Who Climbs the Grammar-Tree: [Leaves for David Reibel] 348430281x, 348430281X, 9783484302815

Table of contents :
Preface
Selected Roots: History of Linguistic Ideas
Studies of English Language and Literature versus Romance Philology in Germany in the Nineteenth Century
E. A. Sonnenschein’s New English Grammar
Reconstructing Seasons Past: Linguistic Change
The Verb speak in Shakespeare’s Plays
Remarks on the Structure of the Verbal Complex in Early 17th Century German
Language Change in Modern German: Feminism and Systematic Gap Production
Pronunciation Rules and Improper Sound Change
Aspects of the Development of the Imperative in Early Modern English
Informal Sources in the History of English and German
On the Historical Development of Greek/Grecian
Aspects of Growth: Language Acquisition and Language Teaching
Product and Process in the Phonology of Foreign Accent: Towards a Linguistic Model of Second Language Sound Acquisition
What Can We Learn from the Case of Pitcairnese?
Variation and Language Acquisition: Psammetichus Revisited
Language Teaching in Utopia: A Retrospective View
The Grammar of ‘Missing Arguments’ in Early Child English
Branches and Twigs: Within and Across Levels of the Grammar
Phonology and Spelling
The Phonologies of the English Auxiliary System
Spelling in Society: Forms and Variants, Uses and Users
Semantics and Pragmatics
What about How about? Or: The Non-Synonymy of How about and What about
Linguistic and Stylistic Considerations Affecting Restrictive Relative Clauses in Spoken and Written English
On the Tension between Properties and Propositional Functions
Sein und Raum: Bemerkungen zur lokalistischen Interpretation des Verbs sein in Heideggers Einführung in die Metaphysik
The English Conditional – Tense, Aspect or Mood?
Syntax
Non-Configurational Properties in a Configurational Language: The Case of English
On Topicalization and Inversion
From Discourse to Syntax: The Case of Concessive Conditionals
English Nominal Gerund Phrases as Noun Phrases with Verb Phrase Heads
The Category of Invariant alles in Wh-Clauses: On Syntactic Quantifiers vs. Quantifying Particles in German
Revisiting -ing -ing
New Seasons
Computational Linguistics: “A Personal View”
List of Contributors

Citation preview

Linguistische Arbeiten

281

Herausgegeben von Hans Altmann, Peter Blumenthal, Herbert E. Brekle, Gerhard Heibig, Hans Jürgen Heringer, Heinz Vater und Richard Wiese

Who Climbs the Grammar-Tree Herausgegeben von Rosemarie Tracy

Max Niemeyer Verlag Tübingen 1992

Die Baumabbildungen wurden mit freundlicher Genehmigung des Verlags Paul Parey, Hamburg und Berlin, dem Band »Pareys Buch der Bäume« (von Alan Mitchell und John Wilkinson, 1987) entnommen.

Die Deutsche Bibliothek - CIP-Einheitsaufnahme Who climbs the grammar tree / hrsg. von Rosemarie Tracy. - Tübingen: Niemeyer, 1992 (Linguistische Arbeiten; 281) ME: Tracy, Rosemarie [Hrsg.] ; GT ISBN 3-484-30281 -x

ISSN 0344-6727

© Max Niemeyer Verlag GmbH & Co. KG, Tübingen 1992 Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlages unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Printed in Germany. Druck: Weihen-Druck GmbH, Darmstadt Einband: Hugo Nadele, Nehren

Table of Contents

Preface

xi

Selected Roots: History of Linguistic Ideas Hans-Helmut Christmann Studies of English Language and Literature versus Romance Philology in Germany in the Nineteenth Century John Walmsley E. A. Sonnenschein's New English Grammar

3

15

Reconstructing Seasons Past: Linguistic Change Anja de Brabandt The Verb speak in Shakespeare's Plays

33

Gerd Fritz Remarks on the Structure of the Verbal Complex in Early 17th Century German

53

Oliver Grannis Language Change in Modern German: Feminism and Systematic Gap Production

67

Fred W. Householder Pronunciation Rules and Improper Sound Change

77

Andrea Nemeth-Newhauser and Richard Newhauser Aspects of the Development of the Imperative in Early Modern English

93

Charles V.J. Russ Informal Sources in the History of English and German

107

VI

Sanford A. Schane On the Historical Development of Greek/Grecian

117

Aspects of Growth: Language Acquisition and Language Teaching Allan R. James Product and Process in the Phonology of Foreign Accent: Towards a Linguistic Model of Second Language Sound Acquisition

125

Robert B. Le Page What Can We Learn from the Case of Pitcairnese?

143

Anne E. Mills Variation and Language Acquisition: Psammetichus Revisited

157

Leonard Newmark Language Teaching in Utopia: A Retrospective View

171

Andrew Radford The Grammar of 'Missing Arguments' in Early Child English

179

Branches and Twigs: Within and Across Levels of the Grammar

Phonology and Spelling Adrian Simpson The Phonologies of the English Auxiliary System

209

Michael W. Stubbs Spelling in Society: Forms and Variants, Uses and Users

221

Vll

Semantics and Pragmatics Klaus-Dieter Gottschalk What about How about! Or: The Non-Synonymy of How about and What about

237

Elsa Lattey and Juliane Moeck Linguistic and Stylistic Considerations Affecting Restrictive Relative Clauses in Spoken and Written English

257

Uwe Mönnich On the Tension between Properties and Propositional Functions

273

Heinrich Weber Sein und Raum: Bemerkungen zur lokalistischen Interpretation des Verbs sein in Heideggers Einßhrung in die Metaphysik

297

Paul N. Werth The English Conditional - Tense, Aspect or Mood?

307

Syntax Hartmut Czepluch Non-Configurational Properties in a Configurational Language: The Case of English

339

H. Bernhard Drubig On Topicalization and Inversion

375

Ekkehard König From Discourse to Syntax: The Case of Concessive Conditionals

423

Geoffrey K. Pullum English Nominal Gerund Phrases as Noun Phrases with Verb Phrase Heads

435

vni

Marga Reis The Category of Invariant alles in Wh-Clauses: On Syntactic Quantifiers vs. Quantifying Particles in German

465

Paul Westney Revisiting -ing -ing

493

New Seasons Franz Guenthner Computational Linguistics: "A Personal View"

507

List of Contributors

519

Who climbs the Grammar-Tree; distinctly knows Where Noun, and Verb, and Participle grows ... John Dryden, The Sixth Satyr of Juvenal (1693. 583-4)

!

~·'«"»-,· »OR··.- tow·--»

Leaves for David Reibel

XI

Preface A tree offered itself as a natural metaphor for this book for several reasons. Most obvious is the long linguistic tradition of representing structures by tree diagrams. Since the book draws on major branches of present-day linguistics, providing a picture of the diversity of theoretical perspectives and research methods, what also comes to mind is the 'tree of knowledge1, that is, the growth of knowledge in a more general epistemological sense.1 Among further associations, there is the ecological aspect: trees, like any living thing (linguists, for example), only grow if the conditions are right. In the case of linguistics, the intellectual and personal climate set the stage for an individual researcher's enterprise. These essays were brought together for a particular occasion:

David Reibel's 60th

birthday. They can be read like rings in a tree, revealing traces of David Reibel's own life as a teacher of linguistics, English as a second language, Russian and Albanian, in the United States from 1959-1962 and 1964-1967, in England from 1963-1964 and from 1967-1975, and, thereafter, in Tübingen, Germany. This book is dedicated to him in respect and appreciation, with particular admiration for his willingness to discuss new ideas and his challenging questions, such as "Why should I believe that ... ?" During all these years, he has provided the inspiration and support for others to grow. The contributors chose topics that can be explicitly related to David Reibel's own interests in the history of linguistics, in the history of English, in specific descriptive puzzles, in language acquisition and language teaching. One particular interest which is addressed by many authors concerns the descriptive and explanatory problems posed by structural or functional variation and the pressures of standardization. The tree metaphor reappears throughout the book, with essays grouped into thematic units on roots, past and future seasons, aspects of growth and individual branches.

Cross-

classification was therefore unavoidable, most noticeably, for instance, between the papers on linguistic change or language acquisition and those in the following unit on individual levels of the grammar. Within each unit, the essays appear according to an alphabetic ordering of the authors' names. The first unit, Selected Roots, comprises two papers. The first one, by Hans-Helmut Christmann, goes beyond linguistics by sketching the general academic climate in the last century when various fields, including English language studies in Germany, emerged.

Cf. K.R. Popper, Objective Knowledge: An Evolutionary Approach (Oxford, 1979) 256ff.

Xll

In the second paper, John Walmsley discusses the English grammar by Sonnenschein as an important link on the way to modem constituent grammars and to unified linguistic terminology. In Reconstructing Seasons Past, eight authors present analyses and descriptions of specific linguistic phenomena in the history of English and German. Anja de Brabandt discusses the use of the verb speak by Shakespeare, thereby revealing interesting structural features of Early Modern English. The following paper on the verbal complex by Gerd Fritz deals with syntactic variation and processes of standardization in early 17th century German. Oliver Grannis turns to a domain of investigation which is, under the influence of the feminist movement, gaining more and more momentum: lexical gaps with respect to gender terms and strategies for coping with these gaps. Concern with variants also plays an important role in Fred W. Householder's discussion of oral vs. non-oral transmission of words and the changes resulting from both types of transmission. In their discussion of the tension between the imperative, the indicative and the subjunctive, Andrea Nemeth-Newhauser and Richard Newhauser trace processes of reorganization within the verbal system in Early Modern English. Charles V.J. Russ, in his treatment of informal sources of data (German and English letters written in the 18th century), draws attention to discrepancies between incipient standardization and private documents. Sanford A. Schane investigates morphophonemic alternations in the history of English, exemplified in the development of Greek/Grecian. The five contributions in the third unit, Aspects of Growth, raise a number of issues concerning first and second language acquisition and teaching. In his paper on the product and process of second language sound acquisition, Allan R. James proposes a model incorporating a distinction between development, learning, and acquisition. Robert B. Le Page's article on Pitcairnese discusses factors involved in the development of a creole language, with individual and group identity playing an important role. He also stresses the need for a critical awareness of stereotypes in defining concepts such as 'language'. Anne E. Mills explicitly addresses the problem of how researchers deal with variation. She criticizes the tendency to ignore differences for the sake of the preservation of theories, and she argues specifically for an integration of the results from so-called pathological and so-called normal studies. Andrew Radford's paper throws new light on children's early word combinations which, because of apparently optional constituents, have troubled researchers for a long time. Rather than endorsing a 'pro-drop1 perspective, he proposes a lexical saturation analysis to account for these structures. Shifting attention from the learner to the wider teaching context, Richard Newmark presents us with his (and David Reibel's) experience of developing a language teaching program for university students. His paper illustrates well the ecological aspects mentioned above, especially as far as prejudice and increasing administrative

Xlll

resistance are concerned, which made the implementation of innovative ideas more and more difficult. The fourth unit, Branches and Twigs, contains thirteen papers whose main theme can be located within a particular level of the grammar or which deal with the interaction between levels. In Phonology and Spelling, Adrian Simpson proposes a polysystemic approach to descriptive problems involving the weak and strong forms of English auxiliary verbs. Michael W. Stubbs focusses on the tension between system and use, between prescriptive and descriptive factors in spelling conventions. Semantics and Pragmatics begins with an article by Klaus-Dieter Gottschalk. He presents the results of an empirical investigation on the pragmatic differences between What about and How about and their complement structures. Stylistic variation is the central topic addressed in the article by Elsa Lattey and Juliane Moeck, who, on the basis of different types of texts, examine parameters affecting the use of restrictive relative clauses in English. Uwe Mönnich presents a critical appraisal of various semantic models, especially the 'possible worlds' semantics and the 'prepositional function thesis'. The chapter on semantics also contains the only essay which was left untranslated: the text by Heinrich Weber, who deals with very subtle localistic meanings of the verb sein in Heidegger's Einführung in die Metaphysik, drawing attention to many insights of Heidegger's which are of considerable linguistic importance. Paul N. Werth applies a model of situational grammar to the description of the English conditional, discussing crucial distinctions with respect to tense, aspect and mood. The section on Syntax begins with an article by Hartmut Czepluch, who takes issue with the belief that English is a configurational and German a nonconfigurational language. Within the framework of Government and Binding Theory, he shows that major contrasts between German and English follow from the interaction of three simple parameters. Also within Government and Binding theory, H. Bernhard Drubig discusses principles underlying and unifying various preposing constructions, such as topicalization, inversion, whconstructions, and focus-preposing. Ekkehard König's paper looks at the English conditional from both a synchronic and a diachronic point of view; it is seen as resulting from the gradual syntactization of discourse structures.

Within the theoretical framework of Generalized

Phrase Structure Grammar, Geoffrey K. Pullum deals with the problem of defining the 'head' of English gerunds, and analyzes these constructions as noun phrases with verb phrase heads. Marga Reis discusses important syntactic and semantic differences between

the

morphologically invariant alles in German wh-clauses and the morphologically variable floating quantifier alle(s), assigning the former to a category of quantifying particles with clitic properties.

Paul Westney's paper investigates multiple adjacent -ing structures in

English complement-taking verbs like continue, keep, start, etc., and deals with variation in grammaticality judgments.

XIV

The final unit, New Seasons, consists of a paper by Franz Guenthner. He outlines the goals of computational linguistics, and, as an illustration, discusses the role of the programming language Prolog as an implementation language. The dissemination of the book in its, present form - an evolution from the original singlecopy Festschrift handbound by Ulrike Grannis - was possible only with the help of many people. I would first like to thank all contributors for their spontaneous and positive response to my incentive, and the editors of the series for helpful criticism of the original manuscripts. I also want to express my gratitude to Carmen Luna of Niemeyer Publishers for excellent cooperation. Special thanks go to my colleague Elsa Lattey for editorial advice throughout the whole enterprise. The book is the result of the common effort of many helpful heads and hands. To name only those involved in the preparation of the final manuscript: Cleo Becker, Diana Gierling, Petra Gretsch, Karin Klett, Dagmar Lalla, Patrick Schindler, Claudia Schlenker, Petra Schulz. For checking the English, thanks go to Catherine Brady and Andrea Kealey. I am indebted to Uwe T. Ruckgaber of CSR-Computer, Tubingen, who solved numerous and time-consuming hardware and software problems, and who took care of the final laser printout. Finally, I would like to explicitly thank Svenja Kuhfuß for her dedication to this project. She provided competent and, moreover, cheerful and never-failing assistance at every stage, and coped with a formidable formatting job in a highly professional manner. The leading quote for this book was taken, tongue in cheek, from Dryden's 'translation' of Juvenal. In Dryden's text2, Who climbs the Grammar-Tree is a relative clause following a head noun in a preceding line: I hate a Wife, to whom I go to School. Who climbs the Grammar-Tree; distinctly knows ... By isolating one line - following an established though problematical linguistic practice - I have created a bias for a certain kind of interpretation, a question or (considering the sequel: ...; distinctly knows / Where Noun, and Verb, and Participle grows) a free relative clause yielding a conditional reading. One might argue that a conditional interpretation would be odd for pragmatic reasons: after all, who climbs the grammar-tree is more likely to come tumbling down than to know where nouns and verbs and participles grow. Linguistics, like tree-climbing, has its hazards, but it also has its rewards. I hope that the book will convey

2

H.T. Swedenberg, Jr., Ed., The Works of John Dryden. Vol.IV: Poems 1693-1696 (Los Angeles, 1974) 183.

XV

some of the excitement which can be found in linguistic puzzles and that it will reflect the spirit in which it grew into a collection of leaves for David Reibel.

Tübingen, May 1992 Rosemarie Tracy

Selected Roots: History of Linguistic Ideas

Hans Helmut Christmann

Studies of English Language and Literature versus Romance Philology hi Germany in the Nineteenth Century1

1.

Introduction

The well-known German Anglist Max Deutschbein once remarked, "Now and then I thank the Lord for not having made me a Romance philologist", because, he said, one single scholar was unable to master all the different languages and literatures comprised in this discipline. Now, as a Romance philologist one cannot but agree with Deutschbein. Furthermore one might feel envious of a neighbouring discipline, namely German Anglistik (studies of English language and literature), which once enjoyed a very high reputation in the world, even in Great Britain. In fact, its reputation was such that a young German who had come to England about the turn of the century to study Old English at a famous English university was told, "The best man for that is Sievers in Leipzig!", and about the same time Henry Cecil Wyld said in his history of the English language that the prior condition for the study of English was to learn German, because the essential textbooks were written in German ... Add to this that German is much closer to English than to French - not to mention the other Romance languages - and you might be inclined to believe that Romance philology in Germany has always been outshone by English studies. Generally, however, it has been just the other way round. Let us take a brief look at modern times: who then were the German philologists of international repute? - Karl Vossler, Ernst Robert Curtius, Leo Spitzer, Erich Auerbach, Hugo Friedrich: Romance philologists all of them, and some of whom- such as Curtius, Spitzer and Auerbach - even also worked on English literature. Looking back at the history of the disciplines in the nineteenth century, one cannot help noticing that Romance philology has always been privileged with respect to English studies.

*· German philologists have often - and not quite wrongly - been accused of turning philology into a 'science of footnotes'. The present occasion being a festive one, I thought it appropriate to do totally without footnotes; but anyone who should nevertheless wish to read some of them is referred to Christmann, Romanistik und Anglistik an der deutschen Universität im 19. Jahrhundert (Akademie der Wissenschaften und der Literatur, Abhandlungen der Geistes- und sozialwissenscliaftliclten Klasse, Jg. 1985, Nr. 1), Mainz 1985.

2.

The wider perspective of the early nineteenth century

The very designations may well serve as evidence.

I do not know exactly when the

designation Anglist first came into existence; but I do know that it is of comparatively recent coinage: I should imagine that it somehow belongs to the period when the review Anglia was founded, that is in 1878.

By contrast the term Romanist appears at least as early as 1840

(while Anglisten had for a long time to be content with makeshift designations like Englischgelehrte, 'scholars of English'). Or let us take the foundation of the first chairs. The first professorship of Romance philology in Germany was established as early as 1822, whereas the first professorship of English philology seems to have been established only in 1866. I insist on the word 'seems', because characteristically we have no precise information. Obviously the holder of the chair was not very keen on being in charge of English Only', but wanted to cover Romance philology as well. The same reserve with regard to their own discipline is shown by the two men who were offered the first full professorships of English officially established at the beginning of the seventies. Let us now go back to the time of the so-called maitres or lectores (people who taught the languages), from the eighteenth up to about the middle of the nineteenth century. In those days, French, English, Italian and Spanish were taught at most universities, often by one and the same man, with French nearly always predominating. During the nineteenth century some of these lectores managed to obtain professorships, a few of them even full professorships, much to the annoyance of their faculties, because most of them entertained good relations with the governments, either as representatives of 'useful' disciplines or for political reasons. The latter applies, for instance, to the full professor (Ordinarius) Adrian of Gießen University whom Heinrich Heine criticized as a nasty censor and government spy. Adrian was one of the few representatives of his guild who laid more stress on English studies, just like his colleague Winterling in Erlangen; for the most part though, French predominated here too, as it did with Kühne in Marburg and Emmert and Peschier in Tübingen. Simultaneously, that is in the first decade of the nineteenth century, there were a number of 'genuine' professors who, coming from other disciplines such as philosophy, history or classical literature, now turned to modern literature and delivered lectures with an aestheticrhetorical approach, covering many peoples and countries. The most famous example being, of course, August Wilhelm Schlegel in Bonn - an example which demonstrates that even with this category of scholar it is not England which is at the forefront, but France. Among other names to be mentioned in this context are Valentin Schmidt in Berlin, reputed to be the most creditable predecessor of Romance philology in Germany, or Karl Ludwig Kannegießer, a Breslau Privatdozent whose renown as a translator of Dante still holds today.

Again such professors who were more interested in English studies were less typical: Victor Aime Huber in Marburg and Berlin, also known for his political activities, and - a case of exceptional rarity - Karl Franz Christian Wagner, full professor of Greek and Latin literature and rhetoric in Marburg, of whom we are told that Greek was his favourite language and that in his opinion English alone was worthy of occupying a place equal to it: English, which he had learned as a private tutor in London and which he wanted to be introduced into German schools instead of French. It is during the first three decades of the nineteenth century that we find early proposals and endeavours for the foundation of regular chairs of modern languages and literatures. For that period we shall have to take German into consideration along with Romance philology and English. There had been no academic teaching of German so far, and so naturally it became a priority among those new disciplines. An official statement from the Faculty of Humanities of Berlin University in 1819 (in Risop 1988:62) reveals the essential principles behind these endeavours (translated into English):

"The Faculty of Humanities is firmly convinced that modern languages and

literatures can only be included in the academic curriculum if a teacher can be found who would treat them on a larger scale and in a wider context, either from a historical-philological or from an aesthetic-literary point of view. Only a man of such quality [...] would be in a position to give full scope to these kinds of studies, whereas knowledge and instruction of one single language and its literature [...] would in our opinion always be too limited [...]." If the academic treatment of any single language or literature was thus rejected in this statement typical of its time, it meant that there could not be any question of creating chairs of English studies for a long while yet. Of course people busied themselves with English studies, but this kind of activity had no claim to scientific treatment unless pursued "on a larger scale and in a wider context", either from a "historical-philological" or from an "aesthetic-literary" point of view. As to the latter approach, we have already mentioned some of its most prominent representatives, especially August Wilhelm Schlegel.. However, the "historical-philological" method which the Berlin faculty thought worthy of mention in the first place, became much more important: in the long run, it was thought to be the only possible way to guarantee a scientific approach, so that it began to dominate the scene for a long time. It was this approach, moreover, that supplied the specific points of argument against the constitution of Anglistik as a separate discipline.

3.

The comparative, the historical, and the philological components

There are three components which can be distinguished in that method. The first is the comparative component, such as had been introduced by Bopp and Jacob Grimm. In this

context English could not possibly form a discipline of its own, but it represented part of the Germanic field; thus it happened that early English, that is, Anglo-Saxon, was treated in Grimm's Deutsche Grammatik (1819 ff.), 'German' appearing - as it did for decades to come as a synonym of 'Germanic1. The comparative component is inseparably linked with the historical one, because in order to discover any common origins it is necessary to go back into history as far as possible. This was in tune with the romantic idea that the older a language level the better and hence the more interesting for research; to use one of Jacob Grimm's (1819:XXVIII) phrases, which by the way all but coincides with the words of Rousseau, "The ancient language and literature are purer, less conscious and nearer to their heavenly origin - hence more magnificent; the new ones have become poor and twisted by the hands of man." In our context this preference for the oldest language levels means neither more nor less than the elimination of the most interesting field of research from Anglistik, even if and when they where eventually recognized as a separate discipline: for the oldest language spoken in England was not considered English at all, but was called Anglo-Saxon; and proper English as a much less interesting, indeed a 'corrupt1 mixed language was not allowed to begin until the thirteenth century. Anglo-Saxon, on the other hand, was considered part of Germanistik (i.e. the study of Germanic languages and literature), indeed of 'German' grammar. It was Jacob Grimm who was at the bottom of this development, and he was followed by Adolf Holtzmann with his Altdeutsche Grammatik (1870) and Moritz Heyne with his Kurze Grammatik der altgermanischen Sprachstämme (1862). This is also the current conception expounded in the first scientific language histories and grammars of the English language by Maennel (1846), Fiedler (1850), Mätzner (1860) and Koch (1863); Fiedler even tells us that the changes of modern English, beginning with the Elizabethan era, really have no place in a language history. So much for the comparative and the historical components of the new method. And now for the third one - the philological component. As far as the development of universities is concerned, this component was probably the most important, because it directly involved the foundation of chairs of the new disciplines, that is to say above all of Germanistik. The great model in those days of neo-humanism was classical philology, a discipline held in highest esteem. It was from classical philology that the method of textual criticism, of establishing 'critical' texts, was adopted, because scholars were anxious to transfer romantic exaltation for medieval literature from the level of dilettantism to one of an acknowledged science. The first professor of Germanic philology was George Friedrich Benecke (1805 in Göttingen), who was specialised in textual criticism. The link with classical philology was so close that several of the early Germanisten at the universities covered German and classical philology simultaneously, as did Lachmann and Moriz Haupt in Berlin; as late as 1854 the Berlin

Germanist Müllenhoff declared it to be "not at all desirable for a student at the university to choose German philology as his main subject. Without preparatory instruction in the old languages the occupation [...] with modem languages and literatures is worth nothing at all [...]" (Müllenhoff in Janota 1980:291). Like the other two components, the third one, that is to say the philological component, of the new method counteracted the establishment of an independent discipline Anglistik, because it was most rewarding for the oldest texts, that is for the so-called Anglo-Saxon ones which were not considered to be English at all. The close connection of Anglo-Saxon editions with German philology is clearly demonstrated by the works of the pioneer in this field, Christian Grein (1857 ff.), who followed the example of Jacob Grimm and of whom his master, the Marburg Germanist Franz E. Dietrich, wrote that he belonged to the GrimmLachmann school and that he had succeeded in "wreaking havoc" among the texts even there where Grimm had already solved most of the problems (Dietrich in Grein 1857:third cover page).

Beowulf was considered essentially as a piece of German poetry by its early

investigator, the historian Heinrich Leo, just as twenty years later by the Germanist Karl Simrock. It hardly needs saying that the early comprehensive outline of literary history by the Germanist Ettmüller under the title Handbuch der deutschen Literaturgeschichte (1847) includes Anglo-Saxon literature (Ettmüller speaks of "the German language in the widest sense"). The new method with its three components - comparative, historical and philological thus had favourable effects on German philology at the outset, so much so that this discipline now rapidly developed and a considerable number of chairs were established. For reasons just pointed out, a simultaneous rise of English philology was out of the question. Romance philology, however, found itself in quite a different situation, comparable to that of German philology, since Friedrich Diez with his Grammatik der romanischen Sprachen (1836 ff.) had created the counterpart of Grimm's Deutsche Grammatik, and since scholars began to concentrate on philological investigations of early Romance texts (the troubadours, Dante etc.). It is true that chairs of Romance philology took longer to be established than those of Germanistik, but soon enough scholars appeared on the scene who - officially or unofficially looked after Romance philology along with Germanistik. Germanic and Romance philologies were taught, for instance, by professors Diez in Bonn (from 1823), Adelbert Keller in Tübingen (from 1841 - he had qualified to give lectures on Germanic and Romance literature in 1835), Konrad Hofmann in Munich (from 1853), Karl Bartsch in Rostock (from 1858) - all of them occasionally lecturing on Anglo-Saxon.

In Halle the Dante specialist Ludwig

Gottfried Blanc was even given a professorship as early as 1822 and was appointed full professor Of Romance languages and literature1 in 1833.

8

Professor Adelbert Keller, one of the scholars just mentioned, gave a theoretical summary of the new knowledge in his inauguration lecture 1842 in Tübingen. He places modern philology alongside classical and oriental philologies and points out as its principal achievement "the historical and comparative study of the two main stocks of modern Europe" (Keller 1842: 9). That same year an "application" was made in the Jahrbuch der deutschen Universitäten "to create professorships of modern languages", with the significant detail that "for the time being three of them would be sufficient: that is, those of Germanic, of Romance and of Slavonic stocks" (Heintze 1842:348).

4.

The influence of secondary schools

As all initiatives coming from the universities in those decades, these were claims on exclusively scientific grounds. In the meantime, however, the subject had begun to become an important field of activity for secondary school teachers, too: they demanded that chairs for modern philology should be created in order to provide teachers of modern languages with a scientific training. Here again classical philology served as a model, which meant in this case the university training and accordingly the social superiority of teachers of Latin and Greek. Thus the well-known educationalist Carl Mager "postulated" in 1843 - just as the university had done - "beside classical and oriental philologies a modern philology: Germanic, Romance and Slavonic" and pleaded for historical and comparative grammar. But since Mager's starting point were the school needs, he suggested, unlike the representatives of the university, for modern philology "three professorships - two of them full professorships, for a Germanist and a Romanist, and one associate professor (Extraordinarius}, which might preferably be reserved for English philology" (Mager 1843:19,110). Thus - unlike the situation of German and Romance philologies - it was the schools which first called for professorships of English philology, but with a marked distance from the other two disciplines and by renouncing to full professorships. Three years later the schoolteachers Herrig and Viehoff referred to Mager's concept of modern philology in the preface of the first review in this field, the Archiv ßr das Studium der neueren Sprachen; soon afterwards they repeated in a special article Mager's claims for relevant professorships, which should be - precisely with regard to school - for German, English and French philologies, adding at once, however, that the Trench philologist' ought to be a 'Romance philologist1, and that the 'German and English philologists' ought to be 'Germanisten in the widest sense of the word', which meant that they should be doing 'historical and comparative' work (Herrig & Viehoff 1846; Herrig & Viehoff 1848:232f.). The greate desire of particularly the school-teachers to prove modern languages and literatures equal to the classical ones can also be seen from the fact that the famous secondary

school-teacher Herrig of Berlin gave an anthology of English literature the title The British Classical Authors (1850, re-edited by Herrig himself eighty-five times, then taken over by Max Förster and published for the last time in 1947). Herrig's "Preface" begins, "If the study of modern languages may lay claim to intrinsic value and an honourable position in our educational establishments, equal to that of the classical languages [...]." But it was going to be a long time yet before the demands of the schools for professorships in Romance (French) and English philologies were satisfied. This was due only in part to the reluctance of the governments to provide the financial means; a more pertinent reason was the fact that for the time being there were no candidates who could have met the academic requirements of the faculties. That explains why in 1863 the Romance philologist K.A.F. Mahn had to repeat the call for professorships, pointing out, however, that what was needed first of all were capable young men as 'Docenten', - chairs then would surely not fail to follow.

He said this at a lecture held before the Germanic-Romance section of the

Association of German philologists and school-teachers, a section which had meanwhile (nota bene) been constituted (Mahn 1863:9f.). Indeed quite a number of Habilitationen, 'habitations', in 'modern languages' or similar were accepted in the course of those years, more and more of the young scholars concentrating on Romance and English philologies simultaneously. I need only mention the early case of Nicolaus Delius (Berlin 1842 and Bonn 1846) and that of his pupil Bernhard ten Brink (Bonn 1866). The still powerful influence of classical philology can be seen in Delius's works on Shakespeare which are esentially works of textual criticism, or in the theses which his pupil Jakob Schipper defended when awarded a doctorate in Bonn in 1867 (the doctoral dissertation on Marlowe was written in Latin): four out of nine theses were on Old English, two on Old French, one on Old High German and one on Old Italian; six of them were textual emendations. Only the ninth thesis dealt with an entirely different subject: "Studium linguae britannicae in gymnasiis optandum est."2 But even in the mid-sixties Albert Stimming as a young man experienced that studying French and English in those days was still "something entirely new, something almost unheard of, just as did young Alois Brandl ten years later with English. In both cases it was decided that the young men should first take up studies of classical and German philologies (Stimming 1967:2; Brandl 1936:79f.). Anyway, there were now more and more Privatdozenten available for the new disciplines, so that from the sixties onward a considerable number of full professorships could be established, precisely those that fulfilled the requests for professorships of French and English 2

On Delius and ten Brink, see Schinner (1970:240f.), Heanicke (1979:passim), and Haenicke (1981:28ff., 189ff); Schipper (1867:46).

10

made in particular by the schools. These requests were supported once more and in an impressive way by the Greifswald lector for modern languages, Bernhard Schmilz, who was later appointed full professor of Romance and English philology, in his Encyclopädie des philologischen Studiums der neueren Sprachen (1859 ff.), and it is significant that he wanted to place the new 'French-English philology1 on a level with the Old twin philology1, 'the Greek-Latin one1. The idea was to create chairs with the double function of French and English. So now the chairs which had so long been requested were finally being established in the form of 'double1 professorships - more than a dozen of them up to the mid-seventies -, but instead of what the schools had wanted, that is professorships for the training of teachers of French and English, they turned out to be professorships of Romance and English philologies with a purely 'scientific' orientation.

This was in fact only to be expected, since the

Privatdozenten who now occupied these chairs had been trained according to the guide-lines of the science of those days, that is to say, they had worked on ancient literary monuments by applying methods of textual criticism and of comparative and historical linguistics. There is another thing which is remarkable about these professorships with a double function: nearly all of the scholars who occupied them were Romance philologists rather than English philologists. Historians of Romance philology have noticed this in retrospect with satisfaction. Carl Voretzsch for instance says, "Luckily most of the occupants of these chairs were Romance philologists, so that Romance philology predominated in teaching as well as in research work." The reason for this course of events is above all the fact that the scholars who had qualified for both disciplines considered Romance philology to be by far the more interesting and diverse of the two. Having made some initial experiences with those 'double' professorships, these very scholars demanded that English studies should be separated, because they wished to devote themselves entirely to Romance philology.

5.

A new era

It took some time before genuine Anglisten also demanded such a separation; a pioneer in this particular case was the secondary school-teacher Karl Elze who was later appointed professor of English language and literature. For Elze, too, scientific aspects were of primary importance, apart from aspects concerning school: the dominating position of the English language in the world, the necessity to give English language teaching a scientific basis, the analogy with classical philology, where there were separate chairs for Greek and Latin, the inclusion of American literature.3 1

Elze (1864:passim); on Elze, see (Haenicke 1979:passim) and Haenicke (1981:43f.).

11

It certainly is not a coincidence that at the same time the concept of 'Anglo-Saxon' was being ardently discussed, with the result that from now on the term Old English' was generally adopted which had principally been introduced by Sweet in England and by Zupitza and ten Brink in Germany. Characteristically enough it was Schmitz who had most strongly pleaded for the adoption of this term. In this way the new professors of English philology had even by name a venerable literature as an important field of activity. Finally Elze (1864:90ff.) refers to "the august patron in order to inaugurate this new era", that is to Shakespeare, in connection with the creation of the German Shakespeare Society. This, too, is not a practical aspect nor one concerning school, but a scientific aspect. The importance of being thus admitted into higher scientific orders can even be seen in the attitude of the publishers Langenscheidt who had so far been known for their practical orientation: in 1896 they were extremely keen on taking over the year-book of the German Shakespeare Society, because after all they also wanted to have 'something distinguished'. As we know, the urgings of the Romance philologists on the one hand and of Elze on the other were finally successful: one university after the other was provided with a professorship of English philology, so that by the end of the nineteenth century the separation from Romance philology was completed everywhere. There is no need for me to go into details, and, referring to what I said at the outset, I just want to point out that the first representatives were anything but happy about their being confined to Anglistik. As far as the first professor of English philology - Nicolaus Delius - is concerned, there is no documented proof for the date of the establishment of a chair; evidence for a chair can only be found at the date of Delius's retirement. It is certain, however, that Delius wanted to be - and to remain - a Romance philologist as well. Something similar happened to the professorship of the English language at Vienna University, which had been created in 1872: in his negotiations with the government, the nominee to this chair Julius Zupitza demanded that it should be extended to one of North-Germanic languages. Bernhard ten Brink, who was nominated professor of English philology at Strassburg University in 1872, carried on a guerrilla war with his collegue Boehmer, the Romance philologist, about his wish to hold lectures and to examine students in French - a war which began with his taking office in 1873 and which went on for years. The fact is that up to then ten Brink had been used to working in both fields (in Marburg), whereas Boehmer, on account of his previous professorship in Halle, took his exclusive devotion to Romance philology as a matter of course.4 So we may presume that when Karl Elze was nominated professor of English language and literature in 1875 at Halle University, where a chair of Romance philology had existed for

1

On Zupitza, see Haenicke (1974:125ff.); on ten Brink, Storost (1961:91ff.) and Haenicke (1974:128ff., 238ff.).

12 more than fifty years, - that it was Karl Elze himself who was the first Anglist to thank the Lord for having made him an Anglist and not a Romance philologist.

Bibliography Brandl, A. (1936): Zwischen Inn und Themse: Lebensbeobachtungen eines Anglisten. - Berlin. Diez, F. (1936): Grammatik der romanischen Sprachen. - Bonn. Elze, K. (1864): Die englische Sprache und Literatur in Deutschland. - Dresden. Ettmüller, L. (1847): Handbuch der deutschen Literaturgeschichte von den ältesten bis auf die neuesten Zeiten, mit Einschluß der angelsächsischen, altskandinavischen und mittelniederländischen Schriftwerke. - Leipzig. Fiedler, E. & C. Sachs (1850,1861): Wissenschaftliche Grammatik der englischen Sprache.- Leipzig. Grein, C.W.M. (1857): Dichtungen der Angelsachsen stabreimend übersetzt. - Göttingen. Grimm, J. (1819): Deutsche Grammatik. Vol. 1. - Göttingen. Haenicke, G. (1979): Zur Geschichte der Anglistik an deutschsprachigen Universitäten 1850-1925. - Augsburg (= Augsburger I - I-Schriften, vol. 8). Haenicke, G. (1981): Biographisches und bibliographisches Lexikon zur Geschichte der Anglistik 1850-1925. Augsburg (= Augsburger I - & I-Schriften, vol. 13). Heintze, J. (1842): "Über die Notwendigkeit auf den Universitäten Professuren der neuem Sprachen zu begründen." - In: Jahrbuch der deutschen Universitäten l, 339-49. Herrig, L. (ed.) (1850): The British Classical Authors. - Brunswick. Herrig, L. & M. Förster (eds.) (1011947): British Classical Authors. - Braunschweig. Herrig,, L. & H. Viehoff (1846): "Vorwort". - In: Archiv für das Studium der neueren Sprachen l, 1-4. Herrig, L. & H. Viehoff (1848): "Wünsche für das Studium der neueren Sprachen." - In: Archiv für das Studium der neueren Sprachen und Literaturen 4, 225-34. Heyne, M. (1862): Kurze Grammatik der altgermanischen Sprachstämme: Gothisch, Althochdeutsch, Altsächsisch, Angelsächsisch, Altfriesisch, Altnordisch. Vol. 1. - Paderborn. Holtzmann, A. (1870): Altdeutsche Grammatik, umfassend die gothische, altnordische, altsächsische, angelsächsische und althochdeutsch Sprache. Vol. 1. - Leipzig. Keller, H.A. (1842): Inauguralrede über die Aufgabe der modernen Philologie. - Stuttgart. Also in: J. Janota (ed.) (1980): Eine Wissenschaft etabliert sich (Tübingen) 263-77. Koch, C.F. (1863): Historische Grammatik der englischen Sprache. Vol. 1. - Weimar. Leo, H. (1839): Beowulf: dasz älteste deutsche, in angelsächsischer mundart erhaltene heldengedicht nach seinem inhalte, und nach seinen mythologischen beziehungen betrachtet. - Halle.

13 Maennel, F.A. (1846): Genesis oder Geschichte der innern und äußern Entwickelung der englischen Sprache. Leipzig. Mätzner, E. (1860): Englische Grammatik. I. - Berlin. Mager, K. W.E. (1843): Ueber Wesen, Einrichtung und pädagogische Bedeutung des schulmäßigen Studiums der neueren Sprachen und Literaturen und die Mittel ihm aufzuhelfen. - Zürich. Also in: H. Kronen (ed.) (1985): Fremdsprachenunterricht. Vol. 2: Mager, K.W.E.: Gesammelte Werke. - Baltmannsweiler. Mahn, K.A.F. (1863): Über die Entstehung, Bedeutung, Zwecke und Ziele der romanischen Philologie. - Berlin. Müllenhoff, K. (1854): "Die deutsche Philologie, die Schule und die klassische Philologie." In: J. Janota (ed.) (1980): Eine Wissenschaft etabliert sich (Tübingen) 277-303. Risop, A. (1910): Die romanische Philologie an der Berliner Universität 1810-1910. - Erlangen. Also in: J. Trabant (ed.) (1988): Beiträge zur Geschichte der romanischen Philologie in Berlin: "Anhang". - Berlin. Schipper, I. (1867): De Versu Marlovii. - Dissertatio philologica Bonnae. Schinner, W.F. (1970): "Die Anglistik an der Bonner Universität" - In: Bonner Gelehrte: Beiträge zur Geschichte der Wissenschaften in Bonn: Sprachwissenschaften, 239-59. Schmilz, B. (1859; 21876): Encyclopädie des philologischen Studiums der neueren Sprachen. (1860-64): Supplement /-///. - Greifswald. Schmilz, B. (1866-72): Die neuesten Fortschritte der französisch-englischen Philologie. Vols. 1-3. - Greifswald. Simrock, K. (1859): Beowulf: Das älteste deutsche Epos. - Stuttgart/Augsburg. Slimming, H. (ed.) (1967): Geheimer Regierungsrat Dr. Albert Stimming, ordentlicher Professor für romanische Philologie. - Wirten-Wengern. Storost, J. (1961): "Dante und die Entstehung der deutschen Hochschulromanistik I: Präliminarien - Eduard Boehmer." - In: Deutsches Dante-Jahrbuch 39, 80-97. Tagliavini, C. (1973): Einführung in die romanische Philologie. - München. Voretzsch, C. (1904): Die Anfänge der Romanischen Philologie an den deutschen Universitäten und ihre Entwicklung an der Universität Tübingen. - Tübingen. Wyld, H.C. (1914): A Short History of English. - London.

15

John Walmsley E A. Sonnenschein's New English Grammar 1.

Introduction

The particular interest of Edward Adolf Sonnenschein's A New English Grammar, first published in 1916, is that it stands at the meeting-point of a number of important movements. Like all works of its kind, it reflects more general aspects of grammatical thinking at a specific moment in linguistic history.

At the same time, it provides insights into the

development of the author's own linguistic thought. Uniquely, it constitutes one milestone in Sonnenschein's crusade for the simplification and unification of linguistic terminology. And, not least, it can be studied from the point of view of its contribution to the long history of teaching- or pedagogical (as opposed to reference-) grammars. Not all of these aspects can be given adequate treatment here. They need to be mentioned, however, since they exert a mutual influence upon each other. Features in one part of the grammar which would otherwise remain unexplained can be shown to be the consequence of decisions made in other areas of the grammar. This paper will concentrate primarily on establishing the position which the New English Grammar (henceforward NEG) occupied in the linguistic thinking of its time. Despite this choice of orientation, it is the intended use of NEG as a grammar for teaching purposes which needs to be remembered when Sonnenschein's treatment of individual grammatical categories is being considered.

NEG belongs to a centuries-old tradition of

teaching-grammars from the Renaissance onwards - a tradition which includes Bullokar, Ratke (Ratichius) and others among the first proponents of vernacular grammar.

This

tradition was predominantly morphological and paradigmatic in orientation. One component of this tradition - parsing - comes out strongly in NEG, though some other kinds of exercise mentioned by Michael (1985), for instance, are less fully represented. Consistent with earlier nineteenth-century developments, considerable weight is also laid in NEG on analysis. From the linguistic point of view, NEG provides clues concerning the movement from dependency to immediate constituent grammars - a move not unconnected with the development of analysis as a technique for teaching and practising grammar. NEG also provides evidence of the impact of diachronic and comparative studies upon the writing of vernacular grammars. Finally, to the pedagogical and the linguistic aspects must be added a third - technological - consideration: namely, the intended role of NEG in projecting the proposals put forward by the Joint Committee on Grammatical Terminology into the practical arena of school-teaching.

16

From this point of view, NEG constitutes one link in a chain which stretched from the foundation of the Birmingham Grammatical Society in 1885 to Sonnenschein's The Soul of Grammar in 1927.

The "simplification and unification of the terminologies and classifi-

cations employed in the grammars of different languages" (On the Terminology of Grammar:3, quoted in NEG Pt. I:Preface) was one of the main areas in which Sonnenschein invested his energies during the last three-and-a-half decades of his life. After the founding of the Birmingham Grammatical Society, the next step towards the achievement of this goal was the production of the Parallel Grammar Series - a set of language textbooks for teaching eight different languages, constructed according to the same three principles: uniformity of classification and terminology; uniformity of scope; and uniformity of size and type. Publication of the series continued until the turn of the century. In 1903, together with Postgate, Sonnenschein founded the Classical Association, with a view to furthering the teaching of classics. It was from a meeting of this association that a proposal for the simplification and unification of grammatical terminology came in 1908, and in 1909 a Joint Committee representing eight associations concerned with the teaching of languages was set up. This committee published the revised and final version of its report On the Terminology of Grammar - in 1911. The reception of Sonnenschein's terminological work in general, and of the Committee's report in particular, was less uniformly positive than Sonnenschein liked to claim. And, as he foresaw, the recommendations of a self-appointed committee, despite enjoying the support of its eight participating associations, would have little impact unless translated into a form which could be used in the schoolroom. The New English Grammar was one vehicle which was supposed to accomplish this, but it was not the only one. Oxford University Press published A New Latin Grammar and A New French Grammar, also by Sonnenschein, as complementary parts of the programme to secure the place of the revised terminology in schools. On the title-page of each of the three parts of NEG appear the words, "Based on the recommendations of the Joint Committee on Grammatical Terminology." Broadly, NEG is divided into three parts, each corresponding to the work for a school year, and subdivided into two sections.

The three parts are separately paginated, but

subdivided into paragraphs which are numbered consecutively across all three parts (for this reason, references are to paragraphs, where possible, rather than to pages). The titles of the three parts are:

"Pt. I, Parts of Speech and Outlines of Analysis; Pt. II, (A) Kinds of

Pronouns, Adjectives, and Adverbs, (B) Forms and Their Chief Meanings; Pt. Ill, (A) Structure of Sentences and Clauses, (B) Uses of Forms". Each of the three consists of an expository section written by Sonnenschein, together with an extensive set of exercises devised by Edith Archibald.

17

2.

NEG and contemporary linguistic thinking

If Percival (1976) is correct, NEG would stand near the crossroads in a series of developments which were taking place in grammatical thinking towards the end of the nineteenth century. Against the background of a more general movement from a synthetic to an analytic approach to grammar, a new and specific tendency to think of grammar in terms of its immediate constituents (I.C.) rather than of its dependency relations was becoming apparent. The initiator of this development was Wundt (1900). The approach taken by traditional grammar up to that time was synthetic in that it began by postulating lexemes or words, which it then combined to make larger units. Wundt rejected this in favour of the analytical approach favoured by Becker, which could be implemented by the successive, binary splitting of constituents from the sentence ("Gesamtvorstellung") downwards. At first sight, NEG appears to follow the conventional pattern: treatment of the parts of speech precedes the treatment of the structure of sentences just as Etymologia had preceded Syntaxis. Part II is a continuation of the treatment of the parts of speech begun in Part I. Despite the title of Part I, however, NEG in fact begins with the sentence as the largest unit, and the first five paragraphs are devoted to analysis. It is as if the first section of NEG Part I - "The Sentence - the Subject and the Predicate - Analysis of Sentences" - had been stuck on to the front of a grammar of the old type as a kind of afterthought, before the grammar proper - (the parts of speech) - began. The impression that Sonnenschein's analytical approach is rather in the nature of an appendage than integrated into the work as a whole is confirmed by the fact that the following sections are not developed along the lines of an analytical model, with each constituent being analysed into successively smaller units. Instead, a jump is made from the sentence (the largest syntactic unit) to the lexeme or part of speech as the smallest. In other words, there is at this stage, at least, no intermediate constituent level between word and sentence. From this point on, the grammar follows the familiar pattern, except that the individual parts of speech are presented within the framework of a table of analysis. In this respect, NEG falls short of the consistency of the Joint Committee's report (in the following referred to as Report), which moved directly from the binary analysis of the sentence, via functions (predicative adjectives, nouns or pronouns; object) - which could be realised by constituents of many types and sizes - to sentence types. The parts of speech per se were only explicitly dealt with in the Report from Recommendation XIII onwards. In NEG, by contrast, sentence types do not appear until Part III, which begins, "The Four Kinds of Sentence." NEG thus stands in the developing analytical tradition, but without exhibiting the feature of binary cutting which was to become one of the hallmarks of I.C. analysis. The relationship

18

between NEG and traditional dependency grammars can be illustrated with reference to the treatment of the phrase.

3.

Sonnenschein's treatment of the phrase

The phrase makes a good test-case because, as Percival pointed out, "equivalence of noun to noun phrase or of verb to verb phrase is absent from traditional grammar" (Percival 1976:233). Traditionally, the attention of teachers was centred on the word [...] the words were parts which identified themselves; they could be removed from the machine for scrutiny; they could be classified; they could be replaced by different but similar words without the functioning of the machine being affected. (Michael 1985:73) In NEG the phrase is clearly recognized and named. In this respect, it marks a step forward from the recommendations of the Joint Committee. There, it will be remembered (cf. Walmsley, forthcoming), the concept of phrase in a more modern sense was present from the beginning, but termed 'equivalent1, as in 'Noun or Noun Equivalent' (Recommendation I). It was only in Recommendations IXff. that the term phrase was introduced and set off from clause, "That part of a Sentence consisting of a group of words equivalent to a Noun, Adjective, or Adverb, but not having a Subject and a Predicate of its own, be called a Noun Phrase, Adjective Phrase, or Adverb Phrase" (Report, Recommendation IX). In NEG, phrase is introduced thus, In analysing sentences we often come across groups of words which go so closely together that they are equivalent to a single word [...] And even when we cannot find a single word which might be substituted for the group of words, we can see that certain groups of words are used like (or are equivalent to or do the work of) a single word. (Sonnenschein 1916:§33) A phrase may hence be defined as a group of words equivalent to a part of speech and not having a subject and a predicate of its own. (Sonnenschein 1916:§44) This statement gives some insight both into the way Sonnenschein perceived grammar, and also into the reason why his system was bound, ultimately, to break down. For a phrase can scarcely be defined as "equivalent to a part of speech". If this were true, it would not be easy to explain why there are not eight kinds of phrase parallel to the eight parts of speech which Sonnenschein postulated in NEG §5. What Sonnenschein really seems to have had in mind was the fact that phrases act as the exponents of grammatical Junctions. He seems to have thought of these functions rather like toy building blocks, which operated as units and could be realised by single lexemes or by larger constituents. They could be shifted around inside the sentence, but they were not - significantly - in his grammar further analysed into their

19

immediate constituents. The tables of analysis from §§37-190 confirm this, in that the analysis never goes below the level of the functional constituent. Constituents are arranged in columns under such heads as "Verb; Indirect Object; Direct Object; Predic. Adj., Noun or Pron.; Adverbial Qualification," but, in the tables at least, such phrases as a very great general, throughout the world (both §37), yellow and red, the Pied Piper's Street, scarce an inch before me (§190) are left unanalysed. This way of thinking explains why some phrase-types current in I.C. and modern grammar appear to be missing from NEG. There are no prepositional phrases (PPs), for instance, nor even verb phrases (VPs) in Sonnenschein's grammar. The reason is clear: prepositions do not operate as exponents of functions in the way nouns, adjectives or adverbs do, and verbs are not normally functionally replaceable by other 'groups of words' which are not themselves verbs. The present-day VP would cover that part of Sonnenschein's predicate which includes the verb and its complements, but not (necessarily) adjuncts. Modern VP thus appears in Sonnenschein's scheme as 'Verb' plus Object1 or 'Predicate Adj., Noun or Pronoun1 and, where relevant 'Adverbial Qualification1. Phrases containing prepositions, on the other hand, almost invariably function as adverbial or adjectival modifiers. A preposition - again, significantly is defined by Sonnenschein as "a word used with a noun or pronoun to form an adverb-phrase or an adjective phrase" (Sonnenschein 1916:§35), - (not, be it noted, "to form a. prepositional phrase"). In other words, 'adverb1 and 'adjective phrase' are being defined essentially in functional terms. This principle is not, however, carried through consistently: if it were, we would expect 'subject-phrase', Object-phrase', etc., named after functions where we in fact only find 'noun-phrase1. The algorithm which this requires the learner to develop in order to perform the task of analysis must look something like this, (1) (2)

identify the constituents at the phrasal level; assign functions to the constitutents (subject; object; indirect object; predicate adjective, noun or pronoun; adjectival; adverbial - noting that the 'adjectivals1 and 'adverbials' need not necessarily contain adjectives or adverbs); (3) label constituents expounding adjectival or adverbial functions 'adjective-' or 'adverb-phrase' respectively; (4) but label constituents functioning as subject, object, indirect object, predicative noun or pronoun 'noun phrase'. Note that it is not even possible to say "label all functions typically realisable by a noun 'noun phrase' [...]". What does one do, for instance, with one day in One day he came to see me! As a constituent it is NP, but it functions here adverbially, and hence, in Sonnenschein's terminology, ought to be an 'adverb-phrase'. This is not, however, the solution he chooses. According to §§168 and 189, one day is an 'adverbial accusative' (adverbial by function, but why 'accusative'?). This must make the parallel PP on a certain day [Sonnenschein's example - J.W.] not a 'prepositional phrase1 (which would be a constituent label), nor even an 'adverb-

20

phrase1 (which would be the functional label), but an 'equivalent to1 an adverbial accusative (cf. §189). The discussion so far shows that a phrasal level intermediate between the lexeme and the sentence was recognised in NEG, but that each phrase was defined and labelled on the basis of its function rather than its head constituent as today. The internal structure of the phrase was expressed not in terms of immediate constituent analysis, but in such traditional dependency terms as government, modification, etc.

4.

Comparative and diachronic philology

The synthetic : analytic and the dependency : immediate constituent axes are only two of the parameters along which NEG can be located. A third, which can only be lightly touched on here, concerns the relationship between contemporary developments in philology and the writing of vernacular grammars. In the discussions which surrounded Sonnenschein's work, much was made of 'scientific1 grammar. The term is used in significant passages putting the case for Sonnenschein's proposed terminological reforms, and it occurs in the lively exchanges which took place on this subject at the meetings of the Classical Association. What it signalled was the impact of nineteenth century philological research on vernacular grammar writing. Sonnenschein made good use of these developments for his own purposes. His reform, he claimed, depends on a recognition of the fact that English is a member of the Indo-European family of languages, and of the belief that harm is done to the study of English, both from the practical and from the scientific point of view, by wresting it from its historical associations. (Sonnenschein 1916 Pt. 11:3) The twin themes - the comparative and the diachronic - are here brought out with full clarity. Their importance for the categories used in NEG was that they were intended to provide legitimacy for a scheme which from a strictly language-immanent, synchronic point of view could scarcely be justified. One can thus, if one wishes, see Sonnenschein in a universalist tradition - which included Ratke (Ratichius), Jespersen and, more recently, Chomsky - as opposed to a structuralist position, which purported to describe every language exclusively on its own terms. Sonnenschein nevertheless differed from Jespersen, his nearest contemporary, as much by his inappropriate method of arriving at 'universal' linguistic categories as by his tendency to project them onto surface structures which could not support them. Nowhere is this more apparent than in the treatment of the secondary grammatical categories (Lyons 1968:274ff.) These constitute the fourth and last dimension along which Sonnenschein's work can usefully be located. They also constitute the areas around which most of the subsequent controversy clustered, from the initial reception of the Joint

21

Committee's Recommendations, through Jespersen's

alternative treatment in his Modern

English Grammar to Sonnenschein's last reply in The Soul of Grammar. Not least, case and mood constitute the greater part of the burden which Sonnenschein bequeathed to future generations. In other words, they are in no small measure responsible for some of the more erratic aspects of grammar which persisted in England for over fifty years after Sonnenschein's death. For these reasons, it is worth looking at his treatment of case, in particular, in somewhat greater detail.

5.

Sonnenschein on Case

Case, it should be remarked at the outset, can and has been defined in a number of different ways. At one extreme, it has been defined as an overt change of morphological form, and at the other as a relation between underlying arguments which may find overt expression in morphological inflection, by means of a preposition in a prepositional phrase, by zero, or through a specified position in a canonical sequence. Sonnenschein's choice among these different possibilities seems to have been motivated by his principle of making the vernacular "the basis on which the study of Latin grammar and of the grammars of other foreign languages may be erected" (Sonnenschein 1916, Pt. 111:4). By postulating five cases for English, Sonnenschein was able to claim that "the pupil who has mastered the uses of the English cases [...] will have little new to learn when he comes to Latin, except that Latin has an extra case - the ablative" (Sonnenschein 1916, Pt. 111:5). The fact that English nouns have no more than two endings to mark case - zero or unmarked and possessive/genitive seems not to have been perceived as a serious obstacle. It meant, of course, that the number of cases postulated would not be constrained by the forms of a word - (the strict Ramist position) - and in defending this view Sonnenschein showed himself an extremely able advocate, drawing on comparative and diachronic evidence when appropriate, and appealing to sympathetic authorities, It has often been said, and is still said by some grammarians, that the system of 'cases', 'tenses', and 'moods' on which the grammars of more highly inflected languages are built up is really inapplicable to modern English. But this contention seems to involve a misconception as to what is the signification of terms like 'case', 'tense', and 'mood'. It is impossible to frame a definition of such terms on the basis of distinctions in forms. They are essentially terms of syntax; that is to say, they denote categories of meaning, not categories of form. And this is just as true of Latin grammar as it is of English grammar. The term 'nominative case', for example, is applied to a great variety of forms in Latin; nor do the different cases of a Latin noun always differ from one another in form. (Sonnenschein 1916, Pt 11:3) It is equally true to say that form alone is not always sufficient to establish the correct case of a noun in a language. Once Sonnenschein had hammered home this principle for Latin, the

22

way was open for him to propose a larger number of cases for English than would have been justified by the forms of the language. In doing this, however, he did not go to the other - Wundtian - extreme. Wundt treated case more as a deep-structure relation between arguments, die Sprache kann gewisse Kasus bloß durch die Wortstellung ausdrücken, ohne daß am Worte selbst irgendwelche Veränderungen eintreten, die mit dem Kasusbegriff in Beziehung stehen [...] es können sich besondere, von dem Wort trennbare Partikeln entwickeln, die als Äquivalente der Kasusformen funktionieren. Solche Partikeln sind die Präpositionen [...]. (Wundt 1900:68-69) This view opens the way to a very large number of cases; it also contains a contradiction. First, the question of how many cases it implies has never been satisfactorily resolved, and it is interesting to note how developments in Fillmore's re-presentation of this view recapitulated the problems raised by Wundt's exposition five decades previously. In principle, it could mean as many cases as a language has prepositions - itself an indeterminate number unless complex prepositions are excluded; or even more, if the fact that some prepositions may signify more than one relation is taken into account. The consequences of this approach had been perceived and discussed by Lindley Murray, and were rejected by successive grammarians throughout the nineteenth century. The contradiction inherent in Wundt's approach stems from the fact that the relations claimed to be expressed by Wortstellung (by which Wundt presumably meant constituent order) are not necessarily the same as those expressed by prepositions. In the eyes of Wundt and his followers, the relations expressed by constituent order were functional, and in this respect in accord with Sonnenschein's view. Sonnenschein was later at pains, however, to draw a clear distinction between using functions as the basis for assigning case-labels to nouns in English on the one hand, and equating prepositions with cases, on the other (Sonnenschein 1927:9). Freeing his conception of case from form permitted Sonnenschein to postulate more than the two cases which present-day English nouns exhibit, without committing him to the large number which a dependence on prepositions would have required. The fact that he finally arrived at five cases for English suggests that he was in reality arguing predominantly in functional terms, and assigned case-labels largely on the basis of functions. There would have to be a vocative case, for instance, to deal with terms of address, and a genitive as the only marked case-form in current English. The three remaining cases were related to the three grammatical functions Sonnenschein believed in - subject, direct object and indirect object. This solution was, of course, highly convenient as a foundation for the teaching of Latin, which, for strategic reasons, was to be as closely linked to the teaching of English as possible (Proceedings of the Classical Association 1923:40).

23 When it came to the details of identifying individual cases, Sonnenschein did not hesitate to put forward arguments based upon diachronic as well as comparative evidence. Pin, for instance, in With sharp blue eyes, each like a pin had to be recognised as a dative (NEG §179). It was dative because it appeared in a phrase functioning as an adverbial equivalent ('like a pin' - the fact that prepositions were supposed to govern the accusative seems not to have been an obstacle). In the same paradigm, me in methought I heard a voice had also to be recognized as a dative, but this time on the grounds that "methought = it seemed to me, from the Old English yncan = 'seem1 [...]" (NEG §180). Nowhere is Sonnenschein's readiness to invoke diachronic evidence clearer than in his treatment of verbs which take two objects. Of these, one was nearly always 'accusative' and the other 'dative1. Some verbs, however, genuinely appear to take two 'accusative' objects, [...] the verb 'ask' may take a second direct object in the accusative case [...] He asked me a question. The verb 'teach' may also be regarded as taking two accusatives in sentences like the following: He taught me a lesson. He taught me French. (NEG:§§487-88) In terms of the kind of grammar Sonnenschein was writing, this would have been a synchronically admissible solution.

Incorporating a diachronic perspective,

however,

introduced a note of uncertainty as to what the case of each object should 'really' be, In Old English there were two verbs meaning 'teach'. One of them took two accusatives, and the other a dative and an accusative. In Modern English the dative never differs from the accusative in form; so that it is impossible to say with certainty whether one of the cases after the verb 'teach' is etymologically the same as the Old English verb which took dat. and accus. (NEG:§488) He then added the footnote, "TXcan took dat. and accus.: l&ran took two accusatives" (loc. cit.). It was one thing for Sonnenschein to arrange these things to his own satisfaction. It was another to persuade the world at large of the Tightness of his theories, and something else again to present them in a way amenable to teaching. It is striking to see how far - in pursuit of these objectives - arguments concerning case dominate the presentation in NEG in general - in the individual Prefaces to the three parts, - and in a special section "Forms and Their Chief Meanings" in §§140-45 (NEG, Pt. II:§§35-38) in particular.

6.

