Analyticity and Syntheticity: A Diachronic Perspective with Special Reference to Romance Languages [Reprint 2012 ed.] 9783110872927, 9783110112450

Citation preview

Analyticity and Syntheticity

Empirical Approaches to Language Typology 6 Editors Georg Bossong Bernard Comrie

Mouton de Gruyter Berlin · New York

Analyticity and Syntheticity A Diachronic Perspective with Special Reference to Romance Languages

by

Armin Schwegler

Mouton de Gruyter Berlin · New York

1990

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter & Co., Berlin.

© Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability.

Library of Congress Cataloging in Publication Data

Schwegler, Armin, 1955Analyticity and syntheticity : a diachronic perspective with special reference to Romance languages / by Armin Schwegler. p. cm. — (Empirical approaches to language typology ; 6) Revision of the author's thesis (Ph. D. — University of California, Berkeley, 1986). Includes bibliographical references and indexes. ISBN 0-89925-332-6 (cloth : acid-free paper) 1. Romance languages — Syntax. I. Title. II. Series. PC201.S38 1990 440—dc20 90-49317 CIP

Deutsche Bibliothek Cataloging in Publication Data Schwegler, Armin: Analyticity and syntheticity : a diachronic perspective with special reference to Romance languages / by Armin Schwegler. — Berlin ; New York : Mouton de Gruyter, 1990 (Empirical approaches to language typology ; 6) ISBN 3-11-011245-0 NE: G T

© Copyright 1990 by Walter de Gruyter & Co., D-1000 Berlin 30. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Printing: Ratzlow Druck, Berlin. — Binding: Dieter Mikolai, Berlin. Printed in Germany.

The aim of science is to seek the simplest explanation of complex facts. We are apt to fall into the error of thinking that the facts are simple because simplicity is the goal of our quest. Alfred North Whitehead: Concept of Nature

Contents Preface

ix

Introduction

xi

Chapter one History of the terms "analytic" and "synthetic"

3

Chapter two Word delimitation: in search of a universal

29

Chapter three Fishing out the baby: The usefulness of the terms analyticity/syntheticity in diachronic description

47

Chapter four The verbal core from Latin to French: Part 1: The marking of person/number. From synthetic to analytic?

75

Chapter five The verbal core from Latin to French: Part 2: Periphrastic cantare habeo, habeo cantatum, and je vais chanter

117

Chapter six Synthesis in non-contiguous elements: predicate negation in Romance

151

Chapter seven On drift, cyclicity, reconstruction, and the motives for changes in the analytic/synthetic spectrum

175

Conclusion

193

Notes

197

Bibliography

245

Subject Index

279

Author Index

287

Preface Since the early 19th century, linguists have been using the terms "analytic" and "synthetic" to characterize the morphological architecture of languages. The concepts of analyticity and syntheticity are frequently used today, but linguists differ widely as to what they regard as a synthetic or an analytic language. This book explores how it is possible that languages such as French are located by different investigators at opposite extremes on the analytic/synthetic scale. Is this disagreement simply due to the lack of clearly defined typological parameters? Or have languages like French recently moved from one end of the scale to the other, so that the different claims merely refer to different diachronic stages? A principal aim of this investigation is to demonstrate the validity of the concepts of analyticity and syntheticity, notwithstanding the vagueness and ambiguity with which the terms have been used. A clearer understanding and more precise delineation of the terms analytic and synthetic—hence of the concepts themselves—is essential not only to avoid the kinds of confusion and oversimplification which mar many typological studies, but also to grasp better certain evolutionary trends in the histories of Romance as well as of other languages. A number of well-documented developments from within the histories of Latin and Romance (with occasional glances at Germanic and other Indo-European languages) will be reviewed here to reveal how once essentially unrelated speech segments undergo synthesis to become semantic and/or grammatical units of an entirely different sort It is hoped that these findings, if corroborated by research from other language families, may serve as a basis for formulating a set of crosslinguistic diachronic principles with respect to the ways in which languages undergo certain morphological changes. Ever since Sapir's well-known excursion into the question of "drift," there has been considerable interest in and debate about the cyclical movement of languages between analysis and synthesis. This investigation purports to shed new light on this phenomenon, and to show that, contrary to a widespread belief, there is nothing hermetic or mysterious about rythmic cycles of analytic and synthetic constructs. This book is a thoroughly revised and considerably enlarged version of my 1986 dissertation, completed under the auspices of the Group in Romance Philology at the University of California, Berkeley. I would like to express my deepest appreciation and gratitude to those who have been a source of inspiration and encouragement to me during the past seven years: Yakov Malkiel, Jerry Craddock, Joseph Duggan, Charles Faulhaber, Ruggero Stefanini, and Karl Zimmer. A particular debt of gratitude also goes to Georg Bossong whose insightful comments on an earlier

χ Preface version of the manuscript have significantly improved both the form and the content of this book. Very special thanks are due to the long list of friends and colleagues whose support, encouragement or cheerful disposition have helped me enjoy the many years that this book was under preparation. Among these let me single out Ella Mae and William Burke, Thomas Büchler, Barbara DeMarco, Virginia Gonzales, Kathryn Klingebiel, John Levy, Albert Muth, Ruedi Müller, Celina Navarro, Salvador Rodezno, Jeff Shimanoff, Victor Simarra, Christopher Stookey, Morteza Tabatabaipour, Norma Thompson, Jeff Turley, and Thomas Walsh. No one has contributed more to the ideas developed in this work, been more supportive, encouraging, and inspiring than Suzanne Fleischman. Her comments and criticisms—with regard to both content and style—prompted me to recognize certain inconsistencies in earlier versions, and helped sharpen my thinking so that I could ultimately find answers even to the knottiest issues of "analyticity and syntheticity." I dedicate this book to her. Last, but by no means least, a special note of thanks is due to my parents and my brother Roland who have remained close to me despite living in my far-off native Switzerland.

Introduction Background of the question The origins of non-genetic criteria for language classification can be traced to the moment when August Wilhelm von Schlegel, whose brother Friedrich had already proposed a twofold classification of languages into those with affixes and those with inflections, suggested in 1818 that inflectional languages be further subdivided into "synthetic" and "analytic" types. Though the new approach generally met with little enthusiasm, at least during the 19th century, and was at the time vehemently rejected by some of the best-known linguists (e.g., Humboldt), August Wilhelm Schlegel's proposal nonetheless turned out to set the tone for many subsequent approaches to language classification. The acceptance of the terms analytic and synthetic is due in large part to the publication of Edward Sapir's much celebrated Language (1921), in which A. Schlegel's terminology is discussed in considerable detail in the chapter 'Types of Linguistic Structure." Recognizing the potential usefulness of the concept, Sapir proposed numerous refinements to A. Schlegel's simple binary classification by incorporating analyticity and syntheticity into a more sophisticated matrix of morphological features. The impact of Sapir's book on subsequent linguistic typology was so profound and long-lasting that the terms analytic and synthetic (as well as Sapir's newly coined "polysynthetic") have since become part of standard linguistic terminology. The ready acceptance of the terms, and hence the concepts themselves, is somewhat surprising in light of the fact that Sapir himself recognized that his scheme, though a vast improvement over Schlegel's, was still open to criticism. Like those before him, Sapir could do no more than provide a scalar concept of syntheticity which relied heavily upon vague and arbitrary criteria. Qualified labels such as "mildly synthetic," "notably synthetic," etc. could not satisfy those who demanded rigorous, unambiguous, and non-subjective parameters; thus the linguistic community, while accepting the terms, remained undecided about how the concepts should be used. While most linguists of the period between 1920 and 1950 recognized the general problem of dealing with a useful taxonomic paradigm which, however, lacked the exactitude and rigor expected of modern linguistic analysis, few were willing to tackle the problem head on in order to eliminate the terminological—and conceptual—"softness" inherent in the analytic/synthetic distinction. It was not until 1954 that a radically different approach was undertaken by Joseph Greenberg. Using the analytical tools of contemporary American structuralism, he revised the Sapirian classification, and introduced the MORPHEME as one of the distinctive units of

xii Introduction morphological typology, deriving the degree of syntheticity by a simple mathematical formula: the total number of MORPHEMES divided by the total number of WORDS (thus lov-ed [2 morphemes/1 word] had a "synthetic index" of 2.00). By plotting the frequency distribution of morphemes per word in several languages, linguists could now establish what many hailed as truly objective indices of syntheticity. Not surprisingly, there was a quick and positive response to the new method; for while Sapir's classification relied on a fairly loose determination of the frequency and degree of formal morphemic cohesion, Greenberg's classification had the advantage not only of showing differences in syntheticity—now quantified—between languages, but also of providing clear guidelines for obtaining undistorted results in calculations involving large numbers of languages. Though the structuralist vogue of the thirties and forties might well have led Greenberg to propose a new and promising way of calculating syntheticity, the structuralist enterprise became, ironically, the major obstacle to the application and acceptance of his technique. As the structures of more and more languages were studied, it became apparent that there is no single, entirely satisfactory working definition of the "word," a state of affairs which, in turn, calls into question the usefulness of Greenberg's formula MORPHEME/WORD. Perhaps in part because Greenberg's approach has a strongly scientific flavor, numerous typologists (e.g., V. Krupa, G. Altmann, E. SlaviCkova, W. Lehfeldt, H. Haarmann, etc.) have continued to use his formula, and have in the past two decades reformulated and extended the classic quantitative approach. Alternative ways for calculating the index of synthesis are being suggested, and some important methodological differences have been proposed. Several typologists have argued convincingly that in order to be meaningful the parameters analytic and synthetic must be applied only to individual cross-linguistic categories and not to languages as wholes. Yet notwithstanding the originality and soundness of these proposals, the central problem of finding a satisfactory working definition of the "word" has been glossed over; and the lack of a universally applicable definition of the word leads directly to the collapse of what has long been hailed as a promising avenue of research: morphological typology. Some have recognized by now the methodological fallacy of plugging in formulas (e.g., MORPHEME/WORD) in which one of the parameters lacks universal extension. If the concepts analytic and synthetic are to have a place in future linguistic discussion, their definitions must not be based on a methodology whose shortcomings cannot be corrected in the foreseeable future. Unless we can find a formula which does not take the word as one of its basic parameters, the idea of operating with a quantitatively based typology of analytic and synthetic languages ought to be laid to rest.

Introduction xiii The aim of this study My curiosity about the terms analytic and synthetic was sparked for a number of reasons. First, I noticed that in much of the literature, even recent literature, oversimplifications and confusion occur frequently in the description of morphological systems, as exemplified by the following contradictory quotes: ... an analytic language like Vulgar Latin, English, and French ... (Lathiop 1980:22) Ce qui est certain, c'est que, même s'il existe un type analytique de langue, ce type n'est en tout cas pas celui du français, qui est une langue synthétique. (Tesnière 1932:64) Secondly, I observed that in discussing certain aspects of Romance morphosyntax, I myself frequently used expressions such as "analytic (go-) future," "analytic person/number marking" etc. without truly understanding the rationale for these labels. Once aware of the conceptual vagueness of the terms, I discovered that many of the so-called analytic constructs (e.g., the "analytic" person/number marking of Fr je parle Ί speak' or the "analytic" Sp vai a hablar 'you are going to speak') have a considerably tighter morphological cohesion (i.e., are more synthetic) than the label analytic suggests. I realized from the onset of my investigation that a clearer understanding and more precise delineation of the terms analytic/synthetic—and hence of the concepts themselves—is essential, not only to avoid the kinds of confusion and/or oversimplification which have marred many typological studies, but also to understand better certain evolutionary trends in the histories of French, Spanish, Portuguese, Italian, and Romansh, as well as other languages. Convinced of the potential usefulness of the analytic/synthetic distinction, I have set the following goals for myself in this study: (1) (2) (3) (4)

(5)

to trace the history of these concepts and offer reasons for their remaining relatively vague and ambiguous; to demonstrate why the concepts as currently used cannot fruitfully be applied in cross-linguistic analysis; to suggest alternative ways in which the analytic/synthetic distinction may profitably be used; to rectify a number of oversimplifications in the analysis of typological change, and offer a more nuanced picture of certain morphosyntactic developments in Romance; to answer the question of why languages—Romance languages as well as IE and non-IE languages—appear to drift back and forth between analysis and synthesis in a cyclical manner.

xiv Introduction Methodology and results This study started as an attempt to answer the above questions, with no preconceived notions as to how the concepts analytic and synthetic would ultimately have to be revised in order to be both meaningful and useful. Hoping to obtain a more balanced picture of how the terms have been applied, both by linguists directly conconed with their definition as well as those who have used them simply as part of standard linguistic terminology, I began by systematically combing a large number of linguistic dictionaries as well as treatises and articles on Romance morphosyntax. Logically, the next step was to determine why there is so little agreement about the morphological typologies of French, Spanish, Italian, etc. How is it possible that the same language is located by different investigators at opposite extremes on the analytic/synthetic scale? Is the disagreement simply due to the lack of clearly defined typological parameters? Or have these languages recently moved from one end of the scale to the other, so that the different claims simply refer to different diachronic stages? Byre-examiningdevelopments in Romance predicate negation and in the verbal core from Latin to French—most notably the switch from post- to pre-verbal person/number marking and the rise of forms like je chanterai Ί will sing,' j'ai chanté Ί sang/have sung,' je vais chanter Ί am going to sing,' I have attempted to answer not just these but a number of other pertinent questions as well. First it will be shown that speech units synthesize according to a series of patterned semantic, morphological and/or phonological changes. In the course of the investigation it will become clear that all changes in the analytic/synthetic spectrum have an underlying semantic basis; that is, only mutually "relevant" items (defined in §3.3.1.1) acquire greater morphosyntactic and/or phonological cohesion. Failure to recognize this has led several investigators to the erroneous belief that the main cause for changes in the analytic/synthetic spectrum is the phonological rapprochement between syntactically conjoined speech units. A major difference between the present approach to the question of syntheticity and analyticity and earlier ones lies in the types of data used. In classifying units such as je parle Ί speak,' voy a hablar, Ί am going to speak,' etc. scholars have generally failed to take into account the changes, crucial for the question of analyticity and syntheticity, which Romance languages have undergone in familiar speech. One of the main concerns of this investigation therefore, will be to show how these often very recent developments have come to change the morphological make-up of Romance vernaculars. Further important differences between this approach and earlier ones will emerge in the course of the discussion. It will become clear that in order to maintain the meaningfulness of the parameters synthetic and analytic:

Introduction xv (1) (2) (3)

their use must be limited to speech units rather than entire languages, the "word" should not be regarded as essential to the concepts, and they must be understood not as quantifiable absolutes but as the rough measure of the overall morphemic interdependent of speech units.

It will be suggested that this measure is best arrived at by taking into account semantic and morphosyntactic as well as phonological criteria. Analyticity will thus be defined as the semantic, syntactic, morphological and phonological autonomy of morphemes within a speech unit, while syntheticity will be characterized as the measure of semantic, syntactic, morphological and phonological interdependency (or relatedness) of morphemes within a speech unit An important point which will emerge from my investigation is that no definition of analytic/synthetic—and, consequently, no method of measuring morphemic interdependency—can be entirely objective and/or language independent. But as I hope to show, the impossibility of arriving at a purely objective definition of these concepts is only a minor drawback, since my approach merely involves an attempt to plot the general direction (i.e., ANALYTIC —> SYNTHETIC, or SYNTHETIC —> ANALYTIC) of a speech unit over its recorded history, rather than to calculate the exact location of such a unit along the analytic/synthetic axis at any given point. One key finding of this study will be that many long-term diachronic changes cannot be grasped appropriately without the notions of analyticity or syntheticity. It will be shown, for instance, that to understand how Lat ego Τ differed from its late OFr descendent je it does not suffice to list the individual phonological and morphosyntactic changes that shaped ego into a different unit. OFr je has a morphosyntactic and semantic profile which in many ways parallels that of its ancestor ego, but it differs fundamentally from ego by having entered into a tighter relation with the verb on several levels. Similarly, early OFr pas (< CL passum 'a step'), when used in conjunction with the negator non, was on the way to becoming a unit of an entirely different sort roughly around the turn of the first millennium. The important differences between ego and je on the one hand, and CL passum and OFr pas on the other, do not, however, become apparent unless we interrelate the individual pertinent diachronic changes which together create this new morphological bond. My aim, then, will be to show, through this integrated perspective which is at the heart of the analytic/synthetic concepts, how essentially unrelated developments conspire over time to change certain elements into units of a different sort. One of the most fascinating and, by the same token, most intriguing aspects of the cyclic move from analysis to synthesis (or vice-versa) is the question of the origin of these changes. Though considerable effort has been devoted to studying this issue in the past few years, it has remained unresolved. The aim of the last chapter of this study is not only to point up some of the weaknesses of earlier in-

xvi Introduction vestigations, particularly those which view analysis and/or synthesis as the result of a somewhat mystical, teleologica! drift, but also to ferret out the motivating factors in the evolution toward new types of morphological structures. Arguing that analysis and synthesis are simply a consequence (rather than a cause) of language change, I will suggest that every innovation ultimately affects the direction a speech unit takes on the analytic/synthetic axis, and that the frequently mentioned drift of IE languages towards greater analyticity merely reflects a statistical trend whose unidirectional momentum has been exaggerated as a result of a widespread failure to recognize signs of synthesis.

Abbreviations A. Terminology acc. ART adj. adv. AUX c. COREF daL EMPH fem. 1. loc. mase. mod. NEG

accusative article adjective advab auxiliary century coreferent/coreferential dative emphasizer feminine line locative masculine modem negator/negative

nom. NP obi. PART pi. P/N pop. PP pres. refi. s. Τ/Α v. VP

nominative noun phrase oblique partitive {dural person/number popular past participle present reflexive singular tense/aspect verse verb phrase

Β. Languages and Dialects Prefixed to an abbreviation, O = old, mid = middle, mod = modern, W = West, Ν = Northern, S = Southern, East = Eastern, Centr. = Central Am. Sp Amp BP Calab Cat CL Eng FSF Fr Gal Germ Gév IE ISF It

American Spanish Ampezzano Brazilian Portuguese Calabrian Catalan Classical Latin English Formal Spoken French French Galician German Gévaudanais Indo-European Informal Spoken French Italian

Lad Lat Lomb Mil Pal Piedm Port Oc PR Rom Russ Sard Sp Surs VL

Ladin Latin Lombard Milanese Palenquero Piedmontese Portuguese Occitan Proto-Romance Romansh Russian Sardinian Spanish Surselvan Vulgar Latin

Chapter one History of the terms "analytic" and "synthetic"

1.1 Introduction This chapter is divided into four sections. The first will show how the concepts "analyticity" and "syntheticity"1 arose in the context of 19th century analyses of morphological typology.2 The second section focuses on Edward Sapir's writings on analyticity and syntheticity, which not only led to the ossification of the already existing terminology, but also produced a thoroughgoing transformation in the application of morphological criteria to language typology. Section three follows the evolution of Sapir's ideas down to our own times. Finally, the fourth section summarizes the main events in the history of the analytic/synthetic dichotomy.

1 . 2 Nineteenth century morphological typology and the rise and fall of the concepts analyticity and syntheticity 1.2.1 The Schlegel brothers Since the beginning of linguistic study, attempts have been made to classify languages into groups. In general, language classification has been approached in four major ways: genetically, areally, sociolinguistically, and typologically.3 The genetic approach, which classifies languages according to their historical antecedents was, from Dante to the beginning of the last century, the earliest and only approach seriously considered.4 When the publications of Friedrich von Schlegel (17721829),5 August Wilhelm von Schlegel (1767-1845)6 and Wilhelm von Humboldt (1767-1835)7 on morphological classification appeared in the early 19th century, they not only marked a definite break in a long tradition, but also set the tone for the subsequent scientific approach to language classification.8 Among the three protagonists mentioned here, August Wilhelm von Schlegel, whose brother Friedrich had already set up a twofold classification into languages with affixes and languages with inflection (see n5), deserves particular attention for being the first to introduce the concepts of analyticity and syntheticity.9 In his Observations sur la langue et la littérature provençales (1818:14) he suggests a classification into three broad categories:10

4

Analyticity and Syntheticity (1) languages without grammatical structure (2) languages with affixes (3) languages with inflection

The first type, exemplified by Chinese, is characterized by roots which cannot be developed or modified. Syntax plays the main role in indicating the relationship among words for the languages of this class. The characteristic trait of the second type, exemplified by the then recently discovered Amerindian languages, is that these vernaculars express secondary ideas and relationships by attaching grammatical elements to other words. The third class, affixing languages such as the classical languages and their modern successors, is typified by the modification of "letters"11 of the roots and by the addition to the roots of derivational suffixes. 12 August Wilhelm von Schlegel's classification does not, however, end here. He further subdivides the inflectional type (group 3) into two subclasses, "synthetic" and "analytic": Les langues à inflexions se subdivisent en deux genres, que j'appelerai les langues synthétiques et les langues analytiques. J'entends par langues analytiques celles qui sont astreintes à l'emploi de l'article devant les substantifs, des pronoms personnels devant les verbes, qui ont recours aux verbes auxiliaires dans la conjugaison, qui suppléent par des prépositions aux désinences des cas qui leur manquent, qui expriment les degrés de comparaison des adjectifs par des adverbes, et ainsi du reste. Les langues synthétiques sont celles qui se passent de tous ces moyens de circonlocution. (1818:8) For A. Schlegel, this subdivision aims above all at distinguishing the classical languages (Greek, Latin, Sanskrit) from their modern successors. The former are synthetic because they express the relationship of one word to another by means of the forms of the words themselves. Analytic languages such as the modern Romance tongues, on the other hand, express these same grammatical relationships by means of separate words or simply by word position. A striking problem in Schlegel's model (a difficulty which, as we shall see, remained largely unresolved until Greenberg's contribution (1960 [1954]) more than a hundred years later) is that the division cannot be applied in absolute terms. A. Schlegel is forced to place the Germanic languages (excluding English which he considers analytic) in an intermediate class since they are, according to him, synthetic in origin but tending heavily to analytic forms: Έ η Europe, les langues dérivées du latin, et l'anglois, ont une grammaire toute analytique,... Les langues germaniques forment un classe

History of the terms

5

intermédiaire: synthétiques dans leur origine et conservant toujours une certaine puissance de synthèse, elles penchent fortement vers les formes analytiques' (p. 17). As a result, the synthetic/analytic dimension, meant to inject rigor into a newly conceived morphological classification of languages, is from its inception beset by the problem of "more or less." 13 Though it may never be fully known what exactly led A. Schlegel to propose the synthetic/analytic distinction, some scholars (particularly Greenberg 1974:38) see a correlation between the then prevailing view that agglutinative languages (Indo-European and Semitic) were superior to the isolating ones and the rise of the new analytic/synthetic subclassification. Like his brother Friedrich who claimed that the apparent richness of affixing languages was in reality an impoverished system with a 'subjectiv sonderbaren und mangelhaften Charakter' (52), August Wilhelm von Schlegel's writings are characterized by a strong value judgment. In his view, isolating languages are sterile and lack the kind of organic life typical of the more developed agglutinating tongues. Even more important, A. Schlegel sees a direct relationship between morphological structure and collective human intelligence. He, in fact, specifically states that the switch from synthetic to analytic constructs was brought about by the cultural decline of the post-classical era: Elles [synthetic languages] appartiennent à une autre phase de l'intelligence humaine: il s'y manifeste une action plus simultanée, une impulsion plus immédiate de toutes les facultés de l'âme que dans nos langues analytiques. A celles-ci préside le raisonnement, agissant plus à part des autres facultés, et se rendant par conséquent mieux compte de ses propres opérations. Je pense qu'en comparant le génie de l'antiquité avec l'esprit des temps modernes, on observera une opposition semblable à celle qui existe entre les langues. Les grandes synthèses créatrices sont dues à la plus haute antiquité; l'analyse perfectionnée étoit réservée aux temps modernes. (27-28) Greenberg (1974:38) points out that A. Schlegel must have been 'disturbed by the fact that the modern Indo-European languages have tended to lose the inflections found in the older stages,' thus apparently reverting back to a more primitive stage. To account for this difficulty, Greenberg believes that A. Schlegel saw himself forced to introduce a further subclass among the inflectional tongues, namely the synthetic for the older type and the analytic one for the modern languages. Although it would be difficult to disprove altogether Greenberg's thesis, there is at least some evidence that the transition from synthetic to analytic structures was perhaps not as disturbing to A. Schlegel as Greenberg would have it. We must keep

6

Analyticity and Syntheticity

in mind that since, on the one hand, A. Schlegel viewed his own culture as inferior to that of the classical period, and, on the other, he saw morphological structure as directly reflecting the cultural genius of a given age, it only made sense to him that the modem tongues showed a less perfect structure than that of their ancestors.14 In a revealing passage, A. Schlegel elaborates on the historical events, namely the barbarian invasions, which led to the breakdown of the once ideal synthetic morphology: Les conquérans barbares... trouvant dans les pays conquis une population toute latine, ou, selon l'expression du temps, romaine, furent en effet forcés d'apprendre aussi le latin pour se faire entendre, mais ils le parloient en général fort incorrectement; surtout ils ne savoient pas manier ces inflexions savantes, sur lesquelles repose toute la construction latine. Les Romains, c'est-à-dire les habitants des provinces, à force d'entendre mal parler leur langue, en oublièrent à leur tour les règles, et imitèrent le jargon de leurs nouveaux maîtres. Les désinences variables, étant employées arbitrairement, ne servoient plus qu'à embrouiller les phrases; on finit donc par les supprimer et par tronquer les mots .... Mais ces désinences supprimées servoient à marquer d'une manière très sensible la construction des phrases, et la liaison des idées; il falloit donc y substituer une autre méthode, et c'est ce qui donne naissance à la grammaire analytique. (24-25) According to A. Schlegel, the transition from a synthetic to an analytic language does not, however, always originate in a contact situation as the one he describes above: Lorsque les langues synthétiques ont été fixées de bonne heure par des livres qui servoient de modèles, et par une instruction régulière, elles sont restées telles; mais quand elles ont été abandonnées à elles mêmes et soumises aux fluctuations de toutes les choses humaines, elles ont montré une tendance naturelle à devenir analytiques, même sans avoir été modifiées par le mélange d'aucune langue étrangère. (18) The two passages above reveal that A. Schlegel may well have opted to subdivide the inflectional languages into an analytic and a synthetic group because two parallel processes (the downfall of the Roman empire and the rise of a more analytic morphology) fit very nicely into his preconceived ideas about the relationship between cultural sophistication and language structure. It is, therefore, unlikely that A. Schlegel's invention came as a response to what Greenberg sees as a

History of the terms

7

"disturbing" fact, namely that inflectional languages—regarded as the acme of excellence and placed at the head of a hierarchy of classes or types of language—had begun to decay. It is much more likely that A. Schlegel took to the new division simply to refîne a classification which, in its essence, had already been proposed by his brother Friedrich.15 One can rarely grasp the full import of intellectual achievements without considering the social and scientific climate in which they were accomplished. This is particularly the case for certain of the ideas of 19th century philologists. At a time when the concepts of analyticity and syntheticity were being discussed, Darwin's theory of evolution was having a profound effect on all areas of science, linguistics included.16 Models devised for biological theories were carried over into theories of language; this led to the notion that languages too had a "life cycle" whose individual stages could be described in evolutionary terms and explained by evolutionary laws. Once it was accepted that languages underwent a "growth" and, like animals and humans, could be traced back to a common ancestor by establishing an genealogical tree, it must have seemed only natural to 19th century comparativists to interpret the newly discovered formal differences among the world's languages as attestations of the intermediate stages through which human speech must have passed. Keeping in mind too the 19th century's sustained fascination with and admiration of Classical Antiquity (and its associated languages), we can understand more easily why a biological model of language was favored in which "primitive," "formless" languages (i.e., those lacking any signs of growth) such as Chinese were placed at the bottom of the scale, 17 while the "formed" (i.e., organically developed) tongues of the IE type were situated at the top. 18 Following A. Schlegel's line of thought, Sanskrit, Greek and Classical Latin now seemed all the more perfect, for—in addition to being the products of high cultures, the words of these languages could be shown to have "roots" from which the synthetic inflections had grown over time. 19 An ideal language, it was thought, was one endowed with words formed "organically," and the more the parts of these forms were intertwined (i.e., synthetic), the more this was interpreted as a sign of healthy growth. Within such a context the "mechanical" agglutination found in languages such as Tibetan or Polynesian naturally represented an intermediate, imperfect stage, while the more analytic systems of the modem IE languages represented the decay of a system once perfect 2 0 Although August Wilhelm von Schlegel clearly perceived that "perfect" synthetic languages could, and in fact do, revert back to more (primitive) analytic modes of expression (cf. his previously cited account of the evolution from Latin to the modern Romance tongues), the overall impression one obtains from reading his

8

Analyticity and Syntheticity

as well as subsequent works of the 19th century, is that, much like biological evolution, language evolution, and in particular the development "analysis —» synthesis," is nonetheless viewed as an essentially unidirectional and irreversible process. Fot as long as the Classical cultures prospered, languages such as Sanskrit, Greek, or Classical Latin could not help but gravitate uniformly towards the more complete synthetic type. Conversely, during the "post-Classical period of cultural decline," Romance and other IE tongues were "naturally" pulled down towards more analytic (and, therefore, less perfect) modes of expression. Such a scheme left no place for language-internal (or other non-cultural) factors which might stop or even reverse the momentum of a language towards greater analysis or synthesis. As the following chapters will show, A. Schlegel's theory of language change contains at least three assumptions which are of prime importance for an understanding of the subsequent history of the concepts of analyticity and syntheticity. He claimed, first, that entire languages (rather than individual structures) undergo the typological changes associated with analysis —> synthesis; second, that analysis precludes synthesis (and vice versa), i.e., the two processes do not take place simultaneously within a single language; and, third, that the change "analysis —» synthesis" is non-cyclical, i.e., languages are not seen as moving back and forth between these two extremes. I would point out that these assumptions still underlie many contemporary studies dealing with morphosyntactic typology. The uncritical absorption of these ideas into modern linguistics has, as we shall see, significantly added to the confusion surrounding the terms analytic/synthetic ever since the concepts were first introduced. 1.2.2 Reactions to Schlegel's analytic!synthetic division The next significant treatment of our topic comes from the pen of Wilhelm von Humboldt. Though explicitly rejecting any Schlegelian theory of historical evolution in which more advanced (i.e., inflectional) languages evolve out of the socalled primitive ones (i.e., those without structure and/or languages with affixes), Humboldt, in the spirit characteristic of German Romanticism, continues in the Schlegelian vein by relating morphological typology to the Volksgeist.2^ Most notably, however, he speaks out against the strict classiñcatory divisions introduced by A. Schlegel. In his Ueber das Entstehen der grammatischen Formen und ihren Einfluß auf die Ideenentwicklung (1822), he stresses the fact that in most languages there exist tendencies for inflection (Beugung) as well as agglutination (Anfügung) and that a systematic separation between them cannot be maintained (1963:46, esp. 334ff). 2 2 Even more important for our purpose are his serious objections to Schlegel's analytic/synthetic breakdown.

History of the terms

9

A. W. v. Schlegel hat diese Gattung der Sprachen mit dem Namen der ANALYTISCHEN, so wie die eines vollständigen organischen und beugungsreichen Baues mit dem der SYNTHETISCHEN belegt, und diese letztere Benennung vorzüglich ist in andere Schriften übergegangen [sie, A.S.]. Ich glaube mit einigen Worten angeben zu müßen, warum ich mich derselben absichtlich nicht bediene. Der Name der synthetischen soll zwar den Unterschied von agglutinirenden bezeichnen, daß die Synthese die einzelnen Theile in Eins verschmelzt, aber jede Synthese setzt immer ein zu verbindendes Mehreies voraus, und wo ist dies, wenn z.B. aus binden ich band wird? eine Lautbeugung, die gerade den feinsten Sprachorganismus vorzugsweise charakterisirt. Die Zusammenschmelzung in Eins läßt sich auch nur gradweise unterscheiden. Man kann nicht sagen, daß sie da sey, oder fehle, sie ist in gewissem Verstände immer vorhanden, nur mehr o d » weniger innig. Der in jede feinste Abschattung der Ideen eingehende Urheber jener Benennungen bemerkt bei den synthetischen und analytischen Sprachen selbst, daß die Gränzlinie nicht scharf zu ziehen ist, und es paßt dies noch mehr auf die synthetischen und affigierenden.... Der Ausdruck ANALYTISCHE Sprachen scheint mir noch wenig» passend. Es geht in den hier genannten Sprachen nicht sowohl eine Auflösung der synthetischen Formen vor, als daß man durch Verbindung einiger, unaufgelöst bleibender, andre entbehrlich macht. (Humboldt, Ueber die Verschiedenheit des menschlichen Sprachbaus [1827-29], ^ ^ Π ^ ΐ δ ) 2 3 By reading further in Humboldt's extensive work, one discerns that the synthetic/analytic division is rejected not simply because it is too vague to be useful but also because, in his view, non-genetic language classification is inherently impossible: Allein einer andren und solchen Classification, wo auch die gar nicht stammverwandten Sprachen nach allgemeinen Aehnlichkeiten ihres Baues zusammengestellt würden, wiederstrebt.... Die einzelnen Sprachen sind nicht als Gattungen, sondern als Individuen verschieden, ihr Charakter ist kein Gattungscharakter, sondern ein individueller. Das Individuum, als solches genommen, füllt allemal eine Classe für sich. (p. 189) Whether Humboldt's rejection of Schlegel's scheme was instrumental in the subsequent work of such prominent linguists as Franz Bopp and August Pott may only be guessed at today. What is striking, though, is that both Bopp (1833) and

10

Analyticity and Syntheticity

Pott (1833) spent considerable effort in making further contributions to the typologization of languages without incorporating the concepts of syntheticity and analyticity into their models. 24 A minor detour is in order at this juncture in order to mention the term "polysynthetic," which in later linguistics frequently appears in conjunction with the Schlegelian pair synthetic-analytic.2^ One would perhaps expect that this neologism was the outgrowth of a further refinement of Schlegel's division. This was not, however, the case. The proposal was made in 1819 (i.e., one year after Schlegel's Observation) by the American-based linguist Du Ponceau. Arguing that it was yet premature to attempt a classification of the world's languages, Du Ponceau (1838:447) introduced the term polysynthetic not to establish a more refined classificatory system but merely to propose a general label for the Amerindian languages familiar to him: 26 Partout où j'ai eu des renseignemens suffisans pour m'assurer de leur caractère, j'ai trouvé que ces langues appartiennent à la classe que j'ai nommée polysynthétique, dans la vue seulement de la désigner et sans y attacher autrement aucune importance. The next one to propose a distribution of languages according to their morphological structure is August Schleicher. Best known for his Stammbaum theory of genetic classification, Schleicher reintroduces the 'old' Schlegelian terminology of synthetic and analytic, not, however, without expanding their scope. As a refinement of the basic tripartite division into isolating, agglutinating, and inflectional languages, Schleicher (1850:9-10,18) proposes to apply the synthetic/analytic subdivision not only to the flexional but also to agglutinative types. It is interesting to note, however, that Schleicher is never explicit about what leads him to extend the subclassification to the agglutinative type; nowhere in his other writings on language classification (Schleicher 1848, 18S9, 1871) does he reiterate these ideas. 27 Schleicher's half-heartedness towards the notions of syntheticity and analyticity appears to have had an impact on his contemporaries concerned with language classification. Thus H. Steinthal (1850, 1860), for instance, who, after first being favorably disposed to the Schlegelian concept (1850:11-12), later refrains from implementing it in his own scheme (see his summary table "System der Sprachen als die Entwicklung der Sprachidee'' [p. 82]). Others, like M. Müller, simply reject the new ideas, arguing that it is not 'necessary to distinguish between synthetic and analytic languages' (1862:325) since morphologically, phrases like Fr j'aimerai (for j'ai à aimer) are still inflectional.28 Whatever the value of the various criticisms, in reviewing pre-1875 works concerned with morphological typology, one cannot fail to notice that Schlegel's

History of the terms

11

synthetic/analytic distinction was given a rough ride. 29 Based on the "fragmentary character'* of word structure, the idea of dividing languages into genetically unrelated groups convinced few, and a once debated idea and its terminology seemed to be well on the way out The lack of scientific precision in Schlegel's "more or less analytic" did not convince the Neogrammarians, who quickly began to make headway in the 1870's; they attempted to found a genuine science based on rigorous methods and exceptionless laws, whence they surely could see no point in establishing a classification into which some languages (e.g., modern Germanic in Schlegel's model) could not be fitted. Furthermore, as Greenberg (1974:40) succinctly points out, in the Neogrammarians' 'concentration on the specific phenomenon of languages ... there was no room for the classification of languages as wholes.' Morphological typology was, to be sure, quickly becoming a part of the past, and was left to a few loners such as F. Misteli (1893)30 and F. Fink (1901 and 1910),31 neither of whom embraced the Schlegel-Schleicher breakdown into analytic/synthetic languages. The net result was that the word and its morphemes, at the center of interest for over half a century, lost its once unchallenged position. Those like Schlegel who had toyed with a possible refinement of the basic tri- or quadri-partite division into synthetic and analytic subgroups, had apparently failed to provide a program that could secure for these concepts a spot in future linguistic discussion: analyticity and syntheticity, never firmly grounded from the day of their inception, were being forced into a quick and—at least from a late 19th century perspective—irrevocable retirement.32 1.2.3

The use of "synthetic" and "analytic" by Georg von der Gabelentz

Gabelentz (1881/1891 [1901]) uses "synthetic" and "analytic" not as classificatory concepts for the description of different language types, but rather as a complementary distinction between two existing systems of grammar coexisting within a single language. 33 The analytic system is fairly traditional and resembles modem structural grammar in that it takes the sentence as the basic unit and segments it into successively smaller constituents, ultimately arriving at the smallest elements of meaning. The synthetic system, on the other hand, rests on what was then an entirely different concept: it sought to establish, inter alia, how a given thought is formulated by the speaker. Put in Saussurian terms, Gabelentz's synthetic grammar thus takes the signifié as point of departure and attempts to show with the help of a series of quasi-transformational rules how and und» what circumstances one arrives at a given signifiant,34 In dealing with parts of speech, for instance, such a synthetic grammar would typically try to elucidate the following questions: Through what means are certain segments of a sentence replaced by other segments? Und«

12

Analyticity and Syntheticity

what circumstances can certain segments be dropped? What determines the choice between alternations of the type Fr hier était le vingt deux vs. le vingt-deux était hier 'yesterday was the twenty-second'? How do such (nearly) synonymous constructions differ from each other? (1878:660-661,1901:101-103) The author never attempts to present his synthetic grammar as a new theory of language. His motivation for positing the analytic/synthetic division is above-all practical for it is meant to facilitate both the description and acquisition of non-IE languages. In his Chinesische Grammatik, where the analytic/synthetic division finds its first large-scale application, he stresses the utilitarian value of his approach, and in so doing explicitly denies any scientific value to his synthetic method: Neu ist dagegen mein synthetisches System. Man kann den praktischen Werth eines solchen sehr wohl empfinden, ohne seine wissenschaftliche Berechtigung anzuerkennen. (1881 :xi) Gabelentz acknowledges that the perspectives of both the analytic and synthetic grammars are essential to understanding and teaching the mechanics of speech production (pp. 479-481), but underlines the importance of the synthetic grammar for handling the finer points of a language: Es gilt, sich in der fremden Sprache richtig, womöglich gewandt ausdrücken zu lernen. So muß man über eine möglichst vollständige Synonymik verfügen und die Unterschiede zwischen den Synonymen möglichst genau kennen. So aufgefaßt ist das zweite grammatische System [i.e., synthetic grammar] geradezu die hohe Schule der Sprachlehre,— sollte sie wenigstens sein. (1888:xii) Gabelentz argues that in addition to the didactic advantages of his system of grammar, its unique division into analytic and synthetic parts provides an alternative and improved way of describing those languages (e.g., Chinese) for which the traditional grammatical model, based on the structure of Latin and Greek, is ill suited. Gabelentz's innovative suggestions, in particular his peculiar use of "analytic" and "synthetic," did not impress his contemporaries, nor was his framework ever adopted by 20th century linguists. Although his idiosyncratic application of the analytic/synthetic distinction is ultimately of little import to the overall history of these concepts, Gabelentz's work is nonetheless important to an understanding of their history. For their use as something other than taxonomic terms makes clear that by the end of the 19th century Schlegel's division of languages into an analytic and synthetic type was all but forgotten.

History of the terms 1.3

13

Edward Sapir and the renaissance of an old idea

The Neogrammarian school had such a profound, long-lasting and negative impact on typological studies that the Schlegel-Schleicher division of languages into synthetic and analytic types was abandoned for over half a century. A new, radically different, and powerful movement which would sweep across linguistic studies— namely structuralism—stood as yet beyond the horizon. With respect to these upcoming developments and the Neogrammarian school, Sapir, in his much celebrated book Language (1921), became a transitional figure between the old and new. Transitional because, inter alia, he returns to an old and by then forgotten idea (analyticity/syntheticity) while introducing hitherto unknown techniques (e.g., the matrix presentation of language features [see below]), which in their basic conception announced future structuralism. Though obviously drawing his analytic/synthetic scheme from Schlegel and Schleicher, he rejects the subjective evolutionary aspects which typified his predecessors.33 His classification of language types is complex and complicated,36 undoubtedly because he realizes that the simple traditional tri- or quadri-partite morphology-based division of the 19th century typologists is a strait-jacket fitting only a few, if any, better known languages.37 He, therefore, distinguishes not one but three separate axes of classification: (1) grammatical concepts, (2) grammatical processes, and (3) firmness of affixation. This third, and most important distinction in our discussion, concerns 'the relative firmness with which the affixed elements are united with the core of the word' (p. 135, emph. mine). Sapir (pp. 135-136) establishes the following three degrees of affixation: (1)

ANALYTIC:

A language that does not combine concepts into single words at all (Chinese) or does so economically (English, French). In an analytic language the sentence is always of prime importance, the word is of minor interest.

(2)

SYNTHETIC:

In a synthetic language (Latin, Arabic, Finnish) the concepts cluster more thickly, the words are more richly chambered, but there is a tendency, on the whole, to keep the range of concrete significance in the single word down to a moderate compass.

14

Analyticity and Syntheticity (3)

POLYSYNTHETIC: More than ordinarily synthetic. The elaboration of the word is extreme. Concepts which we should never dream of treating in subordinate fashion are symbolized by derivational affixes or "symbolic" changes in the radical element, while the more abstract notions including the syntactic relations, may also be conveyed by the word. A polysynthetic language illustrates no principles that are not already exemplified in the more familiar synthetic languages. It is related to them very much as a synthetic language is related to our own analytic English.

Sapir (p. 136) stresses that 'the three terms are purely quantitative—and relative, that is, a language may be "analytic" from one standpoint, "synthetic" from another.' To illustrate the relativity of his classification, Sapir (136 n i l ) proceeds to explain that English, typed as analytic (see above), is analytic only in tendency since, relative to French—at least in certain aspects—it is still fairly synthetic. Sapir, in other words, does not aim at establishing a rigid classification, i.e., one in which a language must fit into either the synthetic or the analytic compartment, but rather seeks to highlight structural tendencies (the famous "drifts") 38 within a given language.39 By employing three sets of distinctions (conceptual type, technique, and degree of synthesis) Sapir offers a wide variety of possibilities for language characterization. 40 His matrix (p. 151, reproduced here as Table 1), with all its complex possibilities of variation,41 contains an unprecedented combination of features, some of which are mutually exclusive. Thus the languages given und» his type A are necessarily analytic. Languages of type C are also predominantly analytic and are, according to Sapir (p. 148), not likely to develop beyond the synthetic stage. Sapir does not limit his attention to merely establishing parameters for his morphology-based matrix. He clearly goes beyond such considerations when he attempts to draw inferences from the relationship between diachrony and structural type of language structure. He argues, for instance: it is interesting ... to note that of the three intercrossing classifications represented in our table (conceptual type, technique, and degree of synthesis), it is the degree of synthesis that seems to change most readily, that the technique is modifiable but far less readily so, and that the conceptual type tends to persist the longest of all. (154)

History of the terms

15

Earlier he comments: it is often illuminating to point out that a language has been becoming more and more analytic in the course of its history or that it shows signs of having crystallized from a simple analytic base into a highly synthetic form. (136) 42 Despite Sapir's vastly improved methodology of basing language classification on non-genetic criteria, and despite the sophisticated and original incorporation of analyticity and syntheticity into a multi-layered matrix, it is easy to see why his scheme might be open to serious criticism. In the tradition of Schlegel and Schleicher, Sapir provided a sliding scale of syntheticity which relied heavily upon vague and arbitrary criteria. His highly subjective qualifying notations of "mildly synthetic," "mildly polysynthetic," "notably synthetic," "very nearly complex purerelational," etc. simply could not satisfy those who continued to demand rigorous and unambiguous taxonomic parameters.43 Moreover, Sapir's strange rejection of his own thoughts on the analytic/synthetic concept on the pages immediately following his exposition thereof, help to show that Sapir himself eventually discarded his initial enthusiasm for the analytic/synthetic division.44 Less obvious shortcomings of Sapir's chapter 'Types of Linguistic Structures" came to light in the years following publication of his book. Velten (193S) pointed out that it was never entirely clear whether semantic, syntactic, or morphological criteria (or a combination of these) were to be applied in determining the degree of analyticity and/or syntheticity.45 Moreover, he objected to Sapir's analytic/synthetic division because it would be wrong to distinguish a synthetic from an analytic language based on the following juxtaposed expressions: reiten - equo vehi, patiner - Schlittschuh laufen, Schimmel - cheval blanc, naschen - to eat dainties surreptitiously (p. 6). For Velten, a language classification based on such principles is inadequate, all the more so because the most fundamental criterion for determining the degree of synthesis and/or analysis is based on a concept, the word, for which a satisfactory definition had never been given (p. 21). 46 Whatever the apparent shortcomings of Sapir's work on language typology, and however vague his use of the terms analyticity and syntheticity, the Schlegelian terminology did, as we shall see in the following section, once again make its way into linguistic discussion; there can be no doubt that the immense popularity of Sapir's Language contributed significantly to the spread of a now very popular terminology which might otherwise have fallen into oblivion.47

16

Analyticity and Syntheticity

Table 1. Sapir's Classification of language types Fundamental Type A (Simple Purerelational)

(Complex Purerelational)

α

m

IV

Technique

Synthesis

Examples

(d)

—

a a,b

Analytic Analytic

Chinese; Annamite Ewe (Guinea Coast)

(b)

—

Isolating Isolating (weakly agglutinative) Agglutinative (mildly agglutinative-fusion al)

Analytic

Modem Tibetan

b,(d) b c b b,d

—

Agglutinative-isolating Agglutinative-isolating Fusional-isolating Agglutinative Agglutinative (symbolic tinge) Fusional-agglutinative (symbolic tinge)

Analytic Polysynthetic Analytic Synthetic Polysynthetic

Fusional

Synthetic

Polynesian Haida Cambodgian Turkish Yana (N. California) Classical Tibetan Sioux polysynthetic) Salinan (S.W. California) Shilluk(Upper Nile)

—

c.d. (b) c

a.b. c

(b)

a ».(b) a b b

—

a,b

—

c

— — —

Synthetic (mildly) Synthetic (mildly

Analytic d,c — d,c,a Symbolic NOTE.— Parentheses indicale a weak development of the process in question. Fundamental Type

η

m

IV

Technique

Synthesis

Examples

(Simple Mixedrelational)

(b) (c)

b c,(d)

a

Agglutinative Fusional

Synthetic Analytic (mildly synthetic)

Bantu French*

D (Complex Mixedrelational)

b.c. d

b

b

Agglutinative (symbolic tinge)

Polysynthetic

b

Nootka (Vancouver lsland)f Polysynthetic(mildly) Chinook (lower Columbia R.) Al gem ki an Polysynthetic

c,(d)

Fusional-agglutinative

c.(d). c,(d) (b) c a c,d c,d c,d —

Fusional Analytic Fusional FusionaI(symbolic tinge) Synthetic

c.b,

c,d

(a)

d,c

c.d

(a)

Fusional (strongly symbolic) Symbolic-fusional

•Might nearly as well have come under D. fVeiy nearly complex pure-relational.

Synthetic Synthetic

English Latin, Creek, Sanskrit Takelma (S.W. Oregon) Semitic (Arabic, Hebrew)

History of the terms

17

1.4 The post-Sapirian era The years following Sapir's gallant attempt to develop a more solid and richer typology are mailed by the slow integration of the terms analyticity and syntheticity into linguistic description; unfortunately, those who adopt the Sapirian terminology inevitably run into the problem of being overly vague. Such is the case, for instance, in Haas' treatment of Tunica where the verbatim use of Sapir's "mildly synthetic" inevitable calls into question the value of her characterization: 'The Tunica language is mildly synthetic in structure. In its technique of synthesis it is for the most part agglutinative but it also employs a limited amount of fusion' (1946:345; emph. mine). Part of the problem in Sapir's and Haas's morphologybased characterization of language is that they typologize languages as wholes rather than distinct features or categories. In other words, they fail to indicate which structures within a language, if not all, are synthetic; without particulars it is impossible to know whether "mildly synthetic" holds for all morphological structures of Tunica or whether it is simply the average, i.e., the overall impression obtained from a variety of differing morphological types. In light of what appears to me to be rather obvious criticism, one wonders why Sapir's synthetic/analytic distinction nevertheless slowly became a well-recognized (though still poorly defined) concept among the linguistic community. Symptomatic of the post-Sapirian era is the coexistence of linguists' criticism of the synthetic/analytic classification and their unwillingness to abandon or redefine it in a more satisfactory manner. A case in point is Bally (1965 [1932]: 143) who remarks that 'les termes de synthèse et d'analyse ont été employés dans des sens si différents qu'il vaudrait mieux les abandonner.' Unlike many, though, Bally at least makes an effort to provide a better definition: On peut aborder la question au point de vue génétique. La pensée non communiquée est synthétique, c'est-à-dire globale, non articulée. Ce sont les besoins de la communication qui poussent à analyser les éléments de la représentation et à les regrouper organiquement Un spectacle insolite peut nous arracher un cri d'étonnement, mais cet étonnement peut aussi s'exprimer par une phrase telle que 'Voilà qui est étrange!'; le cri est synthétique, la [Arase analytique. On peut donc dire qu'un procédé linguistique est d'autant plus synthétique qu'il se rapproche de la nébuleuse primitive, c'est-à-dire de la pensée non communiquée. (143) Bally's attempt fails, however, to provide a more practical working definition, a difficulty which he himself underlines:

18

Analyticity and Syntheticity Mais une pareille définition, admissible en théorie, n'offre guère d'avantages pratiques; c'est là un critère qui se dérobe à chaque instant à la vue du chercheur, lorsqu'il s'attaque à une langue constituée, dont la structure a perdu tout contact avec l'indistinction primitive. Il est facile alors d'attribuer à la synthèse tel procédé qu'un autre baptiserait analyse et vice versa.48 (143)

As we move further into the twentieth century, it becomes increasingly clear that serious discussions of morphological classification can no longer bypass the traditional break-down into analytic and synthetic languages. Leonard Bloomfield (Language 1963 [1933]:207) is no exception; like others he brings up the subject but, rath«- elegantly, avoids a thoroughgoing discussion of the problems involved by restricting himself to a less-than-half page description (rather than a "discussion") of the Sapirian division.49 The ill-defined concepts of analyticity and syntheticity remain unchanged until 1945 when V. Tauli, in an article entitled "Morphological Analysis and Synthesis," broke new ground toward a redefinition , or—more accurately—a refinement of the analytic/synthetic concepts. In a brief but incisive article ignored by some of the linguists who later contribute to the discussion,50 Tauli sets forth a list of eleven criteria on which to base the degree of analysis and synthesis. In his scheme, the analytic or synthetic character of flexional forms depends on the following circumstances: 1. On linearity: a linear form is more analytic than an alinear one. 2. On whether a common or an alternant stem is present: a form with a common stem is more analytic than one with an alternant stem. 3. On whether a common or an alternant morpheme appears: a form with a common morpheme is more analytic than a form with an alternant morpheme. 4. On whether the stem occurs independently and has an independent meaning. 5. On whether the morpheme occurs independently and has an independent meaning. 6. On the phonetic strength of the morpheme: a phonetically strong morpheme, e.g„ a syllable, is more analytic than a phonetically weaker morpheme, e.g., a non-syllabic sign.

History of the terms

19

7. On whether the elements of a form are separable in a context (frequently referred to in the literature as "criteria of displaceability"). 8. On whether the order of the elements may be changed 9. On whether in the case of coordination the morpheme appears with one word or with all words. 10. On whether in the case of subordination there is concord: an expression without concord is more analytic than one with concord. 11. On the order of elements of expression: progressive order is more analytic than regressive. Since the characteristics summarized above may, from the point of view of analysis and synthesis, occur in very different combinations in any given word, a form can be labelled analytic or synthetic (or more analytic or more synthetic) only relatively, i.e., with reference to one or several of the above criteria. A form that is analytic or synthetic in reference to all eleven criteria can naturally be classified with absolute certainty. I find that up to this moment in Tauli's scheme, the basic concept could be illuminating. However, the author falls back into the tradition of purely subjective (and therefore questionable) analysis by suggesting (p. 84) that 'some criteria ... carry so much weight that a form might practically be classed as analytic on the basis of such criteria only.' Tauli never elaborates on exactly how one is to weigh these varying criteria, which leads to a classification which is as subjective and vague (and therefore as inadequate) as those of his predecessors.51

1.5 Joseph Greenberg's method of quantification The quantification of morphological typology as illustrated by Greenberg (1960 [1954]) maries a giant step toward exactitude and rigor in the classification of morphological structures. 32 Recognizing the general dilemma brought about by the terminological—hence conceptual—subjectivism in linguistic analysis, Greenberg remaries: 'Time seems propitious to reexamine the 19th century approach to the problem, discarding what the intervening period of linguistic criticism has demonstrated as invalid and incorporating recent methodological advances in order to reformulate the hypotheses along more rigorous lines' (p. 180). Using the analytical tools of contemporary American structuralism, Greenberg essentially adopts the

20

Analyticity and Syntheticity

Sapirian classification in revised form. However, unlike Sapir, he clearly differentiates between distinctive units (morphemes) by a formal, not a semantic test. 53 Moreover, in place of intuitive estimates based on overall impressions of a few seemingly relevant features, Greenberg makes attempts to define each of the characteristics at issue in terms of a ratio of two units (see Table 2, column 2). The first parameter given is the degree of synthesis, also termed "gross complexity of the word," which is calculated by a simple mathematical formula (total of MORPHEMES divided by total of WORDS (M/W) which yields to ratio of morphemes paword. This measure is called the "synthetic index." Since the word is taken as the upper limit within which all morphemes must fall, the theoretical lower limit is obviously 1.00. Accordingly, a totally analytic language yields a synthetic index of 1.00 whereas a "mildly" synthetic language with an average of two morphemes per word yields an index of syntheticity of 2.00 (2 morphemes/1 word = 2.00). Analytic languages, therefore, give low results on this index, synthetic higher, and polysynthetic highest of all. The question then arises as to where one makes the (admittedly arbitrary) cut between synthetic and polysynthetic languages. After plotting a frequency distribution of eight languages from different families, Greenberg concludes (p. 194) that an analytic language typically shows an index of 1.00-1.99, a synthetic 2.00-2.99, and a polysynthetic 3.00+, which, by and large, confirms the conclusions of previous, nonquantitative classifications. Basic to the synthetic index is, of course, the possibility of segmenting utterances into words. As Greenberg himself acknowledges, an unambiguous definition of the word is clearly essential inasmuch as the index (as well as all his other indices not described here) crucially depends on an acceptable word analysis. The author readily admits that there is no general agreement on this question. The procedure adopted by Greenberg (pp. 192-193) is unconventional.54 Without going into detail here, it should be noted that Greenberg's attempt at word definition, like so many others, is far from being unproblematic. Thus enclitic Latin -que 'and' is part of the word to which it attaches and is counted as part of that word by the present test of synthesis. Despite such shortcomings, Greenberg (p. 192) believes that his procedure for segmenting words leads to satisfactory results because in the languages considered there were relatively few doubtful cases. Since he outlines his procedures for dividing words only briefly, it is difficult to judge whether other languages would also yield a low number of so-called "questionable" words.55

History of the terms

21

Table 2. Greenberg's Typological Index Values56 Sk

Synthesis Agglutination Composition Derivation Gross Inflection Prefìxation Suffixation Isolation Pure Inflection Concord

M/W A/J R/W D/W I/W P/W S/W O/N Pi/N C/N

2.59 0.09 1.13 0.62 0.84 0.16 1.18 0.16 0.46 0.38

AS

Per

2 . 1 2 1.52 0.11 0.34 1.00 1.03 0.20 0.10 0.90 0.39 0.06 0.01 1.03 0.49 0.15 0.52 0.47 0.29 0.38 0.19

Eng

Yak

1.68 2 . 1 7 0.30 0.51 1.00 1.02 0.15 0.35 0.53 0.82 0.04 1.16 0.64 1.15 0.75 0.29 0.14 0.59 0.11 0.12

Swa

Ann

Esk

2.55 1 . 0 6 3.72 0.67 0.30 1.00 1.07 1.00 0.07 1.25 0.80 1.75 0.41 0.40 0.19 0.41

2.72

1.00 0.02 0.46 0.38

M = Morpheme; W = Word; A = Agglutination, lack of morphophonemic change at morpheme juncture; J = Morpheme juncture; R = Root morpheme; D = Derivational morpheme; I = Inflectional morpheme; Ρ = Prefìxation; S = Suffixation; O = Word Order; Ν = Nexus, expression of word interrelations; Pi = Pure inflection; C = Concord. Sk = Sanskrit; AS = Anglo-Saxon; Per = Persian; Eng = English; Yak = Yakut; Swa = Swahili; Ann = Annamite; Esk = Eskimo.

