The Emergence and Development of SVO Patterning in Latin and French: Diachronic and Psycholinguistic Perspectives

The Emergence and Development of SVO Patterning in Latin and French This page intentionally left blank The Emergenc...

Author: Brigitte L. M. Bauer

7 downloads 540 Views 15MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

The Emergence and Development of SVO Patterning in Latin and French

This page intentionally left blank

The Emergence and Development of SVO Patterning in Latin and French Diachronic and Psycholinguistic Perspectives

BRIGITTE L. M. BAUER

New York Oxford OXFORD UNIVERSITY PRESS 1995

Oxford University Press Oxford New York Athens Auckland Bangkok Bombay Calcutta Cape Town Dar cs Salaam Delhi Florence Hong Kong Istanbul Karachi Kuala Lumpur Madras Madrid Melbourne Taipei Tokyo Toronto and associated companies in Berlin Ibadan

Copyright © 1995 by Brigitte L. M. Bauer Published by Oxford University Press, Inc., 200 Madison Avenue, New York, New York 10016 Oxford is a registered trademark of Oxford University Press All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of Oxford University Press. Library of Congress Cataloging-in-Publication Data Bauer, Brigitte L. M., 1960The emergence and development of SVO patterning in Latin and French : diachronic and psycholinguistic perspectives / Brigitte L. M. Bauer. p. cm. Includes bibliographical references and index. ISBN 0-19-509103-5 1. Grammar, Comparative and general—Word order. 2. Historical linguistics. 3. Typology (Linguistics) 4. Language acquisition. 5. Latin language—Word order. 6. Latin language—Grammar, Historical. 7. French language—Word order. 8. French language—Grammar, Historical. I. Title, P295.B36 1995 415—dc20 94-4023

1 3 5 7 9 8 6 4 2 Printed in the United States of America on acid-free paper

To my parents, Marius Bauer and Maria-Claret Bauer-Steinhauser, for their support, dependable and unwavering


Preface

When Antoine Meillet pointed out the importance of the Romance languages for general linguistics, he emphasized the continuity of linguistic change that can be observed in detail in these languages: the development from Latin to French continues the development from Proto-IndoEuropean to Latin. Le principe sur lequel on voudrait attirer Pattention, c'est que lc developpement roman continue lc developpement qui conduit de Pindo-europeen au latin. Il n'y a pas eu deux developpements successifs, mais un developpement continu, tantot lent et tantot rapide, qui va de Pindo-europeen aux parlers romans actuels. (1951:114)

Being an Indo-Europeanist, Meillet was sensitive to developments that go beyond the historical frame of individual languages; he was also attuned to parallel developments in other languages. Well aware of Meillet's observation, I shall attempt in the present study—which contains an analysis of an aspect of Latin syntax and its subsequent development—not to lose sight of the Indo-European perspective, nor of parallel developments in other languages. Fascinated by language change, a number of years ago I set out to examine linear order in Latin hierarchical structures and its development into French. This turned out to be a rich subject because a solid and ample base of linguistic data made it possible to formulate theoretical conclusions that may affect the essence of language. From the outset I want to emphasize that this is a syntactic analysis. Basic order patterns, which are the

viii

Preface

result of syntactic processes, require sui generis syntactic analysis. The assumption that word order is merely determined by semantic, logical, or pragmatic regularities cannot account for the real nature of the linguistic process involved, nor for the regularity in all phrases or the parallel with morphology. In this work I discuss the results of my research, presenting evidence for the hypothesis that the order of grammatical structures in Modern French is the result of a long-lasting evolution that psycholinguistic data can account for. The comparison of psycholinguistic and diachronic data has been proposed by the psycholinguist Slobin (1986) and also by Bichakjian (1987, 1988a), who, in a broader and somewhat different perspective, has attempted to explain the important phonological and grammatical changes in Indo-European—among them word order development—by referring to psycholinguistic data, that is, the chronology in language acquisition. This book represents a translation and revision of my doctoral dissertation, Du latin au francais: le passage d'une langue SOV a une langue SVO, which I submitted to the University of Nijmegcn (the Netherlands) in March 1992. The mores at the University of Nijmcgen do not allow the promovendus or the promovenda to thank his or her dissertation advisor. Consequently this is the first occasion I have to express my thanks to my dissertation advisor, Bernard Bichakjian (University of Nijmegen) for his help and his inspired teaching, which always focused on similarity and patterns in language change. I also wish to thank the staff of the classics department at the University of Nijmegcn, who often permitted me to attend courses on different variants of Latin despite my status as a student of French linguistics. Moreover, I am indebted to the Faculty of Arts at the University of Nijmegen and the Netherlands Organization for Scientific Research (NWO). Their material support allowed me to participate in international conferences and to spend part of the summer of 1990 at the University of Texas at Austin to attend the NEH Summer Institute devoted to "Perspectives on the Indo-European World," organized by Professor Edgar Polome, who made my participation possible and to whom I am most grateful. I also want to thank the other professors of linguistics at this institute: W. P. Lehmann (University of Texas at Austin), Thomas Gamkrelidze (University of Tbilisi, Georgia), Jaan Puhvel (University of California at Los Angeles), Karl Horst Schmidt (University of Bonn), and Werner Winter (University of Kiel). I am indebted to these scholars for the interest they showed in my research, for the extremely useful advise they gave me, and for numerous bibliographical references. I also wish to express my thanks to those with whom I had stimulating discussions and who sent me offprints or copies of articles related to the subject of my research: Melissa Bowerman (Max Planck Institut fur Psycholinguistik, Nijmegen), Robert Coleman (University of Cambridge), Robert de Dardel (University of Groningen), Charles Elcrick (University

Preface

IX

of Texas at El Paso), Ad Foolen (University of Nijmegen), Aafke Hulk (University of Amsterdam), Carol Justus (University of California at San Jose), Dick Klein (University of Nijmegen), Ans dc Kok (University of Amsterdam), Adrian Parvulescu (Montclair State College, New Jersey), Pieter Seuren (University of Nijmegen), Angela della Volpe (California State University at Fullerton), Leo Wetzels (Free University, Amsterdam), and Frederick Schwink (University of Texas at Austin), who kindly sent me books unavailable in Europe but that I needed badly. Finally, I would not have been able to cany out the translation nor the revision of my dissertation without the help of Professor W. P. Lehmann, at whose institute I spent seven months in 1992 to start a new research project, and who kindly offered to read the English version and to comment on the text. His suggestions, based on his exceptionally broad linguistic knowledge and inspired by the concept of the systemic organization of language, have been extremely useful. Dr. Julie Bellquist kindly found the time to read the manuscript and to eliminate infelicities. The kind help of these linguists was both encouraging and fruitful, greatly contributing to the results of my research. The shortcomings of this work are, of course, mine alone. During my stay in Austin, I came to know the staff of the Linguistics Research Center at the University of Texas, who apply and develop phrase structure analyses for advanced machine translation. The friendly and professional atmosphere at the institute had a stimulating effect on my work. Moreover, the Linguistics Research Center generously allowed me to use their computers and other facilities, for which I am very grateful. I am also indebted to the Niels Stensen Stichting, whose grant allowed me to spend 1992—93 abroad to do research at the University of Texas at Austin and at Cambridge University. While preparing this manuscript for publication, I received the kind help of Mr Richard Mendez (University of Texas at Austin), whose knowledge of American and European word-processing systems proved crucial. I am also much indebted to the anonymous readers for Oxford University Press and to senior editor Cynthia Read, whose suggestions, comments, and advice have been extremely valuable. Finally, I am grateful to the Royal Netherlands Academy of Arts and Sciences for accepting me as one of its research fellows, which enables me to continue my research in diachronic Indo-European syntax. My dissertation was dedicated to my parents. Since they have always shown great interest in my work and given me the support I needed, I see no reason to alter my dedication. Nijmegen, The Netherlands Autumn 1993

B.L.M.B.


Contents

1. Introduction

3

1.1 Earlier Studies of Word Order

5

1.1.1 The Erosion Hypothesis 5 1.1.2 Erosion as the Cause of the Shift from SOV to SVO 1.1.3 The Emergence of Fixed Word Order and the Loss of Cases 9

1.2 Attempts at a Broader Approach 10 1.3 The Historical Value of Latin 11 1.4 The Advantages of an Analysis in Terms of Branching and the Definition of Head 13 1.5 The Nature of the Change and Its Explanation 14 1.6 Outline of the Present Work 16 2. The Definition of Head

2.1 Linguistic Typology

18

19

2.1.1 Typological Dichotomy 19 2.1.2 Typological Correlations 20 2.1.3 The Problem of Terminology 21

2.2 Typology and Diachronic Linguistics 22

2.2.1 An Attempt at a Definition 23 2.2.2 A New Attempt: The Notion of Branching

24

7

xii

Contents

2.3 The Definition of Head 26 2.3.1 X-Bar Theory 27 2.3.2 The Identification of Head: Attempts Within the Framework of X-Bar Theory 29 2.3.2.1 Underlying Structure 29 2.3.2.2 Category Consistency 30 2.3.2.3 The Head as Obligatory Element and the Place of Occurrence 32 2.3.2.4 The Function of Head 33 2.3.2.5 The Lacking Definition 34 2.3.3 The Definition of Elements According to X-Bar Theory 35 2.3.3.1 The Distinction BetweenSpecifier Specifier Xn n and Complement X X 35 2.3.3.2 The Place of the Head in the Tree Structure 40 2.3.3.3 The Head of Morphological and Syntactic Structures: Analysis in Terms of Branching 41 2.3.3.4 The Subject and Its Function in the Sentence 44

2.4 The Analysis in Terms of Branching 3. Diachronic Analysis: The Noun Phrase 3.1 The Genitive

45 47

50

3.1.1 The Place of the Genitive in PIE 57 3.1.2 The Genitive in Two Other Italic Languages: Oscan and Umbrian 51 3.1.3 Old Latin 53 3.1.4 From Old to Classical Latin 55 3.1.5 From Classical Latin to Vulgar Latin and Late Latin 59 3.1.5.1 The Letters of Claudius Terentianus 60 3.1.5.2 Two Christian Authors: St. Jerome and St. Augustine 60 3.1.5.3 New Uses of the Genitive 61 3.1.6 Old French 63 3.1.6.1 The Place of the Genitive in Old and Middle French 63 3.1.6.2 The Morphological Change 63 3.1.7 Overview of the Development of the Genitive 64

3.2 The Adjective

65

3.2.1 PIE, Oscan, Umbrian, and Old Latin 3.2.2 From Old Latin Onward 67

65

xiii

Contents

3.2.2.1 Distinctive Adjectives

68

3.2.2.1.1 Antithetical Adjectives 69 3.2.2.1.2 Literal and Figurative Sense 3.2.2.1.3 Predicative Use 71

70

3.2.2.2 Descriptive Adjectives 71 3.2.2.3 The Adjective in Classical Latin: Conclusion 72 3.2.3 Vulgar Latin and Late Latin 73 3.2.3.1 The Letters of Claudius Terentianus 74 3.2.3.2 The Pcregrinatio 74 3.2.3.3 The Writings of St. Augustine 75 3.2.4 Old French 75 3.2.5 The End of the Middle Ages and Later 77 3.3 Determiners 78 3.3.1 Modern French Versus Early Indo-European Dialects 3.3.2 Latin: An Intermediary Stage? 79 3.3.2.1 The Numeral 79 3.3.2.2 The Demonstrative 80 3.3.3 Evidence from Turkish and Japanese 82

78

3.4 The Place of the Determiner, the Adjective, and the Genetive 83 4. Diachronic Analysis: The Verb Phrase

85

4.1 The Ummarked Place of the Verb and its Development 86

4.1.1 PIE

86

4.1.2 The Verb in Oscan, Umbrian, and Old Latin 4.1.3 From Old Latin Onward 89

88

4.2 The Marked Position of the Verb: The Initial Verb 4.2.1 Stylistic Motivation 93 4.2.1.1 Emphasis of the Verb 93 4.2.1.2 Pragmatic Function: Emphasis of the Utterance 94 4.2.2 Syntactic Motivation 95 4.2.3 The Use of the Initial Verb: Conclusion 96 4.3 The Medial Position of the Verb 97 4.3.1 The Elements that Follow the Medial Verb 97 4.3.2 Vulgar Latin and Late Latin 98 4.3.2.1 Petronius 99 4.3.2.2 The Letters of Claudius Terentianus 99 4.3.2.3 The Pcregrinatio 100 4.3.2.4 St. Augustine 101 4.3.2.5 Anonymus Valesianus 101

92

xiv

Contents 4.3.3 The Position of the Verb in Latin: Provisional Conclusion 102

4.4 The Copula 103 4.4.1 The Position of the Copula 103 4.4.2 The Emergence of the Preposed Auxiliary 104

4.5 The Position of the Verb in Old and Middle French 107 4.5.1 4.5.2 4.5.3. 4.5.4

The Position of the Subject 108 SVO Order 109 SOV Order 110 VSO and OVS Orders 111 4.5.4.1 The Direct Object in Clause-Initial Position 112 4.5.4.2 The Adverbial Phrase in Initial Position 4.5.5 Interrogative Clauses 116 4.5.6 The Subject Pronoun 117

113

4.6 The Adverb 118 4.6.1 The Morphology of the Latin Adverb 4.6.2 The Position of the Adverb 722

119

4.7 Provisional Conclusion: The Verbal and Nominal Phrase 725

5. Diachronic Analysis: Early Right-Branching Structures 128 5.1 The Prepositional Phrase 725 5.1.1 The Creation of a Hierarchical Structure 729 5.1.2 The Shift Toward Right Branching 130 5.1.3 The Italic Languages: Residues of Former Postposing 130 5.1.3.1 Oscan and Umbrian 130 5.1.3.2 Latin 131 5.1.3.2.1 Direct Juxtaposition in the Prepositional Phrase 131 5.1.3.2.2 The Sequence [[Complement] Adposition] 132 5.1.3.2.3 Postpositions of Nominal Origin 134 5.1.4 Extension of the RB Prepositional Phrase 137 5.1.4.1 Nominal Declension 138 5.1.4.2 Declension of Nominal Forms of the Verb 138 5.1.5 Conclusion 139

5.2 The Comparative Construction 140

5.2.1 The Morphology of the Comparative 140 5.2.1.1 The Latin Comparative 141 5.2.1.1.1 Cracks in the LB System of Latin 5.2.1.1.2 The Spread of the Analytic Form

141 742

xv

Contents

5.2.1.2 Old and Middle French: The Rise of One Form 143 5.2.2 The Term of Reference and Its Evolution 144 5.2.2.1 The Comparative Construction in PIE 145 5.2.2.2 The Place of the Latin Comparative in the Ongoing Evolution 149 5.2.2.3 The Ablative of Comparison in Latin 149 5.2.2.4 The Construction Featuring Quam 150 5.2.2.4.1 The Original Structure 150 5.2.2.4.2 Motivation of the Use of Quam Versus the Ablative of Comparison 152 5.2.2.4.3 Branching 755 5.2.2.5 A Brief Revival of the Ablative of Comparison 156 5.2.2.6 The Prepositional Construction 157 5.2.3 Conclusion 158

5.3 The Relative 159 5.3.1 The Original Structure: Correlation 160 5.3.2 The Creation of Right Branching 162

5.4 Conclusion

165

6. The Acquisition of Branching

168

6.1 Language Acquisition 171 6.2 The Acquisition of Basic Word Order 173 6.2.1 Japanese, an LB Language 174 6.2.2 Turkish, an LB Language 775 6.2.3 The Acquisition of Basic Word Order in Russian and Polish 777 6.2.4 The Acquisition of Basic Word Order in English and German 178 6.2.5 The Acquisition of the Strict RB Basic Word Order of French 182 6.2.6 The Acquisition of Basic Word Order: Conclusion 182

6.3 The Acquisition of LB Morphological Structures 183 6.3.1 The Acquisition of Inflection in Russian, Polish, and Latvian 183 6.3.2 Prepositional Phrases 186 6.3.3 The Acquisition of Inflection: Conclusion 188 6.3.4 The Acquisition of Turkish and Hungarian Agglutination, and Japanese Case Particles 189 6.3.4.1 Turkish 189 6.3.4.2 Japanese 190 6.3.4.3 Hungarian 191 6.3.5 Prepositional Phrases Versus Postpositions 193

xvi

Contents 6.4 Acquisition of Complex Structures 794 6.4.1 Typological Characteristics of the Relative 195 6.4.2 The Acquisition of RB Relative Clauses in English, French, German, and Polish 197 6.4.2.1 French 199 6.4.2.2 German 199 6.4.2.3 Polish 200 6.4.3 LB Complex Structures: Relative Clauses in Japanese, Korean, and Turkish 202 6.4.3.1 Japanese and Korean 202 6.4.3.2 Turkish 205 6.4.4 Data on the Acquisition of Relative Clauses: Synthesis 206 6.5 Conclusion: The Advantage of Right Branching

7. Conclusion

207

213

7.1 Diachronic Change and Language Acquisition 213 7.2 Left Branching and the Linguistic System 216 7.3 The Broader Perspective 218

References Index

237

279

The Emergence and Development of SVO Patterning in Latin and French


1 Introduction

Many have deplored the loss of Latin, which they considered the epitome of a concise, systematic, and even elegant language. In the past, poets as well as linguists have identified the linguistic system of Latin with the civilization it carried: the language of Rome reflected, it was thought, the elegance and refinement of its civilization. These epithets did not apply, of course, to modern times, nor to their languages. Even in recent history we find examples of the prestige attributed to inflected languages. For Schleicher, for instance, according to whom language was to be considered a living organism, Latin as well as the Ursprache reflected maturity in its full bloom, whereas the Romance development was only the decay and decomposition of a living organism (1873; Robins 1979:181 et seq.). And even in this century Marr, the official linguist of the former Soviet Union, classified the inflected languages as being the most evolved; isolating languages on the other hand were considered primitive and therefore occupied a low place in the linguistic hierarchy. The political reasons were clear as well as the motives that pushed the Soviets to change their official linguistic judgment when the communists came to power in China: Chinese is an isolating language (see Lehmann 1993:38). The aim of the present study is not to bewail the loss of Latin nor to praise Russian inflection. Its aim is to analyze and identify the nature of the development the structures of Latin underwent and to understand the advantage of the new structural organization. Therefore, this study has two parts: first I will examine in detail the ordering of the Latin grammatical sequences and the reorganization they underwent. Then, I will attempt 3

4

The Emergence and Development of SVO Patterning

to show that the diachronic development can be related to findings in language acquisition. The comparison of diachronic and psycholinguistic data will thus enable us to explain a linguistic phenomenon the importance and extent of which have often passed unnoticed: the change, in the development from Latin to French, of the linear order in syntactic as well as morphological structures. As a first step I will prove that the morphological and syntactic structures shifted from one type of linear organization—characteristic of the ancient language—to another: in the Latin structures the complement (or modifier) generally preceded its head and was therefore left branched, whereas in Modern French the order is the reverse: the complement follows; it is right branched. See for instance, [[exercitum] duxit] 2 1 'he led the army'

[il conduisit 1

[P armee]] 2

where the verbal object precedes the head of the Latin phrase but follows it in the French example; or [[can] -ibus] 2 1 'to the dogs'

[aux 1

[chiens]] 2

where the postposed head, the grammatical marker, of the Latin word has been replaced with a preposed adposition in French. Diachronic analysis of grammatical structures will reveal that this reorganization did not occur in all the grammatical structures at the same time, but that there is a linear development. And even if one observes occasional retrograde tendencies, the other Indo-European languages—independently —have undergone the same development (see Lehmann [1974], who observes the shift from (S)OV to (S)VO; see also Bichakjian 1987, 1988a). The universal character of the shift implies that it cannot be attributed to external or ad hoc factors. It is in child language that I hope to find the broader explanatory frame that this type of change requires, relating the regularity of diachronic change to the patterns in language acquisition. The comparison of the rhythms of acquisition of both types of branching may show that the modern, right-branching structures present important advantages over their left-branching counterparts. These advantages provide objective criteria with which to judge the complexity of the structures and permit us to understand or even to account for the diachronic shift toward right branching. Earlier, comparison of diachronic and psycholinguistic data has been proposed by Andersen (1980) and, more recently, by Slobin (1986), according to whom there is a correlation between linguistic features that arc stable over time and the ease of their acquisition (sec chapter 6 in the present study). In an evolutionary, hence dynamic, perspective Bichakjian also correlates diachronic and psycholinguistic data, but he assumes a

Introduction

5

reverse parallel between the diachronic development of language and the order in the acquisition of grammatical structures, consequently claiming that the Indo-European languages are evolving "in the direction of earlier acquired speech sounds and grammatical features . . .: [they] have evolved or arc evolving by eliminating late-acquired features and by replacing them with earlier acquired ones" (1988a:12—13). Only if this parallel can be proven in phonology, morphology, and syntax—but only then—might one try to extrapolate a principle and explain it (for example, Bichakjian [ 1988a] tries to correlate linguistic development and biological evolution). Yet the biological correlation implies a universality in language change that can only be assumed on the basis of a much larger and more varied corpus of language data than Bichakjian has presented. Alternatively, in a more direct linguistic perspective and in line with the idea that synchronic language patterns are systemic, it is legitimate to argue that the various parallel linguistic changes—in morphology, syntax, and lexicon, for example —are integrated in the comprehensive shift from one type of language to another (see Klimov 1986; Lehmann 1990). Once this pervasive change has been demonstrated, one can set out, in a subsequent stage, to account for it. In this preliminary chapter I briefly discuss the earlier descriptions and explanations of the evolution of word order in French. Generally the extent of the change has not been perceived in these studies nor has the diachronic change been related to the chronology of language acquisition.

1.1 Earlier Studies of Word Order Traditionally diachronic studies of French have focused on the disappearance of nominal inflection, the emergence of fixed word order, and the shift from synthetic to analytic forms. Some scholars have related the first two changes, claiming that the disappearance of cases due to phonetic erosion compelled the emerging of fixed word order. This hypothesis takes into consideration only changes that occurred in nominal inflection. Therefore, in order to summarize the entire linguistic development that took place from Latin to French, a second pervasive change had to be postulated: the replacement of synthetic with analytic forms (for instance, Latin laudaveram, which was replaced by French j'avais loue). Yet the observation as such does not integrate the change into a broader frame; consequently, the modifications that took place in the verbal paradigm remain an isolated phenomenon. This explains why the development from Latin to French is generally summarized by assuming two major changes: the creation of fixed word order and the replacement of synthetic by analytic forms.

1.1.1 The Erosion Hypothesis If one tries to explain the development of word order by referring to phonetic erosion, one still has to account for this erosion, that is, the

6


disappearance of case endings. (See Bichakjian [1987:88], who correctly uses this argument against the hypothesis proposed by Vennemann; see also section 1.1.2.) The process of erosion is all the more difficult to understand because case endings, according to this same hypothesis, are supposed to play a prominent role in the sentence. Finally, the hypothesis of the appearance of fixed word order presupposes an indiscriminately free word order in Latin. This presupposition ignores the testimonies of the period and the corroborating data which prove that word order in Latin was not as free as it is often claimed to be. Quintilian, for instance, recommends that the verb be placed at the end of the clause: "vcrbo scnsum cludere multo . . . optimo est" (9.4.26). This admonition does not reflect an obsolete cause as does, for instance, the Appendix Probi; rather, it shows a situation where word order was certainly freer than in Modern French, but where there was a preferred order, in spite of the possibilities offered by a full-fledged inflectional system. In Modern Russian one observes the same situation: notwithstanding the variety that case endings allow, there is a dominating word order that is unmarked (see Jakobson 1963:212). In the introduction to his monumental work, Marouzeau correctly argues that Latin had no fixed word order, but that two different word orders were never synonymous: L'ordre des mots en latin est libre, il n'est pas indifferent. Libre, en ce sens que, sauf exception, il n'y a pas pour chaque terme de la phrase une place attitree, obligatoire. Mais non pas indifferent, parce qu'en general deux ordres possibles ne sont pas synonymes. (1922:1)

Without using the terms "marked" and "unmarked," which were not yet common at that time, Marouzeau indeed distinguishes between the normal and the exceptional, or motivated, order in the Latin structures (1922, 1938, 1949, 1953). The notion of markedness is fundamental to the study of word order because it accounts for the fact that the structures of a language display an unmarked order, that is, the order which the speaker uses without any particular purpose or objective, or in nonspecific contexts. The marked order, on the other hand, is related to specific syntactic and stylistic contexts (interrogative phrases, subordinate clauses, etc.). Analysis of syntactic phrases indeed proves that Latin had an unmarked word order for each sequence; as will be shown in chapters 3 through 5, the elements of each phrase were ordered in earliest times in Latin or before according to the principles of an SOV language. Exceptions to the normal order can be explained by stylistic and syntactic reasons: they present the marked order. Romance languages, on the other hand, feature SVO structures (see Hawkins 1983:320-42; Harris 1978:7). Accordingly, the observation made by Ewert and quoted by Price is not totally adequate because it excludes the existence of an unmarked word order: "[T]he fundamental change in word order from Latin to Modern French . . . consists in its rise as a syntactic device and its decline as a stylistic device" (Ewert 1943, quoted in Price 1971:259).

Introduction

7

Even if word order in French is more strict than it was in Latin, the fundamental change in the development from Latin to French is not the emergence of fixed word order but the shift from an SOV to an SVO language (see also Harris 1978:5-7). This change is part of a comprehensive evolution, which affects not only syntax but also morphology. Since the shift from an SOV to an SVO order is indeed the major change in the development from Latin to French, the erosion hypothesis is refuted as the explanation of the emergence of word order.

1.1.2 Erosion as the Cause of the Shift from SOV to SVO In the 1970s Vennemann took up the erosion hypothesis again, arguing that it can explain the shift from SOV to SVO: A language with S-O morphology tends to be XV; as reductive phonological change weakens the S-O morphology, and does not develop some substitute S-O morphology, the language becomes a VX language. Since in VX languages order is a major grammatical marker, the order becomes increasingly rigid. (1974:359)

According to Vennemann's hypothesis, it is the intermediary position of the verb in an SVO, hence noninflectional language, that distinguishes the subject and the object. This observation might be interesting if one compares only SOV and SVO ordering. But in a VSO language one finds exactly the same concentration of nouns and in the same alignment— SO—even if there are no case endings used to avoid grammatical confusion. Consequently, the alleged distinctive function of the intermediary verb is not convincing. Vennemann's hypothesis has been criticized in extenso by Sassc (1977) and, more recently, by Bichakjian (1987), who argues that structural changes generally occur independently of morphological erosion, and that they are not caused by this process. Indeed, several Indo-European and non—IndoEuropean languages "have developed an SVO order while having quite clear case markers" (Bichakjian 1987:89). Moreover, research by Koch (1974) and Miller (1975) has shown that in Indo-European SVO order chronologically precedes the loss of cases (see Bichakjian [1987:89], who quotes Sasse [1977]). These considerations are substantiated by data from Old French and by synchronic typological evidence. The existence of SVO languages with a case system and, inversely, the occurrence of SOV languages without nominal inflection (see Sasse 1977) imply that there is no necessary causal relation between the loss of cases and the emergence of SVO order. Data from Old French corroborate this conclusion. In Old French the unmarked order was SVO, although the only two cases that still existed to a certain extent were the most important ones in this respect, namely, the nominative and the accusative (see Foulet 1923:36, 257). It is true, of course, that Old French did allow other word orders in specific contexts (see section 4.5) and that the case system

8


showed gaps. Nominal inflection was defective because it was limited to masculine nouns, and in this subcategory there were numerous indeclinablcs (for instance, proper names; see Woledge 1974:80—81). Moreover, declension no longer had the same effect everywhere (Woledge 1974:94; see also Pope [1956:310 et seq.], who gives more geographical details). The status or actual functioning of the case system in Old French and also the various degrees of functioning in different parts of the territory are definitely a matter of discussion. (For the defective use of case in manuscripts, see Cerquiglini et al. [1976]; for the use of cases as a literary means, see Cerquiglini [1983]; see also Fleischmann 1990a, 1990b). Yet independently of the results of this discussion, evidence from Old French shows that the accusative and nominative were the cases that survived longest; one might therefore assume a period of coexistence of case marking and the emerging SVO word order. Consequently the disappearance of nominal inflection cannot have been the ultimate cause of the shift toward SVO. Evidence from Modern Russian corroborates this conclusion: the case system is still well established although the unmarked word order is already SVO (see Jakobson 1963). There is a second consideration on which Vennemann (1974) has based his hypothesis and which has often gone unnoticed. The shift from SOV to SVO is attributed to the ambiguity caused by the topicalization of the object. In a language with SVO word order the topicalization of the object would give OSV, in an SOV language it would give OSV, and in a VSO language it would give OVS. Accordingly, only in an SOV language does the distribution of verb versus noun phrases remain unchanged by the process of topicalization: N[oun]P[hrase] NP V[erb] remains NP NP V. As long as there are case endings, Vennemann assumes, the distribution of the grammatical functions is clear and the speakers know they have to do with a marked or unmarked order. Yet, once the noun declensions have eroded, the sentence is ambiguous and the intermediary position of the verb is needed to distinguish the subject and the object and will thus create a sequence—TVX—which deviates from the normal SOV order. With time, Vennemann argues, the sequence TVX will become the unmarked order with the subject as topic (that is, if it is a nominative language), which will give SVO. Once this order is established, typological consistency is supposed to trigger the total reorganization of the language according to SVO typology, because "much of the word order history of a language is a development toward consistency within its type" (Vennemann 1974:353). This hypothesis implies ipso facto that the removal of the verb marks the beginning of a new typology. Diachronic Italic and Romance data, however, refute this chronology (see chapters 3 through 5) and therefore also the explanation proposed by Vennemann. Moreover, Vennemann's hypothesis is based on the assumption, which he does not discuss, that SXV, SVX, and VSX automatically have XSV, XSV, and XVS as their marked variants. It is legitimate to ask whether topicalization is the only means for emphasizing the direct object. One

Introduction

9

would expect another strategy, more in accordance with the ordering principles of the language under consideration, rather than a construction that—according to this same hypothesis—will not fail to reorganize all the structures of the language. If typological consistency—in casu SOV—is the moving force behind word order change, it would be likely to restrain the movement toward typologically different structures, the more so since marked, hence exceptional sentences are involved. Therefore, other strategies would seem more appropriate; cf, for instance, Dutch, which is a noninflectional language and displays SVO in main clauses but SOV in subordinate ones. The emphasizing of the object by topicalization creates the sequence OVS and not OSV, as Vennemann claimed (1974:355). Compare de jongens zagen het meisje the boys saw the girl

(SVO)

and

het meisje zagen de jongens the girl saw the boys 'it is the girl the boys saw'

(OVS)

As in the case of SOV discussed by Vennemann, the marked variant has the same ordering of phrases (NP V NP) as the unmarked sequence. Yet Dutch, far from changing the structure of its clause, uses another means: it either changes the prosodies of the sentence or uses a passive or periphrastic structure.

1.1.3 The Emergence of Fixed Word Order and the Loss of Cases Whereas Vennemann tried to explain the emergence of SVO order by claiming the importance of the intermediary verb, other linguists, among them the well-known Romanist scholar Edouard Bourciez, have proposed a different chronology between the disappearance of cases and the rise of word order. Despite greater freedom, Latin had, according to Bourciez, a preferred word order (1956:13), and the disappearance of case endings was triggered by the reorganization of word order (1956:22). After this structural reorganization, case distinctions were no longer needed and nominal inflection gradually disappeared: hence, according to Bourciez, the loss of inflection was closely related to changes in syntax (1956:22). Like the linguists who advocate the erosion hypothesis, Bourciez attempts to reconstruct a linguistic phenomenon, but the very first change which dragged along the rest remains to be explained. Although the chronology proposed by Bourciez corresponds better to linguistic reality, the reason why there was a change in word order remains to be found. Moreover, as for phonetic erosion, only the disappearance of the grammatical

10


cases is taken into consideration; the rest of the inflection, that is, the oblique cases, and the conjugation are not taken into account, and the tendency toward analytic forms also remains to be explained. Traditionally the disappearance of Latin inflection has been summarized as being a shift from synthetic to analytic forms which was supposed to be inspired by a need for clarity: the analytic form is considered clearer (e.g., Bourciez 1956:23). The replacement of synthetic forms with analytic ones is undeniable, but not the principal issue: another change was much more important because it affected not only morphological but also syntactic structures. This comprehensive change I will describe and explain in this study.

1.2 Attempts at a Broader Approach In the preceding pages I have demonstrated that neither the erosion hypothesis nor the interpretation by Bourciez can account for the extent of the change that has taken place in the development from Latin to French. The first linguist who set out to integrate the various changes into one development was von Wartburg. He observed that the order of morphological elements in Modern French is not arbitrary, but that the determining (French determinatif) element precedes the determined French determine) element (von Wartburg 1971 [1934]:256). In Latin the order is the reverse. Taking up again this idea of von Wartburg, Baldinger pointed out that the development of the inflectional system of Latin can be summarized as being the substitution of Postdeterminierung by Pradeterminierung: Modern French Pradeterminierung is opposed to Latin Postdeterminierung (Baldinger 1968:88). This was an important observation; yet Baldinger limited his concern to morphological structures without trying to establish a parallel between changes taking place in both fields of grammar. However, a thorough and consistent analysis of the Latin structures and the modern French equivalents reveals the reversal of the order of hierarchical elements in the morphological as well as the syntactic structures. In the archaic organization the complement was placed to the left of its head, whereas in the modern structures it is placed to its right; for instance, laudaverat

vs.

il avait loue

canibus vs. pour les cbiens bellum gerunt vs. Us font la guerre

'he had praised'

'for the dogs' 'they make war' (Caes. DBG 1.1)

In order to describe this organization, which is both linear and hierarchical, I will use the term "branching." This notion was used by Chomsky in 1965 and has been taken up again recently by Bichakjian (1987) to describe the evolution of the ordering of elements in the grammatical structures of Indo-European; he thus proposed a syntactic analysis

Introduction

11

of the diachronic data in terms of branching. The chief advantage of this approach is that it reveals the shift underlying the changes that have occurred in syntax as well as morphology. In this way one can observe, in the history of the Indo-European languages, the shift from archaic structures, where the complement precedes the head—left-branching (LB) structures —to modern structures, where the complement follows the head—rightbranching (RB) structures (cf. Bichakjian 1987:96). This shift can be followed in detail in the development from Latin to French: Modern French, whose right-branching structures are almost exclusive, developed from a language which was basically left-branching. The aims of this study are primarily to analyze and discuss the change of branching in morphological and syntactic structures, and to show the cohesion of the grammatical changes that have taken place in both types of structure, inflected words and syntactic phrases. The comparison of Latin left-branching morphological entities and their right-branching French equivalents will also show the effect of left branching in morphological structures: when left branching, these structures form autonomous words; when right branching, they form phrases; cf. Latin laudabo or French (je) louerai versus (je) vais louer 'I will praise'. Consequently the shift toward analytic forms is no longer an independent phenomenon but is related to branching patterns—in Indo-European languages, that is. In this respect my analysis differs from Harris's (1978). Harris explicitly relates the loss of inflection to the emergence of SVO order, indicating that it is part of the evolution toward structures of the type specified-specifier (1978:16). Changes that took place in the verbal paradigm (for example, laudaveram being replaced with j'avais loue I had praised') are integrated in this same development. Yet in Harris's analysis the shift from synthetic to analytic remains an isolated phenomenon. Harris argues that it is a typical Romance development, independent of word order patterns and determined to a large extent by the tendency toward explicitness (1978:16). A systematic analysis of the morphological structures in terms of branching will inform us, however, of the relation between the ordering of morphological and lexical elements and the synthetic/analytic nature of the structure in question. If one observes the shift toward right branching in all structures, it did not occur in each of them at the same time. Accordingly, the analysis of each structure starts at different, though not arbitrary, periods in history; in order to analyze the nominal phrase, for example, I will have to go back to Old Latin, whereas for the analysis of the verb phrase even data from modern times are needed.

1.3 The Historical Value of Latin Romanists are in the extremely privileged position of studying languages which have a well-documented history. Yet many linguists refer to the gap that would have existed between spoken Latin, from which the Romance

12


languages derived, and written Latin, which is the principal source of our knowledge of Latin (see Pulgram 1950). This kind of consideration may distort an advantage and change it into a disadvantage. If indeed there exist differences between spoken and written Latin—which is a perfectly normal phenomenon—we still are dealing with two variants of one and the same language. Therefore it is difficult to draw their boundaries and even more so to consider these variations of Latin exempt from mutual influence. The authors of certain periods were indeed quite conscious of the literary value of their texts and they were after stylistic effects, but they were not writing in a foreign language—communication remained the chief aim of their activities. Since literate people were very familiar with the texts and the grammatical rules, grammar influenced the evolution of language. But if this indeed slowed language change or perhaps even caused retrograde movements, it did not alter its nature. The written and spoken variants of Latin both evolved in the same direction but at different rates: the written language developed at a slower rhythm than the spoken variant. Accordingly, literary texts show traces of the linguistic changes that finally produced the modern structures of the Romance languages. These findings are corroborated by the observation of Mohrmann, who characterized common and spoken Latin, on the one hand, and written and cultivated Latin, on the other, as being "deux formes d'une seule langue revetant, essentiellement, une meme structure, mais evoluant selon un rythme tres different: la langue cultivee etant fixee par un normativisme rigoureux, la langue courante evoluant plus libremcnt" (1962:94). Absolute dates arc not indispensable to the aim of this study. Consequently, for the analysis and the explanation of the evolution of grammatical structures, it is legitimate to rely upon data from written sources, even—with the necessary precautions of course—from written sources of the Classical period. In addition to these considerations we observe that the Romans themselves say little about the "double" nature of their language. It is therefore legitimate to wonder about the grounds on which the hypothesis of a real Latin "bilingualism" is based. Only a few texts give evidence of the awareness of the Romans of what they call urbanitas and rusticitas. Cicero mentions it, but he has some difficulty defining the difference between these two variations of Latin that distinguish the people who live in Rome from the rest (Brutus 171). This difference shows most clearly in pronunciation and in vocabulary. Yet if people felt the difference, they were not able to describe it. Quintilian also attempted to define the two notions: Nam et urbanitas dicitur, qua quidem significari video sermonem praeferentem in verbis et sono et usu proprium quendam gustum urbis et sumptam ex conversatione doctorum tacitam eruditioncm, denique cui contraria sit rusticitas. (Inst. Or. 6.3.17) [ First, there is urbanitas, which I perceive denotes language with a flavor of the city in its words, accent, and idiom, and further suggests a touch of learning

Introduction

13

derived from associating with well-educated men; in short, it represents the opposite of rusticity.]

This text shows once more that the difference between both variants was especially evident in the pronunciation and in the choice of words, and not in the grammatical structures, which are the object of this study. I am very well aware that neither Quintilian nor Cicero was an experienced linguist, and it is well known that the popular inscriptions show a defective use of the case system and a low occurrence of absolute constructions or of the gerundive; but neither absolute nor gerundive constructions were excluded from official inscriptions nor from dedications. Therefore, there is no conclusive evidence that points to Latin bilingualism.