The appeal to authority

In addition to using comparative and diachronic criteria when they suited his purpose, Sonnenschein in NEG looked back to the work of earlier grammarians to see what support he

24

could find for his theories there. Not surprisingly, he discovered a tradition which treated case as an overt change in termination. Dr. Johnson's definition, which made use of this approach, Sonnenschein dismissed as an "absurdity" (Sonnenschein 1916, Pt. 11:3). The struggle towards the light was evidently slow, "Butler recognizes two cases of nouns, 'rect1 and Oblique'. Ben Jonson also recognizes two, a 'genitive' and what he calls an 'absolute' case. Priestley and Lowth get no further" (Sonnenschein 1916, Pt. 11:4-5). Sonnenschein did nevertheless manage to find one grammarian who appeared to lend support to his approach: Lindley Murray. "Murray", he wrote, after vacillating in his early editions, takes the momentous step of recognizing an Objective' case of nouns, though its form is not different from that of the nominative [...]. In taking this step Lindley Murray rendered English grammar the service of liberating it from the false definition of 'case' by which his predecessors had been hampered; for he thus opened the door to the admission of a dative case - a case which is syntactically distinct from the accusative, though it does not differ from it in form in Modern English'. (Sonnenschein 1916, Pt. 11:5) Sonnenschein saw Murray, in other words, as the pioneer of a new orientation in English grammar - one which first isolated functions, and then assigned case labels to the constituents expounding them. The question is, is this a fair representation of Murray's position? - and, if so, why was it that, after vacillating in the early editions, Murray took his momentous step? Murray's position in his English Grammar (1795) was at first diametrically opposed to Sonnenschein's, and in line with the tradition represented by Johnson, Priestley and Lowth, who Murray himself actually quoted in support. Murray printed a paradigm in his English Grammar which contained only four items - a nominative and a possessive case of the noun, singular and plural, SINGULAR.

PLURAL.

Nominative Case.

A mother.

Mothers.

Possessive Case.

A mother's.

Mothers'.

Murray continued, For the assertion, that there are in English but two cases of nouns, and three of pronouns, we have the authority of Doctors Lowth, Johnson, Priestley, &c. names which are sufficient to decide this point. If case in grammar mean only the variation of a noun or pronoun, by termination or within itself, (as it indisputably does) with what propriety can we distinguish the relations signified by the addition of articles and prepositions, by the names of cases? On this supposition, instead of five or six cases, we shall have a number equal to the various combinations of the article and different prepositions with the noun [...]. (Murray 1798:28)

25

- a conclusion he was clearly rejecting.

Murray's refusal to recognize more than two

inflections of the English noun, however, did not prevent him from identifying grammatical functions correctly, But though, in the following sentence, Ά wise man controuls his passions', we cannot properly say that the noun 'passions' is in the objective case, and governed by the active verb 'countroul1, yet we may with propriety assert, that the noun 'passions' is the object of that active verb [...]. (Murray 1798:29) If such was Murray's position in 1795, what was it that led him to take the 'momentous step' of recognizing an Objective' case of nouns? There is a clue at least at early as the fourth edition of the English Grammar (1798). Here, Murray inserted a further sentence which ran, If, however, any teachers should be of opinion, that the business of parsing may be better conducted, by assuming, for this purpose only, an objective case of nouns, there can be no great objection raised against the practice, provided it be set in a proper light, and clearly explained to the learner. (Murray 1798:37)

7.

The pedagogical dimension

The villain of the piece, then, was the practice of parsing. Requiring pupils to state the case of the noun in English, along the lines of the Latin model, meant that grammarians had to say what cases there were in English, and how they could be identified. This brings out clearly how strong an influence the other dimension Murray and Sonnenschein operated in - the pedagogical - exercised on grammatical decisions. The problem which each of them had to deal with, in other words, was in reality a problem of teaching method. In both schemes, parsing played a prominent part, and this brought each face to face with the problem of how to parse uninflected nouns in English whose Latin equivalents presented no problem. The correct answer to this would have been to keep the functions strictly separate from the caseforms, and accept that English uses an unmarked case for almost every function. This was the approach adopted by Sweet (he called it the 'common' case - a term rejected by Sonnenschein) and by Jespersen. It seems, too, to have been the approach initially adopted by Murray. The possibility of speaking of the object of a verb without therefore having to postulate more than two cases for the noun was claimed by Murray to "answer all the ends of parsing, and of shewing the connection and dependence of words under such circumstances [...]" (Murray 1795:29). Where Murray seems to have gone astray was in equating the unmarked form of the noun with the nominative case. The consequence, as he discovered quite early, was that under this system many nouns could not be sensibly parsed - i.e. have any case assigned to them. He therefore postulated an Objective' (instead of an 'unmarked' or 'common' case) in order to avoid "the irregularity of having our nouns sometimes placed in a situation in which

26

they cannot be said to be in any case at all" (Sonnenschein 1916, Pt. 11:5, quoting Murray's 38th. edition).

8.

Conclusions

Case was only a detail, though an important one, in Sonnenschein's grand design for the simplification and unification of grammatical terminology for some of the most widely taught European languages. Within this programme, NEG has been shown in terms of a number of parameters to occupy an intermediate position: it came after the recommendations of the Joint Committee on Grammatical Terminology, but before The Soul of Grammar, In terms of the synthetic-analytic parameter, NEG adhered in principle to the conventional pattern of the parts-of-speech being followed by the study of constructions - (on the even older pattern of morphology (Etymologia) followed by syntax) - but with a short, preposed section on analysis. The concept of phrase, as a level intermediate between lexeme and sentence is more highly worked out in NEG than in On the Terminology of Grammar, but no obvious use is made of binary splitting inside the phrasal constituent. Finally, the influence of comparative philology makes itself apparent at all levels - from the grand design at the top, to the individual grammatical categories at the bottom. In one respect, however, Sonnenschein stood in express opposition to contemporary developments. NEG shows no evidence of the distinction between synchrony and diachrony, which at the time when NEG was presumably in gestation, was being more clearly developed by de Saussure, and which led to the more precise description of limited synchronic or diachronic fragments of the language. Many of these features are concentrated in Sonnenschein's treatment of case. Sonnenschein established case on the basis of a mixture of formal and functional criteria. Having freed himself from the exigencies of form, Sonnenschein postulated five cases for contemporary English - a number which provided a convenient foundation for the subsequent teaching of Greek and - particularly - Latin. The English cases were given Latin names and labels were assigned (in parsing), on functional grounds. It was here that the problems became most acute, because much time had to be spent trying to justify the fact - and helping pupils to see - that cases which were indistinguishable in terms of form were to be labelled according to functions which were not always easy to identify. Despite differences in motivation, certain parallels between Sonnenschein's treatment and Chomsky's later thinking on case are inescapable. Sonnenschein's solutions to such individual grammatical problems flowed from his approach to grammar as a whole, and hence tend to be repeated in other areas of the grammar. At the more abstract level, the debate was concerned with the question of whether grammatical categories - the secondary grammatical categories in particular - were to be

27

defined in terms of form or notionally. Sweet and Jespersen preferred to reserve secondary grammatical category labels to designate formal oppositions which nevertheless express notional, underlying distinctions in the language.

It is noticeable how frequently

Sonnenschein used denotation as a basis for grouping words together in classes and formulating definitions.

Since Jespersen was himself a committed notionalist (Jespersen

1924:346f.), it is not surprising if subsequent observers erroneously placed Sonnenschein and Jespersen together in the same camp (cf. e.g. Nida 1960:28). Sonnenschein adduced comparative and diachronic evidence in support of his arguments wherever he could find it. The appeal to authority, however, is much more strongly marked in NEG than in On the Terminology of Grammar. It was to play a role, too, in The Soul of Grammar, but its use in NEG is particularly striking since NEG was expressly conceived as a teaching grammar. It surely tells us something of the significance of the controversy in Sonnenschein's eyes, and of the central role of case in the controversy, that this kind of argumentation, together with appeals to the views of Lindley Murray, should appear at all in a book intended for the use of schoolchildren. The position of NEG in linguistic history thus reflects both one step in a number of more general developments, and the culmination of an apparently unique development in the history of English grammar - one man's crusade to co-ordinate the terminology for a number of different languages. Sonnenschein's excursions into comparative philology and diachrony were almost all subordinated to this end. That NEG represents but one milestone in a long tradition of universal grammar from Ratke to Chomsky is clear from the aims and approach it embodies. It is easy to dismiss Sonnenschein's work on the basis of his approach. But those who like to sneer, as Fillmore said, with Jespersen, at his favourite 'bad guy' Sonnenschein, who, unable to decide between Latin and Old English, allowed Modern English teach to be described as either taking a dative and an accusative, because that was the pattern for Old English tcecan, or as taking two accusatives, in the manner of Latin docere and German lehren [...]. (Fillmore 1968:5) would do well to ponder the following quotation from Chomsky, Assume [...] that a verb assigns accusative case to its object and that a preposition assigns oblique case (which may have one or another form) to its object. The case system may be richer and some further variety may appear, but let us take this to be the rudimentary structure of the case system. The cases may be overt, as generally in Latin, or hidden, as generally in Spanish and English, but we assume them to be present, in accordance with these general principles, whether overt or hidden. (Chomsky 1988:102-3) This particular strand of Sonnenschein's grammatical thinking thus appears to be by no means dead.

28

Bibliography Becker, K.F. (1827): Organismus der Sprache. Als Einleitung zur deutschen Grammatik. Deutsche Sprachlehre. Vol. 1. - Frankfurt Becker, K.F. (1836-39): Ausführliche deutsche Grammatik als Kommentar zur Schulgrammatik. 3 vols. Frankfurt. Chomsky, N. (1988): Language and Problems of Knowledge: The Managua Lectures. - Cambridge, Mass. Deutschbein, M. (1917): System der neuenglischen Syntax. - Cöthen. Fillmore, C.J. (1988): "The Case for Case." - In: E. Bach & R.T. Harms (eds.): Universals in Linguistic Theory (London/New York) 1-88. Jespersen, O. (1904-49): Modern English Grammar, l vols. - Heidelberg/Copenhagen/London. Jespersen, . (1924): The Philosophy of Grammar. - London. Jespersen, O. (1925-6): "Die grammatischen Rangstufen." - In: Englische Studien 60,300-09. Mason, C.P. (1858): English Grammar; Including the Principles of Grammatical Analysis. - London. Michael, I. (1985): "Prosing, Transposition and Other Linguistic Exercises." - In: Histoire Epist mologie Langage VII-2,71-85. Morell, J.D. (1852): The Analysis of Sentences Explained and Systematised: After the Plan of Becker's German Grammar. -London/Manchester. Murray, L. (1795; 41798): English Grammar, Adapted to the Different Classes of Learners. - York. Nida, E. (1960): A Synopsis of English Syntax. - Norman: University of Oklahoma Summer Institute of Linguistics. On the Terminology of Grammar: Being the Report of the Joint Committee on Grammatical Terminology (1911) - London. Proceedings of the Classical Association. (1923) - London. Onions, C.T. (1904; ^1970): An Advanced English Syntax, Based on the Principles and Requirements of the Grammatical Society. - London. Percival, W.K. (1976): "On the Historical Source of Immediate Constituent Analysis." - In: J.D. McCawley (ed.): Syntax and Semantics. Vol. 7 (New York) 229-42. Sedgwick, W.B. (1930): "Sonnenschein, Edward Adolf (1851-1929)." - In: J.R.M. Weaver (ed.): Dictionary of National Bibliography 1922-30 (Oxford) 796-98. Sonnenschein, E.A. (1912): A New Latin Grammar, Based on the Recommendations of the Joint Committee on Grammatical Terminology. - Oxford. Sonnenschein, E. A. (1912): A New French Grammar, Based on the Recommendations of the Joint Committee on Grammatical Terminology. - Oxford.

29 Sonnenschein, E.A. (1916): A New English Grammar, Based on the Recommendations of the Joint Committee on Grammatical Terminology, - Oxford. Sonnenschein, E.A. (1927): Tfte Soul of Grammar. - Cambridge. Walmsley, J.B. (1988): "The Sonnenschein v. Jespersen Controversy." - In: U. Fries, M. Heusser (eds.): Meaning and Beyond: Ernst Leisi zum Geburtstag (Tübingen) 253-81. Walmsley, J.B. (forthcoming): . . Sonnenschein and grammatical terminology." - In: G. Leitner (ed.): English Traditional Grammars: An International Perspective (Amsterdam). Wundt, W. (1900): Völkerpsychologie: Die Sprache. 2 vols. - Leipzig.

Reconstructing Seasons Past: Linguistic Change

33

Anja de Brabandt

The Verb speak in Shakespeare's Plays1 0.

Introduction

That which holds for so many features of Early Modern English (EME) can also be said of Shakespeare's use of the verb speak: it strikes modern readers as inconsistent, grammatically 'incorrect1, or otherwise strange, but does not impede our understanding of a given passage. Thus, though we might feel puzzled by the way speak is used in the following example, we have no problems understanding what Rosencrantz means when he says of Hamlet: He does confess he feels himself distracted, But from what cause 'a will by no mans speak. (HAM III. 1.5-6)2 As a comparison of the whole Shakespeare-canon with a corpus of sixty modern British stage plays reveals,3 speak in modern English (ME) occurs considerably less frequently than it did in EME. In relation to three other verbs of linguistic action - talk, say, and tell - speak comprises 7.15% of the examples in the modern plays, whereas it comprises 28% of the examples in Shakespeare's works.4 It is, therefore, reasonable to assume that speak in EME was used in contexts where today say, talk, tell or other verbs of communication would have to be used or would, at least, be more likely. ME speak is, like talk, basically an intransitive verb, whereas say and tell are, with a few exceptions, transitive verbs. Speak and talk can be used without complements, for example: This paper is a shortened version of an unpublished dissertation submitted at the University of Osnabrück in November 1987. 2 References to Shakespeare's plays are to the line-numbering of the Riverside Shakespeare. This edition has been chosen because it served as a basis for Marvin Spevack's Harvard Concordance to Shakespeare. For a listing of the abbreviations of cited plays see section 1 of the bibliography. 3

4

see Dirven et al. (1982), who analysed 60 plays dating from the period 1966-72.

Absolute figures: speak, talk, say, and tell in modem British plays and in Shakespeare (Ggures taken fom Dirven et al. 1982 and from Spevack 1974): speak talk say tell total Modern plays 580 966 4239 2355 8104 Sh.'s works 1702 247 2612 1476 6037 This means that talk in the modern plays occurs 1.66 times as often as speak, say 7.4 times as often as speak, and tell 4 times as often as speak. In Shakespeare's works, however, speak occurs 6.9 times as often as talk, say occurs only 1.5 times as often as speak, and speak is slightly more frequent than tell.

34

He talked and talked. Speak! or can take prepositional phrases such as to/with someone or of/about/on something. Speak and talk with a to- or ννιίΛ-phrase can mean 'address someone1, 'exchange remarks with someone1, or 'converse with someone': Have you spoken to the manager? I talked to him yesterday. Speak/talk of/about/on could be paraphrased as 'discuss, enlarge upon', 'mention', or 'discourse upon, give a speech on': Mrs Thatcher spoke of her plans for the inner cities the day after the election. We have not yet talked about our plans for our upcoming holidays. The range of direct objects that speak and talk can take is very restricted. Combinations of speak and talk + DO are more or less idiomatic expressions, e.g. speak English, speak the truth, speak one's mind, speak one's lines, talk linguistics, or talk nonsense. Speak and talk cannot normally include an explicit mention of the contents of an utterance; that is, direct quotations, embedded statements, κΤζ-clauses and pronouns are normally excluded after speak and talk. Such direct objects refer to the contents - either explicitly (with direct quotations and embedded statements), or in a somewhat more condensed way (with w/i-clauses and pronouns) - which is typical for say and tell, but in standard educated English not possible with speak and talk: *Mary spoke that she had not yet done her homework. * Peter talked what he had seen in Paris. Though ME speak and talk share the same syntactic features, they cannot always be used interchangeably. The basic differences between speak and talk are, in the first place, that talk would tend to be preferred when there is a conversation between people. Talk to usually implies linguistic interaction, whereas speak to often means 'address'. Secondly, talk is more likely to be used when the topic of a conversation (e.g. about one's work) is dealt with in more detail. There is a tendency in ME to use the preposition about instead of of- with verbs like speak, talk, think or dream - when the topic is dwelt on rather than mentioned. In the corpus of modern plays, about-phrases do not often occur after speak but are frequently used with talk (see Dirven et al. 1982:166). With speak, o/-phrases occur more than three times as often as a&ouf-phrases, which suggests that the topic is often 'mentioned1 rather than 'discussed, enlarged upon' (see Vorlat 1982:27-29). Thirdly, speak can suggest a formal address to a group: I hear the President is going to speak on television this evening.

35 1.

Speak in Shakespeare

The following analysis of examples drawn from Shakespeare's plays shows that the use of speak in EME was far less restricted than it is in ME. Bearing in mind, first of all, that ME speak and talk have basically the same grammar but cannot always be used interchangeably, for example: The vice-chancellor spoke at the meeting. ?The Vice-chancellor talked at the meeting. We talked together for hours about politics. ?We spoke together for hours about politics. and secondly, that speak and talk are prototypically intransitive verbs, whereas say and tell are transitive verbs, two generalizing observations can be made: (a)

In contexts where speak in Shakespeare is used intransitively, speak would today still be possible, but talk would tend to be preferred. This is the case, for example, when there is a conversation between people, or when the topic is dealt with in detail.

(b)

The most striking difference between ME and EME, however, is that speak in EME could include an explicit mention of the message, i.e., it could have direct object complements like direct quotations, that- and ννΛ-clauses, and pronouns. Speak was thus used in the sense of 'state; assert; express the view1 and of 'inform1, which today are the basic meanings of say and tell respectively.

In addition, speak could have a far wider range of object noun phrases than it can have in present-day English. In such cases, one would in ME often need an expression containing a non-linguistic action verb.

1.1 Speak + 0 Emma Vorlat (1982:18) found in her corpus of modern plays that speak "favours syntactic surroundings with the negative modals of ability cannot/'couldn't and with was/is unable"; in other words, speak was often used in contexts where someone, for some reason or other, was not able to utter anything or at least not what was expected of him or her. This was also possible in EME: Cassio:

I pray you pardon me, I cannot speak. (OTHII.3.189)

However, more often than not speak was used in contexts where someone did utter something or was asked to do so. In such contexts, say something or tell me/us would often be preferred in ME:

36

Horatio:

It lifted up it head and did address Itself to motion like as it would speak; (HAM 1.2.216-17)

Othello:

Is't lost? Is't gone? Speak, is't out o' th1 way? (OTH III.4.80)

Rene Dirven (1982:43) points out that ME talk can be used without a complement to denote "extensive linguistic action". In a context like the following one would perhaps prefer talk today. Edmund has just told Edgar that their father is terribly enraged. He asks Edgar to come with him to his lodgings where he will provide proof of Gloucester's anger: Edmund:

[...] retire with me to my lodging, from whence I will fitly bring you to hear my lord speak. (LR 1.2.168-69)

Speak could also have the meaning of "mere verbosity, i.e. linguistic action without any serious contents", which today is typical for talk (e.g. "He just talks", see Dirven 1982:47), as illustrated by: Cordelia:

I yet beseech your Majesty If for I want that glib and oily art To speak and purpose not, (LR 1.1.223-25)

Speak could also be used by itself, where we would need a complement such as, for example, about it or to him: King:

1.2

What says he to your daughter? Have you spoke? (AWW V.3.28)

Speak + Adverb of Manner

Just as in present-day English (e.g. speak loudly, rapidly, coherently, glibly, fluently, frankly) speak in EME could be followed by adverbs referring to the tone of voice or speaking rhythm, to the use of a particular language or style of speech, or to the speaker: Desdemona:

Why do you speak so startingly and rash? (OTH III .4.79)

Desdemona:

He speaks well. (OTH IV.3.37)

Lucius:

Dost thou speak seriously, Servilius? (TIM III.2.42)

In addition, speak was followed by adverbs of manner that refer to the contents of an utterance in terms of its truth value, such as true, truly, untrue, false and right. In ME one would have to say speak/tell the truth, lie or tell a lie and be right:

37

1.3

Banquo:

What, can the devil speak true? (MAC 1.3.107)

Macbeth:

If thou speak'st false, (MAC V.5.37)

Edmund:

Th' hast spoken right, 'tis true. (LRV.3.174)

Speak + Preposition

1.3.1 To/mth In view of the low number of ννιίΛ-phrases and the high number of to-phrases in the corpus of modern plays (12 and 160, respectively, see Vorlat 1982:23), one can say that speak with in ME is the marked form, i.e., it is used when one wants to make clear that linguistic interaction is implied. In Shakespeare's works there are as many to-phrases as wiVA-phrases (about 115 of each in the whole canon^). Though both speak to and speak with are used in the sense of 'address' and 'converse with', speak to seems to be preferred when the "address'-meaning is obvious. It occurs, for example, when Lear is about to regain consciousness and Cordelia asks the doctor to say something to him, or when Desdemona has just been stabbed and Othello is afraid of Emilia's coming in and saying something to her: Cordelia:

He wakes, speak to him. (LR IV.7.41)

Othello:

If she come in, she'll sure speak to my wife. (OTH V.2.96) Speak to, however, was also often used in the sense of 'converse with', for example, when Hamlet makes up his mind to have a talk with one of the grave-diggers:6

^ The examples of speak + noun/pronoun/clause + ίο-phrase will be excluded in this analysis, because they do not focus on the linguistic action as such, but on the contents, and therefore say or tell would have to be used in

ME: Queen:

To whom do you speak this? (HAM III.4.131) Horatio: And let me speak to [th'J yet unknowing world How these things came about (HAM V.Z379-80) " The noun speech, too, could have the meaning talk, conversation': Edgar: If e'er your Grace had speech with man so poor, Hear me one word. (LR V.l.38-39) lago: Thus it is, general: Montano and myself being in speech, There comes a fellow crying out for help, (OTH Π.3.224-26)

38

Hamlet:

I will speak to this fellow. (HAM V.l. 117)

In the passive speak to could have the meaning of 'ask, request1: Chamberlain:

We shall be late else, which I would not be, For I was spoke to, with Sir Henry Guilford This night to be comptrollers. (H8 1.3.65-67)

As far as speak with is concerned, it was often used with modals of permission .or verbs of requesting, which in the modern plays occurred frequently with speak to (see Vorlat 1982:24): Horatio:

What are they that would speak with me? (HAM IV.6.1)

Emilia:

I do beseech you That I may speak with you. (OTHV.2.102-03) That speak with could also be used in the passive, even in an imperative, is illustrated by: Horatio:

Twere good she were spoken with, (HAM IV.5.14)

Buckingham:

Be not you spoke with but by mighty suit; (R3 IH.7.46) Here, a passive construction seems strange to modern readers. However, speak with in the passive, though unusual, is not impossible in ME. 1.3.2 About/of'Ion As was pointed out earlier, there is a tendency in ME to use the preposition about when the topic is discussed or enlarged upon, whereas with of the topic is often merely mentioned. On implies some kind of lengthy, systematic treatment. In Shakespeare, this distinction was not yet made. Speak about occurs only three times in his works, in H8 (V.l.84), in WIV (IV.5.34) and in: Banquo:

Were such things here as we do speak about? (MAC 1.3.83) Speak on did not yet have the meaning of 'discourse upon': Pompey:

Ah, this thou shouldst have done And not have spoke on't! (ANT II.7.73-74) As far as the speak o/-examples in the corpus are concerned, in a context like the following speak of would also be appropriate in ME, as speak of could be paraphrased by 'mention1:

39

Edmund:

This is the letter which he spoke of, (LRIII.5.10)

In the following example, however, one expects that the event - the appearance of the ghost will be dealt with in more detail, so that we would prefer to use the preposition about:'1 Horatio:

1.4

Well, sit we down, And let us hear Barnado speak of this. (HAM 1.1.33-34)

Speak + Direct Object

1.4.1 Noun Phrases Just as in ME, speak in Shakespeare could be followed by a reference to a language, a kind of Oblique object1 (= an object as reference to words that does not contain the words actually used), a truth evaluation, and a recitable entity: Host:

Let me speak with the gentlemen; they speak English? (WIV IV.3.5-6)

Costard:

But there are Worthies a-coming will speak their mind in some other sort. (LLL V.2.584-86)

lago:

Yet I persuade myself, to speak the truth Shall nothing wring him. (OTH II.3.223-24)

Hamlet:

I had as live the town-crier spoke my lines. (HAM HI.2.3-4)

However, the range of noun phrases with which speak could occur was far less restricted than it is today. Unlike in ME, speak could have a 'negative' truth evaluation as object: Young Siward:

I'll prove the lie thou speak'st. (MACV.7.11)

Objects like prophecy or speech, which, like one's lines, could also be regarded as recitable entities, would in ME occur with non-linguistic action verbs such as make, and give or deliver: Fool:

I'll speak a prophecy ere I go: (LR III.2.80)

Even a verb such as confer, which implies some kind of discussion or exchange of opinions and which in ME takes and about- or ση-phrase, could be used with of. Edmund: If your honour judge it meet, I will place you where you shall hear us confer of this, (LR 1.2.90-91)

40 Hamlet:

I heard thee speak me a speech once, (HAM H.2.434) As the following examples illustrate, speak was also used with objects like things and words, i.e. objects that refer to the contents of a communicative event. In ME one would use the verb say:6 Cleomines:

You might have spoken a thousand things that would Have done the time more benefit, and grac'd Your kindness better. (WTV.1.21-23)

Horner:

Alas, my lord, hang me if ever I spake the words. (2H6 1.3.197-98) There is a difference between speak with and speak to when used with an object like word. As the following examples show, with speak with the focus is not so much on the object (or contents), but rather on the verb (or communicative event): Emilia:

O, good my lord, I would speak a word with you! (OTH V.2.90)

Quickly:

Hark ye, Master Slender would speak a word with you. (WIV IH.4.29-30) In examples of speak + word + to, however, the emphasis is on the contents, which can be supported by the fact that word is always further specified, either by an adjective or by a determiner: Arthur:

And ne'er have spoke a loving word to you; (JN IV. 1.51)

Austria:

O, that a man should speak those words to me! (JN III.1.130) That speak in EME could occur with a direct object where ME would require an extension in the form of an about- or o/-phrase can be illustrated by: Othello:

Did I but speak thy deeds. (OTH IV.2.76)

Gloucester:

I have receiv'd a letter this night 'tis dangerous to be spoken; (LRIII.3.9-11)

An object like words is not excluded after speak in ME, but then the meaning is different. We could say, for example: Who spoke the words, "Nothing will come of nothing, speak again"? Here, however, the reference is to what could be called a 'recitable entity'. The emphasis is on reciting rather than on communicating. But this is not the case in the examples quoted from Shakespeare.

41

For the second example there are two interpretations: Gloucester might mean that it is dangerous to talk about the letter he has received, in which case we would use speak or talk (in an active sentence) followed by an fl&oid-phrase. Alexander Schmidt, however, glosses 'say'; in this case Gloucester would mean that it is dangerous to make known the fact that he has received a letter. Then we would have to use say because ME speak cannot take clause complements. What is particularly interesting is that speak could occur with noun phrases that say something about the manner of speaking: Hamlet:

I will speak [daggers] to her, but use none. (HAM IH.2.396)

York:

(Within) Open the door, secure foolhardy King! Shall I for love speak treason to thy face? (R2 V.3.43-44) Other combinations of this kind are, for example, speak blasphemy (2H6 V.2.85), speaks holiday (WIV IH.2.68), leprosy [...] speak'st (TIM IV.3.362), speaks nothing but madman (TN 1.5.106-07), speak parrot (OTH II.3.279), speak patience (ADO V.I.27), speak to thee plain soldier (H5 V.2.149), speaks poniards (ADO H.1.247), and speak'st reason (ADO V.1.41). 1.4.2 Pronouns Whereas ME speak can have a pronoun as a direct object only if it refers to a recitable entity (see fn. 8), EME speak could be used with pronouns that refer to any kind of utterance. Pronouns occurring after speak in Shakespeare's works are the quantifying pronoun more, the neuter pronoun it, the demonstrative pronouns this and that, the relative pronoun that, and the interrogative pronoun what: Hamlet:

O, reform it altogether. And let those that play your clowns speak no more than is set down for them, (HAM IH.2.38-40)

Macbeth:

More shall they speak; for now I am bent to know, By the worst means, the worst. (MAC HI.4.133-34)

Regan:

And speak't again, my lord, no more with me. (LR H.4.255)

Benedick:

But speak you this with a sad brow? (ADO 1.1.182-83)

42

Gratiano:

All that is spoke is marr'd. (OTH V.2.357)

Cordelia:

What shall Cordelia speak? (LR 1.1.62)

1.4.3 Direct quotation and clauses Speak in Shakespeare could be followed by a direct quotation, which in ME is typical for say. That Shakespeare could use speak and say interchangeably in this respect is illustrated by: Troilus:

I speak not "be thou true" as fearing thee, For I will throw my glove to Death himself That there is no maculation in thy heart; But "be thou true" say I to fashion in My sequent protestation: (TRO IV.4.62-66)

Reported speech, too, could be introduced by speak: Cassio:

[...] and even but now he spake (After long seeming dead) lago hurt him, lago set him on. (OTH V.2.327-29)

Malcolm:

[...] and 'tis spoken, To the succeeding royalty he leaves The healing benediction. (MAC IV.3.154-56)

Other object clauses that occur in Shakespeare's works are wA-clauses. Speak by itself cannot be followed by wA-clauses in ME; we either have to say speak about or use say or tell together with an indirect object: lago:

It is not honesty in me to speak What I have seen and known. (OTH I V.l.277-78)

Buckingham:

And when you would say something that is sad, Speak how I fell. I have done; and God forgive me! (H8II.1.135-36)

There is one instance in Shakespeare where speak + ννΛ-clause is used in combination with a ίο-phrase: Horatio:

And let me speak to [th1] yet unknowing world How these things came about. (HAM V.2.379-80)

In such a context tell (followed by a prepositionless indirect object) would be the verb most likely to be used in ME because, as Putseys (1982:141) points out, "the basic function of tell

43 is to impart new information to the addressee. Say, on the oher hand, may also be used to express in words what both speaker and hearer already know." As the foregoing analysis shows, EME speak was not basically an intransitive verb, but was used both transitively and intransitively. In its intransitive use speak, like ME talk, often denoted linguistic interaction. Moreover, the topic of the linguistic action could be dealt with in considerable detail, which again today is more typical for talk. In its transitive use, speak could have the same range of pronouns and clause complements with which ME say and tell typically occur. In addition, speak could be used with far more noun phrases than ME speak, talk, say and tell. Since speak cannot be regarded as an isolated item, some aspects of the meanings and usage of EME talk, say, and tell are dealt with in the following.

2.

Talk, say, and tell in Shakespeare

2.1

Talk

Although talk played in comparison with speak only a minor role in Shakespeare's works (there are 247 fa//fc-examples, that is 4% of all the examples of speak, talk, say and tell), it can be said that talk was used basically in the same way as it is used in ME. Its characteristic features were - and still are - a) that it often included a topic (see the relatively high number (117) of of- and on-phrases): Portia:

Ay, that's a colt indeed, for he doth nothing but talk of his horse, (MV I.2.40-41)

b) that it could designate longer lingusitic action when used without a complement: Romeo:

Let's talk, it is not day. (ROM III.5.25)

and c) that it often implied a conversation. In contrast to speak, talk had more m'/A-phrases than to-phrases (35 and 13, respectively). As in ME, both talk to and talk with could be paraphrased by 'converse with': Cassio:

Γ was the other day talking on the sea-bank with certain Venetians, (OTH IV. 1.133-34)

Hubert:

[Aside] If I talk to him, with his innocent prate He will awake my mercy, which lies dead; Therefore I will be sudden, and dispatch. (JN I V.l.25-27)

Talk was accompanied by manner adverbs such as idly and wisely and by adverbial likephrases:

44 Pucelle:

Take heed, be wary how you place your words, Talk like the vulgar sort of market men (1H6 IH.2.3-4)

There is one instance of a substantivized adjective: Leontes:

Thou canst not speak too much, I have deserv'd All tongues to talk their bitt'rest. (WT ΙΠ.2.215-16)

However, pronouns and the kind of noun phrases with which speak was so often used did not occur with talk. Clause complements occur only two times: Gonzalo:

Sir, we were talking that our garments seem now as fresh as when we were at Tunis at the marriage of your daughter, who is now queen. (TMP II. 1.97-99)

York:

Grandam, one night as we did sit at supper, My uncle Rivers talk'd how I did grow More than my brother. (R3 H.4.10-12) As Dirven (1982:76) points out, ME talk "can be used in the sense of 'causing a result by talking'". This was also possible in Shakespeare: Desdemona:

I'll watch him tame, and talk him out of patience; (OTH HI.3.23)

Leonato:

O Lord, my lord, if they were but a week married, they would talk themselves mad. (ADO H.l.353-54) That talk in Shakespeare gave the idea of longer linguistic action is supported by the fact that talk in about 10% of its occurrences was used in the progressive, even in combination with the future, the conditional, and the present perfect, which was rare in EME. The figures in this respect are considerably higher for talk than for the other verbs under discussion. Sebastian:

He will be talking. (TMP II. 1.27)

Orlando:

I take some joy to say you are, because I would be talking of her. (AYLI V.l.89-90)

Desdemona:

I have been talking with a suitor here, (OTH HI.3.42) These observations show that both in ME and in EME speak and talk can and could to a certain extent (with to- and witft-phrases, with about-, of- and on-phrases, with adverbs) be used interchangeably. Talk in ME is often used in contexts where in Shakespeare's plays speak was used.

Speak would today often be considered more formal, and talk in

Shakespeare was often used in less formal language. About one third of the ta/fc-examples in

45

Shakespeare are in prose, and a number of fa/£-examples occur in verse in less formal contexts. Lear, for example, uses speak with when he considers himself a king: Lear:

The King would speak with Cornwall, the dear father Would with his daughter speak, (LR H.4.101-02)

He uses talk with, however, when, already mad, he is on the heath, and at the end of the play when he is happy to be reconciled with his daughter Cordelia:

2.2

Lear:

First let me talk with this philosopher. (LR III.4.154)

Lear:

So we'll live, And pray, and sing, and tell old tales, and laugh At guilded butterflies, and hear poor rogues Talk of court news; and we'll talk with them too (LRV.3.11-14)

Say

As in ME, say in Shakespeare was basically a transitive verb. In the majority of cases say was used with direct object complements such as quotations, embedded statements, whclauses, pronouns (also in combination with of- and otowf-phrases) and pronoun-like items like as and so: Duke:

And may I say to thee, this pride of hers (Upon advice) hath drawn my love from her, (TGV HI.1.72-73)

Alonso:

Arise, and say how thou cam'st here. (TMPV.1.181)

Othello:

What dost thou say, lago? (OTH III.3.93)

Lucio:

And do you remember what you said of the Duke? (MM V.l.330-31)

Imogen:

Will my lord say so? (CYM 1.6.73)

As far as the passive is concerned, both types of passivization that are possible with say in ME were employed by Shakespeare. The first possibility is to have an impersonal subject, it, the verb say being followed by a that- or 0-clause; the second possibility is to have a personal subject, the verb say being followed by a ίο-infinitive. The latter construction occurs twice as often as the former in the corpus of the modern British plays (see Goossens 1982:109), but was rather rare in Shakespeare:

46

Old Man:

Tis said, they eat each other. (MAC II.4.18)

Porter:

Therefore much drink may be said to be an equivocator with lechery: (MAC II.3.30-32) As far as noun phrases are concerned, the most frequently used with say were grace, prayers, amen, and words, which are still possible with say, and truth, which in ME can only be used with speak and tell. Noun phrases like mind, opinion, and farewell (which, however, occurred only occasionally) are not used with say in present-day English: Evans:

That ere she sleep has thrice her prayers said, (WIV V.5.50)

Gloucester:

To say the truth, so Judas kiss'd his master, (3H6 V.7.33)

Hamlet:

[...] and the lady shall say her mind freely, (HAM II.2.324-25) Whereas in ME the addressee can only be introduced by a ίο-phrase, Shakespeare also had the possibility to use a prepositionless indirect object: Bastard:

If thou hadst said him nay, it had been sin. (JN 1.1.275) A to-phrase after say, however, was more frequent: Gloucester:

I go; and if you plead as well for them As I can say nay to thee for myself (R3 HI.7.52-53) Say to was also used in imperatives, where we would use tell: Lady Macbeth:

Say to the King, I would attend his leisure For a few words. (MAC IH.2.3-4)

Polixenes:

Say to me, when saw'st thou the Prince Florizel, my son? (WT IV.2.25-26) There are a few passages that indicate that say, other than in ME, could designate the action of speaking itself, which in ME is typical for speak and talk: Hubert:

Young lad, come forth; I have to say with you. (JNIV.1.8)

Cleopatra:

None but friends: say boldly. (ANTIII.13.47)

Cymbeline:

Prithee say. (CYM V.5.36)

47 King Philip:

When I have said, make answer to us both. (JN H.1.235)

The meaning in the last example is 'when I have finished speaking'; say in this sense is used in the past tense, the present perfect, and in the past perfect (see Oxford English Dictionary s.v.

"say" vb. 3e). Examples of the kind quoted above, however, did not often occur, so that one can probably claim that the uses of say in EME and ME do not differ as much as the uses of speak. 2.3

Tell

Tell in Shakespeare's works was used in much the same way as it is used in ME. It could be followed by direct quotations, that- and κΆ-clauses, pronouns and pronoun-like items like so and as, and on- and o/-phrases: Fortinbras:

Tell him that by his license Fortinbras Craves the conveyance of a promis'd march (HAM IV.4.2-3)

lago:

Emilia, run you to the citadel, And tell my lord and lady what hath happ'd. (OTH V.l. 126-27)

Ford:

When I have told you that, I have told you all. (WIV II.2.220-21)

Caliban:

Why, as I told thee, 'tis a custom with him Γ th' afternoon to sleep. (TMP HI.2.87-88)

King:

You told us of some suit, what is't, Laertes? (HAM 1.2.43)

As in ME, tell was used with noun phrases like truth and lie, with noun phrases that present the contents in a condensed way like news, things, and words, and - when tell was used in the sense of 'narrate' - with tale or story: Warwick:

Father, the Duke hath told the truth; (2H6 II.2.28)

Eros:

My lord desires you presently; my news I might have told hereafter. (ANT HI.5.21-22)

Falstaff:

I'll tell you strange things of this knave Ford, (WIV V.l.27)

Macbeth:

It is a tale Told by an idiot, full of sound and fury, Signifying nothing. (MAC V.5.26-28)

48 Like speak, tell could be followed directly by noun phrases where ME would require an extension in the form of a prepositional phrase (e.g. about my disease) or a wA-clause (e.g. what my disease is): Plantagenet:

I'll tell thee my disease. (1H6 II.5.44)

King:

Good fellow, tell us here the circumstance, (2H6 II.1.72)

King Lewis:

Be plain, Queen Margaret, and tell thy grief; (3H6 HI.3.19)

As far as the passive is concerned, Shakespeare employed both the personal and the impersonal passive. The latter construction, which is not possible in ME, was more frequent in Shakespeare: Constable:

I was told that by one that knows him better than you. (H5 III.7.104)

Polonius:

Tis told me, he hath very oft of late Given private time to you, (HAM 1.3.91-92)

Shepherd:

So, let's see - it was told me I should be rich by the fairies. (WTIII.3.117-18)

There are certain differences between ME and EME as far as the presentation of the addressee is concerned. In ME the indirect object is normally a prepositionless object that is positioned before the direct object. Only exceptionally (for reasons of emphasis, or when the direct object is a pronoun, for example) is the indirect object placed after the direct object, and only in that case is the addressee introduced by means of a to-phrase. This was also possible in Shakespeare: Clifford:

Come, cousin, let us tell the Queen these news. (3H6 1.1.182)

Bates:

He hath not told his thought to the King? (H5 IV.1.99)

In addition, Shakespeare also had the possibility to place a prepositional indirect object before the direct object, and to place a prepositionless indirect object after the direct object: Prince:

Marry, I tell thee it is not meet that I should be sad, now my father is sick, albeit I could tell to thee - as to one it pleases me, for fault of a better, to call my friend -1 could be sad, and sad indeed too. (2H4 II.2.39-43)

49

Prince:

Heard he the good news yet? Tell it him. (2H4IV.5.11-12) The first passage is in prose, so it cannot be said that the to-phrase was required for metrical reasons. The use of to was probably decided on for reasons of emphasis: there is nobody else the Prince could tell about his state of mind. That Shakespeare wrote both tell thee and tell to thee + 0-clause might be considered by modern readers to be inconsistent, but this usage only proves that in EME standards of consistency or correctness that would be comparable to our own had not yet evolved. Whereas in ME the addressee is usually explicitly mentioned, tell in Shakespeare occurred fairly frequently without one: Paris:

Witness the process of your speech, wherein You told how Diomed, a whole week by days, Did haunt you in the field. (TROIV.1.9-11)

Warwick:

Oxford, how haps it in this smooth discourse You told not how Henry the Sixt hath lost All that which Henry the Fift had gotten? (3H6 III.3.88-90)

Othello:

Now he tells how she pluck'd him to my chamber. (OTH IV.l.141-42) In such contexts we would either need an indirect object, or use a different verb, such as say or relate. Generally speaking, however, the use of tell in Shakespeare did not differ very much from its use in ME.

3.

Conclusion

The comparison of speak in Shakespeare's works with speak in ME on the one hand, and the brief analysis of talk, say, and tell in Shakespeare on the other, have shown that speak is the verb that has undergone the most radical changes since the EME period. Whereas ME speak is basically an intransitive verb, it was used both intransitively and transitively by Shakespeare. In its intransitive use EME speak often corresponds to ME talk: it was frequently employed when people were conversing with each other, when the topic was dealt with in detail, and when linguistic action took place for some time. Used transitively, EME speak often corresponds to ME say or tell: it was used when the contents of an utterance were reported or referred to, that is, when the direct object complement was a direct quotation, a that- or >vA-clause, or a pronoun. Moreover, speak could be followed by far more noun phrases than it can in ME: in such contexts we would either often need an extension in the

50 form of a preposition, or have to use different linguistic action verbs or even non-linguistic action verbs. In contrast to speak, the meanings and usage of talk, say, and tell have not changed significantly. Talk implied linguistic interaction; it often included a mention of the topic, and used without a complement, it often denoted longer linguistic action. Say and tell were in most cases used transitively, i.e. with direct quotations, that- and whclauses, and pronouns. Minor differences between ME and EME pertain to the presentation of the addressee and to passive constructions. More significant differences are that say could also be used intransitively, that is, in contexts where speak and talk would be employed in ME; and that tell was not infrequently followed by noun phrases where we would need a prepositional phrase.

Bibliography 1. Primary works The Riverside Shakespeare (1974) ed. G. Blakemore Evans - Boston. - ADO: Much Ado About Nothing - ANT: Antony and Cleopatra - AWW: All's Well that Ends Well - AYL: As You Like It - CYM: Cymbeline - 2H4: King Henry the Fourth, Part Two - H5: King Henry the Fifth - 1H6: King Henry the Sixth, Part One - 2H6: King Henry the Sixth, Part Two - 3H6: King Henry the Sixth, Part Three - H8: King Henry the Eighth - HAM: Hamlet - JN: King John - LLL: Love's Labour's Lost - LR: King Lear - MAC: Macbeth - MM: Measure for Measure - MV: The Merchant of Venice - OTH: Othello - R2: King Richard the Second - R3: King Richard the Third - ROM: Romeo and Juliet - ΉΜ: Timon of Athens - TMP: The Tempest - TN: Twelfth Night - TRO: Troilus and Cressida - WIV: The Merry Wives of Windsor - WT: The Winter's Tale

51

2. Secondary Works Dirven, R., L. Goossens, Y. Putseys & E. Vorlat (eds.) (1982): The Scene of Linguistic Action and its Perspectivization by speak, talk, say and tell. - Amsterdam/Philadelphia. Dirven, R. (1982): "Talk1: Linguistic Action Perspectivized as Discourse". - In: R. Dirven et al. (eds.): The Scene of Linguistic Action and its Perspectivization by speak, talk, say and tell (Amsterdam/Philadelphia) 37-83. Goossens, L. (1982): "'Say': Focus on the Message". In: R. Dirven et al. (eds.): The Scene of Linguistic Action and its Perspectivization by speak, talk, say and tell (Amsterdam/Philadelphia) 85-131. The Oxford English Dictionary (1933) ed. by James A. H. Murray et al. - Oxford. The Oxford English Dictionary (1972-86) Supplement I-IV, ed. by R. W. Burchfield. - Oxford. Putseys, Y. (1982): "Aspects of the Linguistic Action Scene with Tell'". In: R. Dirven et al. (eds.): The Scene of Linguistic Action and its Perspectivization by speak, talk, say and tell (Amsterdam/Philadelphia) 133-63. Schmidt, A. (1874; 51962): Shakespeare-Lexicon. - Berlin. Spevack, M. (1969; ^1974): The Harvard Concordance to Shakespeare. - Cambridge, Mass. Vorlat, E. (1982): "Framing the Scene of Linguistic Action by Means of 'Speak'". - In: R. Dirven et al. (eds.): The Scene of Linguistic Action and its Perspectivization by speak, talk, say and tell (Amsterdam/Philadelphia) 9-35.

53

Gerd Fritz

Remarks on the Structure of the Verbal Complex in Early 17th Century German 1.

Introduction

Speakers of various Southern German dialects may be heard to use two syntactic variants of subordinate clauses which are represented by the following Swabian examples: (1) daß er den net will komme lasse (2) daß er den net komme lasse will Of these two variants of the three-element verbal complex, only the non-dialect counterpart of (2) is accepted as standard modern written German: (3) daß er ihn nicht kommen lassen will In earlier periods of the German language, however, both variants were used by authors of written texts. Simple facts like these raise the question as to how the grammatical norms of the standard written variety of German came into being and what the earlier stages of this standardization process were. This question has been on the agenda of studies in the history of the German language for some time, but still many details of the emerging picture have to be filled in. One of the media that are considered to have had great influence on the development of a supra-regional variant of written German are the periodical newspapers which began to appear in the early 17th century and which reached an increasingly large public in the course of the century. The present paper deals with the structure of three-element verbal complexes of the type represented in the above examples as they are to be found in the first two complete annual copies of weekly newspapers in Germany, the Relation (R), printed in Strasburg, and the Aviso (A), probably printed in Wolfenbüttel, both issued in the year 1609. This paper is meant to serve a double purpose. It is a contribution to the description of the language of the early newspapers and it is a supplement to the valuable work of Hard (1981) who analysed a remarkable cross-section of text types from 1450 to the present without, however, using early newspapers as a source of data. This source provides interesting evidence for a type of text which represents an intermediate level of formality beetween private correspondence on the one hand and literary works and chancery texts on the other.1 The work presented in this paper is part of a project on the language of the early German newspapers which is supported by the DFG.

54

2.

Three-element verbal complexes in subordinate clauses

2.1

Position of the finite verb

2.1.1 A survey In order to provide some background to the special object of inquiry of this paper I shall mention some basic facts about the position of finite elements in subordinate clauses with overt complementizers. By the end of the 16th century clause-final position of a single finite verb is virtually obligatory in written German (cf. Ebert 1986: 105ff.). This applies also to newspaper texts. As far as two-element verbal complexes are concerned it has been shown by a number of authors, including recent work by Ebert, that the sequence non-finite verb form followed by finite form in clause-final position had become more or less the standard sequence in written texts of any kind by the end of the 16th century.2 This result is confirmed by our data. In the whole corpus of the two newspapers (approximately 200,000 words) there are merely six instances of the inverse order (finite + non-finite), five of them modal verbs + infinitive: soll verlieren (R 210,33), sollen haben (A 71,7; R 143,10), sollen betzalen (A 43,15), woll zu stücken hawen (R 193,21), darvon sind kommen (R 189,1). There is no indication of a particular dialectal distribution of these items in our texts, so that there is not much more to be said than that this alternative is still available, probably as a marked option but not ungrammatical. Against this background the data for three-element verbal complexes come as a surprise. As for the position of the finite verb within the cluster, all three possible options are represented: (4) daß sie für vnüberwindlich solle gehalten werden (A 267,19) (5) daß das Defension Wesen jetzo eingestellt werden solle (A 207,6) (6) vnd da er von jhn vberwunden soll werden (R 178,26) For ease of reference I shall in the following paragraphs use 'type 1' for the sequence finite + non-finite + non-finite (cf. (4)), 'type 2' for the sequence non-finite + non-finite + finite (cf. (5)) and 'type 3' for the sequence non-finite + finite + non-finite (cf. (6)). I shall furthermore use Bech's notation for the degree of dependency within the verbal complex, e.g. V^Vj for a structure like liegen bleiben läßt (cf. Been 1955: 63ff.). In those cases where it is useful to formulate the relationships between structures in terms of rules, I shall for practical purposes assume that the basic position of the finite verb is clause-final. This is not meant to prejudice

2

Cf. Hard (1981:18ff.), Ebert (1981).

55

the question whether one is in fact justified in making this assumption for Early Modern German. With respect to the kinds of verbs which appear in finite form the relevant verbal complexes can be separated into two major groups, with either auxiliary verbs (haben, sein, werden) or modal verbs in finite form. Within these groups there are further sub-groups which are characterized by the types and combinations of the non-finite forms represented (infinitives, participles'). Some of these sub-groups will be dealt with in later sections of this paper. In quantitative terms the verbal complexes with a finite modal verb form the most important group. The following table shows the number of verbal complexes listed according to the type of position of the finite verb and the respective newspaper from which the items are taken.

(7)

verb

newspaper

typel

type 2

type 3

(0

dörffen

A R

.

(u)

können

A R

(iii)

mögen

(iv)

müssen

total

2

1

3

2 2

7 6

_

9 8

A R

10 23

12 8

1 1

23 32

A R

1 2

.

1 2

(v)

sollen

A R

33 62

52 32

1 3

86 97

(vi)

wollen

A R

2 4

16 2

1

18 7

A R

49 91

89 51

3 5

141 147

total

A first look at these statistics shows at least two interesting facts. Apart from the verbs dörffen and müssen, where the numbers are just too small to yield significant results, both newspapers contain considerable numbers of both type 1 and type 2 structures, with an additional small number of type 3 examples. The proportion of type 1 items vs. type 2 items is significantly different in A and R, most obviously in the case of sollen, where R shows nearly two thirds of type 1 vs. one third of type 2, whereas A shows nearly the inverse proportion.

56

Before we go into a detailed discussion of these data I shall present the corresponding figures for auxiliary verbs in table (8).

(8)

verb

newspaper

(0

sein

A R

1 1

1 1

(u)

haben

A R

10 7

-

(iii)

werden

A R

9 6

typel

type tj 2

1

type 3

total

2 2 1

11 7

1 1

11 7

The most striking aspect of this table is the virtual absence of three-element verbal complexes with the auxiliary verb sein and the scarcity of examples with haben. An additional fact, not shown in this table, is that all 4 instances of sein occur in the subjunctive mood as exemplified in (9) and (10): (9) gehandelt worden sey (R 207,3) (10) weren außgeschafft worden (R 32,27) The explanation of the absence of structures with sein is the fact that the passive construction sein + worden + past participle generally appears as a so-called afinite construction, i.e. with ellipsis of the finite form of sein (11). It is probably for reasons of semantic perspicuity that the authors of our texts sometimes prefer to use the explicit finite form if the subjunctive mood is involved. Not to the same extent but still very frequently the constructions with haben are used in an elliptical form (cf. (12)). (11) gefangene Vngern (welche bißhero fleissig geprügelt worden) (A 358,26) (12) daß er nicht recht reden können (R 64,12) As a consequence of these observations, we have to extend our typology and include as type 4 the afinite (elliptical) type of verbal complex. It is, of course, useless to speculate which position the finite verb haben or.sein would have taken up if it had been inserted. The afinite construction simply is a characteristic and frequent type of verbal complex sui generis and as such a typical marker of subordinate clauses in texts of this period.3 Of the approaches generally adopted in accounting for this kind of data three will be used in the following interpretation, viz. the analysis of structural factors, the inspection of dialect factors and the search for stylistic factors.

1

Cf. Admoni (1967: 190f.).

57

2.1.2 Structural factors The first structural factor is of course the syntactic category of the finite verb. Leaving aside the afinite constructions we find an overwhelming preponderence of type 1 constructions both with haben and werden which together with sein are generally considered to be auxiliary verbs. In this group there is no significant difference between the newspapers A and R. So the normal word-order type seems to have been VjV^X^ (cf. (13) and (14)), with the marked options VjVjV! (15) and VjV^ (16) and (17). (13) weil der König eo acte hab schließen müssen (A 87,15) (14) daß man [...] wird kommen können (R 47,27f.) (15) da aber [...] suchen lassen würden (A 88,17) (16) welchen J.M. [...] sein Jus fürbringen werden mögen (A 198,13) (17) welche man dergestalt bewilligen hat wollen (A 301,17) In modern standard written German there is the well-known rule - or, rather, combination of rules - that in the pattern main verb (infinitive) + modal verb (past participle) + haben (finite) the past participle of the modal is infinitivized and the finite verb is moved to the left of the first non-finite satellite of the finite verb.4 Whereas both rules are obligatory in the case of dependent modal verbs (V2), the inversion rule is optional for a number of other verbs, including lassen. There is furthermore a similar optional inversion rule for the pattern double infinitive + finite form of future-werden. Applications of these rules are shown in the following examples: (18) *daß man es sehen gekonnt hat (19) daß man es hat sehen können (20) *daß man ihn kommen gelassen hat (21) daß man ihn kommen lassen hat (22) daß man ihn hat kommen lassen (23) daß man kommen können wird (24) daß man wird kommen können Apart from a number of exceptions which will presently be mentioned, our early 17th century data are basically compatible with these rules. In addition to the double infinitive cases (socalled Ersatzinfinitiv) with modals, the cases of haben (finite) + lassen also show infinitivized participle and inversion of the finite verb: (25) weil etliche Türckische Tartanen sich vmb Corsica haben sehen lassen (R 216,21) There are, however, no items of type 2 attested so that one could conclude that the inversion rule is obligatory, if only in the weaker form that produces V3V,V2. In the case of finite werden the inversion rule is not just optional - which of course it is - but indeed the preferred 1

Cf. Behaghel (1924: 366ff.), Reis (1974), den Besten/Edmondson (1983).

58

choice and, therefore, probably the unmarked option. Contrary to modern written German the optional inversion rule also extends to the finite forms of sein, yielding both the structures V3V2V! (26) and V^V;, (27): (26) daß die Confession [...] auch in die Landtaffel einverleibt worden were (A 66,9) (27) das die Jesuitter daselbst auch weren außgeschafft worden (R 32,27) These facts justify a first generalization: in the case of the auxiliaries the inversion rule has a wider range of application than in modern written German. This generalization can even be extended if we take into consideration finite modal verbs. Quantitatively, the most conspicuous difference in our field of enquiry between present day written texts and our 17th century data is the presence of a high percentage of type 1 clusters, as is amply shown in the table 7. In other words, the optional inversion rule also applies to modals, which, as we already noted, is not the case in modern written German. Among the sub-patterns of this pattern there is one that shows a somewhat extreme quantitative distribution, viz. the combination modal verb (finite) + lassen + infinitive (e.g. wollen passiren lassen). Here we find a notable prevalence of type 2 sequences (e.g. tauffen lassen mögen) in A, as the following table shows: (28)

type 1

type 2

type 3

A 4 25 1 R 8 5 1 It is interesting to note that this combination, Hard's subtype VIb, is considered to be one of

the two patterns which tend to be particularly prone to finite-final position in the course of the 17th century (cf. Hard 1981: 90). Why this is so is hard to say. Hard's explanation for the trend he diagnoses is that modal verbs bear more semantic weight than auxiliaries and, therefore, tend to sentence-final position. One might prefer syntactic parlance and consider the factor to be main-verb likeness (cf. den Besten and Edmondson 1983: 190f.). But still the explanation is not very revealing . Anyway, it is remarkable that Hard's second candidate for an early trend towards finite-final position, his subtype III (finite modal verb + past participle + infinitive: solle bezahlt werden), shows a high percentage of finite-first occurrences in our texts. And as realizations of this sub-pattern comprise the majority of three-element verbal complexes in our data the newspaper texts give the general surface appearance of a wellestablished ^ *^ type.

59

2.1.2. Dialect factors The attribution of certain statistical properties of our texts to factors of dialect calls for a very circumspect and cautious analysis. In the first place it is not obvious to what extent regional correspondents, say from Prague or Antwerp, were in fact natives of those regions and, if so, to what extent they made use of or avoided regionalisms. Secondly, it is hard to determine to what extent the editors and printers actually corrected the incoming newsletters from their correspondents and if they did, to what extent they themselves showed influences of regional dialect. A close examination of all three-element verbal complexes in relation to their source of correspondence shows no significant pattern of distribution. On the other hand, the difference between A and R in the relative frequency of type 1 sequences is so striking that it seems justified to attribute the high percentage of type 1 items to the Alsatian dialect of the editor and/or printers of R. A comparison with texts written by authors from Alsatia, though of a somewhat later period (Moscherosch 1660 and the "Lapis mineralis" from 1681), seems to confirm the hypothesis that a high percentage of type 1 sequences is characteristic of an Alsatian version of 17th century written German (cf. Hard 1981: 75ff.). There is a possible competing factor which should at least be mentioned at this point. There is evidence for the middle of the 17th century that type 1 sequences were considered more formal than type 2 sequences (cf. Hard 1981: 90). If this attitude could be traced back to the beginning of the century we might be landed with an additional factor to account for our statistics. This would go well with the generally more formal, booklike appearance of R as compared to the more informal presentation of A. However, this is rather speculative and we are not on very firm ground if we want to make a final decision as to which of the factors or to what proportion of respective influence of both factors the particular statistical data of R should be attributed. If I had to place a bet I should place it on dialect. As for newspaper A, the statistical data are compatible with the data from other contemporary texts from the Augsburg/Nuremberg region, but they are not specific to such an extent that one could count them as a final linguistic confirmation of the Augsburg or Nuremberg provenience of this newspaper.

It looks as if such a confirmation were not

forthcoming on the basis of any of the linguistic variables. There is a further type of data that can be accounted for by the dialect background of the newspapers, i.e. the type 3 sequences which by the end of the 16th century had become a minority pattern in written texts. There is evidence that by 1609 they were considered strongly dialectal and/or archaic in written German. As far as this type of verbal complex does appear in early 17th century texts, it seems to be limited to Bavarian, Swabian and Alemannic texts. Although the number of items in our data is fairly low (11 items), it seems

60

obvious that newspaper editors or printers did not make a systematic effort to get rid of dialectal forms. 2.1.3 Stylistic factors One possible stylistic factor has already been mentioned in the preceding section, the factor of formality. Two more specific factors shall now be taken into account. The first is the factor of rhythm which has generally been considered an important if sometimes slightly elusive factor. In our texts there is one set of data that arouses the suspicion that some factor of rhythm might be involved. Newspaper A shows a distinct disinclination to use type 1 sequences if the finite modal verb is monosyllabic sol(l) as opposed to bisyllabic solle and sollen. There are 3 occurrences of type 1 with sol(l) as compared to 15 of type 2. Unfortunately, closer inspection does not reveal any correlation to particular rhythmical patterns, so that our suspicion cannot be substantiated. We fare better with another factor, length of constituents. Here there emerges at least one distinctive pattern in which preference for early placement of the finite verb can be shown. In coordinative structures with ellipsis of werden in the first conjunct and ellipsis of the finite verb in the second conjunct the finite verb tends to be placed in the first position of the complex, with the past participle following suit and werden in clause-final position:5 (29)

weil offtmal hohe Potentaten durch böse friedhessige Räth können verleitet vnd in Vngelegenheit geßhret werden (A 67,8) (30) daß gemelter Brief den Ständen ehe nicht solle gefolgt vnd in die Landtaffel einverleibt werden (A 206,35) (31) [...] zuberathschlagen / was gestalt die Stadt Thonowerth / wieder vnter denselben Craiß möchte gebracht l vnnd dem Bayerfürsten aus der Handt gerissen werden (A 77,4) The early placement of finite verb + past participle could be described as a kind of 'exbraciation', which is generally considered a means of increasing comprehensibility in structures with complex complements. So in these cases we have fairly good candidates for a stylistic account of the preference for type 1 sequences. Although the numbers are not very impressive, this kind of stylistic strategy could be seen as a minor stabilizing factor in the preservation of type 1 complexes. It is, incidentally, an interesting fact that these examples form the only group worth mentioning of constructions where an additional constituent is placed within the verbal group. There are two more isolated instances of embraced constituents, in one case a pronoun (32) and in the other an adverb (33): 5

Similar constructions are to be found in R 55,1; R 170,22; A 254,27.

