 2.1 Compositional morphology Having rejected the WISC, the first question that needs to be addressed is the formal expression of the relatedness of words, raised in (5): (5) Relatedness: How do we express the phonological, syntactic, and semantic relatedness of words? Take, for example, the systematic pairings of syntactic, phonological and semantic properties illustrated in (6), which every approach to morphology must face:

The conventional strategy of expressing the relatedness in (6) involves compositionality or part-whole relationship, in syntax. That is to say, we assume that the representation of sadden is syntactically composed of the representation of sad and the affix -en. For ease of reference, let us number these two word parts as illustrated in (7): (7) Representations of word-parts: a.

b.

syntax: phonology: semantics: syntax: phonology: semantics:

ADJ sæd MENTAL STATE-i sister node = ADJ, mother node = VERB ən CAUSE TO BE IN X where x = semantics of the sister node.

LINGUISTICS WITHOUT MORPHOPHONOLOGY

143

Given the representations of the word parts in (7), the phonological, syntactic, and semantic relatedness of the words in (6) are captured by assuming the compositionality of words in (8): (8) Representations of words: sad sadden / \ 2.2 Non-Compositional Morphology As I said, both Amorphous Morphology and Projection Morphology reject the hypothesis of Word Internal Syntactic Compositionality, and hence they reject the legitimacy of (8). For instance, Anderson (1992) claims that the 'structure' of discontentedness is not (9a), but (9b) (Chapter 10, section 10.1) (9) a. Compositional representation:

b. Non-compositional representation: [ N content] R d i s → [v discontent] ness → [N discontentedness]

R

ed → [ADJ discontented]

R

It is important to note that what we are dealing with here is the question of representations, not the question of rule systems. Thus, one could treat affix­ ation as a rule, as in process morphology, and still derive a form that indicates its internal syntactic constituency, as illustrated in (9c): c. Compositional representation in Item-Process Model: [N content] R dis → [N dis[N content]] R ed → [ADJ [ N dis[N content]]ed] R ness → [N [ADJ [N dis[N content]]ed] ness] Theories of Non-compositional Morphology reject (9c). How is the relatedness of words captured in non-compositional theories? The solution in Amorphous Morphology is that of 'procedural unidirectional statements' that derive one word from another word, without taking recourse to syntactic composition-

144

TOUBETZKOY' S ORPHAN

ality. The form of the Word-formation Rule is given in ordinary prose in (30). The way this formalism captures the relatedness in (6) is illustrated in (10) and (11): (10)

(11)

Relatedness à la Amorphous Morphology (e.g., Anderson ch. 7, § 7.1.): If there is a word w-i with: phonological properties phon-X syntactic properties syn-Y, and semantic properties sem-Z, change it optionally to word w-j with: phonological properties phon-X' syntactic properties syn-Y', and semantic properties sem-Z'. (Notation: X → Y) Relatedness illustrated in (6) : If there is a word w-i with phonology: X syntax: ADJ semantics: Z then change it optionally to word w-j with phonology: ən syntax: VERB semantics: CAUSE TO BE IN Z

The solution given by Ford and Singh is that of 'declarative bi-directional statements' that establish correspondences between words directly, again with­ out taking recourse to compositionality. The general form of the Word-for­ mation Rule is given in (12). The way this formalism captures the relatedness illustrated in (6) is illustrated in (13): (12)

Relatedness à la Projection Morphology: If there is a word w-i with phonological properties phon-X syntactic properties syn-Y, and semantic properties sem-Z, then there is a word w-j with phonological properties phon-X' syntactic properties syn-Y', and semantic properties sem-Z', and vice versa. (Notation: X ↔ Y)

LINGUISTICS WITHOUT MORPHOPHONOLOGY

(13)

145

If there is a word w-i with phonology: X syntax: ADJ semantics: Z then there is a word w-j with y with phonology: Xən syntax: VERB semantics: CAUSE TO BE IN Z and vice versa.

Ford and Singh concentrate on morpho-syntax and morpho-phonology, and ignore morpho-semantics. I have taken the liberty to include my version of how they might do semantics in order to place their proposals in the larger perspective of linguistic organization within words. Thus, both Projection Morphology and Amorphous Morphology reject the hypothesis of Word Internal Syntactic Compositionality. The crucial difference between them is in the formalism of the rule system they employ. Amorphous Morphology uses a procedural unidirectional formalism, while Projection Mor­ phology uses a declarative bi-directional formalism. I will therefore use the term 'Procedural Non-compositional Morphology' (PNM) to refer to Amor­ phous Morphology, and 'Declarative Non-compositional Morphology' (DNM) to refer to Projection Morphology. It is important to emphasize that what non-compositional morphologies reject is the syntactic compositionality of words, not their phonological or se­ mantic compositionality. Thus, Projection Morphology and Amorphous Mor­ phology would accept the statements in (14a), (14b.i), and (14c), but reject the statement in (14b.ii):1 1 My interpretation of the analysis of noun-verb alternations in words like measure in Kiparsky (1985) is that they presuppose representations such as the following:

NOUN

I MEASURE

Given Kiparsky's current position, the zero morph can be eliminated from these representations: NOUN

I MEASURE

146

TRUBEIZKOY'S ORPHAN

(14)

Three dimensions of compositionality: a. Phonological compositionality. e.g., The word sadden contains two syllables. The first syllable contains two segments; the second syllable contains three segments. b. Syntactic compositionality. (i) Feature Compositionality e.g., The word boys contains the features [+N], [-V] and [PLUR] (ii) Constituency e.g., The word sadden contains the adjective sad and the suffix en. c. Semantic compositionality: e.g., The word sadden contains the meanings of CAUSE TO BE and STATE.

On the basis of these observations, we may identify the fundamental premise of Projection and Amorphous morphologies as (15). (15)

Non-compositional Morphology: The smallest unit of grammar that can be associated with both syntactico-semantic properties as well as phonological properties is the word, not morpheme or morph.

In contrast, compositional morphologies claim that: (16)

Compositional Morphology: The smallest unit of grammar that can associated with both syntactico-semantic properties as well as phonological properties is the morpheme/morph.

I think (15) and (16) yield equivalent descriptions of most individual patterns of morphological relatedness, but how non-compositional morphology will express interaction between different patterns is not clear to me. Take, for instance, the interaction between the reduplication and the affixation of /məŋ)-/ in Malay and Indonesian. Given a root such as tumbuk "punch", affixation yields [mənumbuk], in which assimilation of the nasal and deletion of the voiceless stop change the sequence /ŋt/ to a single [n]. Reduplication of the The representations in Amorphous Morphology and Projection Morphology will be: NOUN

I MEASURE

VERB

I MEASURE

LINGUISTICS WITHOUT MORPHOPHONOLOGY

147

root yields [tumbuktumbuk]. The combination of affixation and reduplication yields two forms: [tumbukmənumbuk] and [mənumbuknumbuk]. I do not see how non-compositional morphology will describe the relatedness among [tumbuk], [mənumbuk], [tumbuktumbuk], [tumbukmənumbuk], and [mənumbuknumbuk]. 3. Morphology and Phonology Compositional Morphology (CM), Procedural Non-compositional Morphology (PNM), and Declarative Non-Compositional Morphology (DNM) yield different types of interaction between patterns of phonology and patterns of morphology. First, let us consider the phonology-morphology interface in Procedural Non-compositional Morphology. In this system, there is an asym­ metry in derivation. Thus, the complex form goodness is derived from the simple form, good. It is therefore possible to distinguish between the underived forms prior to the application of Word-formation Rules (WFR's) and the derived forms which are the output of WFR's: (17)

Procedural Non-compositional Morphology: underived forms → WFR's → derived forms

In such a system, it is possible to distinguish between the regularities observed in root words (roots which are words) and non-root words. Take, for example, the asymmetry between the possibiUty of syllabic lateral in (18a) and (18b) and its impossibility in (18c) and (18d) in English: (18)

a. b. c. d.

[simpl], [fikl], [saikl] [fiklnəs], [saikləs] *[lpin], *[lken], *[lsaid] *[pldin], *[klsen], *[flmein]

The reason why speakers of English reject coinages such as those in (18c) and (18d) can be accounted for if we postulate the syllable structure condition given in (19): (19) [+consonantal] can be syllabic in English only at the end of a root. The generalization in (19) can be stated in Procedural Non-compositional Morphology by stipulating that the principle applies only prior to the application of WFR's. In contrast, there is no way of expressing the counter­ part of (19) in the Projection Morphology of Ford and Singh, or in any Declarative Non-compositional Morphology for that matter. This is because

148

TRUBETZKOY' S ORPHAN

Projection Morphology does not yield a distinction between two kinds of word medial environments, namely, root final and root medial.2 The fundamental tenet of Lexical Phonology is that phonological rules can be interspersed with word-formation rules. Counter to what is frequently made out to be in the literature, Amorphous Morphology (or Item-Process Morpho­ logy in general) is not inconsistent with this tenet of Lexical Phonology. Imagine, for example, a theory of amorphous morphology in which WFR's are grouped into two 'levels' or strata, and the modules of phonology and morphonology, and the relation between them, are organized as given in (20): (20)

Amorphous Morphology + Lexical Phonology Underived words: Underlying Representations

↓ lexical module 1 WFR's lexical module 2 [WFR's

phonological rules

↓ post lexical module

↓ phonetic representations of derived words This model is equivalent to a model of Lexical Morphology with Bracket Erasure at the end of every morphological operation, taking place in morpho­ logy rather than phonology. The phonology that accompanies it can refer to the edges of morphological representations, but not to junctions. That is, it can express the counterparts of the Compositional Morphology statements in (21a) and (21b), but not those of (21c) or (21d): (21) Edge statements: a. Segment x becomes segment y at the end of a form in module z. b. Segment x becomes segment y at the beginning of a form in module z. Junction statements:  Segment x becomes segment  at the end of a form when followed by another form in module z. Ford and Singh's objection to (19) will be discussed later.

LINGUISTICS WITHOUT MORPHOPHONOLOGY

149

d. Segment x becomes segment y at the beginning of a form when preceded by another form in module z. Though restricted in this manner, Procedural Non-compositional Morphology is consistent with a phonological theory that allows phonological rules to refer to morphological information. For example, the model in (20) is perfectly capable of treating English rules like trisyllabic shortening, spirantization, stem final nasal deletion, and the like as phonological rules, rather than as part of WFR's. In sum, Compositional Morphology and Procedural Non-Compositional Morphology make the types of information listed in (22a) and (22b) visible to the phonological module: (22)

a. Compositional Morphology (e.g., Lexical Morphology): Bound forms (affixes and roots which are not words), Roots which are words, and Word internal edges and junctions. b. Procedural Non-compositional Morphology (e.g., Amorphous Morphology): Roots which are words Word internal edges  Declarative Non-Compositional Morphology (e.g., Projection Morphology): None of the above

In Projection Morphology, the consequence of the morphological represen­ tations and morphological rule systems is that the phonological module does not have access to any morphological information. It therefore makes the em­ pirical claim stated in (23): (23) Hypothesis of Autonomous Phonology Phonological rules cannot be sensitive to morphological information. 4. Autonomous Phonology Given the Hypothesis of Autonomous Phonology, there is no such thing as morphophonology. If no phonological rule can refer to morphological infor­ mation, then it follows that any sound pattern which is sensitive to morphological information must be formulated as part of a Word-formation Rule. Hence, the modularity entailed by Projection Morphology is the one in (24):

150

TRUBETZKOY'S ORPHAN

(24) Projection Morphology Autonomous Phonology I WFR's that state semantic, I I phonological rules/constraints syntactic, and phonological → that cannot be sensitive to relatedness | | morphological information | Having worked through the statements of phonological alternation in terms of WFRs, I am persuaded that almost everything that can be formulated as a morphophonemic rule in the phonological module by accessing morphological structure can also be formulated as part of a Word-formation Rule in the mor­ phological module. As far as regularities of alternation are concerned, the pre­ dictions made by autonomous phonology in (25a) and non-autonomous phonology in (25b) appear to be identical. If this is indeed the case, the two positions are not empirically distinct, and hence their differences have no theoretical consequence: (25)

Regularities of Alternation: a. Autonomous Phonology Morphologically sensitive regularities of alternation in pronunciation are part of the module of morphology. b. Non-autonomous Phonology Morphologically sensitive regularities of alternation in pro­ nunciation can be part of either the module of phonology or the module of morphology.

As far as distributional patterns are concerned, the story is different. First, consider the opposing theoretical positions in (26a) and (26b): (26)

Regularities of Distribution: a. Autonomous Phonology Morphologically sensitive regularities of distribution in pro­ nunciation are part of the module of morphology. b. Non-autonomous Phonology Morphologically sensitive regularities of distribution in pro­ nunciation can be part of either the module of phonology or the module of morphology.

A WFR can state a pattern of phonological relationship between two words. However, a WFR cannot state a distributional regularity that exists in proper subparts of words, at least not in Projection Morphology. Hence, the Hypoth-

LINGUISTICS WITHOUT MORPHOPHONOLOGY

151

esis of Autonomous Phonology in Projection Morphology makes the strong empirical claim in (27): (27)

There are no distributional patterns of pronunciation such that they hold on proper subparts of words, but not on words.

This prediction, I believe, is incorrect. In Mohanan (1995), I argue for a num­ ber of generalizations which hold on underived words but not on derived words. These generalizations are listed in (28): (28)

a. In English, adjacent obstruents in a syllable within a morpheme are voiceless. b. In English, a consonantal segment can be syllabic only at the end of a morpheme/stem. c. In English, a coda within a morpheme can have at most three consonants. d. In English, dental fricatives cannot occur after an obstruent in a coda within a morpheme. e. In Malayalam, a single dental nasal followed by a vowel can only occur morpheme initially; in this environment, a single alveolar nasal cannot occur.

Principle (28a) allows tautomorphemic syllables like those in (29a) and (29b), and polymorphemic syllables like those in (29c), but not tautomorphemic syllables like (29d): (29)

a. [spim], [bisk], [lisp] b. [ækt], [bift], [oks] c. [bægd], [livd], [bægz], [feidz] d. * [zbim], *[rizg], *[lizb] e. * He is [lægdiŋ] me; * He is [mebziŋ] me.

Ford & Singh cite lexical exceptions such as Higgs [higz] and AIDS [eidz], and dismiss the generalization as incorrect, but they have no explanation for why the coinages in (29d) are rejected as unacceptable by speakers of English. Nor do they have any explanation for why -ing verbs of the form in (29e) are rejected. Is it an accident that every 'counter-example' that Ford & Singh cite have the same phonological shape of [noun + plural] or [verb + present tense]? One way to deal with forms like Higgs and AIDS will be to assume that

152

TRUBEIZKOY'S ORPHAN

they consist of two 'morphemes' and two 'morphs', but the second morph /z/ does not carry the feature PLURAL or PRESENT TENSE that typically go with the morph /z/ in English. If this analysis sounds farfetched, one may think of them simply as lexical exceptions. At least, principle (28a) has an explanation for (29c) and (29d). Projection Morphology has none. Principle (28b) allows forms like (30a) where the syllabic /1/ is word final, and forms like (30b) where the syllabic /1/ is word internal but stem final. It disallows forms like those in (30c) and (30d), where the /1/ is a sonority peak but not stem final. (30)

a. [simpl, [fikl, [saikl] b. [fiklnəs], [saikləs] c. *[lpin], *[lkon], *[lsaid] d. *[pldin], *[klson], *[flmein]

Once again, Ford & Singh object to the generalization citing forms like Middlesborough and Edinborough, which have the appearance of other bimorphemic forms in English. And once again, they have no explanation for the fact that the coinages in (30c) and (30d) are rejected by speakers of English. In a similar vein, they object to (28c) and (28d) by citing acronyms like Acsts [æksts] ("American Continental Shipping and Transport Services") Uthcs [uteks] ("Universal Turbine Throttle Cleaning Service"). I wonder how many speakers of English have this word in their mental dictionary, and how many of them will accept a past tense verb such as [bætøkst] (as in "Mary has always [bætøkst] John") as an acceptable coinage in English. The asterisks in (29c, d) and (30c, d) do not mean 'not occurring in the corpus/dictionary'. Rather, they mean 'not acceptable to be a speaker of the language'. Ford & Singh's response is that forms which violate the gener­ alizations occur in the corpus of actual words in a dictionary. I find it some­ what surprising that in objecting to the generalizations in (47a-e), Ford & Singh are using the methodology of corpus linguistics whose object of inquiry is the text, not the methodology of generative linguistics whose object of inquiry is the mental linguistic systems of language users. Even a robust generalization such as "Adjacent obstruents in English must agree in voice", which Ford & Singh accept as a robust phonological pattern in English, have lexical exceptions such as midst [midst] and svelt [svelt]. Citing lexical exceptions does not negate the claims of the robustness of a pattern in biological, cognitive, and social systems. Ford & Singh do not provide any evidence to show that their dictionary

LINGUISTICS WITHOUT MORPHOPHONOLOGY

153

entries reflect the 'language potential'. Nor do they make any attempt to provide a counteranalysis to the judgment data which no one whose object of inquiry is the mental linguistic system will deny. I take it therefore, that their objections to (28) have no serious basis. If (28) is accepted, then the claim of autonomous phonology in (25) must be rejected as false. If (25) is false, the phonology-morphology interface that Projection Morphology yields must also be rejected. I would like to close my remarks by pointing to a phenomenon which I think cannot be analyzed in terms of any of the non-compositional morpho­ logies, including Amorphous Morphology. This has to do with bound roots. Since bound roots are not words, if there are distributional regularities that hold on bound roots but not on words, the hypothesis of non-compositionality of words must be rejected. In Malayalam, every verb must be inflected for tense. Thus, (31a-d) are possible words, but not (31e): (31) a. b. c. d. e.

"run" "open" "hear" oot -urn - um keelkk - urn future form oot keelkk present form oot-i keepast form oottam "running" keelwi "hearing" *oot *(l)

In compositional morphology, the forms in (31a-d) will be analyzed as being derived from the bound roots in (31e). I take it that the relatedness of the words in (31a-d) can be expressed in terms of WFR's without appealing to the notion bound root. What is interesting for our purposes are ways in which words and bound roots differ with respect to syllable structure. T. Mohanan (1989) argues for the following generalizations. (32)

Malayalam a. Only single sonorant consonants are permitted word finally. b. Root finally, however, sequences like /kk/, /11/, /rkk/, /lkk/, /mp/, etc. are permitted.  However, root final consonant sequences must obey the sonority hierarchy (= they can only be in the decreasing order of sonority).

Statement (32c) correctly allows 'future' forms of verbs like (33a), and disallows (33b):

154

TRUBEIZKOY'S ORPHAN

(33)

a. [parakkum] [wiyarkkum] "will fly" "will sweat" b. *[paraklum] *[wayapsum]

[keelkkum] [maantum] "will hear" "will scratch" *[kalapmum] *[paraprum]

Non-compositional morphology has no explanation for the unacceptability of (33b). 5.

Conclusions In sum, the evidence I have discussed in these remarks points to the following conclusions, most of which are not dramatically different from tradi­ tional views or from the conclusions I drew from my reading of Kiparksy's contribution to this rountable: (34)

a. Words can be compositional not only along the dimensions of phonology and semantics, but also along the dimension of syntactic constituency. b. The smallest units of words are morphemes, some of which cannot be words by themselves (affixes, bound roots).  A morpheme is associated with phonological, syntactic, & semantic representations. d. There exist morpho-phonological patterns which refer to word internal syntactic constituency.

"Même après le débrouillement il peut rester de la brume": Comments on Ford & Singh* Richard Janda University of Chicago After the initial metaphor — "Avec un peu de débrouillement l'aurore se laisse voir" — it is clear from the second part of Ford & Singh's title — "Quelques avantages d'une linguistique débarrassée de la morpho(pho)nologie" — that, of the three major positions represented at the roundtable discussions collected in the present volume, these two authors have staked out that negative pole which is anchored in morphology. In their view, morpho(pho)nology has no independent existence; it is instead part of morphology, and thus the so-called morphologically conditioned phonological rules of other analysts are for F&S simply parts of morphological strategies which relate entire words to one another (cf. also the earlier advocacy of this view in Ford & Singh 1983, Singh 1986). Once allowance is made for a shift of emphasis from (or within) morphology to the lexicon and its use, this stance is, grosso modo, the same as that taken by Bybee (1996, as well as in such prior works as Bybee 1985,1988). It is also more or less the perspective adopted implicitly by Anderson (1982, 1992, for example) and, again in slightly different forms, explicitly by Aronoff (1976, 1994, e.g.), Zwicky (1988, for instance), and the current author (cf. Janda 1982a,b, 1983a, 1984, 1987, 1994), along with a number of colleagues (see Hoeksema & Janda 1988, Janda & Joseph 1986, 1989, 1992a,b, Joseph & Janda 1988, Janda & Sandoval 1984).1 * For helpful comments and other aid in the preparation of this paper, I would like to express my gratitude to all the participants at the Table Ronde [sur la] Morphophonologie (Réponses contemporaines) held at the Université de Montréal from September 30 to October 2, 1994 — especially its organizer, the more-than-patient R. Singh — as well as to J. Auger, L. Dobrin, J. Goldsmith, J. Sadock, H. Suzuki, and B. Vance. Most translations from the French of F&S's paper are my own — but only to the extent that they do not reflect a related English version which the authors generously made available to me before the Round Table. 1 Because it treats the alleged 'morphologically conditioned phonological rules' of morpho(pho)nology or phonology as purely morphological processes, the general orientation common to these works is often called 'Process Morphology', although the dynamic, derivational implications of the first term are less than maximally appealing to Bybee, F&S,

156

TRUBETZKOY'S ORPHAN

This extensive yet still only partial convergence of views underscores the need for care in evaluating the overlapping but strikingly different claims of researchers who agree that morpho(pho)nology should be assimilated to mor­ phology proper but disagree on many other issues, since there is an under­ standable temptation for scholars to present their approaches as seamless alternative theories, cut from whole cloth, which must be accepted (or rejected) in their entirety. Nonetheless, it has been pointed out by McCawley (1982:1-3) that, in practice, virtually all theories (linguistic and otherwise) include a substantial number of isolable subparts that coexist loosely enough that they can be replaced or eliminated without an overall loss of plausibility or coher­ ence, and this is just as true of morphology in particular (cf. Janda 1983b) as it is of grammar in general (see Mufwene 1992). Thus, however dearly an analyst who denies the significance of a distinct morpho(pho)nology may cherish a particular combination of other theoretical claims within his or her more morphological alternative, there nearly always exists another, related, viable approach where a different status is accorded to one or more of those claims. We do well, then, to bear in mind that F & S's 'Projection Mor­ phology' is actually a set of proposals which do not all stand or fall together. In particular, Ford & Singh supplement their discussion of the advantages that follow from eliminating morpho(pho)phonology from grammar by pre­ senting entirely separate accounts of two issues which, though related, are furthermore arguably independent of each other. On the one hand, they devote several pages of their paper to downplaying the importance of morphemestructure conditions — especially the defense of MSC's offered by Mohanan (1996) — because any demonstration that the morpheme plays a significant role in morphology could be viewed as compromising F & S's claim that mor­ phology is 'radically amorphous'. On the other hand, the authors dedicate the single longest part of their article to giving an analysis of noun incorporation that is intended to replace Sadock's (1980, 1991) 'prelexical' treatment of this phenomenon in West Greenlandic, again because it would require recognizing significant internal structure within words. (Here, F & S attempt to avoid Sadock's (1995b) criticisms of Anderson (1992), whose theory they regard as 'seem[ing] to fail' as a result of its being 'non-radical[ly] amorphous'). Given that they lavish so many pages on the morphology vs. the syntax of incorporation, and so relatively few on morpho(pho)nology per se, one might even assume that Ford & Singh view their previous writings on the subject (and/or those of others) as having already provided ample justification for rejecting both morpho(pho)nology's independence and its inclusion in and other advocates of a more static, relational approach to word structure.

LINGUISTICS WITHOUT MORPHOPHONOLOGY

157

phonology. Indeed, this conclusion is almost inescapable, due to the fact that by far the greater proportion of their paper is concerned with discussing the amorphousness of morphology, rather than with demonstrating the morphologicality of so-called morpho(pho)nology or refuting counterevidence to this position. Yet morpho(pho)nology-as-morphology is now confronted with the putative threat and definite challenge of the counterarguments advanced by Dressier (e.g., 1977, 1985a, and 1996b), Kiparsky (e.g., 1982 and 1996), and Morin (e.g., 1994; cf. also Morin et al 1990). In the face of such op­ position, F&S's lengthy remarks on incorporation seem puzzlingly orthogonal. Considerations of this sort thus provide the justification for belaboring the point here that not all the parts — and hence not all the undefended points — of Ford & Singh's multipartite approach to morphology are shared by other antimorpho(pho)nological theories. Rather, F & S ' s take on morphology represents only one among several approaches that reject the notion of a separate or a phonological morpho(pho)nology, and so the validity of such a theoretical stance does not depend on, e.g., the success of their article's arguments concerning the contemporary MSC's of Mohanan or the incorporation analyses of Sadock (cf. also Baker 1988). The same holds for F & S's treatment of compounding {contra, e.g., that of Lieber 1992), which they view as one of the major innovative proposals in their paper but which again has little to do with (or without) morpho(pho)nology. That is, it is easy to imagine a morpho(pho)nology-free grammar (i.e., one with an expanded morphology) that views both roots (and stems) and their combinations (in compounds) as things while still regarding affixes only as the output of morphological addition-processes which are parallel to — and so may be combined with — processes of subtraction, replacement, and/or permutation (cf., e.g., Janda 1983b). The preceding remarks should not be taken to imply that Ford & Singh provide no arguments in support of the view that morpho(pho)nology is at most a non-significant subpart of morphology. The authors in fact allot several sections of their paper to such issues as their claim that the actual unity between so-called morpho(pho)nological operations and the affixes by which they are often alleged to be separately 'triggered' is shown by "(1) ...[their] nontransfer in contact and (2) ...[their] unattested 'generalization' in diachrony". Indeed, this non-generalizability is claimed to be 'a trivial consequence of nonseparability'. But we will see that certain historical data not considered by Ford & Singh apparently require them to retract or at least revise some of their specific diachronic predictions, while their assertions about second-language acquisition seem to need some fine-tuning. Since the relevant issues involved

158

TRUBEIZKOY'S ORPHAN

in the latter are somewhat clearer, let us consider them next. It is obviously impossible to give real instances of things that do not exist, but one wishes that F&S had provided at least some hypothetical but specific illustrations of the transfer phenomena whose impossibility under language contact is predicted by their model. Considering, for example, that Modern (New) High German umlaut is precisely one of those non-automatic alter­ nations whose status as morpho(pho)nological vs. morphological is still hotly debated (cf. Kiparsky 1996), we are led to ask whether the transferability of this process has ever been investigated by second-language researchers. Ad­ mittedly, there exists no well-known evidence that native speakers of German transfer umlaut into English (at least beyond residual cases where it still survives in English, like man/men — which germanophones presumably produce by learning, rather than transfer). But what of the possible transfer of NHG umlaut into Dutch or Swedish? Here, the presence of so many more additional typological similarities between a first and a second language (e.g., shared instances of verb-2nd word-order in syntax, of front rounded vowels in phonology, and of many formally less divergent suffixes in morphology) makes umlaut interference seem less implausible, at least to the present writer.2 Have such possible cases — where non-automatic alternations might be transferred to typologically close languages — been actively looked for? If not, it may be premature to speak in global terms of the non-transfer of such processes, even if they are indeed purely morphological. In this regard, it is important to emphasize one step in F&S's transferrelated argument — namely, their assumption that non-morpho(pho)nological morphology (e.g., run-of-the-mill affixes) likewise cannot be transferred in language-contact situations.3 From this, there can be constructed a particularly 2 I am emboldened in these suspicions by two reports in Lüssy (1974) (my translations). First (p. 12), "Professor St[efan] Sonderegger reports...that, among the Swiss 'colony' in Stockholm,...several Volvos...are called Völvö". This represents a remarkable imposition of NHG umlaut on a widely-known Swedish word, even if its scope is limited to a borrowing into German, rather than interference in Swedish. Second (p. 162n.8), the Swedish poet Per Atterbom (1790-1855), "after a stay in Switzerland, introduced the Swiss German suffix...[-li] into his [Swedish] poetry, in order to make up for the lack of a generally applicable derivational diminutive in Swedish" (thus coining, e.g.,fågeli "little bird"). Still, umlaut was apparently not borrowed along with this suffix. 3 F&S imply that the non-transfer of alleged morpho(pho)nological rules in language-contact situations would also follow directly from the non-separability of such alternations from accompanying affixes, but this ignores the fact that a putative morpho(pho)nological alternation is often the sole marker of a morphological category. For example, NHG umlaut is the only indicator of plurality in Vater/Väter "father''/"fathers", of past subjunctive in ich hatte/hätte "I had (indicative/subjunctive)", of a derivational relationship in saugen/säugen "to suck"/"to suckle", etc.

LINGUISTICS WITHOUT MORPHOPHONOLOGY

159

straightforward sort of syllogism: Garden-variety morphology fails to cause contact-based interference; therefore, since putative morpho(pho)nological rules are actually just an additional type of morphological process, their nontransfer (if verifiable) can be predicted by the single broad generalization that L1 morphology does not cause interference in an L2. Given this logic, it is crucial that affixes should indeed never (or perhaps at most rarely) be transferred.4 In support of this widely held view, F&S cite the findings of E. Moravcsik (1978) on the borrowing of bound morphemes, but these are flatly contradicted by the three cases discussed in Thomason & Kaufman (1989:1920, 349n.4), and there may well exist at least one further counterexample which involves transfer. I myself have observed anglophone learners of Spanish employing the English possessive-marker -'s/s' in intendedly Spanish phrases like Juan's casa "John's house", though in English the ending at issue attaches to the last element in a noun phrase and so can be analyzed as either an NP-final clitic or at most a phrasal affix. Still, we again require confirmation, not only that researchers have so far failed to find a certain phenomenon whose impossibility has been predicted, but also that they have actively looked for it in a representative sample of the right places. This is so because, if it should turn out that language contact can lead to the transfer of affixes but not of morpho(pho)nological processes, then we will have come full circle, being once more faced with the necessity of recognizing morpho(pho)nology as a distinct part of grammar. If we turn from the contact situation of transfer to that of borrowing, we are again reminded that affixes can certainly be borrowed from an L2 into an L1. If I may again report a personal observation, I once overheard a francophone Québécois of my acquaintance append the English suffix -ing to a French root in the course of a conversation with other native francophones; this com­ bination of placot- "chat" (cf. placoter "to chat") and -ing yielded placoting, where the foreign status of -ing was marked by the fact that its ...i... failed to trigger affrication of the preceding /t/. It is quite well known that such loanaffixes exist, but what about possible loan-morpho(pho)nology? Here, one instance springs to mind almost at once. To wit, so-called velar softening was clearly borrowed into English along with affixes like (the etymon of) -ity in loanwords from Norman French; during the late Middle English period, e.g., Chaucer used felicite(e) and Caxton both lubrik "slippery" and lubrycyte "slipperiness, lasciviousness". F&S might have further supported their paper's theses by discussing such phenomena, since they represent an additional 4 At the very least, this should not be the case for affixes that can occur with alleged morpho(pho)nological rules.

160

TRUBETZKOY'S ORPHAN

language-contact situation where an alleged morpho(pho)nological rule oper­ ates only in tandem with an accompanying affix. Before leaving this topic, it is perhaps worthwhile to consider briefly a final instance where language contact has consequences for morpho(pho)nology — which may be not only undergeneralized but also overgeneralized in secondlanguage acquisition. For example, anglophone learners of German have been observed to overextend umlaut, not only in situations where it is variable, but even to contexts where it is absent from native NHG speech. Thus, e.g., a learner may analogize not only from Bundlbünd-ig "bond"/"conclusive" to Buschl*biisch-ig "bush'"/"'bushy" (vs. native NHG busch-ig), but also from Bub/Büb-lein "lad"/"laddie" to *Büb-i (also) "laddie" (vs. native NHG Bubi), even though -i never triggers umlaut in the standard dialect to which the learner has been exposed (cf. also MutterlMutti "mother'V'mommy"). In such cases, we are dealing with the limited, flawed knowledge of a foreigner, and so they reveal little or nothing as to the status of umlaut in NHG grammar — i.e., whether it be phonological, morpho(pho)nological, or morphological. Unfortunately, similar reservations — or at least cautions — must be expressed regarding F & S's use of diachronic evidence in arguing for the morphologicality of alleged morpho(pho)nology. In a reprise of Ford & Singh (1983:67), the authors explicitly argue that diachrony provides no support for the claim that the 'phonological' part of a so-called morpho(pho)nological operation can be abstracted and generalized. That is, F & S deny that a morpho-(pho)nological rule can 'take over' additional morphological categories other than the one(s) in which it arose and with which it remains intimately connected. In this sense, the actual inseparability of a morpho(pho)nological operation and an allegedly triggering affix is said to be reflected in the fact that "...[such an] alternation... [can]not act independently in the historical evolution of a language... [or be] dynamic in the sense that it can wander from one morphological context to another". F & S thus claim that, e.g., "the mark of the imperfect or the morpheme -los could not [come to] trigger umlaut in German";5 they remain convinced that, "until proof to the contrary is offered, ...historical changes of this form are impossible and that a morphophonological process cannot be mobile in this sense". Incredibly, though, their very next claim (also reprised from Ford & Singh 1983:67) is that "a morphophonological alternation can spread without...the conditions that were present at its appearance", and the example that they cite in 5 F & S append to this statement a cryptic footnote (my translation) — "even if such a phenomenon could exist" — which is mystifying both in its import and in the reference of "such a phenomenon".

LINGUISTICS WITHOUT MORPHOPHONOLOGY

161

support of this prediction is the triggering of NHG umlaut by the diminutive suffix -chen even when it is added to new forms, as in the loanword-based diminutive Elefant-chen 'little elephant".6 It is apparently the rather limited nature of the spread suggested by instances of this sort that leads Ford & Singh not even to consider the possibility that (i) their two major diachronic predictions may be contradictory and (ii) it is precisely German umlaut which simultaneously confirms their second claim and unloads upon the first a virtual avalanche of proof to the contrary. Briefly put, it is precisely the loss of phonetic conditioning that is often (always?) involved in the morphologization of previously phonological rules which tends to make a morpho(pho)nological rule so readily subject to reinterpretation as being associated with another affix than that containing its original trigger — or even with a morphological feature that is otherwise not overtly marked. And, ironically, German umlaut con­ stitutes the paradigm case illustrating such phenomena, since in several morphological categories its former trigger lil has been not only reduced to schwa but also deleted outright, so that speakers have had little choice but to associate umlaut with either a new overt affix or a covert morphological feature. In fact, when German dialect data are brought in, it turns out that umlaut can be extended to new morphological contexts which seem no less surprising than F&S's example of -los as an 'impossible' new umlauttriggering suffix. The first mechanism by which German umlaut was indeed 'extracted' and allowed to 'wander' into new morphological contexts can perhaps be best understood by recalling the Charlie Chaplin movie in which, at a camp line-up, the commander asks for someone to volunteer: the Little Tramp appears to do 6 Earlier in their discussion of NHG -chen, F&S repeat from Ford & Singh (1983:67) two putatively German(ic) forms — Baum-kin > Bäumkin — regarding which greater precision might have been wished. The immediate ancestors of -chen are -chin and ultimately -chi:n, first attested mainly in Middle High German texts from the Middle German area; -chi:n was presumably also the OHG form, its ch- = /x/ having been shifted from earlier *-k- by the Second (OHG) Consonant Shift. The PreOHG etymon of -chi:n can thus indeed be reconstructed as *-ki:n (albeit with a long vowel), which also shows up early in Low German and is attested in even earlier Old Saxon skipiki:n "little ship". But the vocalism of the word for "tree" was o: in OS and then LG, and ou in mid-to-late OHG (and most MHG), so Baumlbaum can only be a Pre- or early OHG form — and recall that -ki:n can only be the former. Moreover, since ¿-umlaut was most likely not yet phonemic in PreOHG, and nonexistent (long) before that, the abovementioned change can most accurately be portrayed as something like the following: PrePreOHG *b[au]m-ki:n > PreOHG *b[äü]m-ki:n. A related problem concerns the dating for the origin of *-ki:n, which reflects a combination of two existing diminutive elements whose reflexes both appear separately in OHG (-k... and -i:n...) — one must be careful, that is, not to reconstruct a form for (or with) *-kin at a time before it was created.

162

TRUBETZKOY'S ORPHAN

so because, when everyone else in his line takes a step backwards, he fails to match his companions and hence stands out, all alone. This is directly parallel to the way in which NHG umlaut was made to look as if it had stepped forward, as it were, when many of the suffixes which had been first its phonological and then its morphological companions stepped back, so to speak, by undergoing first reduction in length and quality and then outright loss. In this way, umlaut has become the sole means of deriving certain words (cf. Note 3) — so that, e.g., the once primarily suffixal contrast between Old High German "ear" o:r-a vs. o:r-i = [ö:ri] "eye(let)" [as if "earlet"] has, via Middle High German o:r-e vs. œ:r-e (where e = [ə]), yielded to the exclusively root-vowel difference in suffixless NHG Ohr vs. Öhr. Similarly, for "apple", what originally was mainly a suffix-marked number-difference in OHG singular apful vs. epfil-i plural has now, via MHG apfel vs. epfel(-e), given way to NHG Apfel vs. Äpfel, in which umlaut is the only mark of plurality and so cannot be triggered by anything but the morphological category Plural (a.k.a. the feature [+plural]). Already in MHG, in fact, the frequent association of umlaut with noun-pluralization led to its introduction into the plural forms of words that had not earlier shown umlaut, like OHG sg. fater YS. fater(-a) pl.> late MHG vater vs. veter > NHG Vater vs. Väter. This process has gone even further in many of the regional (non-standard) dialects of southern Germany, Austria, and especially Switzerland, where a more extensive loss of final reduced vowels and syllables has eliminated a greater number of, e.g., former inflectional suffixes — whose morphological features (categories) are now marked only by umlaut. Thus, where more northerly varieties of Standard NHG have "day(s)" as sg. Tag vs. Tag-e pl. and "dog(s)" as Hund vs. Hund-e, many dialects of Swiss German have Tag vs. Täg and Hund vs. Hiind. In parallel fashion, the past indicative vs. past subjunctive of a so-called strong (= ablauting) verb like "take" is nahm vs. nähm-e in NHG, but na:m vs. nä:m in, e.g., many Swiss varieties of Alemannic Upper German (cf. Lüssy 1974:23-24).7 In cases like these, it may not strictly be true that umlaut has been 'extracted', since it was not 'removed' while an accompanying affix was still present, but that seems trivial in 7 This comes quite close to a development of German umlaut into 'the mark of the imperfect' that F & S predict to be impossible. Indeed, Lüssy (1974:23) mentions that the so-called weak imperfect-marker -t... and the past-subjunctive marker -i have coalesced into a single umlaut-associated suffix -ti in some Swiss German dialects, where "[if he] took" is thus nä:m-ti. Actually, because Swiss German as a whole has lost the imperfect indicative (which is replaced by the present perfect, a periphrastic construction), the imperfect (past) subjunctive is the only form of the imperfect found in any of these dialects — in many of which it is predominantly marked by umlaut alone.

LINGUISTICS WITHOUT MORPHOPHONOLOGY

163

comparison with the fact that umlaut has here indeed been isolated and hence would seem to have been 'abstracted' even in Ford & Singh's sense. And more nearly literal instances of 'extraction' with the opposite orientation can be cited from OHG, where — at a time when umlaut arguably was already morphologized (cf. Janda 1983a) — the umlaut which originally accompanied the dative/genitive sg. -in ending of so-called weak (n-stem) masculine nouns, as in nominative sg. namo vs. nem-in Dat./Gen. sg. "name", was given up in favor of unumlauted -in, etc. by the end of the 9th century. Quite apart from umlaut, similar situations with analogical leveling-out of former morpho(pho)nological alternations are legion, and it can hardly be denied that they show the separability of such alternations, even if they do not involve gener­ alization to new morphological contexts. Returning briefly to examples like the isolating of umlaut in NHG Öhr or dialectal Täg and nä:m, one can legitimately argue that these actually do involve generalization of the sort just mentioned, since they reflect a switch whereby umlaut ceases to be associated with an overt suffix and comes to be linked with a covert morphological category (or feature). Yet even clearer cases can be found without leaving the literature on German umlaut. For example, collective nouns (of neuter gender) could be formed in OHG by addition of an -i suffix — accompanied by umlaut — either with or without simultaneous prefixation of gi-lga-: cf., e.g., "land" lant vs. both ali-lent-i "totality of foreign (allo-) lands" and gi-lent-i "open ground(s)". But the collective -i at issue was in MHG weakened to -e, though it continued to cooccur with umlaut, and it lost its ability to appear without a prefix ge-. In NHG, however, the corresponding collective-formation has a major variant which involves only the prefix Ge-, but this affix is still accompanied by umlaut: cf., e.g., Ast ~ Ge-äst "branch" ~ "collection of branches", Bruder ~ Ge-brüder "brother" ~ "collection of brothers", Darm ~ Ge-därm "intestine" ~ "entrails". Thus, for the forms just listed, whereas it was once clearly a suffix -e (< -i) which showed a link with the umlaut that also marked such collectives, it is now a prefix Ge- which shows such an association. Even more strikingly, Lüssy (1974:163-164) reports that, in at least one variety of Swiss German, the combination of umlaut with an -i suffix that forms neuter diminutives (mainly from masculine nouns) has now been matched with the creation of an -a suffix that forms feminine diminutives and is also accompanied by umlaut. Compare, e.g., "Francis" Franz vs. Fränz-a "little Frances" (plus Fränz-i "little Francis") or "hare" Has vs. Häs-a "little gray (hare-colored) female goat" (plus Häs-i "little gray male goat"). Here, we see umlaut becoming newly associated with a suffix containing an a-vowel —

164

TRUBETZKOY' S ORPHAN

a situation very close to the -los case specifically predicted to be impossible by Ford & Singh, Given such counterevidence to the assertion that morpho(pho)nological alternations cannot generalize so as to become associated with new morphological contexts, two main responses are available. First, it is possible to reformulate F & S's claim so that it prohibits only the extension of a non-automatic alternation like umlaut to a morphological context that lacks any semantic or lexical connection with an existing one. This clearly — and correctly — predicts the spread of umlaut both to Swiss German feminine diminutives in -a (where there was a semantic connection with umlauting neuter diminutives in -i) and to NHG collectives in Ge- (where there was not only a semantic connection with earlier umlauting collectives in Ge-...e, but also a lexical connection between the two once-simultaneous affixes -— or parts of one circumfix). Nonetheless, this is tantamount to a principle preventing one or more speakers from engaging in agitprop of the sort "Hey, guys, let's all introduce this neat randomly-chosen morpho(pho)nological alternation into a completely arbitrary morphological environment with which it has absolutely no connection!". It is not clear that any such specific principle is needed to explain language change, since it seems to follow from general, nonlinguistic factors governing how innovations in soci(et)al conventions arise and spread. Hence the more crucial response to cases like the generalization(s) of German umlaut documented above is, I believe, to recognize that it actually supports Ford & Singh's main thesis: that morpho(pho)nological alternations are not independent and not phonological, but simply a (generally) nonadditive kind of morphology. If we take seriously the vie.v that non-automatic alternations like umlaut are not some kind of affix-accompanying trash, but instead important morphological markers in their own right, with the potential to become the sole indicators of major morphological categories, then we should in fact expect to find such alternations behaving like affixes — which themselves show the same diachronic tendencies to become reanalyzed in their functions, to coalesce (wholly or partially) with other affixes, and the like. For example, the -er ending that is now one of the most consistent markers of plurality in NHG (and is invariantly associated with umlaut) descends from an originally rather empty (non-plural) Indo-European and Germanic stemformant (primarily *-iz-) which came to be reinterpreted as a plural-marker when the phonetically more substantial case/number suffixes following it in the plural protected it from phonological erosion there, but not in the singular. This is exactly parallel to cases like the pluralization of umlaut via loss of a former plural-suffix in examples like German Äpfel.