1.4 The Advantages of an Analysis in Terms of Branching and the Definition of Head I have already referred to the study of Baldinger (1968), who pointed out the shift from Postdeterminierung to Pradeterminierung in the development of morphological structures; his analysis, however, is limited to morphology. Typological studies on word order, on the other hand, are directed at syntactic units (see, e.g., Harris 1978). Diachronic evidence suggests that the reversal in morphology is consistent with the one in syntax, and therefore, that both need a common analysis. This makes the analysis in terms of branching extremely useful because it shows in precise and independently defined terms that the syntactic and morphological changes are not arbitrary or independent, but that they are integrated in one comprehensive shift that started in the early periods of Indo-European and continues in Modern French. In the sixties and seventies inflection had been considered an appendix of SOV typology: "[I]f in a language the verb follows both the nominal subject and nominal object as the dominant order, the language almost always has a case system" (Greenberg 1963:75; see also Lehmann 1974, Harris 1978). Accordingly, the loss of SOV automatically pulls along the loss of inflection. The actual link between the syntactic phenomenon— SOV—and the morphological system, however, is neither indicated nor established. Yet with an analysis in terms of branching, inflection turns out to be a characteristic feature, on the level of morphology, of a leftbranching language, because the lexical element is systematically placed on the left of the grammatical marker. The creation of auxiliaries, and of preposed personal subject pronouns and prepositions, is part of the shift toward right-branching structures. Accordingly, the notion of branching not only describes isolated changes, but also accounts for the coherence of the grammatical change in precise and independently motivated terms. Although the current analysis will be realized in terms of branching, I have used the more conventional notion SVO in the title of this work: the notions SOV versus SVO have become established and arc better known than the concept of branching.

14


Since the proposed analysis will account for the linear organization of the hierarchical elements in grammatical structures, it is necessary to define in precise words the two concepts which are fundamental to this notion, the head and the complement or modifier. The definitions that have been proposed until now are based on synchronic and diachronic analogy. And although the observations arc correct, these definitions are not independently motivated. Therefore, before passing on to diachronic analysis I will first define the concepts of head and complement, using X-bar theory as a starting point for the discussion. With X-bar theory generative grammar finally included a component of phrase structure rules to which the notion of head is fundamental. Although I do not agree with all the interpretations of the theory, I will use some of its notions and structural interpretations. Consequently X-bar theory will be a starting point in my search for a definition rather than a theoretical frame to work in. To have X-bar theory as a starting point underscores the grammatical nature of this analysis, to which I have already referred. In contrast with purely pragmatic approaches, this analysis assumes an unmarked order, which is the result of a grammatical process; exceptions to these basic ordering patterns are accounted for by either syntactic, pragmatic, or stylistic processes (sec also Marouzeau 1992, 1938, 1949, 1953). As such, this study challenges analyses that do not consider word order as the result of grammatical processes but emphasize instead mechanisms of topic/ comment, pragmatics, and so on.

1.5 The Nature of the Change and Its Explanation Once the shift in the linear ordering of hierarchical structures in Latin syntactical and morphological units has been analyzed in well-defined terms, one can try to explain it. First of all it is important to stipulate that this change is not limited to Latin-Romance development. The change in branching also took place in other Indo-European languages and outside this language family. Studies of word order in various languages most commonly use the labels SOV/SVO or OV/VO. In these studies the same conclusion has been arrived at independently—the language under consideration changed from (S)OV to (S)VO, and the loss of inflection was parallel to this development (see Lehmann [1972c, 1973, 1974], who discusses various Indo-European languages; Van Kemenade [1987] on English; Harris [1978] on French; and Givon [1977] for non—IndoEuropean languages). If, indeed, "the overall pattern of changes from P[roto]I[indo]E[uropean] to the dialects is from an OV to a VO structure" (Lehmann 1974:250), it is legitimate to go further and to assert that the evolution of the Indo-European languages is characterized by the replacement of left-branching structures with their right-branching equivalents (see Bichakjian 1987, 1988a). The Indo-European scope of the change is also manifest in its chronology: in some structures the change

Introduction

15

took place very early, in the Proto-Italic period or even earlier (see Bauer in press a). Many explanations have already been proposed to account for the change in word order: the erosion hypothesis, the afterthought hypothesis (see, e.g., Hyman 1975) or the grammaticalization hypothesis (see Bichakjian [1987] for a discussion). Since these explanations are often ad hoc, they do not account for the overall change and they ignore its real nature. In order to find a valid explanation, one must first determine the nature of the shift that took place. The change at issue does not only occur in all the Indo-European languages—and even in non-Indo-European languages (see, e.g., Sasse 1977)—it is unidirectional as well. The cyclical interpretation (see, e.g., Vennemann 1974) is not appropriate, because languages normally—that is, without direct external influence—do not shift from a right-branching to a left-branching organization: or, in other words, from SVO back again to SOV and from noninflectional to inflectional. Yet, the attempts at explanation are a priori based on the assumption that language change is cyclical. Although Lehmann, for instance, argues that the IndoEuropean dialects developed an SVO typology out of an SOV protolanguage, he is reluctant to accept the idea of unidirectionality: "[A] unidirectional change cannot simply be assumed, for external influences may have reintroduced OV patterns" (1974:39). But this latter change is of a different kind. In the history of the Indo-European languages, which are the world's best documented and analyzed languages, one can distinguish two types of change. Eirst of all, there are ad hoc and arbitrary changes that can be explained by external factors and influences—for instance, the use of SOV in the German subordinate clause, which is ascribed to scholarly influence. These modifications are most often language specific. The second type of change, however, occurs in all languages, though at different rates, at various periods, and independently. In addition, even if external factors can accelerate or retain these changes, they are nonetheless unidirectional. This fundamental distinction was made, for example, by Meillet (1926), who opposed the "lignes generales" and "les divergences," as well as by Sapir, who used the notion of "drift" (1949), but it never got the attention it deserved. As Professor Lehmann pointed out (personal communication), we find the notion of directed change already in Bopp, which was then related to the concept of language as a living organism (see the beginning of this chapter): "[W]e should emphasize his [Bopp's] insistence that language be regarded as a structure that undergoes change along lines that are not haphazard" (Lehmann 1991:277). The explicit distinction between both types of language change we observe in Meillet has been taken up again recently by Bichakjian, who uses it as a basis for his evolutionary concept of language change: "the inherent changes" make out the linear development of language (1988a). Without drawing conclusions about evolutionary correlates in the Darwinian sense—for which I am not qualified—I think the distinction between both types of change is

16


indeed crucial, because it reveals the systematic nature of language and its development. The comparison with biology in itself is interesting because although the fundamental principles or the driving power arc still not understood, the concept of evolution is accepted in biology; hence phylogenctic change is considered unidirectional. In diachronic linguistics, the emphasis is too often on changes that are language specific or that are of minor importance. The overall changes as well as their universal character are then easily disregarded. Diachronic analysis, however, reveals irreversible changes in phonology, morphology, and syntax. In phonology, for instance, vowels developed from a quantitative to a qualitative system and complex consonants gave way to noncomplex phonemes. In morphology, aspectual distinctions were gradually replaced with temporal oppositions. This development is parallel in syntax to the loss of middle and the creation of the opposition active—passive. One also observes the emergence of subordination and the shift from left to right branching (for these developments and their theoretical importance, see Bichakjian 1988a). The evolution of word order is, therefore, one of the irreversible syntactic changes. Once the universal and unidirectional nature of the change in branching has been recognized, one can set out to relate this shift to the chronology of first language acquisition. It is legitimate to establish a correlation between language change and language acquisition, because in both fields the chronological order of the emergence of structures is essential. Moreover, on the basis of acquisition data it is possible to judge the difficulty and complexity of morphological and syntactic structures. Consequently, in order to understand the shift toward right branching, it is relevant to compare the acquisition rhythm of left-branching and right-branching structures in a number of Indo-European and non—Indo-European languages. Comparison of diachronic and psycholinguistic data then contributes to a better understanding of language change, and linguistic systems in general.

1.6 Outline of the Present Work The aim of this study is to go beyond the descriptions and explanations of French development that have been proposed until now, and that were for the most part too narrow in scope. I set out to show the comprehensive evolution underlying the changes that on the face of it might appear independent or arbitrary. The analysis of grammatical structures in terms of branching will reveal the principle underlying the ordering in morphological and syntactic units, the cohesion of the grammatical changes, and the shift from left to right branching. Yet, since the nature of branching is at issue, it is extremely important to formulate an independently motivated definition of the notion of head. Accordingly, before passing on to diachronic analysis of the grammatical structures, I will define the concept of head (chapter 2). The analysis of diachronic data will then reveal the shift

Introduction

17

underlying typological change (chapters 3 through 5). Finally, I will compare these findings with psycholinguistic data and argue that right-branching structures present important advantages over their left-branching equivalents (chapter 6). Evidence from child language will thus make it possible to understand or even to explain the evolution of word order, which is characterized by a gradual shift toward right branching.

2 The Definition of Head

In the introductory chapter I indicated the aim of this book, namely, the analysis and explanation of the linear reorganization of grammatical structures from Latin to French. Since these grammatical changes will be accounted for in terms of branching, the analysis is based on the independently motivated definition of the elements between which a branching relation exists, that is, the head and its complement or modifier. In this chapter I give a definition of head that can be used for morphological as well as syntactic structures. As a first step, I briefly discuss the two current approaches in word order typology (2.1) and the analysis of diachronic data in a typological perspective (2.2). Accepting the approach of Bichakjian (1987), who has proposed a description in terms of branching, I apply a number of structural principles from X-bar theory that may account for the regularity observed in grammatical structures and for the identification of the head. Since typology itself has not supplied a convincing definition of head, it is legitimate to turn to generative grammar—that is, a phrase structure analysis—even if it is not the theoretical framework of my analysis. X-bar theory differs from other generative approaches in that the notions of head and syntactic hierarchy are its basic principles. Consequently X-bar theory has been the basis of numerous attempts to identify the head in syntactic and lexical structures (2.3.2). Since these attempts are not convincing in every respect, it will be necessary to turn back to the source, X-bar theory itself (2.3.3). First I discuss several aspects of the traditional model by Jackendoff (1977) that seem important for the definition I want to give. 18

The Definition of Head

19

This discussion will support my proposal to identify a head at all levels of syntactic phrases, which will bring me closer to the current interpretation of X-bar theory (see Speas 1990). I thus borrow from X-bar theory the principle of the analysis in constituents that shows the internal consistency of syntactic phrases, and the notion of head applied to all levels of the phrase. In contrast with X-bar theory, I then propose also to integrate morphological structures into the hierarchical analysis and to identify the head, not by applying the criterion of category consistency, but by defining the syntactic relation with its complement. "Complement" in this context denotes the nonhead in syntactic and morphological structures. I am well aware of the difficulty of this task. Numerous linguists of various schools have tried to find a definition of head; yet no one, as far as I know, has found a single, unambiguous definition that can be used for all grammatical structures. Yet, since the identification of the head is the basis of the description in terms of branching, it is necessary to discuss this thorny question. I do not pretend to succeed where numerous others have tried in vain, but I hope to find a definition that has a theoretical basis and that can be applied to the structures I want to analyze.

2.1 Linguistic Typology Despite the attempts of Henri Weil (1848) and Wilhelm Schmidt (1926), word order typology really became a matter of linguistic interest with Joseph H. Greenberg (1963). Yet, the works of Weil and Schmidt already present the germ of both movements in modern typology, that is, Greenberg's approach and the dichotomous approach of W. P. Lehmann and later of Vennemann. In his study of word order Weil indeed proposes a dichotomy, whereas Schmidt, being more cautious perhaps, limits his concern to correlations.

2.1.1 Typological Dichotomy In his monograph on word order, Weil (whom I will quote henceforth from the 1978 edition) examined several Indo-European languages and Turkish, carrying out a contrastive analysis of ancient and modern languages. He made two fundamental observations. First, he noticed two "movements" in the sentence that are distinct in the ancient languages, "the movement of ideas . . . shown by the order of the words, [and] the syntactical movement . . . expressed by terminations" (Weil 1978:36). In modern languages, on the other hand, the order that expresses the ideas "serves as the same time more or less to express the syntactical relations" (Weil 1978:37). In contrast to modern languages, Greek and Latin were thus characterized by free word order (Weil 1978:54), "[which] is a matter of privilege in inflected languages" (Weil 1978:53). Moreover, Weil argued that French, German, English, and Turkish have

20


fixed word order. He based his statement on his analysis of five constructions: adjective—noun, genitive—noun, noun—relative clause, direct object —verb, and adposition—noun. The ordering of these elements was the basis of his second observation, namely, that languages with fixed word order arc characterized by either a descendant or an ascendent order (Weil 1978: 70 et scq.): "[Sjeveral . . . give the governing word the precedence over the governed word; . . . certain others place the former after the latter" (59). Accordingly, long before other linguists, Weil pointed out the regularity in the organization of the phrases he analyzed: He observed the underlying structural principle of sequences in which there is a hierarchical relation. His approach, therefore, is dichotomistic: the elements A and B in a syntactic phrase are organized in the order AB or BA. The type of sequence that dominates determines the typology of a language. More than one century later other linguists, such as Lehmann (1972c) and Vennemann (1974), took up again a dichotomous description in their analysis of word order. (I will return to this at section 2.2.)

2.1.2 Typological Correlations Like Weil, Schmidt confined his analysis of the "Gcsamtheit der jctzt erreichbarcn Sprachen der Erde" (the totality of the now reachable languages of the world) (1926:386) to a limited number of constituents. He took into consideration the following sequences: direct object—verb, adjective-noun, genitive-noun, the formation of the possessive, and the place of the personal pronoun subject. Although he considered the place of the genitive as one of the dominant features in the typology of a language (1926:384), Schmidt proposed a system of correlations. Therefore, he proposed relations of co-occurrence between a given sequence and another —for example, between the place of the genitive, the presence of pre- or postpositions, and the inflectional nature of the language (1926:382). In contrast to Weil, Schmidt did not attempt to find an underlying principle; consequently the regularities he discovered remain diverse in nature. In this perspective of correlations, we can place the implicational universals of Greenberg, who recognizes his debt to Schmidt (1963:83). Greenberg observes that there are languages that consistently "put modifying or limiting elements before these modified or limited, while others just as consistently do the opposite" (1963:60). He argues, however, that the majority of languages arc not that consistent in this respect, for instance, in English the adjective precedes the noun (as in a red car), whereas the direct object follows the verb (he drives a car). This is why he does not propose one fundamental principle but instead establishes correlations between certain sequences; the result is implicational universals. If the number of languages examined by Schmidt largely exceeds Greenberg's corpus, the number of linguistic features taken into account is larger in Greenbcrg's analysis. In spite of these quantitative differences, the work


21

of Greenberg clearly reminds us of Schmidt's both in its principles and in the way the analysis is carried out. This explains a number of striking resemblances, such as Greenberg's Universal II and the relation established by Schmidt between the preceding genitive, the inflected character of the language, and the occurrence of postpositions (Schmidt 1926:382). Although Schmidt claimed the place of the genitive to be a fundamental feature of the word order of a language, he did not infer a general regularity based on this aspect as I pointed out earlier. Greenberg, on the other hand, applies three criteria to determine "the basic order typology" (1963:60) of the languages he analyzed: the occurrence of pre- versus postpositions; the order of subject, verb, and object; and the place of the adjective. According to these criteria, he distinguishes three basic typologies: SVO, SOV, and VSO. The three basic orders differ from the other sequences in that only these "normally occur as dominant orders" (Greenberg 1963:61) and they are more frequent than the three other orderings, OVS, OSV, and VOS. The structural regularities observed by Greenberg show that SOV and VSO typologies are more consistent than SVO; SOV languages, for instance, feature postpositions and VSO languages "are always prepositional" (Greenberg 1963:62; emphasis added). Most of the SVO languages, however, feature prepositions, but a significant number of them have postpositions. If indeed the typology of an SVO language is less strict, its structural organization bears more resemblances to that of a VSO than to that of an SOV language: SVO and VSO languages are both characterized by the sequence verb-direct object. And since the other phrases also show many parallels, one can reduce the tripartite division of Greenberg to a binary distinction: SVO and VSO on the one hand and SOV on the other (see Lehmann 1972c; and section 2.2).

2.1.3 The Problem of Terminology Whether the study of word order aims at correlations or more general principles, it is basically the analysis of sequences of two elements, mostly in juxtaposition, that are in a relation of dependency in a phrase or subphrase. First of all, it is important to determine the exact nature of this relation, that is, whether it is semantic or syntactic. Often phrases are interpreted as being semantic structures. (See, e.g., Schmidt 1926; Keenan [1979], who uses the terms "function" and "argument"; and Vcnnemann [see section 2.2.1 ]). Purely syntactic interpretations, on the other hand, are rare (sec, e.g., Hawkins 1984). More often the nature of the relation is not indicated. For instance, Greenberg, like Weil, uses the terms "modifier"— "modified," "qualifier"-"qualified," and "governing"—"governed" without defining them. Outside the field of typology, attempts have been made to define the elements of a phrase. Yet for the most part these studies also do not offer precise definitions of the binary hierarchical structures (see Bally 1944;

22

The Emergence, and Development of SVO Patterning

Trubctzkoy 1939). Tesniere, for instance, introduced the structural notion of noeud 'node', which is founded on the dichotomous distinction between le regissant 'governor' and le subordonne 'subordinate'. But since he claims that a structural connection is always combined with a semantic one (1959:44), he proposes the concept of "nucleus" to denote the totality of the structural node including its semantic elements. In such a unit semantic connections apparently prevail, since Tesniere stipulates that the subordinate determines the head: "[L]e subordonne determine le regissant. Le subordonne est alors le determinant, ct le regissant est le determine." Thus, in signal vert 'green sign' the meaning of the subordinate affects the meaning of its governor; it is the adjective vert that determines the governor signal (Tesniere 1959:42-43). Yet Tesniere does not define the syntactic elements subordonne and regissant. This example shows that also outside word order typology clear and unambiguous definitions are not provided.

2.2 Typology and Diachronic Linguistics Word order study was spurred on by the work of Lehmann, who pointed out the importance of integrating diachronic data into a typological frame; this approach revealed the regularity of word order changes. Moreover, on the basis of typological studies Lehmann argues that there are indeed "three patterns of order in government constructions," SOV, SVO, and VSO (1972c:267). But since there is a "remarkable parallelism in patterns of order which has been demonstrated in government and in nominal modifier constructions of VO and OV languages," he distinguishes only two typologies, OV and VO, which have to be accounted for (1972c: 268). Along these lines Lehmann then argues that the place of the verb, which is assumed to be the central element of the sentence (1972c:269 et seq.), determines the typology of a language. Therefore, if the direct object precedes the verb, the adjective precedes the noun and the language displays postpositions. The independent correlations of Greenberg thus turn out to be related and even more so determined by the order of the sequence verb—object. Moreover, since the place of the subject in the analyses of word order "has been a source of trouble," it is not taken into consideration. Consequently, the threefold division of Greenberg automatically becomes a binary distinction, OV versus VO (Lehmann 1973: 51). This reduction of course conceals the inconsistency of SVO typology mentioned in section 2.1.2. Lehmann's interpretation marks a return to dichotomous analysis and attempts to account for language change. The regularity pointed out by Lehmann can be summarized as "the principle of placement for modifiers": #Q[ualifier]V

(NObj)

N(mod)#

QV

(Nobj)

(Nmod)

(Nmod)# for VO Ig.

(Nobj)

VQ# for OV Ig. (Lehmann 1973:49)


23

This principle accounts for the fact that "nominal modifiers ... are placed on the side of the modified element that is opposite its primary accompanying entity. Verbs are the primary entities that accompany objects; therefore qualifiers of objects and of nouns in general are placed before nouns in OV languages" (Lehmann 1973:48).

2.2.1 An Attempt at a Definition Like Greenberg, Lehmann uses such terms as "modifier" and "govern" without explicitly providing a definition. Following Lehmann, Vennemann (1974) also carries out a dichotomous analysis and proposes a principle of serialization, specifying that the operator and operand are ordered in the following way: [operator [[operand]

[operand]] operator]

in OV languages in VO languages (1974:347)

He then sets out to define both terms by applying syntactic as well as semantic criteria, a method which weakens his approach. First of all, he considers the constituent to be a semantic unit: "[S]yntactic structures are not co-ordinate constituent structures but rather function-argument structures, as are logical structures" (Vennemann 1974:346). Then, in order to identify the elements of the structure, he relics not only on semantic criteria but also on the syntactic criterion of category consistency: "[T]he defining criterion, in addition to the semantic one of the operator specifying the operand, was found in category consistency: in a constituent structure [AB], A is operator and B is operand if the entire construction [AB] is in the same syntactic category as B" (Vennemann 1974:347). The terms Vennemann uses, as well as the definition and the analysis that he proposes, indeed show that the relation between operator and operand is, first of all, semantic since the occurrence of the operator has a "specifying," hence semantic effect (Vennemann 1974:347); the syntactic criterion is only secondary, as the terminology also shows. The operator in Vennemann's definition is not a syntactic element because its application has no effect on the syntactic category of the constituent. Yet Vennemann needs some kind of syntactic criterion because semantics alone is not adequate. The definition proposed by Vennemann presents another disadvantage —it excludes the concentration of syntactic and semantic functions in one and the same element. Since the preposition determines the syntactic nature of the prepositional phrase, it is assumed to be the operand of the constituent. Consequently, the noun must be operator, the element "specifying the operand" (Vennemann 1974:347). Yet one can also argue, from a semantic point of view, that the preposition specifies the noun. In Vennemann's approach the only syntactic criterion used is category consistency; it supports a semantic definition. Syntactic notions like "gov-

24

The Emergence and, Development of SVO Patterning

ernment" and "case attribution" arc not taken into consideration. Yet the notion of category consistency itself still needs conclusive definition, and the criteria applied to identify the grammatical category of constituents are still ambiguous (see section 2.3.2.2). Consequently, by exclusively applying the criterion of category consistency—which also lacks definition—to identify the head, one may identify an unknown with another unknown.

2.2.2 A New Attempt: The Notion of Branching In the preceding pages I pointed out that the importance of Lehmann's work lies in the integration of typology and diachrony and the return to the dichotomous analysis of word order data. The typological perspective indeed reveals the regularity of the structures of the language under consideration and of the changes these structures underwent. Taking up the dichotomous approach, Vennemann tried to identify the elements of the constituents. This was an important step even if the result was perhaps not entirely convincing. A new attempt to integrate systematically historical data into a typological frame came from Bichakjian (1987), who has proposed a description of diachronic data in terms of branching. The notion of branching, although used now and then by Chomsky (e.g., 1965 or 1986), has never been very popular, perhaps because of the hierarchy it implies (see section 2.3.1). Moreover, it is used without a precise definition (e.g., Chomsky 1965:13). Branching is a syntactic notion that expresses in one word the linear ordering of the hierarchical elements in a grammatical unit. In a leftbranching structure (e.g., exercitum duxit 'he led the army') the subordinate clement—the direct object—precedes the head of the verbal phrase, namely, the verb. In a right-branching structure the order is the reverse, with the subordinate element following the head (as in he led the army). On the basis of diachronic analysis of the structures of the protolanguage, Bichakjian argues that Indo-European, before splitting up, was characterized by a "left-branching set of [grammatical] structures" (1987: 94). From then on, the Indo-European languages underwent a structural reorganization, becoming "stricto senso more or less right branching" (Bichakjian 1987:96). The analysis in terms of branching relates the changes that occurred in syntax to those that took place in inflectional morphology: they turn out to be part of the same overall development. One observes indeed that the principal change in morphology was not the tendency toward analytic forms, but the reversal of the order of elements: the archaic forms, in which the endings followed the stem, have been replaced by structures in which the grammatical element precedes the lexical element. In an inflected nominal form, for example, the lexeme precedes the grammatical marker, whereas in the prepositional—hence right-branching—phrase, the lexical element follows the head. This development parallels the change in syntax.

25


The comparison of Latin archaic and French modern syntactic as well as morphological structures then indeed shows two series of structures in which the order is exactly the reverse: the structures of the type [[direct object] verb] and [[lexeme] ending] of the archaic language gradually gave way to sequences of the type [verb [direct object]] and [grammatical element [stem]]; for example, the Latin structure [[exercitum] duxit], which we find in French as [il conduisit [I'armee]], 'he led the army', and [[popul] o], which became [pour [le peuple]] 'for the people'. This structural regularity can be observed in syntactic as well as morphological structures. Consider the following table containing Latin examples and French equivalents, which shows that the regularity of the structures and of their development emphasizes the need of an analysis in terms of branching. Indeed, this type of analysis not only accounts for the structural parallel between syntactic and morphological units but also reveals the shift that underlies their development and that passed unnoticed: the replacement of the archaic left-branching grammatical structures with their modern rightbranching equivalents. Latin Archaic Left-Branching Grammatical Structures and Their French Right-Branching Equivalents

Phrases object

verb

object

exercitum duxit 'he led the army'

verb

il conduisit

I' armee

PP

verb

PP

in partes discindere 'to divide in parts'

verb

diviser

en parties

adverb

verb

adverb

leniter ridere 'to smile kindly'

rire

doucement

adjective

copula

adjective

avidus est 'he is greedy'

il est

avide

noun me 'with me'

adposition -cum

preposition avec

noun moi

temporis

causa

a cause

du temps

verb

copula

'because of the time' (continued)

26

The Emergence and Development of SVO Patterning Phrases (continued) adjective

noun

noun

adjective

longissimus truncus 'a very long trunk'

un tronc

tres allonge

genitive

noun

genitive

le present

des dieux

noun

deorum munus 'the present of the Gods'

adjective

noun

pecuniae avidus 'greedy for money'

noun

adjective

avide

d'argent

referent comparative Paulo grandior 'bigger than Paul'

comparative plus grand

referent que Paul

Inflected Forms adjective grand 'bigger'

degree -ior

degree plus

adjective grand

noun

ending

preposition

noun

leg 'with laws'

-ibus

avec

des lots

verb ending am -av-erit 'he will have loved'

aux/pronoun il aura

verb aime

noun conjunction populus -que 'and the people'

conjunction et

noun le peuple

2.3 The Definition of Head Whereas the notion of branching in itself is clear and easy to define, the two elements that are combined in a hierarchical structure are not: the governor and the governee, or the head and its complement. As a first step Bichakjian carefully avoided the problem by stating that "[since] the preposition is the head of a prepositional phrase, then the degree adverb and the case markers must be the head of a comparative adjective and of an


27

inflected noun, respectively" (1986:17). This observation is merely a motivated analogy, based on a parallel and on the regularity of changes. Yet whereas indeed the branching hypothesis might be useful in the analysis of syntactic and morphological structures, conclusive definitions are still lacking and the identification of the head might therefore become highly questionable. It is legitimate to ask, for instance, what argument supports the identification of -ior as head in Latin dulcior 'sweeter', or the identification of -e as head in the French adjective mauvaise 'bad'. Although the grammatical relation between a direct object and the verb is quite clear, the situation is less obvious in inflected forms. The table clearly shows, however, that the development of French is characterized by a reversal of the order in syntactic as well as morphological structures. This change supports the extension of the analysis in terms of branching to morphological units. Since the notions of head and complement are basic to this conception of hierarchical structures, it is of primary importance to define them correctly. In order to find an adequate criterion of identification, I will take up a number of concepts of X-bar grammar, although this linguistic approach does not provide the theoretical framework for the analysis I carry out. First I briefly discuss the principles of X-bar theory (section 2.3.1); then I examine a number of propositions that have been made in the discussion of the definition of head (section 2.3.2). Although not all these attempts at definition have been made in the X-bar framework, they all go back to it in one way or another. I examine them in order to restrict the number of possible solutions and to show that independently motivated criteria to define the head are still lacking.

2.3.1 X-Bar Theory X-bar theory is an approach in generative grammar that is basically a phrase structure analysis. It is based on the assumption that there is a structural parallel between syntactic constituents such as N[oun]P[hrases], P[repositional]P[hrases], A[djectival]P[hrases], and V[erb]P[hrases]. Moreover, in contrast with other generative theories, X-bar recognizes a hierarchical syntactic head in each phrase, being N[oun], P[reposition], A[djective], and V[erb], respectively, which determines the grammatical category of the unit. In addition, each phrase has several levels (e.g., N'", N", N', and N), which accounts for the subconstituents and their relation to the head; see, for example, (an) excellent book versus (an) excellent book on syntax by Delbruck, which needs additional syntactic levels. Each level has its specific subordinate elements (called complements and specifiers); the interpretation of the syntactic status of these elements has changed in time, as we will see shortly (section 2.3.3). X-bar theory goes back to the generative discussion on nominalization in English: problems in the description of derived nouns (e.g., John's refusal of the offer) and gerunds (e.g., John's refusing the offer) affected

28


Chomsky's linguistic conceptions and those of his followers (sec Chomsky 1972). The definite nominal nature of the derived noun, the irregularity of its formation, and occasionally its semantic peculiarity (e.g., hold-holding 'investment') exclude a transformational approach (Chomsky 1972: 15-21). The alternative, lexicalist solution is to generate these nouns in the lexicon and to extend the categorial component. This extension is, however, compensated for in another part of grammar. The gerund and the derived noun are indeed too different to be generated by transformational rules, but the lexical entries—that is, the major categories V, A, and N— "share the same distributional properties" (Chomsky 1972:34). Consequently, instead of claiming a derivational relation between the verb and the derived noun, Chomsky pointed out a structural similarity between the major categories, generalizing the rules: X"

(Spec X) X'

X'

X Comp

where X, referring to the major category—be it A, N, V, or P—represents the head. This diagram "is claimed to constitute a linguistically significant generalization of the structures associated with major categories" (Jackendoff 1977:17). Chomsky set forth this new approach in linguistics, but it was Jackcndoff who really elaborated the concept of X-bar and who came up with a phrase structure analysis. He extended the number of bars from two to three, to account, for instance, for the distinction between restrictive and appositive relative clauses, for their differences in meaning and in movement possibilities. Moreover, Jackcndoff pointed out that each hierarchical level has its own specifiers and complements: "[F]unctional arguments, restrictive modifiers, and nonrestrictive modifiers, [for instance] . . . can be identified with X', X", and X'" complements, respectively" (1977:85). But whereas the complement is represented at all levels, the specifier only comes back under X'" and X". Therefore, one can redefine the structures under consideration: X'" X" X'

Spec X" - X" - Compl. Spec X' - X' - Compl. X - Compl.

After this first elaboration generativists tried diligently to integrate this theory into their linguistic frame and to account for all its consequences. Most of the discussion was about the number of bars to attribute to the various elements of the constituents, about the structure of these constituents, and about the application of generative rules. These questions, how-


29

ever, do not affect the analysis that is planned for this book, the more so since the generative discussion was mostly focused on English. Other aspects to which generative syntacticians paid less attention are more important in this respect. X-bar theory revealed in a consistent way the structural parallel of syntactic phrases; this is its most basic principle. Moreover, it introduced in generative grammar the notion of hierarchical head, which appears not only in constituents, but also in longer structures. Hierarchy is not a characteristic notion of generative grammar. This explains perhaps why the definition of head has never been analyzed in detail. Generative theory represents phrases in tree structures without, however, indicating the internal hierarchy: generative analyses point out the degree of relatedness between the elements of a phrase. This is the main objective of these analyses, because degree of relatedness is the basic factor in the transformations that structures undergo. X-bar theory introduced the notion of head in generative grammar; later, principles of theta-marking provided the notion of head with a real function. Yet, the concept is used without being defined. Moreover, the notion of head seems not always to be related to dependency. Recently still, the phrase pictures of John has been interpreted by Chomsky as [NP [N pictures] [of John]] (1986a:4). This representation does not show the hierarchical relationship. It just indicates the nominal nature of the phrase, which we find also in pictures. Consequently, this element is assumed to be the head of the structure, (see discussion of the criterion of category consistency in section 2.3.2.2.)

2.3.2 The Identification of Head: Attempts Within the Framework of X-Bar Theory X-bar theory triggered the discussion on the identification of the head not only in syntax but also in derivational morphology. For reasons that remain unclear, inflectional morphology did not receive the attention it deserved. In order better to understand the problems related to the identification of the grammatical head, I briefly discuss the principal definitions that have been proposed since X-bar theory (for supplementary criteria and definitions, see Zwicky [1985] and Hudson [1987]). This review may show the complexity of the aim of this chapter—defining head—and, at the same time, reduce the number of solutions one might possibly consider. 2.3.2.1 Underlying Structure When identifying the head of a syntactic phrase linguists often refer to the underlying structure. In the structure the author of the book, for example, Chomsky identifies author as "head noun" (1972:30), explaining his choice by referring to the parallel—which he eventually rejects for theoretical reasons—with the underlying abstract structure: "the one who *auths

30


the book, *auths being postulated as a verb that is lexically marked as obligatorily subject to nominalization" (1972:32). Yet the underlying structure seems to be a weak criterion for the identification of head. In the following sentence, for instance, several of John's proofs of the theorem

John and proof are considered to be heads because "the same selectional feature that associates subject and verb in John proved the theorem will relate these two items" (Chomsky 1972:38). The comparison of two totally different structures, however, may jeopardize the solidity of the analysis. An analysis that is based on the underlying structure will suggest that proof is head in John's proof: the nominal structure indeed corresponds to "John proved . . . " This interpretation fits perfectly well with the generative hypothesis that the verb is the head of the clause. Since the subject is specifier under V"', the parallel with the genitive, which is specifier under N'", is perfect. But the analogy ends when one tries to account for a genitivus objectivus: being a direct object in the deep structure, this element is complement under V and is placed to its right (cf. John's murderer, which goes back to the man who murdereAJohri). The X-bar description of the genitive, however, does not account for the underlying distinction between the genitivus objectivus and the genitivus subjectivus. Finally, numerous phrases do not have clear correspondences in deep structure, which is a practical disadvantage of this criterion. 2.3.2.2 Category Consistency In his structural analysis of words, Williams compared the definition of head in morphology—where this topic is widely discussed (see, e.g., Selkirk 1981)—to the one in syntax "where the head of a phrase is immediately identifiable: it is the subphrase having all the same category features, but one bar fewer" (Williams 1981:248). In syntax the problem apparently does not exist because of the criterion of category consistency. This criterion is thought to be "the syntactic generalization and criterion for distinguishing modifiers from,heads in a consistent way" (Hawkins 1984:122). As long as one is dealing with major categories and their complements the situation is indeed clear and the identification features proposed by Chomsky arc adequate, that is [ ± N] and [ ± V]:


31

Jackendoff added minor categories, the adverb, the particle, the article, and quantifiers. Likewise, he changed the features in [ Subj.], [ Obj.], [ Comp.], and [ Det.]. Yet lexical categories remain the basis of his interpretation and the internal structure of the phrase is not taken into consideration. Consequently, problems arise when one has to identify the head in sequences like the French il a loue I la revue qu'il a feuilletee 'he praised'/ 'the journal he glanced through', or the English he has praised, where the elements share the same grammatical category. Generative grammar does not provide syntactic features that account for subphrases of constituents, as in the preceding examples. One can, of course, propose more detailed features, as did Gazdar and Pullum (1981: 112), who, in order to account for the internal structure of a verbal phrase with a direct object, introduced the feature [+finite] in their analysis of the VP:

where F represents [finite]. This approach carries the risk of circularity because feature correspondence is used to define the head, and, at the same time, the head is referred to as the element that determines the feature correspondence. Moreover, the problem of the auxiliary is not solved: this element features [ + F], just as the phrase it is part of, but it is only the specifier according to X-bar theory. Furthermore, the criterion of category consistency is not of much use in inflectional languages, where the noun phrase, for instance, can display more than one noun: in the Latin constituent patris amor 'father's love' there arc two nouns; consequently the categorial nature cannot be the criterion of identification of the head. Moreover, one cannot say for certain that category consistency can be applied to identify all heads possible. Comparatives and superlatives, on the other hand, show that the criterion of category consistency interferes with other syntactic considerations (see section 2.3.3.1). Some linguists combine the criterion of category consistency with other criteria, such as locality (sec section 2.3.2.3). Williams, for instance, observes a parallel between syntax, where "the head determines the properties of the whole (1981:249), and morphology, where the suffixes determine "the category of the word they are part of" (1981:249). Likewise, the right-hand member in a derived structure determines the category of the whole. Williams, however, docs not specify the criteria that determine the category. Even if the parallel between syntactic and morphological structures is correct, categorial identity then remains a weak criterion. These reservations are supported by the complication observed by Williams: nominal formations based on two non-nouns, such as a push up,

32


which is a noun composed of a verb and a preposition ([push V up P] N), cause serious problems to those who advocate the criterion of category consistency (1981:250). For reasons unknown to me, the relation that is established between syntax and morphology, in the discussion of head, is generally restricted to derivational morphology. The comparison of syntactic and inflected structures is an issue that is seldom discussed. (See, for instance, L. Bauer [1990], who does not exclude the possibility of there being a relation, or Hawkins and Gilligan [1988].) Hawkins and Gilligan argue that the ending is the head of the inflected form. They also support their view with the argument of category consistency, which they also apply in syntax (compare Williams's similar approach of derivational morphology). In derivational as well as inflectional morphology, they argue, there arc a limited number of suffixes; their nature is fixed and it determines the category of the inflected form: [T]he categorial status of a word containing affixes can regularly be computed from the affix, whereas non-affixes or stems will very often have their categorial status changed through the addition of a (derivational) affix. It is therefore feasible to assume . . . that derivational and inflectional affixes are always the heads of their respective lexical categories. (1988:226) Hawkins and Gilligan then propose a parallel with the place of the syntactic head in their Head Ordering Principle (1988:227). Though it seems correct to me to identify the ending as the head of the inflected form, the status of the derivational affix is not certain. By attributing the status of head to both the derivational and the inflectional suffix, Hawkins and Gilligan compare two linguistic processes that are totally different—derivation, which is, a lexical process, and inflection, which is a grammatical process. Apart from the practical problems mentioned, the present discussion also shows that before applying the criterion of category consistency, one first has to have a conclusive definition of the concept that is unambiguous and at the same time easy to apply. This definition is still lacking.

2.3.2.3 The Head as Obligatory Element and the Place of Occurrence The obligatory presence and place of occurrence are two other criteria that are frequently discussed. It is argued that the element that is obligatory is ipso facto the head of the structure (see, Hudson [1987] and Robinson [1970], whose analysis has been discussed by Owens 1984). This criterion, however, is only adequate if related to the two distinct concepts of ellipsis and optionality. In his discussion of the criterion of "obligatory presence," Hudson argues, on the basis of English, that the auxiliary must be the head of the sequence [auxiliary + VP]: the auxiliary can replace the entire phrase (which is the argument of distributional equivalence), and in this same


33

sequence it is VP and not the auxiliary that can be omitted in specific contexts (1987:118-19). Although the identification of the auxiliary as head is interesting (sec section 2.3.3.3), the criteria that have been suggested are perhaps not very adequate. The phenomenon of gapping or of ellipsis is based on principles that vary from language to language. This explains why Hudson's arguments are considerably weakened when one compares his examples with their French equivalents: He will control those penguins. He will. II va controler ces pingouins. Il va.

or: Il veut controler ces pingouins. Il veut. 'He wants to control those penguins. He wants'. Pierre a mange des escargots; Paul n'a pas. 'Peter has eaten snails; Paul has not'.