61

(32) weil [...] nicht betten derselben beywohnen können (A 138,29) (33) daß sie [...] weren bißhero abgehalten worden (A 167,8) There is one statistical reason why the number of candidates for constructions with a nominal complement inserted within the verbal group is not very high, viz. the high proportion of passive constructions among the relevant structures. But it is still surprising that we do not find verbal complexes like (34) or (35) which were not almost certainly ungrammatical: (34) daß sie jhre Söhn sollen zu Padua studieren lassen (35) welche dann sollen im Römischen Glauben vnterrichtet werden So the basic principle seems to have been to have the verbal elements in uninterrupted sequence at the end of the subordinate clause. In other words, restructuring processes within the verbal complex did not involve non-verbal constituents to the extent they do in some of Lotscher's (1978) data from modern Southern German dialects. 2.2

The order of the non-finite elements

In the vast majority of cases the sequence of the non-finite elements is V3V2, either in the combination VjV^ (36) or in the combination V^Vj (37): (36) daß sie für vnüberwindlich solle gehalten werden (A 267,19) (37) der sich nicht sehen lassen darf (A 167,5) There is one example of the inverted sequence V2V3, the VjV^ pattern in (38), as opposed to the normal sequence of this type (39): (38) hat können herfür bringen (R 38,29) (39) hat sehen können (A 35,4) The exceptional item in (38) belongs to the category which Hard (1981: 97f.) notes as the only V!V2V3 pattern to appear with any frequency in 17th century texts, the pattern haben + modal verb + main verb with prefix or complement. It may be noted in passing that this particular item is taken from the only correspondence originating from Erfurt. So it is not altogether unlikely that this example might be accounted for by a Northern German regional preference. It is worth noting that this type shows a superficial resemblance to the fourelement type V!V2V4V3 (hat wollen befinden lassen). Maybe this resemblance contributed to the survival of the otherwise isolated pattern. As we have seen, V3 and V2 are normally in adjacent positions. This makes them potential candidates for the reanalysis operation which has recently been advocated by several authors (cf. den Besten and Edmondson 1983, von Stechow and Sternefeld 1988). Things are slightly more complicated for four-element sequences, where an additional cycle of reanalysis is called for. The most difficult pattern to explain is obviously our type 3. A simple solution

62 could be to assume an additional local movement after V3, V2, and V j have become members of one constituent. But maybe this solution is just too simple.

3.

Four-element verbal complexes

Four-element verbal complexes are not very frequent in our corpus. They are, however, numerous enough for us to be able to differentiate four distinct types. The first type (2 items) has the structure ViV4V3V2: (40)

50000. Taler / so dem Orator / zu Ofen / sollten erlegt worden sein (A 60,25f.)

The finite verb is positioned at the beginning of the cluster, and the non-finite elements follow in inverted order of dependency V4V3V2. In modem standard German the corresponding combination has the sequential structure ν4ν3ν2ν!, viz. the inverted order of dependency is carried through: (41) weil der Brief geschrieben worden sein soll The second type, the most frequent pattern (6 items), conforms to the structure VjV2V4V3: (42) haben wollen befinden lassen (A 180,26) Here the finite verb is also in leftmost position with the next element in the order of dependency following suit. Whereas in modern standard German this sequence is only usual for three infinitives as non-finite verbal forms our data show this sequence also in the case of a past participle in the non-finite cluster:6 (43) welche [...] haben sollen gef hrt werden (R 43,10f.) The third type (2 items) has the structure V4V1V3V2: (44)

da [...l nur in beysein der Herrn directoren verlesen hat werden sollen (R 148,6f.) This type is a close relative of the first type, with the single difference that the basic order of non-finite elements V4V3V2 is interrupted by the interposition of Vj. This yields a four element counterpart of the 'compromise' type 3 ("V^VjV^ among the three-element complexes. Finally, there are also afinite constructions in this category as there are in the category of three-element structures. These show basically the same order of non-finite elements as their counterparts with a finite verb, V4V3V2 (5 items, e.g. (45)) and V2V4V3 (2 items, e.g. (46)):

6

Cf. Engel (1988: 446).

63

Finally, there are also afinite constructions in this category as there are in the category of three-element structures. These show basically the same order of non-finite elements as their counterparts with a finite verb, V4V3V2 (5 items, e.g. (45)) and V2V4V3 (2 items, e.g. (46)): (45) (46)

Als sie sich nicht vertreiben lassen wollen (A 101,17) die man [...] nicht vber Rein wollen passim lassen (A 211,10)

In all cases the ellipsis concerns the finite form of haben. To summarize these findings, the finite verb is generally positioned at the beginning of the verbal complex - with one exception where it is inserted after the first non-finite element -, and the non-finite part always contains the sequence V4V3, with V2 either staying in final position or following the finite verb to the beginning of the complex, however in inverted order VjV2. A comparison with the three-element structures shows that now there is no option for the finite verb to stay in sentence-final position and that V4V3, which corresponds to V3V2 in the three-element structures, is the only pattern represented in our data. The optional position of V2 seems to correspond to the optional position of the finite verb in the three-element complexes.

The main difference between this system and modern standard

German is, of course, that modern German allows the finite verb in sentence-final position, at least in those cases where there is no cluster of infinitives.

4.

Concluding remarks

The general impression conveyed by our data is that in our type of written texts from the beginning of the 17th century there is a fairly stable system of type I/type 2 options for all cases of three-element verbal complexes with a finite modal verb. In those cases where the finite verb is an auxiliary the type 2 (or type 3) option is obviously a marked option. As for the four-element complexes, finite-first within the verbal complex is obligatory.

The

remarkable thing about this system is its stability in face of heavy competition from the oneand two-element verbal structures which almost exclusively conform to the principle of clause-final position of the finite verb. Whereas in the development of standard written German the type 1 option for structures with finite modal verbs lost ground in the course of the 18th century, many German dialects have preserved a system of options comparable to the one represented in our data. As the examples quoted in Lötscher (1978) show, some presentday dialects are, in addition, much less restrictive as to the placement of nominal elements within the verbal chain. So the structure of the verbal complex in our data already looks like the product of a standardization process for supra-regional written German. A stabilizing factor for the type 1 option could be seen in a superficial relatedness to structures in main

64 sentences and subordinate clauses without overt complementizer, as in the following sentence types: (47) (48) (49)

Dieselben Soldaten sollen abgedanckt werden man vermut / die Evangel. Bücher möchten colligirt werden Mandata so allhie von newem sotten publicirt werden

From this point of view the V1V3V2 sequence looks like a small brace construction which could be reinforced by its counterpart in main sentences. Admittedly this is not very strong even as a partial explanation. Even less convincing is an explanation in terms of reduction of perceptual load. The comparison of V^X^ with V3V2V! does not present an impressive gain in perspicuity of construction as in the case of the exbraciation of complex nominal constituents in main sentences which is the prototype of a movement of constituents which can be seen to improve comprehensibility. As for the SOV question, I do not think any far-reaching conclusions can be drawn from the data presented in this paper. The fact that the non-finite verbal forms follow the finite verb seems to argue against SOV. On the other hand, the fact that the verbal complex always takes the form of a continuous constituent without interposition of nominal constituents provides the champion of SOV with the argument that it is the whole complex and not just the finite verb that takes up the characteristic verb-final position.

Bibliography Admoni, W.G. (1967): "Der Umfang und die Gestaltungsmittel des Satzes in der deutschen Literatuisprache bis zum Ende des 18. Jahrhunderts." - In: Beiträge zur Geschichte der deutschen Sprache und Literatur (Halle) 89,144-99. Bech, G. (1955): Studien über das deutsche verbum infmitum. Vol. 1. - Det Kongelige Danske Videnskabernes Selskab. Dan. Hist. Filol. Medd. 35.2. Kopenhagen. Behaghel, O. (1924): Deutsche Syntax. Vol. II. - Heidelberg. den Besten, H. & J.A. Edmondson (1983): "The Verbal Complex in Continental West Germanic." - In: W. Abraham (ed.): On the Formal Syntax of the Westgermania (Amsterdam/Philadelphia) 155-216. Ebert, R.P. (1981): "Social and Stylistic Variation in the Order of Auxiliary and Nonfinite Verb in Dependent Clauses in Early New High German." - In: Beiträge zur Geschichte der deutschen Sprache und Literatur 103,204-237. Ebert, R.P. (1986): Deutsche Syntax 1300-1750. - Bern/Frankfurt am Main/New York. Engel, U. (1988): Deutsche Grammatik. - Heidelberg. Härd, J.E. (1981): Studien zur Struktur mehrgliedriger deutscher Nebensatzprädikate. Synchronie. - Göteborg (= Göteborger Germanistische Forschungen 21).

Diachronie und

65 Lötscher, A. (1978): "Zur Verbstellung im Zürichdeutschen und in anderen Varianten des Deutschen." - In: Zeitschrift für Dialektologie und Linguistik 45,1-29. Reis, M. (1974): "Syntaktische Hauptsatzprivilegien und das Problem der deutschen Wortstellung." - In: Zeitschrift für Germanistische Linguistik 2, 299-327. Schöne, W. (1939): Der Aviso des Jahres 1609- Leipzig. Schöne, W. (1940): Die Relation des Jahres 1609. - Leipzig. von Stechow, A. & W. Stemefeld (1988): Bausteine syntaktischen Wissens. - Opladen.

67

Oliver Grannis

Language Change in Modern German: Feminism and Systematic Gap Production 1.

Introduction

It is often the case that language change results in more complex phonological/ semantic/grammatical systems or subsystems than had existed before, thus allowing for increased differentiation within the systems in question. The tense and aspect system of modern English, for example, is considerably more complex than it was five or six hundred years ago, and most people would probably agree that this constitutes a positive change for the language. Sometimes, however, language change results in systems with less potential for differentiation than previously, or with possibly a different sort of differentiation, one in which previously existing possibilities for differentiation are levelled, thus producing what at least one generation will perceive of as a gap in the system in question. This phenomenon is to be referred to in this article as Systematic Gap Production (SGP). We could describe as SGP, for example, the loss of a second person singular-plural distinction in the pronoun system of English as a result of the extensive changes that took place in the late Old English, Middle English, and Early Modern English Periods. The members of a speech community will often recognize the existence of such gaps and try in one way or another to compensate for them, as evidenced, for example, by the suggestions made during the Early Modern English period to differentiate between you is and you are, or, later, the introduction of such dialectal forms as yez, youse or you all. Similarly, in modern German, which has a systematic gap similar to that in English in that the second person singular-plural distinction, although existing in the informal-intimate du (bist) - ihr (seid) pair, has long since been levelled in respect of the unmarked Sie (sind), which functions as both singular and plural. Speakers of modern German often use ihr (seid) to designate unequivocally plurality even when the communication situation is clearly non-informal-intimate. Speakers of both English and German occasionally feel uncomfortable with the systematic gap in the second person pronouns system and attempt in one way or another to close or circumvent it. Whether such attempts to close a systematic gap are acceptable to the speech community at large, or indeed whether the community is willing to tolerate such a gap coming into existence in the first place, is dependent upon a number of sometimes extremely complex factors. The rest of this article will discuss some of these factors, and SGP as one aspect of language change in modern German.

68

2.

Problems of gender reference

The various changes in English resulting from the feminist movement are fairly well known among educated speakers of the Western European languages and have had a considerable influence on the establishment of similarly motivated changes in German. A term such as chairperson, for example, is not only known passively: it is actively used by some bilingual, i.e. English-German speakers in such an expression as Ich möchte gern wissen, wer die nächste Chairperson wird. would like to know who the next chairperson is going to be.' One of the principal reasons for the fairly widespread knowledge or even adoption of such innovations is their convenience as items that can sometimes fill the gaps created by the changing awareness of gender-scope reference in modern German or simply as items that allow the speaker/writer to avoid being too specific in a situation where openness or vagueness is socially preferable. Chairperson, for example, is a convenient gender-neutral form - just as it is in English - that can be used rather than the more definite masculine der Vorsitzende or feminine die Vorsitzende. Similarly, some speakers/writers of German have been using the English-language letter greeting Dear ... for years: it not only conveys an upto-date, cosmopolitan impression, it allows the letter-writer to avoid having to choose between the relatively formal Sehr geehrte(r) ... or the markedly nonformal Liebe(r) .... In some cases an internal shift within a language system obviates the need for borrowing of the kind under discussion. For example, many German-speaking women feel just as strongly as do English-speaking women that compulsory disclosure of marital status by means of a title is unnecessary and discriminating. For those speakers of English who want to make use of it, the neutral form Ms. is available. English Ms., however, has had little effect on German probably for two reasons: (a) it does not fit into the Fr-paradigm of Frau (traditionally a married woman) and Fräulein (traditionally an unmarried woman), and, no doubt more important, (b) the use of Frau has shifted in meaning in the last fifteen or twenty years so that it is now used to refer to any mature woman, irrespective of her marital status. Fräulein is now seldom used except for younger girls, under seventeen or eighteen, and then often in a bantering or slightly ironic way. The use of Fräulein for older, unmarried women is considered extremely conservative and is increasingly rare. Thus, with the shift of Frau to include unmarried women as well, there is no need to introduce Ms. or an equivalent into the German form-of-address system.1

It is interesting to note that virtually all of the modern English-language dictionaries include Frau and Fräulein as entries, but none of them has taken notice of the shift in usage of these forms in the last two decades. Typical is the entry for Frau in Webster's Ninth New Collegiate: "a German married woman: wife used as a title equivalent to Mrs."

69

The changes in the English language resulting from the feminist movement, although extensive, are of a fairly random nature, amounting largely to the substitution of non-genderspecific terms for items considered to be sexist, i.e. discriminating against or offensive to women. The motivation for parallel language changes in German is similar; their scope, however, is considerably broader: because German is a language with preponderantly grammatical rather than, as in modern English, natural gender, language change of the kind under discussion is bound to be systematic and thus wider ranging. Certainly there are in German isolated lexical items/considered to be sexist or offensive, similar to English mankind, chairman, barmaid, girl Friday, etc., but the principal quarrel most feminists and other culture and language-sensitive speakers of German have with the language is the indiscriminate use of masculine nominal and pronominal forms as generic forms, i.e. forms inclusive of both masculine and feminine with neither, however, being specifically marked, for example der Lehrer 'the teacher', der Student 'the student1, or der Kollege 'the colleague'.2 The Generic use of the der form (i.e. any overtly masculine form, whether it has the article der or not) is for many speakers of German too easily interpretable as a basically masculine and thus femaleexcluding form. Although German has been a grammatical-gender language since its Indo-European origins, there are signs that this is changing and that native speakers of the language are increasingly viewing their nominal system as being characterized by a combination of natural and grammatical gender rather than simply the latter. Thus, although there is (still) no objection to the masculine der in such lexical items as der Tisch 'table', or the feminine die in die Lampe 'lamp', the use of the masculine der Lehrer as a generic with possible or definite feminine reference is becoming increasingly unacceptable. Thus far, as critics of feminism are quick to point out, objection is made only to conflicting forms referring to human beings. Thus der (masculine) Fisch 'fish1, die (feminine) Maus 'mouse', and das (neuter) Pferd 'horse' can refer to male or female without the speakers of the language feeling uncomfortable. The recent impetus to change affects largely the masculine and feminine genders, although a very noticeable shift away from the use of neuter grammatical forms for female referents has been evident for several years. A form such as das Mädchen 'girl', for example, although regularly used without objections as a neuter, is increasingly followed by a feminine pronoun (sie) rather than the grammatically 'correct' neuter es. The increasing reluctance to use - or accept - the der form as a non-gender-specific marker has led to the widespread use of 'splitting1, i.e. the use of the der and the die form together, as See Pusch (1984) for a lucid, detailed account of the feminist point of view. Pusch's account of German as a "man's language" continues to exert a very strong influence on those interested in correcting what are felt to be discriminatory features of modern German.

70

in the following extracts from job advertisements in a German newspaper (Die Zeit, 10 Feb. 1989): (a) Leiter/Leiterin 'manager/manageress1 (b) Direktorin/Direktor 'directress/director' (c) ein(e) Lektor(in) 'a male (female) reader (female)' (d) eine Referentini einen Referenten 'a female official/a male official' (e) eineln Musiktherapeutlin 'a female/male music therapist/female' Although not represented in the foregoing selection, other forms are possible and are frequently encountered, e.g.: (f) Leiterin 'managerEss', or even (g) LEITERIN'MANAGERESS'. The multiplicity of forms in (a) through (g) is a natural result of the relative newness of splitting. There is not yet agreement as to what form it should eventually take.3 Obviously, the majority of the forms (a) through (g) are only possible as written forms. The spoken form in all cases would be something like Wir suchen eine Leiterin oder einen Leiter 'We are looking for a directress or a director'. Still to be found, although far less frequent than the patterns in (a) through (g) are job notices of the following kind, in which a masculine-generic form is used and then followed by a splitting construction: (h)

Bei der Kreisstadt Singen ist die Stelle des Leiters der Allgemeinen Verwaltung zu besetzen [...]. Bevorzugt werden Bewerberinnen und Bewerber, die [...] 'in the county seat Singen there is the position of director (masculine-generic) of general administration to be occupied [...]. Preference will be given to applicants (female) and applicants (male) who [...]' Still less frequent are notices of the kind quite common twenty years ago in which the masculine-generic form is used and nowhere in the text can one find an indication of a preference for, or the exclusion of, masculine or feminine applicants, or a statement that both are acceptable, for example a notice for an Arzt 'doctor' in the same newspaper from which (a)

through (h) above were taken. Such notices are for increasing numbers of people particularly irritating, for it is frequently suspected that behind them lies an unstated preference for male applicants, or that the individual or institution responsible for the notice is blatantly ignoring the conventions increasingly accepted by most native speakers and thus making clear their aversion to feminism and perhaps to women as well. 3

Compare the multiplicity of forms for the 3rd person singular possessive neuter pronoun in Early Modern English. When in the course of the shift from grammatical to natural gender the speakers of English found the neuter possessive his - which had served the language well since Old English - to be uncomfortable because of its identity with the masculine possessive Ais, they tried out a variety of forms such as of it, it's, it and its before finally settling on the last, its, towards the end of the seventeenth century.

71 Firms and institutions more sensitive to current social issues will make an explicit statement of their preference within the body of the text, or, if female applicants are preferred, will, of course, indicate this by means of the feminine form: Ein Kindergarten sucht eine Leiterin 'kindergarten seeks head (female)1. One of the features of the German nominal system that is frequently ignored by the participants in the current debate is that it has never had an exclusively masculine form, as it has an exclusively feminine form. Splitting is a phenomenon to be found in virtually every form of public or institutional communication. For example, in a university most letters and notices not addressed to a specific individual begin with a formula such as Sehr geehrte Kolleginnen und Kollegen Very honoured colleagues (female) and colleagues (male)', and students are addressed as Studentinnen und Studenten. A tendency to carry splitting to what many feel are stylistically uncomfortable if not absurd extremes is to be found in legislative and public administration documents of all kinds, for example in the following extract from the Rundfunkgesetz ('Public Broadcasting Statutes') of Northrhine-Westfalia, Germany, Die Rundfunkkommission wählt aus der Mitte ihrer Mitglieder eine/n Vorsitzende/n. Der/die Vorsitzende erhält eine Entschädigung in doppelter, sein/e ihr/e Stellvertreter/in in eineinhalbfacher Höhe. (par. 53, quoted in Der Spiegel 7,1989) Splitting is being used increasingly in other situations as well when the author wishes to avoid giving offense. Although, for example, the second-person-plural pronominal form Ihr-Euch is acknowledged to be generally appropriate for reference to any group including at least one member with whom the speaker or writer is on familiar terms (i.e. per du), one frequently encounters usages such as the following, intended to avoid insulting anyone with whom the writer is on non-familiar or formal terms (i.e. per Sie), and who could possibly feel offended by the use of the inclusive Ihr-Euch form: Falls Sie/Ihr sich/Euch überlegen/überlegt, Gebrauch von Ihrem/Eurem Wahlrecht zu machen, dann bitte ich Sie/Euch, mir dies demnächst mitzuteilen. 'In case you are considering making use of your right to vote, then please inform me of this soon'. Forms such as the following masculine-generic plurals used as greeting formulas in letters written approximately fifteen years ago indicate a tacit acceptance of the der form as a form frequently unmarked in respect of gender: Sehr geehrte Kollegen an die Studenten im Fachbereich 3 an alle Fremdsprachenlehrer

"Dear esteemed colleagues' 'to the students of..." 'to all foreign language teachers'

In similar correspondence from 1988 and 1989, one finds either splitting - Kolleginnen und Kollegen or Studentinnen und Studenten - or gender-neutral forms that allow for neither a spe-

72

cifically masculine nor a feminine interpretation, e.g. an die Lehrenden im Fachbereich 7, or an die Studierenden 'to those who teach in Faculty 7, to those who study'. The noticeable recent increase in the use of such gender-neutral expressions can be seen as an attempt to avoid the awkward splitting forms. Perhaps the most striking result of splitting is the fact that the der form, for centuries a form that could be interpreted either as gender neutral or as masculine, is in fact being interpreted by increasing numbers of German speakers - not only feminists - as exclusively or predominantly masculine. The frequent juxtaposition of the clearly feminine die form with the masculine and, until recently, neutral der form is, despite the protests of the traditionalists, in the process of bringing about a clear shift to a gender system without a neutral form for nouns referring to persons. An issue of the German newsmagazine, Der Spiegel (45,1986) featured a cover picture of Doris Dörrie, the German cinema directress, with the words: Deutschlands erfolgreichste Regisseurin 'Germany's most successful directress'. Doris Dome is, of course, a woman, and the editors of Der Spiegel decided on the feminine form Regisseurin to describe her rather than the masculine/neutral Regisseur. Had they chosen the latter, many readers would no doubt have protested. The problem is, however, that the superlative 'most successful1 can only refer to female directors, since Regisseurin is exclusively feminine. Strictly interpreted, the phrase can only mean 'the most successful of all female directors' and is thus, at best, misleading. It is clear from the text of the article itself, however, that Doris Dörrie is being compared with all the cinema directors in Germany, male or female, and is considered to be among them the most successful. The opposite interpretation is intended, however, for the following, again the title of a magazine article: Steffi Graf - Deutschlands beste Tennisspielerin 'Steffi Graf - Germany's best (female) tennis player'. In this case the scope of the superlative is intended to be limited to female players only. The interpretation that Steffi Graf is being compared with all German tennis players, male as well as female, would probably be rejected not only by the grammarians but by most tennis fans as .well. The real problem is made evident by the Doris Dörrie example, however: that which until recently was a neutral or masculine form - in this case der Regisseur - can no longer be used without the danger, indeed probability, of its being understood as exclusively male. Thus a not insignificant gap has appeared in the system of grammatical gender.

3.

Avoidance strategies

The following four sentences, (a), (b), (c) and (d), although (still) acceptable to some, are already problematical for others:

73

(a)

Meine Mutter war der erste Arzt in diesem Dorf. 'My mother was the first doctor (masculine-generic) in this village.' (b) Wer ist der beste Koch in eurer Familie ? 'Who is the best cook (masculine-generic) in your family?' (c) Welcher Richter würde so eine Geschichte glauben l 'What judge (masculinegeneric) would believe a story like that?' (d) Ich sehe dort hinten am Horizont einen Reiter. see a (horseback) rider (masculine-generic) back there on the horizon.' Many would claim that (a) is simply impossible or, at best, ungrammatical, that the question (b) is clearly weighted towards the male members of the family, that (c) can refer only to male judges, and that (d) is impossible because no one can identify a horseback rider as male at such a distance. Those who categorically object to the use of a der form as gender-neutral suggest the use of such structures as the following to avoid having to produce such sentences as the foregoing (a) through (d):4 (aa)

Meine Mutter hat die erste Arztpraxis in diesem Dorf 1eröffnet. 'My mother opened the first doctor's office/surgery in this village. (bb) Wer kann in eurer Familie am besten kochen? 'Who can cook best in your family?' (cc) Welche Richterin oder welcher Richter würde so eine Geschichte glauben ? 'What female judge or what male judge would believe such a story?' Note that the nominally masculine-generic form Richter is here given a clearly masculine interpretation by virtue of its juxtaposition with the feminine form Richterin. (dd) Jemand 1reitet dort hinten am Horizont. 'Somebody is riding back there on the horizon. Such 'avoidance strategies' are increasingly common, used not only by those who actively object to a gender-neutral usage of der and related forms, but perhaps just as often by those who want to avoid conflict situations. Particularly the latter tend to make use of what can be called 'token mention1 - an initial splitting, to show an awareness of the problem, followed then by one or more instances of the masculine-generic form. The first of the following two examples is from a newspaper article (Oldenburgische Volkszeitung, 11 March 1989), the second from a letter to the teaching staff of a large university faculty:

"* See Hellinger et al. (1985) for an extensive catalogue of suggestions for emending and/or avoiding sexist expressions in German. Similar, unpublished suggestions for English in a 'policy document on non-sexist language' have been put together by Radio New Zealand to avoid "sexist language [...] in all its communication, be it on air, in written material or in everyday contact with clients." (Since writing this article in June, 1989, I have become aware of the fact that many newspapers and other publications in both Germany and the Engisb-speaking world have begun formulating editorial guidelines, some precise, others rather vague, designed to help their writers avoid what are felt to be sexist (mis)usages of language. In addition, increasing numbers of dictionaries are exhibiting a growing awareness of the problems involved, most strikingly perhaps those dictionaries.

74 (a)

(b)

Der Verwaltungskommission gehören sieben Hochschullehrerlinnen, zwei wissenschaftliche Mitarbeiter, zwei Vertreter der Studentenschaft und zwei Vertreter des nichtwissenschaftlichen Personals an. To the administrative committee belong seven universitiy teachers (masculine/feminine), two academic colleagues (masculine-generic), two student representatives (masculine-generic), and two representatives (masculine-generic) of the non-academic personnel.1 Jede und jeder wird schon wissen, was er für den Kranken tun kann. 'Everyone (feminine) and everyone (masculine) will surely know what he (masculine-generic) can do for the sick person (masculine).'

Occasionally the order is reversed: following several instances of masculine-generic forms, come one or two occurrences of splitting. For obvious reasons of economy, splitting is avoided in newspaper headlines; token mention comes then perforce in the body of the article. A somewhat different problem is raised by the use of the indefinite pronoun, man, in many ways equivalent to English one, as in: Man benimmt sich nicht so, wenn man eingeladen ist. One doesn't behave like that when one is invited out.1 Because of the similarity in orthography and pronunciation to Mann 'man1 and, of course, to English 'man', the German indefinite pronoun man is being increasingly regarded as a sexist expression to be avoided or to be used only in contrast to a feminine form, for example frau, a lowercase form recently introduced as a form supposedly parallel to man. Generally speaking, linguists have no right to expect users of a language to be aware of the etymological origins or structural properties of the expressions they use. The fact that man, as noted in Kluge (1975), for example, as well as in Duden (Drosdowski 1989) originally meant something like irgendein Mensch 'any human being' is unknown and unimportant to the average user of German. Similarly, we cannot expect the average user to be influenced by the fact that man is an indefinite pronoun and Mann a noun, that the indefinite pronoun man, unlike Mann, never occurs with an article, that it is only used as the subject of a sentence and not as object, and that it never occurs as a predicate noun. If for one reason or another the speakers of a language feel uncomfortable with a word or construction, they will avoid it or alter it so as to conform to their sensibilities. Thus there is a tendency in modern German to make increasing use of passives in place of man construction, as in the (a)-(b) pairs below: (a) (b) (a) (b)

Man macht das hier nicht. One doesn't do that here.' Das wird hier nicht gemacht. That isn't done here.' Man sagt, er sei krank gewesen. They say he has been ill.' Es wurde gesagt, er sei krank gewesen. 'It is said he has been ill.'

Where reference is clearly to a woman, man is being used less and less frequently, so that (b) would for many people be preferable to (a):

75 (a)

Will man eine Stelle als Krankenschwester finden ... 'If one wants to find a position as a nurse ..." (b) Will eine Frau ... 'If a woman ..." or, increasingly frequent, Willfrau... 'If 'woman'...' Although frau as an alternative to man is still used only in a fairly self-conscious way, man itself is increasingly being avoided, particularly in writing, where, of course, the style-consciousness level is considerably higher. Other than the increased use of passives as an avoidance strategy, no clear alternative to man is evident. The quasi-coinage frau cannot be seen as a genuine alternative, since frau is even less of a gender-neutral form than man. It is noteworthy that the introduction of frau is apparently in the process of bringing about a similar gap in the pronominal system as was noted with the gender-neutral use of der forms: the feminine-excluding, masculine interpretation of man is becoming dominant while the generic interpretation is disappearing.

4.

Conclusion

In summary, it can be said that language change by no means always leads to a more logical or more complete system than had existed before. Language change has a rationale of its own, with various social pressures being ultimately stronger than the internal structural requirements of a given language. Systematic Gap Production is thus a normal phenomenon to be expected wherever languages are in a process of change.

Bibliography Drosdowski, G. (ed.) (1989): Duden Etymologie: Herkunftswörterbuch der deutschen Sprache. - Mannheim. Hellinger, M., M. Krämer & B. Schräpel (1985): Empfehlungen zur Vermeidung von sexistischem Sprachgebrauch. - Universität Hannover. Kluge, F. (1975): Etymologisches Wörterbuch der deutschen Sprache. - Berlin. Mish, F.C. (ed.) (1983): Webster's Ninth New Collegiate Dictionary. - Springfield. Pusch, L.F. (1984): Das Deutsche als Männersprache. - Frankfurt

77

Fred W. Householder

Pronunciation Rules and Improper Sound Change 1.

Written and oral words

It is a seldom mentioned fact that English vocabulary is not all phonological strings (which may be written down) any more than it is all sequences of printed letters, which may be pronounced. Some words (type 1) are indeed transmitted in a continuous unbroken oral chain, learned by hearing at each link (perhaps by hearing ones parent read, in some cases); but there are others (type 2), of two kinds. The majority are written strings for which at least a mental pronunciation must be devised by the reader,1 since letters without sound are impossible for most of us (except perhaps for the congenitally deaf), and the second type, encountered mainly in a classroom setting, are words or names which are presented simultaneously in written and oral form, or pronounced and then spelled out orally and pronounced again (ell, e, a, dee, [led]). But both kinds are basically non-oral, and will not hereafter be kept apart. Almost all proper names (except a few common first names and the names of people and places of your childhood), belong to these non-oral kinds of words. However, very famous names (like Franklin Delano Roosevelt) may become virtual type 1 items until a later generation says [ddla:nowJ and [ruwzdvelt] - though the latter was okay for Teddy. For the rest of this paper we will refer to the orally transmitted strings as type 1, the others as type 2. Within type 1 there are variations for individual speakers; I may have learnt something at my mother's knee which you only acquired from reading. So a friend of one of my daughters said that "So long" a common and early form of farewell for me, was for him a pure book word. This belongs to a category which we may call "slang and colloquial". A recent example of this is the word lambaste [tembeyst], a common oral expression of my youth, which has now regained circulation as lambast [Ismbaest], obviously by back-formation from the printed past tense lambasted. Conceivably this may now become reestablished as type 1, giving then the specious appearance of a lexically diffused sound change from [ey] to [ae]. Other categories offering the same possibilities are, e.g. technical vocabularies which are oral and early for some, late and written for others: I acquired many farming technical terms (e.g. harness parts) on the farm from my father, and didn't see them in print till years later, but There are also words (especially names) for which the hearer must construct spellings - which he may later mispronounce.

78

most people today either don't know the words at all, or only from printed sources. How many people nowadays know that the 'shafts' of a buggy or rake are [aaevz], as I only learned late that [aaevz] was spelled shafts?2 There does not need to be any generational difference here, as is often the case with slang. Many dialect words are type 1 in the home area, but type 2 elsewhere, or type 1 in a particular social class, type 2 for everyone else; and, in some cases, dictionaries treat a type 2 pronunciation as 'correct' or 'preferred', even though linguistic logic hardly supports them. It is true that writers like Kenyon and Knott (1953) do mention "Spelling Pronunciation" (Kenyon and Knott (1953:xlv-vi), Kenyon (1960:xxiv-xxv)), but in such a way as to suggest that only a small percentage of words are liable to its influence. Kenyon indeed says "a large actual number, though a small percentage", but also says "phonetic change is primarily concerned with the spoken language" i.e. type 1 words, while patently exaggerating the size of the type 1 vocabulary and underestimating type 2. There is one situation in which 'spelling pronunciations' of type 1 words and affixes are fairly common, and that is song. Many singers render the past tense -ed as [ed] in words like tested or hinted, and similarly plural -es as [εζ] e.g. in glasses. Finally, it is only type 1 words which undergo phonological changes (or true neogrammarian style phonological rules)3, and to count as real sound changes they must be type 1 for most of the community; 'changes' in the pronunciation of other words are like the case of lambaste just mentioned, and resemble changes in the modern pronunciation of the written words of a dead language (for students of Latin, Greek, Sanskrit, Gothic, etc.).

Even

technical type 1 words may succumb to spelling-pronunciation; one often hears buoy pronounced [buwiy] even by people who have worked around them from childhood.

2.

Variant pronunciations

As only type 1 words undergo true phonological change, so the only true dialect variant pronunciations are of type 1 words. When one person says aren't in two syllables, while another says it in one, you know they speak different dialects: only Easterners from the New Many dictionaries, even OED and Webster Π fail to mention this, and Wentworth (1944) calls it "rural", citing it from New York, Indiana, Missoury, Virginia, and Maryland. I never heard any other pronunciation than [sasvz] during my boyhood in Rhode Island, Massachusetts or Vermont And "rural" must here have the sense of my type 1, since it is not (like true dialect words) peculiar to any section of the country (or even to the U.S. the English Dialect Dictionary has it under sharf- with silent R - and shaves from 11 different counties). At least Webster III gives it recognition as a normal form. I must confess that I myself have often written as if type 2 segmental variants and changes were phonological variants and sound changes; they are not, and I apologize. This confusion, however, continues, as in Harris, Labov and Wang - and many more. Type 2 words may, however, give evidence for phonotactics.

79

York area are likely to accept aren't you smart as good pun for orange juice mart, and only people from Canada and a few other areas (mainly in Virginia or Martha's Vineyard) are going to centralize the vowels in white house and the like. But if one man says economics with initial [iy] and another with [ε], you learn nothing about their dialect areas. No [iy] in modern dialects is undergoing phonological change to [ε], nor is the reverse happening anywhere. And the great mass of variant pronunciations listed in any dictionary, including the 1100 (or so) words listed on pp lix-lxxvii of Webster II are type 2 words. (Exceptions there are been, sometimes either and neither, fort, garage, gouge, the verb grease, and the adjective greasy, hog, jeans, okra, often, paunch, pecan, roam, route, soot, stirrup, sumac, tomato, toward, were, year, yesterday, and yours, most of which are often on the borderline between one and two.) A large number of the variant pronunciations in any dictionary involve variations in position of the accent (accompanied, automatically, by variations in stressed and weak vowels); no one has suggested a rational process by which this could take place in a natural type 1 language. For instance, banal varies between [beyndl] and [bdnael] but anal [eyndl] does not vary, and neither does canal [kdnael]. We should not, of course, underestimate the ability of speakers to maintain a completely full voweled underlying form, as in languages like Tubatulabal, for instance, but I do not think many illiterate English speakers have very many such base forms in their heads. The second common type of variant is due to ambiguous letters; is g for [g] or [J]? Is ο for [o], [»] or [ow] - or [Λ]? Should stressed y follow the rules for i, or is it always [ay]? Again, some of these variants resemble dialect variants in commanding local loyalties, but many are quite unpredictable. There are, of course, some rules (most of them clearly stated by Kenyon 1960) such as: "g before a is always [g] except in margarine and (British) gaol." But, if we examine the type 2 variants (excluding variations in weak vowels and in palatalization) among the 1100 mentioned above in Webster II, we may note the following frequencies: (1) 'short-long' vowel variation, i.e. [ae] vs. [ey], [ε] vs. [iy], [I] vs. [ay], h>] or [3] vs. [ow] (rarely [Λ] vs. [yuw] or [u] vs. [uw]) is involved in at least 289 of the 1100 words; (2) the position of the stressed vowel or variation between a weak and a secondarily stressed vowel accounts for more than 226, not counting a number of frequent British-American differences in words ending in -tory, -tary, and a few other suffixes; (3) variation in voicing in words written with s and χ (rarely ss or AC), yielding [s] vs. [z] and [a] vs. [έ] accounts for 69 words (this includes one word spelled with -ti-, equation, and one with -ce-, sacrifice; this is the most frequently occurring consonantal variation; (4) other variations are less common; especially important is variation between normal English values and others felt to be 'French1 or 'foreign' and somehow classier, e.g. [ay] vs. [iy] or [I] vs. [iy] for i, [ey] or [ae] vs. [a:] for

80 a, [iy] or [ε] vs. [ey] or [0] for e, and, for -ge or; [fl vs. [ζ], or ch (or -ca'-) as [2] vs. [§]; in addition, 'silent' vowels and consonants belong here, in the main, silent consonants and silent e being Frenchy and classy, which leads to some uncertainty: one announcer says [triystey] for French triste, but another (even Johnny Carson) says [kuw do gra:] for coup de grace. About 74 words fit in this category. (5) A few words involve expanding a more general rule at the expense of a more restricted one, e.g. [ ] or [ ] or [ow] vs. [Λ] (e.g. in hover), [ae] instead of [o] for a before /, [D] or [D] instead of [ow] for ο before /, and various options before r. All these amount to 41. (6) There are, among the 1100, three sets of words showing variations which were probably once true type 1 dialect variants: variation between [uw] and [ow] for ou (rarely ο or oo), between [a] or [ae] and [3], mainly for au (esp. before n), but sometimes a or al, and variation in voicing [Θ] vs [0] for th - all together 36 words. Every other variation is restricted to a much smaller number of words. Between June 9, 1988 and June 8, 1989 I jotted down notable variants overheard on radio or television, totaling about 330 words. If we sort them into the categories we have just listed, the results are generally the same, though one or two new categories do appear. (1) The 'short-long' variation involves 62 words, about a fifth of the group. (2) There are 60 words which show variations in accentuation, again about one fifth of the total. (3) Variation in voicing of s and χ (as well as ci or ft) occurs 17 times. Here we can detect two motivating factors: sometimes voicelessness ([s]) seems to be more elegant medially e.g. in words like Joseph, diesel or crimson, (this may possibly derive from the British pronunciation, e.g. in all words ending in -sia, -sian, -sion), but voicedness ([z]) tends to occur before the stress, e.g. in Isuzu (pronounced [iyzuwsuw]), resources, or emphysema. In a few words [z] apparently seems more Frenchy, as in laissez faire [lezey fer] or rigosum. (4) This brings up the whole category of pseudo-French or 'foreign-sounding' words, which are relatively more numerous here than in Webster, 33, about ten percent (as against 74, less than 7% in Webster). Especially common is the substitution of [z] for []*], which evidently started with a few words ending in -ge (rouge and garage, for instance), but has now extended also to most foreignlooking words containing -j-, or -g- before i or e (raj, Beijing, adagio, Azerbaijan) as well as nearly all words in -ge (including one example of lodge and several of Scrooge - in a movie version of A Christmas Carol). There are also a few pseudo-Italian examples - cicada with initial [c\, and schizophrenia with [ts] or [dz] for the z. (5) Three other types are more numerous here than in Webster: (a) = [g] vs. [fl, usually before e, i or y (no new examples like gaol and margarine except meningococcus with [j], 22 examples, (b) [ey] for Latin ae (regularly [iy], as in Caesar), 20 examples, mostly in the stem-syllable (Aeneas, Phaedra, Linnaeus, Timaeus) but sometimes either nominative plural (larvae, equidae, vertebrae, algae} or genitive singular (scenae frons, curriculum vitae), and (c) letters silent vs. letters pronounced, which is only in part due to pseudo-French but, in the case of e, perhaps to

81

pseudo-English, as, e.g., in aloe (good Greek three-syllable word, both ο and e long, should be [dlowiy] but is normally [aelow]), or contumely, a nice Latin word (contumelia), which would in Latin be accented on the e, and in English should be four syllables, but is often reduced by Hamlets to three. There are about 20 such cases. Among the minor groups are several problems with the sequences -ei- and -ey. In words ending in -eity, e and ι form separate syllables in Latin, and traditionally the ending is [iydtiy] or [iyitiy] in English. One frequent pronunciation today is [eydtiy], though one may also hear [aydtiy]. In my sample deity occurs five times, spontaneity also five times, homogeneity once. Cuneiform behaves in a similar way. Lamprey and parley often occur with [-ey], but not [-ay]. In other words -ei- is originally one syllable, and is normally rendered [ay] (and sometimes given an alternative spelling with -ι-, as in dino vs. deino-, dido- vs. cleido-), but sometimes [-iy-] occurs, as in Pleiades, and sometimes [-ey-]. Are there any type 2 words which are immune to new variations? It is easy to mention a few: (1) if a word ends in -tion or -sion, its accent is always on the preceding vowel; the same is true of words in -ic, -ious, and several other suffixes.

(2) stressed vowels before two

consonant letters which do not form a digraph (th, ph, sh, ch, etc.) are rarely pronounced 'long1. (3) words containing letters or sequences felt to be authentically English (ck, dge, sh, etc.) are rarely given pseudo-French renderings (lodge with [z] was a surprise). Such words are not frequent in type 2. One could perhaps list a dozen rules of similar validity, but more than a few thousands of words are susceptible to random variation.

3.

Objective identification of oral words

How can one identify type 1 words? There are, as far as I know, no lists alleged to contain all and only the orally transmitted words of English. But there are at least two other kinds of lists: (1) frequency counts (e.g. Carroll, Davies & Richman (1971) in which most of the most frequent items are surely type 1; (2) reports of the acquisition of English by children (e.g. Weir 1962, Harwood 1959) in which all words learned before entrance into first grade are probably type 1. These should help us answer our questions about the size of the type 1 vocabulary, and, by subtraction, the type 2. Many writers, like Bloomfield (1927:219), seem to imply that type 2 words are fairly uncommon. This is difficult to test, but we must first note that type 1 words are aurally learned words, not short words necessarily, or common words, though there will obviusly be a considerable overlap in both cases (and some readers may remember Zipf (1949) and his formula linking frequency rank with length). Of course not all aurally learned words (i.e., mainly words in a pre-school child's vocabulary) are type 1, but all type 1 words must be aurally learned. And, for most languages, most or all of the words on Swadesh lists (e.g. Swadesh 1952:546f.) are certainly type 1 (though English

82

definitions like navel and vomit are not). If one compares the words in Weir (1962) alone or both Weir (1962) and Harwood (1959) with the frequency order in Carroll, Davies and Richman (1971), we see an interesting distribution: virtually all the first hundred and twentyfive words of Carroll, Davies and Richman are found in both Weir and Harwood, and over half of Weir's words, 3/4 of those found in both Weir and Harwood, are in the first 2000 words of Carroll, Davies and Richman. From these to the 11,000th word, the children's words are spread out at the rate of one to ten in every hundred. After that point there are still 75 words in Weir, and an uncertain number in Harwood (where less common nouns, verbs and adjectives have been omitted), only a handful of which are peculiarly Australian (like pram, petrol and vegemite). Comparing Weir and Harwood with Swadesh lists shows a very high agreement: only a few (85) of the longest Swadesh list (219 plus) items are not found in Weir, Harwood, or (generally) both. The main basis for making up Swadesh lists is closely related to our basis for type 1 words - continuity over generations. Of course, in many non-literate cultures, all words are type 1; but those which are both type 1 and learned early are the ones which remain type 1 after literacy sets in. (Though sufficient pressure may cause people to alter their pronunciation; in my childhood I switched poor from [powr] to [puwr], but not without a fight). Certainly one of the clues used by Swadesh was frequency, but experience comparing word lists from related languages also counted. Chance and the smallness of the sample are surely to blame for the absence from Harwood or Weir of such words as scratch, skin, moon, rain, few, if, salt, rub, and fight. In fact, almost the only Swadesh words which we would be a little surprised to find in Weir or Harwood are louse, horn, breasts, lake, navel, husband, wife, peirce and vomit. Some words in Weir which we would not expect in a Swadesh list (mainly for cultural reasons) are blouse, bologna, cactus, cider, diaper, necktie, and toilet. If we return for a moment to the suggestion that short words will be especially likely to appear as type 1 words, consider the fact that of some 575 words (list available on request) of one, two and three letters in length known to me (and in most dictionaries), at least 495 are not on Swadesh lists, and do not occur in either Weir or Harwood. Some of these are uncommon, but many others seem quite normal (e.g. act, add, ago, aid, ail, aim, art, awl, axe, bin, bow, bun, etc.). And even these common words mainly feel like type 2. We can confidently say then, that neither frequency nor length in themselves are sufficient markers of type 1 words; the only objective procedure would be more studies like Harwood, involving dozens or hundreds of non-reading children. Subjectively, a Swadesh list is pretty good and a native speaker's intuitions may also be reliable, but neither one can replace objective tests. As we said above, then, only type 1 words are directly relevant in studies of true (neogrammarian) phonological changes, or in describing true (neogrammarian) dialect

83

differences.

Changes in the pronunciation of type 2 words are not really phonological

changes, but new (and often independent) oral renderings of particular spellings or else random borrowings from other reader-speakers. In modern times these renderings may be widely imitated and spread very rapidly through the influence of radio and television, but, when introduced, they are not directly related to previous renderings. Type 1 pronunciation changes are at least possible forerunners of phonological changes, and are changes of the sounds themselves. But type 2 changes (as in King of Prussia, Pa.), are only lexically diffused changes, like a switch from harass to harass or [fo:kn] to [faelkn]. One anticipation of this view which I may cite here is in Labov (1981:289), "The only linguistic data which we must acquire from our parents, and cannot get elsewhere, are the first dictionary entries." I think "must acquire from our parents" is an exaggeration, and siblings and early playmates may also play a role. But Labov does not seem to notice that words like planet and personality are type 2 words. Type 1 words are the domain of neogrammarian sound change and true dialects, type 2 of lexical diffusion and pronunciation dialects. This may, of course, not be true for Chinese; the evidence is not available to me. But a King-ofPrussia variety 'sound change' usually requires (1) a pair of contrasting phonemes existing already in the type 1 vocabulary, and normally spelled the same, and (2) a number of type 2 words containing this spelling. It is always uncertain which direction the 'change1 will move in, though at any time it may be possible to determine a trend. Of course, when one of the variants is wholly new, (introduced, perhaps, for elegance) there is no doubt which direction the change moves in for a while, as in the case of the recent change of Iranian from [avyrevnivdn] to [ivra:nivon] (along with a fair number of similar words). This violates an old descriptive and prescriptive rule that derivatives of foreign loans must be pronounced by the rules for 'classical' or 'native English' words, even if the base is 'foreign'. Here, too, there is of course, no phonological change of [ey] to [a:]. Changes of this type (Wangian changes, lexical diffusion changes, King-of-Prussia changes) are not, as we have said, phonological changes at all, but lexical changes, in which a word (or morph) is replaced by another which differs from earlier forms in only one phoneme (as a rule): e.g. [plaenit] is replaced by [plednit] or the like, and there are a number of other words (e.g. personality) which show similar replacements. It is mainly this parallelism which distinguishes them from simple lexical changes (e.g. the replacement of flesh by meat) although there is also, perhaps, the difference that no contexts are exempt for planet, but some are for flesh. But some words do split, favoring different pronunciations for different meanings, e.g. buffet (furniture vs. snack bar), chamois (goat vs. soft leather), or depot (railroad station vs. storage area). The factors of fashion, prestige and imitation are almost the same in these cases. Words with [td] are classier (in King of Prussia) or more loyal than words with [ae]. Loyalty is often important, as in Canadian [dw] and [dy] before voiceless Cs, or the similar diphthongs on Martha's

84

Vineyard. When the English (and Canadians) pronounce schedule with an initial [s] for our [sk], this is pronunciation dialect supported by loyalty.

4.

Rules for reading

Now we come to the question of the (partially formulated) rules used by readers to assign a pronunciation to a spelling, i.e. to give a mentally storeable form to new words first met in print - a form which may be subject to change due to the influence of a pronunciation heard from others ("Aha, that's how you pronounce it"), especially on television or radio, which leads to a sort of 'dialect1 grouping of type 2 pronunciations. So we say comrade with an [ae], the British with [ey].) There are, in general, besides the passe strategy of consulting a dictionary, two sorts of rules which we may call (in old-fashioned terms) prescriptive and descriptive, and three vocabularies to which they are applied: classical (i.e. Latin and Latinized Greek), foreign (basically perhaps French, but occasionally Spanish, Italian or German), and non-classical, nonforeign, i.e. words which, though now type 2, were once (in a linguistically purer century) type 1: 'native English'. The rules for this last vocabulary may be considered the residual, default or basic rules which apply if there are no special foreign or classical rules to be used. The native speaker first asks "Is it foreign?" and if the answer is "yes", then asks "Do any special foreign rules apply?" If the answer to the first question is "no", then he asks "Is it classical?" and goes on similarly from there. But, since the average speaker no longer has any certain knowledge of foreign languages (ancient or modern), these questions must be replaced by questions about the spelling: "Does it have a non-English spelling?", and "Does it have a classical spelling?" Examples where the first question may receive a "yes" answer are those in which a word (other than a name) ends in ι or u preceded by a consonant, or in eu or au, or j, or k preceded by a vowel, or contains an internal / which is not underlyingly initial, or begins with y plus consonant - and many more (see Venezky 1970, Albrow 1972). In the second case we may see an interconsonantal y or sequences of ae or oe, or ph or chth, or one of a number of familiar prefixes or suffixes, like con-, syn-, apo-, dis-, dys-, -tion, -sion, -tor, -sis, -sia, -ic, -al, etc. These strategies will not, in fact, detect all foreign or classical words. We may illustrate with the letter sequence CH. The default rule says "read CH as [c]", the foreign-French rule says "read CH as [s]", and the classical-Greek rule says "read CH as [k]." For example, the name Charon for Pluto's moon is, in fact, classical, hence [keyrdn], (with the vowel of Mary, whatever your dialect), but was once or twice recently pronounced [saerdn], or [ssran] as if foreign. The word conch is also Greek, hence [k^k], but is often treated as either foreign or native, hence [konc] or [kons], (indistinguishable in most dialects), a pronunciation easily detected in the plural even in print

85

- conches rather than conchs. If no special classical or foreign rules apply, we resort to the default rules. The most complete sources I know for prescriptive (also usually descriptive) rules for neutral, i.e. non-classical non-foreign words (the default rules) are (1) the account in Webster (2nd edition) by Kenyon ("entirely rewritten" from an earlier one by Paul W. Carhart) xxiii to liv, which also considers many words of classical or foreign origin which do not require special rules, and (2) the rules incorporated in the 'MITalk1 system of automatic conversion of text to speech (see Allen et al. 1987). Kenyon's account of purely classical words (including biblical names) is on liv-lv,4 and the following pages to Iviii deal in turn with French, German, Spanish and Italian, and are almost entirely prescriptive. A good deal of Kenyon's discussion treats dialect differences and purely phonetic details which are really irrelevant to our concerns. It is true that some dialects do have more or fewer vowel contrasts (less often fewer consonant distinctions) than others, but the rules must be substantially the same for all. The default rules, both for vowels and consonants, are mostly simple and straightforward; there is no need to give a rule for/(other than the rule that all {CC}5 = [C]), even though the word of is exceptional. No one looks up the pronunciation of type 1 words, except possibly non-native speakers. But we may say something about those letters which require special rules or lists of type 2 exceptions. These include (1) all the vowels, (2) c and t, (3) h, ch and th, (4) g, and (5) s and x. Of the others, a few are sometimes 'silent', but not much other variation occurs. I think that s (and x) are most instructive; I will briefly summarize the treatments of this by Kenyon (1960), Venezky (1970), Albrow (1972) and the writers of MITalk). Kenyon treats 5 on pp.1 and li (paragraphs 217-229), x on liii (par. 259-260) with special rules for Latin words on liv (par. 269) and French on Ivi (par. 272). He gives the treatment of s "everywhere else" first, and then deals with sc, sch, sh and ss, putting together in par. 227228 not only the pronunciations of in, but also all the spellings for [ä] and [z]. He gives one rule ("initial s does not spell z") which is quite general, as well as the general rule for all -s and -es suffixes ("after voiceless sounds except sibilants it is pronounced s [s]."). Otherwise he merely indicates historical reasons for the choice of [z] or [s], including stress, verb vs. noun (or adjective) and assimilation, notes several cases of variation ("usage is divided"), and implies (without actually saying so) that -sion and -sure after stressed vowels are normally He acknowledges his predecessor Walker (1835 - this is only one of many editions of this work), but without specifying title or date. Actually Walker is clearer on several points. A more recent summary of the same set of rules is found in Else (1967), with many examples, especially for the position of stress, and with some improvements in the rules. * Braces enclose ordinary spelling.

86 n] and [Ζβτ]. But that leaves an enormous number of words for which no prediction of [s] or [z] is ever suggested. Let's see how Venezky deals with this problem. Like Kenyon, he notes that initial s (before vowels or consonants) is always [s], and non-inflectional s anywhere before or after/?, t, k, c, ch, th, ph, /is also [s]. He makes a truthful caveat on p. 83, "The correspondences for initial and final s are fairly regular; those for medial s, on the other hand, are highly irregular [...]. The rules which follow [...] in some cases have as many exceptions as examples." Then his first "rule" is a list of 13 words with "silent" s (more or less the same as Kenyon's list, though Kenyon adds a few names). For "final inflectional 5" he has a two-stage sequence: first, he says it "corresponds to [s]", then, under "Alternations" he changes it to [z] after a "voiced morphophoneme", meaning, apparently, any vowel and any consonant except/?, t, k, c, /, th, ph. His correspondence (3) mainly duplicates this for any final s after consonants other than those, and adds final s in as, has, his, is, and was. Then (4) says "final s in all other cases corresponds to [s]", which ignores all the non-plural, non-verbal words ending -ies or es (where e is not "silent") like series, species, Ulysses, (and many other names), aedes, rabies, caries, congeries, anopheles, isoceles, herpes, litotes, which result from the rule for Latin words in English that final s is [s] after any vowel letter except e, when it is always [z]. For the other cases of [z] vs. [s], he gives the following: "Medial 5 [...] in the combinations ease, -vowel +Cse corresponds to [s] except in cleanse," and "when the nouns house and use, and the adjective close are converted to verbs, [s] becomes [z]." The first of these rules is misleading: lease, increase, cease, decrease, grease, surcease, decease, predecease all have [s], but ease, disease, misease, please, displease, pease, appease, and (in many dialects) grease (v.) have [z]. The rule for -Cse is pretty good (and he might have added -Csy) and so is the s to [z] rule for verbs related to nouns or adjective. Albrow's (1972) work didn't come to my attention until I had already noted the distinction between type 1 words and type 2 words (and the more limited class of 'foreign1 words), and it was interesting to see how similar some of his ideas are. Following the Firthian notion of polysystemicity, he proposes three systems (of pronunciation or of spelling), system 1 including mainly words of Old English and Danish origin plus some common words of French origin, system 2 being mainly those of Latin, Greek and French origin, the bulk of our polysyllabic words ending in -ive, -ion, -sis, -ment, -ant, -ent, -able, -ity, -ic, -al, and the like, while system 3 consists of partially Anglicized foreign loans (mostly what Kenyon treats in his French, German and Italian rules). Albrow is at some pains to say (e.g. 45) that it is symbols, not words, that belong to his 'systems', but he tolerates applying the term to words. The main difference between his system 1 and my type 1 is in the definition: roughly speaking, his system 1 words are ones which have been, at some time within the last 400 or 500 years, type 1 words in my sense, even if they are not type 1 (pre-literate) today. But, for

87

his purposes, he can shift some of these arbitrarily into system 2. So, for instance, the noun use belongs to system 1 and the verb use to system 2. Thus he has a general rule for system 1 that final {se} « [s], while for system 2 it » [z]. For my purpose this is circular and arbitrary, and is equivalent to listing all the cases where {se} · [z] or all those where it · [s]. Albrow's only rule for initial {s} words in system 1 is {s} » [s] (22), with a proviso (19) on [i] in sure and sugar (see also 41). Final {s} in as, was, is, has, his (25) · [z]. For final suffixal {s} or {es} (26) he gives the rule of assimilation ("depends on the voicing state of the preceding consonant" - perhaps he should have said "preceding phoneme") (intervocalic {s} (p. 33) m [z].) In system 2 (p. 38-9) he gives initial {s} as · [z] in German names (why aren't they system 3?), and final {s} * [s] in bus, pus, and words in -is and -us, but [z] or [s], "according to environment" in other cases (e.g. lens, species, ethics). Also on 39 he adds "intervocalic {s} · [s], as in philosophy, basin, oasis, etc." I can't find anything about {s} either before the accent or after prefixes. On p. 40 there are some lists of words with prefixes, all preceding the accent, but none involving {s}. Yet there are 20 or 30 words beginning with {de-}, {pre-} or {re-}, all of Latin origin, many with accent on the following syllable, which have {s} · [z], against his rules.

Exceptional (in having [s]) are words like desolate,

desiccate, desultory, presage, research, resource, resuscitate and resurgent, as opposed to desert (as well as dessertl), design, desire, reside, residence, preside, president, resist, desist (so also exist, but not insist) deserve, reserve, preserve (and observe). In fact, the total of all type 2 words (all of which should really belong to Albrow's system 2) with intervocalic {s} · [z] is appreciably greater than those with [s]. And most of the ones with variation between [s] and [z] do not look like type 2 becoming type 1, either: basilisk, meson, quasi, Casanova, usage, unison, etc. Some of these, indeed, may be system 3. I have about 120 medial {s} · [z] words and 100 medial [s] words, all of which seem to be type 2 except perhaps the names Susan and Jesus. The numbers can be reduced somewhat with various subrules involving suffixes, prefixes and accent, but not much. Besides the words for which dictionaries mention variation between [s] and [z], there are others which are familiar to all of us, like exist, exile, luxury, Joseph, basalt, and resource (Albrow's rules for {x} are "shorthand for *{cs}" (21), "initial ... {x} · [z] or in some cases [gz]", " · [s] for some speakers in xi" (38): no mention of words like exist, exhibit). Incidentally, since Albrow did not have any of the reasons in mind for which I proposed type 1, membership in the class of words all of whose segments are according to his system 1 might be an additional criterion for determining my type 1 words. Finally, let us look at one computer system MITalk (Alle, Hunnicut and Klatt 1987), designed to convert printed text to spoken text. In addition to the published book, I have had from Dr. A. I. Rudnicky of Carnegie-Mellon University (whose name was given to me by my colleague David B. Pisoni) a printout of the rules of the system, so that we can examine the

88

treatment of written {s} and its relatives and compare them with those of Albro (1972) and Venezky (1970). As nearly as I can read them, the series of rules for {s} begins with ten which deal with palatalization before various Latin suffixes beginning with {i} or {u}, yielding [δ], [£] (which most of us would regard as irrelevant variation from [s]) in words like extension or censure and [z], then initial {s} goes to [s], {s} before voiceless consonants to [s], before voiced to [z], which agrees with Albrow, but ignores words like asbestos, converts final {se} to [z] (okay for Albrow's system 1, but not for common words, like house, goose), then {ms} to [mz] before voiced segments (damsel, clumsy, etc.), then all intervocalic {s} to {z} (also like Albrow's system 1, but the rule fails for hundreds of words, of which those ending in {-sis} are covered by a listing, but not those beginning with {iso-}, {meso-}, {vaso-}, etc.; about such words as basic, basin, bison, casein, casino, sausage, gasoline, research, etc., I don't know), and finally all remaining {s} go to [s]. Since Dr. Rucnicky did not send me the listing of 'irregular' s-words, I cannot say how complete it is, but obviously it must contain at least 200 words plus a number of compounding stems. Of course the rules can be revised to do much better than this, but exceptions will still need to be listed. It is plausible, for instance, to note that productive prefixes like be-, re- [riy-], a- [ey-], bi-, tri- and co- are followed by initial-type segments, hence {s} is [s]; that most words ending in postvocalic -sVt, -sVnt, -ser, -sVl, -sVn (except for basal, basin, Jason and unison) take [z], that Latin prefixes de- and re- also take [z], and so on. It is obvious, of course that since rules cannot handle the complexity, some of these words must be learned by ear, in part. It is true that most real type 1 words with intervocalic {-s-} are pronounced with [z], but a few, like basin and sausage and gasoline and perhaps moccasin, as well as the name Jason, have [s]. But the type 2 words cause the real problem. If we could say that any type 2 word will naturally have [s] in intervocalic position, things would be okay. But we can't. And if we look at the words over which speakers hesitate (mostly not yet noticed by Webster (2nd edition)) we see both kinds of variation: traditional [s] words like bison, exit, luxury, resource, and quasi are sometimes given [z] pronunciation, while traditional [z] words like merchandise, Joseph, expertise, damson, crimson, diesel, exotic are given [s] pronunciations. The problem is different where it is linked to geography (but cannot be 'dialect' in the traditional sense): a whole class of words in -sia, -sian, -sion are given [s] pronunciations in England, but [z] pronunciations in America (which are surely older). This last group is, I suppose, easily explained by the influence of school teachers, from whom many such words are heard first and often. But the other set seems to depend, in part, on the fact that people do not normally have any rules (such as those briefly sketched above and found in all our sources), but instead work from analogies and paradigms. For bison one could think of reason and season, for exile one might go to the corresponding verb in earlier

89

times when the second syllable was stressed, for exit to exile itself and to exist, for crimson and damson to Thompson and Jimson, and so on. But when such a pronunciation spreads, as we have observed above, no supporting analogy is needed; the prevailing feeling "I must have been wrong all the time" is usually enough. In some cases the new pronunciation has now become type 1, no longer subject to such change.

5.