LINGUISTICS WITHOUT MORPHOPHONOLOGY

165

To sum up, then: while it remains to be seen whether F&S's promising predictions about the non-transfer of alleged morpho(pho)nological alternations in language contact are as accurate as those authors claim them to be, it appears that their predictions about language change are false — but in a way which, backhandedly, turns out to support even more strongly the anti-morpho(pho)nological position that F&S share with many other linguists (including this writer). One is left wishing, however, that F&S had buttressed their diachronic remarks with at least some purely synchronic justification (perhaps even some internal evidence) for the rejection of an independent or phono­ logical morpho(pho)nology, rather than compounding this gap by incorpo­ rating into their paper so much discussion of the more morphosyntactic phe­ nomena of incorporation and compounding. After all, while the latter two spheres are directly relevant to the question of how amorphous the lexical representations of words may be, they are of only marginal interest for the debate regarding the possible existence and probable grammatical locus of socalled morpho(pho)nology. We can thus be grateful to Ford & Singh for clearing away, and thereby letting us see through, some of the fog that often hides (what many of us fervently believe to be) the true, dawn-lit, morphological nature of non-automatic alternations like German umlaut, even if a bit of hazy mist remains. But precisely the history of NHG umlaut in its various manifestations also reveals to us that such alternations are essentially morphological in their ability to be functionally transformed and even to 'wander' from existing contexts to new ones. Still, we can rejoin Ford and Singh in concluding that, when so-called morpho(pho)nology travels, it does not, like the final subject in the poem Aurore by (Paul) Armand Silvestre (1837-1901; cf. the text-setting in Fauré 1897/1990:143-146), do so by "cherchant... des routes inconnues".

Reply to Mohanan and Janda Alan Ford & Rajendra Singh It is a pleasure to acknowledge the opportunity Janda and Mohanan provide us with for clarifying certain things and for addressing questions they believe we did not (sufficiently) address in our paper. As they were both provided with a brief résumé of our position paper, perhaps the best place for us to begin is with that résumé. We reproduce it below: 1. Our radically a-morphous morphology subsumes so-called morphonology under the rule schema [X] a ↔ [X]ß. When a non-automatic alternation is the sole mark of the morphology, it is explicitly treated as such (umlaut, ablaut, guna etc). When it co-appears with an 'affix', it is also explicitly treated as part of a unified morphological operation from which the allegedly 'phonological' part cannot be abstracted away or set up to operate 'in tandem' with the allegedly 'affixal' part (English Trisyllabic Laxing, Velar Softening etc.). The fact that non-radical amorphous theories of morphology seem to fail seems irrelevant to us. 2. 'Morphemes', to the extent that one can speak of them, are processes (and not 'things' with autonomous lexical status), and complex operations involving non-automatic alternations are complex but unitary operations. 3. The non-separability of the allegedly 'phonological part' of a complex morphological operation — an operation in which an 'affix' is tied with a non-automatic alternation — predicts (1) its non-transfer in contact and (2) its unattested 'generalization' in diachrony. Nongeneralizability is, as a matter of fact, a consequence of nonseparability. 4. The basic 'no internal structure' hypothesis of our theory is, in our view, not threatened either by 'MSC's' or by 'incorporation' as some seem to think. The former represent dominant cannonicality but have no constitutive import. The latter has gained currency only because advocates of

LINGUISTICS WITHOUT MORPHOPHONOLOGY

167

'incorporation' pass in silence over the fact that the 'incorporated word' does not generally have the same form as the 'source' word. Besides, the 'incorporation' claim has been withdrawn for 'nonconversive suffixes' even by the advocates of 'incorporation'. As far as 'compounds' are concerned, the hypothesis that one of their 'constituents' is really a constant has not been seriously considered by advocates of 'composition' as a different kind of morphology. The fact that the constant in a 'compound' is isomorphous with a word does not constitute an argument for 'internal structure' anyway. 5. We see no reason to give up the traditional beliefs that words have no internal structure and that the allegedly 'phonological part' of a complex morphological operation is an integral part of it. 6. It is perhaps useful to point out that the adoption of our view of 'morphonology' leads to a simpler and more elegant phonology (of the type proposed in Singh 1987 and 1991). We are grateful to Mohanan for acknowledging that "almost everything that can be formulated as a morphophonemic rule in the phonological module by accessing morphological structure can also be formulated as part of a WordFormation Rule in the morphological module" and for correctly emphasizing the fact that we reject only syntactic compositionality of words. He believes, however, that our view does not provide a proper account of distributional patterns that obtain in 'proper subparts of words'. Since we take up these objections, also raised in Mohanan 1995, in our paper there is perhaps no need to respond to them once again. We must, however, point out that we are surprised at being accused of using "the methodology of corpus linguistics". We quite simply made up things like American Continental Shipping and Transport Services and Universal Turbine Throttle Cleaning Service and the fact that we could make them up is indeed a testimony not only for the non-constitutive import of MSC's but also for 'language potential'. We are also obliged to point out that Mohanan's devotion to his (28a) leads him to the postulation of a morph /z/ without the feature 'Plural' or 'Present Tense'. Notice that this is the only option he has because exceptions that can be multiplied ad infinitum cannot possibly be considered as exceptions. As we provide reasons for not operating with units like 'roots' and 'stems' in Ford & Singh (1985) — though our variable X does sometimes correspond to these Pāninian constructs — we feel it is not necessary for us to refute here (in this text) the 'root-based' analysis T. Mohanan (1989) provides for Malayalam. We must also point out that although we have no objection to Mohanan

168

TRUBETZKOY'S ORPHAN

including his version of how we might do semantics in his comments, we don't think we would do it that way. We are aware of the fact that we do not provide explicit interpretive rules that are constitutive parts of our wordformation strategies, but their absence does not stem from the fact that we ignore what he calls 'morpho-semantics'. As for 'morphophonology', we con­ centrate on it in order to show that it may actually not be there at all. It also seems to us that there is a bit of 'systematic ambiguity' in Mohanan's use of 'rules' and 'information'. It is, for example, not clear to us why only what he calls 'Procedural Non-compositional Morphology' can be said to be consistent with a theory that allows the exploitation of intermodular information. It is possible to argue, for example, that WFS's that are sensitive to constituents such as the syllable or the rhyme guarantee the sort of access he speaks of. The only arguments for such constituents may, in other words, turn out to be morphological (for a demonstration that the other types of arguments don't hold, see Diubalska-Koiaczyk 1995). Even these may not hold, but that is not the point. Janda is quite correct in suggesting that in our paper we assume that 'morpho(pho)nology' can neither be given a separate status nor included in phonology. Although we could not have responded to the papers by Dressier, Kiparsky, and Morin at his roundtable — we did not hear or see them till much after writing our paper — we do respond to Dressier (1985) and Morin, Langlois & Varin (1990) in Singh (1986) and Singh (1996), respectively, papers Janda had, we believe, seen. The reason we decided to concentrate on 'incorporation' was that, until Morin (1994) made his case, it seemed to us to be the most compelling argument against any a-morphous morphology, particularly the sort in which we integrate 'morpho(pho)nology'. As for his reluctance to give up on 'roots' and 'stems' as 'things', we can only invite him to look at Ford & Singh (1985). As for compounds, all we can do here is to restate our hypothesis that a 'compound' is a concatenation not of two vari­ ables, but of a constant, which, in Pāninian terms, is quite like an 'affix', and a variable — the substrings said to make it up do so precisely by loosing their individual wordhoods. It is perhaps useful to add that even Pānini treats compounds together with (other) taddhita derivatives and that the rejection of his atomistic ontology, which reaches its definitive statement in Bhartrhari, pobably began when even Patanjali felt compelled to protest against the denial of nityatva to compounds. Looking for words or variables inside a compound may be a bit like looking for the father and the mother in the child! His point regarding depackaging is, however, well taken, and should it turn out to be necessary for us to give up this or that part of our theory, we won't resist

LINGUISTICS WITHOUT MORPHOPHONOLOGY

169

beyond the point up to which we might actually be expected to! Janda wants us to clarify our lack of transfer hypothesis, developed rather extensively in Singh (1991) Singh & Ford (1987), and Singh & Martohardjono (1988). It is that whereas phonological 'rules' colour interlanguage, morphological 'rules', including, of course, 'morpho(pho)-nological rules', do not. Wode's (1978) classical study of German Inter-English, which shows that German umlauting plays no role in its formation, is just one of the many studies we cite in these papers. The Dutch cases of the type he seems to have in mind are examined in some detail in these papers (cf. Singh & Ford 1987:171, in particular for the point at issue). Actually Janda himself provides part of the clue as to why it seems "less implausible" in cases like that — "less divergent suffixes in morphology" (italics ours). We are tempted to ask him if our analyses of such cases were actually 'actively looked for'? As for the second step in our argument, it is broached in Singh & Martohardjono (1988) and ex­ tensively documented and discussed in Singh & Parkinson (in press), though Janda couldn't, of course, be expected to be familiar with the latter. What he calls the 'transfer of affixes' in the cases we are familiar with is in fact the transfer of lexical items from which the morphology is extracted anew by the host language. Be that as it may. We know of no compelling evidence, vis-àvis transfer, to distinguish between 'affixes' and 'morpho(pho)nogical pro­ cesses'. We don't think we are back to square one. It is just that Janda seems not to have seen alternative analyses of cases of the type he correctly invites us to consider, probably our fault for not making them available to him. We don't, in other words, think it is premature to speak in global terms, partic­ ularly when some of the putative cases are analyzed by Janda himself as supporting our thesis. We are, however, genuinely puzzled by his 'final instance' of 'morpho(pho)nology' — the adaptive rule (cf. Erdman 1973 and Wurzel 1977, discussed in Singh & Ford 1987) of umlauting which he says characterizes Anglophone Inter-German. We agree that what is involved is overgener­ alization for that is what is meant by an adaptive mie, but it belongs to L 2 morphology. L2 morphology, as is well known, is routinely overgeneralized by learners, a phenomenon that materially supports language teachers, who generally do not make a distinction between possible words and 'non-attested' words, all over the world. Actually, the all pervasive overgeneralization of L2 morphology leads Singh and Parkinson to go as far as to claim that all socalled morphological errors are licensed by L2 morphology. We, therefore, humbly invite Janda to reconsider his interpretation of anglophone overgeneralization of German umlauting.

170

TRUBETZKOY' S ORPHAN

Given that the conditions whose erosion we speak of in Ford & Singh (1983) are explicitly identified as 'phonological conditions' in the omitted part, covered by ellipsis, of Janda's citation from Ford & Singh (1983) and that we explicitly allow for lexical generalization within the same morphological category even up to loan-words, we could quibble with Janda's somewhat unorthodox exposition of our views but we won't because in the final analysis it fully supports, as he himself points out, our 1983 claims and, in addition, takes us through an empirically rich data-mine. The dust, we concede, hasn't quite setted yet, but, we remain confident, it will, and it will do so along the patterns we admire as much as Janda does. We hope we have been able to offer some reasons and hints that would help lift the mist from the valley of contact, the place that understandably invited him to pause a little. The fact that, during discussions, he spontaneously provided an analysis of the Twente case presented by Dressier compatible with our own views seems to suggest that he won't stop there long.

Linguistics without Morphophonology Discussion FORD: One of the points that Janda made was that our theory suggests that there is no internal structure at all in words. But of course there is some phonological structure in words and what we are suggesting is that everything can be said in terms of that structure. Whatever we need to refer to in a morphological operation can be said in terms of that phonological structure and morphological category. JANDA: Are you saying though that what's stored in the lexicon is unanalyzed words and then these rules come and analyze it or are the words stored some­ how with dotted Hnes or whatever? FORD: Really, another aspect that you corroborated here is that I am not quite sure that words are stored in the lexicon at all. I think that, as has been suggested by Dr. Bybee here, there are relationships between words. We are to see this more in terms of a network than an actual place where things are stored, so that the sorts of relationship that exist between words will only make reference, as far as morphology is concerned, to phonological structure of words. JANDA: I would see that if there are lines going from one part of the word to another word and lines from the remaining part of the word to another word, that does give a kind of internal structure, because of the association differ­ entially with other words, but maybe that's just terminological. FORD: I was interested by what you were saying about the question of redun­ dancy with regard to predictions that could be made. Obviously, the kinds of redundancies that we do find between morphology and syntax, at least, do make different predictions, so there is no problem about what we are saying about there being a redundancy. The clear places I think might be the English comparative with a syntactic comparative and morphological ways of making comparatives, and they seem to be conflicting at certain stages of acquisition of language, when for example children haven't caught on to the fact that the -er suffix is limited in the number of syllables that can be added to it. JANDA: I think double marking does happen and that is evidence for your

172

TRUBETZKOY'S ORPHAN

theory. When you get the most unkind cut of all, that kind of example has been around for a very long time and it is very common now to have more -er forms. I don't, however, believe you can have both incorporations in a syntactic way and a morphological way because one is supposed to lead to the other and you should take away the one and put it there. To the extent that that's been investigated, this must be a source of evidence, if in Greenlandic there is doubling. FORD: One thing I didn't think necessary to make clear is that there really is a distinction that can be made between morphological transfers and the word. That words are never morphological in content although there is a stage in the evolution of the language where words become potentially morphological processes within that word. But once they become part of morphological processes, and are involved in morphological operations, they lose properties that they had as a word. It's clear in the cases that I was quoting from the Greenlandic, for example, but even in English words lose some of their morphological or even syntactic properties by becoming only constants. SINGH: Maybe two or three brief comments. One, your comment about morpheme structure conditions that there might be specific classes of morphemes where these things hold. You see, the claim that we are making is — we, obviously, can't show that any morpheme structure condition that has ever been postulated is not going to hold. I mean that's an empirically impossible task, the claim that we are making is that any putative case of morpheme structure condition that cannot be restated either as a syllable structure condition or as word structure condition, as Professor Kiparsky has shown for phonological reasons, can have no constitutive import in the language, it can have only a statistical import. It might even be the dominant canonical pattern of that language, but it will have no binding force on the speakers and can be violated on any morning, of any day, of any week, should one choose to do violate it. The second comment, I think, concerns the internal structure hypothesis. The strongest claim is the one that the word is a sort of a mirror image of the incorporation story. In the incorporation story, what shows up as allegedly incorporated noun actually doesn't look like the noun at all, something that Baker, Sadock and everybody else pass over in silence. The compositional case is the mirror image, it looks like a word but it isn't because it has lost some of the properties that it would otherwise have were it a word. And the third clarification I like to offer is about the claim about the diachronic generalizeability: in order to appro­ priately generalize, one has to be able to tease it apart. It isn't that morphophonological processes in general don't generalize. I don't think that's what

LINGUISTICS WITHOUT MORPHOPHONOLOGY

173

we are saying, they clearly generalize lexically, for example. What we are saying is that if the operation contains these two parts, then it seems that you can't just take one part out and then jump over to some other category. It may be empirically false but I just wanted to clarify the claim that we are making. Perhaps you have something to add, Clem. FORD: I suppose that with respect to what you've just said, the example from the umlaut was an interesting one, although we should point out that the umlaut in itself is a morphological strategy. JANDA: Right. FORD: So that from that point of view it is really tough to analyze. SINGH: That's right, so that I don't think your examples are really counter­ examples. We are not saying that there aren't any, I'm simply saying some of the ones that you mentioned really are not counter-examples to the sepa­ rability hypothesis. JANDA: What you mean by separability seems to be that you take it from somewhere and either it's used by itself or it's used with something else. SINGH: If it's used with something else, first you take it out and then you make it jump the category. JANDA: There is no intermediate stage, I mean, it is with something and then you put it with something else, you don't see it in transit and it seems that it goes with the other affixes and in that sense it's generalized to new affixes. But it would have to be a very difficult story to get, that umlaut wouldn't start appearing suddenly with -los and -um or whatever and we talked about this a little bit yesterday. You can imagine a couple of steps intermediarily which then give the impression that it has gone directly to that. With the next steps you can do anything. I would just add that in diachrony it seems to me sometimes we talk about it, as if language change were a volley-ball game, and we play synchronically, stop, and everyone rotates and then we start playing again. But speakers can't stop speaking for this to happen. So when we really have transmission across generations that's different I think from talking about the conditions that we might otherwise want to expect. It could be the hyper-corrections, the various kinds of misunderstandings, or the data are different. So, acquisition has to be divided from other kinds of creations in morphology. Could I ask just about the schema term, and the part on the other, is that a division where you can have QX and Xnik that you really have something on both sides? Number (8) and number (12) both have not just an X, but something else also. It seems to me that kind of recognizes internal structure if you are separating that part of one word and relating it to a word with

174

TRUBETZKOY'S ORPHAN

another part separated out. Or is it, you don't see any problem with that? It's in the same axis. It's the same variable on both sides. JANDA: You're saying the Q of nik is the same as the Q of X and the P of Y, but there are different things, it's [a] in one and [puk] in the other one, in (12), for example. FORD: I thought that this answers your question. There is a form of ambiguity in the use of the word 'variable', so that we are suggesting that morpho­ logical strategies are limited to one variable and it's of the X, Y category which has the semantic interpretation of anything, whereas the other sorts of variables are restricted variables to the extent that they refer either to consonants or vowels or to specific consonants or specific vowels or specific subclasses of consonants or vowels. Of those, you can have as many as you like. DRESSLER: First to this umlaut, I think Janda's most striking examples are simply fossils, for example behende is not even written with etymological umlaut. I think Janda is quite right in pointing to the facts that you have these examples where umlaut was first triggered by suffix, then the suffix was lost and then was triggered by the prefix and I found similar examples in Albanian where the first morphonological palatalization was triggered by a following palatal vowel, the palatal vowel was lost and then it was triggered by a preceding palatal vowel. Another question is bidirectionality of mor­ phological rules which is very important also for morphonology. There is a lot of evidence that there is no equal direction, i.e., clear evidence for a basic form and for derived forms, both in competence and in performance that morphological rules which so to say go from left to right, are different from back formation which goes from right to left. These are entirely different things and they are directional. And I think there is also good psycholinguistic evidence, and since I think your model like Joan Bybee's and my own rather assume that there is a close parallel between competence and performance, But if you assume that, then you can observe what is hap­ pening in performance. One spectacular example that my student Petra had observed in her field-work on a Gipsy language had to do with inchoative formation: inchoative formation is based on the plural of nouns and when she submitted, in testing, nouns in whatever form in the syntactic environment, the speakers invariably said "what do I do?" and made the plural of the noun and then formed the inchoative verb. It's quite clear that it shows that there are intermediate steps and thus directionality. And finally productivity: I quite agree that productivity has extra-linguistic bases, but establishing extralinguistic basis is not equivalent with reducing something linguistic to SINGH:

LINGUISTICS WITHOUT MORPHOPHONOLOGY

175

something extra-linguistic. This is a false view that has also been attributed mainly by Anderson to Natural Phonology and Morphology. But really seeing that productivity has extra-linguistic bases doesn't mean that there are linguistic generalizations to make about them and I think similar to Joan Bybee, I think that it has to be stipulated as a linguistic factor, there are certain rules which are productive. MORIN, Y-C: Do you want to answer? SINGH: 'Answer' is maybe not the right word if we want to talk about it. I think what we are saying about productivity is not that it isn't important or that it doesn't follow from certain reasonably well-understood properties of language systems but only that at the level of the representation of grammar there is no need to make that distinction at that level. Whether you want to talk about it in terms of rules or in terms of processes or in terms of strategies, abstracting way from that, the only claim here is that in the grammar itself, assuming that such positivist entities can and do exist, if we do postulate a grammar of some sort, then, within that grammar, one doesn't have to make that distinction, not that it is to be reduced to one or two simple factors, there are very complex parameters that enter into the definition of precisely what will or will not turn out to be productive, but the rules themselves are not sensitive, can't be sensitive, to that sort of information. And that's why representational decisions about representing morphological processes by one kind of rule either in terms of modularity or in terms of different representational system, that we reject, that you can't distinguish this at the level of rules. The second comment that I'd like to make, given the German umlaut example that you mentioned, it seems to us that the cases of teething the process, allegedly phonological part of the complex morphophonological operations are still relatively very rare if not, perhaps not non­ existent, but that kind of generalization and perhaps some explanation is needed for that fact if ours is not the appropriate one, perhaps another one is needed, but that remains a fact, that there aren't cases, too many cases of that sort where you take the allegedly phonological part of the morphophonogical operation and do with it what you suggest is there in this particular case. Perhaps we need to weaken what we have to say but this is put forward as a point de départ. TIFFOU: Oui, moi je vous amène sur l'autre versant, le versant de la syntaxe, mais j'ai un problème qui met en cause peut-être deux aspects de la théorie de Raj et de Clem; je voudrais savoir comment ils intègrent l'exemple du basque, que je vais donner et qui intéresse la délimitation de la morphologie. Dans une phrase comme eman dizkigutenebila koarentzat diot "je l'ai dit

176

TRUBETZKOY'S ORPHAN

pour faire plaisir à ceux qui sont avec ceux qui nous les ont donnés" nous avons un auxiliaire qui se présente sous la forme d'un composé. C'est donc de la morphologie puisque nous avons affaire à une suite d'affixations. D'autre part, je me demande si c'est vraiment de la morphologie dans la mesure où lorsque j'ai des affixations et que j'ai un ensemble morpho­ logique, je n'ai pas d'explicitation des relations et je les tire, disons, de l'interprétation que j'ai du mot. Par exemple, si je dis armiger en latin, c'est "porteur d'armes", mais le rapport entre armi- d'un côté et -ger de l'autre n'est pas évident; je le tire spontanément, alors qu'il n'y a pas de marqueurs. Je ne peux pas dire en effet *arma ger qui n'est plus un mot composé, tandis que, dans des séquences comme celles du basque, on a effectivement, je sais que vous n'aimez pas ce terme mais je l'emploie pour mon exposé, des 'morphèmes flexionnels' qui explicitent les relations à l'intérieur même des affixations. Nous sommes au coeur du problème qui intéresse la délimitation de la morphologie. Où sommes-nous? Lorsque j'ai des mots comme cela, je n'ai pas l'impression d'avoir un mot tout fait, d'avoir puisé dans mon vocabulaire; là je fabrique à tout moment ces séquences qui présentent la variété d'une phrase syntaxique; en d'autres termes, je dirais que lorsque je vais puiser un mot dans le vocabulaire, c'est un mot global, qui est tout fait, qui est tout configuré, dans la syntaxe, en revanche, je dirais qu'il y a un processus de type analytique où je mets en place mes éléments du discours. Or j'ai l'impression que, lorsque je fais des formes nominales ou verbalesje vais puiser, si je cherche un mot, dans mon stock de paradigmes, mais, quand je contrôle ma syntaxe, j'anticipe et je procède de façon un peu linéaire. Or j'ai l'impression que dans le type d'affixation du basque, c'est de façon linéaire que je procède et je ne fais pas référence au stock. Comment traitez-vous alors ce type d'exemple pour l'intégrer dans la morphologie? Sinon est-on obligé d'admettre qu'il y a un zone floue et qu'il y a des problèmes? Un exemple plus simple: on demande un jour à un petit garçon "avec qui tu joues-tu?", celui-ci répondit: bonetarekilakoarekin "avec le fils de celui avec le béret sur la tête." Peut-on considérer cette expression comme une unité purement morphologique? FORD: Je n'ai pas trouvé de mots de ce type là en basque, donc je ne suis pas en mesure de le dériver, en faisant l'historique en quelque sorte du mot, c'est-à-dire en énumérant dans un ordre déterminé les stratégies de formation de mots qui s'appliquent à ces mots-là. Tous ces mots sont décomposables en sous-mots mais, en même temps, ces sous-mots jouent le rôle de ce qu'on a appelé une constante tout à l'heure au niveau de la construction, une constante morphologique dans une opération. Un mot basque, autant que

LINGUISTICS WITHOUT MORPHOPHONOLOGY

177

n'importe quel mot dans n'importe quelle langue est un ensemble de procédés qui se sont réalisés pour analyser le mot. Jamais des constantes qu'on va trouver à l'intérieur de ce type de mot en basque ne seront identi­ fiées comme des mots ayant une fonction sémantique quelconque. Exacte­ ment la même situation que  inuit. TIFFOU: Oui, mais est-ce qu'en inuit tu as des marqueurs de cas dans la composition? FORD: Mais bien sûr. GROSU: I would like to ask a question about the general philosophy behind your programme, and maybe my puzzlement is due to some misunder­ standing since I arrived late but my impression is that you are opting for a lack of differentiation between certain differences which have been recog­ nized by others, like, for instance, separating morphophonology from the rest of morphology or separating perhaps inflectional from derivational theory. You're arguing for this, I heard on the basis that in languages in contact, for instance, there is no reason to recognize this differenciation. I think that your evidence has to be taken at face-value, in fact I would say that from my own perspective as a syntactician, in current syntax, there is probably no interaction between the way in which certain syntactic processes such as agreement or case marking or whatever are morphologically realized within current models. One wouldn't care within the syntax proper whether they are realized as an instance where the past form is marked by some ablaut in English or by regular affixation or whatever. I don't understand how the demonstration that certain differences within morphology are not needed for certain purposes shows that they don't exist, and also how the lack of separability is used for that conclusion. For instance, if one were to study, let's say, the human body, I suppose the auditory system or the visual system, they never occur in isolation from the body, I don't think that's an argument for not studying them in this integrated system with constellations of properties. So, for instance, as we heard last night from Kiparsky's pre­ sentation, that morphophonological rules and allomorphy rules or morpho­ lexical rules or whatever are characterized by a good constellation of prop­ erties, then even if they don't occur in isolation, I would say that this is a valid differentiation which ought to be recognized, and I don't think that recognizing certain distinctions prevents the overall theory from being unified but, of course, unification that doesn't necessarily take the form of complete levelling. SINGH: Maybe we can both respond to part of it. First, I thank you for making it clear that you did in fact arrive late because the argument was that for

178

TRUBETZKOY' S ORPHAN

internal reasons, we need to do this: we need to present all morphological operations as unified entities. The evidence offered either from contact or from diachrony was a bit of icing on the cake, but that's not what the cake was made of. The second point that I'd like to make is that the idea is not to say that these things are not to be studied but simply to say that there are some properties, that is a cluster of properties, to use your metaphor, that seems to obtain where so-called morphophonology is involved. It seems to operate at a par with what would otherwise be classified as a morphological entity within your metaphor, so there is another cluster of properties, the insistance is not that these things are or not to be studied, the insistance is on what are the reasons for which one wants to say that so-called morphophonological processes are morphophonological or lexical phonological opera­ tions do not belong to morphology. The claim we are making is that they do. Perhaps we are wrong, perhaps there are reasons, but the kinds of considera­ tions you are bringing to bear on what we have to say don't seem to me to be central in any way that I can see. FORD: If I could just reply briefly to that. I think that the basic difference between what we have been trying to say and the point of view that we have been trying to put into question is that we are seeing in a quantum universe what was a particle universe, at a different stage of linguistics and so that we are interested in describing language in terms of the processes that are taking place and not so much in terms of the elements that we are capable of identifying within different subsections. So that kind of fragmentation which goes into saying "this is a morphophoneme" and "this is a morpheme" and "this is a phoneme" has been a bit disastrous for linguistics in the past and that, a way of, at least seeing if we can make a better form of description is to look at the processes that each one of these particles that have been identified in the past are involved in and focussing on that sort of description. MORIN, Y-C: There are many points that I would like to come back to later in the workshop. For now, I have one specific question about the way you conceive compounding as in the example (21). What I don't understand is how the schemas or strategies (I think they are the same things) you propose can account for the speaker's internalized rules. On the model of ouvre-boîte, your schemas predict that one can produce ouvre-patate, ouvre-verre, ouvrechaussette and whatever, but not magane-patate unless there already existed some compounds in the language, with the constant magane. So how do you explain that people can nonetheless do that? FORD: That people can make magane-patatel MORIN, Y-C: Yes. You can take any transitive verb and compound it with any

LINGUISTICS WITHOUT MORPHOPHONOLOGY

179

noun even though there is no prior model with that verb. FORD: I suppose that that sort of statement is the sort of thing that we are putting into question. MORIN, Y-C: I'll put it into a historical perspective. Ouvre-boîte is now attested and we know that it was not in the nineteenth century. Actually no compound beginning with the verb ouvre was attested in the nineteenth cen­ tury, so at one specific time, one person, at least, invented one compound in ouvre-X. What was his model? MEL'CUK: He was crazy. MORIN, Y-C: One must admit that compounds such as tire-bouchon with a different initial verb were used as models to create ouvre-boîte, but this is excluded in your model, as I understand it. FORD: You're quiteright,that's excluded. SINGH: Just two brief comments, one as Professeur Morin has just pointed out, it isn't whether it is attested or not, I think there seems to be a general consensus, unless I'm interpreting it more charitably than I should, that morphology is really responsible for possible words rather than either attested or actually seen words. Given that consensus, what we are saying is not necessarily based on the fact whether it is attested or not attested. As to the second question that you raised, what was the model for the first person who did something for which there wasn't a pattern, that's a question I have no answer for. I'm grateful for it, we need to study how a particular unique act of creativity which might start for various kinds of social historical pragmatic reasons, how it is related to the systematic creativity potentiality made available by a lexicon and then later exploited. MORIN, Y-C: I used the historical evidence as extra-evidence, but contrary to what was implied by Professor Ford's answer, this pattern is very pro­ ductive in fact. This is not a free assertion, you can make any new compound you want with a transitive verb that has never been used as a compound. It's a common observation. MEL'CUK: What is the question again? MORIN, Y-C: Just a while ago, I created the word un magane-chaussure. It's clear to every French speaker what un magane-chaussure means provided that maganer and chaussure belong to his internalized lexicon. It has never been used before, and I know of no other compound with magane in French. So how did I make it? MEL'CUK: You have a rule to make such a compound. MORIN, Y-C: You are right. MEL'CUK: You've just told me the truth, you take any verb, you add a noun

180

TRUBETZKOY' S ORPHAN

and you have it. So there is a rule and that's it, where is the discussion there? MORIN, Y-C: Obviously not with you! MEL'CUK: I didn't understand the question, I don't understand the problem and I don't understand the answer either. KIPARSKY: I want to focus on a question that arises from the proposal that you made, essentially to build in morphophonological processes into the morpho­ logy. The decomposition of these alternations into interacting phonological processes sheds light on their properties. I'll give two examples. The stress alternation in expériment/expérimént is this ostensibly different and perhaps would be for you formally from the alternation in áttribute/attribute. From the perspective of a decomposition into phonological processes, it's really the same phenomenon: in verbs the final syllable counts, in nouns it doesn't. We have here the phenomenon of extrametricality, plus the general rule of stress assignment that is common to nouns and verbs. What I'm wondering is if you lose this insight that comes through the study of stress. Another example is English vowel alternations. If you break them down into a quantitative aspect, which is Closed Syllable Shortening, and a qualitative aspect, which is Vowel Shift, both of which have a domain independent of each other, you gain in understanding those alternations (see Myers 1987). Now are you committed, in fact, to a kind of global treatment of these alternations and how do you respond to the phonological argument? SINGH: I think the response to the phonological argument is that while on the one hand one can say that nouns and verbs, the majority of them actually, not even all of them, behave in this unified way and you gain something but you are losing under that analysis the fact that one of those contrasts actually is a purely morphological one, the classic being the examples used for displacing the stress to the left. The second response to that is that the construct or the idea of extrametricality isn't really without its own problems. So while one might appear to be gaining something, perhaps one needs to look at what is being lost in the process, and I'd like to suggest that perhaps what is being lost is the pure morphologicality. I have two problems. One: how have these two things been put together; two: the exceptions to extrametricality and then exceptions to those exceptions. I also think extrametricality or extraprosodicity has its own problems. So it's a trade-off, an empirical question as to which might actually stack up in a better way, or which generalization in this case is worth losing and which one is worth preserving. KIPARSKY: Can I just add, to sharpen the question a little bit, that the morpho­ logical aspect of difference between the stress in nouns and verbs is precisely in the different availability of extrametricality. There are indeed exceptions to

LINGUISTICS WITHOUT MORPHOPHONOLOGY

181

the generalization, and exceptions to the exceptions, as seems to be typical of morphological behaviour in every domain. There is always a general case, special cases, and special cases to those. This is how things work in the lexicon. On the analysis that I am contrasting with yours, the morphological aspects are still segregated into one very specific difference between nouns and verbs. It's just that they are, in my view, appropriately segregated, because then the rest functions according to phonological principles. On your account, the specificity of the phenomena is lost. That's the criticism, I actually don't think that you have identified any problems with extrametricality. You have hinted that there are some but I'd like to know what they are. SINGH: I mean there is a whole list of them. Which one do you want me to begin with? The problem is that if that is the typical behaviour of morpho­ logical rules, then we are saying that we should recognize that and say this is a morphological fact. If in the majority of cases, given the notion of extrametricality, if this thing falls together with something else, we are saying that it is not, there is no particular interest to keep them together if what is being sacrificed is the explicit recognition that one of these processes is, straight­ forwardly, even given your list, a morphological process. The point is whether this generalization should be combined with the fact that verbs and adjectives perhaps also fall into that kind of a pattern, whether that is worth keeping or not It is perhaps not an appropriate generalization. We are saying that that is an observation which perhaps does not capture any significant generalization about English, unless one makes the assumption that one must provide a single unified module-particular account of accent placement in English. KIPARSKY: I think that's really the wrong way of thinking. But there is no a priori assumption that it has to work one way or another. We are talking about an empirical question. That's chiefly the way I want to look at the question right now. You have a theory that says it must work this way, and the question that I want to ask is how do you get more insight into how these things really work? The claim that you haven't addressed that comes from this phonological research is that you learn more about these phenomena if you take them apart in ways that are governed by phonological principles, and that there is exactly one basic difference between nouns and verbs, which is that the final syllable counts for stress assignment for verbs but not for nouns. For you there is a multiplicity of patterns that distinguish nouns and verbs which, as far as I understood, must be built into the morphology of the derivation. Now I think you could argue on two grounds, but you haven't done either. One, you could argue that some very fundamental

182

TRUBETZKOY' S ORPHAN

principles, that have to be motivated independently somehow, force it to be that way. That was sort of the thrust of your case but I didn't really see exactly how that works. The other is you could say, "I understand stress or morphology better by doing it this way", that would be an empirical argument. Which of these do you actually grant as given? SINGH: I think that what you are saying is that one learns more by looking at accent as one single unified phenomenon. The claim that we are making here is that one actually learns a lot more about what sorts of processes, what sorts of capacities available within the language are exploited by the speakers for morphological purposes. In the final analysis, I think the assumption that Professor Kiparsky is making is that one learns more by making the assumption that there is a unified theory or account possible for all accent placement, we are making the different assumption that one would learn more about languages if one exploited rules as to what kinds of features or properties of the language system are used for making morphological distinctions in the language and which ones are not; to go one step further, for example, we would like to claim, although I think in the written text we didn't, that anything that is exploited in a language, a lexical contrast, whether it is a feature, whether it is a phoneme, whether it is a cluster of phonemes, whether it is a prosodic unit is in principle available for morphological exploitation. KIPARSKY: I can make exactly that claim and yet the other as well, that decomposing the alternations yields more understanding of the workings of the phonology. I think that in this crude accounting, I win. SINGH: Well, if you say so, you must. KIPARSKY: Because I can judge more precisely a phonological phenomenon that is available for morphological use, I mean, with that formulation I agree with that. DRESSLER: Just extremely briefly, I think we should keep, so to say, current in our memory storage the other models that have been proposed also. So I think one should really distinguish problems which exist just for one of the models and not for another. So I would say Morin's question puts a problem much more for Joan Bybee's model than for Singh's model,. The question is: how new rule generalizations may be made by a simple expansion of rules? We have this verb compounding since Late Latin and in many Romance languages. I think child language acquisition gives here more spec­ tacular cases, e.g., Eve Clark found with many American children (at about two years) compounds of the type open-man for the agent opener and openthing for the instrument opener. Now this I would think is really a problem,

LINGUISTICS WITHOUT MORPHOPHONOLOGY

183

where this comes from, if you simply believe in existing relations. PIGGOTT: I think my comment, perhaps a question, relates both to YvesCharles' observation about the origin of ouvre-boîte and then to Professor Dressler's observation about the origins of these compounds in child lan­ guage. This model does not seem to provide any sense of what's possible across languages. Why is it that the patterns that you get fall into these classes cross-linguistically? Why is it that incorporation for example is manifested cross-linguistically in just the way that it is manifested? Why is it that these compounds occur in French, also occur in Spanish, and in Italian? Are we going to assume that something inherent to Romance languages gen­ erate compounds of this sort? If so, how do you explain the fact that these patterns turn up in other languages? It seems to me that this model does not give us a sense of what's possible, of stuctures that are possible in compounds of this type. SINGH: Perhaps a short response given the time, I think the other way to look at the problem is to say that if what you have are these morphemes, smaller units than words, and perhaps that would explain how cases of the type that Professor Morin brings up or if Eve Clark's cases come up, assumes or ignores if you like what I shall refer to as Fabb's problem, meaning that if combination of morphemes or if the construction of words is a problem of morphemic combinatorics then one has to be able to provide an explanation for the other side, which is, given a certain number of morphemes that could perhaps be said to exist in the lexicon, the number of possible combinations is almost infinite and every particular theory of morphology devotes a considerable amount of time into blocking those combinations that seem not to appear, whereas we are taking as le point de départ that the patterns that are possible are in some sense licensed to buy the patterns that are already available, therefore eliminating at least eight thousand seven hundred and twenty-eight possibilities — the numbers are, of course, arbitrary. PIGGOTT: Right, but I don't understand how that deals with the question that I asked. I don't see how that overextends the program to explain why the same pattern occurs in the language next door. Why is it that Spanish, French and Italian have compounds of this sort? SINGH: Let me plead ignorance here and say that I don't understand how the alternative account has anything to say about that question at all. PIGGOTT: That's not the point. I don't even care if the alternative account has nothing to say. I'm putting the question to you. You propose the model, I think the alternative account should have something, that's the point, that's the nature of theories.