Based on Greenberg (1960 [1954]: 193). This time around, the linguistic community did not fail to show a quick and strong response to the new methods for defining analyticity and syntheticity. Unlike the apathetic Sapirian 'twenties, the 'fifties and 'sixties are a propitious time for the mathematically flavored Greenberg proposal. Among the Greenberg supporters I mention Kroeber (1960), who shows no less than enthusiastic support. After summarizing the main ideas expressed in Greenberg's papers, he hails the innovation as a new method which 'is not in the least a question of pedantry through wearing the false plumes of physical science' (p. 176).57 While Sapir's merit rests on the notion that the previous non-genetic classification relied upon a fairly casual rating of the frequency and degree of formal adhesion, Greenberg's originality lies in the elaboration of a quantitative scale of measurement. The real advance of this method over earlier approaches comes to light when one considers that, as regards syntheticity and analyticity, we are now not only capable of ranking each language relative to all other languages, but that the classification is capable of showing in numeric terms the difference between two adjacent tongues.58 This advantage is clearly visible in Table 2, where the difference in degree of synthesis between Eskimo (3.72) and the immediately lower ranking Sanskrit (2.59) is as great as that

22

Analyticity and Syntheticity

between a highly analytic (Annamite: 1.06) and a synthetic (Sanskrit: 2.S9) language. More important, the quantitative approach, unlike the purely terminologybased method, offers the possibility of establishing formulas (e.g., Index of Synthesis = M/W) for objectively calculating the indices. This last point cannot be emphasized enough since clearly defined guidelines are essential for obtaining undistorted results in calculations involving large numbers of languages and, by necessity, the collaboration of numerous linguists. Finally, and perhaps most importantly, Greenberg's determination to delimit formally all the variables involved (i.e., the morpheme, word, etc.) in the calculation of his classification gives ample evidence that his method is anything but "Sapir's scheme translated into currently fashionable jargon" (see n57). It would present a distorted picture to champion Greenberg's approach without also pointing out certain difficulties in his method which remain unresolved even in later publications. It is, first of all, curious that Greenberg has not, as far as I have been able to determine, followed up on these ideas regarding morphological typology. The fact that he did not pursue his ideas, coupled with the sketchy outline of his procedure for isolating words,59 detracts from an otherwise promising approach. Pierce is, therefore, correct in pointing out that 'if Greenberg's typological methodology is to be utilized and expanded, rigid definitions of terms will have to be agreed upon and followed by all investigators to insure comparable results' (1966:4s). 60 Further problems concern the overall conceptualization rather than internal methodological aspects of Greenberg's synthetic/analytic classification. It must be remembered that the average ratio of syntheticity is calculated over a stretch of continuous text for a variety of languages. As Lyons (1968:188) points out, a given language might be, and frequently is, relatively isolating with respect to some classes of words and relatively synthetic with respect to others, thus the ratio of syntheticity might be quite different were it calculated across the entire lexicon of a language with each word counted once only. This observation points to a further problem: in any given language, highly analytic and highly synthetic (polysynthetic) structures may average each other out, the net result being a distortion of facts (a point in case is, for instance, French where synthetic verb forms [cf. j'aimerai] stand alongside more more analytic ones [cf. je vais aimer]).61 These drawbacks have recently been recognized and solutions are being proposed. It is evident, of course, that the difficulty can only be eliminated by calculating indices of syntheticity for individual, language particular, grammatical categories rather than for languages as wholes. 62 Haarmann (1976:60), recognizing the need for such "partial" typologies, reasons as follows:

History of the terms

23

Weiterhin halte ich es methodisch für unbedingt erforderlich, im Raster die Differenzierung des morphologischen Teilsystems kenntlich zu machen (vgl. Systemteil 1 = Nominalsystem, Sysemteil 2 = Pronominalsystem, Systemteil 3 = Verbalsystem). Diese Differenzierung ist notwendig, um Pauschalurteile über den Grad des Synthetismus in verschiedenen verglichenen Sprachen zu vermeiden. Der Grad des Synthetismus ist in den morphologischen Systemteilen durchaus verschieden. Eine Sprache kann daher nicht bezüglich des Synthetismus des gesamten morphologischen Teilsystems, sondern zutreffenderweise nur der morphologischen Systemteile klassifiziert weiden. The remarks at the end of the previous paragraph, though in a sense closing the gate on classification of languages according to an overall degree of analyticity and syntheticity, suggest an important alternative method for applying the traditional terminology analytic and synthetic. What this means is that linguists interested in identifying crosslinguistic similarities and/or diachronic variations in the index of syntheticity must do so by comparing not languages as wholes but individual phenomena. A major advantage of such a procedure is not only the avoidance of gross oversimplifications in language classification but also the unprecedented opportunity to establish possible correlations between linguistic categories, their degree of synthesis, and their rate of change relative to the index of syntheticity. Only then, after assessing the data from this angle, are we able to make firm and reliable conclusions concerning what has generally been referred to as drifts or trends towards a new morphological type. Following the lead of Greenberg's quantitative methods, typologists in the past two decades have recently reformulated and extended the classic quantitative approach. It has become increasingly obvious that one must select carefully a number of different strategies to obtain a more balanced picture. V. Krupa (1965), V. Krupa and G. Altmann (1966), E. SlaviCková (1968), J. Kelemen (1970), G. Altmann and W. Lehfeldt (1973), and H. Haarmann (1976) all subscribe to such a view and propose new avenues. A direct consequence of this search for new strategies in language typology is that alternate ways for calculating the index of synthesis are being suggested. Although the underlying approaches strongly resemble that of Greenberg, there are nevertheless some important methodological differences. The first, and also least revolutionary among those who have modified the standard computations of the index of synthesis is V. Krupa (1965). His approach is to invert Greenberg's formula of M/W to W/M, with the result that the theoretical upper limit (1.00) reflects total analyticity, whereas ranges close to 0.00+ indicate a very high degree of synthesis. The new calculation offers nothing

24

Analyticity and Syntheticity

not already contained in Greenberg's equation, with the exception that now the boundaries of the theoretical fluctuations in the index (> 0.00 - 1.00) are in tune with all other types of indices (where the lower and upper limits are also > 0.00 1.00). The usefulness of this "cosmetic" change is demonstrated most clearly in Krupa and Altmanns article "Relations Between Typological Indices" (1966), which outlines complex calculations correlating two or more indices. A more dramatic change in the formula is given by Kelemen (1970:53), who stresses the need to investigate not only the percentage of characteristic traits (e.g., the index of synthesis obtained by the equation MORPHEME/WORD) but also the proportion of the individual characteristics. What this amounts to in practice is the calculation of an index of syntheticity based not on MORPHEME/WORD but on ANALYTIC WORDS/TOTAL WORDS IN TEXT.63 Thus in the sentence The boy killed the duck-ling Greenberg's index of synthesis will yield 1.40 (7 morphemes/5 words) whereas Kelemen's will result in 0.60 (3 analytic words [the; boy; the]/5 words). 64 The difficulty with this approach is that Kelemen fails to suggest not only what exactly constitutes a word but also at what point a word is to be considered analytic or synthetic. As the author admits, his model is, at this point, hardly more than a "provisorische Annäherung" (p. 57), and it is, therefore, not easy to cull from his article what advantages there are in this novel approach to calculating the degree of synthesis.62 Variations on the Kelemen theme had already been advanced two years earlier by Eleonora SlaviCková in a short paper entitled 'Towards a Typological Evaluation of Related Languages' (1968). The author elaborates a complex schema 'for the evaluation of distinctive and integrative features between related languages' (p. 282). One of the features in her classification deals precisely with the degree of synthesis. 66 SlaviCková proposes to divide the number of root morphemes by the total of morphemes. It must be noted, however, that the morpheme count is obtained by the segmentation of lexical (autosemantic) words only (p. 284). That is to say, grammatical (synsemantic) words form a separate, non-segmented class which do not enter into consideration here. To take the example cited previously (The boy kill-ed the duck-ling), she would divide the 3 root-morphemes (boy, kill-, duck-) by the 5 morphemes of the lexical words in the sentence, which, in turn, yields an index of 0.60. The following hypothetical sentences and their resulting calculations should point up the differences in Greenberg's, Krupa's, Kelemen's, and SlaviCková's approaches:

History of the terms Table 3.

25

Comparison of techniques in the calculation of indices of synthesis as proposed by Greenberg, Knipa, Kelemen, and SlaviCková67

Morphemes Root Morph. Worts Hypothetical morpheme distribution 4 2 1) [m] i M-m-m-m I [m] I M-m-m-m 10 4 4 2) M I M-m I M I M-m-m 7 4 4 12 3) M-m I M-m-m-m I M-m I M-m-m-m M = root-morpheme; m = morpheme; - = morpheme boundary; I = Word boundary; [ ] = synsemantic (morphemes without square bracket are autosemantic) Differences in calculations methods and results Method Greenberg Krupa Kelemen SlaviCková

Example 1 m/W W/m AW/SW RM/m

10/4 4/10 2/2 2/8

= = = =

2.50 0.40 1.00 0.25

Example 2 7/4 4/7 2/2 4/7

= 1.75 =-0.57 = 1.00 =-0.57

Example 3 12/4 4/12 0/4 4/12

= 3.00 =~0.33 = *.** =-0.33

m = morpheme; RM = root-morpheme; AW = analytic word; SW= synthetic word; *.** = faulty calculation due to the nature of the equation Note that Greenberg's and Krupa's indices reflect the difference between exx. (1) and (2) but fail, for instance, to indicate whether the change in the index between (1) and (2) is due to an increase in analytic forms or to a decrease in synthetic constructs. Kelemen's model fails to pick up the difference between (1) and (2) because its calculation yields the ratio of purely analytic over non-analytic (i.e., synthetic AND analytic) structures, thus being insensitive to variations within the synthetic spectrum of the scale. Before coming to the concluding remarks on the history of the problem, I must mention T. Vennemann's recent attempt at eliminating the noted ambiguity of the synthetic and analytic concepts. In an article entitled "Isolation—Agglutination— Flexion? Zur Stimmigkeit typologischer Parameter" (1982), the author sets out to define the terms "isolating," "agglutinating," "inflectional," "synthetic," and

26

Analyticity and Syntheticity

"analytic." Vennemann distinguishes "formal," "functional," and "semantic" parameters, with synthesis/analysis and f u s i o n / c o l l o c a t i o n ^ coming under the first of these. By his definition, synthesis is a 'prozeßmorphologische Ableitung einer Form (nämlich eines Stammes oder eines Wortes) aus einer oder mehreren andern' (p. 330) whereas analysis is simply the opposite. The process of synthesis occurs when the following take place: composition, reduplication, affixation (prefixing, infixing, suffixing), mutation (segment, vowel, consonant, accent, and tone alternation), and subtraction. 6 ' Having established the ground rules for the basic distinctions between the various parameters, Vennemann then proceeds to give the following definitions of synthetic, analytic, and polysynthetic languages: Ich nenne eine Sprache synthetisch in dem Maße, wie ihre Wörter mittels Synthese gebildet sind, analytisch in dem Maße, wie dies nicht der Fall ist .... Unter Polysynthese verstehe ich die Anwendung mehrer ("vieler") Syntheseschritte bei der Bildung eines und desselben Wortes und unter einer polysynthetischen Sprache eine solche mit häufiger Polysynthese. (333) Vennemann's search for definitions of analytic, synthetic, and polysynthetic does not, however, lead to a clarification of the three concepts. First, like A. Schlegel, Schleicher, Sapir, and many others before him, he fails to recognize that a non-quantitative definition will be insufficiently rigorous. Given the repeated criticism in recent literature of ambiguous classificatory terms, one wonders why Vennemann does not heed the premonitions: '"mehrere" ("viele") Syntheseschritte"' is hardly less vague than A. Schlegel's "more or less synthetic," nor is it clear how many synthetic words it takes to characterize a language as either synthetic or analytic. Perhaps most crucially, Vennemann ignores the sobering, though, I believe, healthy conclusion of a hundred and fifty years of research and probings in morphological typology, namely that the parameters synthetic and analytic cannot be applied to languages as wholes in any meaningful way. 70 Seen in this light, it is regrettable that the latest effort to find a satisfactory definition for syntheticity and analyticity reverts back to methods shown to be so detrimental to linguistics in general and morphological typology in particular.

1.6 Summary Although it might appear at first glance from our survey of the history of analyticity and syntheticity that the earliest uses of the terms have very little in common with their mathematically flavored applications of the past thirty years, the two

History of the terms

27

terms are nevertheless consistent in that they take as point of departure the psychological bond which, according to those who have contributed to the definition, exists between the stems and the affixes of a particular language. We have seen that syntheticity and analyticity have played a very important role in language classification from the heyday of comparative linguistic studies in the early nineteenth century down to the present time. Schlegel's incorporation of the analytic/synthetic division into his threefold morphological typology of isolating, affixing, and inflectional languages was an attempt to distinguish the structural differences noted in inflectional languages between the classical (synthetic) tongues and their modern (more analytic) descendants. The initial partition "analytic" and "synthetic" was then extended by Schleicher to apply not only to inflectional but also to agglutinating languages. Our survey pointed out that during the past century, the concepts of syntheticity and analyticity met with swift opposition from a number of scholars (Humboldt, Müller, Steinthal, Pott, and others) and virtually disappeared from discussion for the post-1870 generation of linguists. This was due, in part, to the rise of the Neogrammarian school, but certainly also to the fact that the analytic/synthetic distinction appeared simplistic, unsophisticated and overly vague for a scientific classification of the world's languages. The basis for the renaissance of the analytic/synthetic concepts in the early 1920's was Sapir's Language. For the first time syntheticity and analyticity were incorporated into a multi-layered matrix which permitted a nuanced and more finely graded classification. Yet despite the advances of this method over earlier approaches, a strong dose of subjectivism still attached to these parameters, due to the presence of vague phraseology like "mildly synthetic" and the absence of a rigorous, quantitative method. The first energetic and carefully planned step toward a quantitative method was made by Greenberg. Dividing the number of morphemes by the number of words in a given text, he proposed to calculate the mean word length in a variety of languages. His ideas were generally well-received and laid the foundation for all subsequent quantitatively based definitions of syntheticity and analyticity. Perhaps his greatest contribution was to stress the need for mathematical formulas on which to base calculations. Yet notwithstanding the rigor of his methodology, several problems remained largely unsolved, notably the sketchy outline of his definition of the word, and the less than ideal formula (M/W) for calculating the index of synthesis. One of the major drawbacks of his equation is that the resulting index says much about the average word length but virtually nothing about the possibly varying degrees of syntheticity or analyticity of different grammatical structures within a language. SlaviCková (1968) and Kelemen (1970) have taken initial steps to correct some of Greenberg's shortcomings: besides providing new formulas for

28

Analyticity and Syntheticity

the calculation of alternative indices, they have argued convincingly that in order to be meaningful, these parameters must be applied to individual cross-linguistic categories rather than languages as wholes. Despite the advances just mentioned, the most important task, namely the search for sound operational definitions, remains yet to be solved. It is one thing to perfect quantitative formulas for cross-language comparison, but there still remains the more serious problem of finding a generally acceptable working definition of "the word." If syntheticity and analyticity are to be significant factors in future comparative analysis, linguists cannot continue to gloss over this difficulty, nor circumvent it through simplistic solutions.^1 Boris Uspensky situates the problem of quantification in proper perspective when he states: 'it should be borne in mind that any statistics presupposes a preliminary deterministic theory (before quantification we must define what is to be quantified)' (1968:23). In light of these remarks it is not inappropriate, I believe, to make the definition of the word the core of my next chapter and to conclude this summary with a quote taken from Comrie's book on language universale and linguistic typology: Even in trying to apply the index of synthesis in practical terms, for instance by dividing the number of morphemes by the numb» of words, certain practical problems arise which indicate that still further attention must be paid to the theoretical basis of morphological typology. Perhaps the most obvious, and the most widely discussed in the literature, is the question of establishing word boundaries, and thence the number of words in a sentence. (1981:44)

Chapter two Word delimitation: in search of a universal Pour échapper aux illusions, il faut d'abord se convaincre que les entités concrètes de la langue ne se présentent pas d'elles mêmes à notre observation. Qu'on cherche à les saisir, et l'on prendra contact avec le réel; partant de là, on pourra élaborer tous les classements dont la linguistique a besoin pour ordonner les faits de son ressort. (F. de Saussure, Cours de linguistique générale) 2.1 Introduction In this chapter I shall attempt to answer the question of why an entity such as "the word," taken for granted by the layman, is so difficult to define for the linguist. Why is it that an element of language which the naive speaker recognizes (or feels he recognizes) intuitively in most instances is one about which linguists have so little to say?1 The issue at stake here is, of course, that of the levels of linguistic analysis. Much of our discussion in this chapter will, therefore, attempt to highlight how this analysis has been carried out in the past, and why it is that the search for a universal definition of the word has not been successful. Before proceeding with a survey of definitions, I believe it is necessary to ask what requirements must be fulfilled for a definition of the word to be considered satisfactory. As in any scientific definition, we should ideally be satisfied only with proposals which allow neither exceptions nor marginal cases whose status cannot be decided objectively. This means that (1) if we take the sentence as the outer limit, ALL the phonemes within that sequence must fit within a finite number of words exhaustively, i.e., no phoneme may remain unassigned, and (2) the definition must be applicable to all languages of the world. Finally, we would like the definition to be accompanied by clear and unambiguous procedural guidelines. This last point cannot be emphasized enough since the calculation of indices in typological work crucially depends on the application of uniform parameters.

2.2 What dictionaries of linguistics tell us Recent dictionaries of linguistics seem to be at great pains to define the word. While linguistic units such as the phoneme or morpheme give rise to little dis-

30

Analyticity and Syntheticity

agreement, the same cannot be said of the word. Not atypical are the following observations: 'there are several difficulties at arriving at a consistent use of the term...' (Crystal 1980:383) 'en linguistique structurale, la notion de mot est souvent évitée en raison de son manque de rigueur' (Dubois 1973:327) 'the word as a linguistic unit is ... difficult to define' (Hartmann 1972:256) Comments such as these are typically followed by brief bibliographic references to authors known for having contributed a "solution" to the problem. Among the dictionaries of linguistics I have consulted (Ducrot and Todorov [1979], Heupel [1978], Dubois et al. [1973], Pei [1954], Marouzeau [1943], etc.), those of recent vintage all refrain from citing a standard definition and are essentially of little use.2 2.3 Word criteria 3 Numerous linguists are pessimistic about the possibility of defining the word: the word is either not a universal unit at all (Milewski [1951]) or else must be defined strictly according to intra-lingual, language-particular criteria (Heupel [1978], Krámsky [1969], Bazell [1957], Entwistle [1953]). Since we are interested in a universal definition requiring no complementary, language particular qualifications (all other definitions being inherently useless for typological work), special attention will be paid to those that have proposed universally valid criteria.4 In the past fifty years there has been much eagerness to find such a universally valid definition. In the extensive literature on the subject, one can distinguish a number of criteria which have repeatedly been suggested and which may conveniently be subsumed under the following headings: (1) separability and displaceability,^ (2) isolatedness, (3) the phonetic criterion (acoustic identity of word), (4) the cohesion of the word, and (5) potential pause.6 23.1 The criteria of separability and displaceability Separability and displaceability are closely related but may exist independent of each other. As I hope to show, displaceability is (at least in a number of languages including the Romance group) the most unlikely candidate for being a valid criteria for word separation.

Word delimitation 31 23.1.1 The criterion of displaceability Although in certain languages (e.g., so-called free word-order languages) syntactic displacement of words may not change the basic prepositional meaning of words or word groups (cf. LaL pater filium videt 'father [nom.] son [accus.] sees' vs.filium pater videt 'son [accus.] father [nom.] sees' = "the father sees the son'), 8 this kind of movement is not possible in numerous others, and cannot, therefore, qualify as universally valid criterion for word division. This is especially true of languages in which morphological functions are expressed syntactically and not morphologically; thus displaceability is, for instance, excluded as a possible criterion from Romance and Germanic languages where word position alone can be decisive: le chat mange le souris vs. le souris mange le chat /die Katze frißt die Maus vs. die Maus frißt die Katze 'the cat eats the mouse' vs. 'the mouse eats the cat.' Further reasons for not regarding displaceability as a reliable criterion for word division come to light when one considers that the combinatorial capacity of words is often very limited. Such a limitation is apparent in Romance languages, where word position not only expresses relationships which in other languages are expressed by case inflections, but may also be a factor determining the meanings of individual words in context Thus Krámsky reminds us that 'in French, the criterion of displaceability is not valid in the sequence les braves gens "good people," because when we displace the words to form les gens braves we obtain another meaning: "brave people"' (1969:23). Numerous other such examples could be cited from all other Romance languages. In sum, displaceability, although a factor for word division in some instances and in some languages, cannot stand as a universal diagnostic. 23.1.2 The criterion of separability Separability (also referred to as interruptability) has traditionally been regarded as a major word criterion for a number of languages, particularly French. According to Zaretski (1929), a word may be identified as follows:' A word is a part of speech which can be separated from the surrounding parts and cannot itself be divided. If it can be divided it is more than a word; if it cannot be separated, it is less than a word, (cited in Krámsky [1969:27]) Although less open to criticism than displaceability, separability gives rise to reservations of a different sort. The very notion of "word = inseparable whole" in some instances leads to an unacceptable word analysis.10 Consider the CL verb amamus;

32

Aiudyticity and Syntheticity

due to its interruptability (cf. ama-vi-mus, ama-ba-mus, ama-vera-mus), amawould end up constituting a full fledged word; moreover, anything following the stem would be left as a "residue" which could be assigned neither to the stem nor to the following word. Identical problems arise in French: in (j')aim-ais, the possible intrusion of an additional morpheme (e.g., conditional -r-: aim-er-ais) cuts the "word" somewhere at midpoint, a result few, if any, would be inclined to accept.11 Scholars like André Martinet, who earlier in his career regarded inseparability as most reliable yardstick for word separation, have now recognized the problem in severing the unity of phoneme sequences which should remain together. In an attempt to bypass the problem, Martinet (1949:293) contended that 'dorm-ons reste un mot, bien qu'on trouve également dorm-i-ons, car le signe -/- du subjonctif présent et des divers imparfaits (panni lesquels le conditionnel) n'existe que dans des combinaisons de ce type, et toujours dans cette même position.' He then goes on to claim that je fais constitutes two words because one can say je le fais. Martinet never makes clear, however, why the "infix" le deserves a status different from that of -i- in dormions,1^ Application of the criterion of separability has, at times, led to debate about the interpretation of certain diachronic developments in the French verb phrase. Bally and Dauzat, among others,13 have suggested that some French "pronominal" forms (il as in ii aime/ils aiment) had become, or were on the verge of becoming bound morphemes (ilèm/izèm) in structures not unlike those observed in polysynthetic Amerindian languages. Jespersen, basing himself on the criteria of separability (e.g., tu m'aimes) and displaceability (m'aimes-tu), challenged this view claiming that in French we are indeed dealing with separate words. In Jespersen's view, there is a fundamental difference between the French forms cited here and say, CL amo in that 'we never find anything placed between am and o in the first person' (1964 [1922]:423). The contrast he draws between Fr il aime and CL amat is unfortunate, however, in that while it is true that je can be separated from aime (je t'aime) it is not strictly true that 'we never find anything placed between am and o in the first person, amo' (R. Harris 1972:123). As pointed out previously, in Latin a variety of elements (e.g., tense-mood markers) can be inserted between the stem and the person-morpheme. 1 4 But the merits of this issue need not concern us further. Whatever analysis one adopts of the diachronic events, the main lesson to be learned from the discussion between Jespersen and his contenders is this: as long as those favoring the criterion of separability fail to indicate—and to agree on—what qualifies and what does not qualify as "a separator of words," the deliberation will continue to be fruitless. Switching our attention from French to Portuguese, we come across evidence which further undermines the validity of the criterion of separation. It is well

Word delimitation

33

known that in written Peninsular Portuguese the person morpheme future verb forms such as far-ei may still be separated from the stem by the insertion of object clitics (fa-lo-ei). According to the hypothesis at hand, fa- would thus constitute a word, while/ara would not, a view which is difficult to uphold.1^ Perhaps the strongest evidence against the interruptability criterion comes from German. In that language, verbal prefixes (e.g., überein- in übereinstimmen 'to concord, to agree'), inseparable in the infinitive, are separated and moved to the end of the sentence in a variety of finite forms. Thus, according to the exigencies of separability, in Ich stimme mit deinem Vater nicht überein 41 do not agree with your father,' stimmen and überein would have to be classed as separate word entities. Since überein- is a bound form and, consequently, incapable of constituting a word on its own, the resulting word segmentation is clearly faulty. Discussing the difficulties that arise in analyzing words whose whole, in a variety of cases, becomes separated into widely dislocated segments, Porzig remarks: Nun ist es eigentlich nicht einzusehen, was es für die Einheit Lautung/Inhalt, also für das Wort, ausmachen soll, wenn die Lautung auf zwei Stellen des Satzes verteilt ist. Zum einheitlichen Sinn des Satzes wirken ja ebenfalls alle Glieder zusammen, obwohl sie an verschiedenen Stellen des zeitlichen Verlaufs der Lautung auftreten. Aber hierin liegt gerade die Schwierigkeit. Wie soll man unterscheiden , was Wirkung der Satzeinheit und was Wirkung der Worteinheit ist? (1959:164)

Porzig's comment is well taken and underlines the fact that in a number of languages word segmentation relying on the criterion of interruptability leads to unacceptable results, thus rendering it inadequate as a universal determinant for word analysis. 23.2 The criterion of isolatedness Under this heading falls Bloomfield's well-known definition of the word as minimal free form. In Language we read that 'a word, then, is a free form which does not consist entirely of (two or more) lesser free forms; in brief, a word is a minimum free form' (1963 [1933]:178).16 Unfortunately this definition—despite Bloomfield's intent to bring it somewhat in line with that of our ordinary notion of the word—also has implications not in accord with the "natural," traditional notion. 17 As Bloomfield himself recognizes (p. 181), his criterion cannot be strictly applied: many forms lie on the border-line between bound forms and words, or between words and phrases (cf. for instance Eng the, a, is, was, which are not minimum free

34

Analyticity and Syntheticity

forms, and therefore not words); it is, he says, impossible the make a rigid distinction between forms that may and forms that may not be spoken in absolute position. 18 Some of the outcomes of such borderline divisions are also pointed out by K. Togeby: Les conséquences de la définition de forme libre minimum sont intéressantes. C'est par exemple en vertu de ce critérium que les linguistes américains ne veulent considérer Γ s de génétif de l'anglais comme un mot dans des constructions telles que the king of Englands futí. De même les articles français et certaines prépositions françaises comme de, à et en ne seraient pas des mots. (1949:105)

One interesting aspect of the definition at issue is that French subject and object clitics (as well as preveibal ne) become analyzed as affixes (or, in the case of object clitics, infixes) forming a unit with the verb. The results of such an analysis are particularly important in terms of the calculation of syntheticity. If subject clitics, object clitics, and possibly the negative particle ne are counted as grammatical morphemes in single word complexes, then the French verb phrase would have to be regarded as highly synthetic. 19 Given that scholars have long advocated regarding French "pronouns" (in particular the proclitic subject pronouns) as fully verb-bound (i.e., word-internal) morphemes, an analysis based on the Bloomfieldian word criterion would indeed be attractive. Bloomfield could well have used these arguments in support of his definition of the word. This, however, is not the case. In fact, in one passage he challenges the results of his own criterion stating that 'the conjunct forms [i.e., je, tu, il, etc.], largely because of their parallelism with the absolute forms [i.e., moi, toi, lui, etc.] have the status of words' (1963 [1933]: 179-180). Part of the problem with Bloomfield's word criterion is his own rather inconsistent handling of the term "word." As R. Harris (1972:128) observes, Bloomfield fails to stick to a rigorous use of his own terminology in the passage where he describes au in au roi as a case of 'a single phoneme representing two words' (p. 179). What is striking, and left entirely unexplained, is that according to Bloomfield's own analysis neither au nor what he calls "the two words à and le" could qualify as minimum free forms, nor could he—as in the case of the subject clitics mentioned above—plausibly treat them as "parallel forms." A difficulty of a different sort arises when we consider cases in which a clitic, e.g., Fr vous (in vous allez), is flanked by a lexical homophone (e.g., c'est vous). Thus, according to the criterion of minimum free form, vous allez represents two

Word delimitation

35

words (both vous and allez being minimum free forms [VOUJ as the disjunctive pronoun, allez! as an imperative]), whereas je vais counts as only one (both je and vais failing to qualify as free forms). Few would argue that the resulting word status of the clitics vous and je is entirely arbitrary and unjustifiable. This observation points up at least one advantage of the criterion of separability over that of isolatedness: the former acknowledges not only the parity between vous allez and je vais (both qualifying as two separate words) but also the diachronic facts of French syntax, i.e., the various changes which word order has brought about in the restrictions of the number of possible intercalations.21 Linguists following Bloomfield could not help but notice that the criterion of minimal free form fails to apply to an entire segment of vocabulary, i.e., synsemantics (= grammatical words [typically clitics, auxiliaries, and prepositions] which enter into the sentence only in connection with other words).22 Recognizing this shortcoming, Bloch and Trag«1 proposed a new, less stringent word definition meant to include both auto- and synsemantics: 'Any fraction that can be spoken alone with meaning in normal speech' (1942:54). Unfortunately, these authors failed to make clear what exactly they meant by "meaning" (does it include all grammatical meaning, i.e., all of the synsemantics?) and what was to be considered "normal" speech. Consequently, their definition falls short of being operational even in the best of senses and cannot be considered an improvement over Bloomfield's concept of the word. Finally, the criterion of "minimum free form" calls for an observation of an entirely different sort which, while not invalidating it, nevertheless raises doubts about its value as cross-language word determiner. If occurrence in absolute position is to be our test, numerous French "words" are somehow less complete words than their Spanish or English counterparts. If, e.g., the answer to qu'est-ce que tu cherches? is du pain, and if/arrive is the answer toluenos pour longtemps? neither pain (but du pain) nor arrive (but j'arrive) qualify as free forms (R. Harris 1972:111-118). 23.3 The phonetic criterion If criteria such as those of isolatedness, displaceability, or separability cannot guarantee the identity of the grammatical word, then are there phonetic criteria, we may ask, which can constitute a substantial basis for word division of both auto- and synsemantics? The question of the acoustic identity of the word is one of the most discussed issues in linguistics, and the attention it has received is most likely due to the complexity and difficulty of tackling it in a simple, uniform way. While we have grown accustomed to apprehend written words as clear-cut units within the

36

Analyticity and Syntheticity

phrase or sentence, phonic word identity within SPOKEN discourse is not consistently marked in a number of languages. The reasons for this lack of observable word juncture are manifold and differ from language to language; greatly simplifying, the issue may be summed up by stating that in many cases potential phonological demarcators such as stress, accent, rhythm, etc. do not coincide with "logical" semantic word boundaries.24 While this is not the case, for example, in either Czech or Hungarian, which have initial stress as part of their word-pattems, 25 in Romance, on the contrary, phonemic and morphemic segments need not be congruous. 26 French is particularly notorious for not having consistent phonic word delimitations within the phrase or sentence.27 In French 'whatever congruence there is between phonological and grammatical structure seems to hold over units of higher rank than the word' (Lyons 1968:205).28 Thus, if the French breath group (= utterance between pauses) is longer than a single lexical word, the lexical words composing it are not bounded from one another by any phonological signals, either segmental or suprasegmental (Pulgram 1967:1639). As a result of the lack of grammatical and phonological congruence, the following often cited (albeit bizarre) sentences are phonetically identical though quite distinct lexemically and semantically: 29

Galle, amant Galamment

de la reine, de l'arène

alla, a la

tour Tour

magnanime. Magne à Nîmes.

'Galle, the lover of the queen marched—a noble feat— Gallantly from the arena to the Tour Magne at Nîmes.' (Pulgram 1970:86) Other phonetic features, such as syllabification, can actually lead to the dissembling of word structure. Thus in French and Spanish, the rules of syllabification act independently of word limits and frequently lead to a violation of "natural*' boundaries: un buen autor = un-bue-nau-tor, mis árboles han crecido = mi-sar-bo-leV) san-cre-ci- do. There are numerous other knotty problems with the acoustic identity of the word. The mention of just one of these should further underscore the impossibility of formalizing word identity on the basis of acoustic features.31 We all know that words can be phonetically distorted to the point of incomprehensibility. However, no objective criterion can be established to determine when exactly a word is, or is no longer, recognizable as itself. Comprehensibility is, in other words, purely subjective and differs from speaker to speaker; while one person will understand a

Word delimitation

37

word even in highly distorted form, someone else may fail to identify it with less distortion. In this section I have limited my discussion of acoustic word criteria to a few points. 3 2 Before dismissing the phonetic criteria, a final, and perhaps not unimportant, point needs to be made. Even if no objections could be raised to basing the delimitation of words on phonic criteria, an inconsistency in methodology would nevertheless force us to discard the idea of invoking suprasegmental features to distinguish words. It is wellknown that for computing indices of synthesis, (the often very "meaningful") suprasegmental elements have never been included in the morpheme count. 33 Obviously, making phonological features the criterion for word boundaries while at the same time excluding this criterion from the morpheme count, would be methodologically unacceptable. Whence it follows that unless suprasegmental features can somehow be incorporated into morpheme analysis, phonological criteria—at least when applied in the index of syntheticity—must never be included in the definition of the word. 23.4 Word cohesion The criterion of cohesion is closely related to that of separability in that it takes as its basic premise that of separability. There is an important difference, however, in that cohesion is a less rigid yardstick by which to measure the unity of a word. Unlike separability (which is "absolute" in that a morpheme sequence either is or is not separable), cohesion provides a sliding scale by which to distinguish degrees of connectedness between two or more segments. Underestimating the importance of cohesion has, in some cases, led to oversimplifications. Thus, according to Mikus (19S7), there is no essential difference in the relationship obtaining between the components of Lat cant-o, on the one hand, and those of Eng I sing and Fr je chante, on the other. Mikus further argues that in each category, the only essential fact is that of a relationship between a deteiminandum (sing, chante, can-) and a determinatimi (I, je, -o). Most linguists, however, would presumably argue for a fundamental difference between the degree of possible interruption between Eng / + VERB and Fr je + VERB (the details of which will be discussed in chapter 4); 3 4 since the distinction between analyticity and syntheticity revolves around the degree of tightness among morphological elements of language, then clearly the difference in the cohesion factor between, say, the Peninsular Portuguese example dar-ei cited above (interruptible by object clitics only, e.g. dar-VOS-ei Ί will give YOU') and that of Eng I will sing (interruptible by numerous forms, e.g. I will perhaps never again sing) is significant and should be integrated into an analysis of word separa-

38

Analyticity and Syntheticity

tion and, consequently, also into an index of syntheticity. It should be borne in mind that the degree of integrity or cohesion of the morphological elements of the word differs from language to language. As a result, the criterion of cohesion (rather than simply separation) may be more critical in some languages than in others. Cohesion is greatest in inflected languages where the unity of the word is not, as in many agglutinative languages, maintained specifically by morphological signs. 35 It should also be pointed out that in the extensive literature on word criteria, cohesion has been used in a sense different from that intended here. Cohesion generally refers to the relative degree of interruptability between segments (as in Krámsky 1969). This criterion of tightness is largely independent of semantic considerations. 36 Other linguists (e.g. Zirmunskij 1969) have used the term to designate the semantic bond, i.e., psychological compactness, between two or more segments. Used in this sense, cohesion has been involved in deciding the word status of certain compounds which are extremely difficult to analyze. Historically, compounds evolve from separate lexical words which, through frequent collocation, a strong semantic association, and non-displaceability, become so closely bonded that they eventually merge into a single inseparable segment (i.e., a single "word"). The difficulty in deciding whether a given compound is to be analyzed as one or two words stems from the fact that the two segments may often still stand alone as minimal free forms, which may lead those operating with the criterion of interruptability to classify what has become a compound still as two separate entities. This is particularly true when the seam between the words has not yet become blurred due to phonetic reduction or to morphological regrouping. To cite just one example, according to the traditional criterion of separability, Eng airport must be classed as two words (both air and port occur independently), an analysis many would reject. Taking into account the semantic unity of the group, we would, of course, want to argue that the cohesion is complete and that airport is, accordingly, a single word. 38 Even though one might agree with Krámsky's claim that 'the varying cohesion of word elements is one of the basic factors influencing not only the phonological but also the morphological and syntactical structure of language, and is, consequently, one of the important typological factors as well* (1969:40), the criterion of cohesion may nevertheless have to be discarded as a universal determiner fa' word delimitation. The reason leading to this rejection of cohesion is that linguists have not established a managed barometer for the criterion. If the degree of tightness is to be taken into account in morphological typology, we should, in other words, have a standard of measurement whose extremes are well determined so that the various -3ft degrees of cohesion may be quantified and conveniently placed along its axis.

Word delimitation

39

How difficult, nay impossible, it is to put this concept into practice becomes apparent, however, when one actually attempts to devise such a system. Is a "loose" cohesion one wherein several words of different lexical classes (i.e., pronouns, adverbs, nouns, etc.) can be inserted between the two segments in question? Or should one measure tightness by simply adding up the number of possible intercalations regardless of lexical class? Does distance between the segments, i.e., the possibility of inserting more than one element, outweigh strictly numerical factors? 40 To demonstrate the difficulty involved in formally measuring cohesion, let us briefly lode at some diachronic changes that have taken place between old and modem French.41 There have been various changes in French word-order, particularly in the verb phrase, which might be viewed as instantiating a trend toward reducing freedom of intercalation. For example, in the 12th and 13th centuries, it was possible to intercalate various elements between je [1 s. subj. pron.] + voi ['see,' 1 s.], including a direct object noun phrase (e.g. quant je la demoisele voi 'when I the maiden see')· As late as the 16th century, it was still permissible to separate je from its verb by a relative clause or appositional phrase (e.g. je, qui vous fais ces tant véritables contes, m'estois caché ... Ί , who tell to you [lit. make for you] these entirely true tales, had hidden myself ...'). 4 2 These constructions are no longer possible in Modern French, and in speech the omission of the subject clitic as in je [1 s. subj. pron.] + m' [1 s. refi, pron.] + arrête ['stop'] + et ['and'] + descends ['go down' + 1 s.] (fa' ...et JE descends) is increasingly rare. All these developments point to an increase in cohesion due to increased restrictions on intercalation. Several investigators (Tesnière [1932]; Bally 43 [1965] (1932); Kayne [1975]; M. B. Harris [1978b]; [1980]; Ashby [1977]) have interpreted these changes as evidence in support of the claim that the VP in French has to a large degree become synthetic. While the above developments seem to point to a more synthetic VP, at least one diachronic change involving the VP seems to go against this apparent trend toward tighter cohesion: Modern French requires the presence of both the direct and indirect object clitics between the proclitic subject and the verb (je le lui donne Ί give it to him' (lit- Ί it to him give', but *je lui donne), a restriction which did not apply to Old French (je li done). If we base our criterion of cohesion on the "distance" factor, then it appears not only that the trend towards a synthetic verb construct has been reversed in at least one instance, but also that with the linguistic tools presently available, it would be impossible to formalize objectively— either in numerical or in non-numerical terms—the precise difference in the degree of cohesion between the Old and Modern French verb phrase. 44 While serious in itself, the impracticality of the criterion is not the only reason

40

Analyticity and Syntheticity

for denying its usefulness in deciding word delimitation. "Cohesion" suffers from all the drawbacks mentioned for the criterion of separability: although "cohesion" is an improvement over "separability" in that it differentiates degrees of tightness, it nonetheless fails to help us decide cases of questionable word division such as Fr je le fais Ί do/make it' vs. faites-le! 'do/make it!', Port farei Ί will do/make' vs./βίο-«' Ί will do/make it,' Ger übereinstimmen 'to agree with' vs. ich stimme nicht überein, Ί do not agree with (it),' or air-port. Because its basic premise is that of interruptability, cohesion carries with it the difficulties of "abnormal" wordseparation discussed above under interruptability (§2.3.1.2) 235 The criterion of potential pause Words may be viewed as segments in sequences 'separable by means of pauses' (Jakobson and Halle 1956:20) or 'bounded by successive points at which pausing is possible' (Hockett 1958:167). One could, of course, object immediately that speakers normally do not pause between "words" (Lyons 1968:199) or that it would be impossible to determine empirically those junctures at which all speakers separate segments. But as R. Harris points out, raising such objections may be pointless 'since the criterion appeals not to what pauses do in fact occur, but to what pauses could occur' (1972:121). The difficulty is that for the criterion to be operational, not just any pause will do. We would, for instance, want to exclude "hesitations" (Nida 1949:86 nl5), which clearly do not mark word boundaries. But how then do we identify those, and only those pauses which mark word boundaries? One approach might be to limit "potential pauses" to those which occur only at word junctures, in which case the argument would become circular. As it turns out, no one has come up with a definition of "word pause" which is not itself based on the argument of "potential pause," whence this criterion falls short of providing procedural guidelines for determining word units. Yet another difficulty with "pause" is that in some languages we find instances where, for practical purposes, no pause can be inserted between what would—at least under the criterion of separability—have to be regarded as separate words. Thus the spoken French sentence tu t'y prends mal 'you are doing it wrong' (lit. 'you yourself to it apply poorly') may potentially be divided up into tu / t'y / prends mal, but not between—what most would want to consider—the two words (' and y. 23.6 The "word' within Heger's hierarchy of linguistic units Heger's study (1971) falls outside the main body of traditional attempts at defining the word in that it seeks a solution to the problem by establishing in a rigorously hierarchical ranking of all speech units (from the smallest unit of meaning, the moneme, to entire composita such as literary texts).45 The book analyzes in depth

Word delimitation

41

the linguistic concepts "moneme," "word," and "sentence," with the moneme at the bottom of the scale ("R[ank]l"), followed by the minimal autosememic unit (R2), inflectional forms (R3), compound forms (R4), etc. CL laud- (root of the verb 'praise') corresponds to rank 1, i.e., a "moneme" in the sense of a lexeme not containing an 'exclusively reflexive-metalinguistic' sememe.46 Laud- is elevated to the next higher rank (R2) through the addition of -β-, a lexeme-bound moneme with an exclusively reflexive metalinguistic sememe (a conjugation class marker). The 'autosememic minimal unit' laud-a- turns into a unit of the third rank (R3) when lexeme-bound 'grammemes,' including bound monemes with exclusively reflexive metalinguistic sememes, are added. One thus obtains the 'inflectional forms' (Flexionsformen) laud-a-ba-n-t [root-conjugation class-imperfect and T/A-pl.-3d pers.], laud-a-ve-ra-n-t [root-conjugation class-imperfect and T/A-pl.-3d pers.], lauda-vi-sse-n-t [root- conjugation class-imperfect and T/A-pl.-3d pers.]. Following this scheme, Heger arrives at as many as seventeen levels (see his table, p. 233). Within this complex architecture, level three is of particular interest for it is here that Heger places the linguistic units called "words." As he points out (pp. 59ff) the application of his method yields results not always consonant with traditional notions of the word, e.g. a sentence of the type Fr il la lui a donnée 'he gave it to him/her' represents for him a single word. In the past two decades such an analysis has gained increased acceptance (see chaps. 4-5 below), and may indeed be more in tune with the linguistic facts of modem French. Heger's main concern is not to create a slot for "words" in his overall hierarchy of linguistic units, but rather to establish a coherent system in which all elements of speech have their definitive place. The fact that one of these slots (R3) houses a large portion of what are generally considered "words" merely offers a convenient label for that level of the hierarchy (all other ranks being labelled as well [see p. 233]). Heger underlines that the terminology thus introduced is not intended to settle the dispute about what is or is not a "word": 47 Die hiermit eingeführte Definition von Wort und Vokabel erheben selbstverständlich nicht den Anspruch, eine Antwort auf die untere Bezugnahme auf den vorwissenschaftlichen Gebrauch dieser Termini gestellte Frage "was ist ein Wort?" zu geben. Vielmehr beruhen sie darauf, daß sich innerhalb eines mit dem Bemühen der Kohärenz aufgebauten Systems Stellen ergeben, für deren Benennung sich auf Grund zahlreicher Koinzidenzen und analoger terminologischer Festsetzungen die Termini "Wort" und "Vokabel" anbieten. Daraus folgt, daß es bei Zugrundelegung der hier eingeführten Definitionen in mehr oder minder häufigen Einzelfällen zu Subsumptionen kommen wird, die von dem vorwissenschaftlichen meta-

42

Analyticity and Syntheticity sprachlichen Gebrauch der beiden Termini divergieren. Wem diese Divergenzen so groß erscheinen, daß er die hier vorgenommenen terminologischen Festsetzungen nicht akzeptieren zu können glaubt, dem bleibt es unbenommen, für die Einheiten ERJ· [Vokabel] und ER¡ (Parole) bessere Termini vorzuschlagen. (61, emph. mine)

Given that Heger emphasizes the coherence of his overall system rather than the synchronization of each rank with conventional analyses of linguistic units, it comes as no surprise that some elements which should clearly count as "words"—at least when calculating indices of syntheticity—are classed outside R3, thus putting into question the usefulness of Heger's word analysis for our purpose. Consider, for instance, the compounds Haustier 'home animal = pet,' Hausbesitzer 'home owner,' listed as "Kompositionsformen" in rank 4. 48 Heger's classification fails to provide a satisfactory working definition of the "word" on yet another account. In the transition from an analytic to a synthetic unit, there generally occurs a stage during which one of two (or several) amalgamating elements begins to lose its autonomy. As shown in chaps. 3-7 below, during the initial stages of such a development, synthesis is frequently limited to the semantic plane, with phonological and morphosyntactic changes following later. Examples of such structures abound within all of modern Romance: Sp claro que si, Port claro que sim, Sp desde luego 'of course,' Fr à peu près. It più o meno 'more or less, approximately' have all begun to synthesize (they all form semantic wholes), yet there are no "obvious" formal signs which would betray their current evolution toward units of different sort, i.e., single words. As already mentioned, any method which purports to use analyticity and syntheticity as a measure for typological quantification must provide formal means by which the word status of structures can be decided. Despite the unusually systematic and lucid conceptualization of Heger's scheme, it fails in the case of "transitional" units such as the ones just cited. There are no procedural guidelines for settling these knotty cases, nor does it seem possible that refinements to his analysis could show that expressions like more or less (for 'approximately') differ fundamentally from syntagmatically identical good or bad, white or black, etc.49

2.4 Segmenting "un enfant terrible" In the preceding sections, we have looked at some of the criteria for defining the word which have been put forth since the beginning of this century. Our approach has been to outline each proposal in some detail and to cite examples which reveal

Word delimitation

43

the problems inherent in appealing to that criterion in order to arrive at a universally valid word definition. Taking a single speech segment from a single language (French), in this section we will try to arrive at a valid word analysis by applying the various criteria. The point is to demonstrate that the difficulties in finding a definition do not stem strictly from the fact that a given criterion may be applicable in some but not in other languages, but that even in a single language some segments of speech simply cannot be broken down exhaustively into those smaller units called "words." Take the phrase des enfants terribles. The spelling conventions suggest three separate words. But if we apply phonic criteria, the word analysis will differ. If we take stress as the deciding factor, the phrase turns out to be a phonological "word" with stress on ri: [dezäfäte'rib(l9)\?Q Since this phonologically determined "word" seems to be at odds with our conventional notion of a word, we may proceed to other, non-phonic criteria. "Displaceability" will assign word status to enfants and terribles (cf. des enfants terribles vs. de terribles enfantsand lead to (the discomforting) conclusion that, because it is displaceable, des must be considered a detachable prefix of the "word" desenfants (cf. des terribles...). The interruptability test, on the other hand, will give results more in tune with our intuitive feeling about word juncture. 32 Thus our phrase is made up of three separable words (DE petits ENFANTS si TERRIBLES; or, more colloquially, DES petits ENFANTS si TERRIBLES 'such terribles little children' [lit. 'little children so terrible']). While there would be little disagreement about the word status of enfants and terribles, de(s) is more problematic. One could argue for the word status of de(s) in that it is not only separable but (under certain special circumstances) possibly also a minimum free form. Let us assume for now that we are willing to grant validity to this claim. But how are we to proceed if the phrase is slightly altered to de l'enfant terrible 'of the terrible child'? Should we still accept, as the criterion of separability now leads us to conclude, that the two inseparable elements de and Γ somehow deserve word status much like des did previously? Do we really want to push our analysis that far? We could, of course, appeal to the Bloomfieldian criterion of "minimum free form," but only to find that de Γ, failing to qualify as an independent word, would end up as unassignable "residue." As noted at the beginning of this chapter, one of the underlying premises of finding a universal word criterion is the segmentability of an entire stretch of discourse into chains of individual words, which is tantamount to saying that our last analysis oî de Γ as something other than a full word is simply not satisfactory. Equally unsatisfactory is the claim that there is really nothing wrong with considering del' asan affixed part of the word "de l'enfant," comparable in form to CL infant-is. This last option is untenable largely because of its inter-

44

Analyticity and Syntheticity

ratability (de Γ AUTRE enfant) and relatively "loose" cohesion (cf. DE L'autre grand et fort ENFANT) which make de l'enfant an intuitively different sort of "word" than the uninterruptable infantis. In effect, this example demonstrates that unless we are willing to sacrifice the notion that all parts of a given speech segment must be segmentable into word units, we will always end up with "residue" (particularly clitic items) which are neither full words nor, in the strict sense, prefixed inflections.

2 . 5 Why speakers have an intuitive "feeling" for word boundaries If it is the case that in some languages "words" cannot be defined on the basis of a single criterion, then how is it that speakers of all languages nevertheless seem to have a deep-rooted sense of such units? The answer is that there exists an array of language-particular, often unrelated factors which contribute to making the word the basic unit of language. Among these factors must be grouped those criteria which, as we have had occasion to point out, some linguists believe to be universal determiners of wordhood: isolateability, intenruptability, displaceability, potential phonological markers, and cohesion. But there are also other phenomena which significantly contribute to—what has been claimed to be—a universal word consciousness. These include: versatility of occurrence (boys may be subject, object, etc., whereas -s "plural" occurs only as a suffix), special syntactic relationships (word-order patterns which are not found in phrases may occur within words [cf. the outcast vs. to cast out]), restrictions on phoneme sequence or allophonic occurrences or morphophonemic rules, analogy of doubtful items with items more or less certain (the is more or less non-isolatable, analogous to isolatable this, or -m in I'm here is analogous to am), 53 as well as psychological association (a certain sound complex is felt as one complex even though other word criteria, especially "separation" or "minimum free form," would lead to contrary results; this is particularly the case for compounds [cf. shoe string, Fr pot au feu] and dislocated morphemes [cf. Germ Ich STIMME mit deinem Vater nicht ÜBEREIN]). Since the nature and range of these factors are language particular (some may, depending on the language, never apply, while others, e.g., Turkish vowel harmony, may always be a decisive indicator of word boundaries), speakers of different languages obviously do not base their feelings about word units on the same criteria. Each language has its own set of special "rules," and it would be a remarkable coincidence if it were possible to find general criteria which lead to the same results in several languages. Given these interlingual differences in word criteria, 'the units which pass under the title of "word" in different languages are not exactly the same

Word delimitation

45

sort of unit, though they are similar' (Bazcll 1957:28). The "universal criteria" will, therefore, never suffice. In fact, far from sufficing for all languages, they often do not suffice even for any one language. Even within a language, there may not be a single sufficient criterion of word unity. In one instance the semantic bond between two segments (e.g. pomme de terre 'potato') may be the only recognized criterion of word-status, whereas in another, purely formal, grammatical factors (e.g. separability) may dictate word juncture.54 It would be wrong, however, to believe that all of the linguist's uneasiness with the definition of the word stems from the observed complex interplay and variability of the criteria. Because the inventory of phonemes and phrase structures is limited, we have been driven to the ideal of total accountability. In this quest, we may forget, however, that the assumption that all sentences must be segmentable exhaustively into lower-level units called "words," is an α priori assumption set up solely for the purpose of establishing an ideal theoretical framework. 55 The fact that our criteria invariably fail to account for what I have referred to as "residue" does not invalidate the word status of those segments which clearly are full fledged words. It merely means, as has been said for other linguistic categories, that 'the classes (and subclasses) of elements should not be regarded as "boxes" with clear-cut boundaries but as formations with a compact core (center) and with a gradual transition into a diffuse periphery which, again, gradually passes (infiltrates) into the peripheral domain of the next category ...' (DaneS 1966:12).56

2 . 6 The lack of a universal word definition and its consequences for the concept of syntheticity After generations of 19th- and 20th-century linguists had taken the "word" largely for granted, structuralists set out to define what in popular as well as scientific circles was regarded as the basic unit of speech. In their search for a satisfactory proposal, various unexpected difficulties surfaced. Investigators; cheerfully proposed this and that, readily changing course at each failure. As we have shown, this lively discussion eventually led to the now generally accepted conclusion that (a) the "word" cannot be defined by a single (or for that matter, multiple), common denominator, and (b) not all segments of speech are "words" in the proper sense of the term. Both intellectual passion and the commitment vested in particular ideas (especially that of syntheticity) born long before the general recognition of the "word's" resistance to definition, has, however, led many to avoid the undesirable confrontation with the "facts." Numerous linguists, and in particular those concerned with the concept of syntheticity, continue to use old methods which take the "word" as

46

Analyticity and Syntheticity

the base unit. The reason for their reluctance to accept the inevitable is easy to perceive: admitting the impossibility of defining the "word" in a scientific manner leads directly to the collapse of part of what has long been hailed as a promising avenue for research, i.e., morphological typology. The central question, therefore, boils down to this: do we really want to continue operating with an index as well as a concept of syntheticity whose fundamental assumptions are ill-defined, undefined, or simply undefinable? In a method which supposedly has universal application, do we really wish to keep plugging in formulas (e.g. WORD/MORPHEME = index of synthesis) in which one of the parameters lacks universal extension? Given what we know today about the shakiness of the "word," might not analyticity and syntheticity, as currently used, be simply a terminological subterfuge which leads us to believe that there really exists a rigorously based, empirically determined conceptual opposition suitable to describe the morphological make-up of language(s)? The answer to these questions is clear. Unfortunately, the impossibility of coming up with an adequate word criterion leads to the collapse of not just part but the whole concept of an index of syntheticity. Even if Hockett and others are right in thinking that the calculation of the synthetic index 'is the easiest rough measure of morphological complexity in a language' (1958:181), they still must concede that either there is, oddly enough, no definitive procedure of measuring the morphological complexity of some languages (e.g. French), or else that in some instances the index ought to be calculated in a special way. Since this presumably defeats the purpose of typological comparison, there seems to be no definitive solution to the dilemma. An exception-free working definition of the word is so basic to finding a method of calculating the index of syntheticity that only those future proposals which can offer a universally valid word criterion will deserve attention. Until a solution is found, the idea of operating with a quantitatively based typology of analytic and synthetic languages (or structures) ought to be laid to rest. Given the many years of painstaking efforts invested in the concepts of analyticity and syntheticity, this conclusion is certainly a very somber one. To offset the general tone of pessimism I can offer nothing more uplifting than a quote from Roger Lass who reminds us that it is a commonly held, and erroneous and counter-productive belief, that in order to be justified in criticizing something, you have to have something better to substitute for i t . . . If a totally negative assessment causes the babies to be thrown out with the bath-water, this is no great loss; if the babies to be thrown out are really worth keeping and raising to maturity, someone is bound to fish them out again. (1980:4)

Chapter three Fishing out the baby: The usefulness of the terms analyticity/syntheticity in diachronic description

3.1 Introduction A prime reason for sustained interest in the question of synthesis/analysis is that linguists hope to use the results of the analytic/synthetic gradient scale to formulate inductive generalizations about language—in particular generalizations which will prove useful in typological studies. As shown in the previous chapter, the failure to arrive at both a satisfactory method of quantification and a clear delineation of linguistic units (such as the word) has undermined attempts to classify languages according to their morphological "tightness" and, as a result, has vitiated meaningful typological inferences based on syntheticity and/or analyticity. The fact that endeavors to classify languages according to their morphological structure have to a degree been frustrated should not, however, lead us to neglect the potential usefulness of, nor the wealth of insights that can be gleaned from, the analytic/synthetic concept in areas other than language typology. As I hope to demonstrate in the following chapters, the history of Romance, as well as that of other languages, attests to recurrent and often unrelated diachronic events whose overall effect on morphosyntax cannot be grasped adequately without the notion of syntheticity and/or analyticity. Important differences between the present approach and earlier ones will emerge in the course of our discussion. Thus it will be argued that for the parameters analytic and synthetic to remain meaningful: (1) their use must be limited to speech units rather than entire languages,1 (2) they must be understood not as a quantifiable absolute but as the rough measure of the overall morphemic interdependency of speech units, and (3) the "word" should not be regarded as essential to the concept. Due to the difficulty of quantifying the parameters, the notion of syntheticity/analyticity will be most useful in plotting the general directions taken by a speech unit over its recorded history (i.e., ANALYTIC SYNTHETIC or SYNTHETIC - » ANALYTIC), and least useful in calculating the exact location(s) it has occupied along the analytic/synthetic axis. Prima facie, a method of quantification might seem preferable to ascertaining direction. But, from a diachronic perspective

48

Analyticity and Synthenticity

at least, calculating the syntheticity of a speech unit on some absolute scale is less essential than having at our disposal a flexible method which enables us to place it, often from numerous standpoints (phonological, morphological, syntactic, or semantic), relative to another unit. While such a proposal does not hesitate to reintroduce the Sapirian idea of relativity (i.e., "more or less synthetic"), my approach departs significantly from this-earlier attempt in that, as mentioned above, the parameters analytic and synthetic are claimed to be applicable not to languages as wholes, but to speech units alone. This methodological difference is crucial. As I shall argue in detail, the application of these parameters to entire languages has been one of the major causes of oversimplification of the morphological typology of the Romance vernaculars, as well as the primary factor responsible for the general misconception of the terms analytic and synthetic.2

3.2 Syntheticity and analyticity redefined: the main characteristics Analyticity will be defined as the semantic, syntactic, morphological and phonological autonomy of morphemes within a speech unit. Syntheticity is characterized as the semantic, syntactic, morphological, and phonological interdependency (or relatedness) of morphemes within a speech unit. A unit whose morphemes show a high degree of relatedness on all four levels is highly synthetic; a unit whose morphemes have a very low degree of relatedness on all four levels is said to be highly analytic. Though it may generally hold that strong interdependency among components correlates with a high degree of synthesis, it should not be inferred that synthesis at each level proceeds at an equal rate. The morphemes of a unit may, for instance, become highly related on the phonological level but minimally so on the syntactic level. Examples of such discrepancy are easily found throughout Romance. Take the modern French utterance: on nous a dit qu'ils ne seraient pas à la gare avant huit heures 'we were told they would not be at the train station before eight o'clock.' On the basis of morphological and/or syntactic criteria, one can identify at least ten distinctive units (on ... a dit, nous, que, ils ... seraient, ne ...pas, à, la, gare, avant, huit, heures)? From a purely phonological perspective, the sentence just cited contains at most three units (i.e., on nous a dit / qu'ils ne seraient pas à la gare! avant huit heures). This discrepancy between morphosyntactic and phonological units results from the fact that French is a "cursus language" in which morphological or syntactic units lose their segmental and suprasegmental identity within the breath group. 4 The question of whether the phonological, morphological, syntactic, or semantic component is to be assigned primacy in measuring the syntheticity of a speech

Analytic and synthetic in diachronic description

49

unit is largely subjective and language-dependent. This is no major drawback, however, since the accumulation of minute and often difficult-to-perceive modifications on the various levels tends to push a unit in the direction of either analyticity or syntheticity. As the Romance data will show, significant changes in the analytic/synthetic spectrum are not random, but rather operate according to a set of implicational principles. That is, one finds recognizable, standard patterns of change by which morphological elements alter their overall interdependency. Being familiar with such evolutionary patterns, one can decide, then, with some confidence whether an incipient development ought to be accorded significance in measuring the gravitation of a speech unit toward greater syntheticity or analyticity. Both analysis and synthesis are, by necessity, processes which involve time.^ The two processes resemble each other in that synthesis presupposes "loss of analysis," and analysis presupposes "loss of synthesis." However, while synthesis is achieved through "tightening" of relationships between two or more morphemes, analysis does not, in most instances, occur through a converse process, i.e., the gradual "loosening" of that tightness. In the majority of cases analysis is achieved either through the (lexical) replacement of a formerly more synthetic unit (e.g., the transition from the CL ama-bo- to VL amare habeo), or of amavi by amatu habeo Ί have loved'), or, more rarely, through the final synthesis of a (synthetic) speech unit into a monosememe (see the development of Fr par-ce-que [three morphemes] > /parska/ [one morpheme], illustrated in Figure 1, p. SO), which will thereafter enter into analytic constructions.6 A change in the interrelationship of morphemes of a speech unit towards greater syntheticity always implies analytic —> synthetic; however, greater analyticity in a speech unit does not always entail a change from greater to less«- syntheticity (i.e., synthetic —> analytic on the continuum) and may, as exemplified in Figure 1 (stage S) below, result from the culminating fusion of formerly discrete components as a new single unit The resulting unit may then participate in a fresh cycle of analytic constructions, e.g., CL mane 'in the morning' —> de mane '(tomorrow) morning' —> Fr. demain 'tomorrow' whence demain + matin 'tomorrow morning.' The directionality of speech units is generally not circular in the sense that an analytic formation arises only after having reached total synthesis on the morphological, phonological, syntactic and semantic planes. Speech units may reverse their direction along the analytic/synthetic continuum at any time. As the Romance data to be presented will confirm, such reversal appear to be overwhelmingly more common in the direction of analysis than of synthesis. The issue of whether, as has been suggested (Ashby 1977:34), this is merely a perceptual problem stemming from the fact that analysis is more easily observed than synthesis (since the former

Analyticity and Synthenticity

o 4) •s 8 - i l 62 Ì |

3*

l í & I tif 12s C C U Ρ o G' ε § a O S · 0 e2 ε

® u α d >, s-a ! S

& en ο λ; CA S -Ρ
más grande will be categorized as development in the direction of greater analyticity. 3.32 Measuring syntheticity While semantic synthesis depends on the psychological association a speak«- perceives between elements, morphosyntactic and phonological synthesis reflects the formal interrelatedness of speech units (or parts thereof) in the surface string. The concepts of syntheticity and analyticity, we recall, have often been approached with the assumption that morphological complexity could be arrived at with relatively uncomplicated formulas such as MORPHEME/WORD (Greenberg), ANALYTIC WORD/SYNTHETIC WORD (Krupa), etc. While such proposals have the elegance of simplicity and apparent virtue of methodological clarity, their results have typically failed to reflect important variations in the nature of morphological constructs. Conceptual differences in the architecture of pairs such as Sp hubo/habló, tuve/hablé, fue/habló. Fr nous sommes/nous parlons, je serais/je parlerais etc. were ignored despite their obvious relevance to the question of syntheticity.1^ The model that results from the principles described in the following paragraphs helps to show that to capture important variations in the interrelatedness of units, one

Analytic and synthetic in diachronic description

57

must evaluate and bring together not just a select few, but all of the formal factors which condition the network of relations within a given speech unit Some of the most important parameters—separability, displaceability, and isolatedness—by which formal interrelatedness differs, have already been mentioned in chapter 2, where the difficulties of finding a universal word criterion were highlighted. There are others as well—obligatoriness, semantic and paradigmatic transparency, agreement, etc.—which should be taken into account when tracing the direction of a speech unit on the synthetic/analytic scale. Individual changes in these parameters will not necessarily reflect the direction of a morphological complex at a given point in time. However, when these individual changes are correlated with other, often parallel developments in the entire system of formal interrelations within a speech unit, there will emerge a clear picture of evolutionary patterns in the drift towards a synthetic or analytic construct. As the languagespecific examples in the following chapters illustrate, a variety of disjointed phonological and morphosyntactic developments can interact to cause the synthesis of speech units in a fairly synchronized way so that the passage from independent lexical element to clitic, derivational affix, or even full-fledged inflection may be broken down into a number of largely predictable steps. 332.1 Morphosyntactic criteria The direction of semantically relevant speech units on the analytic/synthetic axis is determined by the following criteria: 3.3.2.1.1 Separability One of the prime factors for morphological synthesis is syntactic contiguity. For relevant lexemes to become joined formally they must be juxtaposed in the surface string. A low measure of separability is, therefore, characteristic of highly synthetic speech units; conversely, consistent separateness is typical of more analytic units. If two relevant units are interruptible, their level of syntheticity depends on (1) the frequency of interruption, (2) the number of insertable word classes (nouns, adverbs, pronouns, etc.), and (3) the size of the insertable word classes. A reduction in the number of intercalatable word classes is generally a much more significant manifestation of synthesis than are large decreases in the overall number of insertable items. The rationale underlying this observation is that the number of intervening word classes proportionally augments the linear distance between the semantically related speech elements regardless of the size of each class. A good illustration of this is provided by the separability of the French subject clitics and verb. In spite of the relatively low overall number of insertable speech elements,

58

Analyticity and Synthenticity

the intercalation of as many as three different word classes (negative particle, direct and indirect object pronouns) nonetheless creates a relatively great distance between the relevant items (cf. JE ne te le DONNE pas Ί NEC you it give NEG'). One should not infer, however, that consistent juxtaposition is a prerequisite for the morphological fusion of relevant units. As the Romance data will illustrate, during the period of semantic reanalysis, relevant units often remain separable, and are conjoined on the level of syntax only when insertable elements fail to be generated in the sentence. As this contiguous word order increases, the higher frequency of the relevant units may then cause the expression to become slowly recognized as an uninterruptable whole. Even though occasional separation of relevant units may have a retarding effect on their eventual synthesis, morpheme separability does not prevent speech units from reaching high levels of synthesis (cf. the relatively synthetic CL canta-b-o whose person/number marker can be separated from the stem by the future suffix -b-). Morphological coalescence is, however, excluded if the separation of two relevant items is an obligatory syntactic feature. The contrast between the differing effects of obligatory separateness and possible separability is perhaps best illustrated by the example of the histories of predicate negation in standard and creole French. 332.1.2

The differing effects of separability and separateness

The rise of the double negation ne ... pas, originally an emphatic construction in which pas 'a step' functioned as emphasizer, dates back to the old French period.16 During the last eight hundred years, both ne and pas have undergone the kind of phonological and morphosyntactic changes (semantic reanalysis of ne and pas, virtually obligatory placement of pas with ne etc.) which in juxtaposed relevant items typically leads to morphological fusion. The obligatory insertion of a verb between the two particles (cf. je ne PARLE pas) has, however, forestalled their morphological coalescence, and it appears that the negative particle ne (which has been eliminated in most registers of modern spoken French) will ultimately be lost without ever attaching itself to pas. A development in contrast to that of standard French has occurred in Mauritian Creole, where ne + pas has become fossilized in the synthetic unit napa(s). The synthesis of ne ... pas was made possible because the speakers of pidgin or creole Mauritian French reanalyzed the standard French syntagm n'a pas '[s/he] has not' (morphemically NEG + VERB + NEG) as simply 'not' (morphemically NEG, used with all persons of the verb), which, in turn, eliminated the obligatorily disjunctive structure of ne ..pas. As a result of this semantic restructuring, the exponent of predicate negation in Mauritian French is now preverbal (cf. mo NAPA tsi done Ί didn't give* [Richardson 1963:11]), and n(e), unlike its

Analytic and synthetic in diachronic description

59

French companion, has been saved from immediate oblivion by being frozen into what has become the new exponent of predicate negation. 33.2.13

Linearity

Linearity is closely related to separability because it also focuses on the segmentation of meaning within morphological constructs. It differs from separability, however, in that linearity focuses on the accumulation and fusion of meanings or functions in a single morpheme. Accumulation and/or fusion occurs when (1) juxtaposed morphemes fuse (as in Sp del - de el 'of the') or coalesce completely (as in Fr au loi = à + le 'to the,' du = de + le 'of the'), or (2) a phonological change results in a functional opposition within a paradigm (see example of Spanish poder below). In entirely linear speech units, grammatical and/or semantic elements are mapped onto morphemes in a one-to-one relationship (cf. Fr les belles femmes /lebel-fam/ = plural + semanteme + semanteme 'the beautiful women'). Alinear forms whose morphological seams become blurred are said to be moving in the direction of greater syntheticity. Good examples of such blurring are ablauted Sp preterites pudo 's/he could,' tuvo 's/he had,' supo 's/he found out.' Here the morphemic succession of semanteme (or stem) + tense/aspect is alinear because the tense-aspect-related vowel alternation in the stem (pod- vs. pud-) overlaps with the root expression 'poder' (i.e., pod- in poder). Even greater synthesis is obtained when such overlapping units fuse further and eventually become layered indistinguishably onto a single—what is often termed—porte-manteau morpheme. Such is the case, for instance, in Sp es or Fr (je) suis where semanteme, tense/aspect, and person/number markers are all amalgamated into a single morpheme. The degree of syntheticity of an alinear unit like suis depends on the number of functions and meanings it contains. Extensive accumulation of functions and meanings denotes extensive synthesis, loss of a function or meaning translates into analysis. While the previous statement has an overtone of simplicity, a look at actual data reveals it to be less obvious than might seem. In the present indicative of stem-alternating verbs like standard Am. Sp poder 'can,' person and number are marked not only by verb-final inflections (cf. pued-O Ί can,' pued-ES, 'you can' etc.) but also by the stem-internal vowel alternation /o/ ~ /wé/ : /of signals [redundantly] 1 pi. [p-O-demos 'we can'], and /wé/ indicates anything but 1 pi. [p-UE-do Ί can,' p-UE-des 'you can,' etc.]. Although the functional load of this vocalic opposition is low when compared to that of the inflectional endings, the vowel change clearly complicates the "ideal" linear one-to-one mapping of semanteme and stem. Now, in some registers of popular American Span-

60

Analyticity and Synthenticity

ish, this paradigmatic irregularity has been leveled out by extending the diphthong /wé/ to the entire present indicative paradigm {podemos —> puedemos)P As a result of this process of paradigmatic realignment, person and number are now signaled exclusively by the verb-final inflections, the greater linearity of the invariant stem morpheme pued- makes the paradigm more analytic.18 As regards synthesis and analysis, the development of popular American Spanish poder reveals that the linearity of a speech unit is not always altered by a unit-internal development. Am. Sp puedo, puedes, puede, and pueden have all attained greater analyticity not because they themselves underwent a change in form, but because a phonological modification in another member of the paradigm (podemos —* puedemos) served to obliterate a functional opposition.19 Having looked at the case of poder, we are now in a better position to discuss the relationship between syntheticity and the phenomenon of "zero expression." Zero expression occurs when one member of a category contrasts with other members by having no overt marker. Examples of zero marking abound in Romance and many other languages; a good illustration is the present indicative conjugation of standard French parler 'to speak': 1. s. 2. s. 3. s.