In addition to the obligatory presence, the place of occurrence is frequently mentioned as an important criterion. Chomsky, for instance, referred to it when he presented the specifier as being "the material in a phrase to the left of the head" (Jackendoff 1977:37). In derivational morphology, in the field of word formation, this criterion underlies the Right-Hand Rule: "In [derivational] morphology, we define the head of a morphologically complex word to be the right-hand member of that word" (Williams 1981:248; see also section 2.3.2.2). Yet the locality criterion presents a number of disadvantages. First, it is not independently motivated. Second, it is a source of considerable trouble when one has to account for a mixed typology or for a typology different from that of English. Finally, it reduces the syntactic phrase to a linear sequence of elements without taking into consideration the internal structural relation. 2.3.2.4 The Function of Head The function of head might provide a criterion of identification. But just as an independently motivated definition of head is still lacking, the function of head has not been explicitly identified. Moreover, the current discussions only deal with the heads of syntactic units. This consideration is an argument against the criterion of the morphosyntactic locus, which specifies that the head "can be the bearer of the morphosyntactic marks of syntactic relations between the construct and other syntactic units" (Zwicky 1985:6). This definition cannot be used in inflectional morphology because the ending itself is the morphosyntactic marker. Therefore, the criterion of the morphosyntactic locus is not adequate for the analysis I want to carry out. Generativists point out that the head of a phrase is the clement that

34


assigns theta-roles and that determines the nature of the specifier and the complement. The function of the head is clearest in the relation between the head and the complement: the head governs its complement; its categorial features determine the nature and the existence of the complement. For elements other than the complement under X' the situation is, however, more difficult and depends largely on the structure itself. One can argue of course that the actual realization of the specifier X" depends on the nature of X: if X is a noun, specifier X" is an adjective; if X is an adjective, specifier X" is an adverb. According to the principle of subcategorization the adjective is thus the head of a structure featuring a degree: in more expensive, expensive is the head. (I will return to this interpretation in section 2.3.3.1.) An adequate functional definition might be a good instrument for identifying the head of a grammatical unit. Yet it is difficult to find the function common to all the heads possible, in syntactic as well as morphological structures. Most often the functions that are dealt with in the discussion are limited to a restricted group of elements—elements that distribute theta-roles, for instance—and exclude all the other elements that have, however, a superior hierarchical function. 2.3.2.5 The Lacking Definition The preceding discussion has shown that an independently motivated criterion of head is still lacking in generative grammar and in linguistics generally. It is remarkable, moreover, that the notion of branching is not taken into consideration in the discussion of head, and that, on the other hand, we find neither the notion of head nor the notion of dependency in the discussion on the concept of branching (see Chomsky 1965, 1986b). The absence of adequate syntactic criteria has led some linguists to omit the notion of head and modifier and to speak only of a dependency relation without including the notion of head as a superior element: "The clustering of items [is explained] by saying that items in a dependency relation cluster together" (Owens 1984:34). In this cluster "the head is the item which contracts relations with all the other items, i.e. the head is the item with the most general distribution" (Owens 1984:36). According to this assumption, the noun, which combines with the adjective, the determiner, the number, and so forth, is supposed to be the head of a noun phrase; likewise, the verb as head of the VP combines with the auxiliary, the direct object, the adverb, and so forth. Since this approach takes only the internal combinatory possibilities into consideration, the syntactic context of the phrase is not considered. Moreover, the application of this criterion is difficult when one has to deal with a preposition, which combines only with a nominal element. Furthermore, there is a contradiction in arguing that the elements of a phrase are in a dependency relation and in claiming at the same time that there is no hierarchically superior element in that phrase. Finally, grammatical structures are definitely hierarchical. Typological studies of grammatical units support these considerations, espe-


35

daily those that arc more than a mere enumeration and that attempt to reveal structural regularities of languages. Therefore, it is necessary to return to the original interpretations of X-bar theory, which differs from other approaches in generative grammar in that it provides an analysis of hierarchical structures.

2.3.3 The Definition of Elements According to X-Bar Theory X-bar theory presents a number of advantages that make it interesting for the analysis I want to carry out: the assumption that phrases present a structural parallel and the notion of head. For the analysis in this book it is important to recognize a head at each level of a grammatical unit. Once this structural principle is established, I can set out to define this head. These two steps are needed to account for the analysis in terms of branching that then shows, in a consistent way, the hierarchical relations as well as the linear change that has occurred in Latin dependency structures, be they syntactic or morphological. In the next part of this chapter I refer to the "classical" X-bar theory of Jackendoff (1977) and to the current model (see, e.g., Speas 1990). In section 2.3.3.1 I briefly discuss the relation between the specifier Xn and the complement Xn, as defined in the traditional variant of X-bar theory, showing that the binary approach of the current model was already manifest in the early stages of X-bar theory. The binary approach can be supported, for instance, with analysis of the comparative construction. Recognizing a head at all levels of a syntactic structure, I then discuss the criteria to identify it in an attempt to relate syntactic and morphological structures.

2.3.3.1 The Distinction Between Specifier X" and Complement X" Traditionally X-bar theory distinguishes three elements in a phrase, which are in a linear order from left to right: the specifier, the major category— the head—and the complement. The complement X is the easiest element to define, because its features correspond to the syntactic predictions of the lexical entry of the head. The head of the structure is the major category, which can be a basic lexical category, and the syntactic nature of which is present at all levels of the structures, in X'", X", and X'. At first Chomsky defined the specifier as "the phrase associated with N', A', V in the base structure" (1972:53); hence the following rules: X" X'

Spec X' X Comp

where Spec N' is Dctferminer], Spec A' is Dcgfrec], and Spec V is Auxiliary].

36

The Emergence and. Development of SVO Patterning

If at first the definition of specifier was syntactic—the place where the element in question was linked to the syntactic phrase was taken into consideration—it became purely distributional afterward: "Chomsky uses the term specifier to refer to the material in a phrase to the left of the head, and complement to refer to the material to the right of the head" (Jackendoff 1977:37). In this context it is important to specify that the phrase structure rules of X-bar theory have a fixed order. In Chomsky's definition the place of occurrence determines the syntactic value of the element under consideration; this explains the problems in languages with inconsistent typologies, such as English (see Jackendoff 1977:84 et seq.). Moreover, a description that only refers to the place of occurrence reduces a phrase to a mere linear sequence, the head of which is the central element. Consequently the three notions that represent a phrase are but distributional concepts. Earlier in this chapter I mentioned Jackendoff, who extended the occurrence of the specifier and the complement to other syntactic levels: the specifier is represented under X'" and X", whereas we find the complement at all three levels of the phrase. This extension had important consequences. With Jackendoff the notions of specifier and complement indeed lost syntactic distinction and status. He does not consider them as constituents: "[T]hey do not move or delete as units, and unlike normal constituents, no part can be designated as a head" (1977:37). Since a definition based on the place of occurrence will cause typological problems (compare, for'instance, equivalent structures in SOV and SVO languages), "the distinction between specifier and complement is to be regarded here as of no theoretical significance, but only as a convenience" (Jackendoff 1977: 37). Jackendoff did not specify the scope of this consideration: whether it affects only the specifiers and the complements at one and the same level or also affects elements at different syntactic levels. JackcndofFs observation is indeed well founded when one considers the elements that occur at the same level: the specifier and the complement under X'" and their equivalents under X". There is, however, a definite distinction between specifier and complement of different syntactic levels. Compare, for instance, a specifier under X" and a complement under X n-1 : (cm) interesting book about linguistics, where interesting is specifier under N", whereas the prepositional phrase is complement under N'. Second, in line with JackcndofPs analysis the specifier differs from the complement in that it is excluded from a certain number of structures, such as N': "[Since] N' is reserved for strictly subcategorized arguments, we are left with two NP specifier positions" (1977:104). Therefore, the use of the notions of complement and specifier is more than simply a matter of convenience as soon as these elements occur at two different syntactic levels. This observation is supported by the plain fact that it is correct to assume—on the basis of ambiguity and movement transformations —different syntactic levels in syntactic phrases. The distinction between complement and specifier is not important when both elements occur at the same level, because it is merely distributional.


37

According to its branching, a language will have more complements —which occur to the right of the head—or specifiers—which occur to the left of the head. French, for instance, will have more complements than English, which has more left-branching structures. Consequently, according to this conception of X-bar theory, an element can be specifier in English but complement in French. Therefore, it is legitimate to do away with a distinction that has no definite independent syntactic value, and to limit the analysis to the notion of complement, the subordinate nature of which is clear. By removing the distinction between specifier and complement, one accepts a binary structure, where each node has a head and a subordinate element:

We find this structure in the current interpretation of X-bar theory, where 1. Every phrase is endocentric. 2. There are three bar levels: X", X', and X. 3. The head term is one bar-level lower than the immediately dominating phrasal node (Speas 1990:37) This description not only specifies that every phrase has a head; it explicitly states that at each level of the tree structure there is an element that is syntactically superior. The former trichotomy of X-bar theory, that made an artificial distinction between specifier and complement (see tree diagram A) has been replaced with a binary analysis. This organization can be shown in tree structure (B), which replaced the traditional representation (A):

38


(According to Spcas [1990:18]: IP represents 'inflectional phrase' and DP 'determiner phrase'.) In this structure I is the former Inflection and the subject is generated in DP under VP. This organization has variants, for instance, the one where the subject in the surface structure is sister of I' (see Speas 1990:17 et seq., 77 et seq.). These interpretations may be important in the theoretical perspective of X-bar, but they do not affect the basic principle of binary analysis; therefore, there is no need to go into that discussion. The importance of a binary distinction on ever)' level is particularly clear in longer structures, for instance, the comparative construction. According to Jackendoff's interpretation, the comparative had the following structure:

(1) more beautiful (2) two feet long (3) bigger than Paul (Spec A" A"; A' Compl. A') This representation raises several questions. First, as a specifier under A'" degree occupies a high position in the tree structure. Yet, according to X-bar theory, hence according to the criterion of category consistency, degree is only a specifier. Therefore, this element allegedly docs not play an important syntactic role; moreover, its place in the tree structure isolates


39

this element from the rest of the structure. On the other hand, the term of reference is Compl. A' under A", which corresponds to the hierarchical dominance of the comparative adjective. The term of reference indeed exists only because of degree; therefore, degree should have a more important syntactic role. According to the criterion of category consistency, however, the adjective is considered to be the head of this structure, because the phrase is adjectival. Yet, the term of reference is not governed by the adjective proper, but by the adjective in the comparative form. Therefore, a binary structure would better represent the syntactic reality because it combines the adjective proper and the degree marker in one subphrase; this structure then governs the term of reference. For instance, more interested in languages than his brother:

where one distinguishes a head on each level: [more interested [than his brother]]. (In section 2.3.3.3 I will come back to this structure and attempt to identify its subphrases.) The comparative structure thus shows that binary distinction better represents the internal coherence of the phrase and its internal hierarchy. Moreover, the example of the comparative structure shows that category consistency is not the well-founded criterion it is often said to be. Binary distinctions can better account for the syntactic coherence, and the distinction of the head at each level better fits the description of syntactic principles: dependency, which sui generis is a binary relation, and internal coherence of the grammatical phrase, as in the example of the comparative more intelligent than Paul, where the unit more intelligent governs than Paul, which can be analyzed in [than [Paul]]; hence [[more [intelligent}] [[than [Paul]]]]. I wish to specify, however, that I am not proposing the so-called principle of binary branching. According to this principle, there are only two elements at each syntactic level, which excludes the combination of a head with more than one complement.

40


Although I accept the principle of binary structuration that we find in the current interpretation of X-bar theory, I cannot accept the criterion of category consistency that X-bar theory then proposes in order to identify the head of the entire phrase. Category consistency cannot be the criterion of identification because grammatical category is an intrinsic quality. The criterion of identification must be based on the relation between head and its complement and must take into account the syntactic context of the structure under consideration. In the following pages I first point out a theoretical augment against the criterion of category consistency (section 2.3.3.2); I then attempt to formulate a criterion that can be used to identify the head of syntactic as well as morphological structures (section 2.3.3.3). Finally, I briefly discuss the role of subject in phrase analyses (section 2.3.3.4).

2.3.3.2 The Place of the Head in the Tree Structure X-bar theory presents category consistency as the criterion to identify the head of a phrase: the complement is the sister of the lexical element X that is the head of the phrase. Elements that have a different syntactic nature play an inferior role. Yet principles in generative grammar itself seem to interfere with this hierarchical conception of the phrase. I refer, for instance, to the principle of binding. If a pronoun is not supposed to C-command its antecedent, then the clement B (sec tree structure [1]) is hierarchically superior to the clement D and E:


41

(where i means coreferential with) (2) * hei greets Paul's' mother (3) Pauli greets hisi mother Therefore, according to the principle of binding, the complement of the head of the phrase is conditioned by an element other than the head, which thus has an inferior syntactic status. In the classical version of X-bar theory the situation was even more complicated. Stating that the lexical element is the head of the phrase, the theory implied that the head occupied a very low position in the tree structure. The syntactic inferiority of this clement is confirmed by other principles of the theory. One can argue of course that the head itself is not concerned—the principle of C-command affects only the relation between a specifier and a complement—but this is not the issue. X-bar theory attributes an inferior syntactic role to an element that in accordance with other principles of generative grammar plays a major role, in casu the principle of C-command. This argument applies to both variants of X-bar theory. It is indeed strange that an element that occupies a high position in the tree structure is only a specifier in the classical variant of X-bar theory and a non-head in the current interpretation. These interpretations are related to the importance attributed to the criterion of category consistency. Consequently, instead of this criterion I will attempt to identify the head on all levels of a phrase, by applying a criterion that can be used in morphology and syntax. 2.3.3.3 The Head of Morphological and Syntactic Structures: Analysis in Terms of Branching Once one accepts the idea that the head is represented at all structural levels, one can come back to the comparison of syntactic and morphological structures (see the scheme in section 2.2.2). As a first step, analog)' might be a criterion to identify the head of grammatical units. Yet it is not very convincing. It is clear that in a phrase like exercitum duxit, exercitum is governed by the verb that assigns its case. Since all other syntactic phrases have their head on the same (right) side, analogy then points out that the element on the right side in the inflected forms is also head. This parallel is supported by the consistent distribution of the morphological elements. In the given inflected examples one indeed observes that the elements that have lexical meaning—dur-, leg-Hois, am-1 aim, populusl peuple—are consistently placed on one side, whereas the endings or particles are placed on the other side. Moreover, in time the order of these elements has been reversed in a consistent way (see also Baldinger 1968). Even if the list of examples presented in section 2.2.2 and the inventories proposed by typologists or by Baldinger are merely empirical, they are based on observation and not merely analogy. Consequently, these are the

42

The Emergence and Development ofSVO Patterning

basis for the structural analysis and one can set out to find the common characteristic shared by all categories and all syntactic functions placed on the one side, and the characteristic that is common to all categories and functions placed on the other side. The relation of head and complement in syntactic phrases is easier to define than its equivalent in morphological structures. There is a general consensus to consider the verb the head of the direct object that it governs and to which it assigns case. Likewise the noun governs the adjective because it determines its gender, number, and case. Yet, although these syntactic considerations are all correct by themselves, one can legitimately argue that they are contingent, as in examples (1) he buys a car and (2) a, red car. In the first sentence a car is direct object, hence a verbal complement, whereas the epithet red indicates a quality in (2). Adjectival qualification and verbal complementation are both syntactic functions, but they express different relations. This observation can be made for all phrases. Consequently, it is important to find the characteristic that is shared by the verbal complement, the epithet, and the other relations expressed in hierarchical structures, be they syntactic or morphological. Therefore, in order to account for all these hierarchical relations of a different nature one needs a broad definition. Before proceeding to morphological units, I will first attempt to capture the common feature of syntactic phrases. In the constituent provinciam nostrum ab Helvetiis dividit (Caes. DBG 1.2.3) 'it separates our province from the Helvetians', the verb assigns a syntactic function to the direct object and the prepositional phrase that owe their existence to it. In a nominal phrase the situation is comparable: according to the criterion of movement transformations and internal coherence one can distinguish several subphrases:

On each level the complement owes its existence and its syntactic function to the head. The subphrase by Delbriick cannot exist on its own because it would be disconnected, and it would not have any syntactic relation with any element in the sentence. I have already discussed the example of the comparative, where the


43

comparative form—and not the adjective proper—determines the existence and the grammatical status of the term of reference:

Likewise in a prepositional phrase that is governed by a verb, a noun, an adjective, or an adverb, the preposition determines the syntactic status of its complement. In all hierarchical structures it is the head that assigns the syntactic function to the other elements of the structure (see Bichakjian 1991). When the head of a phrase is omitted, the complements lose their syntactic function, hence the reason for their existence: the syntactic coherence of the phrase is lost, as well as the syntactic coherence between the phrase and the other elements of the sentence. I do not propose a hypothesis according to which the head is some sort of a "bridge" (see Tesniere 1959), because this would interfere with the analysis in subphrases proposed earlier. It is clear, however, that in the phrase red car, car is the superior element, which assigns a syntactic function to its complement and determines the grammatical value of the phrase of which it is the head. If indeed the head is omitted, the syntactic coherence disappears. Consequently the head is always present, either in its lexical form or in the form of a trace. Just as the head of a syntactic phrase does, the morphological ending also indicates the function of the word in the sentence. The suffix -ibus, for instance in canibus 'to the dogs', shows that the noun is in the dative or the ablative plural: the case ending expresses the grammatical relation between its phrase and the superior head (see Bauer 1987). Likewise, the verbal suffix that indicates person expresses agreement with the subject of the sentence. In these two examples the grammatical values are externally determined: they come from an element that has a superior syntactic status and that is outside the structure under consideration. Yet there are also internal grammatical values (e.g., degree or aspect). In urbemgrandiorem, for instance, the case ending, -em, expresses adjectival agreement with the noun, whereas -ior is a grammatical value that does not depend on external elements. The case ending thus expresses an external grammatical relation,

44


but the Latin degree marker -ioror plus in the French plus grand expresses an internal grammatical value. The problem of the determiner is comparable: in [be buys [a- car]] the indefinite article docs not express an external relation as does the case ending, which indicates the syntactic function of the word and thus expresses an external relation. Yet it is clear that the determiner expresses grammatical values (± definite, ± singular, etc.) of the lexical element it is combined with. The article in German indicates the inherent grammatical value of the noun, gender, as well as grammatical case. This is why it is the head of the sequence article-noun. The auxiliary also expresses two types of grammatical relation: the personal pronoun or the personal suffix expresses agreement with the subject, whereas the auxiliary also determines internal grammatical values, such as aspect, tense, voice, or mood. The distinction I observe in morphological structures is comparable to the problem that arose in the analysis of syntactic phrases: there is a hierarchical relation between the head and the complement, but the actual realization of this grammatical relation varies. Yet what comes back every time in these morphological structures is the grammatical element that indicates and determines the grammatical values of the lexeme. And if one changes this grammatical element, the grammatical status of the morphological structure is affected. On the other hand, in a given syntactic context, one can change the lexical clement without affecting the grammatical relations: whether one says \\consili]o] utor advice-Abl. use-I; 'I use the advice', or [[nlenti]o] utor silence-Abl. use-I 'I use the silence', the syntactic structure does not change (see Garde 1977; Bauer 1987). In the preceding discussion I have attempted to show the hierarchical regularity that underlies the syntactic and morphological structures and that a binary analysis can account for. Yet the hierarchy expresses various relations, and this characteristic stresses the need of a broad definition of head. The definition that Bichakjian has proposed seems rather broad and accounts for a large number of relations expressed by syntactic and morphological structures: "[SJince the head is that item of a grammatical structure that governs the other itcm(s), I suggest Misgoverning be defined as assigning an item its syntactic function or determining its grammatical value'" (1991: 194). Yet in order to account for elements that indicate internal grammatical features, we also have to refer to head as the element that indicates the grammatical value of the unit. In a more detailed analysis one can then determine the expressed relations as I have in the preceding paragraphs. Yet the basis of the syntactic analysis that may identify the head is summarized in these two characteristics: the head assigns a syntactic function to its complement and it indicates, hence expresses, the grammatical value of this element. 2.3.3.4 The Subject and Its Function in the Sentence One element remains to be discussed, the subject, which in Jackendoff's tree structure occupied a high position, that of specifier V". The generative


45

assumption that the verb is the head of the clause (hence V") has a primarily theoretical basis. As I have attempted to show, generative grammar itself carries principles (the principle of C-command, for example) that raise serious questions about the interpretation of highly placed elements under X'"—determiner, auxiliary, and subject—as simple specifiers. Instead, I think there are convincing arguments in favor of the hypothesis that in a nominative language the subject—and not the verb—is the head of the clause, such as (1) the subject—agent as the topicalized item and focus of the language; (2) the subject as the unmarked constituent of the clause and—in contrast with the verb—governed by no other element; and (3) subject—verb agreement. Yet I am well aware of the complex problems related to the matter of subject once we are dealing with crgative and even more so active languages, problems that are also relevant to Indo-European linguistics. Several years ago research into the language system of Pre— Indo-European was given a fresh impetus with the work of Gamkrelidze and Ivanov (1984) and Lehmann (1989a, 1989b, 1993). On the basis of features that we find in early Indo-European languages and that do not fit the pattern of a nominative language, Pre--Indo-European has been reconstructed as a non-nominative, presumably active, language. Consequently, it did not feature a subject-agent. Since Proto-Indo-European originally was a non-nominative and also a left-branching language, the discussion of branching and the definition of head are independent of the notion of subject. Consequently, the analysis of the definition of head can be carried out without analyzing in detail the status of the subject.

2.4 The Analysis in Terms of Branching Although X-bar theory is not the main theoretical framework of this book, it presents several aspects that are important for the analysis in terms of branching: the notions of head and structural hierarchy. It recognizes the structural regularity of phrases that is based on the notion of head. Since linguistic reality does not match the distinction between specifier and complement at the same level, it is more accurate to describe syntactic .phrases in terms of binary structures distinguishing a head and a complement at each level. This then results in a description of syntactic structures in terms of branching, which is a linear as well as a hierarchical notion. In X-bar theory the criterion of category consistency is commonly applied to identify the head. In its classical variant, this was the major category, and it determined the phrase; later functional categories were added. Category consistency is, however, a criterion that is difficult to define and to use and that can neither account for all syntactic relations of a phrase nor morphological structures. Yet inflected forms are an important aspect of the typology of certain languages and they can be accounted for with the same hierarchical analysis in terms of branching. Once one recognizes the necessity to identify a

46


head at each level of a phrase, it is only a minor step to relate syntactic and morphological units; the underlying principle is the same. [[puerttm] [[homin] [[neca-v]

necat] -ibus] -erat]

boy-Ace. kill-Pr.-3sg. men-Dat. kill-pluperfect-3sg.

'he kills a boy' 'for men' 'he had killed'

Evidence from the evolution of morphological forms supports this interpretation: the Latin ending of the oblique cases corresponds to the preposition in French, which is clearly the head of the prepositional, hence syntactic, phrase. The identification of the head can then be accounted for by referring to the function this element has in a given structure: it assigns a grammatical function to its complement, it determines its grammatical value (see Bichakjian 1991), or it expresses the grammatical features. This may then supply the definition of the head in syntactic as well as morphological structures, where the relations, though all hierarchical, are not identical. Accordingly, I borrow from current X-bar theory the idea of structural parallel, the idea of binary structures, and the identification of head at all phrasal levels. In contrast with X-bar theory, I do not apply the criterion of category consistency to identify the head, but I refer to the syntactic role of the head in the structure of which it is part. Moreover, in my view the structural parallel also includes morphological forms, which is in line with the identification of binary structures at all phrasal levels. The approach then results in an analysis of syntactic as well as morphological structures in terms of branching, where the binary distinction between head and complement is basic. With this identification of head in syntactic and morphological structures, I can now analyze the development of grammatical structures in terms of branching. This is the topic covered in the next three chapters, which deal with the evolution of the noun phrase (chapter 3), the verb phrase (chapter 4), and the structures that underwent the shift toward right branching at an early date, the prepositional phrase, the comparative construction, and the relative clause (chapter 5).

3 Diachronic Analysis: The Noun Phrase

In order to account consistently for the underlying regularity of Latin syntactic and morphological structures and their evolution, I will carry out an analysis in terms of branching. Systematic diachronic analysis of word order patterns is not new, of course. (See Lehmann [1971 et seq.] and Adams [1977a], who deals with typological word order patterns in Latin.) In the preceding chapter I therefore indicated in what respect the branching analysis is different from earlier systematic analyses of word order. I elaborated this approach and its relation to typology and other phrase structure analyses and attempted to give an independently motivated definition of head. With this definition it is possible to determine the branching of every grammatical structure. In the following chapters I analyze the linear organization of grammatical structures of Latin and their development. I show with this analysis that the syntactic and morphological structures of Latin shifted from left to right branching. Since the analysis in terms of branching is based on the principle that every unit or phrase has its own head, I will proceed by individual phrases. For each phrase I first determine the original unmarked place of the head and then trace its development. Moreover, the analysis of morphological units is integrated into the discussion of syntactic phrases because the underlying principle of organization is identical in both types of entities. My diachronic analysis of grammatical structures is treated in three chapters. In the first, I discuss the nominal phrase, where the noun can be combined with a genitival complement (section 3.1), an adjective (section 3.2), and a determiner (section 3.3). In chapter 4 I examine the place of 47

48


the verb in the sentence, its position in relation to its complement, and the place of the adverb. Moreover, replacement of the LB inflected verb form by a preposed auxiliary and the preposcd subject personal pronoun is considered in detail. In chapter 5 I deal with the prepositional phrase, the relative clause, and the comparative construction. At face value this combination may appear somewhat strange, but these three structures were among the first to shift toward right branching. Moreover, the relative clause and the comparative construction differ from the other phrases in that they arc complex structures. In complex structures the complement is cither a hierarchical structure itself (for instance, the relative clause) or part of another phrase, as some types of comparatives are. In this way my analysis proceeds from noncomplex (NP, VP, and PP) to complex structures, the early reorganization of which has important theoretical implications (sec chapter 6). Although the change of branching occurred in all phrases, it took place at different periods in the various structures. In some phrases the change has not yet been carried out (e.g., the current replacement of the leftbranching French future je chanterai with its right-branching equivalent je vais chanter 'I will sing'). For other phrases (e.g., the prepositional phrase) one needs to go back far in time to find the archaic LB organization. Therefore, an important part of this book deals with PIE. If I had limited my concern to Latin data, I would have deprived my analysis of an important part of the evolution, which is Indo-European; without these data I would never have been able to determine the real, unidirectional nature of the change. For the analysis of the Latin development we have not only numerous texts, but a considerable number of studies that describe the order of a given syntactic structure in the writings of various authors. These analyses may not be recent, but they are characterized first of all by subtle intuition and a high degree of accuracy: knowing the Latin author very well, the oberver was able to evaluate every deviation from the norm or from the personal style of the author. Yet these analyses took into account only one structure and did not relate it to a more comprehensive syntactic system. Consequently, the relation between the various linguistic phenomena and their evolution passed unnoticed. This lack of perspective is the result of the fact that most often these analyses were carried out without a theoretical frame, which, on the other hand, makes these studies more objective and the evidence all the more reliable. Because of these studies we have a huge amount of data. Their analysis and comparison reveal the underlying regularity of the grammatical structures and their development: these objective and independent data corroborate each other, and in turn support the grammatical regularity I demonstrate later. The same remarks can be made for the studies of Old and Middle French. I trace the evolution to the extent that the data permit, comparing written evidence from Early, Classical, Vulgar, and Late Latin, and from Old and Middle French. I pass directly from Latin to French, disregarding the Proto-Romance period. Proto-Romance presents a double problem.

Diachronic Analysis: The Noun Phrase

49

First, there is disagreement between those who advocate the hypothesis of one proto-language (see, e.g., Hall 1950, 1983) and those who support the other hypothesis, that the Romance dialects split up immediately: "[I]t seems to me venturesome to insist, as does Hall, that there was a single Proto-Romanic spoken throughout the provinces, regardless of the date of conquest, despite the different kinds of Latin imported into various areas at so greatly divergent dates, and despite the variety of substrata superseded by Latin" (Pulgram 1986:102). I do not want to discuss here the validity of either approach (for the state of the art, see Wright 1982; see also Herman [1978, 1983] for the degree of dialectalization in Latin inscriptions in the Propincia); another aspect of this problem is more relevant to this book: the quasi-total absence of syntactic reconstruction, and especially of reconstruction of word order. Studies of Proto-Romance or of earlier linguistic evidence focus especially on phonology, morphology, and the lexicon. Genuine reconstructions, syntactic or other, are, however, very scarce (see, e.g., Hall 1983). In a recent series of articles de Dardel attempted to reconstruct ProtoRomance word order (1985, 1987, 1989). As a first step he proposed a hierarchy of the Romance languages and dialects according to their degree of archaism; then he reconstructed the successive stages of Proto-Romancc: he distinguishes a first period (represented by Sardinian), which is characterized as VSO, followed by an OVS stage and a third period, which features SVO. This hypothesis is based essentially on the analysis of the sequences adjective—noun, the place of the verb, and the structure of compound nouns. Since there are no earlier data, dc Dardel uses exclusively proverbial expressions and medieval texts, which means that there is a considerable gap between the observed data and the proposed hypothesis. From a theoretical and evolutionary point of view these hypotheses seem speculative and ignore known typologies. The chronology de Dardel sets forth includes a period of OVS structures, which is extremely rare typologically; it presupposes a pendular movement—also highly exceptional —between left branching and right branching. In his view Latin SOV gives way to VSO in the first period of Proto-Romance, which is followed by OVS of the second period; finally, left-branching OVS is replaced by SVO. Because of the absence of data, the empirical basis for this hypothesis has not been established according to the strict principles of the comparative method and the method of internal reconstruction. Consequently, the empirical basis does not seem solid enough to accept the contradictions mentioned. The large gap between the last examples of Vulgar and Late Latin on the one hand and the first examples of Old French on the other compels the linguist to limit the analysis to the data available, and hence, to compare the Latin and French data in a diachronic perspective. This approach is justified from a linguistic point of view, because the overall changes that began in Latin, or earlier, continue in Old and Middle French, for example, the increasing importance of transitivity that can be observed in the substitution of gerundive constructions by infinitive structures governing

50


an object, or the shift from an aspect-based to a tense-based verbal system. Like the shift toward right branching, these shifts are part of language change that external factors—such as substratum influence—do not trigger: they arc part of Meillet's "lignes generales." It is useful, however, to notice that in Gaul Latin was exposed to two languages that were typologically different: the Gaulish substratum and the Germanic superstratum (for a discussion of language contact in Gaul, especially Gaulish and Latin, see Schmidt 1983; Polome 1983). In contrast with Lepontic and Celtiberian, which were still OV languages, Gaulish was, as far as we can tell from the little evidence we have, "overwhelmingly VO [and showed a] preference for VO order and head nouns ahead of genitives" (John Koch 1985:66, 267; sec also Karl Horst Schmidt 1976, 1990; Vcndryes 1911). VSO order, on the other hand, was rare in Gaulish (Koch 1985:245). Although our data are fragmentary and few in comparison to, for example, Latin, we may therefore assume that Gaulish was undergoing a shift from left to right branching: it had definite SVO characteristics, but it still featured SOV residues of the previous period (Schmidt 1990:190). In Germanic, which was the superstratum after the period of the great invasions, the situation was totally different. In runic inscriptions, such as the Gallehus inscription (ca. A.D. 325), and in the early texts of other Germanic languages, one observes the sequences object—verb, genitivenoun, adjective—noun, and standard—comparative: "[T]he OV characteristics in the earliest Germanic materials . . . suggest that the parent language, Protogermanic, was OV in type" (Lehmann 1972b:246). The exposing of Latin to these different typologies further demonstrates that the shift I analyze in the following pages cannot be explained by external factors. The change under consideration was under way before the expansion of Rome and even before the Proto-Italic period. Consequently, substrata and superstrata do not have any explanatory power, although they might have accelerated or restrained the reorganization. In the present chapter I analyze the structural modifications of the nominal phrase and account for its reorganization and its shift toward right branching. As head, the noun can be associated with a genitival complement or an adjective. These complements arc examined in the two first parts of the chapter. I then deal with the determiner. In contrast with the genitive and the adjective, the article, which is the result of a more recent development, governs the noun, like the other determiners, because it expresses its grammatical value (see chapter 2). The analysis of these three elements of the NP—the genitive, the adjective, and the determiner —then reveals the shift toward right branching in noncomplcx nominal structures.

3.1 The Genitive The evolution of the genitive is characterized by a double change: in addition to the shift from a synthetic to an analytic form, a change in


51

branching took place. In the following pages I examine the replacement of structures of the type [[avunculi] exitum] (uncle-Gen. death-Nom.) with structures of the type [la mort [de l'oncle]}, where the analytic genitive (that is, de + (article) + noun) is placed to the right of the head noun. As a first step I examine the place of the genitive in the proto-language and in two Italic languages, Oscan and Umbrian. Although the data on Old and Late Latin are clear and unambiguous, indirect evidence is required to determine the position of the genitive in Classical Latin. Finally, the tendency toward right branching increased in Old and Middle French, when the shift toward the analytic form eventually was accomplished.

3.1.1 The Place of the Genitive in PIE According to Delbriick, the place of the genitive in the proto-language "ist dieselbe, wie die des einfachen Adjektivums" (1900:102). Since the adjective preceded its noun (Delbriick 1900:98), the genitive normally did likewise. This is indeed the unmarked order in the oldest Sanskrit prose text, the catapathabrahmana, where the genitive is left branched in unmarked use (Delbriick 1878:42-43). Its unmarked preposing in the proto-language (sec also Lehmann 1974:74) is maintained, for example, in Archaic Greek, in Hittite (Friedrich 1974; Luraghi 1990:73), and in the Italic languages.

3.1.2 The Genitive in Two Other Italic Languages: Oscan and Umbrian Before it was the most important language in what is today Italy, Latin coexisted with a number of other Italic languages—I pass over non-Italic languages, such as Lepontic and Etruscan. Among them Umbrian and Oscan were the most important and best documented. These languages were closely related, but Oscan was more extensive than Umbrian, it withstood Latin longer, and it was even a serious threat to Latin (see Pulgram 1958:319 et seq.). Inscriptions in this language are found on the walls of Pompeii dating from just before the destruction of the city: "(On trouve] des graffiti en cette langue . . . sur des murs dont la decoration cst du dernier style" (Vaananen 1966:15). Since these two languages are closely related to Latin and since they are somewhat archaic in general, they may be expected to present useful information on the first stages of Latin. Oscan and Umbrian evidence shows that the genitive normally preceded the head noun. In her analysis of all Oscan inscriptions Konneker (1975) indeed pointed out a clear preference for the preposed genitive (1975:6). Moreover, the examples of postposing all occur in inscriptions later than 200 B.C. In earlier texts all genitives precede (7 of 7; Konneker 1975: 370). Konneker's statistical data are corroborated by the stylistic arguments of Rosenkranz (1933). In his analysis of various Italic texts Rosenkranz took

52


into account the context to explain the exceptions to the normal order. In the Oscan text Cippus Abdlanus (second century B.C.) on the temple of Hercules, the genitive most often precedes the head noun (Rosenkranz 1933:131), whether it is a noun or a pronoun; for example, senateis senate-Gen. 'the judgment

tanginud judgment-Abl. of the senate.' (8)

This text presents only one example of a postposed genitive. This is clearly marked: it is the first time the subject of the text is mentioned. Compare Sakaraklum Herekieis (11) 'the temple of Hercules' (Rosenkranz 1933: 131). Evidence from the Tabula Bantina (last quarter of the second century A.D.) and the Dedicationes lovilae (third century A.D.) is consistent with these data: in these and other Oscan texts, left branching was the unmarked order: "[D]ie habituelle Stellung des attributiven Genitivs [i.e., preposing] herrscht auch in den iibrigen oskischen Inschriften" (Rosenkranz 1933:133). Instances featuring a right-branched genitive complement are related to a specific use: the complement either is emphasized (see, for instance, in names of temples, the exact indication of which is important in a polytheist society [Rosenkranz 1933:132—33]) or is composed of two codepcnded genitives. Sec, for instance, the Tabula Bantina: sena[teis} tanginud maimas senate-Gen. judgment-Abl. big-Sup.-Gen.sg. 'the judgment of the majority of the senate'

camels part-Gen, (1.6-7)

The Umbrian Iguvine Tables (ca. 200—70 B.C.) show that preposing of the pronominal genitive is the rule in this language (Rosenkranz 1933: 134), as in erar this-Fem.-Gen. 'for her name'

nomneper name-Abl.-for (6 A 23)

Similarly in phrases featuring a nominal genitive, G[cnitive]N[oun] also predominates: of the 196 examples, 141 are left branching, that is to say 72 percent. All exceptions but three are marked: these are either local indications, dedications, or religious references. Thus whenever it is important to indicate clearly the divinity that is worshiped, the postposing of the noun produces the emphasis that is required (Rosenkranz 1933; for a recent confirmation of this explanation, see Konnckcr 1972:44 et scq.). The other instances of postposing can be accounted for by emphasis (in


53

detailed temporal indications, for instance) or by the length of the complement (see the Gesetz der wachsenden Glieder). Finally, the genitive also follows the noun in the names of state servants. This use can also be observed in other Oscan and Umbrian inscriptions and is explained in section 3.1.3 (see Rosenkranz 1933:135-37). The instances of postposed genitives in these ancient languages can be accounted for in terms of special usages, for example, emphasis. We may then conclude that the unmarked genitive was left branched in Oscan and Umbrian. The reverse right-branching order was not excluded, but it was rare and related to specific contexts; it was the marked order.

3.1.3 Old Latin The situation in Old Latin was more or less the same. The genitive was normally left branched, but the beginnings of the imminent change were already manifest. The Senates Cmsultum de Bacchanalibus (henceforth the S.C. de Bacchanalibus) is one of the sources of our knowledge of Old Latin. Like the Oscan Cippus Abdla-nus (see section 3.1.2), which is from the same period, this is a juridical text. Although the S.C. de Batchanalibus is only from 186 B.C., its language is extremely conservative. Except for two instances, the genitive precedes the head (12 vs. 2; see Rosenkranz 1933:137): senatuos sententiad senate-Gen. decision-Abl. 'with the decision of the senate'

(17)

Although the genitive tends to be left branched in the Old Latin texts, its postposing does occur also, but it is not frequent and is only related to specific occurrences: right branching occurs in topographic indications or when the complement is long: ager privatus casteli Vituriorum land-Norn. private-Nom.sg. fortified settlement-Gen. Viturii-Gen. 'the private land of the foritified settlement of the Viturii' (C.I.L. 1.2.584.5) ctedem Castoris Pollucis temple-Ace. Castor-Gen. Pollux-Gen, 'the temple of Castor and Pollux' (C.I.L. 1.1506)

Likewise, the names of the gods generally are right branching, thus marking the difference with other gods, for instance, in the S.C. de Bacchcmalibus:

54


apud aedem Duelonai at temple-Ace. Bellone-Gen. 'at the temple of Bellone' (26.2)

Consequently the hypothesis that applied to Oscan and Umbrian data also accounts for the RB organization of Latin phrases indicating temples: "[B]ei der grossen Zahl der Heiligtiimer kam es auf die genaue Bezeichnung an" (Rosenkranz 1933:138). Noun phrases indicating functions, especially public functions, form another category in which the genitive has followed the noun since the archaic period, as in pater familias 'father of the family', tribunus plebis 'tribune of the people', mater deum 'mother of the gods, Cybil', praefectus urbis 'prefect of the city3. In these structures "liegt ein latenter Gegensatz" (Rosenkranz, 1933:139): the element that expresses the opposition is stressed, hence postposed. This explanation is consistent with Oscan and Umbrian data, which show the same phenomenon (see Konneker 1972: 95). Once these expressions became fixed, they remained in the language. For instance (Rosenkranz 1933:139): tribunus plebis

'tribune of the people' (Liv. 2.33.2)

tribunus militum

'military commander' (Liv. 4.6.8)

tribunus Celerum

'commander of the Celeres' (Liv. 1.59.7)

magister equitum

'chief of the cavalry3 (Varr. L.L. 5.82)

magister navis

'captain of a ship' (Liv. 29.25.7)

In addition to these instance of marked RB word order, patronymic and juridical expressions support the hypothesis of the unmarked preposed genitive in Old Latin. Juridical expressions indeed display the archaic order, as in iuris consultus 'a lawyer3, vitae necisque potestas 'power of life and death', or manus iniectio 'manumission' (for juridical compounds, see also Bader 1962:297). Finally, in patronymic expressions the genitive precedes the noun. These old formations are left branching and one finds numerous examples in dedicatory inscriptions, for instance, C. Fanni M. f. (C.I.L. 1.2.658) 'Gaius Fannius, son of Marcus'. This was the dominating structure in patronymic expressions and it still existed in Classical Latin: in Caesar, for instance, the occurrence of'genitive + films' outnumbers Jilius + genitive' (16 vs. 2; Adams 1977a:75).