184

TRUBEIZKOY'S ORPHAN

DESROCHERS: I have two questions. First, is there any evaluation procedure for the complexity of morphological strategies that allow you to say that some strategies would be favoured by languages in general?. SINGH: I think the answer to that is pretty obvious and that is that if you look at the structural description, the pattern that you can extract, so that if you look at things like, let's say, sing in English as opposed to catch and caught, even take the extreme case of suppletion go and went, one could write some sort of a pattern up to relate these two things but it becomes obvious that some of them would be far more complex in terms of processing complexity and the prediction is the more complex the strategy is, the less likely it is to be exploited. And that's only one of the factors. The catch, caught pattern is probably not going to take off as quickly as some of the other ablauting verbs, strong ablauting verbs in English might. So to answer your question precisely, do we have an algorithm, the answer is no, do we think one can in principle be constructed, the answer is yes. DESROCHERS: My second question is how you evaluate the complexity of a strategy? In this connection, what is exactly the status of any symbol of the alphabet which appears in the strategies and which is not the constant? Do they stand for bundles of features? I'm thinking about the status of phonol­ ogical naturalness in morphological operations. Does it have some theoretical status or are these just convenient abbreviations for features or whatever? SINGH: For the moment, I think they represent, at least, the way we think about them now and as you put it quiterightly,convenient abbreviations for whatever may turn out to be therightrepresentation of phonological material. The intention behind this is not to committ ourselves to any particular view of how phonic material is to be represented. We make no representational com­ mitments. FORD: I want to respond to two interventions. First, the one from Paul Kiparsky. It has to do with the difference that we are trying to establish between morphological accent and phonological accent in English and the need to distinguish between them. I think that, if you take the minimal pair like the words overlie which is a noun and underlie which is a verb, where the accent is in a different position. I don't know if that coïncides with what you were saying about the position of accent with respect to nouns and verbs but it is the case that in English there are suffixes that are accentuated that sort of disrupt the generalizations you could make about phonological positional of the stress, that's all we really wanted to say was that if you are simply saying that the study of accent in English or any other language is a phonological problem, then you are never going to be able to resolve that

LINGUISTICS WTIHOUT MORPHOPHONOLOGY

185

problem because you are leaving out the possibility that the displacement of accent can in fact be a morphological process, that was just a response to that. KiPARSKY: So how do you account for the fact that English doesn't have preantepenultimate stress? Why couldn't your morphology put it any old place? Why does the morphology observe phonological constraints? What makes the morphology obey phonological principles? That's what I don't under­ stand in your approach. SINGH: I would say that that is not mysterious at all because morphological strategies obviously are built on phonological material. I am referring to the fact that morphology is sensitive to phonology. We all know rules and examples of processes that say that if the word weighs more than this, don't do this, or if the word has the following types of the prosodic properties whether one wants to talk about them in terms of syllables, the mora doesn't really matter, so I think there is no denial of the fact that morphology in fact exploits the phonological material that is there. I think what you are trying to suggest is deny the other side, that there is no reason to believe that phonol­ ogy can be sensitive to morphological information, but the first side doesn't appear to me to be mysterious at all. JANDA: I wish I had made two points earlier. Maybe I can make them now. Brian Joseph and I have proposed a device called a 'meta-redundancy rule', when it got farther we called it a 'meta-template', but this would be metaschemata which would then range over rules of your schema in (3) and allow you to recognize the parts without separating them in the rules, so the raised umlaut of German could be recognized insofar as they show the same structural description, these unifications are motivated by idiosyncracies that are shared by particular operations, as in various Sanskrit reduplications. Some of them look more like infixations [ajihista] type, some of them differ in their placement, but they all share this characteristic that the sonorant plus stop is reduplicated with a stop but the resonant plus sibilant reduplicates the sibilant. So that reaches over all these reduplications and seems to unify them in some sense. So if you allow meta-schema then you can have your cake and eat it too, it's one way to look at it. I wanted to suggest that. I want to come back to usage as it's come up. I thought someone was going to ask Joan last night, "well if usage determines representation, what determine usage?" which, I think, was Yves-Charles Morin's question, but compounds are known to be difficult here on the one hand and yet performance doesn't seem to answer all the questions, so if I can just take two problems with this, the second part of what I am saying, in a case like

186

TRUBETZKOY' S ORPHAN

abat-jour, can you generalize that in the right sense, I mean, abat-jour, the lamp-shade? What if you wanted to say "well there is something that blocks the reflection from airplane", could you go abat-avion, I think you would think more that you are going to shoot down the airplane. MORIN, Y-C: I probably wouldn't, but anyhow the question I raised was different. JANDA: But I wonder about that if the compound is idiosyncratic to start with, how does it work? Airplanes do reflect sometimes, I wish I had an abatavion to block the shining. The other side of the things is people do seem to draw conclusions from pragmatic performance factors. I'm thinking of the case of English explain which doesn't have dative movement construction, so explain me that for most speakers of English is out. There are cases of forms, there are systematic gaps. Most people don't have a past tense of forgo in English, they are uncomfortable with forwent and forgoed both. I hesitate to say this, next to my neighbour here, but there are said to be a hundred Russian verbs that have no first person singular present indicative, like laj. People say, Dick Oehrle has argued this in a paper, "if this occurs I should have heard it within a reasonable number of times, I have lived forty years and I've never heard this, it doesn't exist". So, I think, these things can be built into the grammar even if they seem very performance-oriented.

IV Hurch and Piggott MORPHOPROSODY

In order to understand how an entire sphere can be described by a few simple recipes as though nothing else existed, one must first free oneself of the clichés. T. W. Adorno (1984: 123)

Morphoprosody: Some reflections on accent and morphology* Bernhard Hurch Universität Graz 0. Morphoprosody Morphoprosody1 is a rather neglected part of grammar. Whereas morphophonology has attracted the attention of various scholars and theories the systematic study of the interrelation of prosody and morphology is only barely treated in the literature. I willfirsttry to shed some light on the phenomenology of the problem (§4) and second sketch an integration of morphoprosody into a natural theory of grammar. In the line of Dressier (1985a) and of classical Natural Phonology I will discuss the conflicting strength of the different forces acting in this field of grammar (§§5-7). A gradual character of deviation from prosody happens only where the triggering factor is related to prosody (namely in the interaction with segmental phonological processes) whereas the morpho­ logical impact on prosody mostly results in abrupt phenomena. A few diachronic observations will complete this picture (§8). Recent metrical theories (§§15, 16) usually ignore the whole field of morphoprosody and subordinate morphological functions of accent to the functioning of prosodic regularities. In addition to an immanent critique of these theories I will argue that the failure to recognize accent functions different from prosody leads to a proposal for capturing morphoprosody in the frame of Natural Phonology (§18). Morphoprosody is simply intended to cover the interplay which happens to take part in the interaction of prosody with non-prosody. For reasons of space there will be some further limitations. I will mostly treat the various ranges of accent and only randomly comment on other prosodic units. And I will center This paper has benefited from many discussions following presentations at various stages. I am especially grateful to W.U. Dressier and Rajendra Singh (cf., Singh 1996) for enlightening comments. I owe the discussion of the Romani examples to Birgit Igla. I am also grateful to Geoff Nathan for some useful comments and corrections. 1 The term prosodie morphology as used recently in various publications (e.g., Kenstowicz 1994, tracing back to McCarthy & Prince 1990) is intended in a rather different way as I will show in the course of the present paper.

190

TRUBETZKOY' S ORPHAN

the discussion on the dialectic relation between accent and morphology but constantly insist on the relation to other conflicting forces such as segmental phonology and morpholexical properties. 1. Morphoprosody and Stress Types Various distinctions of stress types have been proposed in the literature ranging from a classical structuralist division in the Jakobsonian frame to a strongly formalist taxonomy as in Roca (1992) but none of them sufficiently integrates what will be called in the present paper the morphoprosodic use of accent. With this term I refer to examples (e.g., 1-6 below) where the location of accent crucially depends and/or (co-)signals morphological categories, classes, etc. 7.7 Thus Jakobson (1937)2 offers the following well-known division: a) distinction of words (like Russian ú 'torture' and á 'flower') b) functional divisions showing as 1. cumulative function (Gipfelbildung) 2. demarcative function (Abgrenzung) c) gradation of words in the phrase. From this proposal basically two word accent functions have survived in pre­ sent-day treatments and have been established as accent types: the distinction of words (in the form of what is called lexical stress in current theories) and cumulative function which is theorized, for example, in the core of the various tree and grid models of the generative approach. On the other hand, phrasal accent is usually analyzed on syntactic grounds. It seems generally agreed that the relation between stress and domain (word-) boundaries is not necessarily bidirectional (cf. Latin where the location of the primary accent is established counting from the word boundary but the word boundary is not necessarily recoverable from the accent location) and that demarcation therefore is not a primary characteristic of accent. In formal theories demarcation can only have a derived function as it relies on the directionality and headedness of rules. In typologically related literature the function of indicating word boundaries is frequently attributed to demarcation, especially in languages presenting some degree of agglutination with higher rates of word length.3 2 Jakobson (1937) offers the first concise formulation, but the basic ideas are already present in earlier Praguian writings, like Trubetzkoy (1924, 1936). 3 In fact, agglutinating languages present a higher rate of regular accent systems. A parallel functional explanation for vowel harmony has been proposed in the literature as it is a

MORPHOPROSODY

191

1.2. In most of the recent prosodic literature there is little to be found on different types of stress according to possible different grammatical origins, as basically the same algorithmic devices generate the different stress patterns, and obvious differences in origin are leveled by additional principles and conditions in order to make words fit the algorithms (cf. §15). Roca (1992) is to my knowledge the only attempt to give a systematic overview over not only different results of an algorithm but over distinct types of possible origins and derivations of accent. The author intends to delineate what he calls a 'stress taxonomy' distinguishing between: • Rhythmic stress as divided into: - overt rhythm (covering those systems traditionally called 'fix accent systems' in which the assignment of a rhythmical structure depends exclusively on prosodic principles) - covert rhythm (including those languages which do not show a pattern of secondary stresses and in which the rhythm therefore is covert) • Lexical stress on its turn as divided into: - mixed systems (which are stress systems that show both a variable amount of lexically marked accents and with rhythmically constructed patterns combined through a version of the Elsewhere Condition, or some kind of Faithfulness Condition as stated by Halle & Vergnaud 1987) - pure accent systems (into this group there fall languages — Roca quotes Sanskrit — which do not construct accent contours following rhythmic principles) 1.2.1 Unlike Jakobson's distinctions, quoted above, Roca's stress taxonomy, narrowly speaking, is a taxonomy of stress systems.4 In the section following the above presentation Roca turns to what he calls 'paradigmatic stress', i.e., those forms where we can establish a correlation between the location of stress and the position of a given form in an inflectional paradigm, giving the example of Spanish verbal stress. After presenting the problem Roca opts for a lexical solution by marking the inflectional suffixes individually for their accentual behaviour and ignores a possible direct link between a specific ac­ centuation and a specific grammatical operation. Unfortunately he does not give any principled argument for his choice of analysis. preferably type-restricted phenomenon in agglutinating systems, ceteris paribus, co-signalling word boundaries. 4 It shall only be noted here that his categories overt vs. covert rhythm do not correspond to clearly distinct opposites.

192

TRUBETZKOY'S ORPHAN

2. Stress as Morphological Process There have been various allusions to the problem in the literature which seem not to have had any relevant repercussions in research. In his monograph Language, E. Sapir (1921) mentions variations in accent as "the subtlest of all grammatical processes". The examples he quotes like Navaho ta-di-gis "you wash yourself' (accented on the second syllable), ta-di-gis "he washes himself' (accented on the first) (1921:79) or the Greek verbal accent illustrate this idea. Together with other non-concatenative operations in grammatical processing (like umlaut, pitch changes, or introflection) he establishes the symbolic technique within his relational typology, a term which did not have very much success in the subsequent literature. Sapir does not go more into detail, although he discusses the other grammatical processes at some length, also in view of the following typological chapter V. But this view of the use of accent has largely been neglected in most of the following literature on the subject from all sides: not only from phonological and prosodic, but also from morphonological and morphological research. In the Martinetian framework, Paul Garde (1965) addresses the relation of "Accentuation et morphologie" but concludes proposing that certain groups of affixes have a distinct accentual behaviour. Again the regularity is recognized as being strictly linked to single derivational suffixes. 3. Morphoprosody as Interaction Morphoprosody as intended in the present paper refers to the situation where prosodic variation is connected to morphological processes. This needs some further specifications. By 'prosodie' I mean essentially all prosodie categories, even if the present paper basically is restricted to accent phenomena. The term 'morpho-' here includes its widest range from 'morpho'-lexical down to 'morpho' -phonological. And finally it will be crucial how to conceptualize the rather vague expression 'connected', i.e., to specify how this mutual interaction of prosodie and grammatical processing has to be represented in grammar. 3.1 This latter point seems rather crucial. Classical generative proposals (up to and including Siegel (1974) but also Hooper & Terrell (1976) in the frame­ work of Natural Generative Phonology) make certain stress rules sensitive to morphological classes or categories. Metrical Theories (in the spirit of Hayes 1980) do not treat stress types according to possible different grammatical origins, as basically the same algorithmic devices generate the different stress patterns, and obvious differences in origin are leveled by additional principles

MORPHOPROSODY

193

and conditions in order to make words fit the algorithms. Also in Lexical Phonology stress rules simply apply cyclically in relation to the different strata in the lexical part of grammar. In all these theories grammatical operations seem to have a clear precedence over prosody and it is by no means discussed, how prosody itself may function in grammar. 3.2 The connectedness of prosody and morphology is directly stated in Wurzel's (1980a:313) Principle of morphological accent: "Die Bedingungen für die Festlegung des Akzents sind morphologischer Art. Die Beonung erfolgt in Abhängigkeit davon, welcher morphologischen Klasse ein Morphem angehört. Der Akzent hat die Funktion der Signalisierung der Morphem­ klasse". Although Wurzel formulates this principle for the facts of German compound vs. non-compound verbal accent, it roughly expresses what has to be understood as morphological accent, namely from the side of morphology the fact that there are instances (like in (1) below) where a specific accent location is used as a non-concatenative device for the expression of a morphological class or category, and from the side of prosody, where the accentual structure itself does not follow purely prosodic principles but — on the contrary — fulfills its function best when it is a negation of the language specific syntagmatic prosodic processes. 4. Types of Morphoprosodical Interaction As mentioned above, instances of the interplay between accent and morphology are present gradually across morphology, from the morphological organization of the lexicon through to the interplay and following opacity arising between prosodie and segmental phonology.5 4.1 When talking about lexical distinction by accent I do not refer primarily to Jakobson's Wortunterscheidungsfunktion, i.e., to lexical items individually marked for their accentual behaviour, but rather to morpholexical regularities 5 Dogil (1981) presents various instances of the interaction of accentuation with segmental phonology and morphology. He refers his discussion primarily to the explanation of the (ir-)regularities of accent systems as a whole. Moreover, he does not give a principled account of these interactions either within the different modules or in a diachronic perspective. In this line, he fails to account for the fact that what I call 'deviation' from a prosodie accent below may give rise to irregularity to a differing degree. The metrical model he opted for is of the early generative metrical generation. It is thus difficult and maybe out of interest here to go into detail with the problems he encounters when integrating a 'polycentristic' model (Dressier 1977) with early metrical studies of Halle & Vergnaud (1978). A more detailed discussion of the issue presented in Dogil (1981) is included in Hiirch (1991), esp. pp. 38-43. It should also be noted that Dogil in later metrical studies does not take up his 1981 issue any more.

194

TRUBEIZKOY'S ORPHAN

based on accentual regularities. The inner structure of the lexicon — as opposed to the so-called 'waste-basket-concept' — has been in the center of recent research (cf., for example, Lehmann 1990). The most general lexical distinction expressed through accent is one of lexical strata as in various Romani dialects of the Balkanic countries as described by Boretzky (1989:363). A whole series of differences in gram­ matical behaviour holds for the two basic strata, the inherited lexicon and the loanwords, whereby we have to keep in mind the specific loan situation of Romani. In those varieties the old Romani lexicon presents final accent whereas loanwords are not finally accented as in (1): (1)

Inherited vs. loan words in Romani soró "head" vs. fóro "town" bori "bride" amúni "anvil"

This distinction holds independently of the age of the loan process, and what is more interesting, independently of the accent location in the language of origin (see Boretzky 1989, Igla pers.comm.). Greek karfi which forms part of an older stratum of loan relations (10th/l 1th century), although having final accent and thus conforming to Romani prosodics turns up via karfín, krafín finally as kráfí "nail".6 The only substantive lexical specification which entails a series of specific grammatical behaviours seems to be the one concerning the accent shape. Dressier & Thornton (1991) propose an analysis of verbal inflection in Italian which presupposes two base forms for each verb distinct from each other by absence vs. presence of a thematic vowel and by the placement of lexical accent: "one baseform is a radical base, which bears a prelexical accent on the root and contains no TV [=thematic vowel]; the other is a thematic base, with a TV that bears a prelexical accent", thus /ám-/ vs. /am-á/, /arrív-/ vs. /arriv-á/, etc. According to membership in inflectional classes the two base forms may differ with regard to other characteristics. For a discussion of the advantages of this morphological analysis, cf. the paper by the authors; for the present purpose it may only illustrate one example of the use of accent as an orga­ nizational principle in the lexicon. More clearly this can be seen in those cases, where the relation between the base forms is (strongly or weakly) suppletive and thus one of the alternative forms turns up only with radical accent, the other only with an unaccented root; cf. the alternation of /vád-/ vs. /and/ in the 6 All vowel final words enter Romani as feminines, independently of their original gender (cf. Serbocroat. pívo>píva "beer"; more > mora "sea"). They show final accentuation in the acc.sing. as the result of vowel absorption (see below 4.4.).

MORPHOPROSODY

195

following paradigm (2): (2) Italian uscire lsg. 2sg. 3sg.

present tense indicative verbal paradigm of andare "to go" and "to go out" vádo és  lpl. andiámo usciámo vái ésci 2pl. andáte uscíte vá ésce 3pl. vánno éscono

In Italian, in addition, we find a class of nouns ending in a stressed noninflective vowel, mostly -à and-ù, sometimes -è  -, like università "university", virtù "virtue", canapè "sofa", paltò "coat". This class of nouns, partially loanwords, partially developed out of a subclass of the Latin 3rd declension, includes specific morphological characteristics. The final vowel does not indicate either gender or number; number variation is not expressed on the noun; and as far as a word of this subclass is inherited it is feminine. In German a series of prefixed verbs is distinguished in the infinitive by stress placement, where the variation includes stressing of the prefix vs. stressing of the root vowel; cf., übersetzen "to cross" vs. übersétzen "to translate", übertreten "pass over" vs. übertréten "transgress". This distinction entails some further morphosemantic and morphosyntactic properties. Stress on the first element signals the separability and the literal (prepositional) semantic interpretation of the verbal prefix. But accent does play this role not only in doublets, but this function is a general rule of German verbal prefixing. 4.2 In derivation the use of accent has been discussed at some length with respect to different classes of derivational suffixes in English depending on whether they are stress preserving or stress changing (Chomsky & Halle 1968, Siegel 1974, Halle 1973, and later Lexical Phonology). The dominant solution in the classical generative proposals was that there are rules (like the Lexical Redundancy Rule in Siegel 1974:124) which are sensitive to morphological categories like 'V[erb]' In the Lexical Phonology approach the application of stress rules depends on cycles corresponding to derivational strata. In English, moreover, there seems to be a strong tendency towards a final accent in bisyllabic verbs which results in a series of doublets where the location of the accent functions as the only indication for a noun-verb dis­ tinction; e.g., import vs. impórt, cóntact vs. contáct, discharge vs. dischárge, réfill vs. refill. This matter is treated under various headings (prefixation, conversion) in the literature. Most authors opt for a derivation of nouns from

196

TRUBETZKOY'S ORPHAN

verbs.7 This distinction first is not necessarily linked to prefixing (cf., Bauer 1983:124f.) as is shown by examples like pérmit vs. permit, ségment vs. segmént, etc., and secondly it is not linked to doublets, but despite of a small number of contradicting accentuations,8 bisyllabic verbs tend to show final accent, especially when they have any kind of complex internal word structure. Moreover, this modificatory device seems to be a productive process, as the number of verbs following the iambic accentual pattern is increasing. 4.3 Systematic modifications of accent structure in inflection usually denote categories or paradigm structure. In the Southern Guipuzcoan Basque dialect of Zegama (Amundarain Arana 1990) we find a nominal paradigm as represented in (3). (3) South Guipuzcoan Basque (Zegama) noun paradigm of ardi "sheep" (Amundarain Arana 1990) absolutive ergative dative genitive associative destinative

singular ardíe ardík ardíri ardín ardíkin ardíntzat

plural árdik ârdik ârdi: ârdin árdikin árdintzat

The main and only systematic difference between the determined singular and plural inflection in bisyllabic base forms ending in -i or other vowels is given by accent structure. As a predominantly agglutinating language, Basque does not have declensional paradigms proper, but within this example we have diachronically the complete absorption of the singular definite marker — except in the absolutive — (Standard Basque -a-) which stands between the base form and the case suffix leaving the trace of a stress protracted more to the end of the word. The result is that part of the nominal paradigm shows a systematic difference in the accent patterns of definite singular and definite plural forms. In Spanish future formation we have the cosignaling function of stress as in the examples under (4) as opposed to the imperfect subjunctive. 7 Only Lieber (1981) proposes a separate lexicalization as distinct entries for nouns and verbs. But she obviously misses one generalization. 8 There are some verbs which allow for both accentuations (like conjure), a class of verbs with non-stressable final syllables which therefore carry stress on the first one (like conquer), and finally a small group of items where nouns and verbs are not distinct (like concérn).

MORPHOPROSODY

197

(4) Spanish verb paradigm future and imperfect subjunctive for hablar "to speak" Future Imperfect Subjunctive lsg. hablaré hablára 2sg. hablarás habláras 3sg. hablará hablára lpl. hablarémos habláramos 2pl. hablaréis hablárais 3pl. hablarán habláran

The most adequate analysis of the Spanish verbal system would probably parallel Dressier & Thornton's study of Italian employing a double base form approach. The forms of the future and of the imperfect subjunctive are both derived from the thematic form containing a stressed thematic vowel. But a general restriction states that the last of the lexical or grammatical accents in a word receives main stress (analogously proposed in Hurch 1990 for Italian). It is preferable to distribute stress not individually to all inflectional suffixes of the future in their lexical representation but in the morphological category as a whole. And as inflectional suffixes tend to be on the periphery of the word, they receive the main stress of the word. In this case — contrary to the Basque example in (3) — future is expressed by two concomitant procedures, the addition of inflectional affixes and the modification to a specific accent pattern. In Tongan a final accent indicates definiteness as presented by Condax (1989:425f.) and in (5). (5) Tongan definite accent (Condax 1989:425f.) 'omai ha papa "bring a mat" 'ornai 'a e papá "bring the mat" 'omai hoku papá "bring my mat" 'omai e papá ni "bring this mat" The penult accent of Tongan is overruled by a final accent of definiteness of a noun, derived probably from demonstratives and expressed syntactically by other means like specifiers, possessives, etc. The ordering force of accent within paradigm structure has been recognized for many cases, even if accent as a co-signalling property is frequently nondistinctive. 4.4 In the model of Natural Phonology, prosodic phonology essentially precedes segmental phonology in derivation. But prosodic processes may re-

198

TRUBEIZKOY'S ORPHAN

apply during the whole course of phonological production. This fact may lead to accent variation resulting from the interactions between segmental phono­ logical operations with morphological formations. In the North-Eastern Basque dialect of La Soule (Xuberroa) we find — in contrast to most varieties of Basque — a rather regular accent system probably borrowed from the neigh­ boring Romance languages or dialects, a penultimate accent (Michelena 1957/ 58). A segmental phonological process of contraction reduces vowel sequences (especially when homorganic), even when arising through morpho-logical concatenation as, for example, in the absolutive singular of the mugatua form (determined) where the determiner -a is added to — ceteris paribus — vowel final base forms (cf. the derivation in (6)). The accentuation stresses the penult vowel and a later contraction reduces the vowel sequence to a single vowel carrying accent. (6) Determined and undetermined absolutive forms of vowel final nouns in Suletin Basque determined undetermined alaba alaba "girl" determination alaba + a accentuation alabáa alába vowel reduction alabá I alába surface form Thus, what turns up on the surface is a simple contrast between determined singular and undetermined forms in the absolutive case realized through the accent position. The specificity of such forms lies in their complete phonological/prosodic and morphotactic transparency. The accentuation of a final vowel indicates the application of a segmental contraction process and thus the presence of a determining -a suffix. Cf. also the Ancient Greek genitive plural ending in -ãn carrying final accent, derived from an older -aon (< *-ason) still by contraction, keeping the accent on the suffix (Rix 1976:119, 133f.). The alternating Romani forms lekárka vs. lekarká "medical doctor, fem., nom. vs. acc." (borrowed from Bulgarian into local Romani varieties) are derived from a transparent accusative suffix -a, shifting the accent analogously to the Basque example. This alternation happens in autochtonous as well as in borrowed material (Igla, pers. comm.).

MORPHOPROSODY

199

4.5 In various accent systems the 'lowest' form of interaction between prosody and morphology holds for the selection of the domain of accent location. This point refers to those languages where general accent rules refer not to phonologically or prosodically determinable domains but to morpho-logical ones like the stem. Accent processes which crucially refer to the stem are reported, e.g., for Maori (Hohepa 1967:10f.), Chipewyan (Li 1932), Kwakiutl (Boas 1947:218), Dakota (Boas & Deloria 1939; cf. also the treatment in Halle & Vergnaud 1987:90ff.), Otomi (Blight & Pike 1976), and others. There may be additional interaction with segmental (vowel quality) or suprasegmental (tone, syllable structure, quantity) characteristics, thus morphotactics. 4.6 A loss of transparency of accent may result from the interaction with other prosodic phenomena, like the syllable or the mora. I will not go into the detail of well-known and well-discussed phenomena like quantity sensitivity (cf. the conceptualization, for example, in Hayes (1980); a more refined statement of different types of quantity in Pulgram 1975). Especially when there are no strict limitations on the size of the foot, systems like Latin build accent systems with prosodically productive locations of the primary accent but with decreas­ ing recoverability of, e.g., domain boundaries. Further interactions with the whole range of morphologization may occur. 4.7 On a different level of analysis we find the interaction of accent processes with segmental phonology. Dogil (1981:94f.), without further discussion, quotes the examples of the Moksha Mordvin dialect where only non-high vowels receive stress and the Jazva dialect of Komi where tenseness preferably governs the accent location. But other such systems are not rare, e.g., Azerbaijani (Householder 1965a). A closer analysis should reveal how far those segmental properties have to be defined through their prosodic value. But they produce relatively opaque accent patterns as the location principles are not purely prosodic. 4.8 What is not treated here in (4.1-4.7) are both those instances where the location of accent depends on idiosyncratic specifications of the respective morphological or lexical domains in the lexicon, like the class I and class II suffixes in English, amply treated in Siegel (1974) and, subsequently, in Lexical Phonology and on the other hand prosodie restrictions on morpho­ logical operations as discussed in Kenstowicz (1994) under the heading 'prosodie morphology'. The former do not allow for any further grammatical generalization that goes beyond itself, namely the location of accent, whereas

200

TRUBETZKOY'S ORPHAN

the latter primarily seem to express morphotactic restrictions. The points raised in §§4.7 and 4.8 will not receive further specific attention here as their grammaticalizing force is relatively low. They will be considered only with regard to a gradual deprosodification along the lines proposed in Dressier (1985a). 5. Grammatical and Prosodic Accent The present study follows the basic assumption that accent may be employed by various parts of the grammar for the expression of grammatical relations. As accent is an essentially prosodic unit, prosodie accent itself, i.e., accent which is assigned on purely prosodie principles, plays a major role in all accent systems. And, indeed, also in systems with so-called free (nonphonemic) accent like Italian we easily can detect prosodie preferences for accent locations and accent patterns. The crucial point is that all different uses of accent may interact among themselves and they do interact with prosodie accent insofar as any non-prosodic accent, later in the derivation, enters the syntagmatic prosodie processing, and the grammatical functionality of nonprosodic accent is given only when it is a negation of purely prosodie principles. Hence, at the moment of grammatical elaboration, paradigmatic accentual preferences have been established through paradigmatic accent processes, analogously to the concept of segmental syntagmatic and paradigmatic processes as outlined by Stampe (1973); cf. the more detailed presentation in §18. The relation of grammatical (in a broader sense) accentuations to purely prosodie ones is probably not the same as the relation that holds between segmental phonemics and segmental morphophonemics as outlined in Dressier (1985a). As it will be shown in the following sections, the interaction between prosody and morphoprosody may result in more or less prosodically 'deviant' systems, which to some degree allow a gradual interpretation along the lines of the strength of the interacting modules but the basic difference (see §4) lies in the fact that grammatical accents, diachronically, are not necessarily a product of the grammaticalization of an otherwise prosodie process, i.e., prosodie processes do not gradually loose their prosodicity in moving towards morpho­ logy and maybe on to more distant modules. 6. Derivation, Inflection, and Stress 6.1 Derivational affixes differ from inflectional ones in various respects, among others in their relative position to the stem but also with regard to their phonemic and prosodie (syllabic) shape and their phonotactics. As languages

MORPHOPROSODY

201

like Latin nicely illustrate, derivational affixes more often are bisyllabic and contain vowels or consonants exhibiting quantity, which results in their carrying word stress. 6.2 The way in which inflection may interact with accent is basically threefold. Either we find a single inflectional form (co-)signalled by a specific accent location as in the above Tongan example under (5), or an inflectional category is characterized by a specific suffixal accent as in the Spanish future and im­ perfect subjunctive paradigms in (4) or, finally, two paradigms may just differ in accent location in which case the accent is the only paradigmatic distinction ((3) above). 7. Interactional effects In this direct interaction of the modules or submodules, one more assumption of Natural Phonology and Natural Morphology9 should be confirmed, namely that a prosodically attributed accent pattern reflects the phonologically more natural state than those patterns which arise through the interaction with other triggering factors. This assumption on the formulation of which in terms of markedness Mayerthaler (1981) constructs his version of Natural Morphology, on which Dressier (e.g., 1985a, 1987) builds his semiotic approach covering the different modules of grammar and which is enlarged by the concept of system adequacy by Wurzel (1984a) shows rather different aspects in the analysis of accent phenomena. 7.1 We observe different degrees of morphologization or deprosodification of prosodic processes in correspondence to the modules prosody is interacting with. 7.1.1 The interaction with other prosodic processes and units creates the fewest inconsistencies on a scale of language specific paradigmatic and syntagmatic prosodicity. In fact, systems like Latin where — apart from the subclass of items accented on a final long syllable (cf., Arpinas, nostras) — the location of a primary accent may vary between the penult and the antepenult according to the quantity structure of the penult syllable, are subsumed under regular (traditionally termed phonemic) accent systems. Their regularity con­ sists in the application of cumulating prosodic principles, notwithstanding the point that they are developed on different grounds. Prosodic recoverability of, e.g., the domain boundary, is not exhaustively fulfilled, but no instances of 9 Especially in the versions presented by Dressier (1985a), Dressier, Mayerthaler, Panagl & Wurzel (1987).

202

TRUBEIZKOY'S ORPHAN

ambiguity arise productively. Thus the arising differences are too small and morphonologically too inconsistent to exploit them for grammaticalization. 7.1.2 The interaction with segmental prosodic phonology generally deviates one step further from prosodicity, especially in cases like the Zegama Basque noun paradigm where a later segmental process destroys the previously es­ tablished accent domain size by contraction and subsequent syllable deletion. Here, the accent pattern itself is deviating although the recoverability is perfectly fulfilled, both of the prosodic accent domain and of the grammatical word form signaled by the accent. 7.1.3 A still stronger morphologized prosodic pattern is found when the domain of the application of accent processes is morphologically determined. Languages quoted under this heading usually stress a determinate stem syllable which follows 'prosotactic' regularities, e.g., the first heavy stem syllable in Maori (Hohepa 1967). Prosodic word patterns are directly related to morphotactics insofar as affixes are generally unstressed and stems are stressed and, within the morphological domain, the accent placement itself follows prosodic principles. 7.1.4 Within the range of accentuation in inflection and derivation, the transparency of prosodic processes synchronically is mostly obscured. The examples quoted above show traces of prosodicity to a different degree. What is crucial is that none of the instances where a specific accent location itself is used for morphological purposes is governed or interacts with prosodic principles at the given point of derivation. The question of the extent to which an Early Romance compositional verb accent reflects natural principles is completely irrelevant in a situation where a future tense inflection is grammaticalized as a synthetic form. As mentioned above, there may be systematic differences in the use of accent in inflectional paradigms and in derivation, but they do not concern the prosodicity of the accent location itself. 7.2 A parallel to diagrammaticity (Dressier) or constructional iconicity (Mayerthaler) in accentology should imply that a more highly marked inflec­ tional category corresponds — ceteris paribus — to a prosodically more highly marked accent pattern. Thus, the determined form in the Suletin Basque absolutive case showing final accent deviates from the general pattern of penultimate accent in this dialect. Diagrammaticity is the more appropriate term, as constructional iconicity rather refers to additive procedures only. A synthetic Italian or Spanish future is more marked than the present, and this higher degree of markedness is expressed by the higher marked accent pattern in the future paradigm.

MORPHOPROSODY

203

But this relation may be further obscured when the deviant accentuation is the result of an interaction with a segmental process holding for (or ana­ logically extended to) a complete paradigm as in the Zegama dialect of Basque quoted under (3). The deletion of the singular marker -a(-) produces a more marked accent pattern, but the deletion itself synchronically is not recoverable. Thus, the diagrammatic relation is not a necessary condition on the use of grammatical accent placement. But this mismatch simply expresses the noneasily regularizable behaviour of accent in non-concatenative morphological operations. 8. Morphoprosody and Diachrony 8.1 Only a brief discussion of the diachrony of morphoprosody will be possible here. Any reasonable theory of morphoprosody has to take into account the historically traceable instances of accentuation into the synchronic explanation itself. I will divide the discussion of the present section into two parts, namely, first, the question of how cases of morphoprosody arise at all through diachronic change, and second, how they develop and possibly are leveled. 8.2 The twofold character of morphoprosody as an instance of morphology, phonology, and of prosody, two distinct (sub-) modules of grammar (Hurch & Nathan 1995), suggests a departure from both sides, from morphology and from prosody. And, in fact, according to the basic assumptions of Naturalness Theory, we find two rather distinct origins. 8.2.1 The different effects of grammaticalization of morphosyntactic forma­ tions include the rise of accentual structures in synthetic word forms mapping directly an older analytic construction type. This is the case of the above quoted Tongan definite accent quoted under (5) where definiteness seems to be grammaticalized through cliticizing a demonstrative, which changes the overall accent structure of the word. The specific accent behaviour of the Italian future verb forms carrying systematic suffix accent (cf. the analogous forms in Spanish under (4) originates from a Late Latin/Early Romance analytic construction of the type [[infinite main verb]+[finite auxiliary habere]] in which the second element carries compositional accent. Throughout the process of morphologization the position of the accent remained constant. The subsequent reduction of the auxiliary forms to an inflexional affix resulted in the deviant stress pattern. In Naturalness Theory (Wurzel 1984a, Dressier 1985a) it has been proposed that it is exactly this interaction between an­ tagonistic or at least conflicting forces of grammar which creates complexity

204

TRUBETZKOY'S ORPHAN

and thus more highly marked structures. It is the gain of morphological naturalness of synthetic as opposed to analytic forms which — as one consequence — leads to a loss of prosodic naturalness by introducing specific accent patterns for the new synthetic future. 8.2.2 On the other side also the interaction with segmental phonology or with other prosodic factors may cause morphoprosodic patterns. The collapse of quantity in Latin (starting out already through the Classical period) had as one consequence the breakdown of the accent system, as the material basis of the quantity-sensitive type vanished. Interestingly only in the rarest cases the Romance languages did change the position of the accent in a given word.10 Apart from lexicalizations of specific accent forms and of a series of processes transforming antepenult to penult accentuations by deleting a post-tonic syllable, early Romance languages show a certain amount of grammaticalization which express themselves, among other things, by new and 'unusual' accent locations, e.g., the muta cum liquida sequences create a possible accent position in the preceding syllable, integrum > intero "entire". The context-free paradigmatic (Stampean) loss of quantity corresponds to a natural segmental process, as quantity distinctions render systems more marked.11 The negative effect for prosody again results from the interaction of conflicting teleologies, namely the raising of naturalness of a specific segment inventory versus the prosodic basis of an accent system. As in the case of umlaut in German, the lowering of the trigger vowel or its deletion renders the umlaut opaque and the change of the stem vowel takes over a morphological function. Similarly the vowel contraction in Suletin Basque in (6) is synchronically productive and its transparency is given by the deviating accentuation in the determined singular form. But the absorption of determiner -a- in the Zegama paradigm is traceable rather on diachronic than on synchronic grounds. In contrast to the Latin-Romance development, we deal here with the interaction with a syntagmatic natural process disturbing the transparency of the accent location in a given morphological context. 8.2.3 Obviously in the conflict with tendencies other than prosodic, the latter may also keep stability and both inflectional forms or segmental organization 10 Cf. one of the most striking examples: Latin Augustus shows up in French as aout where everything has been deleted but the accented syllable which remained stable for the whole course of all sound changes. I do not intend to postulate a general stability of accent locations; already the history of the Slavic languages offers quite a different picture. 11 I assume that quantity may be functional on both prosodic and segmental phonological grounds with rather different but not independent functions.

MORPHOPROSODY

205

are then subordinated to prosodic regularities. This happens also in the very same languages, cf. the leveling of accent over inflectional classes in Italian, the frequent vowel deletions in proparoxytones in Late Latin, etc. 83 The development of collapsed accent systems is by no means chaotic and follows basically the same lines as outlined above. The number of lexicalized accents is not arbitrary as it enlarges considerably the amount of information to be stored in our permanent memory and, therefore, tends to be reduced by grammaticalizations and reprosodifications. Again briefly I will sketch some repair strategies in the Romance history (cf. also the presentation in Klausenburger 1974). After the loss of quantity already in the I AD, there seems to prevail a tendency towards penult accent which — with more or less success — dominated all the Central and Western Romance languages. In the Appendix Probi by far the most common type of colloquial substitutions are those shortening a dactyl to a trochee {tabula non labia, auricula non oricla, etc.). A second group of regularizations concerns the lexical specification of derivational suffixes, accent specifications which have survived till today and even govern in part the Latinate lexicon in English. In general, I would make the assumption that languages as a context-free tendency reduce the id­ iosyncratic portion of prosody as effectively as possible towards all other types of partial regularization or even towards complete regularization. The other way round, morphoprosodic systems may be morphologized gradually only through interaction with other prosodic or with segmental phonological regularities or processes. Their interaction with grammar entails a disrupt otherness from prosody which may be functional only if the grammaticalized forms are a negation of paradigmatic prosody. 9. Non-Concatenative Inflection Instances of non-concatenative inflection are discussed only randomly in the literature or focused on within the rigid frame of single theoretical models (cf. the extensive treatment of tone and vowel harmony in autosegmental phonology). But morphological theories in general predict that concatenative inflection is to be preferred over non-concatenative inflection. And there are good external reasons which support this assumption, e.g., coming from cognitive processing, amply discussed in Mayerthaler 1982, Dressier et al. (1987). But in most studies on morphological rule typology the aspect of prosodically governed non-concatenative alternations has been somewhat neglected12 and non-prosodic examples like umlaut, ablaut, vowel harmony or Or — if at all — only randomly discussed as in Carstairs (1988a).

206

TRUBETZKOY'S ORPHAN

internal inflection are usually illustrated. Unfortunately this is also true for Natural Morphology, a theory which otherwise emphasizes discussions of rule typology.13 A. Carstairs (1988a) sheds some more light on some examples he analyzes in view of organizational principles in so-called mixed paradigms, that is, in paradigms where morphological categories are expressed by more than one procedure, umlaut plus affixation in German noun declension, ablaut plus affixation in English verb conjugation, stem consonant alternation plus affixation in Hungarian verbs, and stress alternation plus affixation in Russian noun declension again. All his examples point to the issue that an affixes-only analysis is the only way to guarantee the fulfillment of his 'Paradigm Economy Principle' that states that those analyses are preferred which organize just as few paradigms as necessary in order to make inflections work. But in his argumentation Carstairs refers crucially to paradigm organization in these specific examples and does explicitly accept the morphological functions of non-concatenative operations under discussion. But given the — even if not preferred — existence of non-concatenative inflection there is no reason to exclude the analysis of prosodic means for (co-)expressing morphological regularities from our reflections on the topic.14 Indeed, as the previous sections show, the non-concatenative use of prosodic units in morphology arises as a relatively marked result of the interaction of forces from different (sub-)components of grammar. 10. Prosodic Inflection 10.1 Quantity, as another prosodic feature, may also take part in the organization of morphological alternations and paradigms. It pertains to the same relationality in grammatical and prosodic/phonological organization as is outlined in Natural Phonology and Morphology (and above). Cf. the examples in (7) quoted from Carstairs-McCarthy (1992) with those in (8).

13 Apart from the limited range of phenomena mentioned in Wurzel (1980a), I know of no systematic treatment of prosodic phenomena in Naturalness Theory. Dressier (1985a) does not discuss accent phenomena in any detail. 14 In what is termed "prosodic phonology" in Kenstowicz (1994), who relies nearly exclusively on previous studies by McCarthy and Prince, the author presents a rather different approach. He does not question the function of prosodic units themselves as morphological factors but merely studies the role of prosodic means of description in otherwise additive morphological operations, like reduplication.

MORPHOPROSODY

(7)

New Zealand singular tangata wahine matua

(8)

Extract of Latin 1st and 4th noun declension nominative ablative meensa meensaa "table, nom./abl." puella puellaa "girl, nom./abl." filia filiaa "daughter, nomVabl." nominative genitive "fruit, NomVGen." fructus fructuus domus "house, Nom./Gen." domuus manus "hand, NomVGen." manuus

207

Maori plural formation plural taangata "person/people" waahine "woman/women" maatua "parent/s"

In New Zealand Maori there are some nouns which express plural by lengthening the vowel of the first syllable in the word. In the 1st declension of Latin nouns the nominative suffix -a consists of a simple short vowel, the ablative suffix -aa of the corresponding long counterpart; analogously in the 4th declension the simple vowel in the nominative suffix -us has as a counter­ part long -us in the genitive. The parallel quotation of these two examples does not intend to claim a corresponding analysis of quantity. Whereas in the Maori forms we can postulate a derivation analogous to (9) below, i.e., a processual derivation of a plural from a singular form, the Latin example does not mean to derive a genitive or an ablative suffix directly from the suffix denoting the nominative. These two examples are obviously different in one important respect in that the one from Maori uses vocalic quantity as a modificatory device whereas the Latin noun declension is basically additive and quantity is used here like any other vowel alternation. But both follow a diagrammatical relation as proposed by Natural Morphology: to the relatively unmarked singular (Maori) and nominative (Latin) correspond the simple short quantities, whereas the relatively marked categories plural (Maori) and ablative resp. genitive (Latin) are formed with the long variants. In general, we can assume that modificatory rules prefer the variation of vocalic quantity over the modification of consonantal quantity as the former has more easily repercussions directly on a hierarchical prosodic structure and thus allows for a better cognitive processing.

