Germanic dialects : linguistic and philological investigations 9789027235268, 9027235260

Table of contents :
LANGUAGE AND DISCOURSE: TEST AND PROTEST A Festschrift for Petr Sgall
Editorial page
Title page
Copyright page
PREFACE
Table of Contents
I. SEMIOTICS AND SEMANTICS
"PSG" PAS SI GRAVE
0. INTRODUCTION
1. SIGN AND INTERPRETATION
2. PROBLEMS OF INTERPRETATION
3. POETRY AS THE ART OF THE IMPOSSIBLE
4. THE TURNING OF THE SCREW
5. THE DOUBLE ALIENATION OF THE SIGN: "JE EST UN AUTRE"
6. IMPLICATIONS FOR A LINGUISTIC DESCRIPTION
NOTES
REFERENCES
HOW MANY LEVELS SHOULD A GRAMMAR RECOGNIZE?
ENTAILMENT DESCRIBED IN ORDINARY PROSE
A PROBLEM WITH SEMANTIC INTERPRETIVE RULES A PARTICULAR CASE OF RECIPROCALS
ZUM VERHALTNIS VON LEXEM UND PARAPHRASE
LITERATUR
II. THE SENTENCE AND ITS STRUCTURE
THE DEPTH OF DEEP STRUCTURE
REFERENCES
LA VALENCE: SÉMANTIQUE ET SYNTAXE
NOTES
BIBLIOGRAPHIE
CONTROL IN COMPETING FRAMEWORKS
2. EXTENSIONS AND REFINEMENT OF TIC AND TDC
2.1. Thematic Identity Condition
2.2. The Thematic Distinctness Condition applied to ASK
FOOTNOTE 1
REFERENCES
THE CZECH INFINITIVE IN THE FUNCTIONS OF OBJECTIVE AND THE RULES OF COREFERENCE
NOTES
REFERENCES
ERGATIVITY IN DYIRBAL
REFERENCES
FOUR (SIMPLE) REMARKS ON COORDINATION
REFERENCES
III. BELOW THE SENTENCE STRUCTURE
A NOTE ON THE 'AORIST'
FOOTNOTES
REFERENCES
BEDINGUNGEN FUR DIE AKTUALGENESE DEUTSCHER NOMINALKOMPOSITA
BIBLIOGRAPHIE.
A CURIOUS CASE OF PHONEMIC SUBSTITUTION
REFERENCES
A FORMAL APPROACH TO ERROR TAXONOMY
1. INTRODUCTION
2. ERROR TAXONOMY
3. STRUCTURAL ERRORS.
4. RANK HIERARCHY
5. FINAL REMARKS.
REFERENCES
IV. TOPIC AND FOCUS
LINEARIZATION, TEXT TYPE, AND PARAMETER WEIGHTING
NOTE
REFERENCES
THE "QUESTION TEST" RE-EXAMINED
FOOTNOTES
REFERENCES
FOCUS AND MODALITY
REFERENCES
A NOTE ON THE ORDER OF CONSTITUENTS IN RELATION TO THE PRINCIPLES OF GB THEORY
NOTES
REFERENCES
V. TEXT AND CONTEXT
KONTEXT UND MÖGLICHE WELT (EINE UNTERSUCHUNG DER INDIREKTEN REDE)
ANMERKUNGEN
LITERATUR
QUESTION-ANSWER CORRESPONDENCE AND THE SEMANTICS OF QUESTIONS
REFERENCES
A PURPORTED THEORY OF RELEVANCE
INTRODUCTION
THE RELATION TO 'MUTUAL KNOWLEDGE' ACCOUNTS
CONTEXTUAL IMPLICATIONS: THE MAD-PASSER-BY EXAMPLE
ALTERNATIVE POINTS OF VIEW AND BELIEF SETS
DEGREES OF RELEVANCE
THE THALASSEMIA EXAMPLE
LOCATING THE CONTEXT
COGNITIVE SOLIPSISM
RELEVANCE AND CONVERSATION
BACK TO THE FIRST PRINCIPLE: THE "FULL" PRINCIPLE OF RELEVANCE
CONCLUSION
REFERENCES
BEDEUTUNG UND FUNKTION VON GEGENFRAGEN IM DIALOG
ANMERKUNGEN
ARTICLES, WORD ORDER AND RESOURCE CONTROL HYPOTHESIS
ABSTRACT
ARTICLES VERSUS WORD ORDER
ARTICLES
WORD ORDER
RESOURCE CONTROL HYPOTHESIS
CONTROLLING THE DEPTH OF PROCESSING
ENGLISH EXAMPLES
POLISH EXAMPLES
CONCLUSIONS
REFERENCES
CODE-SWITCHING IN COLLOQUIAL CZECH
NOTES
REFERENCES
VI. FORMAL AND COMPUTATIONAL METHODS
ARRAY LOGIC FOR SYNTACTIC PRODUCTION PROCESSORS
INTRODUCTION
1. CF-GRAMMARS AS NETWORKS OF SEQUENTIAL ENCODERS AND DECODERS
2. PROGRAMMABLE ARRAY LOGIC
3. AN EXAMPLE: THE PROGRAMMING OF AN ARRAY FROM THE TRANSITION TABLE OF A SEQUENTIAL FINITE STATE MACHINE.
4. CF-GRAMMARS AND PUSHDOWN STORE AUTOMATA
5. TRANSITION TABLES FROM CF-GRAMMARS
REFERENCES
MATHEMATISCHE BETRACHTUNGEN ZUR ZUVERLÄSSIGKEIT VON VERFAHREN
1. VORBEMERKUNGEN
2. DEFINITIONEN UND EINIGE ZUSAMMENHANGE
3. EINIGE STATISTISCHE BERECHNUNGEN
COMPUTERUNTERSTUTZTE UNTERSUCHUNG DER GRAPHEM- UND PHONEMEBENE
ANMERKUNGEN
SEMIOTIC ASPECTS OF MACHINE TRANSLATION
1. GENERAL SIGN THEORY
2. ESTHETIC SEMIOTICS
REFERENCES
SOME IDEAS ABOUT DENSITY IN KNOWLEDGE BASES
I. INTRODUCTION
2. SOME EXAMPLES
3. SEMANTICS
4. REPRESENTATION
5. ONTOLOGY
6. PRAGMATICS
7. SOME HEURISTICS
8. ACKNOWLEDGEMENT
REFERENCES
INVESTIGATING HUMAN FACTORS IN NATURAL LANGUAGE DATA BASE QUERY
1. INTRODUCTION
2. METHODS
2.1 SIMULATION
2.2 COMPARISON
2.2.1 Language: Ogden's tables
2.2.2 Language and task: Chauffeurs
2.2.3 Paper and pencil studies
2.3 USER STUDIES
3. QUESTIONS
3.1 LANGUAGE USE
3.2 ERRORS
3.3
COMPLEXITY
4. INTERPRETING THE RESULTS
4.1 SEMANTICS VS. CONTENT
4.2 SYNTAX VS. SEMANTICS
5. SUMMARY
REFERENCES
POSTFACE
15 ANNÉES DE COOPÉRATION AVEC LE GROUPE DE LINGUISTIQUE ALGÉBRIQUE

Citation preview

LANGUAGE AND DISCOURSE: TEST AND PROTEST

LINGUISTIC & LITERARY STUDIES IN EASTERN EUROPE (LLSEE) The emphasis of this scholarly series is on recent developments in Linguistic and Literary Research in Eastern Europe; it includes analysis, translations and syntheses of current research as well as studies in the history of linguistic and literary scholarship.