/2a Ay /¡O)

-

parl-0/ parl-0/ parl-0/

l.pl. 2. pi. 3. pi.

/nu /vu /i(l)

-

parl-ö/ parl-e/ parl-0/

-

In the above examples, number and person are specified by both prefixes and suffixes. While person and number are marked overtly by prefixes throughout the paradigm (the so-called subject pronouns), such consistency is not achieved in the verb endings where only the first and second person plural are suffixally inflected. Given that the absence of verb-final inflection in /pari/ contrasts with /parl-ö/ and /parl-e/ and therefore signals anything but 1 or 2 pl., zero marking cannot be considered meaningless. The question naturally arises, then, of how to characterize zero morphs on the analytic/synthetic scale. According to the concept of linearity as laid out above, the accumulation of meanings or functions within a single morpheme connotes greater synthesis. Given that the zero expression "anything but 1 or 2 pi." is contained entirely within /pari/, one must conclude that the above forms with a zero morph are slightly more synthetic than, say, Eng 'can' which never contrasts with another member of its present tense paradigm.20

Analytic and synthetic in diachronic description

61

A similar but less complex example of the relationship between zero expression and syntheticity is the singular of nouns in the Spanish Creole Palenquero (Colombia). Palenquero libro a-ta aki 'book + tense/aspect + be + here = the book is here' bears no representation for singular in the subject, yet the meaning of the sentence is that only one book is here. Number is left unspecified in the surface string, but is inherently implied because of the absence of the plural marker ma (cf. ma libro a-ta aki 'the books are here'). Just as in the case of /-parl-0/, the non-presence of a marker in libro has a definite meaning. This meaning becomes more apparent when the alinear representation of the singular in Palenquero nouns is contrasted to the more analytic Spanish EL libro está aquí and LOS libros están aquí, where number is signalled overtly in linear fashion in both the singular and plural by the determinate article él, and los. In Spanish nouns the singular and plural signs are not entirely analytic either since the overt plural sign -s in libroS stands in opposition to a zero morph in libro. The representation in Figure 2 highlights the conceptual difference in the marking of number in Spanish and Palenquero nouns.

Zero morph opposition in marking of number in Spanish and Palenquero nouns

Palenquero SINGULAR

PLURAL PLURAL +

NOUN libro

^^ma/^

opposition

NOUN libro

no opposition

Spanish SINGULAR SINGULAR + NOUN el

PLURAL +

libro

PLURAL

0

M

los

•410 opposition

+

NOUN + PLURAL libro

opposition

Figure 2

ri

62

Analyticity and Synthenticity

Involved in the notion of linearity is also the question of grammatical agreement The repetition of identical semantic or grammatical matter in successive morphemes denotes synthesis. The reiteration of gender marking in Sp IA cosA amarillA 'the yellow thing,' for instance, accounts for the relative syntheticity of the expression (note, in contrast, the analytic nature of Fr lA chose jaune /la Soz 2on/ where gender is expressed once only). Similarly, the multiple case marking in Lat CUM homin-IBUS femin-IS et infant-IBUS is more synthetic than, say, Sp CON hombres, mujeres, y niños where case is unmarked altogether. The analysis proposed here is again in accord with our general notion that greater analyticity is achieved when meanings (or functions) and morphemes are mapped onto a unit on a one-to-one basis. The repetition of identical semantic or grammatical information (as in IA cosA amarillA) constitutes a departure from the "ideal" isomorphic norm, and is, consequently, a sign of greater syntheticity. 3.3.2.1.4 Displaceability Displaceable units are more analytic than units which must occur in a fixed order. Fr tu parles is more analytic than Palenquero bo ta abld 'you are speaking' because the subject clitic tu, at least in more formal registers of French, may be moved into post-veibal position (cf. Fr parles-tu? but Pal *ta ablá bo?). 3.3.2.1 J Isolatedness A form which can occur in isolation is more analytic than one which must occur in conjunction with another element. Fr moi and toi are more analytic than je and tu because they may occur without a finite verb. While clitics or affixes are naturally more synthetic than free forms, the measure of isolatability of free forms is not always constant and depends on the syntactic combination in which they occur. Consider for instance Spanish cuatro 'four.' Though clearly a free form in some instances, in a las cuatro, it is no longer an independent unit since its grammaticalness and correct interpretation hinge on the presence of the definite article las (cf. ¿qué hora es? — * cuatro). Failure to recognize the existence of phrasal bonding among units of speech may lead to serious flaws in evaluating the measure of synthesis of relevant speech units. This point, advanced somewhat tentatively here, will be taken up again in the following section, which will bring further evidence to bear on the matter. 3.3.2.1.6 Obligatoriness A linguistic unit which requires the presence of another relevant unit within the same phrase or sentence is more synthetic than one for which no such condition

Analytic and synthetic in diachronic description

63

exists. The relationship between Spanish subject pronoun and verb is analytic because verbs may—and most often do—occur without pronominal person markers (cf. [yo] hablo Ί speak/am speaking'). The contrary is true of Eastern Swiss Romansh (Engadin), where subject pronouns (all free morphemes like their Spanish counterparts) must accompany finite verbs in spite of the fact that suffixal verb inflections mark person and number oveitly in all but the first person singular, eau vend-0, tü vend-ast, ellella vend-a, nus vend-ain, vus vend-ais, els/ellas vend-an Ί sell,' 'you sell,' etc. 22 Other examples of units whose synthetic character originates from the obligatory cooccurrence of other elements within the same clause or sentence are found in what may be termed 'predictable' stock expressions such as Fr peu à PEU 'little by little,' petit à PETIT 'little by little,' and Sp and al fin y al CABO 'in the end.' Though peu, petit, fin, cabo are all free morphemes, in stock expressions these unbound forms are nonetheless relatively synthetic because the syntactic patterning of the initial morpheme sequence (given in lower case) of each expression demands the presence of additional morphemes (highlighted in upper case) in the surface string. The obligatoriness and invariable syntax of these phrases greatly increase their chances of morphological fusion. This claim is easily buttressed by looking at just two such stock expressions of modem French. In AU FUR et à mesure,23 'gradually' and EST-CE QUE [morphemically = interrogative], the morphological seams have become blurred and the semantic content of the capitalized elements are no longer evident to non-linguists. I present these observations on the obligatoriness of certain constitutents within stock expressions primarily to point out that the interrelatedness of speech units cannot be understood simply in terms of coherence on the morphological level. A morpheme like petit may well be a separable, transportable, independent, and, therefore, highly analytic unit in most instances. But we must be prepared to recognize that the syntheticity of petit, peu, cabo, or any other unit also depends on the particular syntactic combination in which it occurs. Morphological factors are clearly dominant in the overall picture of interrelatedness among speech units, but ultimately it is the combination of semantics, morphology, phonology, and syntax which determine the extent to which speech units may eventually coalesce. 332.1.7 Transparency The analytic or synthetic character of flexional forms also depends on paradigmatic transparency. Analytic units typically display great transparency, while synthetic units tend to lack such transparency. We will distinguish between two types, structural and semantic transparency.

64

Analyticity and Synthenticity

33.2.1.7.1 Structural transparency In the above discussion on linearity I upheld the view that isomorphism is characteristic of analytic speech units. The analyticity of such an "ideal" unit is reduced, however, when the form of a morpheme varies. When such variation exists, we speak of an alternation between allomorphs (e.g., the phonotactically conditioned allomorphs /le/ and Λβζ/ in les pommes, les enfants). Allomorphs may be bound or free. In French, for instance, the first person singular has two bound allomorphs, /2(œ)/ je or / and /m(œ)/ me or m', and one free allomorph, /mwa/ moi—and similarly for the second and third persons singular and third person plural pronouns (Hall 1983:194).24 The occurrence of either of these forms is correlated with stress and syntactic position in all but the first and second person plural, where the alternation between the bound allomorphs /nu/ ~ /nuz/ and /vu/ ~ /vuz/ depends strictly on the phonological environment Alternations of allomorphs usually have some fairly general rationale. As seen in the case of the French pronouns above, they can normally be described by a limited set of rules. Alternations between allomorphs are not, therefore, isolated peculiarities of individual morphemes. A given rule does not, however, always prevail throughout a whole paradigmatic set, as is shown by the difference in allomorphic alternations between phonologically conditioned Fr nous and vous, and the syntaxand stress-related je,j', moi, me, etc. As regards the question of syntheticity and analyticity, the emphasis placed on allomorphic alternations is important because the amount of variation in the form of a morpheme directly affects the tightness of a morphological system. Related forms whose paradigmatic association is based not only on semantic or grammatical, but also on (transparent) phonological and structural, similarities obviously have a considerably tighter interconnection than those units which show no formal kinship to the rest of the paradigm. If the analytic/synthetic tension is to measure morphological complexity, then the distinction between a transparent and a "fuzzy" set of related items is surely of considerable interest A high level of syntheticity is reached when a morpheme has an abundance of allomorphs whose alternations can be captured only by a multitude of rules. Furthermore, allophones which bear no phonological resemblance to each other (cf. for instance Fr je to moi) will be considered more synthetic than those with extensive formal similarities (cf. Fr /vu/ to /vuz/). Allomorphic alternations alone do not, however, determine paradigmatic transparency. Paradigmatic transparency also depends on the formal interrelation of the various morphemes constituting a paradigmatic set A comparison between the relatively analytic Palenquero personal pronouns and those of relatively synthetic Clas-

Analytic and synthetic in diachronic description

65

sical Latin (CL) may be best suited to illustrate my claim. Classical Latin 25 SINGULAR 1. 2. 3.

1.

Nom. Gen.

ego mei

tu tui

sui

Dat. Acc. Abl.

mihi me me

tibi te te

sibi se se

PLURAL 2.

nos nostrum. nostri nobis nos nobis

vos vestrum, vestri vobis vos vobis

3. _ sui sibi se se

Palenquero 1. Subj. (, y /y/ non-Subj. mí

SINGULAR 2. 3. bo, o bo, o

ele ele, lo

1.

PLURAL 2.

3.

suto suto

utere utere

ané ané, lo

The most striking difference in the two paradigms is the relative abundance of forms in CL. Abundance of formal alternation increases the syntheticity of each individual form. Consequently, Lat tu, nos, etc. may be said to be more synthetic than Pal bo or suto. In Latin, the personal pronouns are not expressed as subjects, except for distinction or emphasis (Greenough et al. 1979:176). Their alternations are case-related, and are complicated by the fact that the individual changes do not follow predictable patterns throughout the paradigm. One observes, for instance, that the transparent organization of the accusative (me, te, se) and ablative singular (me, te, se) is not matched in the genitive singular where *mui (instead of mei) would serve as a more transparent member within the set mei, tui, sui. Similarly, the observed parallelism between the accusative and ablative singular is not followed through in the 1 and 2 pi. where nos and vos alternate with nobis and vobis, respectively. Further structural fuzziness occurs in vestrum and vestri where the e (instead of o for expected * nostrum and *vostrum) contrasts sharply with the remaining forms in the otherwise unusually parallel series of the first and second persons plural. These irregularities (as well as others which can easily be extracted from our table) show that formal variations within the entire paradigm are not only numerous but also nonrecurrent, that is, they do not follow a general, structurally predictable pattern. It is

66

Analyticity and Synthenticity

ultimately this high measure of non-recurrence, and not merely the fact that Latin pronouns undergo alternation in each of the six cases, which makes the CL ΛΟ

declension of personal pronouns a relatively synthetic one. As already pointed out, the contrast in the amount of formal variation between the Classical Latin (22 variants) and Palenquero (10 variants) paradigms is striking. While this in itself underlines the greater analyticity of Palenquero forms, the fact that some items (suto and utere) show no variation at all contributes further to the overall transparency of the Palenquero system. As our table shows, the Palenquero system is, however, not entirely transparent either. As in Latin, some alternations are related to the syntactic function of the pronoun. The 1 s. pronoun, for instance, changes from itomi when it it is not a subject. The 3 s/pl. pronoun (ele) obeys a similar pattern in that it must take on the form of lo whenever it is not (1) a subject, (2) preceded by a preposition (cf. ku ele 'with him') or (3) placed after a gerund (cf. í ta minando ele Ί am looking at him/them' [Friedemann and Patiflo Rosselli 1983:164, trans, mine]). The alternation between ele and lo—though by no means overly complex—represents by far the most complicated change within the pronominal system of Palenquero. The remainder of the pronominal system is simple: Yo is emphatic, and the change from / to y /y/ and bo to o is tied to phonological factors: y replaces Í when followed by the tense/aspect marker a (cf. y-â ta ablá Ί am speaking'), and o, typically used in rapid speech, appears to be a free alternate of bo (Friedemann and Patiflo Rosselli, p. 160).29 One could build an even stronger case for the transparency of the Palenquero paradigm by arguing that the 1 s. alternation (/ y in the table above really lies outside the domain of the pronominal system. Creoles, including Palenquero, typically lack verbal inflections, and so morphemically free subject pronouns must assume the function of person marking.^0 y /y/ is, however, an exception to the rule since it is used not as a (morphemically free) personal pronoun, but as a clitic preceding the tense or aspect marker a (cf. y-a ten = I + Tense/Aspect + have Ί have'). If we may draw an analogy to the Latin system, y fyl is, then, in a sense much more akin to the inflected person marker -o (cf. am-o) than to ego.31 The main reason for bringing up the question of person marking in Latin and Palenquero is to help focus attention on a recurrent problem in the analysis of morphological systems, namely the written language bias. We take as synthetic (or "belonging together") whatever is written together, and analytic whatever is separated by blanks. Though there can be no doubt that the structure and use of written language is largely derived from the properties of spoken language, it is also true that the concepts of analytic/synthetic have often been developed by linguists on the basis of written language. It is, for instance, commonplace to view the CL pronouns (written as "words") as a paradigmatic set unrelated to inflectional person

Analytic and synthetic in diackronic description

67

markers (written as "part of a word"). While there are strong arguments (minimum free form, displaceability, etc.) in favor of keeping the two separate, we must, nevertheless, be willing to concede that on semantic grounds alone such a division is not altogether justified. Discounting pragmatic differences, there is no compelling reason why ego and the -o in am-o should not be considered merely allomoiphs of one and the same paradigm. The question of how to view the organization of certain morpheme structures is important For on it depends our notion of paradigmatic transparency and, ultimately, of syntheticity and analyticity. It should be sufficiently clear why, from the perspective of paradigmatic transparency, Pal bo stands at the analytic and CL ego at the synthetic end of the synthetic/analytic dimension. Although paradigmatic transparency has, to my knowledge, never been formally incorporated into the question of analyticity and syntheticity, I believe that many of the generalizations (see examples in chapter 4) in the linguistic literature about the so-called analytic or synthetic character of a language have been based in part on an impressionistic analysis of the number of allomorphic alternations within a few select paradigms. Although such an approach to the question of syntheticity cannot be commended, it nevertheless helps to show that paradigmatic transparency is naturally felt to be an intrinsic component of the analyticity/syntheticity tension. 33.2.1.72 Semantic transparency The vocabulary of language may be divided into lexical [semantically transparent] and grammatical [semantically non-transparent] units (cf. lexical house, love and grammatical the, to).32 As noted previously, it has been observed that grammatical elements often were originally independent lexical items. In the transition from a lexical to a grammatical unit, lexical meaning is frequently retained long after a unit has begun to assume a grammatical role. Such units are said to be transparent. Semantically transparent units are more analytic than non-transparent ones. Members of a paradigm typically maintain semantic transparency when their subunits also occur as free lexemes. Consider the following standard Spanish examples: Semantically transparent: voy a hablar ! iba a hablar vas a hablar / ibas a hablar va a hablar I iba a hablar etc. etc.

Semantically non-transparent: había hablado I hubiera hablado habías hablado / hubieras hablado había hablado / hubiera hablado etc. etc.

The measure of separability, transportability, and linearity in the above auxiliary

68

Analyticity and Synthenticity

segments (voy, iba, había, hubiera) is virtually identical; the examples on the left are, however, more analytic because the auxiliaries voy/iba a, though clearly tense markers like había, continue to be associated with the free lexeme voy Ί am going' or iba Ί went' (voy/iba a Madrid)?* Había and hubiera, on the other hand, have no independent meaning, which accounts for the more synthetic character of the paradigms on the right Units whose stem cannot occur independently and has no independent meaning are particularly non-transparent. A major reason for the non-transparency and synthetic character of CL nouns originates from the lack of autonomy of their stems, cf. CL fili-[us] 'son,' where the stem itself has no specific meaning, to Fr [le] fils. The strong analytic nature of Creoles, on the other hand, originates to a large extent from the independence and transparency of their stems. Dependent stems of the type of Fr i- (in j'irais Ί would go') are extremely rare in those languages and, if present at all, are usually ascribable to recent superstrate influence. 33.22

Phonological criteria

According to some linguists, phonology and morphology should be kept very rigidly apart. As argued above (§3.1), morphological tightness cannot, however, be understood properly without taking into account the phonological components which interrelate units of speech. In many languages, including most of Romance, phonological and morphosyntactic units frequently coincide, and the measure of this synchronization is demonstrably related to the tightness of speech units. Evidence for this claim is perhaps best obtained by looking at two relevant lexemes whose phonological development differs when they are used together. The diachrony of the Romance forms cantare and habeo shows, for instance, that when they were juxtaposed to signal futurity, the two items no longer behaved as independent phonological units but instead developed as a single phonological unit Valesio discusses the passage from independent to dependent cantare habeo as follows: It is conceivable that the stress of /habeo/ in the phrases of the /kantáre hàbeo/ type (the equivalents, at the moment of the passage from Early Common Romance to Late Common Romance, of Classical Latin CANTARE HABEO) was phonetically weak»- than, and subordinated to, the stress of the preceding infinitive form; this is the unusual change undergone by the primary stress of a verb which can function independently, when it is used as an auxiliary verb. In a phrase like /hábeo dómu(m)/ Ί have (possess) a house,' where /hábeo/ has its full grammatical and lexical autonomy, the verb has the primary stress. In a phrase like /kántare

Analytic and synthetic in diachronic description

69

hàbeo/, where the verb has grammatically auxiliary function, the primary stress Π becomes a secondary stress Γ/ dependent on the primary stress of the preceding infinitive. With the merger, which gives origin to the future pattern, the stress of the -HABEO form is transformed from secondary to primary, as all the derived Romance paradigms show. (1SS) While the auxiliary habeo gave up its primary accent at a relatively early stage and became bound to the adjacent infinitive, morphemically free habeo appears to have remained accentuated until French developed into a cursus language. Ultimately, some of the accentually different bound and unbound exponents of habere split and the first and second persons plural evolved into -ons (oarlerONS), -ez (parlerEZ) and avons (nous avons), avez (vous avez), respectively. The role of phonological bonding between relevant units has also been important in the history of the definite article in Romance. Pan-Romance evidence suggests that the elements which evolved into the definite article must have become phonologically tied to the adjacent noun at an early stage. The Latin demonstratives ipse, ille etc., when used, either preceded or followed the noun and soon subordinated their accent to that of the nominal elements to which they attached.35 The deictic force of the demonstrative weakened, and, for reasons which I will not attempt to elucidate here, the frequency of the sequential order DEMONSTRATIVE/ARTICLE + NOUN eventually increased to become the standard in most of Romance. That these qualifiers had different phonological contours in this proclitic position than when used in isolation is shown by the widespread apheresis of ille and ipse (or their respective allomorphs). In Old French, for instance, the initial syllable of ille (or its respective allophones) was lost when the quantifier was used prenominally (cf. illu > lo, le·, illa > la\ *illi > li) but was retained when employed as a subject pronoun (*illi > il·, illa > elle)?6 Similarly, the difference in stress between the early Romance atonic demonstratives/articles and tonic subject pronouns illu, *illi, illam, etc. ultimately resulted in the extensive phonological dissimilarities between the modem Portuguese definite articles o(s), a(s) and the subject pronouns êle(s), elafs). No criteria are more language-dependent than the phonological ones. While we have been able to point to a set of criteria (separability, displaceability, etc.) by which to assess morphosyntactic synthesis, no such standard set of criteria can be proposed to measure phonological synthesis. Phonological systems vary greatly from language to language; the most characteristic phonological features of one language are often non-existent in another. Vowel harmony is an outstanding trait of Turkish phonology but virtually absent in modem Romance; sandhi is important in French but quite unimportant in Palenquero; rising intonation is the only means

70

Analyticity and Synthenticity

for signaling interrogation in a Palenquero construction of the type bo ta mind 'are you coming?', but this suprasegmental ingredient is merely an accompanying feature of questions in Romansh (Engadin) where subject-verb inversion is compulsory (cf. vainst tü? 'are you [s.] coming?'). Clearly, neither of these features is common enough to be used as a solid criterion in measuring phonological synthesis. Nor can a phonological trait of a given language be used consistently to assess the synthesis of a speech unit along its historical trajectory. In early Romance, intonation was synchronized to a large extent with morphosyntactic factors in that every free lexical unit carried one primary stress. No such synchronization remains, however, in modern French where word stress has been supplanted by that of the breath group. It is, of course, impossible to discuss the many phonological mechanisms (intonation, syllable structure, sandhi, morphophonemic constraints, etc.) which contribute to the cohesiveness of morphological complexes in Romance and other languages. It will, therefore, be instructive to work through a small set of examples to demonstrate how certain phonological features should be assessed from the perspective of analyticity and syntheticity. First, and foremost, relevant units whose morphosyntax coincides with phonological units are more analytic than those in which such synchronization is not present From an intonational perspective, Fr je, being incorporated into the verbal core, is more synthetic than Engadin Romansh eau which bears primary stress and is phonologically independent within the verb phrase. From a diachronic perspective this means that if two (or more) accentually autonomous relevant items converge phonologically while the language continues to maintain word-stress, the resulting phonetically "tighter" unit will be said to have moved in the direction of greater synthesis. Examples of such a development are Sp /kómo nó/ > /kòmonó/ and Am. Sp /náda más/ > /nádamás/. Because of the written language bias, we tend to think of cómo and no in ¡cómo no! 'of course! ' or nada and más in nada más 'only' (cf. tengo un carro nada más Ί only have one car') as two separate entities. But there is conclusive phonological evidence that these stock expressions have already evolved into synthetic complexes. This is perhaps brought out best by comparing ¡cómo no! to its near homonym "como, ¿no?" Ί eat, don't I?', where there are two primary stresses, /kómo ηó/. In ¡cómo no! /kòmonó/, however, the first element has become destressed and the entire unit is now phonologically compact with a single stress stress on no. The same observation also holds for nada más /nàda más/ 'only' which—unlike nada más /náda más/ 'nothing/anything else' (cf. no tiene nada más 's/he doesn't have anything else')—now receives only one primary stress. Particular attention needs to be paid to suprasegmental features which act as sole

Analytic and synthetic in diachronic description

71

exponents of meanings otherwise expressed by overt markers, e.g., the signalling of interrogation by rising intonation. As is well known, since the Middle French period there have arisen a number of new interrogative patterns. While Old French interrogation almost invariably relied on subject-verb inversion, modern French has come to use two additional modes, est-ce que or intonation alone (cf. est-ce que tu viens?, tu viens? 'are you coming?'). 37 Though the distribution of these three interrogation structures varies greatly with the use of particular registers, today speakers of French show a very strong preference for the purely intonational construction (tu viens?). Of the three interrogation types—inversion, est-ce que, and intonation alone—the est-ce que model is clearly the most analytic. With est-ce que, the correlation between meaning and form is optimal because the signalling of interrogation does not overlap with any other, already semantically loaded elements. On the other hand, in "tu chantes?" the meaning "this is a question" is grafted onto the verbal complex, and it is this accumulation of meanings which ultimately accounts for its greater phonological synthesis. In contrast to Tauli (1945:86), we will not consider phonetic strength as a relevant criterion for measuring the syntheticity and/or analyticity of a speech unit In principle, a phonetically strong moipheme is no more analytic than a phonetically weak morpheme. Fr comment and quoi are equally analytic, and so are Sp qué and cómo. In arguing that the phonetic constitution of morphemes ought to be factored into the syntheticity of speech units, linguists must have been influenced by the fact that a large number of phonetically minimal items are relatively synthetic. The correlation between phonetic substance and syntheticity is, however, largely accidental since the synthetic character of monophonemic items such as Fr y 'there,' en /3/ 'of it/them, from it/them,' on /0/ 'we one (ind. pron.),' etc. derives from their limited moiphosyntactic freedom (they are all bound, and mostly non-displaceable) and not from their phonetic length per se. That the phonetic constitution of a unit is not inextricably bound to morphological tightness is clearly evidenced by monophonemic units (e.g., Fr ou /u/, Sp and It o < CL aut 'or'; Fr et /e/, Sp y. It e < CL et 'and') whose evolution from bulkier to minimal phonic matter has not resulted in the loss of their high level of analyticity.

3 . 4 Chronology of semantic, morphosyntactic, and phonological changes As I have argued at the outset of this chapter, the synthesis of a speech unit does not necessarily proceed at an even rate on the semantic, moiphosyntactic, and phonological levels. But it appears to be the case that in the evolution from ana-

72

Analyticity and Synthenticity

lytic to a synthetic speech units there exists a certain chronological order in which the various levels begin to affect the intenelatedness of morphemes. Initially, synthesis is always of a purely semantic nature. Two or more items enter a state of mutual relevance, and, because of the cultural or cognitive salience of the new semantic association, typically begin to occur with an increased frequency. At this point, syntactic factors begin to play a decisive role. Because of the increased frequency of the relevant items, a particular word order tends to become recognized as a standard syntactic pattern. If the relevant items of this pattern are juxtaposed, the speaker may consider the resulting complex a syntactically cohesive unit, in which case phonological and morphological coalescence is likely to follow. If, on the other hand, the relevant items are always separated because of the prevailing word order, they will be prevented from merging formally. Obligatory syntactic separation necessarily thwarts the possibility of phonological fusion, but it does not preclude the kind of extensive semantic and morphosyntactic synthesis that relevant juxtaposed items often undergo. Phonological and/or morphological fusion is set in motion only after semantic and syntactic association have become relatively frequent. This is not to say that relevant items are exempt from phonological changes at any given point in their evolution. It merely means that unit-specific phonological developments (e.g., loss of word stress) will not materialize before a special semantic and syntactic interrelation has been established. It is important to note that while semantic, syntactic and morphological change are prerequisites for extensive synthesis, phonological coalescence is not It cannot be assumed (pace Ashby 1977:41), that synthesis involves a morphonological state whereby the lexemes in the construction must per force undergo unit-specific phonological changes. Although synchronization between phonological and morphosyntactic units undoubtedly plays a major role in the synthesis of formerly separate items, the history of examples such as Fr maintenir provides evidence that phonological synthesis is not a necessary ingredient in the analytic —» synthetic -» analytic cycle. The tightening of relations in the relatively analytic expression VL manu + tenere, and the subsequent evolution—via the highly synthetic main-tenir (= two morphemes)—to what may now once again be considered an analytic, monomorphemic structure, appears to owe nothing to phonological and everything to morphosyntactic changes. Both main and tenir survive as independent lexical units, but—in part because of the absence of word stress in French—neither contains phonological or phonotactic components which would set them apart from the etymologically related segments contained in maintenir.40 Once semantic and syntactic syntheses have begun to take place, it is impossible to predict whether synthesis will proceed further on the phonological or mor-

Analytic and synthetic in diachronic description

73

phological level. Though synthesis may be set in motion on the phonological but not the morphological level (or vice-versa), unrelated changes may begin to take place simultaneously on both levels. Initially, fusion on the morphological level is generally achieved, in order of sequence, through (a) increased morphosyntactic rigidity of the components of both contiguous and non-contiguous relevant units, (b) obligatory cooccurrence of all relevant constituents of a unit, (c) a decrease in the interruptability of the speech unit, (d) a change in the word class of one (or more) of the components, and (e) a reduction in the morphological variation of one or more of the constituents (cf. CL cantare + habeo where habere could no longer be used in all its temporal and modal options). Languages which pair stress or other suprasegmental features with morphosyntactic units will generally react to semantic and syntactic synthesis by rearranging the intonational contours of the formerly independent elements (e.g., CL manu + tenere [two phonological units] > Sp mantener [one phonological unit]). Other, unit-specific sound changes (e.g., the downgrading and subsequent elimination of unstressed vowels) may then follow. These latter changes tend, however, to postdate the morphosyntactic changes described under (a) through (d).41 I wish to make clear an important point regarding the chronology of the various shifts postulated for the evolution from an analytic to a synthetic unit. The sequence semantic —» syntactic —» morphological/phonological shift is proposed chiefly to facilitate the conceptualization of the change from a morphologically loose to a tight unit, and the various steps should not be understood in terms of discrete chronological stages. Statements such as "semantic synthesis precedes syntactic synthesis" are not meant to imply that one type of synthesis must reach completion before other changes may begin to take effect The chronology provided here merely attempts to establish the order in which one level may activate another. Semantic and syntactic syntheses can, and most often do, continue long after phonological and/or morphological changes have also begun to alter the interrelatedness of morphemes within a speech unit Often, changes on one level will mesh with those on another level, and a single change may have simultaneous consequences on more than one level (cf. the case of Am. Sp podemos > puedemos mentioned earlier). To be sure, the most characteristic feature of the transition from a highly analytic to a highly synthetic unit is not the chronology in which the various levels become activated, but rather the intense simultaneous interaction which takes place between semantics, syntax, morphology, and phonology once semantic and syntactic synthesis have been set in motion.

Chapter four The verbal core from Latin to French: Part 1: The marking of person/number: from synthetic to analytic? En Europe, les langues dérivées du latín ... ont une grammaire toute analytique... (A. Schlegel 1818:17) Rein analytische Sprachen sind ζ. Β. die französische ... (Kuznecov 1956:38) Le français est une langue analytique sortie de la langue synthétique qu'est le latin. Toute l'histoire de l'idiome atteste le décumul des formes synthétiques (Guiraud 1969:44) ... la decadencia analítica del espíritu francés (Cantera 1966:214) Für die indoeuropäischen Sprachen ist es kennzeichnend, daß ihre Entwicklung offenbar vom synthetischen zum analytischen Sprachbau führt; sehr weit fortgeschritten auf diesem Wege sind das Französische und das Englische ... (Schmidt 1967:85) ... an analytic language like Vulgar Latin, English, and French.... (Lathrop 1980:22) ... die beschleunigte Entwicklung des Französischen zu einer analytischen Sprache gegenüber anderer rom. Sprachen (Geisler 1982:26) Das Neufranzösische erscheint weitgehend als „analytisch" gestaltete Sprache mit der grammatischen Schicht der Wortgruppe als bevorzugter Schicht (Eckert 1986:125) Latin is synthetic in comparison with French, French analytic in comparison with Latin (Jespersen 1894:121)

76

Analyticity and Syntheticity L'ancien français était même, dans son lexique ainsi que dans sa grammaire, plus «synthétique» que les autres langues romanes; au contraire, le français moderne est beaucoup plus «analytique.» Plus exactement, le français est la seule langue romane qui soit vraiment «analytique» (Coseriu 1971:29)1 *** m En définitive, la molécule syntaxique française est en train de redevenir aussi synthéthique que la forme fléchie indo-européenne et latine. (Bally 1965 [1932]:301) In this study, we have seen evidence of a trend toward the synthesis of the noun and its satellites, and also the verb and its satellites. If these syntheses are completed, a highly complex polysynthetic, morphology indeed will result for future French. (Ashby 1977:89) Ce qui est certain, c'est que, même s'il existe un type analytique de langue, ce type n'est en tout cas pas celui du français, qui est una langue synthétique. (Tesnière 1932:64) ... after a distinctly analytic period in its history ... French in particular is now, in its popular registers, well on the way to becoming as synthetic, using prefìxation, as Latin was, using suffixation. (Μ. Β. Harris 1980:75) It is true that Bally has succeeded in proving that both tendencies, the analytic as well as the synthetic, exist in French just as in every other language, but the present partial agglutination of auxiliary words with the main word is by no means comparable with the relatively synthetic structure of Latin, and it is questionable whether French will regain the stage of Latin synthesis. (Tauli 1966:283)

Verbal core: Person/number

77

4.1 Introduction The opposing views presented in the above citations pose a number of thought provoking questions: why is there so little agreement about the morphological typology of French, one of the world's better documented languages? How is it possible that the same language is located by different investigators at opposite extremes on the analytic/synthetic scale? Is the disagreement simply due to the lack of clearly defined typological parameters? Or has French recently moved from one end of the scale to the other, so that the different claims simply refer to different diachronic stages? By looking at developments within the verbal a m from Latin to French, I will attempt in the next two chapters to answer not just these but a number of other questions as well. First, and foremost, I will document my claim, advanced in chapter 3, that speech units increase their syntheticity by undergoing a series of patterned semantic, syntactic, morphological, and/or phonological changes. Likewise, I will substantiate the claims that (a) (b) (c)

contrary to M. B. Harris (1980:75), the French verb group has never passed through a 'distinctively analytic period;'2 analytic and synthetic tendencies have existed simultaneously at all stages in the development of the French veib group, and the widespread practice of equating analytic or synthetic relationships between speech units with the overall morphological complexity of these units has led to gross oversimplifications about the morphological typology of French.

Before advancing further, it may be relevant to point out that in assessing the level of morphological tightness of French, scholars—excluding those who have applied quantitative methods—have never looked at the French language as a whole, but have primarily reached their conclusions by limiting their attention to changes within the noun and or verb group. While this is understandable insofar as some of the most important developments have taken place precisely in these two areas, from a methodological point of view such an approach is far from satisfactory. Changes in grammatical agreement—clearly relevant to the issue of syntheticity, e.g., the partial loss of gender and number agreement between adjectives and nouns—have been disregarded across the board. Furthermore, what evidence was adduced to document either the synthetic or analytic tendencies in the verbal core of French—understandably taken from diachronically unstable forms—has always been taken as representative of the entire verbal system, so that many historically more

78

Analyticity and Synthelicity

conservative paradigms (e.g., the imparfait and the présent of irregular verbs) were simply swept undo1 the rug in an assessment of the overall morphological tightness of French. Given that the verb morphology has been used as one of the prime indices of change in the synthetic/analytic tension of French, the reader may wonder why we will here return to the same area of grammar.3 The reason is simple: while a relatively uniform account of the facts of the evolution of the verb group may be obtained, interpreting these diachronic changes from the perspective of analyticity and syntheticity is quite another matter, as seen by the discrepant pronouncements cited at the outset of this chapter. Because of the traditional dichotomy of morphology and syntax as well as the prestigious Classical Latin model of grammatical analysis, until recently only syntactically non-separable elements were considered part of the French verb group. Consequently, a French verb was described as consisting of the sequence STEM + TENSE/MOOD/ASPECT + PERSON/NUMBER marker, but no reference was made to the often obligatory subject clitics. More recent studies (e.g., Ashby 1977, 1982) have begun to stress that an examination of all the components (i.e., even the disjunct morphemes) of the verb group is basic to understanding the nature of the verbal core. Ashby rightly argues (p. 2) that if chanterai cannot occur in French except with a preceding or following je, then surely this is as much a matter of morphology as it is of syntax. Sharing this view, many scholars now tum their attention to the "verb group" (rather than simply the verb stem and its affixed inflections) and look at the verb paradigm not merely from a morphological but also a morphosyntactic perspective. It is from this angle that the syntheticity of French verbal moiphology will be discussed here.

4.2 The verb group in diachronic perspective 4.2.1 Morphological tightness of the verb in Latin While even a very cursory examination of the interrelatedness of the various morphemes of the verbal core in Classical Latin will reveal the relatively synthetic character of the CL verbal system, inventorying some of the most important strategies by which Classical Latin verbs achieve their morphological tightness will help us gain a better perspective on the fundamental diachronic question of whether, or rather how, the Romance forms came to gravitate toward greater analyticity and/or syntheticity. If we except the periphrastic constructions (e.g., amatus sum Ί have been

Verbal core: Person/number

79

loved') for the moment, the high level of morphological tightness of the Latin verb is evidenced by the following factors: (1)

(2)

(3)

(4) (5)

(6) (7) (8)

(9)

the alinearity of morphemes (cf. laud + o = stem + person/ tense/aspect/voice/number, duxit 'he has led/ cepi 'he has taken': overlapping of stem and tense/aspect [cf. c-a-pere/c-e-pit]); allomorphic alternations in the expression of tense/aspect/mood (cf. pres. subj. laud-em '(that) I sing,' mon-eam '(that) I warn,' aud-iam '(that) I hear,'duc-am '(that) I lead'; perf. laud-avi Ί (have) heard,'mon-ui Ί (have) warned,' dux-i /duk-si/'I (have) led,' aud-ivi Ί (have) heard, cep-i Ί (have) taken' and person/numb» (cf. laud-o Ί praise* (ind.), laudav-i Ί praised,' laude-m Ί praise' (subj.), laudo-r Ί am praised'); low measure of separability (the separable elements [e.g., stem + ... + person/number] can be interrupted by only a small number of morpheme classes, and the size of each class is very limited); lack of displaceability (the order of all morphemes is fixed); absence of isolatablity (except for a few isolated cases [e.g., duc-ere —> imp. due 'lead!'], neither the stem nor the inflections can ever occur as free forms); the obligatory cooccurrence of stem + morpheme (including zero morphs) for tense, aspect, mood, person, and number, the total lack of semantic transparency of all inflections (none of the flexional markers have lexical value); lack of structural similarity between person/number markers and the semantically related subject pronouns (cf. discussion in the previous chapter on the formal dissimilarity between the 1 s. allophones -o, -m, -ι, -τ and ego)·, and phonological unity of stem and inflections.

The high level of syntheticity of the Latin veib system is, however, counterbalanced to some extent by the notably consistent linearity of the stem morpheme (only a small number of very irregular verbs show allomorphic alternations throughout most of the tenses, e.g., esse 'to be': sum, es, est: imp. eram, subj. sim, etc.); the mostly predictable recurrence of allomorphic alternations (the allomorphic variation of a morpheme is determined by a small set of factors, namely the four conjugation types -a-re, -e-re, -e-re, -/-re);4 the regularity in the set of morphemes corresponding to the three persons and two numbers (thus: 1 s. = -o or -m, 2 s. = -s, 3rd s. = -t, 1 pl. = -mus, 2 pl. = -tis, 3 pl. = -nt) used in all active tense paradigms except the 1 s. - 3 s. of the preterite; the absence of zero morphs

80

Analyticity and Syntheticity

(all person markers are, for instance, given overtly); the consistent ordering of STEM + ... + PERSON MARKER; and, importantly, the separability, displaceability, and paradigmatic transparency of periphrastic constructions of the type laudatus sum Ί have been praised,' laudatus eram Ί had been praised,' laudatus ero Ί will have been praised.'5 4.2.2 The marking ofperson and number on the verb: diachronic perspective 4.2.2.1 From Latin to Old French 4.22.1.1 Postverbal personlnumber marking in CL In CL, person/number (henceforth P/N) markers form an integral part of the verbal core in that they are an obligatory feature of all finite verbs even when P/N is signalled overtly by a noun phrase or a pronominal element (cf. vir cani-t 'the man sings'; ego cano Ί sing'). 6 Active tense paradigms have a largely regular set of morphemes for the three persons and two numbers: 1 s. = -o or -m, 2 s. = -s, 3rd s. = -t, 1 pi. =-mus, 2 pi. = -tis, 3 pi. = -nt (the 1 s. alternation -ol-m is tense- or mood-related [cf. can-o Ί sing' (ind.), caneba-m Ί sang/was singing,' cana-m '(that) I sing' (subj.)7 The preterite forms a system apart, typically having the endings -i, -iste, -it, -imus, -istis, -erunt. The so-called deponent verbs (cf. sequor Ί follow') and passive forms (laudor Ί am [being] praised') have their own set of endings (see paradigm below), but follow a largely regular and recurrent pattern. It is this high measure of regularity and recurrence which makes the structure of the entire system of P/N a relatively transparent one in spite of allomorphic alternations. Typical examples of P/N marking in CL are shown by the following paradigms: Present indicative

Imperfect

Present (passive)

I praise

I lead

I praised/ lied/ was praising was leading

I am praised

I am led

laud-o lauda-s lauda-t lauda-mus lauda-tis lauda-nt

duc-o duci-s duci-t duci-mus duci-tis duce-nt

laudaba-m laudaba-s laudaba-t laudaba-mus laudaba-tis laudaba-nt

laudo-r lauda-ris lauda-tur lauda-mur lauda-mini lauda-ntur

ducxH" duce-ris duci-tur duci-mur duci-mini ducu-ntur

duceba-m duceba-s duceba-t duceba-mus duceba-tis duceba-nt

CL P/N markers are highly synthetic in that besides being obligatory, they are invariably non-displaceable, bound, semantically non-transparent, and phonologically tied to the verb stem.

Verbal core: Person/number 4.2.2.1.2

81

Emphasis on the category of verbal person and the rise of a new set of subject markers

To emphasize or contrast the subject of an action or state, Latin speakers had recourse to an additional, optional set of person markers, the so-called subject pronouns. Apart from the reflexive se, CL used these pronouns only for the 1st and 2d persons (Elcock 1960:77). Since the morphemically free personal subject pronouns 1 s. ego, 2 s. tu, 1 pl. nos, and 2 pl. vos were not normally used except to signal emphasis or contrast (cf. dabo Ί will give'; ego dabo 'It is I who will give'), they were a semantically (and probably also pragmatically) motivated category which— having (a) semantic content, (b) no bound constituents, and being (c) displaceable, (d) invariable (e) and phonologically independent—may very well epitomize what we defined in chapter 3 as highly analytic forms. Popular speech created a 3d person series pronoun by employing the distal demonstrative ille 'that one' (and in some areas the emphatic ipse 'this (one), (him)self ), thus extending the subject pronoun series to all three persons, singular and plural. The Ule forms differed from the other personal pronouns in that they inflected for gender and number. Consequently, the third person subject pronouns entered Romance as morphologically more complex and, therefore, also more synthetic forms.8 At an undetermined point in the history of Latin, a boost in the semantic relevance of these subject pronouns must have caused gradual increase in frequency. Through their subsequent overuse, ego, tu, nos, and vos began to lose their erstwhile expressiveness—pragmatic unmarking—and the demonstrative meaning once contained in ille became effaced. 9 Evidence of such pragmatic unmarking abounds in Latin texts in which, through conscious design or a slip on the author's part into a more colloquial spoken usage, personal pronouns occur at a much higher rate than in formal CL. In these examples, the subject pronouns are no longer motivated by the need for emphasis or contrast (or, in the case of ille, to signal demonstrative remoteness), and they function as little m we than redundant markers of person and number.1® However, this semantic restructuring, extensive as it ultimately turned out to be, came about only gradually. By the time of Old French [OFr],11 the verb could still occur without a subject pronoun; and use of the subject pronouns still sometimes signalled emphasis or contrast, as in the following examples: (1)

(a)

Contrast Quant tu es mor, dulur est que jo Since you are dead pain is that I 'Since you are dead, it is painful for me to live.'

vif live

(Chanson de Roland, v. 2030)

82

Analyticity and Syntheticity (1)

(b)

Emphasis Tu l'entens

'Toi, fais attention' (Aucassin et Nicolette, XV, v. 12, cit. Brunot and Bruneau p. 221)

Despite the continued use of subject pronouns as emphatic markers, the once primarily semantic motivation behind the use of subject pronouns had by the OFr period begun to yield to essentially syntactic considerations. As Foulet (1968 [1919]:par. 446-480) documents in detail, OFr passed through a "verb-second stage," that is, there was a strong tendency not tc place the verb in first position in affirmative clauses. 12 If there was no element (e.g., nominal subject, relative or demonstrative pronoun) to fill first position In the sentence, then this position would be occupied by the appropriate subject pronoun: e.g., Ele set bien ce est la voie 'She knows well (that) this is the way' (Foulet p. 333). As a result of this stylistic constraint, the subject pronoun was now generally present if no other element occupied the preverbal slot, but was absent almost without exception if some other element preceded the verb. This was a very different situation from that of Classical Latin. The generalization of the verb-second pattern clearly introduced a more rigid patterning of the once syntactically free subject pronouns, and increased the frequency of the word-order pattern "subject pronoun + verb." It cannot be ascertained beyond the point of conjecture whether or not the fact that subject pronouns were used without any particular emphasis to avoid "verb first" word order led to their gradual adoption into other constructions as well. 13 What is beyond doubt, however, is that by the time of OFr, the once purely semantic synthesis between subject pronouns and verb had begun to make inroads into the domain of morphosyntax. When we look back at OFr from a ModFr perspective, one of the most striking features of subject pronouns is their high separability from the verb. Unlike ModFr where only unstressed elements can intervene between the two elements,./«, tu, etc. could precede stressed words even when they had no direct semantic relationship with them: e Oliver, en qui IL tant se FIET (Chanson de Roland, 12th c.) 'and Oliver in whom he so much trusts' (for this and other examples see Franzén [1939:35f]). Before temporarily leaving the subject pronouns, I must make brief mention of the new role which the CL accusative pronouns began to assume. As the personal subject pronouns were losing their emphatic or contrastive value, the function of emphasis or contrast on person was gradually taken over by a new set of forms derived from the Latin object pronouns 1 s. me, 2 s. te, 3 s. illi > *illui (masc.), illi > *illaei (fem.), 1 pl. nos, 2 pl. vos, 3 pl. illos (masc.), and 3 pl. illas (fem.)'

Verbal core: Person/number

83

which evolved into OFr moi, toi, lui (mase.). Ii (fem.), nous, vous, els > eus, and eles respectively.14 Though already found in the nominative case in OFr (cf. quant MOI et LI la mer passames 'when she and I crossed the sea* (Price 1971:145), in this function these forms did not become frequent until MidFr. By the 16th century, the original emphatic or contrastive je, tu, etc. had been supplanted almost entirely by moi, toi, etc. In ModFr, moi, toi, etc. no longer function as the grammatical subject of a verb when not accompanied by the corresponding subject clitic (moi je chante, but *moi chante). Although nominative moi, toi, eux can still express emphasis and/or contrast, in informal spoken French they now primarily hold a pragmatic function (see §4.2.3.5.2 below). The gradual rise of the new function of emphasis and/or contrast in the OFr reflexes of the CL accusative pronouns me, te, etc., is then yet another indication that the erstwhile emphatic Latin subject pronouns ego, tu etc. had given up their referential value. Whether je lost its emphatic or contrastive value because moi had begun to assume a similar role, or whether moi became a substitute for a function which je was gradually losing is really a question of the chicken and the egg type. Several scholars (Wartburg [1963:69], Kukenheim [1967:35], Guiraud [1968:6970]) indeed believe that there is a direct correlation between the grammaticization and concomitant loss of emphatic value of the subject pronouns and the rise of the new emphatic forms moi, toi, etc., but so far their claims have remained pure conjecture. Both the increase in frequency of personal subject pronouns, already observable in some Classical Latin texts, and the stylistic "verb second" preference of OFr were no doubt influencing factors in the eventual semantic recasting of forms like ego, tu, etc. It appears, however, that there was also an external factor, namely Germanic influence, which contributed to the semantic reshaping of what were to become the so-called personal and impersonal pronouns je, tu, il, etc. and on, respectively. Kuen ( 1952:150- 152n3) points out that in the Romance languages the regular positioning of a subject clitic in front of a verb is common only to areas known to have been in extensive contact with Germanic languages (northern France, the upper portion of Italy, and the Rhaeto-Romance regions). A Germanic source is universally recognized for the so-called impersonal pronoun on 'one.' Originally deriving from the nominative singular of the Latin homo 'man,' omlon began to be used in Gaul, presumably on the model of Germanic man 'one,' a reduced form of the nominative singular Mann 'man.' 1 ^ Witness the following example from the Chanson de Roland:

84

Analyticity and Syntheticity (2) Co set HOM ben, n' ai cure de menace (1.293) This knows ONE well NEG I have worry of threat 'This is well known, I am not afraid of threats'

While we need not dwell here on the reasons for the adoption of on into French (and no other modern Romance vernacular), it is worth noting that this form underwent greater semantic reanalysis than any other lexeme which eventually evolved into the ModFr subject clitic. Nowhere in the history of the minimum finite verb group is it easier to see that a semantic change ('man' —» indeterminate subject pronoun 'one') was an essential ingredient for the subsequent boost in morphosyntactic and phonological synthesis between verb and adjoining particle. As homo (+ verb) was losing its semantic specificity, thereby gaining lexical generality, its minimalized semantic content allowed it to enter into semantic association with all verbs, and, consequently, to satisfy the OFr "verb second" constraint 16 Although the members of the OFr paradigm of personal subject pronouns had originated from various domains of CL grammar (i.e., from the personal subject pronouns ego, tu, nos, vos, the demonstratives ille, ilia, etc., and the noun homo) and came to form a coherent set in OFr, it is nevertheless possible to discern a common feature in the history of these lexemes, namely the wholesale effacement of semantic specificity. As the following examples demonstrate, the morphosyntax of OFr je, tu etc. continues, however, to resemble that of their Latin ancestors ego, tu, etc. in that these subject pronouns are still (1) non-obligatory (3a), (2) displaceable (3b-c), (3) separable (3d), and (4) morphologically unbound (3e): (3)

(a)

Por vous passerà, le mer

'For you I will cross the ocean'

(Aucassin et Nicolette, XIII, v. 13-14; cit. Brunot and Bruneau p. 219, trans, mine) (b)

ne pensai faire tel perte

Ί didn't plan on making such a loss'

(Béroul, cit. Foulet, par. 166, trans, mine) (c)

Sire, (¿lege. [= je] bien fait?

'Sire, did I do it well?'

(Chréstien de Troyes, cit. Foulet, par. 162, trans, mine) (d)

Et H, a toz ses oz .... s'en ala

'And he, with his whole army, went away'

(Villehardouin; cit. Price 1971:144)

Verbal core: Person/number (3)

(e)

85

Ja fetes vos vos nés garnir? — Gié? 'Are you already getting your vesselsready?— I?' (Enéas, v. 1679-1680, cit. Brunot and Bruneau, p. 219, trans, mine)

What has been said about the structural transparency of the CL personal pronouns can, by and large, be repeated for OFr. As the following table demonstrates, the 3d P/N—being still inflected for person/number17 AND gender—continue to be more synthetic than the other members of the paradigm, and the noted absence of allomorphic alternations in Latin also subsists: Table 2. Subject pronouns in Latin, Old French, and Modem French Latin

OFr.