208

TRUBETZKOY'S ORPHAN

10.2 One common way of explaining examples like the ones from Maori mentioned above comes from autosegmental phonology, which transfers the problem to an independent tier and links the tier structures to morphology. We can illustrate this through the following diagram (cf. the discussion in Carstairs 1988a): (9)

Possible CV-representation of quantity in a morphological context

In this diagram the two different structures on the CV-tier represent two different morphological categories. This approach has specific undeniable im­ plications for morphological analysis, nicely illustrated by Carstairs. In the context in which I am quoting the example, it is important to note that it does not grasp the specificity of prosodic analysis: the autonomy of prosodic quantity is simply reversed to an abstract segmental CV-tier (cf., (11) below). 10.3 Quantity may in itself take part in cosignalling morphological regularities but not as a characteristic projected higher up in a prosodic analysis. In other words, Maori delivers a suitable example of a distinction in singular and plural paradigms based upon vocalic quantity, but to my knowledge there are no instances where the secondary prosodic function of quantity sensitivity for, e.g., accent rules, is morphologically exploited. Thus, a derived function of a prosodic unit situated on a different level than the unit itself cannot be grammatically functional. 11. The Syllable in Morphoprosody 11.1 The syllable itself does not directly play a role in morphoprosody. There are no examples where distinct syllabifications would characterize or be the bases for the expression of morphological categories or paradigms. It would not be expected that we find grammars in which differences of syllabifications like in strings of CVC.CV and CV.CCV would systematically constitute differentiations in, e.g., singular and plural. The relation that exists between morphology and syllabic structures is indirect and unidirectional. The obser­ vation that in inflexional languages derivational affixes are subject to less structural restrictions than inflexional affixes is a problem of morphotactics and not of syllabic phonology.

MORPHOPROSODY

209

The reason for this exclusion of syllable from morphoprosody is grounded in the fact that — as Donegan & Stampe (1978) have convincingly shown — the syllable is not an underlying unit represented in our permanent memory but entirely derivable; therefore it is not distinctive. 112 In recent phonological literature syllabic restrictions are frequently quoted to explain morphological operations like reduplication (cf. Kenstowicz 1994, §11, including further references). But it is hardly ever the syllable as a whole which reduplicates, but rather only parts of it, which does not really constitute evidence in favour of the syllable itself but rather undeniable evidence for a preferred CV structure of units resulting from reduplication. Moreover, I would claim that such restrictions reflect only indirectly the role of prosody (units, processes) in morphology, inasmuch as prosody has simply a con­ straining function in an otherwise morphological operation and consequently that the syllable or syllabication processes do not actively function as morphological signs. 12. Tone and Morphoprosody Tone is not part of a rhythmic prosodic hierarchy and thus I will exclude it from the present analysis. For the sake of clarity I just want to mention that tone differences and tone processes are rather well established and may best be described among the suprasegmentals used in non-concatenative procedures in morphology, cf. tone reversal in the formation of the 1st. duoplural of the perfective paradigm in Chilcotin (Cook 1989:184). Both in tone and in pitch accent languages such alternations are rather common. 13. Generalizations on Morphoprosody The discussion up to the present point allows for some generalizations which at a first glance seem rather superficial but which — to my knowledge — have not been stated systematically. a) Potentially all underlying prosodic categories may primarily or secondarily be a means of expressing morphological regularities. b) In a given language only nonderivable prosodic categories may have morphoprosodic functions. c) Prosodic units and processes may function or take part as a means of expressing morphological regularities only within their own domain and may not be projected on any higher level. These generalizations perfectly fit the analogies with segmental phonology and

210

TRUBETZKOY' S ORPHAN

the use of only phonemes in morphophonological alternation in modificatory morphological rules and it fits the knowledge we have from phonemicizing, grammaticalizing and deprosodifying historical change. 14. Prosodic Preferences The picture which emerges from the preceding sections suggests a rather different treatment of accent phenomena in the framework of Natural Phonology, or more generally, of Natural Grammar. The crucial point is that in analogy to segmental phonology we postulate a paradigmatic prosody which sketches the language specific prosodic preferences following mainly contextfree prosodic processes. Moreover, all parts of grammar may exploit the resources prosody offers for the expression of grammatical meaning and form. The functionality of this latter use of, e.g., accent is best when in overt contradiction to (i.e., a negation of) the paradigmatic tendencies, e.g., when we find final stress in a language which is otherwise penult. But as all accentuations result in naturally perceivable and pronounceable patterns, those instances of grammaticalized accent which are in a conflicting relation to paradigmatic processes are also syntagmatically natural. The interaction with syntagmatic prosodic processes may still alter them, but not necessarily. The same as other prosodic or segmental processes following accentuation in the derivation may still contribute to opacity and, eventually, may lead to the reapplication of accent processes.15 15. Hierarchical Structure All recent theories on accent and rhythm, both tree and grid models, agree upon the point that a prosodic structure of a terminal element like the word is hierarchically organized, and I would not question this assumption — it has been well established for nearly 2000 years and is a commonplace in related fields like musicology. A structure of a four-four time has an inherent rhythmic pattern of the type represented in (10) in which the first and the third note possess an inherently stronger intensity than the second and the fourth and in which the first beat is stronger than the second one. And the resulting pattern {6060} is one of a series of well-formed rhythmic structures that have found a formal repre­ sentation in occidental musical notation, but which express natural principles of rhythmic perception. A more detailed presentation of a corresponding model will follow below in section 18.

MORPHOPROSODY

211

(10) The inherent hierarchical structure of a four-four time in comparison to a metrical tree and a sequential notation

But recent generative theories make a further claim, namely that a given hierarchical prosodic structure is derived as such any time it presents itself and that the chronology of the derivation of a hierarchical structure exactly matches the structure itself. In other words, a moraic structure is built on vowels, syllables on moras, feet on syllables, word accents on feet, but one layer is not only 'built' on the immediately preceding one, but is also directly derived from it. 1 6

I have presented at some length an immanent critique of this model in my Habilitationsschrift from 1991 and in Hurch 1993 and I will limit myself here to outline just a few aspects of my critique on the strict derivational layering of secondary and primary accents. The discussion of prosodic phonology has changed in the past 15 years, mainly around the question of uniformity of accent derivations, but a primary accent is still derived as 'the intensity relation which holds between secondary accents'. To make this point clear again, I do believe that there is an intensity relation between a primary and a secondary accent, and that intensity relations are hierarchically ordered in time, but this statement does not make any necessary implication concerning the direction of derivation. On the contrary, there are good arguments to assume that what is a primary accent is also primary in derivation. And that this assumption does not only hold for lexical accentuation. One of the major changes from Hayes 1980 to Halle & Vergnaud 1987 is that the number of non-hierarchical primary accents has raised considerably. 16 McCarthy & Prince (1995), cf. also the presentation in Kenstowicz (1994), allow for certain extrametrical syllables which are not dominated by foot structures but directly by the prosodic word. But they still call it a "loose interpretation" of the hierarchy and they do not essentially change the bottom-up directionality of the derivation.

212

TRUBETZKOY'S ORPHAN

But they are explained on the one hand through a higher number of lexicalizations and thus through a more constrained use of the concept of extrametricality, and on the other hand through different constituent restrictions. In a given word accents inherited from somewhere else in the derivation may keep their position through the effect of the so-called 'faithfulness principle', which certainly has a realistic basis but the application of which remains rather unclear. But in the Halle & Vergnaud model there is still no room for what has been described above as morphological accent (another instance of non-hierarchical derivation) and, in fact, comparable instances like the Spanish verbal stress are described as lexicalizations (Roca 1992). A treatment of cases like the Basque dialectal plural formation is not to be found, and I suppose that the only way of integrating it would be through extrametricality, the primary function of which seems to be, to quote Goldsmith (1990:203) "to show, simply and directly, ways in which forms need to be adjusted or modified in order for our simple inventory of rules in metrical theory to work correctly". There is something that prevents natural phonologists, however, from doctoring representations in order to allow for the effect of the phonological rule. But, more importantly, if there is a difference in representation and if this difference is meaningful, then primarily we have to assume that this difference in representation does not happen by chance, and that we cannot transpose it to a different level unless we have good reasons for doing so. The functioning of a rule cannot suffice, as it is not the function of representations to make rules function. But then we have to assume the burden of justification in any instance of the phenomenon and cannot just admit a general and unquestioned procedure for the rest of the analyses. But this is probably one of the main differences between formal and non-formal approaches to linguistics. 16. Bottom-up Analyses Besides the instances of morphoprosody or of grammatical and lexical use of prosodic structures in a more general sense that have been the focus of the previous sections, I offer below a series of comments and observations intended as an immanent critique — in the Frankfurtian meaning — of the simplistic bottom-up analysis of recent metrical theories. These contribute to make a radically new look more desirable. 16.1 There is a series of languages in which variation of secondary accentuation is possible without any meaningful reason. Italian elettricita may serve as an illustration. The two possible accent patterns are elettricita and

MORPHOPROSODY

213

elettricitd11 corresponding to the patterns {60606} and {06006}. The crucial point is that such variations occur only in secondary accentuation and never in the location of the primary accent. In a hierarchical model this means that the variation occurs on the foot and not on the word level. But if the word level was nothing but the intensity relation of the foot level, then we would have to expect the primary accent to be as mobile as the secondary accent. And this simply would be a wrong prediction. 16.2 There are languages in which the locations of the primary and of the secondary accents follow different principles and thus do not stand derivationally in a hierarchical relation. Unfortunately, the most nicely illustrating languages of this type are not treated in the current studies on metrical theory. In Zoque, for example, main stress predominantly falls on the penult syllable of the stress group and secondary stress (which is assigned only in words with three or more syllables) on the first syllable. The stress group in Zoque generally consists "of a single word [...] or a close-knit phrase[...]" (Wonderly 1951:108, cf. also Knudson 1974). Cf. the examples ?d:wat "louse", ho:ho "palm tree", ?a:nd:sa "orange", minge?tu "he came", he?epikpa "he is breathing", mindange?tutih "you [pl.] came again". This prosodic regularity is subject to some further limitations in the case of monosyllabic stems, certain compounds, etc. which are not interesting at the moment and, moreover, it is also subject to syntagmatic segmental processes which may morphologize the prosodic accent location (vowel lengthening like in the first examples above, or rearticulation of the secondary stressed vowel). The crucial point for which I cite Zoque here is that primary accents and secondary accents seem to be located independently from each other, in the sense that they do not show a hierarchical derivation or any relation of dependence upon each other. As examples like ?d:nd:sa and mínge?tu illustrate, these accents may also show up in contiguous positions. In pitch accent languages such situations are not rare but they cannot be captured thoroughly by a strictly hierarchical model. Examples of languages with stress types that posit analogous problems for a hierarchical derivation are not rare.18 16.3 In stress systems like Papago primary stress normally falls on the first stem syllable, whereas secondary stress "is predictable on the basis of syllable 17 In Italian examples like elettricitá I use the Standard Italian orthography. The two following contrastive accent markings obviously have a purely illustrative function. 18 In Hurch (1991) I have listed some more of them.

214

TRUBETZKOY'S ORPHAN

composition and the phonetic characteristics of contiguous consonants" (Saxton 1963:33). In a VV sequence the second element receives secondary stress if the first vowel is high and the second non-high, otherwise it falls on the first vowel (if unstressed). Single vowels are secondarily stressed if they "precede lenis or sonorant consonants which in turn precede phonetically unstressed vowels" (Saxton, loc.cit.). Thus, ?d?ai ."vari-directional", cícíwíi "to play", tdtcua "to desire", ddapiù "to smoothe", wámí'gi "to raise". Secondary stress also falls on all unstressed /u, a, o/,βdhtdhpi "fish". Such stress phenomena as in Papago in more recent versions of metrical theory would probably be derived non-hierarchic ally in the sense that we deal here with different accent domains for the primary and the secondary accents, but the puzzling fact is that both 'levels' of stressing seem to follow independent 'overt' types, i.e., they are both purely prosodic with the respective domains. 16.4 I shall conclude this series with a brief summary of some of the additional arguments discussed in Hurch (1991) for a more detailed presentation. - The way that quantity sensitivity of accent structures is conceptualized in current metrical theories is oversimplified as quantity is never mapped equally on higher levels of representation in the hierarchy. In Latin, for example, quantity counts only for the foot which — on the word level — carries main stress, i.e., for the last foot of the word, but its relevance is never equally distributed over the complete foot level. There are also languages that make use of quantity for the secondary accents, but I know of no example of equal distribution. - The treatment of languages lacking secondary accentuations as 'covert rhythms' is highly unsatisfactory. The suppression of non-primary accent through whatever procedure like conflation is equally questionable, as derived foot structures are not easily controllable and frequently do not match locations of secondary accents, as in the emphatic accent in French. - Van der Hulst (1992) argues against a simple bottom-up model of stress assignment as there are languages that he calls 'count systems' in which the directionality of foot structuring depends on a global analysis of given stress unit on the word level. In MalakMalak for example, stress falls on the first syllable if a word has an even number of syllables and on the second if it is odd numbered.

MORPHOPROSODY

215

17. Naturalness The prosodic reasons for looking for radical alternatives thus are multifold. Both theoretical and theory immanent inconsistencies of the proposed bottomup studies are the motivation. In the remaining sections I will briefly outline a model of accentuation in the framework of Natural Phonology and give an account of treatment of morphoprosody. One of the starting points of Natural Phonology is that whatever is produced in natural language is natural. And the same holds for prosody.19 That means any given accentual form in any language is natural and corresponds as such to a well-formed rhythmic pattern. Most scholars would agree on this point as long as systems are concerned which in one theory are called overt, in another regular or in still another systems with a prosodically fixed accent. It is not necessary here to quote the standard textbook examples like Czech, Hungarian, Polish, Latin, or more exotic ones like Weri, Southern Paiute or Garawa. If we separate the principles employed by those proposals which offer an overall prosodic analysis from their theoretical specificities, then I think that not only there is no essential difference between tree and grid models in generative metrical theories as Goldsmith (1990) convincingly points out, but probably proponents of the most different schools would subscribe to these principles as the metrical fundamentals employed there are much older than the theories themselves. What is meant here is that for example a stress alternation condition does essentially the same work as a principle of bound vs. unbound feet, a restriction against stress clashes and stress gaps, or as a preference statement over preferred foot size. I will thus mainly concentrate on the treatment of what might be called non-prosodic origin of stress, in particular morphologically conditioned accentuations. 18. Motifs 18.1 The model I am proposing here is rather simple. In its basic idea it follows Stampe's repeated but rather cryptic claim, that no accents are put on words but words on accents or, more exactly, words are put together with accents. However, this formulation is not complete, as words are not put on simple accents but on accent structures. Certain prosodic patterns are perceived as rhythmically even, others not. From here on I will call the former rhythmic motifs, following Riemann's famous musicological classification from 1903. 19

Loanwords may behave exceptionally but the degree of integration then depends, ceteris paribus, on their integration into a prosodic system.

216

TRUBETZKOY'S ORPHAN

We perceive these motifs as rhythmically well-formed independently of whether linguistic or other 'material' is mapped on them. Motifs are rhythmic patterns with differing size, which range from a simple binary foot up to a complete phonological phrase. The size of a motif is not a priori restricted, except through perception: a whole melodic string may be comprised in one motif if we perceive the rhythmic pattern as a unit. Obviously there are prosodic preferences for the size of motifs. The trochee has been claimed to be the least marked foot structure and this is reflected in word size preferences (cf. Dressier 1985b). But type-specific regularities may con­ siderably alter these conditions. Perceptual strategies like figure-ground prefer the sharpest contrast in patterns and thus simple distinctions between accented vs. unaccented, long vs. short, etc., over more refined scalar grading like primary vs. secondary vs. unaccented, long vs. half long vs. short. The syntagmatic changes to paradigmatically preferred structures parallels the situation well-described in segmental phonology. Allophonic processes may well give rise to phonetic representations containing units which — as such — are unpronounceable on phonological grounds (Donegan & Stampe 1979). 18.2 Such motifs are primes, in the sense that they do not arise in the course of grammatical or syntagmatic prosodic processing but that they are present in­ dependently from specific words. They arise in the paradigmatic (or prelexical in the terminology of Dressier) part of phonology by the effect of prosodic processes. I have been presenting these in form of preferences with a more de­ tailed discussion, internal and external evidence, etc. in Hurch (1991) and (1996):20 • binary foot structures are preferred over ternary, the latter over quaternary, etc. • falling foot structures are preferred over rising ones • a binary intensity relation is preferred over a ternary, and a ternary over a quaternary, etc. 20 Each of the following preferences corresponds to one or several processes which all fulfill the function of building and structuring metrical feet. They are not intended to express constraints as recent theories (e.g., Optimality, Prince & Smolensky 1993) suggest. The teleology of the first of the following preferences is fulfilled, for example, both by processes which reduce ternary to binary foot structures as the examples from the Appendix Probi under 8.3. above indicate, but also by simple introductions of secondary accents like in Czech, by anti lapse reactions, as well as by the addition of (silent) beats in stressed monosyllables, etc. The conceptualization as preferences will not imply that such preferences — in contradistinction to Vennemann's (1988) use of preference laws for syllable structure — themselves are part of grammar.

MORPHOPROSODY

217

• peripheral foot structures are preferred over foot structures in non-peripheral position • vowels possessing one of the intrinsic properties that phonetically realize stress (esp. length, tone) are accented with preference over vowels which lack these properties.21 Accentual and rhythmic processes fulfilling these preferences simply delimit the possible concatenation of accented and unaccented elements of varying size to a set of perceptually preferred patterns. Moreover, these preferences and the resulting motifs give a straightforward account for a series of observational facts like the higher type frequency of penult accentuation than of final position which results from the (unordered) application of processes corresponding to the first two preferences. 18.3 Rhythmic patterns exist at different sizes, and the language-specific choice is due to lexical restrictions on actual word length, typological and morphological preferences, etc. The relation between linguistically prosodic and extralinguistic rhythmic patterns is not entirely clear, for example, the relation which exists between word accent patterns and other rhythmic expressions, like folk music, dances, etc. A tentative model of paralleling rhythm in language and in folk music has been established by Boroda (1991, 1992, 1973 n.v.) who operates with 'metrorhythmic segments', deduced from Riemann's basic principles of the rhythmic structuring of classical music. But it is important to notice that the linguistic rhythmic perception is just a subpart of a more global human capacity. In other words, speakers of, let's say, Czech would perceive a motif of the type {60060} as rythmically wellformed even if it is not part of their linguistic prosody. 18.4 The mapping of specific words with prosodic patterns selects those patterns which reflect the size of a given word and then follows the languagespecific restrictions which may be only prosodic — in case of a language with so-called fixed accent—, or they may follow morphological and/or lexical principles or idiosyncratic accent specifications. This mapping can probably be best compared to a tier model. The rhythmic motifs are present on a separate metrical tier and words are associated to them 21

Donegan (1978) formulates an analogous "rich-get-richer" principle for segmental phonological processes which states that, e.g., high vowels which intrinsically possess more colour tend to become tense more strongly than non-high vowels, low vowels which already have higher sonority tend to become lax, etc. In Dressier's (1996a) semiotic model this is a subcase of figure-ground contrast sharpening.

218

TRUBETZKOY' S ORPHAN

according to size, inherited lexical or morphological specifications or finally purely prosodic principles. Paralleling the above representation of quantity, let me illustrate this mapping with examples already mentioned. (11)

Mapping of moiphoprosodic accentuations

These two examples differ in the source of accentuation. In the former, we find a fixed lexical main stress, which together with the number of syllables chooses two possible accent motifs. No other motifs are available for Italian given the lexical specification. And indeed, these are the two possible accentuations of this word in Italian. In the second example the two different patterns have morphological implications. As stated above, in the Zegama dialect the singular noun paradigm is accented on the second syllable, the plural on the first syllable of the base form (stem).22 We cannot state generally that one of the two patterns itself carries morphological implications. For the sake of ease of derivation we should assume that the singular form is derived prosodically, the plural morphoprosodically. But the pattern {600} by itself has as little morphological meaning as a German vowel u or ö or ail or whatever. In Basque, which is agglutinating, both the singular and the plural are derived from the un­ determined mugagabe paradigm, the processual implication thus does not exist so much between the two forms quoted here but rather with the base from ardi, corresponding to the undetermined absolutive. It is a stress retraction to the first stem syllable which forms and signals the plural. In this moiphoprosodic process of mapping, thus, it is neither the word itself, nor the prosodic pattern itself, but just the mapping of the specific word form /ardintzat/ with a specific accentual pattern {600} which produces the plural. If in a language like Italian in segmentally identical strings as ancora, once mapped with {600} and once with {060}, we find two patterns; they cannot both be attributed to the same source. Formally both are derivable prosodically even with some plausibility. Italian has a strong prosodic tendency towards penult accent. But we also might suggest a proparoxytone structure in case the antepenult syllable is heavy. Without going into detail I consider Italian a 22

For further details cf. (3) above.

MORPHOPROSODY

219

language with moderate quantity sensitivity23 in which a heavy antepenult syllable may exert some attracting but no blocking force. The quantity of the penult syllable possesses both. Paradigmatic processes of Italian, other things being equal, establish the {060} motif. Everything else must be mapped on non-prosodic grounds even if itself it {600} turns out as a syntagmatically well-formed pattern. 18.5 For more clarity, take the Italian pentasyllablc elettricita again with a morpholexically specified final accent. This specification has a series of morphological characteristics co-signaled by the deviant accent structure (formal invariability, gender). Among the well-formed pentasyllablc motifs there are some which do not enter the group of linguistically relevant motifs for non-derived or non-inflected words in Italian, like {60060} or {60600} and others. Pentasyllables show five possible motifs and this is due to the number of syllables which allow for more variation in the location of secondary accents within the pattern than at a lower rate of syllables. Altogether a corpus of 4000 non-derived word patterns under investigation presents 334 pentasyllables; the overwhelming majority of them (264) show the pattern {60060} besides a small number of 8 items, still with penult accent of the type {06060} which thus should perhaps be better integrated as a subgroup into the first pattern type. The bigger the size of the pattern, the stronger becomes the tendency towards penult accent. I do not have a readily tested explanation for this fact.24 The number of antepenult accented pentasyllables decreases to 42 items which 23

As I have proposed elsewhere, quantity sensitivity is not a binary type in the sense that a language is either fully quantity sensitive or quantity insensitive, but prosodic systems may be quantity sensitive to different degrees. In Latin a heavy penult syllable attracts accent, a heavy antepenult attracts accent in case the penult is light. In Italian, a heavy penult attracts accent too, but a heavy antepenult does not. This difference has caused rather diverging descriptions of Italian. In a CV.CVC.CV structure any other accentuation than penult would be paradigmatically impossible. An oxytonic pattern would have to need a lexical or a morphoprosodic specification. A proparoxytonic structure would simply be impossible. Insofar quantity blocks certain accentuations. But a long antepenult syllable does not attract stress. In CVC.CV.CV, quantity does not block a penult accent. In such structures Italian stresses the penult syllable. It may simply facilitate the violation of the penult accent. And this simply results from the rich-get-richer in prosody and is thus a consequence of a natural prosodic process. And this process may operate to a varying degree and, in accordance with those effects, we may classify languages as more or less quantity sensititve. 24 I suspect that a higher rate of syllables per word favours a more rigid application of paradigmatic prosodic processes which facilitates perceptual processing. From cognitive processing we know that the more complex a given structure is, the more we rely on the dominant principles of decoding.

220

TRUBETZKOY' S ORPHAN

is less than one sixth of the corresponding penult items.25 The finally accented patterns split into two interrelated subgroups, namely {60606} and {06006}, together they amount to 20 items, again a small number dependent on the parallelism to lexicalized suffix accent. The only morphoprosodic specification a word like elettricita carries concerns the location of the primary accent. It then fits both of these types and in fact both of them can be realized invariably. It might be adequate to subsume the two final accentuated pattern as subtypes of one motif. The difference between these two pattern is subphonemic, so to speak, or subprosodic, as no distinctive accentuation is based on the different location of secondary accents in Italian. It is rather astonishing that generative tree or grid models must postulate essentially different structuring principles in order to derive similar results. 18.6 The simplest stress systems lack any kind of context sensitivity, i.e., they lack any interaction with segmental or other prosodic categories, and chose only those patterns which are produced by the preferences of binarity, prominence direction, structural depth, and peripherality. A simple regular system, like Czech as described by Jakobson (1971[1926]), would thus generally map bisyllabics onto {60}, trisyllables onto {600}, tetrasyllablcs onto {6060} with a possible secondary accent on the third syllable, pentasyllables onto {60060}, hexasyllabics onto {606060}, etc. Even if such a system is the most easily describable for all theories, a difference must be assumed between the proposal made in the present paper and current tree theories. The latter make claims about prosodic or non-prosodic shapes using principles like headedness or left-to-right vs. right-to-left application of pro­ cesses. But they miss the explanatory force of why a system like Czech corresponds to more natural restrictions than other completely regular systems, i.e., that there is a preference of binary over non-binary, that there is a preference of peripherality over non-peripherality, of falling over rising. And Czech corresponds to most of these preferences. 18.7 Paradigmatically, all the possible patterns (or groups of patterns) or motives of the Italian pentasyllables respond to what we perceive as prosodically natural. But this does not mean that all these patterns have equal effects in the syntagmatic mapping, as only part of the syntagmatically realized accentuations correspond directly to one language specific paradigmatic pat­ tern, namely those instances which are also syntagmatically derived on purely prosodic grounds. And it is important to note here that all languages, also those In the groups of the tri- and tetrasyllables, the relation is about 3:1.

MORPHOPROSODY

221

like Italian which possess what we call traditionally a free accent, have clear syntagmatic prosodic preferences (e.g., basically the paroxytonic rhythm; see above). This view can directly be paralleled in the segmental part of phonology. Any allophonic realization X a of a phoneme Yp theoretically can be phonologized and a phoneme Xp with the corresponding phonetic/phonological properties must be a realistic output of paradigmatic (prelexical) processes. Thus, if an allophone X a turns up, it is the result of a natural process and as such natural, even if Xp in this specific language is not among the linguistically (phonologically) relevant underlying entities. Segmental morphophonemic rules only have phonemes as their output, not allophones. Correspondingly, also morphoprosodic rules may produce only patterns which have some kind of distinctive value. This holds also for other levels of prosodic analysis: for a language in which vocalic quantity is only allophonically distributed and, for example, secondarily associated with stress (as in Italian), quantity cannot be an in­ strument of encoding morphological regularities. For accent in Italian this means that the last three syllables of the word are the paradigmatically (prelexically) possible positions of the primary accent (for prosodic and lexical principles) and morphoprosodic accentuation is limited to this 'window' (in a slightly varied use of this term borrowed from Halle & Vergnaud 1987). The functionality of morphological accent is best if it is a negation of prosodic accent. Morphoprosodic accentuations have the characteristic that there are other possible accent forms of the same string of segments which are not only possible but which are prosodically preferred. This obviously does not mean that a morphological accent cannot be 'neutralizing'. The point that accents of lexical or morphological origin are syntagmatically as natural as purely prosodic accents finally is also demonstrated by the fact that segmental phonological processes refer to accentual units as contexts independently of the origin of the latter. Only a model like the one presented here allows for a thorough explanation of why lexical and morphological (i.e., nonprosodic) accents are prosodic.

Another view of Prosody and Morphology: Comments on Hurch Glyne Piggott McGill University Hurch's paper is both a critique of standard generative theories of prosody and a proposal for a new model of stress/accent determination. The criticism of the standard models identifies many failures. Among them, the following is perhaps the most glaring. First, the paper criticizes 'recent generative prosodic theories' for failing to properly integrate into linguistic descriptions the fact that stress and accent can be associated with a variety of grammatical functions. Good examples of such a correlation are the well-known Spanish patterns in which stress can be penultimate in some paradigms (e.g., dmo "I love") or fixed either before the tense marker (e.g., amdbamos "we used to love") or on the tense marker (e.g., cantare "I will sing", cantaremos "we will sing"). This paper even suggests that the approaches represented in works such as Halle & Vergnaud (1987), Roca (1992) and Hayes (1994) are inherently incapable of capturing prosodic features that are associated with morphological patterns. Other failures of the standard model identified in this paper follow from the inappropriate treatment of primary stress as nothing more than the enhancement of secondary stress. Such an approach obscures the fact that secondary stress placement can vary without any variation in primary stress. It also results in inadequate treatment of the different principles that regulate primary and secondary stress placement in some languages. The paper attributes the failure of the standard generative model of prosody to two things. First, it claims that in the standard model the prosodic units are not recognized as autonomous. Secondly, it asserts that standard metrical representations are inappropriately constructed from the bottom up. The paper then gives a rough outline of a prosodic model that is supposed to provide a better treatment of 'morphologically conditioned accentuations' and other aspects of prosodic structure. Before commenting on Hurch's theory of prosody, I would like to challenge his claim that prosodic units are not treated as autonomous in the standard generative model of prosody.

MORPHOPROSODY

223

McCarthy & Prince (1990) outline a theory of how prosodic structure interacts with morphology, especially in the description of templatic phenomena such as reduplication and circumscriptional phenomena like infixation and truncation. They identify a set of autonomous prosodic units, the mora (μ), the syllable (a), the foot (F) and the prosodic word (PWd) which can all be associated with morphological functions. Their proposal has generated considerable research activity, resulting in a substantial number of published and unpublished papers (e.g., Inkelas 1989, McCarthy & Prince 1990; Archangeli 1991; Black 1991; Lombardi & McCarthy 1991; Crowhurst 1992). The evidence for the autonomy of each of these units and its role in morphology is now quite impressive. I will review some of it. In Choctaw, a completive verb form called the y-grade is marked by the gemination of a consonant. (1)

Base tayakci kobaffi atobbi binili falama

Completive tayyakci kobbaffi attobbi binniili fallaama

"to be tied" "to brea" "to pay" "to sit" "to return"

Lombardi and McCarthy relate the geminates to a morphological process of prefixation to a circumscribed portion of the base. Both the prefix and the circumscribed portion of the base are prosodically defined; the former is a mora, the latter the (binary) foot (2a). This prefix is then realized by spreading the first consonant of the circumscribed material leftward (2b).

The vowel lengthening which occurs in the plurals of certain Maori nouns (e.g., tangata "person" vs. taangata "people", wahine "woman" vs. waahine "women") seems to be a consequence of mora-prefixation. Several examples of syllable affixation are cited in the literature on re­ duplication. Pangasinan, an Austronesian language, provides a good example of plural marking by syllable prefixation.

224

(3)

TRUBETZKOY' S ORPHAN

bdso ba-bdso lápot lo-l όpot bálbas ba-bálbas

"glass" "rag" "beard"

According to some analyses, French provides an example of a different use of the syllable in morphology. It is considered to be the feminine desinence, responsible for the appearance of some word-final consonants (e.g., petit [pəti] vs. petite [pətit] "little"). A more convincing example of the syllable in morphology is provided by the Mukogean language, Koasati. Data in Martin (1988) show that roots in certain plural verb forms are systematically shorter than in the corresponding singular. (4)

Singular pitaf tiwap misip tipas atakaa icoktakaa

Plural pit tiw mis tip atak icoktak

"slice up the middle" "open something" "wink" "pick something off "hang something" "open one's mouth"

The truncation of the last syllable of the root accounts fully for the realization of the plural. Notice that at the segmental level the truncation affects the final VC and not the final CVC sequence. This falls out naturally from a morphologically-driven process of final syllable truncation.

The segments that are 'stranded' as a result of syllable truncation get syllabified if they can be accommodated by the remaining syllable(s). Consider, next, the evidence for the foot in morphology. The foot is usually associated with stress assignment. However, McCarthy & Prince (1990) propose an analysis of broken plurals in Arabic which makes it clear that foot structure is independent of stress. According to this account, the broken plural is a manifestation of the canonic iambic foot which consists of a sequence of light and heavy syllable. McCarthy & Prince (1990) identifies the Nicaraguan language, Ulwa, as manifesting another interesting use of the iambic foot in morphology. In this language, the base to which the possessive

MORPHOPROSODY

225

marker is attached has a variable shape which can only be described in prosodic terms. (6)

a. has suulu b. sana siwanak c. sapaa anaalaaka karasmak

bas-ka suu-ka-lu sana-ka siwa-ka-nak sapaa-ka anaa-ka-laaka karas-ka-mak

"hair" "dog" "deer" "root" "forehead" "chin" "knee"

The suffix is attached to a portion of the root which has a variable shape: CVC, CVV, CVCV, CVCVV, CVCVC. All of these shapes instantiate the iamb, which, in a sense, subcategorizes the possessive suffix. One might think of this suffix as having the iambic foot as one of its lexical properties and the roots to which it is attached must conform to this lexical requirement. Additional examples of the foot in morphology can be provided from crosslinguistic patterns of reduplication. The only other prosodic category which is reflected in morphological alternations is the minimal prosodic word. The hierarchical structure of prosod­ ic representations requires that the minimal word contain at least one binary foot, which can be satisfied by a single heavy syllable or a sequence of two light ones. Some of the best evidence for the autonomy of this category is provided by augmentation patterns which apply in languages like Mohawk, Choctaw and Lardil (Piggott 1993). Augmentation has the appearance of an affixation process that increases the size of words to at least two syllables. In Lardil, for example, uninflected stems like ter and wun surface as tera and wunta, respectively. This derived shape is predictable from the enforcement of a minimal word requirement on every word. A different exemplification of this category is found in Woleaian (Sohn 1975). This language in which syllables are normally open is affected by a general process which devoices word-final vowels and sometimes results in their complete loss. However, in underlying bisyllabic words, the vowel reduction is accompanied by a lengthening of the remaining vowel. (7)

a. ita mata b. ita-la mata-la

[iite]/[iit] [maate]/[maat] [itale]/[ital] [matale]/[matal]

"name" "eyes" "Ms name" "his eyes"

This vowel lengthening makes sense, if there is a unit of structure, independent

226

TRUBEIZKOY'S ORPHAN

of segmental content, that regulates the size of words. Since the minimal word must contain a binary foot, it must contain at least two moras or two syllables. In a number of reduplication and truncation patterns discussed in Piggott (1993) and It6 & Mester (1992), a morphological unit identified as the minimal word plays a role. Pangasinan provides an example, again involving a pattern of plural formation. (8)

Singular pus a lusor manok tamuro paŋando

Plural pusapusa lusolusor manomanok tamutamuro paŋapaŋando

"cat" "cup" "chicken" "forefinger" "middlefinger"

The fact that the prefix has a bisyllabic shape is imposed by the minimal prosodic word. Note that there are no systematic patterns of reduplication in any language in which the affix is trisyllabic. In addition to its role in affixation and truncation, Demuth (1994) points out that in Sesotho the minimal word can be a constraint on the realization of functional categories in early grammars of both English and Sesotho. There is, therefore, fairly compelling evidence that all prosodic categories can express 'morphological regularities'. These categories even include the syllable, contrary to Hurch's claim. His suggestion that the syllable is "not an underlying unit represented in our permanent memory" is not consistent with the evidence from reduplication and truncation. It is also contradicted by the findings by Petito & Marentette (1991) that the manual babbling of deaf children contains a units which has similar properties of the syllable in the vocal babbling (ba-ba-ba...) of hearing infants. The hypothesis that the prosodic categories are primitive linguistic units has informed a considerable body of research for almost a decade. It should, therefore, not be a surprise to find that these units may be active lexically, morphologically, syntactically and phonologically. If primitive non-prosodic units like segments and features can have morphological functions, we would expect the same behaviour from primitive prosodic units. The occurrence of a paradigmatic stress in Spanish should, therefore, not be unexpected. If we assume that stress identifies the head of a prosodic unit (e.g., head of a foot, head of a prosodic word), the occurrence of fixed stress on the syllable preceding the tense marker in imperfective forms reduces to the lexical subcategorization of the tense marker by the head of the prosodic constituent. It is also possible for the location of primary stress to be determined independently

MORPHOPROSODY

227

of secondary stress placement; primary stress identifies the head of the prosodic word usually in relation to an edge. The fixed primary stress on the final syllable of the Italian word which varies phonetically between èlettricitd and elettricitd reflects a lexical restriction requiring the head of the prosodic word to coincide with the last syllable of the word. The location of the heads of other feet (i.e., secondary stress) may then vary independently of placement of primary stress. Such an approach is, in fact, reflected in the non-derivational analyses of stress in Optimality Theory (see McCarthy & Prince 1993, Prince & Smolensky 1993). The criticism of the generative treatment of word prosody is directed in part at a model in which metrical structure is built from bottom to top. However, bottom-up constructionism has been dead for some time, and, even if it is not yet buried in some minds, it cannot be brought back to life. Let me now comment more directly on how the paper proposes to account for prosodic organization. It proposes a model in which words are mapped to prosodic patterns called 'rhythmic motifs'. These rhythmic motifs are primes "in the sense that they do not arise in the course of grammatical or syntagmatic prosodic processing". But the selection of motifs cannot be free, because the stress pattern of words is not completely idiosyncratic. How is the selection determined? The paper proposes to regulate the shape of motifs by a set of preference 'laws' which among other things rank binary feet higher than ternary and quaternary ones, select the trochaic pattern over the iambic, single out heavy syllables as inherently stressable and attract stress to the edges of words. Careful consideration of the 'preferences' reveals a striking similarity to generally accepted constraints on prosodic organization in current metrical theories. The shapes of motifs in Hurch's model of prosody would correspond to the prosodic shapes that would be generated by the principles of current generative prosodic theory. Consequently, it appears that much of the machinery required by Hurch's theory of prosody already exists. Even the reference to 'preferences' is not foreign to current prosodic theory. Prince (1990) recognizes the equivalent to 'preferences' in the form of hierarchies of well-formedness of prosodic structures, according to which some rep­ resentations are better than others as instances of particular prosodic categories. It is possible to identify some differences between Hurch's proposal and current generative theories of prosodic structure. The entities that are equivalent to the prosodic categories of the latter do not seem to be defined in absolute terms. Thus, this paper refers to possible foot shapes that are binary, ternary, quaternary, etc. A more restrictive theory of foot structure is generally adopted in current generative theory: the foot is maximally binary. Hurch's theory of

228

TTUJBETZKOY'S ORPHAN

foot structure seems to allow for too many possibilities. I know of no language that systematically selects quaternary foot shapes. In general, there is a salutary conclusion to be drawn from this reading of Hurch's paper. At a fundamental level, there is a convergence of ideas on what a theory of prosody must account for and surprising similarity in proposals about the shape of the theory that is required.

Reply to Piggott Bernhard Hurch Glyne Piggott filters out of my contribution essentially two argumentational strategies which undoubtedly reflect the aims of my paper. Most of his response is dedicated to showing that generative theories of prosody — pace my objection — do not fail to recognize prosodic units as autonomous. Secondly, he briefly summarizes the approach within the framework of Natural Phonology concluding that most prosodic principles elaborated here, in one form or in another, are also present in generative treatments and that essential portions of the theories do not seem to be as distant from each other as is frequently assumed. The parallelism I draw in §15. with musical notation and the explicit statement that the basic principles of metrical structure have been known for about 2000 years renders exactly this point, and equally, as Goldsmith (1990) demonstrates, the different models of tree and grid representations address very much the same issues, although the algorithms look different (cf. §17). Apart from minor doubts about the explanation presented in the approach that is covered by the term 'prosodic morphology' in generative theory, there is one major point in my exposition which is completely neglected in Piggott's Response, namely that it is essentially part of the theory of naturalness to relate phonological and morphological naturalness in a theory of morphoprosody coined on Dressler's (1985a) framework, i.e., the interaction of the two components can best be understood as a reflection of antagonistic tendencies on formal and functional grounds (Wurzel 1980b, 1984a). What has been outlined here under the heading of 'morphoprosody' thus is not simply an enriched version of prosody including those instances which previously have been neglected or ignored, but the attempt of describing in a principled way the particular position this subcomponent holds, i.e., to il­ lustrate how a morphological function may lower prosodic productiveness and opacity, how these instances arise in diachrony, how morphoprosodic phenomena can be related to 'phonological' prosody, how syntagmatic and paradigmatic prosodic processing are to be related analogously to phonemic

230

TRUBETZKOY'S ORPHAN

and allophonic parts of segmental phonology, etc. So far, I see rather crucial differences between the issue presented here and prosodic morphology as developed recently in generative theory. The same critique Stampe (1973) and Donegan & Stampe (1979) direct towards lumping together phonology and morphophonemics holds in the present context for not distinguishing between prosodic principles exploited for morphological functions and prosody proper. I doubt that the bottom-up analysis, well-established in generative theories of prosody, "has been dead for some time, and, even if it is not yet buried in some minds, it cannot be brought back to life" (Piggott). In the picture generally assumed, those instances which are not derivable hierarchically must be derived autonomously, but what can be described hierarchically, even if in a 'loose' hierarchy (Kenstowicz 1994), is derived as such. And much of the discussion going on in and around generative prosodic morphology is rather a discussion of morphotactics, more specifically, the integration of prosodic units (in addition to segments) into the description of the shapes of morphological units. As a last point let me turn to the existence of the syllable as underlying unit which is an argument Piggott insists upon. The position of Natural Phonology (Donegan & Stampe 1978) that we have little or no evidence for the syllable as a non-derived underlying unit does not mean that syllables do not exist altogether. On the contrary, it is assumed that syllables do arise in processing, and once they have been established, they do have a central function as prosodic units. But still their 'autonomous' character, advocated for anew by Piggott, is not presented here in a convincing fashion. In fact, none of his examples is simply based on the syllable, and I assume that these are the examples bearing the strongest argumentational power. In Pangasinan the plural reduplicates the first syllable but only if it has a CV structure, otherwise it only copies the core. Thus, what is prefixed has syllabic status, but it is not the syllable itself which is prefixed. The in­ terpretation of French phonotactics as linked to a specific feminine syllable structure could hardly hold a more general derivation. It is French consonantism which is crucial, and this influences upon the syllable. But I cannot follow how to construct the argument in favour of the syllable. In Koasati truncation plurals finally count one syllable less than the corresponding singular forms. But what is truncated hardly corresponds to what is termed a syllable. In fact, the analysis proposed by Piggott with reference to Martin employs the stranding of segments which are systematically part of the deleted syllable and which are not deleted. Again, I would not doubt that we can employ prosodic

MORPHOPROSODY

231

categories for the explanation of this truncation but the evidence for the syllable does not 'fall out naturally'. I do agree with Piggott that approaches to prosody have more resemblances across the different theories than we would expect at a first glance. But still the discussion of the differences might contribute to a mutual criticism, appreciation and, hopefully, improvement.