Rules for latinate words

We now return briefly to Latin and Greek words (especially names) in English, which was my original starting point for this paper. Kenyon (1960) is not quite as clear as his original source, Walker (1835), and ordinary people without any classical education may need a few tips (see also Else 1967). (1) Accent in complete Latin or Greek words of three or more syllables (with no endings dropped) depends on the quantity or 'weight1 of the penult (quantity anywhere else is totally irrelevant to English pronunciation). We can identify 'long1 or 'heavy' syllables ourselves only if either (1) they contain a diphthong - ae, oe, au, ui, eu - or (2) the vowel is followed by two consonant letters, of which the second is not h, r or /. Otherwise even Latinists or Hellenists may have to consult a dictionary, unless they can remember a piece of verse containing the form in question. But once the quantity of the penult is known (not necessary for two-syllable words, which must be accented on the penult anyway), we can say, accent it if it is 'heavy' or 'long', otherwise accent the antepenult. Next we determine vowel quality for accented vowels. Kenyon here introduces an extra step, syllable division, saying that it "follows the English method [...] thus, det'i-net." This almost certainly will confuse the reader; first, he means this rule to apply only to accented antepenults, where it means that these are considered 'closed', if they are, as he says "followed by a consonant in the same syllable," in which case they receive English 'short' (i.e. checked) vowel pronunciations, [ae,e,Ip,A] except before {r}. If he just left the matter of syllableclosing out, he could say that stressed vowels, if followed by one consonant in the antepenult or by two or more in the penult, have 'short' values. But, even so, he overlooks an important exception, to which Walker correctly calls attention several times. If the penult consists of just the vowel {i} (followed usually by -a, -us, -urn, or -o, then all the antepenult vowels except {i} (and {y}) have 'long' values - alias, Decius, Photius, juria, but Lucilius, Syria. (Incidentally, {y} is always a vowel letter in these words, and is merely an allograph of {i}, as Kenyon says.) Final {e} in Latin and Greek words is always a vowel and has the 'long' sound [iy]. But, of course, the ignorant modern reader can't be sure this {e} is ancient; very often when ancient -us, -a, etc. are dropped, their place is taken by silent {e}. In Latin a real final {e} is either an ablative ending (as in bona fide 'in good faith') or neuter nominative as in simile

90

'likeness'), but in Greek words it is most often a feminine nominative, as in apostrophe, hyperbole, Zoe, Chloe and aloe vera ('true aloe') a phrase which neither cosmeticians, herbalists or nursery men ever pronounce 'correctly' today. Just as {y} is an allograph of {i}, so {ae} and {oe} are allographs of {e}, and always to be pronunced as {e} would be in the same position. In most words, these have been replaced by {e} in English spelling, anyhow - eternal, penitent, ceremony, levoratatory, cenozoic, chimera, scene, epicene, cemetery, penalty, homeopathy: in all cases the first {e} represents either Latin {ae} or Latin {oe}, which, in many cases, represent Greek {AI} or {OI}. The modern deviations from this rule (that {ae} and {oe} are pronounced exactly like {e} in all positions) are mostly in case endings (where {oe} does not occur), where the tendency is to pronounce {ae} as 'long' {a}, i.e. {ey} - vertebrae, scaenae frons, larvae, equidae, algae (so also in some names - Judaea, Aeneas, Phaedra, Linnaeus, Timaeus, Aesop), or else with high school Latin pronunciation, as [ay] ('long' {I}) - larvae, algae, antennae, Lycaenidae. There is no doubt, however, that {ey} is more frequent than [ay]; in fact, I have several times heard vertebrae (with {ey}) used as a singular. The other common deviation from Walker's and Kenyon's rules concerns the variant pronunciations of {g} and {c} before 'front vowel' letters - i, e, ae, oe and y. The rule in this case is 'soft1 pronunciation, i.e. [J] and [s] respectively, except for British fungi and longitude with 'hard' [g] (this is also Albrow's rule for 'system 2', incidentally). Deviation in the case of {c} is rather limited: one case of pseudo-Italian pronunciation of cicada (regularly [sikeydö]) as [clkädd], and two cases where {cc} is rendered as [s] instead of [ks]: accessory and accipiter. But [g] against the rule is more common: Two words, gibbous and ginglymus were already so rendered in Webster's day; gynecology is nearly always pronounced with first syllable [gayn-] today. The use of 'long1 [ay] for {y} occurs in other words as well, especially in British use.

6.

Conclusions

We may now sum up: (1) only type 1 words, those which have for two or more successive generations been acquired before literacy, can possibly undergo normal sound changes; (2) type 2 words vary in the amount of aural influence, but are always subject to new spellingpronunciations or what Bloomfield calls dialect borrowing; (3) as yet we cannot clearly determine how many type 1 words there are in English, but several possible clues have been examined: (4) one form of dialect borrowing is looking the word up in a dictionary, another is following pronunciation rules, but people are especially likely to be influenced by what they hear on TV; (5) the branch of linguistics which deals with synchronic variation in type 2 words is not comparative phonology or dialect geography; (6) the branch which deals with

91

historical changes in type 2 vocabulary is not historical phonology; (7) in fact no one has yet clearly separated the study of type 2 vocabulary, though many studies of the past two centuries would belong to this field.

Bibliography Albrow, K.H. (1972): The English Writing System: Notes Toward a Description. - London (= Schools Council Programme in Linguistics and English Teaching, Papers Series II, Vol. 2). Allen, J., M.S. Hunicutt & D. Klatt, with R.C. Armstrong & D.B. Pisoni (1987): From Text to Speech: The MITalk System. - Cambridge/New York. Bloomfield, L. (1927): "On Recent Work in General Linguistics." - In: Modern Phonology 25, 211-30. Carroll, J.B., P. Davies & B. Richman (1971): The American Heritage Word Frequency Book. - Boston/New York. Else, G.F. (1967): "The Pronunciation of Classical Names and Words in English." - In: The Classical Journal 62, 210-14. Harris, J. (1989): "Towards a Lexical Analysis of Sound Change in Progress." - In: Journal of Linguistics 25, 35-56. Harwood, F.W. (1959): "Quantitative Study of the Syntax of the Speech of Australian Children." - In: Language and Speech 2, 236-74. Kenyon, J.S. (1960) (revised from an account by P. W. Carhart): "A Guide to Pronunciation." - In: Webster's New International Dictionary (second edition) (Springfield, Mass.) xxii-lxxviii. Kenyon, J.S. & J.A. Knott (1953): A Pronouncing Dictionary of American English. - Springfield, Mass. Labov, W. (1981): "Resolving the Neogrammarian Controversy." - In: Language 57, 267-308. Swadesh, M. (1952): "Lexical-Statistic Dating of Prehistoric Ethnic Contacts." - In: Proceedings of the American Philosopical Society 96,452-63. Venezky, R.L. (1970): The Structure of English Orthography. - The Hague (= Janua Linguarum, Series Minor NR. 82). Walker, J. (21835): "A Key to the Classical Pronunciation of Greek, Latin, and Scripture Proper Names." - In: N. Webster: An American Dictionary of the English Language (fifteenth edition) (New York) 941-1011. Wang, W.S.Y. (ed.) (1977): The Lexicon in Phonological Change. - The Hague. Webster's New International Dictionary (21960) - Springfield, Mass. Weir, R. Hirsch (1962): Language in the Crib. - The Hague (= Janua Linguarum Series Maior, XIV). Wenrworth, H. (1944): American Dialect Dictionary. - New York.

93

Andrea Nemeth-Newhauser, Richard Newhauser

Aspects of the Development of the Imperative in Early Modern English 1.

Introduction

There is a wide range of opinions concerning mood and modality. It is generally accepted that mood is an attribute of the verb and that it expresses the speaker's "mental attitude towards the fulfillment of the action or state expressed by the predicate" (Poutsma 1926, II.ii:161). There are, however, conflicting views about how mood is formally expressed, whether through "a form of the finite verb or a verb-group" (Poutsma 1926, II.ii:161), through "a grammatical form, or the function of such a form" (Jespersen 1949, VII:623), or through "verbal inflections or ... syntactic contrasts" (Harsch 1968:12). A different point of view altogether was taken by Palmer (1977:2), who suggested that one should: distinguish the terms "mood" and "modality," mood being essentially grammatical, such that Latin has a subjunctive mood, while modality relates rather to the speaker's commitment. In this sense English has no moods of the subjunctive/optative type (though the term might be applied to the modal verbs), but we are free to discuss modality in relation to Latin, Greek or English. A similar view is expressed by Khlebnikova (1976:3-4): Modality is a conceptual (semantic) category, a type of meaning or a complex of meanings, with various reflexes in the language [...]. [Modality] establishes the character and content of the relation between the subject and the predicate of the statement [...]. Modality is not a syntactic category, since the linguistic means of its expression as a semantic category will be separate subclasses of words and phrases, the morphological category of the modal moods (imperative and conjunctive) and modal verbs. Thus, it appears that there are essentially two schools of thought concerning mood: one which regards it as a semantic category with various possibilities of formal expression, and another which defines it as a morphological category within the larger semantic category of modality. The linguist's definition of mood also determines his view of the number of moods in English. Those who define mood as a semantic category and include various modal verbs in the expression of mood may arrive at a number as large as 16 (Deutschbein 1931) or 15 (Nida 1957). Others who regard mood as a form of the verb characterized by inflection exclusively deny the existence of the subjunctive mood in English altogether (Palmer 1977). Most authors, however, distinguish the indicative, the imperative and the subjunctive (sometimes called optative) (e.g., Trnka 1930, Harsch 1968, Jespersen 1949, etc.).

94 Occasionally the conditional is added to this list, either as a separate mood (Poutsma 1926) or as a syntactically conditioned variant of the subjunctive (Khlebnikova 1976). For the purpose of this essay we will accept the view that mood is a morphological category expressed by verbal inflection or by analytical forms in which the modal verb has lost its lexical meaning. This will enable us to examine the distribution of the modally marked forms of the verb in Middle and Early Modern English. This distribution will form the basis for categories in which other possibilities of expression can be examined later. The point of departure in distinguishing various moods is the modally unmarked form, the indicative. The indicative is said to state reality (Trnka 1930), or to state a fact (Harsch 1968). Poutsma (1926) labels it a "predication of certainty," and Jespersen (1949) calls it a "fact-mood." The indicative occurs the most freely, in main clauses and subordinate clauses as well, even when the modality of the utterance is non-fact. In a standard form of the East Midlands dialect of Middle English (which served as the foundation for Early Modern English), a simplified paradigm of the present-tense indicative, subjunctive and imperative would appear as follows: (1)

(a) Indicative Ich pu he, sehe, hit we, 3e, hy/pei

(1)

Ich am bu art/best he, sehe, hit is/beb we, 3e, hy/bei aren/beob/ben/be.

here here(n)

bo/be ben

her heres/hereth

be beth

(b) Subjunctive singular plural

(1)

here herest hereb herenl-es

(c) Imperative singular plural

In Early Modern English the indicative of the verb in the present tense had the following forms: (2)

I, we, you, they come thou comest he, she, it cometh

I am thou art he, she, it is we, you, they are. Along with these possibilities in present tense usages in Early Modern English, one also finds modally marked forms (amounting in effect to the imperative and subjunctive), namely (1) occurrences of the base form of the verb for all persons (which is identical in appearance with the infinitive without to) and (2) various analytical forms, i.e., combinations of modal

95

auxiliaries and the infinitive.

In the following we will examine the distribution of the

imperative measured against the unmarked verbal category of the present tense indicative in Early Modern English. We have attempted to gather material from various sources. Since our major interest is in determining what means of expression the language had at its disposal during the Early Modern period, rather than to what effect these means were used in literature, we have included, whenever possible, evidence from non-literary prose texts.1

2.

The Base Form of the Verb

There are two major categories in Early Modern English in which the base form of the verb occurs: the imperative and what is customarily termed the present subjunctive.

The

distinction between the two categories is not always clear. It is often said, for instance, that the subjunctive is used as a substitute for the missing forms of the imperative. Thus, we find that sentences like: (3)

Some officers take them away (LR V.iii.l)

are sometimes treated by linguists as imperatives and sometimes as subjunctives. Although historically (at an earlier medieval stage of the language) the subjunctive was used in nonsecond person commands and exhortations, we hope to show that with the disappearance of differences between the two verb forms of the subjunctive and the imperative by the time of Early Modern English there is no longer any reason to call this usage subjunctive. Some Middle English examples relevant to the Early Modern English forms will demonstrate the close proximity of the subjunctive and imperative in non-second person usage in the later medieval development of the language. The erosion of inflectional endings during this period, predating Early Modern English, only sets the stage for the rapid encroachment of the imperative on earlier subjunctive functions: (4) (5) (6) (7) (8)

'jet firre,1 quob be freke, 'a forwarde we make' (SGGK, 1105). 'Kefourme we oure forwardes, er we fyrre passe' (SGGK, 378). if ony man hungrith ete he at home (Edinburgh, University Library MS. 93, fol. 103r). So bo hit euer in unker sibe bat bo bo sori, & ich blibe! (ON, 993-94). God yow for3elde\ (SGGK, 2429).

1 Among the citations from Shakespeare we have tried to use only those which most closely resemble prose style in order to avoid distortions of the pattern resulting from considerations of meter and rhyme. For a listing of the abbreviations of cited works see section 1. of the bibliography.

96

(9)

3.

If any freke be so feile to fonde bat I teile, Lepe Iy3tly me to, and lach bis weppen, I quit-ciayme hit for euer, kepe hit as his auen (SGGK, 291-93).

The Imperative

The imperative mood expresses a command, a request, a prohibition or a warning. It is, in other words, a wish that an action be or not be performed by the person(s) spoken to, the speaker together with the person(s) spoken to, or by a third person. Imperatives occur exclusively in main clauses.

Originally, the verb in Old and early Middle English had a

second person plural imperative form ending in -b, -th, -s or -z, and a singular form without ending (see above, (lc)). The plural inflection was also used as a polite form to address a superior. However, plural imperatives without ending can be found in texts as early as the Ancrene Riwle (thirteenth century), and by the mid-fifteenth century this became the established usage (cf. Visser 1:15). A few examples of second-person usage will illustrate the possibilities of expression in Middle English: (10) (11) (12) (13)

Ta now by grymme tole to be, And let se how bou cnokez (SGGK, 413-14). Tas yow j>ere my cheuicaunce, I cheued no more (SGGK, 1390). Tas yow lust nu hider! (ON, 955). 'Lustep,' heo cwab, 'latep me speke!' (ON, 1729).

In Early Modern English the base form of the verb was used for the imperative without person, number and tense distinction: (14) (15)

Make muche of your dearest frendes, and do not procure anie Enemyes (Warnings, no. 14). Therefore prove thyself in this wise [...] (Latimer 6:357).2

The presence of a second person reflexive pronoun, thyself, in (15) above suggests that the underlying structure of subjectless imperatives contains the second person pronoun as subject. Thus, the You-Deletion transformation proposed for present day English (Culicover 1982, p. 177) can also be included preliminarily in an Early Modern English grammar (in a revised form, as what must be called the You/thou-Deletion), ordered after that of Reflexivization. Merely modeled on You-Deletion, this rule would appear as follows:

The difference between singular and plural forms in the imperative is also observable, for instance, in German, French, Latin and Italian. Finnish and Hungarian, on the other hand, have imperative forms for every person except first person singular. In Hungarian, the imperative paradigm is used for all "subjunctive" functions, whereas Finnish has a separate subjunctive paradigm.

97

(16)

yoM/fAow-Deletion you] \ thou j

-

1

VP 2

=>

0

2

Condition: in main clause only This transformation is triggered by the absence of an auxiliary in the deep structure, i.e., it applies only to strings that contain no tense and no modal verbs.3 The deep structure would then appear as follows:

(17) NP

VP

you

make

you/fAoM-deletion is an optional transformation in Early Modern English. Sometimes the second person pronoun thou or you, or even thee or ye is retained and placed after the verb. This usage was common from Old English through the 18th century; today it survives only in dialects or petrified forms like mind you (cf. Visser 1:16-17): (18) (19) (20) (21) (22)

drynkeye all herof (Edinburgh, University Library MS. 93, fol. 102r). I schal bynde yow in your bedde, bat be je trayst (SGGK, 1211). Hermia, sleep thou here (MND II.ii.35). Look thee here boy (WT III.iii.116 - Visser 1:17). Fetch me this herb and be thou here again Ere the leviathan can swim a league (MND II.i.173-74).

Placing the pronoun before the verb was a phenomenon frequent in Old and early Middle English, one which entirely disappears between 1475 and 1695 (Visser 1:17), and which reappears at the end of the seventeenth century: (23) (24) (25)

Forby bow lye in by loft and lach byn ese (SGGK, 1676). 'Vnwi3t!' ho sede, 'awei bu floV (ON, 33). bis penaunce now $e take (SGGK, 897).

The reason for the presence of the pronoun may be contrast, as in the following in today's usage: Because of surface forms like: Pass the salt, will you? Be quiet, won't you? the following imperative transformation has also been proposed (Katz and Postal 1964): you tense will VP 1 2 3 =>003 For the purposes of the present paper this rule would have to be modified so that in certain cases only the deletion of tense-»«// would be obligatory.

98

(26) You go ahead; I'll wait here. On the other hand, as Curme (1931:436) suggests, adding (or retaining) the pronoun takes the edge off the command and makes a request more polite. At any rate, it is definitely a marked form: a preliminary count of imperative sentences in A Midsummer Night's Dream yielded eleven times as many occurrences without the pronoun as with it. The entry bring in the Shakespeare Concordance - a likely candidate for many imperative sentences - contains 8 affirmative imperatives followed by a personal pronoun, contrasted with 158 without it. Although the auxiliary verb do is very frequent in Early Modern English in present and preterit affirmative sentences, the do-V sequence virtually never occurs in affirmative imperatives. We find either the second person pronoun or but between do and the verb: (27)

You have her father's love, Demetrius, Let me have Hermia's; do you marry him (MND I.i.93-94). (28) Do but behold the tears that swell in me [...] (LLL IV.iii.32). In Old and Middle English, one can note, the form do-V occurred both with the imperative ending and without it. This involved, however, not the periphrastic do so frequent in Early Modern English, but rather an emphatic do similar to present day usage: (29) Dos teches me of your wytte (SGGK 1532). (30) Now hastily dofecche a book (Chaucer, CT B662 - Visser 111:1519). Visser lists altogether four instances of do-Pron-V (111:1520) before the sixteenth century, and no instance of do but before 1505. According to Elleg rd (1953:174-78) the reason for the absence of the do-V sequence in Early Modern English is "that in the present and past tense the do-form could develop in analogy with the modal auxiliaries, whereas in the imperatives this analogy fails." If we derive imperatives from tenseless structures, as it has most often been suggested (see above, (16)), then the impossibility of Do-Support can account for this lack of affirmative imperatives of the do-V type.4 However, the occurrence of do in surface sentences presents a problem when the verb is preceded by another word. The only viable solution here seems to be an ad hoc do-insertion rule, as proposed, for instance, by Culicover (1982:176). Dahl (1956:93) suggests that the presence or absence of do in affirmative imperative sentences is structurally conditioned: when the verb is preceded by weakly stressed words or 0 then do

Do-Support, as discussed by Culicover (1976:123), inserts do in the appropriate context. The transformation presupposes the presence of Tense in the deep structure: X Tense Υ 1 2 3 = > 1 # Λ > + 2#3. See Culicover (1976:122-26) for a full discussion of the relative merits of an analysis containing either DoSupport or Do-Deletion.

99

is omitted. This suggestion can only be treated sufficiently in a more detailed examination than we have space for here. In any case, affirmative imperatives with do seem to be rare. In A Midsummer Nighfs Dream only two occurrences of such forms containing do were found, both followed by you (see above, (27)). The imperative in Early Modern English is often introduced by go, come, pray, prithee, etc.:

(31) (32)

Come bring me to some private place (PER IV.vi.90). Pray bring her in And let's see how she is (TNK IV.iii.101). (33) Prithee bring him in, and let him approach (WT IV.iv.211). (34) Go bid the huntsmen wake them with their horns (MND IV.i.140). Another subjectless structure using the base form of the verb may be included here, the kind that Jespersen (1924:315) calls the "imaginary imperative": (35)

Weigh oath with oath and you will nothing weigh (MND III.ii.131).

Although this structure is not strictly a command, formally there is no difference between it and the imperative. Furthermore, the above sentence can also be derived from a tenseless underlying structure through You/thou-deletion. In negative imperatives we find forms both with and without do, as was already the case in Middle English (cf. Visser 111:1540-41): (36) Do not say so, Lysander; say not so (MND II.ii.108). (37) Good Hermia, do not be so bitter with me (MND III.ii.306). (38) Be not afraid [...] (MND III.ii.321). When the second person pronoun is retained, its place in a negative imperative varies. If do is not present, the pronoun either precedes or follows not: (39) Her Highness is in safety, fear you not (JN III.ii.8). (40) Uncleanly scruples! fear not you (JN IV.i.7). When do is present, the pronoun follows not: (41) (42)

Donotyoufear(WTll.ii.63). Do not you grieve at this (2H4 V.v. 76).

Negative imperatives containing the verb be are virtually never paraphrased with do. The only occurrence of do not be in Shakespeare is MND III.ii.306 (see above, (37)). The pronoun, if retained, can either precede or follow not: (43) Be not you spoke with but by mighty suit (R3 III.iii.46). (44) But cheer thy heart, and be thou not dismay'd (R3 V.iii.174). This construction containing the verb be was considered by Visser (111:1542) to be a true innovation in Early Modern English. He has no examples before 1553 without do and none

100

before 1590 with do. While it may be true that "negative exhortations, commands, etc. with be appear for the first time in print in the sixteenth century," as he maintained, there is evidence that such collocations were not unknown in Middle English: (45) Bolde burne, on bis bent be not so gryndel (SGGK, 2338). The kinds and frequencies of negative imperatives in Shakespeare's dramas contrasted with some short prose works of the period can be seen in Table 1. Table I. Frequency of various types of negative imperative in Early Modern English works Shakespeare

Good Advice

Warnings

6th Sermon

3rd Sermon

3

3

V-not

398

2

17

do-not-V

153

1

1

V-PN-not

11

V-nof-PN

9

3

2

do-not-PN-V 6 be-not

65

do-not-be

1

be-not-PN

1

be-PN-not

1

It is apparent that the forms without do are far more frequent than those with do, and the use of the pronoun is also fairly restricted. The prevalent pattern seems to be of the type go not and be not. This coincides with the findings of other linguists (Dahl 1956): in the Marprelate Tracts, in the writings of Robert Some, and in those of John Percy, Dahl found no occurrences of do in negative imperatives; in the other works he investigated, the form without do was more frequent. His findings also suggest that certain verbs are more often paraphrased than others, even in statements. For example, in the works of Robert Some and John Percy, the verb doubt is never negated with do. The data reveal the following possible structures for imperative sentences with second person subject in Early Modern English:

101

(Go!) (Go thou!) (Go not!) (Go thou not!) (Go not thou!)

V V-Pron Vnot V-Pron- not Vnot-Pron

(46)

(Do thou go!) (Do not go!) (Do not thou go!)

V do-Piondo-notV do-not-Ρτοη- V

Two generalizations can be made on the basis of the data: (1) the imperative sentence always begins with a finite verb or the auxiliary do, (2) the presence of the personal pronoun is optional. Thus, there must be a rule which moves V to the beginning of the structure or alternatively inserts do in this position, and there must be an optional pronoun deletion rule. The following rule was tentatively proposed for ybu/fAou-Deletion above (cf. section 3):

(16)

YoM/f/ioM-Deletion

LthouJ

1

2

=>

0

2

Condition: in main clause only However, it is now apparent that the structural description of this rule is too restricted. The deletion of the pronoun from the surface structures V-Pron, V-Pron-nof, V-nof-Pron and donof-Pron-V would yield equally correct surface forms. The string do-Pron-V is the only exception, because here the pronoun deletion is not allowed. The rule must be modified so that it can optionally delete the second person pronoun in tenseless main clauses except when the pronoun immediately follows do. It appears from the data that starting from a deep structure (noi)-Pron-V, the following derivations are possible:

102 Affirmative Imperative

(47)

Hermia, thou sleep here

(b)

(a)

Hermia, do thou sleep here

Hermia, sleep here

Hermia, sleep thou here (MND II.ii.35)

Hermia, sleep here Where (a) = pronoun deletion, (b) = V-movement, and (c) = do-insertion. The rules cannot be ordered, as is apparent from the fact that the application of either (a), (b), or (c) yields acceptable surface forms. Moreover, (a) can also apply to the output of (b), because its structural description is met. Thus, the form of the imperative without either the pronoun or do can be arrived at in two different ways, and this, in fact, corresponds to its much greater frequency in the language.5

(48)

Negative Imperative * not thou fear

(b)

(c)

fear not thou

do not thou fear

"Was ist mit Ihnen?" (47) "What about Molto?" asks Lip. "You figure he was in?" (p. 359) ->" Was ist mit Molto?" (48) "And what about Molto? Did you ever hear anything about him?" (Sabich on p. 378) --> "Und was ist mit Molto?" In (49), Lip explains to Sabich the evidence on which he was supposed to be convicted: (49)

" What about your phone records, if somebody should think to look?" (p. 418) --> "Und was ist mit den Telefonprotokollen?" In the translation of (41) above, the indefinite NP a report has been changed to a definite NP: "Sind die Akten [...] angefordert?" In some other cases, full sentence translations gloss over the distinction between How/What about: (50)

[...] Larren is upset. " What about this, Mr Delay Guardia?" (p. 228) --> "Was haben Sie dazu zu sagen [...]?"

(44) --> "Und warum habe ich keinen fairen Prozeß gekriegt, Raymond?" As to the intended speech acts, How about has been appropriately translated as Wie war's and What about as Was ist mit (although the latter would not be stylistically acceptable in all contexts as a standard translation). Wie steht's mit apparently corresponds to What about more often than to How about. Because of this match between Wie steht's mit and What about, the translation of (45) is somewhat infelicitous: (45)

"Do we have a secreter?" I ask. --> "Wie steht's mit Sekreten?" The German translator, Christa E. Seibicke, cannot be held responsible for misrepresenting Sabich's approach to the investigation by this detail of translation, of course. The choice of

255

How/What about happens to be only one minor device among several others for structuring the plot.

Bibliography Akmajian, A., R. Demers & R.M. Harnish (21979): Linguistics: An Introduction to Language and Communication. - Cambridge, Mass. Fischer, W.L. (1973): Äquivalenz- and Toleranzstrukturen in der Linguistik: Zur Theorie der Synonyme. München. Fowler, H.W. (1959): A Dictionary of Modern English Usage. - Oxford. Gottschalk, K.-D. (1977): "What about...? - Eine Fehlerbewertung." - In: anglistik und englischunterricht l, 6179. Gottschalk, K.-D. (1982): "How to Describe Idiomatic Language Within the Framework of Universal Grammar" - In: R. Dirven & G. Radden (eds.): Issues in the Theory of Universal Grammar (Tübingen) 159-196. Radford, A. (1988): Transformational Grammar. - Cambridge. Rombauer, I.S., M. Rombauer Becker, M. (1973): Joy of Cooking. · New York. Turow, S. (1988): Presumed Innocent. - London. Turow, S. (1988): Aus Mangel an Beweisen. - München (translation by Christa E. Seibicke)

257

Elsa Lattey and Juliane Moeck

Linguistic and Stylistic Considerations Affecting Restrictive Relative Clauses in Spoken and Written English1 1,

Introduction

The differences between spoken and written discourse are often discussed in terms of factors such as permanence, speed of production, reviewability, or intonation. Less frequently, investigators have considered the role of specific grammatical structures in the one or the other type of discourse. The present study takes a look at specification of an entity mentioned in the discourse (an NP) by means of postmodification.

One of the principal clausal structures used by

speakers/writers in such postmodifications is the relative clause. In this construction, there are several available options: use of the relative pronoun that or of wA-relatives such as who, which, whose, etc., absence of a relative pronoun, and absence of both a relative pronoun + the auxiliary be (cf. (l)-(7)). T2 W W 0 W Z T

(1) He's using the power that California gains from being [a] textbook adoption state. (radio panel = B^) (2) She soon put the city slickers who dismissed her in their place [...] (non-fiction = L) (3) And he would not see that the tints which he spread upon the canvas were drawn from the cheeks of her who sate [sic] beside him. (fiction = E) (4) In the continual war [0] he waged on complacency lies one of his greatest strengths, (non-fiction = L) (5) They'll use the phone in the house where the hoods are that they're with [...] (informal interview = C) (6) [...] nowhere in England or Scotland or Ireland is there a form of educated speech [Z] sufficiently unified and [Z] used by a sufficient number of speakers to form a rival standard, (non-fiction = L) (7) Kathy Craig [...] wrote this article ώα* was talked about on the show, (informal interview = C)

The choice among these options appears to reflect a number of linguistic and stylistic considerations - linguistic considerations such as the (semantic) role of the antecedent or of

* An earlier version of this paper was presented at the 17th Annual University of Wisconsin-Milwaukee Linguistic Symposium, 1988. 2

Τ = that, W = any of the »Wt-relatives: wlio, whom, whose, which, where, wlien, why, 0 = absence of a relative pronoun, Z = absence of a relative pronoun and of the auxiliary be.

^ These letters refer to the styles considered (cf. 2. The data for details).

258

the relative pronoun and stylistic considerations such as the type of discourse and the level of formality (cf. also Danielewicz 1988). In the absence of a meaning analysis of that, the whwords and 0, we can in this paper only discuss the choice among the relative clause alternatives in terms of the nature of the signal, i.e., its form. We will try to point out the significance of facts relating to this form for certain aspects of communication. Before we focus on the relative clause construction itself, we would like to fit it into a more global picture of postmodification possibilities, including in particular other clausal constructions

and prepositional phrases.

And since our objective is to investigate

specification of an NP by postmodification, we shall restrict the discussion to defining (limiting, restricting) postmodification, to mention just a few of the terms under which these structures have been subsumed. The current investigation includes 2693 restrictive relative clauses from two studies at the University of Tübingen, one a collection of relative clauses per se gathered from written and oral sources, the other a more general collection of restrictive postmodification, including relative clauses. Contrary to popular opinion, which expects fewer relative clauses in spoken than in written language (e.g. Ochs 1979, Chafe 1982), spoken contexts that resemble written ones communicatively (i.e., with respect to conveying information about finding the referent or the subset of referents intended by the speaker) do contain relative clauses - and in greater number than corresponding written language samples.

Since previous counts of relative

clauses (cf. for example, Chafe 1982) have been based on the number of relative clauses occuring in relation to the number of words in the sample, we carried out a similar count for a portion of our data, with the following results: Table 0: Comparison Count

Non-fiction (articles from English Today, The Listener, The London Times, &. Time) Fiction (short stories by James Joyce, Woody Allen, Ernest Hemingway, William Sansom) Interviews (face-to-face conversation) Radio call-ins (non-face-toface conversation)

Total no. of words

No. of Rel.Cl.

Rel.Cl./ 1000 words

6102

58

9.51

6391

33

5.16

6452

48

7.44

5421

93

17.16

These results differ noticeably from those derived from Chafe's counts of dinner-table conversation vs. academic papers (Chafe 1982:36 and 44):

259

Total no. of words Dinner-table conversation Academic papers

Rel.Cl./ 1000 words

9911 12368

9.7 15.8

This discrepancy suggests very definitely that the nature of the communicative (interaction needs to be taken into account in any such comparison. A count of something like relative clauses in relation to total number of words seems to be somewhat inappropriate to begin with, there being countless factors that can affect the total number of words without being relatable to relative clauses (e.g. exchanges of greetings or situational utterances like "Pass the salt, please" can take up quite a bit of space in a dialogue without contributing a potential locus for a relative clause). It seems to us that relative clauses should be counted in some way relative to the number of potential antecedents, i.e. to the number of entities that could be modified by a relative clause. That is what we have done. Our general postmodification data allow us to compare clausal and prepositional postmodification relative to type of source, and the results suggest a first approximation of communicative considerations that influence the choice between them. Table 1 shows a total of 2401 cases of postmodification, divided into clausal and phrasal modification4 and differentiated by type of linguistic production. Table 1: Postmodification Clauses

Face-to-face conversation Non-face-to-face conversation Ads (radio) Live radio reporting Fiction Non-fiction

Prep. Phrases N

%

N

%

376 201 16 45 178 205

54 52 50 36 34 32

322 184 16 78 345 435

46 48 505 645 66 68

1021

1380

= 2401

The clauses include: relative, infinitive, -ing, as well as participial, adjectival and adverbial clauses. The phrases consist of preposition + NP. The small N here makes generalization difficult.

260

We can see from these data that the speakers and writers of different styles - different contexts of language production - avail themselves of the postmodification types to varying degrees, which nevertheless correlate nicely with a preliminary hierarchy we might suppose to exist among the styles. We refrain from naming this hierarchy for the present, though we feel that the possibility of 'degree of formality1 or 'degree of communicative closeness' or something along similar lines is worthy of consideration. What is striking about the scalability of these percentages is that the more directly interactional the communication becomes, the more clausal and the less phrasal postmodification occurs. It seems that in the direct communication situation, the weight of a whole clause - a whole proposition about the antecedent - is useful, either as an aid to processing for the hearer or as a means of buying think time for the speaker.6 So far we have included very little breakdown in this comparison, and so one may be tempted to say we have only a basic distinction between written and spoken language here but the suspicion that further types of data will allow an even more satisfying scalar assessment once the hierarchy is given the proper name remains (cf. Biber 1986 as well as Biber and Finegan 1988 for a step in this direction). That the channel of communication is a factor to be considered in assessing language production and understanding would seem to be a claim not easily dismissed.

2.

The data

The study comprises a total of 2693 relative clauses in two corpora. Corpus I:

1988 restrictive relative clauses gathered from:

spoken A conversation among friends (cassettes, dialogue between blues musician and audience) B conversation among colleagues (taped radio panel, transcript of panel discussion) C informal interview (radio interviews on non-academic, non-formal matters, interviewees are not professional talkers) D formal interview (radio interviews on academic or formal matters, speakers are professional talkers)

6

Our data are all from adult speakers. Although with very small N, the data in Wenzell (1988) appear to show the opposite tendency for children.

261

written E fiction, dialogues or first-person narratives F personal letters H7ads I letters to the editor: college paper, Stars & Stripes J letters to the editor: Radio Times, Ebony K letters to the editor: Guardian, Economist L reports, academic prose Corpus II: 2401 instances of restrictive postmodification (clauses and phrases) gathered from the following sources: spoken B, C, D: radio talk shows, panel discussions, live commentary of Constitution Day parade, interviews conducted as part of a research project on soap operas written E fiction: short stories L non-fiction: newspaper and magazine articles H ads (commercials) broadcast on radio The 705 relative clauses included in these data were combined with those from Corpus I and coded to match them for the following features: dependent variable: choice of κΆ-, that, 0 relative, Ζ (= 0 relative + omitted be) independent variables: 1. role of antecedent in matrix clause 2. role of relative in its clause 3. intervening material between antecedent and relative 4. humanness of antecedent 5. style (cf. A - L above) 6. number of antecedent (singular/plural) 7. definiteness of antecedent

3.

Generalities8

3.1

Antecedents tend to be non-subjects

The first general observation that should be pointed out is that the antecedents of relative clauses tend to be non-subjects. In our combined data (we refer always to the combined data unless otherwise specified) there are only 12% subject antecedents - 13% if we include passives - while 33% of the antecedents occur in prepositional phrases, 23% as objects and 21% as predicate nominals.

There is no category G. 8

These pertain to the independent variables 1 - 3 (cf. 2. The data.)

262

This is not really surprising when one considers the fact that only one subject per clause can be eligible for postmodification, whereas there are multiple chances (in principle) for each of the other mentioned categories. Quirk (1957:104) already observed that "the majority of relative clauses are placed after the completion of the clause containing the antecedent; this means in effect that relative clauses normally qualify the object or complement rather than the subject of a main clause." This may be due to a general tendency reported also by Taglicht (1977:100) that lighter constructions tend to come earlier and heavy ones later, but it may also derive from the fact that the subject is normally the entity in focus and thus already specified in such a way that the hearer has no difficulty identifying its referent. It is in the "activation state" that Chafe (1984:9) terms "active," i.e. it is "lit up," or "in a person's focus of consciousness," at that point in the discourse, which, in effect, makes it a poor candidate for restrictive postmodification. In contrast, entities that have not already been focussed on are likely to require specification. 3.2

Relatives tend to be subjects

The relatives themselves, on the other hand, tend to be subjects of their clause. We find that 46% (56% if we include passives, but there may be good reason not to do so, as we shall see below) are subjects, i.e., in focus in their clause. This is in keeping with the flagging function of the relative clause, its signalling "look here, this is the way you can find the referent of the antecedent" (or in the case of an indefinite antecedent, as in (8), "this is the way to restrict the set in terms of which the referent is to be understood"): W

3.3

(8) [...] an irritating performance from an actor who is perfectly capable of better things, (non-fiction = L) Intervening material

As far as the factor of intervening material is concerned, two observations will suffice: You don't get much material intervening between the antecedent and its postmodifying relative clause - and apart from a few superordinate and coordinate relatives, an occasional adverbial or interjection, and a vanishing number of verbs, what you do get is prepositional phrases, mostly of+ NP: T

(9) That is only one of the dramatic aspects of her that fascinates Hollywood mythmakers [...] (non-fiction = L) Contrary to what Quirk et al. (1972:871) suggest, which is that intervening material favors which, in our data intervening material is pretty equally distributed between following that and H-A-relatives.

263 With 0 relatives or when both the relative and be are omitted (what we have labelled [Z]) intervening prepositional phrases occur in much smaller numbers. Here, even more than with other NP + PP combinations, the analyst becomes acutely aware of the problem of deciding whether the PP is indeed intervening material or part of the antecedent nominal. At any rate, one expects intervening material much less with the omission constructions (cf. (10)) than when there is an expressed relative pronoun, because perception of the link between the antecedent and the relative clause is already more difficult. Z

3.4

(10)These were the findings of a medical team after a study of Pepto-Bismol's effectiveness against symptoms of upset stomach [Z] associated with overindulgence in food and drink, (ads = H) Omission

One further general point to be made is that it seems more plausible to view the omission cases not as 'deletion of a relative pronoun' but as an equally valid and viable choice. Perhaps one should rather consider the occurrence of that and the tv/i-relatives as the realization of a relative pronoun than to see 0 and Z as deletion of one (the difficulty in many cases of deciding just which pronoun has been 'deleted' is just one argument in favor of this).9

4.

Specifics

4.1

Role of antecedent: role of relative

When we consider the interaction of the role of the antecedent in the matrix clause with the role of the relative in its own clause, we find that when the antecedent is the subject - the entity in focus - there is a carryover of this role to the relative 60% of the time. Τ

(!!)[...], the superpowers that share the awesome ability [...] will have an actual commitment [...] (non-fiction = L) This is true also if the antecedent is a predicate nominal (i.e., in cases of double mention of the antecedent), where the relative is subject of its clause 58% of the time. W

(12) [...] a good arbitrator in sports is someone who'd know very little about baseball, (call-in radio talk show, not a professional talker = Q This tendency, considering that we have only 12% subject antecedents all told, is of course also strengthened by the general predominance of subject relatives.

This applies to Z (0 + omitted be) as well.

264

4.2

Role of relative : humanness of antecedent

Although most relatives function as subjects in their clause - that is, they are in focus - this overall patterning derives from a rather skewed distribution when we consider the interface of the role of the relative with the humanness of the antecedent, as we can see from Table 2. Table 2: Interface of role of relative with humanness Human

non-Human

Institution10

N

%

N

%

N

%

Subject

629

82%

566

37%

45

60%

Object

61

567

8

Passive

36

225

6

inPP

36

164

11

762

1522

70

There is a clear correlation between a human antecedent and a subject relative. This correlation is motivated by the humanness factor, for we can say that given a human antecedent, there is a very strong likelihood (82%) that the relative will be functioning as subject in its clause, but not the other way around. Given a subject relative, chances are only slightly better than 50-50 (cf. 629 and 566) that the referent will be human. Being human ourselves, we are wont to talk about human beings, and to attribute to them the importance of being entities in focus. What do these observations (in 1. and 2. above) imply for choice of relative? Since both subjecthood of the antecedent and humanness of the antecedent favor subjecthood of the relative, we might expect to find importance attributed to that relative as well. In the top half of Table 3 we see that 77% of the subject relatives that refer to humans are wh-forms. The κΆ-relative allows us to exercise the most precision because it offers the option of distinguishing focus from non-focus (who vs. whom) and forces a choice between human and non-human (who vs. which, etc.). The that is less precise (while nevertheless clearly pointing, through its mention), and omission is least precise of all. Therefore the choice of who

Institution' includes antecedents such as family, society, government and corporation.

265

attributes due importance to its referent and makes it perfectly clear that the entity in focus is also one of the hallowed lot of human beings. Table 3: Crosstabulation of role of relative with humanness Non-human Subject

T 402 71% W 145 26% 0 4 1 % Ζ 15 3% Tot. 566

Object

T W 0 Z Tot.

197 29 341 0 567

35% 5% 60% 0%

Human

Institution

Totals

106 485 8 30 629

17% 77% 1% 5%

28 12 Ο 5 45

62% 27% 0% 11%

536 642 12 50 1240

43% 52% 1% 4%

11 8 42 0 61

18% 13% 69% 0%

l 0 7 0 8

13% 0% 88% 0%

209 37 390 0 636

33% 6% 61% 0%

There is another correlation based on the role of the relative that is, however, independent of humanness: given an object relative, there is a very strong likelihood that it will be rendered as 0 (61%) or that (33%) (cf. Totals column in Table 3, bottom half). This is also in keeping with the communicative tack we have been taking, for it lies in the nature of discourse that non-focus elements needn't be especially mentioned or, if they are mentioned, that they can be referred to with less precision. 4.3

Role of relative : style

The remaining aspects of relative distribution that we wish to consider have to do with the interaction of role of the relative with style (i.e., with the different text sources). In Table 4, we are struck by a patterning that is at first sight unexpected. 4.3.1 Passive subjects behave as logical objects We find that the grammatical subjects of passive relative clauses do not pattern like active subjects. The active subjects - especially in the written styles - show a strong tendency to be expressed as ννΛ-relatives, but the passives pattern like object relatives in that they exhibit little affinity for w/i-forms. Passive relative subjects tend rather to be omitted (in this case with concomitant omission of be, i.e. our [Z] examples) in the written varieties (cf. (6), repeated below for convenience).

266

(6)

[...] nowhere in England or Scotland or Ireland is there a form of educated speech [Z] sufficiently unified and [Z] used by a sufficient number of speakers to form a rival standard, (non-fiction = L)

In three out of four spoken varieties (B, C and D) that is the form that occurs predominantly in passive relative clauses (cf. (7), also repeated here).11 (7)

Kathy Craig [...] wrote this article that was talked about on the show, (informal interview = C)

The affinity of object relatives to 0 is even stronger, in all styles except A (conversation among friends) - for which we have very little data all told, which may be due to the fact that conversation among friends is an unlikely breeding ground for relative constructions, the major flagging function of relatives being least required when the interlocutors know each other well, sharing a good deal of their world knowledge and experience.

Cf. Table 4 for

details. (Derwing 1988 also mentions the effects of this shared experience on discourse.) Table 4: Crosstabulation of role of relative with style Subject

Object

Passive

Prep.Ph.

Adv.Ph.

Totals

Spoken

A:

39% 46% 11% 4%

19 0 11 0 30

63% 0% 37% 0%

1 20% 1 20% 1 20% 2 40% 5

10 0 6 0 16

63% 0% 38% 0%

1 7 3 0 11

9% 64% 27% 0%

49 29 26 4 108

45% 27% 24% 4%

Tot.

18 21 5 2 46

B: T ConverW sation 0 among Z colleagues Tot.

32 47 1 1 81

40% 58% 1% 1%

8 30% 1 4% 18 67% 0 0% 27

5 56% 1 11% 0 0% 3 33% 9

3 0 1 0 4

75% 0% 25% 0%

0 4 3 0 7

0% 57% 43% 0%

48 53 23 4 128

38% 41% 18% 3%

T W 0 Z Tot.

155 73 4 7 239

65% 31% 2% 3%

29 8 21 0 58

50% 14% 36% 0%

8 47 32 0 87

9% 54% 37% 0%

272 132 145 13

48% 23% 26% 2%

T W 0 Z Tot.

104 42% 136 54% 1 0% 9 4% 250

17 13 17 0 47

36% 28% 36% 0%

6 3 8 0 17

35% 18% 47% 0%

Conversation among friends

C: Informal interview

D: Formal interview

1

T W 0

z

64 2 88 0 154

42% 1% 57% 0%

52 5 53 0 110

47% 5% 48% 0%

16 2 0 6 24

67% 8% 0% 25%

22 13 0 14 49

45% 27% 0% 29%

562 201 170 79 23

42% 36% 17% 5%

473

The spoken varieties also include passive [Z] examples like (i) But don't buy the first policy [Z] offered to you. or the single occurring example of passive [0] from our data: (ii) There's a club down there [...], which was actually an Irish pub, you know, [0] was named after an old Irish fiddle tune. (A)

267

Subject

Object

Passive

Prep.Ph.

Totais

Adv.Ph.

Written

E: Fiction

72 48 0 11 131

55% 37% 0% 8%

14 8 62 0 84

17% 10% 74% 0%

0 2 0 7 9

0% 22% 0% 78%

1 8 8 0 17

6% 47% 47% 0%

0 11 10 0 21

0% 52% 48% 0%

87 77 80 18 262

33% 29% 31% 7%

4 11 0 1 16

25% 69% 0% 6%

10 3 31 0 44

23% 7% 70% 0%

2 29% 1 14% 0 0% 4 57% 7

0 2 4 0 6

0% 33% 67% 0%

0 3 2 0 5

0% 60% 40% 0%

16 20 37 5 78

21% 26% 47% 6%

T W 0

z

20 53 0 2

27% 71% 0% 3%

27% 15% 58% 0%

1 6 2 0 9

11% 67% 22% 0%

0 2 1 0 3

0% 67% 33% 0%

32 69 18 18 137

23% 50% 13% 13%

75

4 4 0 16 24

17% 17% 0% 67%

Tot.

7 4 15 0 26

T W 0

21 69 0 4 94

22% 73% 0% 4%

9 4 29 0 42

21% 10% 69% 0%

3 17% 2 11% 0 0% 13 72% 18

2 7 3 0 12

17% 58% 25% 0%

0 1 1 0 2

0% 50% 50% 0%

35 83 33 17 168

21% 49% 20% 10%

23 55 0 10 88

26% 63% 0% 11%

12 4 13 0 29

41% 14% 45% 0%

5 7 0 24 36

14% 19% 0% 67%

1 12 1 0 14

7% 86% 7% 0%

0% 0 5 100% 0 0% 0% 0 5

41 83 14 34 172

24% 48% 8% 20%

51 142 1 3 197

26% 72% 1% 2%

11 5 38 0 54

20% 9% 70% 0%

6 14 0 55 75

8% 19% 0% 73%

2 23 2 0 27

7% 85% 7% 0%

6% 65% 29% 0%

71 195 46 58 370

19% 53% 12% 16%

47 9 0 0 56

84% 16% 0% 0%

3 8% 1 3% 33 89% 0 0% 37

0 0% 2 100% 0% 0 0% 0 2

51 12 34 10 107

48% 11% 32% 9%

547 664 12 50 1273

43% 52% 1% 4%

903 923 535 204 2565

35% 36% 21% 8%

T W 0

z

Tot.

F: Letters to friends

T W 0

z Tot.

I: Letters to editor J: Letters to editor

z Tot.

K: Letters to editor

T W 0

z

Tot.

L: NonFiction

T W 0

z

Tot.

H: Ads

T W 0

z

Tot.

Totals

T W

0

z Tot.

209 37 391 0 637

33% 6% 61% 0%

1 9% 0 0% 0 0% 10 91% 11

65 24% 47 18% 1 0% 154 58% 267

0 0% 0 0% 1 100% 0% 0 1

66 79 66 0 211

31% 37% 31% 0%

1 11 5 0 17

16 96 65 0 177

9% 54% 37% 0%

268 This patterning of passives with objects may well be a natural concomitant of their being logical or semantic objects, having been promoted to grammatical subject status as a function of discourse considerations. In the following discussion, we group passives with objects to make the semantic distribution pattern clearer. 43.2 Subjects, humanness and style Looking at the effects of the interaction of the role of the relative and the style of the data on the choice of relative, we can see from the Totals column in Table 5 that for subject pronouns - this table includes only subject pronouns - it makes a decided difference whether the source is spoken or written data. Table 5: Crosstabulation of humanness with style (for relative subjects) Human Spoken

T W 0

z

Tot. Written

τ

W 0

z

Tot. Totals

T W 0

z

Tot.

Non-human

Institution

Totals

87 236 8 14 345 [56%]

25% 68% 2% 4%

210 36 3 3 252

83% 14% 1% 1%

12 5 0 2 19

63% 26% 0% 11%

309 277 11 19 616

50% 45% 2% 3%

23 267 0 16 306 [47%]

8% 87% 0% 5%

199 112 1 12 324

61% 35% 0% 4%

16 8 0 3 27

59% 30% 0% 11%

238 387 1 31 657

36% 59% 0% 5%

110 503 8 30 651

17% 77% 1% 5%

409 148 4 15 576

71% 26% 1% 3%

28 13 0 5 46

61% 28% 0% 11%

547 664 12 50 1273

43% 52% 1% 4%

The written variety prefers wh- decidedly (cf. 59% vs. 36% in the totals column) - and the spoken variety prefers that - but not by much, while both varieties virtually demand an expressed relative pronoun (cf. the vanishingly low percentages of 0 and Z in the totals column). This is particularly interesting when we consider the fact that there are fewer human antecedents in the written variety (47%) than in the spoken data (56%). If there is a human antecedent, in writing this antecedent is referred to with who 87% of the time (that following a very poor second with 8% and there being no 0 at all), while in

269 spoken discourse, these antecedents rate a who only 68% of the time, more closely followed by thai at 25% and with 2% 0 and 4% Ζ [Ο relative + omission of be] occurring (cf. Table 5). These data nicely reflect a developing tendency in the spoken language to replace who with that - a tendency that is particularly marked in more casual styles, such as our informalinterview and conversation-among-friends categories (cf. Table 6). Table 6: Crosstabulation of humanness with spoken style (detail of "spoken" from Table 5) Human

Non-human

Institution

Totals

T W 0 Z Tot.

7 17 4 2 30

23% 57% 13% 7%

11 3 1 0 15

73% 20% 7% 0%

0 1 0 0 1

0% 100% 0% 0%

18 21 5 2 46

39% 46% 11% 4%

converT W sation 0 among colleagues z Tot.

3 45 0 1 49

6% 92% 0% 2%

28 2 1 0 31

90% 6% 3% 0%

1 0 0 0 1

100% 0% 0% 0%

32 47 1 1 81

40% 58% 1% 1%

57 67 3 5 132

43% 51% 2% 4%

95 4 1 2 102

93% 4% 1% 2%

3 2 0 0 5

60% 40% 0% 0%

155 73 4 7 239

65% 31% 2% 3%

20 107 1 6 134

15% 80% 1% 4%

76 27 0 1 104

73% 26% 0% 1%

8 2 0 2 12

67% 17% 0% 17%

104 136 1 9 250

42% 54% 0% 4%

conversation among friends

informal interview

T W 0

z

Tot. formal interview

T W 0

z

Tot.

In object contexts this tendency to replace who with that may be attributable to the fact that that is 'stylistically neutral,' whereas with the wA-form one must decide between 'colloquial' who and 'formal' whom (Ungerer et al. 1984:204). 4.3 J Objects and omissions On the other hand, we can see from Table 7 below that when the relative is object in its clause, omission takes on an important role, by far the greatest in the written varieties, and one virtually as strong as the role of that in the spoken varieties.

270

Table 7: Crosstabulation of role of relative with style12 Subject Written

Totais

Totals

36% 59% 0% 5%

87 59 221 129 496

18% 12% 45% 26%

325 446 222 160 1153

28% 39% 19% 14%

Tot.

238 387 1 31 657

T W 0 Z Tot.

309 277 11 19 616

50% 45% 2% 3%

187 25 171 25 408

46% 6% 42% 6%

496 302 182 44 1024

48% 29% 18% 4%

T W 0

547 664 12 50 1273

43% 52% 1% 4%

274 84 392 154 904

30% 9% 43% 17%

821 748 404 204 2177

38% 34% 19% 9%

T W 0

z

Spoken

Object

z

Tot.

The 0 suggests that the speaker finds it unnecessary to reorient the hearer by 'pointing towards' the referent again. In these cases there is another entity involved (the subject of the relative clause) and the referent of the antecedent is to be identified via its relationship to this other entity. Since it is the crux of the identification procedure, it is natural for it to be placed in focus in the discourse via the subject role. When we took a look at what kind of subjects follow 0 (object) relatives in our Corpus II data, we discovered that 90% of them are personal pronouns (cf. (13)): (13)

[...] and the truth [0] he embraced became a falsehood, (fiction = E)

and 64% of these are first and second person references, as in (14)

So that's why he doesn't know all the people [0] Fm talking about when I'm trying to refer back to that particular soap, (informal interview = C)

i.e., references to orienting entities that are most pertinent to the discourse because they are most readily identifiable. As Chafe (1984:11) put it, "First and second person referents acquire the active status naturally from the conversational context itself."

Table 7 is a condensation of the data quantified in detail in Table 4.

271

For passives - included with objects in Table 7 - the picture is only slightly different: what the omitted relative (the logical object) is being placed in relation to is the verbal proposition, the event rather than an entity. In (15) and (16) it doesn't matter who writes or offers the insurance. What matters is that it was written in the U.S. or that it is being offered. (15) (16)

90% of all term insurance [Z] written in the U.S. never pays off. (call-in radio talk show, professional talker = D) But don't buy the first policy [Z] offered to you. (call-in radio talk show, professional talker = D)

Although there is this difference in what the relative is being related to, for both objects and passives the omission options underscore the fact that identification of the antecedent's referent is to proceed via that referent's relationship with something else - either another entity or an event.

This strategy appears to be independent of style (cf. the similarity in the

percentages in Table 7). So, while we feel that we have demonstrated the existence of recognizable patterns in the distribution of relatives, and while we are of the opinion that these patterns cannot be adequately described and explained by referring to the simple distinction between oral and written discourse alone - i.e. that they must be seen in light of the nature of the linguistic sign and its interaction with communicative and stylistic considerations - we recognize that this is an area of empirical research not yet clearly defined. The observable patterning derives from the fact that speakers and writers are, after all, human beings with something to say that they wish the hearer/reader to understand, i.e. common strategies are useful. The fuzzy edges come from the fact that individuals may also fall back on very idiosyncratic strategies to achieve their communicative goals.13

1·* Although a number of tendencies are recognizable that appear to be dependent on stylistic factors, as we have seen, we must not lose sight of the fact that style may also be influenced by factors that lie outside the scope of linguistic generalizations. We would like to mention just two of these: the first is that publishers sometimes dictate peculiarities of style - some of the leading American magazines and newspapers have an editorial policy that says, "restrictive relative (non-human) = that I non-restrictive relative = which' And second, there is in the last analysis no getting around personal style, idiolect, if you will - there may be individual speakers that have a distribution pattern all their own. Certain preferences can be noticed even on a cursory glance: in our fiction data, Edgar Allen Poe's preference for which as a non-human relative pronoun was outstanding - his short story "The Oval Portrait" contains 18 restrictive wA/c/ies, no thats and no cases of 0!

272

Bibliography Biber, D. (1986): "Spoken and Written Textual Dimensions in English." - In: Language 62,384-414. Biber, D. & E. Finegan (1988): Neither Speech Versus Writing Nor Orality Versus Literacy: Abandoning the Garden Path. - Contribution to 17th Annual University of Wisconsin-Milwaukee Linguistics Symposium. Chafe, W. (1982): "Integration and Involvement in Speaking, Writing and Oral Literature." - In: D. Tannen (ed.): Spoken and Written Language: Exploring Orality and Literacy (Norwood, N.J.) 35-53. Chafe, W. (1984): Cognitive Constraints on Information Flow. -Berkeley (Berkeley Cognitive Science Report No. 26). Chafe, W. (1985): "Linguistic Differences Produced by Differences Between Speaking and Writing." - In: D.R. Olson, N. Torrance & A. Hildyard (eds.): Literacy, Language and Learning (Cambridge) 105-23. Chafe, W. (1988): Information Flow in Speaking and Writing. - Contribution to 17th Annual University of Wisconsin-Milwaukee Linguistics Symposium. Danielewicz, J. (1988): Sentences in Children's Spoken and Written Language. - Contribution to 17th Annual University of Wisconsin-Milwaukee Linguistics Symposium. Denying, B.L. (1988): Orthographic Aspects of Linguistic Competence. - Contribution to 17th Annual University of Wisconsin-Milwaukee Linguistics Symposium. Ochs, E. (1979): "Planned and Unplanned Discourse." - In: T. Givon (ed.): Syntax and Semantics 12: Discourse and Syntax (New York) 51-80. Poe, E.A. (1986): "The Oval Portrait." - In: E.A. Poe: The Fall of the House of Usher and Other Writings (London) 290-92. Quirk, R. (1957): "Relative Clauses in Educated Spoken English." - In: English Studies 38,97-109. Quirk, R., S. Greenbaum, G. Leech & J. Svartvik (1972): A Grammar of Contemporary English, - New York/London. Taglicht, J. (1977): "Relative Clauses as PostmodiGers: Meaning, Syntax and Intonation." - In: W.-D. Bald & R. Ilson (eds.): Studies in English Usage: The Resources of Present-Day English - Corpus for Linguistic Analysis (Frankfurt/Main) 73-107. Ungerer, F., G.E.H. Meier, K. Schaefer, & S.B. Lechler (1984): A Grammar of Present-Day English. - Stuttgart. Wenzel!, V. (1988): Differences in the Spoken/Written & Unplanned/ Planned Classroom Language of Native and Non-Native English Speaking Children. - Contribution to 17th Annual University of WisconsinMilwaukee Linguistics Symposium.

273

Uwe Mönnich

On the Tension between Properties and Prepositional Functions1 The present essay2 centers on a case of doctrinal bifurcation. It concerns the claim that a certain theoretical framework, the framework which construes properties as prepositional functions, threatens the prospect of an otherwise promising attempt at a solution of the problem presented by propositional attitudes. I hope to be able to show that this claim is predicated upon a confusion between internal and external semantics and that even within the realm of external semantics technical solutions for the propositional attitudes problem can be worked out in which the notion of a function is taken as a primitive concept.

1.

Attitudes towards possible worlds

A little story recently told by Hans Kamp (1990) makes particularly vivid the problem posed by the propositional attitudes. Calvin and Kevin, two friends, each ten years of age, are left by Kevin's father at the entrance to Disneyland. They are allowed to spend the afternoon together and Kevin's father provides each of them with twenty dollars. At the end of the afternoon, on their return to the entrance, Kevin, who is feeling very hungry, walks up to a hamburger stand and orders one of those appealing items which, as a sign informs the potential customer, cost $1.50 a piece. When he goes to pay for the ordered hamburger, he discovers to his embarrassment that he has only 60c left and thus cannot pay for the much coveted food. Calvin, who is just as hungry as his friend, spares himself the embarrassing experience at the hamburger stand because he knows that he has only 600 left. During the afternoon the two friends spent their money on the same attractions. From the point of view of folk psychology there does not seem to be any problem with this little story. Calvin is aware of a fact which keeps him from making the futile attempt at buying a hamburger which puts his friend into such an awkward situation. Both friends act as rational creatures, guided in their practical reasoning by principles which must have

It is a great pleasure to contribute to a volume dedicated to David Reibet. I remember with gratitude the many discussions on linguistic matters we have had during the last fifteen years. The enormous influence they had on me was not only due to the particular points raised by David but also to the specific way in which he tackled linguistic issues. I consider David's attitude towards linguistics a clear example of an unbiased theoretical liberalism in the sciences. 1 am very much indebted to Elsa Lattey for comments and for stylistic matters.

274

constrained human behaviour since times immemorial and which were first codified by Aristotle. According to these principles, we expect our fellow human beings to act in such a way that, given a certain set of beliefs, their relevant desires are furthered to the utmost. We have been fairly successful in our predictions or explanations of the behaviour of people we have to deal with and our little story provides us with a telling example. The justified belief that Calvin has and that Kevin is lacking explains the difference in the course of action each boy thinks suitable for the maximization of his desires.