Mod. Fr. 18

ego tu ille ilia

gié, jo, je tu il ele

homo

om

on

nos vos illi illae

nos vos il eles

nous vous ils elles

je tu il elle

42.22 Postverbal marking of P/N from Old to Middle French In the passage from Latin to Middle French [MidFr], as a result of the normal sound changes of VL, Proto Gallo-Romance and OFr, as well as a number of analogical tendencies, the CL Latin system of P/N markings on the verb underwent numerous changes. Leaving aside, for the moment, the perfectum past (laudav-i) which, as I pointed out earlier, did not comply with the P/N endings of the remaining paradigms, final -o (CL cant-o) and -m (CL cantaba-m) were reduced either to zero (cf. OFr \je] chant) or to schwa (graphically e) when the final vowel was /a/ (cantarti > chante)}** The P/N marker for the 2 s., -s, survives throughout the verbal system (CL cantas > chantes-, CL dormis > dors), and so does 3 s. -t (CL cantal > chantet) prior to the 12th century. Thereafter -t was progressively effaced, first in the present tense of -er verbs (cantai > chante), and subsequently in 3 s. forms (dormet > dort /dor/). Except in the preterite, the 1 pl. CL -mus was replaced by -ons, an analogical formation presumably derived from OFr sons 'we are' (CL

86

Analyticity and Syntheticity

sumus) (Price 1971:178). In the 2 pl., OFr ended up with essentially two, phonologically conditioned forms: -ez (chantez) and -iez, the latter occurring after a stemfinal palatal only (pecchiez < CL peccatis 'you sin'). However, by late OFr, -ez had replaced -iez almost entirely (Price p. 179). Finally, the CL 3 pl. -nt (chantent < contant) may, as Pope (1952 [1934]:340) argues, have remained stable, though it is possible that already in OFr -nt was no longer pronounced. Despite the brevity of our survey of the diachrony of P/N markers from Latin to OFr, it should be clear that even after the phonological reorganization in the suffixes had taken place, the endings remained distinctive enough to assure the functionality of the P/N system. Witness, for instance, the following, not atypical early 12th-c. present tense paradigms in which each person/number marker is signalled either overtly or, as in the case of the 1 sg. of durer 'to last' and dormir 'to sleep,' by a "zero" opposition to the rest of the paradigm:

1.

2. 3. 1.

2. 3.

durer

dormir

(je) (tu) (il)

dur dur-es dur-e(t)

doim dor-s dor-t

(nous) (vous) (il)

dur-ons dur-ez dur-ent

dorm-ons dorm-ez dorm-ent (Pope p. 338)

It is often claimed that the rise of subject pronouns stands in direct relationship to the eventual loss of the P/N inflections shown in the above paradigms. Ewert, for instance, remarks that 'the gradual breakdown of flexional distinctions makes its use [i.e., that of personal subject pronouns] necessary' (1966 [1933]:par. 240). Yet upon closer analysis one finds that during the period in which these phonological markers of P/N were still overtly given on the verb, as many as two thirds of the finite verbs already showed subject pronoun insertion (Price 1971:147). Thus, the rise and spread of the personal pronouns in French predated the fall of the verbal endings, 20 and the highly synthetic verb-final P/N markers began to disappear only after the combination subject clitic + verb had, in fact, itself become relatively synthetic, as we shall see shortly. Verb-final inflections persisted as the basic exponents of P/N throughout not only the OFr but also much of the MidFr period, and, as such, they retained their high level of syntheticity within the verb group (in written French, the distinctive endings have been preserved to this day). Given that the ever more frequent subject

Verbal core: Person/number

87

pronouns duplicated functional information already contained in the verb endings and thus reduced—from the perspective of analyticity—the "optimal" one-to-one mapping of function/meaning and form, it must be concluded that the signalling of P/N during the transitional period from the post- to pre-verbal marking in French was no less synthetic than in CL. 21 Viewed from this angle, the widely held belief that the French verb phrase at some point in its pre-fifteenth-century history passed through a markedly analytic period finds little or no support Although I have argued repeatedly that the function of person/number marking was signalled in highly synthetic fashion within the verbal core throughout CL, Old and much of MidFr, it should, of course, not be inferred that the "new" preverbal forms which came to express this function with increasing frequency did not change their interrelatedness vis-à-vis the verb. While the syntheticity of the verb phrase was being maintained largely on account of the survival of postverbal inflection, the descendents of CL ego, tu, ille, etc. proceeded to move in the direction of greater syntheticity, and by the time the inflectional system began to break down irretrievably during the MidFr period, these forms had—as I shall shortly demonstrate—definitely lost their once distinctly analytic character. 42.23 Postverbal marking of PIN in Middle French To understand the dramatic changes in the structural importance of subject pronouns, we must briefly look at developments that took place in postverbal inflections. In the course of the thirteenth through sixteenth centuries, the phonological oppositions which had previously distinguished the verbal endings were slowly effaced, either through phonological reduction or analogical developments. This wave of changes, and not the phonological developments of Vulgar Latin or OFr, radically reduced the efficacy of the personal suffixes and only the 1 pi. and 2 pi. continued to distinguish crucially both tense and person/number.22 In the absence of any other overt signs, the preverbal markers thus became the only P/N indicators in many cases. This switch from post- to pre-verbal person marking did not, however, affect all tenses of MidFr the same way, as one is often led to believe (e.g., from Price [1971], M. B. Harris [1978b]). Due to the person/number related stem alternations of the irregular auxiliaries être 'to be' and avoir 'to have,' the passé composé continued to signal person/number in redundant—and, therefore, also highly synthetic—fashion throughout MidFr and ModFr. Similarly, in the future tense, phonological reduction did not eliminate the overt postverbal inflections, and in the imperative inflections remained the sole exponents of person/number.23 Witness the following paradigms in which all persons except the 2 and 3 s. have a distinct postverbal inflection (by the late MidFr period, 2 and 3 s. es/est, asla had become homophonous due to effacement of final consonants):

88

Analyticity and Syntheticity Passé Composé (être)

Passé Composé (avoir)

Simple Future

je tu il elle

suis es est est

parti parti parti partie 24

j' tu il die

ai as a a

fait fiait fait fait

je tu il elle

parierai parleras parlera pariera

nous vous ils elles

sommes ¿tes sont sont

partis parti(s, es) partis parties

nous vous ils elles

avons avez ont ont

fait fait fiait fait

nous vous ils elles

parlerons parlerez parieront parleront

1 pl. parlons

2 pl. parlez

Imperative: 2 s. parle 4.2.2.4

Morphosyntactic tightness between subject pronouns and verb in the Middle French period

By the time the personal subject pronouns become the only markers for P/N in all s. and the 3 pi. forms of most tenses, these pronouns had lost much of their analytic character. In particular, they had already (1) attained virtually obligatory status in many instances, (2) lost their morphological independence (i.e., they had become bound forms), (3) given up their syntactic freedom, (4) formed a phonologically coherent unit (i.e., a "cúrseme") with the verb, and (S) lost much of their former semantic transparency. It is these five points that I will now try to document 4.22.4.1 Obligatoriness By the time of MidFr, the OFr "verb second" constraint had become generalized so that the subject pronouns became mandatory in most circumstances.25 By the early 17th century, the deletion of the subject pronoun, as in the following 15th-century example, would thus result in ungrammaticality: (4) Et pour sa beauté et doulce maniere ESTOTT moult desiree and for her beauty and gentle manner SHE WAS much desired 'And because of her beauty and gentle manner she was much desired' (CiL in Rickard 1976:93,1.14) It is worth noting, however, that MidFr subject pronouns never reach the measure of obligatoriness observed earlier for the CL P/N inflections. Unlike the highly

Verbal core: Personlnumber

89

synthetic CL affixes, these clitics were not felt to be an indispensable part of the verb group, and in some cases they were consistently left out: no pronominal marker was used when the subject was expressed nominally (cf. le flateur est anemy de toute vérité 'the flatterer is the enemy of all truth' [Gardner and Greene 19S8:11], trans, mine); imperatives lacked preverbal P/N marking altogether (cf. attendez moi a la fontaine... 'wait for me at the fountain' [ibid, p. S3]); and a P/N marker needed not be repeated in a single clause containing more than one verb (cf. et je m'y consens, et vous prometí espouser le matin 'and I consent that I will marry you tomorrow' [ibid, p. 91]). 4.22.42 The evolution from free to bound form In OFr, we recall, subject pronouns enjoyed morphological autonomy, that is, they could occur as free forms in either stressed or unstressed position. Thus they could occur in any of the following circumstances: (a)

as coordinate subjects: e JO e VOS i irum 'both you and I will go' (Chanson de Roland);

(b)

when subjects are contrasted: TU es trop tendre e IL trop dur 'you are too tender and he too hard'; when separated from the verb: et IL, a toz ses oz ..., s' en alia 'and he, with his whole army ..., went away' (Villehardouin)

(c) (d)

when the verb is not expressed: et qui i sera — JOU et TU 'and who will be there? — you and I* (Froissart)

(e)

with meïsmes 'themselves': et IL meïsmes mengierent... 'and they themselves ate ...' (examples from Price 1971:44-45)

Already in OFr, the original oblique stressed forms moi, toi, lui, li, eus, eles occasionally functioned as subjects (e.g., in a 13th-century text quant MOI et LI la mer passames 'when she and I crossed the sea'). Over the course of the next centuries, the use of these forms became more prominent, so that by the 16th or 17th centuries the original subject pronouns,)«, tu, etc. could no longer occur as stressed forms, and their presence became contingent on the cooccurrence of a coreferential finite verb within the same clause. 26 These elements had, then, changed their linguistic status from pronoun to clitic. 27 These new restrictions had two important consequences. First, the preverbal person/number markers in constructions (a) through (e) above now all had to be expressed by the formerly oblique forms (i.e., MidFr MOI et VOUS irons for OFr JO e VOS irum; MidFr et qui y sera—MOI et TOI for OFr JE et TOI·, etc.). And, second, the linear distance between subject clitic and verb was now perforce reduced signifi-

90

Analyticity and Syntheticity

cantly to a maximum of three (invariably unstressed) syllables (cf. je NE LUI EN donne pas Ί + NEG + to him + of it + give + NEC = I don't give him any [of it]'). 28 4.22.43 Paradigmatic and semantic transparency The lack of autonomy of the erstwhile subject pronouns did not manifest itself to the same extent throughout the paradigm. Because of homophony between some of the formerly oblique forms and the original subject clitics, several of the stressed and unstressed P/N markers coincided, as shown by the underlined forms in table 3 below: 29 Table 3. Unstressed and stressed subject pronouns in Middle French Unstressed

Stressed

je tu il ells

moi toi lui eile

on nous vous ils 3 0 elles

-

nous vous eux elles

The development of unstressed subject pronouns from free to bound form was unquestionably the major step in the direction of greater morphosyntactic tightness between the verb and the original subject pronouns. There was, however, another development, namely the loss of paradigmatic transparency, which caused the original pronouns to become more synthetic. As can be extracted from the above table, there is very little phonological similarity between the allomorphs jelmoi, tu!toi, il/lui, and ils/eux. Since throughout the Old and MidFr periods moi, toi, etc. could function as the (stressed) grammatical subject of a verb without having to be accompanied by the corresponding unstressed pronouns (cf. MidFr moi et li la mer passames Ί and she the ocean crossed = she and I crossed the ocean' [Price p. 145], but ModFr moi et lui NOUS avons passé la mer/*moi et lui avons passé la mer), the alternations created by the rise of a new set of stressed subject pronouns accounted for a significantly more complex system of P/N marking. Looking at MidFr, we find that P/N could be signalled in one of the following ways, in order of importance:31

Verbal core: Person/number (1) (2) (3) (4)

91

through a preverbal subject clitic (il chante /ilsat/) through strictly postverbal inflectional marking (chantez! /Sate/ 'sing!') through a stressed preverbal subject pronoun (lui chante) through stem-internal flection ([je] suis).

Allomorphic alternations were, however, not the only factor responsible for the decrease in the transparency of MidFr P/N marking. One of the most dramatic effects of the split between stressed and unstressed forms was the reduction of semantic transparency in the unstressed preverbal clitics. No longer capable of standing in isolation, je, tu, il, ils and on were now functionally downgraded to mere grammatical markers whose semantic motivation had essentially become blurred. Though still different in a number of respects, the MidFr subject clitics definitely began to be as synthetic as the CL inflections -o, -s, etc. (cant-o, canta-s). 4.22.4.4 Syntactic constraints The syntactic constraints to which the je, tu, etc. were subjected by the 16th century w o e the result of a number of independent developments which are best seen against the background of the pan-Romance trend towards a more rigid word order. In CL the speaker had relative freedom in the way the arrangement of sentence elements, with word order presumably expressing discourse functions. As Romance lost much of its inflectional system, word-order became increasingly more important for signalling grammatical relations within the sentence. Already in OFr there was appreciably less flexibility than in Latin, although the rigid subject-verb order of ModFr, for instance, had not yet been reached. At any rate, looking back from the MidFr period, one of the most important changes in the syntactic freedom of subject pronouns took place when OFr came to depend primarily on the use of verb-subject inversion to signal interrogation (cf. est-eie ivre? 'is she drunk?'; est morte m'amie? 'is dead my girl friend = has my girl friend died?' (for these and other examples see Foulet (1968 [1919]:par. 336-337).32 This particular interrogation pattern was commonly used for several centuries, only to be rivaled by an alternative construction, est-ce que, as early as the 12th century: que est iço que est avenud a Said (Price 1971:267) 'what is this which (has) happened to Saul?'. 33 First used with partial interrogation, and later also with total interrogation, this stock formula became the standard means of signaling questions by late Middle French. Speakers' preference for the use of est-ce que over S-V inversion resulted in a much less frequent movement of subject clitics from preverbal into postverbal position. This, in turn, further cemented the already semi-formulaic SUBJECT CLITIC + VERB combination, and reduced the commutability of je, tu, etc., to a few, rather insignificant archaizing constructions.34

92

Analyticity and Syntheticity

In the domain of interrogation there was yet another development which considerably affected the tightness between subject clitic and verb. The S-V inversion remained productive despite the noted predominance of est-ce que, but underwent a change which, although rooted in OFr, did not become widespread until MidFr. When a nominal subject was part of a total interrogation, in MidFr a (»referential clitic had to be placed immediately after the verb: the OFr example cited above est morte mon amie would, in MidFr, have to be expressed as est-ELLE morte mon amie? ¡mon amie, est-ELLE morte?35 While the rise of this pattern reinforced the transportability of subject clitics and may consequently have slowed down the fossilization of the syntactic formula CLITIC + VERB, it undoubtedly fostered the (now nearly obligatory) cooccurrence of these two elements, and perhaps provided the model for the similar, now frequent repetition of a P/N marker in sentences containing a nominal subject (cf. ISF il vient ton pèrelton père il vient 'your father is coming' [see §4.2.3.2.4 below]). Among the various developments which contributed to the overall fixing of word order in French, none was perhaps more consequential than the cliticization of je, tu, etc. As these items lost their ability to occur in stressed position and became phonologically bound to the verb, their syntactic freedom became extremely limited. Having lost much of their commutability because of the trends we have observed in the domain of interrogation, and being no longer separable from the verb through insertion of stressed elements, subject clitics were now essentially restricted to a preverbal slot which, though not always immediately juxtaposed to the verb (cf. je ne parle pas), had to be within its immediate vicinity. 4.2.25 Intermediate summary We have seen that well before phonetic attrition had undermined the functionality of the inflectional P/N system, the use of preverbal pronouns had become firmly established. Although at least three factors—separability, displaceability, and incomplete grammaticization—continued to make the relationship between subject clitics and verb a relatively analytic one, many of the traits associated with synthesis were already present by the time (late MidFr) je, tu, etc. in numerous instances became the sole exponents for P/N. Thus we have witnessed that subject clitics (1) could no longer occur as free forms, (2) were obligatory in many instances, (3) became severely limited in their syntactic mobility, (4) had given up their phonological autonomy (i.e., they were unable to receive stress and had to be phonotactically integrated with the verb), and (S) failed to maintain semantic transparency (i.e., they had evolved from lexical to grammatical morphemes). While these observations should suffice to defend my earlier claim that in the transitional period of MidFr P/N marking was never distinctively analytic, there are

Verbal core: Person/number

93

other arguments which bespeak the relative syntheticity of the system. As observed previously, in several instances phonological reduction failed to obliterate the highly synthetic postverbal inflectional morphology. The future tense, the passé composé (both the je suis + VERB and the j'ai + VERB types), the imperative, and the 1 and 2 pi. forms of all verbs continued to mark P/N overtly. Also, some of the so-called analytic subject pronouns were far less analytic than is generally claimed. VL ille, illi, illae etc. were porte-manteau morphemes which, expressing GENDER/NUMBER/PERSON in alinear fashion, maintained their distinctively synthetic character during the entire history of Old and MidFr. By arguing that P/N marking gave the verbal complex a relatively synthetic character in both OFr and MidFr, I am, of course, not trying to deny that the highly analytic relationship between the CL pronouns ego, tu and the verb evolved in the direction of greater syntheticity. Nor am I refuting that CL -o, -s, etc. are relatively synthetic when compared to MidFr je, tu. What I am trying to oppose is the widespread belief that MidFr subject clitics were analytic units. It is crucial to understand that the RELATTVE analyticity of MidFr je, tu vis-à-vis CL -o, -s cannot— as is so often done—be translated into an "absolute" polar opposition analytic vs. synthetic. MidFr je, tu, and particularly alinear il, elle, ils, or elles, are somewhat analytic when compared to CL -o, -s, -t, but quite synthetic in comparison to the morphologically free Palenquero yo, bo, and ele. By the same token, it is incorrect to claim that after the (partial) loss of inflectional endings the French verb phrase was analytic. It is, however, correct to say that in the transition from post- to pre-verbal marking the French verb phrase did become more analytic—or, better, less synthetic. When speaking, say, of analytic MidFr il chante or synthetic CL canto we must keep in mind that these concepts are relative. 4.2.3 Syntheticity of "subject clitics + verb" in modern informal and formal spoken French The overall impression which emerges from the previous summary is that the descendents of ego, tu, ille, etc.—by all accounts semantically, syntactically and morphologically autonomous units as late as the 13 th century—lost their independent existence as "words" and underwent grammaticization by the end of the MidFr period. 36 As the following discussion will bear out, this process continued into present-day French, so that the relationship between subject clitics and verbs has by now gravitated toward the synthetic end of the analytic/synthetic scale. Before we proceed, a few theoretical and terminological questions must be addressed. The fact that the history of French subject pronouns has—by necessity—

94

Analyticity and Syntheticity

been based exclusively on written data should induce caution about some of the claims made about the diachrony of person/number marking in Old, Middle, as well as Modern (spoken) French. Although numerous features of spoken French are undoubtedly of fairly recent vintage, it cannot be assumed that some of the "innovations" to be discussed here were not already prevalent during the MidFr period. For this reason the reader is urged to interpret the subsequent data, especially those of informal spoken French, from an essentially synchronic perspective. In discussing modem French, we will look primarily at non-written varieties, informal and formal spoken French (ISF and FSF respectively), variously referred to in the recent literature (e.g., Lambrecht [1981] and [1986], Ashby [1977]) as "standard" or "formal" vs. "non-standard French," "français populaire," "colloquial French," etc. It is not always possible to draw a clear line between these two. While one speaker may label a given construction as simply colloquial, another may characterize it as substandard. That the ISF/FSF distinction is justified, despite this occasional fuzziness, is argued convincingly by Lambrecht: one striking example confirming the intuitive validity of the SF/NSF distinction is the case of the lpl subject pronoun .... Virtually every speaker of modern French knows that the first person plural in the French verb paradigm is expressed by the subject pronoun nous followed by the verb stem to which is attached the suffix -ons .... Thus every speaker has learned that the verb parler is conjugated je parle, tu parles, illelle parle, nous parlons, vous parlez, ils/elles parlent, and will be able to reproduce this paradigm upon request. However every speak» of French also has the 'implicit' knowledge that the first person plural is in fact on parle, which form he will almost invariably use in actual speech unless special circumstances obtain. (1981:5)37 Unless noted, grammatical features (in FSF or ISF) analyzed here are not geographically limited within the French-speaking community. Though regional and sociolinguistic differences exist, the properties of interest here have wide cross-dialectal application. Following Ashby (1977), I assume a priori that, historically, FSF has been more conservative than ISF. However, no diachronic link between FSF and ISF should be inferred. Though the replacement of "nous + VERB+ons" by "on + VERB" (nous parlons —» on parle) is a "fact" of ISF, it cannot be assumed automatically that on parle will some day oust nous parlons from FSF (the arbiters of bon usage in French being notoriously conservative), nor that nous parlons (vis-àvis earlier parlons) was at one time considered substandard.38

Verbal core: Person/number

95

42.3.1 The final "desyntacticization" of subject clitics in ISF The order and position of subject clitics has now become fixed in French. An important innovation of ISF that maries a further step towards total synthesis of clitic + verb is the elimination of virtually all S-V inversion. In MidFr, we recall, S-V inversion was possible in a number of circumstances. In FSF one still finds them in yes-no questions (Sa), partial interrogation (5b), and after certain adverbs (5c): (5)

(a) (b) (c)

As-tu mangé? Quand pars-tu? Ainsi est-il arrivé.

'have you eaten?' 'when are you leaving?' 'thus he arrived'

ISF rarely marks yes-no questions by inversion, but rather by rising intonation (5d). Similarly, in ISF partial interrogation the subject clitic remains in the preverbal slot and the Q-word is placed in sentence-initial or sentence-final position (5e):*> (5)

(d)

(e)

Tas mangé?40 Γ comprend? Tu pars quand τ41 Quand tu pars? Quand c'est que tu pars?

'have you eaten?' 'does he understand?' 'when are you leaving?'

One consequence of the innovation of the new patterns (5d-e) is a decline in frequency in ISF of other interrogative strategies, notably est-ce que, a syntactic device often used in FSF to avoid the more formal inversion strategy. As the following examples show, ISF maintains a number of interrogation devices in addition to rising intonation, but in all of these, the subject clitic remains before the verb: 42 (6)

(a)

Elle vìent-ti?

(b) (c)

Quand est-ce qu'il est arrivé? Quand ¡¿esl qu'il-est arrivé?

(d)

Quand que vous venez?

'she is coming+Interrogative = is she coming?' 'when did he arrive?' 'when did he arrive?' (Lambrecht, p. 33) 'when are you coming?' (Price 1971:270)

In (6a), an interrogation particle ti, unknown to FSF, has been created. From 3 s. interrogative forms such as ton père, vient-il? and parle-til?, the segment /ti/ has been extracted and grammaticized for some speakers into a marker of yes-no in-

96

Analyticity and Syntheticity

terrogation unmarked for gender or person. 43 In (6c), the est-ce segment of (b) occurs in its non-inverted form c'est. Whether this constitutes a reinversion of est-ce que (and is, therefore, yet another sign of the trend towards non-inversion of subject and verb) or a new, unrelated interrogation pattern is difficult to assess. Finally, type (6d), also a specifically ISF feature, is avoided as a vulgarism by many who would use the i'ai mangé? type (Price 1971:270). As regards S-V inversion after certain phrase-initial adverbs (e.g., ainsi, aussi, peut-être), the ISF trend is clearly toward elimination of such structures (cf. ISF il va venir lui aussi?; comment tu vas? for comment vas-tu? 'how are you?'). 44 The preverbal positioning of subject clitics has eliminated the possibility of displacingye«), t(u), etc. Just as the suffixal P/N markers of CL invariably occupied the final position in the (inflectional) verb complex, ISF clitics now invariably occupy the FIRST SLOT in the verb complex. This rule applies even when other veibbound elements (i.e., object clitics and/or the FSF negative particle ne) form part of the verb phrase (cf. t' y vas, literally 'you there go' = 'you're going' but not *y tu vas). 4.2.32 Separability of subject clitics and verb in FSF and ISF 4.23.2.1 Separability of subject clitics and verb in FSF In OFr, the preverbal subject pronoun was separated from the verb when the negator ne (or its stressed variant non) was present (cf. OFr jo non/ne ferai Ί won't do [it]'). This particular word order still survives in FSF, though it is well known that the predicate negation pattern has become obligatorily bipartite (cf. je NE parle PAS). In ISF, the redundant, semantically empty ne has been virtually eliminated. In the corpus of spoken data analyzed by François (1974), ne is dropped in over 95% of its potential occurrences. Other ISF texts show less frequent deletion (see Ashby 1977:62), due in part to differences in speakers' ages (younger informants tend to delete more often). At any rate, the trend is clearly toward elimination of ne, which naturally decreases the linear distance between subject clitics and verbs. A close correlation has been suggested between the elimination of ne in ISF and the current status of subject clitics (M. B. Harris 1978b:118; Ashby 1977:62). The ousting of ne, Harris argues, indicates that the verbal complex is striving towards greater morphological tightness, i.e., a single unit composed of the verb plus a bound préfixai P/N marker. However, the elimination of ne in Occitan (cf. sabe ges/pas '[s/he] knows not = s/he doesn't know'), where subject pronouns have never been obligatory, strongly suggests that the development of préfixai P/N morphology and the decline of the preverbal negative particle are not causally related (see Schwegler 1983:307f).45

Verbal core: PersonJnumber

97

Interestingly, in ISF the trend toward greater inseparability between subject clitics and the verb has been reversed in a number of instances. The major obstacle to the complete bonding of subject clitics and the verb in ModFr has been the possibility of inserting other clitic elements between the subject clitic and the verb (cf. FSF tu me le donnes 'you + to me + it + give = you give it to me.' As the following list shows, interposable clitics are not numerous.46 Due to their combinability, the linear distance they create between je, tu, etc. and the verb is nevertheless considerable (the possible syntactic combinations among the various clitics are limited to those contained in the same rectangle; thus me le is a valid sequence, *me lui, *te leur, etc. are not): Figure 1. Combinability of object pronouns in FSF (from Sandfeld 1970 [1928]:3)

me te se nous vous

le la les

lui leur

Compare the following valid and invalid FSF examples (from Lambrecht 1981:360: (7)

(a)

Il me la présentera.

(b) Il la lui présentera. (c) *Il me lui présentera. (d) *Il me leur présentera.

'He + to me + her + introduce + FUTURE = He will introduce her to me' 'He will introduce her to him' 'He will introduce me to him' 'He will introduce me to them'

In FSF, the ungrammatical preverbal placement of luitleur in (7c/d) must be circumvented by moving the indirect pronoun into the stressed postverbal position: (7)

(c') Il me présentera à lui/elle. (dO II me présentera à eux/elles.

Though all pronominal elements in the above diagram could occupy a preverbal slot already in OFr, their sequential order was not fixed, nor was their frequency preverbally comparable to what it is in modFr. While the changes in the relative ordering of direct and indirect object are not relevant hoe, inasmuch as they do not

98

Analyticity and Syntheticity

affect the actual distance between subject clitic and verb, changes in thefrequency of preverbal positioning are important to the question of syntheticity. In OFr, emphatic object pronouns could follow the verb so that both OFr el(l)e aime mei 'she loves ME* and el(l)e m'aime were possible. 47 Because these OFr stressed pronouns were never placed between subject clitic and verb, the presence of an object pronoun did NOT automatically lead to the linear detachment of subject pronoun and verb, as occurs in both FSF and ISF. Since the OFr period, the tendency has been to prepose object pronouns (except in affirmative imperatives: cf. mange-le! 'eat it!'), so that in unemphatic sentences the elle m'aime pattern eventually became standard. Today FSF emphatic object pronouns can still follow the verb (see [8b] below), though in most instances they require a (»referential preverbal clitic:48 (8) (a) (b) (c)

SF tu me le donnes tu me le donnes à moi (?) tu le donnes à mofô

Similarly, in OFr two stressed object pronouns (9a) or a stressed pronoun plus a noun (9b) could be placed after the verb without a correlating preverbal (unstressed) pronoun: (9)

(a) OFr Dius beneie vous et lui 'God bless you and him' 5 0 (b) OFr Je haz li et sa compaignie Ί hate her and her followers' (cit. by Ashby 1977:74, trans, mine)

As in the previous instance, FSF again requires unstressed coreferential object clitics (italicized in the examples below), so that the OFr examples given in (9) would be rendered in FSF as in (10): (10)

(a) (b)

Dieu vous bénisse, vous et lui Je les. déteste, elle et toute sa compagnie

There is evidence that the FSF trend toward use of coreferential object clitics— and, consequently, the separation of subject clitics and verb—is receiving additional impetus from three ISF innovations: (1) combinations of previously non-combinable clitics, (2) reinterpretation of the pronominal adverb y, and (3) placement of additional coreferential object clitics.

Verbal core: Person/number

99

4.23.2.2 Combinations of previously non-combinable clitics As shown in Figure 1 above, the combinability of object clitics in FSF is fairly limited. In ISF, this combinability has been greatly increased. Compare the parallel FSF and ISF sentences in (11), noting that the permissible clitic sequences of ISF cancel the FSF requirement for a prepositional complement cfter the verb: 51 (11)

ISF

FSF

(a) Tu TE ME rappelles, TuTErappelles(de)MOI, quand f avais 20 ans quandf avais 20 ans 'You remember me when I was 20 years old' (b) Je VOUS ME montrerai Je ME montrerai à VOUS Ί will show myself to you' (c) ¡I ne faut pas que je ME LUI casse. Il ne faut pas que, pour LUI, je ME casse. Ί must not break my neck for him' (Ashby 1977:76; trans, mine) 42.3.23 The reinterpretation of the pronominal adverb y in ISF In FSF, lui or leur—the two least combinable pronouns—can function only as indirect object, i.e., dative clitics (with animate referents). When encoding oblique (animate) complements pronominally, FSF must have recourse to a postverbal prepositional phrase. In the (b) sentences in (12), the relationship between penser and its complement is conceived as oblique. As a result, preverbal lui (12d) is ungrammatical: (12)

(a) (b) (c) (d)

Il parle à son frère Il pense à son frère Il LUI parie *Il LUI pense —> (e) Il pense à LUI

'he talks to his brother' 'he thinks about his brother' 'he talks to him' 'he thinks about him'

The distinction between indirect and oblique complements does not apply to the inanimate clitic y, so that if the complement of the verb penser is animate, y replaces lui, yielding grammatical (13b). Note, however, that the [-animate] constraint on FSF y renders (13d) ungrammatical (or at least derogatory): (13)

(a) (b) (c) (d)

FSF II pense à sa voiture FSF II y pense FSF II pense à son firère FSF *Il y pense

'he thinks about his car' 'he thinks about it' 'he thinks about his brother' 'he thinks about him'

100

Analyticity and Syntheticity

An important innovation has taken place in ISF, whereby y can replace not only oblique ([¿animate]) but also ([±animate]) indirect complements, thus competing in the latter function with lui and leur (as well as with the dative or indirect object clitics me, te, vous).52 Compare the FSF and ISF examples in (14): (14)

Il pense à son (a) FSF (b) ISF

frère

II pense à lui *Il y pense

II pense à ses frères II pense à eux *Il y pense

II y pense

II y pense

The development just described has the interesting result that in ISF, where certain constraints on the combinability of non-subject pronominal markers have been lifted, the possible sequences of object pronouns are now those shown on the left in Figure 2: 5 3 Figure 2. Combinability of object pronouns in ISF and FSF informal French formal French me te le se la y nous les vous

me te se nous vous

le lui la leur les

Only clitics contained in the same square can be combined and they must appear in the linear order of the diagram. With the substitution of y for lui/leur in ISF, the constraints on possible combinations of the two sets shown on the right are cancelled and the three sets can be freely combined (provided that no two clitics with the same case function appear, a constraint that is independent of the issue presented here). Contrast the following FSF and ISF examples: (15)

(16)

FSF (a) Il me φ) Il MR (c) *Il ME LUI/LEUR

présentera à Pauliaux élèves présentera à LUI/EUX présentera

ISF (a) lime (b) Il M! IL

présentera à Paul/aux élèves présentera (à lui/à eux)

'He will introduce me to Paul/the students' 'He will introduce me to him /them*

Verbal core: PersonJnumber

101

The substitution of postverbal object pronouns such as (à) lui/eux by preverbal y in ISF has increased both the linear distance and the frequency of separation between subject clitic and verb. This innovation seems to suggest that despite the long trend towards the obligatory use of subject clitics (see §§4.2.3.2.4-4.2.3.3.3 below) speakers of ISF do not yet interpret the complex "subject clitic + verb" as a largely inseparable (synthetic) whole. This impression is further buttressed by the observation that the frequency with which subject clitic and verb become separated has also increased significantly in FSF on account of several other factors, namely: (1) obligatory conferencing of object or dative pronouns in constructions like tu ME le donnes à MOI or Dieu VOUS bénisse, VOUS et lui, (2) the replacement of intolerable FSF sequences such as me lui, te lui, me leur, te leur, and (3) the combining of previously non-combinable clitics (e.g., ISF je VOUS ME montrerai). Separability being one of the prime determinants of analyticity, it might seem, at first, that these—apparently fairly recent—changes are interfering with the historically strong overall trend of French preverbal P/N markers to attach themselves directly to the verb. But as I will try to elucidate in the following section, a more careful analysis reveals that the increasing insertion of object or dative clitics between subject clitics and verb is not pushing the verbal complex of French towards greater overall analyticity. 4.2.3.2.4 The placement of preferential object (dative, oblique) and locative clitics Subject clitics and verbs are being separated increasingly because coreferential object and locative clitics are inserted when object or locative noun phrases are emphasized or topicalized. In the following examples, the uppercase clitics are coreferential with the (right-dislocated) topics, likewise in small caps: (17) (a) Object clitics (i) Jean V achète, LE LIVRE. (p. 78) John COREF (acc.) buys THE BOOK 'John buys THE BOOK' (lit. John buys it, the book) (ii) PIERRE j[L]UIai donné le livre hier (p. 59) PETER I- COREF (dat.) have given the book yesterday Ί gave the bode to PETER yesterday* (lit. Peter, I gave him the book yesterday) (iii) Pierre i- mYprésentera, À TON ONCLE (p. 37) Peter he me COREF (obi.) will introduce TO YOUR UNCLE 'Peter will introduce me TO YOUR UNCLE* (lit. Peter, he will introduce me to him, your uncle)

102

Analyticity and Syntheticity

(17) (a) Object clitics (cont.) (iv) IIEN bois tous les jours, DU VIN he COREF (acc.) drinks every art day PART WINE 'he drinks WINE every day'

(p. 85)

(lit He drinks [some] every day, wine [that is]) (b) Locative clitics (i) LA PLAGE i- faut Yaller quand i- fait chaud THE BEACH one must COREF (loe) go when it is hot' 'You have to go TO THE BEACH when it's hot out* (p. 53) (lit. The beach, one must go there when it is hot out) (ii)

Il EN revient aujourd'hui, DE LA PLAGE54 he COREF (loe) returns today FROM THE BEACH 'He will come back FROM THE BEACH today* (lit He is coming back from there today, [from] the beach)

Having analyzed a large corpus of spoken French, Ashby (1977:75) finds that these types of coreferentiality are not yet frequent in FSF, but exceedingly common in ISF. François (1974:433) claims that in her ISF corpus, object clitics are often used as redundant coreferential markers for nominal topics. Similarly, Ashby (1982:33) calculates that in the ISF materials he recorded in Tours, 47% of all lexical object NPs are accompanied by a coreferential preverbal clitic. The apparently increasingly common use of coreferential object and locative clitics has prompted a number of linguists (e.g., Heger 1966, 1982b, Rothe 1966, Llorente/Mondéjar 1974, Bossong 1979,1981) to argue that in addition to the wellrecognized subject conjugation, French (and other Romance languages)55 also possesses an "object conjugation" not unlike that known from the more "exotic" synthetic verbal complexes of Basque, Georgian, Quechua, Bantu, etc. 56 According to such an analysis, a sentence like Fr il la lui a donné 'he gave it to him/her' (which includes as many as three incorporated clitics) consists of a single grammatical unit, i.e., a verb. Independent, coreferenced nomináis are then regarded as expansions of the function indicators already marked on the verb, as shown graphically in the following attested, often cited example (Tesnière 1966 [1959]: 175; Heger 1982b):57

Verbal core: Person/number

103

(18) expansion 1 expansion 2 expansion 3 ι 1 il la lui adonnée, à Jean, son père, sa moto COREFERENTS has given to John his father his motorcycle 'his father gave the motorcycle to John' Compare also examples (17a[iv]) and (17b[i])> repeated here as (19), with expansions marked: (19)

^ (a)

· , expansion 1

Il en bois tous les jours,

^ du vin

expansion 1 (b)

La plage

i-faut

y aller quand i-fait chaud

The question of whether we view—as I think we should—all coreferential object and locative clitics as an integral, i.e., incorporated, part of the verb is directly relevant to the problem of analyticity/syntheticity. If we accept the notion that, in addition to an already well-developed subject conjugation (see §4.2.3.3 below), several Romance languages have evolved an object conjugation which forms an integral part of the verb, it is no longer logical to argue that the increasingly frequent incorporation of these agreement markers (and the resulting detachment of the semantically relevant P/N markers je, tu, illelle, etc. from the rest of the verbal complex) decreases the syntheticity of the French verbal complex (see my remarks in this regard in §4.2.3.2.3 above). Quite the contrary, it seems more sensible to argue that the transition of various types of object clitics to incorporated inflectional markers means that French is reaching, or has perhaps already gone beyond, the level of syntheticity encountered in highly synthetic verb systems such as that of Classical Latin. 5 8 42.33 The further spread of subject clitics in FSF and ISF 4.2.33.1 The rise of coreferential subject clitics in ISF There is strong evidence from ISF that the tendency to incorporate coreferential object clitics within the verb is accompanied by the rise of yet another type of coreferentiality. Lambrecht (1981 and 1986) analyzes the discourse functions and formal properties of the characteristically ISF sentences given in (20b-c):

104

Analyticity and Syntheticity

(20)

(a) Les Romains sont fous (b) Les Romains ils sont fous (c) Ils sont fous, les Romains

'The Romans are crazy'

He suggests (1986:316ff) that the choice between strategies (a) and (b)/(c) is to a large extent determined by the degree of topicality the referent of the NP has in the discourse. 59 This difference in topicality generally correlates with the difference between foregrounded and backgrounded parts of the discourse, S V(0) constructions (type [20a] above) typically having low topic status (lower topicality of a referent entails that the referent is a less salient protagonist in the discourse). 60 The pragmatic devices shown in (20b-c), on the other hand, tend to be used as foregrounding devices by which referents are singled out as more salient participants in the discourse.61 Additional examples of sentences containing topic or what Lambrecht (1981) refers to as "antitopic" structures (left- and right-dislocations respectively) include (arguments and their «»referential clitics are underlined): (21)

(a)

Topic

(i) L¿ garçon il attend devant la porte (p. 61) The boy COREF waits in front of the door 'The boy is waiting at the door' (lit. The boy, he is waiting at the door) (ii) Mon portemonnaie il est devenu lourd (p. 65) My wallet COREF has gotten heavy 'My wallet has gotten heavy' (lit. My wallet, it has gotten heavy) (iii) Li cimetière. où il est, f vous prie? (p. 66) The cemetery where COREF is I you ask 'Where is the cemetery, please?' (Lit. The cemetery, where is it, please?) (b)

Antitopic (see n.61) (i)

¡I attend devant la porte, li garçon (p. 84) COREF waits in front of the door the boy 'The boy is waiting at the door' (lit. He is waiting at the door, the boy)

(ii)

l s' y intéresse pas, mon frère (p. 82) COREF REFL in that is interested NEG my brother 'My brother is not interested in that' (lit. He is not interested in that, my brother [that is])

(iii)

Il a voté pour Giscard, C£L imbécile (p. 88) COREF has voted for Giscard this jerk 'That jerk voted for Giscard' Oit. He has voted for Giscard, the jerk)

Verbal core: Person/number

105

Lambrecht (1981) convincingly demonstrates that we must distinguish (at least) two competing basic sentence types in ISF, "topic" (including antitopic) sentences and "subject" sentences, distinguished by the presence or absence of (»referential subject clitics. 62 He does not, however, speculate about the relative frequency of the three sentence types, but merely notes (on several occasions) that subject sentences (i.e., those without (»referential clitics) are relatively rare in ISF, an observation which is confirmed by François' corpus of spoken data (see esp. François 1974:356). Taking Lambrechts monograph as point of departure, Ashby (1982) attempts to answer just that question. By analyzing an extended corpus of spoken French, he finds that over 50% of 3rd person verbs exhibit a coreferential subject clitic (Ashby's claim has recently found further support in Lambrechts own 1986 statistical analysis).6^ Ashby's study reports a revealing correlation between demographic groups and frequency of subject pronoun coreferentiality: the use of coreferentiality is markedly higher among younger, lower class speakers—most likely the vanguard of linguistic change, thus supporting my earlier statement that ISF has now gone another step further in making use of pronominal coreferentiality. If this practice of including coreferential subject clitics spreads further and generalizes to all sentences containing a nominal subject, the entire set of preverbal P/N markers will not only move toward an even higher level of syntheticity, but will also undergo an important change in morphological stature. Because of their boundness and non-obligatoriness in the 3 s. and pi. with a lexical NP subject, il, elle, etc. have appropriately been ranked as "clitics" within the organizational hierarchy of the language. However, as we have just witnessed, the synchronic data of ISF suggest that colloquial French is approaching a stage where ALL verb stems (except imperatives) will attach preverbal P/N markers, and that these morphemes will occur obligatorily in the appropriate context. Should such tightness between P/N morphemes and the verb indeed be reached, the currently still clitic je, tu, il, etc. would then Γα the definition of inflections (see §3.3.1.3). 4.233.2 The subject inversion construction and the dummy subject clitic i(l) There exists in ISF a construction which is known in the literature as subject inversion. Employed as a presentational or event-reporting device, it serves to mark an NP grammatically as a non-topic (Lambrecht 1986:3000- Given the constraint in French against initial position for verbs, in most instances subject-verb inversions are preceded either by the dummy subject morpheme il (22[a] below) or by a non-subject constituent (e.g., qu[e], quand, comment, or adverbial phrases such as tout de suite, ici, etc.) which always lies outside the minimal clause (cf. quand and où in [22b] below). Witness the following examples:

106

Analyticity and Syntheticity

(22) (a) Subject-Verb inversion with preverbal il (i) un beau soir IL' descend une de mes VOISINES one nice evening comes down one of my (female) neighbors 'one evening one of my (female) neighbors drops in' (ii) ...après IL' meurt son FILS, IL' meurt sa BELLE-FILLE afterwards dies his son dies his daughter-in-law 4 ... (and) then his son dies, his daughter-in-law dies' (iii) Γ

va être construit deux IMMEUBLES. FUT be constructed two buildings (Lambrecht 1986:303) 'there are going to be two buildings built'

(b) Subject-Verb inversion without preverbal il (i)

i' travaille puis QUAND arrive le MOMENT DES VACANCES he works then when arrives the moment of vacation Γ dit ah he says ah 'he works and then when vacation time comes he says ah ...'

(ii) c'est OU habitait le DIRECTEUR DE CHEZ GARDY-WATTEZ that's where lived the DIRECTOR OF GARDY-WATTEZ 'that's where the manager of Gardy-Wattez used to live' Since the actual usage of these subject-verb inversions is complex, I will limit myself here to a few summary remarks.64 According to Lambrecht (1986:3000. in ISF the construction is constrained both syntactically and pragmatically to such an extent that it has a relatively low frequency, at least when compared to subject-veib inversions in Spanish or Italian. Concerning the occurrence of the subject-verb inversion involving the dummy subject i(l) (22a-c), it is important to observe that despite a superficially similar structure between this construction and that found in the topic or antitopic structures discussed in the preceding section (§4.2.3.3.2), there exist a number of fundamental differences—syntactic and pragmatic—between them. 65 Lambrecht reports that his analysis of several corpuses of ISF suggests that the (/-inversion construction is restricted to verbs 'which are inherently non-agentive and/or non-volitional, or which are pragmatically interpreted as non-agentive/nonvolitional (1986:304).' 66 Considering that such verbs can also appear in subjectverb inversions not containing the dummy subject clitic (cf. [22d]), it cannot be determined from the literature presently available exactly what syntactic and/or pragmatic factors trigger the inclusion (or deletion) of anaphoric il.67 This lacuna,

Verbal core: Personlnumber

107

plus the absence of statistical data on the relative frequency of both subjectinversion with and without the dummy marker i(l) makes it impossible to assess whether the frequency of dummy il in subject-inversions is presently increasing. Given the observed general tendency of French toward obligatory collocation of SUBJECT CLITIC + VERB, it would not be surprising, however, if such a development woe indeed taking place. 4.2333

The obligatory placement of subject clitic + stem

That the ISF subject clitics are more and more felt to be an indispensable part of the verb phrase is also seen from the repetition of these elements where their inclusion was previously optional (cf. et je m'y consens, et vous prometí espouser le matin 'and I consent that I will marry you tomorrow,' §4.2.2.4.1 above). Except in the imperative, FSF finite verbs rarely occur without an immediately preceding subject; clitics are generally repeated with every new finite verb (cf. ISF moi / mange et f bois beaucoupJ.68 4.23.4 Analytic tendencies in PIN marking of FSF and ISF In spite of the relatively long and vigorous trend toward obligatory preverbal P/N marking, subject clitics are still deleted in a small but significant number of cases (other than those already mentioned for nominal "subject sentences" [see §4.2.3.1 above]). Best known among these are relative clauses with qui 'who' (cf. le père qui est venu 'the father who came' vs. ungrammatical *le père qu'il est venu (but see n70 below). 69 From the standpoint of ISF this behavior may be considered conservative or anomalous. According to Givón the absence of 3 p. subject clitics in relative clauses is apparently not unusual if viewed in the context of universal tendencies. He explains: Ά personal pronoun is rarely present in a position relativized if that position is a basic subject one* (1979:320).70 No appeals to universal tendencies can be made in the case of the ISF deletion of clitics with "verbes impersonnels" il faut 'one must,' il paraît 'it appears,' il y a 'there is/are,' il s'agit ¿te 'it has to do with.' In ISF 'these non-referential subjects tend to be dropped systematically, though not obligatorily' (Lambrecht 1981:27). Typical examples are: 71 (23)

(a)

ISF FSF

va avoir une hausse sur le tabac il va y avoir une hausse sur le tabac 'There is going to be an increase in the price of tobacco'

(b)

ISF FSF

faut tout nettoyer il faut tout nettoyer 'everything has to be cleaned'

108

Analyticity and Syntheticity

Finally, emphatic subject pronouns lui 'he' and eux 'they' (masc.) still occasionally act as the sole subject of a verb. Ashby (1980) reports that in approximately 2000 potential cases only eight actually occurred (e.g., lui est chargé des émissions 'he is in charge of the programs' (p. 200), and, importantly, these are restricted to the 3rd person (cf. *moi suis/*toi es chargé Ί am/you are in charge'). Though it cannot be denied that this type of clitic deletion is still a factor in preserving a measure of analyticity in subject clitics, seen in diachronic perspective this construction is no more than a last remnant of a once prevalent pattern which can be expected to vanish completely. 42.35 The continued semantic bleaching ofpreverbal PIN markers in ISF In conjunction with phonetic erosion, a certain measure of semantic reanalysis has taken place in ISF with respect to the signalling of gender, number, and person. The most widely cited case is the extension of the indefinite on to mean 'nous.' 4235.1

The new role of ISF on and moi, toi, etc.

As we have already seen (§4.2.2.1.2), OFr on distinguished itself from other subject clitics in having undergone a significant semantic change ('man' —> ind. pron. 'one'). In ISF, on has once again changed meaning so that it is now the stylistically unmarked 1 pi. form (cf. [nous], on y va toujours 'we always go there'). 72 Because of the ISF preference for on + V rather than nous + V, the non-linear (synthetic) reiteration of P/N marking (NOUS parl-ONS) has been eliminated for the 1 pl., leaving the 2 pi. form [parl-EZ] as the only consistently non-linear (disjunctive) form. 73 4.2.35.2 The semantic bleaching of moi, toi, lui, etc. The FSF emphatic subject clitics are only of indirect concern to the question of semantic and morphosyntactic synthesis between CL ego, tu, etc. and the verb. They nevertheless deserve to be treated here because their recent history illustrates well the relative chronology by which analytic lexemic structures move toward greater synthesis. In chapter 3, I have argued that an increase in semantic relevance (with a consequential increase in frequency) and subsequent semantic weakening are necessary preconditions for morphosyntactic synthesis. It appears that moi, toi, etc. are presently undergoing such a development. In ISF there is an increasing tendency to use the formerly emphatic moi, toi, etc. in combination with the subject clitic in cases where no emphasis is intended.

Verbal core: Person/number

109

Typical examples are: moij-sais pas/je sais pas moi Ί don't know,' toi tu sais ...¡tu sais toi... 'you know ...,' eux ils font ça tout le temps 'they do this all the time,' nous on fait pas ça 'we don't do that' Analyzing a large corpus of ISF data, Ashby (1980:200) finds that the "pleonasm" stressed pronoun + subject clitic occurs in almost 10% of all 1 s. verbs. The frequency of this construction suggests that moi is no longer invariably associated with forceful assertion. In fact, used as a subject pronoun, moi is clearly emphatic no more than 50% of the time. What is remarkable about the recent history of ISF disjunctive pronouns is that within this same morphosyntactic realm a virtually identical process had already occurred: the evolution which led to the semantic change of ego, tu, etc. Thus, one finds that in both Latin and ISF: (a)

(b) (c)

(d)

the disjunctive "subject" pronouns originally enjoy a high measure of syntactic freedom (cf. ISF tu sais, MOI je vais à la maisonJMOI, tu sais, je vais à la maison, je vais à la maison, MOI, tu sais, etc. 'you know, I am going home'); the disjunctive "subject" pronouns are originally used for emphasis or contrast; changes in the semantic relevance, frequency, and function of disjunctive "subject" pronouns antedate phonological and/or morphosyntactic synthesis; and the non-emphatic use of disjunctive "subject" pronouns is typical of informal, non-stylized speech.

Notwithstanding these similarities, there exists an important difference between the disjunctive P/N markers of Latin and ISF. In Latin, emphatic subject pronouns were normally separated from the inflectional P/N markers by the verb stem (EGO cant-O). Consequently, the morphological tightening that has taken place between the French subject pronoun and verb has not resulted in a phonological synthesis between "new" and "old" P/N markers (the "old," postverbal markers have been lost in regular verbs in all but the 2 pl.). In ISF, however, moi, toi, etc. are frequently placed in immediate proximity to the "old" markers (moi je parle). Should the function of ISF moi, toi, etc. shift to one of mere grammatical morphemes (a development which is not unlikely to take place judging from the history of CL ego, tu, etc.), the morphosyntactic preconditions would then be such that future French could amalgamate the old and new morphemes into unitary morphological complexes (moi-j[e], toi-t[u], etc.).

110

Analyticity and Syntheticity

42.3 J J The partial loss of gender in ISF 3d person clitics As mentioned earlier (§4.2.2.1.2), the VL 3d person pronouns illi, illae, etc. and their descendente have always been more synthetic than the remainder of the subject pronoun paradigm because they express P/N and gender in alinear fashion. There is indication that this alinearity is being somewhat eroded. In the case of (»referential clitics, the gender distinction has been neutralized for certain speakers in favor of the masculine form, as in the examples in (24): (24)

(a)

La date [fern.] U [mase.] n'était pas dessus 'The date was not on it'

(b)

Ma femme [fem.] il [masc.] est venu 'My wife came' (cit. by Frei 1929:145-146)

Lambrecht (1981:40) cites similar examples from the François corpus (1974): (25)

je repique I pick again

les tomates. the tomatoes

¿ they

η' ont NEG have

pas, NEG

Ί again prick the tomatoes, they are not..., is ont pas, they have not

ça i because they

sont encore are still

vertí, green

they are not..., because they are still green, i sont they are

pas NEG

crevés dead

(François p. 778 [1,43], trans, mine)

they are not dead' In this example too the agreement markers are all masculine (verts should be vertes), despite the feminine gender of les tomates. Lambrecht (p. 40) also observes a tendency in ISF—which my own intuitions for Swiss French confirm—to use the masculine /i/ (= il, ils) before consonant and /il/ or /iz/ (= illils) before vowel, instead of the feminine e(s) (elle[s]) to refer to feminine antecedents not given in the same clause as the clitic: i font du bruit (i = women in another room). The importance to be attributed to the above cases of gender neutralization in subject pronouns is, at this point, not possible to assess. Although my own intuitions about informal Swiss French lead me to believe that this kind of grammatical simplification is relatively infrequent, a careful study of ISF may well reveal that the phenomenon is more prominent than assumed.

Verbal core: Personlnumber

111

We should not leave the discussion on the developments in subject clitics without mentioning a further important change taking place in ISF. In a number of instances where FSF il is simply a dummy subject, (as in il est vrai 'it is true'; il pleut 'it is raining'), il has been replaced by c' (< ce) and ça (< cela): ça pleut, c'est vrai. Originally a strictly disjunctive pronoun, ISF ça can now function both conjunctively and disjunctively. Free and bound ça may even be placed in immediate juxtaposition, as in the first of the following typically ISF stock phrases (I put disjunctive ça in small caps for easier recognition): ÇA ça va pas 'that won't' work; that's not OK'; ça joue ÇA? 'is it working/going all right?'; Ç'a ¿té? 'that has been? = did it go well?'. There has lately been some dispute about the morphemic nature of ISF il(s) (and elle[s]). M. B. Harris (1978b:121) argues that because of the switch illelle —» ça or c \ as well as a number of other changes within the pronominal system, il and elle are increasingly equivalent to 'he' and 'she'; that is, the principal factor in the selection of subject pronouns is 'sex' (i.e., male/female) rather than the grammatical category 'gender' (i.e., masculine/feminine). Lambrecht (p. 42f) takes issue with this interpretation of the synchronic "facts" of ISF il and elle, claiming that such a change has not (yet) occurred. As principal evidence he cites the example of il est beau, le livre 'the book is nice,' in which il clearly signals gender and not sex. The complexity of both Harris' and Lambrechts arguments is such, however, that I cannot discuss the matter further. As for the question of the syntheticity of the subject clitic system, I merely note that the (undisputed) partial takeover of the function of il (morphemically PERSON/NUMBER/GENDER) by ça or c (morphemically PERSON/NUMBER) increases the linearity of 3d person clitics (see §3.3.2.1.3 above). Should the trend continue and lead to the total elimination of gender marking in third person subject clitics, French would then return to the position of the CL system in which the inflections -t and -nt signalled P/N but not gender. 423.6 Paradigmatic and semantic transparency In looking at the paradigmatic transparency of CL P/N marking (§3.3.2.1.7.1), we noted that even though the system was not overly complex, lack of semantic transparency in the inflectional endings gave it a distinctly synthetic character. One would expect that the post-Classical developments which led to the loss of verbal inflections in non-periphrastic paradigms, the rise of a new set of disjunctive and conjunctive P/N markers, and extensive phonological change, would have made the transparency of the ModFr P/N paradigm radically different from that of CL. This is not, however, the case. But before drawing analogies between CL and ModFr, let us first look at the synchronic picture of FSF and ISF. As Table S illustrates, a measure of morphophonemic remodeling has taken

112

Analyticity and Syntheticity

place in the ISF system. Except for /ty/, /vu/, and the 'new' conjunctive /sa/, before consonants all ISF subject clitics have been reduced to a single phoneme (jl¡ Τ ; Ν 'he, they*; /ε/ 'she' /Ö/ 'we'). Characteristic for both FSF and ISF is a certain amount of allomorphic alternation. These alternations are relatively simple in that they are invariably conditioned by a single phonotactic factor, i.e., the presence or absence of a following vowel. A comparison between the two FSF columns in the above table reveals that FSF je, on, nous, vous, its, and elles all have alternants. ISF eliminates the first of these ( j e f j ' -> ISF j), but adds one in the masc. and fem. s. (/i/ vs. /il/ and /ε/ vs. /εΐ/). Table S. The morphophonemics of clitics in FSF and ISF FSF [Disjunctive pronoun] [moi] [toi] [lui] [elle]

ls. 2s. 3s.

ISF

/- c

/-V

/-C

/-V

2œ

l

l

ty il εΐ 0

ty il εΐ On

ty i ε

1 t il εΐ

[ça] [ça] [nous] [vous] [eux] [elles] Note:

1 pl 2 pl. 3 pl.

nu vu il εΐ

nuz vuz ilz εΐζ

sa

s(a) s(=c'

0 vu i ε

ôn vuz iz εζ

(a) before an unvoiced consonant of the III of the 1 s. form may unvoice: /2parl/ - /Spari/.

(b) 3 s. ISF /si only occurs before /ε/ (ça c'est bon) The FSF and ISF paradigms are relatively transparent in that the allomorphic alternations occur consistently throughout the verbal paradigm (except in the imperative which has no markers at all). This relative structural transparency is, however, more than outweighed by the lack of semantic transparency in several of the FSF and ISF clitics. In FSF only two verb-bound forms (nous and vous) also occur as free forms, thereby providing a link between mere grammatical markers and semantically transparent morphemes. This number is further reduced in ISF through the replacement of clitic nous by on. Consequently, the ISF morphemes /2/, 2 s. N, 3 s. fij, 3 s./pl. /ε/, and 1 pi. /Ö/ acquire meaning only when brought into association

Verbal core: Personlnumber

113

with a verb stem. We are then left with two French paradigms of P/N markers which are not unlike that of CL. Both systems are characterized by (1) the low semantic load of their morphemes, (2) the bound and (almost) obligatory nature of these morphemes, and (3) a number of relatively simple and RECURRENT allomorphic alternations. Though these allomorphemic alternations differ in that in French they are phonologically conditioned and CL they are morphologically conditioned (i.e., tense, aspect, or mood-related) (the CL 1 s. allomorph, we recall, is -o in the present indicative, -m in the imperfect, etc. [for details see section §4.2.2.1]), these differences are not substantial. Significantly different is only the non-linear gender marking in the 3 s. illelle and ils/elles. Considering that ISF shows signs of eliminating the gender distinction, it is, however, not clear just how much importance should be attributed to this divergent feature.