Morphoprosody Discussion HURCH: Let me just make some brief remarks. The first argument I am about to take up is the one on the syllable. The theory which I am presenting here would not negate in general the existence of the syllable. It negates the syllable as an underlying unit, and thus the use of the syllable in morpho­ logy. Looking at your examples, here under number (2) and (3), I am not exactly sure how to arrive at these syllable divisions, like the third example in the number (2), for example, balbas. Same problems in the number (3), I think. Till there are more general remarks at the end, I would say that the number, obviously, of motifs is limited. It is not that there is an unlimited number. I don't think the length is unlimited, I don't think the length of words is unlimited. I think there are very clear preferences among the structure of motifs which also limits the number of possible rhythmic motifs. The unit 'Word' is not a unit in grammar which is completely open. There are preferences for word size depending on typological affiliation, etc. I think there is simply not an open number of motifs. I think that the restrictions which exist on the number of motifs are partly morphological or in the nature of lexical structures and partly restrictions which are prosodic in nature, in the sense that probably the one unit should not exceed the number of seven syllables, one of the proposed magic sizes for units with an internal prosodic structure. The point of the solution in the variation of secondary accents, what I have proposed in the Italian examples elèctricitd,èlectricitá,is that there is a lexical accent in the last syllable, which is inherited in the prosodic analysis, an assumption which I think you would subscribe to. This lexical accent has a whole series of morphological implications, like words that have final accent in Italian are feminine, they don't have separate plural suffixation, as far as they are pluralizeable, electricita isn't, but onecita for example. This, finally: accented nouns have very specific morphological behaviour. This accent is somewhat inherited and comes as a lexical accent into morphology and from all the patterns of pentasyllable motives which Italian might choose, if it is given that it's the last syllable which has the primary accent, then there are only two possible motifs to choose from. This

MORPHOPROSODY

233

is how I would treat it. PiGGOTT: Can I comment just briefly? Just three points, just clarifications. The syllable that is manifested in prefixation is an instance of the core syllable. Syllables come in various types, of course, so this looks like the smallest syllable you can get. In the terminology of standard morae theory, a core syllable has an onset and the nucleus contains one vowel The reason for including balbas is to prevent anyone here from assuming that in fact, what you're doing is simply repeating the first syllable of the root. That's not what it is. It is prefixation of a prosodic category, which you then have to satisfy in a certain way. It's not just the first syllable of the root. And the second example is even clearer, the Kowasati case. However, let me ask a question about how you determine the length of any word in any language. How do you know the length of a word in Ojibway? How do you know the length of a word in Tubatulabal? How do you know the length of new words in Tubatulabal, given the possibility of a phenomenon like compounding? Those are questions that will have to be addressed if you assume that the fact that there is a finite number of morphemes determine the length of words. But I don't think that that is really a crucial issue. I think, the more crucial problem comes with how to choose a motif, how do motifs get chosen for a language. How do you decide how to choose motif x over motif y, both of which are motifs for a given language? Now, with respect to your analysis of the two patterns for Italian, the question I am asking of you, which is the question that I would ask even of standard generative theory: why is it not the case that you have a fixed accent at the end of that word and then you choose two of the patterns that are compatible with the fixed accent? Why can that choice not be determined morphologically? HURCH: Because secondary accent is much less reliable than primary accent and, for that, is a much worse means of expressing a generalization in a different module of the grammar. It is saving us. I think this would be the simplest explanation. PiGGOTT: It's not formal, it's functional. HURCH: Functional, yes. I mean it's translatable, one into the other in this sense, form and function. You could say whatever inner hierarchy further up has more salience translated in this very same way. Whatever has more pluses is more salient. PIGGOTT: If I may just respond very briefly, syllable has more salience than just a single consonant, at some acoustic level, psychologically. That doesn't exclude suffixation of single consonants as a way expressing morphological difference. So salience is not going to be the only answer.

234

TRUBETZKOY'S ORPHAN

HURCH: No, salience is not absolute, is not an absolute criterion. I mean, you have a gradation of salience. We are working in completely different levels. The syllable is a unit at the prosodic level and the segment isn't. I mean, this would be tantamount to claiming, why don't you use syllables as features for segments. You can't mix up the levels of analysis, I think. I suppose it was ironical, your question. PIGGOTT: It was. KIPARSKY: Before I ask my question, I have a comment on the role of sec­ ondary accent. There are lots of languages in which it is precisely the secondary accent which is morphologically irregular. In Finnish, for example, the primary accent is mechanically on the first syllable, and the secondary accent is a complicated mark of the morphological organization, and all the complexity of Finnish accent... PIGGOTT: I knew you would have an example. KIPARSKY: ... lies there. The same is true of Estonian. My question concerns what I think might be a contradiction in your approach. As a Natural Phonologist and a Natural Morphologist, you are, I think, committed to a processual approach, as opposed to, let's say, a constraint-based approach of the sort advocated by Professor Singh. On the other hand you have been arguing for a model of stress assignment, or, I should say, of prosodic structure in general, where rhythmic motifs are imposed at different levels of analysis, and not necessarily according to some bottom-up derivational scheme. Your argumentation converges very nicely with recent argu­ mentation in Optimality Theory. I am the first to mention this. (This is the first workshop or conference I have attended in the last two years where Optimality Theory hasn't been mentioned!) The necessity for top-down (or middle-outwards, or simultaneous) imposition of constraints at different levels has usually been associated with constraint-based models. It is in fact taken as an argument for constraint-based as opposed to processual models, so I am wondering exactly how does one reconcile in your model the processual approach with the kind of top-down construction you propose. Now, one way in which that might be done in the case of primary stress is for it to have some kind of representational independence from secondary stress. Your remarks on Italian seem to me to hint that primary stress is just some entity that is represented independently of, let's say, foot structure or secondary stress, so that you could assign it lexically or morphologically by some template or motif that is independent of other stress mechanisms in the language. That seemed to me then to be in conflict with your idea that stress is in fact hierarchically organized as in your picture (6). So how do you, in

MORPHOPROSODY

235

fact, see this contradiction resolved? HURCH: Well, the point is that I think that what I call motif is produced by prosodic preferences or processes which I didn't have the time to explain here. I think that there are preferences like preferences on foot structures, that binary structures are preferred over ternary structures, that ternary are preferred over quaternary etc., that there is a preference which would probably cause more discussion, that there is a preference for falling over rising foot structures, that there is a preference which might correspond to what Patricia Donegan called the 'rich-get-richer' principle in segmental phonology, that is, that those units which have an inherent quality of what corresponds to the realization of accent like in certain pitch accent language tone registers, that accent in the presence of these units preferentially falls on one syllable which already carries this characteristic. There would be a preference that a binary relation of intensity between elements is preferred over a ternary relation, a ternary over a quaternary. There is a series of preferences like this, expressible as processes which produce the motifs, and I think that motifs, I would call them 'primes', not 'primes' but rather 'primitives'. Primitives in Natural Phonology are considered to be the output of processes. That is the basic idea, I think in Natural Phonology and what is then left as being non-processual is simply this action of the mapping of the words onto the patterns, and I would not say that all locations of primary accent are somewhat inherited from somewhere or are marked clearly. Languages like Italian, in this model, I would propose that they have certain accent positions which are their prosodically preferred accent positions. Particularly for Italian, all those stress patterns for bisyllabics, trisyllablcs, tetrasyllablcs, pentasyllables, etc., which have an accent on the penult syllable, are the ones which are preferred in the absence of other specifications. KIPARSKY: To me a preference has something to do with a choice you make between alternatives, and you seem to me to be equating a preference with a process, or saying that you can interpret a preference as a process. But a process means doing something to something else, that is producing an output from an input. That is, it seems to me to be an operation. Isn't that the sense in which process is used? HURCH: This is the sense. KIPARSKY: Okay, so what is the input and the output of the process that produces these patterns? HURCH: I would say a string without intensity relations, for example. The output might be a string containing intensity relations.

236

TRUBETZKOY' S ORPHAN

KIPARSKY: And are the processes ordered? Do they apply one after the other? HURCH: They are partially ordered. I would say, for example, that the binarity of full structure is clearly ordered before falling or rising. KlPARSKY: But then you can't have Italian. Because if you represent primary stress as prominence at the top of the hierarchy then you will have to assign the hierarchical full structure before primary stress. And so you will return to the paradox of bottom-up assignment as soon as you introduce processual terminology or processual methods. HURCH: The point is, I don't deny that motifs themselves have a hierarchical structure, but what I deny is that this hierarchy is derived individually with every derivation of an accent pattern of a single realization of a word. KlPARSKY: Yes, I understood that, but you derive the hierarchy itself through a series of processes, isn't that correct? What I don't understand is how you can derive a hierarchy by processes without going bottom to top. What does the first process actually do if it doesn't construct the bottom layer of the hierarchy? How can you construct the top layer of the hierarchy without having the bottom? What does it actually do? HURCH: I would construct a motif hierarchically. That's not the problem. It's just that I don't need the relation which exists in a word and then represents exactly or reflects this very same hierarchy „ SINGH: Just two brief questions. One, you said that the syllable doesn't seem to play any role in your view in these morphological processes. Is there any reason for which it doesn't? Let put this in contrast with what Professor Piggott was trying to point out, that apparently everything that is available in matters of phonological organization,— I took the thrust of Professor's Piggott's examples here as showing that practically anything and everything that can be legitimately said to be a matter of phonological organization is available in an autonomous way for the morphology,— I had the impression that you wanted to say that this is true about other things but perhaps not for the syllable. Suppose that were empirically true. Is there any reason for which it must be true? HURCH: I don't know if I get your question right, but I would say that syllables as syllables are not underlying units present in our permanent memory. SINGH: AS opposed to these other types. HURCH: As opposed to these other units which may be, which are not necessarily, but which may be present. SINGH: What would be the reason for that distinction in your theoretical view? Suppose, as I say, that the distinction is empirically correct, but the mere fact

MORPHOPROSODY

237

that it is empirically true does not explain why it must be true. HURCH: I would probably refer to very classical principles like distinction, distinctiveness. SINGH: And you think that would be sufficient and to say that these other things exist as autonomous units and the syllable doesn't. JANDA: I want to suggest a possible mating of what you're proposing with the notion of template. You brought in patterns, but if we use template we can access some notions that are already there. We have to talk about, I think, partial templates or local templates rather than global templates and that gives us advantages in analyzing languages like Spanish that have what you would call 'paradigmatic' or what I would call 'columnar stress'. I think if you say 'paradigmatic', there could be paradigms that were parallel where the stress didn't line up in the column, but I think it's clear when you say 'columnar', and the crucial notion here is that if simultaneous tiers of structure prosodically are present you can then view the morpheme as a lexical diacritic, as a morpheme, saying where the stress is and no other content than that. So, in Spanish the paradigm has a morpheme for stress location and in Spanish the wonderful thing is that it marks a morphological category, and you have to have in Spanish two special morphemes of stress location for exceptional cases, so hablamos doesn't fit to habla, hablas, but hablamos, that will be the one that will be most resistant to change in the theories that aren't paradigmatic, aren't columnar. We'll be able to say that never changes, that maybe Spanish is not columnar that way, but otherwise the present has the syllable containing the last vowel before the stem stressed. The past and non-finite forms like gerunds have the first syllable containing the first vowel after the stem and the subsequent forms, the futures and the conditionals, have the syllable containing the second vowel and then these line up. There are then morphemes of subsequentness, of pastness or of non-finiteness and presentness that can be viewed as just the location of the accent, but they don't care about the rest of the word, so there's a local template that has three dots before it and three dots after it. It's just anchored on that morphological boundary. It seems to me that simultaneously present levels allow you to state that, so it's a morpheme with so little content that it's just where the stress is. But once you have that, then it has implications for other kinds of stress and also for phenomena that we were arguing about yesterday, about trisyllabic laxing and umlaut and so forth, there are languages where a morpheme causes a stress to be somewhere outside itself, so you can have a morpheme affix, let's say, that is associated with a prosodic structure where it's a vowel two or three over. So these are complex morphemes of an affix

238

TRUBETZKOY' S ORPHAN

and a location of stress. Now if you can locate stress there, then you can also locate a lax vowel, a fronted vowel and the response is going to be, "but then aren't you going to lose generalizations, why is it better to have one trisyllabic laxing or one umlaut". If you believe in redundancy rules, you can answer that objection because there will be something to cross the various rules and to end within the justification of that and again to tie in with something from yesterday. Joan Bybee brought in something about semantic idiosyncracies tied with combinations of morphemes. No one ever really answered that. Take the English example dogs, meaning "feet" like My dogs are tired. It is very rare to say My left dog is tireder than my right dog. If you don't like dogs use guts, which means something different in that than in gut, You can't really put the idiosyncracy only in the plural and say that dog means "dog", but when we add -s it means "feet". It's a little bit weird to say that dog means "foot", but it only gets that meaning when you add -s to it. It really is a fact about the combination. And you can get much longer example like suppository or antidisestablishmentarianism, so if you have combi­ nations of things you are going to have redundancies. Again I come back to this as the main difference between a lot of the theories here: whether you have redundancy rules or not. It seems to me that a network model, the Ford & Singh model, I don't know, Professor Dressier can speak to that, as to whether it has redundancy rules or not, but if you have redundancy rules you say things separately and you can use these levels of structure to have then specification of an affix and an idiosyncracy somewhere else in the morpheme, and of course it doesn't allow this and can't handle that and so I think your model has good implications for treating Spanish, at the very least. HURCH: I am looking forward to seeing your paper on columnar stress. DRESSLER: In answer to your last question, my model has in fact redundancy rules, it has phonological ones, morphological ones and to a certain extent morphonological ones. But here comes my question to the speaker. Due to his being very brief at the end I don't know whether he really has something like prosodic morphonology, so let me make an extremely simplistic pro­ posal and ask you what you think about that. So I would say that similar to segmental things, there would be phonoprosody and this might be Czech, Hungarian, etc. There would be morphoprosody like the Spanish columnar accent as we have heard, and then there would be in between and maybe, some people would try to limit it very discreetly, morphonoprosody. We will say this something like the Latin declension type arpinaslarpinatis where your final accent, which in formal prosody of Latin is not allowed, but in this

MORPHOPROSODY

239

one tiny class you get this and as is typical in the relationship between a naturalist theory and certain generative theories, a generative reanalysis of that would be in terms of formal prosodies, say that you have of course in arpinas, somebody coming from Arpinum, like Cicero. You have a penultimate accent only then after assigning the stress, the last syllable or peak or whatever is lost. This typical relation of an abstract phonological analysis versus a naturalist morphonological analysis and then you would say this, then, is the case, with a certain class you get this, you get opacity in diachronic change and therefore you have then a proper domain of morphonoprosody. Is this similar to what you intend? HURCH: This is similar to what I intend and this arpinas example is very similar to what I intended to show with the Basque example, not the one on the handout, but the alabar example, where you have a regular stress, I mean in traditional terms, they have a regular stress assignment which somehow is getting opaqued through a rule, a phonological process of contraction of two vowels. Where a morphological distinction arises, as ultimate versus penultimate accent. I would subscribe to what you propose on these three levels. I was too short about this. TIFFOU: Moi, j'ai une petite question d'Information. C'est ponctuel. C'est dans votre hand-out, le premier exemple, est-ce que le paradigme du nom angipuspwa est á deux termes ou trois termes, parce qu'on a varo labourdin, il yale déterminé,... HURCH: II y a trois termes. TIFFOU: Et dans le troisiéme terme, I'accent est le même au singulier qu'au pluriel. Voila ce que je voulais dire. HURCH: Dans ce dialecte, il y en a trois, le déterminé au pluriel. KAHEYA: I would like to get back to the underlying notion of the underlying representation of the syllable and just mention that converging evidence from neurolinguistics and psycholinguistics shows, actually I'm referring to ex­ periments currently under way in Australia, where they're looking at stroke patients with aphasia who have a specific deficit affecting prosodic patterns. These patients however can recognize syllables and they can also syllabify words. There is also psycholinguistic evidence that shows that the syllable is a recognizable unit and I am just wondering what kind of implications this kind of evidence would have for positing the existence of an underlying representation of the syllable. HURCH: What is proposed in the theory of Natural Phonology with respect to the syllable is not that the syllable does not exist at all. That the syllable is a recognizable unit couldn't be doubted, that the syllable somewhere in the

240

TRUBETZKOY'S ORPHAN

course of derivation comes into being. The syllable is, for example, a unit of grouping together sounds, this could not be doubted at all. The syllable gets some kind of prosodic point, this couldn't be doubted. It's really just the question of whether for underlying representations, we need this level. What are the generalizations concerning the syllable in terms of permanent mem­ ory. It's not like the rejection of the feature, but the rejection of the syllable as notion of long-term storage. DRESSLER: If I may just add that there is a variant of Natural Phonology where the syllable is very unimportant. For example, aphasics are aware of the number of syllabic peaks, and this is in permanent memory, so this is rhythmic. But for the exact delimitation of syllables, I have not seen any evidence. If you look at how aphasics syllabify, they are very often influenced by the morphemic make-up. So this is not aphasic evidence for the syllable. SINGH: I get the impression from the paper that apparentiy your intention is to relate these motifs, because you mentioned musical notation, to other rhythmic creative exploitation. I would like you to say a bit more about how would you see that relationship. Are these motifs going to be the things on which not only words but perhaps other things are also going to be mapped on because that partially answers the question about how limited or how large these motifs are going to be. HURCH: The motifs clearly have the intention, one of the ideas behind the assumption of motifs is clearly what I mentioned briefly at the end, that the capacity we have for prosody, what we perceive as prosodically well-formed isn't limited to how we use it in the language, you're completely right in this. The Czech speaker in his language puts only primary accents on first syllables, this doesn't mean that he wouldn't perceive a motif or a template like an Italian tadadadadd as something that is rhythmically well-formed, even if it is not exploited in his language, and the idea of assuming motifs and proposing that model is that linguistic motifs are just a subset of a bigger number of motifs, clearly comes from the idea or observation that we make use of such rhythmic patterns also in other fields. Let me give you just one example from the field of memorization. We memorize things more easily if we combine them with a rhythmic motif, for example, phone numbers. SINGH: Just a very brief comment. Coming from an oral tradition I don't have to be convinced of that. MORIN, Y-C: You mentioned the fact that syllabification was not or could not be used to make morphological distinctions. There is a problem that puzzled me for quite a while in some French dialects, including Standard French. It's

MORPHOPROSODY

241

a case where syllabification does appear to make such distinctions. They are systematic in some Walloon dialects as described, e.g., by Warnant. The facts are clear, but I will give you some examples from Standard French. Future forms like montrerai are regularly pronounced with two syllables. This pronunciation has never been mentioned in the literature but it's quite frequent. People say je te mont'rai [mct.re] ga demain and I very often perceive a distinctive difference of syllabification with the imperfect montrais of the same verb, roughly je montrais [mo.tre] ces choses hier vs. je te mont'rai [mot.re] ga demain. In fast speech, however, the distinction gets completely blurred, and the future or conditional forms become identical to those of the imperfect. HURCH: And you think that this is a morphological rule or a later reduction. MORIN, Y-C: It appears to be a reduction, but the facts are not clear. HURCH: Yes, but synchronically, when you are writing French grammar, would you put this in the morphology or... MORIN, Y-C: In Standard French the variant [mot.re] could be derived from trisyllabic [mctrcre]. In the Walloon dialects where it has been described, there is no such variation, and the morphological distinction is uniquely related to syllable differences. HURCH: Just a suggestion, that you might consider the first consonant as ambisyllabic, that the [t] is ambisyllabic, so that the difference between them would be just the question of gemination. MORIN, Y - C : Whether it is or not, it would still be a difference in syllabification. MORIN, J.Y.: Still, with the syllable there is a problem, from my point of view of the representation of knowledge and it's the difference between stored units and computed units. I had assumed, I suppose with everybody, spe­ cific syllables are always computed. Well, maybe not in some languages. Maybe in Chinese, you have a list of possible syllables and you have them, but I suppose most phonologists would admit that you have a list of segments, you have a list of syllable patterns, you have a list of morphs, morphemes, whatever but you don't have a list of specific syllables. You have the patterns but you don't have the syllables, like in syntax, you might have a list of phrase patterns but you obviously don't have a list of the phrases in your mind. You have to compute them. So this would be equiv­ alent to having a list of syllables, but in the case of syllables, they are in finite numbers, so eventually it's quite possible for a language to have a finite list of possible syllables and this reflects on the actual descriptive practice of traditional grammars. In Chinese they give you the syllables but in most

242

TRUBETZKOY'S ORPHAN

languages you don't find that. I mean, in Malagasy grammars, there are lots of syllables, they don't give you all the syllables. I would like that because there are so many of them in a word. PIGGOTT: I think there is a confusion that seems to be pervasive here, the distinction between the syllable as a unit and syllabification and I think that the arguments are addressing two different things here. I think your point in the paper was that syllabification, that is differences in syllabification, do not play a morphological role. My point and my actual response is that this is probably true, but the syllable as a unit does play a role and I don't see any alternative to it. HURCH: As a prosodic unit. PIGGOTT:Yes, as prosodic unit. That case from Koasati, the Muskogean language, seems to me to be fairly compelling evidence that you need to recognize something like that. HURCH: No, I would agree for the differences which came up in the discussion here. DESROCHERS: Natural Phonology talks about phonetic teleology because there are physiological and limitations preferences either on perceptual grounds or articulatory grounds, and I am wondering how you justify the preferences. Is it only on statistical grounds or do you have a principled way to say that some stress patterns are preferred over others? HURCH: I think there are two basic preferences which I haven't had the time to talk about. There are two basic context-free preferences and the context-free preferences should be ordered before the context-sensitive preferences, which are binarity of foot structure and rising over falling which would clearly together produce a likelyhood of systems with word-initial or domaininitial accent and systems with domain-penult accents and this converges statistically, but the point of departure is not statistical reflections, but orientation on the preferences. FORD: I just wanted to insert a word about top-down prosody and the place for the fact that an Australian aborigine can sing his way around Australia and never get lost and always know where he is, and that I have a Chinese friend who can sing me passages of the I Ching and always identify passages by the rhythm that is actually in what he is listening to. He knows, he can reconstruct what he heard simply by the rhythm that he got that belongs to a particular part of the I Ching. People who practice music have always identified a passage as part of a particular symphony, reconstruct the whole symphony. So, in language, especially in fields like argument-structure, we know where people are in their development in a passage by the information

MORPHOPROSODY

243

that they are using at that particular time. Can you say something about what position that would have in your theory of prosody? HURCH: The position that my theory has is that I would like to know more about it, and that I think that these two things are not so clearly measurable. For example, I know one study of Italian interjections, and already interjections in Italian would not necessarily follow a prosodic pattern or intonative pattern of a corresponding declarative phrase, for example. It's an intriguing question what's the relation between a linguistic-prosodic model of the language and prosodic models a language uses in folk music, in folk dances etc. I have to admit that I haven't found out a clear one-to-one relationship which would allow a solid statement. I am very happy of any indications, any reference, any example which could help settle some kind of relationship between linguistic and non-linguistic factors in prosody.

V

Bybee and Goad PRODUCTIVITY AND THE LEXICON

We simply need to know more about the lexicon before we can make further progress in other fields. D.L.BoUnger (1973:9)

Productivity, Regularity and Fusion: How language use affects the lexicon Joan Bybee University of New Mexico Alongside developments in generative grammar in the last three decades, alternate views of grammar have been increasingly gaining empirical support. A convergence of cognitive and functional studies with computer modeling has led to the view that language use is not just an irrelevant factor of 'performance', and thus of no interest to grammarians, but rather that language use is the primary determinant of structure. This point has been argued for syntax by Hopper & Thompson (1984), DuBois (1985), Langacker (1987) and others. For morphology the point has been made by Bybee (1985), Stemberger & MacWhinney (1986, 1988), Stemberger (1994) and others. The basic insight is that patterns that are frequently used become conventionalized or grammaticalized and even obligatory under certain conditions. The particular properties of universal and language-specific grammar are thus explained in terms of how they come into being. Such a theory is much richer than its generative predecessor, which has no explanation for grammar, but must rather view it as innately given. This basic theoretical program can also be applied to morphophonology: much of what we analyze as morphophonology is fossilized sound change from bygone eras. To the extent that such residue can be considered 'structure,' it is structure that arose for phonological reasons, but now either is simply residual or has been reanalyzed as expressing morphological categories. Morphophonology thus presents an excellent example of the conventionaliza­ tion of items from use: production characteristics of a word which are origi­ nally phonetically-conditioned become conventionalized as part of the represen­ tation of the word in the lexicon, creating alternations that have morphological or lexical conditioning. A fundamental problem for morphophonology is distinguishing between diachronic residue routinized in words in the lexicon and synchronically viable patterns. Generative phonology and morphology chose to view any pattern a

248

TRUBETZKOY'S ORPHAN

linguist could discern as structure, but in the perspective taken here, the question of what is a viable synchronic pattern is taken as a serious empirical question. Decades of research in generative grammar, with its emphasis on competence, have failed to turn up answers to the basic questions we will address here: how do language users acquire, internalize and generalize over morphological and morphophonological patterns? The core issue in synchronic morphophonology is the nature of productivity and what factors determine the relative degrees of productivity present at a given stage of a language. In this paper it will be argued that 'productivity', defined as the likelihood that a pattern will apply to a new form, is a direct reflection of the type frequency of that pattern. In connection with productivity, we will examine the related notion, 'regularity', defined as the relative lack of lexical idiosyncrasy, and a third notion, sometimes thought to be related to the others, 'fusion', defined as the extent to which the phonological shape of two morphemes are co-mingled or co-determined. In natural language — and thus in most theories of natural language — productivity, regularity and lack of fusion tend to be characteristic of the same patterns, and their opposites, lack of productivity, irregularity and high fusion tend to characterize another set of patterns. I will argue that the links among these three properties of morphological and morphophonemic patterns are diachronic in nature and that synchronically they are relatively independent of one another. In the associative Network model argued for here, synchronic productivity and regularity result from the way language is used, and the difference between productive, regular patterns and unproductive, irregular ones is a quantitative rather than a qualitative difference. This model will be compared with Level-Ordered morphology and the Dual-Processing model proposed by Pinker, Marcus and others, both of which hypothesize innate, structural differences among types of morphological and morphophonological rules. 1. The Network Model In the model I have proposed in various works (Bybee 1985, 1988, 1991), irregular, unproductive morphology and regular, productive morphology are not categorically distinguished, but rather represent two ends of a continuum. The representation of these types of morphology is similar — whatever generalizations exist emerge from lexical patterns — and the productivity of any pattern is predictable from its distribution in the language.1 In many ways 'Distribution' includes more than simple type frequency; it also must take account of

PRODUCTIVITY AND THE LEXICON

249

this model resembles a connectionist model since it proposes that generalizations arise from the patterns in an associative network; however this model is more complex than existing connectionist models in that it incorporates the notions of lexical item with a wide variety of inherent properties, lexical strength due to token frequency, and morphological analysis resulting from sets of interconnections among items. The three relevant properties of the model for present purposes are explained below. Note that the first and third of these properties relate language use to properties of the lexical representation. (i) Words entered in the lexicon have varying degrees of lexical strength, due primarily to their token frequency. Words with high lexical strength are easy to access, serve as the bases of morphological relations and exhibit an autonomy that makes them resistant to morphophonological change and prone to semantic independence. (ii) Words entered in the lexicon are related to other words via sets of lexical connections between identical and similar phonological and semantic features. Parallel phonological and semantic connections constitute morphological relations. These connections among items have the effect of yielding a morphological analysis, as shown in Figure 1. Figure 1: Network representation of regular affixation

(iii) Sets of words having similar patterns of semantic and phonological connections reinforce one another and create emergent generalizations describable as schemas. New items or items whose connections are not known or are weak can be fit into these schemas. The likelihood of the schema being extended to new items is directly dependent upon the defining properties of the schema and its strength, the latter property lexical features such as syntactic category, phonological shape and other subdivisions of the lexicon, such as borrowed vs. native items.

250

TRUBETZKOY' S ORPHAN

being derivable from the number of items that reinforce the schema. Thus productivity is a direct consequence of type frequency. A consequence of the notion of lexical strength is that morphologically complex words can be represented lexically if they have sufficient token frequency. Thus some regular and productive formations are represented lexically, in association with the forms to which they are related. This means that there are two ways of producing a regular form: by direct lexical access of that form as a lexical unit, and by access of a base form and the schema for the morphological pattern (Bybee 1985, 1994). Given that morphological patterns are represented as lexical connections, there is no particular advantage to having, for instance, affixes rather than stem changes, since they are both represented in the lexicon (rather than in a rules component) by lexical connections (Compare figures 1 and 2). On the other hand, high type frequency will give whatever patterns do exist a strong representation, making them highly available for new formations. Rubba (1993) presents an associative network analysis of Modern Aramaic which demonstrates that many of the problems encountered in analyzing Semitic languages with their interwoven lexical and grammatical morphemes are easily accommodated in an associative network model. Figure 2: Network representation of a schema for a vowel change class

2. Productivity, Regularity and Fusion 2.1 Productivity Morphological productivity is the extent to which a morphological pattern applies to new forms. Relative productivity may be measured among different morphemes or different allomorphic patterns. At the morphemic level, word-

PRODUCTIVITY AND THE LEXICON

251

formation devices (or derivational patterns) can differ in their productivity. A well-known example of a productive derivational suffix in English is -ness, which is far more productive than the derivational suffix -dom, for instance. Productivity of allomorphic patterns is important in inflection. Inflections are by definition obligatory, so for every inflectional morpheme, there must be at least one productive pattern, one way of inflecting new words. Different patterns for the same morpheme can be more or less productive; English plural -s is the only productive pluralization allomorph, though other expressions of plural in nouns exist. Generative theories provide no way of predicting the productivity of a given morphological or allomorphic pattern. The two generative theories to be discussed below (Level-Ordering and Dual-Processing) make distinctions among rule types that correlate partially with productivity, but these dis­ tinctions are structural in nature and do not provide an explanation for why one pattern is productive and another is not. On the other hand, MacWhinney (1978), Bybee (1985, 1988) and Baayen & Lieber (1991) claim that pro­ ductivity is directly related to the type frequency of the pattern. Bybee (1988, 1995) notes further that the openness of the defining features determining the domain of the pattern also contributes to productivity; that is, a pattern with no phonological or semantic restrictions is more likely to be or become productive than one with such restrictions. Thus in the Network model, productivity is directly related to properties of the use of a pattern, primarily to the number of different types to which the pattern applies. The type frequency is represented in the strength of the schema: schemas that apply to large numbers of items are highly reinforced and thus highly available for future uses. 2.2 Regularity Often the terms 'regular' and 'productive' are used interchangeably, and while productive patterns are usually regular, it is worth pointing out a difference in focus in the two terms, especially given the use of the regular/irregular distinction in some recent theories (such as the DualProcessing model, see below). 'Regular' is contrasted with 'irregular', where the latter indicates that a pattern is characterized by lexical idiosyncrasies. Lexical idiosyncrasies may be measured in degrees on two dimensions: the extent to which the pattern is applicable only to arbitrary lexical classes and the extent to which it deviates from the regular pattern. If English Past tense -ed is regular (despite phonologically conditioned allomorphy), the verbs which use this suffix and a vowel change (kept, slept, left, etc.) are somewhat irregular,

252

TRUBETZKOY'S ORPHAN

verbs that use only a vowel change (bit, drove, struck) are more irregular, and verbs that have both vowel and consonant changes are the most irregular (thought, taught, went). Thus irregularity is defined language-internally in relation to the most regular pattern. 2.3 Fusion Some morphological patterns involve a greater degree of fusion between the stem and the grammatical morpheme than others, which are more agglutinative in nature (Bybee 1985). The other end of this scale is called 'morphotactic transparency' by Dressier (1985a). When I use the term 'transparency' in this paper, I will be referring to morphotactic transparency. Some indicators of fusion (lack of transparency) are allomorphic changes conditioned in the stem by the affix, or the reverse, allomorphic changes conditioned in the affix by the stem, as well as actual phonological fusion at the boundary between the two. Even greater fusion would be indicated by the use of a stem-change to express the grammatical category, or the interweaving of lexical and grammatical morphemes. The most extreme case of fusion would be the total replacement of the stem, as in suppletion. It should be noted that the distribution of high fusion in a language is not arbitrary. Two gradient factors affect fusion. One is the semantic relevance of a grammatical category to the stem — higher relevance categories are more fused with the stem (Bybee 1985). The other is degree of grammaticization; as grammaticization proceeds over time, affixes become more fused with stems, as we shall see below. 3. The Relations among Productivity, Regularity and Fusion The hypothesis presented here is that the common co-occurrence of productivity, regularity and transparency in the same morphological patterns is a result of diachronic patterns of development and is not a psychologically valid correspondence that should be modeled for synchronic morphology. However, because these three properties tend to be associated in natural language such that morphological or allomorphic processes that are productive tend also to be lexically regular and exhibit a relatively low degree of fusion, while unproductive ones tend to be irregular and characterized by higher fusion, some theories build parts of this association into their models. I will argue that a model, such as the Network model, that does not insist on any of these correlations is more accurate than ones that do. The most complete association of the three factors is hypothesized by Dressier in the theory of Natural Phonology and Morphology, which correlates

PRODUCTIVITY AND THE LEXICON

253

morphotactic transparency (lack of fusion) with productivity, and with regularity (lack of arbitrary allomorphy) (1985a:316-22). According to Dressier, transparent morpheme combinations are easier to process (319), which, I surmise, in his view leads to their greater productivity. Lack of allomorphy and lack of fusion both result from the same general principle of semiotic transparency, or the one-to-one relation between meaning and form. Operations that are more natural on one scale will also be more natural on other scales. Thus the three properties discussed here all correlate in Natural Morphology. In a model with level-ordering of phonological and morphological rules, both phonological and morphological rules are sequenced in such a way that the ones with the most lexical restrictions are also the ones that have greater access to the stem, and thus create the greatest fusion (Kiparsky 1982a). In this model, then, irregularity and fusion are correlated. It is also true of this model that the rules in later levels are more productive, a property derivable from the generality of rules. By the Elsewhere Condition more specific rules apply first and more general rules apply later to all forms not affected by previous rules. A more recent model which I will call the 'Dual-Processing model' makes a strict distinction between regular and irregular morphology and claims that the two types of morphology are processed in entirely different ways (Pinker 1991, Marcus et al 1992, Marcus et al 1995, Clahsen & Rothweiler 1992). Regular morphology is handled by concatenating, symbol-manipulating rules which countenance no lexical restrictions, while irregular morphology is represented in the lexicon by means of associative networks, as proposed in Bybee & Slobin (1982b) and Bybee & Moder (1983). The regular rules apply in various 'default' or 'emergency' circumstances, characterized as cases in which the inflected form is unknown (as in the case of new or derived forms, memory lapses, etc.). In this model, regularity and productivity are equated. However, no way of predicting or explaining which patterns are productive and which are not is offered. Nothing explicit is said about fusion, but Marcus et al (1995) do characterize regular rules as 'symbol-concatenating' compu­ tations, which suggests that highly fusional patterns, such as those involving stem change, do not qualify for such rules. 3.1 The Diachronic Source of Morphology The vast majority of affixes in the languages of the world evolve from independent words by the gradual process of 'grammaticization' or 'grammaticalization' (Heine & Reh 1984, Heine et al 1991, Bybee et al 1994). In the progression from a lexical morpheme to a grammatical one, changes occur

254

TRUBETZKOY'S ORPHAN

in the phonological shape of the morpheme, its meaning and its grammatical behavior. A well-documented instance of this type of change is the devel­ opment of the future tense in Romance languages such as Spanish and French. A periphrastic construction in Latin consisting of an inflected auxiliary habere "to have" and an infinitive yielded a meaning of obligation or predestination: amare habeo love+inf aux+ls "I have to love, I am to love" The auxiliary reduces phonologically and comes to consistently appear after the infinitive (where previously it could occur in various places in the clause). In Old Spanish we find the construction indicating future: amar he love+inf aux+ls "I will/shall love" The auxiliary is written separately from the infinitive because at this stage other morphemes could come between the two; for instance, the object pronoun: amar lo love+inf him "I will love him"

he aux+ls

Later this possibility disappears and the auxiliary becomes an actual suffix to the verb: lo amare "I will love him" In this process the grammaticizing morpheme undergoes phonological reduction (e.g., from habeo to he to e), its position becomes fixed, it fuses with the verb, and the whole construction takes on a more abstract, grammatical meaning. A similar process leads to the development of derivational affixes. However, in this case the process begins with compounding. If the same element occurs in a number of compounds, it can reduce phonologically and change semantically in such a way that it becomes a derivational affix. For instance, the Modern English suffix ~ly derives from a noun, which in Old English was lic(e) meaning 'body.' The compound mann-lice originally meant 'having the body or appearance of a man' whence it generalized to 'having the

PRODUCTIVITY AND THE LEXICON

255

characteristics of a man,' the modern sense of manly. Since -lic was used in so many combinations, it lost its stress and reduced to -ly by losing its final consonant. Its meaning had already generalized in Old English to sometimes just mean 'pertaining to' as in the form heofon-lice 'heavenly.' In Modern English -ly is used to derive adjectives, as in friendly, and to derive manner adverbs, as in cleverly, but it occurs in many other uses as well: consider daily, weekly, cowardly, possibly and so on. Most derivational affixes in English and other languages can similarly be traced back to independent words where evidence is available. The process of grammaticization is not discrete, but continuous; grammaticization in the form of semantic change and further phonological reduction and fusion continues even after grammatical status is achieved, and even after affixation occurs. This means that we can categorize morphemes for their 'degree of grammaticization'. Non-affixed forms such as auxiliaries are less grammaticized than affixes; affixes that are more reduced (e.g., shorter), that cause changes in the stem or undergo changes caused by the stem are more grammaticized than morphemes that are unvarying. As one instance of this continuing development, consider the Spanish future forms discussed above. Some time after affixation had occurred, the new suffixes began to condition changes in certain verbs. Thus the combination venire "I will come" changed to vendre', querere "I will want" changed to querre; tenere "I will have" changed to tendre. Such changes can be taken to indicate increased fusion between the stem and the suffix. 3.2 Inflection In considering how productivity, regularity and degree of fusion are related, let us take up inflection first. A newly developed inflection was formerly a full word separate from the stem, so it will exhibit low degrees of fusion at first, and only when the affix and stem have been joined for some time will phonological processes begin to take their toll, gradually eroding the morphotactic transparency of the combination. Similarly, new formations will be relatively free of lexical idiosyncrasies, as these also take time to develop.2 Thus regularity and lack of fusion correlate for diachronic reasons. Productivity can be said to be a prerequisite for the development of 2 There is at least one type of instance in which lexically conditioned allomorphy can be built in to the grammaticization process from the beginning: where two auxiliary constructions were present from the beginning, but merge into the same grammatical construction. Thus the use of both 'have' and 'be' auxiliaries in perfect and perfective constructions in French, German, Dutch and related languages could be regarded as a lexical idiosyncracy or irregularity of some verbs.

256

TRUBETZKOY' S ORPHAN

inflection: affixation of a grammaticizing construction does not occur until that construction has achieved some measure of lexical generality, which implies that it can apply to new bases. When a new grammatical category is developing it is usually the case that it becomes fully general, applying to all lexical items of a general class (such as all verbs or nouns), and totally replacing older inflections. However, there are also cases where the new construction divides the lexical domain with an older one but without signaling any semantic difference. The 'layering' (Hopper 1991) thus created yields allomorphic variation between the old construction and the new one. An illustration may be found in the English Past tense system, where the older means of signaling past remain in the verbs that have vowel changes in the stem, despite the development of a now synonymous construction in which the Preterite form of the ancestor of the verb do is suffixed to the verb.3 The newer construction is more productive, more regular and more transparent than the older one. Evidence suggests that the new construction began as a pe­ riphrastic causative construction, verb + dyde, and gradually spread to more verbs, losing its causative meaning, but retaining its past sense. Its uses in­ cluded providing a Past form for new verbalizations and borrowings. Because the two means of forming Past in English represent an older and a younger layer, the former pattern is irregular, largely unproductive and has a greater degree of fusion. Thus these properties correlate for diachronic reasons. Layering is a fairly common phenomena in inflectional languages. Another interesting example occurs in the Saharan language, Kanuri (Hutchison 1981). Compare the Class 1 verb bu- "eat" with the Class 2 verb fəle- "show, point out". The Class 2 verb root is easily separable from the suffixes, while the Class 1 root is not. Class 1: bu- "eat"

3

Class 2:

"show, point out"

The existence of ablauting verbs in Germanic or vowel and consonant changes in Semitic languages does not constitute a counter-example to the claim that all morphology develops via grammaticization. In all cases where the sources of stem changes are known, phonological changes conditioned by affixes create stem changes and then delete (e.g., umlaut in English and German). It is reasonable to assume that stem changes existing since prehistoric times have the same type of source.