55

3.1.4 From Old to Classical Latin Although the genitive normally preceded the noun in Old Latin, statistical data from Classical Latin onward (Adams 1977a) show clearly that a change in branching took place after the Old Latin period. Evidence covering a period from Plautus to Vegetius was gathered by Adams (1977a); its analysis shows that the LB genitive no longer was the unmarked position from Plautus on. Yet preference for post- or preposing of the genitive is not clear in any of the analyzed writers, except in Caesar, where the 437 examples of N[oun]G[enitive] outnumber the 319 instances of GN in De Bello Gallico (1-3.32) and Tacitus (Annales 14), where we find 311 examples of NG and only 212 of GN (1977a:77). Only the later authors (Petronius, Pseudo-Quintilianus, and Juvenal) tend to make the genitive follow the noun (see the statistical data of Adams 1977a:77-78). These numerical data show not only that neither prc- nor postposing prevails in the given period, but also that the work of a given author may show various preferences (for instance, the preference for LB genitives [63.6 percent] in Plautus's Aulularia, whereas right-branching genitive nouns dominate [60.5 percent] in his Bacchides [Adams 1977a:77-78]). Yet since preposing of the genitive in Aulularia often occurs in fixed expressions, the preceding statistical data may suggest a randomness that is not consistent with reality. Instead of giving up statistics as some do (e.g., Adams 1977a:78), it may be interesting to analyze the data in greater detail in order to determine the actual difference between the occurrences of GN and NG. The irregularity that the preceding data reveal shows first of all that the order in noun phrases is not determined by a single strict principle. Yet the order of the elements was not indiscriminately free. The context (marked or unmarked) and the syntactic nature of the genitive—nominal or pronominal—had an impact and this may account for the apparent irregularity. In the first literary Latin texts—those of Plautus and Terence, for instance—the nature of the genitive turns out to lay an important role in the organization of the phrase. Bennett (1914) collected a large number of phrases featuring a genitive (ca. 1250 in all) from the period before 100 B.C., which he considers the period of Early Latin (1910:iii). Analysis of these examples, most of which have been taken from the works of Plautus and Terence, indeed reveals regularities in the linear organization of these structures. The genitives in question are the possessive, subjective, and objective genitives; genitives of quality; and genitives in free use (of the type libertas orationis [PL, Batch. 168] 'freedom of speech'). The number of their occurrences is presented in the table. Statistical data from this table show that the numerical distributions of GN and NG are nearly equal, although there is a slight preference for left branching (53.8 percent). Yet, if one takes into account the nature of the genitive, one observes that the pronominal genitive clearly tends to pre-

56

The Emergence and Development of SVO Patterning Numerical Distribution of GN and NG in Bennett's Corpus (1914)

Total

GN

NG

Total Numbers According to Categories PRONOMINAL GN NG NOMINAL

1250 673 577 % 53.8 46.2

233

167 71.7

66 28.3

1017

GN

NG

506 511 49.8 50.2

Sequences Genitive + Noun, According to the Nature of the Genitive TOTAL GN NG

673 (53.8%) 577 (46.2%)

PRONOMINAL . 167 (24.8%) 66 (11.4%)

NOMINAL 506 (75.2%) 511 (88.6%)

cede the head noun (more than 70 percent); the nominal genitive, on the other hand, shows a slight preference to follow the head noun. Since this difference is less than 1 percent, it is unimportant. The numerical preference for left branching in these examples is accordingly due to the high occurrence of pronominal genitives. The pronoun tends to precede the head in all but one type of genitive phrase (the exception being the free uses of the genitive). The distributions of nominal NG and GN are roughly equal in the time of Plautus and Terence. Consequently, the quasi-absolute dominance of the LB nominal genitive as observed in Italic and Old Latin no longer exists in the early literary period, only the pronominal genitive is still regularly left branched. Since preposed and postposed nominal genitives were in nearly equal distribution, a more detailed analysis of the data might show whether GN or NG has a particular stylistic value. There are indeed indications that suggest that the RB genitive was less marked than its LB equivalent. Explaining the left-branched genitive as a result of either stylistics or context, Adams argued that NG was the unmarked order in Classical Latin (1977a:78 et scq.). His stylistic argument is based on examples of antithesis, which are, of course, clearly marked. Other instances of preposing can be accounted for by context: sequences with adjective and genitive governed by the same noun, as in the example from Caesar (DBG 1.19.3): summam omnium rerum fidem highest-Ace.sg. all-Gen. affairs-Gen. confidence-Ace. '(in whom) he had total confidence in all matters'

habebat have-Impf.-3sg.

In this type of construction the adjective and the genitive are often placed at the same side of the noun, resulting in the sequence AGN or NGA. In this organization the place of the adjective determines the ordering of the phrase: when the adjective precedes, the genitive tends to do likewise (for the [marked] preposing of the adjective, see section 3.2). An important number of preposed genitives can thus be explained (cf. Caesar's example).


57

This assumption is supported by Elcrick's (1991) statistical evidence. In his analysis of a number of Cicero's speeches (Pro Caecina, In Catalinam 1-3, Philippicae, and Lex Manilia) Elerick (1991:318) found 71 instances of nouns with dependent adjective and preposed genitive. In all except four examples, the adjective also precedes the noun. The examples featuring a postposed genitive do not display this type of regularity (75 AN vs. 19 NA). Similarly, when the adjective modifies the genitive and normally precedes the noun (see section 3.2), both elements generally precede the noun governing the genitive (see Adams 1977a:78— 82): magni ponderis heavy-Gen. sg. weight-Gen. 'rocks of heavy weight'

saxa rocks-Ace.

(Caes. DBG 2.29.3) tantae altitudinis machinationes SO-great-Gen.sg. height-Gen. engines-Ace. 'engines of such a great height' (Caes. DBG 2.3 1.2)

Often these periphrases with genitive are equivalent to superlative adjectival structures, such as gravissima,saxa. and altissimae machinationes, which arc marked. Accordingly, these instances show that the sequence AGN, where the genitive precedes its head noun, can be explained by the marked use of the adjective. As in the previous structure where the noun governed a genitive and an adjective, Elerick's statistical data support the hypothesis of the decisive role of the adjective in this matter: of 82 examples of genitive— noun, 64 are of the type [[adjective] genitive] -noun, whereas the type [genitive [adjective]]— noun only occurs 18 times. For the postposed genitive these numbers are nearly equal (Elerick 1991:318). The phenomena I brought up in the preceding paragraphs show that preposing of the genitive is motivated either by emphasis or by the occurrence of the adjective with which it is combined. This indeed means that the left-branched genitive is marked. That the preceding genitive was the marked variant can be further demonstrated by another form of emphasis characterized by repetition, and by the change of order in a specific type of structure that can be traced in the history of Latin. A specific but common type of emphasis consists of repeating the noun in question, which takes the genitive form. These structures are left branching (see Ernout and Thomas 1951:50): reliquiarum reliquias leavings-Gen. leavings-Ace. 'the remnants of the remnants' (PL Curc. 388)

58

The Emergence and Development of SVO Patterning divum deo gods-Gen. god-Dat. '(to) the god of gods' (Carm; Varr. L.L. 7.27)

From Christian times onward these expressions shifted toward right branching: vanitas vanity-Norn.

vanitatum vanities-Gen.

(For the status of these expressions and the possible Hebraic influence, see section 3.1.5.) Adams mentions another phenomenon, which he does not relate, however, to the structural change in Classical Latin (1977a:74) but is a telling indication of the shift. At the beginning of the Classical period we find fixed expressions that often go back to Old Latin and that changed branching in the course of time. These new structures with a reverse rightbranching order are a terminus antequermof the shift toward right branching: Adams noticed a word order change in phrases featuring senatus, "[which] precedes its governing noun in the ancient senatorial expressions senatus consultum, senatus auctoritas and senatus sententia" (Adams 1977a: 74; sec also Bennett 1914:51—55). Cicero generally uses senatus auctoritas 'the decree of the senate'; Caesar never uses the reverse order, "whereas already in Livy it is auctoritas senatus that predominates" (Adams 1977a: 73). The change in branching can also be observed in other expressions featuring the noun senatus: acta senatus 'records of the senate', decretum senatus 'decree of the senate' (Tac.Ann. 3.37; 4.44; 6.11; 14.49), sententia senatus 'decision of the senate.' The same shift took place in other juridical expressions and in other nominal phrases, for example, deum pax, which Plautus uses instead of fax deum, which one finds in Livy. Finally, the following example clearly shows that the postposed genitive is the unmarked order. It displays marked left branching and forward gapping, which is characteristic of a phrase that in an unmarked order is right branching. sub under

Veneris Venus-Gen.

regno kingdom-Abl.

vapulo, suffer-Pr.-lsg.

non sub Iovis not under Jupiter-Gen. 'I suffer under Venus's rule, not Jupiter's (PL Pseud. 15) The omission of the second regno, instead of the first, points to unmarked right branching. Moreover, the preposing of the genitive stresses the antithesis (Veneris vs. lovis), and thus supports the hypothesis of marked preposing: this example corroborates Leumann and Hofmann's idea that


59

GN was possible especially in emphasis: "[I]n guter Zeit ist Voranstellung des Gen[itivs] nur bei Betonung moglich, bes. im Gegensatz" (1928:617). The preceding evidence suggests that in Classical Latin the genitive was well on its way in changing from left to right branching. In number nominal postposed and preposcd genitives are equal; the pronominal genitive, however, still tended to precede the head noun. This was a residue of the archaic organization. It is the actual use of the nominal genitive—not its frequency—that reveals its unmarked position: preposing of the nominal genitive was related to emphasis or to a specific context, whereas the unmarked position was after the head noun. One observes the same principle in some Germanic languages: the unmarked RB genitive coexists with its marked, left-branching variant. In German, for instance, the synthetic nominal genitive precedes the head noun when denoting a proper name (for instance, Englands Konigin 'the queen of England'), whereas postposing of synthetic or analytic genitive, which is the unmarked order, is not motivated. 3.1.5 From Classical Latin to Vulgar Latin and Late Latin In Classical Latin it was the specific value of the preposed genitive that revealed its marked nature; in Vulgar Latin and in Late Latin the high frequency of the postposed genitive confirms the tendency toward unmarked right branching. In the following pages I first compare data from typical Late and Vulgar Latin texts then bring up new uses of the genitive that became increasingly popular in Late Latin and that corroborate the observed tendency toward right branching. The observed shift in Classical Latin becomes more obvious in Vulgar and Late Latin. Pertronius, for instance, prefers postposed genitives: "[Wir finden] 53% aller Genitive hinter das zugehorige Nomen gestellt, 47% zeigen Voran- oder Zwischenstellung" (Feix 1934:26). Preposing, on the other hand, is motivated: emphasis is used to prevent misunderstandings. In the Peregrinatio right-branching structures outnumber their LB equivalents (15.5 vs. 1; see Haida 1928:2). These are once again explained by emphasis or stylistics: "[L]a postposition etant dc regie, 1'anteposition signifie une mise en relief ou un enchainement" (Vaananen 1987: 107). Likewise in the Chronicle of Anonymus Valesianus II postposing dominates: only 10 of the 107 instances of noun phrases including a genitive are left branching. Moreover, "most of the 10 examples of anteposition in II are in invariable formulae ... or in expressions with a formal or official ring," for example, aquae ductum (71) and patriciatus dignitatem (38; Adams 1976:140). In the writings of Gregory of Tours and Fredegar—two interesting authors, whom one must be cautious about, however, because of their bilingualism—"NG occurs almost without exception" (Adams 1977a:73). If indeed in the work of Gregory of Tours postposing of the genitive is normal and is part of the objective style, the order [[genitive] noun] represents the stressed or literary'variant: "[Lj'an-

60


teposition semblc etre le signe d'une certaine emphase . . . ou bien il correspond a une affectation, unc 'enflurc' litteraire fort sensibles" (Herman 1970:86). In the following pages I analyze the use of the genitive in the earliest Vulgar text, and in the writings of two Christian authors who wrote for two different types of public.

3.1.5.1 The Letters of Claudius Terentianus The letters of Claudius Terentianus (first quarter of the second century) arc among the earliest sources of Vulgar Latin. In addition to phonological data, they present evidence of morphological and syntactic modifications. This Roman soldier was probably the son of a mixed marriage; that would explain his Greek—Latin bilingualism. The lack of Latin interference in his Greek texts suggests that the author normally spoke Greek, which does not mean, however, that he was not "completely fluent in Latin" (Adams 1977b:3). He wrote his letters in Greek and in Latin. It turns out that in this correspondence the postposed genitive clearly dominates (ca. 80 percent in the Greek and 70 percent in the Latin texts). The Latin examples of left-branched genitive are either formal or fixed expressions; such a pattern stresses the archaic nature of this type of organization; for instance, the fixed expression deorum beneficio (467.26) 'by the help of the gods' (Adams 1977b:70-71).

3.1.5.2 Two Christian Authors: St. Jerome and St. Augustine St. Jerome's translation of the Old Testament for a large and general public provides important data on the position of the genitive. In the first four chapters of Genesis the 101 examples of postposed genitive contrast sharply with the single instance of GN (see Adams 1977a:82). St. Jerome most certainly based his translation on the (original) Hebraic version and stayed as close as possible to the original text (see Labourt 1949). Since Old Testament Hebrew features the structure [Noun [Genitive]], the dominating RB genitive may well result from the interference of this language. Yet St. Jerome intended to achieve two things. First, the diversity of questionable and bad translations easily affected the unification of the dogma and the unity of the church. This explains why he respected the original text as much as possible. Second, he simply wanted to communicate the word of God to a large number of people, and this desire prevented him from differing unduly from the everyday language. This explains why Jerome, even if he respected the original text, could not move too far away from popular speech in order not to fail in his aims. We may therefore conclude that the postposed genitive was common, not only in Hebrew but also in Latin of that time. This hypothesis is supported by evidence from the Peregrinatio, which is from the same period. Hebrew is also important for expressions of emphasis by means of


61

repetition. These expressions display a genitive and are commonly attributed to St. Jerome's translations. Although the influence of Hebrew is undeniable in this type of formula, it also reflects the popular tendency: caeli caelorum 'heaven of heavens', saecula saeculorum 'times of times', vanitas vanitatum Vanity of vanities'. In section 3.1.5,1 pointed out that Latin in an earlier period preferred prcposing in this type of formula. Consequently, we here have another temporal indication of the change, a terminus ante quem. Although St. Jerome shows a definite preference for the right-branched genitive in his translation, his letters arc considerably different: Adams pointed out that the preposed genitive dominates by far in the first letter (28 vs. 16; 1977a:82). This contrast between the letters and the translation can be accounted for by referring to the difference in audience: the Vulgata was meant to be used by common people, whereas St. Jerome's letters were written for his learned and scholarly colleagues. Moreover, they treat religious and dogmatic subjects. Accordingly, Adams was correct in concluding "that in the spoken Latin of the lower classes the stylistically marked variant GN was rarely used" (1977a:82). St. Augustine's Latin displays the same characteristics as St. Jerome's. Although the Sermones and the Confessiones show a preference for postposed genitives (3 vs. 1; cf. Balmus 1930; Muldowney 1937:55-56), the tendency to postposing was definitely less strong in Books 1 and 21 of De Civitate Dei and in the letters written to his learned colleagues, in which postposing does not exceed two thirds (Muldowney 1937:55). The Sermones and the Confessiones are generally considered less scholarly and less refined than De Civitate Dei, which is a religious treatise. In addition to the frequency of the postposed genitive, it is important to point out the context in which left-branched genitives appear in the texts of St. Augustine and others: in disjunctive constructions. More than half of the preposed genitives are found in [preposition—genitive—noun ] sequences. The postposed genitive, on the other hand, is rarely integrated in a disjunctive construction. Accordingly, preposing can be explained by stylistic expressions, by emphasis, or by the attempt to form a structural unity (cf. Muldowney 1937:56-58). 3.1.5.3 New Uses of the Genitive In addition to the tendencies of individual authors that show an increasing use of the postposed genitive, one also observes in Late Latin new uses of the genitive that give us information on the preferred branching. Late Latin indeed displayed characteristic structures that were right branching. First of all, noun phrases with an adjective could be replaced with an abstract noun followed by a genitive (Bonnet 1968:548 et seq.): in sanctitate continentiae viveres in holiness-Abl. continence-Gen. livc-Subj.-past-2sg. 'that you should live in the holy' state of continence' (Aug. Ep. 220.12)

62


The noun—genitive phrase replaced in sancta continentia in holy-Abl.sg. continence-Abl. In the works of Gregory of Tours we find instances where a genitive with adjective replaces a superlative. This same structure—with genitivus qualitatis —occurred earlier, but it became frequent in Late Latin; cf.: homines . . . magnaeque virtutis men-Ace. big-Gen.sg.-and valor-Gen. 'men ... of a great valor also' (Caes. DBG 2.15.5)

Ernout and Thomas ascribe its popularity to Hebraic influence (1951: 46). Yet the tendency existed before Hebraic influence: instead of an adjective one finds a genitive in, for example, deliciae pueri (Pl. Pers. 204) 'darling boy'. The replacement of a simple apposition with a genitive is part of a more general change that can also be traced in geographical names. In Old and Classical Latin, apposition was common in these expressions: for example, urbs Roma, which gave way in the course of time to urbs Romae. This use, which is not considered proper, occurred especially in familiar Latin (Riemann 1927:112). At the same time another substitution developed: the construction of the type arbor olea 'olive tree', common in Varro (e.g., R.R. 1.2.20), gave way to structures of the type arbor sicomori (Per. 8.3.) 'sycamore' (examples from Ernout and Thomas 1951:43). Far from being an isolated phenomenon, these expressions are integrated in the development of subordination (see Bichakjian [1982] on structures of the type la ville de Vienne]: the attributive element, which is appositive in the archaic structure, becomes grammatically marked as complement of the head. This structure is never found in Caesar but became common after him (Ernout and Thomas 1951:43). The Peregrinatio offers numerous examples that often go back to examples in the Vulgata. Later the genitive was replaced by a prepositional phrase and the development toward right branching was achieved, as in [la ville [de Rome]] 'the city of Rome', where the left-branching case ending is replaced with the RB prepositional phrase. The last stage of this development was accomplished in Old and Middle French. In Classical Latin the use of the genitive revealed its unmarked position. In Late Latin the preference for right branching also clearly shows in its high frequency. Moreover, new uses of the genitive are characterized by right branching. These unambiguous data of Late Latin support the hypothesis that in Classical Latin the shift toward right branching was well under way.


63

3.1.6 Old French It is in French that the shift toward RB analytic genitive constructions was accomplished: the eventual replacement by prepositional phrases (hence the choice of the preposition) was a medieval development. Gradually the structure without preposition of the type la fille le roi 'the king's daughter' gave way to the prepositional phrase.

3.1.6.1 The Place of the Genitive in Old and Middle French In the Old French noun phrase the genitive directly followed the head noun (Foulet 1923:17; see also Moignet 1973:349). One still finds examples of the type pro deo amor et pro Christian poblo (Serments de Strasbourg 1) 'for God's love and for the Christian people' or liDeo inimi (Eulalie 3) 'the enemies of God', but these are archaic structures that survived in fixed expressions, such as la Dieu merci (Queste 40.17; Ch. 60.2) 'God's pity'. These expressions were still common in the thirteenth-century chansons de geste (Foulet 1923:18). These and earlier texts provide instances of the genitive preceding the head noun not only as an absolute construction, but also as a prepositional phrase (e.g., Moignet 1973:349). They are, however, highly marked: in medieval French the genitive followed the head noun in unmarked order. In Middle French the place of the genitive is no longer an issue for grammarians. Immediate postposition became a stricter rule: it was no longer possible to separate the complement and the noun as occasionally had occurred in Old French (Marchello-Nizia 1979:319). Postposing of the genitive, therefore, was achieved at the end of the Latin period; in the subsequent, French, period the choice of the preposition was accomplished.

3.1.6.2 The Morphological Change In Old French the genitive could be expressed generally in two ways: either with an absolute construction, of the type la fille le roi 'the king's daughter', or with a prepositional phrase. This phrase was introduced by de when the complement noun indicated nonhuman animates or objects; when it referred to human beings, the genitive construction was introduced by the preposition a, which was probably related to the former Latin possessive dative. In the absolute construction, which was the original structure, the noun had an oblique ending and followed the head noun immediately. The use of this construction was related to linguistic and social conditions (definite reference, expression of human relations, hierarchically important person; cf. Foulet 1923:15—17). Moreover, a syntactic condition stipulated that the nominal complement be an underlying subject, as in le courouz le roi 'the king's anger, the king is angry', which contrasts with le courouz de son

64


roi 'the anger against the king'. This purely syntactic requirement further demonstrates the marked nature of the absolute structure. Finally, its marked character shows also in the nearly exclusive use of the singular. The restriction of the complement to a limited number of well-defined persons might account for, according to Foulet (1923:22), the very low frequency of the plural. Yet the preference for the singular proper—the unmarked number—demonstrates the marked nature of this construction. The conditions of the use of the absolute construction were roughly the same in Middle French, only stricter. The construction occurred exclusively in the singular. Moreover, fourteenth- and fifteenth-century French typically featured prepositional phrases even in contexts where the absolute construction was possible. This was rather rare in Old French (MarchelloNizia 1979:319). The occurrence of the absolute construction was declining. Kinship relations, for instance, were almost exclusively expressed with a prepositional phrase from the fifteenth century on. Possession was expressed in the same way: the absolute construction was very rare after the fourteenth century (Marchello-Nizia 1979:319—20). Moreover, the prepositional construction dominated the expression of social hierarchy from early in the fourteenth century. Finally, although the agent relation was still expressed in absolute construction in the fourteenth century, its use was limited to fixed expressions in the following century and to the person of God: it was rarely used with proper names (Marchello-Nizia 1979:320). Statistical data clearly reveal the decline of the absolute construction, which increasingly was replaced with prepositional phrases introduced by de or a (Marchello-Nizia 1979:321). In the fifteenth and sixteenth centuries the absolute construction only occurred in fixed expressions, where it survived into modern French, for example, in photel-Dieu 'hospital'. Gradually the prepositional genitive phrase prevailed in all linguistic contexts. First the choice between de. and a was motivated by the grammatical feature fa human], which is consistent with its original use. But from the twelfth century on de came to prevail. In this way, the archaic possessive construction, which featured a dative of possession and which goes back to Proto-Indo-European, finally gave way to the modern structure with de.

3.1.7 Overview of the Development of the Genitive The evolution of the genitive is one of the few changes that can be traced in the history of Latin: in the course of time the unmarked order of the Latin genitive structure shifted from GN to NG. This development took place in three stages: in the first period the normal, unmarked order was [[genitive] noun], and the reverse order was marked. Then the sequence [ noun [genitive] ] lost its marked character and was used without specific connotations. The preposed genitive, on the other hand, which was still relatively frequent, came to be marked, whereupon its frequency dramat-


65

ically decreased in Late Latin. In French, finally, the RB analytic form introduced by de eventually became the only construction; this change achieved the shift toward right branching. The shift from left to right branching on the syntactic level was, therefore, a Latin phenomenon, whereas the choice of de was a French accomplishment. Latin and French data thus reveal the clear and unambiguous development of the genitive that is characterized by the gradual and consistent shift from left to right branching.

3.2 The Adjective In the following pages I analyze the place of the adjective in the noun phrase. In general its unmarked position is more difficult to determine than the unmarked place of the genitive. Compared to the adjective, the genitive noun construction is a clear-cut example of a syntactic element in subordinate function; it has been replaced in the course of time by a prepositional phrase, which clearly is a syntactic constituent. This assumption is supported by Benveniste's analysis of the Latin genitive that relates the noun phrase featuring a genitive to the verb phrase: "[L]a fonction du genitif se definit comme resultant d'une transposition d'un syntagme verbal en syntagme nominal" (1966:148). Since the genitive is an example of syntactic government and since it is less expressive, its place generally is a better indicator of language typology than that of the adjective. Although the evolution of the adjective is more complex than that of the genitive, the tendency toward right branching is clear. In this analysis I distinguish between the distinctive adjective (in French adjectif distinctif), which denotes a characteristic quality of the noun, and the descriptive adjective (adjectif descriptif), which expresses a judgment or an appreciation and which features degree. Since this is a specifically French distinction that has no equivalent in English, English grammatical terminology does not provide a proper term; I, therefore, use the literal translation of both terms and refer to the corresponding usage with "descriptive value" and "distinctive value." The distinction between these two subcategories, which goes back to Marouzeau (1922), is fundamental not only for the analysis of Latin but also for later developments: the place of the adjective is closely related to the value of this element. For a recent discussion of these different types of adjectives—determiners and qualifying adjectives—in Latin and French, see Touratier (1991), who accepts the distinction propose by Marouzeau. As I did in the analysis of the genitive, I will trace the evolution of the place of the adjective, but the period of change is much longer.

3.2.1 PIE, Oscan, Umbrian, and Old Latin Originally in PIE the simple adjective as well as the derived adjective preceded the head noun (see Delbruck 1990:98, 100; Lehmann 1974:69).

66


This was, however, not a strict rule. Predicative use and emphasis of either the noun or the adjective could create a deviant order. This order was still in existence in Greek where postposing of the adjective was always marked by the repetition of the article, whereas the preceding adjective, which represented normal order (see Brunei 1964:14), was not. This characteristic illustrates "die Ursprunglichkeit der Voranstellung des Adjektivums" (Delbruck 1900:94). The opposition between normal preposing and marked postposing can also be observed in Germanic, Iranian, and Vedic Sanskrit (see Delbruck 1878:35, 1900:94-102). From this situation, where preposing of the adjective was the normal order and postposing was marked, a development took place in the direction of unmarked right-branching adjectival structures. This was already partially accomplished in the Italic languages: in Umbrian and Oscan "the adjective regularly follows its noun, but may precede it if emphatic" (Buck 1979:224), for instance, in the following sentence from the Cippus Abellanus: ligatuis Abell[anuis] inim ligatuis Nuvlamiis legates-Abl. of-Abella-Abl.pl. and legates-Abl. of-Nola-Abl.pl. 'with legates of Abella and Nola' (6-7)

Similarly, in the Tabula Bantina we read: ceus Bantins citizen-Nom. of-Bantina-Nom.sg. 'citizen of Bantina' (4.19) But with emphasis: mu[inikad} tanginud common-Abl.sg. agreement-Abl. 'by common agreement' (Cipp.Ab. 15)

Buck's point of view is supported by the more recent analysis of Konneker (1975), who concludes that "[Oscan and Umbrian] descriptive adjectives normally follow the noun" (1975:367, where "descriptive" is French qualificatif, as opposed to adjectif' determinatif, 'determiner'). Indeed, in all the Oscan inscriptions, Konneker found 72 examples of the sequence [noun [adjective]], but only 5 instances of left branching (1975:370; for details on Umbrian, see Konneker 1972:44, 96 et passim). Considering the situation in Oscan and Umbrian one might expect that the adjective shifted to postposition in an early period in Latin. And indeed in the Twelve Tables, which goes back to the earliest Latin period, "adjectives . . . are positioned according to the same rules as those of


67

Classical Latin" (Adams 1977a:90): the postposed adjective was a common feature. The Twelve Tables (450—400 B.C.) is a juridical text, the style of which is cautious, hence somewhat archaic. Yet, its principal aim was to guarantee the rights of the plebeians and it was displayed at the Forum for everyone to see. Consequently, the text "may be presumed to have been composed in a straightforward style, reflecting popular usage fairly accurately" (Warner 1980:253-55). Among the phrases featuring a postposed adjective, one observes the use of religious fixed formulae and archaic expressions (Adams 1977a:89), as in [noun [ Campus

[adjective]] [Martius ] ]

This KB organization reminds one of the—often religious—Old Latin structures where the genitive complement that carried emphasis followed its head noun in marked order (see section 3.1). These fixed archaic formulae, therefore, presuppose an earlier period where the adjective was normally left branched as it was in PIE (Adams 1977a). To support this hypothesis Adams brings up fixed expressions that reflect the archaic unmarked order, for instance, the archaic expression erilis fillius 'the master's son', which became fixed in this order and occurs in later texts (e.g., Ter. And. 602 [1977a:89-90]).

3.2.2 From Old Latin Onward Having demonstrated that in Old Latin the adjective followed the noun, I now set out to determine the place of the adjective in Classical Latin and in the writings of Plautus and Terrence. For this period of Latin one needs to distinguish between the two types of adjective identified previously. This distinction is indeed relevant, because distinctive adjectives normally followed the head noun, whereas descriptive adjectives tended to precede the head (Marouzeau 1922:13 et seq., 1953:1 et seq.). See, for instance: infinita multitudo est men-Gen. be-Pr.-3sg. countless-Nom.sg. crowd-Nom. 'the population is very dense'

hominum

(Caes. DBG 5.12.3)

But: effugisse a vita marituma escape-Pf.Inf. from life-Abl. seafaring-Abl.sg. '(that I) escaped from a seafaring life' (Pl. Bacch. 342)

68


That the distinction between distinctive adjectives and descriptive adjectives was important can be demonstrated by the descriptive versus distinctive value of one and the same adjective in pre- or postposed position: praetor urbanus 'praetor for Roman citizens'

where the postposed adjective simply expresses a function, in contrast with urbanus praetor 'a witty praetor' where the adjective conveys a judgment (examples from Marouzeau 1922:14). Since the distinction descriptive versus distinctive is relevant to the place of the adjective in the noun phrase, the comparison of their use may reveal the unmarked place of the adjective. 3.2.2.1 Distinctive Adjectives Distinctive adjectives in Latin may be classified in two subcatcgories: 1. Adjectives that are derived from proper names: names denoting persons, groups, or geographical places 2. Adjectives that express a kind of affiliation (social, natural, racial, etc.) or a definition (material, local, or physical quality, etc.) Adjectives derived from proper names follow the head in unmarked order: "[L]a postposition . . . est de regie toutes les fois qu'il s'agit d'un enonce pur et simple, qui n'appelle aucun commentaire" (Marouzeau 1922:17). Similarly, adjectives expressing affiliation or definition are normally right branched. Exceptions are due to rhetoric (for example, chiasm as in campum Vaticanum . . . Martium Campum [Cic. Aft. 13.33a]) or to the emphasis of the adjective that supports argumentation, reasoning, antithesis, or expression of feelings (1922:17—32, 34—56). Compare Punica fide Punic-Abl.sg. faith-Abl. 'in bad faith' (Sall.jug. 108.3)

which contrasts with simple postposition, as in litteris Punicis letters-Abl. Punic-Abl.pl. 'in Punic characters' (Cic. Verr II 4.103)

Sometimes antithesis—even if it is not explicit—goes beyond the sentence. It then relates paragraphs, parts of the reasoning, and so forth, and

Dictchronic Analysis: The Noun Phrase

69

is also expressed by preposing, as when Cicero in one of his letters moves from his private life to public affairs (Marouzeau 1922): Nunc Romanas res Now Roman-Acc.pl. affairs-Ace. 'Now hear the news of the town'

accipe learn-Imper.-2sg. (Att. 4.15.4)

That the sequence [ [distinctive adjective] noun] is marked is also clear in the use of antithetical adjectives and in the figurative and predicative use of preposed adjectives. The unmarked nature of right branching, on the other hand, is further demonstrated in the use of the adjective borrowed from Greek and in forward gapping. 3.2.2.1.1 Antithetical Adjectives. Some adjectives expressing affiliation and definition carry the idea of antithesis. The opposition is explicit in pairs like iuvenilis-senilis 'young—old', equester-pedester 'equestrian—pedestrian', meretricius-nuptialis 'loose—marital', as in meretricios loose-Acc.pl.

amores amours-Ace.

nuptiis marriages-Abl.

conglutinas combine-Pr.-2sg. 'you combine their loose amours by marriages'

(Ter. And. 913)

The opposition is binary, and consequently often implicit, in examples like sinister versus dexter 'left—right', superior versus inferior 'superiorinferior', oriens versus occidens 'east—west', privatus versus publicus 'privatepublic'. In order to underline these antitheses the adjective in question is placed in first position in the noun phrase, either in implicit opposition, as in ad inferiorem partem fluminis to lower-Ace.sg. part-Ace. river-Gen. 'to the lower part of the river Rhine'

Rheni Rhine-Gen.

(Caes. DBG 1.1.6)

or in explicit opposition: Tu, ... You-Nom.

in sinistrum in left-Ace.sg.

cornum, corner-Ace.;

tu . . . in dexterum you-Nom. in right-Acc.sg. 'You ... to the left corner, you ... to the right' (Ter. Eun. 775)

(For more details on these adjectives, see Marouzeau 1922.)

70


3.2.2.1.2 Literal and Figurative Sense. The specific value of the preposed distinctive adjective is also clear in figurative use: the literal meaning of the adjective can change into a figurative sense when the adjective precedes the head noun. Preposing of the adjective thus corresponds to its symbolic connotation, as in viri Socratici 'Socrates's disciples', but Socratici viri 'men of virtue'. Or: genus Sullani regni sort-Norn. Sulla-Adj.-Gen.sg. reign-Gen.sg. 'a sort of Sulla's reign / Machiavellian polities' (Cic.Att. 8.11.2)

This process can also be observed in common phrases featuring nonderived adjectives (see also Modern French un pauvre homme 'a poor fellow' and un homme pauvre 'a man who is poor'): aquae dulcis '(of) fresh water' (Cat.Agr.104)

versus dukes voluptates 'sweet-flavored pleasures' (Cic. Fin. 2.30) vinum lene 'mild wine' (Cat.Agr.157.13)

versus leni animo 'with indulgence' (Ter. And. 262)

In line with this principle, preposing of the adjective can express a pejorative connotation, for example, in the use of servilis, which normally follows the noun, but precedes it in a pejorative context. These examples show the regularity with which postposing of the distinctive adjective corresponds to its neutral and basic sense. The symbolic or figurative meaning, on the other hand, is related to its preposing: the literal versus the figurative meaning of the distinctive adjective depends on its preposing (figurative sense) or postposing (proper sense). Consequently, postposing is the unmarked order of the distinctive adjective. The situation is the reverse in the category of the descriptive adjectives (Marouzeau 1922:53-56).


71

3.2.2.1.3 Predicative Use. A number of adjectives — medius, summus, and others — take a predicative value when they precede the noun; this pattern is an additional argument for the marked nature of preposing: media aestate 'in the middle of summer' (Cic. Pomp. 35) summus mons 'the top of the mountain' (Caes. DBG 1.22.1)

In the preceding discussion I set out to show that preposing of the distinctive adjective was marked. Two additional phenomena suggest that the distinctive adjective followed the noun in unmarked order: the place of the borrowed adjective and the syntactic process of forward gapping. Since speakers are less sensitive to words borrowed from a foreign language, the borrowed adjective normally follows the noun. See, for instance, Marouzeau (1922:93): sermonibus morologis words-Abl. extravagant-Abl.pl. 'with extravagant words' (Pl. Pseud. 1264)

When the Latin sentence features two identical nouns, both combined with an adjective, one of the substantival elements is often omitted. Marouzeau stipulates that generally the second noun disappears (1922:1718). This preference for forward gapping reveals an underlying RB structure. See Graeco Greek-Abl.sg.

ritu rite-Abl.

sacra, ceremonies-Acc.,

non Romano facere not Roman-Abl.sg. do-Inf. '(to) perform the ceremonies in the Greek, not the Roman fashion' (Varr. L.L. 7.88)

In this example, the emphasis of the preposed antithetical adjective, Graeco, as well as the process of forward gapping, show that postposing is the unmarked position of the distinctive adjective. 3.2.2.2 Descriptive Adjectives If in general the value of the distinctive adjective is relatively clear in a given context, the value of descriptive adjectives is less so. They are more expressive, and consequently, in contrast with distinctive adjectives, they display comparative and superlative forms. The descriptive adjective nor-

72


mally precedes the head noun (Marouzeau 1922:75). Yet postposition is not excluded and can be explained by "differences de valeur, d'emploi" (1922:76). Unmarked preposing of the descriptive adjective is important not only because of the situation in Latin but also because of later development: descriptive adjectives form a subcategory that over time continued to be left branched; for example, magnus 'big', longus 'long', and malus 'bad', (for statistical data on these and other adjectives, see Delbruck 1900; Marouzeau 1922; sec Walker [1918] for adjectives expressing quantity and dimension). The LB adjectives in Modern French form only a small group. They arc the last residues of the former Latin subcategory that gradually decreased in number. More and more these elements shifted to right branching. From Latin to French one observes thus a linear development: the LB adjective gradually gave way to right branching. Whereas the Latin preposed distinctive adjective expresses a judgment or a specific meaning, it is postposing of the descriptive adjective that expresses a distinctive value or an emphasis, for example, in res odiosa (Att. 14.13.2), where the predicatively used adjective is stressed: 'the business is unpleasant'. Likewise there is contrast in ingentem multitudinem

'a numerous force' (Sall.jug. 107.4)

gaudium ingens 'an immense joy' (Sail.jug. 55.1)

The increased expressivity of the postposed descriptive adjective matches the high incidence of comparatives, but even more so of superlatives in the totality of postposed descriptive adjectives (see the data of Delbruck 1900; Marouzeau 1922). In addition to emphasis, the length of the adjectival complement can account for the occurrences of RE descriptive adjectives: the adjective is postposed when the complement is composed of more than one adjective, of an adjective and its complement, or of an adjective with an adverb (cf. Marouzeau 1922:97): civitas magna atque magnifica city-Nom. big-Nom.sg. and sumptuous-Nom.sg. 'the great and sumptuous community1 (Sail. Cat 51.5)

3.2.2.3 The Adjective in Classical Latin: Conclusion Evidence presented in the preceding pages clearly illustrates that the distinctive versus descriptive value of the adjective was fundamental for the position of this element in Latin nominal phrases. The distinctive adjec-


73

tive, which is an objective modifier, was normally right branched, but preceded the head when marked; the descriptive adjective, on the other hand, normally preceded the head noun. This regularity can be accounted for by the subjective nature of this type of adjective. Alternatively, when the descriptive adjective in question was emphasized, it followed the noun. In purely synchronic perspective this seems to be a paradoxical situation. Yet the diachronic perspective can account for this apparent idiosyncrasy. Earlier, I pointed out that originally the adjective preceded the head in unmarked order. In marked order the adjective followed its noun. The change in branching first affected the distinctive adjectives, which subsequently followed the noun except when they were marked; the descriptive adjectives did not follow this change in a first step. This explains the somewhat peculiar situation we find in Latin: [Noun [Distinctive Adj.]] in unmarked order; AN in marked order [[Descriptive Adj.] Noun] in unmarked order; NA in marked order

Diachronic perspective thus reveals a shift from archaic left branching to right branching. The preposing of a subcategory of Latin adjectives can be considered an archaic feature and can be explained by its non-distinctive value. In postposition the adjective had a distinctive value, which shows that the sequence [noun [adjective]] was the unmarked order in Latin (for the unmarked postposition of the adjective, see also Coleman 199 la: 326). This conclusion is based on the comparison of independent though corroborating studies of the use of adjectives by individual authors or groups of authors. Finally, the Latin data show that the preposed adjective is not a Germanic innovation, as has often been assumed by way of explanation. By contrast, it is part of an intrinsic linguistic shift. The Germanic superstratum might have had a consolidating affect, but it did not trigger the change.