Founding Editor: John Odmark General Editor: Philip A. Luelsdorff

Volume 19

Jacob L. Mey (ed.) Language and Discourse: Test and Protest A Festschrift for Petr Sgall

LANGUAGE AND DISCOURSE: TEST AND PROTEST A Festschrift for Petr Sgall

edited by

Jacob L. Mey

JOHN BENJAMINS PUBLISHING COMPANY AMSTERDAM/PHILADELPHIA 1986

Library of Congress Cataloging in Publication Data Language and discourse. (Linguistic & literary studies in Eastern Europe (LLSEE), ISSN 0165-7712; v. 19) Bibliography: p. 1. Linguistics. 2. Sgall, Petr, 1926. I. Mey, Jacob. II. Sgall, Petr, 1926. III. Series. P26.S43L36 1986 410 86-6882 ISBN 90 272 3526 0 (alk. paper) © Copyright 1986 - John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher.

PREFACE

PREFACE

At the time of publication of this volume, Petr Sgall will be the youngest sexagenarian among European linguists. Petr was born on May 27, 1926, and has recently authored two new books, while two others are ready to go into print. It may, therefore, seem somewhat Festschrift

premature to publish a

in his honor; however, as evidenced by the

number of contributors to this volume, many of his colleagues do not share this view, and have decided to offer him their homage, long before his productivity starts declining (if it ever does!). As one who had seen his world break down, and his father and many of his relatives perish in the Nazi Holo­ caust, young Petr might easily have yielded to the then current illusions about an easy rebuilding of a so much better world. However, his experiences have brought him to recognize the importance of obeying, and even transcending, the wisdom of the old adage Audiatur to accept things only cum grano

salis,

et

altera

pars;

and

if at all.

In the beginning of his career, Petr had as his teachers people from the old Prague School. He soon began to protest against their theories of poetic language, as well as against some other structuralist tenets. Also, when he studied Indo-European linguistics (with KuryXowicz and others), soon he started protesting against the one-

viii

PREFACE

-sidedness of the diachronic view. In his sociolinguistic work, Petr protested against the official attitudes towards the culture of language, as embodied in the views of many Czech linguists. In text linguistics, he protested against the unwarranted parallelism of treating the structure of discourse as if it were similar to that of the sentence. When Chomskyan linguistics made its appear­ ance on the scene, Petr saw it as his duty to raise his voice in protest against the notion of non-relationally based transformations and against the neglect of semant­ ics; also, rather than succumb to a one-sided influence, he called for mutual respect and understanding, asking for linguistic ideas to be allowed to cross the Atlantic ocean in both directions. In the same vein, Petr has attempted to convince computer people that they may have some use for linguist­ ics, and linguists that an active interest in computers and computation will widen the perspective of their own science. He has tried to persuade European structuralist linguists that the generative attempts at formalization are not the devil's own work, and to convince Barbara Partee that variables and parentheses belong in logic, rather than in linguistic representations, and finally to have Chomsky accept the view that it was a pity to abandon Cartesian Linguistics altogether! Petr has formulated the principles of a relatively economical type of linguistic description, by postulating and working out a language-specific level of meaning that includes a detailed representation of topic and focus. In addition, he has founded a research team which he infected with his own spirit of workoholism to make them become enthusiastically engaged in formal grammar, as well as in

PREFACE

iX

language comprehension and other areas of computational linguistics. Basically always having been a heretic, Petr has not been prevented by this attitude from broadening the scope of his interests. Far from it: his concerns range from linguistic typology to theoretical semantics, from sentence structure to orthography, and from mathematical methods in linguistics to psycholinguistics, not to speak of possible uses of computers in linguistics for a variety of purposes. The present volume bears testimony to this multifar­ ious and variegated character of Petr's background and current activities. If it may strike someone as being too broad in its orientation, it is only to be piously wished for that this kind of variety will contribute to broader and deeper understanding between linguists of various backgrounds, and bring together different disciplines that share, as their common orientation, the study of cognition and communication. Finally, as the grateful Editor of this volume, I would like to express my sincere thanks to Eva Hajičová for her invaluable help of all kinds.

Jacob Mey

TABLE OF CONTENTS

PREFACE

vii

I. SEMIOTICS AND SEMANTICS Jacob Mey "PSG" Pas si grave David G. Hays How Many Levels Should a Grammar Recognize?

25

Andrzej BogusXawski Entailment Described in Ordinary Prose

31

Irena Bellert A Problem with Semantic Interpretive Rules. A Particular Case of Reciprocals

43

3

Gert Jager Zum Verhältnis von Lexem und Paraphrase

57

II. THE SENTENCE AND ITS STRUCTURE Olga Mišeska Tomić The Depth of Deep Structure Victor Yu. Rosentsveig La valence: sémantique et syntaxe Rudolf Rủzička Control in Competing Frameworks

101

Jarmila Panevová The Czech Infinitive in the Functions of Objective and the Rules of Coreference

123

75 95

xii

TABLE OF CONTENTS

Milan Bílý and Thore Pettersson Ergativity in Dyirbal

143

Petr Piťha Four (simple) Remarks on Coordination

163

III. BELOW THE SENTENCE STRUCTURE Hartmut Haberland A Note on the 'Aorist' Herbert E. Brekle Bedingungen fur die Aktualgenese deutscher Nominalkomposita Josef Vachek A Curious Case of Phonemic Substitution

173 185 205

Wojciech Buszkowski and Philip Luelsdorff A Formal Approach to Error Taxonomy

217

IV. TOPIC AND FOCUS Nils Erik Enkvist Linearization, Text Type, and Parameter Weighting Frantisek Danes The "Question Test" Re-examined Ferenc Kiefer Focus and Modality Eva Hajičová A Note on the Order of Constituents in Relation to the Principles of GB Theory

245 261 287

313

V. TEXT AND CONTEXT Anita Steube Kontext und mogliche Welt (Eine Untersuchung der indirekten Rede)

327

Elena V. Paduceva Question-Answer Correspondence and the Semantics of Questions

373

TABLE OF CONTENTS

xiii

Yorick Wilks and Chris Cunningham A Purported Theory of Relevance

383

Rudi Conrad Bedeutung und Funktion von Gegenfragen im Dialog

419

Janusz S. Bien Articles, Word Order and Resource Control Hypothesis

433

Louise B. Hammer Code-Switching in Colloquial Czech

455

VI.

FORMAL AND COMPUTATIONAL METHODS

Helmut Schnelle Array Logic for Syntactic Production Processors

477

Jürgen Kunze Mathematische Betrachtungen zur Zuverlassigkeit von Verfahren

513

Ilpo Tapani Piirainen Computerunterstutze Untersuchung der Graphemund Phonemebene

541

Paul L. Garvin Semiotic Aspects of Machine Translation

555

Walter v. Hahn Some Ideas about Density in Knowledge Bases

565

Magdalena Zoeppritz Investigating Human Factors in Natural Language Data Base Query

585

POSTFACE Patrice Pognan 15 années de coopération avec le groupe de linguistique algébrique

609

I. SEMIOTICS AND SEMANTICS

JACOB MEY Odense

"PSG" PAS SI GRAVE Hommage léger á un, pour qui la soixantaine n'est qu 'une certaine cardinalité passee en passant mais pas sans passion et PASSU

GRAVI

"PSG"

"O passi graviora, dabit deus his quoque finem" (Vergil, Aeneid

I:199)

0. INTRODUCTION One of the terms that occur most frequently in the works of Petr Sgall is that of level. As a linguist and as a scientific worker, but also as a philosopher in the pre-Socratic tradition, one of his main standing worries is how to account for, not just save, the pheno-