PRODUCTIVITY AND THE LEXICON

257

Class 1 contains about 150 verbs, many of which have irregularities, and the class is unproductive; Class 2 comprises the rest of the verbs of the language, which are regular, and this class is productive. Class 2 was formed by adding the forms of the Class 1 verb ngin "say, think" to the verb root. Hutchison argues that the construction originated with onomatopoetic forms, parallel to English go boom, but with the form "say boom". The construction provided a means of forming verbs from adverbs and nouns, and was used with compounds, derived forms and loan words from Arabic, Hausa and English. Because Class 2 represents a newer formation, it is more regular, less fused and more productive than the older Class 1. The Kanuri case is very similar to the case of English Past tense. Two factors determine productivity: type frequency and the lack of restrictions, both phonological and semantic, on the application of a pattern. When a new construction is developing, it does not always have a high type frequency from the beginning. However, if it is free of phonological and semantic restrictions, it will be available to apply to new words, derived and borrowed, and its type frequency will in this way increase, making it more and more productive. 3.3 Derivational Morphology For derivational morphology, as mentioned above, productivity applies to individual morphemes. Relative productivity is determined by type frequency and the relative lack of phonological and semantic restrictions on application. Since derivational morphology does not have to be obligatory, it is possible to have derivational patterns that apply in a very restricted domain, such as the English suffix -dom, which forms an abstract mass noun from a concrete one indicating a person's position or status. The new abstract noun indicates either the domain, realm, or condition bestowed by that status (kingdom, serfdom). In a few cases it is used on adjectives to produce an abstract noun (freedom, wisdom, boredom). Its productivity is restricted by its meaning and the semantic conditions on its application. It does show some productivity, as evidenced in a recent issue of Time magazine, where the word supermodeldom was coined to describe the status and domain of a 16-year-old model, whose perks included getting invited to more parties. In comparison, the well-studied suffix -ness seems to have very few semantic, phonological or morphological restrictions concerning the adjectives to which it can apply, and it is thus one of the most productive of all English derivational affixes. In the case of derivation, then, type frequency will be closely tied to the extent to which there are restrictions on the application of an

258

TRUBETZKOY'S ORPHAN

affix to a base. There is also layering in derivation: older affixes will be more tightly fused both semantically and phonologically to their bases, having undergone more sound change and semantic change; they may also be less productive and more irregular than newer ones. For derivation, however, the relation of productivity to transparency may work in two ways: new formations that are quite transparent may be highly restricted, but gaining in productivity; older formations that are fused may be losing productivity and thus gaining arbitrary lexical restrictions. 3.4 The Synchronic Relations Among Productivity, Regularity and Transparency The preceding sections have demonstrated that there is a diachronic explanation for the fact that the most productive morphological patterns are also usually the most transparent and regular. The question to be discussed in this section is whether or not there is any synchronic psycholinguistic relation among these properties, such that e.g., patterns that are more transparent have a higher degree of productivity because they are somehow easier to process and acquire (Dressier 1985a:318). Berman & Clark (1987) have argued that Hebrew derivational patterns that exhibit the greatest degree of stem change are also the most difficult for children to acquire. However, it is also the case that such patterns have the lowest type frequency, and it could be their lower type frequency that makes them more difficult. Since it is generally true, for diachronic reasons, that greater fusion and low type frequency correspond, it is difficult to know which factor is impeding acquisition. To review, we are considering the following three properties, one of which breaks down into two determining features: . . procuctivity: regularity: transparency:

(i) (ii) (iii) (iv)

lack of motivated restrictions4 high type frequency lack of arbitrary lexical idiosyncrasies lack of fusion

There is evidence, in terms of synchronic distribution and experimental data, that some pairs of these properties are independent of one another. First, consider a case which exhibits lack of fusion and lack of motivated restrictions, but has low type frequency and arbitrary lexical distribution: the 4

'Lack of motivated restrictions' is shorthand for lack of phonological and semantic restrictions.

259

PRODUCTIVITY AND THE LEXICON

German plural in -s. German has several pluralization patterns, as shown in Table 1. The one with -s has a very low type frequency, but could be regarded as the one with the least fusion (since it never has an effect on the stem) and the one most free of phonological restrictions. This pattern is only marginally productive: Kopcke's (1988) nonce-probe task showed it to have a relatively low rate of use. It is, however, used on about half of recent loan words (Kopcke 1988) and in pluralizing proper names and words with very unGerman phonological structure (Marcus et al 1995). This -s plural in fact was introduced into German through loan words. Janda (1990) predicts that it will become the most productive pattern for plural in German, taking over the territory of its competitors. This case shows that high type frequency and lack of restrictions do not always go together; however, type frequency is likely to increase in this case since the pattern does not have phonological or semantic restrictions. Table 1: German noun plural formation in the 200 most frequent nouns

Affix -(e)n

type frequency singular 42% die Strasse die Frau das Bett -e der Hund 35% (+ umlaut) die Kuh 12% der Daumen zero (+umlaut) die Mutter das Leben -er 10% das Kind (+umlaut) der Wald 1% -s das Auto der Park

plural die Strassen die Frauen die Betten die Hunde die Kiihe die Daumen die Mutter die Leben die Kinder die Wälder die Autos die Parks

gloss "the street" "the woman" "the bed" "the dog" "the cow" "the thumb" "the mother" "the life" "the child" "the forest" "the car" "the park" (based on Janda 1990)

Thus (i) and (iv) are each independent of (ii) and (iii). Bybee & Newman (1994) have separated transparency from productivity experimentally and shown that there is no particular preference for morphotactically transparent formations, provided that the fused formations have suffi­ cient type frequency. The experiment consisted of four artificial mini-languages in which sixteen nouns and their plurals were learned by English-speaking subjects. In each language, half the plurals were formed with suffixes and half with stem changes. In one language, both patterns were the same for all the nouns they applied to (i.e., they were both regular); in another condition, the

260

TRUBETZKOY'S ORPHAN

suffixes were regular, but the stem changes exhibited four different patterns (i.e., they had a low type frequency and lexical idiosyncrasies); in the third language, the reverse was true, i.e., the stem changes were regular and the suf­ fixes were irregular, exhibiting four allomorphs. In the final condition, both stem change and suffixes were irregular. After the subjects learned the sixteen nouns of the language, they were asked to supply plausible plurals for eight new nouns. We hypothesized that the subjects would not necessarily generalize the suffixes, falsifying Dressler's theory that suffixes (transparent formations) are more natural than stem changes (fused formations), but that subjects would use whichever pattern was regular in the language they had learned. The results showed no particular favoring of the suffixed plurals; subjects supplied stem changes and suffixes in approximately equal numbers (overall 48.75% suffixes and 50.25% stem changes). The subjects supplied new stem change or suffixed plurals in approximately the proportions they were present in the input data — about half and half. This result supports a usage-based model. Moreover, we found that the regularity of the suffix affected productivity: in the conditions in which the suffix was regular (had no lexically arbitrary allomorphy), more suffixes were supplied for new forms; in the conditions in which the suffix was irregular, more stem changes were supplied for new forms. This result shows that productivity (based on type frequency and lack of restrictions) is independent of the degree of fusion of the pattern. Are (i), lack of motivated restrictions, and (iv), lack of fusion, inde­ pendent? There are formations that have motivated restrictions that are not fusional, but transparent, as the Kanuri "say" verb at an early stage, or any newly grammaticized construction. Patterns that are highly fusional but lacking in motivated restrictions appear in Semitic languages, as for instance, the iambic plural of Arabic (McCarthy & Prince 1990). Are (ii), high type frequency, and (iii), arbitrary restrictions, independent? In a sense they are not, because any restriction reduces type frequency. Of course, there are systems where there are several patterns each having fairly high type frequency, but with each one applying to an arbitrary portion of the lexicon, for example, German plurals or Hausa plurals (Lobben 1991). Synchronically, then, these properties appear to be independent of one another. In the Network model, this is just how they are treated, but in two other current models, the Dual-Processing model and the Level-Ordering model, these properties are treated as though they are structurally related to one another. The following two sections describe these models and how they view productivity, regularity and fusion.

PRODUCTIVITY AND THE LEXICON

261

4. The Dual-Processing Model A proposal emerging from the child language and psycholinguistic fields sets up a strict distinction between regular and irregular morphology, treating them in two distinct modules of the grammar (Pinker 1991, Marcus et al 1992, Marcus et al 1995, Clahsen & Rothweiler 1992). Irregular morphology is treated as in the Network model with irregular forms listed in the lexicon and organized into patterns describable by emergent schemas. Regular forms are derived by a 'symbol-concatenating rule' that acts in a generative fashion on a base form to produce the regular derived form. While irregular patterns are highly affected by actual lexical distribution, the regular ones are not. Prasada & Pinker (1993) have shown experimentally that the lexical distribution of English irregular Past forms such as rung, strung, struck, etc. affect subject's responses to nonce verbs, while the lexical distribution of the regular Past does not affect nonce form application. However, since regularity and type frequency are confounded in English, it is possible that this effect is due to the high type frequency of the regular Past, plus its lack of motivated restrictions. Marcus et al (1995) argue that the case of the Germans-plural serves as an example in which regularity and type frequency are independent. However, in this case, despite the lack of phonological restrictions, a strong effect of lexical distribution can be observed. Kopcke (1988) found that the s-plural was much more likely to be applied to nonce words ending in full vowels than to nonce words ending in schwa. This result reflects the lexical distribution in which almost all words ending in full vowels have a plural in -s. Clahsen & Rothweiler (1992) have argued that the case of the German Past Participle also separates type frequency from regularity. Here the productive affix is -t as opposed to -en. Clahsen and Rothweiler argue that both affixes have approximately equal type frequency. Their counting method takes verbs with separable prefixes to be distinct types, as though one were counting each of the following English verb-particle combinations as separate verbs: break, break up, break down, break out, break in. If the German verbs are counted as one would count English verbs, that is, counting break only once, then the distribution of regular and irregular types is similar to that of the English Past tense: out of 1258 base verbs listed in Ruoff (1981), there are 150 Past Participles in -en and the remainder use the suffix -t (Bybee 1994). In this case as well, type frequency predicts the productivity of -t. This model provides no link between productivity and language use. The best it can offer as to why a certain pattern is treated differently from others (handled as a symbolic rule) is that human beings are endowed with an innate neural architecture that makes two types of processing possible. This model

262

TRUBEIZKOY' S ORPHAN

treats the English s-plural and the German s-plural both as symbolic rules and provides no means of accounting for the difference in their productivity. As I mentioned above, the Network model also is 'Dual-processing' in a sense: a regular form may either be accessed whole from the lexicon or derived by applying a schema to a base, depending on the token frequency of the form. The Network model does not base the difference in derivational mode on a structural difference, but rather on a usage difference — high frequency forms are accessed whole and low frequency are not (see Losiewicz 1992 for experimental evidence that this distinction is based on word frequency). 5. Level-Ordered Morphology Level-Ordered morphology is based on the insight that both morphological and phonological patterns have different degrees of involvement with the lexicon (Kiparsky 1982a). We have already seen that for morphological pat­ terns this is a result of the way grammaticization proceeds, creating layers of new morphology on top of old morphology. For phonological rules there is a diachronic explanation as well. Phonological rules begin as phonetically-moti­ vated processes and gradually become more and more involved in the morphol­ ogy and lexicon. They tend to lose their phonetic motivation and remain only as fossilized alternations in assorted morphological environments. Thus phonological processes also create layers of newer patterns on top of older ones. Level-Ordering models this diachronic layering by recreating it in a synchronic grammar. This works fairly well for English where the deepest level is largely comprised of morphological patterns that entered English through the borrowing of French words, many of which were already morphologically complex and thus carried with them fossilized phonological processes that had occurred in French. Thus certain old phonological processes of French (like Velar Softening) as well as the older sound changes of English (like the Great Vowel Shift) affect words at this level and not at later levels, where the more productive, largely Germanic patterns are described. However, in other languages, which may not have this bifurcation in the lexicon, it is not so clear that rules are positioning so neatly on distinct levels. (See Kaisse & Hargus 1993a for an overview of the recent literature, which seems to show that there is no general agreement about how phonological and morphological rules interact.) Even in English there are problems, especially with affixes that have properties of both levels, such as -ity (Aronoff & Sridhar 1987). Still, given the way language change proceeds and the way new phonological processes and grammatical affixes are added, any theory that

PRODUCTIVITY AND THE LFXICON

263

recapitulates diachronic development will need to show some type of interleaving of morphological and phonological rules. In addition to the fact that there is no consensus on the internal structure of the theory, there exist a set of problems that affect all theories of this type. As in other versions of generative grammar, Level-Ordered morphology makes the highly improbable claim that underlying forms of high frequency items, such as damn, are affected by low frequency items such as damnation. The underlying representation of damn with a final /n/ is set up to accommodate the derivation of damnation, just as an underlying representation for sign as /sign/ is set up to accommodate the derivation of signature (Borowsky 1993). I consider it very unlikely that a learner would modify her/his underlying representation for an established and highly frequent word simply because s/he has now acquired a low frequency word which might be related to the established word. Productivity is modeled by the Elsewhere Condition, whose premise is that more specific rules apply before, and block, more general ones. The Elsewhere Condition has an effect similar to the criterion of the openness of the schema, described above for the Network model. But the effects of type frequency, the most significant factor in determining productivity, cannot be represented in a model in which rules exist independently of the forms to which they apply and which does not take into account the way language is used. Another serious flaw in the theory is the lack of a way to account for the semantic composition of morphologically complex words. It is often observed that words formed with affixes at the deeper levels do not have compositional meaning. This should be taken as sufficient evidence that such words are listed in the lexicon as units and not decomposed morphologically. However, since practitioners of this theory still want to do morphological decomposition, then they are obliged to come up with a semantic theory that can predict the results of morpheme combination. In my view, Level-Ordering reconstructs diachronic development of form, while neglecting meaning, and produces analyses that have no synchronic reality and no prospect of synchronic verification. This is precisely the reason that it is so difficult to find a model of level-ordering that works consistently. A usage-based model, such as the Network model, takes into account the way forms are used synchronically and is constructed to bear a direct relation to the surface forms of the language. In the following I compare the Network model to Level-Ordering models. In the Network model it is use rather than structural criteria that determines representation. A lot of what is studied as morphophonology is diachronic

264

TRUBETZKOY'S ORPHAN

residue and does not need a synchronic explanation, unless there is evidence of productivity. Thus all morphology and morphophonology that is put in Level 1 for English by the Level-Ordering model is not compositional in the Network model at all; rather, the 'derived' words are listed in the lexicon. The evidence in favor of this position is the simple fact that most of these derivational formations have unpredictable meaning. Many also have unpredictable form (including the irregular inflection), because alternations that are fossilized have many exceptions and idiosyncrasies of application. To the extent that there are valid morphological relations among such lexical forms, these can be captured in the lexical connections, as shown in Figures 1-3. One insight of Level-Ordering is that more peripheral phonological and morphological rules do not have access to the internal structure. This is modeled in Level-Ordering by erasing the morphological bracketing at the end of each level. Thus each new morphological or phonological application treats the item it is applying to as an unanalyzable whole. This is precisely what would be expected if the input to new morphological formations were items that were already stored in the lexicon. In fact, only because generative models insist upon maximal decomposition of words does it come as a surprise that outer, newer patterns do not have access to older, more internal composition. In the Network model where words are entered into the lexicon as they are formed, there would be no reason to expect that the diachronic composition of a form would affect its synchronic behavior. Kiparsky (1982a) has pointed out that even when irregular forms are the heads of compounds the compound as a whole is often inflected as though it were regular. Thus in the Toronto Maple Leafs the plural is regular rather than the irregular *Maple Leaves. In the Network model, when a compound is formed that is a noun, it is associated with the schema for plural nouns, and treated like an unanalyzable noun. The irregular plural [livz] is listed in the lexicon along with [lif], but it is not accessed for the new compound plural. A verbalized noun is associated with the verb schemas, and the strongest schema for Past tense is the -ed pattern. For instance, the verb to ring formed from the noun ring has no access to rung, but will rather participate in the regular Past schema. Thus the phenomena that Level-Ordering accounts for are also accounted for if words are stored in the lexicon as they are created and associated with appropriate morphological schemas. (See my discussion of Heather Goad's contribution later in this volume for a further explanation of how these facts are accounted for in the Network model.) Another fact easily accounted for by the lexical storage of derived words is the minor tendency for irregular plural nouns to appear in English compounds

PRODUCTIVITY AND THE LEXICON

265

while regular plural nouns usually do not. Thus mice-infested is possible because mice is a lexical item, but rats-infested is not acceptable because it contains the regular plural marker.5 Note that the irregular plurals of English all designate objects that tend to occur in groups of more than one: mice, teeth, feet, geese, oxen, or they are plurals of extremely high frequency nouns: children, women, men. Thus the irregular plurals are not only listed in the lexicon but they tend to be highly accessible due to their token frequency, either in absolute terms or in relation to their singulars. It should be added, however, that compounds with plural nouns in them, even irregular plurals, are extremely rare in English. In addition the Network model provides a way of accounting for two usage-based phenomena in morphology not accounted for in generative models. One is productivity, which we have already discussed: productivity is directly determined by type frequency, and type frequency affects the lexicon by strengthening the schemas that are used with greater numbers of distinct items. The other usage-based phenomenon is the maintenance of irregularity in forms with high token frequency. Because the use of items affects the lexicon, items that are frequently used have strengthened representations and forms that are less often used have weaker representations and may even fade from memory. Irregular inflections that are highly available in the input will be easily accessible and will not be regularized. Weaker, less accessible irregular forms are more likely to be replaced by regular formations (Bybee 1985). 6. Language Acquisition An interesting controversy that distinguishes the structural or generative models from the associative or usage-based models has arisen in the field of child language development. The linguistic behavior of little children seems to point to the existence of structure of certain types, but since structural theo­ reticians do not believe that young children have enough exposure to data to acquire the structure in question, they propose that these structures are innate. The usage-based approach would claim that children have ample exposure to store certain items in their lexicon, or more generally have representations in memory of words and phrases they have heard and used. The 'structures' in question do not have to be innate or acquired by the child, because in fact they do not exist. Gordon (1985), following Kiparsky (1982a), observed that even though English compounds can contain irregular plural nouns (mice-infested), such compounds are extremely rare and the use of the singular form in the The same account is given in Marcus et al. 1992.

266

TRUBETZKOY'S ORPHAN

compound is much more common. This means that young children will have been exposed to many compounds with singular nouns, but few, probably none, with irregular plural nouns. Gordon hypothesized that if children cor­ rectly assigned irregular pluralization to Level 1, then irregular plurals could be used in children's compounds, since compounding is assigned to Level 2. Gordon created an experiment in which children were exposed to plural nouns in the context of multiple objects and asked to label a puppet as an X-eater. The results showed that children could use irregular plurals in the compound, e.g., teeth-eater, but not regular plurals; that is, they produced bead-eater not beads-eater. Since there is little chance of exposure to compounds such as teeth-eater, and since children produced these forms (given a bias in the experiment towards plurals) but did not produce regular plurals in compounds, Gordon reasons that they have correctly placed irregulars at Level 1, and that they also correctly use Level 1 as input to compounding. Given the lack of appropriate input to the child and the fact that children between the ages of 3 and 5 gave consistent responses, with no apparent learning taking place, Gordon suggests that the Level-Ordered structure might be an innate property of the lexicon (p. 87). A much simpler explanation for the facts that does not require so much innate architecture is that children have stored the irregular plurals in their mental lexicon and can access them for compound formation. This is the account given in the Network model and also in the Dual-Processing model (Marcus et al 1992:142). Since the experimental situation primed the children with plural forms rather than with singulars, and since for English irregular plurals there is some phonological distance between singular and plural, the plural was more available in the situation and was thus used. Several factors could explain why these same children did not use regular plurals in their compounds even though in the situation regular plurals were primed also. First, the children probably know and use a number of compounds of the relevant type, none of which includes an internal regular plural marker. The formation of new compounds will be based on existing compounds; in the experiment the children accessed their compound-forming schema, which does not include any plural inflectional affixes. Second, these children were successful at forming compounds, which suggests that they understand the basics of compound formation, which is that the component words of compounds, especially the non-head element, is decategorialized (Hopper & Thompson 1984). In the case of nouns this means that the noun is unable to refer, and thus does not carry the inflection that a referring noun

PRODUCTIVITY AND THE LEXICON

267

would. The lexical noun in a compound represents the semantic essence of a class of entities, but without referring to a particular entity. After being primed with regular plural nouns, we might think of what the children do as removing the inflectional affix. After all, they will not have been exposed to compounds with plural -s in them (e.g., it is Cookie Monster, not Cookies Monster, despite the fact that he eats lots of cookies). Evidence for a strategy such as this is the fact that 69% of the responses for scissors singularized it to scissor to put it in the compound. If there is a prohibition against a plural affix (rather than plural meaning) in the compound, that would explain why irregulars with affixes such as children and oxen do not sound acceptable in compounds: child-eater is better than children-eater, or, to compare across items, mice-eater is better than children-eater. Unfortunately, Gordon did not include the items children and oxen in his experimental material.6 Gordon also included the pluralia tantum nouns, clothes, pants, glasses and scissors in his experiment. Level-Ordering predicts that these nouns will retain their lexical plural marker, since it is derived at Level 1. The Network model would predict some variation with these items, since they do not have singular forms, and yet they include the regular plural marker which does not appear in compounds. If the children are using the strategy of removing the inflection before forming the compound, then they will run into trouble with clothes which would be [klouð] with the inflection removed, and with pants and scissors, which do not have singular bases, and even with glasses, whose singular base has a different meaning. In the Level-Ordered account, there is no conflict: these plurals are derived at Level 1 and should fit nicely into the compound, just as mice and teeth do. Gordon's results on the pluralia tantum do not have the uniform and categorical nature that one would expect if Level-Ordering were innate. Instead, clothes and pants tend to go into compounds as 'plurals', while glasses and scissors tend to have the 'inflection' removed. The Level-Ordered account fails for scissor, since there is no such singular. One would have to posit an inflection-stripping strategy, as I have for the Network account, but if an inflection-stripping strategy is to be invoked, then it might as well be invoked to explain the difference between regular and irregular nouns, and level-ordering is not necessary. Another problem for the Level-Ordering account (acknowledged by 6

Sternberger 1994 argues that regulars and irregulars in English are not parallel phonologically and Gordon's attempts to separate them using real English words fails because the morphological is confounded with the phonological.

268

TRUBETZKOY'S ORPHAN

Gordon) is the fact that Dutch and German do allow some plural inflections in compounds. This can be viewed as a simple difference between English and her sister languages in terms of the patterns for compound formation; in Dutch and German there is no restriction against a plural affix in a compound. For Dutch, the most common and productive plural affix is -en, and this affix is found in compounds such as paardendief "horse-PLURAL#thief\ and tandenborstel "tooth-PLURAL#brush". Such compounds can be produced if all Dutch plurals are lexical, as the Network model would predict, and Dutch does not restrict plural affixes inside compounds. German also allows plural affixes in compounds, with the exception of -s, which never occurs in compounds (Autobahn, not *Autosbahn). Clahsen et al. (1992) argue that the facts of German can be described by positioning the different plural affixes of German at different levels: -er and -e are at Level 1, -en at Level 2 and -s at Level 3. Compounding takes place at Level 2, so all affixes except -s can appear in compounds. Evidence for Level-Ordering can be found, they claim, in data on compounds from 19 dysphasic children they studied. These children never put -s in compounds, as would be predicted since adults never do either, but eight of them left -n out of a compound requiring it at least once. Clahsen et al. argue that the omission of -n in the compound can be accounted for by assuming that these eight children had placed -n on Level 3 rather than Level 2. Six of these eight children overregularized with -n more often than with -s, a finding which supports a Level 3 assignment of -n. None of these children omitted the other plural markers -e and -er from compounds, as predicted by the assignment of these affixes to Level 1. Thus Clahsen et al. argue that children, even dysphasic ones, have access to Level-Ordering, which, then, must be innate. One problem with this account is that none of the children always omitted -n from compounds. Assignment of a morphological rule to a level is a categorical decision, not a probabilistic one. Thus these children would have to have been moving -n pluralization from one level to another during the period of the study. Another problem is that the one nonimpaired child studied in Clahsen et al. did in fact produce a compound that omitted the plural -er. Simone (Miller 1976) produced both the correct form Bilderbuch "picture book" and the incorrect Bildbuch. A general problem with applying LevelOrdering to child language is that Level-Ordering predicts discrete, categorical behavior, which is rarely found in children, or adults for that matter. Constructing a full Network account would require more information about the frequency of occurrence of compounds of different types in German than is

PRODUCTIVITY AND THE LEXICON

269

available to me at the moment. However, it is possible, even in the absence of that data, to outline a description of the acquisition of compounding. In the Network account, existing compounds have lexical representations, in keeping with their unpredictable semantics and their propensity to undergo further semantic and phonological change as a unit. Children acquiring German compounds will have some with singular and some with plural nouns in them. If these are not yet all strongly represented, there could be retrieval errors involving omission of inflections internal to the compound. Children will also be producing new compounds using existing patterns as schemas. This means that some of their formations will have singular nouns in them and some will have plural. Why would -n be commonly omitted from compounds but not -er and -el Here is the point at which the token frequency of the nouns and compounds in question becomes important. First, a highly frequent compound is easy to retrieve and is not likely to be produced incorrectly. It might turn out that the compounds with -er and -e used by children are of high token frequency. Second, a highly frequent plural, even inside a compound, might be less likely to lose its plural marker. Again, the relevant vocabulary statistics are not available, but they are likely to show that the more lexically restricted -er and -e tend to occur on words of high token frequency, while the more productive -n occurs on words of a wide range of token frequencies. Thus, the Level-Ordering account, which describes rules as more or less deeply embedded in the lexicon, can be matched by an account without ordered rules, which represents the lexical involvement of morphological patterns as a matter of direct lexical representation. 7. Conclusion In contrast to structuralist and generativist theories, the more modern usage-based theories of language find significance in the fact that linguistic behavior is continuous and not discrete, probabilistic and not categorical, and dynamic rather than static. Language does not exist in a mental prison, insulated from real-world factors of meaning and use. Language is a social instrument that is in constant use, and the varying details of this use have an effect on storage and processing, creating and recreating the mental associations that we study as grammar. Morphophonology is no different from morphosyntax in this regard, and its study in this framework could lead to answers to some of the persistent questions of the field — what determines productivity, what is its relation to regularity, and how are processing and storage affected or not affected by the degree of fusion among morphemes.

Productivity, Derivational Morphology and Atypical Populations: Comments on Bybee* Heather Goad McGill University 0. Preliminaries In her paper "Productivity, Regularity and Fusion: How language use affects the lexicon", Bybee argues for an associative network model where the difference between regular and irregular morphological patterns is not discrete. Instead, relatedness across forms is captured through sets of associations of varying strengths. Productive patterns are extended to new forms through the product-oriented schemas which emerge from these associations, i.e., on the basis of shared semantic and phonetic content. Asssociative network models stand in direct contrast to generative approaches in which at least regular alternations are argued to be rule-governed. Rules have a status independent of items listed in the lexicon, and apply categorically to inputs which meet their structural description. The theory has no direct way of predicting productivity, and it is primarily for this reason that Bybee rejects it. In these brief remarks, I begin by discussing some of the advantages of Bybee's 'Network Model' which follow from the hypothesis that the lexicon is highly structured; specifically, the idea that subregular verbs are stored in family resemblance classes. In Section 2,I turn to a problem for the Network Model which follows from Bybee's view that "representation does not depend on structural criteria, but rather is directly determined by language use". Since this model does not encode the derivational history of lexical items, it cannot capture the fact that irregular verbs exhibit regular, rule-like behaviour in denominal constructions (see esp. Pinker & Prince 1988). A compromise position is discussed in Section 3, Pinker's (1991) hybrid model of inflectional morphology, where productive irregular patterns are argued to be stored in associative networks while regular morphology is derived by rule. In Section This research was supported by an interdisciplinary grant from the Medical Research Council of Canada and the Social Science and Humanities Research Council of Canada. I would like to thank Nigel Duffield and Eva Kehayia for useful discussion and comments.

PRODUCTIVITY AND THE LEXICON

271

4,1 provide evidence for the hybrid model from 'specific language impairment' (SLI), a developmental language disorder which is characterized most significantly by problems with inflectional morphology. Based on earlier research (Gopnik & Crago 1991, Goad & Rebellati 1995), it is argued that individuals with SLI are unable to decompose inflectionally complex words into their constituent parts; and, as a result, they store stem + affix complexes as unanalysed chunks. In the absence of rules which build inflectionally complex words, the mechanism with which individuals with SLI extend patterns to new forms is analogy. If this is an accurate characterization of the impairment, then counter to the Network Model, unimpaired individuals must have access to a qualitatively different system for regular inflection. I conclude with the hypothesis that the Network Model characterizes regular inflection only in the absence of a rule-based system. 1. Frequency and Productivity Let us turn directly to some of the advantages of the Network Model. First, phonetically-related irregulars are argued to be stored in clusters. Since most rule-based theories assume that the lexicon is unstructured, the fact that phonetically similar irregulars pattern together cannot be expressed. It thus becomes fortuitous that the past tense of many English irregular verbs is marked by the rhyme /A/ + final nasal and/or velar (see Bybee & Slobin 1982b). If this were merely a historical artefact, it would be of no con­ sequence. However, these patterns affect the way present day speakers learn forms. Rule-based theories are deterministic: rules apply categorically to inputs which meet their structural description. When extended to irregulars, the results are less than satisfactory (cf. Hoard & Sloat 1973). For example, a rule which converts lil to [A] in the past when the following consonant is /ŋ/ will overapply to bring and think and will incorrectly not apply to spin or stick. Extending the structural description to include coronal nasals and velars of any quality will incorrectly include grin and lick. Thus, rule-based theories cannot express the fact that, in the case of irregulars, membership is determined by 'family resemblance', approximation to and not identity with the conditions of target and trigger (Bybee & Slobin 1982b, Bybee & Moder 1983). Family resemblance is a dynamic notion and, as a result, it permits variation in the system, both within and across speakers. First, patterns with high type frequencies readily generalize to new forms and may, in fact, compete with one another. Berko (1958), for instance, remarks that in addition to glinged, 75% of her adult subjects provided glang or glung as the past of the

272

TRUBETZKOY'S ORPHAN

novel verb gling. More comprehensive studies report similar findings (e.g., Bybee & Moder 1983; see also Kim et al 1991). Along the same lines, the verbs in (1) have been added to the /a/ + {nasal, velar} class since Old English (Bybee & Slobin 1982b; from Jespersen 1954 [1942]). Those in (lb) are restricted to certain nonstandard dialects (Bybee & Moder 1983; from Mencken 1936). (1) a. fling - flung string - strung b. bring - brung

hang-hung stick - stuck sneak- snuck

sling-slung dig - dug shake - shuck

sting-stung strike - struck drag - drug

Second, if lexical entries are clustered according to family resemblance, variation should be most apparent for those items which are some distance from the cluster's prototype. This has, in fact, been demonstrated experi­ mentally by Bybee & Moder (1983) for the /Λ/ + {nasal, velar} class: irregular past responses were more readily supplied for novel verbs which were phonetically closest to the prototype. All of these phenomena can be straightforwardly captured in the Network Model. Yet, while Bybee argues that the strength of the schemas is directly tied to productivity (see esp. Bybee 1985), I suggest that the facts may prove to demand a more explicit account of the relationship between the potentially competing factors of type frequency within a pattern and phonetic relatedness across patterns. Thus, we might expect subregulars which form their past in /t/ to be more readily subject to regularization. For those items which do not require stem-vowel shortening, (2a), the breakdown of the /t/ past in American English is particularly conspicuous.1 (2) a. smell-smelt/smelled spill-spilt/spilled burn-burnt/burned b. dream-dreamt/dreamed leap-leaptAeaped

spell-spelt/spelled spoil-spoilt/spoiled learn-learnt/learned kneel-knelt/kneeled

dwell-dwelt/dwelled earn-(earnt)/earned lean-leant/leaned

What is striking about the class in (2) is that it contains a relatively large number of items, most of which have token frequencies comparable to the members of the productive /Λ/ + {nasal, velar} class (from Dahl's 1979 corpus of spoken American English). In addition, the verbs in (2a) have two 1 It appears that regularization of the verbs in (2a) has happened quite recently in American English. Jespersen (1954:32) notes that, in contrast to Britons, Americans tend to prefer the regular pasts for these verbs, both in speech and writing; see also Bybee & Slobin (1982b).

PRODUCTIVITY AND THE LEXICON

273

properties which Bybee & Moder (1983) suggest are partly responsible for the productivity of the strung class. The pasts have a well-defined phonetic shape: a rhyme containing a non-back vowel + sonorant consonant; and the past forms have 'high cue validity': the rhyme sequences virtually always signify past tense.2 If the strength of the schemas is partly determined by phonetic distance from the most pervasive -ed pattern, it may contribute to an explanation for the direction of change. For the ablaut class in (1), high type frequency, pho­ netically well-defined pasts, pasts with high cue validity, and perhaps distance from -ed bear on productivity: change involves movement into the class, but no movement toward regularization; see (3). (3)

bring - brought → bring - brung *bringed drink-drank → drink-drunk *drinked sing - sang → sing - sung *singed

However, the first three factors do not seem to play a significant role in 'protecting' the forms in (2) from regularization. Perhaps similarity to regular inflection must be taken into consideration in the determination of productivity. This may prove to be a promising avenue for future research. 2. Denominal Verbs We turn now to a problem with the Network Model which concerns the relationship between productivity and derivational morphology. As we have seen, the model relies on productivity to determine the behaviour of subregulars. However, short of encoding the derivational history of a word, the model cannot obtain the effects of level-ordered morphology. Specifically, it cannot capture the fact that irregular verbs exhibit regular, rule-like behaviour in denominal constructions. Examples such as those in (4) have been used by Kiparsky (1982a) to motivate Lexical Phonology and by Pinker & Prince (1988) to argue against the connectionist model of Rumelhart & McClelland (1986). In Lexical 2 To demonstrate that the rhymes in (2a) are good cues of pastness, all phonetically parallel non-past monosyllabic words which are frequent enough to appear in Dahl's (1979) spoken corpus and/or KuCera & Francis' (1967) written corpus are provided below (data were initially extracted from Wood 1936). There are no non-past words which end in the sequence [oilt] or [a-nt] in Wood's dictionary, [elt]- and [iltj-fmal words to the left of the line appear in Dahl's corpus. Those to the right of the Hne also appear in KuCera & Francis (with a frequency of only 001-004). [elt] belt, melt svelte, veldt [lit] guilt gilt, hilt, milt, tilt, wilt

274

TRUBEIZKOY' S ORPHAN

Phonology terms, these verbs are derived from nouns at Level 2; irregular pasts, whether stored or derived at Level 1, are thus no longer accessible to them. (4) Heflied/*flewout to centre field They joy-rided/*joy-rode all night She braked/*broke the car suddenly He high-sticked/*high-stuck the goalie One might attempt to attribute the regularity in (4) to lack of productivity: these ablaut patterns are not strong enough to attract new verbs, so -ed is the only past tense form available. Lack of productivity, however, cannot be appealed to in the case of the strung pattern in (1) which, as we saw in (3), resists regularization. The relevant data are provided in (5a). If productivity were a factor, we should at least expect to get varying judgements for these examples. Yet, unlike with the non-derived cases in (5b), speakers are firm in their judgements for (5a) (cf. Kim et al 1991). (5) a. You ringed/*rang, *rung the bathtub with dirt He kinged my man in checkers I winged/wung the model plane so it could fly b. She winged/wung it in class yesterday gling - glinged/glang/glung Bybee provides the following explanation for the regularity exhibited in (5a): "[a] verbalized noun is associated with the verb schemata, and the strongest schema for past tense is the -ed pattern. For instance, the verb to ring formed from the noun ring has no access to rung, but will rather participate in the regular past schema". This, however, is inconsistent with the general thrust of the argument. It is inconsistent with the diachronic trends observed in (1) and (3), for instance. The main advantage of associative networks is precisely that they allow productive irregular patterns to show rule-like behaviour. If the new verb ring must be associated with ringed, then the novel word gling must similarly be realized in the past as glinged and never as glang or glung. One last alternative for the regular behaviour exhibited in (5a) would be to appeal to semantic blocking: rang is not available as a past for derived ring because the two verbs to ring have different meanings. However, on this account, we would never have expected the grammar to tolerate w-initial wring as irregular, (6a) (cf. Pinker & Prince 1988); nor would we expect the novel kinitial kling to have klang and klung as possible past tense forms, (6b).

PRODUCTIVITY AND THE LEXICON

(6)

a. ring-rang (or rung) wring - wrung

b.

275

cling-clung kling - klang/klung/klinged

To summarize, the Network Model's reliance on productivity to explain the behaviour of novel forms fails to express speakers' knowledge of the derivational history of lexical items: [ring]v → [rang] versus [ring]N→ [ring]v → [ringed]. Any attempt to explain the different behaviours of ring in terms of semantic blocking is inconsistent with what is observed across phonetically identical non-derived strings: [rij] → [rærj] or [rΛŋ] "to make a resonant sound" versus [rin] Λ [rnj] "to squeeze tightly". 3. A Hybrid Model of Inflectional Morphology To summarize, we have seen that irregulars cluster together into phonetically similar families and that this follows directly from the Network Model. However, the fact that these very same verbs exhibit regular, rule-like behaviour in denominal constructions is problematic. In response to ob­ servations such as these, Pinker and his colleagues have argued for a hybrid model of inflectional morphology, one where common irregular patterns are stored in associative networks while regular morphology is derived by rule (see esp. Pinker 1991, Marcus et al 1992). If regulars and irregulars are computed differently, they should exhibit different effects. In normal individuals, we find support for this hypothesis from morphological priming: irregulars such as sing do not prime sang, while regulars such as wash do prime washed (e.g., Stanners et al 1979, Kehayia et al 1992). In addition, regulars and irregulars should dissociate in impaired populations. This has been demonstrated experimentally by Ullman (1995): Broca's (agrammatic) aphasics have more difficulty with regular inflections than with irregulars, while Alzheimer's patients exhibit the opposite pattern, Finally, we would expect to find a population where, in the absence of a rulebased system for regular morphology, regulars and irregulars are treated identically. In specific language impairment (also called developmental dys­ phasia), we find exactly this: regular inflectional morphology is impaired with the result that regulars are treated no differently from irregulars (see below). In fact, in lexical decision within morphological priming paradigms, inflectional complexity does not affect response time in dysphasics, unlike in unimpaired individuals (Kehayia 1995). This suggests that, in the impaired population, morphological complexity in regularly inflected words is not processed during word recognition.

276

TRUBETZKOY'S ORPHAN

4. Specific Language Impairment Specific language impairment is a genetic disorder which exists in the absence of general cognitive problems, hearing loss, and other deficits (cf. Tallal et al 1991). In Gopnik & Crago (1991), it is argued that SLI is characterized by the inability to decompose inflectionally complex words into their constituent parts. As a result, individuals with SLI store stem + affix complexes as unanalysed wholes. In the absence of rules which build inflectionally complex words, impaired individuals extend patterns to new forms through analogy. However, if this accurately characterizes SLI, un­ impaired individuals must have access to a qualitatively different system for regular inflection, to one which is rule-governed. On this view, the Network Model characterizes regular inflection only in the absence of a rule-based system. The following discussion focusses on some of the results from a Wug Test on plurals (see Goad & Rebellati 1995). Not surprisingly, dysphasic individuals exhibit different behaviour when they provide plurals for words they already know as compared to when they have to construct them spon­ taneously. If this difference were merely quantitative, it could be accounted for in the Network Model as follows. It could be the case that when a new item enters the lexicon, individuals with SLI have difficulty finding phonetically related existing forms with which to associate it. Alternatively, perhaps they have difficulty extending schemas to new words. However, I argue that neither of these explanations can be maintained, as dysphasic individuals can readily provide plural-like forms for new words. From Table 1, it can be seen that Va, an adult impaired individual, provided plural-like responses for non-sibilantfinal nonce words 70% of the time. REAL WORDS

NONCE WORDS

Non-sibilant final

Sibilant-final

Non-sibilant-final

Sibilant-final

95%

58%

70%

28%

1

Table 1. Broad Analysis for Va: Percent Plural-like Element Provided

A more detailed analysis of Va's plural-like responses reveals a rather different profile from that in Table 1. From Table 2, it can be seen that only 11% of this 70% sounded truly unimpaired. The number for her normal-sounding re­ sponses for non-sibilant-final real words, though, remains relatively high: 71%. We thus come to the first problem. Va can quite successfully provide plural-like responses for both real and novel items, but she seems to be using different mechanisms for the two categories of words.

277

PRODUCTIVITY AND THE LEXICON

REAL WORDS

NONCE WORDS

Non-sibilant-final

Sibilant-final

Non-sibilant-final

Sibilant-final

71%

27%

11%

0%

Table 2. Narrow Analysis for Va: Percent Correct-sounding Plurals

In Table 3, Va's plural-like responses are divided into two types of compen­ satory strategies. The first, 'real word substitution', involves the substitution of a sibilant-final real word for the plural of a nonce word. Typical examples are provided in (7a.i): bronze, for instance, substitutes for the plural of [bram]. Real word substitution is exactly the type of strategy that is predicted to occur in the Network Model. The dysphasics search in their lexicons for a sibilant-final real word which is phonetically close to the target nonce word. REAL WORDS Non-sibilant-final \

||

Sibilant-final

A. Compensatory Strategies i. Real Word Substitution

—

|

—

ii. Explicit Plurals 24% | ' 31% B. Omission 5% | 42% C. Other 0% | 0%

NONCE WORDS Non-sibilant-final

|

|

Sibilant-final

A. Compensatory Strategies i. Real Word Substitution 22% 30% ii. Explicit Plurals 30% | 6% B. Omission 27% | 72% C. Other

2%

|

|

|

0%

Table 3. Impaired Responses by Category for Va

The existence of the second compensatory strategy, 'explicit plurals', poses more of a problem for the Network Model. These are forms which Goad & Rebellati argue arise from applying an explicitly learned rule.3 For present purposes, it is sufficient to say that this strategy results in plurals such as [wAgs] in (7a.ii), where there is no voicing assimilation between the stem-final consonant and the plural, and in [kizis] where the plural defines its own domain for stress.