3.2.3 Vulgar Latin and Late Latin Although statistical data might sometimes suggest differently, the regularity observed in Classical Latin had not changed essentially in Vulgar and Late Latin. Analysis of word order in Petronius's work shows a preference for preposed adjectives: approximately 65 percent of all adjectives precede the head noun (Feix 1934:27). On the face of it, this tendency seems to indicate a retrograde shift toward unmarked preposing, but the data need to be specified. First Feix takes into account all adjectives as well as determiners, which he considers adjectives. In fact determiners tended to precede the noun (see section 3.3). As to qualifying adjectives, one observes that Petronius generally did not change fixed expressions and that the length of the complement was an important factor. Moreover, derived adjectives, hence distinctive adjectives, display an important regularity: in spite of a few exceptions, denominative adjectives follow the noun (Feix

74


1934:28). Exceptions can be accounted for, for example, in the figurative sense (Bauer 1992:102), and are thus marked. These considerations show that the position of the adjective was not random: the adjective followed the noun in an unmarked order. In addition to Petronius's novel, other works inform us about the place of the adjective in Vulgar and Late Latin, such as the letters of Claudius Terentianus, the Peregrinatio, and the writings of St. Augustine.

3.2.3.1 The Letters of Claudius Terentianus Claudius Terentianus's letters confirm the general tendency to postposed adjectives: the 67 instances of postposing clearly outnumber the 24 examples of preposing (Adams 1977b). The preference for preceding adjectives in his Greek letters (11 vs. 8) shows that there was no interference from this language and even more so it illustrates the importance of postposing in Latin. Moreover, preposing turns out to be marked: "[I]t is chiefly emphatic adjectives of size or quantity which occupy the marked position before the noun" (Adams 1977b:71). The majority of these adjectives are descriptive, for example, magnus 'big', maximus 'biggest', paucus 'few', and minor 'smaller'. These data are consistent with the evolution I pointed out.

3.2.3.2 The Peregrinatio Evidence from the Peregrinatio also confirms the tendency revealed in the texts discussed in these paragraphs. In general, distinctive adjectives denoting material follow the noun, as do temporal adjectives (Haida 1928:11). As to descriptive adjectives, the situation is more difficult. Postponing dominates; preposing, though frequent, is marked and expresses figurative or affective meaning (see Vaananen 1987:108). Haida's statistical data show, moreover, that adjectives that commonly precede the head in Modern French are the same as the ones in Late Latin (1928:4—12), even if grandis 'big' follows the noun (Vaananen 1987:108). Consequently, the instances of preposing of the French adjective have their origin in the earlier period. Adjectives that arc integrated in a more extended unit form an RB phrase. These are adjectives that combine with an adverb or adjectives that are part of an adjectival phrase. Plenus 'full', pertinens 'appropriate', and aptus 'appropriate', for example, and their complements follow the noun. See, for example, psalmus unus pertinens psalm-Norn. one-Nom. appropriate-Nom.sg. 'a psalm appropriate to the occasion'

ad rem to occasion-Acc, (10.7)

Diacbrank Analysis: The Noun Phrase

75

Data of this text therefore support the tendency I pointed out and foretell the situation in Old French.

3.2.3.3 The Writings of St. Augustine Analysis of the adjective in St. Augustine's work is interesting since it shows that quantitative criteria alone are not sufficient to determine the unmarked place of the adjective. Analysis of De CivitateDei, Confessiones, Sermones, Epistulae, and De Beata, Vita has revealed a numerical preference for LB adjectives (Muldowney 1937:23-28). Preposing indeed dominates, but it represents, nevertheless, the marked order that can be explained by emphasis, antithesis, contrast, or other stylistic effects. Aeternus 'eternal', for example, occurs most often in preposition "[because] in practically ever)' instance ... an emphasis or contrast of some kind is present and is evidently intended" (Muldowney 1937:38). Precise analysis of the data demonstrates that "pre-position is the more expressive, the more emphatic order, and therefore deliberately adopted by St. Augustine to suit his purpose" (Muldowney 1937:38). This same expressivity explains St. Augustine's preference to have comparatives and superlatives precede the noun (see Muldowney 1937:46). Moreover, the adjectives that precede the noun turn out to be a small group: often the same adjectives recur. Of the totality of 134 instances of temporal adjectives, for example, 100 are either aeternus 'eternal', sempiternus 'eternal', temporalis 'temporal', futurus 'coming', or cottidianus 'daily' (for these data, see Muldowney 1937:32). The hypothesis of unmarked postposing is further supported by the place of the descriptive adjective used with a distinctive value. Normally preposed (330 vs. 122), these adjectives switch to right branching when they are used with distinctive value: "[W]hen they serve to distinguish a class or category . . . they take the position as well as the force of determinatives" (Muldowney 1937:40-41). Finally, descriptive adjectives in combination with another adjective, with an adverb, a relative clause, a nominal phrase, or a prepositional phrase, normally follow the head noun: 97 of the 142 instances display postposing (see Muldowney 1937:46). Postposing of the adjective was thus the unmarked order.

3.2.4 Old French Adjectives may seem to have been relatively free elements in Old French: they could precede or follow nouns (Moignet 1973:345). Yet more precise data reveal a regularity that fits the earlier observed evolution. The statistical data of Moignet (1973) indeed show that the place of the adjective was not indiscriminately free: in La Queste del Graal (pp. 1—9) and La Mort Artu (cap. 1—25), most of the preposed adjectives (92 of the 108 instances) are part of the group granz, biaus, bons,gentix, haux, and so forth.

76


In general the preposed adjectives vary little indeed and they have a descriptive value, indicating the object or person "[dans sa] grandeur, valeur, puissance, bcautc, cadre affectif, rang social, age, saintete, legitimite" (Moignet 1973:345). In postposition, the adjective characterized the noun and attributed "une caractcrisation particularisante": it had a distinctive value (Moignet 1973:345). This difference reminds us of the Latin distinction between descriptive and distinctive adjectives. If indeed the nature of the adjective determined its position in the noun phrase, it is important not to restrict the analysis to numerical data, but to determine what adjectives were normally preposed. Mere statistical data would point out that the adjective in Old French was left branched. In the Quatre Livres des Rois, for instance, there is a clear preference (83 percent) for the AN construction (Haarhoff 1936:73). This order is also preferred in the Chronique de jean d'Outremeuse (Nissen 1943) and in Le Roman de Berinus (Lewinsky 1949). In Books I, III, and VI of the Chronique, 89.2 percent of the adjectives precede the noun (1203 instances of AN vs. 145 instances of NA); the percentage is 95.33 percent in Berinus (2270 AN vs. 118 NA), which is of the same period (see Lewinsky 1949:171). Comparing these data to evidence from earlier work, one observes an increasing number of preposed adjectives: see, for example, La Vie de Saint Leger (82 percent), La Vie de Saint Alexis (75 percent) (Volcker 1882:39), and the Chanson de Roland (63 percent) (Morf 1878, quoted in Volcker 1882:39). This high frequency of AN in novels of the fourteenth century seems to violate the tendency toward right branching. Yet although preposing was important in Old and Middle French, the actual number of adjectives that preceded the noun was limited. Analysis of the detailed data from Berinus and the Chronique (e.g., Lewinsky 1949:169-70; Nissen 1943:132-33) reveals an important regularity. Of the 3736 adjectives, 3473 are left branched, but lexically these represent only 26 adjectives. Moreover, of the 1203 examples of AN in the Chronique, 722 arc part of the subcategory of adjectives that always preceded the noun and that continue to do so in Modern French. In Berinus the corresponding numbers are 1693 of 2270 instances. This class of adjectives is nearly always preposed. Preposing was thus more and more limited to a small group of adjectives, the number of which was diminishing. This distinction is consistent with the Latin regularity pointed out earlier. Indeed, in Old and Middle French, it still existed: the postposed adjective expressed a differentiating and perceivable quality (Nissen 1943:134). Alternatively the preposed adjective expressed a specific value, a subjective appreciation (Nissen 1943: 135). In line with this evidence we can therefore assume that the unmarked place of the adjective in Old and Middle French was postposing. In time, the number of adjectives that could precede the noun—generally descriptive adjectives—diminished. The preference for adjectives expressing an appreciation and the little variation in highly frequent adjectives


77

account for the high incidence of left branching in Old and Middle French.

3.2.5 The End of the Middle Ages and Later At the end of the medieval period the situation had changed little: most of the adjectives followed nouns, and adjectives that could precede nouns were not numerous. These were especially descriptive adjectives and comparatives expressed by an inflected form (meilleur 'better,' pire 'worse', grandre/greignor 'bigger', mendre 'less'). Moreover adjectives denoting color and adjectives of the type loyal 'loyal', divers 'diverse', mauvais 'bad', merveilleux 'great', and petit 'small' were generally preposed but their postposing became less rare (Marchello-Nizia 1979:318). In general the old distinction between the distinctive and the descriptive adjective continued to exist: the descriptive adjective tended to precede the noun, whereas the distinctive adjective followed the noun most often, except when it expressed an opinion. In Modern French the number of adjectives that are left branched is very low: the former subcategory of descriptive adjectives that preceded the head in Latin now consists of only a small group. Other adjectives follow the noun in unmarked order in Modern French. If an adjective precedes the noun, it is highly marked and it has a non-distinctive value. Moreover, a number of adjectives sometimes follow, sometimes precede the noun, for example, un furieux menteur 'a pathological liar' and un menteur furieux 'a liar who is furious'. Yet the unmarked place of these adjectives is postposition; preposed these adjectives are non-distinctive and their meaning is extended (for a systematic analysis of these semantic processes, see Waugh 1977:86 et seq.). As long as French displays this kind of sequence—even in very reduced number and in highly marked use—the shift toward right branching has not been fully carried out. Since postposing of the adjective is an early change in Latin, the adjectival shift toward right branching is a long and complex development—a longer development than that of the genitive. In contrast with the genitive, which is more syntactic, the adjective is sensitive to subtleties .of meaning, hence its frequent stylistic uses. Moreover, one has to bear in mind that the shift from LB to RB genitive was paralleled by the shift from an LB case form to a RB prepositional phrase. Although analyticity did not trigger the syntactic change of the genitive, it definitely supported it. Alternatively, the adjective, being a noncomplex element, can more easily precede the noun than a complex subordinate element, such as a prepositional phrase (see chapter 6); see also, for example, the English LB adjectives and synthetic genitives, which are residues of OV typology in an otherwise RB language. These considerations may well explain why the adjective is one of the very few elements that occasionally are left branched in Modern French syntax, which strictly applies right-branching principles in other phrases.

78


3.3 Determiners In the first two parts of this chapter I showed the gradual shift toward right branching of the genitive and adjective, which are nominal complements. In the following pages I discuss this change in the determiner, which is a grammatical element. The term "determiner" refers to the possessive pronoun, the demonstrative, the article, the indefinite, the interrogative, and the cardinal numerals. Although they form a varied group, determiners are characterized by their status of head (see chapter 2): they express the grammatical features of the noun—gender, number, case, and indefinite/definite nature. Because of this grammatical function the determiner differs from the adjective and the genitive, which are elements that are governed by the noun and thus have a different syntactic status. In the following pages I focus on possessives, demonstratives, and numerals. Since several stages of their development have been analyzed in earlier studies, these are the determiners that lend themselves best to the analysis of branching.

3.3.1 Modern French Versus Early Indo-European Dialects In Modern French, determiners normally precede the noun: the noun is right branched. This type of branching has been the rule ever since Old French, even if postposition of the possessive was at that time not completely excluded: the article, the possessive, and the demonstrative preceded the noun (Moignet 1973:343). Cardinal numerals also preceded the noun. Moreover, possessives generally were preceded by a definite or indefinite article, which strengthened right branching (see Moignet [1973: 343—45] and Foulet [1923:117—29] for numerous examples of possessives and demonstratives; for right-branching determination in Middle French, see Marchello-Nizia [1979:305-15]). This right-branching determination in French is the result of a long and complex evolution, which also includes the development of the definite article from demonstratives. In the history of Indo-European languages, determiners arc late, which explains why in the most ancient IndoEuropean languages, Vedic and Hittite, "demonstratives are infrequent; nouns . . . have no accompanying determinative marker unless they are to be stressed. The demonstrative then precedes" (Lehmann 1974:82). Therefore, when stressed, the determiner preceded the noun. Instead of a possessive, Hittite displayed a personal pronoun in the genitive. Early Hittite, however, had a clitic possessive element that later survived in assimilated form "in neuhethitischer Zeit" (Friedrich 1974: 66). The old possessive structure in Hittite, which is the oldest attested Indo-European language, was thus left branching. Finally, numerals in Indo-European usually preceded the nouns (Brugmann 1922:685).


79

Data on the evolution of the determiner are very diverse. On the one hand, the situation in French is clear and unambiguous: determiner structures are right branching. In the oldest Indo-European periods, on the other hand, determination is far from being a fully developed system. Moreover, analysis of the demonstrative is especially tricky because its use implies, more than that of the other determiners, emphasis. Evidence suggests, however, that the possessive normally followed the noun, whereas the demonstrative preceded it (for resemblances to other LB languages, sec section 3.3.2). In the following pages I analyze the place of the possessive, of numerals, and of demonstratives in Latin and the development they underwent. I then relate these data to evidence from two other strict LB languages in an attempt to demonstrate the particular situation of the demonstrative.

3.3.2 Latin: An Intermediary Stage? The possessive seems to be the most stable determiner in Latin: it normally followed the noun (Marouzeau 1922:133; see also Leumann and Hofmann 1928:617). Yet statistical data vary from author to author and even from work to work. In Varro's Res Rusticae, for instance, postposing dominates, but the dialogues of Plautus and Terence show a preference for preposed possessives (Marouzeau 1922:147). Only the analysis of the context made it possible to sort out the differences in use: Marouzeau indeed pointed out that the preposed possessive carries the emphasis of adherence, affection (first person), politeness (second person), and so on (see also Muldowney 1937:61-62; Feix 1934:30). The marked nature of this organization is also clear in the habit of preposing this clement when it is in the intensified form, featuring the suffix -met or -pte (Marouzeau 1953:15): meamet culpa

'my own blame' (Pl. Poen. 446)

tuopte your-own-Dat.

tibi you-Dat.

consilio plan-Dat.

occludunt linguum tie-Pr.-3pl. tongue-Acc, 'they tie your tongue using your own plan' (Pl. Mil. 605)

Later, from Nepos and Vitruvius on, from the first century A.D., nonemphatic preposing spread (Muldowney 1937:60, 63).

3.3.2.1 The Numeral In unmarked order cardinal numerals preceded the noun: as a simple enunciative expressing "un chiffre, connu, banal" (Marouzeau 1922:190;

80


see also Leumann and Hofmann 1928:617; Feix 1934:31). On the other hand, when the numeral contained important information or when it was used for explanation, evaluation, comments, order, measures, and so on, it followed the noun, for example (Marouzeau 1922:192 et seq.): divisa in panes tres, quarum divided-Norn.sg. in parts-Acc, three, of-which-Gen.pl. 'divided in three parts, (one) of which . . .' (Caes. DBG 1.1.1.)

3.3.2.2 The Demonstrative In section 3.3.1,1 mentioned the particular status of the demonstrative in PIE: its use was by itself emphatic. In Latin, which has no articles, it is difficult to evaluate the use of the demonstrative, because its deictic function already implied an emphasis. Moreover, since the demonstrative gradually developed and ended up as a definite article, the distinction between its demonstrative and definite use is occasionally difficult to establish. The demonstratives—hic, iste, ille—"assez naturcllement" followed the noun (Marouzcau 1953:17). When emphasized, they preceded their complement. Yet thirty years earlier, Marouzeau had concluded exactly the opposite: he then argued that the demonstrative generally preceded the noun (1922:155; see also Leumann and Hofmann [1928:616] and Muldowncy [1937:73 et seq.], who support this idea). These opposite views illustrate the difficulty of the question. Statistical evidence indeed corroborates the hypothesis of unmarked preposing: in De Bello Gallico (640 vs. 4), in Cicero's speeches (486 vs. 28), in St. Augustine's works (596 vs. 84), in Aulus Gellius (1427 vs. 453), and in Petronius's work, the proposed demonstrative clearly dominates (see Muldowney 1937:73; Feix 1934:29-30). In a detailed analysis of the use of the demonstrative by a large number of classical authors, Fischer (1908) concluded that preposing was the unmarked order of the Latin demonstrative (hic, ille, iste, ipse, idem, and is, which was not a proper demonstrative, but could be used as such). Fischer based his conclusion not only on quantitative data (1908: passim), but on the marked as well as unmarked use of the demonstrative. He pointed out that the reverse order, [[noun] demonstrative], only occurred when the noun was emphasized. Alternatively, when the demonstrative carried emphasis, the order remained right branching. This can be demonstrated, for instance, by its antithetic use. When two demonstratives are in contrast, they precede the noun; yet when two nouns are opposed, the demonstrative follows: illo that-Abl.

miserrimo tempore . . . ; miserable-Sup.-Abl.sg. time-Abl.

neque nor

Diachrmic Analysis: The Noun Phrase

81

hoc tempore this-Abl. time-Abl. 'during the most miserable period. . . , nor the present time' (Cic. Fam. 13.71)

But magis more

me me-Acc.

iudicio trial-Abl.

hoc this-Abl.

quam than

morte inimici laetatum death-Abl. enemy-Gen. rejoice-PartPf.-Acc.sg. '(that) I rejoiced more in this trial than in the death of my enemy' (Cic. Fam. 7.2.2)

Similarly, the demonstrative follows the noun when referring to a person, or an object to which one wants to draw attention in a presentation, a definition, and so forth. This indeed is emphasis of the noun (Marouzeau 1922:160 et seq.). Fischer's conclusion is further corroborated by evidence from Oscan and Umbrian, in which the phrase [demonstrative + noun] was right branching (Konneker 1972:92). Moreover, diachronic data provide additional evidence in favor of the hypothesis of unmarked preposing of the Latin demonstrative: the gradual and slow development from demonstrative to article, which is a preposed element in French, suggests that the article was the result of a gradual change in meaning, not of a structural reversal. It is indeed often difficult to determine the exact value of the element in question. When Me and is precede a noun that is known to the speaker (see also Muldowney 1937:81), they very much resemble an article, whence the typical preposing of is in this context (Marouzeau 1922:149). Gradually this use expanded as we can see in the Vulgata, where the preposed demonstrative with the value of an article is very common (Ronsch 1875: 419-24). Although Latin demonstratives most often preceded the noun, they had a left-branching morphological structure: iste, for example, is composed of the particle is and the demonstrative -te, the exact value of which (remoteness or proximity) is not clear; in Me the clement l, the first or the second, marks remoteness; finally, in the paradigm of hic we find is, related perhaps to Sanskrit ayam, which in Indo-Iranian denotes the closer object. In hic this "demonstrative" is followed by the epideictic particle -c, that is found also in ecce 'see' (e.g., Ernout and Meillet 1959; Meillet 1964:325-27). Before Plautus this particle was more manifest in the paradigm: for example, harunc or horunc in the genitive plural. In a later stage it again became very important under the form of ecce. This element preceded the former Latin demonstratives to create French right-branching demonstratives: ecce + hoc > ce; ecce + ille > cel; ecce + iste > cist. Therefore, on the morphological level as well we observe a shift toward right branching.

82


3.3.3 Evidence from Turkish and Japanese Preposing of the demonstrative is exceptional in a basically LB language, as were Latin and other early Indo-European languages. Since Japanese and Turkish represent the strict variant of an LB language (see also chapter 6) comparison of their data may be relevant. Like Latin, Turkish and Japanese display a demonstrative "triad" denoting (1) proximity, (2) remoteness, near the interlocutor, and (3) remoteness, near a third person. In Turkish this organization is even reflected in the etymology of the forms, which are closely related to the singular personal pronouns, cf. the personal pronouns ben 'I', sen 'you', o 'he', and the demonstratives bu "das beim Sprecher Befindliche," u "das beim Angeredeten Befmdliche," and o "das bei einem Dritten Befmdliche" (Kissling 1960:133). In Japanese the three elements are kono 'this', sono 'that', and ano 'that over there' (Hinds 1986:137). In spite of the left-branching structure of the language, Turkish and Japanese demonstratives precede the noun. Only Turkish has a number of idiomatic expressions featuring a postposed demonstrative, but it is not certain whether this is an archaic structure or not (see Lewis 1967:71). The Turkish possessive, on the other hand, has the form of a declinable particle that follows the noun: it comes after the noun and its form varies according to number, person, and case; see, for example, kiz 'daughter', kizim 'my daughter', kizimin 'of my daughter' (Kissling 1960:29—42). In Japanese the situation is different and is of little use for our analysis of possessives. Because of the lack of pronouns, Japanese has no proper possessive pronoun: instead, the genitive particle no is added to the noun that has a pronominal value. According to the left-branching principles of the language, this form, which is a genitive construction rather than a possessive, precedes the noun (Hinds 1986:136), as in anata -no you -Particle-Gen. 'your daughter'

musume daughter

Evidence from Turkish and Japanese reveals that even in these strictly left-branching languages the demonstrative precedes the noun. Preposing of the Turkish, Japanese, and Latin demonstrative is difficult to account for, the more so since few data exist from earlier periods. The formal correspondence between a number of Indo-European languages (see Meillet 1964:326 ct seq.) suggests that demonstratives were one of the earlier determiners. This chronology rules out the explanation that suggests that the emergence of the demonstrative coincided with the shift toward right branching. Moreover, if this hypothesis were true, the possessive, which was created later, should have been right branching also. There may be another factor that can account for preposing of the demonstrative. The occurrence of a demonstrative marks an emphasis ipso facto. It is thus possible that markedness of this element accounts for its


83

right branching in a basically left-branching language. This assumption is supported by the fact that the demonstrative occupied the same position in marked or unmarked context. Yet in line with this reasoning, demonstratives should be left branching in Modern French, and this is not the case. Therefore, there is not enough evidence to substantiate this hypothesis.

3.4 The Place of the Determiner, the Adjective, and the Genitive The sequences in Modern French [noun [adjective]], noun [[noun-complement]], and [determiner [noun]] are perfectly consistent, because they all arc right branching. Yet the analysis of the determiner is difficult, especially from a typological point of view. First, typological data for an important number of languages show that the combinations we find in the Romance languages are exceptional and that they seem to violate typological regularities (see Hawkins's inventory of typological data 1983:320— 42). Generally, typologists relate the place of the adjective and the demonstrative, for example, Greenberg's Universal 18, which reads: When the descriptive adjective precedes the noun, the demonstrative, and the numeral, with overwhelmingly more than chance frequency, does likewise. (1963:68)

More recently, Hawkins has reformulated the relation between the place of cardinal numerals and the demonstrative and the place of the adjective: when the demonstrative or the numeral follows the noun in a prepositional language (a language that features prepositions), the adjective does likewise (1983:82, 109). Postposing of the adjective, on the other hand, does not presuppose a specific organization (see Hawkins 1983:71). Typological data also deny Lightfoot's hypothesis according to which "demonstrative specifiers of N' should occur on the opposite side of N to genitive complements, producing DemN & NGen or NDem & GenN" (1979, discussed in Hawkins 1983:194). Although Lightfoot's hypothesis is integrated in the X-bar framework, according to which the specifier is not the head of a phrase (see also chapter 2), the application of his principles by Hawkins might reveal a possible correlation between the place of the demonstrative and the place of other elements in syntactic phrases. Hawkins's evidence shows that the place of the demonstrative is neither related to the branching of the verbal phrase featuring an auxiliary, nor to the organization of a nominal phrase featuring a genitive. Only relatives display some correlation: the combination DemN and NRel outnumbers DemN and RelN (46 versus 32; Hawkins 1983:193). These data clearly show that the use of the demonstrative is not determined by strict rules; its irregularity is indeed an accepted fact in linguistic literature. In an attempt to reformulate Greenberg's Universals through observations of a larger data corpus, Hawkins noticed a hierarchy in the stability of the elements of a nominal phrase in prepositional languages: "[T]he

84


demonstrative and the numeral are more unstable than the adjective, the adjective is more unstable than the genitive, and the genitive is more unstable than the relative clause" (1983:75). This Prepositional Noun Modifier Hierarchy accounts for the high frequency of sequences demonstrative —noun in prepositional languages; preposing of the demonstrative, which Hawkins considers a modifier, is more frequent than preposing of the adjective, and so on. The relative clause is at the base of the hierarchy because it nearly always follows the noun in prepositional languages. Alternatively, it often occupies a deviant place in postpositional languages (see also chapter 6), whereas the demonstrative precedes the noun systematically in this type of language (Hawkins 1983:96). Hawkins ascribes these characteristics to the Mobility Principle and the Heaviness Interaction Principle. These principles account for the mobility of "light" elements and for the tendency of heavy elements to follow the head: "[T]he quantities of preposed and postposed noun modifiers are therefore a function of both heaviness and mobility, operating in conjunction with the basic noun modifier ordering" (1983:97). One has to consider, however, the scope of these principles: what structures arc determined by this "basic noun modifier ordering" and what are the criteria? And if indeed the length of the elements is a factor important enough to interfere with the "basic ordering," why does it not completely determine the ordering principles? The unsuccessful attempts of Lightfoot as well as Hawkin's typological data and discussion of the irregularities illustrate the complexity of the problem of the determiner, and of the demonstrative in particular. Currently we lack typological data to solve the problems that have been raised in this analysis. The lack of information is probably related to the characteristics of the element under consideration: the history of Indo-European shows that the determiner was late and that it underwent important changes. Its typological analysis requires therefore that the exact role and value of the element in question be determined. Most often the data required are absent in typological analyses, which moreover often include exotic, hence poorly known languages. As a grammatical element, the French determiner is the head of the determiner—noun sequence. Since the nominal phrase is introduced by the determiner and the adjective and noun—complement are placed to the right of the noun, the French nominal phrase is right branching. This is the result of a complex evolution, the major modifications of which had already taken place before the French period.

4 Diachronic Analysis: The Verb Phrase

In the detailed analysis of the position of the Latin genitive, adjective, and determiner I have demonstrated that the noun phrase underwent a structural reorganization whereby left branching gave way to right branching: comparison of data of various periods of Latin and even earlier reveals the gradual shift from archaic LB to modern RB structures. The changes that occurred in the Medieval French nominal phrase were but the accomplishment of that same evolution. The verbal phrase, which is analyzed in this chapter, underwent the same change, but at a different rate. Its structural reorganization extends as far as modern times: the relative clause featuring a final, hence LB, verb only disappeared in the seventeenth century, whereas even today the French verb still displays a left-branching future, for example, je chanterai 'I will sing', although this form is giving way to the right-branching je vais + infinitive, as in je vais chanter. The present chapter presents an analysis of the place of the verb in PIE and the development it underwent in the Italic languages and afterward. I demonstrate that the late changes mentioned illustrate the delay of the verbal phrase with respect to the noun phrase, which underwent the shift toward right branching much earlier. The delay we observe in the Italic languages contrasts with the chronology proposed for German and English by Lehmann (1971) and Vennemann (1974), who argue that the verb plays a decisive role in the reorganization of word order typology. Studies of Latin word order for the most part focus on the three positions of the verb—final, initial, and medial—without taking into considcr85

86


ation the nature of the other elements of the sentence. This purely distributional approach leaves out the syntactic nature of the context and does not take into account that the verb is the head of its phrase. On the face of it this traditional distinction does not seem to be useful for the analysis in terms of branching. Yet as long as one deals with the final or initial position, the restriction does not interfere with the current analysis, because the elements that depend on the verb automatically precede or follow it. This leaves the medial position: a detailed analysis of the studies in question can inform us, however, on the syntactic context of the medial verb, hence on the branching of the verb phrase. As a first step, I discuss the unmarked position of the verb in Latin and in the Pre-Latin period (section 4.1). I then analyze the initial position of the verb, which is clearly marked (section 4.2). The notion 'initial verb' places the verb either in the very first position of the clause, or in the position following a initial sentential adverb. Third, I analyze the medial position of the verb, for which the nature of the context is of primordial interest given the Romance perspective of this study (section 4.3). In contrast with the preceding chapter, I first analyze the various positions of the Latin verb and only then set out to analyze the verb in Old and Middle French (section 4.5). Finally I discuss the morphology and the syntax of the adverb in the verbal phrase (section 4.6).

4.1 The Unmarked Place of the Verb and Its Development In the following pages I discuss the unmarked place of the verb in PIE and in the Italic languages. Although the place of the PIE verb is a matter of ever ongoing discussions, there are convincing arguments that support the hypothesis that the verb had an unmarked position, which the Italic languages inherited.

4.1.1. PIE According to some linguists, word order in PIE was free. Leumann and Hofmann, for instance, argue that the verb had no fixed position and that it occurred in initial and final position (1928:613). Yet Latin, which is also supposed to have had free word order, clearly shows that one has to be very careful about these assumptions. Alternatively, Delbruck concluded his chapter on the place and the accentuation of the Indo-European verb, arguing that "Das Verbum stand in dcm unabhangigen Aussagesatz am Ende und war schwachbetont" (1900:83). In addition to this unmarked position the verb occasionally could take another position: when it was emphasized, it occurred in initial position and was heavily stressed (Delbruck 1900:83). We find this order, for instance, in imperative clauses. Seventh-five years later Lehmann supports Dclbruck's point of view,

Diachronic Analysis: The Verb Phrase

87

arguing that "whatever the varieties, the normal unmarked form of the PIE sentence has the order (S)OV'' (1974:114). In his comparative analysis of various phrases in early Indo-European dialects Lehmann had shown that PIE was indeed OV in type (1972d). Delbruck's hypothesis had been confirmed earlier by Watkins (1964). Whereas Dclbruck limited his analysis to the verb and the direct object, Watkins has taken into account four elements: the sentence connective, the pronominal clitic element, the preverb, and the finite verb. Analysis of these elements in a number of ancient Indo-European languages—Vedic, Hittite, Latin, Greek, and Old Irish— revealed data that are important for the current discussion: they inform us on the reconstruction of the original VP structure and its development, which turns out to be parallel in the five languages under consideration. First, one observes that the structure of the verbal phrase is basically identical in the five languages under consideration. Since the structures in these languages cannot have had mutual influence, it is legitimate to assume that they were inherited from the proto-language (Watkins 1964:1041). On the basis of the comparison of these ancient structures, Watkins has proposed that in PIE (1) the simple verb was final in unmarked order, but initial in marked order and (2) the verbal element of the compound verb (i.e., verb and preverb) was always final, directly or not preceded by the preverb (1964:1041). These structures "repeat themselves with extraordinary regularity in the ... IE languages under consideration" (Watkins 1964:1037), even if in some of the languages their occurrence is stricter than in other (for example, in Hittite), or the beginnings of the subsequent development are manifest (for example, in Greek). These data suggest, indeed, that the unmarked position of the verb in the proto-language was clause-final. This assumption is supported by the fact that in the five languages under consideration the verbal element of the compound verb always occurred at the end of the clause. In addition to a common structural origin, Watkins's data suggest that the languages under consideration all roughly evolved in the same direction independently from each other. This evolution includes the loss of the dominance of the verb in clause-final position and the process of univerbation that was accomplished for Greek in the pre-Myccnean period and for Latin on the eve of the historical period. Consequently, well-established and independent empirical data support the hypothesis of the existence of an unmarked word order in PIE, that is, ... V#: the verbal phrase was left branching in unmarked order. Recent research on the active origin of Proto-Indo-European syntax further supports this conclusion because active languages favor SOV, or in my terminology, left branching (see Klimov 1977; Karl Horst Schmidt 1979; Lehmann 1990). Among those who advocate a dominant word order in PIE are some who assume unmarked VSO order. According to this hypothesis VSO gave way to SOV, which was replaced with SVO in modern languages (for instance, Miller 1975). This hypothesis is consistent with the cyclical con-

88


ception of language change assuming VSO SOV SVO VSO (for instance, Vcnnemann 1974). This presupposes an alternation of structures that are symmetrically opposed: right-branching VSO is replaced by leftbranching SOV, which eventually gives way to SVO, which is once more right branching. In this perspective the occurrence of negation and other particles in sentence-initial position is considered to be the indication of an earlier unmarked order #PV . . . #. In the assumed typological shift "the verb was moved to the end of the sentence leaving its particle stranded in initial position" (Miller 1975:37). Accordingly, the occurrence of the initial verb in Indo-European is then considered to be a residue of the former unmarked VSO order (see Miller 1975:35). Yet Watkins's data deny categorically the unmarked nature of the initial position of the compound verb. And if indeed the simple verb in initial position were a residue of a former unmarked word order, then we must be able to find examples of compound verbs occurring in the same position. The verbal element, however, is always in final position. In addition, the hypothesis of particle movement is based on the idea that in the course of time the compound verb split into two distinct elements (hence underwent a process of disintegration). Yet Indo-European data (e.g., Watkins 1964; Chantraine 1953 for Greek) reveal not a process of disintegration but of univerbation: the verb and the particle, which were originally distinct elements, became a compound form in the course of time. Consequently, instead of being posterior, as proposed by Miller (1975), the sequence #(...) P ... V# precedes both the unmarked #... PV# and the marked #PV ...#. Therefore, the arguments on which Miller based his hypothesis are ill founded. In 1976 Watkins characterized the typological discussion of the OV/VO nature of PIE as "purely and simply a pseudoproblem: . . . the facts are such, . . . that both word order patterns, marked verb-initial and unmarked verb-final, must be reconstructed for the proto-language, IndoEuropean itself (1976:315-16). Although Latin reflected basically the ancient structure of the proto-language, it introduced in the course of time a new structure, which was RB: the medial verb (see section 4.3). Yet before discussing these Latin data, I first analyze the position of the verb in two other Italic languages, Oscan and Umbrian, and in Old Latin.

4.1.2. The Verb in Oscan, Umbrian, and Old Latin The position of the verb in Oscan and Umbrian corroborates the analysis of the position of the verb in Old Latin. Generally, however, the place of the verb in Italic languages has not been analyzed in detail. Poultney, for instance, noticed that in the Iguvine Tables, the verb "commonly stands at the end of its clause. . . . But there are many exceptions (1959:155). He did not specify the exceptions to this regularity, however. The analysis of Konneker (1975), on the other hand, is more detailed.


89

She demonstrated that in Oscan, Umbrian, and Old Latin "the verb positions characteristic of Proto-Indo-European are maintained. That is, in both main and subordinate clauses the verb normally follows its object and appears last in its clause" (1975:367). Analysis of all dated Oscan inscriptions reveals, indeed, that the verb phrase was LB: the direct object and the infinitive or participle (in structures featuring an auxiliary) precede the finite verb. This organization is consistent with the left-branching verbal morphology in Oscan. The first examples of VO are later than 200 B.C. and they only occur in main clauses. Moreover, it turns out that the tendency toward left branching is stronger in subordinate than in main clauses. This is a general tendency that can also be observed in Umbrian (Konneker 1972:92, 1975:370), in Latin, in Medieval French, and in other languages. Accordingly, we can conclude that in Oscan and Umbrian there is "no fundamental difference from the Latin order, the resemblance being closest with the style of early prose such as that of Cato or the inscriptions" (Buck 1979:223). This assumption is supported by ancient law texts, which are nearly without exception verb-final. The Twelve Tables, the S. C. de Batchanalibus (see also chapter 3), and the text on the Columna Rostrata, from the end of the Second Punic War, show that Latin continued the Proto-Indo-European organization of the verb phrase: the final verb was the unmarked order (see Watkins 1964:1039; Adams 1977a:92). The OV order occurs almost without exception. This tendency is confirmed in the use of prepositional groups with adverbial value, which in a overwhelming majority precede the verbal head (36 to 2); see Adams (1977a:91). Comparison of data from Latin and from the two other Italic languages presents corroborative evidence: the oldest Latin texts show indeed that the verb in final position was unmarked and that "the older the text, the more regular this appears" (Watkins 1964:1039). Consequently, Latin had inherited the archaic organization of PIE displaying the unmarked final and the marked initial verb. In section 4.2 I demonstrate that the initial verb in Latin is related to given syntactic contexts or that it occurs in less objective texts, which more easily express stylistic differences: it represents the marked order.