4

JACOB MEY

mena in the Pythagorean expression) . A prime tool in this work of accounting is the concept of level, as a representation of what is there, but cannot always be straightforwardly accounted for in terms of representative and (indeed) countable linguistic units. The notion of level thus creates a multi-dimensional perspective on the linguistic real­ ity: levels create depths, even though perhaps the exact nature of what has come to be called, by a sweeping generalization, 'the' deep structure , is something not all are agreed on, and not every theory 2 can provide a satisfactory definition of. What most level-based theories do have an agree­ ment about, however, is that their levels sooner or later can, and have to, be reduced to certain repres­ entational primes, be they of the more common (e.g. morphophonemical) kind, or of a more abstract nature, such as the levels postulated in 'Functional Generat­ ive Grammar 1 , the model of linguistic description and explanation so skilfully propounded and vigorously developed by Peter Sgall and the group of linguists inspired by his ideas. The question I want to raise in the present paper connects in various ways with this idea of represent­ ation and representational levels. However, the primes I want to focus on are not of the purely traditional, linguistic kind. In a way, it would perhaps be justif­ ied to include them under the heading pragmatics (since they have a lot to do with the people who use them); on the other hand, the term prime itself, used

"PSG" PAS SI GRAVE

5

in this connection, may appear to be too restrictive, given its connotations of unique divisibility (by the identity and by itself), and some of the other more general features characterizing mathematical primes (such as their undecidability: for one thing, primes cannot be enumerated and thus, in a very precise, but narrow sense, cannot be accounted for). What, then, as a linguist, can one say about those further levels which traditionally are considered, and named, 'extra-linguistic', i.e. outside of linguistics? If we choose to use the term 'pragmatic' here, we should be careful not to use it in ways already de­ cried by Goethe: "Where'er concepts fail, a word springs quickly to the mind". And let's not forget, either, that pragmatics has indeed been called a practical waste-basket, no more: to inspect the contents of such a basket may be reminiscent of grave-digging! Or, to use another picture, venturing upon a quest for those arcane beasts of the extralinguistic realm could end up as a conjurer's trick, whereby certain items are produced from the depth of a pragmatic hat, where they never were in the first place. To keep our feet on the ground, or to maintain at least one of them within the confines of the linguist­ ically relevant, it is important that we see what is there, on the pragmatic level. Being linguistic in nature, pragmatic items have a meaning; that part of the sign which we can directly see or hear, is what Saussure has called the signifiant. On the level of the linguistic sign, we must insist on this 'letter'; however, taking the signifiant seriously (in accordance

6

JACOB MEY

with Lacan's advice: literally ("a la lettre"; Lacan 1966a: 495)) means that we must insist also on its transliteral, extralinguistic properties: 'trans-literating', so to speak, its possible or realized signi­ fies in what properly may be called a t r a n s c e n d e n t a l , cognitive operation: "l'instance de la lettre" urges the letter of the signifiant to render its transcend­ ental, 'extra-linguistic' content. 1. SIGN AND INTERPRETATION As signs, linguistic items are there to be inter­ preted: "signum ad interpretandum", the School would say. However, just as wrong it would be to take the sign by itself as the primordial, original and static unit, as 'the primitive prime', of signification, just as difficult will it turn out to realize an interpret­ re­ ation of the sign as a non-isolated element, as presenting a major, transcendental significance. Yet it is the transcendence which gives life to the individ­ ual sign, subsuming it under its interpretation, and allowing for its 'triangulation', to borrow Deleuze & Guattari's expression (1972: 61-2). Any interpretation which goes beyond the strictly ('technical') linguist­ ic level will have to face this problem sooner or lat­ er. As Foucault puts it, "the life of interpretation is ... to believe that there are only interpretations" (1966: 312, my emphasis); however, no interpretation is possible without assuming that there is something to interpret, something which is stable, primary, original - even though such a stability may come close to an original state which is nothing but immobility,

"PSG" PAS SI GRAVE

7

carrying with it the death of interpretation, "la mort de 1'interpretation", in Foucault's phrase. Traditionally, the former treatment of the ling­ uistic sign has been assigned to the unique domain of linguistics proper, while the latter has been thought of as belonging to the study of literature. However, as already Hjelmslev pointed out, the sign is one, and linguistics cannot be separated from semiotics - in fact, linguistics is semiotics, and a true semiotics has to be linguistic (Hjelmslev 1959: 38). 2. PROBLEMS OF INTERPRETATION Literature is not alone in having problems with interpreting the sign: neither literature nor linguist­ ics can do without a higher-level interpretation, a 'supreme interpretant'. A possible explanation might be that, at some point in the interpretational hier­ archy, both disciplines obey the laws of a higher authority (some, mainly linguists and psychoanalysts, prefer to use the other polarity of the metaphorical dimension, viz. 'deeper'; but that need not concern us here). The question is: Who is that authority, and what is its role? Two concepts will be useful: (a) the original 'sender' of the text (as opposed to its immediate, visible or audible origin); (b) the inevitability of the text as text. As to the first, the question is: What does the text say that is not directly conditioned by its own significance? (The text as exceeding its own context). As to the latter, we must ask: What is it that constrains the text to meaning exactly this, and nothing else? (The text as bound up in its own context).

8

JACOB MEY

Clearly, the first aspect is the transcendental, the second the immanent. In the first, we stress the contingent character of the text's necessity: a text is always necessarily someone (or something) else's. The second aspect emphasizes the necessity of the text's c o n t i n g e n c y : a text is always just this text, and has to be read and interpreted as such. It is this latter character of the text, and of text production, that has fascinated so many linguists, as well as a good number of literary theoreticians (e.g. those of the 'New Criticism'). However, the former character, too, deserves to be explored, especially in a linguistic ambience; the recent interest shown, by a number of linguistics, in Freud's work, as well as many psychoanalysts' pre­ occupation with linguistics (Lacan and his followers are the most outstanding example) are symptomatic, and should be taken as having its own significance. Just as Wittgenstein has taught us that the sense of the world lies outside the world itself, so the sense of the text, its true (or 'deeper') significance, lies necessarily outside the text. It is the latter necessity (the first aspect highlighted above) that I want to explore here, be it ever so superficially; shallow re­ marks on a deeper level, as it were. 3. POETRY AS THE ART OF THE IMPOSSIBLE The Danish poet Per Højholt has once remarked (1969) that, "if politics is the art of the possible, then art is the politics of the impossible". Now, what do we mean by the word 'possible' (or

"PSG" PAS SI GRAVE

9

'impossible1, for that matter)? First of all, one cannot deal with possibilities except in general, in the abstract mode. As the late Thomas Ballmer used to say: "Impossible worlds? No way!" ("Unmőgliche Welten? Nee, das gibt's nicht"; see Mey 1978: 979) - which, on reflection, is true for possible worlds as well. There is an equivocation here whose result is the above quoted paradox: that which is possible, is not, by definition; however, the impossible cannot he defin­ ed, except by way of the existing, possible set of objects. So, we cannot define the impossible except by force of abstraction: in the moment of truth, when the so-called impossible turns out to have happened all the same, that becomes reality. We can define the impossible essentially (like in a classical, semantic distinctive feature analysis of the Katz-Fodor type (e.g. 1963)), but we cannot ever pronounce ourselves existentially about it, except when it does happen, which is when it isn't impossible any longer. Hence, the impossible is existentially possible, while essent­ ially impossible. This paradox has a dual: the possible cannot be defined except by abstraction from the existing: "de esse ad posse valet illatio", the Scholastic adage tells us. But what is abstractly (i.e. essentially) defined as possible, is existentially impossible: when the moment of truth arrives, the possible world 'becomes' the real, existing, one, and is no longer (merely) possible. The perfect parallel here is the case of Utopia: in the moment of its coming into exist-