3

While this rule is explicitly learned, it does not correspond to what is explicitly taught in remedial classes. Dysphasic individuals report that they are taught to add -s to a noun to form its plural. While this often corresponds to what they produce for novel plurals,, this is not the case for past tense. They report that they are taught to add -ed and yet, parallel to their plural productions, they in fact most often add [t] (see Ullman & Gopnik 1994).

278

TRUBEIZKOY'S ORPHAN

(7) a. Nonce words: i. Real word substitutions: Root: Va's Plural: [bram] bronze [branz] [LΛnt] lunches [ianciz] ii. Explicit plurals: Root: Va's Plural: [wAg] [wAgs] [kiz] [kizts] [splow] [splow's]

b. Real words: i. Correct-sounding plurals: Root: Va's Plural: comb [kowmz] cart [ka:ts] ii. Correct-sounding plurals Root: Va's Plural: leg [legz] hose [h6wziz] crow [krowz]

Goad & Rebellati argue that a single explanation underlies both the [WAgs] and [kizts] patterns in (7a.ii): the derived structures do not involve affixation, but instead, reflect compounding. Briefly, forms such as [wΛgs] violate Greenberg's (1978a) markedness constraint requiring tautosyllabic obstruent clusters to agree in voicing. Minimally, then, the plural must not be in­ corporated into the stem-final syllable. The latter receives additional support from vowel-final stems such as [splow] in (7a.ii) where, in the plural, the vowel is lengthened in anticipation of a voiced consonant or pause, and yet the consonant which follows is voiceless: [splows]. If SLI is attributed to the absence of functional categories4 — what some have argued represents the initial state in child language acquisition (e.g., Radford 1988, Guilfoyle & Noonan 1992) — the structure in (8a.i) for inflectionally complex words cannot be built. The only mechanism available for pluralizing novel items is compounding, as the structure in (8b.ii) shows. Importantly, the real word data in (7b) sound completely normal: they exhibit voicing assimilation and normal stress. Separate mechanisms must then be responsible for the differences in the plural-like outputs produced for real words versus those produced for nonce words. Given our view that impaired individuals have no mechanism with which to build inflectionally complex words, plurals of real words must be stored by the dysphasics as in (8b.i). While (8b.i) is clearly interpreted as plural, it lacks internal morphological structure and, in this way, presumably corresponds to the way unimpaired individuals store irregularly inflected words. 4

Clahsen (1991) and Rice (1994) argue instead that the problem is restricted to Agreement. They judge the dysphasics' performance on plural marking to be inconclusive: as we have seen, instead of commonly omitting plural marking, dysphasic individuals often provide plural-like responses. However, once the absence of voicing assimilation and the presence of compound stress are noticed, it becomes apparent that the derived structures do not involve affixation.

PRODUCTIVITY AND THE LEXICON

(8)

279

a. Unimpaired Representations:

Real word substitutions involve accessing representations along the lines of that in (8b.i). If a lexical search fails, however, the dysphasics are forced to derive new plurals on line. In this case, the stem + plural complex involves compounding, as the structure in (8b.ii) shows for wugs. Compare this with the representation in (8a.ii) which unimpaired individuals have for true compounds such as wughouse. To summarize, the behaviour of individuals with SLI on real words is qualitatively different from that on nonce words. I have argued that this is due to the dysphasics inability to build inflectionally complex representations. The mechanism with which impaired individuals extend patterns to new words is analogy. SLI behaviour is in this way consistent with the Network Model. If this is the case, however, then normals must have access to a qualitatively different system, i.e., to one which is rule-governed. The Network Model, it seems, characterizes regular inflection only in the absence of a rule-based system.

A Reply to Goad Joan Bybee Goad's response brings up two data sets that she argues present problems for the Network model: the first involves English denominal verbs, which have a strong tendency to be regular, and the second involves the treatment of inflectional morphology by individuals with Specific Language Impairment. I will briefly evaluate these data sets and suggest treatments consonant with the structure of the Network model. 1. English Denominal Verbs Kiparsky (1982a) observed an interesting phenomenon in English morphology: new denominal verbs seem to be protected, at least for a while, from the influence of the irregular schemas. Thus a denominal verb fly, as in He flied out to centerfieldhas regular inflection, rather than the inflection of the established verb to fly. Of course, this comes as no surprise, since the new verb to fly has a different meaning from the old one, and there is no reason to expect that it would participate in anything but the regular (strongest) schema. On the other hand, one might expect that the semi-productive class of verbs (such as the strung class) would have some influence on denominal verbs, such as to ring ("form a ring around") or to wing. Goad suggests that the lack of influence is due to the irregular verbs being stored in the lexicon (or derived at Level 1 of Lexical Phonology), while the denominals are derived in a separate rule component that has no access to the lexicon. First we should evaluate the facts. The results of denominal derivation are actually somewhat variable. Consider Goad's examples in (4). While there is no doubt that flied out and braked the car are much better than flew out and broke the car, it should be borne in mind that the new verb is unrelated in meaning to the old one (except in the first case by a long series of steps). In contrast, I do not find joy-rided acceptable at all, and much prefer joy-rode. I also do not agree with all the judgments Goad reports for (5). In particular, I do not find wung acceptable in either example, unless it is a jocular rendering, in which case it is equally acceptable in both.

PRODUCTIVITY AND THE LEXICON

281

My own judgments aside, it must be reported that certain denominals are irregular, for example hamstrung, as in The new regulations really hamstrung George's operations. And there is a tendency for denominals to eventually find their way into the irregular classes: stick, stuck is a verb derived from a noun. Despite the variability, I would not deny the phenomenon: newly derived verbs from nouns have a strong tendency to be regular. Moreover, there is something correct in Kiparsky's treatment of these items — it is precisely their relation to the noun that protects them from being fully incorporated into the verbal network. In Kiparsky's framework, this idea is expressed by sets of bracketing: [y[N wing]N]v]. In the Network model, this same phenomenon can be expressed in the degree of association of the new verb with the noun from which it derived. After all, its meaning is directly based on the meaning of the noun, so the semantic connection will be strong:

When the verb is activated, the closely related noun will be activated much more strongly than any of the phonologically similar, but semantically unrelated, verbs of the strung class. (Morphological evidence indicates that semantic relations are much stronger than phonological ones, Bybee 1985). I predict that when the verb to wing changes its meaning enough to become emancipated from the noun from which it derived it will become susceptible to influence from the semi-productive strung schema. 2. Inflectional Morphology in Individuals with Specific Language Impairment Gopnik (1990) and Gopnik & Crago (1991) describe Specific Language Impairment (SLI) as a speech and language disorder that disrupts the ability to form morphosyntactic rules for tense, number and gender, in the absence, as Goad puts it, "of general cognitive problems, hearing loss or other deficits". More recent research with the same individuals that Gopnik tested reveals other areas of grammar are affected, as well as general intelligence (affected family members have verbal and performance IQ scores on the average 18-19 points below those of other family members). In addition, affected individuals have "grossly defective articulation of speech sounds, and, further, a severe extralinguistic orofacial dyspraxia" (Vargha-Khadem, Watkins et al. 1995).

282

TRUBETZKOY'S ORPHAN

Gopnik (1990) reported that these individuals with SLI produced irregular past tense correctly, but failed to use regular pasts correctly. This report was based on only four verbs: was and went were the irregulars and kissed and walked were the regulars. Obviously there was no attempt to control for frequency, Subsequent tests by Vargah-Khadem et al (1995) using a much larger set of verbs found no significant difference between the subjects' performance on regulars and irregulars. I mention these conflicting reports in order to highlight the difficulties of studying a language disorder even in the same individuals in the same family. Given that we know very little of the source of the deficit or the extent to which it affects linguistic and cognitive domains, it seems to me to be pointless to try to draw conclusions about a theory of normal language processing from data obtained from individuals with this deficit until much more is known about the deficit However, I will comment on Goad's account of one individual's attempt at pluralizing nonce nouns in order to draw attention to the inconsistencies in her account. Goad's subject Va is able to produce correct-sounding plurals of real English nouns from 71% of the time for non-sibilant final nouns to 27% of the time for sibilant-final nouns (see Table 2), but for nonce words her per­ formance is much worse: only 11% of the non-sibilant final nouns sound correct and none of the sibilant-final ones do. This suggests that Va is able to store plural nouns but is unable to extract a pattern from them. This could either be because she does not associate them with the appropriate singulars and consequently does not analyze the plural morpheme, or because she is not able to associate them with other plurals and extract a generalized plural pattern, or both. What Va is able to do, according to Goad, is to form a kind of compound of the nonce noun with an [s] ending (as though she had been taught that plurals are formed by adding s or es). This demonstrates that a certain amount of explicit concatenation is possible for this individual, perhaps resembling a syntactic operation more than a morphological one. To this point it is easy to agree with Goad's account, but she goes on to make two claims, one that is false and one that contradicts her own account. First, she claims that dysphasics store regular plural words unanalyzed, which "presumably corresponds to the way unimpaired individuals store irregularly inflected words". It is of course not true that unimpaired individuals store irregularly inflected words unanalyzed: they form connections with related words in the same paradigm and in other paradigms that show morphological similarities. Schemas emerge from these connections.

PRODUCTIVITY AND THE LEXICON

283

The second claim is that "the mechanism with which impaired individuals extend patterns to new words is analogy" (Goad, 1996:299). In the previous paragraphs Goad has argued that such individuals use compounding to form new inflected words. If they were able to use analogy, the results would sound normal. Also in order to use analogy, they would have to be able to analyze stored inflected forms and extract patterns from them, an ability that Goad claims that they are lacking. It appears rather that they are unable to use analogy, and must resort to a more syntactic type of operation — concatenating two independent units, a noun and final [(I)s]. Thus it is not true that the deficit in individuals with SLI is more easily explained in the Dual Processing model (the hybrid model) than the Network model. In the Dual Processing model one can hypothesize that SLI individuals are unable to use the rules component, but in the Network model the hypothesis is also simple: individuals with SLI have an impaired ability to form associations among stored words and/or an impaired ability to extract generalizations from these relations.

Productivity and the lexicon Discussion BYBEE: I would like to point out that the arguments against the connectionist model with regard to the notion 'lexical item' don't apply to my model, where there is in fact a notion of lexical item. The way that the connectionist model that Lieberman and Prince were arguing against worked, you just had masses of phonological connections among forms, and ring the noun isn't any different from ring the verb. Their argument, which is quite correct, is that you have to have some notion of a stored image. I think all of us would recognize that. In accounting for compounds such as joyride, the idea is that once joyride is formed as a word, it accesses the regular schema, in order to get joy rode you have to take the past tense of ride, which is not as available. GOAD: May I use the case with ring, because it's one morpheme? If the network associates ring and rang, and you were to introduce a new verb ring, which is presumably the case when you derive it from the noun, then wouldn't it be just like a new verb ring and would it be associated with it? BYBEE: That's a good point. However, you do have to realize that stick/stuck was originally formed from the noun stick', string is another one. They came from verbalizations and they did enter a strong class, so the question is whether they entered it right away or not. Now, one of the things about stuck and strung is that they have initial consonants that favour this ablaut class. The only major point that I would like to make is that the deficit in these subjects is the inability to form the appropriate connections among the inflected forms. They have learned a plural but have not made the association of that plural to other plurals and may not even have a good connection of that plural to the singular form. In my model they have the ability to learn new words, or an ability to build up a sizeable vocabulary, but lack the ability to form connections. GOAD: It just seems to me that some subjects favoured compounding much more. BYBEE: You mean the kisses kind of compounding. GOAD: I still don't see how you could get that kind of structure, because it

PRODUCTIVITY AND THE LEXICON

285

seems to me that the only way you would end up with a compound with that kind of structure would be for these patients to actually map the representation of kiss onto their other representations of compounds. BYBEE: Are these children or adults? GOAD: The cases that I'm citing here are from an adult, aged 40. BYBEE: Well, just another hypothesis about what might be going on is that she could have learned a strategy. Obviously these people spend their lives trying to figure how to cope with this language thing and she could have learned this strategy of compounding. TiFFOU: You see, I am interested in your theory but I want more information. It is nice to have a good description of what is going on but I would like to know if you have some strategy to see how there production happens. You see, I was told in the Paul Kiparsky lecture that there is a certain productivity. In your exposition, I got the impression, it is static but perhaps I do not understand, and I think you have some ideas about that. In derivation you can't form new words but from time to time you notice words but you can't make them, and I don't know why. For instance, you have in French boulangerie, cordonnerie, et alors on forme des mots comme pantalonnerie, croissanterie, but I don't say vinerie, where I might buy some wine, I can't create journalerie. Why is it impossible, given that there are a lot of cases like that? So, I think, in theory we have to be a fanatic for description and, for my part I am convinced by your lecture, but lack of productivity has to be real and if not, how can you explain this impossibility of applying certain processes? BYBEE: So, part of your question, as I understood it, is about actual processing, or did you just mean productivity? TIFFOU: Productivity according to predictivity. If you see this is going on, you can see that we have this possibility or that possibility. There is, you see, a relation between productivity and predictivity. BYBEE: I certainly agree that there is a tremendous amount of detail that needs to be described, and that there are lexical gaps. Perhaps those will eventually yield to analysis. I'm not trying to tell people not to describe languages, because naturally that is the right thing to do and other factors constrain derivation. KIPARSKY: I want to ask three questions. The first one is actually connected with the question of productivity that you discussed in the commentary and I will preface that with a comment. One of the criticisms that you directed at the Lexical Phonology and Morphology model is that there is no account of productivity. But in fact on any account that incorporates something like the

286

TRUBEIZKOY'S ORPHAN

Elsewhere Principle, the organizing principle is that for every morphological process there is always an elsewhere case, the unrestricted form which picks up all the cases which are not provided for by more specific rules and processes. So, intrinsic to Lexical Phonology is the Elsewhere condition. That relates to the question of -s which you brought up of a case where high frequency seems at odds with productivity. BYBEE: German plural -s? KIPARSKY: Plural -s in German, yes. The other plurals that you mentioned are assigned to Level I, and all of them have specific phonological and/or morphological conditions on their occurrence, -s is the one German plural that does not have specific phonological or morphological restrictions on its occurrence. You see that -s is the only Level II plural suffix quite in­ dependently of its productivity in the fact that it is the only one that can't appear in compounds as in Strasse, Strassenbau, Film, Filmemacher, Kind, Kinderpflege. At least I am not familiar with any nouns like *Autosfahrer in German. The ones that do occur are a sort of s which is not a plural, because it occurs on nouns that form their plurals in other ways, as in Liebeserkldrung. The question then that I would like to ask in that connection is what exactly was your explanation for the productivity of this suffix in German. You seem to argue that it has something to do with the fact that it is the favoured or I would say even more strongly the near-obligatory suffix on words that end in full vowels, which is true for native German words like Papa but also foreign ones, like Auto. But what accounts for its productivity in words ending in consonants, e.g., die Jungs, die Gophers? BYBEE: I'm sorry, that didn't quite come clear. What I am claiming is that type frequency is very important to productivity and what you actually see in the German plural affixes is that -s is not very productive. It does not appear in loan words that are more like German words, i.e., that don't end in a full vowel. It is also influenced by lexical distribution because it is more likely that -s will apply to nonce words ending in a full vowel, exactly that type of word that has -s in the German lexicon. Thus it has the property of being influenced by lexical distribution, which is supposed to be a property of irregular morphological categories. What I'm claiming is that it is as productive as it is because it has a totally open definition, it can apply to any kind of word, but it's not very productive because it has low type frequency. With regard to the compounds, actually we need more data on that. The trouble with Clashen's paper on that topic is that he is working with the same kind of dysphasic children that were just reported on here, and he. claims that the ones who use -n as their regular plural don't put it in compounds.

PRODUCTIVITY AND THE LEXICON

287

However, there are normal children who also omit plural inflection from compounds even where it belongs, so I feel like before I could take a stand on that I would really need to have a lot of information about German. KIPARSKY: Fair enough, except the answer is very similar to the answer that would be given in Lexical Phonology, that is the productivity of -s such as it is depends on its unrestricted character, the absence of constraints. BYBEE: What we would expect, what I was trying to explain was the different degree of productivity in German -s and the English plural -s which is much, much more productive because of its high type frequency. What is your next question? KIPARSKY: Well, the next question relates to a criticism that was made against Rumelhart and McClelland's connectionist account of plural formation by Pinker and Prince, which I thought was quite convincing. I just wanted to understand whether yours was open to that or not. It was that because they consider phonology itself, or phonological-morphological principles, as emergent properties of the model, and didn't build in constraints on rules, they didn't have an account for why any arbitrary formal relation between forms couldn't be the basis of a productive formation. If we just gave enough examples of it, the model would, for example, give metathesis. So the question is whether you know the reason, intrinsic to this model, why you wouldn't form plurals by reversing the phonemes, so that the plural of step wouldn't be pets! That is, do we assume some universal phonological principles as a basis which are wired into the mind? BYBEE: That's a good question. What you need for this model to work is some notion of phonological or phonetic similarity, because all these connections or relations are of identity or similarity. Clearly there are relations of segments independent of their ordering in sequence. Words are whole el­ ements and the order of segments is just as important as their particular phonetic characteristics, so I would need to think about how to rule that out, a reversal if in fact that should be ruled out. It is possible that you could show experimentally that people could learn something like metathesis (actually I doubt it) and thus it should not be ruled out. A lot of our relations of similarity have a diachronic source and they are governed by what sorts of changes can occur. There is no diachronic source for the reversal of phonemes. It is important to distinguish those two aspects, the actual synchronic psycholinguistic constraints and the constraints due to possible diachronic sources. KiPARSKY: The similarity itself is a pretheoretical notion and you have to

288

TRUBETZKOY'S ORPHAN

define the space and the parameters that make things similar. What a computer or connectionist network might consider similar might actually be very different from what a human being would consider similar. The third question is about your criticism of Lexical Phonology relying on instances of alternations that were exemplified by just a few forms. You cited damn, damnation etc., but in fact the constraint of which the exclusion of mn is a core example in English is extremely general. It doesn't apply just to that particular cluster and it isn't in any way marginal but it is a very fundamental principle of English syllable structure. I don't think at least currently anybody would argue for a specific rule that says drop n after m, rather this is just one instantiation of a constraint on English syllable structure. BYBEE: But you would still have the n at the end of damn even though we don't say it there; and you have to postulate a relationship between these words. You're saying there is a relationship between damn and damnation that is productive. KiPARSKY: Well it certainly looks like there is a relationship between those words. The point is you need to say nothing except just state the syllable structure of English in order to account for that. SINGH: First, a couple of things, maybe three. Number one, it seems that the discussion of productivity that you provided us with actually could be paraphrased as breaking down the dimensions of any rule system of any kind that could describe these morphological relationships, which means that in some sense you are in fact arguing that there is no reason for any grammar to actually talk about it. I mean, even Kiparsky's Elsewhere Condition might not be necessary if the position of these parameters can be read off the structural complexity of the system of rules by which this morphological system is to be described. At the level of the system, there is no need to talk about it at all. Whatever kind of rule system you use they will have a certain degree of complexity which you can break down into various categories and you can decide how these sort of stack up or pair up against each other, which dimension plays what role, whether the degree of fusion can be read off in terms of the phonological composition of the string and so forth. So, in essence, nothing is really needed at the level of representation, at the level of grammar. We needn't say anything about this at all. Since it doesn't matter what kind of rule system you use because these are apparently dimensions that can be read off directly from any sort of rule system whatsoever, so why should it be necessary to talk about this at the level of the system? BYBEE: Because I'm not making a distinction between competence and performance. I'm not just describing the forms and their regularities, I'm

PRODUCTIVITY AND THE LEXICON

289

really asking how this works with real people, and they don't have these metrics unless you build them into the system. So for example, high type frequency: if you just take an ordinary generative grammar, then you can go to the lexicon and count up type frequency and that will predict productivity, but that's not going to tell you exactly how the real speaker creates a new word, for example, so I feel it has to be built into the system. SINGH: But if it has to be built into the system, aren't you giving it the same representational status that I thought you had argued against? You see, your quarrel with the last study was that some representational status is being given to two kinds of morphological processes, one represented this way and the other represented some other way, which particular way doesn't really matter, and then you're saying that these things really need to be built into the grammar, the precise way you want to represent them is really not the issue, rather than being read off the rule system, whatever rule system you choose to employ. I don't see the distinction at that level of abstraction, if you are building this into the system of grammatical description. You could build it in any number of ways, and then the question really becomes which is the right representation but both of them are actually providing a representation status to that distinction. BYBEE: Okay. DRESSLER: I have two objections to the interpretations of these German plurals. I have two objections for these German plurals. The first thing is that you can describe with some success in a language like German the difference between regular and so-called irregular verbs in a similar way as in English. Although I would challenge you that you would have serious difficulties to explain in your terms what Andreas Bittner has done in a very clear description of the diachrony of how so-called German irregular verbs are diachronically slowly regularizing in a certain type of way, but if we come to the nouns then clearly I think you cannot do it in this way and the Elsewhere Condition won't work either because the German nouns have to be described like Latin nominal declensions. You must simply say there are that many classes and each class has its properties and you can say what within this class is regular and what is not, with the Elsewhere Condition, but not for the whole declension. So, if you are talking about these -s plurals, then this is a specific class, and this specific class happens to be open to many foreign words, but still they can be put into another class. So, you have Pizza, Pizzas but you also have people who treat it like Villa and say Pizzen. What is interesting for diachrony, is that there is one class which slowly expands diachronically, and this is a class you do not have here and this is the class

290

TRUBETZKOY' S ORPHAN

der Mops/die Mopse, which a hundred years ago was der Mops/die Mopse, der General/die Generate, which a hundred years ago was, with the exception of Austria, der General/die Generate. So, if learners of German very often make the error der Hund/die Hünde, then they're simply more advanced than today's native speakers, because I think that in a few hundred years this might be standard in German. This is a type of productivity but this type of productivity is one which lends itself to an interpretation in terms of schemas. I think this is also a big problem for Kiparsky and rule analysis if I might say so, referring back to your paper, because this works in a schematic way. There are new findings by Kopcke who shows that the probability of having an umlaut plural is loaded according to properties. The more semantically animate or human the word is, the more likely it is to have an umlaut. The more consonants you have in a monosyllabic word, the more likely it is to get an umlaut. And this is something you cannot describe with any morphonological or lexical phonological rule and which is really schema­ like. This is very different from the other type of rule productivity. As to the so-called (compound-)internal plurals, in German they are all interfixes, they have come diachronically either from genitive singulars or from plurals. Synchronically, both have to be treated in the same way and therefore I think this problem of plurals does not apply here. I have a second very general question. And this general question is: could you explicitly say what your paper has to do with morphophonology? BYBEE: No. DRESSLER: It seems that you are treating this simply as a sort of irregularity and you show how irregularity versus regularity is treated in your frame­ work. Does that mean that there is nothing specifically for morphophonological plus/minus regularity, nothing specifically for morphophonological plus/minus productivity and for opacity, 'fusion' as you call it? BYBEE: So, you mean is there a component? DRESSLER: Not a component, but if there is something common to what we call 'morphonology'. BYBEE: Well, I still believe that morphophonology consists of phonological alternations that have morphological conditioning or some kind of morpho­ logical import, so that certainly means there is some morphophonology. I suppose I could have extracted all the various things that I call morpho­ phonology and talked about their properties but I feel I have already done that in the past when I talked about the fact that stem changes are more often conditioned by highly relevant categories than they are by more peripheral ones. That is something that goes under the heading of 'morphophonology'.

PRODUCTIVITY AND THE LEXICON

291

DRESSLER: Does that mean that you stick to the views in your first book? BYBEE: Well, no, not entirely, and there are various reasons for that. That was a kind of generative model, and it turns out, as we have already seen, to be very difficult to decide whether we do allomorphy by replacement or se­ lection. Eventually I think those questions turn out to be non-questions and it is simply time to look at things in a different way. DRESSLER: SO we may apply the Elsewhere Condition to you, saying what you haven't changed in having this Network model, there still your first book would hold for you? BYBEE: I don't know. I haven't done a check to find out. KEHAYIA: It seems to me that you mentioned that lexical representations are not dependent on structural criteria but rather are determined by language use. BYBEE: I said that. KEHAYIA: So then I'm trying to figure out how lexical connections are built. I'm looking at figure (1) and figure (2); there's plan and then there's planning. How are these lexical connections? BYBEE: Well, what I assume is that when a child is learning the forms of his or her language, with every new word in input, there is an attempt to relate it to something already stored. So whether plan or planning or hop or hopping comes in first doesn't especially matter, what matters is that given that one is already stored and one is new coming in, relationships of similarity, both phonological and semantic, are sought and when those relations are found the lexicon takes on this kind of structure. KEHAYIA: What's the other alternative you had in mind? If internal word structure were to be represented then they could possibly relate higher than the common root. BYBEE: I don't see the difference between that and what I'm doing here. KEHAYIA: I'm just asking what are the determining criteria for linking up words and there is actually two questions. By looking at figure one, it seems that the affixes are also linked to each other. BYBEE: Yes. KEHAYIA: There is quite a bit of data available from psycholinguistic experiments, priming experiments and lexical decision ones, that show that there is no priming between affixes, so if planning is used to prime talking there is no prime at all. The conclusion that they draw is that actually affixes do not prime each other, that is one thing I would like to bring to your attention. The other is that it is also known from psycholinguistic exper­ iments that irregular verbs do not. So the present tense does not find its past

292

TRUBETZKOY'S ORPHAN

as well as the others. To conclude this comment on psycholinguistic exper­ iments that I mentioned earlier, we have found that in genetically dysphasic children, whereas normals would take longer to recognize an inflected word, these are children and adults, actually treat them as a chunk. So washed get an identical or quicker reaction time than wash. It is exactly the contrary from normals who seem to be taking longer when recognizing inflected or derivational complexes. BYBEE: Well, to start with your comment about the priming of irregulars versus regulars, that was actually one of the original things that got me thinking this way, because I figured that if sat doesn't prime sit as well as hop primes hoped, that means that there is more distance between sit and sat than there is between hop and hopped. The phonological difference affects the degree of relatedness of the words. My 1985 book argues that degree of phonological relatedness goes along with expected types of morphophonemic change and with certain types of categories and conditions. It's also the case, you'll probably recall, that if you ask someone to give you the past tense of a verb and it's sit versus walk, then sat is actually faster: you access sat faster which also suggests to me that it has a strong lexical representation. KEHAYIA: Very strong means frequency, it's a high frequency word? BYBEE: And it has its own representation. KEHAYIA: If sat has its own independent representation... BYBEE: Not independent. KEHAYIA: Then what about walked! BYBEE: Well, it may have its own representation too, but it depends on its frequency. What we don't know about the priming experiments is whether or not we could get a real frequency effect. KEHAYIA: We do know with inflected forms, regular and irregular. BYBEE: Then that would support what I was saying. KEHAYIA: Actually frequency effects do exist. However, structurally motivated effects also exist and these effects in some experiments seem to override frequency effects. BYBEE: I could offer something from the experiment, but let's just leave it. JANDA: Three quick comments. I don't see why the model can't defend itself against ringed and so forth, it's known from diachrony that nouns often cause analogy in verbs and lead to the ironing out of alternations. In Spanish huebla generalizes from hueble to cases where there wasn't diphthongization on the verb and there are many analogical cases like that. So you can have a representation that can have a link to a noun which would be distinguished, so I propose an experiment where you would create glain as a noun and then

PRODUCTIVITY AND THE LEXICON

293

talk about it for a while, and say "Gee, I think that person does glain so much" and "I think they glain all time" and then you can find out what they do when you suddenly switch it, so representationally you have a way to distinguish the ringed cases from the other cases. And then to the German: it is also quite common in diachrony when there are many competing patterns, perhaps it comes through the Latin pattern, everything is divided up, it's as if the different declensions divide up the world and say I'll dominate this phonological zone and you have that one, it's hard to find an Elsewhere case, but one often comes in as the Elsewhere, the general case spreads, often some of the little ones spread at the same time, so as the tide goes out for the other ones, often they can leave seaweeds on some rock and you get the generalization of the game plural in English, where you have a lot of animals have zero plurals and that has survived and you can say now "I went out and bagged two cow" if you want to, if you're talking about cowsas game so because generals and other humans spread umlaut, I don't think that protects the spread of the -s, so I just want to end with this third observation. I won't stake my life on this because I'll be dead by then, but I do think what I predicted in this article is that -s will eventually win as the German plural. It has spread mightily in Dutch. Once it gets on the genitives, watch out! That will really be a sign that -s is going to go in. It comes in as a beggar and then becomes a mugger or a kidnapper, gradually it gets nouns from other areas, you can see this thing spreading and you have got to have a default eventually, I think, even if languages of this other type tend to divide every­ thing up, foreign words come in, phonological reductions take place and I'll end with this diachronical and morphophonological aspect, perhaps. Shwa, Shwa+r and shwa+n don't have a future as endings over time but a very strange generalization, languages with initial stress don't aspirate their -s's, they will do anything besides that, whereas languages that have pitch accent and calculate stress from the end of a word aspirate their s's and lose them. German is the right to keep the s around or at least turn it into something with more substance, so sub specie aeternitatis, -s will win. [NGH: One minor clarification, talking about the English plural. If I heard you right I thought you suggested that the frequency of these plurals in the lexicon is important. Is it possible that what really matters is not that but the fact that the phonological contexts in which this plural shows up have nothing in common. The generality, in other words, is a result of the fact that the words with which it does appear share nothing phonologically at all. ANDA: That's lack of motivated restrictions, that's all. YBEE: That's what I was thinking, the more open definition. I believe that's

294

TRUBEIZKOY' S ORPHAN

right: the speaker doesn't come up with a restricted definition of where that -s goes. SINGH: Actually it's not that he comes up with a restricted one, he is sort of almost forced to come up with the most unrestricted one, just by the fact that it is there. DRESSLER: I would like to answer to Janda. I would say that your test should be still more complicated because what you cannot explain is the following: you tried to explain why a verb to ring, which is derived from a noun ring doesn't undergo ablaut, but note that if you derive by conversion a noun from a verb, then according to you this should somehow diminish verbal ablaut. For example, you can form the sink from to sink. This should diminish the use or the the possibility of the ablaut in to sink, but this is not the case. So I think this is a very good case where one can say that such noun-verb relations are not really bi-directional but you must know which base is converted into which derived word, and this speaks for the rule model. GOAD: If a noun is linked to a verb, and it's precisely for that reason that you get regular rather than irregular, when it does regularize over time, you would have to delink your verb from your noun representation. DRESSLER: Very briefly, I think what is important is that you differentiate between the two different types of irregularity. I have called the first one 'arbitrary lexical restrictions', a question of generality, and the second, 'regularity', where regularity means that you can write it in a simple rule. I think that these two things are really not related. What I was getting at and here you were not correct in citing me, is that you get a preferential relationship between productivity, regularity and due to the fact that they are indirectly correlated because they are derived from the same semiotic principles, but this is a very indirect and only preferential relationship. Now, how can one try to empirically differentiate? Well, diachrony is a very good test, because it is not true that nearly all of morphology has this source. Morphological rules come also from phonology as we have seen. They arise due to reanalysis of something else, like resegmentation and then they behave very differently from the original state, as in language contact, e.g., in Middle English from the many Romance words, certain rules have been extracted.

VI Desrochers and Dasgupta SOME ADDITIONAL CONTRIBUTIONS

But the issue is not clear: in actual fact, we are probably in covert disagreement as to what our overt disagreements are about C.F.Hockett (1955:8)

Issues in Morphophonology: A view from the floor Richard Desrochers Universite de Montreal 1. Introduction In response to the invitation to contribute a view from the floor, I should perhaps begin with what some may be wondering about: why the question of morphophonology should come up again at all, since the main issues in what are seen as dominant phonological theories don't focus on even related questions, like abstractness? The fact remains that phonological practice has changed over the last thirty years in ways sometimes essentially and tightly linked with what stands for 'morphophonology', whether one speaks in terms of Lexical Phonology (Kiparsky 1982a, 1985, 1993, 1996 and Mohanan 1990), Natural Phonology or Morphology (Dressier 1984, 1985a, 1996b) or Integrated Morphology and Phonology (Singh 1987, 1990b, Ford & Singh 1991). Even for models that don't seem to settle domain decisions, as may be the case with constraints-based models (Paradis 1993, Prince & Smolensky 1993, Scobbie 1993, McCarthy 1993) and underspecification theories (Kiparsky 1985, 1993, Archangeli 1988), morphophonology is bound to show up under one or another form probably sooner than we may expect.1 1

One of the main problems towards this admission is our reliance on ill-defined concepts and the understandable tendency to oppose atheoretical notions like 'naturalness' and 'abstractness'. When Singh (1986:345) says "one of the reasons we have as many theories as we do is that the debate on the need to have something called morphophonology was somewhat prematurely shut off, and that 'abstractness' is "a crippling pseudo-issue, and must...be dropped forever" (353), he misses the opportunity to point out that it is actually because labels like 'abstractness' and 'naturalness' have been applied to intuitive content with hyperformalized representation. So-called 'abstract' analyses are still a common practice, if by 'abstract' we mean the direct and straightforward way to grasp 'regularity' at the expense of 'automaticity'. The postulation of final schwas in French or descriptions of French liaison with floating consonants are to me a phonological guise for something which pertains to morphological or lexical conditions having nothing to do with the optimality of CV syllables or preferred syllable structures. The insistence on 'representationability' is at the core of the confusion Singh talks about between domains. If phonemic systems look now closer to those obtained by classical phonemic principles, this is because the 'abstractness' of the analysis has been transferred out of the segmental representation onto syllabic

298

TRUBETZKOY' S ORPHAN

My intention is to expose my reasons for believing that there must be some kind of distinction between morphophonology and phonology, that is, some significant, i.e., qualitative one, at some point, between the two. I should, however, turn my attention to some specific issues that, in my opinion, will have to be raised before we can make sense of questions about morphophonology. Before I proceed, I should clarify some of the basic ingre­ dients of what I take a satisfying account of phonology to be.2 • phonology concerns automatic alternations and is constituted of (i) a small set of obligatory processes and (ii) a considerable repertoire of phonetic, option­ al processes, both with articulatory (and maybe perceptual) motivation. • there is nothing phonological in Velar Softening in French or English, French nasalization, French liaison or German umlaut, although I don't know what they are. • rules are not a very appropriate means to formalize processes, because they look like something to add to a grammar, which, as Stampe showed, they are not. On the contrary, they are things to substract from the grammar. WFCs of the Singh type make more sense in that they are language particular provisos on universal processes (for which, in Singh's model, universal repair strategies stand for).3 • the mechanism of WFC/repair strategies, as to the choice of repair, is governed by a preference for allophony and the 'minimality' repair, everything else depending on the formulation of the WFCs and on the satisfaction of other repair strategies (cf. Desrochers 1994a). constituents, floating segments, empty nodes, feet, extrasyllabicity and extrametricality. It is not abstractness which is a matter of debate, it is the fact that abstractness was or still is a convenient disguise for a dirty word like 'morphological', and the question should really have been "How morphological is phonology?" 2 Except when, of course, definitions of some domain within some specific model or for some particular individual are discussed, I will use terms like 'phonetic alternation' or 'phonetic rule' for alternations which are non-automatic, optional, 'low-level' and whose application are either of a probabilistic order, or unpredictable in their precise, 'gradual' effect. I will call 'phonological' any obligatory, automatic alternation whose conditioning is 'transparently' phonological. I will then call 'morphophonological', alternations which either have exceptions, or that may be described straightforwardly as having morphological or lexical conditioning. These are neither formal definitions nor really unexpected ones and they are not wirhout problems. But since the question is to find out criteria defining domains, with the meaning intended by Singh for that term (1986:343-45), these approximate definitions will suffice. 3 But the optimal representation, if possible, would be more accurately anti-processes, because inhibition of a process is what a child acquires, not the trigger for (another) process. What a child learns in the phonology of his/her language is not the processes he/she must apply, but those he/she must not, and the eventual restrictions on those he/she can.

299

SOME ADDITIONAL CONTRIBUTIONS

It is not a mere accident that I have left aside precisely what constitute morphophonology: it is because I have not found a place where it would fit naturally, although I know where it does not. I can now take up the three models that sought to provide explicit answers to what and where morpho­ phonology is, or what and where it is not. 2. Three Models of Morphophonology The desire to constrain relations ruling between the surface level and a deeper level has led some to return to a rather classical phonemic level, and some others to limit possibilities and choices of analysis for the elements at the deep level, if not to limit rules themselves. The rule format, still preferred by Lexical Phonology (LP) and Natural Phonology/Morphology (NPM), may have been replaced in other models by constraints, of a more abstract nature, but the only one among constraints-based models to be explicit about the place of morphophonology is Generative Phonotactics/Word and Process Morpho­ logy (GP/WPM) (Singh 1987, 1990b, Ford & Singh 1991), so that the com­ parison should be limited to these three models. Each of these models happens, in a more or less direct fashion, to say something about both word-formation and low-level processes, so that each of them may be considered as a theory of the form of grammars. Some of their particularities could roughly be summa­ rized as in (1), assuming the informal definitions I have already given: (1) Word-formation Morphophonology

LP unidirectional rules morpholexical rules lexical rules

Phonology

lexical rules

NPM unidirectional rules MP and AMPrules P rules

Phonetics

post-lexical rules

(stylistic) P-rules

GP/WPM bidirectional rules part of WF rules well-formedness conditions stylistic rules

Some important differences are not displayed in this figure, especially the motivation for rules. They are not necessarily explicit in some cases. Lexical Phonology is still a rule-based model, with considerations of naturalness mostly in a 'markedness' framework, so that evaluation of rules, segment types (default values for features) and alternations is empirically founded on generalizations more often having a statistical basis than a physiological one (cf. Desrochers 1990). In the case of NPM/NM, there is a scale from WFRs to PRs of decreasing semiotic motivation and increasing pronounceability/

300

TRUBETZKOY' S ORPHAN

perceptibility. This last type of motivation occurs in GP/WPM for the last two levels, and pronounceability seems the main motivation for the phonological level. Another important difference, partly linked with the first, is underspecification, admitted by LP, not to say considered indispensable, and closely related to markedness. The other two models reject underspecification, both roughly on the same grounds (Dressier 1984:32; Singh 1986). One is tempted to say that in LP, 'default rule' is a derived concept, the 'primitive' being markedness, whereas in the other two models, 'process'/'WFC and repair strategies' are primitives. In the latter cases, though, primitives are also founded partly on empirical generalizations (but much less on segment types than on 'process type'), in addition to the 'phonetic teleology' criterion. One important difference is also the gradience/discreteness parameter in the properties of rules (or what stands for them). Dressier (1985a) proposes, as is well-known, a scale along which rules have a more or less phonetic function and conversely a more or less semiotic function according to the combination of different criteria (generality, productivity, phonetic distance between input and output, indexicality, etc.), though he speaks of "typical" PRs or MPR's. Kiparsky (1996) admits of many "fine-grained distinctions", but maintains that there are "core properties", and that these override, at least methodologically, the blurring cases. Singh (1987, 1991) and Ford & Singh (1991) admit only a clear-cut distinction and allow only for three types of alternation: stylistic optional processes, phonological processes and the alternations that constitute part of morphological operations. The first question presents itself as a choice between these three positions: 1° morphophonological alternations are of the same nature as automatic, phonological alternations, but some alternations (morpholexical) would be part of morphology. Different criteria may serve to attribute alternations to one or the other component (Kiparsky 1996). 2° morphophonology is a transition between phonological rules and wordformation rules. Differences along this scale are more functional than formal: P-rules display a high degree of iconicity, low indexicality and the least degree of semiotic function, whereas morphological rules have a high degree of semiotic function and low degree of physiological motivation. MP-rules are somewhere between these extremes, and they may display more or less of a P-rule quality and more or less of M-rule quality. 3° morphophonology is part of word-formation, and any 'rule' or statement accounting for morphologically bound alternations is part of a WFR. Before I take up considerations that were presented for constituting MPP as

SOME ADDITIONAL CONTRIBUTIONS

301

semi-autonomous or as part of word formation, I should perhaps turn to my reasons for thinking that MPP is not phonology. 3. The Non-Natural Basis of Morphophonology Kiparsky (1973) had already mentioned the difficulties for morphophonological rules or units that would treat morphophonological alternations without a Naturalness Condition of the kind proposed by Postal (1968). I would agree with the proposal that the naturalness of MPP 'rules' or alternations is a mere accident (cf. Ford & Singh 1983), that is, it is most of the time inherited from phonology, and some have lost anything that could be ever called 'natural'. A first case that comes to mind is the celebrated Velar Softening rule common to English and French. An alternation like k ∞ s can never be 'allophonic' from what we know about sound systems: opposition between k and s is nearly universal and when it is not exhibited in a system, we do not expect it to be 'intrinsic allophony'. Technically, there is no natural process that would change k to s, because there is no violation involving k and, say, a front vowel, that could not be repaired by changing k into some segment more similar to k, even if it can be as distant as t or c. The most extreme case I know synchronically is a k ∞ č alternation in French Acadian dialects (Lucci 1968). If we look at this alternation in terms of the features involved and without recourse to underspecification, markedness conventions or the like, we have a change in the values of five features at least, and this would make this alternation impossible to state elegantly without a supplementary device, which calls for strong independant motivation.4 Phonetic distance between input and output is not a sufficient condition to exclude putative morphophonology from phonology since many MPRs look really natural and do not involve more than minor changes whereas a number 4

Worse cases can be found, as in French with words of the type réductionN/'réduirey, constructionN/construireV, inductionN/induireV. Any phonological analysis would have to state a k ∞ i alternation in a more or less direct fashion. There is no happy way to maintain that a rule describing this alternation would respond to any criterion of naturalness, especially in its formulation in or out of the standard SPE format, without assuming the massive use of linking conventions or default rules. This assumption may look well-founded, since the aim of this mechanism is to 'simplify natural rules'. This is indeed a perfectly circular argument: such rules have to be simplified because they don't look like natural rules, and they don't look like natural rules because, quite simply, they are not, and not, as SPE pretended in its chapter 9, because of an excessively formalistic approach: markedness theory, on this score, adds insult to injury, and nowhere could anybody find such formal excess as in the theory of Kean (1980). Naturalness of rules should then be measured, if the simplicity criterion is to be maintained, by a thorough comparison of the rules that never needed linking rules or default rules with those that crucially need it. The alternation would be morpholexical for Kiparsky (1996), since it is not limited to one segment.