4.1.3 From Old Latin Onward Already in Plautus's writings the dominance of the final verb is no longer exclusive: although OV still clearly dominates, the sequence VO also occurs and dominates even in the first five hundred lines of the Miles Gloriosus. In this same text the verb phrase featuring a prepositional phrase with adverbial value also tends to be right branching (87 occurrences of V-PP vs. 78 instances of PP-V) (Adams 1977a:90-98). In spite of these minor exceptions to the common order in the Miles Gloriosus, the verb in final position still clearly dominates in Plautus: 234 (59 percent) versus 163 (41 percent). These LB verb phrases contrast with

90


the RB tendencies of the noun phrase in the writings of this same author: the genitive and the adjective follow the head noun, whereas the object displays a preference for preceding the verb. Although Adams argues that "the evidence . . . does not explicitly support Lehmann's view that the position of the object is crucial in determining the position of other limiting elements" (1977a:98), he tries to find indications of a VO order in spoken Latin of the period of Plautus. Evidence of this VO order is present, according to Adams, in the high frequency of inversion in the sequence [passive verb—subject]. From a psychological point of view this order indeed corresponds to the order [verb [object]]; Adams accordingly concludes that the shift from OV to VO must have taken place before Plautus and that "in spoken Latin of the informal varieties VO was already established as the unmarked order" (1977a:97). In order to enhance the literacy of his writings, Plautus, it is thought, preferred OV, which was the "prestige order" (Adams 1977a:97), but he gave himself away in passive constructions This concern for literary prestige is amazing when it comes from this famous Roman comedian, who was well known for his everyday language and who wrote for the people. Moreover, it is legitimate to wonder why Plautus limited his literary concerns to the verb (compare the high frequency of postposing in noun phrases). In addition, it is important to emphasize that the literary texts of Classical Latin display the same difference: the verb is still in final position, whereas the noun phrases already feature right branching. Consequently, the LB nature of the verb phrase in Plautus is not a stylistic feature, but part of the linguistic evolution. The delay between the LB verb phrase and the tendency toward right branching in a number of other phrases can be observed in all periods and also in other Italic languages. In 1975 Konneker had already invalidated the implicit chronology— based on the fundamental role of the verb—in the hypotheses of Lehmann (1971, 1972c) and Vennemann (1974). The left-branching nature of the Italic verb phrase indeed contrasts with the right-branching pattern of the nominal phrase, in which the adjective, for instance, was right branched in Oscan and Umbrian (Konneker 1975:367) and in Latin (see chapter 3). For the early shift toward right branching in the prepositional phrase, see chapter 5 and Bauer (in press a). The unmarked final verb remained the basic structure in Classical Latin; it was the most common order and it was used without particular value or emphasis (Marouzeau 1938:47). Marouzeau's observation is supported by every analysis of the Latin verb that preceded his (Kroll 1912; Lindc 1923; Perrochat 1926) and that followed his (Watkins 1964, 1976; Fankhancl 1938; Adams 1977a; Elerick 1989a, 1989b, 1989c, and forthcoming). In Classical Latin the final position is most frequent in the writings of Caesar, to whom Linde refers as the "Fanatiker der E[nd]st[ellung]" (1923:154). Since Caesar's style is sober and neutral—hence his use of the third person singular to refer to himself—the high incidence of verb-final is supplementary evidence for its unmarked nature. Elcrick's recent studies confirm the high incidence of the final verb in


91

Caesar's prose as well as its unmarked nature. His analysis of the De Bello Gattico 1—7 and Bellum Civile reveals that in main and adjectival clauses featuring a subject, a direct object, and a finite verb "85% . . . show the SOV [360 examples] order or the alternate OSV [120 examples]" (1989b:l; sec also forthcoming). The other statistical data are SVO, 21; VOS, 27; VSO, 6; OVS, 33 (Elerick, (forthcoming:7). Whereas the order SOV "is clearly dominant and neutral," OSV is always motivated: "[A]ll instances ... are non-neutral linearizations showing the motivated leftward displacement of O. Although this order is non-basic and non-neutral, it is part of the dominant SOV linearization pattern of Latin" (Elerick 1989b:2). These data indeed show not only that OSV was a marked order in Latin, but also that the object was emphasized by extrapolation in initial position. Although the verb in final position was fundamental, its frequency diminished with time as shown by statistical data presented by Linde (1923), Haida (1928), Leumann and Hofmann (1928), Adams (1976, 1977a, 1977b), and Vaananen (1987). These independent analyses corroborate and reveal first of all that postposing of the verb dominates in Classical writers in main as well as in subordinate clauses. We observe this tendency in the period covering Classical writers and including St. Augustine. Yet, although the final verb clearly dominates, its dominance decreases with time: the written, even literary sources of Latin show a decline of the verb in final position. This observation is an important argument in the discussion on the alleged gap between written and spoken Latin and sustains the point of view of Mohrmann referred to in the introductory discussion (see chapter 1): both variations of the same language turn out to have undergone the same development albeit at different rates. Data from the detailed analyses mentioned show, moreover, that the decline of the final position of the verb was much less evident in the subordinate clause—where OV continued to dominate—than in the main clause. This gap but increased in the course of time: with only one exception, the incidence of the final verb in the subordinate clause extends 60 percent, whereas this percentage goes down to 33 percent in main clauses. Consequently, the shift from SOV to SVO has always been quicker in the main clause than in subordinate clauses. Moreover, although the occurrence of OV in subordinate clauses is only 37 percent in the Peregrinatio, half of the examples of [[object] verb] in the first six chapters appear in subordinate clauses. Gradually, the use of SOV was limited to this type of clause (for the parallel development in German, see Wells 1985:253; Bauer in press a). Consequently, for the analysis of word order of Latin, Old French, but also Germanic and other languages, it is extremely important to determine the syntactic nature of the clause, which has often been neglected in the Latin studies, although the final verb was more frequent in the subordinate clause, as Linde correctly observed (1923:154). The conservative tendency of the subordinate clause is confirmed in

92


Perrochat's analysis (1926) of the second book of Tacitus's Historian: in 86.6 percent of the subordinate clauses the verb is in final position. His study, moreover, shows that this conservative tendency varies according to the type of subordinate clause. The occurrence of OV is least frequent in subordinate clauses that are not marked as such by a relative pronoun or a conjunction, but rather by a subjunctive, for example. On the other hand, in Classical as well as Late Latin, the tendency to use OV is most manifest in relative clauses. Although in the Peregrinatio the initial verb dominates in relative clauses (307 instances of V-initial vs. 153 instances of V-final) and V-final is sometimes more frequent in other types of subordinate clauses, nearly half (40 percent) of the subordinate clauses featuring a final verb are relatives (for data, see Haida 1928:34). The relation between the degree to which left branching is maintained and the nature of the subordinate clause is also interesting in a diachronic perspective, because the analysis of Old French reveals the same tendency: in Old French as well the frequency of the verb in final position is related to the type of subordinate clause (see section 4.5). In addition to frequency and markedness, other syntactic phenomena support the hypothesis of the unmarked final verb in Classical Latin: gapping and the place of the verb in two coordinated main clauses. Elerick's analysis of Cicero's speeches revealed that "gapped or elided sentences . . . are overwhelmingly SOV, which I claim to be the unmarked order, or have the common alternate OSV order" (Elerick 1989a:569). Consequently, whereas in Modern French, a consistent RB language, forward gapping is common, in Cicero's prose "leftward gapping OV structures [are] . . . the rule" (Elerick 1989a:570). This regularity points to a left-branching verbal phrase in Latin. In addition to backward gapping, we observe in Latin interesting combinations in verb position when two main clauses are coordinated in one sentence: four combinations are possible; the most common, however, is the one where the verb appears in final position in both clauses (Leumann and Hofmann 1928:614; and, earlier, Lindc 1923:164-65): #. . . V-final, . . . V-final# #V-initial . . . . . . . V-final# #. . . V-final, V-initial . . .# #V-initial . . ., V-initial . . .#

common frequent in Petronius frequent in Petronius and Bellum Africanm very uncommon

These data not only reveal that the unmarked position of the verb was final; they also strongly suggest that the verb in initial position is definitely marked. This structure is analyzed in the following pages.

4.2 The Marked Position of the Verb: The Initial Verb Independent word order studies of works of various authors present corroborative evidence and demonstrate the unmarked nature of the final


93

verb: they briefly discuss the evidence of the final verb to focus more on the medial and even more on the initial position, which is the marked order. The occurrence of the verb in initial position was motivated either by stylistic or syntactic reasons. The stylistic motives are complex and pluriform; Marouzeau (1938) distinguished between the emphasis of the verb, on the one hand, and the emphasis of the utterance (pragmatic function), on the other hand. Syntactically, the occurrence of the initial verb was often related to a given context, for which we find parallels in other Indo-European languages.

4.2.1. Stylistic Motivation Prcposing of the verb violated the common word order, Marouzeau argues, and is meant to indicate the exceptional nature of the sentence (1938:82). In the following pages I again take up the distinction proposed by Marouzeau between emphasis on the verb as opposed to that on the utterance, because it reveals the undoubted markedness of the initial verb and thus points out the basic difference between the initial and the medial verb: the verb in medial position was never motivated by the emphasis of the verb proper. 4.2.1.1 Emphasis of the Verb Emphasis of the verb displayed multiple forms. The verb in initial position was meant, for example, to underscore its grammatical features: mood, tense, or voice. Alternatively, the grammatical features of the verb may explain its occurrence in initial position. It is difficult, for instance, to determine whether the imperative frequently occurs in initial position for reasons of emphasis or for cognitive reasons. It is clear, however, that the imperative—in its proper form or in the form of the hortative subjunctive—is marked in initial position. See for instance: dic

mihi

tell-Imper.-2sg. 'tell me'

me-Dat.

(Ter. Eun. 349;360)

When the imperative is not in initial position, it is because another element needs to be emphasized, as in constructions featuring two contrastive clauses: facito. . . , Hoc this-Acc. do-Impcr.-3sg. 'do this. . . . , avoid that'

hoc fugito this-Acc. avoid-Imper.~3sg. (Tcr.Ad. 417)

94


In addition to emphasis, the initial verb is common in enumerations or in contrastive structures featuring irreconcilable verbal forms, such as past and present forms or past and future forms. See the following example featuring an enumeration of antitheses (Marouzeau 1938:53): vicit triumph-Pf.-3sg.

pudorem libido, timorem shyness-Ace. lust-Nom. fear-Ace.

audacia rationem amentia impudence-Norn. sense-Ace. madness-Nom. 'lust triumphed over shyness, impudence over fear, madness over sense' (Cic. Clu. 6.15)

The initial verb is also used for means of attention in negations or in categoric affirmations, declarations, or warnings (Marouzeau 1938:5758). Similarly, it also occurs in sentences displaying literal or synonymous repetitions. The popularity of this stylistic device, hence of this type of V-initial structure, in poetic texts underlines its markedness (for details on its application, see Marouzeau 1938:61). The markedness of the initial verb is further shown in examples in which the verb expresses a wish, a will, a certainty, or intense emotions. These verbs indeed often occur in initial position and display a high incidence of preposed personal subject pronouns—the use of which is marked in Latin: more than half of the instances of ego feature verbs denoting 'want', 'think', and 'know' (Marouzeau 1938:55). These verbs often govern a following subordinate clause, an infinitive, or an Accusativus Cum Infinitivo (henceforth ACI). Therefore, these instances of prepositing are motivated by emphasis and supported by syntax; the phrases mentioned were the first to be right branched (sec section 4.3). 4.2.1.2 Pragmatic Function: Emphasis of the Utterance In addition to emphasis, the verb-initial position could have a pragmatic function: it introduced a definition, an utterance indicating surprise in animated narrations, and so on. The initial verb occurs, for instance, in poetic evocations and characterizes, according to Marouzeau, many of the great verses of Ennius and Virgil (1938:65). Historians, on the other hand, like to use the initial verb in combination with the historical present when describing an important event, such as an intervention or a mission. In comedies, on the other hand, unexpected entrances are often marked by means of the initial verb: Ei!

Video

uxorem

Dear me! See-Pr.-lsg wife-Acc. 'Dear me, I see my wife' (Tcr. Ph. 797)


95

Similarly, in the works of historians as well as orators, the initial verb stresses a quick and continuous succession of events. The preposed verb, therefore, clearly has a narrative function and we find it often in the writings of historians, novelists, and epic poets (Leumann and Hofmann 1928:614)—all authors of the narrative genre. Similarly, the high frequency of verbs in initial position in the much later Peregrinatio (Vaananen 1987:105) goes well with the lively style of the narration. The preceding brief survey of the use of the initial verb suggests that it was a highly marked order: the initial verb emphasized cither the verb proper or the utterance it introduced.

4.2.2 Syntactic Motivation In addition to stylistic uses, the initial verb could also have a syntactic motivation when related to a particular context. In a main clause preceded by a subordinate clause the verb tended to occur in initial position. The occurrence of the historical present, the narrative infinitive, or even adverbs that stress a particular aspect of the action shows that this structure was not uncommonly used in emphasis. Yet initial position of the verb was motivated by syntax. Marouzeau argues that the prototype of the subordinate-main clause construction was the sentence preceded by an ablative absolute construction (1938:80). This interpretation, though difficult to verify, is supported by the subordinate value of the ablative absolute and its eventual replacement with a subordinate clause. In yet another, though related context the verb tended to occur in initial position: when the sentence was introduced by a negation or an adverb. This same tendency can also be observed in Old French, for instance (see section 4.5), or in a number of modern Germanic languages, such as German and Dutch. In these languages also the verb occurs in initial position when a subordinate clause, an adverbial phrase, or another element introduces the sentence. Because of these characteristics these languages are often referred to as V2 languages (see also section 4.5.4). Considering the popularity of this type of structure in languages that are evolving from SOV to SVO—Old French, Modern German, and Modern Dutch—it is legitimate to relate this phenomenon to the typological change that was or is taking place (see section 4.5). Consequently it is important to determine the context of this phenomenon in Latin. Evidence from several novelists and historians presented by Mobitz shows that the verb is preposed after conditional but especially after temporal subordinate clauses (1924:120). The construction was characteristic of narrative authors, and in particular of novelists, but it was rare in the writings of historians (Mobitz 1924:121). Although the construction also occurs in other texts (for instance, Ovid's Metamorphoses [3.6.26]), it is uncommon (Kroll 1918; Mobitz 1924). In Apuleius, on the other hand, there are numerous instances of this construction, in particular with temporal (20 instances) and conditional

96


(12 examples) subordinate clauses. Similarly, of the 256 instances of initial verb in Apollonius's novel 50 arc in main clauses preceded by a subordinate clause (Mobitz 1924:121). In Pctronius's work the construction is particularly frequent after temporal subordinates (Kroll 1918:117) but occurs also—though less commonly—after a conditional subordinate clause (Mobitz 1924:121). Moreover, although this structure was already frequent in the writings of Apuleius and Petronius, Mobitz noticed a clear increase in its use in Minucius Felix and Tertullian, who lived in the same period as Apuleius (1924:122). Consequently, this phenomenon can be considered characteristic of a level of speech rather than of the novel, hence a linguistic rather than a literary feature. This hypothesis is supported by findings presented by Linde, who argues that the popularity of this structure in everyday speech is reflected in Petronius's work (1923:164). Yet in spite of the high frequency of initial verbs in this context, we usually find the normal, SOV, order in Pctronius's work (Kroll 1918:118). Occurrence of the verb in initial position in the context mentioned was a tendency that developed during the course of Latin and that gradually spread in popular texts. In Old French the verb no longer occurs in initial position after a subordinate clause; however, it still does so after adverbial phrases (see section 4.5).

4.2.3. The Use of the Initial Verb: Conclusion Examination of the use and the function of the verb in initial position in a large number of Latin writers suggests (1) that in the subordinate clause the verb rarely occurs in initial position—in the course of time its use increased in some texts; (2) that in the main clause the preposed verb is much more frequent and is motivated by stylistic or syntactic reasons. The initial verb either carries emphasis or has a pragmatic function, introducing a description or marking the progress of action or narration. The use of a given tense often supports this function: progress in action or in narration is frequently expressed with a historical present, a perfect tense, or an infinitive (Marouzeau 1938:79; Kroll 1918:119); description, on the other hand, is generally characterized by the imperfect past tense. This rupture of the regularity in the use of tenses illustrates the marked character of the initial position of the verb. In addition to these stylistic motivations that can be observed in all authors, the verb occurred in initial position when the main clause was preceded by a subordinate clause, an ablative absolute, an adverb, or an adverbial phrase in sentence-initial position although we find this structure in the writings of all authors, its frequency varies considerably and increases in later texts. Since the occurrence of the initial verb in Latin was related to stylistic and syntactic factors, it clearly was a marked and exceptional order (see also Marouzeau 1938:81—82). In the course of time the literary value of


97

the verb in initial position was lost, whereas the syntactic and pragmatic functions remained and even became increasingly important. The incidence of the initial verb is very high in the Peregrinatio, for instance, where its use has a narrative function or is related to a preceding subordinate clause or an adverbial complement (for the situation in Old French, see section 4.5.4). The increasing importance of the syntactic motivation accompanies Latin's gradual shift from left to right branching.

4.3 The Medial Position of the Verb Since the verb in final position was unmarked and the verb in initial position was marked, Latin reflected the archaic order of PIE, which it had inherited. The medial position on the other hand was an innovation. We find instances of it in the earliest Latin texts: in votive and other inscriptions. Its occurrence, however, is related to given contexts: the occurrence of the verb in medial positions was characteristic of sentences featuring numbers; this pattern explains its high frequency in inscriptions. The importance of numbers is further demonstrated in Cato's practice: in his texts, for instance, the verb is normally in final position except when the sentence displays a number; the verb then occurs in medial position. Whereas the verb in initial position expressed a specific function, the verb in medial position did not. Marouzeau even believes that the choice between final and interior position was as it were indiscriminately free (1938:82). Without specifying the authors or the period he refers to, he even claims that in some texts the incidence of the medial verb is high, coming close to the frequency of the final verb (1938:82). Yet evidence suggests that the medial verb remained uncommon for a long time and that the elements following it were not random. In contrast with that of the initial verb, its use was, however, never stylistically marked. This is an important observation, because the structure was going to be the dominating feature of the Romance languages.

4.3.1. The Elements That Follow the Medial Verb Since the Latin clause was characterized by the unmarked final verb and the marked initial verb, the medial position never received the attention it deserves. Only in analyses of later texts does it become a matter of discussion. The analysis of this position of the verb is complicated, moreover, because most studies focus on the three positions of the verb—initial, medial, and final—without considering the nature of the elements concerned. Yet for the medial verb it is extremely important to determine the syntactic context, because it shows the actual branching of these structures. Marouzeau is one of the few Latinists to give some information on the nature of the postposed elements without, however, determining their frequency: the verb in medial position is followed by an enumeration of subject nouns or direct and indirect object nouns; by a relative, infinitive,

98


or participial clause; by an ablative absolute or a complement featuring a gerund or supine (1938:87-88). Evidence gathered by Linde (1923) gives us more details. The postposed elements most frequently are adverbial expressions indicating goal or aim. These are often prepositional phrases: tile . . . aquam poposcit he-Nom. water-Ace. demand-Pf.-3sg. 'he called for water to wash his hands'

ad for

manus hands-Acc, (Petr. 27.6)

Second in the order of frequency are direct objects with or without an adjective or an adverb. Other frequently postposed elements are constructions featuring a gerundive or a supine, as well as infinitives and ACIs. Postposing of the subject alone, on the other hand, is uncommon. Its occurrence is even less frequent than that of adjectives or genitives, which also rarely follow the verb (see Linde 1923:175-77). Moreover, the postposed adjective or genitive is a stylistic means to emphasize the elements concerned: disjunction is a common stylistic device of emphasis. By this process the verb is inserted in another syntactic—often nominal—phrase separating thus the nominal head and its complement. Consequently, the adjective, the genitive, or the participle follows the verb and is in clausefinal position. The medial position of the verb is then used for emphasis, but in contrast to the initial position, it is not the verb proper or the utterance that carries emphasis, but rather another element of the clause. These data do not imply that the grammatical elements I just mentioned generally followed the verb: they show that the phrases that followed the medial verb were not arbitrary. On the contrary, the elements that followed the verb in medial position were most often verbal complements. Accordingly, the medial verb in Latin, which was still fairly uncommon, was a right-branching structure. Evidence gathered by Linde (1923) is based on the analysis of a large number of texts of various literary genres—speeches, comedies, letters, and so on. As I demonstrate later, these findings are corroborated by data from the work of Petronius (Feix 1934) and that of other Vulgar Latin and Late Latin authors.

4.3.2 Vulgar Latin and Late Latin In the course of time the nature of the postposed elements did not change, but the frequency of the medial verb increased significantly. In Petronius, for instance, 25 to 30 percent of all verbs are in medial position. Similarly, in Apuleius's novel the verb occurs in medial position in 35 percent of the main and in 28 percent of the subordinate clauses (Bernhard 1927:17). This tendency continued to increase, as the following analysis of Vulgar and Late Latin texts points out: Petronius's novel, the letters of Claudius Tercntianus, the Peregrinatio, the Chronicle of Anonymus Valesianus, and the texts of St. Augustine.


99

4.3.2.1 Petronius The postposed elements in the narrative fragments and dialogues are complements of the verb in question: complements of direction, prepositional complements or infinitives, integrated or not in an ACI (see Feix 1934:13 et seq.). Accordingly, evidence from this Vulgar Latin text corroborates the tendency I observed in Classical Latin: the medial verb was an RB structure. Moreover, as in Classical Latin, one finds instances of disjunction which mark emphasis: leviorem facere light-Comp.-Acc.sg. make-Inf. 'to make a slight loss'

iactuam loss-Acc, (14.4)

The artificial nature of this structure shows also in its use: whereas the structure is rather frequent in the narrative parts, it is less common in the speech of educated people and it only appears once in the fragments of daily speech (Feix 1934:20). 4.3.2.2. The Letters of Claudius Terentianus Of the 20 sentences featuring a nominal subject and object 10 have the order SVO, 4 feature VOS, and 3 feature SOV (the other orderings each occur once; Adams 1977b:74-75). These data corroborate the hypothesis of the spreading of right branching. In sentences without an explicit subject, the tendency toward right branching is even stronger: VO clearly dominates OV (40 vs. 14; in the Greek letters these figures are 33 VO vs. 16 OV [Adams 1977b:68]). These data are based on occurrences of phrases featuring finite verbs and phrases with infinitives. The preference for VO over OV is more important in subordinate (9 vs. 1) than in main clauses (31 vs. 13; Adams 1977b:69). At face value this characteristic distinguishes this text from all other Late Latin texts. Yet the occurrence of OV structures and the occurrence of the verb in final position, though closely related, are not exactly the same phenomena. Sixteen of the 43 subordinate clauses feature a verb in final position, whereas in main clauses these numbers are 20 versus 70. Apparently, the verb in final position is more popular when the clause does not contain a direct object. This observation is consistent with the observed frequency in Classical Latin of objects in medial verb contexts. Yet it does not account for the low frequency of OV in subordinate clauses. Since "the order OV does not appear to have been especially common in subordinate clauses in Greek" (Adams 1977b:69), it may be tempting to ascribe the high frequency of VO in subordinate clauses to the influence of Greek, which was the native language of the writer. But this explanation is at variance with the high frequency of the final verb in subordinate clauses (sec the preceding data). Therefore, we can only con-

100


elude that the use of the medial verb increased, even in the subordinate clause; the high incidence of VO, moreover, emphasizes the fact that the medial verb definitely was right branching. The increasing tendency of VP to become right branching is also manifest in constructions featuring an ACI (21 of the 25 instances are RB), in structures featuring an infinitive which always follow the head verb (Adams 1977b:71), and in prepositional phrases with adverbial value, which follow the verbal head in 47 of the 63 instances.

4.3.2.3 The Peregrinatio In the Peregrinatio declarative sentences show a predominance of SVO over SOV (35 instances of SVO against 22 of SOV [34 vs. 21 percent]; Vaananen 1987:106). The other orderings are represented by the following figures: VOS, 22 (21 percent); VSO, 15 (14 percent); OSV, 6 (6 percent); OVS, 4 (4 percent). Moreover, in sentences without an explicit subject VO occurs 99 times, whereas OV is represented 53 times (data found in Vaananen 1987:106). As in Terentianus's letters, not only is VO more frequent than OV, but SVO is also the most common sequence. Its predominance is even stronger than in the letters of Terentianus. Yet the other orderings arc still relatively important. Consequently, we cannot conclude that SVO predominates in this text—it does not, but it is legitimate to say that right branching is very firmly established, because all three orders, VSO, VOS, and SVO, are RB. Just like SVO, the sequences VSO and VOS both share postposing of the verbal complement, which accounts for the high percentage (69 percent) of right-branched complements in sentences featuring a subject and an object. Evidence from the Peregrinatio further illustrates that length of the constituents had become a criterion. In shorter sentences, the verb tends to occur in final position (Haida 1928:18). Alternatively, the verb in medial position is generally followed by rather long attributes, enumerations, and appositions. Similarly, relatives entail VO structures when they refer to the direct object of the verb, for instance, leget

ilium

locum

read-Fut.-3sg.

that-Acc.

passage-Ace. which-Nom.sg.

qui

scriptus

est

written-Nom.sg.

be-Pr.-3sg. in Gospel-Abl.

in evangelio

'he will read the passage that is written in the Gospel'

(29.5)

The constituents that are right branched are syntactically still the same: the predicative noun and the complement indicating goal or direction tend to follow the verb. Exceptions to this regularity are motivated, which stresses the increasing importance of right branching (see Vaananen

101 101

Diachronic Ana-lysis: The Verb Phrase

1987:106-7). Similarly, the infinitive normally follows the finite verb, except when combined with doleo and habeo, in the sense of 'must', which is a motivated use (see Bauer 1993b). Evidence from the Peregrinatio strongly suggests the increasing importance of postposed verbal complements: the verbal complement increasingly tends to follow the verb. This is also manifest in participial constructions: in the large majority of instances the complement follows the verbal adjective (24 vs. 10). The 10 exceptions to this regularity are all, except 2, more or less fixed expressions: for example, gmtias Deo agentes (10.7) 'thanking God'. When the complement is a prepositional phrase, the tendency toward right branching is even stronger: there are 34 instances of [present participle [prepositional phrase]] against 10 instances featuring the reverse order, which often mark emphasis. See, for example: Sic as much

de about

viris, men-Abl.,

sic as much

de about

mulieribus women-Abl.

requirens dicit inquire-PartPr.-Nom.sg. say-Pr.-3sg. 'he makes the same inquiry of both men and women' (45.4)

4.3.2.4 St. Augustine Although the number of verbs in medial position in St. Augustine's texts (Muldowney 1937:121-22) is lower than in the Peregrinatio, the tendencies are the same: complements of verbs of movement and nouns, followed or not by a relative, are normally placed to the right of the verb. These are thus right-branching structures. The other instances of verbs in medial position are accounted for by emphasis, stylistics (disjunction), or rhythm (e.g., avoidance of hiatus or elision) (cf. Muldowney 1937:120-28). 4.3.2.5 Anonymus Vcdesianus Evidence of the Chronicle confirms the importance of the syntactic nature —main or subordinate—of the clause in the rate of word order change: VO dominates in the main clause (ca. 59 percent), whereas OV is still the preferred order in subordinate clauses (ca. 78 percent). When both the subject and the object arc expressed, the subject normally precedes the object (24 vs. 1). The one exception is a marked structure. In these 24 examples, SVO order clearly predominates: 16 instances against 6 of SOV, 4 of which occur in subordinate clauses. The other two examples present VSO and are syntactically motivated, because they arc preceded by temporal expressions (data from Adams 1976:137). These numerical data suggest, moreover, that "in a main clause the object is more likely to follow the verb when the subject is expressed than it is when the subject is unexpressed" (Adams 1976:137). Generally this

102


tendency is attributed to possible ambiguity. In sentences without a subject, context as well as inflection are thought to indicate the function of the noun in question (see Adams 1976:137). Inflection is correctly thought to be pertinent, because if it were not pertinent, the sequence [noun—verb] would be interpreted as subject-verb. Yet, if inflection is pertinent in OV/VO sequences, it must also play a major grammatical role in other contexts, hence in clauses featuring object and subject. Consequently, ambiguity cannot be the underlying reason for the frequency of SVO in these contexts. Right branching also dominates in other verbal structures: the verb combined with an infinitive, a prepositional phrase, and a participle. In subordinate clauses the prepositional phrase still precedes predominantly, as does the direct object (35 vs. 10), but this complement is more frequently postposed in main clauses (106 vs. 88; Adams 1976). Finally the tendency toward right branching is manifest in constructions featuring a participle (present participle most frequently): we find 18 examples of postposed objects against 8 featuring the reverse order. On the other hand, the object of the infinitive is generally LB, because it often occurs in subordinate clauses or as an accusative in an ACI construction.

4.3.3 The Position of the Verb in Latin: Provisional Conclusion Displaying the unmarked final verb as opposed to the marked initial verb, Latin had inherited the organization of the PIE verb phrase. In Classical Latin the initial position of the verb continued to be highly marked: obviously less frequent than the final verb, the initial verb either had a stylistic or a syntactic motivation. The medial position, on the other hand, was already more frequent in Classical Latin than it was in Old Latin, and one observes that the element that followed the verb was indeed most frequently its complement. In the few stylistically marked instances of medial verb (e.g., disjunctions) it is never the verb that carries the emphasis, but rather another element of the sentence. By contrast, the occurrence of the verb in initial position, when stylistically motivated, was meant to emphasize cither the verb or the pragmatic function of the utterance. In the course of time the unmarked final position gradually lost its predominance. The verb in final position survived longest in subordinate clauses. Meanwhile, the medial verb and also the initial verb became more frequent. In Late Latin the verb in initial position was an important narrative feature and became more and more popular; the same observation can be made for the initial verb after a subordinate clause, an adverb, or an adverbial expression. Although the initial verb was perhaps more common, its occurrence was always related to a specific use or syntactic context: it was a marked order. To a lesser extent, these tendencies can still be observed in Old French, where postposing of the verb occurred basically in subordinate clauses and where the verb occurred in initial position after adverbial clauses.


103

Alternatively, the medial position of the verb was not related to specific situations, neither in Latin nor in Old French. Moreover, it created a rightbranching structure without interfering with the principle according to which the subject in a nominative language occupies the first position of the clause because it is the unmarked topic. This is probably why SVO eventually became the unmarked order in the Romance languages.

4.4 The Copula The position of the copula is relevant for two reasons: it informs us of the unmarked position of the Latin verb (sec section 4.4.1) and its history is closely related to the development of the proposed auxiliary, which is part of a RB typology (section 4.4.2). Both aspects underline the importance of the copula for the analysis of the grammatical shift toward right branching.

4.4.1 The Position of the Copula In Old Latin texts the copula occurs—with a few exceptions—behind its complement and in final position of the clause: the verbal phrase featuring a copula was left branching. The data gathered by Warner (1980) covering the period from 450 B.C. (the Twelve Tables) until 186 B.C. de Bacchanalibus) show that the copula predominantly follows the predicate nominative (59 vs. 13), the adjective (45 vs. 13), and the participle (25 vs. 6) (see Warner 1980:253-54). Warner therefore concludes that "the unmarked order in [very] Old Latin is weak SAV, strong NSV and rigid SPV" (1980:261). The high incidence of N[oun] Subject] V[erb] (36 instances against 18 instances of SNV and 5 of NVS) indicates, according to Warner, that NSV was the unmarked and the original order in Latin (1980:257). Yet the fact that sentences featuring a direct object in "the most archaic Old Latin [are] clearly SOV" (Warner 1980:258; see also section 4.1) raises serious doubts about this interpretation. The unmarked SOV order and the dominance of [subject-adjective/participle-copula] sequences (see the preceding data) both confirm the unmarked status of the subject in initial sentence position, hence also of SNV (for the unmarked initial position of the subject in Latin, sec de Jong 1989). The statistical data on the first Latin texts presented leave no doubt about the tendency to postpose the copula. This conclusion is supported by Cato's use of right-branching copula constructions: in his texts RB occurs only "to introduce a definition, a localization in time or place, or when the attribute is complex" (Muldowney 1937:130). The verb phrase with a left-branched attribute is consistent with transitive phrases in which the direct object precedes the verb in an unmarked order (see section 4.1). Consequently both types of phrases present corroborative evidence for the unmarked left-branching verb phrase in Latin.

104


Alternatively, when the utterance is a definite affirmation or a negation, the copula precedes its attribute, as in the next example, where the emphasis is very clear: Mors Death-Norn.

misera miserable-Nom.sg.

non est. not be-Pr.-3sg.

Aditus Transition-Nom.

ad mortem est miser to death-Ace. bc-Pr.-3sg. miserable-Nom.sg. 'It is not death itself that is miserable, but the transition to death that is miserable' (Com Inc. 203 Ribb; Marouzeau 1938:18)

4.4.2 The Emergence of the Preposed Auxiliary When combined with a past participle the copula originally also occurred in final position in unmarked order; examples are numerous (Marouzeau 1938:24). The reverse order was marked and emphasized a negation or an affirmation. When the speaker or writer wanted to digress on the identity or the qualities of the person or the object in question, the order also was the reverse (on the use of the sequence copula-attribute, see Marouzeau 1938). Whereas the copula was frequently combined with a past participle, it was rarely used with a present participle. This type of phrase only spread in the everyday language, where since the time of Old Latin the occurrence of the present participle with sum instead of the finite verb was no exception. This periphrasis stressed the stative aspect, the duration of the action, or modal subleties (Leumann and Hofmann 1928:605—6). See, for example: scientes esetis 'that you be aware' (S. C. de Baccb. 23)

(on the popular nature of this expression and its spread, see Lofstedt 1911:245-49; Ernout and Thomas 1951:232-33). Although this periphrasis developed in the imperial period, it was in Late Latin that it became a means to avoid the noncompound forms of the subjunctive, which were weakly represented in everyday speech (Leumann and Hofmann 1928:605-6; Lofstedt 1911:249). Like the attribute and the past participle, the present participle formed an LB structure with the copula (Marouzeau 1910a:lll, 1938:27), for instance: te carens . . . fui 'I missed you' (PL, Capt. 925) The regularities observed in the preceding pages on the use of esse also apply to other copulas: the sequence [[attribute] verb] was the unmarked order in Latin, as in sanus fieri 'to heal'.


105

The frequency of the sequence [[participle] copula] gradually, though slowly, diminished in the course of time. Esse, for example, frequently occurs in popular dialogues in Petronius, but it is still generally postposed (60 percent of the instances feature an LB verbal phrase [Feix 1934:24]). Yet Muldowncy's data on a number of authors of various Latin periods demonstrate the constant decline of [[factus] est] in favor of its RB variant. The sequence factus est is exclusive in the Twelve Tables and in the S. C, de Bacchanalibus; the proportions arc 15 to 1 in Cato and 10 to 1 in the writings of historians and orators, whereas in the more popular texts of Plautus and Terence these preferences are less important (4 to 1) (Muldowncy 1937:130). These data foretell the situation in Late Latin, where the sequence [copula [present participle]] was the normal order (Leumann and Hofmann 1928:615). The unmarked nature of this type of sequence clearly shows, for instance, in examples from the Vulgata (see Plater and White 1926:109-10). stellae erunt decidentes

'stars will fall' (Mc. 13.25)

erat expectans

'he was waiting'

(Mc. 15.43; Lc. 1.21)

The subsequent development of the position of the copula is related to the changes in the inflectional system. The Latin verb combined in one form the ending—the grammatical element—with the preceding verbal— perfective or imperfective—stem, which expresses the lexical meaning: [[laudav] -erat] 'he had praised'. Yet the chief feature of this form is not its synthetic nature. The Latin inflected verbal forms were characterized by a postposed grammatical clement; hence they were LB forms. Accordingly, the development of the verb from Latin to French is not merely the replacement of synthetic with analytic forms, although this observation is not incorrect. The comparison of Latin and French forms reveals another, more important, change: the emergence of habeo as auxiliary, which was paralleled by the reversal of the order of elements: [[laudav]-erat] [[laudav]-erit] [[laudav]-ero]

[ilavait [loue]] 'he had praised' [qu'il ait [ l o u e ] } 'that he has praised' [j'aurai [loue]] 'I shall have praised'

The development of the Latin verb is indeed integrated in the twofold change one observes in Indo-European languages: (1) the shift from an aspect-based verb system to a verbal system based on distinctions of voice and tense and (2) the shift from left- to right-branching structures. Al-

106


though the opposition perfective—imperfcctive was basic to the Latin verb (for the development of this distinction, see Bichakjian 1988b) and although Latin still featured residues of the archaic system—such as the gerundive construction (sec Bauer 1993a)—the Latin verb presented a major innovation, "[celle] d'exprimer la notion de 'temps' par le theme" (Meillet and Vendryes 1924:249-50). This marked the beginning of the new verbal system. Since the aspectual distinction was fundamental to the Latin verbal system, it featured a stative auxiliary, hence esse. The emergency of habeo— a nonstative verb—as an auxiliary marks the change to a verbal system based on temporal distinctions. Parallel to the loss of aspect, the order of elements in the verbal structures also changed: instead of the archaic leftbranching synthetic organization, the new verbal form is characterized by the RB sequence [auxiliary [lexical element]], whereby the auxiliary is the head of the compound form. In the course of time, laudatum habeo lost its aspectual value and habeo became a real temporal auxiliary (see, e.g., Bichakjian 1988b). The evolution of Latin shows not the creation of the auxiliary as such, but rather a change in the nature of the auxiliary clement and in the place it occupied. In chapter 2,I attempted to demonstrate the hierarchical status of the auxiliary: its function is to express grammatical features—tense, aspect, and so on—of the verbal compound form and to express a syntactic relation: being an inflected element it carries the personal marker and thus expresses subject agreement. As such, the auxiliary is the head of the compound form. In an LB language the lexical form of the verb precedes the auxiliary, which generally has the form of an ending (see also section 4.7 and 5.4); in an RB language, on the other hand, the auxiliary is preposed and it is an independent form. The tendency toward RB also affected suffixes and infixes that express specific values, such as the inchoative and frequentative. They were replaced with periphrases featuring RB prefixes or adverbs that follow the verb (Bauer 1993b). In the development from Latin to French one thus observes the emergence of a new type of auxiliary and parallel to this change, the structural reorganization of the VP. This twofold grammatical shift was long-lasting and was not achieved before the French period. Whereas the ordering habeo / sum + participle prevailed in twelfth century Old French, the development toward the modern structure, where the direct object follows the compound verb, [[habeo / j'ai [participle]] [direct object]], is a development of Middle French (see Bauer 1993b). Accordingly, the structure (a) epistulam scriptam habeo I je possede une lettre ecrite 'I possess a written letter' changed first into (b) habeo epistulam scriptam / j'ai une lettre ecrite, and then into (c) j'ai ecrit une lettre 'I wrote a letter', where the direct object follows the auxiliary and the participle. This change can be represented in a tree structure in the following way:


107

In structure (b) the perfect participle, being a verbal adjective, depends on the noun which it follows in unmarked order. In (c) the participle has lost its perfective value and constitutes a compound form with the head, the finite verb. The compound form governs the direct object. The modern structure featuring a preposed auxiliary developed partly in the period of Old and Middle French, as did the personal subject pronoun, which eventually replaced the personal ending. Since the emergence of the pronoun really started in Middle French, I discuss it in section 4.5.

4.5. The Position of the Verb in Old and Middle French Although the Latin verb phrase inherited its LB organization from PIE, one observes a change toward right branching in the texts we still have. In the course of time, this tendency became increasingly strong, even if the change was slower than the parallel shift in the noun phrase. Old French data reveal the same development I pointed out in Latin: the gradual and ongoing disappearance of left branching, the delay of this shift in the subordinate clause, the emergence of the preposed auxiliary and subsequently of the personal subject pronoun, and the occurrence of the sequence verb—subject in sentences introduced by an adverbial phrase (see section 4.5.4). Although the change toward right branching started in the Latin period, its last alterations were carried out in French: in Old French, SVO dominated in the main as well as the subordinate clause, but other sequences were still well represented. The following analysis shows that these orderings were related to specific syntactic contexts and that they gradually gave way to SVO. In the discussion on the position of the verb in Old and Middle French numerous analyses use the notion complement without indicating whether this terminology refers to the verbal or the sentential complement. Yet the precise nature of this element is essential to the current analysis. I have therefore attempted to identify the elements in question. Often it is the verbal complement that is being referred to, but the precise identification of the complement sometimes remains uncertain. In these instances and in the direct quotations I will use the notion of complement or C in combination with S or V. As a rule I have only used data and figures from studies

108


that actually published the lists of data or gave representative examples from the text they dealt with. I first compared this evidence and then, only when the structures under consideration were indeed identical, decided to use it for my research. The Old French sentence displayed four sequences, SOV, VSO, OVS, and SVO, which from the Middle Ages on was the most common construction (Foulet 1923:257). It occurred in main clauses, where it coexisted with VSO and OVS, as well as in subordinate clauses, in which we also find SOV (Foulet 1923:268). Whereas SVO dominated in main (Brunot 1924:257) and subordinate clauses, the three other sequences were related to a specific type of clause: SOV was limited to subordinate clauses, and OVS and VSO occurred especially in main clauses (see Foulet 1923:248). Accordingly, since SVO was the nonmotivated order, it is legitimate to consider it the unmarked order in Old French. In the following pages I demonstrate how this preferred order in Old French spread in the course of time, thereby supplanting the other sequences in all instances.

4.5.1 The Position of the Subject Brunot argued that in Old French the subject generally preceded the verb and that it occurred in clause-initial position (1924:265). Brunot's general remark is confirmed by data from more detailed studies. In the Quatre Livres des Rois, for instance—the translation of the second part of the twelfth century of the Latin text—71 percent of all sentences featuring a subject have it preposed (Haarhoff 1936:59). Herman's data on chapters 7-14 of the third book of this work also reveal this tendency: 70.5 percent of the nominal subjects and 95.7 percent of the pronominal subjects precede the verb (1954:353). Alternatively, 70 percent of the 209 sentences featuring a postposed subject arc introduced by an adverbial phrase (for this marked structure, see section 4.5.4). The twelfth century translation of the Psalms presents another convincing example of the tendency toward SV. In the Cambridge manuscript one observes 45 instances where the subject occurs at a position different from that of the original text: in 42 instances the translator made the nominal subject—which was postposed in the original text—precede the verb (Herman 1954:82). Moreover, the comparison of both manuscripts—the Cambridge and the Oxford versions—reveals that whenever the Latin sequence VSC or VCS was reversed in the French text, it always became SVC (Herman 1954:83). In Old French the preposing of the subject was more frequent in the subordinate than in the main clause. In the Quatre Livres des Rois, for example, the percentages are 66.8 percent (nom.)/88.2 percent (pron.) in main clauses and even 84.2 percent (nom.) and 100 percent (pron.) in subordinate clauses (Herman 1954:535). This numerical difference can be accounted for by referring to specific word order variants occurring in each


109

type of clause: although SVO was the dominating order in the main as well as the subordinate clause, SOV was the "second" order in the subordinate clause, and VSO, featuring subject inversion, was the most common variant in the main clause. Yet this widespread phenomenon of the Middle Ages, the inversion of the subject—limited to the main clause—diminished in the course of time (see Lewinsky [1949:13—15] for statistical data). Since inversion is closely related to the position occupied by other elements of the clause, I come back to this problem while discussing the sequences VSO and OVS (section 4.5.4).