10

JACOB MEY

ence, it is no longer the 'real thing', whereas the really existing {"real existierende") object of our dreams is not at all like the possible world that was conceived of in our optimistic philosophy. A rather enlightening (and for linguists, perhaps, amusing) illustration of the above paradox can be found in the case of Chomsky's classical notion of the (semantically) 'ill-formed sentence'. Where Chomsky first got the idea of well-formedness from was, of course, formal logic. Here, an expression can be defin­ ed purely in terms of the formal and the essential: by combining properties that are (formally) non-combinable, or (essentially) mutually exclusive, such as 'possess­ ing a certain color' (e.g. 'green') and 'not possessing any color at all' (hence 'colorless), or 'characterized by quiet' (e.g. 'sleep') and 'behaving in a non-quiet manner' (e. g. 'furiously'), and so on, Chomsky produc­ ed what, at the time, seemed like an impossible sent­ ence: impossibility here defined on the respective essences of its terms. Yet, when people started using this famous sentence ("Colorless green ideas sleep furiously")and others of the same ilk, not only to quote it in the Chomskyan spirit, but for purposes quite alien to its inventor's mind, such as poetic effect (appealing to its metaphoric beauty), or perhaps just in order to show off one's familiarity with the in-trends in one's science, the impossible became the real, existing; by the same token it stopped existing as the impossible and became existentially possible. (See Chomsky 1957: 15). Conversely, once one wants to realize the possible,

"PSG" PAS SI GRAVE

11

one has to decide on one particular possibility among the many given (in our grammars or in our imaginations). Accepting the necessity of such a choice restricts us, but at the same time it is the only way towards our expansion and growth, both as language users and as humans. The person who forever keeps hovering on the brink of the not-realized, the possible, will discover that out of all those possible pipe-dreams only one can be existential - and that if one waits too long, that possibility's chances are likely to decrease with the decline of one's own existence, to be replaced in the final stage, by the universal 'end solution': the im­ possibility of choice, the existential fact of non-existence, Death. However, the possibility of choosing is not with­ out its own serious (and likewise existential) condit­ ions. This holds for our choosing a way of life as for our choice of words and our production of texts. The possible is something made possible by something that is not merely possible, but existing: our life in society and our life as individuals. Our human essence is caught in an existential screw. 4. THE TURNING OF THE SCREW "Wo es war, soll Ich werden": Freud's famous phrase (1933) may give us a clue as to where to look for the conditions of necessity that I referred to in the previous section. In a level-based terminology, the question can be phrased as: At what level of inter­ pretation can we find the significance that is not contained in the text-as-text, but determines the text

12

JACOB MEY

from what is neither a pure outside field of forces, nor a mere internal affair? Let's consider the Freudian expression as to its form. It says something about 'being' and 'becoming' ("war" and "werden"): something was, and becomes some­ thing else: The Ego replacing the Id. Any formula of the type S is/becomes x has two readings: one that I will call the essential, the other the existential, The essential reading stresses the latter part of the formula: S is x; the existent­ ial reading emphasizes its former part: S is x. The difference can be described as follows: When I say S is x, I stress the essential aspect of S's being: true, possible, probable, necessary, etc. This is the reading of our truth-tables, whether we operate within a true-false logic (two-valued), or add more logical values to our tables, allowing (as in deontic logic) for the additional value 'necessary' or 'obligatory', or, as in other types of modical logic, for still further values. By contrast, the reading S is x stresses the existential aspect: I'm interested in how S relates to being true or false (or necessary, etc. whatever). S is seen both as an independent subject, and as one who depends on others to be/become the one he/she is. It is not enough that "es ist"; we want "Ich werden" as well. Or, to take an analogy from art: Faced with the Mona Lisa, our proper reaction should not be the common tourist's ("She is just like I've seen her on the pict­ ures": es ist), but instead, we should become engaged

"PSG" PAS SI GRAVE

13

in Leonardo's work, and re-create his art within our4 selves: Ich werden. Some people may feel that the above stresses Freud's dictum to the breaking-point. Still, I want to maintain that the usual interpretation of the sign in linguistics has concentrated on the essential aspect: signs in an abstract world. Linguistics has neglected the existential character of the sign: signs as related to people, to the users of the sign and their world. By no accident of fate, the Freud who coined the famous formula "Wo es war, muss Ich werden", was also the first to read the linguistic signs as something more than what they 'stand for', in any of the garden variet­ ies of reference; and Freud did this precisely by exploring the sign's necessary ties with the deeper levels of the Self, In a pragmatic framework, this Self is considered to be basically dependent on the others: Lacan has given this notion an even deeper perspective by taking in not just the other users of language as such, but also what he calls 'The Other': the Great Signifier, the institutionalized creator of the sign. (See especially 1966: 524). 5. THE DOUBLE ALIENATION OF THE SIGN: "JE EST UN AUTRE" The above quote is from a work by Arthur Rimbaud. 'Une saison en enfer'). Said (1971: 129-130, which is where I found the Rimbaud line) beautifully encapsulat­ es the frustration of self-expression as torn by the split between essence and existence, which has been discussed in the preceding section: "What, as a human

14

JACOB MEY

being I mean when I say "I am me" is not really what I am, but what I formulate myself to be in order to exist historically". This basic alienation, condition­ ed by, and contained in, our language and our use of it (witness the Rimbaud quote), is what will be the topic of the present section. The subject, in producing a text, expresses him­ /herself in the medium of language. Basically, what the text does is to verify the subject's existence for the subject him-/herself, by making him/her into an object: "je" becomes "moi". But furthermore, this "moi", the result of the subject's text-producing activity, is another than, and in fact alien to, the original "je". The price of the signifying symbol is the sacrif­ ice of the signified-as-such: an analogy to what happens in the Saussurean relation between the two parts of the linguistic sign, signifiant and signif'ié. It is at the deeper level of self-understanding that the basic alienation of the language is realized; as Denis Vasse has expressed it, "le sujet, par la paro­ le, s'interdit d'etre 1'interlocuteur du moi pour devenir ce qu'il est, 1'inter-Locuteur de 1'Autre". (Vasse 1983). From 'interlocutor', simple partner in a conversation, to 'inter-locutor', a partner who is thwarted, baffled ('interloque'), caught up in the basic mode of contradictory self-expression under the aegis of the existential Other: an admirable resume of what Said has called "Lacan's version of the univers­ al neurosis of mankind". (1971: 121). Thus, the universal existential question: "Who am I?" can only be answered e x i s t e n t i a l l y , as in the

"PSG" PAS SI GRAVE

15

Vedic formula: Tat tvam asi "Thou art that". That, or i t , the other element, is the alienated definiens of the questioning I; it is not the I itself, nor is it any kind of essential, definitory extract of the I. By contrast, questions such as: "What is it?" ask for a definition, an enumeration of abstract, universaliz­ ing, essential features; the answer to such a question is a positive (or, as the case may be, negative) inform­ ation on the presence or absence of such 'distinctive features' (as linguists are wont to call them): "It is it"; the bottom line carries the right figure, Mona Lisa has the right smile, "just as in the picture". The following schematic representation may serve to illustrate what just has been said. (See Fig. 1 ) .

Fig.

1

In Fig, 1, existential questions are asked to the left, essential ones to the right. However, there is the possibility of 'crossing over': giving an essential answer to an existential question (and vice versa). Most of the questions asked by philosophers, linguists, psycho­ logists, and other cognitive scientists are of the essential, descriptive kind. Even if they ask the exist­ ential question: "Who am I?", such a question is real­ ly not too different from the essential one: "What am

16

JACOB MEY

I?", and receives an essential answer, confirming the queried What? by a deserved tautology, as in the philo­ sophical "I am I". However, an existential answer to an essential question may also be given, and this answer incorporates the affirmation of the Other in both the querying instances: "I am It", Ego's alienating partner, 5 or as Rimbaud has it, "Je EST un autre". Ultimately, then, the significatory function of linguistic units depends not so much on their essential identity with, or conformity to, a descriptive norm. Rather, it is their existential dependence on what is being said, in the concrete world, that furnishes the context of human discourse. The text is not so much a representation, as it is an interpretation of that world: the sign is first of all a symbol. Or, if one wants to call it that, the sign is primarily a pragmat­ ic, not a semantic entity. 6. IMPLICATIONS FOR A LINGUISTIC DESCRIPTION The sign's character is determined by the role it plays in discourse.