4.3 Summary and conclusion Our survey of the history of French verbal agreement marking has shown that for two millennia the overall momentum of the construction "subject pronoun + verb" has been in the direction of greater syntheticity. We began our investigation by focusing on the CL Latin system of postverbal inflection, noting that the relation between the verb stem and the P/N morphemes was distinctly synthetic. Next we concentrated on the semantic bleaching of ego, tu, Ule, etc. which took place as popular Latin speech developed the series of unemphatic and demonstrative subject pronouns which were later to evolve into mere grammatical markers of person and number. The switch from post- to pre- and post-verbal P/N marking was a very slow process. As our OFr data revealed, even as late as the 12th century, subject pronouns were still independent, non-obligatory, displaceable, stressable, and relatively mobile lexical units which gave only limited, and certainly no compelling, indication that they were about to begin a new era by entering into extensive morphosyntactic and phonological synthesis with the verb. Initial signs of this incipient synthesis were, first, the "verb-second" stage during which the presence of je, tu, etc. was often conditioned by rhythmic considerations. Second, finite verbs were more and more accompanied by a subject pronoun even when emphasis was not intended. And, third, the new role of moi, toi, lui etc. as subject pronouns hinted that the loss of emphasis in je, tu, il etc. had progressed sufficiently to warrant a new and fully emphatic set of subject pronouns. It was the transitional period between OFr and MidFr which saw the extensive proliferation of subject pronouns. Perhaps the single most important step towards a

114

Analyticity and Syntheticity

more synthetic verbal complex occurred when je, tu, etc. lost the ability to function as autonomous, stressable units which could no longer be placed outside the cursus of the verb (cf. *je et tu parlons —» moi et toi [nous] parlons). The effects of this development were manifold. First, the originally independent subject pronouns were now clitics dependent on the presence of a verb. Second, with subject clitics now invariably being drawn into the verb phrase, je, tu, etc. became an increasingly standard feature of the verb phrase. Third, the inability of these clitics to be placed outside the verb phrase greatly reduced the linear distance between them and the verb. Finally, due to the now obligatory replacement of je, tu, and il, etc. by moi, toi, lui, etc. in stressed position, subject clitics lost much of their semantic transparency. By the end of the MidFr period, the use of subject clitics had been largely generalized so that they became mandatory in most instances where no overt nominal subject was given. By creating an obligatory bond between 1st and 2d person subject pronouns and the verb in all verb forms except the imperative, French thus took the first steps toward the morphologization of the preverbal P/N markers. Cliticization of the subject pronouns had, of course, already reduced the syntactic mobility of items such as je to a minimum. The syntax of these MidFr clitics became even more rigid as the est-ce que interrogation pattern began to rival the standard S-V inversion, thereby reducing the frequency of clitic transportation into postverbal position (cf. parles-tu?). Importantly, the MidFr period also saw the beginnings of obligatory coreferentiality. In questions where a nominal element acted as subject, the corresponding subject clitic now had to cooccur in the sentence (e.g., MidFr est-ELLE morte, mon amie). The obligatory application of this rule marked a crucial step toward the synthesis of third person subject clitics, for it meant that purely functional, non-semantic considerations could now demand the presence of il(s) or elle{s) in the surface string. It is at this juncture in the history of French that the widespread loss of verbal endings lifted the functional importance of subject clitics to new heights. Being the sole exponents of P/N in many—though not all—tenses and not all persons their role now shifted from one of redundancy to one of necessity. While the period from OFr to MidFr is characterized by the evolution of subject pronouns from non-obligatory to obligatory, and from free to bound forms, the modem era is marked by four entirely independent developments: (1) the final desyntacticization of subject clitics, (2) the rise of a new type of clitic coreferentiality, (3) changes in the separability of subject clitics, and (4) a series of semantic changes. As our synchronic account of ISF has shown, the syntactic mobility of MidFr and modern FSF subject clitics—largely contingent on inversions of the type par-

Verbal core: Person!number

115

les-tu? or aussi viens-tu—has been eliminated. The ultimate fixing of subject clitics in preverbal position was achieved through the adoption of new interrogation strategies (tu viens?, tu viens quand?, quand c'est que tu viens?, tu viens-ti?), as well as the simple realignment of the "déviant" V-S order after certain adverbs (cf. peut-être vient-il —> peut-être il vient). Though the separability of ISF je, tu, etc. has been reduced through elimination of the redundant negative particle ne, there are no signs that French is conspiring to ban the interruptability of the complex "subject clitic + verb." On the contrary, certain developments within the pronominal system of ISF suggest that the drift toward linear proximity between P/N markers and the verb stem has been halted and, to some extent, even reversed. This is particularly evident in constructions of the type ISF je vous me montrerai (for FSF je me montrerai à vous) and ISF il y pense (for FSF il pense à lui) where the combinability or function of these items has been widened so that clitics now occur more often preverbally. Notwithstanding such instances of increased separateness of subject clitic and verb, there is ample evidence from other domains of FSF and ISF grammar that P/N markers continue to move in the direction of greater syntheticity. Particularly relevant in this respect is the widening practice of using 3d person subject clitics even when the subject is expressed nominally. A similar effect is obtained by the common ISF insertion of coreferential object clitics in constructions such as il la lui a donnée, à Jean, son père, sa moto. As previously mentioned, coreferentiality was first applied in interrogative sentences (ton ami, vient-il?). It later arose in other constructions where it eventually became a grammatical necessity (cf. FSF Dieu vous bénisse, vous et luí). Coreferential subject clitics are, however, by far most frequent in ISF. Used in more than half of all sentences, they are rapidly becoming a hallmark of informal spoken French, and their function is essentially a pragmatic one. Yet another feature which attests to the syntheticity of the combination "clitic + verb" is found in the now obligatory ISF repetition of P/N markers in coordinated constructions such as (moi) f mange et f bois. Our analysis of the ModFr system of P/N marking has revealed a number of semantic differences between FSF and ISF. On, already subject to semantic change during the early periods of French, has extended its meaning from the indefinite 'one' to ISF 'we.' Equally weighty is the observation that ISF "emphatic" or "contrastive" pronouns are now in the process of shedding their emphasis, thereby repeating a process which CL ego, tu, etc. began two thousand years earlier. The combinations moif, toi t(u), lui i(l) etc. have become so prominent in ISF speech that these semantically relevant pairs may themselves conceivably synthesize in the future. Finally, it has been argued that a new kind of semantic change, namely the effacement of the masculine/feminine distinction, appears to be affecting the 3d per-

116

Analyticity and Syntheticity

son subject clitic series. Though our survey has not been able to confirm the frequency of examples such as i font du bruit 'they [the women] are making noise' or i sont encore verts les tomates, occasional attestations such as these suggest that a blurring of the gender distinction in 3d person clitics is indeed underway. It has been shown, then, that in the evolution from an exclusively post-verbal to a predominantly pre-verbal system of P/N marking, a series of morphological units have undergone the kind of changes which in the previous chapter were typically associated with synthesis. These units, once fully displaceable, morphologically independent, non-obligatory, and semantically transparent, became progressively more fixed syntactically, more dependent morphologically, obligatory, paradigmatically more complex, and semantically less transparent, thus closely matching the morphosyntax which once characterized the CL postverbal P/N markers. The history of these preverbal markers demonstrates that changes in the nature of morphemes cannot be grasped appropriately without the notions of analyticity or syntheticity. To fully understand how ego differed from its late OFr descendent jo (> je) it does not suffice to list the individual phonological and morphosyntactic changes that shaped it into a different unit. Clearly, OFr jo maintained a morphosyntactic and semantic profile which in many respects still parallels that of its ancestor ego, but it differed fundamentally from ego by entering into a tighter relationship with the verb on several levels. This important difference between ego and jo/je does not, however, become apparent unless we seek to relate the pertinent individual diachronic changes which together create this new morphological bond. It is this global perspective, then, which is at the basis of the concepts of analyticity and syntheticity, and it is through this perspective that we can understand how unrelated developments conspire over time to transform lexical items into units of a different sort—grammatical and/or pragmatic.

Chapter five The verbal core from Latin to French: Part 2: Periphrastic cantare habeo, habeo cantatum, and je vais chanter

5.1 Introduction In the previous chapter we saw how an entire paradigm of analytic forms, CL subject pronouns, developed in a highly unified way and over a period of two millennia, into synthetic, verb-bound units. In this chapter we will examine how CL habere 'to hold, to possess, to have' underwent a similar process, during roughly the same time span, by entering into a strong semantic and morphological bond with certain verbal units. In tracing the history of habere, we will again attempt to isolate the main steps through which synthesis has been achieved, and elucidate how contrastive word order (OV versus VO), syntactic rigidity, and separability played crucial roles in determining the morphological tightness of the modern reflexes of habere + cantatum on the one hand, and cantare + habere on the other.

5.2 The rise of a new perfective: habeo +

cantatum

In CL, perfectivity was marked in a number of ways: the infix -v- (canta-v-it 'he sang/has sung'), reduplication (cad-it 'he falls' ~ ce-ci-dit 'he fell/has fallen,' -s(scrib-it 'he writes' ~ scrip-s-it 'he wrote/has written'), or a different stem allomorph (f-a-cit 'he does' ~f-e-cit 'he did/has done'). These structurally complex forms eventually came to be replaced in one of their functions by periphrastic constructions, all based on a finite form of habere plus a past participle, e.g., habeo cantatum Ί have sung' (> modFr j'ai chanté). This more analytic structure ultimately provided the model for an entire set of new exponents for the perfective categories (cf. habebam/habui factum [> Fr j'avais/eus fait] which replaced the CL past perfect feceram Ί had done'). 52.1 The beginnings of heightened relevance between habeo and PPs There is general agreement that the origins of structures such as j'ai chanté go back to Latin constructions in which habere 'to hold, to possess, to have' governs a (pro)nominal object which is modified by an adjectival PP, as in:1

118

Analyticity and Syntheticity (1) (a)

(b)

multa bona bene parta habemus many things well procured we have 'we have many well-procured things'

(Plautus)

qui eum vinctum habebit who him tied will hold '(he) who will hold him in bonds'

(Cicero)

Since the original function of adjectival PPs is to modify NP objects (and not verbs), the level of semantic integration between PPs and verbs is minimal. The low relevance of PPs vis-à-vis verbs is evidenced by the fact that the deletion of the former neither affects the meaning of the latter, nor yields an ungrammatical sentence. However, early on examples occur in which the verb and the adjectival PP are semantically integrated to a greater extent. Compare (2) where the PP cannot be dropped without jeopardizing the meaningfulness of the sentence: (2) nam hominem servom suo s domitos habere oportet oculos for aman servant his controlled have it behooves eyes et and

manus hands

(Plautus)

'for it behooves a man servant to control his eyes and hands' Although domitos still modifies the nominal object oculos, the PP is here tightly connected with habere. At this stage, however, the increased relevance of PPs vis-àvis habere has not yet affected the basic lexical meaning of the respective constituents. A f t » this initial situation, in sentences such as (1) or (2) above there occurs a reanalysis which leads to the gradual reduction of lexical value of habere, and the concomitant reorientation of PP's away from the nominal object and toward habere. This reanalysis, not at first visible in surface structure, draws the PPs into the verb phrase itself, and—for the first time in the history of Latin—creates a strong semantic link between them and habere. Witness the following "transitional" example where it must be determined from context whether invitatum modifies the noun object or the verb: (3)

episcopum ? [episcopum ? [episcopum]

... invitatum ... invitatum] ... [invitatum

habes [habes] habes]

(Gregory of Tours)

(a) 'you now have (with you) the bishop who has been invited' (b) 'you have invited the bishop'

Verbal core: Periphrastic tenses

119

As noted above, in early Latin PPs invariably modified (pro)nominal objects. By the time of Cicero, an embedded predication could, however, also act as object of the syntagm habere + PP (Pinkster 1987:204-205). Witness how in the following examples it is no longer possible to interpret the sentential object solely as the object of habere: (4) auditum habemus heard we have

quod that

...

(Vulgate)

'we have heard that' (5)

cum cognitum habeas quod sit summi rectoris ... numen when realized you have what is of the supreme ruler ... the will 'when you realize the will of the supreme lord' (Cicero)

Although the precise function of PPs in transitional examples of habere + PP + OBJECT must often be determined from context, there exist a few Latin examples where the PP is unambiguously part of the verbal complex. In these cases habere + PP lacks an appropriate object, and habere, having evolved into a mere auxiliary, has completely lost its lexical meaning. Witness (6) below where the PP is an integral part of the verb phrase, and forms a single semantic unit with habere: (6)

quem ad modum de ea re supra scriptum habemus which in the manner of that thing above written we have 'as we have written above on the matto-' (Vitruvius)

During the post-CL period, one increasingly encounters instances in which habere + PP unambiguously forms such a semantically unitary complex even when a nominal object is present. Such is the case in (7), where the absence of gender and number agreement in probatum (for CL [omnia] probata) reveals the strong semantic integration between auxiliary habere and between the PP:^ (7)

([Aree omnia] these all (things)

[probatum habemus]) check we have

(Oribasius)

'we have checked all these things' We have seen then that the reanalysis of habere as a grammatical marker rather than a full lexeme involved a change at a deeper level, that is, the semantic synthesis between the PP and habere did not manifest itself immediately in the surface string. Semantic synthesis preceded all other types of synthesis, and it was not until the habeo + PP construction had become sufficiently bonded on the semantic level that the newly relevant items could begin to tighten their phonological, mor-

120

Analyticity and Syntheticity

phological, and/or syntactic relations as well. It is important to realize, however, that the reanalysis of the PP away from the object and toward the verb cannot in and of itself have been responsible for the subsequent morphological and/or phonological tightening of relations between habeo and the PP. Though the semantic interrelation between habeo and the PP—crucial for the question of syntheticity and analyticity—was now established, the meaning of this new periphrastic construction was, as we shall now see, initially too restricted for it to achieve a level of frequency which would trigger an analytic/ synthetic cycle. 5.2.2.1 The widening meaning of the auxiliary habeo + PP construction: the importance of semantic evolution to the question of synthesis* In order for a form to occur frequently enough to undergo synthesis, it must have a meaning that is widely applicable. It must, furthermore, be communicatively useful enough to guarantee a high frequency of occurrence. Infrequent items—however great their semantic relevance—will invariably fail to spur synthesis. There are, of course, multiple factors (e.g., stratal, sociolinguistic, language-internal, etc.) which may contribute to a sudden increase in the occurrence of a rare form. None, however, tends to boost the frequency of an item more than semantic expansion. Morphemes with a widening meaning generally have a greater chance of being used in an increasing number of functions and/or environments. If the new and old functions happen to be expressed by disjunct items (e.g., habeo + PP), their increased collocation may then lead speakers to recognize them more readily as a tight and inseparable morphosyntactic whole. That semantic expansion and concomitant frequency increase promotes synthesis is a well-attested phenomenon (see Bybee 1985:155ff). In the case of habeo + PP, there is ample evidence that here too the widening of its functions has played a major role in bringing about higher frequency and a concomitant tightening of morphemic relations. Once habeo + PP had evolved into a verbal periphrasis, its applicability was rather heavily restricted. Scholars are in general agreement that the early function of the construction was to signal aspect, that is, the new complex structure was used only to refer to present states resulting from previous actions. Because habeo + PP was excluded from situation types other than those in which the verb referred to a situation which obtained as a result of a previous action (typical examples include habere + paratum 'to have ready,' clausum 'closed,' occultum 'hidden,' scriptum 'written,' etc.), other situation types—e.g., states [know the man, admire the artist] which inherently describe stable situations that do not involve change— were initially never expressed by the periphrasis.4

Verbal core: Periphrastic tenses

121

According to M. B. Harris (1982b:49), this remained the common situation until the break up of a common Romance stock. Modem Sicilian and Calabrian still reflect this stage. There, habeo + PP has retained the purely aspectual role it hekl in VL, and is, therefore, never used to describe past action, no matter how recent (thus Calab I'aju fattu Ί have done it' has no real temporal value since the focus is entirely on the present state [ Ί have done it, therefore I am now experienced in it'] rather than on the temporal location of the past event). Given the extremely restricted range of functions of the periphrasis, it comes as no surprise that in southern Italy habere + PP is found very infrequently. After this initial phase, the habeo + PP construction embarks on a semantic drift which leads to the progressive widening of its functions. The history of this drift is traced in detail by M. B. Harris (1982b) who, basing himself on a comparative analysis of modern Romance reflexes of habeo + PP, distinguishes three additional steps in the semantic evolution of post-VL habeo + PP. During the next two stages the complex structure begins to evolve in the direction of a perfect capable of referring to all types of past events which occurred at an earlier moment and which are either still going on or which continue to be considered relevant to the present situation. Unlike in VL, situation types other than resultant states can now be expressed with habeo + PP (thus the earlier example of stative "admire(d) the artist" is now combinable with "have"). However, during the early stages of this semantic expansion, there are still a number of aspectual restrictions which severely limit the applicability of the construction. One such restriction, still found in standard Portuguese, is that the use of the complex paradigm is in general limited to 'instances where a continuous state, or alternatively a continuous or repeated series of events, which began at some earlier moment, is still in progress either now or at some period of time which includes now, i.e. has not fully elapsed' (M. B. Harris, p. 51). As a result, the presence of adverbs which relate a past event to the present moment (e.g., 'just now,' 'so far,' 'today,' etc.)— i.e. adverbs which typically trigger the use of the perfect—is not sufficient reason to warrant the use of present perfect in standard Portuguese nor in American English. 5 With the beginning of the third phase—found today in Occitan, regional varieties of French, and Peninsular Spanish—these semantic restrictions are lifted altogether and the compound formation comes to be used to indicate all past events with "present relevance." Unlike in Latin where the habeo + PP periphrasis could never be used in conjunction with past time adverbials ('a second ago,' 'just now,' etc.), formations such as Spio he hecho esta tarde Ί have done it this afternoon* are now permissible.6 However, even by the end of this third stage, a numb«' of restrictions still obtain which continue to limit the applicability of the structure.

122

Analyticity and Syntheticity

Thus, two or more past events, no matter how relevant these may be to the present situation, cannot yet be expressed with the perfect if their order is indicated by a sequential time adverb (*esta tarde he comprado la pintura y LUEGO he pintado el cuadro ""this afternoon I have bought the paint and then I have painted the picture'). With the advent of the fourth phase, even these constraints are lifted. In French, as well as Northern Italian, standard Rumanian, and Catalan, there occurs what Fleischman (1983:199) and others consider a predictable semantic shift from perfect to preterite. In these languages the "present relevance" restriction has been relaxed to such an extent that habeo + PP begins to express, in addition to its perfect meaning, the temporal functions previously held by the simple past. Compare (8) to (9) where the aspectual differences between the perfect and preterite are marked explicitly in the Spanish and English but not the French equivalents: (8) Aujourd hut je n'Ai rien FAIT de la journée Hoy no HE HECHO nada en todo el día 'Today I HAVEn't DONE anything all day' (9) Hier je n'Ai rien FAIT de la journée Ayer no HICE nada en todo el día 'Yesterday I DIDn't DO anything all day' The temporal/aspectual polyvalency of French j'ai fait Ί have done/did' originated no later than the 16th century. Since then, the complex past (fai fait) has almost entirely ousted the "older" simplex past (je fis) from the spoken registers, thereby becoming not only one of the most common periphrastic constructions in all of French but, consequently, also a prime candidate for extensive morphosyntactic and/or phonological synthesis. 5.2.2.1.1 The interrelation between multiple functions, continued relevance, and synthesis I have previously argued that because of their natural propensity for high frequency, relevant forms with multiple functions are ideally predisposed to undergo synthesis. But there is yet another reason why functional multiplicity favors synthesis. In chapter 3,1 suggested that continued or prolonged relevance is a critical factor in the synthesis of related speech units because it increases the likelihood that separated items will eventually become juxtaposed. Moreover, given the slow rate at which even invariably juxtaposed speech units tend to synthesize, extended semantic relevance often directly determines the magnitude of morphological and phonological synthesis. Now, it is a well-known fact that old functions are often

Verbal core: Periphrastic tenses

123

expressed by new forms, a case in point being CL cantavi —> habeo factum Ί have sung' or MidFr je chantai —» fai chanté Ί sang.' The point I wish to make here is that, all things being equal, semantically polyvalent complexes have a better chance for prolonged existence, and, consequently, more extensive synthesis. For, unlike monovalent structures, they are not automatically subject to elimination upon loss of function. Thus it is not accidental that multi-functional Lat cantavi survived the take-over of its "perfect" function by habeo + PP, while the monovalent reflex of the same form, post-MidFr je chantai Ί sang,' was swiftly ousted from informal spoken French once fai chanté came to express also the preterit function. From what has been said in this and the previous sections, it should be clear now that the semantic reanalysis of habeo + (OBJECT) + PP not only caused habeo and the PP to become semantically interdependent, but also crucially increased the frequency and functional range of the construction. Having come to understand the pertinence of semantic relevance, frequency of occurrence, and functional multiplicity to the question of overall synthesis, we can now understand why the French complex past has unusual potential for achieving synthesis on all levels of grammar, and for evolving into an inflectional paradigm not unlike that of CL cantavi. The extent to which such a development has already occurred is discussed in §5.4.2 below.

5.3 The rise of a new periphrastic construction: cantare

habeo

One of the betta documented and most talked about syntheses in the Latin/Romance history is the formation of a new future tense based on a construction of the type cantare + habeo. In CL, futurity was expressed through a number of explicitly future strategies— synthetic inflectional (e.g., cantabo Ί will sing,' cantabitur 'it will be sung, cantarero Ί will have sung'), analytic ACCUSATIVE + INFINITIVE construction (e.g., dico me hoc FACTURUM ESSE Ί say that I will do it'), or simply the praesens pro futuro (imusne sessum 'shall we take a seat [are we going to sit]?'). 7 Non-CL texts as well as the modern Romance forms suggest that in later popular Latin and early Romance, the explicitly future strategies became progressively replaced by a series of constructions all involving the infinitive and a conjugated auxiliary verb. Among these constructions are:

124

Analyticity and Syntheticity (1) cantare (2) (3) (4)

habeo habeo habeo habeo

(5) (ß) (7)

venio volo debeo

cantare al cantare

[quid [bona

[habemus dicereH [parta habemus}]

As in the case of habeo + PP constructions, habeo is unambiguously part of a complex verbal expression whenever the previously mandatory object is no long» present, as in: (17) habes erubescere, cum venerit you-AUX will blush if he comes

(Augustus)

'you will blush when he comes' While aspect first dominated over tense in habere + PP, modality (rather than tense) first prevailed in the INFINITIVE + habere construction. Thus, according to Fleischman, 'the obligative meaning is almost universally acknowledged to have been the antecedent of subsequent semantic developments' (1982:56). Witness the following sentences where habere denotes obligation, albeit with a strong overtone of futurity, particularly in the third example: (18)

quid habemus adorare? 'what do we have to adore?'

(19) nihil habeo ad te scribere nothing I have to you to write

(Cicero)

Ί don't have anything to write to you' (20) et eum acdpere habetis and him receive you have 'and you are to receive him'

(Augustine)

Verbal core: Periphrastic tenses

127

As in the case of habere + cantatum, the shift from a lexical collocation to a grammatical unit involves the semantic extension of an originally more specific form. The meaning of obligation which the CL phrase came to possess evolves to a clear temporal meaning, as evidenced by earliest synthetic Romance future forms such as darás in (13) above. We have already seen that the new, "tense" meaning of the habeo cantatum periphrasis was the logical result of an original extended secondary meaning (i.e., completedness) of habeo + PP. In the case of cantare habeo, semantic reanalysis was achieved by an identical process, namely by bringing to the fore a logical entailment of obligation, i.e., prospection or futurity. As Fleischman points out, 'implicit in an utterance expressing obligation is the notion that the predicated action will logically be carried out, if it is to be carried out at all, at some future (or posterior) moment' (p. 59). In Romance, this temporal dimension has come to dominate, as can already be seen in the following late Latin example: (21) quod sum, essere abêtis

(Imperial inscription)

'what I am, you are to be (i.e., you must become)' It cannot be established from textual evidence just how much—if at all—the weakening of the obligative value increased the generality of the structure, thereby fostering a higher frequency and a concomitant acceleration of morphosyntactic synthesis. It seems reasonable to argue that the switch from a largely aspectual to a more temporal habeo + cantatum construction was far more consequential for the ultimate morphologization of the respective elements than was the temporalization of modal habeo + INFINITIVE. For, already in its function as a marker of obligation, habeo was combinable with all verbs, and, as such, it did not need to be freed from the kinds of (semantic) combinatory restrictions which were shown above (§5.2.2.1) to once have limited the applicability of habeo + PP to non-resultati ve verbs.12 5.33 An intermediate summary We may, at this point, briefly sum up the most important parallels between the early histories of habeo cantatum and cantare habere. First, the origin of perfectum and future forms lies in habere + OBJECT + INFINITIVE/PP constructions. Early on, the infinitive and PP both modified the object rather then habere, and as such their semantic relevance to habere was minimal at best For reasons not discussed here in detail, there subsequently occurs a deep-level semantic reanalysis which assigns new, essentially temporal values to both constructions. During this process, the following additional changes take place: (a) the PPs and infinitives develop strong semantic ties with habeo and come to form a semantic whole; (b)

128

Analyticity and Syntheticity

habeo loses its lexical value 'to hold, possess, have'; (c) habeo is demoted from main verb to auxiliary; (d) both periphrases greatly increase their frequency; and (e) the French, N. Italian, Romanian, and Catalan perfects achieve functional bivalency (perfect/preterite).

5 . 4 Morphosyntactic synthesis of habeo periphrases: the failure of habeo cantatum to fuse Despite the numerous parallels between the semantic reanalysis of habeo cantatum and that of cantare habeo, and despite the fact that from a semantic perspective both constructions were ideally predisposed to proceed towards high morphosyntactic syntheticity, the development of the future periphrasis was ultimately quite different from that of the new perfect. While the future eventually synthesized on all levels of grammar, the perfect—though undergoing most of the changes typically associated with synthesis—has maintained much of its analytic character. 5.4.1 Morphosyntactic synthesis of the Romance future The intermediate stages in the morphological fusion between INFINITIVE + habeo are, unfortunately, only partially attested. The scattered evidence from Latin and Romance nonetheless suffices to substantiate that the developments which I considered characteristic of synthesis in chapter 3—phonological fusion, reduction in separability, transportability, and isolatablility—have all contributed to the morphological tightness of the Romance future. 5.4.1.1 Separability As shown by the CL examples (10-11) above, early habeo + INFINITIVE, though often syntactically conjoined, were not always contiguous. At a time when the infinitive was already an adjunct of habeo (rather than of the noun object), a wide range of elements could still be inserted between it and its complementary infinitive. As synthesis progressed, the separability of the two elements became more restricted. Romance evidence suggests that the decrease in separability followed a selective path, that is, some words were less capable of standing between cantare habeo than others. One deciding factor appears to have been stress, so that full lexemes—by definition carriers of stress in Romance—but not unstressed oblique pronouns, were first ousted from the intermediary position. Thus, in all of the early Romance vernaculars (except French), beside the more common agglutinated constructions one finds non-agglutinated future formations of the type:

Verbal core: Periphrastic lenses (22) OCat

trobars' find-FUT REFL

ichthere

a P/N

(= s' hi trobard)

ai P/N

(= telo donarai)

129

'it will be found there' (23) OOc

donar give-FUT

lo it

t' to you

Ί will give it to you* (24) Port

darIhehas give-FUT to him/ho- P/N 'you will give it to him/her' (cit. in Fleischman 1982:73)

The final synthesis of infinitive and habeo was undoubtedly retarded by their potential separability on the syntactic level. In several dialects, the occasional separation of cantare habeo-like constructions persisted well throughout the medieval period, surviving longest in Peninsular Portuguese where it can still be found in formal written registers to this day (cf. canta-lo-â 'he will sing it'). The progressive agglutination of cantare habeo may then be invoked as an illustration that, as in the case of French subject pronouns, the semantic association between two relevant units draws these into ever closer syntactic proximity, until they eventually coalesce into a single, inseparable syntactic element. 5.4.12 Transportability In Latin, habeo could be placed both before and after the infinitive, although this variation in word order presumably reflected stylistic, pragmatic, and also possibly semantic differences (for examples see Thielmann [1885b] and Pinkster [1985]): (25) sic habeo queri (Quintilianus) 'thus I will/have to accuse' (26) de ... somniis quid habemus ácere of ... dreams what we have to say 'What can we tell about dreams?'

(Cicero)

Although no reflexes of habeo + cantare are attested in OFr texts, comparative evidence suggests that this alternative order persisted through the common Romance period, and may also have been used occasionally in the dialects) which became OFr. Thus, OSp texts, though abundantly illustrating the INFINITIVE + have construction, still occasionally display a habeo + INFINITIVE pattern (cf. OSp los que han lidiar 'those who will (have to) fight' [Cantar de mio Cid, 3523]). Moreover, a wide range of other Romance vernaculars have maintained—side by side with

130

Analyticity and Syntheticity

postverbal, agglutinated reflexes of habeo—disjunctive formations that ultimately go back to habeo cantare, with or without an intervening preposition (ad or de):^ (27) (28) (29) (30) (31)

Gal It Fr OSard OLomb

hey llegar/hey de llegar ho da cantare fai à travailler aet pagare aportare

Ί must sing' Ί must w o k ' 'he will pay' 'he will cairy'

It would seem from the comparative evidence that the synthesis of cantare habeo was in no way inhibited by the alternative, preverbal syntax of habeo. Old Spanish, for instance, clearly shows that INFINITIVE + habeo had already evolved into a single lexical unit at a time when constructions such as han lidiar were still in use. Bourciez' assertion (1967 [1910]:269) that syntactic variation in the sequence of lexemes within a given syntagm forestalls synthesis is, therefore, not justified. Rather, it seems that once two or more juxtaposed relevant elements have fused into a semantic whole, their ultimate synthesis depends above all on their frequent, and therefore, largely formulaic use. 14 Though it cannot be established with certainty why French was first among Romance languages to give up the habeo + INFINITIVE structure, a comparison between the morphosyntax of non-lexical Lat habere and that of its earliest French reflexes makes clear that the grammaticization of habeo into a morpheme of future tense must be held responsible for the progressive syntactic rigidity of temporal habeo.15 5.4.13 Isolatedness In chapter 3,1 argued for a direct correlation between the morphological synthesis of two or more elements and their relative isolability. At first glance, the history of the future tense appears to contradict my claim insofar as reflexes of habere continue to stand in isolation well after constructions such as cantare habeo began to signal futurity (cf. modFr j'ai [< habeo] un livre Ί have a book'). The contradiction is, however, only apparent, and has its origin in the— from a synchronic perspective—accidental homophony of the elements ai = '(I) have, possess, hold' and -ai = [1 s.]. Where ai is a mere morpheme signalling P/N, it no longer is a free form and its presence depends on the cooccurrence of a verbal root That non-lexical habeo must have lost its ability to stand alone at an early date is shown not only by Latin texts, where it is consistently placed in conjunction with an infinitive (or participle), but also by its status as a completely bound morpheme in all of modern Romance. Observe the following contrastive examples where the future tense mo--

Verbal core: Periphrastic tenses

131

pheme (given in small caps) is verb-bound in French and Spanish, but not English: (32) Eng (33) Fr (34) Sp

WILL you write? EcriRas-tu?16 EscribiRás

—Yes, I WILL — *Oui,Rai — *Sl, Ré

but Oui, j'écriRai but: Sí, escribiRé

As in the case of subject pronouns, the history of cantare habeo substantiates that a strong semantic relationship between two separate elements not only tends to bring these into syntactic contiguity, but also causes one (or both) to become morphologically dependent This dependence is more difficult to detect when a free, historically related morpheme remains homophonous with the "new" clitic. But this does not vitiate the claim that extended semantic synthesis will tend to reduce and ultimately prevent the occurrence in isolation of mutually relevant items. 5.4.1.4 Phonological synthesis of INFINITIVE + habeo The phonological development of the Romance synthetic future has been treated most extensively by Valesio (1968, esp. 1280. according to whom the sound development of disjunctive cantare ... habeo is not the same as that of juxtaposed cantare habeo Ρ Rather, the phonological evolution of contiguous cantare + habeo suggests that soon after habeo became a mere functor of tense, the two lexemes were not treated as two separate words, but instead as a single phonological unit (cursus). As phonologically independent units, both cantóre and hábeo carried a primary accent As they entered into a strong semantic relationship and began to be juxtaposed frequently, the primary accent on hábeo was gradually reduced to a secondary accent (hábeo), thereby suggesting that the form was on the way to becoming a clitic. Subsequently, as synthesis progressed and the two forms merged further, cantóre and hàbeo coalesced into a single lexeme carrying a single primary stress (*cantarâjo). According to Valesio (p. 1SS), the following prosodic-morphological series of forms can thus be reconstructed for Late Common Romance: »/hábeo/ - » */hàbeo/ -> */-ájo/ In this schema one moves from a situation of full independence of habeo to a situation of full dependence. In the middle situation the form is still relatively independent; in the situation on the right, i.e., the future pattern, the reflex of habeo— though bearing the main stress—has lost all of its independence and become a mere clitic. While the treatment of *cantar-ájo as a single word carrying a single stress is

132

Analyticity and Syntheticity

undoubtedly the major phonological sign of the progressive grammaticization of habeo, there is also at least one other development which points to phonological synthesis. In autonomous infinitives, the final vowel of the desinence (-re) was maintained in Common Romance, but elided when followed by habeo. Since this development was not part of a general phonological tendency, i.e., the reduction of word-final -e, but an evolution generated by a specific morphological and morphotactic context, i.e., the merging of the two verbal forms, it appears that speakers may consider a construction as a morphological whole even though its individual, semantically relevant morphemes are still separable. 5.4.1 J Morphological restrictions on awdliary habeo The change from a lexical to a grammatical morpheme reduced the morphological variability of habere. As long as habere did not have a predominantly temporal function, its morphological behavior remained identical to that of lexical habere in that it could be inflected for tense and mood like any other finite verb. Witness (35) where habere is used in the imperfect subjunctive: (35) quia cum since with

bestiis haberet pugnare animals he had to fight

(Ignatius)

'since he had to fight with animals' But as habere came to be applied regularly, either as a functor of future or futureof-the-past, temporal and modal variation for the auxiliary greatly diminished. In the end, only the present and imperfect indicative forms of habeo could be attached to infinitives, and these eventually provided the endings for the modem synthetic forms chanterai (cantare habeo) and chanterais (cantare habebam) (in Italian the perfect rather than the imperfect was attached to the infinitive: cf. It. canterebbe < cantare habuit). 5.4.1.6 Summary of the characteristic features of synthesis in the Romance future We have seen that the trajectory of habeo from a lexical and fully independent element to a highly dependent inflection is characterized by the following features: (1)

The semantic reanalysis of habeo: The meaning of habeo 'to hold, possess, etc.' evolved from 'obligation,' to 'prospection,' and finally to 'futurity.'

Verbal core: Periphrastic tenses

133

(2)

The loss of morphosyntactic freedom: When the relation between habeo and infinitive was still one of verb and complement, the two elements displayed great syntactic independence. However, as the semantic synthesis progressed, this order became increasingly more rigid. Once habeo came invariably to be drawn into the verb phrase, either before or after the infinitive stem, it evolved into a bound form. Eventually, this clitic lost its transportability and separability altogether, becoming fixed in postverbal position in some, and in preverbal position in other Romance languages.

(3)

The phonological binding of INFINITIVE + habeo: The whole construction became phonologically bound, carrying a single primary stress (*'cantar-âjo). Moreover, the agglutination of the two elements in Proto-Romance caused the—from a synchronic perspective unusual— apocope of "final" -e in infinitives.

(4)

The decrease in morphological variation of auxiliary habeo·. Bound habeo no longer inflected for mood, and its tense-aspect possibilities became limited to the present and, depending on the area, imperfect or perfect

While semantic reanalysis must have occurred prior to any other type of synthesis, it is clear that diachronically these processes overlapped and interacted. Though the morphological and phonological conditions were decisive in shaping the morphemes of modern Romance future forms, without continued semantic relevance between cantare and habeo the construction would never have conflated into a single, inflectional lexeme. 5.4.2 The partial synthesis of habeo cantatum Since the morphosyntactic history of habeo cantatum largely parallels that of cantare habeo, I will examine it only summarily. My account will attempt to highlight the claim that, notwithstanding its relative analyticity in present-day French, this structure too has undergone many of the developments typically associated with synthesis. One of the main concerns of the following sections is to search for the reasons and possible causes for the continued analyticity of fai chanté. 5.4.2.1 Morphologiealfreedom ModSF easily demonstrates that 'have' has become a bound morpheme, i.e., auxiliary ai, as, etc. (< habeo, habes, etc.), which can no longer be equated with its

134

Analyticity and Syntheticity

syntactically independent homophone. Note the following sentences where the auxiliary ai and PP lu 'read' cannot stand in isolation: (36) As-tu

un livre?

—Oui, j'en ai un

'Do you have a book?

—Yes, Ido*

As-tu lu le livre?

—*Oui,jel'ai

[avoir = lexeme]

[avoir = bound auxiliary]

— Oui, je l'ai lu 'Have you read the book? —Yes I have* The level of boundness of auxiliary ai in the above example is, then, comparable to that of future -ai (chanter-aï). Though the syntactic positions of auxiliary as and inflectional -as differ vis-à-vis the verb, both depend on the presence of an additional verbal element. 5.4.2.2 Transportability As shown by (37), Latin habeo could precede or follow the PP. (37)

(a) ancilla ... quae habeat cottidianum familiae servant ... who have everyday for the family coctum cibum (Plautus) cooked food 'a servant who every day must have the food cooked for the family* (b) ... cibum tibi food to you

et and

familiae to the family

curet she must take care

uti coctum habeäi (Cato) so that cooked she has 'she must keep a supply of cooked food on hand for you and the servants' In Romance, this syntactic freedom became progressively more restricted, so that OFr constructions of the type PP + [...] + habere were increasingly less common: (38) Cele mesme That one herself

conté m' told me

a has

(La Chastelaine de Vergi)

That one has herself told me' According to Kukenheim (1968:66), the order of auxiliary and PP became fixed towards the end of the 16th century, though the tendency to place the PP after the

Verbal core: Periphrastic tenses

135

auxiliary already prevailed in the earliest written French records. In ISF, largely because inversion is no longer used in interrogative patterns, subject clitics have become fixed in preverbal position (see chapter 4), there no longer exists any flexibility in the sequencing of periphrastic avoir + PP. The PP now invariably follows the auxiliary. 5.42.3 Separability The high degree of separability of CL habere and PP (cf. [37] above) persisted until relatively late, surviving through Old and even MidFr. Witness the following OFr construction where a noun can still intervene between the two verbal elements: (39) Jo I

ai have

païens heathens

veüz seen

(Roland)

Ί have seen heathens' Since MidFr, this syntactic freedom of the two elements has, however, become limited to the extent that nominal complements (and most other things) are no longer insertable. Moreover, innovations in the domain of interrogation (treated in chapter 4), have resulted in a situation such that subject clitics in informal spoken French no longer interrupt the sequence avoir + vu (cf. as-tu vu le film? —> t'as vu le film? with rising intonation). But despite the progressive reduction in the number of insertable words classes, separability remains the major factor arguing for a degree of analyticity in avoir + PP construction. A large number of frequently occurring adverbs or sentence qualifiers (e.g., pas 'not,' jamais 'never,' toujours 'always,' presque 'almost,' etc.) can still be placed between the auxiliary and PP (cf. je n'ai PAS compris votre réponse Ί did not understand your answer'). These insertable elements are often concatenated so that the linear distance between the auxiliary avoir and the PP becomes considerable (cf. je n'ai TOUJOURS PAS TRES BIEN compris votre réponse Ί still can't quite understand your answer'). Even more important is the fact that in many instances French has no alternative word order which would allow the speaker to place these adverbs or sentence qualifiers outside the periphrastic construction. Compare the syntax of pas 'not' in the following grammatical and ungrammatical ISF sentences: (40) l'AS pas COMPRIS la réponse? *t'AS COMPRIS la réponse pas? *t'AS COMPRIS pas la réponse? * tu pas AS COMPRIS la réponse? *pas t'AS COMPRIS la réponse?

'Didn't you understand the answer?'

136

Analyticity and Syntheticity

5.4.2.4 Phonological unity between auxiliary and PP Given the distance created between the AUX and PP by the insertion of a string of adverbs, it is not surprising that ai and chanté do not always occur within the same cursus (cf. toi t'as toujours / très bien chanté where "Γ indicates a potential cursus boundary). However, when the PP and auxiliary are contiguous, they are always joined phonologically, and 'even when there is an adverb, the entire syntactic unit is more often uttered as a single cursus than not' (Ashby 1977:81). The phonological behavior of juxtaposed avoir + PP suggests, then, that French aspires to replicate on the level of phonology as well the strong semantic relationship inherent in the construction, but that this phonological unity is often overridden by stronger, syntactic considerations. Although the trend toward phonological unity no doubt favors the overall synthesis of avoir and the PP, the syntactic rigidity of the potentially intervening elements suggests that French will not complete the synthesis of auxiliary + PP in the near future. 5.4.3 Causes for habeo cantatum's failure to synthesize It has been suggested (e.g., Pinkster 1987:213-214) that the periphrastic verb constructions habeo cantatum and cantare habeo must have existed concurrently at a relatively early point in Latin, that is no later than the first century B.C. An early date for both structures is desirable because the major Romance languages have all developed constructions which are clearly based on the Latin model. But given the roughly simultaneous nature of these developments, one must wonder why the future synthesized completely while the perfect only did so to a certain extent. 5.43.1 Valesio's phonological hypothesis Valesio (1968) believes that the differing accentual patterns of hábeo cantàre and càntare hábeo lie at the center of the problem.18 According to him, primary stress on the preposed auxiliary (hábeo cantàre) prevented coalescence of the two elements because hábeo thereby continued to be recognized as a separate unit By contrast, primary stress on postposed càntare hábeo—in line with the accentual pattern of other verb endings—encouraged fusion between the elements so that the verbal complex came to be viewed as a single unit. 5.43.2 Fleischman's morphological hypothesis While not denying that stress may have been a factor in preventing the agglutination of habeo cantare, Fleischman (1982:115) proposes an alternative, more transparent, and purely morphological explanation. According to her, postposed habeo

Verbal core: Periphrastic tenses

137

could fuse with the infinitive because none of the necessary grammatical information would thereby be sacrificed. Since the important P/N information in *cantareajo was not threatened by the welding of -re and -ajo into -rajo, synthesis was possible. By contrast, with the auxiliary proposed, any type of fusion between the adjoining elements *ajo- and -cantare would have reduced or obliterated the P/N information signalled by the syllable which was to make contact with the juxtaposed infinitive.1^ Moreover, the genesis of habeo-cantare would have brought about the—from the Latin or Romance perspective—anomalous situation of having prefixed or infixed inflections. Implicit in Fleischman's view is the assumption that the synthesis of habeo cantatum would perforce have led to the phonological fusion (rather than simply agglutination) of the auxiliary and the infinitive. But from a synchronic, late Latin or even early Romance perspective, there is no compelling reason for such a supposition. The two elements could have agglutinated (but not fused) phonologically without sacrificing grammatically crucial matter. But even if phonological reduction had obliterated the final, unstressed vowel(s) of the auxiliary, the danger of a possible syncretism would still not have been imminent. As shown by the hypothetical, phonologically reduced paradigm in the right-hand column below, each form would have remained distinct, thus maintaining its functionality: CL and Early Common Romance

habeo habes habet habemus habetis habent

Late Common Romance (cantare) + habere (synthesized)20

Late Common Romance habere + {cantatum) (synthesized, with apocope of final, unstressed vowel)

7-ájo/ */-ájs/ */-ájt/ */-ájmu(s)/ •/-ájtis/ •[-ájnt]/[-áwnt]

*/aj-/ */ajs-/ Vajt-/ */ajm(s)-/ Vajts-/ *[ajnt-]/[awnt-]

Though it is true that prefixed inflection would have been somewhat of an anomaly in Late Romance, there are good indications that such a solution was not an impossible choice. As Fleischman herself shows (1982:73), several dialects (including Old Lombard, Sicilian, Old Sardinian, Rumanian), eventually preferred the preverbal over the postverbal pattern (cf. OSard aet < [habet] pagare 'he will pay'), thereby suggesting that there was no inherent aversion to proposed P/N marking in late Latin or Romance in analytic auxiliary structures.21

138

Analyticity and Syntheticity

5.4.3.3 An alternative explanation for the difference in syntheticity between je chanterai and j'ai chanté I would like to suggest that the boundness of chanterai beside failure of fai chanté to agglutinate is linked to (a) the contrastive ordering of these elements in Latin, (b) the chronological interval in the grammaticization of the two constructions (Fleischman 119ff), and (c) the progressive rigidity of late Latin and/or Romance word order. As Fleischman (p. 1200 convincingly argues, the contrastive ordering of habeo in the constructions cantare habeo and habeo cantatum must ultimately have resulted from a difference in the chronology of their grammaticization. Cantare habeo became grammaticized at a time when the OV word order still predominated, thus naturally placing cantare into preverbal position. By contrast, the consolidation of habeo cantatum occurred after the basic syntax of Latin had switched from VO to OV, thereby naturally channelling the originally objective PP into postverbal position. I depart from Fleischman in that I do not subscribe to the view that morphophonemic conditions (i.e., the potential obliteration of P/N marking in habeocantare) were instrumental in blocking synthesis of the perfect periphrasis. Rather, I would argue that syntactic conditions alone were responsible for the continued relative analyticity of j'ai chanté: (a) the high degree of separability of habeo cantatum prior to and during its grammaticization, and (b) the greater rigidity in word order during the grammaticization of the periphrastic perfect. Keeping in mind the chronological gap between the grammaticization of the Romance future and the periphrastic perfect, and considering the syntactic changes which were profoundly altering late Latin and early Romance word order, one is led to understand that the syntactic conditions under which habeo cantatum became a semantic whole must have differed greatly from those which cantare habeo experienced during its phases of agglutination. Most importantly, during the grammaticization of the Romance future, the inflectional character of earlier Latin must have still permitted greater syntactic flexibility, such that potentially intervening items could readily be placed outside the ever more "relevant" cantare habeo complex. Given the progressively more rigid syntax of later Latin or Romance, habeo and cantatum, on the other hand, must have synthesized in an environment where alternative positions for intervening items were less readily available. Consequently, elements separating habeo cantatum could not be squeezed out easily, and the kind of morphosyntactic synthesis which had earlier brought together cantare and habeo so effectively was either delayed or, in some instances, halted altogether. The difference in morphological tightness between the future and periphrastic perfect is, then, ultimately a function of their differing ability to avoid separation.

Verbal core: Periphrastic tenses

139

For, as shown previously, the other developments typically associated with synthesis were all carried out by both constructions: two, at one point independent and mobile lexical items came to form semantic and phonological wholes whose components are now completely bound and non-transportable. While the history of these two paradigms illustrates that semantically highly relevant items tend to proceed along similar paths, it also points up, above all, the primacy of the criterion of separability over those of phonological fusion and/or boundness.2^

5.5 The new "analytic" future tenses of French and Spanish In several of the Romance languages, periphrastic constructions of the type Fr je viens de chanter Ί have (just) sung,' je vais chanter, je suis en train de chanter Ί am (in the process of) singing,' etc., are currently showing signs of incipient morphological synthesis. The various developments which have occasioned this new strengthening of morpheme interdependence may well turn out to be indicative of how synthesis proceeded during the early, largely unattested stages of the habeo cantatimi and cantare habeo constructions. Particularly telling in this respect are the "new" French and Spanish go-futures, to which attention will now be turned. 55.1 The new periphrastic future in French 5 J.12 Semantic changes in aller + INFINITIVE In spoken French, the analytic construction je vais chanter 'I'm going to sing' is making consistent inroads into the territory of synthetic je chanterai as an expression of futurity, to the extent that it may eventually oust it altogether.24 Semantic restructuring, usually one of the first and most important reactions to an increase in the mutual relevance of two once independent items, has certainly affected the new future periphrasis. Early examples of aller + INFINITIVE, found sporadically in the 14th, and frequently by the 17th century (Flydal 1943:109), make clear that aller initially always retained its literal meaning (i.e., spatial movement). 25 Just as the syntagms INFINITIVE + habere/habere + PP previously had values other than tense, so too aller + INFINITIVE started out with a non-temporal (spatial motion) meaning. The temporality of aller is particularly apparent in instances where, for contextual reasons, it cannot possibly be associated with its lexical meaning 'to go,' as in (41)-(42): (41) Si tu vas rester ici, assieds-toi! 'If you are going to stay here, sit down! '

140

Analyticity and Syntheticity (42) II va faire chaud demain 'It's going to be warm tomorrow'

Periphrastic aller did not, however, take on its temporal role by passing directly from a lexeme to a tense functor. Just as habeo in habeo cantatimicantare habeo developed from a LEXEME ('have') into a MODAL MARKER ([obligation]), and finally into a TENSE FUNCTOR ([future]), so too aller changed from a lexical ('go'), to an aspectual ([present relevance]), and, finally, to a primarily temporal marker ([futurity]). It should, however, not be concluded that the shifts 'go' -> 'PRESENT RELEVANCE' —> 'FUTURITY' occurred in discrete steps. Rather, the increased temporality of the go-future entailed a concomitant weakening but not loss of the aspect of present relevance.26 Current usage of aller + INFINITIVE suggests, in fact, that the three potential functions of aller still coexist, and that the primary meaning of aller is often context-dependent, as in the following examples: (43) Primarily lexical [spatial movement] Je vais toujours manger au centre Ί always go eat downtown' (44) Primarily temporal/aspectual [present relevance] Madame va vous voir tout de suite 'Madame will see you shortly'

(Simenon)

In many cases, the lexical/aspectual or the temporal/aspectual meanings overlap. This is particularly evident in (45) where the originally spatial meaning of vais has still not been effaced altogether: (45) Je vais m'asseoir là-bas Ί am going to sit down over there' It has often been suggested that the future/go-future contrast centers around temporal proximity (i.e., future vs. immediate future). Fleischman argues against such an analysis and demonstrates that along the tense axis the future and go-future are currently equivalent, that is, 'in respect to logical or real time, a go-future may be used to refer to an event situated at the same point on the time line, i.e. at the same real-time distance from the speaker, as an event marked by a simple future' (1982:83). Witness (46-47) where both an imminent and remote event are rendered by either future:

Verbal core: Periphrastic tenses (46)

141

(a) Je te téléphonerai demain (b) Je vais te téléphoner demain Ί will/am going to call you tomorrow'

(47)

(a) La comète ne reviendra pas avant Γ an 2050 (b) La comète ne m pas revenir avant l'an 2050 'The comet will/is not going to come back before the year 20S0'

But even though (46a) and (47a) above place the event at the same location on the time axis as (46b) and (47b) respectively, these sentences are not interchangeable. Fleischman shows that in examples such as these the go-future retains, as an overtone to its basic future meaning, a link to the speaker's present. No such connection is, however, implied in the simple future. Thus in (47a), the speaker views the return of the comet as a future action which is detached from the present worldstate. In (47b), however, the speaker underscores the (psychological) connection he feels between the distant future event and his present situation. In other words, in the (b) sentences above, the future event—regardless of its remoteness—is somehow perceived as being relevant to the "here and now" or to the discourse situation. If, as Fleischman suggests (p. 154), the semantic trajectory of Romance periphrastic habeo constructions is—as it seems to be—an indicator of how the new future periphrasis will proceed, it is not unreasonable to assume that periphrastic aller will, some day soon, shed its optional aspectual meaning of present relevance altogether and become a mere marker of futurity. By relaxing the aspectual specificity of temporal aller, the periphrastic future tense could then potentially be used to express all future events, thereby encroaching on territory now still held by the simple future. The ensuing increase in the frequency of aller + INFINITIVE would enhance further the already formulaic character of the structure, thereby possibly eliminating the interruptability which currently makes it a relatively analytic unit. The inherently "non-terminal," "non-resultative" aspect of certain situation types (e.g., 'know the answer') originally limited the applicability of the habeo + PP construction (see §5.2.2.1. above). The restrictions were lifted once the construction began to assume a temporal character. Interestingly, the aspectualization/temporalization of aller had virtually identical consequences. As long as aller had not yet been pressed into service to express aspect and/or tense, its semantic specificity prevented it from combining with a range of infinitives. A good illustration of these combinatory limitations is provided by the still strictly goal-meaning of modlt andare 'to go.' Note how the semantic clash between vado a and the infinitives stare and ritornare causes the ungrammaticality of (48b- c):

142

Analyticity and Syntheticity (48)

(a) Vado a studiare stasera in casa del mio fratello Ί am going to study at my brother's tonight' (b) *Vado a stare qui

Ί am going to stay here'

(c) *Vado a ritornare

Ί am going to return'

By taking on a temporal meaning, aller expanded its generality to a point where it could enter into semantic association with all verbs. This was a crucial step, for it

27

led to a considerable increase in the frequency of aller + INFINITIVE structures. This then triggered the kinds of phonological and morphosyntactic changes one typically finds in the history of frequent, semantically interrelated forms. 55.1.3 Morphosyntactic and phonological changes in aller + INFINITIVE In OFr, lexical aller + INFINITIVE were still separable by numerous word classes, including adverbial and nominal complements of the type given in (49): (49) Il mi avant la maisun aprester {La Vie de Saint Alexis) 'He goes on ahead to prepare the house' Subsequent to the OFr period, the separability of aller + INFINITIVE was progressively reduced so that today only a limited number of adverbs (pas, jamais, toujours, etc.), and object clitics may interrupt the syntagm: (50) Il (ne) m pas encore déjeuner 'he is not going to eat breakfast yet' (51)

Tu mi le faire 'you are going to do it'

Just as habeo in the Romance future construction eventually lost its ability to be conjugated in any but the present, imperfect, or simple past indicative tenses, so too periphrastic aller underwent tense restrictions. As a morpheme of futurity it can only appear in the present indicative (il va faire beau) or the imperfect (il allait faire beau 'it [the weather] was going to be nice'), the latter signalling futurity from the perspective of the past. Where aller + INFINITIVE is used in any other tense, aller retains its spatial meaning, as in (S2-S4): (52) L'année prochaine fin¿ travailler en Suisse 'Next year I will go work in Switzerland'

Verbal core: Periphrastic tenses

143

(53) Si f avais eu un permis de travail, je serais allé travailler en Suisse 'If I had had a work permit, I would have gone to work in Switzerland' (54) Je veux que tu ailles travailler en Suisse cette année Ί want you to go work in Switzerland this year' Note how in (SS-S6) the use of tenses other than the present or imperfect indicative leads to ungrammaticality whenever aller cannot have a spatial meaning: (55)

(a) (b) (c) (d)

Je wis manger dehors f allaii manger dehors f izó. manger, dehors Je serais allé manger dehors

Ί am going to eat outdoors' Ί was going to eat outdoors' Ί will go outdoors to eat' Ί would have gone outdoors to eat'

(56)

(a) (b) (c) (d) (e)

Je vais être content Ί am going to be happy' f allaii elee, content Ί was going to be happy' *J' irai être. content Ί will go to be happy' *Je serais allé être coment Ί would have gone to be happy' *je veux que tu ailles être content Ί want you to be happy'

It is conceivable that, as Elcock (1960:109) seems to suggest, the temporal limitations on an auxiliary may eventually play a role in the morphologization and/or synthesis of the components of a periphrasis. If an auxiliary such as aller remains homophonous with and syntactically equivalent to its lexical counterpart, the speaker may continue to regard it as an autonomous, self-contained, and, therefore analytic unit. This, in tum, may slow down or even inhibit phonological and/or morphological synthesis, regardless of the strong semantic relevance which otherwise binds the respective elements. According to Ashby (1977:84), there is also another type of morphological levelling which points to the morphological synthesis of aller. By analogy to the 2s. and 3s. persons, some non-standard French dialects (not specified, unfortunately) have now rendered homophonous all singular forms of the present tense of aller (i.e., /va/: je va, tu va, il va, etc.). 28 Aller + INFINITIVE has also been heading in the direction of greater phonological synthesis. Formerly always a separate phonological unit, non-lexical aller—like the PP in tu as chanté—is now invariably bound to the infinitive whenever nothing intervenes (cf. ISF il va aller /ivale/). The progressive increase in phonological, morphological, and semantic synthesis of aller + INFINITIVE has induced at least one investigator, Pulgram (1967), to predict that the construction is likely to become entirely synthetic, and that Fr /va/ may well develop into an invariable morpheme of future tense in both the singular

144

Anaiyticity and Syntheticity

and the plural forms, as in the following ISF paradigm (hypothetical forms are staned): /Zvaparle/ Avaparle/ /ivaparle/ /Ovaparle/ •/vuvaparle/ ivaparle/ Although Pulgram's forecast may some day become a reality, there are no indications that such developments are imminent. As long as French maintains the syntactic inflexibility of adverbs such as pas, jamais, toujours (cf. ISF i va PAS manger), the chances of the total morphosyntactic and phonological fusion of these elements will remain slim. As in the case of auxiliary avoir + PP (j'ai chanté), potential separability of semantically interrelated items together with their lack of syntactic freedom has created a disequilibrium between morphosyntax and phonology which may well prevent French from adopting the kind of preverbal inflectional morphology which is, as we shall see below, presently being implemented in Spanish. 55.2 The new periphrastic future in Spanish We will now temporarily leave French in order to see how important the criterion of separability can be to the question of synthesis. In addition to confirming our claim that semantically relevant units aspire to morphosyntactic and phonological synthesis in a highly uniform way, the Spanish go-future (voy a hablar) also provides a prime illustration of how paradigmatic homophony, grammatical tradition, and spelling conventions have conspired to obscure the already strong syntheticity of a once analytic construction. In Modern Spanish, synthetic hablaré Ί will speak,' coexists beside the newer future form voy a hablar. Parallel to French, in many Spanish dialects the go + INFINITIVE construction is now used to express a future event, and the periphrastic futures differs from the simple future in that, in addition to futurity, it expresses present relevance, a nuance of meaning which is lacking in the simple future. Although the morphosyntactic history of the Spanish go-future mirrors almost perfectly that of aller + INFINITIVE in that the colloquial usage of voy a has become lexically empty, syntactically rigid, non-transportable, and morphologically simple (like auxiliary aller it only inflects for present and imperfect), its chances for a complete semantic, morphosyntactic, and phonological synthesis are greater than

Verbal core: Periphrastic tenses

145

those of the French go-future. Underlying this claim is the observation that in colloquial usage, bound voy a, unlike French aller, can never be separated from the infinitive,^ 9 thereby forming a syntactic unit not unlike that of synthetic cantaré, cantarás, etc. Witness the following grammatical and ungrammatical colloquial Spanish sentences (the French counterpart is given for ease of comparison): (57)

(a) No voy-a-volver nunca (b) Nunca voy-a-volver (c) *No voy nunca a volver

'I'm never going to return*

(d) Je ne vais jamais y retourner (58)

(a) Le voy a dar algo (b) Voy a darle algo (c) *Voy a le dar algo

Ί am going to give him something*

(φ Je vais lui donner quelque chose Ironically, just as the variable placement of object clitics in Old Spanish slowed down the synthesis of the "old** Spanish future (cf. OSp dar le haslle darás 'you will give him'), so the flexible syntax of OSp pronouns must later have allowed the consistent juxtaposition of increasingly more relevant ir + INFINITIVE. By having at its disposal alternative word-order patterns for pronouns and adverbs, Spanish was able to eliminate the kind of infixation which can still separate temporal French aller from its infinitive (cf. [57d] and [58d] above), thus bringing together in surface syntax elements whose meanings have combined into a semantic unit Given the complete semantic, morphological, and syntactic cohesion of a structure such as voy-a-volver, one must wonder why many scholars continue to regard the Spanish go-future as analytic. At least three factors suggest themselves. First, the differences in graphic representation between consolidated forms such as iré and segmented voy a ir have undoubtedly reinforced the impression of different degrees of morphological cohesion. Second, the longstanding tradition of synthetic postverbal inflection has probably prevented analysts from recognizing functionally analogous but preverbal morphology, in particular the now relatively synthetic preverbal morphology of the new Spanish future. And, third, the complete homophony existing between the paradigms of lexical and temporal ir may suggest that the seams of morphemic composition in voy a + INFINITIVE can still be cracked open. Though I would not deny that the formal correspondence between voy a Chicago and voy aira Chicago increases the feeling of relative analyticity for the temporal ir construction, it is—as I hope to have shown, particularly for colloquial Spanish—time to view the Spanish go-future as a construction well on its way to

146

Analyticity and Syntheticity

becoming a synthetic unit Notwithstanding the fundamentally sound argument that temporal voy a + INFINITIVE construction has evolved into a synthetic unit, it would be wrong to claim that it has given up all characteristics of analytic constructs. Gapping is possible (cf. vamos a comer y beber todo eso 'we are going to eat and drink all this'), and vamos a + INFINITIVE—neither element is entirely destressed (vàmos a comér)—is generally not a phonological unit. 30 Recent studies of certain Central-American dialects reveal, however, that at least one of these last analytic tendencies shows signs of weakening. According to Anderson (1979), in colloquial Panamanian Spanish, the ir a + INFINITIVE paradigm has already undergone a number of sound changes suggesting that synthesis is now progressing on the phonological level as well. Compare the following standard and non-standard future paradigms which he cites: standard Spanish voy vas va vamos van

a a a a a

dormir dormir dormir dormir dormir

colloquial Panamanian Spanish yo v(o)adormir tu vAdormir él vAdormir nosotros vamos adormir ellos vaji adormir

Taking 2s. and 3s. να α as point of departure, 31 Panamanian speakers have assimilated the a following va (va a dormir > va: dormir), thus causing the [a] of να to be lengthened. Since final -n in van is articulated only weakly, the entire paradigm might eventually regularize to the va form. Anderson (p. 25) attempts to account for the synthesis of the Panamanian go-future by invoking the stress hypothesis described earlier for cantare habeo (§5.4.3.1). He argues that since và a cantâr has a similar accentual pattern to that of càntare hábeo, i.e. main stress on the second element, synthesis to vacantar is possible. Anderson's treatment of the morphological coalescence of the 'new' Spanish future is flawed in its implication that phonological factors are the ultimate cause of synthesis. By claiming that 'an analytic form becomes synthetic by means of a phonological fusion' (p. 26), he fails to recognize (a) that units can and do become synthetic without ever fusing phonologically (cf. Sp completamente, menospreciar), (b) that not all units which interlock on the phonological level are necessarily synthetic (cf. separable, and, therefore, relatively analytic modFr il va aller ... /iva:le/), and (c) that synthesis must ultimately always be linked to an increase in the semantic bonding between two or more units. Anderson's hypothesis is weakened, furthermore, by the fact that it fails to answer the crucial question: why is

Verbal core: Periphrastic tenses

147

voy, but not quiero (quiéro comér), puedo (puédo comér), etc., subordinating its accent to the adjoining infinitive? The answer is that the progressively stronger semantic integration between temporal ir a and the infinitive has caused Spanish speakers to reanalyze the construction as a single unit of meaning. Since such units typically carry no more than one primary stress, the phonological restructuring is only a logical and largely mechanical response to a semantic and morphological synthesis which began long ago. Clearly, the new morphologically conditioned sound changes are not a primary, but a secondary process, and their completion now hinges almost exclusively on the continued relevance and uninterruptability of ir a + INFINITIVE.