4.5.2 SVO Order In the first French texts one observes a preference for OV, but these texts are basically poetic, with the exception of the Serments de Strasbourg. From the middle of the eleventh century on (e.g., La Vie de Saint Alexis] we notice a predominance of the sequence verb—object; a large number of independent studies show the increase of this predominance over time (Volcker 1882:30; Nissen 1943:74-75; Dill 1935:41; Lewinsky 1949:84; Haarhoff 1936:54). Moreover, in theQua-treLivres desRois right branching predominates (by 84 percent) in sentences featuring an object as well as a subject, either as SVO or VSO (14 percent); left branching occurs only in 14.2 percent either in the sequence SOV (8.6 percent) or OVS (5.6 percent; Haarhoff 1936:59-60). Finally in the fifteenth century Cent Nouvelles Nouvelles the object follows the verb in the large majority of instances, while its preposing is very rare: of 4379 main clauses with and without subject only 47 feature the order O-nom.V or O-pron.V (Dill 1935:39,42-49). The tendency to place the object to the right of the verb can also be observed in divergences in translations, such as the translation of the Psalms (twelfth century). In addition to the numerical dominance of VO, the Cambridge and Oxford manuscripts present 34 examples in which, in the translation from Latin into French, the order in the verbal phrase has been reversed: in 31 instances the preceding direct object in the Latin text follows the verb in the French translation (Herman 1954:90). This same type of change occurs in verbal phrases featuring a prepositional phrase: the Latin sequence [stressed word (i.e., subject, prepositional complement, or direct object) + prepositional complement + verb] is conveyed in French by the sequence [stressed word + verb + prepositional complement] (17 instances; see Herman 1954:90). These data are cited to show that the regular change that occurred was in one direction only: it is the complement + verb that was changed into verb + complement and not the other way around. In the main clause SOV order was very uncommon, whereas the sequence OV—featuring a nominal object—was more frequent when the subject was not expressed (Brunot 1924:268). This reminds us of the

110


Vulgar Latin text of Anonymus Valesianus (see section 4.3.2). Earlier in this section we saw that OV increasingly gave way to VO in the development of French. In Middle French postposing of the object is a plain fact, but it is less frequent when the subject is not expressed. In Berinus VO dominates in main and in subordinate clauses, but the occurrence of OV is greater in subordinate clauses without subject than in the ones with a subject (ca. 38 percent vs. 5 percent) (Lewinsky 1949:112). Similarly, whereas SOV does not occur in main clauses in this work, one finds examples of OV in main clauses without a subject, albeit at a very low rate: 29 instances of a total of 1160 (2.5 percent; Lewinsky 1949:109,112). In section 4.5.3 I return to this stronger tendency toward OV in clauses without a subject. Because of this ever increasing tendency toward right branching of verb phrases and because of the dominance of the subject in initial position, the order SVO increasingly became the established word order in Middle French (see also Brunot 1924:480). Statistical data indeed show the increasing predominance of SVO in main clauses featuring subject, verb, and object (Lewinsky 1949:86). In addition to SVO, which increasingly dominated in main and subordinate clauses, Old and even Middle French featured two second orders: one for subordinate clauses, SOV (see section 4.5.3), and one for main clauses, which is characterized by subject inversion, VSO and OVS (see section 4.5.4).

4.5.3 SOV Order More frequent in the first French texts (Brunot 1924:264), SOV is very uncommon in main clauses in later texts, where it has, moreover, a clear poetic connotation (Foulet 1923:266). On the other hand, this order is preferably used in subordinate clauses (Foulet 1923:255). The tendency toward left branching was strongest in relative clauses, more than in any other type of subordinate clause, where SVO is more frequent (see Foulet 1923:257; Moignet 1973:362). This difference between the relative clause and other types of subordinate clause was general and can be observed in data of all analyses of word order. Although, on the one hand, the order SOV was almost completely excluded from the main clause, it was, on the other hand, the only order to coexist in subordinate clauses with SVO, which from Old French on was the most frequent order in this type of clause: of the 1854 examples of subordinate clauses in Berinus, for instance, only 95 feature SOV. The majority of these instances of SOV feature, moreover, a subject relative pronoun (89 of the subordinate clauses arc relatives) (Lewinsky 1949:90). This tendency can be observed also in other medieval texts: already in Old French conjunctions, especially the relative pronoun, triggered postposing of the verb at the end of the clause (Foulet 1923:255). The same can be said about Middle French; in the Chronique de Jean d'Outremeuse, for


111

instance, the large majority of subordinate clauses featuring SOV are relatives (16 of 23; Nissen 1943:109; see also Price 1961). Another text of the second part of the fourteenth century, the Chroniques of Froissart, shows that SCV(C) exclusively had become the order of the relative clause that featured a subject relative pronoun. In other subordinate clauses the percentage of SCV remains below 1 percent (vs. 89 percent of SVC) (Price 1961:43). By contrast, we observe that almost a quarter of the 1065 relatives with subject relative pronoun in this text display the order SCV(C); the order SV(C) is represented by 76.2 percent (Price 1961:41-44). The preference for SOV in subordinate clauses introduced by a subject relative pronoun can be related to the tendency to make the complement precede the verb when the subject is not expressed (see section 4.5.2). In a relative clause featuring a subject pronoun, the postposing of the verb does not trigger a sequence of two nouns (NNV). This assumption is supported by the fact that the remaining six instances of subordinate clauses featuring SOV in Berinus display a pronominal subject, hence exclude a sequence of two nouns (cf. Lewinsky 1949:91). Having pointed out a clear tendency I still have not given an explanation. The revealed regularity shows that the OV order was most resistent in contexts where there is no succession of two nouns, hence no ambiguity. Since the case system was defective ever since Old French (and already before), the grammatical distinction—in contrast to Latin—may provide an explanation here. SOV order, the use of which was limited basically to subordinate clauses in Old French, decreased in frequency in Middle French, where it was almost exclusively used in relative clauses. These retained this construction for a long time (Brunot 1924:481), and it was only in the seventeenth century that the prcposed complement eventually disappeared (Brunot 1922:671).

4.5.4 VSO and OVS Orders Paradoxically, SVO dominated more in subordinate than in main clauses, because in the main clause we also find another rather common order that is characterized by inversion of the subject. Subject inversion was still relatively frequent in Middle French, and even in Modern French syntax it continues to play a role, albeit a modest one. SOV in the subordinate clause was a relic of the archaic unmarked leftbranching structure of PIE and one observes a parallel in Modern German and Dutch. VSO, on the other hand, is right branching. In Latin, where the complement was left branched in an unmarked order, the verb, in initial position and followed by its complement and the subject, probably was highly marked. In Old French, where the normal order was SVO, the distinctive feature of the VSO variant was not the different branching of the verbal phrase, but the postposing of the subject. Subject inversion is also characteristic of OVS sequences. Whereas both

112


orders were much used in poetry, VSO in contrast to OVS was also frequent in prose (Foulet 1923:267). Yet OVS and VSO arc related constructions, because the occurrence of a complement at the beginning of the sentence triggered inversion of the subject (Foulet 1923:249). This complement could cither be an adverbial element (ainsi, encores, etc.), an adverbial complement, or a direct or indirect object. This gives cither the sequence [... VSO] or OVS. The decisive importance of the introductory element shows in the low percentage of VSO sentences without initial complement: 12 percent in the Quatre Livres des Rois. In the same text 69 percent of all clauses featuring subject inversion have an introductory element (Haarhoff 1936:41). The decisive influence can also be observed in the low frequency of VSO in subordinate clauses in Old and Middle French, where we rarely find an adverbial complement after the conjunction (Lewinsky 1949:98-9). Yet subject inversion in Old French was related to one condition: the elements preceding the verb had to be part of the clause. In contrast with Latin, a preceding subordinate clause was no longer a triggering element in Old French. In the Latin main clause that was preceded by a subordinate clause the verb often occurred in initial position (see section 4.2.2). In the Quatre Livres des Rois, for example, 65 sentences feature a preceding subordinate clause but only 2 (3 percent) display subject inversion; the Latin original, on the other hand, shows subject inversion in two thirds of the examples (Haarhoff 1936:45-46). Similarly, other Old and Middle French texts feature very low frequencies of inversion after an initial subordinate clause (sec Lewinsky 1949:57; Nissen 1943:62).

4.5.4.1 The Direct Object in Clause-Initial Position Whereas the subordinate clause could trigger preposing of the verb in Latin, it rarely did so in Old French. At that time only the adverbial phrase and the direct object triggered inversion of the subject. Since the first position of the clause was occupied either by the subject—most commonly —or by an adverbial phrase or direct object, Old French can be considered a V2 language. I will come back to this characteristic and its correlates later in these paragraphs. The direct object was the next element that did not automatically trigger inversion in initial position. The occurrence of the direct object in initial position was rare and related to emphasis and stylistic variation. This order was extremely marked, not only because the object occurred in the very first position, but also because the object was left branched. From the twelfth century on, and especially in Middle French, the complement in initial position was often referred to by a pronoun later in the clause: Les

deux

premiers

mesagiers

il

les

the

two

first-Acc.pl.

mcsscngers-Acc.

he

thcm-Acc.

Diafhronic Analysis: The Verb Phrase first

113

entrer

make-Passe def.-3sg. enter 'the first two messengers, he made them come in' (Berinus 1.304.4)

The use of the pronoun, which is part of the process of dislocation, shows that "plain" object-initial sentences were no longer common in French. From the thirteenth century on, a preposed direct object no longer automatically triggered inversion of the subject: therefore, in Berinus OVS and OSV are nearly equally frequent (Lewinsky 1949:102). This tendency can also be partly related to the extension of the use of the personal subject pronoun, which generally precedes the verb (Lewinsky 1949:47). With only one exception, the subject in the OSV examples is indeed always pronominal in Berinus (see Lewinsky 1949:102—3). Price also has pointed out the importance of the pronominal nature of the noninverted subject: the verb can be preceded, according to Price, by two semantic entities if one of them is relatively light, which explains the occurrence of CSpron.V. and the nonoccurrence of CSnom.V [1961:46-47]; where C is a sentential or verbal complement). These details inform us on the chronology of the change. The chief development that the preceding data reveal is, however, the disappearance of a highly marked structure: the context in which subject inversion occurred became more and more limited. In contrast to the direct object, the adverbial phrase, interpolated clauses, interrogative clauses, and imperatives continued to trigger inversion in the main clause for a long time.

4.5.4.2 The Adverbial Phrase in Initial Position Whereas the preposed subordinate clause as a rule no longer triggered subject inversion in the Old French main clause, adverbial phrases in initial position still did (see, for instance, Haarhoff [1936] on this matter in the Quatre Livres des Rois). Yet in later texts, such as Berinus, only a noncompound adverbial phrase in clause-initial position, for instance, a noncompound adverb, always triggered inversion (Lewinsky 1949:37). When subject inversion occurred, two orders were possible: (1) [introductory element--Verb—Subject—Object] or (2) [introductory elementVerb—Object—Subject], which was much less frequent. The difference in frequency is related to the pronominal subject, which in postposition always immediately followed the verb in Old French. Consequently, structure (2) is exclusively nominal, hence its lower frequency. Moreover, from a typological point of view also, the order VOS is extremely rare (see Greenberg 1963:61). In the course of time, inversion after adverbs in initial position disappeared, except in a residue of fixed expressions. The percentage of inversion after an initial adverb is relatively low in Berinus, for instance, 70.9

114


percent, compared to the Chanson de Roland, where the percentage is 87 percent (Lewinsky 1949:33). This type of inversion became more and more unusual in the fifteenth century (Lewinsky 1949:59), which marks the important change in Middle French, where SVO gradually became the normal order in every context, not only in the main clause after a preceding subordinate clause, but also after longer adverbial phrases (see the data presented by Price [1961] and also the discussion in section 4.5.4.1). The length of the introductory element was an important factor in the rate of disappearance of subject inversion. At a given moment in the development of French, adverbs, in contrast with prepositional phrases, for instance, most frequently occurred in sentences with inverted subjects (Lewinsky 1949:16). Whereas the initial adverb generally triggers subject inversion in Berinus (70.9 percent), the adverbial phrase does so only in 25.2 percent of the occurrences (Lewinsky 1949:37). The decisive influence of the length of the phrase is also illustrated in Froissart's text, for instance: comparison of the SV and V(S) examples in sentences introduced by an adverbial complement reveals a relatively high occurrence of CSV when C is a relatively long constituent. When C is a short element, CV(S) is preferred (Price 1961:46; for corroborative data from the Quatre Livers des Rois, sec Herman 1954:362). Since in the majority of the instances the noninverted subject is pronominal, the disappearance of subject inversion must also be related to the emerging of the subject pronoun (see section 4.5.6). The preceding data show that subject inversion became restricted to contexts with shorter complements. Price (1961) accounts for this tendency by referring to the pause that is possible after longer complements: when the complement was long enough, there was a rhythmic break and the sentence proper had the normal order, that is, subject—noun or pronoun—in initial position followed by the verb. Yet, even when adverbs introduced the sentence, subject inversion was no longer automatic in Middle French. In sentences featuring a nominal subject and a direct object, the presence of an adverb in initial position did not change the normal order of the constituents (sec Lewinsky 1949:19). In addition, inversion was more and more limited to a restricted number of adverbs. At face value subject inversion following an adverb seems very high in the Chronique (82.2 percent; see Nissen 1943:53). Yet this frequency is related to a limited number of adverbs. Only a few, highly frequent adverbs (e.g., adonc, jamais, si) always triggered subject inversion (see Nissen 1943:53; see also Lewinsky [1949:32], who observes the same, though weaker tendency in Berinus). Alternatively, Nissen's data show that the majority of adverbs in initial position trigger inversion in only 50 percent of the instances. This ongoing reduction of the number of adverbs that trigger inversion was another change in the pervasive shift toward dominating SVO in French. Subject inversion related to an introductory element indeed gradually disappeared. It was much less frequent


115

in Middle than in Old French, and it continued to diminish. (Lewinsky 1949:98). In Modern French the number of adverbs or adverbial expressions that require subject inversion is very limited. The occurrence of SOV in the subordinate clause and subject inversion in the main clause introduced by a direct or adverbial complement has a parallel in Modern German and Modern Dutch. Since these languages are codified, their grammatical rules arc much more strictly applied than in Medieval French. Yet Old French, Modern German, and Modern Dutch are all characterized by the shift toward right branching. The observed parallel cannot be arbitrary, but it is still difficult to capture its linguistic implications and to understand this stage in the development. The verbal phenomena we observe in Old French, Modern German, and Modern Dutch can be referred to by the notion of V2 (the verb is bound to occur in second position), which originally is a generativist but quite useful terminology, hence generally accepted. Generativists have tried to account for these phenomena and related Old French and Modern German structures. Working in a generative framework, Marianne Adams, for instance, considers that strong accent and preposing of the verb are the two conditions of V2 (1988:34). The loss of the initial accent is supposed to have triggered the disappearance of V2 in French. To support this hypothesis, Adams proposes a parallel with the Scandinavian languages, claiming that the disappearance of V2 would not have occurred in Old French had the only motivation been the underlying SVO order. This motivation alone could not explain the situation in the Scandinavian languages, which feature SVO, but remained V2 (Adams 1988:24, 34). This explanation raises more questions than it solves. Formulated as such, it does not explain the preposing of the verb in Old French, and the cause of the prosodic change remains to be explained also. In fact, it is legitimate to ask whether prosodies do not support syntax rather than determine it. Focusing on the loss of V2, generative analyses—however elegant they may occasionally be—ignore the other end of the evolution, SOV. Although SVO dominated in Old French, there were still a number of residues of the archaic organization (for example, the occurrence of SOV in subordinate clauses). In the main clause the verb occurred in second position and was preceded by the subject—the normal order—or the complement. The occurrence of the—LB—object in sentence-initial position is marked not only by its stressing (strong initial accent), but also by subject inversion. In this perspective, the sequences OVS are part of an SOV typology that is gradually disappearing, rather than part of a SVO typology, which allows less freedom in the ordering of elements. In French the tendency toward SVO was strong enough to achieve the reorganization of the verbal phrase; by eliminating the possibility of having an LB initial complement, it restored the order subject—verb. It is revealing, indeed, that the direct object was the first element to give way to this change: the syntactic

116


relation between the verb and the direct object is stronger than that between the verb and the adverbial complement; therefore, the preposed direct object is more marked. Moreover, languages of the type VSO, which is an intermediary stage in the development from SOV to SVO in a number of languages, "show no reluctance in placing [verb modifying adverbs and phrases as well as sentence adverbs] before the verb so that the verb does not necessarily begin the sentence" (Greenberg 1963:62). I do not propose a VSO stage in the development of French, but I point out a parallel that seems to support the hypothesis that Old French marks a transitory stage during which SVO was clearly the unmarked order, but where right branching was not yet strictly observed, either in the main or in the subordinate clause. Variations, however, arc clearly marked—as subject inversion shows, for example—and arc residues of the former typology. This same phenomenon can also be observed in Modern German and Modern Dutch, which also undergo the typological shift toward right branching (for the Germanic languages, see Lehmann 1971; Gerritscn 1984).

4.5.5 Interrogative Clauses In Middle French we find sentences featuring initial verbs only in interjections where the order #VS (pronominal or nominal) was compulsory, and in imperatives (Marchello-Nizia 1979:333); these structures were inherited from Old French. The direct interrogative sentence showed a different pattern: more and more often it was introduced and marked by the preposed phrase est-ce que. Although this structure existed already in Old French, it really started to spread in the fourteenth and fifteenth centuries and even occurred in indirect questions (Marchello-Nizia 1979:169). In the beginning its use was limited to pronouns, interrogative adjectives, and interrogative adverbs, but since the sixteenth century the construction has also introduced questions referring to the entire subject—predicate construction and could be used independently (Marchello-Nizia 1979:169). This gradual development illustrates that in interrogative strategies also, the new spreading structures feature SVO; this emphasizes the unmarked nature of SVO and its more and more prevailing dominance. The ever-increasing use of this type of interrogation does not imply that interrogative inversion was completely lost: subject inversion remained a possible strategy in Middle French with nominal or pronominal subjects, even if VSpronominal was henceforth impossible in other contexts (Marchello-Nizia 1979:333). Later, inversion became exclusively limited to pronominal subjects. The interrogative strategies that remained feature right-branching structures: the phrase est-ce que, which is followed by the sentence proper, and interrogation marked by intonation, which also does not change the SVO order of the sentence.


117

4.5.6 The Subject Pronoun The developments examined previously show the ongoing evolution toward the dominance of right branching in Modern French, where the predicate is right branched to the subject and where the direct object follows the verb. Evidence from Old and Middle French shows that exceptions to this organization were highly marked and that they gradually disappeared. Though analyses of the pronominal subject in Old and Middle French do not always corroborate each other, they agree in indicating that subject inversion was more frequent than statistics might suggest. This hypothesis is somewhat doubtful since it is based on data in absentia: subject inversion is claimed to be "hidden," because it is assumed that the pronominal subject is deleted in inversion (see, e.g., Foulet 1923:252). Yet this interpretation is difficult to verify. It is true, however, that the totality of instances of postposcd subjects show a very low percentage of pronouns (sec, e.g., Foulet [1923:265], where these percentages vary from 5.34 to 32.24 percent; also Franzen 1939:130). On the other hand, the percentage of preposed pronouns is very high, for example, in the Quatre Livers des Rois, where 223 of the 233 main clauses with a subject pronoun show the order pronoun—verb (sec Herman 1954:368). Moreover, in the sequences of the type [adverbial complement—subject—verb], the percentage of pronominal subjects is high (see Herman 1954; Price 1961) and the use of this structure continued to increase (Price 1966:478). This frequency is often accounted for by rhythm. Being a weak element, the subject pronoun is said to be readily inserted between the introductory element of the sentence and the verb (Price 1961). On the other hand, it is said that the personal subject pronoun did not add to the clarity of the verbal form and it was used only to fill up the first stressed position of the sentence (Herman 1954:369—70). Consequently, when a nominal subject or a complement was absent, it was the personal pronoun that filled the gap (Herman 1954:371; see also Franzen 1939; Price 1966:477-78). Although these considerations might account for the preposing of the pronominal subject, they definitely do not account for its absence after the verb. Consequently, there has to be another reason. V2 strategies may account for [#pronoun—verb] sequences but cannot explain the high occurrence of [adverbial complement—pronominal subject—verb] sequences mentioned. Generativists try to explain this phenomenon by referring to the direction of government in Old French, which is supposed to determine the principles of pro-drop (Vance 1988; Dupuis 1988). Yet it is important to emphasize that the subject pronoun was only emerging in Old and Middle French. In the translation of Psalms 1—50 (twelfth century), for example, there are 355 instances where a subject pronoun appears in the French text without having its correspondent in the original text

118


(Herman 1954:86). Since the subject pronoun was in the process of being introduced into the language, it occurred in the unmarked position of the subject, that is, before the verb. In the Psalms the subject pronoun precedes its verb in 322 examples (Herman 1954:86). At that time the verbal form was still clear, even without a pronoun (see also Herman 1954:369), because the category of persons was still expressed in verbal conjugation. Consequently the subject pronoun did not have a distinctive function in medieval French and there was no reason to reinforce left branching by putting the pronoun behind the inflected and LB form. The preposed pronoun, on the other hand, had a distinctive feature: it was right branching. Its rise therefore was integrated in the creation of RE structures: as the preposed auxiliary, which replaced the ending, the preposed personal pronoun became increasingly important. The creation of the subject personal pronoun, therefore, marks one of the last stages in the reorganization of the verb phrase. Having inherited the unmarked final position and the marked initial position from PIE, the Latin verbal phrase gradually underwent a structural reorganization. The ever-increasing frequency of SVO at the expense of other variants in Old and Middle French, the appearance of the prcposed subject pronoun and of the sequence [auxiliary-past participledirect object], as well as the current replacing of the LB future form with an RB periphrasis: all these changes mark the achievement of the development toward right branching. In the following section I analyze the development of the other element of the verb phrase, the adverb.

4.6 The Adverb In section 4.5 I discussed the effect of the adverb in sentence-initial position, but I have said nothing on the position of the adverb in the verbal phrase, nor on its morphology. In the following pages I analyze the position and the morphology of the Latin adverb and the evolution this element underwent. The adverb is a pluriform and multifunctional clement: it can be combined with a finite or nonfinte verb, an adjective, or another adverb. Yet the grammatical relations with these elements vary. There is a distinction between the verbal adverb and the degree adverb, which is combined with an adjective or another adverb. In contrast with this element, the verbal adverb is governed by the verb: [[adverb] verb]. The adverbial element in tres elegant or plus elegant Very elegant' versus 'more elegant', on the other hand, has an entirely different function, because it is the grammatical element that expresses the syntactic function of the phrase (see chapter 2 and section 5.2). In these structures the adjective is the subordinate clement. Similarly, in the analytic plus elegamment 'more elegantly', plus is the chief element. In synthetic forms, the ending has the function of head: {[elegant] -issimus] or [{elegant} -ius{. The following analysis takes into account only verbal adverbs, which


119

determine the way the action is carried out and which are governed by the verb. The development of the degree adverbs is analyzed in the following chapter, in section 5.2.

4.6.1 The Morphology of the Latin Adverb However diverse its original may seem—nominal, adjectival, prepositional, or other—the Latin adverb is characterized by morphological regularity. In Latin one distinguishes roughly two types of adverbs: adverbs in -e or -ter, which are derived from adjectives, on the one hand, and adverbs in -tus and -(t)im, which go back to nouns, on the other hand. Both types of adverbs were left branching because the grammatical marker followed the lexical element, which was left branched: the suffix -e was combined with the root of adjectives in -o, such as parvus 'small'; the suffix -ter was the morphological marker of adverbs derived from adjectives of the third declension; and adverbs of nominal origin were characterized by the suffix -tus (e.g., medullitus 'thoroughly') or -im preceded by verbal nouns, as in cursim 'quickly'. It is generally acknowledged that the adverb that has a nominal or adjectival origin often goes back to an inflected case form that became fixed. The adverb in -e, for instance, which features in all Italic languages, is said to be a former ablative (see Buck 1979:136), just like adverbs in -o, which are much less frequent. The suffix -im (< -em) goes back to a former accusative. This formation was limited to the supine, which was the basis for verbal nouns, for example, raptim (< rapio, ruptum 'capture') 'greedily5

The origins of the suffixes in -ter and -tus, on the other hand, arc unknown: -(i)ter has been related by some to the Indo-European suffix -tero, which marked an opposition (Meillet and Vendryes 1924:360; Leumann and Hofmann 1928:299; Karlsson 1981:29 et seq.). Others account for it by referring to the ancient noun iter 'way' (see McCartney 1920:227; Lindsay 1963:549). Although decisive data are lacking to confirm either of the two hypotheses, the absence of analytic structures in iter makes the second interpretation less plausible. Whatever the origin of these endings may be, the Latin adverb was formed according to consistent morphological left-branching processes. Among these formations, the adverbs in -e had the most elaborate morphology, because they could be combined with degree markers to form comparatives and superlatives. Generally other adverbs did not feature degree. The LB formation of the adverb was productive in the entire Latin history, which is illustrated by the application of various adverbial suffixes also to other grammatical categories and subcatcgories: even in Classical Latin some adverbs could feature two suffixes: humanus 'human', humane

120


and humaniter 'humanly'. Similarly, adverbs in -im were taken as an example for other adverbial formations, whose basis was a verbal root without deverbatives, a nonverbal noun, or even a pronoun: viritim (< vir 'man')

'individually'

(Caes. DBC 7.71.7) tuatim (< tua.) 'after your manner'

(Pl. Am. 554)

New and relatively recent formations, as urcecrtim (Petr. 14.18) 'in pailfuls', further show that the adverbial derivation in -im was still an active process in this period of Latin (see also Meillet and Vendryes 1924:446). Moreover, the suffix -tus, which originally was used to derive adverbs from nouns, was added to ancient prepositions (subtus [< sub] 'underneath') and even to adjectives. In Vulgar Latin and Late Latin the adverb was still commonly formed according to the process described (e.g., the synthetic adverbs in the Vulgata, such as granditer [Ronsch 1875:147—53; Karlsson 1981:34]). Even in the period of Gregory of Tours the morphological process was still productive, but in the writings of this author one also observes a generalization of the adverb in -e,even to adjectives or participles which did not carry this suffix in Classical Latin (Bonnet [1968:468], where we find examples like manifeste 'evidently', rauce 'hoarsely', or rare 'sparsely'). On the other hand, adverbs in -iter, which are also very numerous in the texts of Gregory of Tours, seem to go along with the tendency to use very long adverbs (Bonnet 1968:468). One observes the same tendency to use long suffixes in everyday speech and in the inscriptions in Pompeii (Karlsson 1981:38-39). With the derivational criteria thus disappearing, the morphological regularity of Latin became more opaque and gradually the synthetic forms completely disappeared. According to some the disappearance of the synthetic adverb can be accounted for by the heterogeneity of its formation, especially in non-Classical Latin, where several suffixes coexisted, the use of which was sometimes inconsistent. Others integrate the loss of the adverbial formation in the total disintegration of the inflectional system (for a discussion of both hypotheses, see Karlsson 1981:35—38). Yet the loss of the synthetic adverb was closely related to the existence of an alternative, which already existed in Classical Latin. In Classical Latin and even earlier, we find periphrases with adverbial value, for instance, vicem, or ad vicem 'after the manner of, which combined with a genitive or an adjective, as in Sardanapalli

vicem

Sardanapalus-Gen.

after the manner of

[ in suo

in his-Abl.sg.

Diachronic Analysis: The Verb Phrase lectulo]

121 121

mori

bed-Abl. die-Inf. 'to die the way Sardanapalus did [in his bed]' (Cic.Att. 10.8.7) or modum, ad modum, in modum, or modo 'in the manner of "[which was] used in adverbial phrases more frequently than any other word" (McCartney 1920:222-23), in servilem modum 'like slaves' (Caes. DBG. 6.19.3) Generally, the periphrasis consisted of a noun in the ablative (cf. the common ablative origin of the adverb) accompanied by an adjective. The adverbial value of the expression is especially clear in sentences in which it is juxtaposed to an adverb (McCartney 1920:214), as in the following example from Cicero's Tusculanae Disputationes, aguntur leniter et mente tranquilla play-Pass.-Pr.-3pl. gently and mind-Abl. calm-Abl.sg. 'they are played gently and with a mind at peace' (4,25.55) Adverbial periphrases were numerous and diverse: they featured nouns that refer to abstractions (modus 'manner') or body parts that indicate the seat of state of mind (cor 'heart', mens 'mind') or the instrument (manus 'hand'). The expression in mente, which was to remain in the Romance languages, was at the beginning only one of the numerous periphrastic structures (sec McCartney 1920:215—29): "[It was] slow in getting a start toward adverbial usage, but it gathered momentum in course of time" (McCartney 1920:216). This was the beginning of the new adverbial form that was going to become dominant in the Romance languages, but that is characterized by the same type of branching as the former synthetic form (for an explanation of the choice of mente as adverbial suffix in the Romance languages, see Karlsson 1981:43 et seq.). Of the 91 examples of analytic adverbs gathered by McCartney that feature a noun indicating a body part, 71 present the structure [[adjective] noun], as in timido pectore (Pl. Epid. 533 'in shyness'). There are other examples in Karlsson (1981:45). This order can be explained by the descriptive nature of the adjective that occurs in this context and that normally preceded the noun (see section 3.2). These data permit us to account for the LB pattern of the French adverb: the analytic form became again a synthetic form before the descriptive adjective changed to right branching.

122


As a result of this process, ment(e) became a left-branching adverbial suffix. The adverb is one of the last residues of the former left-branching morphological organization. Its history shows that left-branching units tend to become synthetic forms (see also section 6.5). This process of synthesis must have occurred before the eighth century. In the Glosses of Reichenau, the item singulariter 'particularly', for instance, is rendered by solamente, whereas in the Lex Salica (507—511 A.D.), the form is still analytic, sola menu (Karlsson 1981:47-48). In the first French text, the Serments de Strasbourg, one also finds the other French left-branching synthetic form, the future tense prindrai. Both processes of synthesis must have taken place at roughly the same time before the change in branching. The French future tense and adverb, which are the very few remaining left-branching forms of Modern French, were thus originally analytic structures that became synthetic forms. The development of French therefore shows that morphological postposed heads tend to become suffixes. At face value the development of the past participle seems to be an exception to this tendency, because in structures of the type epistulas scriptas habeo the verbal elements did not become synthetic forms. Yet these constructions were different. In the beginning the past participle was not governed by the finite verb, but depended on the noun that was the direct object of habeo: [epistulas [scriptas]] habeo (see also section 4.4.2). Alternatively, the infinitive and the adjective depended right from the beginning on the clement that was going to be the suffix, habeo and menu, respectively. Although the LB adverb is still very productive in Modern French, one often observes periphrases that arc not only analytic but RB as well, for instance, de facon + adjective.

4.6.2 The Position of the Adverb Morphologically the Latin adverb was an LB element and the suffix originally was an ending; in the Romance variant the origin of the suffix is lexical, but the structure is still synthetic and left branching. Whereas the morphological analysis of the Latin adverb is clear and unambiguous, it is more difficult to determine the unmarked position of this element. One of the reasons might be the occurrence of the adverb at several syntactic levels: it can determine the meaning of an entire sentence, a syntactic unit, or simply a word, as in te spectat et audit dulce ridentem you-Acc. look-3sg. and listen-3sg. sweetly laugh-Part.-Acc.sg. 'looks at you and hears you sweetly laughing' (Catul. 51.5)


123

When the Latin adverb was part of a longer unit, its place was indiscriminately free (Marouzeau 1949:13). On the other hand, the adverb that was associated with only one element was less free. Only in these instances (Marouzeau 1949:33) is it possible to distinguish the normal or usual, and unmarked order; inversion of this order marks an exception (Marouzeau 1949:14). This tendency shows especially in local adverbs (intro ibas, Ter. Eun. 87, 'you came in') and in temporal adverbs, which—very frequent and very general in meaning—determine circumstances. These adverbs, as well as those that go back to descriptive adjectives and had a very general meaning (bene, male, ridicule, etc.) preceded the word they modified, be it a verb, an adjective, or an adverb (Marouzeau 1949): non recte iudicas de Catone not rightly judge-Pr.-2sg. over Cato 'your judgment about Cato is not correct' (Cic. Lael. 2.9)

This unmarked order is reflected in fixed expressions and formulae: ego me bene habeo I-Nom. me-Acc. well have-Pr.-lsg. 'I am well' (Tac.Ann. 14.51.3) bene ambula et redambula well walk-Imper.-2sg. and walk-back-Imper.-2sg. 'a pleasant walk and walk back' (Pl. Capt. 900) bene vale 'be well; goodbye' (Pl. As. 606; Mil. 1352; Am. 499)

quid faciemus puero ? ridicule rogas what do-Subj.-lpl. boy-Abl. absurd-Adv. ask-Pr.-2sg. 'what are we doing with the boy? An absurd question' (Ter. Hec. 668)

Similarly, compounds featuring an adverb are left branching: benedicere 'praise', maledicere 'curse' or male facere 'harm', male facio or malefacio, satisfacio 'satisfy', malevolens 'ill disposed', maleloquor 'speak evil', and their derivatives. In addition to these formations, observations of Latin grammarians support the hypothesis of the unmarked preposing of the adverb proposed by Marouzeau (1949): Charisius, for instance, stipulates that the adverb precedes and modifies the verb: "advcrbium est pars orationis quac praeposita verbo significationem eius implet atque explanat" (2.13; emphasis added; quoted from Keil 1961:1 181).

124


Finally, statistical data from more recent analyses corroborate the hypothesis of the unmarked preposed adverb; sec Pinkster (1971) and Adams (1976). Pinkster has demonstrated that especially adverbs of adjectival origin (graviter 'heavily', celeriter 'quickly', etc.) tend to precede the head (81 percent) and that they precede it directly in 63 percent of the instances (1971:124-25). Although Pinkster's statistical data only take into account a limited number of adverbs, it is clear that the "descriptive" adverb was the most regular adverb in Latin. Whereas in Classical Latin the unmarked adverb preceded the verb, the reverse order marked emphasis—cither of the verb or of the adverb (see Marouzeau 1938, 1949). Yet the preference for the preposed adverb diminished in the course of time. Data presented by Adams, for instance, show that prepositional phrases with adverbial value modifying the verb shifted from preposing in the first Latin texts (the Twelve Tables, S. C. de Bacchanalibus) to postposing in the Vulgar Latin writings of Valcsianus and Terentianus (1977a:90—91). Similarly, in Petronius's novel the occurrence of the adverb in clause-final position is "remarkably" frequent (Feix 1934:31), especially in the dialogue of noneducated people. Parallel to what I pointed out when discussing the place of the direct object, we observe that preposing of the adverb was more resistant in subordinate clauses, where "the lingering tendency for the verb to occupy final position causes anteposition to predominate" (Adams [1976:139], discussing the text of Anonymus Valesianus). This development that had started in Latin continued and was even consolidated in Old French. In the Quatre Livres des Rois, for instance, the simple adverb generally follows the predicate (Haarhoff 1936:70). The analysis of Buridant (1987) of two translations of the Chronique des Rois de France, which originally was a Latin text, offers more precise information on the shift from AdvV to VAdv. The first manuscript (A) is from the end of the thirteenth century or the beginning of the fourteenth century and was the work of a poorly educated transcriber; the second manuscript (ms B) is from the end of the fifteenth century and the transcriber was more meticulous (Buridant 1987:31—32). In manuscript A we find 12 instances where the adverb precedes the verb, whereas the order is the reverse in manuscript B (Buridant 1987:53). Since the verb in question is either a past participle or an infinitive, these instances illustrate that the tendency to postpose the adverb was strong, because it even applied to compound verbs. These data, therefore, suggest that in the fifteenth century the verbal adverb normally followed the verb. Inversion still occurred, but only in marked contexts. This conclusion corroborates Dill's findings: in the Cent Nouvelles Nouvelles the position of the adverb is that of the modern adverb. Exceptions can be accounted for by rhythmical needs or emphasis (1935:54-55). Yet traces of an intermediary stage during which the verb is inserted between the auxiliary and the participle/infinitive can be found even in the


125

seventeenth century. This development parallels the change of the direct object, which as a first step also was inserted in the right-branching sequence [auxiliary [participle/infinitive]] (see section 4.4.2). Oudin (seventeenth century), for example, who barely accepts the insertion of a simple adverb between the nominative and the verb, argues that a simple adverb must be placed between the auxiliary and the verb (but closest to the participle) (Oudin; quoted in Brunot 1922:682-83). At the end of the century these grammatical rules finally changed (for stylistic variations and grammatical judgment, see Brunot 1924). Evidence of the Latin adverb confirms the left branching of the verbal phrase and its gradual evolution toward right branching. The place of the adverb is well defined when it is part of a cohesive syntactic group: the development of this adverb can be traced in detail and is important for the current analysis. The adverb that modified a sentence or a clause was much freer, even in Modern French, which is a strict right-branching language. On the morphological level, the Latin adverb changed its marker, but not its branching. The future form and the adverb are the few morphological LB structures that are left in Modern French: they differ from the imperfective, for instance, in that they went through an attested analytic stage.

4.7 Provisional Conclusion: The Verbal and Nominal Phrase Although my analysis of French grammatical structures and their Latin antecedents is not yet complete, examination of the verbal and nominal phrase suggests a number of conclusions. First, the analysis of these phrases demonstrates that the order of syntactic elements was neither indiscriminately free nor arbitrary in Latin; this conclusion refutes the tenacious hypothesis of Latin's alleged free word order. Instead of an indiscriminately free word order, the nominal and verbal phrases displayed an unmarked order and a number of variants, which were stylistically or syntactically motivated. My comparative analyses of various stages of the evolution of these grammatical structures revealed, moreover, a fundamental shift in their internal organization: in the archaic structures the complement of the verb and of the noun preceded its head; in the modern structures this subordinate element follows. The more one goes back in time, the more these grammatical structures are left branching. Consequently, the analysis of their evolution reveals the shift in branching, which underlies syntactic as well as morphological changes. Although in morphology the creation of analytic structures is undeniable, it is less important than the change in branching, which affects the hierarchical organization proper and which exceeds the limited scope of morphology. Analyticity is only a secondary development, which is especially well illustrated in the history of the adverb and the future. These forms show that left-branching morphological structures tend to become synthetic forms, whereas right-

126


branching elements remain separate (I will come back to this tendency in section 5.4). Analysis in terms of branching reveals the consistent shift in morphological structures which parallels the change in syntactic phrases. Consequently, the notion of branching assigns to morphology the position it deserves: instead of being a simple appendix of SOV typology, inflection now is recognized as featuring the same structural organization as syntax. The regularity of the morphological processes and the parallel with syntax underscore the nonrandomncss of the grammatical sequence and challenge the point of view of Herman, for example, who claims that Latin never was an SOV language and that it is a "grossiere simplification" to consider the history of Latin word order as a shift from OV to VO (1989:147). His conclusion illustrates that the pragmatic approach alone docs not account for the analysis of a syntactic phenomenon such as word order. Yet a pragmatic description of word order is not excluded as such, but can only be carried out after having determined the unmarked order of the language in question. In Latin this was SOV, or in my terminology left branching. The advantage of a grammatical analysis over purely pragmatic studies, in my view, is that a grammatical analysis includes all constituents, hence also constituents that are less sensitive or not sensitive at all to topic/ comment mechanisms. Second, it includes morphology, where we also observe a change of ordering patterns over time. Moreover, a syntactic analysis in terms of branching based on an independently motivated definition of the grammatical terms involved seems stronger to me than purely pragmatic analyses, the terminology and interpretations of which arc not always conclusive or do not include all structures. Finally a large corpus of data, including evidence from other, independent and highly specialized studies, as well as the diachronic perspective—in both directions, IndoEuropean and Romance—allow one to grasp the ordering patterns in the early Indo-European languages (and also PIE with an analysis based on comparison and internal reconstruction) and their subsequent evolution. Although the shift in branching took place in syntactic as well as morphological structures, it took place at different moments in the various structures. In general, morphological development is slower than syntactic change. Whereas postposing of the genitive, for instance, was a Latin development, the analytic structure introduced by de was only carried out in French (although, of course, the tendency toward analyticity also had started in Latin). Moreover, in noun phrases featuring an adjective, the shift toward right branching took place before Old Latin, whereas morphological elements, such as the adverb or the future, arc still synthetic LB structures in Modern French. These examples arc also symptomatic of the delay of the verb phrase with respect to the noun phrase. This gap refutes the decisive role attributed to the verb. Considering the verb the head of the clause, Lehmann (1972c) and Venncmann (1974) argue that the changes that occur in the verb phrase trigger the entire typological


127

reorganization. These interpretations arc essentially based on the development of word order in Germanic languages, which is characterized by the late introduction of right branching in nominal phrases (for data on English, sec also Canalc 1976). Yet Italic data and, more precisely, evidence from Latin refute this chronology and raise serious doubts about the generalization of the regularity revealed in the Germanic languages and the dominant role of the verb. The divergence between Latin NPs and VPs, which contrasts with the chronology in Germanic languages, suggests that on the level of the noncomplcx phrase, there is no specific chronology in the structural reorganization we observe in the IndoEuropean languages. In the next chapter I examine structures whose reorganization took place early in the history of Indo-European languages: the prepositional phrase, the comparative, and the relative clause. In contrast with the other phrases analyzed in this book, the comparative and the relative clause arc complex structures and their early reorganization has important theoretical implications.