Hopper & Thompson remark that the

'prototypicality' of a noun like fox

does not so much

depend on the fact that it 'denotes'a visible (etc.) entity ...; rather, the important fact is that m u s t the

p l a y

a

c e r t a i n

d i s c o u r s e

f i g u r e s

in

i t

r o 1 e

w h i c h

in it

. Whether this fact is designated to be

a semantic or a pragmatic one is a matter of terminology." (1984: 708). Of this, I agree with the first part: discourse is all-important at all levels of linguistic description (not only the semantic and pragmatic).6 I strongly

"PSG" PAS SI GRAVE

17

take objection, however, to the latter part of the Hopper & Thompson quote: there is much more involved here than a terminological quibble. If one understands, by 'pragmatic', a reference to the user(s) of language, in the widest possible sense of the term 'use', then this reference can be made either to the user in the classical, Chomskyan way (the competent native speaker that everybody knows from Aspects) - in which case the difference between 'semantic' and 'pragmatic' is truly not worth arguing about; or to the user in some extend­ ed sense of the term. But what would such an extension be all about? The user's competence can be extended in two directions, corresponding, if one wishes, to the two levels on which discourse is necessarily and always excercised: on the one hand, the level of society, whose discourse is governed by the 'invisible hand' (as the classical economists called it) of the socio-economic infrastructure; on the other hand, the level of the psyche, whose discourse obeys the unwritten laws of the unconscious, as studied and explicated by psychoanalysis. (Or rather, instead of calling these laws 'unwritten', we should say that they are written in another script, in a different writing: "cette prime écriture" (Foucault 1966: 56), the Writ of the Other). Even though we distinguish between these two levels, we cannot (and should not) separate them; in fact, what linguistic psychoanalysis has revealed as being the "discourse of the Other", to use Lacan's term (1966b: 549), is, in its ultimate reality, the

18

JACOB MEY

voice of our Masters, the rulers of society. It is this latter property which characterizes, in the final analysis, the pragmatic approach: There is level of analysis which is not a priori accessible to the purely linguistic techniques that are practic­ ed in isolation from the use of language. At this level, people say what they do not intend to say: their subconscious speaks for them. Conversely, they cannot say what they want to (or should want to): the power that our society invests in language and its use is restricted to society's powerful. As already St. Paul experienced it, "that which I do I allow not: for what I would, that I do not; but what I hate, that do I For the good that I would do I do not; but the evil which I would not, that I do" (Rom. 8: 15-19) - and why? Because the Law (the 'Voice of the Other', "cette Loi-Langage", in Foucault's (1966: 386) expression) has established this collective bur­ den (which for the Apostle is defined as Sin) upon the individual humans. Salvation (linguistic and otherwise) is neither in a despairing rejection of all language and its use ("Who shall deliver me ...?"; Rom. 7: 24), nor in an overly optimistic view of the Other's language as containing magic formulas for expressing that which I cannot say (" the Spirit itself makes intercession for us with groanings which cannot be uttered." (Ibid. 8: 26)). Thus, whereas 'straight' linguistics may be like­ ned unto politics: viz. the art of the possible, of what can be described and said, using some appropriate technique, pragmatics, on the other hand, is like Art itself: in IIøjholt's expression (see above, section 3 ) ,

"PSG" PAS SI GRAVE

19

the "politics of the impossible". Politics, we recall, is the way we organize our lives in society; it is the realization of our individual being in a societal framework. Or, in linguistic terms: it is the voice of the Ego elevated to the level of social communicat­ ion on the force of the Other's wording; a truly im­ possible task, because of the alienating and repressive aspects of such a verbalization - yet an unavoidable and necessary one. Without it, the I remains forever I, its own mirror image ("I am I", see Fig. 1 ) , never attaining its true objectivization, as in the Tat tvam asi of the Vedic sages: "Thou art That". The differences between a psychoanalytically oriented and a pragmatically founded linguistics can be summarized as follows: Psychoanalysis endeavors to determine human dis­ course as, primarily, "1'instance de la lettre". It takes discourse literally, letting the letter (Saussure's signifiant) guide one's interpretation, as seen in the treatment of neuroses, where slips of the ton­ gue and other parapraxes are considered to contain the key to the particular, individual universe of the patient. In linguistics, on the other hand, we try to pin down the interpretation of discourse to the letter as the bearer of some higher, collective instance, explain­ ing the sense of the communicative act by identyfing what is ultimately responsible for this insistence on the letter: "la lettre de 1'instance". Pragmatic linguistics expands the universe of dis­ course by underscoring the role of the participants as

20

JACOB MEY

not limited to the letter as s u c h , to the papel of the 7 writing, but as determined by the whole scene of the discourse.

That scene is more than a theater in the

regular sense; on the one hand, it is a place where the forces of society meet and measure their mettle; on the other, it is Freud's "anderer Schauplatz", the theater of the Other, where the individual players find themselves and discover whom they really represent and whom they obey, at the deepest level of their groanings:

passi gvave, graviova patientur. NOTES 1 The expression is usually attributed to Eudoxus of Cnidus, pupil of the Pythagorean Archytas. How­ ever, it seems that the latter has a certain right to it as well, through his master Philolaos, who is said to have seen the problem in the first place. See: Jacques Chevalier, Histoire de la pensee, I: La pensee antique (Paris: Flammarion, 1955: 6 1 2 ) . 2 The occasion being a Festschrift, I may per­ haps be permitted to be a bit more personal, and slight­ ly more rambling than is usual in linguistic product­ ions (my own not excluded). Although most American linguists will associate the notion of 'level', first of a l l , with the school of linguistics which carries that concept's name, in its label, v i z . 'Stratificational Grammar', it behooves us to remember that the notion itself is basically a very European one. Scholars such as Tesniére and Hjelmslev had thought along those lines long before any American linguist had. (Not to speak of the Prague tradition, of course, to which the dedicacee of this article so eminently belongs) . It is also interesting to note (and here I am close to being very personal) that Petr Sgall's connections with a number of the most influential American linguists were a product in time of certain happy circumstances, of which I had the pleasure of being the instrument. Thus, at a conference in Magdeburg, GDR, in August of 1964, I had the honor of introducing Petr to David G.

"PSG" PAS SI GRAVE

21

Hays, whom I had met the summer before, during a comput­ ational linguistics seminar at the Rand Corporation, Santa Monica, California. The contacts between Petr and Dave (they struck it off like m a t c h e s ) , proved to be valuable and have lasted unto this very day. Hays' important article on dependency grammar (Hays 1964) was one of the factors that helped clear the way for a bet­ ter understanding of European thought in American ling­ uistic circles, and thus was a welcome counterweight against the dominant pattern of thinking of those days, Early (some call it Vintage) Transformational Grammar. As to Stratificational Grammar, this was viewed - if possible - with even greater distrust by the common American linguist than all those other European-inspir­ ed brands of theories. And in a way, Sidney Lamb was the one to take most obviously advantage of certain ideas, conceived on the other side of the Atlantic. But when I met Sid for the first time (during the same, in retrospect, so portentous, 1963 Summer Seminar), he seemed to be unaware of the existence of Hjelmslev's work. He listened very carefully when I explained to him that Hjelmslev, already decades earlier, had been speculating along level-oriented lines; and how these ideas of Hjelmslev's had come to the fore in his 1954 major article 'La stratification du langage'. Thus, it seems no idle speculation that Lamb's reading of this article and the naming of his theory may have something to do with each other; references to Hjelmslev's works start to appear in the former's writings after 1964. As to Hays, his notions of an alternative model, based on dependency, were a valuable addition to, and inspir­ ation for, the Prague workers, who integrated the best of all the diverse level-oriented theories in their own theoretical framework, incorporating also the earlier work of the Prague school in all its generations. (For a major statement of the whole theory of 'Functional Generative Grammar', as it is now called, see Sgall et al, 1985 - a work which I am proud to have been a kind of godfather figure to, be it on ever so modest a scale.) 3 I owe this quote to Lisbeth, who sat through a first presentation of some of the ideas contained in this paper one afternoon in November of 1983, in the smoke-filled meeting room of an old castle in Jutland (Second Nordic Winter Symposion on Semiotics, Sostrup Slot, November 11-14, 1 9 8 3 ) .