They are faced with two alternative

courses of action at a certain moment of their lives: either proceeding with the purchase of a hamburger to soothe their feeling of hunger or refraining from such a purchase. In Calvin's case his well-founded belief about the sum remaining in his pocket is causally relevant for his decision not to walk up to the hamburger stand. Any respectable theory of belief should be able to do justice to this explanation derived from the concepts of folk psychology and it should explicate the special role the notion of belief as a causally relevant agent plays in this explanatory account. The leading semantic model, the possible worlds theory, fails in precisely this. According to this model, which within linguistics has found its canonical incarnation in Montague Grammar, sentences express propositions and propositions are identified with the possible worlds or situations in which they are true. Now, if we attribute to Kevin the belief that he has spent some of his money on the different attractions in Disneyland that he has visited together with his friend Calvin, then this belief is the same as the one which spared Calvin the embarrassing experience at the hamburger stand: since the second belief is logically entailed by the first belief, the two are indistinguishable with respect to the set of possible worlds in which they are validated. Anyone having the first belief, so the possible worlds account goes, must by the same token be credited with the second belief, whose content specified the exact sum of the money which was left in the boys' pockets. We are deprived by possible worlds semantics of any difference between the belief held by the two boys which would explain their different behaviour. The theory of possible worlds is not to be dismissed lightly on account of one recalcitrant example. As was remarked above, the general account of practical reasoning which regards humans as rational agents who seek to maximize their desires as far as the circumstances and their overall conception, i.e. belief-structure, of the world permit has been a fairly reliable guiding principle in the assessment of the behaviour of our friends and foes. Underlying this model of human behaviour is the intuition that we can make informed decisions when confronted with different options for further action. We classify possible future courses of events according to their conformity with our pursuit of happiness. It is a sad fact that we often make mistakes in interpreting our beliefs and desires and that even more often we com-

275

pletely miscalculate the intentions of others, but mistakes can only be made against the background of a satisfying general account, and the possible world account, through its identification of (the meaning of sentences expressing) beliefs and desires with sets of possible worlds - which for the sake of argument can be understood as sets of possible courses of action - seems to offer just such a general account. Unfortunately, though, it turns out that the method I have used for constructing a counterexample to the possible worlds construal of propositions is completely general in the sense that every substitution for each other of logical equivalents leads to a problem for the possible worlds framework when applied to the propositional attitudes. The crucial step in Hans Kamp's little story has been the logical equivalence of the belief that the sum total of money spent on rides, icecream, shows etc. equals $19.40 and the other belief that only 60l). As long as a formula or term does not contain an occurrence of an abstract term, the truth definition and the denotation definition go through in the familiar fashion. In the case of abstracts the route to the expected reading is barred, however. λζ[Α] does not denote the function which sends an arbitrary individual a into the proposition denoted by A(a/z), the result of substituting every occurrence of the variable ζ by a, but rather denotes that element in the family of properties which is the value of the algebraic expression in the model structure as which λζ[Α] can be analyzed in a unique way by means of the Quinean predicate functors. For illustrative purposes take the abstract Xz[Bel(j,Spy(z))] standing for the property of being believed by John to be a spy. The corresponding algebraic expression has the following form: (13)

RelPred(Pred(Xzy[Bel(z,y),j]),Xx[Spy(x)])

where Pred stands for the predication and RelPred for the relative predication relation respectively: the technical complexity of this simple example grows into disproportionate dimensions once one considers expressions of a more involved structure. What is needed in this situation is a conception of content that goes beyond a mere homomorphism between a term and a content structure as illustrated by Beater's translation of his Quinean abstracts into elements of a suitably chosen meaning algebra. The clue to the solution of this problem was already indicated when we talked about abstracts as names for functions. As adumbrated by our little example, if we revert to a Fregean terminology and treat senses of predicate expressions as functions which send the meanings of their argument expressions into propositions, we have circumvented the complicated instrument of predicate functor logic. Indeed, by connecting the familiar formal structure of the formulae of a first order logic with abstracts to its interpretation by way of recursion clauses which make the deviation through the functorial expressions superfluous, the usual generation tree of syntactic objects can be retained as a climbing-frame for the semantic procedure. Or so it seems. The relevant literature contains some arguments which appear to show that despite its technical advantages, the correlation between properties and prepositional

282

functions leads to serious logical difficulties. As far as I can tell, these arguments reduce to two general types, one of them originating with Anderson (1986) and the other with Bealer (1989). Stripped of all irrelevant details, the first argument boils down to the following: This argument avails itself of the notion of a constant propositional function, a function that is characterized by the fact that it expresses the same proposition on application to any argument whatsoever. Let us suppose the existence of such a constant function by the following comprehension schema: (14)

(3f)(x) (f(x) = A)

If A stands for the proposition that Socrates flies, then f, when applied to an arbitrary object, say, the number nine, denotes that very proposition. This state of affairs is not changed when the number nine is substituted by a certain hummingbird. Given its definition, the expression 'f (9)' stands for the same proposition as 'f (a)1, where α is a name for the selected hummingbird. Both expressions have exactly the same meaning, that is, the proposition that Socrates flies. Now it should be obvious that this is unacceptable. If one is to believe, be certain of, or know a priori a simple predication, and this simple predication can be rendered in two different ways without affecting the truth value of the attitude sentence, then its constituents must be the same. That is to say, a proposition that has the form of a predication, pred(f,a), should be regarded as having an internal structure from which the subject a and the predicate/can be recovered: (15)

pred(f,a) = pred(g,b) -> f = g & a = b

The switch from the notational form 'f (a)', indicating functional application, to the predicational notation 'pred (f,a)' was intentional in order to forestall the objection that the debate is preempted by a suggestive notational device. Whether properties are functions and predication is function application is the issue the constant function argument is meant to decide. But, we have reached a stage in that argument where the position of the propositional function thesis, the thesis that propounds a strong relation between properties and propositional functions, appears to be desperate. By our comprehension principle, someone who believes that f, representing the property, i.e. constant function, of being a z such that Socrates flies, can be truly predicated of a certain hummingbird a, may be said, with equal justification, to assent to the proposition to the effect that the number nine is f, an absurd conclusion.

What this seems to show is that we have to choose between the following

alternatives: We can identify properties with propositional functions if we introduce a new predication relation Pred that is to be distinguished from functional application. Once we explain the predication relation as a nexus between a special type of saturated objects, unary relations or

283

properties, and their arguments this nexus can be imagined as resulting from functionally applying Pred to an ordered pair of arguments. Retaining both the prepositional function thesis and the identification of function application with predication leads to a levelling of content structure that is unfaithful to the requirements of a semantics for prepositional attitudes: someone who staunchly defends the status of properties as saturated objects has an easy way out. He can always claim that Pred introduces an additional structural element and that consequently Pred(f,a) and Pred(f,g) are different propositions. Or, we can exploit the distinction created by the different order in which the argument positions of a two-place relation like following can be discharged.4 The order of the steps in partially saturating an n-ary predicate (n >1) must be recoverable from the syntactic pattern of the resulting expression. As will become evident in a moment, the syntatic structure codetermines the prepositional content prepositional attitudes are able to reflect. Let 'R1 now be a two-place predicate and a and b two different objects. Comprehension allows us to introduce by stipulation two properties P = Xx[R(x,b)] and Q = Xy[R(a,y)]. If properties are indeed identical to prepositional functions and if the general characteristics of functions find an adequate expression in the lambda-calculus, then the following equations are instances of the rule of -conversion and of the intersubstitutivity of definitional identities: (16) P(a) = Xx[R(x,b)](a) = R(a,b) = Xy[R(a,y)](b) = Q(b) To illustrate the fact that these equations model a far too coarse-grained notion of proposition, let R express the relation of following and a and b stand for Jane Fonda and Rajneesh, respectively. Then we can stipulate that (17) and

Being an χ such that χ rajneeshes = being an χ such that χ follows Rajneesh (=P)

(18)

Being a y such that y fondalees = being a y such that Jane Fonda follows y (=Q)

But this seems wrong. When a person consciously and explicitly thinks that Jane Fonda rajneeshes must that person be consciously and explicitly thinking that Rajneesh fondalees? It certainly does not seem so. The order of predication appears to be a relevant factor of prepositional identity. At first sight, this untoward result gives the impression that it is avoidable by reverting to an 'intensional1 model in which the way an operation is given constitutes part of its meaning. In such a model the 'subject-first1 and the Object-first1 R could be distinguished even though both operations would have the same value for all pairs of objects z: 4

Cf. Section III of Bealer and M nnich (1989).

284

(19)

- Γ(ζ) (Subject-first R(z) = Object-first R(z)) -> λζ [Subject-first R(z)] = Az[Object-first R(z)]]

This intensional notion of a function that incorporates aspects of sequential behaviour is of no help with respect to our problem of the internal structure of propositions in prepositional attitude contexts. If P(a) and Q(b) are the same proposition, then the fact that the dynamic behaviour of Ρ is to be analyzed in terms of Object-first R and the dynamic behaviour of Q is to be analyzed in terms of Subject-first R gives us no handle on the resulting proposition unless these sequential stages can be discerned as parts of the final value. But this is denied by the presupposition that the input-output behaviour of the two operations is the same. Proponents of the propositional-function thesis seem to be caught in a dilemma. On the one hand, they have an elegant solution to the worries that have plagued those scholars who tried to reconstruct intensional objects out of the primitive elements provided by the possible worlds approach.

They have the option of relying upon a principle of individuation for

intensions that is finer than logical equivalence, the latter notion being reduced to identity of truth value in all possible worlds in the case of propositions. It is certainly true that anything that is triangular in a possible world is trilateral in that same world. But this does not imply that for any object χ the proposition that χ is triangular is the same proposition as that χ is trilateral. When a person consciously and explicitly thinks or believes that χ is triangular, must that person be consciously and explicitly be thinking or believing that χ is trilateral? It certainly does not seem so. The propositional-function thesis does not entail the unwanted identifications implicit in the possible worlds semantics. Thus, on this score, there is no objection against an interpretation that models properties as propositional functions and interprets the operation of predication as functional application. On the other hand, the propositional-function thesis still seems to imply a sort of unwanted identification illustrated by the constant function and the sequentiality of argument discharge problems. Both problems involve the comprehension principle that licences the formation of abstract terms and the principle of -conversion that gives the result of functionally applying the function-abstract to its argument.5 Both principles are operative in the semantic conflation of syntactically distinct expressions. Since I see no way of tampering with either lambdaabstraction or functional application and their attending principles without violating in essential respects the basic intuition that guides our understanding of the notion of function, the propositional-function thesis appears to be mistaken and its technical advantages outweighed by its factual shortcomings.

* In the case of the constant function, β - conversion assumes the trivial form Xx[A](a) = [A] since, by hypothesis, χ does not occur free in A.

285 3.

Internal versus external semantics

We have seen above that, within the possible worlds tradition, semantics was carried out directly within set theory.

Semantic analyses, on this view, are to be developed in the

theoretical framework of set theory enriched by certain atoms such as individuals and possible worlds. The universe of sets provided all the structure that was needed to spell out the meaning of expressions and to specify the relations between them. The technical system of set theory was used to model the meaning of the language described. As Barwise (1987) emphasises in his notes on Situation Theory, we must not forget about the difference between a model and the theory it is supposed to model. There is no danger of confusion in the case of mechanical devices.

No one would take an orrery for the solar

system whose theory it is meant to exemplify, let alone for the theory itself. The danger of confusion increases when the theory whose models we are studying serves as an account of the abstract structures underlying our understanding of the semantic functions of both natural and artificial languages. In this abstract realm, it is not uncommon for set theoretic models to be taken for the ultimate theoretical framework that cannot be transcended.

To put it

differently, it is not always easy to tell whether the properties of the model belong to the intended interpretation of the theory the model is to visualise, or whether these properties result from artifacts of the modeling process. Barwise, who stresses the artificiality of model building, illustrates the fallacious identification of parts of a certain model with elements of the intended interpretation by means of the prepositional-function thesis, "For example, when we say that a model of something or other is an ordered triple, we normally understand that the thing itself is not to be considered an ordered triple, but rather, that it has three distinct components that we need to isolate.

However, when one says that properties are to be

modeled by functions from possible worlds to functions from individuals to truth values, it is not uncommon for this to be misunderstood as a claim that properties really are such functions, rather than that they are being modeled by them." It is this misunderstanding that generates the tension between properties and prepositional functions. Once we are clear about the role of model theory and its use for the construction of a theory of properties, relations and propositions, the tension will have lost its threatening power. Remember that the technical superiority of prepositional functions over the construal of properties as certain unstructured objects was presented in connection with a special algebraic model structure. The point here is that such a meaning algebra is nothing but an awful set theoretic η-tuple. Now, nobody will want to identify properties, relations, or propositions with sets and for the same reason it does not make any sense to identify the predication operation with a particular single-valued set of ordered pairs in the framework of this algebraic model. Set theoretic semantics can serve as a useful guide in our investigation of

286

the realm of PRP, nevertheless. This is true, e.g., for the purpose of proving a completeness theorem. In the end, however, the set theoretic ladder has to be kicked away and supplanted by an internal semantics specified directly in terms of a property theory. In this sense, by distinguishing clearly between the external semantics of the set theoretic model and the internal significance of the theory of properties, it would be equally implausible to equate properties with functions or with certain sets. These equations are to be invoked for no other purpose than to serve as an external criterion for certain formal properties of the property theoretic framework. Neither of the two types of external modeling can claim any advantage over the other from this perspective. The contrast beween external semantics and internal significance echoes a distinction between two views of looking at the relation of language and its logic to content. Within the area of logic this distinction is often referred to as the difference between logic as language and logic as calculus (cf. Hintikka 1988). According to the first view, logic has content in its own right and the notion of validity is dependent on the precise specification of the nature of the things postulated by a particular logical theory. This dependence assumes the form of basic rules or principles and has therefore led to the association of this doctrine of logic with the axiomatic approach. This doctrine can be traced back to Leibniz1 conception of a characteristica universalts and finds its modern expression in the work of Frege, Russell and Whitehead. Because logic on this view has content of its own, the language in which it is formulated cannot be reinterpreted, the world-view enshrined in its fixed interpretation confining its user to some sort of conceptual prison. The chapter on the connection between this logical doctrine and the well-publicized strand in modern linguistics that has gained a certain notoriety under the name of linguistic relativism has still to be written. The most important characteristic of these historically parallel developments in logic and linguistics is the insistence on a concept of logic and language that sees in them a philosophical framework that must ultimately fend for itself. It goes without saying that on this view of logic as language no general metalogic was possible and semantic explorations had to rely on indirect hints that presuppose a prior understanding on the side of the reader. Perhaps the best way to get a feeling for the axiomatic approach to logic is to contrast it with the view of logic as calculus. This second view is closely tied to the development of model theory in the second half of the fifties. Its historical origin was Hubert's conception of the nature of interpretation assigned to his axioms of geometry and Husserl's ideas of the multitude of Lebenswelten. What a specific axiomatization of some kind of geometry does, according to Hubert, is to delineate the class of structures in which they are true. The nature of the elements populating the structures is completely irrelevant. What matters is only the variety of interpretations over domains of different sizes. Similarly, a logic and its language

287

do not have a content of their own. The cornerstone of this type of semantics is given by the relation of a sentence to the family of models in which it is satisfied. In the particular circumstances where these models are the space of possible worlds, the meaning of a sentence assumes the familiar form of a set of possible worlds. Although the various interpretations on the view of logic as calculus are given in terms of a set theoretic semantics, set theory itself is not necessarily to be ranged with this second doctrine of logic.6 The iterative or mathematical concept of set, which characterizes sets as having their being in their members, is grounded in stable intuitions about the cumulative hierarchy and has led to a group of axioms everybody accepts as a partial description of their intended domain. Set theory is by now a comprehensive framework within which one can organize most mathematical activity. In particular, set theory has turned out to be a powerful domain in which to define the possible variations of interpretation to which logical calculi lend themselves. Set theory may be investigated by model theoretic means, but this method of investigation does not turn set theory into a calculus as long as one keeps in mind the cumulative hierarchy which forms the intuitive background for its axiomatic treatment. The same is true of the theory of properties, relations and propositions. PRP is intended as a revival of logical realism and its accompanying theory of universals: properties, relations and propositions are regarded as primitive entities and PRP's axioms are to be understood as an attempt to assemble the salient formal traits of its intended realm of significance. PRP thus belongs to the view of logic as language and is endowed with content in its own right. Set theoretic interpretations can only serve as an external guide to the internal constraints on its specific theory of logical form. It should now be understandable why I said above that the type of set theoretic interpretation that forbids us to talk about the same properties in different models is totally alien to the spirit of PRP. This point can be further clarified by discussing the completeness problem of a specific version of PRP. This is the version that takes intensional entities to be identical if they are necessarily equivalent. This theory is certainly inadequate for matters concerning the prepositional attitudes, but the basic intuition behind it, i.e. our practical ratiocination about the different courses of events made possible by our actions, has to find an explanation in the framework of property theory if that theory aims at playing a foundational role in logic, philosophy, natural language semantics and psychology.

In this connection the question

immediately arises whether one can present a complete, recursive axiomatization of the theorems pertaining to this conception of logical form. In case of a positive answer the semantics that is employed in carrying out the completeness proof serves only an ancillary 6

Reinhardt (1980) emphasises this aspect and points to a similar interpretation of Montague's predicational intensional logic.

288

function and has to be removed in favour of the axioms and rules of the logic in question. Since this logic is regarded as an example of the logic as language view, it cannot be otherwise. The logic has content in its own right and the external semantics supporting the metalogical property of completeness is fully characterized by the function it was designed to fulfil in establishing this feature of the inherently meaningful logic. Incidentally, when one interprets predication as membership and imposes a maximality condition on the model in the sense that it should contain every possible extension of its predicates, then an incompleteness result is forthcoming with respect to validity in these maximal models. The reason for this negative result is the implicit second-order element in the maximality constraint. If instead of using the class of all the algebraic model structures we use the subclass of the maximal structures, a first-order necessity sentence is equivalent with a second-order sentence with all its predicates universally quantified away. It is well known that when validity is specified for second-order logic in terms of maximal models, the set of valid formulae is not recursively enumerable. The weak point in this reasoning is the hypothesis that standard maximal models provide an adequate semantics for this type of logic. The decision to base one's investigation of the completeness problem on this type of semantics involves a number of assumptions that are difficult to justify from the perspective of logic as language. The fundamental question that is preempted by the use of maximal models concerns the modeling of the notion of predication. If predication is given its own representation that bears no resemblance to membership in a set, the justification for an incompleteness theorem disappears suddenly. Membership, as we have seen, is what defines the mathematical notion of set, and once we reject the idea that the varying extensions of properties and relations are given in terms of mathematical sets, we are free to use the logical notion of set instead. Classes or sets in the logical sense have their being in the properties or concepts whose extensions they are. If we interpret the extensions of properties and relations as logical sets, the analogon of a necessity sentence in the property theory of necessary equivalence is not a sentence in standard second-order logic but a sentence in predicative second-order logic. Again, it is well known that with respect to this logical theory a completeness theorem can be proven.

In other words, the alternative

specification of validity which accords predication a primitive representation leads to a semantics that is not essential for the internal meaning relations of the property-theoretic incarnation of the notion of necessary equivalence. We can kick away the ladder and stay within the basic philosophical framework of property theory, whose semantics can be treated on its own terms. I started the discussion of the completeness problem with a remark concerning its relevance for the prepositional-function thesis. We saw that the identification of properties with prepositional functions is untenable if the latter are in turn modeled as functions from

289 individuals into truth values. This is so because the double dependence of properties on the sets of possible individuals and on the sets of possible worlds makes it impossible to talk about the same property in two different models. By imposing the maximality constraint on the models of the theory of necessary equivalence, we have been able to rid ourselves of the dependence on a primitive set of possible worlds, these being now identified with the maximal variation in the assignment of extensions to predicates. The dependence on the set of possible individuals, though, appears to be a factor we have to accept. This, fortunately, is only an illusion created by our concentration on the notion of necessary equivalence. That properties and relations are identical when they have the same extension in every possible world holds good only of theories that represent predications as membership. With predication as fundamental concept, we are free to define a relation of necessary identity that reduces to necessary equivalence. But this relation of necessary identity has to be kept apart from identity proper. Two properties may apply to the same objects in all possible worlds without therefore being the same property, as shown by the attributes of triangularity and trilaterality. Two models can contain the same property even if the two domains of possible individuals they incorporate have an empty intersection.

4.

Three ways out

The contrast between the views of logic as language and logic as calculus removes the sting from the tension between properties and prepositional functions. PRP is a theory of logical form in the tradition of logical realism and, as such, is avered by the doctrine of logic as language. In this doctrine, it will be recalled, you cannot re-interpret the language in which a particular logical theory is expressed. The ultimate criterion of significance is the internal content the language is supposed to have in its own right, and the different set-theoretic modelings have no other purpose than to serve as a directing principle to the internal characteristics of the meanings forming part of this independent content structure. Whether we interpret properties as saturated objects or as functions is then to be decided on the respective merits of these two interpretations as directing principles with respect to the realm of basic intensional entities associated with the universale of logical realism. We have seen that a model that uses a domain of prepositional functions and analyses predication as functional application carries with it undesirable constraints we must be careful not to confuse with properties of the things modeled. The arguments of the constant function and of the inverted order of saturation of a two-place predicate have shown that models of this kind lead to a levelling of content structure that is in disagreement with the demands of an adequate prepositional attitude semantics. This state of affairs indicates a branching point in the external modeling technique of the theory of properties, relations and propositions which calls

290

for a careful consideration of the consequences certain approaches to constructing models of the theory have for our understanding of the things postulated by the theory. The systematic variation of modeling procedures can be of great help in this line of research, and the example of set theory shows how model-theoretic considerations lead to insights into the structure of the cumulative hierarchy. Looked at from this perspective, the tension between properties as objects and properties as functions constitutes a potential source of new discoveries about the hidden universe of properties, relations and propositions. The actual presentation of the formal details that are necessary to carry out an in-depth comparison of the pros and cons of the propositional function thesis would lead us beyond the scope of this paper. I shall therefore conclude by sketching some of the paths into the modeltheory of PRP that do not end at an abyss created by the erosional effects of the attitudes. To take up the constant-function argument first, it is perhaps fair to say that at least part of the burden of proof lies with the detractor of the prepositional-function thesis. We charitably granted him the assumption that the principle allowing the postulation of a function that has for every argument in its domain the same value, a proposition expressed by a closed sentence of PRP's language needs no further justification. But this does not seem to be the case. There is a long debate on the meaning of vacuous variable binding in the history of logic, and a related debate has been taking place in linguistics, triggered by Chomsky's strictures against a similar device within the framework erected on general principles and specific parameters. The reasons brought forward against the vacuous binding of variables have to be countered by someone who opposes the prepositional-function thesis before he has won his case. The time and space allotted to this paper prevent me from going over the ground of vacuous binding. Be it here stated without further argument that the case against the unrestricted use of variable binding seems to be a particularly strong one. When the unrestricted process of variable binding or what amounts to the same thing, abstract formation, is taken for granted, the dilemma created with the constant-function is of a piece with the one illustrated by means of the fondalee/rajneesh-example. In both cases the principle of ß-conversion barred us from recapturing the predicate and the argument(s) that make up the respective propositions. We know that we cannot uphold both the propositional function thesis and the identification of predication and functional application together with its regulating principle of ß-conversion; but this restriction on the activity of model building does not preclude the retention of the prepositional-function thesis when complemented by a suitable substitute for the operation of function application. I know of three ways to proceed in formally implementing this idea.

As an envoi I will try to give an inkling of the

constructions which lie behind them. The first method takes its inspiration from Carnap's account of intensional isomorphism. According to that account, the meaning of an expression consists of the meaning of its parts

291

combined in a structure that is a homomorphic picture of its syntactic analysis. Cresswell (1985) has shown how to incorporate this account into a formal semantic theory that belongs to the possible-worlds paradigm. The solution he presents, however, can be described independently from the assumptions of possible-worlds semantics. What makes his solution interesting for our discussion is the fact that Cresswell's theory of meaning, besides its allegiance to the possible-worlds paradigm, employs a form of categorial semantics that uses functional application as its only form of semantic composition. In this type of analysis, expressions are partitioned into various categories in consonance with their semantic functions. There are two basic categories, the category of sentences, indicated by 0, and the category of names, indicated by 1. The complex categories are constructed out of these two basic ones according to the schema that if τ and σ1?...ση are categories, then (x/σ^...σ,,) is a category. An expression assigned the category (τ/σ^.,.ση) concatenated with η expressions of category at , ... , and category ση, respectively, yields an expression of category τ. Its meaning is an η-place function that gives as value a meaning of category τ when applied to the meanings of the η expressions of categories σ^.,.σ,,. Once we are given two sets, one consisting of possible individuals Dj and the other of sets of possible worlds DQ, the additional meaning types are easily specified.

Expressions of category (τ/σ^.,.σ,,) are

interpreted as (possibly partial) functions from the Cartesian product Dal χ..·ΧΟσ1η into DT. To take an example, the names Miriam and Socrates would be assigned the category 1, the activity of running would be viewed as an expression of category (0/1) and the attitude of believing would be considered an expression of category (0/1,1), combining a name and another name to yield a sentence. The sentence (20)

Miriam believes that Socrates runs

would be analysed as follows: (21)

believe(0/u) (Miriam!, that(1 0) (r

In this analysis the complementizer assumes the role of a nominalizer which sends elements of DO into Dj.

This role makes it possible to assign the same category to believe in

constructions where it governs an embedded sentence and in constructions where it is followed by a noun phrase. Suppressing all morphological details, the above analysis says that believe is a function that sends a certain ordered pair into a set of possible worlds, this ordered pair having as first component an individual, the person named Miriam, and as second component the individual value of the complementizer function that applied to the proposition that itself results from functionally applying the activity of running to the individual named Socrates. Needless to say, this elegant functional analysis runs into all the problems that beleaguer every semantic account in the possible worlds framework.

292

Cresswell's solution consists in assigning to the complementizer that a potentially infinite number of categories. In this way he is able to retain the essential aspects of the syntactic structure of the sentence expressing the content at which the attitude is directed. In our example, that could also have been assigned the category (1/(0/1),1). An expression of this category combines an intransitive verb and a name to yield another name. Intuitively, this that can be considered a sort of pairing operator.

On the semantic side, this alternative

categorization leads to an analysis that says of Miriam that she believes of Socrates and of the. activity of running that he is engaged in the latter. Since in this type of semantic account Miriam is supposed to be mentally related to all the meaningful constituents of the complement sentence, the account is not taken aback by the lexical decomposition examples. The relation of Miriam to Robin's winning (in (3) and (4)) has a much simpler macrostructure than her relation to something that contains besides Robin such meaningful components as the logical operations of negation and universal quantification and the activities of losing and competing.

By reverting to a suitable complementizer of category (Ι/σ^ ... , a„) we can

analyse Miriam's belief in such a way that it becomes responsive to all those structural details. Logical equivalents are intersubstitutable salva veritate only if they display the same structure, if they are intensionally isomorphic.

Similar remarks apply to the fondalee/

rajneesh-example. In this case it is advisable not to climb down the syntax tree to its ultimate constituents but to stop at the level where the defined predicates of fondaleeing and rajneeshing label its nodes. The meaning, i.e. the function assigned to these intransitive verbs, is certainly different, and this gives us a handle on how to distinguish a belief in Jane Fonda's rajneeshing from a belief in Rajneesh's fondaleeing. If the respective beliefs are made responsive to sentences expressing them, there is no danger of mistaking somebody's following Rajneesh for Jane Fonda's following somebody. Cresswell's solution shows that it is technically feasible to arrive at a suitably fine-grained meaning structure in a model that identifies properties with propositional functions. As his solution is a proposal for a specific type of model building, an objection taking exception to the fact that attitudes are interpreted as functions from individuals and η-tuples of meaningful entities to propositions is certainly beside the point. We are free to use any artifact of the modeling process as long as it gives us the components we need to isolate. But whether Cresswell's analysis in terms of the notion of an ordered η-tuple discerns the right components is a totally different question. Remember that we insisted on the computational character of mental states. Mental states, especially attitudinal states, have to have some sort of symbolic structure that supports our practical deliberations. The domain of the logic of the attitudes is a particularly murky area strewn with pitfalls for the unwary, and even the outlines of a generally accepted logical system are still to be developed. What seems to be clear, however, is that any attempt to base such a system on a symbolic structure of η-tuples is doomed to

293

failure. What is needed is a technical implementation that is faithful to the predicational form of statements without barring the later recapture of the arguments and relations that make up the statement. Specifically, we are looking for a model that contains a functional Pred that sends pairs of prepositional functions and individuals into propositions and that satisfies the following principles: (22) (23)

Pred (P,a) = Pred (Q,b) ->P = Q&a = b Pred (P,a) P(a)

(22) echoes principle (15) and stipulates that predicating P of a generates the same proposition as the result of predicating Q of b only if P and Q are the same properties and a and b are the same individuals. (23) says only that predicating P of α ushers in a proposition that is materially equivalent with, i.e. has the same truth value as, the functional application of Ρ to a. It was shown in Dealer and M nnich (1989) that a function satisfying the principles above can be added to a type of functional structure that was invented by Aczel (1980). The construction of Pred makes use of a standard inductive definability approach. Starting from a functional model of the lambda calculus, the usual logical operations and the functional Pred may be defined in such a way that they form the clauses of a monotone operator on the model. The least fixed point of the operator specifies a model of the logical operations and Pred. It is fairly straightforward to guarantee by the appropriate formulation of the defining clauses that in the extended model the logical operations and Pred exhibit the expected characteristics. Although this construction is not subject to the objection levelled against Cresswell's solution, it still suffers from a major drawback. The fixed point of the inductive model construction contains both the new functional Pred and functional application, which was built into the functional structure from the very beginning. Depending on the intended area of investigation, the semantic analysis of a statement has the option of'taking recourse to predication or to functional application. But this constitutes a clear case of having lost one's semantic innocence. Why should a statement occurring after a verb standing for a prepositional attitude be composed of one set of elements when that 'same' statement occurring after a sentential operator like it is true that exhibits a different set of semantic constituents?

As Bealer (1989) has observed, the thought that the glass is half empty is

different from the necessarily equivalent thought that the glass is half full. This is not so for conditions in the world. The glass's being half empty is the same state of affairs in the world as the glass's being half full.

This distinction between fine-grained and coarse-grained

intensions gives rise to the question of what reasons there are for thinking that this distinction has to be modeled by two different combinatory processes that glue an individual to its

294

property. Upholding the propositional-function thesis by giving up a unified treatment of intensions seems too high a price to pay. This last dilemma points to a general defect in the attempts at modeling the propositionalfunction thesis.

When we introduced the thesis we took as our point of departure the

considerable technical advantages that come with the treatment of the predication operation in terms of functional application. By tying the simplification in the semantic analysis of intensions to the propositional-function thesis we have closed the door to other implementations of functional accounts of meaning structure. One such implementation of which I can present just the barest contour lines starts from a resolute turning of tables. It looks in a top-down procedure first at the structured objects and only afterwards attempts a description in functional terms of these objects. The abstract framework in which to carry out such a project has only recently become available. By combining ideas from category theory and abstract algebra, Peter Aczel (1990) has developed the beginnings of a theory of structured objects.

For this purpose he defines two central notions, the notion of a

replacement system and the notion of an ontology. A replacement system is a universe of structured objects, each object having components which are again objects of the system. An object can be changed by replacing some or all of its components with other objects. Such a replacement operation satisfies a number of axioms. An ontology can be thought of as a universe of templates for the formation of the objects of a replacement system.

These

templates have markers instead of components. These markers may be exchanged for other markers, but markers in contradistinction to components need not be elements of the ontology. The notion of an ontology generalizes the concept of a signature, where a signature consists of a class of symbols with their associated arities. By interpreting the symbols of a signature as operations of their respective arity on a class, we arrive at the notion of an abstract algebra. Aczel proves a general theorem to the effect that for every replacement system there is an ontology, such that the replacement system is an algebra of the associated ontology. This theorem suggests the following project: start with the universe of intensions as your initial replacement system, construct the associated ontology and regain the replacement system as an algebra of the ontology. Since an algebra is defined as a class together with a family of operations on this class, the only combinatory process which remains is that of functional application. For different intensional replacement systems the resulting algebras will embody a variety of identifications. Some will contain properties in the guise of one-place functions, others will contain them as elements of the class on which the operations are defined. This flexibility is to be applauded and testifies to the power of this type of top-down model building. The account I have given of the tension between properties and prepositional functions has led to a principle of tolerance in the model theory of PRP. Beginning with the difficulties that

295

confront the leading semantic paradigm, the possible worlds approach, we have seen that a theory of properties that constitutes an incarnation of basic tenets of logical realism runs into grave technical complexities in connection with the semantic analysis of abstract terms. The identification of properties with prepositional functions offers itself as a way out of this technical quandary. The proposed identification must not be misunderstood as an essential semantic characterization of properties but serves only the function of an element in an external modeling process. Under this perspective we are free to investigate the strengths and weaknesses of various modeling techniques. This has enabled us to espouse an attitude of liberalism that was hailed in the beginning as a distinguishing feature of the scholar to whom these lines are dedicated.7

Bibliography Aczel, P. (1980): "Frege Structures and the Notions of Proposition, Truth and Set" - In: J. Barwise, HJ.Keisler, & K. Kunen (eds.): The Kleene Symposium (Amsterdam) 31-59. Aczel, P. (1990): "Replacement Systems and the Axiomatization of Situation Theory." - In: R. Cooper, K. Mukai & M. Perry (eds.): Situation Theory and Its Applications (Stanford University) 3-31. Anderson, A. (1986): "Some Difficulties Concerning Russelian Intensional Logic." - In: Nous 20,35-43. Barwise, J. (1987): Course Notes on Situation Theory. - Stanford University. Bealer, G. (1989): "On the Identification of Properties and Prepositional Functions." - In: Linguistics and Philosophy 12,1-14. Bealer, G. & U. Mönnich (1989): "Property Theories." - In D. Gabbay & F. Guenthner (eds): Handbook of Philosophical Logic Vol.IV, 133-251. Cocchiarella, N. (1988): "Predication Versus Membership in the Distinction Between Logic as Language and Logic as Calculus." - In: Synthese 77,37-72. Cresswell, M.J. (1985): Structured Meanings. - Cambridge, Mass. Hintikka, J. (1988): "On the Development of the Model-Theoretic Viewpoint in Logical Theory." - In: Synthese 77,1-36. Kamp, H. (1990): "Prolegomena to a Structural Theory of Belief and Other Attitudes." - In: A. Anderson & J. Owens (eds.): Prepositional Attitudes (Stanford University) 27-90. Mönnich, U. (1983): "Towards a Calculus of Concepts as a Semantical Metalanguage." - In R. Bäuerle, C. Schwarze & A. von Stechow (eds.): Meaning, Use, and Interpretation of Language (Berlin) 342-60. Quine, W.V. (1960): Word and Object. - Cambridge, Mass.

See footnote 1 above.

296 Reinhardt, W. (1980): "Satisfaction Definitions and Axioms of Infinity in a Theory of Properties with Necessity Operator." - In: A. Arruda, R. Chuagui & N. da Costa (eds.): Mathematical Logic in Latin America (Amsterdam) 267-303. Thomason, R. (1980): "A Model Theory for Prepositional Attitudes." - In: Linguistics and Philosophy 4,47-70.

297

Heinrich Weber

Sein und Raum: Bemerkungen zur lokalistischen Interpretation des Verbs sein in Heideggers Einfährung in die Metaphysik 1.

Fragestellung

Philosophische und wissenschaftliche Untersuchungen gehen seit der Antike oft von dem alltäglichen Vorverständnis aus, wie es in den Einzelsprachen niedergelegt ist, um von da aus zu einem neuen Verständnis der behandelten Gegenstände weiterzuschreiten. Diese Methode verwendet auch Martin Heidegger in der im Sommersemester 1935 in Freiburg gehaltenen, aber erst 1953 publizierten Vorlesung Einführung in die Metaphysik, in der er die folgende Frage untersucht: "Warum ist überhaupt Seiendes und nicht vielmehr Nichts?" Bei der Suche nach einer Antwort greift er explizit auf grammatische und etymologische Tatbestände zurück, nimmt selbst eine Analyse des alltagssprachlichen Gebrauchs von sein vor und setzt das Verb zu vergleichbaren Ausdrücken in Beziehung. Außerdem paraphrasiert er sein in einer Weise, die linguistisch interessant ist. Heideggers Leitfrage braucht uns hier nicht zu interessieren. Ihre Beantwortung ist Aufgabe der Philosophie oder, wenn wir Carnaps Kritik an dieser Ausprägung von Metaphysik folgen wollen, bloßer Ausdruck eines Lebensgefühls, das mit dem der Dichtung vergleichbar ist (vgl. Carnap 1931:238-41). Was uns hier allein angeht, ist Heideggers Beitrag zur Beschreibung des Verbs sein. Diesen Beitrag wollen wir vorstellen und kritisch beleuchten, indem wir zunächst die Forschungslage zum Verb sein skizzieren und anschließend Heideggers Gesichtspunkte referieren und diskutieren.

Abschließend wird darauf hingewiesen, daß

Heideggers Neuansatz in der Metaphysik gerade mit einer Neuinterpretation der Bedeutungsstruktur von sein zusammenfällt, bei der ein Aspekt ins Zentrum der Überlegungen gestellt wird, der in der Tradition vernachlässigt worden ist.

2.

Forschungsstand

Das Verb sein ist eines der Wörter im Deutschen, die am häufigsten und vielfältigsten verwendet werden und deren Beschreibung eben deswegen am schwierigsten und interessantesten ist.

Das gleiche gilt für die Entsprechungen von sein in anderen Sprachen.

Die

Philosophie hat schon früh die Analyse dieses Verbs zu einer ihrer Aufgaben gemacht. In der Metaphysik (Met.) arbeitet Aristoteles die verschiedenen Verwendungsweisen von on und etnai, von seiend und sein heraus; weitere Hinweise finden sich an vielen Stellen seines Werkes. In dieser Hinsicht ist sein bzw. seiend ein attributionsanaloges Wort, dessen Inhalt

298

von der Kategorie des Prädikats abhängt, das sich aber, wenn auch in verschiedener Weise, auf Eines bezieht, nämlich die Substanz, das Einzelding (vgl. Met. IV,l:1003a32ff.). Nach einer anderen Auffassung, die sich auf eine Stelle in der Hermeneutik (De Int. 16b22-26) stützen kann, ist sein nur Kopula:. Das Wort "bezeichnet zusätzlich eine Verbindung, die ohne das Zusammengestellte nicht zu denken ist". Die Logik von Port Royal aus dem Jahr 1685 führt diese Auffassung weiter, indem sie sagt, die Hauptfunktion des Verbs sei die "Bezeichnung der Bejahung" (Arnauld 1972:100f.), und diese Hauptfunktion käme in der dritten Person ist rein zum Ausdruck. Eine dritte, neuere Auffassung von sein besteht darin, daß es primär zur Bezeichnung der Existenz diene, etwa in dem Sinne, wie es in Descartes' berühmtem "cogito ergo sum" zum Ausdruck kommt. Sein als allgemeinstes Prädikat, sein als Kopula und sein als Existenzbezeichnung: das sind die wichtigsten Auffassungen in der philosophischen Diskussion (vgl. auch Weber 1988:510-15). Auch in der Sprachwissenschaft ist das Verb Gegenstand eingehender Analysen gewesen. Der sowjetische Germanist Wladimir Admoni unterscheidet bei dem 'kopulativen Verb' sein erstens die Bedeutung 'leere Form des Seins', die durch das Hinzutreten von Prädikaten konkretisiert wird, zweitens die lexikalische Bedeutung 'existieren, sich befinden', drittens die Funktion als Hilfsverb, bei der die lexikalische Semantik völlig verloren geht (Admoni 1982:167). Die historische Sprachwissenschaft des 19. Jh. hat die Etymologie von sein untersucht.

Sie ist dabei - nach Auskunft des 1899 erschienenen Artikels "Sein" im

Grimmschen Wörterbuch - zu dem Ergebnis gekommen, daß die Formen des Verbs auf die Wurzel es- mit der wahrscheinlichen Grundbedeutung 'leben, atmen', auf die Wurzel bhu- mit der Grundbedeutung 'werden, entstehen, wachsen1 und auf die Wurzel ves- mit der Bedeutung Verweilen, wohnen, bleiben' zurückzuführen sind (DWB 1984, Bd. 16:228-29).

Eine

syntaktische Analyse des Verbs hat sich mit Form und Funktion der Ergänzungen zu beschäftigen, mit denen es verbunden werden kann. Eine solche Beschreibung wird beispielsweise im Valenzwörterbuch von Engel und Schumacher (1976) versucht.

Die

Anwendung

die

der

Wortfeldtheorie und

der strukturellen

Semantik ermöglicht

Zusammenstellung von Gruppen von Verben, durch die sein - in Abhängigkeit vom jeweiligen syntaktischen Kontext - substituierbar ist, und damit die Erfassung der Stellung von sein im lexikalischen System der Sprache (vgl. Erben 1978, Bald 1972).

Für den

Sprachvergleich ist die von John Verhaar herausgegebene Schriftenreihe über das Verb be und seine Synonyme von besonderem Interesse; genannt sei die umfassende Monographie von Charles Kahn aus dem Jahr 1973, in der das Verb etnai im Altgriechischen umfassend behandelt wird.

299

3.

Zur " Grammatik und Etymologie"

Martin Heidegger geht, wie er selbst sagt, beim "Fragen der Seinsfrage" von "sprachlichen Überlegungen" aus (1953:39). Am Anfang steht für ihn die "Tatsache, daß uns das Sein in der Tat fast nur noch ein Wort und dessen Bedeutung ein verschwebender Dunst ist", sowie die Feststellung, "daß überhaupt die Sprache verbraucht und vernutzt ist, ein unentbehrliches, aber herrenloses, beliebig verwendbares Mittel der Verständigung, so gleichgültig [...] wie die Straßenbahn, in der jedermann ein- und aussteigt" (1953:38). Um für das Wort den "Rest eines Bezuges" zu finden, untersucht er zunächst Grammatik und Etymologie. Er analysiert das Sein als Infinitiv form, als ptosis oder gklisis aparemphatikos, d.h. als eine Form, die - in Heideggers Worten - zum einen ein "Fallen, Kippen und Sich-neigen" ausdrückt, ein "Abweichen vom Aufrecht- und Geradestehen", zum ändern aber, als Infinitiv, "nicht mehr zum Vorschein [bringt], was das Verbum sonst offenbar macht" (1953, 46 und 52). Daraus wird die Folgerung gezogen, daß man sich bei der Bedeutungsanalyse an die fmiten Formen halten solle. Bei der Beschreibung der Etymologie folgt Heidegger der konventionellen Einteilung, bietet aber originelle und eigenwillige Umschreibungen der Grundbedeutungen: 1. 2. 3.

Das Stammwort es- bezeichnet "das Leben, das Lebende, das, was von ihm selbst her in sich steht und geht und ruht: das Eigenständige". Das Stammwort bhu, bheu bedeutet, wie griechisch phyo, "aufgehen, walten, von ihm selbst her zu Stand kommen und im Stand bleiben." Der Stamm wes- wird wie im Grimm durch "wohnen, verweilen, sich aufhalten" wiedergegeben (1953:54-55).

Heidegger verwendet also, obwohl die etymologische Forschung zunächst keine Anhaltspunkte dafür bietet, zur Umschreibung von es- und bhu- Ausdrücke, die dem Bereich der Befindlichkeit und Bewegung im Raum entnommen sind; man vergleiche 'stehen, gehen, eigenständig, aufgehen, Stand, kommen, bleiben1. Die Wurzel ves- ist bereits von der sprachwissenschaftlichen Etymologie in diesem Sinne verstanden worden.

4.

Das "Wesen des Seins"

Im dritten Kapitel stellt Heidegger, wie die Überschrift sagt, "Die Frage nach dem Wesen des Seins". Man kann diese Frage als Frage nach der Grund- und Richtbedeutung verstehen, welche die Einheit der drei Stämme von sein ermöglicht. Er versucht diese Bedeutung zu finden, indem er beim "Sagen des Seins" ansetzt und die folgenden Beispiele untersucht (vgl. 1953:68):

300 (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14)

Gott ist (wirklich gegenwärtig) Die Erde ist (ständig vorhanden) Der Vortrag ist im Hörsaal (findet statt) Dieser Mann ist aus dem Schwäbischen (stammt her) Der Becher ist aus Silber (besteht aus) Der Bauer ist aufs Feld (hat seinen Aufenthalt verlegt, hält sich auf) Das Buch ist mir (gehört) Er ist des Todes (verfallen) Rot ist backbord (steht ßr) In Rußland ist Hungersnot (-) Der Feind ist auf dem Rückzug (-) In den Weinbergen ist die Reblaus (-) Der Hund ist im Garten (treibt sich herum) Über allen Gipfeln / ist Ruh (herrscht?, liegt?, waltet?)

Heideggers methodisches Vorgehen besteht darin, daß er den Ausdruck ist in den aufgeführten Sätzen durch andere Ausdrücke zu substituieren oder zu expandieren sucht. So wird beispielsweise in (3) sein durch stattfinden und in (13) sein durch sich herumtreiben substituiert. In (1) wird sein durch wirklich gegenwärtig sein expandiert. Beim letzten Beispiel, dem Goethe-Gedicht, lehnt Heidegger eine Substitution oder Expansion ab; Paraphrasen wie "Über allen Wipfeln herrscht Ruh, liegt Ruh, waltet Ruh" erscheinen ihm nicht als treffend. Heidegger kommt zu dem Ergebnis, daß ist im Sagen zwar eine reiche Mannigfaltigkeit der Bedeutungen bekundet, daß diese aber nicht beliebig ist. Er schreibt: Das im "ist" gesagte "sein" bedeutet: "wirklich gegenwärtig", "ständig vorhanden", "stattfinden", "herstammen", "bestehen" "sich aufhalten", "gehören", "verfallen", "stehen für", "sich befinden", "herrschen", "angetreten haben", "auftreten". [...] Die Begrenzung des Sinnes von "Sein" hält sich im Umkreis von Gegenwärtigkeit und Anwesenheit, von Bestehen und Bestand, Aufenthalt und Vorkommen. (1953:69) Heideggers Analyse ist in linguistischer Hinsicht durchaus beachtenswert. Sie stützt sich auf den alltäglichen Sprachgebrauch, verwendet - wenigstens implizit - die strukturalistische Substitutions- und Paraphrasemethode und stellt eine Art Wortfeld zusammen, in dessen Mittelpunkt sein steht. Trotzdem ist die Analyse nicht recht befriedigend. Zum einen ist die Auswahl der Beispiele außerordentlich einseitig. Zwei Drittel der Beispiele (neun von vierzehn) weisen einen Satzbauplan auf, in dem ist mit einer - meist räumlichen Präpositionalergänzung verbunden ist. In der Alltagssprache liegt der Anteil dieses Satzbauplans bei sein unter zwanzig Prozent.1 Die Konstruktionen ohne Ergänzung (Gott ist, die * Die quantitativen Angaben stützen sich auf ein kleines Corpus von Belegen, die unter dem Gesichtspunkt der vollständigen Auswertung von Textabschnitten gesammelt wurden. Die drei häufigsten Gebrauchsweisen verteilen sich wie folgt: sein + Adjektiv 230 (49 %) sein + Substantivgruppe im Nominativ 160 (34 %) sein + Präpositionalgefüge 80 (17 %) Die sonstigen Gebrauchsweisen, z.B. der Gebrauch als Hilfsverb, wurden bei dieser Zahlung nicht berücksichtigt

301

Erde ist), mit Dativergänzung (Das Buch ist mir) und mit Genitivergänzung (Er ist des Todes) sind in der Alltagssprache ziemlich selten. Als einziges Beispiel für die Konstruktion mit prädikativem Adjektiv steht der Satz Rot ist backbord. Wir haben hier jedoch nicht den normalen Gebrauch der Konstruktion, ist drückt hier die Beziehung zwischen Bezeichnendem und Bezeichnetem im Zeichen aus. Die beiden häufigsten Verwendungsweisen von sein in der Alltagssprache bleiben unerwähnt; Beispiele wie die folgenden fehlen völlig: Das Auto ist grün Hans ist Lehrer Sie bilden in der Alltagssprache etwa drei Viertel der Belege. Versucht man das ist in diesen Beispielen zu paraphrasieren, so erhält man eine Reihe von Paraphrasen, die nicht in den Umkreis von Gegenwärtigkeit und Anwesenheit verweisen. Man vergleiche: Das Auto sieht grün aus, das Auto bleibt grün, das Auto wirkt grün Hans betätigt sich als Lehrer, Hans hat den Beruf eines Lehrers, bei Hans handelt es sich um einen Lehrer Unberücksichtigt bleiben bei Heidegger auch die Gebrauchsweisen, die von alters her die wissenschaftliche Diskussion geprägt haben, z.B.: Sokrates ist ein Mensch, Sokrates ist sterblich Der Mensch ist ein Lebewesen Der Mensch ist ein vernünftiges Lebewesen Der Gebrauch des Verbs beim Klassifizieren und Definieren von Wörtern oder Sachen kommt nicht ins Blickfeld; seine Rolle beim Denken bleibt ausgespart. Die einseitige Auswahl der Beispiele mindert stark den Wert von Heideggers Versuch, die allgemeine Bedeutung von sein zu bestimmen. Zum zweiten ist es methodisch fragwürdig, die anstelle von sein substituierten Ausdrücke als Bedeutungen von sein auszugeben. In dem Satz Der Hund ist im Garten sage ich nicht, daß er sich dort herumtreibt; er könnte dort genauso sitzen, stehen oder liegen, ruhen oder bellen.

Sein ist allgemeiner als alle diese Ausdrücke, ist Gattung oder Oberbegriff.

Heidegger hat nicht Bedeutungen von sein angegeben, sondern Ansätze zur Beschreibung eines Wortfeldes geliefert, das durch den Kontext 'Subjekt und Raumergänzung' definiert ist und das Verben der Befindlichkeit im Raum enthält. Von diesen Verben ist sein das allgemeinste:

Es ist Archilexem. Sein ist jedoch nicht nur Verb der Befindlichkeit im

Kontext einer Raumergänzung, sondern tritt auch in anderen Kontexten auf und konstituiert dort andere Wortfelder. Diese anderen, häufigeren und wichtigeren Gebrauchsweisen werden von Heidegger ausgeklammert. Wie bei der Untersuchung von Grammatik und Etymologie wird auch bei der Untersuchung des Alltagsgebrauchs der räumliche Aspekt überbetont.

302

5.

Die "Beschränkung des Seins"

Im vierten Kapitel untersucht Heidegger "Die Beschränkung des Seins" (1953:71). Er verfolgt, wie er sagt, "die Unterscheidungen des Seins gegen Anderes". Dabei kommt er zu folgendem Ergebnis: Das Sein [...] muß [...] eine bestimmte Bedeutung haben. Die Bestimmtheit des Seins wurde durch die Erörterung der vier Scheidungen vor Augen geführt: Sein ist im Gegenhalt zum Werden das Bleiben. Sein ist im Gegenhalt zum Schein das bleibende Vorbild, das Immergleiche. Sein ist im Gegenhalt zum Denken das Zugrundeliegende, Vorhandene. Sein ist im Gegenhalt zum Sollen das je Vorliegende als das noch nicht oder schon verwirklichte Gesollte. Bleiben, Immergleichheit, Vorhandenheit, Vorliegen - sagen im Grunde alle dasselbe: ständige Anwesenheit: onalsusia. (1953:154) Diese vier Unterscheidungen werden von Heidegger philosophisch begründet; "Sie haben", wie er sagt, "mit dem Anfang des Fragens der Philosophie angefangen" (1953:72). Die Unterscheidungen Sein und Werden, Sein und Schein gehören den Anfängen der griechischen Philosophie an. Die Unterscheidungen Sein und Denken, Sein und Sollen gewinnen ihre zentrale Bedeutung erst in der Neuzeit. Diese Unterscheidungen sind somit nicht unmittelbar sprachlich zu begründen; allerdings meint Heidegger, alles Wissen, Tun und Sagen sei von ihnen durchdrungen. Tatsächlich würde eine sprachliche Interpretation dieser Unterscheidungen als Oppositionen, mit denen man die Bedeutung von sein eingrenzen könnte, nicht zum Ziel führen. Eine sprachliche Opposition besteht nur zwischen sein und werden, bis zu einem gewissen Grade auch noch zwischen sein und scheinen, aber nicht zwischen sein und denken, sein und sollen. Man vergleiche die folgenden Beispiele: (1) (2)

Hans ist krank Hans ist krank -

Hans wird krank Hans scheint krank Hans scheint krank zu sein (3) Hans ist krank "Hans denkt krank Hans wird als krank gedacht Er denkt, daß Hans krank ist Er denkt Hans als krank (4) Hans ist fleißig 'Hans soll fleißig Hans soll fleißig sein Sollen modifiziert sein; scheinen kann sein modifizieren, kann es aber auch ersetzen. Das Verb denken führt zusätzlich das Agens ein, das die Verbindung zwischen Subjekt und Prädikat des gedachten Satzes herstellt. Eine Substitutionsmöglichkeit besteht allenfalls im Passiv, weil da das Agens nicht genannt zu werden braucht.

303

Man muß Heidegger vom Sprachlichen her zustimmen, wenn er einen von den Oppositionen zum Werden, Scheinen, Denken und Sollen begrenzten Seinsbegriff für zu eng hält. Er fordert: [...] das Sein, das durch sie [die vier Scheidungen] eingekreist ist, muß selbst zum umkreisenden Kreis und Grund alles Seienden verwandelt werden. (1953:156) Diese Forderung soll hier jedoch nicht diskutiert werden. Es sei nur noch hervorgehoben, daß die vier Scheidungen für Heidegger im Grunde gar keine sind: Er führt sie zurück auf eine Formulierung, die im Wortfeld der Befindlichkeit im Raum anzusiedeln ist: auf "ständige Anwesenheit".

6.

Paraphrasen von sein

Die Untersuchung des "Sagens des Seins" hat Heidegger auf die Grammatik des Infinitivs, die Etymologie des Verbs, den Gebrauch in der Alltagssprache und auf zentrale philosophische Unterscheidungen geführt. In allen diesen Untersuchungen hat er ein Seinsverständnis expliziert, wie es vom Verb sein und anderen Verben des Wortfeldes der Befindlichkeit im Raum ausgedrückt wird. Dieses Seinsverständnis findet bei Heidegger aber auch einen indirekteren Ausdruck. Es zeigt sich in den Paraphrasen, die Heidegger verwendet bei der Erklärung dessen, was Sein ist. Auf solche Paraphrasen sind wir bereits mehrmals gestoßen; im folgenden seien einige weitere besonders prägnante angeführt: Dieses aber, das in sich hoch aufgerichtete Da-stehen, zum Stand kommen und im Stand bleiben, verstehen die Griechen als Sein. (1953:46) Das In-sich-da-Stehende aber wird, von der Betrachtung her gesehen, zum Sich-darStellenden, das sich in dem, wie es aussieht, darbietet. Das Aussehen einer Sache nennen die Griechen eidos oder . Im etdos schwingt anfänglich mit, was auch wir meinen, wenn wir sagen: die Sache hat ein Gesicht, sie kann sich sehen lassen, sie steht. Die Sache "sitzt". Sie ruht im Erscheinen, d.h. Hervorkommen ihres Wesens. (1953:46) "Sein" besagt im Grunde für die Griechen Anwesenheit. (1953:46) "Sein" sagt für die Griechen: die Ständigkeit in dem Doppelsinne: 1. das In-sich-stehen als Ent-stehend (physis) 2. als solches aber "ständig", d.h. bleibend, Verweilen (usiä). (1953:48) Das Aussehen eines Dinges ist das, worin es sich uns, wie wir sagen, präsentiert, sich vorstellt und als solches vor uns steht, worin und als was es an-west, d.h. im griechischen Sinne ist. Dieses Stehen ist die Ständigkeit des von sich her Aufgegangenen, der physis. (1953:138) Der Mensch ist das in sich offene Da. In dieses steht das Seiende herein und kommt zum Werk. Wir sagen daher: das Sein des Menschen ist, im strengen Sinne des Wortes, das "Da-sein". (1953:156)

304

Wir wollen diese aus dem Zusammenhang gerissenen Zitate nicht im einzelnen interpretieren, sondern auf ihre sprachliche Form hinweisen. Heidegger verwendet zur Paraphrasierung von sein durchgängig eine Raum-Metaphorik, die hauptsächlich um das Verb stehen und die zugehörige Wortfamilie angesiedelt ist. Man vergleiche: dastehen, Stand, darstellen, Ständigkeit, entstehen, ständig, vorstellen Auch andere Ausdrücke der Befindlichkeit oder Bewegung im Raum werden verwendet: Anwesenheit, zum Stand kommen, im Stand bleiben, hoch gerichtet, darbieten, stehen, Hervorkommen, Erscheinen, Verweilen, aufgehen, da, hereinstehen, kommen, Dasein Die Raum-Metaphorik wird auch dort verwendet, wo es um das Aussehen geht, d.h. um Sachverhalte, die normalerweise durch Sätze wie Das Auto ist grün, Das Gesicht ist rund ausgedrückt werden. Da Heidegger das Aussehen mit griechisch eidos und idea in Verbindung bringt, den Bezeichnungen für die Art und das zugrundeliegende Wesen der Dinge, wird die Raummetaphorik auch auf das Verständnis von Sätzen wie Sakrales ist ein Mensch, der Mensch ist vernünftig, ein Lebewesen übertragen. Heidegger gründet sein Seinsverständnis also auf einen bestimmten, zwar weit verbreiteten, aber keineswegs dominierenden Alltagsgebrauch von sein. Er versteht "Sein" nach dem Paradigma von Sätzen wie In den Weinbergen ist die Reblaus, Der Hund ist im Garten. Die Tradition folgte anderen Paradigmen, z.B. dem Paradigma Der Mensch ist ein vernünftiges Lebewesen oder Gott ist. Heideggers philosophische Einsichten stehen hier nicht zur Diskussion. In linguistischer Hinsicht kommt ihm das Verdienst zu, die Bedeutung einer Gebrauchsweise von sein herausgearbeitet zu haben, die vor ihm kaum beachtet worden ist.

7.

Die lokative Interpretation von sein

In der neueren Sprachwissenschaft gibt es Ansätze, die Heideggers Verständnis bestätigen. Sie haben den Raumaspekt von sein, den Heidegger herausgearbeitet hat, auf ihre Weise aufgegriffen und weitergeführt. John Lyons unterscheidet in seiner Einführung in die moderne Linguistik vier Funktionen des Verbs sein. Neben der existentiellen, der identifizierenden und der attributiven bzw. prädikativen Funktion führt er die lokative Funktion explizit auf und weist auf die Verwandtschaft von existentieller und lokativer Funktion hin: Für einen Zusammenhang zwischen existentiellen und lokativen Konstruktionen spricht der Umstand, daß in den existentiellen Sätzen einer ganzen Reihe von europäischen Sprachen ein ursprünglich lokatives (und noch genauer: ein deiktisches [...]) Adverb zur Anwendung kommt: vgl. deutsch da (in ist da, vgl. auch das Dasein), englisch there (in there ist are [...]), französisch y (in il y a), italienisch ci (in ci sono u.s.w.). (Lyons 1971:398-99)

305

Noch deutlicher arbeitet Charles Kahn in seiner Untersuchung von griechisch einai den Raumaspekt heraus. Folgende Gesichtspunkte seien hier hervorgehoben: 1.

Kahn sucht eine "Urbedeutung" für idg. es-, aus der einerseits der "vitale Gebrauch" (z.B. er ist = er lebt} und andererseits der kopulative Gebrauch abgeleitet werden kann.

Er findet sie in einem strengen oder prägnanten lokativen

Gebrauch, bei dem das Verb als Kopula konstruiert ist und zugleiche eine Art existentielle Funktion hat, z.B. Er ist anwesend, ist vorhanden. Die etymologische Hypothese wird allerdings insoweit relativiert, als Kahn die Annahme bezweifelt, daß Urbedeutungen notwendig konkret seien. (Kahn 1973:373-85) 2.

Kahn betrachtet die Raumvorstellung - nun synchroniser! gesehen - als grundlegend für das Denken im allgemeinen und speziell für den Begriff der Existenz. Er weist darauf hin, daß die Raummetaphorik bei Philosophen durchgängig dazu verwendet wird, die Existenz auszudrücken. (Kahn 1973:385-88)

3.

Kahn sieht das Verb sein in einem System, das durch die Oppositionen "statisch vs. veränderlich" und "prädikativ vs. lokativ vs. possessiv" gekennzeichnet ist (vgl. Kahn 1973:389): statisch

veränderlich

prädikativ

X ist klug, Präsident

X wird klug, Präsident

lokativ

X ist in Chicago

X kommt/geht nach Ch.

possessiv

X hat Geld

X bekommt Geld

In diesem System ist sein zweimal vertreten: in der prädikativen und der lokativen Zeile. Wir können also abschließend festhalten, daß Heidegger den lokativen Gebrauch von sein in seiner grundsätzlichen Bedeutung entdeckt und herausgearbeitet hat. Die Folgerungen, die sich für die Philosophie ergeben, übersteigen den Rahmen dieser Überlegungen. Es ist aber durchaus beachtenswert, daß die Reflexion des vorher meist übersehenen lokativen Gebrauchsaspekts beim Verb sein zur Entwicklung eines neuartigen philosophischen Ansatzes beigetragen hat.

Bibliographie Admoni, W. (41982): Der Deutsclie Sprachbau. - München. Aristoteles (1831-70): Aristotelis Opera. Ex recensione I. Bekkeri ed. Academia Regia Borussica. - Berlin.