5 Diachronic Analysis: Early Right-Branching Structures

In this chapter I discuss three structures that underwent the shift from left to right branching at an early time: the prepositional phrase, the comparative construction, and the relative clause. As a first step I analyze the rise of prepositional phrases, which were RB in Latin, although one still finds instances of the former LB period. I then set out to analyze the comparative construction and the relative clause: the modern RB sequences are the result of a long and complex evolution. In contrast with the nominal, verbal, and prepositional phrase, the relative clause as well as the comparative construction are complex structures, and their early structural change is important in an explanatory perspective. In chapter 3 I gave a definition of complex structures that I briefly repeat here: in complex structures the complement is (1) itself a hierarchical unit, such as the relative clause (section 5.3) or the clause introduced by the comparative particle quam (section 5.2.2.4); or the complement is (2) part of yet another structure. This latter variant is illustrated in a number of comparative constructions (see section 5.2.2).

5.1. The Prepositional Phrase In the French prepositional phrase the complement follows the head; it is an RB structure. This is the result of an evolution that started at a very early period and that is characterized by the creation of a hierarchical structure, and subsequently by the change in branching. In the following pages, I first analyze the creation of the hierarchical structure (section 128

Diaehronic Analysis: Early Right-Branching Structures

129

5.1.1), which subsequently shifted from left to right branching (see section 5.1.2). I then set out to determine the place of the Italic languages and especially of Latin in this development, attempting to discover the underlying regularity in the use of prepositions and to evaluate the exceptions (section 5.1.3). The tendency toward right branching, which began very early, has been extended later in the comprehensive substitution of prepositions for oblique cases, not only in nouns but also in nominal forms of the verb (section 5.1.4).

5.1.1 The Creation of a Hierarchical Structure The Indo-European adposition generally goes back to an adverbial element or to a PIE particle, which was a highly autonomous and mobile element. It is possible that prepositions originally were ancient fixed case forms, but their formation is too remote to be clearly analyzable (Meillet and Vendryes 1924:479). From the PIE particle two very different elements originated: the preverb and the preposition (for the parallel development in various IndoEuropean languages, see Coleman 199la). The prevcrb ultimately formed a cohesive unit with the verb, which is the head of the verbal phrase, the syntactic status of which does not change. The preposition, on the other hand, is a grammatical element: it is the head of the prepositional phrase and governs the noun. In Indo-European languages it is either postposed (archaic languages) or prcposed (modern languages): hence the term "adposition," which denotes the element without indicating its position. Although the creation of the adposition is later than the creation of the preverb, both elements formed a unit with the verb and the noun, respectively, from the fifth century B.C. on (sec Schwyzer and Debrunner 1950:420 et seq.). Homer's language still reflects the looseness of the period of transition; the elements in question occur in their original varied usage. According to Chantraine, the Homeric text offers numerous examples indicating that the rise of the adposition was not yet fully accomplished. In these instances the specific status of the element is not clear; it is functioning as either a preposition, an adverb, or a preverb (1953:84). Moreover, although a number of Greek adpositions governed only one case, most of them occurred with more cases, which further illustrates that the Homeric preposition was not yet fully developed. In a multicase system, each case reflects a specific value of the preposition (for an overview in Greek, see Chantraine 1953:87-124); this hybrid structure (a productive and linguistically relevant LB case ending combined with a preposition) shows that the relation between the preposition and the noun was not yet strong enough for the preposition to determine the case of the noun. Recently, Coleman has stressed the archaic status of multicase complementation: "In fact Ancient Greek with only three oblique cases represents the most conservative situation anywhere in IE, with many prepositions occurring

130


with two and some with all three cases" (1991a:330). More evolved than Greek, Latin had a reduced number of prepositions that governed two cases (see section 5.1.3.2).

5.1.2 The Shift Toward Right Branching The postposed adposition is an archaic feature: "[Q]uand la rection s'est constitute, la preposition se trouvait souvcnt placee apres son regime' (Meillet and Vendryes 1924:480). Early Indo-European languages indeed displayed this organization. In Sanskrit, for instance, we find numerous postpositions whose cognates are prepositions in more modern IndoEuropean languages. Moreover, whereas postpositions dominated in Tocharian, adpositions were exclusively postposed in Hittite, which is a more archaic language (Friedrich 1974:129). The instances of preposition we find in the Hittite texts are Akkadian (see Friedrich 1974:178-80). Since a number of these Hittite and Sanskrit postpositions occur as prepositions in Greek and in other languages, we may conclude that proposing is an Indo-European innovation: for example, compare the Hittite postposition katta to the Greek preposition KCITCX 'downwards'. In Greek, which was an archaic language in numerous respects, the prepositional phrase, though RB, displayed a number of ambivalent features which disappeared over time. In Homeric Greek direct proposing was more common than were postpositions or discontinuous prepositions. Yet although the adposition generally preceded the noun (Chantraine 1953:83), one still finds numerous instances of postposition. These examples are not related to a specific category of a given context as in Latin. Accordingly, postposing was freer and more common in Homeric speech, but in Attic Greek anastrophy (that is, permutation of the preposition and the object) was only used for 'round about' (Meillet and Vendryes 1924:480). Adpositions of nominal origin, which were postposed, represented a specific category (see section 5.1.3.2). In Greek, therefore, the tendency to use prepositions gradually became stronger, consistently with the general development of Indo-European languages.

5.1.3 The Italic Languages: Residues of Former Postposing Although the RB prepositional phrase was an early creation, one still finds residues of the former organization in Italic languages. Whereas Latin, in contrast to Umbrian and Oscan, did not have genuine postpositions, the Romans postposed their prepositions in given contexts.

5.1.3.1 Oscan and Umbrian The Indo-European evolution toward RB prepositional phrases became stronger in the Italic languages. Yet Oscan and especially Umbrian still featured a number of postpositions, which we find in Latin as preposi-

Diachronic Analysis: Early Right-Branching Structures

131

tions. Of the 16 attested adpositions in Urnbrian sources, 5 were LB. The adpositions -a(r), -per, and -co(m)/ku(m), for example, were postposed in Umbrian, whereas their Oscan and Latin cognates (ad, pro, and cum) preceded their nominal complements (see Buck 1979). Moreover, although -en was a postposition in Umbrian, it occasionally followed its complement—even nominal—in Oscan, but was always prcposcd in Latin. Accordingly, although in Italic languages the prepositional phrase was generally RB, one still finds genuine postpositions. In Umbrian, the most archaic of the three languages discussed here, postpositions represented one third of all attested adpositions. They were, however, residues of the archaic structure. We can therefore conclude that in the prepositional phrase the shift from left to right branching occurred at an early time, in the Proto-Italic period or even earlier.

5.1.3.2 Latin In contrast with Greek, Latin prepositions normally governed only one case, even if Latin had a number of prepositions that governed an accusative or an ablative according to the presence or absence of movement. Moreover, the syntactic status of the Latin forms was clear: they were mainly prepositions or preverbs, although their use as adverbs was not excluded. In the following pages I attempt to show (1) that the Latin preposition was indeed an element that was juxtaposed to its complement and (2) that instances of postposing were confined and context-related and that they eventually gave way to the ever ongoing tendency toward RB. 5.1.3.2.1 Direct Juxtaposition in the Prepositional Phrase. In accordance with the tendency in Italic, the preposition already preceded its complement in Old Latin (Leumann and Hofmann 1928:495). Moreover, the preposition preceded the complement directly (Kuhner and Stcgmann 1955:583; see also Leumann and Hofmann [1928], or Ernout and Thomas [1951]). Although insertion of an element between the reposition and the complement was "extremely rare" (Marouzeau 1949:42; see also Leumann and Hofmann 1928:495), it sometimes did occur, but only in specific contexts: The inserted element was either the clitic -que 'and' or -ve 'or'. Yet the occurrence of clitics in the given context has less to do with discontinuous prepositional phrases than with the process of clitization itself. In addition, in formulae of adjuration, in curses, and in exclamations the adposition per, always preposed, was often separated from its complement by an interjection. The formulaic status of oaths gives the structure a definite genuineness and emphasizes its ancient character. We still find examples of it in Classical Latin, but they occur especially in writings of poets and archaistic authors (see Marouzeau 1953:62). Prepositional disjunction in Latin was therefore an uncommon phenomenon that occurred only in specific and archaistic contexts and registers.

132


Data from Latin and, as we saw earlier, from other Italic languages suggest moreover that the prepositional phrase was definitely right branching. This organization is illustrated in Latin formations, as in invicem (< in + vicem) 'in turn', denou (< de + nouo) 'again', sedulo (< sine dolo) 'sincerely', or, with a pronoun, postea 'thereafter' (see Ernout and Meillet 1959). Yet Latin still featured residues of the former situation, such as eapropter 'therefore' or quoad 'as long as', which coexisted with the more modern propter ea and adquo. Consequently, it is important to determine the status of these examples by analyzing the context in which the adposition followed its complement. 5.1.3.2.2 The Sequence [[Complement] Adposition]. Whereas preposing was the normal order in Latin, postposition was either a stylistic strategy or the residue of an old use. There is a definite distinction between anastrophy, which is the transposition of the preposition, and actual postposition. Postposition is a grammatical subcategory with its own syntax; anastrophy is a stylistic process. Lucretius, for instance, introduced the fashion to use anastrophy in prepositional phrases "in kunstlicher Weise" (Leumann and Hofmann 1928:495; emphasis added). For example, tempore de mortis period-Abl. from death-Gen, 'from the period of death' (3.1088)

This is definitely a literary expression, stylistically marked and highly influenced by Greek: anastrophy therefore is not part of language change. In addition to this stylistic phenomenon, postposition was also a residue of the typological pattern of PIE and only occurred in highly exceptional contexts (Marouzeau 1949:44). The adposition was generally postposed, for example, when its complement was a relative pronoun, a situation we also observe in Modern German: er geht nach Paris mit he-Nom. goes to Paris with 'he goes to Paris with the car'

dem Auto the-Dat. car

versus das Auto, womit the-Nom. car that-with 'the car he takes to Paris'

er he-Nom.

nach Paris gebt to Paris goes

In Latin this tendency is illustrated in compounds, such as quoad,, quaad, quocirca 'wherefore', or quapropter. The antiquity and genuineness of this type of formation arc illustrated in the use of i in quicum 'with whom', which occurs in Latin texts from Plautus until Caesar and which represents

Dictchronic Analysis: Early Right-Branching Structures

133

a former masculine, feminine, and neuter ablative. The creation of quocum, quacum, and quibuscum is more recent: these forms are attested from Plautus and Terence until Late Latin and coexisted with the ancient forms for a long time. Yet even the well-attested use of the sequence [[relative pronoun] postposition] became less popular in the course of time (Leumann and Hofmann 1928:495). Whereas in Plautus, for instance, every adposition followed the relative pronoun, disyllabic adpositions preceded this element in Terence. Moreover, plurisyllabic postposed adpositions basically are attested only in early times and in poetic texts (Marouzeau 1949:47). Cum has been longest in use as postposition in this context, even if cum quo, cum qua, and cum quibus coexisted with quocum, quacum, and quibuscum from the Classical period on (Ernout and Thomas 1951:101). Combined with a noun, cum normally preceded it, but it formed an LB structure with a pronoun: cum iumiintum

'with a mule' (C.I.L. 4.8976)

versus quocum 'with whom'

Although preposing was very uncommon in a pronominal phrase with cum (Marouzeau 1953:64), exceptions to this rule suggest the gradual reversal of the structure: in early inscriptions, for example, we find instances where a second, preposcd cum reinforces the LB pronominal sequence (see 1949:46): con quicu 'with whom' (C.I.L. 11.5779) con qua com 'with whom' (C.I.L. 6.164.14)

In th Vulgata, moreover, we find examples of cum qui. In addition, writings of authors of the same period and the same genre, or even writings of one and the same author show that the place of cum was highly variable. See, for example, Caesar or Sallust, who use postposed cum regularly with plural pronouns but rarely with singular pronouns (Marouzeau 1953:64; for data on poets, see Marouzeau 1949:45). This irregularity of its use is due to the disappearance of this usage (Marouzeau 1953:64). The data presented suggest indeed not only that the use of postposed cum was limited to a given context, but also that it gradually disappeared. Whereas cum was in a transitional stage, other adpositions had developed more quickly. De, for instance, followed the relative pronoun only in fixed formulae of juridical language (see quo de agitur 'the point in question' [Cic. Verr. II 2.12.31]). Postposiiig of ex in structures featuring a

134


relative pronoun occurs only in Plautus and Ennius and is rarely attested in later periods. The postposing of yet other adpositions (ad 'to', ab 'from', per 'through', pro 'in front of) was even more restricted and was exclusively limited to political, juridical, or administrative language (sec Marouzeau 1953). In phrases featuring personal and demonstrative pronouns, cum once more occurred longest in postposition. Cicero even ascribes the compulsory postposition of cum to reasons of propriety: once the sequence nobis cum was formed to avoid con nobis, it spread, according to Cicero, to the entire paradigm (Or. 154) (Marouzeau 1949:48-49). Whereas the difference between cunnus Vulva' and connobis (< cum nobis} might be clear enough to prevent any ambiguity, it does not seem appropriate to discuss these sensitivities with Cicero but it would be excessive to explain the entire paradigm by referring to rules of etiquette, the more so since postposition also occurred in other contexts and with other adpositions. Moreover, there is a grammatical, diachronic explanation for this phenomenon, which is preferable. Postposing of cum in pronominal context is further illustrated in fixed formulae, mecum, tecum, vobiscum, and so on. Even structural symmetry could not interfere with this regularity: cum Caesare nobiscumque with Caesar-Abl. us-Abl.-with-and 'with Caesar and with us' (Cic.Fam. 9.9.1)

Yet like other adpositions, cum eventually gave way to the RB organization. The preposition precedes its complement in all contexts in Modern Italian, Portuguese, and Spanish. Spanish conmigo 'with me', which goes back to cum mecum, is a relic of the transitional period. In Ercnch, cum eventually was replaced with apud followed by hoc, which gave the RB preposition avec. 5.1.3.2.3 Postpositions of Nominal Origin. In Latin we find a group of adpositions that differ from all other adpositions in several respects. They were recent formations and of nominal origin: generally they were a fixed nominal form in the ablative. The original noun generally continued to exist. Moreover, in contrast with "historical" adpositions, these adpositions were postposed and most commonly governed only one case, the genitive, which can be accounted for by the nominal origin, as in: bominum causa men-Gen, for the sake of 'for the sake of men' (Cic. Nat. Deo. 2.158)

Diachronic Analysis: Early Right-Branching Structures per through

135

mare umbilici fine ingressi sea-Ace. waist-Gen, to wading-Nom.pl.

terram petebant land-Ace. go-Impf.-3pl 'wading waist-high through the sea, they made for the land' (Bell. Aft-. 85.1)

Alternatively, tenus 'as far as', which is an accusative or a nominative, governed a genitive or an ablative: lumborum tenus 'as far as the loins' (Cic. Arat. 83) Tauro tenus 'as far as the Taurus' (Cic. Dei 13.36)

This same type of adposition also existed in other Italic languages (Buck 1979:211), in Greek, and, though much later, in German (see Lehmann 1971). In all these languages these adpositions were late formations and governed a genitive, as in Greek 'for the sake of' or German wegen 'because of (see Bauer in press a). Whereas the use of the genitive can be explained by the nominal origin of these adpositions, their postposing is more difficult to account for: these elements of nominal origin followed their complement in Latin at a period when the prepositional phrase was already RB and the genitive normally followed the head noun, [head noun [genitive]]. Consequently, the formation of these postpositions must have taken place before the reversal of word order in nominal phrases. The ambivalence of the form, which was at the same time an adposition and a noun, might explain why this type of phrase did not change into an RB structure. Being postposed, the adposition was thus distinct from the regular noun. Yet since the exact date of the creation of these adpositional elements is still lacking, we cannot support this assumption with conclusive evidence. The preceding data show that, once again, postpositions formed a specific subcategory: the postposed adposition was, therefore, a marked order in Latin. Later these adpositions of nominal origin either disappeared or gave way to the tendency toward RB. Gratia 'for the sake of and causa, for example, occur in Modern French as elements of complex prepositions (grace a, a, cause de) and in the Peregrinatio, gratia precedes its complement (Lofstedt 1911:220). Since the nature of either the adposition or the complement (nominal vs. pronominal) accounts for left branching in Latin prepositional phrases, it is legitimate to assume that this was the marked order in Latin. Consequently, the place of the adposition was less arbitrary than might appear at face value. Some linguists indeed assume that the Latin adposition was a free element, and they support their claim by referring to adpositional

136


insertion: sometimes the adposition is inserted between the elements that constitute its complement, for example, arbusta per alta timber-trees-Acc. through tall-Acc.pl. 'through tall timber-trees' (Enn. Ann. 187)

According to Marouzeau, this structure reflects the hesitation of the speakers of Latin between preposing and postposing (1953:67). Yet closer examination of this phenomenon provides not only a more convincing explanation, but another argument in favor of the hypothesis of the unmarked preposition. Initially, the construction was practically excluded— at least in prose—when the elements of the adpositional complement were coordinated (sec Marouzeau 1953:57). The situation was different when these elements were in a dependency relation, for example, a noun and an adjective. When the noun came first, insertion of the adposition was rare and at first occurred only in texts in verse (Marouzeau 1953:67). Later, during the Roman Empire, prose writers also used this construction, which, however, was confined to literary style. On the other hand, insertion has always been frequent when the noun followed the adjective (Marouzeau 1953:68). Compare (A) arbusta per alta (Enn. Ann. 187),

which was rare, with (B) pauca in verba few-Ace.pi. in words-Ace. 'condense in a few words'

confer condense-Imper.-2sg.

(PI. POT. 661) which was much more common (sec Marouzeau 1949:58—60). These two structures differ not because of the alleged arbitrary use of the adposition (see the interpretation of Marouzeau), but for other reasons. In the structure of the type (A) the adposition follows its complement (arbusta) that is head of the nominal phrase. In structure (B), on the other hand, which is characterized by direct preposing of the particle, it is the adjective that does not occur in its normal unmarked position (see section 3.2). Since this structure was frequent when the modifier itself was an intensive (Marouzeau 1953:68), it is indeed legitimate to conclude that the "whimsical" preposition was not responsible for this sequence, but instead, that the modifier occupied an unusual place for reasons of emphasis. Consider the following example featuring a descriptive, hence LB, adjective:

Diachronic Analysis: Early Right-Branching Structures Magna great-Abl.sg. curari

cum with

137

cura . . . ilium care-Abl. him-Acc.

polo

care-Pass. Inf. want-Pr.-lsg. 'I want him to be cared for with much care'

(PI. Men. 895)

The determinative role of the adjective in this context is further demonstrated in structures featuring a relative pronoun, which nearly always occurred in initial clause position and which thus triggered postposing of the preposition (see qua ex pane [Cat. Ayr. 136] 'in proportion to the share'). This type of structure is frequent in all types of Latin, whereas the adjectival variant can be observed especially in poetic writings (Marouzeau 1949:60). Consequently, these constructions, which at face value might illustrate the arbitrariness of the adposition, are consistent with precise rules and support, in fact, the unmarkedness of the RJB prepositional phrase in Latin. The sequence [noun-adposition-adjective] was extremely uncommon, because it featured a postposed adposition. Alternatively, in the sequence [adjective-adposition-noun], which was much more frequent, the adposition normally precedes the noun. The left-branching adjective or more precisely its leftward extrapolation, is less surprising than the leftbranching adpositional complement (see section 3.2). The place of the Latin preposition, therefore, was far from arbitrary. Although it is possible to explain the unequal frequency of both types of disjunction, we still have to account for a syntactically unusual order, because the preposition is found in the middle of its phrase. This syntactic irregularity can perhaps be explained by prosodies. Whereas -cum comes close to being a clitic particle in vobiscum, cum in magna, cum cura may well have the same value. This assumption implies that people said magnacum cura rather than magna cumcura. The data presented suggest that ever since Old Latin the complement followed its head immediately in unmarked order in prepositional phrases. Exceptions to this regularity are well defined. In addition to stylistic expressions, postposing also occurred as a residue of the former structure or was confined to adpositions of nominal origin. Yet these instances of left branching, which were highly marked, all gave way to the change that was taking place, either because they were replaced by a right-branching preposition or because they became prepositions themselves.

5.1.4. Extension of the RB Prepositional Phrase Once the RB prepositional phrase was well established, the tendency toward this type of construction underwent a second expansion in Vulgar Latin and Late Latin, where the prepositional phrase replaced an important number of LB structures.

138


5.1.4.1 Nominal Declension The most obvious change was of course the substitution of prepositions for case endings. This change was twofold. Preposed adpositions replaced the old LB case endings, on the one hand, and the accusative, on the other hand, became the only oblique case, marking the loss of the LB morphology. The beginnings of this development are to be dated very early, when the accusative of movement and the ablative of separation were reinforced by the appropriate prepositions; when the dative tended to be replaced with ad + accusative, already in Plautus (see Vaananen 1963:111—21); and when the accusative was increasingly used as a universal case in everyday speech, for instance, in the inscriptions in Pompeii, such as cum saddles (C.I.L. 4.221) 'with comrades' and cum iumiintum (C.I.L. 4.8976) 'with a mule' (Vaananen 1966:121). Yet only in Late Latin did the replacement of the LB inflection by an RB structure become a really widespread phenomenon. In the Vulgata, for example, we observe a high incidence of the preposition instead of a simple case and a considerable number of instances where the case governed by a preposition, an adjective, or a verb is no longer dictated by grammar, but by the personal preference of the author. In these instances, the use of the accusative instead of the ablative is indeed more common than the reverse (see Ronsch 1875:406-15). Finally, the important inflow of exotic and indeclinable names, due partly to Christianity, did not contribute to the survival of the case system. Accordingly, "the great irregularity in the employment of cases [in late writers]" (Grandgent 1907:42) was symptomatic of the breakdown of the case system. For the replacement of case endings with prepositions, see Vaananen (1963:119-22; 1966:115-19), Grandgent (1907:42-48), and Muller and Taylor (1932:65). The replacement of inflection by prepositions cannot be summarized merely by referring to the analytic development of language, even if the modern structure is indeed analytic. The chief issue is the replacement of the postposed case ending, which formed an LB morphological structure with the lexical root, by a preposed particle, which results in an RB analytic structure. Consequently this substitution is integrated in the comprehensive shift toward right branching.

5.1.4.2 Declension of Nominal Forms of the Verb The prepositional structure was also gradually introduced in the verbal paradigm. The supine, for example, was replaced not only with the infinitive, but with ad followed by the gerund (gerundium). Moreover, Aalto (1949) presents numerous examples in which prepositions with gerund replace the oblique case of the same form. Prepositions arc also frequent in constructions featuring a verbal adjective in -ndus (gerundivum) (1949:105—15). In the course of time the gerund and its adjectival form eventually were replaced, in part, with the infinitive. Yet the gerund and the infinitive did not display the same syntax. Since the infinitive is charac-

Diacbronic Analysis: Early Right-Branching Structures

139

tcrizcd by verbal syntax, it governs a direct object, whereas the gerund in early times did not. In order to express the patient of the action referred to by the verb, Latin did not use the gerund, but instead an adjectival construction featuring the form in -ndus. In two thirds of the instances the verbal adjective followed the noun in these constructions, except in the genitive, where the order was the reverse (Aalto 1949). Postposing of the form in -ndus is consistent with the position of the distinctive adjective and of the participle, which is also an attributive element. Whereas in gerundive constructions the genitive and the dative could be expressed only by case marking, the preposition often was combined with the ablative, and with accusatives its use was even compulsory. This particle always preceded its complement, which was thus RB (see the corpus of ca. 220 instances in Aalto 1949:109-15). When the adjectival form followed the noun, the structure was completely RB: [preposition [noun ob regnum because-of kingdom-Ace. 'to protect his kingdom'

[gerundive]]] tutandum protect-Ger.-Acc.sg. (Sail. Jug. 102.12)

Yet in the course of time the gerund developed verbal syntax and the construction with the verbal adjective (gerundivum) could be replaced with the gerund followed by a direct object (see Bauer 1993a). Whereas in Aalto's corpus postposing of the verbal adjective dominates, the order is the reverse for the gerund: "[D]as Objekt steht fast immer nach dcm Gerundium" (Norberg 1943:208). The direct object was, therefore, RB in the beginning at least. From the seventh century on, however, the object was frequently placed between the preposition and the verb (Norberg 1943:208-9). This change is parallel to the explosive extension of the infinitive combined with a preposition, which eventually replaced the gerund, but which was rare in an earlier period. In combination with the infinitive, the direct object was LB for a long time. Yet this was the period of Latin bilingualism; the language of that period docs not lend itself to the analysis of language change. Moreover, this feature does not affect right branching of the prepositional phrase. The preceding data on the syntax of the gerund and its development are brought up to show the use of preposed adpositions with verbal adjectives, infinitives, and gerunds, on the one hand, and the ever increasing use of prepositions at the expense of the nominal synthetic form of the verb, on the other hand.

5.1.5 Conclusion We have enough data from Indo-European languages to trace the comprehensive development from the former mobile particle to the modern RB preposition, which precedes its complement.

140


The creation of the Indo-European preposition has known three stages: as a first step the originally independent particle developed into an adverb and adposition. Originally, the adposition formed an LB phrase with its complement. Only later did it become a preposed element. Whereas this new syntax was still somewhat uncertain in Umbrian, it was well established in Latin. The Latin adposition, governing only one case, directly preceded its complement. Yet we still find instances of postposing, either in literary contexts (anastrophy) or in specific syntactic contexts (e.g., pronominal complements). Accordingly, postposing was the marked order in Latin. Once the new RB structure came into being, the tendency toward RB prepositional phrases was once more reinforced in Vulgar Latin and Late Latin, when, in a comprehensive change, oblique cases were replaced with prepositions.

5.2 The Comparative Construction In the development of the prepositional phrase we notice the progressive replacement of the case ending with the RB preposition. Similarly, in the history of the comparative construction the term of reference, or standard, also underwent a change whereby the former case ending—the ablative— came to be replaced by a structure introduced by a particle. This is but one aspect of the twofold evolution of the comparative phrase that started in the early period of Indo-European, but not before these languages split up, as Professor Lehmann kindly reminded me, because then, of course, the various dialects would not vary in expression. The development of comparison is characterized indeed by two changes, which, though distinct, are clearly related and which affect (1) the expression of degree and the place of the corresponding morphological marker and (2) the expression of reference and the place of the referent. These two changes resulted in the replacement of the archaic LB structure of the type [[lumine] darior] 'lighter than light' by the modern [plus clair [que la lumiere]]. In the first part of this section I analyze the replacement of the comparative marker -ior, which Latin had inherited from PIE, by a preposed adverbial element, and the gradual rise of a unique periphrastic expression (section 5.2.1). I then examine the change that the term of reference underwent: the initial nominal structure, the ablative of comparison, eventually was replaced by the structure featuring a particle (section 5.2.2). Since this structure can be used in complex contexts, this analysis is related to the change that will be examined in section 5.3, the rise of subordination.

5.2.1 The Morphology of the Comparative In this analysis of the expression of degree and its evolution, I first discuss the Latin system, in which the first weaknesses in the synthetic and LB morphology already are apparent (section 5.2.1.1). These cracks were the


141

first occurrences from which the analytic form increasingly spread. It was only in French proper that the unique form was imposed (see section 5.2.1.2).

5.2.1.1 The Latin Comparative The Latin comparative was characterized by left branching, which was original. The suffix -ior (-ius for neuter) was the morphological marker of the comparative and was preceded by the adjectival root; sec, for example \\lon0] ior] 'longer', {[grand] ior] 'bigger'. The suffix goes back to IndoEuropean *-yes, an intensifier, expressing "d'une maniere intense la qualite signifiee par les adjectifs" (Mcillet and Vendryes 1924:379; sec also Benveniste 1948:121—25). Accordingly it is the referential value and not the separative value that is realized in the term of reference. Luce clarior, therefore did not mean 'light parting from light', but 'light by referring to light'. The term of comparison was therefore a reference (Benvcniste 1948:130), which in Latin was expressed with the ablative. Indo-European had yet another comparative suffix, *-tero, which is a secondary suffix and which as a comparative marker is attested in only a limited number of Indo-European languages, where it has a "differential value" (Benveniste 1948:119). Since the comparative in *-tero is always opposed to a different form featuring the positive degree (Benveniste 1948:117), it easily became the marker of comparison in certain languages, such as Greek. In Latin this element, which is probably related to the archaic grammatical distinction of dual, only occurred in a number of antinomies, such as laevus-dexter 'left'—'right', unus—alter 'one'—'other', and so on. The original value, moreover, was lost when these pairs were replaced by oppositions in which both elements were marked by the same suffix: sinister-dexter or alter-alter (Benvcniste 1948). Consistent with the morphology of the comparative, the formation of superlatives in Latin was also characterized by left branching. The marker of the superlative, -mus, which was realized in Latin under the form of -tumus, -issumus, issimus, goes back to *-simo. The origin of this element is not clear, but it is at any rate Italo-Celtic according to Meillet and Vendryes (1924:361). Under its weakened form -is, the intensive *yes was added to mus, which resulted in -issimus, -issumus (sec Kuhner and Holzweissig 1978:551). I omit the phonological processes that underlie the various actual realizations but do not affect the principle of branching itself. More important for the current analysis are the instances that display a different organization and that mark the beginning of the new structure, which is not only analytic but RB as well. 5.2.1.1.1 Cracks in the LB System of Latin. The morphology of the Latin comparative showed several weaknesses. First, it was defective because neither the comparative nor the superlative of inequality expressing the

142


lesser and least degree of comparison displayed proper morphological markers, nor did the equativc forms. Consequently, these degree forms featured the adverbs minus, minime, and tarn (for the effect of the comparison of equality on the development of the referent, see section 5.3.2). Second, Latin had a number of adjectival roots featuring a final vowel, the comparative forms of which would create phoiiologically complicated sequences: the adjectives in -ius, -eus, and -uus. Consequently, these comparatives also were expressed by analytic forms. Instead of a synthetic form featuring three successive vowels, a periphrastic form was created in which the adjective in the positive degree was preceded by the adverb, e.g., magis (Ernout 1953:78). Once this structure was created, the superlative with maxime could develop, for instance, dubius 'uncertain', magis dubius 'more uncertain', maxime dubius 'the most uncertain', or arduus 'high', magis arduus 'higher', maxime arduus 'the highest'. Yet the analytic form is a relatively recent formation because archaic authors used synthetic forms: "[Vjetustissimi tamen comparativis etiam huiuscemodi sunt [est quando] usi" (Priscian 3.8). Hence, perpetuius 'more universal', innoxiiorem 'more blameless' (examples from Priscian 3.8 [Keil 1961:11 87-88]). 5.2.1.1.2 The Spread of the Analytic Form. Once the analytic form was accepted for adjectives featuring a vocalic root, its use spread to adjectives that had full-fledged synthetic forms. The increasing use of adverbs went along with the loss of the precise meaning of the inflected form. This change is especially well illustrated in examples in which an adverb is added to the comparative form (e.g., magis melior [Vitr.], plus magis [Me. 6.51]). Moreover, one finds numerous instances of semantic and formal encroachment. These can clearly be observed in translations, which offer assured examples (as in the Vulgata, carissimus [Me. 9.8], which translates the Greek positive b'beloved' [Plater and White 1926:67]). Although, according to Priscian, originally magis was the only comparative adverb (sec also the inventory presented by Ernout and Thomas 1951:78—79), a second element came to be used also, plus; its use only became established in the imperial period, especially from Tertullian onward (Vaananen 1963:126; Wolfflin 1884). Yet even in Late Latin the periphrastic expression was not common in all texts. This somewhat slow spreading, however, did not prevent Juvenal from explaining improbior 'more wicked' by plus improbus (Wolfflin 1884:100). Moreover, in the Vulgar Latin letters of Claudius Terentianus "[the] genuine synthetic superlatives (adjectival or adverbial) . . . arc confined to epistolary formulae" (Adams 1977b:56), and in the Chronicle of Valesianus, the superlatives appear especially in formulae of respect, such as nobilissimus (39) (Adams 1976:70-71). In addition to plus and magis, the use of which remained more or less constant, there was a complete series of adverbs that came up and disappeared again without leaving many traces (Wolfflin 1884). Numerous


143

adverbs "were of such plebeian character that they were never admitted into the literary language" (Arts 1927:43); others were characteristic of everyday speech. A number of combinations, however, survived in modern Romance languages and are characterized by right branching. The Spanish tamano, for example, goes back to tarn magnus, which appeared for the first time in Vulgar Latin texts of the first quarter of the second century, where it has the value of a reinforced superlative (Adams 1977b:56; see Lettr. Cl. Ter. 471.27 or Petronius 42.3). The emphatic value of preposed bene in Italian and of bien in French also goes back to a Latin use we already observe in Plautus's writings, as in bene morigerus (Copt., 966) 'much obedient'. Later, even fixed expressions were created: "[C]ombinations comprising bene + adjective or adverb arc common in colloquial texts" (Adams 1977b:58).

5.2.1.2. Old and Middle French; The Rise of One Form Whereas in Late Latin the comparative was formed with various adverbs, the definitive choice was established in Old French. Generally the comparative of inequality (that is, the comparison expressing inferiority) was expressed with mains (< from its Latin equivalent minus 'less'), and the comparative of superiority was formed by plus. With few exceptions the corresponding superlative was not formally distinct from the comparative. Only later, in middle French, the distinction between the comparative and the superlative was introduced again, when the preposed, hence RB, article was used to mark the superlative: [plus [jjrand]] 'bigger', [le [plus grand]] 'biggest'. Moreover, the RB nature of the modern degree form was reinforced in Old French by a series of preceding adverbs: plus 'more', pis 'worse', mieux 'better', and so on (see Martin and Wilmet 1980:137); si and aussi marked inequality; and tres marked the absolute superlative and was only rarely used in the earliest times (see Moignet 1973:270). In addition to these analytic RB structures Old French also had inherited a number of synthetic comparatives and superlatives, among them mieudre (< buens 'good'), graindre (< granz 'big'), bailor (< haus 'high'), forgor (

The Emergence and Development of SVO Patterning in Latin and French: Diachronic and Psycholinguistic Perspectives

Mouse Development: Patterning, Morphogenesis, and Organogenesis

Mediaeval Latin and French bestiaries

Systemics of Emergence: Research and Development

Systemics of Emergence: Research and Development

Systemics of Emergence: Research and Development

Cognitive Models of Speech Processing: Psycholinguistic and Computational Perspectives

Dependency and Development in Latin America

Nominalization in Asian Languages: Diachronic and typological perspectives

Emergence and Adaptation: Studies in Speech Communication and Language Development

Cognitive Models of Speech Processing: Psycholinguistic and Computational Perspectives

Literacy and Development: Ethnographic Perspectives

Child Development in Context: Voices and Perspectives

Nanolithography and Patterning Techniques in Microelectronics

Tourism and Responsibility: Perspectives from Latin America and the Caribbean

Quantity and Prosodic Asymmetries in Alemannic: Synchronic and Diachronic Perspectives (Phonology and Phonetics, 5)

Adolescent Emotional Development and the Emergence of Depressive Disorders

Remittances and Development: Lessons from Latin America (Latin American Development Forum) (Latin American Development Forum)

Current Trends in Diachronic Semantics and Pragmatics

Neurolinguistic and Psycholinguistic Perspectives on SLA (Second Language Acquisition)

Langue and Parole in Synchronic and Diachronic Perspective

The Role Of History In Latin American Philosophy: Contemporary Perspectives (Suny Series in Latin American and Iberian Thought and Culture)

Plant Patterning: Structural and Molecular Genetic Aspects

Lexical semantics and diachronic morphology: The development of -hood, -dom and -ship in the history of English (Linguistische Arbeiten)

Communism and the Emergence of Democracy

Strong and Weak Emergence

Environment And Development In The Caribbean: Geographical Perspectives

Environment And Development In The Caribbean Geographical Perspectives

Cosmopolitics and the Emergence of a Future

Crime and Violence as Development Issues in Latin America and the Caribbean (World Bank Latin American and Caribbean Studies)

The Emergence and Development of SVO Patterning in Latin and French: Diachronic and Psycholinguistic Perspectives

Mouse Development: Patterning, Morphogenesis, and Organogenesis

Mediaeval Latin and French bestiaries

Systemics of Emergence: Research and Development

Systemics of Emergence: Research and Development

Systemics of Emergence: Research and Development

Cognitive Models of Speech Processing: Psycholinguistic and Computational Perspectives

Dependency and Development in Latin America

Nominalization in Asian Languages: Diachronic and typological perspectives

Emergence and Adaptation: Studies in Speech Communication and Language Development

Cognitive Models of Speech Processing: Psycholinguistic and Computational Perspectives

Literacy and Development: Ethnographic Perspectives

Child Development in Context: Voices and Perspectives

Nanolithography and Patterning Techniques in Microelectronics

Tourism and Responsibility: Perspectives from Latin America and the Caribbean

Quantity and Prosodic Asymmetries in Alemannic: Synchronic and Diachronic Perspectives (Phonology and Phonetics, 5)

Adolescent Emotional Development and the Emergence of Depressive Disorders

Remittances and Development: Lessons from Latin America (Latin American Development Forum) (Latin American Development Forum)

Current Trends in Diachronic Semantics and Pragmatics

Neurolinguistic and Psycholinguistic Perspectives on SLA (Second Language Acquisition)

Langue and Parole in Synchronic and Diachronic Perspective

The Role Of History In Latin American Philosophy: Contemporary Perspectives (Suny Series in Latin American and Iberian Thought and Culture)

Plant Patterning: Structural and Molecular Genetic Aspects

Lexical semantics and diachronic morphology: The development of -hood, -dom and -ship in the history of English (Linguistische Arbeiten)

Communism and the Emergence of Democracy

Strong and Weak Emergence

Environment And Development In The Caribbean: Geographical Perspectives

Environment And Development In The Caribbean Geographical Perspectives

Cosmopolitics and the Emergence of a Future

Crime and Violence as Development Issues in Latin America and the Caribbean (World Bank Latin American and Caribbean Studies)

Recommend Documents