5.6

Conclusion

Having looked at the history of French subject clitics, periphrastic habeo, as well as the French and Spanish go-futures, we are now in a better position to answer the question of why there is so little agreement about the morphological typology of French. The disparity of opinion is probably attributable to a variety of reasons. First, many investigators continue to operate under the assumption that the graphic separation of morphemic units can be equated with analyticity. As a result, contrastive pairs such as CL/modFr canto/je chante are believed to be paradigm cases for the synthetic/analytic distinction even though the phonological, morphological, and syntactic cohesion of the elements jelchante and cantío is, as we have seen, virtually identical. Disagreement about the morphological typology of modFr also results from a lack of uniformity in the types of data used to substantiate the various claims. Upon perusing the more recent literature on morphological typology, one cannot help but notice that those who place French at the synthetic end of the scale (e.g., M. B. Harris, Ashby, Pulgram) all recognize the—for the question of syntheticity and analyticity—important changes which the veib-phrase has undergone in familiar speech. By contrast, defenders of "analytic French" (e.g., Guiraud, Cantera, Lathrop, etc.) rarely, if ever, seem to take into account such developments as the frequent occurrence of co-referential clitics (mon père IL habile Paris 'my father lives in Paris'), non-transportability and/or limited separability of subject clitics (SF habites-tu Paris? —> ISF t'habites Paris?·, FSF je n'habite pas Paris —> / habite pas Paris), etc. Given the difference in French morphosyntax between colloquial and formal registers, it is understandable that investigators using these different databases will arrive at different views about the analyticity and/or syntheticity not

148

Analyticity and Syntheticity

only of je comprends Ί understand' or fai compris Ί (have) understood,' but of the language at large. Many of those advocating the analyticity of French judge the morphological tightness of certain speech units exclusively against the backdrop of functionally similar, but etymologically only indirectly related, CL units, oblivious to how arbitrary the results of such an approach may be. Illustrative cases are ISF je compare and j'ai comparé, typically juxtaposed to CL comparo and comparavi. Operating with the strictly binary opposition analytic/synthetic, historical linguists quickly reached the "logical" conclusion that je compare and fai comparé are more analytic, and that French is, consequently, an analytic language. The shortcomings of this reasoning become evident once CL is no longer taken as the natural point of comparison. Let us assume, for instance, that neither CL nor pre-12th-century French had ever been attested. The same investigator who previously categorized French as analytic will now have to conclude that the syntactically, morphologically, and phonologically rather synthetic ISF je compare and j'ai comparé—contrasted not to CL comparo or comparavi but to the syntactically, morphologically, and phonologically highly independent OFr je compare and j(e) ai comparé—are quite synthetic, and that French is, consequently, a synthetic language. But the real problem with the above approach is not just arbitrariness. Having translated the difference in syntheticity between speech units A and Β into the binary opposition analytic vs. synthetic, many linguists have automatically assumed that the result could be equated with the overall morphological complexity of the respective units. That the tightness between semantically related units is not always—and, in the case of French, rarely—a reflection of their overall morphological complexity is born out by practically all of the so-called analytic constructions. Fr il a may be analytic compared to CL habet, but neither of these highly alinear and nontransparent morphemes (il = [PERSON/NUMBER/GENDER]; a = [PERSON/NUMBER/TENSE/MOOD] is an analytic structure. Similarly, modFr elle va rentrer 'she is going to return' is relatively analytic compared to elle rentrera, but its overall morphological make-up (elle = [PERSON/NUMBER/GENDER]; va = [FUTURE/PERSON/NUMBER]) is not nearly as analytic as the label "analytic go-future" might suggest. Because many investigators have come to equate the relative morphological tightness of constructions such as CL canta-v-i (synthetic) and j'-ai-chanté (analytic) with the overall morphological complexity of these units, the typologization of modFr has been excessively slanted in the direction of analyticity. Though French may, in relation to CL, possess a number of analytic structures (i.e. the "go-future," avoir + PP constructions, etc.), its overall morphological architecture simply is not, and has never been, sufficiently linear or transparent to qualify as

Verbal core: Periphrastic tenses

149

what typologists have generally considered to be an analytic language. Herein lie the reasons why scholars like Tesnière objected—rightly—to the label "analytic French," arguing that 'ce qui est certain, c'est que, même s'il existe un type analytique de langue, ce type n'est en tout cas pas celui du français, qui est une langue synthétique' (1932:64).

Chapter six Synthesis in non-contiguous elements: predicate negation in Romance

6.1 Introduction 1 It is relatively uncontroversial to suggest that synthesis has invariably been associated with phonological and morphological tightening between syntactically conjoined elements. Analysis, on the other hand, has typically been taken to involve the replacement of a single synthetic unit by two (or more) syntactically separate, i.e., isolable items. In the present chapter I will attempt to show that some relevant units may undergo the analytic/synthetic cycle without ever being juxtaposed and/or replaced by new analytic constructs. Examples of such a development are readily found within the domain of predicate negation in several Romance vernaculars. Besides illustrating how synthesis can occur in disjunctive relevant units, the study of Romance negation strategies is also instructive for at least three other reasons. First, it presents a prime example of "convergence" (what Sapir called "drift") in showing how in a single area of grammar a number of related vernaculars have developed, independently, more synthetic formations, but in a highly uniform fashion, thus substantiating my earlier claim that the synthesis of relevant units tends to proceed in a well-patterned, and, therefore, largely predictable way. Second, the data presented will lend further support to my claim that word order and syntactic flexibility are the most crucial factors in determining the level of synthesis in semantically interrelated items. And, finally, the history of Romance negation substantiates the idea that loss of semantic relevance between two (or more) separate items will necessarily thwart further movement toward phonological and/or morphosyntactic synthesis.

6.2 Predicate negation in Latin and Proto-Romance Although there has been considerable debate about the reasons for the rise of new negation strategies in Romance (see §6.3.5 below), scholars generally agree that in Proto-Romance (PR), reflexes of Latin non 'no, not' were used as the unique negator (e.g., Togeby 1980), and that the descendents of other Latin negatives (e.g., numquam 'never') were used more as negative polarity items than as functional

152

Analyticity and Syntheticity

predicate negation panicles. In Latin, predicate non always stood to the left, and often in immediate juxtaposition to the negated verb (e.g., difficile est saturant NON scribere [Juvenal] 'it is difficult to not write satire'). But within its preverbal position, non enjoyed a good measure of syntactic freedom, often appearing at a considerable distance from the negated verb:2 (1) NON haec sine not these things without

numine divom EVENIUNT the will of the gods they occur

'These things do not occur without the will of the gods.' (Aen., cit. Greenough et al. 1979:203) In addition to its displaceability, non showed other morphosyntactic features typical of analytic units. As absolute negator, it could stand in isolation (vidistine fratrem Chaeream? — NON 'did you see my brother Chaerea? — No' [Väänänen 1967:161]), though such usage was rather rare because answers to yes/no questions generally repeated the verb of the interrogative (iam dedit argentum? — non dedit 'has he already given the money? — No, he has not* [ibid]).3 Non was, furthermore, phonologically independent, and in the absence of allomorphic alternation it was structurally simple.4 62.1 Reinforcement of predicate negation in Latin Texts written in a more colloquial style suggest that already around the time of Plautus (c. 254-184 B.C.), spoken Latin occasionally reinforced non through nominal elements expressing minimal value or insignificant quantity (Väänänen p.

1620: (2)

(a) NON MICAM not a crumb

mentis of mind

sanae habere healthy have

'To be entirely out of one's mind' (b) NON licet TRANSVERSUM D1GTTVM discedere not it is permitted width of a finger go away 'one is not permitted to go away a single step'

(Cicero)

Although there are no pre-Classical attestations of nominal reinforcers, the formation of non—which is said to have involved the amalgamation and subsequent fusion of the IE negator ne 'not' and the emphatic oenum 'one [thing]'—suggests that this practice was not unknown in earliest Latin. The morphological and phono-

Synthesis: Predicate negation

153

logical synthesis of ne with oenum, formerly an object of the verb, was possible because of Latin's verb final structure, which left the two particles contiguous in preverbal position:5 (3) NE OENUM dico

Ί don't say one [thing]'

After Latin non (< ne oenum) had lost its emphatic value, it in turn attracted new strengthened such as micam, transversum digitum ([2] above). Importantly, during the period in which these new emphatic constructions rose to prominence, Latin was in the process of evolving from an XV to a VX structure, which, by relocating nominal objects such as micam or digitum from the left to the right of the verb, thwarted the juxtaposition of non and its emphasizers, as in the following Late Latin example: (4) VL NON vales UNO COCO [= unum coccum] 'you are not worth the peel of a fruit = you are not worth anything' [Väänänen, p. 163; trans, mine]).

6.3 The rise of new negation strategies in French 6.3.1 Semantic changes in nominal emphasizers By the PR stage, the VO syntax had become firmly entrenched so that nominal emphasizers such as micam, coccum, digitum, etc. invariably came to be placed in post- rather than pre-verbal position. Examples of OFr usage as illustrated in the earliest texts include:6 (5) Altrement Otherwise

ne not

m' amerai me will love

il he

Otherwise he will not love me (at all)' (6) Ne-l devez PAS blasmer not-him you should EMPH blame

MIE EMPH

(< mica 'a crumb')

(iChanson de Roland) (pas < CL possum 'a step')

'You should not blame him (at all)' (7) N' i at not there is

GIENS de hontage (giens < gentium 'of the people') EMPH disgrace

'there is no disgrace (at all)'

(Pélérinage de Charlemagne)

154

Analyticity and Syntheticity (8) de mon non ne savroiz vos of my name not know you

POINT (< punctum 'a point') EMPH

'you wouldn't know my name at all'

(Chrétien de Troyes)

The main distinction between negatives such as pas and mie was dialectal, pas being characteristic of the center (including Paris) and west of the French-speaking area, and mie of the north and east, though both forms occur in many texts (Price 1971:252). In all these areas, mie, pas, giens, etc. have completely lost their original lexical meanings in many instances (including [5-8] above) by the time they are first attested as emphasizers at the beginning of the 12th century. Some perhaps more archaizing texts reveal, however, that this semantic transformation was of fairly recent vintage, and that mie, pas, etc. must first have come to function as emphasizers only with a select group of verbs. In the Vie de Saint Thomas Becket (1174), for example, pas 'step' is still restricted to use as a complement of measure with verbs of motion (i.e., a quantifier), and cannot yet function as a full-Hedged negative emphasizer. Consequently, in every one of the 151 instances of ne ... pas, pas retains its nominal value: (9) N' not

irai I will go

un a

PAS step

avant forward

Ί will not go forward one step' Basing ourselves on texts such as the one just cited, it may be safe to conclude that pas was first used as negative emphasizer with verbs of motion exclusively, and that it was allowed to enter into association with other verbs only after it had begun to give up its lexical meaning. Since the original value of (emphatic) pas was adverbial, it is no surprise that throughout the OFr period the use of ne ... pas was restricted to cases where the verb did not command a direct object (thus transitive *il n'avoit PAS d'espouse 'he did not have a wife' had to be negated with substantival mie, point or any other such particle: il n'avoit POINTIMIE d'espouse). Only by the 17th century could pas be applied as the general predicate negative particle, and even then point still outnumbered it in those cases where the verb was transitive (Price 1962:34f). 7 Semantic restrictions must, at one point, also have limited the use of the other nominal emphasizers. As long as they retained shades of their original lexical meaning, the potential semantic clash prohibited them from entering into combination with the majority of verbs, and this, in turn, kept them from occurring at a frequency high enough to trigger grammaticization. In the case of pas there is ample evidence that greater semantic generality led to the rapid increase of the ne ... pas construction. According to Wartburg (1937:120), in the 13th century pas—still retaining overtones of its erstwhile meaning 'a

Synthesis: Predicate negation

1 SS

step*—accompanied ne in only 1 out of 10 instances,8 but by the ISth century by which time pas had lost all of its original nominal lexical meaning, ne ... pas had already become the norm. 6.32 Loss of emphasis and increase in semantic unity of ne ... pas With the increasing collocation of a postverbal emphasizer together with the original preverbal negator ne, the semantic feature of negation became attracted from the preverbal to the postverbal particle so that the latter eventually lost its emphatic value and itself became a negative proper, as in the following MidFr example: (10) aultrementla ballade N1 est Ρ AS bien composée 'otherwise the ballade is not well composed' (Gardner and Greene 1958:124) Despite the continued syntactic separation of ne and pas, by the MidFr period the construction had clearly evolved into a semantic whole whose meaning paralleled that of OFr ne ? Just as the strong semantic bond between reflexes of cantare habeo, habeo cantatum, ego canto, etc. brought about morphosyntactic and phonological changes typically associated with synthesis, so too did the progressive increase in semantic relevance between ne and pas, as we shall now see. 6.33 Morphosyntactic and phonological synthesis of nz... pas In addition to increased semantic unity, the greater synthesis of the OFr reflexes of CL non ... possum is evidenced by the following features: (1) Syntactic restrictions: the progressive restriction of the syntactic independence of ne and pas is observable in the records. As mentioned above (§6.2), CL preverbal non enjoyed considerable syntactic mobility. Similarly, possum—like any other CL object noun—could occupy various positions within the sentence. But by the MidFr era, the placement of both ne and pas had become sufficiently restricted that neither element—when used as a predicate negator—could appear outside the verb phrase. Pas was now less often separated from the verb by a lexical item, 10 and ne could be detached from the finite verb only by a limited series of object clitics (il N'en perdit PAS ung 'he did not lose [a single] one of it' [Gardner and Greene 1958:124]). (2) Obligatoriness: After the 17th century, pas rapidly became increasingly more obligatory feature of predicate negation. Although Haase (1964 [1898]:250) reports instances of the older, strictly preverbal pattern in written French well into that century, scholars agree that in the spoken language, embracing or postverbal

156

Analyticity and Syntheticity

negation had long become a regular feature of general predicate negation. Despite strong competition from mie in OFr and point in MidFr, pas eventually emerged as the standard negative particle in modern spoken French. (3) Morphological freedom: As already noted (§6.2), CL passum and non could both occur in isolation. By the time of MidFr, neither ne— the unstressed reflex of CL non—nor (negative) pas was capable of standing by itself. Both particles are now dependent on the cooccurrence of other elements in the sentence, and neither can function as absolute negato* (cf. FSF As-tu faim? — *ne/*pas, but non). (4) Reduction in morphological variability: There is ample evidence too that, as in the case of habeo in cantare habeo and or vais in je vais aller, one of the elements of the new negation construction was reduced in morphological variability. Once passum, micam, etc. began to give up their lexical content and develop into mere markers of emphasis and then negation, their grammatical status shifted from nouns to sentence qualifiers. 11 As a result, pai, mie, etc. lost their ability to inflect for number, and, like all other French sentence qualifiers, became invariable. (5) Phonological unity: Once French developed into a cursus language, ne ... pas, ne ... mie, etc. became always joined phonologically when no adverb intervened (cf. FSF il n'a pas compris Alnapaköpri/). But even when there is an adverb, ne and the postverbal negative particle are m we often than not part of a single cursus (cf. FSF il n'a toujours pas compris /ilnatuzurpaköpri/).12 Given the parallelism in morphosyntactic history between ne ... pas and other highly relevant units (e.g., the reflexes of cantare habeo, ego canto, etc.), it appears reasonable to assume that consistent syntactic separation does not prevent semantically interrelated elements from entering the analytic/synthetic cycle. Just as OFr je chante is overall more synthetic than CL ego canto, so too OFr ne ... pas displays a tighter semantic and morphosyntactic profile than CL non passum. To understand the rapid post-OFr evolution of ne ... pas toward a marker of predicate negation, one must do more than look at the individual mutations in the morphological and/or phonological evolution of the individual constituents. Only by interrelating the pertinent historical changes which collectively helped ne ... pas become a unit of a different sort can we grasp why French ultimately switched to a postverbal negation strategy.

Synthesis: Predicate negation

157

63.4 The importance of continued semantic relevance It is well known that spoken French shows (and presumably has shown for some time) a strong tendency to discard preverbal ne, leaving pas as the sole, postverbal negator (cf. ISF il est toujours pas rentré? 'has he still not returned?')» thereby aborting the analytic/synthetic cycle before ne ... pas has a chance to become juxtaposed. 13 To understand why ne ...pas never synthesized like earlier ne oenum, cantare habeo, etc., it does not suffice, as I have done in §6.2.1 above, to invoke the basic VX syntax of post-CL and/or the prevailing rigidity of French wordorder. 14 FOT the real question is not why these elements failed to be conjoined, but why ne was dropped before additional syntactic changes could bring semantically interrelated ne ... pas into juxtaposition. To answer this question, we must, however, examine what prompted French to develop postverbal emphasizers in the first place. 63 J Origins of postverbal negation in French: cyclicity in negation The changes in French negation structure, i.e. from NEG + VERB —> NEC + VERB + EMPH - » VERB + NEG (henceforth NEG1, NEG2, and NEG3 respectively) has given rise to much debate. Five different solutions have been offered for the move from pre- to postposition: (1)

(2) (3) (4) (5)

phonetic reduction of unstressed ne, which created the need for a reinforcing particle, i.e. pas, which is in turn in the process of ousting ne (cf. Regula 1966:257); redundancy of ne, givra pas, prior to the phonetic reduction of ne (Ionian and Manoliu 1972:373); German adstratum influence (Lockwood 1968:208); typological pressures resulting from a basic word-order change from XV to VX (Vennemann 1974:366fï); and 'a shift in socio-linguistic conditions, with weakening of tradition, such that the originally emphatic "forclusir [i.e., pas] was reanalysed as the vehicle of negation, to the detriment of the weakened non negator* (Posner 1985a:194).15

In Schwegler (1983) and (1988) I attempt to show that none of the above hypotheses can account for the word-order change in predicate negation in French as well as in other IE languages, and offer an alternative proposal.16 Since my counter-arguments are complex, I will not reiterate them here, but instead limit myself to explaining why, in my view, French ultimately turned pas into the sole predicate negator.

158

Analyticity and Syntheticity

Already Jespersen (1929 [1924]) observed that the general history of negative expressions in some of the best-known languages presents a curious fluctuation: the negative adverb of a given language is first given emphasis (or contrast) by some additional word, which gradually takes on the negative meaning, after which it in turn must be strengthened by some additional word, which in its turn may come to be felt as the negative proper. As we have seen, the Latin/French developments ne —» ne oenum —» non —» non possum —» ne pas —» pas support this notion of a cyclic development in negation. I would like to suggest that this kind of cyclicity occurs for the following reasons: (1)

First, there seems to be a constant and universal psycholinguistic need 17

(2)

(3)

(4)

for negative emphasizers; The frequent adjunct of an emphasizer to a predicate negator typically results in the semantic feature of negation being extended to the emphatic element, such that the latter eventually becomes a negative proper, A significant increase in the frequency of use of an emphasizer often results in a gradual loss of its emphatic value—what has been referred to as "pragmatic unmarking" (M. B. Harris 1978b)—whence the particle tends to become a simple negative maiker; Languages tend to eliminate non-functional redundancy.

The negative cycle [1] NEG - * [2] NEG + EMPHASIZER -> [3] NEG + NEG -» [4] NEG is kept in motion independent of other developments in word-order. If the emphatic particle happens to be contiguous with the predicate negator during step [2], the potential for eventual phonological and morphological fusion of the elements is obviously great. It is important to note, however, that the cycle will not become activated unless (a) the syntagm NEGATOR + EMPHASIZER is of high cognitive and cultural salience, and (b) the emphasizer has a relatively high lexical generality. If the emphatic negation structure fails to satisfy either of these conditions, the construction will simply not occur frequently enough to become grammaticized, and the emphasizer will fail to take on a non-emphatic, i.e., purely negative meaning. If a cycle involves separate elements (such as ne ... pas), their strong semantic relevance will tend to bring them into syntactic contiguity. Rigid word order as well as other competing developments in syntax may, however, inhibit their drive toward greater proximity. If juxtaposition is not achieved before the "new" emphatic particle evolves into a negator proper, complete morphological synthesis will not occur, for reasons which I shall now try to elucidate, using ne ... pas by way of example.

Synthesis: Predicate negation

159

By the time the French negation cycle had reached step three (ne ... pas 'not') around the 15th or 16th century, pas had become both lexically empty (it had evolved into a mere grammatical marker of negation) and functionally identical to ne. As a result, the strong semantic relationship which once helped bond ne and pas was now absent, and the cultural or cognitive motivation to bring these disjoint members into syntactic contiguity was lost. The continued cooccurrence of ne ... pas was now merely the result of a leftover linguistic habit whose erstwhile semantic and/or functional rationale no longer obtained. Consequently, ne ... pas ceased to strive toward greater syntheticity, and—given ne's lack of lexical meaning and non-functional redundancy—the original negative particle eventually began to be dropped.18 It should be clear from the above that the extent of synthesis between the "old" and "new" negators is to a large degree determined by the rate at which the negation cycle advances. If the cycle turns slowly enough to assure prolonged semantic relevance between the elements concerned, chances are that syntactically separate items will eventually become juxtaposed and coalesce into an indistinguishable phonological, morphosyntactic, and semantic unit If, on the other hand, syntactic contiguity is not achieved by the time the emphatic particle has itself become a negator proper, synthesis will be blocked. In either case (i.e., coalescence [ne oenum > non] or elimination of the old negative particle [ne pas —» pas]), the language returns to an analytic expression of predicate negation, thereby preparing the ground for another negation cycle as well as new synthetic structures.

6.3.6 Reduction in structural complexity of ModFr

pas

Before we leave French negation, a few words should be said about changes in the structural complexity of the "new" negative particle pas, changes which are currently increasing its analyticity. Where negation does not bear directly on the verb, pas by itself does not always function as sole negator, and, at least in one construction type, the presence of additional non may be required. Compare (11-14) where pas—when present—can be used without accompanying non in all cases except (11) (examples are from Price 1971:254f): (11) Two elements are presented as being in opposition Il était NON Il était NON PAS

fatigué, mais malade fatigué, mais malade

'he was not tired but sick*

160

Analyticity and Syntheticity (12)

The second element is negated Il était fatigué, NON Il était fatigué, NON PAS Il était fatigué, PAS

malade malade malade

'he was tired, not sick'

(13) Two elements are presented as being alternatives Fatigué ou NON, il faut qu'il vienne Fatigué ou PAS, il faut qu'il vienne (14)

'tired or not, he must corne'

A single element is being negated NON loin de París... PAS loin de Paris ...

'not far from Paris'

The use of pas (rather than non) in non-predicative function is of fairiy recent vintage. Nowadays, where there is a choice between non and pas ([12-14] above), the former is typically found in more formal, the latter in more familiar style, but even in more formal speech pas is now infringing on the territory of non in a number of expressions (in [14], the difference is simply one of spoken vs. written language). Although pas has not yet become the all-purpose negator that non was in Latin, there are strong signs that pas may also soon come to exhibit a high degree of structural simplicity. If evidence from southern French dialects may be taken as an indication of future use of pas, we might expect it to appear eventually in all negative constructions, even when negative polarity items such as jamais 'never,' (ne) ... que 'nothing ... but,' rien 'nothing,' etc. are present. Compare the following examples from the Gévaudan dialect in which pas invariably accompanies negative polarity items: 19 (15) JAMAIS never

acô this

s' ero PAS bist REFL had been NEG seen

'This had never been seen* (16) Manjabo PAS QUE he ate NEG but

deslebres rabbits

'he ate nothing but rabbits' 6 J.7 Intermediate summary It has been shown so far that the continued obligatory separation between two semantically related elements, ne and pas, did not stop these from undergoing several

Synthesis: Predicate negation

161

of the semantic, morphosyntactic, and phonological changes typically found in contiguous relevant items. Unlike synthetic canto (~*je chante) or chanterai (—» je vais chanter), the mildly synthetic MidFr predicate negation ne ... pas gave up its syntheticity without being replaced by a new, more analytic structure. Rather, one of the once obligatory elements, ne, began to be dropped because of its functional redundancy and lack of semantic bond with pas. The dissolution of the attraction between these two items was gradual and began when emphatic pas came to be recognized as a negator proper. The history of French predicate negation shows then that a change of the type synthetic —» analytic need not be abrupt, and that the analytic/synthetic cycle may reach completion even if the morphemes forming a semantic whole have not amalgamated or fused. 20 Our discussion of French predicate negation has also illustrated the centrality of word order to the question of synthesis. If ne and pas had been juxtaposed during the letter's transformation from a marker of emphasis into a negative particle, the construction would most likely have have developed into an inseparable, highly synthetic unit That such extensive synthesis never occurred was the result of two factors, namely (1) the switch from XV to VX syntax which removed nominal emphasizers from the immediate vicinity of ne, (2) the relatively quick pace at which the negation cycle turned pas into a negator proper and thereby eliminated the semantic attraction that might have united ne and pas at a later stage in the history of French.

6 . 4 Predicate negation in other Romance languages: parallelism in the analytic/synthetic cycle It is often assumed (e.g., M. B. Harris [1978b], Vennemann [1974]) that French is the only Romance language to have developed a new type of predicate negation pattern. The following documentation on verbal negation in several Romance speech areas, crucial among which are Rhaeto-Romance, Portuguese, Palenquero (Colombia), Aragonese, dialects of Catalan, Occitan, and Northern Italian, should make clear that the view of French postverbal negation as an isolated phenomenon within Romance is unfounded. The language data examined hereafter suggest that a negation cycle predictably causes the synthesis of two or more previously independent items. As a given item comes to be frequently used as a verbal emphasizer and thereby loses its original meaning or function, the growing semantic relevance between the negative and emphatic elements causes the construction to undergo the kinds of morphosyntactic changes previously observed in the Lat non ... possum: reduction in (a) transporta-

162

Analyticity and Syntheticity

bility, (b) separability, (c) morphological freedom, and/or (d) morphological variability. Although the extent of morphological tightening varies greatly and depends on a number of components, among which are the pace of the drift in negation (i.e., how quickly an entire "negative cycle" is completed), phonotactic compatibility of the respective elements, stratal influence, etc., syntactic contiguity during grammaticization will again be shown to be the most crucial factor in the overall synthesis of "negative" speech units. Although there is considerable variation in the kinds of items that can acquire negative meaning, the Romance data suggest that a switch in negative pattern always involves four basic steps: [1] NEG + VERB —» [2] NEG + VERB + EMPH —> [3] NEG + VERB + NEG -» [4] NEG. The similarities in the domain of negation command special attention because the various developments occurred independently of one another, i.e. each took place entirely within the history of the respective language. Although PR provided the basic patterns for reinforcing verbal negation by making occasional use of nominal emphasizers, none of these emphasizers had yet lost its primary lexical meaning by the time the various vernaculars began to split off from the common Romance stock. Given the cross-language parallelism in negation cycles, it would be senseless to demonstrate in detail how each individual language contrived to tighten the relation between certain elements. I will, therefore, only provide data necessary to illustrate the major semantic changes in the drift towards a new negation strategy, and— unless deemed essential to the argument—omit discussion of the progressive morphosyntactic and/or phonological restrictions which inherently accompany these developments. The reader should also infer automatically that in disjunctive negative constructions (e.g., ne ... pas) synthesis peaks between the stages NEG + V + EMPH and NEG + V + NEG, and that the subsequent development (i.e., NEG + V + NEG —» V + NEG) marks the return to a more analytic negation pattern. Contiguous negative elements (e.g., Lat. ne oenum), on the other hand, generally reach their highest level of synthesis after the completion of a negation cycle because morphological and/or phonological fusion continues to tighten their interrelation even after their coalescence into a single semantic unit. 6.4.1 Types of "new" negation structures in Romance Two basic types of new (i.e., not Latin-derived negation structures) can be distinguished. The first, already familiar, originated when a formerly nominal emphasizer became a negative. The second type, probably of fairly recent vintage and little known even among Romance comparativists, involves copying of the "old" negative particle, as in:

Synthesis: Predicate negation (17) Port

Eu I

NÀO fab NEG speak

portugués Portuguese

163

NÂO NEG

Ί don't speak Portuguese' 6.42

Negation cycles involving nominal emphasizers

6.42.1 Occitan Modern Occitan (MOc) has fully completed the cycle NEG1 > NEG3. It maintains side by side two negatives derived from earlier emphasizers, namely pas and ges (< gentium 'of the people'): 21 (18) Oc

Sabe GESIPAS

Ί don't know'

Since ges was lost in French, and since derivatives of gentium survive in dialects south of the French border, it seems doubtful that pas, and postverbal negators in general, entered Oc through French, as is often suggested (e.g., Price 1962b:267). In the absence of compelling evidence, there is little reason to assume that MOc pas and ges are not indigenous descendants of PR forms, or that Oc did not perform a negation cycle on its own. In OOc, NEGl could also be reinforced, as in other Romance languages, by ges, mica, ren 'a thing,' and sometimes dorn 'a small measure, a bit' (Anglade 1969 [1921]:364>.22 Data from post-lSth-c. texts confirm that Oc underwent changes similar to those observed in French. Although pre-positioned ne could still function without an emphasizer as late as the 16th century (19), in MidOc the use of emphasizers was already common (20) (data from Lafont [1970:84,88]; trans, mine). (19) N'anes de tous grans mots ma Princess eichante 'Don't go scaring my princess with your big words' (20) Io NO parlari PAS de la goerra Troiana Ί will not speak of the Trojan war' The shift from emphasizer to primary negative marker is clearly under way; already in the 17th century, pas can negate by itself (examples cited in Pansier [1973]): (21) Deve PAS ieu prendre une corde? 'Shouldn't I take a cord?'

(p. 262)

The embracing negation, however, is still in use:

164

Analyticity and Syntheticity (22) f f es PAS du tout ren difficile 'It is not at all a difficult thing'

(p. 265)

By the 19th century, the majority of cases show postverbal negation: (23) Τ agradan PAS titrate? 'Don't you like the traitors?'

(Aubanel 1963:206)

Although the embracing structure (NEG2) is still found in 19th-century written Oc, it is doubtful whether it survived in the spoken language. At any rate, today it has been supplanted entirely by strictly postverbal negation, thus completing the cycle. 6.4.2.2

Evidence of a third cycle in parts of the Languedoc, Massif Central, and Gascony

According to map 89 of the Atlas linguistique de la France [ALF] (il n'y a pas de source[s]), within the areas of the Languedoc, Massif Central, and Gascony there exist several pockets in which a nominal element—ges in the Languedoc/Massif Central and cap (< CL caput 'head') in western Languedoc and Gascony23— must have come to be placed as emphasizer alongside pas once the latter had evolved into a mere negator. 24 Judging from modern synchronic evidence, these additional emphasizers have now themselves become negators proper, thereby concluding yet another negation cycle. On account of the continued VO word order, this latest cycle has brought the "old" and "new" negators—both derived from nominal elements— into juxtaposition to the right of the verb:

(24)

(a) (b)

ai

PAS GES denfanP5

I have NEG NEG

child

Ί don't have a child'

ya PAS there is NEG

de sources springs

(ALF, map 89, point 780) 'there are no springs'

KAP NEG

Looking further at map 89 of the ALF, one also notes that in the Department of the Basses Pyrenées (southwestern corner of France), a few isolated dialects have not yet dropped the preverbal negator in ne ... pas constructions despite the addition of a third negative element, nada (< CL [rem] natam 'a bom thing'): (25)

NU y a PAS NEG there is NEG

NADA NEG

(ALF, map 89, point 693) 'there is nothing'

Given the syntactic contiguity of the postverbal double negatives in (24-25) above, and knowing the extent to which "new" and "old" negative particles tend to

Synthesis: Predicate negation

16S

interrelate semantically, it would be no surprise to see constructions such as pas ges, pas kap, pas nada synthesize completely into an unanalyzable semantic, morphological and phonological whole. Unfortunately, neither the ALF nor the grammars which I have consulted (for a list see the bibliography in Schwegler [1986]) make clear to what degree such a synthesis has already begun to take place. 6.4.23 Postverbal negators on Spanish soil 6.42.3.1 Aragonese Aragonese has introduced several nominal postverbal negators. While the majority of these are purely emphatic, at least one, pas (< Cat/Occ pas (?) < CL passum 'step'), no longer expresses emphasis and is used as a negative proper (26a). Witness how brenca 'little branch, splinter, chip,' and mica 'crumb,' when used in conjunction with preverbal no, have lost their original semantic content and come to mean 'at all' (26b-c): (26)

(a)

NO foré yo NEG will do I

PAS NEG

Ί won't do (it)'

(b) NO la tostarás BRENCA ista NEG it you will taste EMPH this

coca tait

'You won't taste this tart AT ALL' (c)

ixo NO lo this NEG it

feremos MICA we will do EMPH

'This we won't do AT ALL'

(Nagore 1979:164-166)

6.4.2.3.2 Catalan: semantic and grammatical change from nominal CL caput 'head to Cat. cap 'not' The use of cap as a negating element links Catalan with Occitan. Since the negation cycle appears to have turned at a slower rate in many areas southeast of the French border, Catalan provides some insight into how the semantic reanalysis of cap may have proceeded. In the lexicon of Mod. Catalan, cap (< *capu),26 in addition to its original meaning 'head,' has evolved into a negative adj. or pron. and, in Ύ1 a few dialects, also an adverb of predicate negation: Typical examples are: [a] CAP = adj., pron. (= Fr. aucun,point de...) (27) no sab CAP joch de cartas not know ANY game of cards Ί don't know any card games.'

(Balarí y Jovany 1929:268)

166

Analyticity and Syntheticity (28) CAP d 'NONE

ells no ha vingut of them not has

(Fabra 1956:83) come'

'None of them has come.' [b] CAP = adv. (Fr. ne ... pas [du tout], ne ... point) (29)

No ho sé CAP, no é CAP ventai not it I know NOT (AT ALL) not is NEG/EMPH true Ί don't know (at all), it is not true (at all).' (Coraminas 1981:524)

(30) No ho hai CAP fet not it I have AT ALL done

(Alcover 1935:856) (Andorra)

Ί haven't done it (at all)' 6.42.3.2.1 Variation in the use of cap The use of cap as a negating particle differs notably depending on the speech area. 28 Today, the adjectival or pronominal cap ([27- 28] above) shows considerable vitality throughout Catalonia, a situation which no longer holds for the Occitan domain. Cap's greatest stronghold is in the central Pyrenees. This domain distinguishes itself from other cap prominent areas in that here the particle has evolved into a negative adverb of intensity (i.e., 'not...AT ALL'), or, in a few instances, into a simple, non-emphatic predicate negator meaning 'not' ([29-30] above).29 6.4.2.3.2.2 Diachrony of *capu The type of semantic and grammatical shift which *capu has undergone from noun to negative particle appears to mirror almost perfectly that of pas. Given that these events started in the pre-literary period, the trajectory of the negator cap cannot be reconstructed exactly, but judging from comparative evidence, was likely to have paralleled semantically the following hypothetical series of steps:3® (a) I don't have a HEAD of cattle = Ί don't have a HEAD of cattle' (b) 1 don't have a HEAD [of cattle] = ' I don't have ANY cattle (AT ALL)' (c) I don't have a HEAD = Ί don't have ANY AT ALL' Somewhere between (a)-(c), cap lost its original nominal meaning of 'head,' becoming merely a negative adverb or pronoun; as a result, it could be used in new, formerly impossible contexts:

Synthesis: Predicate negation I don't have a HEAD ofbread I don't see a HEAD of books I don't see a HEAD

167

= Ί don't have ANY bread' - Ί don't have ANY books' = '¡don't see ANY'

Due to its increasingly frequent association with the predicate negator non as well as to the reduction in lexical specificity, Pyrenean cap then must have made the final step, becoming grammaticized as a negative particle in the embracing construction non...cap. The entire process thus involved the following steps (bracketed numbers refer to illustrative examples above):

(a) (b) (c) (Φ (e)

I don't I don't I don't I don't I don't

have a HEAD of cattle have a HEAD of cattle have a HEAD have a HEAD have a HEAD

= = = = =

Ί Ί Ί Ί Ί

don't have a HEADof cattle* don't have ANY cattle (AT ALL)' [2η AT ALL' [28] don't have ANY don't have AT ALL'[29-30] don't have' [29-30]

To be sure, closer inspection of the various uses and distributions of cap would reveal a number of complexities. These are, however, irrelevant to the point at issue. What matters here is that in some varieties of Catalan, cap, like pas, went through a negation cycle in which it developed considerably tighter relations with the "old" negative particle no(n). That the synthesis did not lead to the welding of no(n) and cap must, once again, be ascribed to (a) the prevailing word order which kept the particles separate at all times, and (b) the pace of the drift in the negation cycle which resulted in the elimination of redundant no before juxtaposition could be achieved 6.4.2.4 Swiss Romanst?1 In talking about Swiss Romansh, one must draw clear distinctions between the three main dialect areas: (A) Western Romansh: Sursilva (Β) Central Romansh: Sutsilva and Surmeir (Ç) Eastern Romansh: Upper and Lower Engadine, Val Müstair Among these, (A) and (B) have introduced new negation patterns. As shown by (31-33), postverbal negation in modem Eastern and Central Romansh is expressed by britch, bretga, buc(ca) and derivatives thereof, depending on the dialect 3 3 Although the embracing construction (NEG + V + EMPH) of Central Romansh corresponds pragmatically in many instances to non-emphatic ModFr ne ... pas, it can still be used for emphasis:

168

Analyticity and Syntheticity (31) Ke co NU fatschi that this NEG do

BRITCH NEG

Ί will not do this at all* In the Central region, postverbal negation is the predominant pattern (33), though the embracing pattern is not uncommon; in Western Romansh V + NEG is the rule, with the old preverbal particle having been eliminated altogether (32): (32) Western Rom. Nus we

mwein BUCA die NEG •We don't die'

(33) Central Rom.

Eu I

sai BEKA know NEG Ί don't know'

Unfortunately, in Western Romansh the development from NEGl to NEG3 had been completed by the date of appearance of the earliest written records so that it is not possible to determine how, nor in what environments) buc(c)a or beka first rose to prominence. Thus in an early document of Sursilvan (1625), buc(c)a already occurs as the sole predicate negator (34) Christ ha BUCCA mai sponds sieu song Christ has NEG ever spilled his blood 'Christ never spilled his blood' Even though none of the oldest documents provides direct evidence of the gradual synthesis between the Latin negator non and emphatic particles, the modern synchronic situation of Central and Eastern Romansh strongly suggest that NEG3 in modern Western Romansh arose by way of NEGl —» NEG2 —> NEG3, and that the postverbal negator was once part of a more synthetic construction. That the tightening relations between NEG and EMPH did not result in a stronger synthesis is undoubtedly due to the consistent linear separation between old and new particle. 6.425 Northern Italian dialects Several Northern Italian dialects provide evidence that changes in negation strategies must have gone hand in hand with the rise and fall of synthetic negation structures. Standard Italian and Northern dialects are illustrated in (35), and (36-40) respectively:

Synthesis: Predicate negation (35) Questa (36) Sa Sta (37) (38) Akelo (39) Akella (40) Kwela This

donna dona fomna jfeme fremo huma woman

NON ία li M2 there NEG

mi me m me m m' me to me

piace pyas pyaz pyay agrad pyas pleases

NEN MIA PÄ REN MIGA NEG

169

[158] [238] [160] [170] [248]

Ί don't like this woman' Example (40), from the Northeast, the transitional area between NEG1 and NEG3 (Central and Southern Italy have NEG1), represents the intermediate stage NEG2 (i.e., double negation) through which all other Northern dialects with postverbal negation must have passed. Note the variety of negative particles which were all derived from nominal emphasizers: mia < mica 'a crumb,' ren < rem nata 'a born thing,' nen < ne ente 'not a thing.' Since the last negation cycle occurred in a VO environment, all these negators naturally came to be placed in postverbal position, away from the original preverbal negator non.·*4 6.4.3 Negation cycles involving the copying of the "old" negator The circumstances under which the "old" negative particle came to be repeated, first for emphasis and later as part of a double negative, are not well known. There is no evidence that Latin ever used this technique to reinforce its predicate negation, and given the historical unrelatedness and geographic dispersion of the dialects which today make use of it—crucial among which are Central Ladin (Austrian-Italian border), the dialect of Gévaudan (southern France), Peninsular and Brazilian Portuguese, the Spanish-based but Portuguese-derived Creole of Palenque (Colombia), and Chocó Spanish (Northwestern Colombia)—there is no compelling reason to assume a common origin.3^ What is certain is that in some of these languages the negation cycle led the negative elements to develop an increasingly strong semantic and morphological bond with their respective emphasizers so that cooccurrence of NEG + EMPH eventually became a common, even obligatory feature. Unlike nominal emphasizers, which in a VO syntax predictably came to be placed postverbally, emphatic non (or derivatives thereof) was not constrained to appear on one side of the verb or the other. 36 As (41-45) illustrate, there is, however, a strong tendency in these languages to place the second negative in sentence-final position, with no intonational break between this element and the rest of the sentence: γί (41) La NO ven (NO) 'you are not coming C. Ladin (Fassano/ (at all)' Livinallonghese) (Siller-Runggaldier 1985:74)

170

Analyticity and Syntheticity Ί am not going to the party (at all)'

Pen. Portuguese38

(43) (NAO) vou à festa (NÄO) Ί am not going to the party (at all)'

Braz. Portuguese'.39

(44) I (NU) ta kandá (NU) Ί am not singing I NEG T/A sing NEG (at all)'

Palenquea)40

(42) NAO vou â festa (NÄO)

(45) Yo NO sé ΝO

Ί don't know*

Chocó Spanish (Montes Giraldo 1974:425)

If, as I have argued, the prevailing syntax of the various languages did not dictate an a priori positioning of the "new" negator, at least two pertinent questions must be asked: why is the second negator in (41-45) above sentence-final? And why, given the "natural" semantic attraction between the two negators, were the negative elements not juxtaposed in the first place? 41 The following hypothesis suggests itself. Numerous languages, including English, German, and French, tend to reinforce negation by positioning the absolute negator at either extreme of the sentence. Compare the following semantically similar constructed examples from colloquial speech responses to the question "Do you want to go to X?": (46) (a) (b) (c) (φ

English: German: French: Spanish:

Nah, I don' wanna go there Ne, ich hab' keine Lust da hinzugehen Non, fai pas envie d'y aller No, no tengo ganas de ir allí

These particles complement the predicate negative elements required by the grammar. As Jespersen (1917:72) put it: 'Something is added in a negative form with the obvious result that the negative effect is highlighted.' For additional stress, these supplementary negatives may also be placed in sentence-final position, often as an afterthought:'.42 (47)

(a) (b) (c) (d)

I don' wanna go,... nah I'm not gonna go, ...no way J' vais pas y aller,... moi, non, (non) (Yo) no voy a ir allí, ...no (no)

As such sentences become increasingly more common, two developments may take place, both decisive for the synthesis of the two negators. First, the intonational break separating the reinforcing particle from the rest of the sentence is eliminated so that the 'supplementary negative may be felt as belonging within the sentence,

Synthesis: Predicate negation

171

which accordingly comes to contain two negatives' (Jespersen 1917:72), thereby setting off a negation cycle. Second, the increased semantic relevance between the ever more cooccurring negatives may lead to consistent juxtaposition of the negative particles, and thus create favorable conditions for extensive morphological and phonological synthesis. If the "old" negator is not, or only occasionally, utterance-final, the consistent juxtaposition of the "new" and "old" negative particles may be achieved later in at least two ways: (1) by "freezing" or generalizing an occasional pattern in which the elements are contiguous ([48] below), or (2) through repositioning one of the particles into the other's immediate vicinity ([SO] below). There exist at least two Romance dialects, Gévaudanais (southern France) and N. It. Ampezzano, which appear to bear witness to such developments. Note how Gév non, repeated here as (48), is sentence-internal and directly adjacent to pas:43 (48) La

Marjano manquet PAS NON RES de Marie-Jeanne forgot NEG Ν EG NOTHING of

ço que that which

'Marie-Jeanne did not forget anything /' abio to him had

dich said

(Camproux

1959:484)

she had told him' I would like to suggest that emphatic Gév non was originally not sentence-internal but sentence-final like Eng nah. Germ ne, Fr non, etc. in (46) above. The repositioning of non from a final into a sentence-internal slot must have been caused by the generalization of a pattern in which pas and non were "naturally" adjacent. Such juxtaposition occurred particularly frequently in short simplex sentences lacking (postverbal) nominal objects, as in: (49) La Marjano manquet PAS NON

'Marie-Jeanne didn't forget'

Once speakers began to recognize pas non as a coherent semantic and morphosyntactic whole, the construction must have resisted separation so that direct objects or any other elements following pas came to be placed aft» non. Ampezzano is the second vernacular which supports my claim that disjunct negative morphemes tend to seek syntactic contiguity. This dialect is spoken in the same Ladin area as the Fassano and Livinallonghese dialects which, as shown above (41), also reinforce predicate negation by copying no. However, as the following example demonstrates, Ampezzano distinguishes itself from neighboring vernaculars by juxtaposing the "old" and "new" negators:

172

Analyticity and Syntheticity (SO) Ampezzano: Gaidinese: Badiotto: Fassano: Livinallonghese:

Ra La Ara La La You (s.)

NO (NO) vien NE ven NE vegn NO ven NO ν en NEG NEG come

(NO) (NO) (NO) ψ0) NEG

'You are not coming (at all)' (Siller-Runggaldier p. 74; trans, mine) To the best of my knowledge, there is no historical evidence showing Amp no in sentence-final position. There are, however, a number of strong arguments favoring the hypothetical reconstruction of an earlier postverbal Ampezzano construction. First, the consistent utterance-final placement of the second no in neighboring dialects points strongly to an earlier pan-Centr.-Lad non + VERB + ... + no pattern. Moreover, there exists a functional parallelism between reinforcing Centr.-Lad no and sentence-external negators which suggests that emphatic no entered the respective dialects through a process identical to that postulated for Peninsular and Brazilian Portuguese, Palenquero, Chocó Spanish, etc. Like utterance-final Eng nah, no. Germ ne. Fr non, or Sp no (cf. [50] above), the scope of the second Centr.-Lad no is always the entire proposition rather than individual components (Siller-Runggaldier p. 74), thus hinting that it, too, was originally not included in the main body of a negative proposition. But even if the functional similarities between the second Centr.-Lad no and external negators of other languages were no longer present to support my hypothesis that Amp no 'at all' originally lay outside the main frame of the sentence, the comparative evidence from numerous other negation cycles involving copying of the old negator in sentence-final position (see Schwegler 1988 and MS) would lead me to believe that Amp no + no + VERB arose via an intermediate no + VERB + no. Although it cannot be established at this point just how and in what environments postverbal no was first moved into preverbal position, the ultimate motivating factor for this syntactic change must be sought in speakers' desire to unite disjunct but semantically coherent units. Should the Ampezzano negation cycle proceed along its projected path and render obligatory the now optional double negation, no no would then undoubtedly become a prime subject for swift morphological and phonological synthesis. Whether such synthesis will occur hinges, above all, on the frequency of the expression, as well as on its ability to ward off the spread of other emphatic predicate negation constructs, as well as the rate at which the language will drop one of the negative particles to eliminate (future) functional redundancy. The centrality of the last of these factors is evidenced by Milanese in particular, where the rapid pace of a nega-

Synthesis: Predicate negation

173

tion cycle must have led to the loss of "redundant" preverbal no before it could first become juxtaposed and then amalgamated to the "new" negator. Witness (S 1) where postverbal nò is now the sole exponent of negation: (51)

Mil

A to

scòla school

ghe there

voo I go

NÒ NEG

Ί am not going to school.*44

6.5 Summary and conclusion It has been shown in the present chapter that psycholinguistic factors, namely the constant need for negative emphasizers and their semantic weakening during a frequency increase are responsible for the cyclical renewal of predicate negation. Because a negation cycle always leads to the semantic (and, in some instances also morphosyntactic and/or phonological) rapprochement of independent speech units, innovations in the domain of predicate negation inherently trigger a number of changes in the analytic/synthetic spectrum. The history of the various Romance negation strategies confirms several of my earlier claims about the nature of synthesis. First, in all the vernaculars shown to have undergone a negation cycle, semantic synthesis not only preceded but also paved the way for all other types of synthesis. New negators, whether derived from a nominal element (CL passum, mica, caput, etc.) or copied from an existing negator (Port nâo, Amp no, etc.), invariably acquired a new emphatic function (i.e., emphasis) before morphological and/or phonological synthesis were allowed to take place. Once the semantic association between negator and emphasizer was firmly established, each language then predictably moved to tighten the morphological and/or phonological relation between these elements so that they eventually developed into units of an entirely different s o n To avoid excessive repetition, I have limited the detailed analysis of semantic, morphosyntactic, and phonological syntheses in negation strategies to French, and discussed those of other Romance languages only summarily. Although a closer look at Romansh, Catalan, Occitan, etc. would have yielded similar results, the French evidence suffices to show that semantic synthesis alone does not automatically trigger morphosyntactic or phonological synthesis. Rather, speech units tighten their relations on all levels only if an expression achieves a certain minimum frequency. Thus, only the most common among the several hundred nominal reinforcers of OFr (see n6) eventually succeeded in establishing tighter connections with preverbal ne. Perhaps the most transparent example of the

174

Analyticity and Syntheticity

relationship between frequency of occurrence and synthesis is the case of ne ... pas. Infrequent and relatively analytic before the 12th century, the construction synthesized progressively as it gained wider currency during the MidFr period. The foregoing analysis of the diachrony of negation structures, much like that of habeo + PP, has also reestablished that synthesis takes place regardless of the linear separation between relevant units. OFr ne ... pas or Cat no ... cap, or Chocó Span no ...no developed rather tight relations despite failing to become juxtaposed. While this observation reconfirms my earlier claim that the analytic/synthetic notions should not only be applied to contiguous items, it is not meant to downplay the importance of syntactic proximity to the question of synthesis. For, particularly in the domain of negation, juxtaposition often determines not only the extent of synthesis, but also the lifespan of speech units. Fr ne, Oc no, W. Rom no(n), etc. were eliminated not just because they had become functionally redundant, but also because the syntax of the new negators kept them from attaching to pas, cap, or buc(c)a, respectively. Perhaps the most interesting aspect of the interplay between cycles of negation and synthesis is their combined effect on innovations in word order. I have postulated that a word-order change took place in Gévaudanais because a negation cycle first led to the reinforcement of pas by non, and then the synthesis of occasionally juxtaposed pas non brought about the inseparability as well as obligatorily sentence-final placement of emphatic non.4^ In the literature dealing with syntheticity and analyticity it is generally implied (e.g., Ashby 1977:33) that synthesis is a process, whereas analysis is an event, i.e., synthesis is viewed as a slow evolution leading to the union of formerly independent units, whereas analysis is seen as a rather abrupt change in which morphologically bound items are "replaced" by morphologically free lexemes. 46 Although it is certainly true that the continuous back and forth movement between analysis and synthesis generally takes place in such a manner, my account of the negation cycle has hopefully made clear that analysis can also be achieved through loss rather than simply replacement of a form. Neither preverbal BP nao nor Pal nu is presently being substituted by new forms, yet the ongoing shift to strictly postverbal negation, i.e., the elimination of the preverbal negative particle, is clearly moving BP and Pal back to a more analytic way of expressing negation.

Chapter seven On drift, cyclicity, reconstruction, and the motives for changes in the analytic/synthetic spectrum

7.1 Introduction Throughout the last three chapters, I have focused attention on the analysis of numerous diachronic changes. Although these changes have occurred in various areas of the respective grammars (P/N marking, negation, future, perfect, etc.), they have all had a common result, namely the recurrent cycling between analytic and synthetic formations. While one could pursue this exercise ad infinitum to illustrate how speech units go through the analytic/synthetic cycle, I trust that it has been made sufficiently clear how such changes come about so that a number of more theoretical questions may now be addressed.

7.2 The analytic/synthetic cycle and the question of drift When A. Schlegel first classified languages into the analytic and synthetic types, he did not envision a system in which languages could be typologized along a continuum (see chap. 1). But already Humboldt attempted to establish that differences in the morphological make-up of vernaculars reflected discrete stages along an evolutionary scale, i.e., he believed that highly inflectional (synthetic) languages had evolved from a more primitive, isolating (analytic) type. While subsequent 19th and early 20th-century linguists refused to view language development in such an evolutionary framework, the notion that languages gradually "drift" from one type to another has been widely upheld. Many see the direction of such a drift as a diachronic "target," that is, languages are believed to move purposefully yet inexplicably towards a typological goal in a non-haphazard fashion. It has been conjectured (e.g., Lakoff 1972:passim), or at least implied (e.g.. Jeffers and Zwicky 1980:58, Hodge 1970), that overall analyticity (or, conversely, syntheticity) is precisely the sort of target toward which languages tend to strive. To feel our way better towards an understanding of what is generally implied by "drift," it may be appropriate to cite a passage from chapter 7 of Sapir's Language (1921), where the concept was first introduced:

176

Analyticity and Syntheticity Language is not merely something that is spread out in space, as it were—a series of reflections in individual minds of one and the same timeless picture. Language moves down in a current of its own making. It has a drift (150) We must now return to the concept of "drift" in language. If the historical changes that take place in a language, if the vast accumulation of minute modifications which in time results in the complete remodeling of the language, are not in essence identical with the individual variations that we note on every hand about us, if these variations are born only to die without a trace, while the equally minute, or even minuter, changes that make up the drift are forever imprinted on the history of the language, are we not imputing to this history a certain mystical quality? Are we not giving language a power to change of its own accord over and above the involuntary tendency of individuals to vary the norm? And if this drift of language is not merely the familiar set of individual variations seen in vertical perspective, that is historically, instead of horizontally, that is in daily experience, what is it? Language exists only in so far as it is actually used—spoken and heard, written and read. What significant changes take place in it must exist, to begin with, as individual variations. This is perfectly true, and yet it by no means follows that the general drift of a language can be understood from an exhaustive descriptive study of these variations alone. They themselves are random phenomena, like the waves of the sea, moving backward and forward in purposeless flux. The linguistic drift has direction. In other words, only those individual variations embody it or carry it which move in a certain direction, just as only certain wave movements in the bay outline the tide. The drift of a language is constituted by the unconscious selection on the part of its speakers of those individual variations that are cumulative in some special direction. This direction may be inferred, in the main, from the past history of the language. (154-155)

Sapir's main point then is that there are factors intrinsic to language that cause it to drift in certain directions. These directions cannot be inferred from its synchrony, but become visible by inspecting the diachronic changes in morphosyntax over an extended period of time. Sapir never makes clear just what this "drift" consists of, whether it is to be considered a universal phenomenon, or what motivates it.1 When Sapir exposed his ideas on drift, he apparently did not have in mind changes in the analytic/synthetic tension. The examples of drift which he offered were all from the history of English: the loss of case endings, the rigidification of

On drift, cyclicity, and the analytic/synthetic spectrum

177

word order, and an evolution toward the invariable word. Later scholars, notably Lakoff (1972), have argued that these "drifts" are in fact part of one and the same phenomenon, namely the move from synthetic to analytic morphosyntax. Lakoff has observed that these drifts, 'strange and inexplicable as they are within the history of English, still more strangely and inexplicably occur again and again, independently, in the related Indo-European languages' (p. 179). Thus, the rise of subject non- emphatic pronouns in English, French, German, etc., the spread of obligatory articles in many IE languages, the switch from inflectional to prepositional case marking as well as the development of periphrastic tenses in Romance are all seen as substantiating such long-term trends. What has been particularly intriguing to scholars like Lakoff is the fact that the drift from a synthetic to an analytic language (or vice-versa) cannot, in their estimation, be described by talking about individual changes in transformation rules, in the phrase-structure component, in the lexicon, or indeed, in any part of the grammar. Rather, drift must be viewed as an inexplicable metacondition on the way the grammar of a language as a whole will change. Neither Lakoff nor any other scholar who views the analytic/synthetic drift from a teleological perspective has been able to identify exactly what this metacondition consists of. 2 Lakoff readily admits that she has no plausible explanation for the existence of such a drift, but adds, justifiably, that 'the fact that we do not understand what is happening should not blind us to the existence of this metacondition' (p. 178).3 That the analytic/synthetic drift is an extremely widespread phenomenon among the languages of the world has been demonstrated repeatedly. In the case of Egyptian, Hodge (1970) shows that over a period of 4000 years the language has apparently completed the full analytic/synthetic cycle twice.4 Tauli (1966) demonstrates a similar development for the Uralic languages, and argues (1958:178ff) that changes in the analytic/synthetic dimension can be postulated for all language groups. Pulgram (1963) provides palpable evidence of more than one link in the chain of typological repetition by pointing to the development of the future tense from early Latin to Romance (*ama bho Ί am to love' > amabo Ί will love' —> amare habeo > Fr ( f ) aimerai —» je vais aimer). Given the solid documentation of recurrent analytic and synthetic tendencies in various language families, one would be hard-pressed, Lakoff argues (1972:173, 192), to reject the notion that there exists some principle—whatever that may be— by which languages move back and forth between analysis and synthesis. If we do not postulate such an underlying principle, we would then have to assume that the histories of all IE as well as many other languages have been subject to a staggering number of coincidences. But the truly baffling question is this: Since the drift

178

Analyticity and Syntheticity

can move in both directions (towards analyticity or syntheticity) and, therefore, does not appear to be motivated by an inherent universal psycholinguistic preference for analytic or synthetic units, why is there such a linguistic pendulum built into language? Is the direction of this pendulum accidental, or is it—as Sapir would have us believe—prefigured in each dialect in a somewhat mystical fashion? Must we, like Lakoff and others, accept as a given without explanation the notion of a rythmic back and forth between analysis and synthesis? Or is there a motivating force behind it all which has heretofore been overlooked? And if there exists such a force, why do speakers of different generations feel compelled to pursue a similar typological goal? Furthermore, if, as is widely held, linguistic innovation generally takes place in infinitesimally small selective steps, how do speakers know which innovations are in line with the general drift? And why is this synthesis or analysis not achieved more quickly? All these questions are ultimately related to the possible motives for changes in the analytic/synthetic dimension. Several proposals have been advanced over the past decades, and it may be propitious to summarize the most important ones before offering my own view on the subject.