22

JACOB MEY

4 Just like Freud himself did, in his well-known interpretation of Mona Lisa and her enigmatic smile. For Freud, the facial expression on Leonardo's painting is a reflection of the artist's mental split. Mona-Lisa's smile mirrors the dual view that Leonardo held of his mother: contemptible because of her weak­ ness to Leonardo's father, yet sensuous and attractive because of her femininity. It is this blend of subordin­ ation (the Ego under the Super-Ego) and rebellion (the Ego yielding to the Id) which constitutes the particular tension and charm of the painting. It tells us that the two poles of the human being, the sensual and the spir­ itual, never can be united except in a continuing struggle for the establishment of the person. Yet, the sensual and the spiritual cannot be realized existentially in a static mode: to be, but only in a dynamic one: to become; and the reason is that they essentially contradict themselves. (Freud 1910; see also Olsen & Køppe 1981: 2 0 6 f ) . 5 Or, in Lacan's words: "... je ne suis p a s , là où je suis le jouet de ma pensée; je pense à ce que je suis, là où je ne pense pas penser" (1966a: 517; in connection with Descartes's famous Cogitó) . Cf. also Foucault: "... comment peut-il se faire que l'homme pense ce qu'il ne pense p a s , ..." (1966: 3 3 4 ) . 6 Witness also the old name for 'word-classes' (a category strictly subsuming grammatical objects, such as nouns, verbs, e t c . ) : partes orationis (Quint i l i a n ) , translated into French as les parties du dis-

cours,

into English as parts

of

speech.

The formal

appellation in fact supports Hopper & Thompson's point, even apart from all discussions on semantics vs. pragmatics, 7 Recall that Spanish papel stands for both 'paper' and the writing on the paper, as directive of the player's acting on a scene: a 'role'.

REFERENCES Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton, Deleuze, Gilles & Félix Guattari, 1972. L'Anti-OEdipe. Capitalisme et schizophrénic. Paris: Minuit.

"PSG" PAS SI GRAVE

23

Freud, Sigmund. 1910. "Eine Kindheitserinnerung des Leo­ nardo da Vinci". Gesammelte Werke 8. 128-211. Freud, Sigmund. 1933. "Neue Folge der Vorlesungen zur Einftührung in die Psychoanalyse". Gesammelte Werke 15. Foucalt, Michel, 1966. Les mots et les choses. Paris: Gallimard. Hays, David G. 1964, "Dependency theory: A formalism and some applications". Language 4. 511-525. Hjelmslev, Louis. 1953. Prolegomena to a theory of language. Bloomington, Ind.: University of Indiana Press. [1943 ] . Hjelmslev, Louis, 1959. "La stratification du langage". Essais Linguistiques by Louis Hjelmslev. 36-68. Copenhague: Munksgaard. [1954]. Hopper, Paul J. & Sandra A. Thompson. 1984. "The dis­ course basis for lexical categories in universal grammar". Language 60. 703-752. Højholt, Per. 1969. "At least 18 theses on art and poetry - at least 18!" Mak 3. (In D a n i s h ) . Katz, Jerrold J. & Jerry A. Fodor. 1963. "The structure of a semantic theory". Language 39. 170-210. Lacan, Jacques. 1966a. "L'instance de la lettre dans 1'inconscient, ou la raison depuis Freud". Ecrits by Jacques Lacan. 493-528. Paris: Seuil.[l957] Lacan, Jacques. 1966b. "D'une question preliminaire à tout traitement possible de la psychose". Ecrits by Jacques Lacan. 531-583. Paris: Seuil [1958] . Lamb, Sidney. 1966. Outline of stratificational gram­ mar. Washington, D.C.: Georgetown University Press. Mey, Jacob L. 1978. Review of: Sigfried J. Schmidt, ed., Pragmatik/Pragmatics II: Zur Grundlegung einer expliziten Pragmatik. Munchen: Fink 1976. Language 54. 977-982. Olsen, Ole & Simo Køppe. 1981. Freuds psykoanalyse. K^benhavn: Gyldendal. (In D a n i s h ) . Said, Edward W. 1971. "Linguistics and the archeology of mind". International Philosophical Quarterly 11. 104-134, Sgall, Petr, Eva Hajičová & Jarmila Panevová, 1985. The meaning of the sentence in its semantic and pragmatic aspects. Dordrecht: Reidel/Prague: Academia. Vasse, Denis. 1983. Le poids du rēel, la souffranee. Paris: Seuil,

DAVID G. HAYS New York

HOW MANY LEVELS SHOULD A GRAMMAR RECOGNIZE?

In Prague some years ago, during the discussion period following a lecture, Professor Sgall and I arriv­ ed at a statement that I think of as the Sgall-Hays hypothesis of indefinitely discriminable levels between the deepest linguistic forms and the phonological sur­ face. To the best of my knowledge, no one has published this hypothesis, nor made any attempt to find evidence bearing on its validity or usefulness. If I now choose to write about this joint contribution without Professor Sgall as coauthor, it is perhaps because I doubt that he ever took it quite seriously. The general approach of many linguists has been to operate on representations. An operator may replace part of a representation with something or with nothing, or permute parts. A representation may contain segment­ ing brackets or relational links, and the brackets or links may bear labels. Ordinarily the operators respond to the brackets or links, and to their labels, if they are present. As regards the application of operators to represent­ ations, two major policies have been popular. Either the operator applies to a complete representation and

26

DAVID G. HAYS

yields another, or else the operator applies to a pair of representations, one of them complete, and enhances the incomplete member of the pair. The first policy is typical of transformational theories, whereas the second is stratificational in flavor, and I shall henceforth call the policies by the names of the theories that tend to favor them. A transformation is an operator that applies to a complete representation, including labeled brackets or links. It yields a new representation. The prior repres­ entation vanishes with the snows of yesteryear. The logical sequence of transformations flows along, each operator applying to the final result of all preceding operators. The operators of a stratificational theory do not flow so smoothly, but rather surge along. Given some initial representation, operators from a first set apply until a complete second representation appears. Operators of a second set apply to the second representation, collectively yielding a third. And so on to some final stage. The several representations of a stratificational theory may well be said to belong to different levels; and one may say that the initial representation belongs to the deepest level, the final representation to the surface. A linguist may choose to make the vocabularies of the several levels disjoint, and to use a different structure (network, tree, string) for each. If some element of a certain level seems to reappear on the next, a trivial operator can make the formal conversion. The more numerous representations of a transformat-

HOW MANY LEVELS SHOULD A GRAMMAR RECOGNIZE?