306 Aristoteles (1925): Kategorien: Lehre vom Satz (Peri hermeneias). Übs. von E. Rolfes. - Hamburg. Aristoteles (1979): Metaphysik: Schriften zur ersten Philosophie. Übs. von F.F. Schwarz. - Stuttgart Amauld, A. (1972): Die Logik oder die Kunst des Denkens. - Darmstadt (franz. "1685: La logique ou Van de penser. Amsterdam). Bald, W.-D. (1972): Studien zu den kopulativen Verben des Englischen. - München. Camap, R. (1931): "Überwindung der Metaphysik durch logische Analyse der Sprache." - In: Erkenntnis 2, 21941 (Neudruck Amsterdam 1967). DWB (1854-1960; 1984): Deutsches Wörterbuch. Von Jacob und Wilhelm Grimm. 33 Bände. - Leipzig/ MUnchen. Engel, U. & H. Schumacher (1976): Kleines Valenzlexikon deutscher Verben. · Tübingen. Erben, J. (1978): "Über 'Kopula'-verben und Verdeckte' (kopulalose) Ist-Prädikationen." - In: H. Moser, H. Rupp & H. Steger (Eds.): Deutsche Sprache: Geschichte und Gegenwart. Festschrift für Friedrich Maurer zum 80. Geburtstag (Bern/München) 75-92. Heidegger, M. (1953; 41976): Einßhrung in die Metaphysik. - Tübingen. Kahn, C.H. (1973): The Verb 'Be' in Ancient Greek. - Dordrecht/Boston. Lyons, John (1971): Einführung in die modernene Linguistik. - München (engl. 1968: Introduction to Theoretical Linguistics. Cambridge). Verhaar, J.W.M. (Ed.) (1967ft): The Verb 'Be' and Its Synonyms: Philosophical and grammatical studies. Part 1-4 & 6. Dordrecht/Boston (=Foundations of Language, Supplementary Series). Weber, H. (1988): "Zur Inhaltsstruktur des Verbs 'sein'." - In: J. Lüdtke (ed.): Energeia und Ergon. Vol. 3: Das sprachtheoretische Denken E. Coseruis in der Diskussion (2) (Tübingen) 501-22.

307

Paul N. Werth

The English Conditional - Tense, Aspect or Mood?

1.

Introduction

In this paper I will apply a model of English grammar that I have been working on for some years (Werth 1981a, 1984, 1985, 1986, 1989) to some problems in the semantic explanation of the English verb-paradigm system. The model is a discourse-level account which I am now tentatively calling Situational Grammar since the notion of situation is central to it. I take the notion of explanation rather seriously - i.e., I do not mean by it anything like 'internal consistency1, though this notion in itself is of course a necessary and important one in any model-building activity. I take an explanation to be an account of the functionality of a given phenomenon in the model, i.e. what it does and what value it has in comparison with the other elements of the model. An advantage of any new approach, of course, is that it enables you to look at old categories in a new light, and from the new arrangements and relationships which your model throws up, to get fresh insights into old problems which have always seemed intractable, or which other approaches have comfortably managed to ignore or shelve. As a case in point, I shall be considering the so-called conditional in English, and specifically the semantics/ pragmatics of the verbal forms with would. I shall be looking at their conditional use together with other uses of complex verb forms in would. A random glance through the literature reveals a basic lack of agreement in the initial categorisation of this phenomenon: Traditional school grammars (such as Nesfield 1917) talk about the conditional mood, and alternate the term 'conditional' with the term 'subjunctive'. Thus the conditional is in the mood system together with 'indicative1 and 'imperative1. English language teaching grammars (e.g. Thompson and Martinet 1980) either talk blandly of conditional tense, or else (e.g. Stannard Allen 1974, Hornby 1954, cf. also Pearce 1975) avoid the question by talking about conditional clauses as a kind of adverbial (which ducks the issue of what exactly the verb forms are). Scholarly grammars (e.g. Quirk et al. 1972,1985) treat conditionals by taking their use of would, would have as a modal form like any other, while specific linguistic studies (e.g. Hopper 1982, Lo Cascio and Vet 1986) hardly mention the conditional at all, on the whole, assuming like Quirk et al. that it is wholly covered by the account of modals (there are, of course, exceptions to this generalization, such as Palmer 1965, 1974, 1977, 1979, Huddleston 1977, 1979, N. Smith 1981, Werth 1981b, Haegeman and Wekker 1984, all of which are to some extent specifically devoted to the conditional; many of these tend to follow the modern orthodoxy, nevertheless, in treating would forms as modal verbs pure

308

and simple, and rather few of them are centrally concerned with the semantics of these forms, rather than the syntactic constructions they appear in). The traditional term 'mood1 (cf. Lyons 1977:746) is a notional category which nowadays falls under the general heading of speech-act, with a three-way distinction into indicative, subjunctive and imperative, indicating respectively factuality, hearsay and command (in the case of the subjunctive, there is a syntactic distinction also, which is sometimes mentioned, namely the fact that it classically occurs in subjoined clauses, hence the name. But this distinction is in fact neither necessary nor sufficient to identify subjunctives or, a fortiori, conditionals). Tense' strictly speaking refers exclusively to temporal deixis, although the term is often used loosely to refer to a member of verbal paradigm, i.e. as a term denoting a verbal form rather than a verbal meaning or function. Semantically, it is conventionally agreed that there are three tenses, Past, Present and Future, though it is usually remarked that English has only two differentiated simple forms to denote tense distinctions, namely Past and non-Past. I shall subsequently present a somewhat more complicated model (though not unduly so), which I claim more sensitively handles the expression of time in the verb. 'Aspect' is said to be a non-deictic category (cf. Lyons 1977:705), denoting such distinctions as duration, instantaneity, frequency, completion. "Aspects are different ways of viewing the internal temporal constituency of a situation" (Comrie 1976:3).

I shall be

modifying both the claim of non-deicticity and the list of phenomena which are conventionally said to be aspectual in English. In summary, therefore, the facts about the conditional seem clear enough - in the sense that there is no serious disagreement over data. But there is nothing approaching consensus in respect of the status of this phenomenon, nor has there been much interest in explaining it. I therefore wish to look at conditionals in relation to other verbal forms, and specifically other forms with would; I want to provide a textual explanation of their behaviour; and finally, from this explanation, I should like to settle the question of the status of the conditional. I shall first give an outline of an approach to grammar in general, and verbal function in particular, which I find these notions naturally fit into. Then I shall come back to the conditional to see what insights the new model has generated.

2.

Situational Grammar

A discourse is a fusion of a text and its relevant context. The discourse participants go into a discourse-event with a certain amount of 'baggage' which we shall here refer to monolithically as {K}, the propositions constituting a person's knowledge-base. Which of these propositions count as 'relevant' in the context is defined by the text, in this sense: the text itself consists of

309

a set of propositions {P}.

But these propositions have many notional links with other

propositions which are unexpressed, but nevertheless Present. Some of them {PE} are more or less directly connected with the expressed propositions as entailments. Others {PjJ, probably the majority, are pragmatically connected in that they relate the propositions actually expressed {P} to speaker and listener knowledge, {Ks} and {K^}.

All of these

notional links, then, constitute, at the start of the discourse, areas of 'potential relevance' to the propositions of the text, and as the text proceeds, some of these areas are actually activated by the text, white others are disactivated. Notice that, for the listener at least, the whole process is discourse-driven, while for the speaker, the discourse is the sole means by which the context (s)he wishes to construct may be conveyed. It is perhaps worth pausing for a moment to review the actual process of relevanceformation here. So far, we have three sets of discourse-specific propositions: {P}, {PE} and {PK}. At the start of the discourse, the content of {P} is, of course, extremely limited, and therefore the same is true, comparatively speaking, of {PE}. {PK}, on the other hand, is still very large, since it consists of the (pragmatic) intersection of {K} with only one proposition. As new propositions are introduced into the discourse, {P} grows in direct proportion, {PE} also grows, though not necessarily in direct proportion, since some of the entailments of earlier propositions may be blocked by later propositions and vice-versa. The set {PK}, on the other hand, gets more and more restricted, since successive intersections give ever smaller subsets. Given that in conversational discourse the function of the speaker typically alternates between the participants, we can see that the context in such a case is being jointly negotiated by the participants. It is perhaps less clear that this is still the case even in monologues - but since any set of propositions 'put forward' by the speaker in such a case nevertheless has to pass through the filter of the listener's knowledge, it seems reasonable to. hold that even then there is an important element of 'negotiation' - albeit rather more passive - taking place (note that this position fits well with narratological accounts of the function of the reader in the interpretation process, e.g. Riffaterre 1966). The set of propositions actually expressed {P} contains a subset of deictic terms {D}, denoting place, time, relationships with the speakers, nominated entities and knowledge relating to all of these categories. This deictic subset I call the text-world, and it defines the conceptual space into which the discourse refers. The set {P} of expressed propositions, together with their entailed propositions {PE} and the propositions they activate {PK} from the potentially relevant propositions in knowledge {K}, all of these together constitute the Common Ground of the discourse. Thus the Common Ground (henceforth: CG) is equivalent to the set of acutally relevant propositions in that discourse (for a very different view of what constitutes 'relevance', see Sperber and Wilson 1986).

310

The propositions expressed refer to, though they do not necessarily in themselves completely define, the situations which constitute the foreground of the text. A situation is a complex unit made up of entities in relationships; it takes its deictic framework from the context (specifically, the text-world), although it may contain further deictic elements which serve to fine-tune the general orientation provided by the text-world - we shall see cases of this in the system of verbal tense. The situations thus explicitly expressed must necessarily be possible situations in the text-world of the discourse: I take this to mean that they must, at least be consistent with the propositions defining that text-world, but also, and more strongly, they must cohere with those propositions. The requirement of coherence is assured by a coherence constraint (cf. van Dijk 1972:128f., Werth 1984:89ff.), the reflex of which is informationflow. Information flow regulates such surface-structure phenomena as constituent-order and sentence-stress (see Werth 1984: passim).

3.

The tense system

Like much contemporary work on tense, the system I will use is based on Reichenbach 1947, in that it analyses verbal time-expressions into a Speech Time (ST) component, a Reference Time (RT) component, and an Event Time (ET) component. Reichenbach himself and most later users of the system (e.g. Hornstein 1977, C.S. Smith 1978, de Vuyst 1980) have assumed that the three terms apply linearly (indeed, much of Hornstein's paper is devoted to the linear 'syntax' of his notation). Other studies (e.g. Bertinetto 1985, Rohrer 1985, Adelaar and Lo Cascio 1985) argue that a single RT is inadequate: they propose a temporal 'localiser' function when RT=ET, and a Reference Time 'proper' when RT^ ET (Rohrer, in addition, proposes three more varieties). I will show, however, that it provides a much more satisfying explanation if we view the application of the terms as being layered. One of the arguments for this is its observational and descriptive adequacy: it predicts only nine separate tense possibilities, whereas the 'undiluted' Reichenbach system allows 13, and the Hornstein system, unconstrained, allows 22. Hornstein's system of constraints brings his total down to 9, but oddly enough these are a different nine from mine! I will come back to this comparison after I have presented my system. Let us first deal with the 'layered' nature of my approach. Taking the three Reichenbachian 'times' in turn, ST marks the deictic starting point of the discourse, RT places the principal time period of the situation concerned, while ET represents the actual time of the event itself. Here are some examples to distinguish the three concepts: (1)

Pete had finished by 4 o'clock ET (Pete finishes) is before RT (4 o'clock) is before ST (now)

311 or alternatively: ET ET->RT

'Future-in-thePast'

Intentional Future

Future + adv.

RT/ET

When ST=RT, then the text-world is in the Present time-zone, and the foreground situations relate to the Present time. This means that they are likely to refer to the immediate situation (i.e. as perceived by the participants). This is the explanation for the frequent gloss for the Present Perfect of 'current relevance1 or 'having a present result'. It also explains the presence in Diagram 2 of the Present time-zone tense which I have called the 'Intentional Future'. I regard this as a Present tense since it is a projection of the Present state of mind of the subject, whereas the so-called True Future' represents a prediction, speculation etc. based in the Future time-zone. When RTST, the text-world is in the Future time-zone. In both of these cases, the foreground situations will refer into the wider context (i.e. as remembered, imagined etc. by the participants). So, the ST/RT relationship gives the general time-zone (and in the text this might be expressed by adverbs or by the preceding sentences), while the RT/ET relationship fine-tunes the more precise time of the situation within that time-zone. To represent this layering, I shall henceforth bracket the notations as follows: ET

C. would have intervened

——

(pAST EPISTEMIC +

III

REMOTE)

Let me now provide a longer text containing several examples of different kinds of conditionals in different configurations. I will not diagram the various conditional combinations, since the results will closely resemble one of the above three diagrams anyway. I will rather comment on the textual circumstance which differentiate between such natural manifestations of the phenomenon and the isolated sentence we have dealt with so far. The text is taken from a play by David Mercer. The character Tom, backed up by his wife Marion (Isobel's sister), has proposed that his firm should take over Isobel's company.

Isobel is

exploring the implications of this proposed alternative world.

(53) Tom: Isobel: Marion:

Tom: Marion: Tom: Isobel: Tom: Isobel: Tom: Isobel:

My company has the spare money. We want to use it. We want to help your firm expand. Because, well, our motives are wonderful. We happen to believe in you. Yes, well, that's jolly good. It's just the FORM of the arrangement. Isobel's worried about the idea of a board. (Tom frowns) Oh, I see. She's used to owning her own firm. Yes, but surely you'd expect us to protect our own investment (He stands for a moment, genuinely puzzled). I don't think there's anything sinister in that. It's pretty normal practice. Your company would own us? Well, yes, indirectly. They wouldn't interfere. After all, you'd have a board of your own. Of which you'd be chairman? Technically. And of which I would be one single member? (TOM frowns again)

332

Tom: Marion Tom: Isobel: Tom: Marion: Tom: Marion: Isobel:

Marion:

You'd also be managing director. (Firmly): Isobel, Tom is President of Christians in Business. I think that makes it pretty clear he's a man you can trust. We meet six times a year. We try to do business the way Jesus would have done it. You mean, had he come to earth in a polyester suit and with two propelling pencils in his top pocket? I'm sorry? Isobel's making a joke. Oh, I see. Tom is out there in the community. He runs all these schemes. For youth. Don't you dear? Of course, Tom's honesty is not at issue. It's just I fear I'd be losing control. (Isobel shakes her head, frustrated now, beginning to get angry) Why don't you just GIVE me the money? I find that question unforgivably naive. (David Mercer)

(53) is, of course, part of a much longer text. Nevertheless, it has its own integrity, since it establishes a separate text-world: a possible Future, in which Tom's company in effect takes over Isobel's. The propositions defining this text-world are contained in Tom's first speech (and we must also infer from Isobel's questions that the subject has been raised prior to this scene, but outside our hearing, as it were - these inferences also constitute world-defining propositions). The definitions of this world include: (54)

(a) (b) (c) (d) (e) (f) (g) (h)

Tom's company wants to use its spare money They want to help Isobel's company Tom's company will own Isobel's company There will be a controlling board There will be a second board within Isobel's company Tom will be chairman of the second board Isobel will be one member of the second board Isobel will be managing director of her own company

Of course, there are many other pragmatic implications arising out of general knowledge: the likely social relationships and the tensions associated with these, implications connected with the business world, the contrast between Tom's philanthropic words and Christian claims and the ruthless and barely concealed commercial advantage he wants to exert, and so on. Of the propositions in (54), (c-h) represent the proposed situation, while (a-b) relate to the overt, expressed motivation for the proposal. There are two general points: first, all of the conditionals, with only one exception, are statements, restatements or clarifications (i.e. including entailments) of (54c-h); secondly, there is not one single occurrence of if in the text: there is in fact only one protasis clause at all, which accompanies the sole exception to the previous point, but is not part of the same sentence as its apodosis nor is it even uttered by the same speaker. The actual occurrence of conditionals in a text then is a good deal freer than

333

the manufactured examples such as (38-40) which are usually discussed in this connection. This is because the condition(s) of the world-within-a-world represented by i/-clauses are not in fact on the sentential level, except accidentally. They need be stated only once in the text, if at all. In the case of text (53), they are not even stated explicitly - the reader or listener is left to infer the appropriate i/-clause: 'If Tom's proposal to help Isobel's company were implemented.' The Situational Grammar model therefore assumes that the world-within-aworld conditions always occur at the discourse level, rather than the text level; in this case they form part of the inferential and background knowledge content, rather than of the explicit linguistic content of the discourse - the latter being what is abstractly represented by the text. The single alternative-Past conditional is distributed across two sentences and two speakers, as already pointed out. Its protasis is, in fact, an ironic definition of the world in which its apodosis (do business the way Jesus would have done it) would fit. In all cases of conditional constructions in (53), I claim, the verb forms are Remote, marking the hypothetical nature of their world. The single protasis is the Remote equivalent of the non-hypothetical: (55)

(a) When Jesus came to earth ...

while its apodosis is the Remote equivalent of the Past epistemic: (b) he (certainly) will have done business this way. The other conditionals are all Remote equivalents of the forms given in (54c-h) or entailments thereof. The only question remaining, in my mind at least, is: what exactly are the will forms in (54c-h), true Future, intentional Future or epistemic?

It seems to me that since they

represent proposals rather than predictions or assessments of probability, the correct answer must be that they are intentional Futures, i.e. they are properly to be considered in the Present time-zone, rather than in the Future or modal zones.

This would mean that the non-

hypothetical equivalent of the presumed i/-clause would be something like: (56)

I accept your proposal

To conclude, then, what is the answer to the question we set ourselves at the outset? Our story has been that conditionals occur in a world-within-a-world, set up in the discourse, and their consequent hypotheticality is marked by the persistent use of the Remote. Consequence clauses often represent epistemic conclusions, but can alternatively be statements of intent (I suspect that predictions, as such, are not possible, given the hypothetical framework, which would tend to provide assessments of probability rather than confident predictions).

This

arrangement of senses suggests that aspect has nothing to do with the question whatsoever. Conditionals, I suggest, are a combination of tense and modality features: tense in that the whole Remote system is parasitical upon the tense system, and also in that the basic ST/RT

334

configuration, which gives the time-zones, seems to underlie the distinctions which are possible; modality in that the epistemic sense appears to be the most frequent meaning of would. There is another point about Remoteness which seems to bring us back to the virtually discarded term 'mood1: in its use to denote hypotheticality, Remote seems to suggest one of the principal traditional functions of the subjunctive, namely to denote that the speaker cannot vouch for the truth of the statement in question.

Bibliography Adelaar, M. & V. Lo Cascio (1985): "Temporal Relation, Localization and Direction in Discourse." - In: V. Lo Cascio & C. Vet (eds.): Temporal Structure in Sentence and Discourse (Dordrecht) 251-97. Bertinetto, P.M. (1985): "Intrinsic and Extrinsic Temporal References: On Restricting the Notion of 'Reference Time'." - In: V. Lo Cascio & C. Vet (eds.): Temporal Structure in Sentence and Discourse (Dordrecht) 4178. Bossuyt, A. (ed.) (1986): Functionality in Linguistics. - Brussels. Brown, P. & S.C. Levinson (1987): Politeness. - London. Comrie, B. (1976): Aspect. - London. Daalder, S. & M. Gerritsen (eds.) (1980): Linguistics in the Netherlands. - North Holland. Dahl, O. (1975): "On Generics." - In: E.L. Keenan (ed.): Formal Semantics of Natural Language (Cambridge) 99-111. de Vuyst, J. (1980): "Lexical Tense Representations". - In: S. Daalder & M. Gerritsen (eds.): Linguistics in the Netherlands (North Holland) 225-34. Fillmore, C. (1975): Santa Cruz Lectures on Deixis. - Indiana. Haegeman, L. & H. Wekker (1984): "The Syntax and Interpretation of Futurate Conditionals in English." - In: Journal of Linguistics 20,45-56. Hopper, P. (ed.) (1982): Tense-Aspect: Between Semantics and Pragmatics. - Amsterdam. Hornby, A.S. (1954): A guide to patterns and usage in English. - London. Hornstein, N. (1977): "Towards a Theory of Tense." - In: Linguistic Inquiry 8,521-77. Huddleston, R. (1977): "Past Tense Transportation in English." - In: Journal of Linguistics 13,43-52. Huddleston, R. (1979): "'Would Have Become1: Empty or Modal WUT?" - In: Journal of Linguistics 15, 335-40. Jackendoff, R.S. (1983): Semantics and Cognition. - Boston. Joos, M. (1964): The English Verb: Form and Meanings. - Madison. Lo Cascio V. & C. Vet (eds.) (1986): Temporal Structure in Sentence and Discourse. - Dordrecht

335 Lyons, J. (1977): Semantics. - London. Nesfield, O. (1917): Outline of English Grammar. - London. Palmer, F.R. (1965): A Linguistic Study of the Engish Verb. - London. Palmer, F.R. (1974): The English Verb. - London. Palmer, F.R. (1977): "Modals and Actuality." - In: Journal of Linguistics 13,1-24. Palmer, F.R. (1979): Modality and the English Modals. - London. Pearce, RA. (1975): "Teaching Conditional and Related Clauses." - In: English Language Teaching Journal 29, 206-13. Quirk, R., S. Greenbaum, G. Leech & J. Svartvik (1972): A Grammar of Contemporary English. - London. Quirk, R. (1986): Words at Work. - London. Quirk, R. (1985): A Comprehensive Grammar of the English Language. - London. Reichenbach, H. (1947): Elements of Symbolic Logic. - New York. Rescher, N. (1975): A Theory of Possibility. - Oxford. Riffaterre, M. (1966): "Describing Poetic Structures: Two Approaches to Baudelaire's Les chats." - In: Yale French Studies, 200-42. Rohrer, C. (1985): "Indirect Discourse and 'consecutio temporum1." - In: V. Lo Cascio & C. Vet (eds.): Temporal Structure in Sentence and Discourse (Dordrecht) 79-98. Smith, C.S. (1978): "The Syntax and Expression of Temporal Expressions in English." - In: Literature and Psychology 2, 43-99. Smith, N. (1981): "Grammaticality, Time and Tense." In: Philosophical Transactions of the Royal Society of London B295, 253-65. Sperber, D. & D. Wilson (1986): Relevance. · Oxford. Stannard, A. W. (1974): Living English Structure. - London. Thompson, A.J & A.V. Martinet (1980): A Practical English Grammar. - London. van Dijk, T. (1972): Some Aspects of Text Grammars. - The Hague. Werth, P.N. (1981a): "The Concept of 'Relevance' in Conversational Analysis. - In: P.N. Werth (ed.): Conversation and Discourse (London) 129-54. Werth, P.N. (1981b): "Tense, Modality and Possible Worlds." - In: Rapport des Actf/vir^s de l'Instüute de Phon4tique 16,17-30. Werth, P.N. (ed.) (1981c): Conversation and Discourse. - London. Werth, P.N. (1984): Focus, Coherence and Emphasis. - London. Werth, P.N. (1985): Anaphora and Coherence. - MS.

336 Werft, P.N. (1986): "A Functional Approach to Presupposition: Pulling the Plug on Holes and Filters." - In: A. Bossuyt (ed.): Functionality in L inguistics (Brussels) (^Belgian Journal of Linguistics 1) 239-79. Werft, P.N. (1989): On Presuppositions, French Kings and Other Imaginary Creatures. - MS.

Syntax

339

Hartmut Czepluch

Non-Configurational Properties in a Configurational Language: The Case of English The paper presents a new view of the syntactic structure of Complements and Adjuncts in English. If feasible, the analysis throws new light on the recent typological discussion about configurational and flat languages.

0.

Introduction

It has been a standard assumption of Generative Syntax during the last 25 years or so that English sentence structure is configurational in two major respects: first, English has a Verb Phrase constituent excluding the Subject position; and, secondly, inside the Verb Phrase, the verbal head and its lexical Complements form a minimal phrasal domain to the exclusion of Adjuncts, i.e., Complements and Adjuncts necessarily appear on different structural levels. While the first claim, that English has a NP-VP (or: Subject-Predicate) structure, is beyond doubt, the second claim may be called into question on empirical and theoretical grounds. In fact, it is contended here that basically the English Verb Phrase has a flat structure. The structure of the paper is as follows:

Section 1 sketches the common core of

descriptive assumptions for layered VP structures in English.

Section 2 and 3 discuss

empirical and theoretical problems, respectively, for a configurational VP in English. Section 4 provides an alternative account within the flat VP view for fixed word-ordering effects and the existence of structural subdomains in English. Section 5 compares this with the arguments for a hierarchical VP in German and addresses the question of how these somewhat conflicting results may be reconciled within a modular model of grammar. The final section sums up the major results of the investigation and draws some conclusions as to the configurationality issue.

1.

Standard assumptions about English VP structure

It is nowadays widely assumed that the phrasal structure of constituents is an X-bar projection over lexical categories X° according to the schema (1):

340 (1)

XP I

I

I

Specifier

X' I

X°

Coipleient(s)

The head X° and its Complement(s) form an intermediate phrasal category X', whereas X1 and specifiying elements make up the maximal phrase XP (cf. Chomsky 1986b). While linear ordering relations between specifier, head and Complement(s) may vary parametrically in languages, the hierarchical relations are taken to hold uniformly across categories and across languages (cf. Travis 1984). In its strongest form, X-bar theory holds that there are no category or language-specific PS rules at all (cf. Stowell 1981).1 1.1

Levels of hierarchy in the English verb phrase

It goes without saying that the basic X-bar schema does not exhaust the structural possibilities in English or other languages.2 For the English VP, there are two slightly differing analyses, shown in the tree diagrams (2) and (3) for the sentence John played the tune on a guitar for three hours in Carnegie Hall on Tuesday. Both analyses agree that the verb and its lexical Complements - here: played the tune - form a minimal domain to the exclusion of adverbial Adjuncts. It is in the analysis of Adjuncts where the proposals differ:

* X-bar theory was introduced in Chomsky (1970) as a substantive universal that gave a restrictive definition of the format that language-particular PS rules must conform to. In the context of the Lexicalist Hypothesis, it aimed to capture cross-categorial generalizations about grammatical functions and the application of rules (e.g., in S and NP). Last but not least, it answered the critical observation of Lyons (1968:331) that a system of PS rules without some metatheoretical constraint does not preclude nonexisting rules like *NP —> V NP or *VP —> Del N PP. Clearly, Stowell's (1981) arguments for a category-neutral base that ideally only contains the X-bar schema extended the explanatory force of X-bar theory. 2 Only three types of categories are assumed to exist: lexical and functional heads X°, intermediate phrases X' and maximal phrases X" = XP (cf. Stowell 1981, Muysken 1982, Chomsky 1986b). To account for Adjuncts, the intermediate category X' needs to be recursive. Heads and maximal projections may also be recursive: for XP, this follows from Adjunction-to-XP processes (Extraposition, Heavy NP Shift etc.); for X°, incorporation and reanalysis processes justify complex lexical categories. Opinions on SpecVP position differ: according to Akmajian, Steele and Wasow (1979), the aspectual auxiliaries are VP specifiers; similarly, Roberts (1987) and Radford (1988). Fukui (1986) and Speas (1986) take the θ-subject of a verb to originate in SpecVP (cf. also Koopman and Sportiche 1988). And Pollock (1989) associates the position with VP-initial adverbs.

341 (2)

Jackendoff (1977):

VP ι

ι

(Spec)

V i

John played the tune on a guitar for three hours in Carnegie Hall on Tuesday

(3)

Radford(1988): VP ι

\

(Spec)

ι V

ι V ι

V I

I V

ι V John played the tune on a guitar for three hours in Carnegie Hall on Tuesday

(a) ... and Bill did ySO too] (b) ... and Bill did on Monday] δ (c) ... and Bill did in the Royal Albert Hall on Monday] . ν ° ; ; (d) ... and Bill did yiSO for half an hour in the Royal Albert Hall...] (e) ... and Bill did on a trombone for half an hour...] (f) *... and Bill did [VP[vso] a folk song on a trombone ...] Jackendoff (1977) distinguishes three types of post-verbal elements on the basis of syntactic and functional considerations:^ functional arguments (i.e., Objects and subcategorized Adverbials), restrictive (e.g., Means, Manner, Instrument, etc.) and nonrestrictive modifiers (e.g., Place, Time, VP adverbials, etc.). Each type appears at a different level of structure, whereas elements of the same type appear at the same level. Structure (2), then, gives the maximal hierarchy of a VP. - As to Radford's structure (3), each Adjunct projects a phrasal

Structure (2) is a more modern X-bar adaptation of Jackendoff s original Fixed-Level Analysis, which also incorporated the Uniform Three-Level Hypothesis and the 5=VP Hypothesis (1977:54 et passim).

342

level of its own. The VP may thus have any degree of hierarchical depth, depending on the possible number of functionally distinct Adjuncts. Both analyses have their merits: Jackendoffs analysis nicely accounts for unmarked constituent ordering of V < arguments < restrictive modifiers < non-restrictive modifiers, in line with Williams' (1975) notion of "natural distance to the verb"; and also that ordering among elements of the same hierarchical level is (relatively) free. - Radford's analysis is better suited for the effects of grammatical processes like VP Deletion, Coordination or so Substitution. Consider, e.g., the possible so Substitutions given below structure (3): the proform so may exclude any single Adjunct, as in (3b-e); only the Argument domain of the verb cannot be broken up in this way - hence, the ungrammaticality of (3f).4 Taking all these possibilities together, each Adjunct seems indeed to be realized at a separate structural level. 1.2

The alternative

Here the extreme opposite view to both analyses will be taken -as shown in (4) for the example sentence:

(4)

.. VP ... |

(Spec 1

V

f° I

HP PP , 1 . .Ι I I I

ΙI IΙ

I

I

I

PP

PP

PP

I

ΙI IΙ

I

ΙI I Γ

I I

John played the tune on a guitar for three hours in Carnegie Hall on Tuesday

The claim that English has a flat VP basically does not preclude that there may be more structure. But such additional structure only arises as the effects of grammatical processes (cf. Sect. 4.1). Since such a flat structure runs counter to what nearly everybody else takes for granted,5 the view will be substantiated in the following by empirical, theoretical and comparative considerations. The combined effect makes it unlikely that we are concerned with a purely theory-internal aspect of present GB theory.

4 By (3a), so appears to be a V'-proform uniformly, which makes it completely parallel to the one substitution pattern in NP, one exclusively substituting for N' constituents. In (3a) too would have to be regarded as a VPfinal specifier and non-restrictive Adjuncts as V' daughters rather than VP daughters (cf. (2) and (3)). 5

A notable exception is Napoli (1988).

343 2.

Empirical problems for hierarchical layering

In this section we will look at some empirical phenomena in English which appear to be problematic for a strict configurational distinction between Complements and Adjuncts. Before doing so, it may be appropriate to introduce a terminological distinction between (syntactically) un-lmarked and (pragmatically) un-lnatural constituent ordering. English is commonly assumed to be a language with a (relatively) fixed word-order. Thus, for sentences, with the 3-place verb put, sentence (5a) represents the syntactically unmarked ordering. Constituent orders deviating from canonical ordering do occur: an NP object may appear finally if it is 'heavy' in some sense, as shown by the contrast in (5b), or if it carries contrastive stress, as shown by the contrast in (5c): (5)

(a) John [vp[vPut tne book on the table]] (b) John fvp[vpput e\ on the table] [the book he had bought the day before] J *Jonn put on the table the book (c) John [vp[vpput et on the table] [the book]J (not the term paper) *John put on the table the book (not into the drawer)

This has been taken as indicative of a Scrambling process which extracts a phrase from its canonical position and Chomsky-adjoins it to the first maximal projection. Adjunction structures thus leave gaps in Complement positions.^ Moreover, Adjunction seems to be ungrammatical unless licensed by some extra property of the postposed phrase. The absence of such properties then indicates that a sentence represents a syntactically unmarked order. Thus, the well-formed structures in (5b) and (5c) may be said to be syntactically marked because the application of Adjunction effects a deviation from canonical constituent order; but that they represent natural orderings with respect to pragmatic factors, such as the 'constituent weight' or 'marked intonation1. 2.1

Manner adverbiale

With this much as background, let us consider as a first empirical phenomenon the position of Manner adverbials in sentences such as (6): (6)

(a) John put the book on the table carefully (b) John put the book carefully on the table

One would expect the adverb carefully

to appear unmarkedly after the θ-marked

Complements of put in final position, as in (6a). The interesting fact is that native speakers at least my informants - prefer (6b) over (6a). Since the Manner adverb is an Adjunct, in a " According to the Radford-type analysis, where there are no syntactic criteria regulating the distribution of Adjuncts, Adjunct reordering should not leave a gap.

344 configurational VP, the order of (6b) would have to be analyzed as a VP-Adjunction of the Locative phrase; cf. (7): VP

(7)

I

1 PPj

I

VP I

1 V

1

1

1

carefully on the table

I

1

IIP

1

PP

put the book ej And a configurationalist would have to argue that (6b)/(7) is just natural because the postposed PP is either heavier than the one-word adverb, or because it carries contrastive stress. Such an analysis appears to be unmotivated, though. Neither of the requirements on Adjunction is satisfied by (6b): the sentence need not have a contrastive reading at all.7 Nor can the order of (6b) be attributed to the heaviness of the PP. This can easily be tested with sentences like (8): (8)

(a) John put the book on the table quickly but carefully (b) John put the book quickly but carefully on the table

Coordinating two adverbs should make the adverbial at least as heavy as the PP Complement. In a configurational VP, there should then be a clear preference for the (a)-sentence. But my informants still prefer (8b) with the internal Adjunct. So it is likely that the (b)-sentences of (6) and (8) represent syntactically unmarked orderings - hence, that an Adjunction analysis for them is unwarranted. If so, there are English sentences for which the structural distinction between Complements and Adjuncts does not hold on empirical grounds.8 2.2

Prepositional objects

My second empirical example concerns a similar ordering phenomenon. Verbs like argue, speak or talk, which take two prepositional objects, may realize them in any order, as shown in ex. (9): ' The fact that Focus intonation falls on carefully in (6a) and on table in (6b) is regular under normal sentence intonation. 8

Admittedly, Manner adverbials are special in this respect, and the fiat VP analysis per se does not explain these ordering facts. But since a flat structure more easily admits word order variation, the unmarked orders of (6/Bb) represent less of a problem than for a configurational VP. Moreover, the flat VP hypothesis does not falsely claim that (6/8a) represent the unmarked orderings.

345

(9)

(a) John argued with Mary about politics (b) John argued about politics with Mary

Under the flat VP hypothesis, the prediction is that with these verbs, an adverbial may appear in any postverbal position. The prediction is borne out by the ex. in (10): (10)

(a) John argued with Mary about politics vehemently (b) John argued with Mary vehemently about politics (c) John argued vehemently with Mary about politics

All three orderings are grammatical. Actually, my informants prefer (b) or even (c). As before, there are no good reasons to assume the marked properties characteristic of Scrambling.

So again, empirical facts suggest that Complements and Adjuncts need not

necessarily be distinguished hierarchically. 2.3

Pseudo-passives

As a final phenomenon to be discussed here, Pseudo-Passives may pose a problem for a layered VP.

In English, objects of prepositions passivize rather freely, leaving behind a

stranded preposition.9 Although this process is semantically and/or pragmatically constrained (cf., e.g., Couper-Kuhlen 1979, Davison 1980), it has always been assumed that structural factors do play a role,10 and only these will be of concern here. In GB theory, pseudo-passives are commonly analyzed in terms of a Reanalysis process, which takes as input a V-PP sequence and incorporates the preposition into the verb. The NP

9

According to Davison (1980), prepositional objects (see (a)-(b)), spatial and directional Locatives (see (c)-(d)) and Instrumentals (see (e)-(f)) passivize freely: (a) The clown was laughed at [^pe] by everybody (b) The decision has been waited for [fjpe] a long time (c) That bed has been slept in [^pe] (d) The valley has been marched through [j^e] in two hours (e) That knife has been cut with [^pe] too often without being sharpened (f) This spoon has been eaten with [j^pe]

Time adverbials (see (g)-(h)) and expressions of Cause (see (i)-(j)) are generally exempt from PseudoPassivization: (g) 'Dinner was quarrelled before/during/after [tipe] by John and Sue (h) *The night was typed through [^p£] by Susan (i) »Cowardice was run away from [^pe] by John (j) *The superintendent's negligence was fallen because of [NPe] by John 10

E.g., Hornstein and Weinberg (1981) correlate the pseudo-passive differential with the distinction between

VP- and S-adverbials; Rad ford (1988) gives it as one criterion to motivate the structural distinction between Complements and Adjuncts.

346

left-behind thus becomes the Object' of a complex verb.11 The most restrictive, and hence preferable, version of V-P Reanalysis requires V and PP to be adjacent and sisters (cf. Kayne 1981): (11)

...[V.V [PpP NP ]...] ... ==> ...[V{VV-P] NP ...]...

That is, there may be no intervening material, nor a category boundary between V and PP for Reanalysis to be applicable. With these conditions in mind, consider the Locative passive (12c): (12)

(a) They have [vp[v«aten (food)] at the table ] (b) Food has been eaten [^pe] at the table (c) The table has been [yeaten at]

Apparently, the Locative passive is formed from the objectless variant of (12a). Of course, if eat takes a Direct Object, only the passive form (12b) is possible. For lexical reasons and by standard constituency tests, the Locative phrase is an Adjunct in both variants of (12a). As such, it should be outside the minimal V domain in a configurational VP, as indicated in (12a). But V sisters are generally assumed not be accessible to passivization, which seems to require an Object'-like relationship: the Locative Adjunct should be a V° sister with respect to V-P Reanalysis. Thus, pseudo-passives from Adjunct position pose an empirical problem for the assumption that Complements and Adjuncts have to be realized on different levels of structure.12 The problem vanishes if arguments and nonarguments may be realized in a flat VP structure.

Reanalysis is motivated with respect to the Case absorption analysis of passives: without V-P Reanalysis, passive movement would leave behind a Case-marked trace, which is illicit; with V-P Reanalysis, the object NP is Case-less if the V is a participle. 2 To reconcile this state of affairs with a configurational VP, one would have to drop the sisterhood requirement on Reanalysis. But this would make V-P Reanalysis such a powerful mechanism that other problems are sure to arise. Moreover, a configurational basis does not seem to be the right track anyway: although nonrestrictive Adjuncts are generally exempt from Pseudo-Passivization, while restrictive Adjuncts quite freely admit it, there are exceptions for the latter type. Thus, other factors have to be appealed to anyway, and these properties would draw the appropriate distinctions in a flat VP also.

347 3.

Theoretical considerations

As a starting point to my argumentation from theory, it should be kept in mind that in the modular GB theory properties of syntactic structure are no longer fixed in an a priori way.13 Rather, such configurational properties as there are arise by the modular interaction of lexical properties and (parametrized) principles of grammar. The relevant question then is: how much structure do the principles of grammar determine? The contention is that GB theory actually does not determine hierarchical distinctions between Complements and Adjuncts. 3.1

Lexical Information

Let us begin with the lexical input into syntax. For ease of exposition, I assume that basically lexical entries are unordered θ-grid representations - i.e., they only specify how many arguments a lexical item is construed with, and which kinds of semantic relations hold between the lexical item and its arguments; cf. (13):14 (13)

(a)

put:

[+V,-N] [Agent, Theme, Location]

(b) mend:

[+V,-N] [Agent, Theme]

Apart from a category specification, lexical entries contain a minimum of syntactic information: the Agent roles in (13) are designated to be realized as syntactic subjects. Although idiosyncratic categorial properties of arguments may be specified, normally this does not seem to be necessary. The mapping of lexical structures onto syntax is subject to the Projection Principle: (14)

Projection Principle (PJP) Representations at each syntactic level (i.e., LF, D- and S-structure) are projected from the lexicon, in that they observe the subcategorization properties of lexical items. (Chomsky 1981:29)

^ The history of Generative Grammar is characterized by two major conceptual shifts: firstly, the abstraction from language to grammar (Koster 1978); secondly, the shift of emphasis from rule systems to systems of principles (Chomsky 1981). In the latter view, structural properties of linguistic expressions are exclusively determined in terms of independently motivated parametrized principles of grammar. 14

θ-role grids represent generalizations of the multi-level lexical subcategorization of the Aspects theory (Chomsky 1965). The syntactic categories of lexical arguments are predictable, at least in the unmarked case, if following Chomsky (1986a), θ-roles have Canonical Structural Representations (CSRs). For instance, it is normal for lexical arguments to be referential expressions, and it is normal for R-expressions to be NPs syntactically. - Independent reasons for the reduction of categorial and syntactic information are given in, e.g., Freidin (1983) and Czepluch (1984,1987a,b).

348

The PJP relates the syntactic levels of representation in a structure-preserving way in that it requires a Id-correspondence between lexical arguments and their syntactic exponents.15 From this it follows that D-structure is a 'pure' representation of thematic information in some sense (cf. Chomsky 1982:9).16 Clearly, the PJP does not say anything about the realization of Adjuncts. If we look for a lexical motivation for the structural Complement-Adjunct distinction, we have to go back to the Aspects theory, where lexical insertion into syntactic structure is used as a filter mechanism to eliminate unwanted sentences.

Because of the strict subcatego-

rization features in (15c and d), the verb mend cannot be inserted in the V° position of (15a), and/wrt cannot appear in the monotransitive context of (15b):17

(15)

(a) ...

V Ln

V°

'

IIP

...

(b)

...

'

PP

(c) put: +[ HPPP] ---> ...,-[ HP],... (d) lend: t[_ HP] ---> ...,-[_ HP PP], ...

*^ Thus, the verb put must be construed as a 3-place verb in syntax, as in (a): (a) John [ypput the book on the table] (b) »John [ypput the book] (c) *John [vpput °n the table] (d) *John [ypPut his friend the book on the table] (e) *John [vpput the book for three hours] Ex. (b) and (c) are deviant because of a missing argument; ex. (d) is ungrammatical because it has one argument too many; and (e) violates the thematic requirement that the PP Complement be interpretable as a Locative expression. 1

^ This does not mean that D-structure only realizes the θ-roles of lexical heads. Adverbiale will be represented too, their thematic interpretation being inherent rather than being thematically dependent on a head (cf. Larson 1985).

*' Chomsky (1965:99) states this as a rule of Complex Symbol formation in the syntactic base: A —> CS / α _ β, where αΑβ is a Γ, where, furthermore, Γ is the category symbol that appears on the left in the rule Γ —> ... A... that introduces A. By the strict locality requirement, CS will contain two features: a categorial feature of the lexical category and a strictly local context feature depending on the particular option of the rule V —> V ... chosen; cf. (15a-b), Given the entries (15c-d), lexical redundancy rules add mmws-features for any context feature not specified positively. Lexical insertion, i.e., the merging of CS features and lexical features of full entries, is blocked by conflicting values for a given feature. - Although the dummy-symbo/-alternative of lexical insertion also discussed in Chomsky (1965) differs technically from the CS version in that lexical context features are interpreted as structural indices for local insertion transformations, it yields the same results.

349

For this procedure to work, lexical subcategorization features must correspond to the strictly local domain of a head. Specifically, the procedure motivates a minimal phrasal domain comprising all and only lexical arguments of a verb to the exclusion of adverbial Adjuncts. Thus, lexical insertion in the Aspects theory yields different structures for superficially similar sentences like those in (16), depending on the properties of lexical heads: (a) John will [vp[v Put the car in the garage ] tomorrow ] (b) John will [vpfvlv mend the car ] in the garage ] tomorrow ]

(16)

Note, though, that the motivation for such a subcategorization domain does not hold in GB theory any longer: by the PJP, it is only required that the θ-arguments of a head be uniquely identified in syntactic structure. Take this to mean that the θ-indices of a lexical entry have to be 'matched' by indices of syntactic arguments, in the sense of, e.g., Stowell (1981) or Borer (1983):

(17)

1

1

1

V

HP

PP

ιΙ

ι ι

ι. ι

I

1. John* will, put the car} in the garage^ toiorrou . [M^TH^LQC*] John1 will lend the car 5 in the garage toiorro»

If so, the lexical properties of a head may be satisfied irrespective of whether Adjuncts occur as V°-sisters or not, and lexical mapping per se no longer requires a structural distinction between Complements and Adjuncts. In this respect, the sentences in (16) may well have the same 'flat' structure (17), differing only as to the thematic identification of arguments. 3.2

X-bar projection

As to the syntactic realization of phrasal structure, it is now widely assumed that it arises as a bottom-to-top projection over head categories, instead of the top-to-bottom generation by the earlier PS rules. The classical formulation of the relevant X-bar principles is given in Stowell (1981:70):18

° (18a) expresses the core idea of X-bar theory, where endocentricity is now interpreted to mean that phrasal categories are categorially derivative from head categories - i.e., the bottom-to-top view of phrase structure; (18c) has become known as the Head Parameter, which allows heads to be either initial or final in minimal phrasal domains; (18d) provides for the proper augmentation of zero-, single- and double-bar levels - e.g., precluding that maximal projection of a head is included in a single-bar projection of the same head; and (18e) requires that only maximal phrases appear in the domain of a head - which is redundant, though, given that Specifiers and Complements themselves are subject to bottom-to-top projection.

350 (18)

(a) Every phrase is endocentric. (b) Specifiers appear at the X" level; subcategorized Complements appear within X'. (c) The head always appears adjacent to one boundary of X1. (d) The head term is one bar-level lower than the immediately dominating phrasal node. (e) Only maximal projections appear as non-head terms within a phrase.

For the realization problem, only statement (18b) appears to be of immediate relevance. It precludes that Specifiers appear as V°-sisters and that Complements are realized at the Specifier level. It does not, though, require a structural differentiation between Complements and Adjuncts.19 It is not surprising that X-bar theory is quite neutral as to the realization of Complements and Adjuncts since it merely provides a structural template to be 'filled in1 according to the requirements of more specific realization principles. 3.3

Case-marking and θ-role assignment

The primary determinants of X'-syntactic realization relate to Θ- and Case theory. The principles of Θ- and Case theory interact with the PJP to regulate various aspects of the HeadArgument realizations in syntax.20

Since V is potentially recursive anyway, according to standard analyses, statement (b) as such does not even require that Complements appear as V°-sisters only, although this is probably meant by Stowell. - Our analysis of the German VP will show that Complements per se are not subject to a sisterhood constraint 20

According to the PJP (see (14)), the lexical arguments of a head need to be realized in syntax, θ-theory requires that each θ-argument be uniquely realized at LF: ^-criterion Each argument bears one and only one θ-role, and each θ-role is assigned to one and only one argument (Chomsky 1981:36) The PJP extends this LF condition to all levels of syntactic structure. Case theory deals with the distribution of NPs in the domain of a head. Since English Case-governors assign one Case only and do so adjacently (cf. Jaeggli 1982 and Chomsky 1981), the V-DO relation is singled out For English, this has the consequence that Complements are realized as V°-sisters and that the DO argument precedes other Complements, which, in turn, precede Adjuncts.

351

There is no doubt that in English thematic Complements happen to be realized as V°-sisters (except for the SU-argument21). Thus, it is generally assumed that θ-role assignment and Case-marking are subject to a head-government requirement22, as given in (19): (19)

(a) Q-role assignment: θ-role is assigned to an Α-position under head-government by a lexical head, (b) Case-marking: An NP is Case-marked if head-governed by a Case-assigner.

But again: since neither of these principles says anything about the Head-Adjunct relationship, they do not impose any restriction on the syntactic realization of Adjuncts. In particular, Case and θ-theory do not require that Adjuncts be realized outside the minimal V°domain. Thus, also from this angle, the sentences in (16) may well have the flat structure (17) rather than the standard configurational analyses. 3.4

Other principles of grammar

Besides the core principles of syntactic realization discussed so far, other modules of grammar have been used to motivate structural properties of sentences. Of these, we will just consider aspects of the Empty Category Principle (ECP) and of Binding theory.23 3.4.1 The Empty Category Principle The ECP requires the traces of movement to be governed in a particular way - called Proper Government:

21 We do not address the question of whether the external argument is directly projected into the SU position [ΝΡ,ΙΡ] - as traditionally assumed - or as an element of VP which has to be externalized - a more recent alternative in works such as, e.g., Fukui (1986), Speas (1986), Koopman and Sportiche (1988). Since the latter view is associated with modifications in the X-bar projection of lexical vs. functional categories which would unnecessarily complicate the presentation, we adhere to the more conventional point of view of SU-6-role assignment 22

Cf. Davis (1987:313): "a head-governs β iff α is X° [...] and β is a sister of a" Head-government is a special, restricted case of the core notion of government (Aoun and Sportiche 1983), according to which a head X° governs all elements in its maximal projection XP.

Hence, Case and θ-role assignment appear to be subject to the same formal properties. It is not clear, though, that the Case Filter should in fact be subsumed under the θ-Criterion, as Chomsky (1981) proposes. While Case theory determines the distribution of NP positions for arguments, θ-marking applies to nominal and nonnominal arguments alike, the latter not being subject to the Case Filter. 23

For additional discussion see the appendices.

352

(20)

An element is properly governed if it is: (a) (G-)governed by a lexical head, or (b) coindexed with the closest nonargument binder.

(= lexical government) (= antecedent-government)

For our concern the relevant aspect is that ECP effects due to lexical government have been interpreted as to whether extraction sites are V°-sisters or not. From the classical thai-trace phenomena,24 the ECP account has been extended to Superiority contrasts in multiple questions. Consider first the Subject-Object asymmetry in

(21): (21)

(a) Whoj [ip e^ said what2 ]? (b) *What2 did [IPwho! say e2 ]?

As question operators, both w/t-phrases in a sentence must have wide scope at the level of Logical Form (LF). Since only one tvA-phrase can move in the syntax, the w/t-phrase in situ must undergo Operator-movement at LF. The traces of LF-movement are not antecedentgoverned because of the syntactically moved w/t-phrase in SpecCP. Hence, they must be licensed for the ECP by lexical government. LF-movement of the Object in (21a) leaves a lexically governed trace, but not for the Subject-trace in (21b). Hence, (21b) represents an ECP violation at LF. - More interestingly, the ECP account has been taken to comply with the hierarchical Object-Adjunct distinction. Consider ex. (22): (22)

(a) why3 did he [vp[veay what2] *3]? (b) *what2 did he [vp[vsay h\ why3]? Adjuncts behave like Subjects with respect to LF-movement: the ungrammaticality of (22b) seems to indicate that LF-movement of why leaves a trace not lexically governed.

Hence,

there appears to be evidence for a structural distinction between Complement and Adjunct positions. But the configurational interpretation of such contrasts is problematic when we take nonrestrictive Adjuncts into account; cp. (23):

4 The classical phenomena for the ECP account are given in (a)-(c): (a) Who2 do you think [CP e2 (that) [jpMaryj loved e2 ]]? (b) »Whoj do you think [cp el that [!Ρ e1 loved Mary2 ]]? (c) Whoj do you think [Q> el [jp e^ loved Mary2 ]]? As in (a), extraction of an Object from an embedded clause is well-formed whether there is a complementizer or not, because the Object trace is always lexically governed by V°. Extraction from the Subject position of an embedded clause cannot be licensed by lexical government. Hence, the Subject trace must be antecedentgoverned. But this is impossible if a complementizer intervenes between the Subject trace and the trace in SpecCP - which gives the ungrammaticality of the that-trace phenomenon in (b). Apparently, if the complementizer is missing, as in (c), antecedent-government may hold. For technical accounts of the relevant blocking mechanism cf. Aoun, Hornstein and Sportiche (1981) and Lasnik and Saito (1984).

353

(23)

(a) When4 did he [vptv-say what2] r4]? (b) What2 did he [vp[vsay h] when4]? If the LF-gap of the restrictive Adjunct in (22b) violates the ECP because of lack of lexical

government, then the LF-gap of the nonrestrictive Adjunct in (23b) should be out for the same reasons. But it is well-formed.25 The ECP should apply to extraposition phenomena also. Consider the contrast in Relative Clause extraposition in (24): (24)

(a) [flpthe [N{N.man] e ]] came to see me yesterday [who had pinched my book] (b) *[NP[NpJohn] e ] came to see me yesterday [who had pinched my book]

If restrictive and appositive relatives are to be analyzed as indicated, the deviance of (24b) may be taken to be an ECP effect: as Adjunctions to a full phrase, extraposed appositive relatives leave a gap that is not governed. Hence, it cannot be properly governed and violates the ECP. The gap of the extraposed restrictive relative in (24a) is governed by the nominal head man, but it is not head- or lexically governed. Since it satisfies the ECP, a weaker version of the government requirement seems to be called for. This is corroborated by Adjunct extraposition in (25a):26 (25)

(a) [n,pa [^student] e ]] came to see me yesterday [PPwith long hair] (b) *[NPa [Nstudent e ]] came to see me yesterday [PPof physics]

But note that if the Adjunct gap in (25a) satisfies the ECP, the Complement gap in the deviant example (25b) should do so, too, since it is properly governed on any account.27 25

More suspicion about an ECP account for Superority effects comes from examples like (a)-(d). Consider the

pair (a) and (b): (a) what2 did he give e2 to which girl3? (b) *which gir!3 did he give what2 to e3? The contrast is unexpected under an ECP account: (b) should be well-formed, since the trace of wliat·^ is properly governed by V after LF-movement. Apparently, there are ECP contrasts even when there are no hierarchical distinctions that preclude lexical government. - And in (c) and (d) both Object gaps are lexically governed by tell and pinch, respectively: (c) Whot did he tell t, [CP that Bill had pinched which book2 ]? (d) * Which book2 did he tell which girl, [Cpr2 that Bill had pinched t2 ]? Nevertheless, a Superiority contrast obtains, which the ECP alone seems to be unable to account for. 2

° The examples in (25) have been taken by Radford (1988) to descriptively motivate a structural AdjunctComplement distinction, though he does not appeal to the ECP.

2

' Furthermore, the contrast in (25) does not cany over to VP structure, as the examples (a) and (b) seem to indicate: (a) John [yp[v-[y e and carries a special semantics, or

" This section relates to 1.5 above.

459 more probably something more complexly restricted, so that the syntactic difference between determiners that do and do not allow N[BAR:1] anaphora can be captured). NGPs do not have N[BAR:1] constituents, so there is no N[BAR:1] anaphora. Yet VP ellipsis is also incapable of yielding strings like */ was amazed at Stacy's being so eager, and at Morgan's too, because of the well documented condition that null VPs in English are permitted only after finite auxiliary verbs (see Gazdar, Pullum, and Sag 1982). It is not clear to me that VP ellipsis is entirely excluded from NGPs. VP ellipsis is often thoroughly unacceptable after Present Participles (because of the prohibition on unstressed nonfinite auxiliaries discussed by Zwicky and Levin (1980), for example), but to the extent that VP ellipsis is found after any Present Participle, it seems to me that it can occur in NGPs as well: (59)

(a) ?You may not remember that he was being ridiculed by everyone, but he was being . (b) ?I am shocked at your having done this, and even more so at Lee's having .

4.5.2 One anaphora is similarly impossible in the NGP, again because the anaphor in question is an N[BAR:1] anaphor. It may be that the word one simply has N[BAR:1] as one of the categories it belongs to lexically, and the syntax of one anaphora has nothing more to say, or it may be that a special rule N[BAR:1] -> e is needed; but simply by making the familiar assumption that one substitutes for N[BAR:1] but not for other categories (see the extended illustration built on this assumption by Hornstein and Lightfbot 1981), we predict that a VP head of an NP will not antecede anaphoric one.

5.

Comparison with Abney's analysis

There is one previous analysis of the NGP that has factual coverage and accuracy tantamount to that of the analysis defended here, and that is Abney (1987). As far as I have been able to determine, the factual consequences of Abney's description are correct wherever they can be tested, to the same degree as my analysis. All the properties summarized in (22) either follow or can readily be made to follow under Abney's analysis. It would be interesting to find factual evidence that distinguished between these two analyses, but I have not been able to do this. The only distinction between the two analyses that I am aware of was pointed out to me by Lauri Karttunen: Abney needs to state, either in a rule or in the lexical entry for a special null determiner, that the complement of the type of determiner that occurs in NGPs is a VP, whereas in the analysis presented above no such stipulation is necessary, the VP category of the constituent in question being predictable from a single morphosyntactic feature specification, [VFORM:prp], given in the rule.

460 Apart from this minor observation about partial predictability of feature composition, only the principle of Null Licensing militates against Abney's analysis (which posits numerous invisible heads). Making the choice in a fully motivated way will call for additional research. Even a choice between frameworks, made on other grounds will not settle the issue unless the other grounds are truly overwhelming, because broadly speaking it is quite easy to cast Abney's analysis in the generalized phrase structure grammar framework used here. It may also be possible to devise a government-binding analysis of NGPs that has the characteristics of my description. It is harder to be sure of this because of the vagueness of governmentbinding attitudes toward features and the notion 'head1, but if there is a clear principle that precludes VP heads of NPs under government-binding assumptions, I am not aware of it.

6.

Conclusion

I have shown that it is possible to analyze the NGP in English in a way that (i) maintains the principle that no syntactic rule affects word- internal structure (Strong Lexicalism), (ii) preserves the generalization that every phrase has a head (Endocentricity), and (iii) respects the normal criterion for the postulation of empty categories, namely that they should be either syntactically or semantically motivated (Null Licensing). These three attributes are not simultaneously present in any other syntactic analysis of the English NGP that I have seen. To the extent that the three desiderata are indeed desirable in a theory of grammar, the analysis suggested here represents an advance. My analysis shows some support for a hitherto almost unnoticed claim of the GKPS theory of syntax, namely that there can be single-headed constructions with heterocategorial heads. The analysis of coordinate structures adopted in GKPS and further elaborated in Sag et al. (1985) shows clearly how, if conjuncts are treated as multiple heads that can differ in feature structure in ways controlled by the HFC, interesting data can be explained (for example, the contrast between John is a Republican and proud of it and the ungrammatical *John met a Republican and proud of it). We have seen in this paper that there are also non-coordinate constructions analyzable as having a single head of a different major category from its mother. Constituents that act as heads for the purposes of the passing of feature specifications by the HFC normally receive their values for the features N or V via this principle, but they do not have to. Although it is not the default situation, there can be NPs with VP heads. It should be noted in passing that it is not at all clear how my analysis could be reconstructed under any theory in which the Head Feature Convention or its analog is an absolute principle rather than a default. One example of a theory with an explicitly absolute HFC is the head-driven phrase structure grammar outlined by Pollard and Sag (1987), but there are probably many others. Most varieties of generative grammar seem to assume an

461

absolute HFC to the extent that they make their assumptions clear on this score. GKPS is apparently the only detailed presentation of an HFC with default status that has been put forward. It might seem that any analysis that involves forcing feature disagreement between mothers and heads must involve a disappointing weakening of the claims associated with the notion of head in syntax. I do not think this is so, and the main part of my reason is to be found in section 3.1, where it is shown that at least three features have their values predicted by the HFC even though the values for N and V on the head are coerced into disagreement with the mother. But there is another relevant observation to be made. If the analysis given here is on the right track, we would expect to occasionally find other examples of constituents with heterocategorial heads, identifiable by virtue of having all the external syntactic characteristics of one category but internal syntactic characteristics suggestive of the structure of some other category. However, contrary to what some have suggested concerning clines of properties from category to category (cf. e.g. Ross 1972, 1973), we would not expect to find arbitrary mixtures of syntactic characteristics from different categories in any constituent type. Instead, the phrasal head of some types of phrase may be sharply and consistently of different type from the type we would expect from the usual effects of the Head Feature Convention. In those cases where the grammar enforces a special value for N or V on a head, we would get not an odd blend of syntactic properties but rather a head with sharply and consistently different behavior from what would be expected. Two other constructions in English might be worth investigating in an effort to further explore the idea just advanced. One is the partitive construction, where in a structure of the form [xpsome^of the bricks]] the constituent labelled

looks like a PP[o/J in the position

where we would expect the head of NP, suggesting a rule like (60). (60)

N[BAR:2] -> Art H[PFORM:0/)

The relevant data are very complex, and previous studies such as Jackendoff (1977) and Selkirk (1977) have generally tried to finesse the problem of the preposition of by assuming (without a lot of conviction or explicitness) that it is inserted by a transformation in some way that does not create a PP node. It might be interesting to attempt a reanalysis of the partitive and related constructions along these lines. The other construction that might repay a close look is the prenominal gerundive modifier seen in some [γ quietly sleeping] children, where the constituent labelled γ, in attributive adjective phrase position, could be an adjective phrase with a VP head. It is suggestive that addition of post-head modifiers in the phrase γ makes a structure just as ungrammatical as adding a complement to an attributive adjective; thus *some [γ sleeping all day] children is

462

exactly as bad as *some [γ tired all day] children. Perhaps a description based on a rule like (61) might be appropriate: (61) A[BAR:2] -> H[VFORM:prp] This would yield structures containing an AP with a gerund VP head (a type of construction which, as Alec Marantz has pointed out to me, is also found in Russian). I leave the investigation of this construction, and of the partitive, for future research.

Bibliography Abney, S.P. (1987): The English Noun Phrase in Its Sentential Aspect. - unpublished MIT Diss, Cambridge, Mass. Akmajian, A. & T. Wasow (1975): "The Constituent Structure of VP and AUX and the Position of the Verb BE." - In: Linguistic Analysis 1, 205-46. Baker, M. (1985): "Syntactic Affixation and English Gerunds." - In: J. Goldberg, S. MacKaye & M.T. Wescoat (eds.): Proceedings of the West Coast Conference on Formal Linguistics. Vol. 4 (Stanford, Cal.) 1-11. Borsley, R. D. (1983a): "A Note on the Generalized Left Branch Condition." - In: Linguistic Inquiry 14,169-74. Borsley, R.D. (1983b): "A Welsh Agreement Process and the Status of VP and S." - In: G. Gazdar, E. Klein, & G.K. Pullum (eds.): Order, Concord, and Constituency (Dordrecht) 57-74. Borsley, R.D. (1984): "On the Nonexistence of VP's." - In: W. de Geest & Υ. Putseys (eds): Sentential Complementation (Dordrecht) 55-65. Chomsky, N. (1957): Syntactic Structures. - The Hague. Chomsky, N. & M. Halle (1968): The Sound Pattern of English. - New York. Emonds, J.E. (1976): A Transformational Approach to English Syntax. - New York. Emonds, J.E. (1986): A Unified Theory of Syntactic Categories. - Dordrecht. Farkas, D. (1986): "On the Syntactic Position of Focus in Hungarian." - In: Natural Language & Linguistic Theory 4,77-96. Gazdar, G., E. Klein & G.K. Pullum (eds.) (1983): Order, Concord, and Constituency. - Dordrecht Gazdar, G., E. Klein, G.K. Pullum & I.A. Sag (1985): Generalized Phrase Structure Grammar. - Cambridge, Mass. Gazdar, G., G.K. Pullum, R. Carpenter, E. Klein, T.E. Hukari & R.D. Levine (1988): "Category Structures." - In: Computational Linguistics 14,1-19. Gazdar, G., G.K. Pullum & I.A. Sag (1982): "Auxiliaries and Related Phenomena in a Restrictive Theory of Grammar." - In: Language 58,591-638. Harlow, S.J. (1989): "The Syntax of Welsh Soft Mutation." - In: Natural Language & Linguistic Theory 7,289316.