7.3 Standard hypotheses for the analytic/synthetic cycle 73.1 Semantic and phonological weakening With regard to the prehistories of many languages, it is widely held (e.g., W. P. Lehmann 1974) that a system of inflectional affixes and, to some extent, perhaps also derivational affixes, reflects one stage in the analytic/synthetic cycle that proceeds from free (i.e., analytic) morphemes to clitics and thence to affixes. These originally free morphemes are believed to weaken semantically, become unstressed and subsequently phonologically reduced, so that they gradually lose their autonomy, both lexical and morphological, and become attached to adjacent accented words.5 Additional phonological bleaching then threatens the functionality of these morphemes, so that they eventually come to be replaced by new items, whence the cycle begins once again.6 If any credence is to be given to this theory, the question must be raised why certain words are demoted phonologically in the first place. Janson seems to suggest that words become clitics because they are 'semantically so weak that they are not allowed to carry a main accent' (1976:242). It would be difficult to argue, however, that all forms which have been known to synthesize were semantically weak, a case in point being Lat * alacre mente —» Sp alegremente 'happily.' (cf. also Gmc

On drift, cyclicity, and the analytic/synthetic spectrum

179

wis a 'manner, way' —> Germ [gleicherweise 'in the same manner'; Eng [like]wise, [otherwise, etc.)· While it is clear that phonological weakening is a possible consequence of synthesis, it clearly is not the driving force behind the analytic/synthetic cycle. 7.32 Expressiveness According to Pulgram, 'the replacement of synthetic by analytic forms, often issuing from a redundant construction, aims at a kind of greater explicitness, emphasis, preciseness; in all of them there is a tendency to be repetitive, redundant, lengthy; and all arise... from a desire to achieve certainty of communication through prolixity rather than brevity, or from fear not to achieve communication unless the hearer is told repeatedly and abundantly and insistently' (1963:41). Although Pulgram may be right in his claim that analytic formations are in some instances a therapeutic response to phonologically and semantically "weak" constructs, there are many cases in which the replacement of a synthetic form cannot be ascribed to an inherent semantic or phonological weakness. It cannot, for instance, be plausibly argued that Fr vous ferez is being substituted by vous allez faire because of phonological or semantic demotion, nor can it be said that there is any tangible difference in semantic preciseness or explicitness between the two constructions. Furthermore, and more importantly, if certainty of communication is indeed a primary concern, why then are analytic formations ever allowed to synthesize? 733 Slobin's "four charges" According to Slobin (1977:186ff), there are four basic ground rules to which a communicative system must adhere if it is to function as a full-fledged human language. The system must: be (1) clear; (2) humanly processible in ongoing time; (3) quick and easy; (4) expressive. The first charge—to be clear—means that the surface structures of a language must not be too different in form and organization from the semantic structures which underlie them. The second charge—to be humanly processible in ongoing time—means that a language must conform to viable strategies of speech perception and production, that is, that there are certain sets of linguistic features (e.g., the positioning of nominal and verbal modifiers on the basis of dominant word order) which typically cohere in languages of a given type. The third charge—to be quick and easy—forces the speaker to get in as much information as possible in an economical way. The fourth charge—to be expressive—has two aspects: to be semantic and rhetorical. By "semantic expressiveness" Slobin means that a language must be capable of expressing a set of universal basic conceptual cate-

180

Analyticity and Syntheticity

gories. Logical propositions and referential information are at the core of semantic expressivity. Rhetorical expressivity, on the other hand, serves to provide alternate ways of expressing notions, that is it provides the speaker with a means to stylize the information. According to Slobin (1977:192), 'the first two charges—clarity and processibility—strive towards segmentalization (i.e., analysis). The other two charges—temporal compactness and expressiveness—strive toward synthesis. As a result. Language constantly fluctuates between the poles of analyticity and syntheticity, since none of the charges can be ignored.' Since, as Slobin himself suggests (1977:212), 'the human mind has the capacity to consistently maintain and adjust Language so that it remains in consonance with all these goals,' it is not clear what exactly prompts speakers to give in alternately to one or several of these pressures, thus allowing analysis or synthesis to dominate temporarily. Furthermore, in order to strengthen the hypothesis, one would also like to explain why some but not other units tend to synthesize with each other. Can we really assume—as Slobin seems to suggest—that formations which have repeatedly gone through the analytic/synthetic cycle (e.g., the Latin/Romance future) were communicatively less "ideal" than the historically more stable expressions (e.g., the Romance imperfect)? And if, as Slobin asserts (1977:186), the needs and constraints of speaker and listener determine the structure of language, why then does an expression not become stable and immune to further change once it has reached an ideal form? And if, as Slobin observes, languages tend to move away from oversynthesis, why does the first charge ("to be clear") not halt "excessive" synthesis before unacceptable levels of semantic and phonological fusion aie reached? Without rejecting the notion that Slobin's four charges may be important factors in language change, I believe the hypothesis leaves unanswered too many questions to be considered a satisfactory explanation for the constant back and forth between analysis and synthesis. Any theory which fails to answer the questions why some but not other units tend to synthesize, and why the morphosyntactic and phonological rapprochement of two (or more) independent speech items becomes activated at moment X (but not Y) in history of a language, is clearly inadequate to solve the puzzle. 73.4 Geisler's principle of self-regulation Geisler's principle of self-regulation strongly resembles Slobin's "four charges" in that his hypothesis also posits that changes in syntheticity are triggered by the speakers' communicative needs. In his view, languages are homeostatic (or "self-

On drift, cyclicity, and the analytic/synthetic spectrum

181

regulating") systems which are kept in operation by two opposing principles. The first of these is the tendency to create a maximally efficient system: its effect is to eliminate redundancy, shorten the syntagmatic length of words and utterances, and to pack multiple functions into as little phonic substance as possible (also known as the "principle of least effort"). The other principle is the speaker's desire to be understood: to guarantee successful communication, the speaker must create redundancy at various levels.7 These opposing tendencies are then an attempt to optimize (or facilitate) encoding on the one hand, and decoding on the other. Encoding is handled more efficiently in synthetic languages because, among other things, they typically possess polyfunctional morphemes which economically pack the maximum amount of information into the minimum amount of phonic matter.8 Decoding, on the other hand, is less efficient in synthetic modes of expression because the complex grammatical structures complicate the relation between meaning and form, thus rendering perception more difficult For analytic languages, the opposite holds: while encoding is more complicated because it requires greater phonological and combinatory effort (speech segments are longer due to the one-to-one relation between meaning and form), decoding is simpler (i.e., more economical) because of the optimal transparency between meaning and form. In addition to the two opposing tendencies just described, Geisler (1982:20f) posits a third principle—and this is at the heart of his theory of "self-regulation"— which states that in order to guarantee the functionality of any given language, the analytic and synthetic tendencies must at all times be harmonized. That is, neither analysis nor synthesis is permitted to go beyond a certain (non-specified) level of optimization. Whenever either of these two tendencies becomes excessive (i.e., threatens successful communication), the principle of self-regulation then triggers a reversal of the prevailing trend. It is this kind of scenario which leads to the formation of analytic and/or synthetic structures: speakers optimize their encoding strategies, thereby introducing more synthetic formations, until, communicatively speaking, excessive levels of synthesis are reached.9 At that point, analytic modes of expression are introduced to improve upon and eventually replace these communicatively "deficient" strategies. As an illustration of such a reversal from synthesis to analysis, Geisler cites the history of the heavily inflected verb endings of IE: Das Idg. wird gewöhnlich als flektierende OV-Sprache eingestuft und miißte demnach postdeterminierende Morphologie im Verbalbereich aufweisen. Dies ist auch der Fall, da eine Reihe von Verbal „Suffixen" bzw. Endungen vorliegen. Diese weisen jedoch einen derartig hohen Synthetisierungsgrad auf, daß eine Segmentierung in MOD-TMP-KONGR im obigen Sinne nicht mehr möglich ist. (1982:222, emph. mine)

182

Analyticity and Syntheticity

Even though both coding strategies are assumed to be active at all times, they are not always in an equilibrium. Rather there exist phases in the history of a language during which one or the other strategy may dominate, thus leading to distinctively analytic or synthetic periods (he recognizes one such period between Latin and French when he argues that, in comparison to other Romance vernaculars, French has evolved more quickly towards the analytic type [p. 26]). Given the strong similarity between Slobin's "four charges" and Geisler's "principle of self-regulation," it follows that the criticisms directed at the former apply to Geisler's hypothesis as well. To avoid repetition, I shall not reiterate them here, and merely add a few observations which will further highlight the difficulties of purely perception-based explanations for changes in the analytic/synthetic spectrum. Although we may (pace M. B. Harris [1982a: 189] and many others) agree that perceptual difficulties can cause innovations in language structure, Geisler's view that all changes in the analytic/synthetic spectrum are ultimately related to perceptual factors remains unproven. Like others before him, he isolates a few developments within the history of a language and, not unconvincingly, fits these into the framework of his theory. However, this selective analysis yields a slanted picture of typological changes (and language evolution at large) because it leaves out important data which seriously undermine the validity of his hypothesis. I am thinking here in particular of instances where highly synthetic structures were replaced, not by analytic, but by (almost) equally synthetic structures, as in the substitution of the past subjunctive endings -asel-iese by -aral-iera in spoken (American) Spanish (cf. hablASE/comlESE —> hablARA/comiERA '[that] I spoke/ate'). Similarly, analytic constructions have replaced equally analytic ones (cf. the competition between MidFr ne ... mie 'not' by ne ... pas 'not' [see §6.3]). Since, at least from a perceptual perspective, these developments neither facilitated nor undermined the already available coding mechanisms, it follows that these changes were clearly not repair strategies to perceptual problems. Perceptual difficulties may, I repeat, cause speakers to replace highly complex forms in some instances, but these factors alone cannot be responsible for all the changes in the analytic/synthetic spectrum. There is yet further criticism to be levelled at Geisler's and Slobin's hypotheses. While it seems intuitively valid that to maintain successful communication languages must not exceed certain threshold levels of syntheticity, linguists have never been able to establish such thresholds empirically. The linguistic tools presently available are incapable of measuring objectively, for a given language, which speech segments fall within the range of "perceptually inadequate," and it is doubtful whether this state of affairs will change. The fact that analytic forms tend to frequently replace highly synthetic ones in some languages merely hints that per-

On drift, cyclicity, and the analytic/synthetic spectrum

183

ceptual problems may be at the root of some of these changes, but it is not the kind of conclusive evidence that will satisfy sceptics like Lass (1980:35) who makes the following critical remark about "explanations" based on the notion of "optimal" and "non-optimal language": In general then "optimization" can't be predicted; it can only be recognized when, given the prior definition, something occurs that can be said to be an optimization. If it fails to occur, then "the things weren't equal." This doesn't leave optimization—for all the fuss that has been made about it, including its deification in King (1969)—a very exciting notion. Languages are apparently quite free not to optimize. This unruliness underlines the absurdity of notions like optimal and non-optimal languages, which at one time at least could apparently be talked about with a straight face (e.g. Chomsky and Halle 1968:410-11; cf. Lass 1975:491). The problem of "refusal to optimize" will continue to haunt us. (1980:35) One final point: to my knowledge, it has never been proven that analytic structures are indeed easier, and synthetic structures more difficult to decode. 10 Conversely, it remains to be seen whether the encoding of analytic languages in fact requires greater effort on the speaker's part. We are here dealing with aprioristic claims which, though perhaps correct, require further careful research before they can be accepted wholeheartedly. 7.3 J The cultural-mentalistic hypothesis Although the cultural-mentalistic hypothesis has not enjoyed much popularity in recent decades, for the sake of completeness I shall nonetheless briefly state its basic tenets. The hypothesis is rooted in the belief that language structure directly reflects the level of cultural sophistication (or lack thereof). Thus Kuen, speaking of the analytic character of French, argues that 'dieser "analytische," verstandesmäßige zerlegende Geist findet sich mehr oder weniger auch in den anderen romanischen Sprachen, aber er hat nirgends so stark gewirkt wie im Französischen' (1958:7, emph. mine). In a similar vein, Vossler (1929 [1919]) and Lerch (1933) interpret the so-called "logical" analytic word order of French as a direct reflection of the supposed clarity (and superiority) of the French mind.11 The evolutionary prejudice which underlies the works of Lerch, Vossler, etc. has, of course, been criticized extensively so that there is no need to reiterate the weaknesses of their hypothesis. While it is clear that cultural factors (i.e., the lasting spread of a new negative emphasizer by a culturally prestigious minority) can crucially determine whether synthesis may take place, it is now equally well accepted that language structure and cultural genius are—strictly speaking—unrelated.

184

Analyticity and Syntheticity

7.4 Multiple causation: the source of the analytic/synthetic cycle 7.4.1 The problem with "drift" A main objective of diachronic linguistics has generally been to trace the particular steps by which a language changes, and to then hypothesize about the possible reasons for these changes. In cases where clusters of seemingly unrelated developments have had a single effect, some linguists have proceeded to search for a SINGLE underlying cause. Such, as we have seen, is the approach taken by Lakoff, who appeals to the notion of non-haphazard, i.e., goal-oriented "drift" in attempting to account for the noted IE tendency toward greater analyticity. The constant fluctuation between analysis and synthesis cannot, however, be explained through recourse to a teleologica! drift, nor to any other theory which has as its basis a single, common cause. In searching for the possible origin(s) of analysis and/or synthesis, there has prevailed the misconception that a largely unidirectional drift towards analysis changed the morphological structure of IE languages in a fundamental way. It has, however, never been documented that the structure of a language in toto has passed through the evolution "synthesis —» analysis —» synthesis, or analysis —> synthesis —> analysis," 12 nor has it been demonstrated that analysis ever prevailed to the exclusion of synthesis (or vice-versa). Rather—and this is born out by the Romance data presented in chapters 3 through 6—both tendencies, the analytic as well as the synthetic, continually coexist. ^ As regards the question of synthesis/analysis, then, no palpable evidence has ever been offered that speakers unconsciously select and grammaticize only those variations that cumulatively point in a particular direction (e.g., synthesis —» analysis). By arbitrarily limiting the evidence to a few select developments—the rise of subject pronouns, prepositions, periphrastic tenses, etc.—and by, consciously or unconsciously, overlooking those changes which yielded more synthetic formations, scholars have propagated the erroneous impression not only that there existed a distinctively analytic period in the histories of IE languages, but also that this analytic current inevitably dragged the entire morphological structure of these veraculars away from synthesis. Though a more rounded-out investigation of analytic and synthetic trends from IE to its modern descendents would certainly have yielded a less slanted picture, it is only fair to point out that a purely perceptual problem in detecting synthetic tendencies can easily lead investigators to false conclusions. Unlike analysis, which almost always involves the easy-to-observe replacement of one form by another (see chapter 3), synthesis always consists initially of a deep-level change (i.e., the semantic rapprochement of two or more independent speech units) which is essen-

On drift, cyclicity, and the analytic/synthetic spectrum

185

tially imperceptible. The first "obvious" signs of synthesis—either in the form of phonological and/or morphosyntactic changes—typically do not manifest themselves in the surface structure even centuries after once independent units have begun to establish tighter relations. What is more, synthesis is often aborted before it triggers any change in surface structure, in which case it may go entirely unnoticed. 14 Given this perceptual difficulty, it is easy to understand why analysis appears to be far more common than synthesis, and why speech units seem to reverse their momentum along the analytic/synthetic continuum with overwhelmingly greater frequency in the direction of analysis. 7.42 Causalfactors in analysis and synthesis Just because two or more languages of common ancestry undergo many parallel changes in their morphological structure, or just because a series of diachronic changes ultimately converge to yield similar results, we should not feel obliged to search at all cost for a single motivating factor for all these changes. We must, for instance, be prepared to accept that different mutations within different areas of the grammar may—and, often do—converge to increase the number of analytic constructions in language for a variety of unrelated reasons. As Lightfoot lucidly points out in his discussion on drift (1979a:392), there is no reason to assume that various changes should be formally related simply because of their overall effect. My main contention, then, is that the series of century-long, often minute and, from a synchronic perspective, virtually imperceptible changes towards analyticity or syntheticity have never in any way been goal-oriented, but have merely been the consequence of the many changes which, for one reason or another, have altered the shape of languages. Once analyticity and/or syntheticity are no longer viewed as the result of a long-range conspiracy to acquire a new type of morphological structure but rather as the outcome of complex and potentially unrelated historical changes, and once it is fully understood that speech units are liable to undergo analysis or synthesis at all times, it will follow that the direction of the observed IE trend towards analysis (i.e., the so-called drift) was, in a sense, coincidental. From the standpoint of analyticity/syntheticity, it is purely by chance that the more analytic subject pronouns, the periphrastic tenses such as the habeo future, and/or prepositions, all arose during roughly the same period. Similarly, and again strictly from the perspective of analyticity/syntheticity, it is merely accidental that various Romance languages concurrently formed more synthetic negation patterns at the close of the Vulgar Latin era. And it must be considered equally fortuitous that at least two Romance vernaculars, Gévaudanais and Brazilian Portuguese, have recently formed mildly synthetic double negation patterns by drawing after-thought

186

Analyticity and Syntheticity

(postverbal) nò, non, and näo, respectively, into the main body of the sentence. By denying any broad-gauged, overall teleological directionality to analytic/synthetic drifts, and by viewing them—where they appear to accumulate— merely as a reflection of an accidental, purely statistical trend, I am not, of course, implying that synthesis and/or analysis may not be provoked by specific factors. As I have argued repeatedly in the previous chapters, synthesis, for example, is always dependent on the semantic relevance between speech units, and both semantic and morphosyntactic reanalyses typically play a crucial role in the formation of synthetic constructs. But just as there normally is multiple causality in most longterm language changes, so, too, the formation of analytic or synthetic constructs depends on multiple factors, often unrelated. These may be intra- as well as extragrammatical, and either foster or inhibit the diffusion of analytic and/or synthetic tendencies. In chapters 3 through 6 we have had occasion to discuss a number of mostly internal factors—phonotactic compatability, separability, transportability, etc.—which crucially determine the extent to which units synthesize.15 Less attention has been paid to external factors, though these are just as important to our question. Stratal influence, for instance, may trigger the adoption of more analytic modes of expression, as has perhaps occurred in French and Romansh where Germanic possibly served as the model for the new preverbal P/N marking. Or, a change in the social structure of a community may intervene and halt the generalization of a particular feature headed for synthesis (e.g., the rise and eventual dominance of Parisian French drastically reduced the frequency of postverbal emphasizers other than pas, thereby slowing down and eventually blocking their potential synthesis with ne). There are, of course, many possibilities for change entailed by intra- and extralinguistic factors, and it would be presumptuous to attempt to identify them all here. In each case where a particular series of developments has yielded a more synthetic (or analytic) formation, a fine-grained study of the various changes is needed to determine the causes for the development There are no grounds to suppose that a speech unit must travel down a determined hierarchy of changes, a kind of prescribed causal slope, to arrive at a new level of syntheticity (or analyticity). Quite the contrary, given the many variables in language change, it would indeed be surprising if two or more speech units ever acquired their synthetic (or analytic) character for precisely the same reasons. By claiming that in order to arrive at a causal explanation of analytic/synthetic fluctuations we must study the source of each individual change in the trajectory of a speech unit from analysis to synthesis (or vice-versa), I do not mean to imply that we can thereby arrive at the ultimate causes of these changes. For, despite respectable advances in linguistic science, the ultimate cause of structural (or, for

On drift, cyclicity, and the analytic/synthetic spectrum

187

that matter, any other) language change are not yet known, nor do linguists even agree on the possibility of "explaining" in historical linguistics (cf. Lass 1980). While an increase in frequency may, for instance, be said to cause (or, perhaps better, condition) the synthesis of two relevant juxtaposed items, the real reason for such frequency increases remains unexplained. Similarly, certain types of phonotactic conditions frequently appear to induce the phonological fusion of relevant juxtaposed items, but here, too, the ultimate cause of such a development remains unknown. The "causal explanations" one can offer for changes in the analyticity or syntheticity of a speech unit are, then, in a sense no more than inductive generalizations based on the study of recurrent diachronic events. The fact that we are as yet unable to determine the ultimate cause(s) of language change should not deter us, however, from trying to understand the mechanics of synthesis and/or analysis. As demonstrated by our investigation into a number of Romance developments, a careful study of synthesis and/or analysis can, for instance, not only give us a better insight into language change at large, but also help us discover a set of principled reasons for why some, but not other, units have altered their morphological character. We were, for instance, able to establish that semantic relevance, frequency of use, syntactic flexibility, etc. all crucially determine the extent to which an analytic unit may undergo synthesis. 7.43 Prediction and reconstruction Just because it is not known what ultimately causes languages to change does not mean that certain developments cannot be predicted or reconstructed. As Lass (1980:13) correctly points out, the fact that an event can be predicted does not mean it can be explained, though any (correct) explanation inherently involves correct prediction. With the linguistic tools presently available, the most promising approach to non-explanatory prediction and/or reconstruction of analytic/ synthetic developments appears to be that based on inductive generalizations. By studying changes in the morphological structures of historically well-documented languages, it is possible to arrive at rough taxonomic generalizations about how and when analysis and synthesis occur. These findings can then be used to project the long- term behavior of morphological constructs. Thus, we can agree with Sapir (1921:155) that the direction of certain changes can, in a sense, be inferred from the past history of a language. Where two (or more) speech units show strong signs of semantic, phonological, and/or morphosyntactic synthesis (e.g., in Gév pas non 'not (at all)'; ISF J'comprends Ί understand'; Sp voy-a-salir Ί am going to go out') it may be

188

Analyticity and Syntheticity

claimed with reasonable certainty that these will eventually fuse into an unanalyzable whole. No matter how large the data base of inductive generalizations, the hypothetical claims about past or future analytic/synthetic tendencies must, however, always be considered highly probabilistic and speculative. Given the virtually unlimited variables in language change, projections and/or reconstructions based on taxonomic generalizations alone are really no more than better-than-average guesses endowed with certain statistical reliability. One cannot, for instance, assume that because a number of Egyptian, IE, or Uralic speech units have all given evidence of a repetitive linguistic cycle from analysis to synthesis (or vice versa), it may be concluded with absolute certainty that this type of change has affected every word class of these languages. Nor must we believe, as early Indo-Europeanists like Bopp did, that the inflectional forms of Sanskrit, Greek, Latin, etc. are necessarily based on earlier syntactic construction groups of independent words. For although it is certainly true that in the vast majority of cases morphology recapitulates the syntax of an earlier stage (Givón 1979), there are sufficient counterexamples to the claim that ALL bound morphemes derive from separate words.16 In this connection it must also be noted that morpheme order does not always faithfully reflect the syntax of the word from which it was derived. Comrie (1981:210) shows some examples, albeit very rare, where the order of morphemes changed after they became fused together. A similar development is discussed by Bybee (1985:41) with regard to some verb forms in Pengo, a Dravidian language. By reconstructing earlier syntax on the basis of morphology one also runs the risk of reconstructing something which may not at all reflect the dominant, most characteristic syntax of a formerly free morpheme. To cite just one example, Gév. pasnon 'not (at all)' (cf. the previously cited La Marjano manquet PAS NON RES de ço que l'abiô dich) leads one to posit an earlier syntax in which the two elements were predominantly juxtaposed, thus allowing the hypothetical sequence * [SUBJECT] VERB - PAS NON · OBJECT. But as I have argued in 6.4.3, there are good reasons to believe that independent non was obligatorily sentence-final (as in "'[SUBJECT] VERB - PAS - [OBJECT] - NON) and, therefore, generally not adjacent to pas. It was only after the generalization of a pattern in which pas and non were "naturally" contiguous (a situation which obtained only in short simplex sentences lacking [postverbal] nominal objects) that non could be placed before objects. Since by that time pas and non must already have evolved into a single complex, it cannot plausibly be argued that independent non was ever sentence-internal, as the morphosyntax of modern non-pas appears to suggest So while modern non pas does not really vitiate the claim that syntax gives rise to morphology (in short, simplex sentences independent pas and non must, after all, have been juxtaposed occasionally), the ex-

On drift, cyclicity, and the analyticlsynthetic spectrum

189

ample illustrates not only that the most characteristic feature of independent non (i.e., its sentence-final placement) cannot be reconstrued on the basis of the modem reflex alone, but also that morphology does not necessarily originate from the fossilization of a formerly predominant word order. "Exceptional" cases such as pas-non or those discussed by Bybee and Comrie seriously undermine the validity of reconstructing earlier syntax solely on the basis of morpheme order. Even stronger reservations must be raised with regard to attempts at determining the age of the structure of speech units on the basis of their level of syntheticity. In his discussion of the structural tendencies in Uralic languages, Tauli makes the following pertinent remarks: It is a well-known fact that within the borders of a language family the isolated and peripheral languages are conservative and at the same time more complex and more synthetic than other languages of the same family, whereas the newer traits occur in general in the center of the language family. It may be concluded from this that a relatively synthetic or complex character is in general a distinctive feature of archaism. The occurrence of common structural traits in large contiguous areas also in unrelated languages, as well as the unbroken transition of structural features between genetically unrelated neighboring languages, indicate that the principles which are valid within the borders of a language family, are also applicable to the whole of the world. Consequently the peripheral strongly synthetic languages on the linguistic map are in general more conservative and archaic, whereas the structure of the languages in the central areas is younger. (1966:290, emph. mine) Despite the fact that Tauli couches his hypothesis in non-universal terms, serious objections must be raised against it. For one, the Romance data in chapters 4 through 6 unambiguously show that the rate at which an analytic/synthetic cycle turns varies so greatly, depending on the speech unit, that syntheticity cannot possibly serve as a reliable indicator of structural archaism. Consider, for instance, the post-16th-century French Creole negator napas 'not' (< Fr n' a pas = NEG has NEC 'he/she/it does not have') whose relatively synthetic character is not at all synchronous with its "young" age. 17 Similarly, the relatively synthetic modSp voy-air Ί am going to go* cannot be said to be an archaic formation, nor could it be determined from the structures of modSp iré and voy-a-ir that the origins of the former predate those of the latter by almost two thousand years. Tauli's hypothesis loses further credibility if we recall that analytic/synthetic cycles have transformed the morphological structure of some Romance words several times over while the

190

Analyticity and Syntheticity

structure of others has remained entirely unchanged. Contrast, for instance, the relatively recent formation of the synthetic Gascon double negatives pas-ges or pas-cap 'not'—both the product of a third post-CL negation cycle (see §6.4.2.2)—with the highly conservative yet analytic It non 'not' (< CL non). If the application of Tauli's hypothesis to the Romance data cited above yields contrary-to-fact results, this does not mean that Tauli's claim about the relative archaism and conservatism of the world's synthetic languages is inherently incorrect It simply means that such correspondence may be purely accidental, and not directly related to the question of analyticity and/or syntheticity.

7.5 Summary and conclusion In this chapter I have argued that no single factor can account for changes in the analyticity and/or syntheticity of speech units. Except for the culture-based hypotheses and Lakoffs teleologica! explanation, I have not rejected previous proposals regarding the origin of analysis and/or synthesis—Slobin's "four charges," Pulgram's "expressiveness," or theories anchored on phonological arguments—but have sought to point out their inadequacy because they take into consideration only a select few of all potential causal factors. Viewing analysis and synthesis simply as a consequence (rath»' than a cause) of language change, I have suggested that every innovation, no matter how minute or insignificant it may be in the overall picture of language development, ultimately affects the direction a speech unit takes on the analytic/synthetic axis. Because each change must have causes of its own, it follows that if we accept the premise that analysis and synthesis are a consequence of change, then there must be just as many causes for analysis and synthesis as there are reasons for change. As regards Lakoffs discussion of analysis and synthesis, we saw that the puzzling "facts" about the so-called IE drift towards analysis are far more amenable to explanation (and, therefore, more easily divested of their mystical qualities) once they are brought into proper perspective. Analytic and synthetic tendencies have occurred at all times in the histories of IE languages, and the observed directional drift merely reflects a statistical trend whose unidirectional momentum has been exaggerated as a result of a widespread failure to recognize signs of synthesis. In the latter section of this chapter I have concluded that only modest claims can be made concerning the possibility of "explaining" why languages change their morphological structure. Though careful investigation of long-term diachronic trends helps us understand better what conditions (semantic, phonological, and/or morphosyntactic) favor, or perhaps even induce, synthesis or analysis, the ultimate

On drift, cyclicity, and the analytic/synthetic spectrum

191

causes of these changes remain unknown. Consequently it has been suggested that current hypotheses about the origins of these developments are best viewed as no more than "reasonable" attempts at explanation. Because the origins of language change remain unknown, and because there appear to be so many factors involved in the formation of analytic and/or synthetic units, I have concluded that it is impossible to reconstruct or predict with certainty fluctuations in analyticity/syntheticity. If we seriously consider the proposition that changes in analyticity and syntheticity originate with stimuli from the structure of a language as well as from society, we should not, however, be surprised that truly reliable predictive or reconstructive hypotheses are not readily available. As Weinreich, Herzog, and Labov rightly point out, 'this is a problem common to all studies of social behavior' (1968:186), and there is currently no reason to assume that we may be able to overcome this difficulty in the near future.

Conclusion I began this investigation by focusing on the history of the concepts analytyic and synthetic, specifically, how these have been applied to the evolution of Romance morphology, and more broadly, with respect to their usefulness as parameters for morphological typology. One important thing this survey has uncovered is the striking vagueness and ambiguity with which the terms ANALYTIC and SYNTHETIC, hence the concepts themselves, are used and understood in the literature. Though the reasons for this confusion (summarized below) are multiple, I have sought to show that the application of analytic and synthetic to entire languages rather than simply to individual speech units lies at the core of the problem. Having examined how the lack of a clear delineation of linguistic units such as the word perforce thwarts any attempt at satisfactorily quantifying languages according to their morphological "tightness," I then offered reasons in chapters 4 through 6 for why these frustrated endeavors should not lead us to neglect the potential usefulness of, and wealth of insights that can be gleaned from, the analytic/synthetic concepts in areas other than language typology. Applying the parameters analtyic and synthetic not as quantifiable absolutes but as the rough measure of the morphemic interdependency of speech units, I have attempted to demonstrate that the overall effect of recurrent and often unrelated diachronic events cannot be grasped adequately without appeal to the factor of syntheticity/analyiicity. Using changes in the P/N marking from Latin to French by way of example, chapter 3 revealed, for instance, how the application of our concept affords a more global view of the differences between OFr je and CL ego, and how these differences together conspired to crucially transform subsequent French verbal morphology. The usefulness of our concept was brought out again in chapters 5 and 6 where the tying together and interrelating of individual pertinent diachronic changes provides a better understanding of how over time lexical items like CL habeo, Sp voy, Fr pas, etc. have evolved into units of an entirely different sort— both grammatical and pragmatic. In analyzing a number of analytic/synthetic cycles within Romance, a set of principles begins to emerge by which semantic, morphosyntactic and phonological changes proceed to alter the interrelatedness of morphemes. Semantic synthesis has, for instance, been seen always to precede all other types of synthesis. One important point this investigation has brought out is the marginal importance of phonology to the question of synthesis. Although the data analyzed in chapters 3 through 6 strongly support the notion that long-term synthesis usually involves the interaction of all three levels—semantic, morphosyntactic, and phonological—my discussion of examples like Fr maintenir or Sp alegremente provides incontrovert-

194

Conclusion

ible evidence that, contrary to opinio communis, phonological synthesis is not a prerequisite for extensive synthesis. Syntactic factors, on the other hand, have invariably been shown to play a crucial role in the morphologization of once independent units. Particularly clear examples of the centrality of word order have been given in the chapter on predicate negation, where we have seen how the linear distance between negators and emphasizers largely determines the rate at which these elements fuse. A primary aim of this study has been to offer a more nuanced picture of morphosyntactic developments in Romance, and to rectify certain gross oversimplifications in the analysis of typological change. Testing the widely-held assumption that Romance, and particularly French, has shown a virtually unidirectional tendency— or, to use Sapir's term, a "drift"—towards analysis, we have found that the data simply do not support such a claim. Recurrent negation cycles, for instance, have introduced numerous changes in the directions of both analysis and synthesis; similarly, the discussion of recent changes in the ISF system of P/N marking illustrates that synthetic trends (e.g., the substitution of ISF il y pense for FSF il pense à lui) occur side-by-side with strongly analytic ones (e.g., the spread of coreferential subject clitics). As shown by the numerous quotes presented at the outset of chapter 4, the oversimplification and confusion in the description of morphological systems has been so extensive that modem French, one of the world's better known languages, is currently being labeled as both analytic and synthetic. By examining diachronic as well as synchronic descriptions of French, we have attempted to show that the causes for this terminological sophism are multiple. In classifying units such as j'ai fait as analytic, various scholars have failed to take into account the—for the question of syntheticity and analyticity—important changes which the verb-phrase has undergone in familiar speech. In addition, many of those advocating the analyticity of French have judged the morphological tightness of certain speech units exclusively against the backdrop of functionally similar, but etymologically only indirectly related, CL units, oblivious to how arbitrary the results of such an approach may be. But unquestionably the most serious "abuse" of the terms analytic and synthetic has been shown to stem from the widespread practice of translating differences in syntheticity between speech units into a non-relative, binary opposition analytic vs. synthetic, and then equating the result with the overall morphological complexity of the respective units. It is through just such practices that potentially useful labels such as "analytic future" or "synthetic future" lose their meaningfulness, and linguistic terminology—if not clearly defined and understood—runs the risk of obscuring or even distorting the facts of language. Given that practically all post-19th-century discussions of analyticity and/or

Conclusion

19S

syntheticity have—however briefly—addressed the question of the possible causes for the recurrent changes in morphological interrelations, it is only Atting that almost an entire chapter be devoted to this issue. Viewing analysis and/or synthesis as a consequence rather than a cause of language change, I have suggested that there is no single causal factor responsible for these processes. Rather, all causes of language change—both external and internal—contribute to altering morphological interrelations. Because the ultimate origins of such changes remain unknown, investigations of long-term diachronic trends can merely help us understand better the conditions that favor analysis and/or synthesis; they cannot advance us any further in solving the actuation puzzle. Though the main objective of the present investigation has been to rectify and/or clarify much of what has been said about analyticity and syntheticity, I do not claim that the terms analytic and/or synthetic are now precise or objectively defined. As I hope to have demonstrated, these concepts can only be used meaningfully in relative terms, and any attempt to apply them as quantifiable absolutes necessarily leads to the kinds of rigid morphological taxonomy which have confused the issue ever since the concepts were first introduced.

Notes Chapter one - History of the terms "analytic" and "synthetic" 1. 2.

3.

4.

5.

6. 7.

8.

I am not concerned here with the terms "analyticity" and "syntheticity" as applied in semantic theory (see Katz 1964 and 1974). What I refer to as "morphological typology" (or simply "typology") is labelled "morphological classification" in 19th-century linguistic discussion. The two terms are virtually synonymous, the former is first attested in 1928 (see Greenberg 1974:36). For a description of the various techniques of language classification see Home (1966:1-8). For general summary reading on typological linguistics see Manoliu (1985), Current Trends in Linguists: Diachronic, Areal, and Typological Linguistics, volume 11, esp. 11-48. For a chronological list of pre-scientific attempts at language classification see Meillet (19S2 [1924] rxviii-xxiii). A more extensive discussion of early language classification is given in Robins (1973). In Ueber die Sprache und Weisheit der Indier (1808), F. Schlegel establishes two classes of languages: those with internal change of the root (Sprachen durch Rexion) and those with affixation (Sprachen durch Affixa). Schlegel's usage of the terms affix and inflection are not in accordance with generally accepted usage. We may perhaps be excused, says Greenberg (1974:37), if we fail to grasp clearly what von Schlegel intended by his terminology when we read in a letter from von Humboldt to Friedrich's brother August (Dec. 30, 1822) that he could never understand clearly what Friedrich meant by inflection. Observations sur la langue et la littérature provençales (1818). For details of his classification see below. Humboldt published thoughts on "Klassifikation" for the first time in 1822 in "Ueber das Entstehen der Grammatischen Formen und ihren Einfluß auf die Ideenentwicklung" (= Humboldt 1963:31-63, esp. p. 46). He once again exposed these ideas in the famous Introduction to his major work Ueber die Kawisprache auf der Insel Jawa (1836-40), published posthumously. The often cited introductory chapter is entitled "Ueber die Verschiedenheit des menschlichen Sprachbaus und ihren Einfluß auf die geistige Entwicklung des Menschengeschlechts." For more detailed information on early language typology I refer the reader to Arens (1969 [1955]), Kuznecov (1956), Home (1966), Schmidt (1967), Robins (1967), Entwistle (1953:230ff), Coseriu (1972), Haarmann (1976),

198

9.

10. 11. 12.

13. 14.

15.

16.

Notes and Comrie (1981). Although the Schlegel brothers and Humboldt were clearly the main figures in the development of morphological studies, the beginnings of language typology go back to the 18th century. For particulars see Haarmann (1976:12-13) and especially Monreal-Wickert (1976:215). A quarter of a century earlier, R. A. Sicard had already used the term analytic in his Cours d'instruction d'un sourd-muet de naissance pour servir à l'éducation des sourds-muets. For an insightful discussion of the rise of this term, see I. Monreal-Wickert (1976). Monreal-Wickert's summary table of 18thcentury works which most likely influenced A. Schlegel's Observations is reproduced in Haarmann (1976:13). For a recent, more extensive elaboration on these basic types see Comrie (1981:39-50). Early linguists typically spoke of "letters" rather than "sounds." A distinction between the two was not consistently made until much later. Schlegel's tripartite division has survived practically unchanged down to our time. His "langues à inflexions" is rendered by later linguists under a variety of terms. Among the better known names are "inflective" (Sapir 1921:123), "inflecting" (Lyons 1968:189), "inflectional" (Bynon 1977:262), and "fusional" (Comrie 1981). For discussion of a number of inconsistencies in A. Schlegel's analytic/synthetic parameters see Lehfeldt and Altmann (1975:61). The following remark regarding the superiority of the classical languages reflects a view found throughout A. SchlegePs writings: 'Je l'avoue, les langues anciennes, sous la plupart des rapports, me paraissent bien supérieures (p. 25).' In a revealing article, Coseriu (1972:115-116) suggests that A. Schlegel may have culled the basis for his more refined analytic/synthetic distinction from Adam Smith's mid 18th-century Dissertation on the Origin of Language (for a discussion of its debated date of publication see Coseriu p. 47n3). Though Adam Smith did not use the terms analytic/synthetic, the ideas expressed by A. Schlegel are, by and large, already present in Smith's writings. One of the major differences between the two is that A. Schlegel's distinction applies only to a specific subgroup of languages (the agglutinating). It is noteworthy that despite Schlegel's acquaintance (1818:25, 85n6) with Smith's work, he apparently did not consult the Dissertation while writing his Observations. On the Origin of the Species appeared in 1859. When Darwin's theory reached Germany in 1860, it was at first rejected by most established scientists who were und«- the influence of the earlier nature-philosophy. Not surprisingly, younger scholars were more sympathetic towards the new ideas, and it was

Chapter one - History of the terms

199

through their positive reaction that evolutionary theory was extended to spheres outside biology. For further reading on this topic see Mason (1975 [1956], chap. 34). 17. A. Schlegel states that the words of formless languages such as Chinese consist merely of "sterile roots" (p. 14). He further argues that De telles langues doivent présenter de grands obstacles au développement des facultés intellectuelles; leur donner une culture littéraire ou scientifique quelconque, semble être un tour de force; et si la langue chinoise présente ce phénomène, peut-être n'a-t-il pu être réalisé qu'à l'aide d'une écriture syllabique très-artificiellement compliquée, et qui supplée en quelque façon à la pauvreté primitive du langage. (1818:14) 18. The idea that languages develop like living organisms was not new to the 19th century. Already three centuries earlier, Joaquim Du Bellay (Chamard 1948:24) used analogous vegetation metaphors to describe the development of French (vis-à-vis Latin) into a language of culture. 19. "... parce qu'elles [= the "organic" Classical languages] renferment un principe vivant de développement et d'accroissement, et que'elles ont seules, si je puis m'exprimer ainsi, une végétation abondante et féconde' (A. Schlegel 1818:15). 20. Despite this apparent decay, A. Schlegel did not consider the modern IE languages to be "decadent." He in fact classed them among the world's most ideal tongues, fully capable of expressing the needs of their societies: 'Je l'avoue, les langues anciennes, sous la plupart des rapports, me paroissent bien supérieures. Le meilleur éloge qu'on puisse faire des langues modernes, c'est qu'elles sont parfaitement adaptées aux besoins actuels de l'esprit humain dont elles ont, sans aucun doute, modifié la direction' (1818:25). 21. On the philosophy of language of German romanticism in general see Fiesel (1929). Cassirer (1945:115ff) shows that the typological concepts embraced by Humboldt are ultimately related to Goethe's idealistic view of the morphology of organic nature. 22. As Coseriu (1972) points out, for over a century there has been a misinterpretation of Humboldt's writings on morphological classification. It has generally been claimed that Humboldt actually adopted the Schlegelian tripartite schema of isolating, flexional, agglutinative languages to which he then supposedly added a fourth category, incorporating languages. Coseriu shows that Humboldt has merely furnished the terminology of the well-known division, and even though Humboldt did furnish the terminology, he did not suggest such a quadripartite a typology. Coseriu further points out that the propagation of the false interpretation of Humboldtian views is probably due to

200

Notes

Schleicher's later influence. For a further, brief discussion on the subject see H a a r m a n n (1976:55). A very recent treatment of Humboldtian typology is given in Ramat (1984:165ff) "La tipologia linguistica di Wilhelm von Humboldt." 23. The passage just cited here is not the only place where Humboldt rejects Schlegel's views on syntheticity and analyticity. In Aus dem Lateinischen hervorgegangene Sprachen (1963:640-649), though he refers neither directly to A. Schlegel nor to the concepts of analyticity or syntheticity, Humboldt emphasizes that fundamentally the Romance languages are in no way typologically different from their Latin ancestor ( ' . . . e s entsteht keine reine neue Schöpfung, die veraltete Sprache dauert...' [p. 646]). Though it is true that he speaks of a "degeneration" (Verfall and Zertrümmerung) of the Romance tongues as well as of structural differences, he maintains (p. 643) that the Romance languages have an inflection system not unlike that of their Classical Latin ancestor. As proof of his claim, he cites the grammatical unity of the widely dispersed Romance tongues. Given the arguments expressed here, we may, safely I believe, regard Humboldt's refusal to make a typological distinction between classical Latin and the modern Romance tongues as yet another rejection of Schlegel's analytic/synthetic division. 24.

Basing his classes on root forms, Bopp avoided the terms synthetic and analytic as well as those of "flexion" and "agglutination." It would be wrong, however, to conclude that Bopp was altogether unfavorably disposed to A. Schlegel's division. He remarked (1833:112) that 'wenn nun also Fr. v. Schlegel's Sprach- Eintheilung ihrem Bestimmungsgrunde unhaltbar ist, so liegt doch in dem Gedanken an eine naturhistorische Classificirung der Sprachen viel Sinnreiches. Wir wollen aber lieber mit A. W. v. Schlegel drei Klassen aufstellen.' Bopp never explained explicitly why he did not adopt the synthetic/analytic division. Pott, on the other hand, chose to operate with A. Schlegel's tripartite division to which he then added a fourth, the incorporating type. Pott, rather than Humboldt (as has generally been claimed [see Coseriu 1972:135]), thereby became the first to use the quadripartite division which to this day remains the standard morphology-based classification. For a brief summary of Bopp's and Pott's arguments see Home (1966:15-16) (note, however, that references to Pott in both Steinthal [1860:10] and Home [p. 23] are wrong). Schlegel's proposal did not fare much better in the United States. Du Ponceau (1838:447), noted for his studies on Amerindian languages, stated in a letter to the Société Philosophique Américaine that 'la science n'est pas encore arrivée à un point de maturité, sufisante pour permettre d'entreprendre une

Chapter one - History of the terms

25.

26.

27.

28.

29.

30.

201

clasification [sic] exacte et complète de toutes les langues qui existent.' In a footnote on the same page, the anonymous translator of Du Ponceau's letter adds the following, more optimistic personal remark: 'M. le baron Alexandre de Humboldt observe avec raison que la division des langues en analytiques et synthétiques n'est pas satisfaisante . . . . Cela n'empêche pas qu'il n'y ait des langues analytiques et des langues synthétiques, et qu'on ne puisse les distinguer ainsi en attendant mieux.' The concept of polysynthesis was made famous above all by Humboldt (1963 [1835]:528) who termed it Einverleibung. In the twentieth century, Saputi 921) contributed much to its further spread (see §1.3 below). It is noteworthy that the term has frequently come to be used interchangeably with "incorporating" but that recent scholars (e.g., Vennemann [1982:333]; Comrie [1981:42]) have begun to make a clear distinction between these two. In their works, incorporation is a special, more complex subtype of polysynthesis which allows several lexical morphemes in one word. According to Steinthal (1850:12-13) the term "polysynthetic" is first used by Du Ponceau in (1819). However, the source quoted by Steinthal (Transactions of the American Philosophical Society, 1819, vol. 1:399) does not exist Even in Die Sprachen Europas, Schleicher's only publication to mention the analytic/synthetic division, the author's exposition is hardly more than fragmentary. After hinting that he will apply a subdivision to both inflectional and agglutinating languages (p. 10), he never really returns to the topic again. His later remark that 'was früher in einem Worte gesagt wurde, löst sich in mehrere Wörter auf (z.B. lat. matri ital. alla—aus ad la—madre, franz. à la mère', ...), weshalb man auch diese Sprachen, im Gegensatze zu den älteren, analytische genannt hat (p. 18)' is not followed up by a more in-depth discussion on the subject and provides nothing that is not already contained in Schlegel's writings. A similarly negative attitude is displayed a few years earlier by A. Pott (1833:154). Given that Pott falsely attributes the analytic/synthetic distinction to both Schlegel brothers (rather than just to August), it may also be the case that he was not entirely familiar with the details of the new typological division. As far as I have been able to determine, A. Schlegel's multi-page exposition remained the only substantial, non-fragmentary treatment of the subject for almost exactly a hundred years. Misteli revised the typology of Steinthal, who, as mentioned above, had rejected Schlegel's analytic/synthetic breakdown. He significantly redefined some of Steinthal's concepts but without making reference to Steinthal's dis-

202

31.

32.

33.

34.

35.

Notes cussion on analyticity and/or syntheticity. Not surprisingly, in Misteli's work the terms analytic/synthetic are absent altogether. For a brief summary of Fink's work see Home (1966:25-26). A very critical review of Fink's ideas is given in "La classification des langues" in Benveniste (1966). The attempt at grouping languages according to non-genetic criteria was so unsuccessful that linguists of the caliber of Antoine Meillet (1954 [1924]: 1) later proclaimed: 'La trop fameuse classsifìcation en langues isolantes, agglutinantes et flexionelles ne se laisse pas poursuivre exactement, et, pour autant que'elle se laisse formuler, elle n'a ni portée scientifique ni utilité pratique. La seule classification qui ait une valeur et une utilité est la classification généalogique, fondée sur l'histoire des langues.' Though the author had discussed the need for an analytic and synthetic grammar in two previous publications (Gabelentz 1875 and especially 1878), it is in his 1881 and 1891 works that his ideas are laid out most fully. The chapter in which the question of analysis and synthesis is addressed in greatest detail ("Bestandteile des grammatischen Wissens; die beiden Systeme") underwent only minor revision between the first (1891) and second editions (1901) of the Sprachwissenschaft. Gabelentz (1972), a reprint of this second edition, contains a useful introduction by Eugenio Coseriu to Gabelentz's less than transparent pronouncements about language. As Coseriu points out in the introduction to Die Sprachwissenschaft (1972 [1891/1901]:30), Gabelentz' synthetic grammar resembles a modern transformational grammar not only because of similarities in the overall approach, but also because of a certain parallelism in terminology (see especially Gabelentz's chapter on "Verwandlung der Sätze in Satztheile" [1901:463-70] in which terms like Verwandlung, verwandeln are frequently used, albeit without formalization implicit in the transformational rules of earlier models of generative grammar). Already as a young scholar, Sapir objected to the evolutionary theory. As early as 1911 he states: It was quite customary formerly to look upon the three main types of morphology as steps in a process of historical development, the isolating type representing the most primitive form of speech at which it was possible to arrive, the agglutinating coming next in order as a type evolved from the isolating, and the inflecting as the latest and socalled highest type of all. Further study, however, has shown that there is little to support this theory of evolution of types. (224; emph. mine)

Chapter one - History of the terms

36.

37.

38. 39.

40.

41.

42.

43.

203

In Language (1921) we read: 'There is a fourth reason why the classification of languages has generally proved a fruitless undertaking. ... This is the evolutionary prejudice which instilled itself into the social sciences towards the middle of the last century and which is only now beginning to abate its tyrannical hold in our mind' (Sapir 1921:130). Despite Sapir's strong feelings against the evolutionary bias, his own scholarship, though in a different way, is at times reminiscent of the romanticism typical of early 19th-century linguists. This is particularly true with regard to Sapir's partial rejection of his own ideas on analyticity and syntheticity (discussed below). For details not specifically related to his work on the analytic/synthetic dichotomy, see especially chapter 6, "Types of Linguistic Structures." A good summary is given in Home (1966:27-36). 'Various classifications have been suggested, and they all contain elements of value. Yet none prove satisfactory. They do not so much enfold the known languages in their embrace as force them down into narrow, straight-backed seats.' (Sapir 1921:129) On the Sapirian concept of drift see §7.2 and §7.4.1 below. Sapir's emphasis on flexibility in his typology is particularly evident when, after first classifying the inflecting Latin and Greek as synthetic rather than analytic, he then goes on to describe them as not notably synthetic if viewed relative to other languages that resemble them in broad structural respects (p. 137). I deliberately say "characterization" and not "classification" because careful reading of Sapir's chapter 'Types of Linguistic Structure" reveals that he never meant to provide an all-encompassing typology with pre-established slots for each language. He specifically states: 'we are too ill-informed as yet of the structural spirit of great numbers of languages to have the right to frame a classification that is other than flexible and experimental' (p. 149). His table displays as many as twenty-six different language types. Even the most elaborate earlier typologies (Steinthal, Misteli, Fink) never exhibited more than eight language classes (Home 1966:32). Sapir claims that the former process is demonstrable for English, French, Danish, Tibetan, Chinese, and a host of other languages. The latter tendency may be proven for a number of Amerindian languages, e.g., Chinook or Navaho. Witness Ben veniste (1966:113) who assesses Sapir's typology in the following critical manner 'Mais la difficulté est dans le maniement de cette classification, moins pour sa complication qu'à cause de Y appréciation subjective qu'elle comporte en maints cas' (emph. mine).

204

Notes

44. After proposing two separate axes of classsification (degree of synthesis and technique [e.g., fusion, agglutination]), Sapir falls back into a vein of romantic Humboldtian scholarship and rejects these two axes, calling them purely external and material, and moves on to what he considers a much deeper and more reliable classification 'based on the nature of the concepts expressed by the language' (p. 144). 45. I translate Velten's "grammatisch" by "morphological." He states: 'die Unzulänglichkeit der Termini Analyse und Synthese [beruht] zunächst darauf, daß oft nicht klar genug ausgedrückt wird, ob von semantischer, syntaktischer o d » grammatischer Analyse die Rede ist.' (1935:3) 46. Velten (p. 21) flatly denies even the possibility of finding a cross-linguistic definition for the word: 'Ferner ergibt sich, daß es für den Begriff des Wortes eine allgemeingültige Definition gar nicht geben kann; denn er ist von dem Wort der klassischen Sprachen abgezogen, paßt schon auf eine ganze Reihe von modernen indogermanischen Sprachen nicht mehr, und läßt sich nur auf die wenigsten nicht-indogermanischen Idiome sinnvoll anwenden.' 47. Before ending the discussion of Sapir's work, it is only fair to point out that many linguists have subsequently (ab)used the reincarnated terminology in a way never intended by the author of Language. One such abuse common among modern linguists (for references and examples see chapters 3 and 4) is to apply Sapir's term "synthetic" (which, as mentioned above, he considered useful in defining certain drifts) as an absolute and to speak of a "synthetic language" (rather than tendency) as if syntheticity were a well-defined category within an overall typology. 48. As in the works of numerous other authors, further on in his Linguistique générale et linguistique française. Bally returns to the old practice of using the problem-ridden terminology despite his own earlier objections to it: 'Le français semble animé de deux tendances contradictoires: l'une engendre des formes analytiques, l'autre les condense et rapproche la langue du type synthétique' (p. 191). 49. Witness also Tesnière (1932) who rightly points out that the concepts of analyticity and syntheticity frequently hinge on arbitrary orthographic word-divisions. He argues that taking into account strictly orthographic criteria has led linguists to the erroneous claim that French, for instance, is an analytic language. Tesnière then suggests that French be considered synthetic (rather than analytic) because structures such as je le lui ai dit are essentially a single phonetic word (je-le-lui-ai-dit). His short article, although insightful in several respects, unfortunately does not, however, elaborate on criteria for non-orthographic word division and thus fails to provide better working definitions for

Chapter one - History of the terms

205

analyticity and syntheticity. 50. See for instance Vennemann (1982). 51. One wonders whether it is not ultimately this impracticality which later led Tauli to reformulate his definition. In Structural Tendencies of Languages (1958), where ample use is made of the analytic/synthetic terminology, he states: The analytic tendency in a broader sense of the word is manifest in several tendencies, viz. in all those tendencies which result in making the expression more analytic in accordance with the eleven points laid down by me in a previous paper. In the present chapter we are using the term "analytic tendency" in a stricter sense, understanding it mainly as a tendency toward forms of expression constructed by means of auxiliary words. (156) Regarding syntheticity he (p. 82) says: The synthetic or agglutination tendency occurs, to a greater or smaller extent, in all languages and at all times. This is due to the linear character of the language: the succession of linguistic signs.... Synthesis is complete if the agglutinated words form a unity which cannot be analyzed by the language instinct, a so-called unmotivated or arbitrary simple word. (82) Further on Tauli (p. 173) adds: If this tendency toward unanalyzable arbitrary lexemes is considered from the point of view of lexemes, which formerly could be analyzed into stems and derivational affixes, it may be characterized as a synthetic tendency resembling agglutination. (173) 52. Grcenberg's article appeared first in the obscure Festschrift for Wilson D. Wallis, Methods and Perspective in Anthropology: Papers in Honor of Wilson D. Wallis, ed. Robert F. Spencer (1954), and was reprinted in 1960. All citations are from the 1960 reprint According to Home (1966:32n27), unpublished versions were previously presented at the annual meeting of the Linguistic Society of America in 1949 and in a talk given before the Linguistic Circle of New York in 1950. 53. We recall that Sapir's semantically-based division led him to adopt the socalled "relational concepts." 54. Greenberg's main criterion is a version of "separability" (discussed in chapter 2). For particulars I refer the reader to Greenberg (1960:192-196). 55. As far as I have been able to determine, Greenberg returns only once to the is-

206

Notes

sue of word-division in his later writings (Greenberg 1963 [1957]: chapter 11, "The Definition of Linguistic Units"). The most serious drawback is not the questionable cases of word segmentation (one which comes immediately to mind is the problematic analysis of discontinuous morphemes of German verbs like anfragen where prefix-like an- is dislocated to the end of the sentence: ich FRAGE sie oft AN) but the failure to provide an explicit step by step outline (with examples) of his complex procedure for those wishing to test and eventually adopt his technique. 56. For similar calculations on other IE languages see Cogwill (1966:124). 57. Others like Ebeling (1960:135) see the approach as an operational definition which is 'easily the most successful of all attempts made thus far.' Contreras (1963:267), after applying Greenberg's formula to seven Romance languages (French 1.46, Sardinian 1.55, Rumanian 1.76, Spanish and Catalan 1.81, Italian 1.82, Portuguese 1.87), reaches the conclusion that the technique is appropriate for the classification of relatively similar or related languages. Using Greenberg's technique, Cogwill (1966) obtains the index of syntheticity reproduced in Table 2b; he admits (pp. 120-121), however, that arbitrary word and morpheme separation as well as a number of other problems seriously undermine the accuracy of his calculations. Table 2b. Cogwill's Analytic/Synthetic Index (Formula: Morpheme/Word) RV Asoka NT

OP

Go

PCS Horn

Hitt

Bg

2.56

2.41

2.31

2.29 2.07

1.95

1.90 1.82

2.52

2.45

NGk

RV = Vedic Sanskrit; Asoka = Asoka's Rock Edicts; NT = New Testament Greek; OP = Old Persian; Go = Gothic; OCS = Old Church Slavic; Horn = Homeric Greek; Hitt = Cuneiform Hittite; Bg = Bengali; NGk = Modern Greek (based on Cogwill 1966:125). According to Winter (1967:546) Glazov (1965) also applies the formula MORPHEME/WORD to obtain an index of syntheticity for Tamil and Uigur (Western China). Still others (e.g., Pierce 1966) approve of the new dimensions introduced but withold judgment about their usefulness. It would be wrong to conclude, however, that Greenberg's approach has met with unanimous approval. Most outspoken among the dissenters is André Martinet who, in a footnote, describes the new method as one that simply 'translates Sapir's scheme into currently fashionable jargon' (1967[1962]:67n2). This criticism seems overly harsh and fails to recognize the extensive differences in method, material, and interpretation between the earlier methods and Greenberg's (since Martinet had earlier dismissed Sapir's classification as a 'nearly tragic illustra-

Chapter one - History of the terms

58. 59.

60.

61.

62.

63.

207

tion of the pitfalls of psychologism' [1962:39], his assessment of Greenberg is not entirely surprising). Among the less virulent critics of Greenberg one might cite Arens (1969 [1955]:500), according to whom the 'sogenannte „innere Formlosigkeit" gewiss» Sprachen eine Illusion ist' He rejects the analytic, synthetic, or polysynthetic division as being 'zu rein quantitativ' (p. 504). For a discussion of the advantages of quantitative over earlier typologies see Lehfeldt and Altmann (1975:55-56). According to Ebeling, the sketchiness of Greenberg's methodology regarding word-division is not the only problem that besets his word analysis. Ebeling (1960:135) makes the valuable point that Greenberg establishes "the word" on the basis of features of arrangement of the constituent morphemes, but that these features are exclusively formal in nature. Since, however, morphemes have not only a formal, but also a semantic aspect, it would seem advisable to take into account also their arrangement in the semantic whole. The difficulties which arise in applying the formula proposed in Greenberg's pilot project are brought out in a study by Winter (1967). In one of his seminars, a group of students engaged in a test of Greenberg's method. For a 100word block in the second chapter of Bloomfield's Language, a group of eleven students obtained analytic/synthetic indices ranging between 143 and 175. This 32-point spread represents about a ten percent deviation on either side of the mean, and only two out of ten counts fell within a five-percent deviation from the mean. Given the sketchiness of the "word," the variance in results was not surprising. Of even greater concern is the fact that these incongruities were obtained when the language and text were identical for the entire group of students. Had Winter's test contained text samples from different languages, the practical infeasibility of Greenberg's formula would have been confirmed a fortiori. In the absence of ground rules for what is and what is not an analytic structure, it is difficult to cite an unambiguous illustrative example. Je vais aimer is here considered more analytic because, unlike aimerai, numerous constructs may interrupt vais aimer (e.g.,je vais toujours l'aimer). I shall return to this point on several occasions in subsequent chapters. This remark also holds for SkaliCká's (1966/1967) probings into the formulation of an "ideal typology" (which would include an analytic/ synthetic index). For discussions of SkaliCká's complex formulations see Haarmann (1976:5659); Lehfeldt and Altmann (1975:71-76). Kelemen's actual formula reads: 'Proportion der analytischen und synthetischen Wortkomplexe (p. 62).' To the readers' regret, the author fails to

208

Notes

explain what exactly is meant by "Wortkomplex." 64. The example and the calculations are mine. 65. One possible advantage is that this index would avoid at least one pitfall of the Greenberg index, namely the potential distortion of the degree of synthesis in cases where we find polysynthetic and purely analytic structures side by side. An illustrative example is the French verb phrase where tense, aspect, and mood are expressed by both synthetic and analytic constructs (cf., aimerai vs. vais aimer). In such a language, Kelemen's formula provides a breakdown into the ratio of strictly analytic vs. synthetic words, an analysis which obviously cannot be obtained from Greenberg's calculations (for an illustration of the differences in these two approaches see Table 3 below). 66. Though SlaviCková does not actually call her index "synthesis," it is obvious that she is operating with that concept. Accordingly, other writers (e.g., Altmann and Lehfeldt [p. 109]) discuss the passage under the heading "Synthetismus." 67. All examples cited are mine. 68. I translate German Kollokation as collocation, the term used by Vennemann. The term used by Sapir to denote the same concept is "juxtaposition." In light of Vennemann's statement 'dieser Ausdruck scheint mir treffender als Sapirs .Juxtaposition," der die Auffassung nahelegt, nur affigierend verbundene Elemente könnten fusionieren, während tatsächlich auch mutierte Elemente, nämlich Akzente und Töne sowie Töne mit Segmenten und Segmente mit Akzenten, fusionieren können' (p. 331), I prefer to adopt Vennemann's terminology. 69. As Vennemann points out (p. 330n7), his list of processes leading to synthesis is based almost exclusively on Humboldt (1963:31-63). 70. Vennemann's effort is even less comprehensible when one considers the fact that he cites some of the sources (e.g., Haarmann 1976) which point to the futility of non-quantitative definitions. The idea that languages ought no longer to be assigned to types as wholes, is of course, not an entirely recent phenomenon. For a brief account on this subject see Greenberg (1973:173-174). 71. Altmann and Lehfeldt (1973) recognize the difficulty in finding an acceptable word definition but maintain that in texts this problem can be solved rather easily by accepting the traditional graphic criterion of word separation: 'Die einzige Lösung, die die wenigsten Probleme aufwirft, scheint die Verwendung der graphischen Grenzen bei der Segmentierung des Wortes zu sein (p. 72).' Quine's point is well taken that 'the principles behind the printers' use of spaces are dim, and the relevance of such principles to any considerations of our own are doubly so' (1960:13-14). As I shall argue in chapter 2, the sim-

Chapter one - History of the terms

209

plistic use of printers' practice leads to serious distortions when applied to the Romance languages. Altmann and Lehfeldt are not the oily ones to have proposed "easy" solutions for word juncture. Hiorth (1958:9-20) concludes that 'linguists have not succeeded in formulating a set of useful specific properties of words, but that the ambiguities of "word" generally have no detrimental effects in linguistics as the ambiguities are mostly well known.' Chapter two -Word delimitation: in search of a universal 1.