27

ional grammar do not belong to as many different levels, surely. Yet nothing abstract prevents us from postul­ ating a level for each operator. Were we to do so, we should find that almost every item passes from level to level unchanged; and we could recognize the transformat­ ional principle that what a transformation does not touch it leaves alone as an abstraction of the stratificationalist's trivial conversion operators. The transformational theory thus appears as a special case of the stratificational system, with one operator per level. We must now ask why a linguist might want to have fewer levels, with more operators to each. The primary reason for grouping operators into levels is the power one gains, making for fewer and simpler operators. Each operator can apply according to characteristics of both the lower representation and the partial higher representation that it will enhance. The characteristics of the lower level indicate what must be represented. The partially formulated characteristics of the higher level constitute the conditioning context. But this argument alone would probably force only small groupings; context does not link together large numbers of operators. A second reason for grouping operators is the covariance of form and function across the operators them­ selves. As an example of a single function, take the introduction of case endings, with perhaps

some pre­

positions. The conditions that select among these operators are much alike, so these operators have rough­ ly the same form. Another function is the establishment of order in accordance with pragmatic values - communic-

28

DAVID G. HAYS

ative dynamism. These operators apply on conditions that are altogether unlike those of case marking, but this group of operators again has a common form. Such con­ trasts and similarities often lead to the segregation of three to six levels. Thus we have a scale on which grammars can be ranked, from the purely transformational type, with hundreds of operators and a level for each, to the purely stratificational type with as many operators but only a few levels. Any number of intermediate types of grammar might be placed between, as the linguist takes into account the mutual influences of the operators and their degree of similarity by form and function. And this is the Sgall-Hays scale of depth discrimination. How should a linguist choose a position on this scale? In fact, most adopt positions near the polar types. Formal criteria such as economy (simplicity) or symmetry might favor intermediate positions. For example, suppose that the operators in a group have a common form. The linguist writes the form once and then tabulates the conditions and effects of the several operators. This style of statement makes for compactness. If this style is acceptable and this compactness is desirable, then intermediate types are probably superior to polar. Taking a broader viewpoint, the linguist can look for correspondences between the grouping of rules and the covariance of syntax with semantics, pragmatics, and phonology. Indeed, those linguists in Prague and else­ where who have worked toward the stratificational pole have emphasized these correspondences. Further, the typologist may ask whether certain

HOW MANY LEVELS SHOULD A GRAMMAR RECOGNIZE?

29

groups of rules identified by form and function are limited to the languages of certain families or regions. The evolutionist may ask whether new forms and functions appear at higher cultural levels. The psycholinguist has another list of questions. Do the groups of rules appear at successive stages of individual growth? Can a whole group of rules be blocked out by experimental manipulations? How do the forms of syntactic rules match with the forms of other cognitive processes such as Piaget's concrete and formal operat­ ions? We live in a world of verbal subtlety and elegance. Many of our contemporaries have recourse to irony, meta­ phor, indirection, shifts of theme and viewpoint, and a great many other tricks of linguistic art. These phenomena are densely packed in literature, but they appear everywhere. Besides psychopathology, there is a poetics of everyday life. Today's linguist is sophistic­ ated enough to understand that the reality of human speech cannot be captured in a grammatical snapshot. The Sgall-Hays scale measures characteristics of language that have been neglected. I think it deserves serious consideration.

ANDRZEJ BOGUSŁAWSKI Warszaw

ENTAILMENT DESCRIBED IN ORDINARY PROSE

The following is a tentative representation of the sense of entailment in simple non-technical terms used in all kinds of talk. It is my intention to be faithful to the usual understanding of "entailment": the formulation is supposed not to narrow or broaden the customary scope of phenomena covered by our term. I take entailment to be equivalent to the relat­ ion of logical or semantic implication or consequence. A technical characterization of it reads: "for any model M, it is not the case that p is satisfied in M and q is not satisfied in M". Another account of our relation runs as follows: entailment is such a case of the falsity of the ante­ cedent of a material implication combined with the negat­ ion of its consequent where the falsity is guaranteed by the very meanings of the expressions involved (the falsity may only be avoided by attributing other mean­ ings to them than those taken of the outset). A variant of this account presents entailment as such a case of falsity where the conjunction of the antecedent and the negation of the consequent is self-contradictory or yields a contradiction. Notice, however, that "yield­ ing a contradiction" apparently involves, again, the

32

ANDRZEJ BOGUSŁAWSKI

relation of entailment itself, the entailment, namely, between the indicated conjunction and some self-contra­ dictory statement, something which removes the definit­ ional solution indefinitely. The last mentioned formulations are likely to evoke the following notion couched in undeniably simple terms: "what is said in q is said in p" or "what is said in q is identical with the whole or a part of what is said in p". This is no doubt an indispensable, per­ haps even fundamental, metalinguistic concept. Still, it is narrower than entailment as normally conceived; therefore, it cannot render the properties of entail­ ment faithfully, as required. For example, we obviously wish our notion of entailment to cover the relation between some Fa as the antecedent and the statement of the truth of Fa as the consequent; but in this case some­ thing is said about Fa(namely, that Fa is a true sent­ ence) which is not said in Fa (in Fa something is said exclusively about a, and not about F a ) . This is not to deny that for the most part examples of entailment can as well serve as examples of the inclusion of what is said in q in what is said in p; thus, "This is a square" entails "This is a rectangle" (provided the same refer­ ence is being made in both sentences), where in saying "a square" we do also say "a rectangle". The formulation of entailment between some entail­ ing p and some entailed q (indeed of what is entailed by "q is entailed by p") which I want to offer reads: (E) it is not the case that both I. someone understands "p, q" and "p, not-q" in the way "p, q" and "p, not-q" are understood here.

ENTAILMENT IN ORDINARY PROSE

33

and II. it is not the case that either 1, he is prepared to say A. about certain things F that F are said in •p, not-q" is a true sentence" B. that a. it is not the case that both he is prepared to say some F and he is not prepared to say certain other things G about the same objects as the object about which that F is said or b, it is not the case that both he is prepared to say some F or some G and he is not pre­ pared to say certain other things H or c. it is not the case that both he is prepared to say some F or some G or some H and he is not prepared to say certain things I about sentences where F or G or H are said or 2. S is such that there are parts of S which are contradictory. Some explanations are necessary, "not-q" stands for the proper negation of q or for the claim of the falsity of q.

"S" stands for all F, G, H and I. "Contra­

dictory" applies to either of the things said in two sentences in their relation to each other iff the two sentences are such that (a) nothing but one of those things is said in either of them, and (b) one of the sentences is a proper negative counterpart of the other. It is easily seen that entailment is construed here as material implication between (a) understanding "p, q" in a certain way and (b) preparedness to say certain things in relation to the notion "true sent-

34

ANDRZEJ BOGUSŁAWSKI

ence" as applied to "p, not-q", things among which there are contradictory things in the most element­ ary (syntactic) sense. In other words, entailment is equated with the absence of the cooccurrence of (a) and the absence of (b). Clearly, our relation E is formulated in read­ ily understandable terms reproducing the non-contro­ versial senses of expressions of everyday life; many of them (including the pivotal ones) are in all likeliness true (and absolute) semantic primitives ("say", "is prepared", negation, "part"); others are their modifications or close derivates (e.g. "understand"). It should be emphasized that the relation E does not require that any speaker who understands two sentences in a given way, such that E holds for them, and who accepts E for them must be ready to produce on demand all the corresponding sentences referred to in the formulation or even the necessary contra­ dictory sentences. All our formulation requires is that he concedes that such sentences exist. Of course, the proof of his claim can only consist in producing particular sentences and showing the preparedness to say them as proper to anyone who shares his understand­ ing of "p, q". But his and other persons' "prepared­ ness to say" those things can as well be shown by their readiness to accept to their "preparedness to say" the corresponding sentences as soon as the sentences are pointed out to them. On the other hand, someone may claim that E holds for certain sentences and not be prepared even to accept the necessary contradictory sentences simply because he understands the initial sentences in such

35

ENTAILMENT IN ORDINARY PROSE a way that E does not in fact hold between them.