463 Horn, G. (1975): "On the Nonsentential Nature of the Poss-Ing Construction." - In: Linguistic Analysis 1, 33387. Hornstein, N. & D. Lightfoot (1981): "Introduction." - In: Hornstein, N. & D. Lightfoot (eds.): Explanation in Linguistics: The Logical Problem of Language Acquisition (London) 9-31. Jackendoff, R.S. (1977): XSyntax. · Cambridge, Mass. Komai, A. & O.K. Pull urn (1990): "The X-bar Theory of Phrase Structure." - In: Language 66,24-50. Koster, J. (1978): "Why Subject Sentences Don't Exist." - In: S.J. Keyser (ed.): Recent Transformational Studies in European Languages (Cambridge, Mass.) (=Linguistic Inquiry Monograph 3) 53-64. Kuno, S. (1973): "Constraints on Internal Clauses and Sentential Subjects." - In: Linguistic Inquiry 4,363-85. Lapointe, S.G. (1991): "On Deriving the Government Constraint for Incorporation and Inflection." - To appear in: E. Schiller & E. Steinberg (eds.): Developments in Autolexical Syntax (Berlin). Lees, R.B. (1960): The Grammar of English Nominalizations. - The Hague. Marantz, A (1978): "Embedded Sentences Are Not Noun Phrases." - In: M.J. Stein (ed.): Proceedings of the Eighth Annual Meeting of the North Eastern Linguistic Society (Amherst, Mass.) 112-22. Milsark, G.L. (1988): "Singl-ing." - In: Linguistic Inquiry 19,611-34. Pollard, C. & I.A. Sag (1987): Information-Based Syntax and Semantics. Vol. 1: Foundations. - Stanford, Cal. Postal, P.M. & G.K. Pullum (1982): "The Contraction Debate." - In: Linguistic Inquiry 13,122-38. Pullum, G.K. (1975): "People Deletion in English." - In: Working Papers in Linguistics 14,95-101. Pullum, G.K. (1979): Rule Interaction and the Organization of a Grammar. - New York. Pullum, G.K. (1985): "Assuming Some Version of X-bar Theory." - In: Chicago Linguistic Society 21,323-53. Pullum, G.K.& D. Wilson (1977): "Autonomous Syntax and the Analysis of Auxiliaries." - In: Language 53, 741-88. Pullum, G.K.& A.M. Zwicky (1988): "The Syntax-Phonology Interface." - In: F.J. Newmeyer (ed.): Linguistics: The Cambridge Survey. Vol. 1: Linguistic Theory: Foundations (Cambridge) 255-80. Pullum, G.K. (1991): "English Nominal Gerund Phrases as Noun Phrases with Verb-Phrase Heads." - In: Linguistics 29,763-99. Pullum, G.K. & A.M. Zwicky (1991): Condition Duplication, Paradigm Homonymy, and Transconstructional Constraints. - Paper presented at the 17th Annual Meeting of the Berkeley Linguistics Society. To appear in the proceedings. Rosenbaum, P. S. (1967): The Grammar of English Predicate Complement Constructions. - Cambridge, Mass. (=Research Monograph No. 47). Ross, J.R. (1967): Constraints on Variables in Syntax. - Unpubl. MIT Diss., Cambridge, Mass. (Published 1968 by the Indiana University Linguistics Club and 1986 as Infinite Syntax! by Ablex, Norwood, New Jersey.)

464 Ross, J.R. (1972): "The Category Squish: Endstation Hauptwort" - In: P.M. Peranteau, J.N. Levi & G.G. Phares (eds.): Papers from the Eighth Regional Meeting (Chicago) 316-28. Ross, J.R. (1973): "Nouniness." - In: O. Fujimura (ed.): Three Dimensions of Linguistic Theory (Tokyo) 137258. Sag, I.A., G. Gazdar, T. Wasow & S. Weisler (1985): "Coordination and How to Distinguish Categories." - In: Natural Language and Linguistic Theory 3,117-171. Scalise, S. (1984): Generative Morphology. - Dordrecht. Schachter, P. (1976): "A Nontransformational Analysis of Gerundive Nominals in English." - In: Linguistic Inquiry 7,205-241. Selkirk, E.O. (1977): "Some Remarks on Noun Phrase Structure." - In: P.W. Culicover, T. Wasow & and A. Akmajian (eds.): Formal Syntax (New York) 285-316. Stowell, T. (1981): Origins of Phrase Structure. - Unpubl. MIT Diss., Cambridge, Mass. Stucky, S. (1983): "Verb Phrase Constituency and Linear Order in Makua." · In: G. Gazdar, E. Klein, & G.K. Pullum (eds.): Order, Concord, and Constituency (Dordrecht) 75-94. Zwicky, A.M. & N.S. Levin (1980): "You Dont Have to." - In: Linguistic Inquiry 11,631-36.

465

Marga Reis

The Category of Invariant alles in Wh-Clauses: On Syntactic Quantifiers vs. Quantifying Particles in German1 1.

Introduction

German wh-clauses often contain an invariant expression alles, which has two salient properties: First, alles is related to the wh-phrase (its antecedent) in a quantificational way, cf. (1): (1) Wen (alles) hat Hans besucht? If alles is added, some kind of universal quantification takes place: the question is taken to imply that all the people visited by Hans must be identified in the answer, and moreover, that Hans visited a plurality of people. Without alles, wh-questions are not so understood: there is at most an existential implicature, and the answer(s) given must be true, but not necessarily exhaustive. Second, alles may appear in variable positions: either immediately adjacent to its antecedent or in various nonadjacent positions to the right, the quantificational effects always being the same: (2)

(a) Wen (alles) hat er (alles) gestern (alles) besucht? (b) Was (alles) für Leute (alles) hätte er denn (alles) dort (alles) treffen können? Both these properties are shared by the universal quantifier alle(s) which belongs to the class of syntactic DP quantifiers that participate in socalled quantifier floating (QF),2 cf.(3): I am indebted to the members of my Tübingen research group on wh-intenogatives as well as various audiences in Rendsburg, Stuttgart and Lund for valuable discussion of earlier versions of this study. Special thanks are due to H. Altmann, V. Molnär, J. Pafel, and I. Rosengren, whose comments and criticisms have clarified (and sometimes changed) my views in important respects. Thanks also to D. Le Claire and F.-J. d'Avis for much-needed technical assistance in preparing the final version of this paper. Regarding the nature of QF, I shall proceed from the by now standard position that nonadjacent positions of quantifier and antecedent are not a result of rightward 'quantifier floating1, but - if a movement analysis is adopted at all - of leftward movement of the antecedent DP.

For a recent account along these lines cf.

Sportiche (1988), for its extension to German cf. Giusti (1990, 1991a), and, making use of the revision of Sportiche's account by Shlonsky (1991), Giusti (1991b). While this account may be criticized in various respects (see Bhatt (1990:211-213), Drubig (1991:52f.), see also the non-movement analysis of Pdrennec (1988)), these criticisms are immaterial for the following discussion. - Regarding the scope of QF, what matters here are just the totality quantifiers' with definite antecedents (see below 2.2, 2.7), to which alleys) belongs. There is no need to take a stand on the status of split constructions involving bare antecedents and quantifiers like viele, einige (cf. Fanselow (1988)), which have sometimes also been treated under the heading of QF, but may be instances of a different construction.

466

(3)

(a) (Allen) seinen Freunden (allen) hat er (allen) gestern (allen) Bescheid gesagt, (b) (Alles) das (alles) hat er (alles) gestern (alles) für uns (alles) mitgebracht.

Hence, it seems natural to treat the cases in (1-2) also as instances of this quantifier, instances which lack agreement with their antecedent, but are otherwise unremarkable. This is exactly what we find in the literature on QF in German (Link (1974), Vater (1980), Reis/Vater (1980), Fehlisch (1986), Kniffka (1986), Müller (1986), Perennec (1988), Giusti (1990, 1991a)), even in the one case (Giusti (1991a)) where the syntactic properties of invariant alles are dealt with at some length. In other words, it is commonly presupposed that alle(s) as in (1-2) and (3) are instances of the same lexical item: the floating quantifier alle(s), and that their different agreement behavior is a mere morphological quirk, on a par with the case of copular sentences with a predicate nominal, where floating invariant forms of alle(s) also occur: (4)

(a) Hans, Fritz, Otto, Uwe sind schließlich alles Schüler von ihm. (b) Diese Typen da drüben sind alles Linguisten.

In this paper, I should like to argue that this view is mistaken. To this effect, I shall first present evidence (section 2) that invariant alles in wh-clauses (henceforth l-alles) and QF-cases involving alle(s) (henceforth QF-alle(s)) differ in so many syntactic and semantic respects that, in spite of the parallels mentioned at the outset, they cannot be instances of the same phenomenon. Hence, the assumption of categorial identity between l-alles and QF-alle(s) must be given up. This gives rise to the question, what the categorial status of l-alles could be, keeping in mind that not only the differences but also the parallels to Q¥-alle(s) have to be accounted for. The answer that I shall try to make precise and give substance to (section 3), is that l-alles belongs to a hitherto unrecognized class of quantifying particles with rather special properties. In the concluding section, I shall draw attention to the many questions this first attempt at descriptive adequacy concerning l-alles has to leave open.

2.

Differences between QF-alle(s) and l-alles

2.1 Agreement? Whereas QF-alle(s) typically agrees with its antecedent phrase, l-alles is always invariant. However, as should be pointed out from the start, this difference is not decisive, for invariant alles in copular sentences such as (4) will turn out to be bona fide instances of QF-alle(s) (see below 2.4). Still, being uninflected is a necessary property of l-alles (as for right-adjacent inflected cases of all- that seem to share some of the properties of l-alles, cf. Link (1974:106, 124 note 7) and below, 3.4, note 24).

467

2.2

Range of participating elements

Besides all-, the following quantifiers participate in QF in German: beid-, sämtlich- (for those speakers that treat sämtlich- as an adjectival quantifier, and not just as an adjective), jed-, QF of the latter being more or less restricted to the invariant form jedes (see Kniffka (1986:65), Fehlisch (1986:89ff.)). (5) Die Kinder haben alle/beide/sämtliche/jedes schon ein Eis bekommen. But none of these may occur in wh-clauses replacing l-alles: (6) Wer hat alles/*beides/*sämtliches/*jedes schon ein Eis bekommen? If l-alles is an instance of QF-alle(s), then, obviously, the reasons for its uniqueness cannot be syntactic. However, semantics does not offer an easy way out either: Since sämflichvirtually has the same meaning as all- (in terms of Reis/Vater (1980), it is not only [+distributive], but also [+total]), at least this quantifier should be semantically admissible, if QF were involved. Hence, it appears as if this difference between l-alles and QF-alle(s) could not be reduced to independent factors. 23

Focussing properties

Whereas QF-alle(s) may bear main stress, which means it may be minimally focussed, this is impossible for l-alles, cf. (7) vs. (8): (7)

(a) Das hat er alles für dich getan. - Die Leute sind alle interessiert. (b) Das hat er ALLes für dich getan. - Die Leute sind ALLe interessiert. (8) (a) Was hat er alles für dich getan? - Wer ist alles interessiert? (b) *Was hat er ALLes für dich getan? - *Wer ist ALLes interessiert? Since the possible focus-background structures of wh-clauses and declarative clauses are normally parallel (see Rosengren (1991)), there is again no way of systematically reducing this difference to the difference in clause type and/or antecedent phrase involved. 2.4

Modifying elements

While QF-alle(s) may combine with modifying elements that are typical for quantifiers: fast (alle), so gut wie (alle), (alle) bis auf einen, l-alles may not, cf. (9) vs. (10). On the other hand, l-alles may cooccur with 'nonexhaustive1 so (an unstressed particle that indicates nonexhaustive listings/unexact descriptions in declaratives, and licenses nonexhaustive answers in wh-interrogatives), which is impossible with QF-alle(s), especially if it is focussed. (9) (10)

Diese Kollegen möchte er fast alle/ *so ALLe kennenlernen. Wen möchte er *fast alles/ so alles kennenlernen.

468 With this additional piece of evidence we have already sufficient criteria to identify invariant alles in copular sentences as instances of QF-alle(s). As (41) shows, it may be stressed/ focussed, it is interchangeable at least with invariant beides, and it may combine with modifiers such as fast, but not with nonexhaustive so. (41) Diese Typen da drüben sind (*so) ALLes/beides/ fast alles Linguisten. As can be easily verified, these findings are confirmed by the differences to be discussed in 2.5ff.: wherever structurally possible, invariant alles in copular sentences invariably sides with QF-alle(s) rather than with l-alles. Hence, it is certainly no accident that inflected alle/beide may be used interchangeably with alles/beides in this context, cf.(4"): in copular sentences, the uninflected forms are really no more than variants of QF-alle(s) (for additional observations on the use of uninflected alles in copular sentences, cf. Giusti (1991a:337ff.). (4") 2.5

Diese Typen da drüben sind (*so) ALLe/beide/fast alle Linguisten. Syntactic restrictions on admissible antecedents

As has already been observed by Link (1974:107f.) QF-alle(s) is often incompatible with PPs,3 and always incompatible with genitive DPs, as distant antecedents: (12)

(a) ?*Mit diesen Leuten hat er allen schreckliche Probleme gehabt, (b) "Von diesen Studenten wurde er allen nachgeahmt. (13) *Dieser Kollegen/*Unser hat er sich aller angenommen. However, the corresponding wh-cases with l-alles are always good: (14)

(a) Mit wem hat er alles schreckliche Probleme gehabt? (b) Von wem wurde er alles nachgeahmt? (15) Wessen/Welcher Kollegen hat er sich alles angenommen? This shows that the ways in which QF-alle(s) vs. l-alles are structurally related to their antecedent phrase must be significantly different, otherwise the interaction with the same grammatical principles could not produce such different effects. The only attempt I know of to come to grips with these contrasts, at least regarding the PP cases, is by Giusti (1991a:344f.). Following Sportiche (1988) in postulating adjacency for QF-alle(s) and its antecedent DP in deep structure, she tries to explain the difference between (12) and (14) by positing different base positions for alle(s) according to whether the Acceptability judgements seem to vary with the preposition involved and its respective properties (the case it assigns, its being selected or not), albeit in ways that are as yet unclear (cf. also Giusti (1991a:332f.), Pdrennec (1988)). There can be no doubt, however, that there are signiflcant structural restrictions turning on the PP status every theory of QF has to account for.

469

'antecedent' DP is pronominal (Q right-adjacent) or not (Q left-adjacent), cf. (16) vs. (17); Ialles is of course subsumed under QF-alle(s).

DP

(16)

DP

(17)

Q

DP

DP

alle

diese Studenten

wir wer

alle alles

(16·) DP

diese Studenten t Extending Sportiche's account of distant QF constructions, which result from leftward movement of the DP, cf. (16'), to QF constructions involving PPs, the difference between (12) and (14) is then accounted for by appealing to the Left Branch Condition. This condition would always be violated by the necessary intermediate PP structure (18') to be derived from (18) in the case of (12), but would not be violated in the case of (14), where Q starts out on the right branch, cf. (17).

(18')

(18) PP DP DP

von

allen

diesen Studenten

von

e *

diesen Studenten —χ

allen *

However, even disregarding that structures like (18') are not always unacceptable (cf. Link (1974:107)), this account clearly does not work. First of all, it cannot be extended to the genitival cases, cf. (13) vs. (15), where intermediate structures with rightward moving Q are as

470

unnecessary as with accusative and dative antecedents. Second, it is on the wrong track even in the PP cases, cf.(12'), (141): (121)

(a) ?*Mit uns hat er allen schreckliche Probleme gehabt. (b) *Von ihnen wurde er allen nachgeahmt. (a) Mit was für Leuten hat er alles schreckliche Probleme gehabt?

(14')

(b) Von welchen Leuten wurde er alles nachgeahmt?

As (12') and (141) show, constructions with QF-alle(s) are still ruled out, even if the PPantecedent is pronominal, and constructions with l-alles are still ruled in, even if the antecedent wh-PP is nonpronominal.4 Hence, the contrasts between (12-13) vs. (14-15) cannot be explained as long as they are uniformly held to be instances of the same structural configurations. Rather, the reasons for the difference must be located in the different nature of QFalle(s) vs. l-alles itself.

2.6 Syntactic-semantic restrictions on admissible antecedents As is well-known, QF-alle(s) presupposes definite antecedent phrases, that is phrases that denote a set with its membership fixed in advance, cf. (19a) vs. (19b,c). This restriction also shows up with respect to relative clauses, where (contra Sportiche (1988:438)) QF-alle(s) fares well only with appositive relative antecedents, cf. (20a) vs. (20b), and forces the appositive interpretation for relative constructions that would otherwise be ambiguous, cf. (21). (19)

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

Die/Diese/Unsere/Pauls Studenten/Sie hatten alle eine Vorliebe für Syntax. *Studenten hatten alle eine Vorliebe für Syntax. *Keine/*Manche/*Viele Studenten hatten alle eine Vorliebe für Syntax. (20) Diejenigen Studenten, die den Test ?*alle nicht bestehen, müssen ihn wiederholen, (b) Diese vier Studenten, die den Test alle nicht bestanden haben,... (21) Die Fakultäten, die Vogt alle überprüfen will, richten sich auf sein Kommen ein. As for l-alles, we have so far concentrated on occurrences with interrogative wh-phrase antecedents. However, although this is by far the most important class of possible antecedents, there are others: exclamative wh- and d-phrases, with l-alles typically in distant position, cf. (22-23), restrictive relative phrases (although much more marked), cf. (24), and, marginally, also wh-phrases appearing in 'unconditional constructions' (see Zaefferer (1991)), cf. (25).

This problem is recognized by Giusti, who tries to account for the different behavior of l-alles with respect to PPs by attributing it to the wh-fearure of its antecedent (1991a:345). It is doubtful that this is more than a restatement of the facts, but if it is, it is refuted by PP-versions of examples such as (28-28') to be discussed below (2.7), in which QF-alle(s) has a wh-antecedent, too, cf. (i-i1): (i) *Von wessen Häusern habt ihr allen Aufnahmen gemacht? (i1) Von wessen Häusern habt ihr alles Aufnahmen gemacht?

471

(22) (23) (24) (25)

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

Wen der nicht alles kennt. Wo die schon alles gewesen ist. Was für Leute der schon alles in Rezensionen verrissen hat. Die Leute, die der alles kennt. Die Orte, wo/an denen der schon alles gewesen ist. ?Diejenigen Studenten, die alles den Test nicht bestehen, müssen ihn wiederholen, ?Such die Ingenieure raus, die alles Cahuilla sprechen. ?Was für Leute/Wen auch immer du alles zum Essen mitbringst, ich bin gerüstet.

Note that all these antecedents belong to a highly specific syntactic class: they are operator phrases that may appear in SpecC position (i.e. the initial position of finite verb final clauses). That this is a necessary condition for l-alles antecedents is borne out by the fact that l-alles is incompatible with XP antecedents that are excluded from SpecC position, cf. (26), no matter whether they are definite or indefinite, cf. (19'): (26) (19')

*Hans weiß, die/diese/unsere/Pauls/keine/manche/viele Studenten stets eine Vorliebe für Syntax hatten. (a) *Die/*Diese/*Unsere/*Pauls Studenten/*Sie hatten alles eine Vorliebe für Syntax. (b) *Studenten hatten alles eine Vorliebe für Syntax. (c) *Keine/*Manche/*Viele Studenten hatten alles eine Vorliebe für Syntax.

This leaves as an interesting question what happens in the case of definite SpecC operator phrases. Applying Hawkins1 notion of 'definiteness', by which a definite expression denotes a set that is identical to an 'antecedent set' independently given via discourse or situation (see Hawkins (1978, 1991), cf. also

(1991:9)), the only viable candidates are appositive rela-

tive pronouns/phrases, which, unlike all other XPs appearing in SpecC position, "semantically function as anaphoric pronouns /[phrases]" (Zimmermann (1991:264)). We have already seen in (20-21) that they are compatible with QF-alle(s). In contrast, l-alles is next to incompatible with them, cf. (20'b), in any case much better with restrictive relative antecedents as in

(24). (20')

(b) ?*Diese Studenten, die alles den Test nicht bestanden haben, ...

This suggests the generalization that only indefinite SpecC operator phrases occur as antecedents of l-alles. In other words, l-alles antecedents do not correspond to (an expression containing) an anaphoric pronoun but to (an expression containing) a variable. There is another type of SpecC phrases that has been called 'definite1 in the literature: exclamative wh-phrases as in (22) and (27), which are definite in the sense that they refer to a uniquely identifiable set, the identity of which is known to the speaker (see Rosengren (1992:283).5 5

Given the interchangeability of exclamative wh- and relative constructions, cf. welche Ärzte der kennt - die Ärzte, die der kennt, it will be obvious that this notion of definiteness is more like Russell's, underlying his term 'definite descriptions'. As for the relation between this and Hawkins' notion of 'definiteness', see Vater (1984:32ff.) and Hawkins (1991).

472

(27)

(a) Wenderkennt. (b) Von welchen Künstlern der Autogramme besitzt. (c) Wie ruhig es hier ist.

They are, however, not definite in Hawkins' sense specified above, in that there is no anaphoric or deictic/situational link to an independently established antecedent set. Since exclamative wh-phrases do allow for l-alles, cf. (22), it is obviously Hawkins1 notion of definiteness that is relevant here, since it yields the generalization just cited.6 The respective restrictions on antecedents of QF-alle(s) vs. l-alles may accordingly be formulated as follows: (28)

(a) QF-alle(s) requires that its antecedent phrase be definite, i.e. denote 'closed sets', (b) l-alles requires that its antecedent phrase be indefinite, i.e. denote Open sets', that is: l-alles is related to the variable expression that interrogative, restrictive relative, etc. operator phrases denote.

Obviously, (28) marks a decisive difference between QF-alle(s) and l-alles, which it is hard to imagine how to explain away. 2.7

Contrasts in antecedent relations

While the evidence presented in 2.1-2.6 shows that l-alles markedly differs from QF-alle(s), there is still a credibility gap: QF-alle(s) and l-alles have so far only been observed in complementary distribution. This gap is closed by examining the rare cases in which QF-alle(s) and l-alles occur in the same environment, the best one being provided by wh-clauses with possessive wh-phrase antecedents, cf. (29-29'): (29) (29')

Wessen Bücher wurden alle von Reich-Ranicki schlecht rezensiert? Wessen Bücher wurden alles von Reich-Ranicki schlecht rezensiert?

There is a clear contrast in the way QF-alle(s) and l-alles relate to the antecedent they apparently have in common: while QF-alle(s) is related to the antecedent phrase as a whole, Ialles is related to the possessive wh-specifier only. In other words, (29) and (29') do not ask the same thing: in (29) the question is about authors whose entire book production has been negatively reviewed by Reich-Ranicki (and there might be no more than one unlucky author 6 A further case calling for comment is provided by Pafel (1991:166-169), according to whom interrogative wer/was- and welch- phrases, all of which occur freely with l-alles, are definite, because they are said to behave exactly like noninterrogative definite DPs regarding word order and relative scope interaction. However, no matter which concept of definiteness is adopted, they do not constitute viable counterexamples. Regarding wer-lwas- phrases, Pafel's data seem too weak to support the conclusion (the word order contrasts (ibid.:167) are at best minimal, and could also be accounted for if wer/was were assumed to be indefinite interrogatives; as for scope properties, wer-lwas- phrases seem to be different from welch- phrases, and closer to indefinite NPs at that). Regarding welch- phrases, the data seem correct, but what is definite about them is arguably irrelevant for l-alles, cf. the discussion in 2.8 below.

473

like this), whereas (29') is about all the authors (and there is supposed to be a plurality of them) of books (one or more) that have been negatively reviewed by Reich-Ranicki, no matter whether other books by some such author exist and/or have been negatively or positively reviewed by Reich-Ranicki. These facts can be brought into line with (28) as follows: possessive phrases such as in (29-29% although they contain a possessor phrase which, according to (28b) above is indefinite, are, as a whole, still definite enough to allow QF-alle(s). To see that this is not ad hoc, cf. the parallel non-interrogative indefinite cases in (30): (30)

(a) Eines Kindes Träume können gar nicht alle in Erfüllung gehen, (b) Sogar eines Enkels zahlreiche Freundinnen mußte sie alle einladen.

Obviously, what makes them definite in the required sense is not the identity of the possessor phrase, but the possessive construction as a whole.7 Hence, a possessive wh-phrase is one of the rare cases, where one and the same syntactic phrase provides simultaneously a definite DP (the whole phrase) and a separately identifiable indefinite wh-DP (the possessor phrase), thus allowing (28a) as well as (28b) to be satisfied, albeit in different ways. l-alles and QF-alle(s) exhibit the same difference in antecedent relations in the dialectal possessive dative construction, which may also appear in colloquial standard, cf. (31-31'): (31) (3 )

Wem seine Bücher hat der Kerl alles verrissen? Wem seine Bücher hat der Kerl alle verrissen?

This is much less obvious than has been commonly assumed in the literature (e.g. by Olsen (1989) and Bhatt (1990)). That possessive DP constructions are not invariably definite, has been argued by Zimmermann (1991:35ff.) on comparative grounds. But it is also true for German, where there are not only indefinite possessive phrases with the possessor phrase in postnominal position (Bücher meines Sohnes, ein Geschenk Pauls, etc.), but also indefinite possessive phrases with a prenominal possessor phrase, cf. the (in)definiteness effect in diagnostic right dislocation and es-structures illustrated in (i-ii) (see also Grimshaw (1990:55) for similar effects in English): (i) Wir haben sie belächelt, *?eines Kindes Träume/??die Träume eines Kindes/des Kindes Träume/die Träume des Kindes, (ii) Es wurden Spuren eines Kindes/eines Kindes Spuren/?die Spuren eines Kindes/??des Kindes Spuren gefunden. This raises, of course, the question why (30) above is acceptable at all. A natural guess would be that 'specificity* is involved (in the sense of (1991), in which indefinite noun phrases linked to a previously established set are 'specific*). However, considering the nonspecific nature of the respective DP in (30a), which is at least as good as the specific case (30b), and the fact that (i) is bad, although the indefinite DP is clearly specific, an explanation along these lines seems implausible. At present, there seems no way around concluding (a) that the (in)definiteness effects may be different in different domains, cf. (30) vs. (i-ii), (b) that a certain definiteness value must be attributed to the possessive construction as such, no matter how it is induced. While this leaves many questions unanswered, it does not impair the argument 2.7 is about.

474 Since their indefinite non-interrogative counterparts likewise admit QF-alle(s), these effects can be made compatible with (28) in the same way as in the case of the standard possessive construction.8 These data constitute an insoluble dilemma for a uniform QF-analysis including l-alles: in bona fide cases of QF the distant quantifier always has a full, independent DP as antecedent, never a partial DP, which means it is always c-commanded by its antecedent-DP (see Sportiche (1988:432)), and this remains true even in cases with invariant Q-forms, cf. (32): (32)

(also was Meier, Müller, Schubert, Schlägle, und Krotz angeht,) (a) [[deren]; Kinder^ haben jedes .^j ein Eis bekommen. (b) [[derenji Kinder]] sind alles.j/j Intelligenzbestien. Hence, any syntactic-semantic theory of QF that does justice to this central fact about bona fide cases, must rule out the antecedent relations entertained by l-alles in (29'-3 ), or vice

versa. This shows beyond doubt, first, that QF-alle(s) and l-alles must be elements of different categories, and, second, since l-alles does not relate to its antecedent operator phrase as a surface syntactic category (cf. the irrelevance of pied-piped syntactic material in (29131')), that the categorial difference must be considerable. In other words: l-alles cannot just belong to a different subclass of the syntactic DP quantifiers. 2.8

Contrasts in meaning effects

This is not to deny that QF-alle(s) and l-alles are semantically similar in important respects: both have universal quantificational force and are furthermore distributive, both require accordingly that their antecedent DPs allow for a referentially plural interpretation.9 Hence, sentences like (33-331) are good, whereas sentences containing referentially singular antecedents (34-341) are out, likewise sentences containing referentially plural antecedents, if the group reading is required, as it is in (35-35"), where a paper jointly written by four authors (Brandt/Reis/Rosengren/ Zimmermann 1992) is referred to. It is, moreover, obvious, how this characterization can be extended to cover wh-phrase antecedents in the case of l-alles, which

° Note that the overt possessive pronoun does not by itself ensure or signal dcfiniteness. If the possessive genitival phrases in (i) (see previous note) are substituted by the corresponding dative, the results are just as bad. The same is true for the exceptional possessive construction involving preposed vo/i-phrases (von Hans die Bücher, von einem Freund die Bücher), the interrogative counterparts of which, however, are too marginal to be used in this context of discussion. - As for welch- phrases as antecedents for l-alles vs. QF-alle(s) see the discussion in 2.8 below. 9 Regarding the differences in distributivity between all- und jed-, cf. the observations in Reis/Vater (1980:386ff.)

475

are frequently unspecified for number (potentially plural), and mass noun antecedents in the case ofQF-alle(s) (uncountable, hence not singular either),10cf. (36): (33) (33') (34) (34') (35)

Diese Bücher hat er alle gelesen. Welche Bücher hat er alles gelesen? *Dieses Buch hat er alles gelesen. "Welches Buch hat er alles gelesen? *Diese Linguistinnen haben alle "Satztyp, Satzmodus und Illokution" geschrieben. (351) * Welche Linguistinnen haben alles "Satztyp, Satzmodus und Illokution" geschrieben? (36) Das Geld hat Müller alles bei Meyer abgeliefert. Beyond these similarities, however, there are significant differences: First, there is no l-alles counterpart to QF-alle(s) cases with mass noun antecedents, cf. (36) with (361): (36') Was hat Müller alles bei Meyer abgeliefert? (36") implies that Müller has handed over to Meyer a countable plurality of things; accordingly, an answer that consists only of Das Geld is inappropriate. Given (28), this difference comes as no surprise: since variable expressions cannot have the semantic properties of mass nouns, the requirement that the antecedent DP be plural and distributive is necessarily stricter in the case of l-alles than with Qf-alle(s). Second, inasmuch as not only l-alles but also QF-alle(s) may have plural we/cA-phrases as antecedents, they refer to this antecedent in semantically different ways, cf. (37-37'): (37) Welche Bücher hat Max alles für die Prüfung gelesen? (37') Welche Bücher hat Max ?alle für die Prüfung gelesen? While (37) containing l-alles is a question about the totality of individual books Max read for the exam, (37') containing QF-alle(s) can only be interpreted as a question about the subclassics) of books all of which Max read for the exam (see also Sportiche (1988:438)). As a consequence, the respective appropriate answers also differ: suppose, for example, that (3737') is asked in a situation in which Max has to read three books each on syntax, semantics, and phonology for a linguistics exam. An appropriate answer to (37) would have to be exhaustive (see section 1) but otherwise could list any (plural) subset of these nine books, including 'mixed lots' as in (38a,b). In contrast, an appropriate answer to (37) could only refer to natural subclasses as in (38') but need not list them exhaustively. As a consequence, due to the individual:subclass distinction, in a situation in which (38b) is an appropriate answer to (37), (371) could only be answered by (38'b/c). On the other hand, due to the different " Only neuter mass nouns, however, are possible antecedents (cf. Vater (1980:241); see also Porennec (1988:96f.)). There is no convincing explanation for this restriction.

476

requirements on exhaustiveness, (38'b) and/or its more natural equivalent (38'c) may be an appropriate answer to (37) even in a situation where Max has read exactly the five books referred to in (38b), whereas in the same situation (37) could only be answered by (38b). (38)

(38')

(a) Satzverschränkung im Deutschen, Semantic Structures, Wahr neben Falsch, Metrical Phonology, (b) Bausteine syntaktischen Wissens, Barriers, Satzverschränkung im Deutschen, Semantic Structures, Metrical Phonology. (a) Die zur Syntax und die zur Phonologie. (b) Die zur Syntax. (c) Die zur Syntax, das ist ja sicher das einzige, was dich interessiert, hat er durch.

This difference also manifests itself in the fact that QF-alle(s) and l-alles may in principle (although uneasily) cooccur in the same question, cf. (39), which is a question about all subclasses of books Max has read in their entirety, with the understanding that there is more than one such subclass: (39)

Welche Bücher hat Max denn alles für die Prüfung schon ?alle gelesen?

It should be possible to bring this difference into line with (28) and the relational difference between l-alles and QF-alle(s) with respect to possessive phrases noted in 2.7. As a first step in this direction let us try to paraphrase welch- clauses containing l-alles vs. QF-alle(s) in the usual way: (40) (40')

(a) Welche Bücher hat Max alles gelesen? (b) Welche Bücher hat Max ?alle gelesen? (a) What are all , a book, such that Max has read x? 'Name all the books which Max has read.1 (b) What are the , a book, such that Max has read allx? 'Name the books all of which Max has read."

It will be obvious that, in order for (40b) to be a meaningful utterance, something must be done about the italicized part of its literal meaning (40'b). (This explains why welch- clauses containing QF-alle(s) are at best halfway acceptable.) What happens, apparently, is an accommodation such that a definite plural antecedent for QF-alle(s) is extracted from the restrictor phrase,11 yielding roughly the following meaning: In other words, the question is interpreted as ranging over the sum objects denoted by the (definite) plural DP (for this interpretation of plural DPs cf. Krifka (1992)) contained in the welch- phrase. Note that, contrary to common belief (see La. Pesetsky (1987)), welch- is not inherently definite (or inherently 'discourse-linked1), cf. uses as the following, where no previously established set is referred to: (i) Welcher Idiot hat denn die Lampe montiert? Consequently, the deGniteness usually characteristic of welch- phrases must be a derived effect, which is largely based on the (denotation of the) NP welch- is in construction with (perhaps interacting with the entire construction itself). If so, the parallel to the wessen cases is even more apparent (see also the following note), and the claim that l-alles is incompatible with definite antecedents can be easily upheld.

477 (40")

(b) What are the X, X a class of books, such that for all y, y a book and y in X, Max has read y? 'Name the classes of books Max has read all members of.'

Note, moreover, that we/cA-X-phrases combining with l-alles allow an attributive paraphrase and disallow the partitive construction, cf. (40a-41), while with QF-alle(s) it is just the other way around, cf. (40b-41'): (41) (4l1)

(a) (b) (a) (b)

Welche Dinge, die Bücher sind, kennt Max alles? ?*Welche der Bücher hat Max alles gelesen? ?*Welche Dinge, die Bücher sind, hat Max alle gelesen? Welche der Bücher hat Max alle gelesen?

If we think of partitive questions like (41b), (41'b) as questions about a subclass of books, whereas attributive questions like (41a), (41'a) are about the subclass of entities that are books, then the parallel to the possessive cases may be drawn as follows: with QF-alle(s), the definition of the set X the we/cA-question is about and QF-alle(s) quantifies over involves X=Bticher from the start, i.e. the syntactic antecedent wh-phrase as a whole, whereas with Ialles, welch- may be just associated with the set of entities which then are specified as 'Bücher'.12 While this needs more working out, it seems to be on the right track, cf. (42-42'): (42) (42')

*Welche drei Bücher hat er alles gelesen? Welche drei Bücher hat er alle gelesen?

(42-42') are not questions about entities that are books, but about one triplet of books from the start. Hence, l-alles is out from the start, because its plurality restriction is violated, whereas QF-alle(s), which can be accommodated to quantify over the three members of this subclass, is possible. Let us note, finally, the most telling semantic difference of all: the meaning of QFalles(s), which can be focussed and modified, is part of the truth-conditional content, that is part of the asserted proposition in declaratives, and of the questioned proposition in interrogatives; as a consequence, QF-alle(s) may reappear in answers.

In contrast, the

meaning of l-alles is a conventional implicature: it may not reappear in the answers to whquestions, and, if used in restrictive relative clauses, it does not add to the restriction itself.

2 Paraphrases of wessen-cases containing l-alles vs. QF-alle(s), in which the relevant parallels to uv/c/t-cases are brought out, are given in (i-i1): (i) (a) Wessen Bücher hat Max alles gelesen? (b) Wessen Bücher hat Max alle gelesen? (i1) (a) What are all , a person, such that Max has read books of x? 'Name all persons Max has read books of.' (b) What are the , a person, such that for all y, y a book of x, Max has read y? •Name the persons for whose books it is true that Max has read them all.'

478 Also, the meaning of l-alles may neither be focussed nor modified (see 2.2-2.3), which is typical for conventionally implicated meanings. In sum: there are considerable differences in meaning effects between l-alles and QFalle(s). Since they cannot be reduced to different environments, this is final evidence that Ialles and QF-alle(s) in fact have to be assigned to different categories.

3.

The category of l-alles

What then is the category of l-alles? I take it to be an undisputable fact of German noun phrase structure that there is some separate category of syntactic DP quantifiers, to which QFalle(s) belongs.13 The behavior of QF-alle(s) as evidenced in 2.1-2.8 directly reflects the syntactic-semantic properties of this category. If so, the conclusion immediately following from 2.1-2.8 is that l-alles is not a syntactic DP quantifier. Still it is quantificationally related to DPs; hence, in determining its category, we must look for other types of elements that may bear such a relation. Since l-alles is uninflected, the only possible candidates are adverbs and particles. In the following, I shall argue that only the latter is a viable alternative. 3.1

\~alles: a quantificational adverb?

If l-alles were an adverb, we would expect it to behave like quantificational adverbs denoting totality such as allesamt, samt und sonders, insgesamt, durchweg, ausnahmslos, vollzählig, for these adverbs may quantity over DPs in various syntactic positions, even including PPs,

cf. (43): (43)

(a) Diese Vorschläge wurden samt und sonders/ausnahmslos/insgesamt/durchweg abgelehnt. (b) Die Studenten erschienen vollzählig. (c) Er hat unsere Vorschläge samt und sonders/... verworfen. (d) Von diesen Vorschlägen ist samt und sonders/... nichts zu halten. However, there are a number of decisive differences. First, quantificational adverbs may be focussed and modified, their meaning clearly belonging to the truth conditional content of their clause: (44)

(a) Die Vorschläge wurden Ausnahmslos abgelehnt, (b) Die Studenten waren nicht/fast vollzählig erschienen. Second, they may not appear in the typical l-alles position within the DP they quantify over: " S. Vater (1984). For a recent (and to my mind convincing) discussion of the phrase structural properties of this class, cf. Giusti (1991b). In order to avoid misunderstandings, I always refer to this class of so-called quantifiers as 'syntactic quantifiers'.

479

(45)

Was (* ausnahmslos) für Studenten (* ausnahmslos) würde man empfehlen können?

Third, in the case of complex wh-phrase antecedents, they relate to the antecedent phrase as a whole, just like QF-alle(s), and not to the possessor phrase, as does \-alles, (cf.2.7): (46)

Wessen Kinder haben samt und sonders/ausnahmslos/... bei Max Unterricht?

Likewise, if the quantificational adverbs are distributive (which, for example, insgesamt is not), they relate to welch- phrase antecedents like QF-alle(s) rather than l-alles, (cf.2.8): (47)

Welche Vorschläge fanden ausnahmslos keine Zustimmung?

Fourth, although there are configurations like (48), in which the wh-phrase is said to be quantified by the adverb of quantification in the matrix clause above it (see Berman (1991)), l-alles does not act this way, as shown by its cooccurrence with other adverbs of quantification in these configurations, and the corresponding reading, cf. (49): (48) (49)

Hans dürfte ausnahmslos wissen, wen Susi eingeladen hat. = Tür alle, die Susi eingeladen hat, dürfte Hans wissen, daß Susi sie eingeladen hat.1 Wer schon alles dürfte meistens wissen, wen Susi eingeladen hat ( - ihr wißt schon, wer: natürlich Max, Fritz und Lilo.)14 = 'Für alle, die Susi eingeladen hat, dürften Max, Fritz und Lilo meistens wissen, daß Susi sie eingeladen hat1. = 'Für die meisten, die Susi eingeladen hat, dürften Max, Fritz und Lilo wissen, daß Susi sie eingeladen hat1.

In sum: if anything, there is a clear overlap in crucial semantic properties between quantificational adverbs and QF-alle(s). The syntactic and semantic properties of these adverbs and Ialles are way too different to put them into the same category. This leaves only one possibility: l-alles must be a particle with quantificational force. 3.2

l-alles : a quantifying particle

We can give substance to the claim that l-alles is a particle by taking a well-established particle class such as modal particles as a paradigm case of particle behavior. Syntactically, the main characteristics of modal particles (as of other particle classes) is that they fall outside the X-bar schema, which means (i) they cannot be modified or expanded (although there can be a combination of them), (ii) they are not maximal, which bars them from canonical XP

" In order to make comparison possible, I have forced a rhetorical, hence assertive reading for the wh-a//es construction (which, due to the quantiflcational force that l-alles has over its antecedent, is necessarily specific). (One of the problems inherent in Herman's approach is that it cannot easily be extended to cases where the matrix clause has a bona fide nondeclarative interpretation.)

480

positions (such as the 'initial field' in verb-second clauses).

Rather, (iii) modal particles

appear in special ('clitic') positions, which are primarily (a) the right margin of the so-called 'Wackernagel position', their base position, but also (b) the position right adjacent to the whphrase in the initial field. Semantically, the property that is specific to modal particles is (iv) that they convey 'modal meanings', i.e. they have functions pertaining to sentence mood, discourse, and communication situation in general. The general property, which extends also to other particles (for example, to the socalled 'focus particles') is (v), that their meanings do not figure in the truth-conditional content of the proposition, but rather as implicatures. Likewise, (vi) the content of modal particles as such cannot be focussed/accented, unless exceptional conditions of interaction with sentence mood obtain.15 Moreover, to add first things last, (vii) they are uninflected. It will be obvious that l-alles shares almost all of these properties, which are, one the whole, general particle properties:

(i), (ii), the (b)-part of (iii), (v), (vi), (vii). Hence, by

comparing it with modal particles, most properties of l-alles fall out as characteristic particle properties. Naturally, l-alles is not a modal particle (hence (iv) does not apply), but a quantifying particle with a distinctive property: it quantifies over variable expressions denoted by operator phrases. As such, it is not unique, cf. quantifying elements such as genau, exakt, ungefähr, allein, so,16 which also take operator phrase antecedents, cf. (50), and share at the same time other salient properties of l-alles: they appear in the right-adjacent position to their antecedent as well as in the distant positions, cf. (50), which they may do in various combinations, cf. (51), (thus completing the parallel to modal particles with respect to (i)), and they also share the most critical properties of l-alles, including the diagnostic relational effect with possessive phrase antecedents, cf. (52), and the impossibility of modification and focus, cf. (53), which underlines that their contribution to sentence meaning is by way of implicature. (50)

(a) Was (genau/exakt/ungefähr) wird eigentlich (genau/exakt/ungefahr) gemacht? (b) Was (allein) für Leute (allein) kommen (allein) dafür (allein) in Frage?

" Regarding accented modal particles, see Ormelius (in prep.), and Meibauer (1992:Kap.3). l" I am neither claiming that this list is exhaustive (cf. elements and phrases such as (wer) sonst/außer ihm, etc.), nor that the properties of the items listed are totally identical: in the absence of any detailed studies of these elements in this perspective, I simply do not know. In fact, it may well turn out that some of the items form (or belong to even cross-cutting) subclasses. - This still leaves open the possibility that members of the same particle class have idiosyncratic properties. This might be claimed, for example, for the absence of so from the 'clitic' position in (50c), see also 3.4, or for überall, which optionally substitutes for l-alles in locative phrases and shows a preference for distant position. Given the fact that even the rather well-established particle classes (focus particles, modal particles) are beset with idiosyncrasies, this option can never be ruled out

481 (51) (52) (53)

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

Mit wem (?*so) habt ihr denn (so) in der Kneipe (so) gesprochen? Womit alles hat er sich denn so beschäftigt? Mit welchen Leuten genau hat er sich denn so alles beschäftigt? Wessen Bücher genau habt ihr denn für die Prüfung lesen müssen? Wessen Bücher kommen denn so/genau als Lesestoff in Frage? *Was sehr genau/ganz ungefähr wird eigentlich gemacht? 'Womit (geNAU) hat er sich (geNAU) beschäftigt? [* in the relevant meaning of genau]

Obviously, these elements overlap with l-alles in crucial properties that are typical particle properties at the same time, thus allowing to set up a class of 'quantifying particles'. If so, the claim that l-alles is a quantifying particle is a rather substantial descriptive claim. Taking a closer look at the relationship betweem l-alles and the other quantifying particles, we note that genau, ungefähr, etc. (but not so) may also occur with non-operator phrases, cf. (54): (54)

(a) Genau/ungefähr/exakt/allein diese Bewerber ziehen wir in Betracht, (b) Auf genau solche Fälle/exakt drei Fälle stützt sich seine Behauptung.

Since these phrases must be focussed, it is obvious that genau, ungefähr, etc. act as focus particles in these cases (as 'Gradpartikeln1 of the sort discussed in Altmann (1978), cf. also König (1991)). Clearly, these uses of genau/ungefähr are related to their uses with operator antecedents, suggesting that l-alles and so form a subclass of quantifying particles of their own. It does not suggest, however, that there is no special class of quantifying particles at all, for in cases like (50-52) ungefähr, genau, etc. do not act as focus particles: the operator phrase antecedent need not be (and as a rule is not) focussed, cf. also (55), right-adjacent position of the particles is perfectly possible, and distant position more or less all right, whereas with non-operator antecedents both are in any case worse, and frequently questionable, cf. (56-56'): (55) (56) (56')

(a) (b) (a) (b)

Wen genau/ungefähr würdest DU in Betracht ziehen? ?Diese Bewerber genau/ungefähr ziehen wir in Betracht, ?Auf solche Fälle genau hat sich seine Behauptung gestützt. ??Diese Bewerber ziehen wir genau/ungefähr in Betracht. ??Auf solche Fälle hat sich seine Behauptung genau gestützt.

This shows that appearing in a quantificational relationship with operator phrase antecedents constitutes a 'particle class function' of its own. We shall see in the following section that the locus where this class function originates is the clitic DP position. Before taking up this issue, a final remark concerning quantifying vs. modal content of particles is in order. Since quantifying particles most frequently occur with interrogative and exclamative operator phrases, which both strongly interact with sentence mood, it comes as no surprise that they also may get involved in modal and discourse functions. Thus, in universally quantifying over the variable expression in unembedded wh-interrogatives, l-alles also gives rise to the implicature that the answer must be exhaustive, and, furthermore, that there should be more than one true answer. This automatically excludes that the negative

482 answer is among them. As a consequence, wh-questions containing 1-alles always implicate not only that there is an

fulfilling the interrogative proposition, but that there is more than

one (which in most contexts is strengthened to 'many'). This in turn leads to specific effects, when l-alles is employed in rhetorical questions or exclamatives.17 Much the same is true for the other quantifying particles (the implicatures varying, of course, with the respective quantificational force). But this does not make them modal particles, of course, for these effects can be derived from their basic quantifying function as indicated. Note also that Ialles/so/genau, etc. occur as uninhibited in embedded wh-clauses as in unembedded ones, which is by no means true for modal particles, cf. (57). This is conclusive proof that modal and quantifying particle function must be kept strictly apart. (57)

3.3

(a) Heutigen Eltern ist egal, mit wem ihre Kinder alles/genau/so verkehren. (b) Heutigen Eltern ist egal, mit wem *denn/*nur/*schon ihre Kinder verkehren. l-alles : a quantifying wh-phrase clitic

It is clear from the preceding section that quantifying and modal particles are different kinds of particles. What we shall have to ask now is what exactly the difference consists of. Could it be that the difference is just semantic - modal particles contributing a modal implicature, quantifying particles a quantificational implicature, with all else being the same? The answer is no, as can be shown by carefully examining their positional behavior. At first glance, modal and quantifying particles seem to have the same positional options: both occur adjacent to operator antecedents in the initial field, both appear in variable positions in the middle field, cf. the apparently parallel distribution of l-alles and denn in (5858'): (58)

Was (denn) für Leute (denn) hat er (denn) damals (denn) bei dir (denn) treffen wollen? (58') Was (alles) für Leute (alles) hat er (alles) damals (alles) bei dir (alles) treffen wollen? At second glance, however, there are decisive differences. We observe, first, that quantifying particles occur freely as wh-phrase clitics, no matter whether the wh-phrase is in the initial position of verb-second or verb-final clauses, cf. (59). This is true for Standard German and dialects alike. In contrast, the occurrence of modal particles in this position is highly marked in verb-second clauses, and practically impossible in

*·' The effect of l-alles on exlamatives is insightfully discussed in Rosengren (1992:291ff.). As for rhetorical questions, as well as for the derivation of the respective implicatures in general, see Reis (in prep.).

483

verb-final clauses, cf. (59'); in dialects, the occurrence of modal particles as wh-phrase clitics is practically ruled out altogether (cf. Brandt/Reis/Rosengren/Zimmermann (1992:75-77)). (59) (59')

(a) (b) (a) (b)

Wen (alles) hat er (alles) ans MIT geschickt ? Man fragte ihn, wen (alles) er (alles) ans MIT geschickt habe. Wen (?schon/?denn) hat er (schon/denn) ans MIT geschickt? Man fragte ihn, wen (*?schon/*?denn) er (schon/denn) ans MIT geschickt habe.

Second, the positional options of modal particles in the middle field depend on focus-background-structure: all material intervening between the Wackernagel position and the modal particle must be unfocussed. As a consequence, sentences such as (60) are not good. Exactly parallel sentences containing l-alles, however, are unobjectionable, cf. (60'): (60) (60')

(a) (b) (a) (b)

?*Was hat er heute dem CHEF denn von dem Krach erzählt? ?*Wo hat ihm FRANZ gestern denn Fehler nachgewiesen? Was hat er heute dem CHEF alles von dem Krach erzählt? Wo hat ihm FRANZ gestern alles Fehler nachgewiesen?

On the other hand, l-alles seems to be unacceptable in the position it occupies in (61), whereas a modal particle is allowed there, cf. (61'): (61) (6l1)

*Wen hat er der Prüfung alles HEUTe ausgesetzt? Wen hat er der Prüfung denn HEUTe ausgesetzt?

Obviously, the distribution of modal and quantifying particles in the middle field is governed by different regularities: the distribution of modal particles is best explained by positing the right margin of the Wackernagel position as their base position, over which other constituents can scramble (for a more detailed discussion of this hypothesis, cf. Brandt/Reis/Rosengren/ Zimmermann (1992:71ff.)). Since focussed constituents do not scramble,18 this accounts for all the ordering facts of modal particles presented in (58-6 ) (the prediction for (58) being that the various versions are good if the main stress is located either on constituents preceding the middle field proper, or on constituents following the modal particle in its respective position). In contrast, the distribution of quantifying particles seems to correspond to the positional possibilities of the antecedent wh-phrase, which include (a) its base position, in which it may

18

For this restriction, which goes back to Lenerz1 (1977) 'rheme condition', cf. von Stechow/Sternefeld (1988:466), Grewendorf/Sternefeld (1990:15), and Fanselow (1990:115ff.). A weakened version of this restriction allowing for minimal focus on scrambled phrases is adopted in Geil ruß (1991:23), but, as far as I can see, it is not borne out by the facts. Note that elements moved to the Wackernagel position may be focussed (including full NP subjects that may precede personal and reflexive pronouns). At least from the perspective of focus-background structure, it seems plausible, then, that this movement does not belong to scrambling at all.

484

be scrambled over, (b) all the positions it may reach by scrambling itself.19 In other words: Ialles in the middle field seems to be licensed by an adjacent wh-trace of its antecedent. This generalization, which has first been suggested by Pafel (1991) (without, however, taking a stand on the status of alles), covers all the l-alles data presented above (with no focussing restrictions on (58'), excepting perhaps bei dir), in particular the ungrammatical case (61): with aussetzen, the basic order is accusative object > dative object, which even under favorable scrambling conditions (main stress on the accusative, relative weight of accusative constituent, for example) can hardly be reversed. Hence, l-alles in (61) is in a position, which it can neither have reached by scrambling, since scrambling is always leftward, nor by having been scrambled over by the dative object, which in this particular case is disallowed. As is to be expected, an overt wh-phrase in this position is also impossible, cf. (61"): (61") * Wer hat der Prüfung wen (alles) HEUTe ausgesetzt? Let us assume, then, that Pafel's generalization is basically correct. If so, two conclusions can be drawn. First, the base positions of modal and quantifying particles are different. Hence, the difference between these classes must also be syntactic. Second, the base position of quantifying particles must be a clitic position right adjacent to the operator phrase, otherwise the observed distribution would not make sense. If so, we must assume that there are (optional) clitic base positions to the right of wh-phrases, in which quantifying particles can be base-generated. In other words, quantifying particles are wh-phrase clitics from the start.20 As a consequence, modal vs. quantifying particle clitics to wh-phrases in the initial field would be accounted for in different ways: a quantifying particle would get there by being (optionally) pied-piped along with its host wh-phrase. A modal particle, however, being base-generated in a different position, would have to undergo an extra process of cliticization to the wh-phrase it appears with. As to how and where this process takes place, there is not much evidence to make use of: multiple question cases such as (62) show that modal 19 Following Fanselow (1990:116-118), it has frequently been claimed that wh-phrases do not scramble, see von Stechow/Sternefeld (1988:466), Grewendorf/Sternefeld (1990:14). The data used to support this claim, however, are on the whole unconvincing (thus, Fanselow's main example *wie hat was der Mann gestern repariert? (ibid, p.117) is already out for a more general reason: no object may scramble over subjects in this case). As far as I can see, if factors like accent/focus, definiteness, grammatical status, heaviness of the constituents involved and other factors known to influence scrambling are carefully controlled, wh-phrases behave just like comparable phrases in the middle field with respect to scrambling. See also Geil fuß (1991:29ff.). 20 Calling them clitics is motivated by two criteria: they are subcategorized to attach to specific types of syntactic phrases, their "hosts', and they must show liaison with them (see Spencer (1991:377), following Klavans (1985)). What is unusual about 'quantifying particle clitics' is that they may remain attached to a phrasal host position that is phonologically empty.

485

particles can also cliticize in the middle field (although the results are highly marked), and there is some indication that they cannot be left behind, cf. (62'a), where the position of the second wh-phrase by itself is legitimate, cf. (62b). (62) (621)

(a) Wer schon hätte damals wen schon fürchterlich ernst genommen? (b) Wer schon hätte wen schon damals fürchterlich ernstgenommen? (a) ?*Wer schon hätte wenj damals t; schon fürchterlich ernstgenommen? _ ?* —l

If so, modal particles either cliticize to their host wh-phrase in the middle field before whmovement, and obligatorily pied-pipe along, or they cliticize to their host wh-phrase later on (in Surface Structure or Phonetic Form). In order to decide between these hypotheses, one would have to check the behavior of modal particles with respect to wh-phrases that are basegenerated in initial position but, unfortunately, the evidence from the scope-marking was-construction, as the only case in point, is far from conclusive: (63)

(a) Was meint er denn/??Was denn meint er, wer die Wahl gewinnt? (b) ??Was wird er schon meinen/*Was schon wird er meinen, wer die Wahl gewinnt?

All we know for sure, then, is that modal particles can only be cliticized to overtly realized wh-phrases.

This by itself, however, is a considerable difference to quantifying particles.

Note, moreover, that the l-alles cases parallel to (62) are perfectly acceptable, likewise the ones parallel to (62'), whereas l-alles relating to the scope-marking was is squarely ruled out, cf. (64) vs. (65):

(64) (64') (65)

(a) (b) (a) (a) (b)

Wer alles hat damals wen alles fürchterlich ernstgenommen? Wer alles hat wen alles damals fürchterlich ernstgenommen? Wer alles hat weni damals t; alles fürchterlich ernstgenommen? Was meint er, wer alles in Betracht kommt. *Was alles meint er/*Was meint er alles, wer in Betracht kommt.

This confirms that in fact two different cliticization phenomena, based on different base positions, are involved, thus also confirming that modal and quantifying particles are syntactically different categories. 3.4

Defending the analysis

While the conclusion that modal and quantifying particles are syntactically different categories seems to be established beyond doubt, the particular clitic analysis given for quantifying particles in 3.3 still has to face at least two problems. The one problem is (at least according to the intuitions of some speakers) the existence of apparent counterexamples to Pafel's generalization concerning nonadjacent l-alles.

l-alles

may appear in certain positions of the middle field under conditions in which the respective

486 wh-phrase is much worse or disallowed: (a) in the position immediately adjacent to the Wackernagel position, cf. (66-66'); (b) in the position immediately preceding the nonscrambling elements in the rightmost position(s) of the middle field, cf. (67-67') and (6868'):21 (66) (66') (67) (67) (68) (681)

Welche Politiker/Wen HAT denn alles Karl so verleumdet? ?*Wann HAT denn welche Politiker (alles)/wen (alles) Karl so verleumdet? Wer hat denn die Mutter im Krankenhaus alles beSUCHT? ?*Wann hat denn die Mutter im Krankenhaus wer (alles) beSUCHT? Wer hat denn die Mutter damals alles nach Hause beGLEItet? ?*Wann hat denn die Mutter damals wer (alles) nach Hause beGLEItet?

It seems that these asymmetries can only be accommodated by assuming an exceptional movement of l-alles in the case of (a), and an exceptional tolerance of l-alles towards being scrambled over in the case of (b). On a more general level, however, these exceptions are not really ad hoc but rather in tune with the present analysis. As for (a), note that the exceptional movement looks like movement to the modal particle position, not just exceptional scrambling of object-related l-alles over the subject, cf. the ungrammaticality of (69-691). In other words, the exceptional movement seems to be licensed by the properties that modal particles and quantifying particles have in common, which most likely is their clitic property. If so, (a) testifies to a cliticization process, with l-alles cliticizing to the right of bona-fide modal particles. (69) (691)

*Welche Politiker/*Wen hat denn mit an SICHerheit grenzender Wahrscheinlichkeit alles Karl verleumdet? *Wann hat denn mit an SICHerheit grenzender Wahrscheinlichkeit welche Politiker/wen (alles) Karl verleumdet?

As for (b), this exceptional tolerance towards scrambling is only possible, as long as no focussed material is scrambled, cf. (70), which is, of course, exactly parallel to the condition modal particles in the same position are subject to (see 3.3). Note also that in this exceptional position, l-alles may cooccur with modal particles, but must be ordered after them, cf. (71). This, too, is most easily understandable on the basis of the exceptional movement process underlying (a) that makes l-alles part of a specific particle sequence, thus confirming that (b) is also a typical form of particle behavior. As is to be expected, corresponding Q¥-alle(s) and its antecedent may not appear in these positions either, cf. (i-ii1), thus adding to the evidence presented in section 2 that l-alles and QF-alle(s) are indeed distinct. ? »Diese Pol itiker/Sie HAT j a alle Karl so verleumdet. ?*Damals HAT ja alle diese Politiker /sie alle Karl so verleumdet (a) ??Die Skatfreunde haben die Mutter im Krankenhaus alle beSUCHT. (b) ??Die Skatfreunde haben die Mutter alle nach Hause beGLEItet (ii1) (a) *Dann haben die Mutter im Krankenhaus alle die Skatfreunde beSUCHT. (b) »Dann haben die Mutter alle die Skatfreunde nach Hause beGLEItet. For related observations, see Geilruß (1991: 54).

487 (70) (71)

(a) ?*Wer hat denn die Mutter im KRANKenhaus alles besucht? (b) ?*Wer hat denn die MUTTer damals alles nach Hause begleitet? (a) Wer hat die Mutter im Krankenhaus denn alles/?*alles denn beSUCHT? (b) Wer hat die Mutter damals denn alles/?*alles denn nach Hause beGLEItet?

What we find, then, is a kind of mirror image relation between modal and quantifying particles. Whereas modal particles originate in the right margin of the Wackernagel position, that is as VP- or I-projection clitics, and exceptionally appear as wh-phrase clitics, with quantifying particles it is just the other way around. I should like to interpret this in the way already indicated: on the one hand, modal and quantifying particles are alike in being clitics, a property not regularly shared by other particle types; this accounts for their being able to appear in the same positions. On the other hand, their behavior with respect to these positions is different, which is in line with the assumption of different base positions for modal vs. quantifying particles (see 3.3 and below). If taken this way, the ordering possibilities of Ialles extending beyond Pafel's generalization are satisfactorily accounted for. The other problem one might raise against analyzing quantifying particles as wh-phrase clitics is exemplified by so, and has already been illustrated in (50c) above: while so undoubtedly is a quantifying particle related to wh-phrases, it does not freely appear in the position immediately adjacent to them, cf. (50c) above and (72), thus lacking the most typical feature of wh-phrase clitic behavior. Cases like (72') are no evidence to the contrary, for the real wh-phrase clitics in these cases are l-alles and zum Beispiel, cf. (72"), which so has combined with and is dependent on:22 (72) (72') (72")

?* Wen so könnte man sich auf diesem Posten vorstellen? Wen so alles/Wen so zum Beispiel könnte man sich auf diesem Posten vorstellen? Wen alles/Wen zum Beispiel könnte man sich auf diesem Posten vorstellen?