302

TRUBETZKOY'S ORPHAN

of phonological rules involve drastic changes, like nasal consonant assimilation. Among the battery of criteria proposed by Dressier (1985a), one could think of 'process matching in a relevant environment', and of 'child phonology matching' (these two conditions being of course among the necessary conditions for a 'rule' to be given the status of a process). The first one would then admit something like 'palatalization in a palatalizing environment' or 'voicing in a voiced environment' and would not favor a k∞s alternation in a palatalizing environment (nor, for that matter, in any environment) or a k ∞ i alternation in a non-palatalizing, non-vocalic environ­ ment. The second one would also exclude these two alternations from phonology, unless they are attested among the extremely rich stock of processes of child phonology. What stands for independent motivation in the case of a rule like Velar Vocalization is the mere fact that there are other words to which it could be said to apply, as for the alternations inlait/lact-[lε/lakt-] or huit, oct. But every occurrence of this 'rule' is tied to some other very idiosyncratic rule(s) applying to the same base form or word, and no speaker will ever apply it to a word without being able to apply all the other rules that have to apply to this word: this rule is not extractable (see also Singh 1996). Forms like or are excluded, just as forms like [lεkte] or [lajte], for lacté, are. And if we want to break alternations like k ∞ ƒ into k → → č, č→ ž, ž.→, j → i (since we are easily to find alternations requiring those), we would have to justify the separate existence of these rules by showing that any of them could fail to apply in the same words in either pathology, spooner­ isms, word games or the like. These alternations don't have any existence outside these words: they are syntagmatically and paradigmatically indivisible. Now, we could say that construction is an underived lexical entry, and leave the speaker to state outside his grammar the relationship between ükt and qiz, or admit that the relationship holds between lexical entries, not related by phonological rules with natural classes, but taking as such arbitrary chunks of sounds in words and relating them. This option will be considered later. MPRs can even be highly counter-natural. Ford & Singh (1991) give an example of such an alternation in French, relating adjectives ending in -a with adverbs in -ma where the adjective ending is denasalized, as showed by their rule: (2) [Xã]A ↔ [Xamã]Adv savant/savamment, intelligent/intelligemment, fréquent/fréquemment, constant/constamment,courant/couramment,méchant/méchamment

SOME ADDITIONAL CONTRIBUTIONS

303

This alternation is perfectly regular, and to formulate a phonological rule for it is certainly easy, but it would be difficult to maintain such a process as a 'natural' denasalization in a nasal environment. We should also consider the fact that there is another word-formation strategy for these adverbs, using the feminine form of the adjective, and formulated here à la Ford and Singh: (3) [X]Afem↔[Xmã]Adv (i) fortement, grossièrement, légèrement,lourdement,franchement (ïi) justement, amèrement, durement, tendrement, gravement The feminine adjective ends with a consonant (in oral speech) (3-i), and this strategy is used also for adjectives with afinalconsonant in the masculine form (3-ii). It happens that some words ending with ãC use this second strategy (grandement, présentement, lentement). It is clear that a phonological rule would have to tell apart words like présentement from words like fréquemment, but it would have to be an abstract analysis, since there is no obvious way to make a distinction between, for example, putative suffixed forms (sav/ant, intellig/entvs.fréquent, lent, méchant) that could be of any help. But it also happens that there is a nominal 'suffix' -ma that can be added to verbs and that never triggers denasalization: (4) manquement, rendement, dérangement, enfantement, enchantement which can only mean that the morphological information is indispensable, whereas we've seen that the phonological information is not a determinant for the choice of adverb formation. Last, we could take a look at the pattern exemplified in (5) below: (5) grammaire grand-mère grand-maman

[gram(m)εr] [grämer] [grãmamã], [gramma]

"grammar" "grand-mother" "gran'ma"

It is unlikely that we could motivate the choice of nasal/oral vowel with a dissimilation rule which would, in any conception of 'naturalness', account for what we observe in adjective/adverbs, verbs/nouns and the specific nouns above. The fact that two identical derivational suffixes behave so differently and that the putative denasalization rule does not show any tendency to generalize to nouns indicate how tight is the association between the adverbial suffix and the alternation. These observations point to the fact that some rules behave quite differently from what we mean by 'natural rules'. Phonetic distance, close association to

304

TRUBETZKOY'S ORPHAN

other rules and to word-formation and discrepancies between output and environment speak sometimes against the collapsing of automatic alternations and more or less regular alternations. If some morphophonological rules look natural, it is because they very often come from phonological rules: phonetic and phonological alternations are the main sources for morphologization, and as soon as an alternation is morphologized, it won't go astray in every arbitrary direction. Lack of physiological motivation does not mean lack of any motivation. Velar Softening in French or English or the k ∞ / alternation may look like extreme cases, but if they look like 'rules', it is because of their origin, and there is no reason for /k/ to alternate with, say, /m/ in a grammar if diachrony cannot itself find the way to do it. Alternations like k ∞ s and k∞/ may look like rules mainly because we know from history how to relate them and because there are correspondences of a similar nature between distinct languages of common origin, but we never expect phonetic or phonological alternations like those in a grammar. Morphologization implies the loss of phonetic teleology, or the switch from iconicity to indexicality, and this 'qualitative' switch is not predicted anywhere in particular along the scale from PRs to WFRs in Dressier's NPM, but is predicted by the uncompromising model of Ford and Singh. 4. Switch from Iconicity to Indexicality Dressier would maintain that defining criteria for rule types do not switch in a block, and that the switch for each criterion is independent of the others. A simple switch in phonetic distance, for example, makes an alternation less morphological than a switch in phonetic distance and environment, and the more the switches, the closer we get to morphology. The fact that an alter­ nation is not in the phonology does not mean, according to Dressier, that it belongs to morphology. It is hard to see how something that had lost a bit of naturalness can be represented by a rule, or whatever device, that could translate these in-between properties. Let's compare two examples from Dressier, Surface Palatalization and Postalveolar Formation in Polish. A formulation of the first one, here in the standard format for ease of exposition, would look like (6): (6) The relation between input and output respects biuniqueness, since there is no neutralization here, and also indexicality and diagrammaticity. I could illustrate

SOME ADDITIONAL CONTRIBUTIONS

305

this with a figure like (7), where the letters stand for features, and where [a, b, c] = [k], [a, b, z] = [k] and [x, y, z] = [i]: (7)

It would be possible to state the degree to which the relationship between the two levels, for a given sequence, is (a) diagrammatic (b) bi-unique (i.e., (b1) iconic or (b2) indexical). Since the combination of the two matrices in (7) is 'automatic', that is, one can not find the sequence [a, b, c] [x, y, z] on the surface, the relation is highly diagrammatic; and since the combination of features [a, b, z] is not found on the phonemic level, the relation is biunique, although not as much as the one on the right. Cases of neutralization would of course be lower on the scale of bi-uniqueness, and opaque cases lower on the scale of diagrammaticity. We can see that an evaluation procedure in terms of feature-counting, taking into consideration the potentialities of the specific system, would not only include what was aimed at when applied only to rules, but would not even need any rules to make the counting possible: every principle of 'classical phonemics' could be subjected to such an evaluation. The problem is that feature-counting has meaning only when we have a set of primitives, e.g., features, to which it can be applied. What then of morpho­ logical rules and rules having a very low score on diagrammaticity and/or biuniqueness? Obviously, indexicality would only be the reverse of iconicity, and there is no way to represent 'morphological indexicality', except if we have a finite, universal, and formally definable set of morphological features. For example, we could formulate a somewhat simplified version of Postalveolar Formation in a way that would look partly like (not relying on any form of redundancy):

If such a rule is a legal construct, I cannot help wondering how speakers

306

TRUBEIZKOY' S ORPHAN

would acquire it, although what it states is rather simple. If this is what Dressier (1996b) means concerning the difficulty of telling apart rules that have the appearance of (vestigial) phonological function, I hardly see any difficulty in this case. What is important in this rule, and what is straightforward, is the environment. Nothing else has anything to do with what goes on. The vestigial phonological function is no function at all: features only happen to be able to describe the input and the output, but like in the case of Velar Vocalization, the phonetic distance is of no interest for the indexicality. We have here a case that could be represented as follows, where capital letters stand for morphological features: (9)

The combination [a, x, z] is present in phonemic representations, which is the normal case for morphologically conditioned rules. If there can be something like an allophonic morphologically-conditioned rule, the output would be something like [a, b, w], where w is a non-distinctive feature in the context [a, b, — ]. We can also have cases where the environment requires phonological information, something of the type of (10): (10)

The fact that this situation is much more common than an allophonic MP-rule should be an indication that what happens to features in MP rules is only accidental, since if it had a phonological function, least phonetic distance should still be more common and favored, or these rules subject to sim­ plification in terms of features, which does not seem to be the case. The French prefix dé- is subject to a regular allomorphy rule spelling it débefore a word (mainly verbs) beginning with a consonant and dés- [dez-]

SOME ADDITIONAL CONTRIBUTIONS

307

before a word beginning with a vowel. There is only one exception, admitted recently, the noun déodorant (there is no corresponding verb form). There is no constraint at work here, for inside morphemes there is no intervening consonant: déesse, déifier, déambuler. So, for verbs, the situation found can be represented as (11):

(11)

Here the morphological bits of information are A ('in the prefix dés-') ànd B ('in a verb'), and d stands for 'consonant'. But 'in the prefix dés-' is not a member of a finite set of features like 'plural' or 'feminine', etc. A good example of a 'rule' that is at the edge of a phonological rule, without really being one, is found in Québec French. In final position, /a/ is realized by /α/ (itself realized [α], [ɔ] or [Λ]), with a very few exceptions, that are given here in (12c): (12)

a. chatte/chat "cat" masc./fem., garçon/gars "boy'V'guy", rate/rat "rat (f./m.)", plate/plat "flat (f./m.)", abattent/abat "strike down" 3pl./sg., éclate/éclat "blows out'V'a blow". b. passe/pas "pass by'V'step", rase/ras "shave'V'short cut", tasse/ tas "push aside'V'pile (noun)", basse/bas "down(f./m.)". c. la, fa (note) "A, F", tata "bye-bye", tarla "jerk", ga ( i and ue > e" Diachronica 3.185-201. Mencken, Henry L. 1936. The American Language: An inquiry into the development of English in the United States. London: Kegan Paul, Trench, Trübner & Co. Michelena, Luis. 1957/58. "À propos de l'accent basque". Bulletin de la Socié­ té de Linguistique de Paris 53:2.204-233. Miller, M. 1976. Zur Logik der fruehkindlichen Sprachentwicklung. Stuttgart: Klett. Mohanan, K. P. 1986. Lexical Phonology. Ph. D. dissertation, Cambridge, Mass.: MIT. . 1990. The Theory of Lexical Phonology. Dordrecht & Boston: D. Reidel. ,1995."TheOrganization of the Grammar". Goldsmith 1995.24-69. Mohanan, Tara. 1989. "Syllable Structure in Malayalam". Linguistic Inquiry 20.589-625. Monteil, Pierre. 1970. Elements de phonétique et de morphologie du latin. Paris: Nathan. Moravcsik, Edith A. 1978. "Language Contact". Greenberg 1978b I, 93-122. Morin, Yves-Charles. 1979. "La morphophonologie des pronoms clitiques en français populaire". Cahiers de linguistique 9.1-36. ,1994."The Place of Morphology in Grammar as Evidenced by Linguistic Change". Paper presented at the Roundtable on Morphophonology, October 1994, Université de Montréal, Montréal. & Jonathan Kaye. 1982. "The Syntactic Bases for French Liaison". Journal of Linguistics 18.291-330.

352

TRUBETZKOY''S ORPHAN

--, Marie-Claude Langlois & Marie-Ève Varin. 1990. "Tensing of Word-Final [ɔ] to [o] in French: The phonologization of a morphophonological rule". Romance Philology 43.507-528. & Etienne Tiffou. 1989. Dictionnaire complémentaire du bourouchaski du Yasin. Paris: Peeters/SELAF. Mufwene, Salikoko S. 1992. "Why Grammars Are not Monolithic". The Joy of Grammar: A festschrift in honor of James D. McCawley ed. by Diane Brentari et al, 225-250. Amsterdam & Philadelphia: John Benjamins. Myers, Scott. 1987. "Vowel shortening in English". Natural Language and Linguistic Theory 5.485-518. Nevis, Joel A. 1986. "Decliticization and Deaffixation in Saame: Abessive taga", Studies on Language Change ed. by Brian D. Joseph, 1-9. (= Ohio University Working Papers in Linguistics 34.) Newmeyer, Frederick J. 1992. "Iconicity and Generative Grammar". Lan­ guage 68.756-796. . 1994. "Competing Motivations and Synchronic Analysis". Sprach­ theorie und Forschung 47.67-77. Noske, Roland. 1982. "Syllabification and Syllable Changing Rules in French". Van der Hulst & Smith 1982, vol. II, 257-310. Paradis, Carole. 1993. "Ill-formedness in the Dictionary: A source of constraint violation". Canadian Journal of Linguistics/Revue canadienne de linguistique 38:2.215-234. Parret, Herman. 1983. Semiotics and Pragmatics. Amsterdam & Philadelphia: John Benjamins. Peirce, Charles S. 1965. Collected Papers. Cambridge: Harvard Univ. Press. Petito, Laura & Paula Marentette. 1991. "Babbling in the Manual Mode: Ev­ idence for the ontogeny of language". Science 251.1493-1496. Pichon, Edouard. 1935. "L'enrichissement lexical dans le français d'aujour­ d'hui". Le français moderne 3.209-222, 323-344. Piggott, Glyne. 1993. "Satisfying the Minimal Word". McGill Working Papers in Linguistics 8.194-233. Montreal: McGill Univ. Pinker, Steven. 1991. "Rules of Language". Science 253.530-535. & Alan Prince. 1988. "On Language and Connectionism: Analysis of a parallel distributed processing model of language acquisition". Cognition 28.73-193. Plunkett, Kim & Virginia Marchman. 1993. "From Rote Learning to System Building: Acquiring verb morphology in children and connectionist nets". Cognition 48.21-69. Postal, Paul. 1968. Aspects of Phonological Theory. N. Y.: Harper & Row.

REFERENCES

353

Prasada, Sandeep & Steven Pinker. 1993. "Generalisation of Regular and Irregular Patterns". Language and Cognitive Processes 8.1-56. Prince, Alan. 1990. "Quantitative Consequences of Rhythmic Organization". Papers from the Parasession on the Syllable in Phonetics and Phonology ed. by Anthony Bruck et al (= Papers from the Chicago Linguistic Society, 26:2, 355-398.) & Paul Smolensky. 1993. Optimality Theory: Constraint interaction in generative grammar. Piscataway, N.J. (= Technical Reports of the Rutgers Univ. Center for Cognitive Science, 2.) Pulgram, Ernst. 1975. Latin-Romance Phonology: Prosody and metrics. München: Wilhelm Fink. Radford, Andrew. 1988. "Small Children's Small Clauses". Transactions of the Philological Society 86.1-46. Rice, Mabel L. 1994. "Grammatical Categories of Children with Specific Language Impairments", Specific Language Impairments in Children ed. by Ruth V. Watkins & Mabel L. Rice, 69-89. Baltimore: Paul H. Brookes. Riemann, Hugo. 1903. System der musikalischen Rhythmik und Metrik. Leipzig: Breitkopf & Härtel. (Repr., Vaduz: Martin Sandig, 1985.) Rix, Helmut. 1976. Historische Grammatik des Griechischen: Laut- und Formenlehre. Darmstadt: Wissenschaftliche Buchgesellschaft. Roberge, Paul T. 1980. Morphologization of Phonological Alternations: A theoretical study based on the evidence from Germanic. Ph.D. dissertation, Univ. of Michigan, Ann Arbor. Roca, Iggy. 1992. "On the Sources of Word Prosody". Phonology 9. 267-287. Rubba, Johanna E. 1993. Discontinuous Morphology in Modern Aramaic. Unpubl. Ph. D. dissertation, San Diego: Univ. of California. Ruke-Draviņa, Velta. 1959. Diminutive im Lettischen. Lund: Lindstedt. Rumelhart, David E. & James L. McClelland. 1986. "On Learning the Past Tenses of English Verbs". Parallel Distributed Processing: Explorations in the microstructure of cognition. Vol. II: Psychological and Biological Models ed. by James L. McClelland, David E. Rumelhart & the PDP Research Group, 216-271. Cambridge, Mass.: MIT Press. Ruoff, A. 1981. Häufigkeitswörterbuch gesprochener Sprache. Tübingen: Max Niemeyer. Sadock, Jerrold M. 1980. "Noun Incorporation in Greenlandic: A case of syntactic word-formation". Language 57.300-319. ---. 1991. Autolexical Syntax: A theory of parallel grammatical rep­ resentations. Chicago: Univ. of Chicago Press.

354

TRUBETZKOY'S ORPHAN

--------. 1995a. Review of Anderson (1992). Natural Language and Linguistic Theory 13:2.327-341. .1995b."Hierarchies All the Way Down". Ms., Univ. of Chicago. Sapir, Edward. 1921. Language: An introduction to the study of speech. New York: Harcourt, Brace & Co. Saxton, Dean. 1963. "Papago Phonemes". UAL 29.29-35. Schane, Sanford. 1968. French Phonology and Morphology. Cambridge, Mass.: MIT Press. Scobbie, James 1993. "Constraint Violation and Conflict from the Perspective of Declarative Phonology". Canadian Journal of Linguistics/Revue cana­ dienne de linguistique 38:2.155-167. Shannon, Thomas F. 1987. "The rise and fall of final devoicing". Papers from the 7th Conference on Historical Linguistics ed. by Anna Giacalone Ramat et al., 545-559. Amsterdam & Philadelphia: John Benjamins. Siegel, Dorothy. 1974. Topics in English Morphology. Ph.D. dissertation, MIT. (Repr. New York: Garland, 1979.) Singh, Rajendra. 1985. "Prosodic Adaptation in Interphonology". Lingua 67.269-282. . 1986. "On Finding a Place for Trubetzkoy's Brain-Child". Canadian Journal of Linguistics/Revue canadienne de linguistique 31.343-363. [Re­ view of Dressler (1985a)] 1987. "Well-Formedness Conditions and Phonological Theory". Phonologica 1984 ed. by Wolfgang U. Dressler, Hans Luschützky, Oskar Pfeiffer & John Rennison, 273-285. Cambridge: Cambridge Univ. Press. . 1990a. "Do We Really Want a Syntax-Organ?: A note on phonology, morphology, and the autonomy thesis". Indian Linguistics 49.109-114. , 1990b. "Vers une théorie phonotactique générative". Revue québé­ coise de linguistique 19:1.131-163. [New enlarged version of Singh 1987.] --- 1991. "Interference and Contemporary Phonological Theory". Language Learning 41:2:157-175. . 1996. "Natural Phono(morpho)logy: A view from the outside". Hurch& Rhodes 1996.1-38. & Alan Ford. 1980. "Flexion, dérivation et Pānini". Progress in Linguistic Historiography ed. by Konrad Kærner (= Studies in the History of Linguistics, 20), 323-332. Amsterdam & Philadelphia: John Benjamins. & 1982. Remarks on Lexical Phonology. Unpubl. Ms., Univ. de Montréal. &.1987. "Interphonology and Phonological Theory". Sound Patterns in Second Language Acquisition (= Studies on Language

REFERENCES

355

Acquisition, 5.) ed. by Allan James Sr Jonathan Leather, 163-172. Dordrecht: Foris. & Gita Martohardjono. 1988. "Intermorphology and Morphological Theory". Linguistic Theory and Second Language Acquisition ed. by Wayne O'Neil and Susann Flyn, 362-383. Dordrecht: Kluwer. & David Parkinson. In press. "L 1 , L2, and Intermorphology". Rivis­ ta di Linguistica. Sohn, Ho-min. 1975. Woleaian Reference Grammar. Honolulu: Univ. of Hawaii Press. Spencer, Andrew. 1991. Morphological Theory. Oxford: Basil Blackwell. Stampe, David. 1969. "The Acquisition of Phonetic Representation". Papers from the Chicago Linguistic Society 5.443-454. . 1973. A Dissertation on Natural Phonology. Bloomington: Indiana Univ. Linguistics Club. (Repr., New York: Garland, 1979.) Stanners, Robert F., James J. Neiser & Scott Painton. 1979. "Memory Repre­ sentation for Prefixed Words". Journal of Verbal Learning and Verbal Behavior 18.733-743. Stemberger, Joseph P. 1994. "Rule-Less Morphology at the PhonologyLexicon Interface". The Reality of Linguistic Rules ed. by Susan D. Lima, Roberta L. Corrigan & Gregory K. Iverson, 147-169. Amsterdam & Philadelphia: John Benjamins. & MacWhinney, Brian. 1986. "Frequency and the Lexical Storage of Regularly Inflected Forms". Memory and Cognition 14.17-26. &1988."Are Inflected Forms Stored in the Lexicon?". Hammond & Noonan 1988.101-116. Tallal, Paula, Jeanne Townsend, Susan Curtiss & Beverly Wulfeck. 1991. "Phenotypic Profiles of Language-Impaired Children Based on Genetic/ Family History". Brain and Language 41.81-95. Thomason, Sarah G. & Terrence Kaufman. 1989. Language Contact, Creolization, and Genetic Linguistics. Berkeley, Calif.: Univ. of California Press. Tremblay-Villa, Diane. 1977. Fusion des voyelles aux frontières de mots en français québécois. M.A. Thesis, Univ. de Montréal. Trubetzkoy [Troubetzkoy], Nikolai S. 1924. "Zum urslavischen Intonationssystem". Streitberg Festgabe, 359-366. Leipzig. . 1929. "Sur la morphonologie". Travaux du Cercle Linguistique de Prague 1.85-88. . 1931. "Réflexions sur la morphophonologie". Travaux du Cercle Linguistique de Prague 4.160-163.

356

TRUBEIZKOY' S ORPHAN

. 1936. "Die phonologischen Grenzsignale". Proceedings of the 2nd International Congress of Phonetic Sciences ed. by Daniel Jones & Dennis B. Ivy, 45-49. Cambridge: Cambridge Univ. Press. . 1957[1949]. Principes de phonologic Traduction de Jean Cantineau. Paris: Klincksieck. Ullman, Michael. 1995. "Inflection Morphology in Adult-Onset Brain Dam­ aged Patients and in Developmentally Language Impaired Subjects". Paper presented at the 6th Annual Meeting of TENNET, Université du Québec à Montréal, May 1995. & Myrna Gopnik. 1994. "The Production of Inflectional Morphology in Hereditary Specific Language Impairment". Paper presented at the 19th Annual Boston University Conference on Language Development, Novem­ ber 1994. van der Hulst, Harry. 1992. "The Independence of Stress and Rhythm". Paper presented at the Internationale Phonologietagung, Krems, Austria. & Norval Smith, eds. 1982. The Structure of Phonological Repre­ sentations, 2 vols. Dordrecht: Foris. Varga-Khadem, Faraneh, Kate Watkins, Katie Alcock, Paul Fletcher & Richard Passingham. 1995. "Praxic and Non-Verbal Cognitive Deficits in a Large Family with a Genetically Transmitted Speech and Language Dis­ order". Proceedings of the National Academy of Science 92.930-933. Vennemann, Theo. 1974. "Words and Syllables in Natural Generative Phonol­ ogy". Chicago Linguistic Society Parasession on Natural Phonol-ogy, ed. by Anthony Bruck et al, 346-374. Chicago: Chicago Linguistic Society. . 1988. Preference Laws for Syllable Structure and the Explanation of Sound Change. Berlin: Mouton de Gruyter. Walker, Douglas C. 1984. The Pronunciation of Canadian French. Ottawa: Univ. of Ottawa Press. Wiese, Richard. 1990. "Über die Interaktion von Morphologie und Phono­ logie: Reduplikation im Deutschen". Zeitschrift für Phonetik, Sprach­ wissenschaft und Kommunikationsforschung 43,603-624. . 1994. "Phonological vs. Morphological Rules: German umlaut and ablaut". Recent Developments in Lexical Phonology ed. by Richard Wiese (= Theorie des Lexikons: Arbeiten des Sonderforschungsbereichs, 282.). Düsseldorf: Heinrich Heine Univ. Wiik, Kalevi. 1990. Suomen kielen morfologian historia II, Nominien taivutus 1820-1845. Turku: Turun Yliopisto. Wode, Henning. 1978. "The Beginnings of Non-classroom L2 Phonological Acquisition". International Review of Applied Linguistics 16.109-126.

REFERENCES

357

Wonderly, William. 1951. "Zoque II: Phonemes and morphophonemes". IJAL 17.105-123. Wood, Clement, ed. 1936. The Complete Rhyming Dictionary. Garden City, N.Y.: Doubleday & Co. Wurzel, Wolfgang U. 1977. "Adaptations Regeln und heterogene Sprach­ systeme". Phonologica 1976 ed. by Wolfgang U. Dressler & Oskar E. Pfeiffer, 175-182. Innsbruck. . 1980a. "Der Deutsche Wortakzent: Fakten-Regeln-Prinzipien. Ein Beitrag zur natürlichen Akzenttheorie". Zeitschrift für Germanistik 1. 299318. . 1980b. "Sprachsystem und Dialektik". Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 33:1.165-175, . 1984a. Flexionsmorphologie und Natürlichkeit: Ein Beitrag zur Morphologischen Theoriebildung. (= Studia Grammatica, 21.) Berlin: Akademie-Verlag. , 1984b. "Was bezeichnet der Umlaut im Deutschen?". Zeitschrift für Phonetik, Sprachwissenschaft und Kommunikationsforschung 37.647-663. ,1995."Alterserscheinungen in der Morphologie ". Natürlichkeitstheorie und Sprachwandel ed. by Ν. Boretzky, W. U. Dressler, J. Orešnik, K. Teržan & W. Wurzel, 67-91. Bochum: Norbert Brockmeyer. Zwicky, Arnold M. 1967. "Umlaut and Noun Plurals in German". Studia Grammatica 6.35-45. Berlin. . 1985. "How to Describe Inflection". Berkeley Linguistic Society Papers 11.372-86. , 1986. "The General Case: Basic form vs. default form". Berkeley Linguistic Society Papers 12.305-14. . 1988. "Morphological Rules, Operations, and Operation Types". Proceedings of the 4th Eastern States Conference on Linguistics (1987) ed. by Ann Miller & Joyce Powers, 318-334. Columbus, Ohio: Ohio State Univ. & Geoffrey K. Pullum. 1987. "Plain Morphology and Expressive Morphology". Berkeley Linguistic Society Papers 13.339-342.

Index Α. Albanian: 84, 96, 174 Algonquian: 13 Amundarain Arana: 196 Andersen, Henning: 95 Anderson, Stephen: 14, 26, 46, 57-58, 99, 101, 143-144, 155-156, 175 Arabic: 224, 257, 260 Aramaic: 250 Archangeli, Diana: 223, 297 Aristotle: 120 Aronoff, Mark: 16, 18, 262, 316 Austronesian: 223 Azerbaijani: 199 B. Baayen, Harald: 251 Baker, Mark: 129, 157 Bangla: 320-322, 325 Basque: 175-176, 196-198, 202-204, 212, 217-218, 239 Baudouin de Courtenay, Ian: 13 Bauer, Laurie: 196 Bazell, Charles: 323 Berko, Jean: 271 Berman, Ruth: 258 Bhat, D.N.S.: 70 Bittner, Andreas: 289 Black, H. Andrew: 223 Blight, Richard: 199 Bloomfield, Leonard: 4, 13-15 Boas, Franz: 199 Bohm, David: 120 Booij, Geert: 82 Boretzky, Norbert: 194 Boroda, M. G.: 217 Borowsky, Toni: 263 Bougaïeff, André: 313 van Bree, Cor 83, 114 Bromberger, Sylvia: 69

Bulgarian: 198 Bybee, Joan: 7, 113, 119, 137, 155, 171, 174-175, 192, 238, 247-248, 250ff. 259, 261, 265, 270ff., 281 C. Cardinal, Pierre: 313 Carstairs(-McCarthy), Andrew: 22-23, 30, 74, 83, 98, 206-208. Caxton, William: 159 Cech, Petra: 82 Celtic: 73 Chaucer, Geoffrey: 159 Chilcotin: 209 Chinese: 241 Mandarin ~: 96 Chipewyan: 199 Choctaw: 223, 225 Chomsky, Noam: 4, 195, 323 Chomsky, Noam & Halle, Morris: 1 Clashen, Harald: 253, 261, 268, 286 Clark, Eve: 182-183, 258 Closs-Traugott, Elisabeth: 79 van Coetsem, Frans: 78 Compes, Isabel: 79 Comrie, Bernard: 75 Condax, Ioanna: 197 Cook, Eung-Do: 209 Corbin, Danielle: 18 Crago, Martha: 271, 276, 281 Crowhurst, Megan: 223 Czech: 215, 217, 220, 238, 240 D. Dahl, Hartvig: 272 Dakota: 199 Darden, Bill 83 Dell, François: 62 Deloria, Ella: 199 Demuth, Katherine: 226

360

TRUBETZKOY'S ORPHAN

Desrochers, Richard: 298-299, 311-312 Dickens, Charles: 131 Dogil, Grzegorz: 199 Donegan, Patricia: 210, 215, 230, 235 Dressler, Wolfgang: 13, 16, 25, 67ff., 157, 189, 194, 197, 200ff., 216, 229, 252-253, 258, 297, 300, 302, 304, 306, 310-311, 314, 316ff. DuBois, John: 247 Dutch: 83, 95, 114, 158, 268, 293 Dutt, Madhusudan: 320 Dziubalska-Kolaczyk, Katarzyna: 67, 82, 105 E. Encrevé, Pierre: 311 Engels, Friedrich: 81 English: 25, 30, 34, 38. 49, 58ff., 62, 71-72, 95, 103, 130. 132-133, 135, 138, 147, 149, 151-152, 158-159, 172, 177, 180-181, 184ff., 195, 199, 205-206, 226, 238, 251, 254ff., 259, 261-262, 264ff., 268, 271-272, 280, 287ff., 293, 298, 301, 304, 327ff. Middle ~: 159, 294 Estonian: 234 F. Fabri, Ray: 49 Fant, Günnar: 92 Fauré, Gabriel: 165 Finnish: 19, 24, 30, 234 Ford, Alan: 86, 93, 114, 121, 136, 138, 140-141, 144-145, 148, 151-152, 155ff., 160-161, 163ff., 179, 238, 297ff., 309-310, 314-315, 322ff. French: 45, 51, 56, 62, 72, 78, 95, 103, 134, 179, 183, 214, 224, 230, 240241, 254, 262, 285, 298, 301-302, 304, 306, 310-311, 313ff., 319 Acadian ~: 301 Norman ~: 159 Québec~: 307-311, 313 G. Garawa: 215 Garde, Paul: 192 German: 4, 19, 25-27, 30, 56-57, 71-72, 75, 78, 80, 83, 103-104, 114, 158, 160ff., 175, 185, 193, 195, 204, 206,

218, 259ff., 268-269, 286-287, 289290, 293, 298 Alemannic Upper ~: 162 High ~: 95, 160ff. Swiss ~: 163 Austro-Bavarian ~:71,108 Twente-: 83, 113-114 Germanic: 164 Gipsy: 174 Goad, Heather: 264, 271, 276ff., 280, 281ff. Goldsmith, John: 49, 212, 215, 229 Gopnik, Myrna: 271, 276, 281, 282 Gordon, Peter: 265ff. Gósy, Maria: 99 Greek 8, 23, 192, 194, 198, 79, 104 Ancient ~: 79 crétois 7 Greenberg, Joseph: 278 Greenlandic see Inuttitut Guilfoyle, Eithne: 278 H. Hall, Christopher: 67, 112 Halle, Morris: 1, 4, 49, 69, 92, 95, 191, 195, 199, 211, 221-222 Hargus, Sharon: 44, 262 Harris, James: 25 Hausa: 257, 260 Hayes, Bruce: 53, 192, 199, 211- 222 Hebrew: 258 Heine, Bernd: 253 Hiley, B.: 120 Hindi: 320 Hoard, James: 271 Hoeksema, Jack 155 Hohepa, Patrick: 199, 202 Hookway, Christopher 72 Hooper see Bybee Hopper, Paul: 79, 247, 256, 266 Householder, Fred W.: 199, 316 Hudson, Grover: 15, 58 Hungarian: 23, 74, 81, 104, 113, 206, 215, 238 Hurch, Bernhard: 197, 203, 211, 214, 216, 222, 226ff. Hutchison, James: 256-257 I. Icelandic: 27

361

INDEX

Igla, Brigit: 194, 198 Indo-European: 164 Indonesian: 147 Inkelas, Sharon: 223 Inuttitut: 122, 125ff., 156, 172 Italian: 19, 25, 78, 80, 183, 194-195, 197, 200, 202, 205, 212, 218-221, 227, 232ff., 243 J. Jakobson, Roman: 13, 49, 92, 190-191, 220 Janda, Richard: 26, 80, 93, 95-96, 102ff., 115, 155, 157, 163, 174, 259 Jespersen, Otto: 272 Joseph, Brian: 79, 95, 155, 185 K. Kahn, Daniel: 30 Kaisse, Ellen: 44, 262 Kanuri: 256, 257, 260 Karmiloff-Smith, Annette: 102 Karpf, Anne-Marie: 68, 81, 102 Kaye, Jonathan: 311 Kehayia, Eva: 275 Kenstowicz, Michael: 199, 209, 230 Kilani-Schoch, Marianne: 316 Kim, John: 272, 274 Kiparsky, Paul: 15, 25, 32-33, 35ff., 43, 45ff., 69, 99, 140, 157-158, 172, 177, 184, 253, 262, 264-265, 273, 280, 285, 288, 290, 297, 300-301, 317, 320, 322, 324-325 Klaus, Georg: 70 Klausenburger, Jürgen: 205 Knudson, Lyle: 213 Koasati: 224, 230, 242 Köpcke, Klaus-Michael: 259, 261, 290 Koutsoudas, Andreas: 79 Kowasati: 233 Kruszewski, Paul: 13 Kuryîowicz, Jerzey: 1, 99 Kutscher, Silvia: 79 Kwakiutl: 199 L. Langacker, Ronald: 247 Lardil: 225 Latin: 7, 71, 78, 135, 176, 182, 190, 201, 204, 207, 214-215, 238. 289,

293 Late ~: 205 Latvian: 27 Lehmann, Christian: 194 Li, Fang-Kuei: 199 Lieber, Rochelle: 17, 18, 129, 157 Lieberman, Philip 284 Lindner, Gerhart: 70 Linell, Per: 314 Lobben, Merritt: 260 Lombardi, Linda: 223 Loporcaro, Michele: 78 Losiewicz, Beth: 262 Lucci, Vincent: 301 Lupas, Liana: 79 Liissy, Heinrich: 163 M. MacWhinney, Brian: 247, 251 Maiden, Martin: 71, 80-81 Malagasy: 242 MalakMalak: 214 Malay: 147 Malayalam: 34, 151, 153 Malkiel, Yakov: 113, 136 Manaster-Ramer, Alexis: 78 Mao Tse-Tung: 96 Maori: 199, 202, 207, 223 Marantz, Alec: 52 Marcus, Gary 248, 253, 259, 261, 266, 275 Marentette, Paula: 226 Marslen-Wilson, William: 70 Martin, Jack: 224, 230 Martinet, André: 1, 4, 319 Martohardjono, Gita: 138 Matthews, Peter: 14, 63 Mayerthaler, Willi: 72-73, 201, 202- 205 McCarthy, John: 18, 51, 98, 223-224, 227, 260, 297 McCawley, James: 156 McClelland, James: 273, 287 McCormick, Robert: 78 Mencken, Henry: 272 Mendez-Dosuna, Julian: 78, 136 Merlini-Baraberesi, Lavinia: 67 Mester, Armin: 226 Michelena, Luis: 198 Miller, M.: 268 Moder, Carol: 253, 271ff.

362

TRUBEIZKOY'S ORPHAN

Mohanan, Κ. P.: 1, 48, 49-50, 53, 130, 131, 132, 133, 156, 157, 297 Mohanan, Tara: 153 Mohawk: 129, 225 Monteil, Pierre: 7 Moravcsik, Edith: 138-139, 159 Mordvin (Moksha): 199 Morin, Yves-Charles: 78, 95, 103, 112, 136-137, 157, 179, 182-183, 185, 309, 311, 313, 315 Mufwene, Salikoko: 156 Mukogean: 224, 242 Myers, Scott: 30, 180 N. Nathan, Geoffrey: 203 Navaho: 192 Nevis, Joel: 95 Maori: 206, 207 Newman, Jean: 259 Newmeyer, Frederick: 77 Nicaraguan: 224 Noonan, Michael: 278 O. Oehrle, Richard: 186 Otomi: 199 P. Paiute (Southern): 215 Panini: 13, 141 Pangasinan: 223, 226, 230 Papago: 213-214 Paradis, Carole: 297 Parkinson, David: 138 Parret, Herman: 72 Peirce, Charles: 72, 110 Pensado, Carmen: 78, 136 Petito, Laura: 226 Philippaki-Warburton, Irene: 79 Pichón, Edouard: 312 Piggott, Glyne: 225-226, 229ff. Pike, Kenneth: 199 Pinker, Steve: 248, 253, 261, 270, 273, 275, 287 Polish: 68ff., 75, 76, 80-81, 109, 215, 304 Portuguese: 80 Postal, Paul: 301 Prasada, Sandeep: 261

Prince, Alan: 18, 51, 85, 100, 223, 224, 227, 260, 270, 273, 284, 287, 297 Pulgram, Ernst: 199 R. Radford, Andrew: 278 Rebellati, Catherine: 271, 276ff. Reh, Hugo: 253 Riemann, Hugo: 215, 217 Rix, Helmut: 198 Roberge, Paul: 80 Roca, Iggy: 190-191, 212, 222 Romance: 182-183, 198, 202ff., 294 Romani: 82, 194, 198 Ronneberger-Sibold, : 114 Rothweiler, Monika: 253, 261 Roturnan: 53 Rubba, J.: 250 Rudorf, Carmen: 79 Ruke-Draviŋra, Velta: 27 Rumelhart, David: 273, 287 Ruoff, J.: 261 Russian: 4, 75, 81, 103, 109, 190, 206 S. Saame: 95 Sadock, Jerry* 85, 122-123, 125, 127128, 156-157 Sandoval, Maria: 155 Sanskrit: 13, 19, 25, 30, 185, 191 Sapir, Edward: 13, 192 Saussure, Ferdinand de: 330 Saxton, Dean: 214 Scobbie, James: 297 Selkirk, Elisabeth: 62 Semai: 72 Semitic: 250, 260 Sesotho: 226 Shannon, Thomas: 78 Siegel, Dorothy: 192, 195, 199 Singh, Rajendra: 3, 9, 16, 32, 81ff., 86, 93, 98, 114-115, 121, 136, 138, 140141, 144-145, 148, 151-152, 155ff., 160-161, 163ff., 234, 238, 297ff., 309-310, 314-315, 317, 322ff. Skalička, Vladimir: 81 Silvestre, Armand: 165 Slavic: 30, 108 Sloat, Clarence: 271 Slobin, Dan: 253, 271-272

INDEX

Smolensky, Paul: 85, 100, 227, 297 Sohn, Ho-min: 225 Sommerstein, Alan: 63 Spanish: 21, 25, 57, 113, 119, 135-136, 138, 183, 191, 197, 201ff., 212, 222, 226, 237-238, 254, 292 Andalusian ~: 78 Old ~: 254 Spencer, Andrew: 16 Sridhar, Shikarpuri 262 Stampe, David: 1, 28, 69, 200, 209, 215, 230, 298 Stanners, Robert: 275 Stemberger, Paul: 63, 247 Stone, Gerald: 75 Swedish: 158 T. Tallal, Paula: 276 Terrell, Tracy: 192 Thompson, Sandra: 247, 266 Thornton, David: 80, 194, 197 Tiwa: 122 Tongan: 197, 203 Tremblay-Villa, Diane: 313 Trubetzkoy, Nikolai: 3-4, 7, 9, 329 Tubatulabal: 233 Turkish: 23, 81, 113 U. Ullman, Michael: 275 Ulwa: 224 van der Hulst, Harry: 214 van Marie, Jaap: 22 Vargah-Khadem, Faraneh: 281-282 Vennemann, Theo: 1 Vergnaud, Jean-Roger: 191, 199, 211, 221-222 W. Walker, Douglas: 48, 51ff.53, 58, 60, 103, 105, 313 Walloon: 241 Warlpiri: 19, 30 Warnant, : 241 Watkins, Clavert: :281 Weri: 215 Wiese, Richard: 72 Woleaian: 225 Wonderly, William: 213

363 Wunderlich, Dietrich:: 49 Wurzel, Wolfgang: 57, 80, 193, 201, 203, 229 γ , Ζ. Yiddish: 78, 108 Zoque: 213 Zwicky, Arnold: 14, 17-18, 86ff., 155