Although some linguists deny the validity of the "word" as a universal linguistic category (thus Milewski [1931:248-249], who, working with Amalean Indian languages, concludes: 'syntactic groups occur in all languages of the world but not all languages contain words ...'; but see the critique in Harweg [1964:27]), few, if any, would deny that the concept of word is a salient one to the speakers of all languages. Recent remarks to this effect are found in Crystal ('WORD = A unit of expression which has universal intuitive recognition by native speakers' [1980:383]), Hartmann ('all native speakers of a language seem to have an intuitive idea of what is meant by the term "word" in its general sense' [1972:256]), and Krámsky ('we do not regard it possible that a language could exist which does not distinguish words ...' [1969:76]). The most widely cited observation regarding the universality of "the word" comes from Sapir (Language 1921:34-35): it is true that in particular cases, especially in some of the highly synthetic languages of aboriginal America, it is not always easy to say whether a particular element of language is to be interpreted as an independent word or as part of a larger word. These transitional cases do not, however, materially weaken the case for the psychological validity of the word .... No more convincing test could be desired than this, that the naive Indian, quite unaccustomed to the concept of the written word, has nevertheless no serious difficulty in dictating a text to a linguistic student word by word.

2.

It may be relevant to point out that Milewski (who, as mentioned, rejects the universality of the word) also suggests that 'French has a decidedly much less precise conception of the word as there is no uniformity in the morphological structure of syntactic groups (p. 250)' and that 'French must therefore be considered as a language essentially wordless containing only remnants of the former word character' (emph. mine). Apparently, earlier generations were somewhat more optimistic about finding

210

3.

4.

5.

6.

Notes a universal definition for the word. Marouzeau, for example, uncritically adopts Meillet's proposal (which later came under heavy criticism): 'Un mot résulte de l'association d'un sens donné à un ensemble de sons donné susceptible d'un emploi grammatical donné' (1943:143). Useful material on word criteria is given in Krámsky (1969), Zirmunskij (1966), Bazell (1957), and, especially for French, in the excellent article by R. Harris (1972). Although somewhat outdated, Juilland and Roceric's analytic bibliography is still a useful introduction to the vast literature on the subject. The following discussion is drawn largely from these sources. It should be noted, however, that the so-called universal criteria are typically stated in universal terms but often tested against a very small number of languages. Moreover, the various theories have not, to my knowledge, been rigorously tested across a broad range of languages. Krámsky (1969:22) also mentions a third criterion, replaceability, which he considers closely related to the two mentioned here. Since what is replaceable is, by definition, also separable or displaceable, I can see no justification for distinguishing between "replaceability" and "separability"/"displaceability" on the other. Interestingly, after citing replaceability in the section heading, Krámsky never returns to this criterion. These criteria are frequently subsumed under three basic levels: the PHONOLOGICAL WORD bounded by pauses, the SYNTACTIC WORD defined in terms of its position in the sentence, and the LEXICAL WORD as a vocabulary item with a specific meaning. "Semantic criteria" are sometimes cited as another determinant for "wordhood" (for a sympathetic account of this position see Ullmann [1959:50-54]). Even though we must agree with Bolinger (1963:113) that 'as a reaction to popular unawareness of certain of the niceties and asymmetries of word meaning, linguists have combatted the idea that words "have meaning'" and that 'it is precisely the relative fixity of word meaning that makes it a viable unit of communication,' it is also true that due to the absence of a working definition of "word-meaning," the various semantically flavored proposals for a universal word definition are part of an entire family of bogus word criteria (as I shall argue below [§2.5], this is not to say that the speaker's "innate" feeling about words is not, to some extent, based on semantics). A survey as well as critical discussion of various semantic criteria can be found in Krámsky (1969:19-22, 72-74), who also concludes that 'meaning cannot be the criterion of the word' (73). The strongest rejection of semantic criteria comes perhaps from R. Harris who argues that:

Chapter two - Word delimitation

211

These [semantic criteria] are usually based on the untenable thesis that features of the word as a segmental unit are merely incidental properties of expression of a semantic unit. Since no semantic theorist has ever given a reasoned account of "word-meaning" which would enable a commutation test on sentences to be used as the basis of a word analysis, the matter does not merit serious discussion. (1972:122) 7.

Among proponents of displaceability Krámsky (1969:23) mentions V. Mathesius (1947) and J. Vachek (1942). I have not been able to verify these sources and therefore rely on Krámsky's account (for particulars see p. 22« 18). Jespersen (1964 [1922]:424) also appears to consider "possibility of transport" as a valid criterion for word separation, citing the example he is ill vs. is he ill?. However, he does not qualify his proposal as universally applicable. 8. It is often claimed that in free word-order languages (such as Latin), syntactic displacement of words has pragmatic motivation. Found in approximately 4% of the world's languages (Mallinson and Blake 1981), free word-order languages figure among the least well represented basic word order types. 9. Supporters of separability include well-known members of the Prague School (Vachek, Jakobson) as well as American structuralists, including Bloomfield who, although best known for his "a word is a minimum free form" (for details see §2.3.2 below) must also be grouped here since he states that 'this latter principle, namely that a word cannot be interrupted by other forms, holds good almost universally' (1963 [1933]: 180). For particulars see Harweg (1964:22-23). 10. I shall limit my discussion to Romance and Germanic. Krámsky (1969:2425), focusing on Slavic, discusses a number of cases in which the criterion of separability is not airtight either. 11. In French the problem with separability is not limited to the "hyperdissection" of elements which should remain whole words. We find that the opposite extreme, "hyper accumulation of 'words,"' occurs, e.g., in a variety of adverbial expressions: à quatre heures = a single, uninterruptable word /akatïcér/. 12. R. Harris attacks Martinet for precisely that point: If it is maintained, as by Martinet, 'que l'on peut, sans dissocier le mot, intercaler, entre deux signes qui le composent, un autre signe, à condition que ce dernier n'existe jamais que dans ce type de combinaison,' what is to decide whether it is legitimate to count e.g. le in je le fais and in faites-le as occurring in the same "type de combinaison"? (1972:124)

212

13.

14.

15.

16.

17.

Notes Criticisms of this type must have had an effect on Martinet's line of thinking, thus his later publications on the subject reflect a markedly more conservative attitude towards the possibility of finding a truly unproblematic word definition (compare for instance his 1965 article "The Word" where we read: 'All efforts to give the term "word" a really scientific status ran up against the fact that aside from clear-cut cases, there are others where none of the usable criteria permit a yes or no reply' [p. 51]). These include Siitterlin (1902), Tesnière (1932), Bally (1965), Kayne (1975), M. B. Harris (1978b), (1980), and Ashby (1977), Bossong (1979), (1980), (1981). For a list of additional, pre-1950 authors favoring Bally's analysis, see Togeby (1949:105). The history of the problem is assessed in Hunnius (1977). Even though the number of possible insertions in French may exceed those of Latin, such interpolations are not all that abundant in French either (only preverbal ne in formal discourse: je N'aime pas Ί do not love'; periphrastic tense-aspect markers: je VAIS aimer Ί am going to love'; and a variety of object clitics: je Τ'aime/je L'aime Ί love you/I love him/her/it')· As in the case of the French examples above, part of the trouble with the criterion of separability is not just that what should obviously be whole, inseparable word units becomes separated, but that in the process of splitting words like farei we are left with unassigned "residue" (-ei) in the final break-up of sentences into words. To "free form" Bloomfield contrasts "bound form," which he defines as one which is 'never spoken alone' (p. 160); all citations from Bloomfield are from the 1963 reprint. Bloomfield had first expounded his ideas on word separation in the 1926 article "A Set of Postulates" (discussed critically in Newmark [1957: esp. pp. 25-26]). Bloomfield's definition of the word has generally been accepted by linguists and some have modeled their own definitions on his to the point of near identity (cf. Benveniste [1964:269]: 'Le mot peut... se définir comme la plus petite unité signifiante libre susceptible d'effectuer une phrase et d'être elle-même effectuée par des phonèmes.' I am not suggesting that a valid criterion for word segmentation must be in total accord with the intuitive feelings of naive speakers. I do believe, however, that an operational test must yield "natural" results in cases where word segmentation is beyond dispute (e.g., in words like four, beautiful, warm, Paris, etc.). If the procedure we choose is invariably capable of cutting phoneme sequences at these "natural" word junctures, we might then be justified in appealing to it in unclear and difficult cases (e.g., door knob·, innkeeper; manhunt, etc.). The view expressed here is basically identical to

Chapter two - Word delimitation

18.

19.

20.

21.

22. 23.

213

that of Chomsky (1965:19) who holds that if a proposed operational test does not meet with the empirical condition of conforming, in a mass of crucial and clear cases, to the intuition of the native speaker concerning such elements, the formula is without value. Despite such reservations, Bloomfield's concept of "minimum free form" has been retained by many; e.g., Elson and Pickett (1965:76): 'This definition [Bloomfield's] is so convenient and so useful that it would be unwise to abandon it, even though there are problems with its application in many languages.' In light of the various complications attaching to the criterion of "minimum free form," one may wonder what led to Elson and Pickett's conclusion that the 'word as defined above is applicable to every language' (p. 76). Note how much the adoption of a particular word criterion affects the degree of syntheticity of a given construction. Contrary to the result obtained using the Bloomñeldian criterion, using separability as a word divider yields a rather analytic French verb phrase. In this context, R. Harris (1972:128) convincingly shows that Bloomfield's argument of parallelism does not suffice for several uses of modem French determiners. There is, Harris argues, a parallel between le and ce (le roi/ce roí); but ce is not a free form. The "nearest" free form is ceci 'THIS (one),' and we cannot have *ceci roi, but only ce roi-ci 'THIS king.' Thus the chain of analogy required to include le as a word in French is a conspicuously long and tenuous one. But at least a chain can be traced, which is more than can be said in the case of à. For neither in Old French nor in Modem French does à ever stand in absolute position; nor does it stand in any suitable relation to a form which does. E.g., in the 12th and 13th centuries, it was still possible to intercalate nominal elements between a subject pronoun and a verb, including full noun objects (e.g., quant je la demoisele voi 'when I the maiden see' [R. Harris 1972:131]). Such a construction is no longer possible in modem French. The development outlined, together with a number of other changes not discussed here, has resulted in constraints on the separability of the verb phrase in modem French, which has, in tum, resulted in tighter cohesion between subject clitic and verb. For a more detailed discussion see R. Harris (1972:130-133). On the distinction between syn- and autosemantics see Givón (1984:48ff), Krámsky (1969:19n7), and Benveniste (1964:266-293). R. Hams, recognizing the earlier drawbacks of the traditional criterion of interruptability, suggests that we treat separability as a gradation or cline, with complete freedom of interruption at the top end of the scale and paradigmatic-

214

Notes

consequential restriction to a single morph at the bottom. Its utility as a criterion in the case of any individual language will depend, says Harris, on whether and where well demarcated "breaks" occur on the scale of types of separability found in a language. In the present context, it will be of particular interest to see whether a clear distinction is apparent between separability which is paradigmatically restricted to a few individual morphs, and separability which admits the members of a large distributional class of morphs, or constructions of a common type. After elaborating further on this fairly complex scheme of gradation, Harris concludes, however, that his technique may not be adequate for a cross linguistic definition of the word; like others before him, he realizes that word structure is determined by language particular factors which do not easily lend themselves to cross-linguistic generalizations: '... French has no unit which corresponds exactly to the "word" of Latin (or of English)' (1972:126). 24. 'They [phonological units] are seldom, if ever, entirely in agreement with grammatical criteria over the whole of the vocabulary of a language, and where there is conflict between grammatical criteria ... and phonological features normally coinciding in delimiting the same stretch of speech, the grammatical criteria must be allowed to carry the day' (Robins 1964:198). From a historical point of view, the result of a variety of apparently independent phonological changes must have caused the widely acknowledged non-correspondence between the "phonological" and "grammatical word." R. Harris remarks that in French this discrepancy 'can be traced back to the thirteenth century ... [and] ... as early as the ninth, there is enough evidence of the prevalence of proclisis and enclisis to cast doubt on the existence of juncture phonemes as word markers' (1972:122). 25. Bazell (1957:27). Another language which marks words phonically is Turkish, where the principle of vowel congruence is regarded as a sufficient (though not necessary) criterion of word unity. It must be noted with respect to phonological demarcation that linguists frequently overestimate the degree of predictability of word boundaries. Pulgram (1970) shows that even though one may affirm that stress, for example, acts as a boundary marker (especially in languages where stress is predictable: Hungarian, Czech [on first syllable]; Polish [on penult]; Latin [according to the quantity of the penult]; French [last syllable, unless it ends in schwa]), it is not quite true that stress always reveals where a word begins or ends. For particulars see Pulgram (1970:31nl3). 26. For details on the discrepancies between contour boundaries and word distribution see Garvin (1954:346). On the notion that phonetic criteria are not di-

Chapter two - Word delimitation

215

rectly tied up to the frame of the word also consult Rosetti (1947:18-19). A recent summary overview of the "phonological word" and "phonological phrase" in the major Romance languages can be obtained from B. Agard (1984: chaps. 20-21). 27. Although it is true that in French autosemantics spoken in isolation carry stress on the last syllable, this stress is nevertheless not a marker of word boundaries, since in discourse only the stress of the last word in the phrase or breath group is perceptible, i.e., word-stress is subordinate to group stress. The so-called accent d'intensité is not a word-stress, and it generally does not coincide with the predictably stressed (last) syllable of a word; rather it is a feature of the intonation contour typically characterized by an extra high pitch. For a summary of stress patterns in other Romance languages see Agard (1984:208). 28.

According to Bloomfield, some French word boundaries are predictable on the basis of vowel length for 'the mere presence of a long vowel in French ... indicates that the next consonant or consonant-group ends a word' (1963 [1933]: 110). R. Harris (1972:117), however, takes issue with Bloomfield's statement, arguing that 'it seems doubtful whether the generalization holds.' 29. For an in-depth discussion on the interrelationship between phonological criteria and word delimitation in French see Pulgram (1970). 30. The measure of distortion depends, of course, on the initial and final phonemes of adjacent words. Agard (1984:183), in his definition of the phonological word, seems to suggest that (in Romance) there exists a one-to-one relationship between the grammatical and phonological components of language: The phonological word: composed of one or more syllables and constituting the smallest link between the phonological and the grammatical component of a language, in the sense that every lexical or grammatical word, regardless of its internal structure, is in one-to-one correlation with a matching phonological word. I believe that the few Spanish and French examples cited here suffice to cast doubt on Agard's statement. 31. For further insights into the difficulties of applying phonic word criteria to French, Czech, Polish and English consult Garvin (1964:32). 32. For a variety of other problems with phonic criteria see Krámsky (1969:2037). 33. The reason for leaving suprasegmental elements out of account is undoubtedly the difficulty in formalizing them. In light of their importance (they, after all, often express the "meanings" also carried by segmental morphemes [e.g.,

21
qu[ej /k/> is followed by the corresponding subject clitic (the paradigm is from Frei [1929:189-190] who qualifies it as "français avancé"; the parentheses in qu[ej are mine, and are meant to indicate the strictly conso-

Chapterfour - Verbal core: Person/number

233

nan tal quality of que): (a) (b) (c) (d) (e) (f)

C'est moi QU(E) JE paie for FSF 'It is I who pay' C'est toi QU(E) TU paies 'It is you who pay' C'est lui QU' IL paie etc. C est nous QU' ON paie C'est vous QU(E) VOUS payez C' est eux QU' ILS paient

C'est moi QUI paie C'est toi QUI paies C'est lui QUI paie C'est nous QUI payons C'est vous QUI payez C'est eux QUI paient

Lambrecht, also a speaker of Swiss French, confirms the use of the relativization pattern illustrated here within French-speaking Switzerland. He rightly observes that example (d) above is not only by far the most frequent construction, but also that the pattern c'est NOUS que NOUS ... (which Frei includes in his paradigm; cf. c'est NOUS QUE NOUS payons) is never used in SwissFrench dialects (the same might be true of other varieties of French as well). Lambrecht interprets this fact as good evidence for the replacement of clitic nous by on, so that at the time nous payons was replaced by on paie 'the traditional pattern in qui was already so rare that it was not combined with the new verb form paie of the 1 pi.' (1981:30«22). Whatever the usage and frequency of the constructions illustrated here may be, the fact that they are attested in some dialects constitutes additional evidence that the subject clitics are increasingly being reinterpreted as grammatical markers and as an ever more obligatory part of a highly synthetic verbal complex. 71. On possible reasons for clitic deletion in these and similar examples see Lambrecht (1981:28). 72. Several scholars (M. B. Harris 1978b, Lambrecht 1981) interpret the replacement of nous + V+ortj by on + V as a development in line with the French move toward préfixai inflection. Of course, ISF on can still be used as a "generic" impersonal pronoun. Thus ISF on y va à pied may mean 'we're walking (as opposed to driving)' or 'you get there by walking.' 73. The final step in regularizing French verbal inflection would be to eliminate the -ez suffix. There is at least one dialect (Old Mines, Missouri) in which such a change is reported to have taken place (cf. vous se porte bien 'you are doing well' [Thogmartin 1970:69, trans, mine]). It should also be kept in mind that the rise of on + VERB for nous + VERB eliminates a certain amount of alinearity in stem alternating verbs (cf.

234

Notes FSF je sais/nous savons Ί know/we know'; ISF je sais [/íse/]/on sait [/öse/] Ί know/we know').

Chapter five - Verbal core: Periphrastic tenses 1.

2.

3.

4.

5.

6. 7. 8. 9.

Examples are from Thielmann (1885a) (translations mine) and Pinkster (1985 and 1987). The history of habeo + PAST PARTICIPLE has most recently been discussed by Ramat (1984:143ff) and Pinkster (1987:196ff), whose contributions have sharpened my thinking about the interrelationship between the early semantic changes and the subsequent morphosyntactic synthesis in periphrastic habeo constructions. Once the association between PPs and habere had been fiimly established, the motivation for grammatical agreement between nominal objects and PPs was lost, and, as a result, -um became the unmarked form of the PP in periphrastic habeo constructions. This development is still reflected in modern Romance, where we find Fr j'ai écrit la carte, Sp he escrito la carta (instead of *j'ai écritE la carte, *he escritA la carta). Note, however, that there is still agreement in constructions of the type j'ai ma lessive faitE à la lingerie Ί have my laundry done at the cleaners,' or—at least in formal French—la lettre que j'ai écritE 'the letter which I wrote/have written'. In Portuguese, the PP could agree with the object as late as the 14th century (cf. todos meus dias tenho perdudos Ί have lost all my days' (Graal, cit. Ramat 1984:148)). In Spanish, the situation was essentially the same, with cases of PP agreement being attested through 15 th century. For recent discussions of the semantic evolution of habeo + PP constructions in Latin and/or the Romance see Fleischman (1983), M. B. Harris (1982b), and Ramat (1985). For definitions and/or discussions of "situation types" see Vendler (1967) and Dowdy (1972). The interaction between aspect and situation types has recently been discussed in Smith (1983). As a result of these aspectual restrictions, a sentence like esta tarde èie tern bebido tres litros de vinho 'this afternoon he has [so far] drunk three liters of wine' is only grammatical if the subject is likely to continue his drinking. On the use of the perfect together with specification of the time of the past situation see Comrie (1976:54) and Fleischman (1983:197). For a summary of the various strategies see Fleischman (1982:330· For further discussion of the rise of the periphrastic future tenses in the various dialects see Lausberg (1966:311-314) and Elcock (1960:106-108). Since there is good reason to believe that the major steps in the synthesis of

Chapter five - Verbal core: Periphrastic tenses

10.

11. 12.

13. 14.

15.

16.

235

cantare habeo were similar, if not identical, to those of cantare habebam, I will limit my discussion to the periphrastic future. By giving preference to habeo cantare I am not, however, implying that the functional evolutions of the future and future-of-the-past periphrases were necessarily parallel, nor that habebam cantare was modeled after cantare habeo. As Fleischman (1982:590 reminds us, there are strong arguments suggesting that the past-based forms developed first, and that cantare habeo gained ascendancy over its model only after the synthetic CL future began to decline. French and Occitan appear to have retained a few isolated forms of the CL future tense (Occ ier < CL ero Ί will be'). It has, however, been suggested (Ashby 1977:36) that these cases are merely later borrowings from CL rather than archaisms. The parallelism in the semantic development of the Romance future and perfect is discussed in Fleischman (1983) and Pinkster (1987). It is possible that greater semantic restrictions on PPs caused habeo + PP to grammaticize after cantare habeo. To the best of my knowledge, the lack of generality in PPs has never been suggested as factor in the relatively late spread of the perfect periphrasis. Unfortunately, the available data can neither support nor refute such a theory. Additional examples from various domains of Romance are given in Rohlfs (1922, passim). That the cantare habeo model must have been much more common in Spanish or French is supported by the observation that today neither language makes use of the alternative, preverbal construction to express strict futurity (Sp haber de 'to have to' [cf. he de cantar Ί must sing')—besides being used only in most formal registers—has a strongly modal, i.e., obligative meaning). The fact that Old French but no other Romance language failed to display disjunctive future forms has often been interpreted as evidence that French was first in synthesizing the analytic cantare habeo. Since the disjunction of habeo in other regions is, as mentioned previously, entirely contingent upon the placement of an infixed pronoun, it may be argued that the absence of disjunctive futures in northern France may simply have resulted from an aversion to pronoun insertion. Until now I have spoken of habeo and its reflexes as a "morpheme of futurity." While this analysis may be correct for the earliest stages of its tensed use, it is certainly incorrect for the Romance period. With fusion of habeo and the adjoining verb a reanalysis of the morphological components took place. The erstwhile infinitive marker -r- came to be regarded as part of the ending (i.e., the tense-mood inflection), and the reflexes of habeo as the P/N marker.

236

Notes

17. For an assessment of Valesio's work see Butler (1969) whose criticisms are, in turn, rebutted in Valesio (1969). 18. For a similar argument see C. G. Lyons (1978). 19. Maintaining the information contained in the final syllable of habeo was crucial insofar as Latin or early Romance did not normally signal P/N with subject clitics. As we saw in chapter 4, grammaticization of these clitics occurred only in the MidFr period. 20. Based on Valesio, p. 159. 21. If one presumes, as does Bossong in his review of Fleischman (1982)—that the genesis of the Romance future took place essentially after Latin had switched from OV to VO, then the pre-verbal "future" formations in Old Lombard, Sicilian, Old Sardinian, or Rumanian would—at least from the perspective of word order typology—present the "normal" (i.e., expected result), French, Spanish, Portuguese, etc. the "abnormal" (i.e., unexpected) result On this topic see also Ineichen (1980). 22. Stengard (1985) argues unconvincingly against the view that the switch from a VO to an OV syntax lies at the center of the divergent morphosyntactic developments of the future and perfect in Romance. 23. If one keeps in mind the numerous changes which have driven the habeo cantatum paradigm towards greater syntheticity, it becomes apparent that the passé composé is less analytic than one tends to believe. Spelling conventions, i.e., the fact that this paradigm continues to be written as two words, no doubt contribute significantly to the widespread but impressionistic analysis of j'ai chanté as an analytic complex. 24. Many normative grammars of French, e.g., Grevisse (1964 [1950]:60), hold that the periphrasis aller + I N F I N I T I V E expresses proximate futurity, whence the name "futur proche." As Fleischman (1982:95f) shows, this is a misleading term because the primary value of aller + I N F I N I T I V E is not to indicate that the action will occur in the near future, but to indicate that a connection has been made in the speaker's mind between the present situation and the future action. 25. Fot the diachrony and examples of aller + INFINITIVE constructions see Flydal (1943). 26. On the semantic diachrony of go-futures see Fleischman (1982:84-85). 27. Fleischman (1982:82) notes that tentative beginnings of a temporal meaning for the go-construction can be traced to the 13th and 14th centuries. Already by the 15th century, the construction had become generalized in colloquial Spanish, French, and Portuguese. 28. Judging from similar developments in other Romance areas it would not sur-

Chapterfive- Verbal core: Periphrastic tenses

237

prise if French were eventually to distinguish phonologically between lexical and auxiliary aller. Witness, for instance, the following Catalan paradigms where the lpl. and 2pl. of anar 'to go* are no longer homophonous (in Catalan, the 'go' + INFINITIVE construction developed into a past rather than future tense. A future periphrasis, vaig a cantar Ί am going to sing,' also exists but is, apparently, a caique on Castilian voy a cantar. (For further discussion of these and other Romance prefixed inflection markers see Champion [1980]): vaig vas va anem aneu van

'Igo' 'you go' 's/he goes' 'we go' 'you go' 'they go'

vaig vas va vam vau van

cantar cantar cantar cantar cantar cantar

Ί sang' 'you sang' 's/he sang' 'we sang' 'you sang' 'they sang'

29. In std. Spanish, the go-future may still be interrupted: ¿va Vd. a comer? 'are you going to eat?'. In colloquial usage this is no longer possible. The pronoun must either precede or follow the entire construction: ¿Vd. va a comer?! ¿va a comer Vd.? Similarly, in non-colloquial registers highly emphatic negative particles may still be interposed between voy a + INFINITIVE: No voy nunca a volver Ί am never going to return.' 30. If an infinitive begins in /a/ (e.g., amar), 3s. va a—and, in areas where final -s has been reduced to zero, also 2s. vas—is, however, frequently reduced to ν-, thereby forming a phonological whole with the stem (va a amar /bà:màr/ esta casa 's/he is going to love this house'). 31. In Panamanian Spanish, vas has been reduced to να. Chapter six - Synthesis: Predicate negation 1. 2. 3.

4.

The primary data presented in this chapter are drawn largely from Schwegler (1983,1986, forthcoming a). Fot a more detailed treatment of Latin negation see Szantyr (1965:447-455). The Latin practice of reiterating the verb in answers to yes-no questions is still reflected in modern Portuguese where näo rarely stands by itself: fala portugués? — nâo falo/falo näo 'do you speak Portuguese? — No (I don't).' In CL, non was originally not the universal predicate negator, particularly in concessive or prohibitive clauses, ne was used instead of non (ne sit summum malum dolor ... 'granted that pain is not the greatest devil...'; however, according to Väänänen [p. 161], in VL non began to rival ne at an early point, ousting it altogether by the time of PR).

238 5. 6.

7.

8.

9. 10.

11.

12. 13.

14.

Notes The phonotactic compatability of the morphemes involved must also have played an important role in the synthesis of ne oenum. For statistics and the different uses of such emphasizers see Yvon (1948:22). A more recent investigation of the affective reinforcement of the non in OFr is Möhren's exhaustive study (1980:252-256) which lists over 250 nominal emphasizers in order of frequency. The earliest example of negative substantival pas cited by Yvon (1948) is found in Corneille's Horace (1639): Non, non, n'embrassez PAS de VERTU par contrainte 'No, no, do not embrace virtue against your will.' Price's calculations are somewhat higher. According to him, 'by the end of the OFr period, negation is expressed by ne + PARTICLE rather than ne alone in about 25% or 30% of the cases where the particle must be used today* (1971:254). Cf. Gardner and Greene: 'In our period [i.e., MidFr], ne... pas and ne ... were identical in meaning' (1958:124). In negative-interrogative sentences, pas was regularly separated from the verb: et ne le VERRONS nous PAS? 'and will we not see him?' (Gardner and Greene 1958:127). In ISF, pas can still be detached from the verb through the insertion of a lexeme: il l'a PRESQUE pas vu 'he almost didn't see it.' Some scholars, notably Vennemann (1974), treat pas as an adveib rather than sentence qualifier. For an argument against this analysis see Schwegler (1983:297). For a detailed analysis of the cursus in French see Pulgram (1970:85ff). The frequency of ne deletion varies considerably depending on register. M. B. Harris (1978b:26) claims that today ne is virtually ousted from ISF, while Ashby (1976:120) reports that in Parisian upper middle-class speech the particle is dropped at an overall rate of 44%. For an extensive treatment of the subject, see also Pohl (1968:1446-1453) and especially Sturm (1981, passim) who provides an extensive corpus of spoken data containing negative constructions. It must be mentioned in this context that ne and pas are not always separate. In constructions such as Ne pas y aller... 'not to go (there)...' the two particles are juxtaposed. But the frequency of this structure is far too low to trigger analogical formations of the type *je NE PAS chante. Also, in MidFr, the combination was sometimes written pas ne before a verb: pas ne suis des plus jones Ί am not one of the youngest' (Gardner and Greene 1958:124). This word order, seemingly unattested in OFr might possibly be seen as a holdover from the earlier Latin V X syntax, though it is more likely an example of MidFr poetic license which failed to strike root in the spoken language.

Chapter six - Synthesis: Predicate negation

239

15. Posner never makes clear just exactly why changes in sociolinguistic conditions (e.g., the split of Catholic Sursilvan speakers from their Romansh neighbors who became Calvinists in the first half of the 16th century), should have prompted the adoption of new negation strategies. See also Posner (1985b). 16. My 1983 article was already in press when I saw Posner (1985a/b), virtually on the same topic. Posner's contribution has not changed my views, and the rebuttal of previous themes presented in Schwegler (1983) applies to Posner's theses as well. 17. "Psycholinguistic proclivity" is not an explanation. It is really an unknown for a causal factor whose existence must, however, be acknowledged for an understanding of the process involved. 18. It is still unclear to me why (Romance) languages, if confronted with two negative exponents, predictably eliminate the old one. Jespersen observed that the old negative particles are always of small phonetic bulk; given the choice between two particles in an embracing construction, it is perhaps the speakers' need for emphatic devices that unconsciously motivates them to favor the particle with greater phonic substance. Particularly interesting in this respect is the Vermes dialect (W. Switzerland) where the preverbal negator η (< ne) is maintained consistently while pas is reduced to [p] after vowel, and to [pcfe] after consonant: ta η è ρ fé kom mwá = tu n'as pas fait comme moi 'You didn't do it like me' (for details see Butz 1981:181-185,244). If phonetic attrition proceeds along its trajectory and reduces [p&] to [p] in all environments, it will be highly interesting to see which of the two particles of equal phonic substance will ultimately be eliminated. Unfortunately, the Vermes dialect is threatened by extinction, and will most probably die out before such observation will be possible. 19. For further examples see Camproux (1959:477020. An analytic/synthetic cycle may be considered to near completion when (1) the relevant items must cooccur, and (2) their combination produces an unanalyzable semantic whole. In this sense, j'ai chanté, je chante, and FSF ne ... pas are now all approaching the end of the cycle. 21. The etymology of ges has long perplexed scholars. Walsh (1981:89-99) argues convincingly for gentium. 2 2 . FOT a discussion and examples ofreinforcingdorn see Schwegler (1986:294). 23. An interpretive map of cap is given in Schwegler (1986:295). 24. It could, of course, be argued that French influence and not a new negation cycle led to the formation of pas cap or pas ges. In other words, pas could have been inserted between southern VERB + caplges after these particles had

240

Notes evolved into mere predicate negators. In addition to the arguments already cited against French ad- or super-stratum influence, there is also concrete evidence that southern French vernaculars continued to undergo new negation cycles CHI their own after cap, pas or ges had become negators proper. Thus in the Gévaudan dialect a reflex of the CL negator non has come to be placed as emphasize in conjunction with pas, as in the following example (for further discussion of this negation pattern see §6.4.3. below): La

Marjano manquet PAS NOUN RES Marie-Jeanne forgot NEG EMPH NOTHING

de of

ço that

que which

l' abiô dich to him had said 'Marie-Jeanne did not forget anything she had told him' (Camproux 1959:484, trans, mine) 25. Since pas ges or pas cap never appears on map 899 of the ALF (nest pas encore), it is probably safe to conclude that—much like OFr and MidFr pas— these "new" double negatives are restricted to substantival use. 26. For the numerous offshoots of *capu see FEW, 334ff. 2 7 . FOT details see Balarf Jovany (1929:2680; Alcover (1935:856); Badia Margarit (1981:327); Corominas (1981:524); and Griera (1926), vol. 3, map 4 0 2 . Unfortunately, except for map 402 (No s' hi cap pas; No si cabrem) Griera's atlas (ALQ contains no instances of cap = adj., pron. 28. In light of the few linguistic maps available for the Catalan area (Griera's atlas and the ALPI are notorious for their lack of syntactic data), it is difficult to obtain an accurate picture of the frequency, distribution and/or exact usage of cap. 29. For examples of rare non-emphatic cap see ALF vol. 4, map 899 point 790; Seguy vol. 1, maps 276-277 points 790, 790S, 790SE, 791N; and Griera vol. 3, map 402 point 6. For the geographic distribution of cap and other negating particles see ALF vol. 4, maps 896-898, 900. 30. *Capu's development to 'head of cattle' is attested in a variety of GalloRomance vernaculars (cf. e.g., Charente-Inférieure ché 'pièce de bétail' (FEW 2,1:336). 31. I am here limiting myself to Swiss varieties of Romansh. There are other archaic Alpine vernaculars which are presently undergoing a negation cycle. For a general overview of new postverbal negation patterns in Dolomitic Ladin see Plangg (1980). 32. According to the Dicziunari Rumänisch Grischun (1946: vol. 2, 488-507), Eastern Romansh at one point appears to have made use of at least some of

Chapter six - Synthesis: Predicate negation

33.

34. 35.

36.

37.

38.

39.

40.

41.

241

the nominal emphasizers which later evolved into negative particles in Central and Western Romansh. For reasons not known, this practice was later abandoned almost entirely. The etymologies of these particles are disputed, though there is general agreement that they are derived from nominal elements. Surs. buc(c)a is probably from bucca 'mouth, mouthful,' and may first have been used in constructions with the verb 'to eat.' For a summary see Dicziunari Rumänisch Grischun (1946:506-507) and Avery (1978:35-46). For these and other examples see Jaberg and Jud (1940): vol. 8, maps 1647, 1665,1679; the bracketed numbers refer to points on these maps. This is not to say that the negation strategies of some of these vernaculars, particularly Palenquero, Brazilian Portuguese, and Chocó Spanish, are unrelated. I am merely suggesting here that some of these developments, e.g. those of Central Ladin and Palenquero, cannot possibly have arisen from a common source, and that the various negation cycles must, therefore, have been set in motion by the psycholinguistic factors mentioned in §6.3.5 above. Arguments for the genetic relationship between Palenquero, Brazilian Portuguese, and Chocó Spanish negation patterns are given in Schwegler MS. In Schwegler (1983) and (1988) I reject the view that word order changes in negation are triggered by typological pressures, as envisioned by Vennemann (1974:366-370) or Posner (1985b). I do not, therefore, believe that the placement of the "new" negator is a response to a change from OV to VO. According to Siller-Runggaldier the second no is always sentence-final. Its usage is, therefore, similar to that of the negators in (Brazilian) Portuguese, Palenquero, and Chocó Spanish given below. The Pen. Portuguese double negative appears to be largely restricted to answers to yes/no questions: voce vem? — venho näo 'Are you coming? — No I am not (coming).' For details see Schwegler forthcoming a. Both NEG2 and NEG3 are extremely frequent The use of all three negation patterns depends on pragmatic, sociolinguistic, grammatical, as well as other factors (see Schwegler forthcoming a). In Schwegler (MS) I suggest a genetic relationship between Brazilian Portuguese and Palenquero negation. The most striking common features to both languages are: (1) concurrent use of all three negation patterns; (2) restrictions placed on NEG1 under virtually identical pragmatic conditions (NEG1 cannot be used to negate a previous positive assertion); (3) the more conservative behavior of negation in subordinate clauses; and (4) the higher frequency of NEG23 vis-à-vis NEG1. Recall that an important hypothesis which I have adopted from Bybee (1985)

242

42.

43.

44.

45.

46.

Notes is that semantically highly relevant items tend to occur in close proximity (see chapter 3). In certain languages a negative tag is frequently used to transform a declarative sentence into a question: Tas faim, non? 'you're hungry, aren't you?'. This additional negator in sentence-final position (e.g., voce está com fome, näo?) probably furthered the spread of the postverbal negation under discussion. As Camproux cites no further examples of pas non, nor does he discuss its syntax, it is impossible to determine whether the construction has already fused into a single, inseparable complex. Milanese also has preverbal negation: a scòla NO ghe voo (archaic). Nicoli's brief account of Milanese predicate negation patterns (pp. 238f, 282f, 332) does not make clear whether remnants of no ...no survive to this day. There is synchronic evidence from at least two other vernaculars presently undergoing a switch from NEG1 to NEG3 suggesting that the frequent and, therefore, largely formulaic placement of certain elements alongside a clause- or sentence-final negator can lead to the kind of syntactic reanalysis found in Gév non pas. In Brazilian Portuguese and Palenquero, the final negator is often juxtaposed to the verb: BP êtes falam espanhol? — falam näo 'do they speak Spanish? — (no), they don't'; Pal i nu quelé nu Ί don't want [it].' By analogy to this "VERB + NEG" pattern, some speakers now consider the juxtaposition of "VERB + NEG" acceptable even though additional information may be given after the postverbal negator, as in BP näo GOSTO NAO deste disco näo Ί really don't like this record at all' or Pal. i nu KELÉNU ablá español nu Ί really don't want to speak Spanish (at all)' (for further examples and discussion of such structures see Schwegler forthcoming a and Schwegler MS). Although both Palenquero and Brazilian Portuguese presently give no indication of abandoning the sentence-final syntax of the postverbal negator, the frequent collocation of "VERB + NEG" may some day trigger a change in word-order similar to that postulated for Gév pas non. The substitution of the synthetic unit is, of course, also a process in the sense that the old and new constructions cooccur for quite some time before the synthetic one dies out, leaving its analytic "replacement."

Chapter seven - On drift, cyclicity, and the analytic!synthetic spectrum 1. 2.

Sapir's famous chapter 7 and its reverberations in modem linguistics are discussed in extenso in Malkiel (1981). Since some, but not all, languages undergo changes in the analytic/synthetic spectrum, Lakoff (p. 192) rightly rejects the notion that greater intelligibility

Chapter seven - The analytic!synthetic spectrum

243

or other psychologically relevant notions can be advanced as "explanations" for the drift As Malkiel (1981:560) points out, Lakoff despairs of grasping why drift exists and thus admits defeat. By admitting so frankly the existence of so many unknowns, she thus psychologically approaches Sapir's position in 1921. In 1933, Sapir removed some of the ideas on drift contained at the beginning (notably p. 172) of chapter 8. Hodge's evidence is taken exclusively from the verbal system. In §5.5.2 we also saw that Anderson (1979) posited phonological reduction and concomitant loss of autonomy as the cause for the synthesis of Panamanian Spanish tu vAdormir. Similarly Sapir (1911). The rise of prepositions in Romance is, for instance, often viewed as a therapeutic response to the phonological weakening of the word-final suffixal case system. Sapir is among the most celebrated scholars to propose phonological reduction as the main factor in the formation of new morphological types: Strange as it may seem at first blush, it can be demonstrated that many, perhaps most, changes in grammatical form are at last analysis due to the operation of phonetic laws. Inasmuch as these phonetic laws affect the phonetic form of grammatical elements as well as of other linguistic material, it follows that such elements may get to have a new bearing, as it were, brought about by their change in actual phonetic content; in certain cases, what was originally a single grammatical element may in this way come to have two distinct forms, in other cases two originally distinct grammatical elements may come to have the same phonetic appearance, so that if circumstances are favorable, the way is paved for confusion and readjustment ... In the long run, not only may in this way old grammatical features be lost and new ones evolved, but the entire morphologic type of the language may undergo profound modification. (1911:213-214) Geisler is not the first to propose such a model. The best-known formulation of the principle of economy is found in Martinet: "L'évolution linguistique en général peut-être conçue comme régie par l'antinomie permanente des besoins communicatifs et expressifs de l'homme et de sa tendance à réduire au minimum son activité mentale et physique" (1955:94). Similar views are also expressed by Lyons (1968:90). What is new in Geisler's contribution is the notion that these principles are the actual cause of the analytic/synthetic tendencies.

244 8.

9.

10.

11.

12.

13. 14.

15.

16.

17.

Notes According to Geisler (p. 22), assimilatory sound reduction leads to ease of articulation, which improves the combinability of segments while at the same time reducing the number of segments, thus leading to more compact language. The direct effect of such long-term optimization is the formation of synthetic languages. Geisler (p. 23) recognizes that reversals may take place even though the dominant coding strategy has not yet pushed the language to either extreme of the analytic/synthetic scale. This is not to say that synthetic structures are as easy to learn as analytic ones. It remains to be seen, however, whether a synthetic system, once fully mastered, is in fact more difficult to decode (or easier to encode). Similarly, though somewhat more cautiously, Tauli expresses that 'the different types of WO [= word order] might be explained by the different mentality of the peoples who speak or have spoken the languages' (1958:75). According to Tauli (1966:283) the latter schema appears to be discernible to a certain extent in the history of Chinese; however, this development seems to have been proved only with regard to some phenomena, and to some extent. For examples of concurrent analytic and synthetic tendencies see Bally (1965 [1932]) and Tauli (1966) for French and the Uralic languages, respectively. An example of such a "abortion" was seen in the case of Romance negative emphasizers (e.g., Fr ges. East. Swiss Romansh bitch) which were dropped before they had a chance to synthesize on all levels. There are many other internal factors which undoubtedly play a decisive role in the life cycle of synthetic and/or analytic constructs. Among the most important is the often-noted tendency of speakers to avoid oversynthesis. Comrie (1981:210) discusses one such case. Final a in Estonian jalga 'foot' is a partitive singular suffix derived not from an independent word, but from part of its original stem (jalka-a). See also Green (1976), Comrie (1980), and Bossong (1982). Napas is, of course, relatively synthetic only from a diachronic perspective. Synchronically, it is entirely analytic, since there exists a one-to-one correspondence between units of meaning and of form.

Bibliography Agard, Frederick B. 1958 Structural sketch of Rumanian. Language Monograph. 26. Baltimore: Linguistic Society of America. 1984 A course in Romance linguistics, 1. A synchronic view. Washington, D.C.: Georgetown University Press. Alcover, Antoni M. 1935 Diccionari català-valencià-balear. 1. Palma de Mallorca: Alcover. Altmann, Gabriel, and Werner Lehfeldt 1973 Allgemeine Sprachtypologie. Prinzipien und Messverfahren. Munich: Fink. Anderson, Eric W. 1979

Anglade, Joseph 1969 (1921) Arens, Hans 1969 (1955) Ashby, William J. 1974 1976 1977 1980

1982 1988

The development of the Romance future tense. Morphologization II and a tendency toward analyticity. Papers in Romance 1:21-35. Grammaire de l'ancien provençal. Paris: Klincksieck. Sprachwissenschaft. Der Gang ihrer Entwicklung von der Antike bis zur Gegenwart. Freiburg: Alber. "II parle" or "iparl"? Prefixed inflection in French. Semasia 1:83-93. The loss of the negative morpheme "ne" in French. A syntactic change in progress. Language 57:674-687. Clitic inflection in French. An historical perspective. Amsterdam: Rodopi. Prefixed conjugation in Parisian French. Italic and Romance linguistic studies in honor of Ernst Pulgram, ed. Herbert J. Izzo, 195-207. Amsterdam Studies in the Theory and History of Linguistic Science. Series 4: Current Issues in Linguistic Theory. Amsterdam: Benjamins. The drift in French syntax. Lingua 57:29-46. The syntax, pragmatics, and sociolinguistics of left- and rightdislocations in French. Lingua 75:203-229.

246

Bibliography

Aubanel, Théodore 1963 Le Pain du péché, le rapt, le pâtre. Ed. E. Aubanel, 1-2. Avignon: Aubanel. Avery, J. B. 1978

A problem of etymology in Swiss Romantsch: "buca" 'not*. Semasia 5:35-46.

Badia i Margarit, Antoni M. 1981 Grammàtica histórica catalana. Bibl. d'Estudis i Investigations, 4. Valencia: Soler. Bailard, Joëlle 1981

A functional approach to subject inversion. Studies in Language 5:1-22.

Balari y Jovany, José 1929 Diccionari balari. Barcelona: Elzeviriana. Baldinger, Kurt 1968

Post- und Prädeterminierung im Französischen. Festschrift Walther von Wartburg zum 80. Geburtstag, ed. K. B„ 1:87106. Tübingen: Niemeyer.

Bally, Charles 1913 Le Langage et la vie. Geneva: Atar. 1965 (1932) Linguistique générale et linguistique française. Francke. Barnes, Betsy 1983

1985 Barri, Nimrod 1977

Berne:

Left-dislocation in French: A multi-level model of topic. Paper presented at the Minnesota Regional Conference on Languages and Linguistics. The pragmatics of left detachment in standard spoken French. Pragmatics and Beyond, 6:3. Amsterdam: Benjamins. Giving up word formation in structural linguistics. Folia Linguistica 11:13-37.

Bartsch, Renate, and Theo Vennemann 1972. Semantic structures. A study in the relation between semantics and syntax. Athenäum-Skripten Linguistik, 9. Frankfurt: Athenäum.

Bibliography Bauche, H. 1928 (1920) Bauer, Laurie 1978

Bazell, C. E. 1957

Benveniste, Emile 1946 1964

1966 1974

247

Le Langage populaire. Paris: Payot. The grammar of nominal compounding (with special reference to Danish, English and French). Odense: Odense University Press. On the historical sources of some structural units. Miscelánea homenaje a André Martinet: Estructuralismo e historia, Diego Catalán, ed. Tenerife: Universidad de la Laguna. 1:1929. Structure des relations de personne dans le verbe. Bulletin de la Société de Linguistique de Paris 53:1-12. Les Niveaux de l'analyse linguistique. Proceedings of the Ninth International Congress of Linguists (Cambridge, MA: 1962), ed. Horace G. Lunt, 268-275. Janua Linguarum. Series Maior, 12. The Hague: Mouton. Problèmes de linguistique générale. Paris: Gallimard. Mutations of linguistic categories. Directions for historical linguistics, ed. W. P. Lehmann and Yakov Malkiel, 126-136. Austin: University of Texas Press.

Bloch, Bernard, and George L. Trager 1942 Outline of linguistic analysis. Baltimore: Waverly Press. Bloomfield, Leonard 1963 (1933) Language. New York: Holt, Rinehart and Winston. Bloomfield, Maurice. On adaptation of suffixes into congeneric classes of substan1891 tives. American Journal of Philology 12:1-29. On the so-called root-determinatives in the Indo-European 1894 languages. Indogermanische Forschungen 4:66-78. Bolinger, Dwight L. The uniqueness of the word. Lingua 12:113-136. 1963

248

Bibliography

Bopp, Franz 1833

Bossong, Georg 1979

1980 1981 1982

Vergleichende Grammatik des Sanskrit, Ζend. Griechischen, Lateinischen, Litthauischen, Gothischen und Deutschen. Berlin: Dümmler. Prolegomena zu einer syntaktischen Typologie der romanischen Sprachen. Festschrift Kurt Baldinger zum 60. Geburtstag, ed. Manfred Höfler et al., 1:54-68. Tübingen: Niemeyer. Aktantenfunktionen im romanischen Verbalsystem. Zeitschrift für romanische Philologie 96:1-22. Séquence et visée. L'expression positionnelle du thème et du rhème en français parlé. Folia Linguistica 15:237-252. Historische Sprachwissenschaft und empirische Universalienforschung. Romanistisches Jahrbuch 33:17-51.

Bourciez, Edouard 1967 (1910) Eléments de linguistique romane. Bowen, J. Donald 1976

ed. Paris: Klincksieck.

Structural analysis of the verb system in New Mexican Spanish. Studies in Southwest Spanish, ed. J. D. B. and Jacob Ornstein, 93-124. Rowley, MA: Newbury.

Brunot, Ferdinand, and Charles Bruneau 1969 (1887) Précis de grammaire historique de la langue française. Paris: Masson. Butler, Jonathan L. 1969 Remarks on the Romance synthetic future. Lingua 24:163180.

Butz, Beat 1981 Bybee, Joan L. 1985

Morphosyntax der Mundart von Vennes (Tal Terbi). Romanica Helvetica, 95. Berne: Francke. Morphology. A study of the relation between meaning and form. Typological Studies in Language, 9. Amsterdam-Philadelphia* Benjamins.

Bibliography Bynon, Theodora 1977 Calonge, Julio 1981

249

Historical linguistics. Cambridge: M.I.T. Press. Implicaciones del género en otras categorías gramaticales. Logos semánticos (studia linguistica in honorem Eugenio Coseriu), ed. Christian Rohrer, 4:19-28. New York: De Gruyter.

Camproux, Charles 1959 Etude syntaxique des parlers gévaudanais. Paris: Presses Universitaires de France. Cantera, Jesús 1966 Cassirer, Emst 1945 1964 (1953)

Características y tendencias del francés actual. Filologia Moderna 23-24:207-235. Structuralism in modern linguistics. Word 1:99-120. The philosophy of symbolic forms, 1. Trans. Ralph Manheim. New Haven: Yale University Press.

Chamard, Henri, ed. 1948 Du Bellay, Joachim. La Deffence et illustration de la langue francoyse. Paris: Didier. Champion, James J. 1980 A note on prefixed inflection in Romance. Italie and Romance linguistic studies in honor of Ernst Pulgram, ed. Herbert J. Izzo, 97-102. Current Issues in Linguistic Theory, 18. Amsterdam: John Benjamins. Chang, Kun 1967 Chaurand, Jacques 1969

Descriptive linguistics. Current Trends in Linguistics 2:59-90. Histoire de la langue française. "Que sais-je?", 167. Paris: Presses Universitaires de France.

Chevalier, Jean Claude 1966 Eléments pour une description du groupe nominal. Les prédéterminants du substantif. Le Français Moderne 34:241253.

250

Bibliography

Chomsky, Noam 1965 1966 (1957)

Cogwill, Warren 1966

Coleman, R. G. G. 1971 Comrie, Bernard 1976 1980

1981 Connors, Kathleen 1981 Contreras, Heles 1963 Coraminas, Joan 1981 Coseriu, Eugenio 1968

1972

Aspects of the theory of syntax. Cambridge, MA: M.I.T. Press. Syntactic structures. Studia Memoriae Nicolai Van Wijk Dedicata. Janua Linguarum. Series Minor, 4. The Hague: Mouton. A search for universale in Indo-European diachronic morphology. Universals of Language, ed. J. H. Greenberg, 114141. Cambridge, MA: M.I.T. Press. The origin and development of Latin "habeo" + infinitive. Classical Quarterly 21:215-232. Aspect. Cambridge: Cambridge University Press. Morphology and word order reconstruction. Problems and prospects. Historical morphology, ed. J. Fisiak, 83-%. Trends in Linguistics, Studies and Monographs, 17. The Hague: Mouton. Language universals and linguistic typology. Oxford: Blackwell. Rev. of Ashby 1977. Romance Philology 35:434-436. Una clasificación morfo-sintáctica de las lenguas románicas. Romance Philology 16:261-268. Diccionari etimologie i complementari de la llengua catalana. Barcelona: Curial. Adam Smith und die Anfänge der Sprachtypologie. Wortbildung, Syntax und Morphologie. Festschrift Marchand, 46-54. Janua Linguarum. Series Maior, 36. The Hague: Mouton. Ueber die Sprachtypologie Wilhelm von Humboldts. Beiträge zur vergleichenden Literaturgeschichte. Festschrift für Kurt Wais zum 65. Geburtstag, ed. Johannes Hösle, 107-135. Tübingen: Niemeyer.

Bibliography

251

Crews, Cynthia M. 1935 Recherches sur le judéo-espagnol dans les pays balkaniques. Société de Publications Romanes et Françaises, 16. Paris: Droz. Crystal, David 1980 DaneS, Frantisele 1966 Dauzat, Albert 1910

A first dictionary of linguistics and phonetics. Cambridge: Cambridge University Press. The relation of center and periphery as a language universal. Travaux de Linguistique et de Litérature, 2:10-12. La Vie du langage. Paris: Colin.

Dessaintes, Maurice 1960 Eléments de linguistique descriptive. Namur: La Procure. Dicziunari Rumänisch Grischun 1946 Chur: Società Retorumantschà. Dowdy, David 1972.

Studies in the logic of verb aspect and time reference in English. Dissertation: University of Texas, Austin.

Du Ponceau, M. P. 1838 Mémoire sur le système grammatical des langues de quelques nations indiennes de l'Amérique du Nord. Paris: Phian de la Forest. Dubois, Jean 1966

Essai d'analyse distributionelle du verbe (les paradigmes de conjugaison). Le Français Moderne 34:185-209.

Dubois, Jean, and Françoise Dubois-Charlier 1970 Eléments de linguistique française: Syntaxe. Paris: Larousse. Dubois, Jean, et al. 1973 Dictionnaire de linguistique. Paris: Larousse. Ducrot, Oswald, and Tzvetan Todorov 1979 Encyclopedic dictionary of the sciences of language, trans. Catherine Porter. Baltimore: Johns Hopkins University Press.

252

Bibliography

Ebeling, C. L. 1960 Ebneter, Theodor 1973

Eckert, Gabriele 1986.

Elcock, W. D. 1960

Linguistic units. The Hague: Mouton. Das bündnerromanische Futur. Syntax der mit 'vegnir' und 'habere' gebildeten Futurtypen in Gegenwart und Vergangenheit. Romanica Helvetica, 84. Berne: Francke. Sprachtypus und Geschichte. Untersuchungen zum typologischen Wandel des Französischen. Tübinger Beiträge zur Linguistik, 265. Tübingen: Gunter Nair. The Romance languages. 2nc* ed. London: Faber and Faber.

Elson, Benjamin and Velma Pickett 1965 An introduction to morphology and syntax. Santa Ana, CA: Summer Institute of Linguistics. Entwistle, William J. 1953 Aspects of language. London: Faber and Faber. Ewert, Alfred 1966 (1933)

The French language. London: Faber and Faber.

Fabra, Pompeu 1956

Gramática catalana. Barcelona: Teide.

Faine, Jules 1937

Philologie créole. Etudes historiques et étymologiques sur la langue créole d'Haïti. 2n(* ed. Port-au-Prince: Imprimerie de l'Etat.

Feldman, David M. 1964 Analytic vs. synthetic. A problem in the Portuguese verbal system. Linguistics 10:16-21. Fiesel, Eva 1929 Fink, F.N. 1901 1910

Die Sprachphilosophie der deutschen Romantik. Tübingen: Mohr. Die Klassifikation der Sprachen. Marburg: Elwert. Die Haupttypen des Sprachbaues. Leipzig: Teubner.

Bibliography

253

Fleischman, Suzanne 1982. The future in thought and language. Diachronic evidence from Romance. Cambridge: Cambridge University Press. From pragmatics to grammar. Diachronic reflections on com1983. plex pasts and futures in Romance. Lingua 60:183-214. Flydal, Liev 1943.

Foulet, Lucien 1968 (1919) François, Denise 1974

Franzén, Torsten 1939 Frei, Henri 1928 1954

"Aller" et "venir de" suivis de l'infinitif comme expressions de rapport temporels. Avhandlinger det Norske Videnskaps (Akademi i Oslo), 1-120. Petite syntaxe de l'ancien français. Paris: Champion. Français parlé. Analyse des unités phoniques et significatives d'un corpus recueilli dans la région parisienne, 1-2. Paris: S.E.L.A.F. Etude sur la syntaxe des pronoms personels sujets en ancien français. Uppsala: Almqvist and Wiksells. La Grammaire des fautes. Paris: Geuthner. Critères de délimitation. Word 10:136-145.

Friedemann, Nina S. de, and Carlos Patiño Rosselli 1983 Lengua y sociedad en el Palenque de San Basilio. Publicaciones del Instituto Caro y Cuervo, 66. Bogotá: Imprenta Patriótica. Gabelentz, Georg von der 1875 Weiteres zur vergleichenden Syntax. Zeitschrift für Völkerpsychologie und Sprachwissenschaft 8:129-165, 300-338. Beitrag zur Geschichte der chinesischen Grammatiken und 1878 zur Lehre von der grammatischen Behandlung der chinesischen Sprache. Zeitschrift der Deutschen Morgenländischen Gesellschaft 32:601-664. 1881 Chinesische Grammatik. Mit Ausschluss des niederen Stiles und der heutigen Umgangssprache. Leipzig: Weigel. 1891 Die Sprachwissenschaft. Ihre Aufgaben, Methoden und bisherigen Ergebnisse. Leipzig: Weigel Nachfolger.

254

Bibliography

1901

Die Sprachwissenschaft, ihre Aufgaben, Methoden und bisherigen Ergebnisse. 2n(* ed. Leipzig: Tauchnitz. Repr., with an introduction by Eugenio Coseriu, ed. Gunter Narr and Uwe Petersen, in Tübinger Beiträge zur Linguistik, 1. Tübingen, 1972.

Gamillscheg, Ernst 1957-1958 Historische französische Syntax. Tübingen: Niemeyer. Gardner, Rosalyn, and Marion A. Greene 1958 A brief description of middle French syntax. Chapel Hill: University of North Carolina, Department of Romance Languages. Garvin, Paul L. 1954 1964

Delimitation of syntactic units. Language 30:345-348. On linguistic method. Selected papers. Janua Linguarum. Series Minor, 30. The Hague: Mouton.

Gauger, Hans-Martin 1971 Durchsichtige Wörter. Zur Theorie der Wortbildung. Heidelberg: Winter. Geisler, Hans 1982

Studien zur typologischen Entwicklung. Lateinisch - Airfranzösisch - Neufranzösisch. Romanica Monacensia, 17. Munich: Fink.

Gilliéron, Jules, and Edmond Edmont 1902 Atlas linguistique de la France. 8 vols. Paris: Champion. Givón, Talmy 1979 1984

On understanding grammar. New York: Academic Press. Syntax. A functional-typological introduction, 1. Amsterdam: Benjamins.

Glazov, Leonid Borsovich 1965 Κ probleme tipologiCeskogo schodstva dravidijskich i tjurkskich jazykov (na materiale tamil'skogo i ujgurskogo jazykov). LingvistiCeskaja tipologija vostoinye jazyki, ed. L. B. Nikols'skij, 205-212. Moscow: Nauka.

Bibliography

255

González Muela, José, ed. 1970. Alfonso Martínez de Toledo. Arcipreste de Talavera o Corbacho. Madrid: Castalia. Graff, W. L. 1929 Grandgent, C. H. 1928 Green, John N. 1976.

The word and the sentence. Language 5:163-188. Introducción al latín vulgar, trans. F. de Β. Moll. Madrid: Hernando. How free is word order in Spanish? Romance syntax. Synchronic and diachronic perspectives, ed. Martin B. Harris, 732. Salford: Salfoid University Press.

Greenberg, Joseph H. 1957a The nature and uses of linguistic typologies. International Journal of American Linguistics 23:68-77. 1957b Essays in linguistics. Chicago: University of Chicago Press. 1960 (1954) A quantitative approach to the morphological typology of language. International Journal of American Linguistics 26:178-194. 1963 (1957) Essays in linguistics. 2n