What so does seem to share, however, is the characteristic positional pattern of quantifying particles in the middle field (see Pafel (1991)), which means that so apparently cliticizes to wh-traces. As far as I can see, we have to acknowledge these facts as they are. We do not have to acknowledge, however, that they constitute a serious problem: (i) so is about the only quantifying particle exhibiting this behavior, which suggests that lexical idiosyncrasy is at stake (see

The same kind of dependency is exhibited by nicht, cf. (i-ii), which, however, is not a quantifying particle. Regarding the exclamative interpretation forced by pied-piped nicht alles as in (iib), cf. Rosengren (1992). 'i) 'Was nicht hat er sich (alles) nehmen lassen, ii) (a) Was hat er sich nicht alles nehmen lassen. Was nicht alles hat er sich nehmen lassen.

488 also note 16);23 (ii) the distribution of so in the middle field does allow to treat it as basically a wh-phrase clitic, the idiosyncrasy thus being restricted to the impossibility of pied-piping; (iii) comparable idiosyncrasies involving distant vs. adjacent position show up in some other cases, cf. the focus particle ebenfalls, which, unlike all other focus particles, must not be in a position immediately adjacent to its domain, cf. (73a), or wh-interrogatives involving partitive constituents, which are necessarily discontinuous, cf. (73b): (73)

(a) (*Ebenfalls) Ole Jensen ('ebenfalls) ist natürlich (ebenfalls) blond, (b) WaSj (*Interessantes) habt ihr denn gestern t; (Interessantes) erlebt?

With the apparent counterevidence disposed of, the analysis of quantifying particles as whphrase clitics presented in 3.3 seems pretty well established, too. An additional argument in its favor, it should be recalled, are the facts about homonymous focus vs. quantifying particles discussed in 3.2, for they suggest strongly that the position right adjacent to wh-phrases (and DPs in general) is intrinsically related to quantifying particle use.24 I should like to end this section with briefly speculating on how the syntactic differences between quantifying and modal particles worked out in 3.3 and 3.4 may be related to their different functions. Modal particles are in the service of functions which sentences fulfill as a whole and whose formal manifestations (notably manifestations of sentence mood) are located in the left peripheral positions: the C- and/or I-projections (for a more detailed account, cf. Brandt/Reis/Rosengren/Zimmermann (1992)). Hence, it seems natural for modal particles (a) that their base position is on the borderline between these projections and the Vprojection, from which they have scope over the latter, i.e. over the proposition, (b) that they

" In the case of complex antecedent welch- and wessen- phrases, the right-adjacent position of l-alles is also highly marked, if not unacceptable. The reasons for this are, however, clearly systematic: particles cliticizing to a wh-phrase want to cliticize to the wh-element directly (cf. the difference in markedness between was alles fir Leute and ?was ßr Leute alles), which in welch- and wessen- phrases is syntactically impossible. According to Giusti (1991a) complex tvo-pronouns (womit, worauf, etc.) behave the same way, but judgements seem to vary considerably. 24 This is, of course, also suggestive of historical developments. The most interesting case in point in this context is perhaps inflected right adjacent all-, which is usually unstressed, cf. note 2, and, in the rare cases, in

which it may have the same antecedents as QF-alle(s) and l-alles, seems to behave like l-alles, cf. (i-ii): (i) (a) Welche Bücher alle hat Max für die Prüfung gelesen. (b) (c) (ii)

(a) (b) (c)

Welche Bücher hat Max alles für die Prüfung gelesen. Welche Bücher hat Max ?alle für die Prüfung gelesen. (ia)=(ibV(ic) Wessen Bücher alle hat Max für die Prüfung gelesen. Wessen Bücher hat Max alles für die Prüfung gelesen Wessen Bücher hat Max alle für die Prüfung gelesen.

Thus, right adjacent alle acts as a kind of synchronic 'bridge' leading from QF-alle(s) to l-alles. Whether this also reflects the diachronic passage, is a question I cannot go into here.

489

may cliticize to the initial wh-phrase, which - by itself and by overtly realizing the +wh feature - is the primary formal correlate of interrogative sentence mood.25 Furthermore, it seems natural (c) that modal particles may appear in the main clause of interrogative was-constructions, see (63) above, for the initial was-phrase still marks the scope of the specific whphrase(s) and wh-interrogative sentence mood at the same time, although the specific whphrase(s) are located in the embedded sentence. Quantifying particles, in contrast, do nothing of the sort: they are just related to wh-phrases and the like inasmuch as they denote variable expressions over which quantification is possible.26 It stands to reason, then, that they are closely related to these operator phrases from the start, which is brought out by the above analysis as wh-phrase clitics. The fact that quantifying particles may also appear in typical modal particle positions should then be taken as an analogical extension that is syntactically licensed by the clitic property both types of particle share, and functionally motivated by the prominent role wh-phrases and the like also play in specifying sentence mood and discourse functions.

4.

Final remarks

In the foregoing discussion, two points have been reasonably well established: first, QFalle(s) and l-alles are instances of different categories; second, l-alles is a quantifying particle of a special sort, presumably a wh-phrase clitic. While these points are rather specific, they had to be argued with reference to a number of broader issues (in particular the syntax and semantics of DP-quantifiers, of particle types, and wh-constructions), which in turn makes them relevant for pursuing these broader issues further on.

In conclusion, it should be

stressed, however, that the discussion offered in this paper is far from complete in many respects.

What is, for example, the structure of wh-phrases containing quantifying particle

clitics? Exactly which elements belong to this particle class, and what are the regularities governing their cooccurrence with other particles in the respective 'clitic1 positions? Is it really justified to speak of clitics in all these cases? On a more general level, what is the semantics and pragmatics of quantifying particles, and how exactly do they interact, for example, with the semantics and pragmatics of wh-interrogatives? How do they interact with other types of operators? Are there parallels to quantifying particles in other languages, and,

25

Note that clitic modal particles in the middle field presuppose that the same particle occurs as a clitic to the initial wh-phrase, cf. the examples (62)f. discussed in 3.3.

26

Some wh-phrases do not allow for quantification, cf. Twam alles/*warum alles, *wieso alles, *wie alles, *wie (teuer/groß, etc.) alles. The reasons for this are partly obscure.

490

if not,27 how is this to be explained? Finally, how did the quantifying particles, which all have cognates in other syntactic classes, develop, and in what way do the historical data bear on the descriptive analysis? The answers to these questions no doubt hold further surprises in store, the discovery of which, however, will have to wait for another occasion.

Bibliography Altmann, H. (1978): Gradpartikel-Probleme: Zur Beschreibung von gerade, genau, eben, ausgerechnet, vor allem, insbesondere, zumindest, wenigstens. - Tübingen (=Studien zur deutschen Grammatik 8). Berman, S.R. (1991): On the Semantics and Logical Form of wh-Clauses. - unpubl. Ph.D. Dissertation, University of Massachusetts, Amherst. Bhatt, C. (1990): Die syntaktische Struktur der Nominalphrase. - Tübingen (=Studien zur deutschen Grammatik 38). Brandt, M., M. Reis, I. Rosengren & I. Zimmermann (1992): "Satztyp, Satzmodus und Illokution." - In: I. Rosengren (ed.): Satztyp und Illokution (Tübingen) 1-90. Drubig, H. B.d (1991): Fokusstruktur und Fokusprojektion im Englischen. - MS. Tübingen. [To appear in: Sprachtheoretische Grundlagen für die Computerlinguistik: Arbeitspapiere des Sonderforschungsbereichs 340] , M. (1991): "The Semantics of Specificity." - In: Linguistic Inquiry 22,1-25. Fanselow, G. (1988): "Aufspaltung von NPen und das Problem der 'freien' Wortstellung." - In: Linguistische Berichte 114,91-113. Fanselow, G. (1990): "Scrambling as NP-MovemenL" - In: G. Grewendorf & W. Sternefeld (eds.): Scrambling and Barriers (Amsterdam/Philadelphia) 113-40. Fehlisch, U. (1986): "Jeder: Stellungs- und Referenzeigenschaften." - In: H. Vater (ed.): Zur Syntax der Determinantien (Tübingen) 83-122. Geilfuß, J. (1991): "Scrambling und Pseudoscrambling." - In: J. Geil fuß: Verb- und Verbphrasensyntax (sSprachtheoretische Grundlagen für die Computerlinguistik: Arbeitspapiere des Sonderforschungsbereichs 340, Nr. 11) 19-57. Giusti, G. (1990): "Floating Quantifiers, Scrambling and Configurationality." - In: Linguistic Inquiry 21,633-41. Giusti, G. (1991a): "The Syntax of Floating alles in German." - In: W. Abraham, W. Kosmeijer & E. Reuland (eds.): Issues in Germanic Syntax (Berlin/New York) (=Trends in Linguistics: Studies and Monographs 44) 327-350. Giusti, G. (1991b): "The Categorial Status of Quantified Nominals." - In: Linguistische Berichte 136,438-54. Grewendorf, G. & W. Sternefeld (1990): "Scrambling Theories." - In: G. Grewendorf & W. Sternefeld (eds.): Scrambling and Barriers (Amsterdam/Philadelphia) 3-37. As far as I know, there is no parallel to l-alles in the other Germanic languages (suggestive parallels in Hungarian and Chinese have been pointed out to me by V. Molnar and Li Jie, respectively).

491 Giewendorf, G. & W. Sternefeld (eds.) (1990): Scrambling and Barriers. - Amsterdam/Philadelphia (^Linguistik Aktuell 5). Grimshaw, J. (1990): Argument Structure. · Cambridge, Mass. (=Linguistic Inquiry Monographs 18). Hawkins, J. (1978): Definiteness and Indefwiteness: A Study in Reference and Grammaticality Prediction. London. Hawkins, J. (1991): "On (Indefinite Articles: Implicatures and (Un)grammaticality Prediction." - In: Journal of Linguistics 27,405-42. Klavans, J.L. (1985): "The Independence of Syntax and Phonology in Cliticization." - In: Language 61,95-120. Kniffka, G. (1986): "Zur Distanzstellung von Quantoren und Qualifikatoren im Deutschen." - In: H. Vater (ed.): Zur Syntax der Determinantien (Tübingen) 57-82. König, E. (1991): The Meaning of Focus Particles: A Comparative Perspective. - London. Krifka, M. (1992): "Definite NPs Arent Quantifiers." - In: Linguistic Inquiry 23,156-63. Lenerz, J. (1977): Zur Abfolge nominaler Satzglieder im Deutschen. - Tübingen (=Studien zur deutschen Grammatik 5). Link, G. (1974): "Quantoren-Floating im Deutschen." - In: F. Kiefer & D.M. Perlmutter (eds.): Syntax und Generative Grammatik. Vol. 2 (Frankfurt) 105-27. Meibauer, J. (1992): Modaler Kontrast und konzeptuelle Verschiebung. Studien zur Syntax und Semantik deutscher Modalpartikeln. [Habilschrift, preliminary version]. - MS. Tübingen. Müller, M. (1986): "Zur Verbindbarkeit der Determinantien und Quantoren im Deutschen." - In: H. Vater (ed.): Zur Syntax der Determinantien (Tübingen) 33-55. Olsen, S. (1989): "Das Possessivum: Pronomen, Determinans oder Adjektiv?" - In: Linguistische Berichte 120, 133-53. Ormelius, E. (in prep.): Die Modalpartikel schon. - To appear in: I. Rosengren (ed.): Satztyp und Illokution. Vol. 2 (Tübingen) (= Linguistische Arbeiten 279). Pafel, J. (1991): "Zum relativen Skopus von w- und Q-Phrasen (w/Q-Interaktion)." - In: M. Reis, I. Rosengren (eds.): Fragesätze und Fragen (Tübingen) 145-73. Pdrennec, M.-H. (1988): "Von floatenden Quantoren und topikalisierten Nominalphrasen." - In: Cahiers d'etudes Germaniques 14,91-104. Pesetsky, D. (1987): "Wh-in-Situ: Movement and Unselective Binding." - In: E. Reuland & A. ter Meulen (eds.): The Representation of (In)defmiteness (Cambridge, Mass.) 98-129. Reis, M. (in prep.): w-a//es-Konstruktionen im Deutschen: Zu ihrer Syntax, Semantik und Pragmatik. Reis, M. & Vater, H. (1980): "Beide." - In: G. Brettschneider & C. Lehmann (eds.): Wege zur Universalienforschung: Sprachwissenschaftliche Beiträge zum 60. Geburtstag von Hansjakob Seiler (Tübingen) (= Tübinger Beiträge zur Linguistik 145) 360-86. Reis, M. & I. Rosengren (eds.) (1991): Fragesätze und Fragen. - Tübingen (^Linguistische Arbeiten 257).

492 Roscngren, I. (1991): "Zur Fokus-Hintergrund-Gliederung im Deklarativsatz und im w-lnterrogativsatz." - In: M. Reis & I. Rosengren (eds.): Fragesätze und Fragen (Tübingen) 175-200. Rosengren, I. (1992): "Zur Grammatik und Pragmatik der Exklamation." - In: I. Rosengren (ed.): Satztyp und Illokution. Vol. l (Tübingen) 263-306. Rosengren, I. (ed.) (1992): Satztyp und Illokution. Vol. 1. - Tübingen (linguistische Arbeiten 278). Shlonsky, U. (1991): "Quantifier Phrases and Quantifier Float." - In: Proceedings of the Annual Meeting of the North Eastern Linguistic Society 21,337-50. Spencer, A. (1991): Morphological Theory. - Oxford/Cambridge, Mass. Sportiche, D. (1988): "A Theory of Floating Quantifiers and its Corollaries for Constituent Structure." - In: Linguistic Inquiry 19, 425-49. Stechow, A. von & W. Sternefeld (1988): Bausteine syntaktischen Wissens. - Opladen. Vater, H. (1980): "Quantifier Floating in German." - In: J. van der Auwera (ed.): The Semantics of Determiners (London) 232-49. Vater, H. (1984): "Determinantien und Quantoren im Deutschen." - In: Zeitschrift für Sprachwissenschaft 3,1942. Vater, H. (ed.) (1986): Zur Syntax der Determinantien. - Tübingen (=Studien zur deutschen Grammatik 31). Zaefferer, D. (1991): "Weiß wer was? Wer weiß was? Wer was weiß ... w-Interrogative und andere wKonstruktionen im Deutschen." - In: M. Reis & I. Rosengren (eds.): Fragesätze und Fragen (Tübingen) 7793. Zünmermann, I. (1991): "The 'Subject' in Noun Phrases: Its Syntax and Semantics." - In: I. Zinunermann (ed.): Syntax und Semantik der Substantivgruppe (Berlin) (=Studia grammatica 33) 33-68. Zünmermann, I. (1992): "Der Skopus von Modifikatoren." - In: I. Zimmermann & A, Strigin (eds.): Fügungspotenzen (Berlin) (=Studia grammatica 34) 251-79.

493

Paul Westney REVISITING -ing -ing1

1.

Introduction: the problem

Multiple adjacent -ings, as in (1) It's just starting raining have long attracted interest, both descriptively, since native speakers often detect a certain oddity about such examples (particularly when their attention is drawn to them) but rind it difficult to state the nature of the problem, and theoretically, since it is unclear how far the restriction relates solely to surface structure. The initial full-length investigation of the question by Ross (1972: here 61) takes off from the problem posed by the following set: (2) (3) (4) (5)

It continued to rain It continued raining It is continuing to rain *It is continuing raining

Since continue takes both infinitival and -ing complements without obvious restrictions, the challenge is to find a more adequate explanation for the oddity of (5) (and, similarly, (1)) than a general stylistic/aesthetic resistance to the sound or appearance of two -ings side by side. The paper by Ross (1972), followed by those by Milsark (1972), Herman (1973), Emonds (1973) and Pullum (1974), led generally to the conclusion that this is a surface structure phenomenon, restricted to cases where the second of the -ing forms is in the complement of the first; and this remains the view presented by McCawley in his overview of English syntax (1988:306-8). There seems to be no restriction on the grammatical function of the first -ing form, but there is some evidence that the second must be a participle, and not a gerund; there is, however, no consistency in the literature over the use of the terms 'participle' and 'gerund1. Pullum & Zwicky (1991a) present a number of arguments against distinguishing between different V-ing categories, and argue that the 'Doubl-i/ig' constraint applies to a single grammatical category which subsumes functions such as are traditionally associated with gerunds and present participles, and is thus a condition on a single syntactic rule.

M am grateful to Geoffrey Pullum foi letting me see, and quote from, two papers prior to publication (Pullum & Zwicky 1991a, 1991b).

494

2.

A phonological issue?

Aside from the studies mentioned above, there have been two further investigations that have contributed to descriptive knowledge of the problem area. Firstly, Bolinger (1979) produces a wide array of data that considerably complicate the question. His conclusions are, in sum, as follows: (a)

there is a general restriction on the proximity (as opposed to immediate adjacency) of verbal -ing suffixes; (b) this restriction is eased by the degree to which the two items are separated by words, pauses or grammatical boundaries; by familiarity, especially through lexicalization or colloquial status; and, above all, by prosodic considerations ("the problem is basically phonological", Bolinger 1979:56). Bolinger's conclusions thus reassert, with qualification, the traditional notion that the sound of two adjacent -ings is the basic problem; at the same time they suggest that a broad perspective is needed for adequate discussion of the phenomenon, and that the whole question is a matter of degree rather than of rule-bound restrictions. It has been proposed that the constraint may have originated as a "grammaticization of a phonesthetic dispreference for 'jingling' sequences of similar-sounding endings" (Pullum & Zwicky 1991b:6); the crucial question, quite clearly, concerns the synchronic status of such a constraint.

3.

The special case of being -ing

Secondly, Halliday (1980) deals with the question of being + -ing in phrases like being teaching. This concerns a notorious gap in the non-finite system of the English verb: the participle/gerund that corresponds to an infinitival form like to be teaching is teaching, whereas an initial being might logically be expected, as in being teaching. This means that teaching has to serve for the senses that correspond to both to teach and to be teaching. Halliday draws on a collection of naturally-occurring spoken data and cites examples like on the grounds of being teaching and being moving around so much. He is able to show that the 'normal' reduced forms would not express the exact senses required, and suggests that the gap in the non-finite verb system is in process of being filled. Quirk et al. accept the implications of these data, and give ... on the grounds of being teaching as a rare example showing that the troublesome being forms are "not quite impossible" (1985:154); however, Warner's outright rejection of Halliday's data as "strictly ungrammatical" (1985:74) no doubt represents a more general reaction. The being + -ing problem is commonly distinguished from the double -ing question being discussed here: whichever analyses of these phenomena are adopted, and whether they are

495

given a unitary or distinct accounts, it is usually felt that, in contrast to general double -ing, the restriction on being + -ing is absolute, or at least clearly defined. This subject will therefore be left out of the discussion here.

4.

The data

Two points stand out if one considers the evidence adduced in discussions of the double -ing problem: (a) (b)

the data are remarkably vague; the hard cases centre on a very small number of complement-taking verbs.

I shall look at these points in turn. The first question is well illustrated by Ross's paper, where the double -ing constraint, having been stated, is reformulated so as only to claim that no surface structures not having the phrase-structural and derivational properties specified in the double -ing constraint produce violations; the converse, that all structures having the properties specified are ungrammatical, is found too strong. The reasons for this are, firstly, that certain combinations do not produce the predicted violation, and, more seriously, that there is a hierarchy of acceptability among predicted violations, as shown in (just to give the adjacent verbal forms): avoiding contacting, lattempting breathing, ??trying swimming, ?* continuing fondling and ^keeping resisting (1972:78-79). Ross's judgements are not in fact radically contradicted elsewhere; and Bolinger has suggested several reasons for the various degrees of acceptability such pairs of items may be felt to have. However, there are some striking judgemental variations to be found in the literature. Thus, immediate adjacency is generally seen as a necessary condition for double -ing violation, so that Ross contrasts the following, where topicalization involving fronting the second -ing form predictably produces an acceptable sentence: (6) (7)

*I'm not particularly keen on trying kissing this moray eel Kissing this moray eel I'm not particularly keen on trying (1972:68)

By contrast, van Riemsdijk & Williams (1986:338) argue that violation occurs even after movement, as if the moved item were still in its deep structure position: (8) (9)

*The police are beginning searching the car *My car, searching which the police are beginning e today

A further illustration involves the verbs keep and keep on. It is generally claimed that keeping -ing, but not keeping on -ing, can cause violation (e.g. Bolinger 1979:47); Emonds, however, treats the following as a normal double -ing violation, with the intervening particle on not affecting the grammaticality status of the -ing + -ing sequence (1973:40):

496

(10)

*She is keeping on knitting sweatshirts

Such disagreement may seem surprising, but essentially serves to show how vague the data can be; and it does not affect the basic question why beginning searching and keeping (on) knitting should be at least a little odd, a matter over which there is no disagreement. However, consideration of the literature on the phenomenon in question makes it clear that there is a marked dichotomy between scholars who have, or at least describe, a grammatical double -ing constraint, and those for whom the issue is essentially fuzzy.

5.

The really hard cases

The second of the prominent issues mentioned in section 4 concerns the fact that the verbs involved - or at least those cited - in the hard cases of violation are strikingly few in number. It has often been noted that aspectual verbs are prominent here, along with the pair try and attempt. The following is a list of all the verbs that appear with asterisks in -ing + -ing combinations in Ross (1972) and the four other papers quoted above that followed it (the same examples recur throughout this literature): (11)

be, go, start, begin, continue, keep, stop, try, attempt

The only other items here - apart from the aspectuals and the try/attempt pair - are be and go. The case of being + -ing was discussed in section 3. Go may also be considered a special case, since the normality of going shopping, drinking, etc. as opposed to the oddity of going working, painting, etc. is to be related to a restriction on the use of go in this idiomatic 'expeditionary1 sense (Bolinger 1979:50), and has nothing directly to do with the use of -ing forms as such. We are, therefore, very clearly left with the aspectuals (which could, but do not, include cease) and the roughly synonymous pair try and attempt. I shall concentrate on these seven items in the ensuing discussion.

6.

Two crucial factors

I shall argue here that close examination of the data on these verbs suggests that two related factors may be crucially important in determining the oddity of certain double -ing combinations: (12)

(a) (i) the syntactic availability of alternative complement structures, and (ii) possible semantic distinction between these; and (b) the possibility of semantic anomaly when two -ing forms are adjacent.

497

In talking of oddity, I follow Bolinger in assuming that 'acceptability' is highly contextdependent (1979:42), so that it is difficult, and perhaps pointless, to insist on any specific judgements. Any sentence that seems odd is marked with a '?', unless it is being quoted from the literature, in which case its assigned status is retained.

7.

Continue

I shall start with continue. The implied argument relating to Ross's initial set, quoted above as (2)-(5), is that this verb (13)

(a) can take both -ing and infinitival complements without significant restriction or distinction; (b) can itself occur in simple or progressive forms without significant restriction or distinction; but (c) cannot occur in the progressive with an -ing complement.

All three points can be questioned, and I shall now examine them as they apply to continue. Subsequently I shall look at the other items listed in (11), omitting be and go for reasons given above. 7.1

Infinitival vs. -ing complements

The semantic distinction between infinitival and -ing complements with aspectual verbs is often seen as minimal, and even Quirk et al. (1985:1192) do not record such a possibility with continue. Significant differences between the semantics of the two types of complement with continue have, however, been observed. Boertien (1979) argues that -ing complements are semantically progressive, and are thus excluded in stative sentences, where infinitives are normal, e.g.: (14) (15)

He continued to own property until his death *He continued owning property until his death (Boertien 1979:44)

Kilby (1984:156) claims that, with continue, -ing complements refer to "a continuation of the same process" (which points to the duration sense of progressive aspect), while infinitives refer to "another, or several other, processes of the same sort". This claim is illustrated by the following set of two pairs: (16) (17) (18)

?After his first chapter had been rejected by the publishers, he continued to write the book After his first chapter had been rejected by the publishers, he continued writing the book After his first book had been rejected by the publishers, he continued to write books that were never published

498

(19)

??After his first book had been rejected by the publishers, he continued writing books that were never published

The acceptability judgements here depend on a rather fine distinction between the same process (with -ing) and a series of similar processes (with the infinitive). The combined evidence of (14)-(19) suggests that continue + -ing is associated with dynamic or progressive verb meanings (the semantics of infinitival complements do not directly concern us here). The following data provide further illustration. The senses of the verb be in the sentences He was reckless as a child (stative) and He was being reckless this morning (dynamic) are reflected in the natural use of infinitival and -ing complements, respectively, in the following: (20) (21) 7.2

He was reckless as a child and continued to be (?being) so all his life He was being reckless this morning and continued being (?to be) so all day Simple vs. progressive tense forms

Of the three components of progressive meaning that are often singled out (e.g. Quirk et al. 1985:198) - 'duration1, 'limited duration' and 'not necessarily complete' - the first is most obviously relatable to the use of continue in the progressive. If we now go back to the original problem case (5), the coexistence of the two -ing forms continuing and raining, being both semantically progressive, suggests redundancy, rather as would happen if still was added to (4) (It is still continuing to rain). To paraphrase this in simple terms, to say It is continuing raining is to say that a process is in process, while to say It is continuing to rain is to say that a state of affairs is in progress. This argument means that (5) is odd or anomalous as it stands; but it does not mean that continuing -ing is absolutely blocked. If we return to (17) and (18), the 'good' examples from the set (16)-(19), and consider them as as responses to What was he doing while you were building the garage? the following would be possible natural responses, respectively: (22) Well, during that period he was continuing writing the book (23) Well, during that period he was continuing to write books The double -ing in (22) can be justified by the fact that, exceptionally, both the main verb and the complement verb 'have to1 be in -ing forms. We may also consider the following example from Bolinger (1979:46): (24) But I don't think that continuing expanding is the answer to our economic woes This is noted as better than: (25)

(?)But I don't think that the answer to our economic woes is continuing expanding (= is that of continuing expanding)

499 Bolinger sees the relative prominence of the problem phrase (continuing expanding) in (24), as opposed to (25), as decisive here. One could, however, also see these examples in terms of available options. In (25) continuing has the alternative to continue available without significant semantic distinction (cf. (12a)). By contrast, in (24) to continue expanding (in subject position) would be rather odd. All this suggests that, to return to point (13c), the question of the claimed ungrammaticality of sentences like (5) must be relativized.

8.

Keep

The next verb to consider is keep, which is semantically close to continue and could, roughly, replace it in (3), (24) and (25). There are, however, considerable differences. Keep counts as a catenative (Quirk et al. 1985:147), and is much less versatile than a full verb. It is normally stative, and in many typical uses (e.g. keep smiling, he kept ringing us up at midnight) would hardly accept progressive aspect (e.g. *be keeping smiling, *he was keeping ringing us up at midnight); and it does not accept infinitive complements. McCawley illustrates the double -ing restriction with the following set (1988:306-7): (26) We kept working hard (27) We kept the employees working hard (28) *We were keeping working hard (29) We were keeping the employees working hard Example (28) cannot be considered exactly parallel to (5) since, for one thing, there is no infinitival alternative available. Further, the contrast of (28) and (29) is intended to suggest that immediate double -ing adjacency creates the deviance; but insertion of on after keeping hardly improves (28). It is in any case doubtful whether the keeps in (a) (26) and (28) and (b) (27) and (29) have similar enough meanings to be compared, a point supported by the fact that on, in the required sense, cannot be inserted in (27) and (29). Finally, (28), as opposed to the other sentences, is hard to contextualize. Despite these complications, the basic problem with (28) appears to be the double expression of progressive meaning, just as was suggested above for (5). Another problem case discussed is the following (Ross 1972:67): (30) *His keeping chanting ads bugs me In this connection, Bolinger sees keeping -ing as an 'especially sensitive1 combination (precisely because keep on is an available alternative), but nevertheless offers the following as acceptable (1979:47): (31)

Her keeping interfering all the time is what bothers me

500

The reason Bolinger gives for (31) being good, but not (30), is the fact that the former, but not the latter, has adequate rhythmic contrast, an illustration of Bolinger's insistence on taking the prosodic aspect of the double -ing question into account.

9.

Start and begin

Start and begin can be considered together, since their syntax and semantics are, for our purposes, identical. McCawley (1988:306) gives the following set, comparable in essentials to the set (2)-(5) discussed in section 7. (32) It began to rain (33) It began raining (34) It was beginning to rain (35) *It was beginning raining As with the continue set I shall look at the complement and tense contrasts in turn. The discussion happens to mainly concern start, but it applies equally to begin. 9.1

Infinitival vs. -ing complements

A contrast between 'potentiality' and 'performance' is noted as a possible distinguisher between the use of infinitival and -ing complements, respectively, with aspectual verbs of 'beginning', as in the following pair (Quirk et al. 1985:1192): (36) He started to speak, but stopped because she objected (37) He started speaking, and kept on for more than an hour The 'potentiality1 vs. 'performance' contrast can be further refined. While in (37) the 'speaking1 occurred over time, it is not clear to what extent the action was realized in (36): the subject could have merely opened his mouth in preparation. In the following similar pair, from Palmer (1988:176), however, the speaker in (38) clearly did perform: (38) He started to speak, but was soon interrupted (39) He started speaking, and kept on for hours This suggests that the contrast lies in something like the difference between incomplete and complete realization of the action denoted by the complement verb (regardless of whether or not the action is 'completed' on a specific occasion - in the present case, whether or not the speaker said all he intended). This can be illustrated further in the following pair from Kilby (1984:157): (40) (41)

The Queen started to collapse The Queen started collapsing

501

Kilby points out that while both are compatible with an iterative interpretation, (40) is "much more natural" with a single event. This can be shown as follows, with disambiguating extensions to suggest (a) repetition and (b) uniqueness: (40)

The Queen started to collapse

(41)

The Queen started collapsing

(a) at open-air events (b) but someone supported her (a) at open-air events (b) ?but someone supported her

In (40), both iterative and unique interpretations are possible, since in neither case is complete realization required. By contrast, in (41) complete realization is denoted by collapsing, but this is at odds with the unique reference suggested by (41b). 9.2

Simple vs. progressive tense forms

If we go back to (36) and (37) and replace the simple tense forms in the main verbs by progressive forms, the following results, reflecting the situation shown in, for example, (34) and (35):

(42) (43)

He was starting to speak, but stopped because she objected *He was starting speaking, and kept on for more than an hour

We must now, therefore, consider the semantics of start and begin in relation to progressive tense uses.

Such verbs denote transitional acts (Quirk et al. 1985:207-9), and thus "the

meaning of the predication effectively excludes duration", so that there is bound to be a semantic clash between the use of a progressive form of such a verb and a complement that suggests 'performance1 or 'actuality', since this denotes at the same time the transition in process, which is thus incomplete - here, he was starting - and the settled state of performance - here, speaking. The same comments apply to examples (1) and (35), and to progressive forms of (39) (He was starting speaking) and (41) (The Queen was starting collapsing). This discussion suggests that problems of semantic incompatibility are much greater with starting/be ginning -ing than is the case with continue and keep.

10. Stop Stop, of the aspectuals in (11), remains to be considered. This item only accepts -ing forms in complements: infinitival clauses express purpose or result (e.g. He stopped to smoke, He arrived to find the meeting in full swing). The following example from Ross, and variants of it, have been widely discussed: (44)

(*)The police are stopping drinking on campus at midnight

502

It is claimed that this is only good if the drinking is by other agents than the subject; this is one of the examples used to support the relevance of underlying structure in determining acceptability or otherwise. Yet cases where the subject of stopping and of the -ing form coincide are attested: (45) (46)

California shows no signs of stopping growing (quoted by Bolinger 1979:44) Stopping smoking can reduce the risk of lung cancer (on cigarette packets; supplied by Adrian Simpson)

Disputing grammaticality judgements is an invidious business, but the case for any general block on stopping -ing of the kind claimed is untenable. It is relevant that, as noted above, stop has no alternative complement structure available (cf. point (12a)).

11. Try and attempt The verbs that remain are try and attempt, which form another closely synonymous pair. An example of trying -ing was given above as (6); other cases are: (47) (48)

The air ministry is trying keeping the books in balance *I was attempting playing the "Minute Waltz" with my nose (Milsark 1972:542; Ross 1972:69)

Try takes infinitival and -ing complements, the former meaning something like 'make an attempt' and the latter 'do, as an experiment1. This distinction is standardly treated as clear-cut (a unique exception being COBUILD (1990:188), where no semantic contrast between the two complement structures is made), while attempt is normally only allowed the infinitive option; exceptionally, the LDOCE (1987) gives / attempted walking along the rope; but this clearly means the same as / tried to walk... The real problem with the verbs try and attempt is with the examples themselves. In (6), trying kissing does not seem likely to mean kissing and seeing what happens; if it does mean this, it is hard to see what is wrong with it, and if it does not, then the sentence is indeed odd in its choice of the -ing form. Similarly with (47), if this means keeping the books in balance for a change, to see if it makes a difference, then this is surely a good sentence; but it seems more likely to have the infinitival sense of make the effort (with no mention of success), in which case it will be odd. With (48) the sense of 'do, as an experiment' seems very unlikely, and in the other sense the infinitive is the normal option. Basically, there seems nothing odd about an exchange like What's he doing about his diet? - He's trying leaving half of everything on his plate, and it seems to be working quite well. In terms of (12), both (ai) and (aii) apply to try, while it appears that attempt is simply rarely used with -ing complements.

503 12. Conclusion This investigation has concentrated on the verbs that have been cited as being unacceptable in double -ing combinations. My aim has been to suggest that, whatever other considerations come into this question (and if we follow Bolinger, there are several), adequate account should be taken of the syntactic and semantic factors noted in (12).

If this is done, the

evidence for a syntactic restriction of the kind widely proposed in the literature is very much weakened; so perhaps it is time to start considering stopping bothering so much about multiple adjacent -ings.

Bibliography Herman, A- (1973): "Triple -ing." - In: Linguistic Inquiry 4,401-03. Boertien, H.S. (1979): "Toward a Unified Semantics of Aspectual Verbs with to and ing Complements." - In: Papers from the Regional Meetings of the Chicago Linguistic Society 15,42-52. Bolinger, D. (1979): "The Jingle Theory of Double -ing." - In: D.J. Allerton, E. Carney & D. Holdcroft (eds.): Function and Context in Linguistic Analysis (Cambridge) 41-56. COBUILD (1990): Collins COBUILD English Grammar. · London. Emonds, J. (1973): "Alternatives to Global Constraints." - In: Glossa 7, 39-62. Halliday, M.A.K. (1980): "On Being Teaching." - In: S. Greenbaum, G. Leech & J. Svartvik (eds.): Studies in English Linguistics (London) 61-64. Kilby, D. (1984): Descriptive Syntax and the English Verb. · London. LDOCE (21987): Longman Dictionary of Contemporary English. - London. McCawley, J.D. (1988): The Syntactic Phenomena of English. Vol. 1. - Chicago. Milsark, G. (1972): "Re: Double -ing." - In: Linguistic Inquiry 3,542-49. Palmer, F.R. (21988): The English Verb. - London. Pullum, O.K. (1974): "Restating Doubl -ing." - In: Glossa 8,109-20. Pullum, G.K. & A.M. Zwicky (1991a): "Condition Duplication, Paradigm Homonomy, and Transconstructional Constraints." - In: Proceedings of the 17th Annual Meeting of the Berkeley Linguistics Society. Pullum, G.K. & A.M. Zwicky (1991b): Subjacency between Present Participles. - MS., Center for Advanced Study in the Behavioral Sciences, Stanford, California. Quirk, R., S. Greenbaum, G. Leech & J. Svartvik (1985): A Comprehensive Grammar of the English Language. London. Riemsdijk, H.C. van & E. Williams (1986): Introduction to the Theory of Grammar. - Cambridge, Mass. Ross, J.R. (1972): "Double -ing." - In: Linguistic inquiry 3,61-86. Warner, A.R. (1985): The Structuring of the English Auxiliaries: A Phrase Structure Grammar. - Bloomington.

New Seasons

507

Franz Guenthner

Computational Linguistics: "A Personal View"1 1.

If Computational Linguistics is the answer, what was the question?

I would like to start with a few notes on the name of this (relatively new) discipline, with some examples the tasks this discipline has addressed in the recent past, which are not necessarily inherent in the name 'Computational Linguistics1. This paper has three parts: first, I would like to make a suggestion concerning a possible basic predicate of this discipline, and how this predicate could be specified so as to partitition the discipline into sub-areas, each with its own specific task, i.e. its specific predicate. Then, I will use this predicate to characterize three goals of Computational Linguistics that I consider to be of central importance. At the end of this note, I will make some remarks about the role of the programming language PROLOG in the computational processing of the predicate and in the concrete realization of the above goals.

2.

Computational Linguistics: 'Computable Linguistics'?

Computational linguistics is a young discipline with many names2 and to this day a discipline with almost as many different methodological and theoretical assumptions as there are practicioners. How are we to understand 'computational1? Two interpretations - at least - are possible: one addresses the concrete utilisation of computers and computer technology, the others involves concepts like 'enumerable' or 'programmable1. Both interpretations are, of course, related to each other in very direct ways. But as we shall see, it is the computability aspects that make the discipline interesting and it is the question of what exactly we want to compute in linguistics that this note deals with. In an extremely interesting paper, the late David Marr (1977) gave the following characterization of scientific progress in artificial intelligence:

I thank David Reibel for the many discussions we had when our offices were door to door on the 4th floor of the Neuphilologische Fakultät in Tübingen. This note is a summary of a talk given in the spring of 1988 at the Universität München. Bob Kowalski's book Logic for Problem Solving was an important source of inspiration for my views on the topics discussed here. 'Natural Language Processing', for example, is a typical alternative in English speaking communities. As the original version of this paper was written in German, part of its aim was to explore what possible interpretation of various German denominations (e.g. what the 'Computer' in 'Computerlinguistik') might mean.

508

Strictly speaking then, a result in Artificial Intelligence consists of the isolation of a particular information processing problem, the formulation of a computational theory for it, and a practical demonstration that the algorithm is successful. This way of looking at how work in AI should be evaluated has always impressed me and I think that it applies equally well to work in computational linguistics. I will thus discuss - in a relatively abstract manner - how the goals of research in computational linguistics can be characterized along the lines mentioned in Marr's quote, what some of these goals are and what progress has been made to date in reaching them. From the beginning, modern linguistics has emphasized the role of 'computability1. It was certainly one - if not the greatest - merit of Noam Chomsky in the fifties to have introduced many fundamental notions of computational linguistics, which - as is well known - have turned out to play a role in linguistics and computer science ever since. Above all, Chomsky stressed the concept of 'syntactic grammaticality' as a fundamental notion in linguistics. What Chomsky underestimated and apparently oversaw altogether was that the enumeration of syntactically well-formed structures is only a part of what needs to be computed in linguistics. Not only are we interested in enumerating syntactic structures, we want to enumerate those syntactic structures which have a 'meaning1. It was the philosopher and logician Richard Montague who in the late sixties first provided the basic techniques with which meanings could be correlated to forms in a systematic way.3 As a result, many, but by no means most linguists agree at least on one thing: a central goal in characterizing linguistic objects consists in defining a relation between forms of utterances and meanings of utterances. In any event, practically all linguistic theories that are fashionable these days adhere to this slogan in one form or the other. This relation, which I will call syntax-semantics, applies to pairs , where χ is a syntactic description of an utterance and y a semantic description of the same utterance.4 The syntax-semantics relation is probably the greatest common denominator of theoretical work in linguistics in the past thirty years. Even though this relation has been extremely productive over the years, I feel it is too restrictive both for theoretical and for computational linguistics. A more adequate basic predicate for the description of linguistic communication should pay attention in addition to As both Chomsky's and Montague's achievements in formalizing the respective notions in syntax and semantics are well-known, a detailed discussion is not needed here. Let us point out however that to this day no 'Montague semantics' has been provided for a 'Chomsky syntax' and vice-versa even though of course the programme of Montague semantics has been carried out on the basis of a variety of other kinds of syntactic theories and similarly, but to a much lesser degree, for Chomsky syntax. Usually (but not always) a syntactic description of an utterance is an 'analysis tree'; and a semantic description a logical form'. The exact nature of these depends of course on the details of the syntactic and semantic theories under consideration.

509

syntactic and semantic properties of utterances also to their pragmatic properties, which concern the conditions under which information is transmitted from speakers to hearers in specific contexts.

Moreover, the basic predicate should also pertain to sequences of

utterances just as much as to individual ones, it should also apply to dialogues and it should not be restricted to natural language communication only.5

3.

What is to be computed?

What we want to compute is thus a relation between four arguments, which describes how a linguistic utterance extends or modifies an existing context to a new context, and which linguistic reaction, that is, which answer, is appropriate to this utterance. This predicate is referred to as process in the following. process(Context, Utterance, Answer, Context1) Utterances and answers are sentences or texts (sequences). It remains for us to define what kind of objects contexts could be and, above all, how the relation which we are characterizing by the predicate process is to be defined (i.e. to be axiomatized). I would like to propose the following definition of process: the processing of a sentence changes a context C to a context C' if the sentence can be represented by a logical form and if the incorporation of the logical form into the context leads to the respective change of context. More formally:6 process(C, Sentence, Answer, C') -> represent(Sentence, Logical_Form, Pragmatics) incorporate(C, Logical_Form, Pragmatics, Answer, C'); Since we wish to apply this formulation of process to sequences of sentences, the definition must be applied recursively to the individual sentences of a text or a dialog. This can be achieved by reformulating the above predicate in a recursive manner:7

^ For instance, programmes should fall under this definition as well. 6

The notation Al -> A2... An is to be read alternatively as: Ά1 holds if A2.... A" hold' or as 'if A2.... A" hold, then Al holds too'. The '_' is a linking element.

' The notation A.B is to be understood as referring to a list of objects beginning with the object A followed by the list of objects B. refers to the empty list. The second definition of the process clause thus says that the context doesn't change whenever no utterance is produced.

510

process(C, Sentence.Text, Answer.Text', C1) -> represent(Sentence, Logical_Form, Pragmatics) incorporate(C, Logical_Form, Pragmatics, Answer, C') process(C, Text, Text', C1); processed, ,, Q ->; The definitions of represent and incorporate are to be understood as theories. interaction of the two theories thus derives infinitely many utterances of the form:

The

8

processiC, U, A, N) whereby the arguments C, U, A, and N are names of concrete values. To what extent these theories must be formulated in an integrated or in an autonomous way remains unsolved.9 We can however at least specify what the concepts 'integrated' and 'autonomous' mean in our specification. Theories (and the theoretical concepts defined in them) are 'autonomous1, if no predicate of one theory must be used to define predicates in the other. Otherwise the theories are 'integrated1. The predicate represent in the second line of the definition of process indicates a relation between an input sentence, a logical form and a pragmatic function. This is comparable to the syntax-semantics relation we know from linguistic theories. It functions somewhat differently in our specification, as it is computed only with respect to the predicate process - as a sort of corollary, so to speak. This predicate must be defined primarily with concrete syntactic and semantic theories. This is certainly not the place to go into the pros and cons of the numerous syntactic and semantic theories available today. Of course, we will have to base a more detailed definition of represent on the insights of one or another of these syntactic and semantic theories. I find it equally clear, although this is often disputed, that no existing syntactic or semantic theory is so precise, logically consistent and covers a large enough fragment of a natural language that it could be used in lieu of the predicate represent with no or little reformulation. In fact, the opposite is the case: theories which spell out a relation between sentences, their logical form and their pragmatics still remain to be formulated for larger fragments of natural languages. They do not already exist by any means.10 Furthermore, it is only through algorithmically applicable representations that we can test which relations our theories actually characterize. ° These are the theorems of the (computational linguistic) theory. " In spite of the many pages written on the topic of 'autonomous syntax' in certain circles. 10 Detailed descriptions of many construction have of course been carried out. A significant attempt at large scale descriptions is the work by Maurice Gross and his associates in the context of what they call lexiquegramma ires'.

511

The specification of the relation represent can be divided into four important sub-problems: represent(Sentence, Logical_Form, Pragmatics) -> syntactic_analysis(Sl) leaves(Sl, Sentence) semantk_analysis(Sl,S2) logical_form(S2, Logical_Form, Pragmatics); A sentence is assigned a logical form together with a pragmatic function, if there is a syntactic analysis (an analysis tree) whose leaves are identical with the sentence (the list of words), and if the syntactic analysis can be assigned a semantic analysis, and if this semantic analysis leads to the logical form. I will return to the parameter Pragmatics later. The individual predicates in this specification must fulfill the following: the predicate syntactic_analysis is defined by a set of grammatical rules (a grammar). Its task is to enumerate the set of wellformed syntactic trees of the respective grammatical theory. The predicate leaves defines a relation between these syntactic trees and actual sentences: this predicate is thus essentially a logical specification of a 'parser' or a sentence recognizer. The predicate semantic_analysis translates syntactic trees into semantic trees and the predicate logical_form translates the semantic tree into the intended logical form. Anyone working solely within one grammatical framework, for example Lexical Functional Grammar (LFG), could rely on the exact spelling out of the theory, i.e. the so-called c-Structures (for the syntax) and the f-Structures (for the semantics). Which parsing strategies and which logical forms should be applied cannot be answered by LFG alone. This is not the task of such a theory. This abstract specification of the relation represent requires two further comments: first, a conceptual division between semantic analysis and logical form is made here. It would be possible to relate sentences directly with a logical form, as it is done in many current linguistic approaches. The reason for these two semantic forms of representation is the following: those elements of a sentence which are meaningful in a linguistic sense are both very complex and also not equally relevant for the different contexts in which a sentence must be processed; that is, one and the same semantic analysis can come into play in very different ways according to further processing (in our specification according to the definition of incorporate). The adjustment of the logical form to the respective contextual form is a large - if not the largest - problem, which must be overcome in any systematic analysis of natural language. A simple example to illustrate this problem: no one would disagree that a question such as Where do you live? would only require one analysis at the syntactic and semantic levels respectively (even if this analysis looked very different in the different theories). This question can be answered in many different ways, according to who asked the question and where it was asked: eg. over there; in Main St.; at 50 Main St.; in Manhattan; in New York; in

512

the U.S.; etc. Each of these answers relates to a specific correlation between a logical form and a context. The translation of a semantic analysis into logical form is one method of attending to the intended context of an utterance. A more general solution to the problem is not in sight. The second comment refers to the algorithmic structure of our specification. As we have defined the predicate represent, the specification is not an executable programme, at least not in an efficient way. The intention behind this specification is indeed logical, but if we want to associate sentences with a logical form, then we cannot wait for the predicate syntactic_analysis to enumerate all possible syntactic structures, until it has found a structure whose leaves are identical to the sentence.

The execution of the programme calls

syntactic_analysis and leaves must be 'co-routined', that is, they should not be executed one after the other, but rather concurrently. There are methods which are well suited to this problem.

We can use so-called 'lazy' or 'partial evaluation' methods offered in some

programming languages.

This confirms that we need efficient processes for syntactic

analysis, that is, for the recognition of grammatical sentences in order to come closer to the goal of mechanical analysis of natural language. We could even go a step further: both the syntactic and the semantic structures we assign to sentences do not fulfill any direct goals. They serve to make those interpretations (i.e. logical forms) visible, which are necessary for dynamic information processing (both production and reception). Thus, we should demand that possible interpretations of a sentence be accessible for the calculation of syntactic and semantic sub-analyses as soon as possible. In other words, syntactic and semantic analyses, which do not lead to a 'meaningful* i.e. incorporate interpretation, should be left aside immediately. This recommendation would lead to the full "co-routining1 of all predicates in the definition of process.

It should be noted that this recommendation can only be

implemented in areas where the logical forms and interpretations are indeed efficiently enumerable.

4.

Sub-areas of 'Computational Linguistics'

The definition of the relation represent is much more complex than can be adequately described within the scope of this paper.

This definition embraces those sub-areas of

Linguistics and Computer Science which are directly relevant for Computational Linguistics and which have already demonstrated many important results. The existing theoretical and partly implemented approaches to a computable definition of this relation point to a division of labor which appears in this schematic representation:

where should we obtain our

intuitions and adequacy requirements for syntactic and semantic analyses of sentences if not from our knowledge of existing syntactic and semantic theories in Linguistics?

513 It does not have to be emphasized that mathematical logic must play the same role in the definition of the 'logical forms' as syntactic and semantic theory do in the syntactic and semantic analyses. This is also a part of the tradition of the past decades. If we wish to take advantage of the syntactic and semantic recommendations of linguists transparently and efficiently and actually use them in different applications, syntactic analyses must be translatable into efficient parsing strategies and semantic analyses into (deductively) processable logical forms.

For this reason, it is not surprising that the sub-areas of

Computational Linguistics such as parsing methods and parsing strategies, as well as knowledge representation and automatic theorem proving are focal points of current research programmes and have developed a tremendous impetus which is not to be underestimated. Even though linguists concentrate more often than not on one sub-part of the theory of the relation represent, there is more to compute. All utterances have more than just a logical form (i.e. interpretation), they are made or used to achieve something. This pragmatic or informative function is present in the specification of process in the form of the predicate incorporate. This predicate refers to the effect which an utterance has in a concrete context. Which effects an utterance has - according to its pragmatic function - in a context, plays just as important a role, if not a more important one, in the definition of the basic predicate process as the other components of the specification. We view the predicate incorporate as a relation between a current context, a logical form and its pragmatic function, an answer and a new context.

The simplest cases of providing information which are characterized by

incorporate, are those utterances, whose pragmatic function is to demand an answer, that is interrogative sentences (questions). That there are many different kinds of questions is selfevident (there is ample literature on this topic). incorporate(C,L,P,A,C) -> question(P) prove(C,L,P,R) verbalize(R^); This specification states that a context C remains unchanged, if we are concerned with a question (that is, the pragmatic function of the respective utterance is a question), and if the proof (that is the execution) yields the result R as the utterance's logical form, which is then to be verbalized as the sentence or answer A. The logical form underlying the answer can be proven (or answered) in a semantic or in a deductive way. For the definition of the predicate prove several proof-theoretical techniques are equally well suited. Which one we will use in a concrete formalisation of the predicate incorporate depends essentially on the form of the context parameter, that is, on the concrete application we have in mind. I will return to this issue in the next section.

514

The third predicate in our specification (verbalize) can be regarded as a kind of inverse of the predicate represent. Since the result argument R in prove is also a logical form, we want to generate a sentence from a logical representation. Generating utterances from meaning representations has become a subdiscipline of computational linguistics in recent years. It should also be clear that the characterization of the predicate verbalize is important not only in the context of the process predicate. Two other areas of computational linguistics are concerned here. On the one hand, a version of verbalize (if appropriately parametrized) makes up one-half of the specification of a translation relation. Even if most existing automatic translation systems are more or less syntactically based in that translation precedes from syntactic trees to syntactic trees, it is clear that more adequate translation systems cannot be constructed without including semantic representations as well in the translation relation. On the other hand, and I consider this a still more important consideration, it is very likely that the systems involving mainly generation will have a greater range of applications than systems based on verbalize. How can this be? The reason is quite simple: in man-machine communication there are many practical alternatives to natural language access for the human user (e.g. graphic interfaces, menus, database languages, programming languages, etc.) which can be used depending on the specific purposes in mind. In contrast, if the machine wants to communicate the results of its workings to the human, the options are not as large and not as easy to program. In many applications such as expert systems, intelligent tutorials and the like, there is simply no alternative to natural language as the language of explanation for the machine. In other words, it is more important that the machine communicates in natural language with human users than the other way around. The interaction with existing contexts is not restricted to the asking of questions. Contexts can not only be questioned,11 they can be modified in various ways; they can even be revised. There is thus a second rule for incorporate, which is intended for the treatment of indicative sentences, so-called assertions: incorporate(C,L,PAC') -> assertion (P) prove(C,L,P,R) incorporate(C,R,C) verbalize(R^); A new context C' results from the incorporation of a logical form L in C with the answer A, if R is the result of the proof of L in C (the information value of L in C in other words); C1 is 11

An interesting question to consider is how asking (in addition to answering) questions affects contexts of utterance. Contrary to the clause above, the context changes, even if minimally.

515 an up-date of C if R is appropriately dealt with in C; finally, A is a verbalized 'feedback1 of the transmission of information. The information value of a logical form in a context can take on various forms and these have led to a number of developments that seem to be quite involved from a logical point of view. There is room to briefly mention only a few of the cases that arise: a) the assertion may be totally or partially redundant in the actual context, b) the assertion is independent of the context and c) the assertion contradicts the actual context. The systematic analysis of how new information relates to a given context is the subject of a new area of philosophical logic and artificial intelligent now popular under names like 'nonmonotonic1 logic, 'default reasoning', 'belief-revision', etc. Here, too, it is an open question if algorithmically useful results can be obtained in the near future.

5.

Goals of Computational Linguistics

Many detailed remarks could be added to the general outline of the landscape of computational linguistics.12 I have also said nothing so far about how the notion of context that plays a central role in our predicates should be regarded. Nor did I say anything concerning the relation between these predicates and actual goals of computational linguistics or their concrete implementation in running systems. I would like to address these desiderata by describing three practical goals of computational linguistics. I say 'practical' because many implementations of systems designed with these goals in mind exist already; some even in the form of commercial products. It is a further advantage of the way we have structured the theoretical predicates that it allows a rather natural classification of typical applications in terms of what happens with the context parameters. Besides automatic translation - the first well-known application of computational linguistics - there is an application area which benefits quite directly from all previous work in syntax and semantics: namely, systems for interrogating databases in natural language. Such systems were already conceived in the late sixties and today they belong to the most sophisticated products of research in computational linguistics. Typical database questionanswering systems realize a version of the basic predicate process(C,question_part,answer_part,C) ->....

2 For instance, we have said nothing so far about the place of speech (input or output) in the general picture, about the role of morphological and lexical descriptions, etc. They are implicitly present in the represent predicate, but merit of course their own theoretical predicates in a more comprehensive treatment.

516

where C is an instance of a database. Such systems 'process1 queries expressed in natural language (instead of in an artificial programming language) and yield answers that are 'true' in the database. The dialogues that are involved are typically question-answer pairs, where the answers are, in general, simply term expressions referring to objects in the database in addition to yes and no. What is more important for our classification is that the context parameter C (that is, the database) does not usually change during an interaction with the system. The information in the database remains the same. The computation of question-answer pairs does not seem to be particularly spectacular at first sight. Nevertheless, I consider these systems to be the most successful application of linguistic and computational linguistic methods. These applications13 have a well-founded theoretical basis: formal semantics (both from a model-theoretic and a database point of view) and the corresponding theory of denotation conditions. The type of application allows us to give an answer to the question concerning the nature of the context parameter: the context here is a relational database. A relational database amounts to a collection of relations, and these relations are nothing more than a set of facts. How queries expressed in natural language are evaluated in such a database is the main subject under investigation here. Let us call the retrieving of information from already existing databases a goal and an application of type I.14 If existing databases may systematically be interrogated in natural language, then an obvious goal (type II goals and applications) might be the inverse of this procedure, namely, to construct data bases or (knowledge) bases from natural language input. This goal is much more complex. The main difficulties lie in the determination of the exact form of the initial context and of the target context that is induced by the natural language input text as well as in the systematic treatment of the many context-dependent constructions in natural language. There are only a few systematic attempts to develop a general method for the translation of natural language texts into knowledge bases. One such attempt is the so-called Discourse Representation Theory due to Hans Kamp, where we also have clear answer as to what the source and target contexts in the processing of texts are taken to be. Contrary to the static databases of the type I systems used in database interrogation systems, more dynamic structures are manipulated here.

The elements of these structures (called Discourse

13

There are, of course, also systems aiming at the same purpose that are designed and programmed with very different techniques.

14

1 will return to the question of how such systems may evolve and how they may differ amongst themselves is a question addressed below.

517 Representations Structures in Kamp's theory) are not elementary facts but rather general predicate logic like formulas. The processing of a text transforms an initial context structure C in an extension C which contains the information contained by the text. Thus, a type II theory might be regarded as a version of process(C,Text,OK,C) ->... where OK is simply a list of eventual error messages concerning the parts of the texts that could not be treated in an adequate way. Since the elements of C are no longer facts, semantic operations on C cannot be of the evaluation type as in type I. These context structures require deductive manipulation techniques, in other words, theorem proving techniques for the representation languages in question. Applications of type II are far from being realized, even though in restricted settings some hope for progress is in order. The theoretical and practical developments in type I and type II applications naturally lead to type III systems which are nothing more than a combination of the former two. Genuine dialogue systems would be the result, so it would seem. But we are more than far removed from working systems in type III than in type II if such are the goals. Just as there are ways to extend the complexity of database interrogation systems, there are ways to work on restricted versions of type II and III applications. Database interrogation systems can be extended to systems with deductive databases instead of relational ones; this would exhance the expressive power of such systems considerably. Similarly, the notion of context of such systems could be extended to include not only the database of facts and rules, but also aspects of the dialogue context in which queries are asked. In systems with local contexts of this sort, various kinds of contextual phenomena (e.g. anaphora, ellipsis, and the like) could be dealt with in more depth. Systems of type II and HI can be restricted in many ways and need to be if more than adhoc implementations are envisaged. For instance, database updates are a way of relating systems of type I and II. Here the structure of the context is determined in advance and the theoretic problems are mainly related to the semantics of the up-dates. Such systems could then be scaled up to systems dealing mainly with factual knowledge.

6.

Prolog as the implementation language for Computational Linguistics

I would like to finish this note with a remark concerning the implementation of the predicates discussed earlier.

518

There are several reasons why the programming language Prolog15 is closely related to the tasks mentioned above. First of all, the 'abstract machine1 which underlies Prolog has striking similarities to the mechanism inherent in the process predicate. Secondly, we said earlier that the various subpredicates of represent corresponded to particular theories.

A Prolog

specification of a grammar or a semantic representation language is an ideal combination of a syntactic rule format, on the one hand, and a logical theory on the other. A Prolog grammar is in fact an axiomatized theory of a language. And from the point of view of the underlying operational semantics, the Prolog grammar programme implements a non-deterministic parsing technique. Thirdly, it is, in fact, only a very small step from abstract specifications of top-level predicates like process or represent above to executable Prolog code. There is no other programming language where the connection between specification and execution is so close.

Finally, the expressive power of Prolog consists in the combination of non-

deterministic programme execution and built-in unification procedures. The very first Prolog implementation of man-machine communication systems in the early seventies were the first grammars using unification as a central programming technique. During the last few years newer generations of Prolog have seen the day; in these, the techniques of unification are considerably extended. For fifteen years, unification was little more than simple syntactic equality between Prolog terms; more recently, many specific theories have been 'built in1 - so to speak - in the unification mechanism. As a result, it has become more straightforward to encode the details of linguistic knowledge about strings, logical forms and contexts than ever before.

Bibliography Kowalski, B. (1979): "Logic for Problem Solving." - In: Artificial Intelligence 7. Marr, D. (1977): "Artificial Intelligence - A Personal View." - In: Artificial Intelligence 9,37-48.

5 In various projects at the SNS in Tübingen and the CIS in München systems of type I and II have been efficiently implemented in the programming language Prolog.

519

List of Contributors Christmann, Hans-Helmut Romanisches Seminar Universität Tübingen Czepluch, Hartmut Englisches Seminar Universität Göttingen de Brabandt, Anja Fachbereich Sprach- und Literaturwissenschaft Angl isti k/Ameri kanistik Universität Osnabrück Drubig, H. Bernhard Seminar für Englische Philologie Universität Tübingen Fritz, Gerd Deutsches Seminar Universität Tübingen Gottschalk, Klaus-Dieter Seminar für Englische Philologie Universität Tübingen Grannis, Oliver Fachbereich Sprach- und Literaturwissenschaft Anglistik/Amerikanistik Universität Osnabrück Guenthner, Franz Zentrum für Informations- und Sprachverarbeitung Universität München James, Allan R. Engels Seminarium Universiteit van Amsterdam Householder, Fred W. Department of Linguistics Indiana University

520

König, Ekkehard Institut für Englische Philologie Freie Universität Berlin Lattey, Elsa and Moeck, Juliane Seminar für Englische Philologie Universität Tübingen Le Page, Robert B. Department of Language and Linguistic Science University of York Mills, Anne E. Instituut voor Algemeine Taalwetenschap Universiteit van Amsterdam Mönnich, Uwe Seminar für Englische Philologie Universität Tübingen Nemeth-Newhauser, Andrea and Newhauser, Richard Department of English Trinity University, San Antonio Newmark, Leonard Department of Linguistics University of California, San Diego Pullum, Geoffrey, K. Cowell College University of California, Santa Cruz Radford, Andrew Department of Language and Linguistics University of Essex Reis, Marga Deutsches Seminar Universität Tübingen Russ, Charles V.J. Department of Language and Linguistic Science University of York

521 Schane, Sanford A. Department of Linguistics University of California, San Diego Simpson, Adrian Seminar für Sprachwissenschaft Universität Tübingen Stubbs, Michael W. Fachbereich Anglistik Universität Trier Walmsley, John Fakultät für Linguistik und Literaturwissenschaft Universität Bielefeld Weber, Heinrich Deutsches Seminar Universität Tübingen Werth, Paul N. Engels Seminarium Universiteit van Amsterdam Westney, Paul Seminar für Englische Philologie Universität Tübingen