I shall now adduce some examples which will hopefully suggest very strongly that (E) does match the intuitive range of entailment. As is well known, self-contradictory sentences entail any sentence whatsoever. Consider (1) "Warsaw is the capital of Poland and Warsaw is not the capital of Poland" entails "Einstein died". "p, not-q" for (1) is (1') "Warsaw is the capital of Poland and Warsaw is not the capital of Poland, Einstein did not die" is a true sentence. There are, among other things, the following F's for

(1'): (F'l') "Warsaw is the capital of Poland" is a true sentence. {F"l') "Warsaw is not the capital of Poland" is a true sentence. There is the following H for (F"1'): (H1') "Warsaw is the capital of Poland" is not a true sentence. (F'l') and (H1') are contadictory. Take now a more trivial example: (2) "This is a square" entails "This is a rectangle", "p, not-q" for (2) is (2') "This is a square, this is not a rectangle" is a true sentence, (with the provise that

this

refers to the same object throughout). We may say that there are the following F's said in (2'): (F'2') This is a rectangle. (F"2') This is not a rectangle.

ANDRZEJ BOGUSŁAWSKI

36

which are contradictory. Or take (3) "This is a dog" entails "This is not a cat". "p, not-q" for (3) is (3') "This is a dog, this is a cat" is a true sentence. Assume that there are such F's for (3') as (F'3') This is a dog. (F"3') This is a cat. Assume, furthermore, that there is the following G for (F'3') : (G3') This is a dog, this is not a cat. For(G3') we have the following H: (H3') This is not a cat. (F"3') and (H3') are contradictory. However, one may raise doubts on whether (F'2'), (F"2')f (F'3'), (F"3') are actually said in (2'), (3'), respectively. We need not insist on this. We may con­ fine ourselves to the claim that (to take only example (3)) there are the following F's for (3'): (F'"3') "This is a dog" is a true sentence. (F""3') "This is a cat" is a true sentence. We then obtain the following H's for (F'"3'): (H'3') "This is a dog, this is not a cat" is a true sentence. (H"3') "This is not a cat" is a true sentence. (H'"3') "This is a cat" is not a true sentence. (F""3') and (H'"3') are contradictory. Let us now turn to the entailment between general negative-existential sentences and particular sentences,

ENTAILMENT IN ORDINARY PROSE

37

cf.: (4) "Dogs which are not mammals do not exist" entails "This dog is a mammal". "p, not-q" for (4) is (4') "Dogs which are not mammals do not exist", this dog is not a mammal" is a true sentence. We have the following F's for (4'): (F'4') "Dogs which are not mammals do not exist" is a true sentence, (F'4') "This dog is not a mammal" is a true sentence. It is reasonable to say that "something exists" means that the corresponding expression occurs essentially in a positive extensional sentence which is true. This of course requires comments for which there is no space here. In addition, let us abbreviate the above definition to "occurs essentially in a true sentence". We may now say the following H in connection with (F'4'): (H'4') It is not the case that the expression "dog which is not a mammal" occurs essentially in a true sentence. For (F"4') we obtain, on the other hand, the following H: (H"4') "This is a dog which is not a mammal" is a true sentence, and for (H"4'), the following I (cf. "c" in (E)): (I'4')""This is a dog which is not a mammal" is a true sentence" is such that the expression "dog which is not a mammal" occurs essentially in a true sentence.

ANDRZEJ BOGUSŁAWSKI

38

But on the basis of (H'4') we have to say the follow­ ing I about the same sentence (H"4'): (I"4') ""This is a dog which is not a mammal" is a true sentence" is such that it is not the case that the expression "dog which is not a mammal" occurs essentially in a true sentence. (I'4') and (I'M') are contradictory. Consider now the entailment between a particular sentence and a general positive-existential one, as in (5) "This is an elf" entails "At least one elf exists". "p, not-q"

for (5) is

(5') "This is an elf, no elf exists" is a true sent­ ence. Here, we have the following F's: (F'5' ) "This is an elf" is a true sentence. (F"5') " N O elf exists" is a true sentence. In a

way similar to the analysis of (4) we obtain

the following contradictory I's about (F'S'): (I'S') ""This is an elf" is a true sentence" is such that the expression "elf" occurs essentially in a true sentence. (I"5') ""This is an elf" is a true sentence" is such that expression "elf" does not occur essential­ ly in a true sentence. Notice that what is said in "This is an elf", on the one hand, and what is said in "There are no elves",

ENTAILMENT IN ORDINARY PROSE

39

on the other hand, are not immediately contradictory things. The logical way from "This is an elf" to "There exists at least one elf" crucially depends on the introduction of the notion of "true sentence" as applied to both sentences, the introduction which is inherent in the concept of entailment. The way from Fa to "F exists" is not the same as the way from not-Fa to "Fa is not a true sentence" and other element­ ary transformations (some of which we applied above). Finally, a word on tautologies as entailed by any sentence. According to (E) we have to negate them; this trivially leads to contradictory sentences within not-q; but these are sufficient for (E) to hold since there are no constraints in it as to where the contradictory parts of S should be found. This is a mirror image of what we stated for contradict­ ions as entailing everything. Notice that our view of entailment does not require a roundabout way of demon­ strating the all-embracing entailing force of contra­ dictions by means of first ascribing them certain tautologies as entailed by them. It is worthwhile to make a final remark relating to the properties of entailment as applying to tautologies and contradictions. Both tautologies as entailed by an arbitrary sentence, cf. (6) "It is raining" entails "Either Peter is taller than John or Peter is not taller than John". and contradictions as entailing an arbitrary sentence, cf. (1), give the impression of fairly bizarre state­ ments and provoke questions about their relation to

40

ANDRZEJ BOGUSLAWSKI

what is apprehended as more normal statements, such as (3), where p and q are intuitively connected with each other in a certain way, and, indeed, about the substance behind the generalization covering all of them, the generalization which is in fact attached to the logical notion of entailment and accepted here in a special form (of (E)). I think the following interpretation answers these questions. Some combinations of states of affairs depicted in particular sentences are barred on linguistic grounds as violating the law of the necessary choice between the absence and presence of negation with a given sayable, the law which is at the heart of language as such. Entailment in its broadest scope appears to be the most general form of stating the void character of just any combinat­ ion of expressions having that sort of "antilinguistic" nature, and having it in spite of the appearances which might make one think of the combination as a product language is basically expected to supply, i.e. as a true sentence. Viewed in this way, entailment can be paralleled by factual general hypotheses of non-existence like (7) Elves do not exist. Such sentences entail more specific

statements

like (8) Elves with Einstein's IQ do not exist. (9) Elves whose IQ is other than Einstein's do not exist. which are redundant in relation to the more general

ENTAILMENT IN ORDINARY PROSE

41

statements, but quite legitimate and even indispensable for certain purposes. The meaningfulness of (E) does not preclude the possibility of constructing other, more restrictive concepts. One of them can be obtained from (E) by postulat­ ing that S' resulting from the operations described in (E), but applying to "p, q" rather than "p, not-q" comprises no contradictory sentences. Such a richer notion would exclude combinations where p is self-contradictory without at the same time excluding comb­ inations with tautologies taken as q. In addition, we can postulate that not-q by it­ self does not yield any analogon of S comprising a contradiction. This would also leave combinations with tautologous q's outside the scope proper to (E). The resulting relation would hold between p and not-q where neither p nor not-q by themselves comprise contradictions: the contradictions would be due exclus­ ively to some special combinations of non-contradictory sentences. No doubt this is the most interesting relat­ ion between sentences. It is this relationship which draws particular attention when one talks about entail­ ment. We may wish to call the former case of more re­ strictive relations based on entailment "seminarrow entailment", and the latter - "narrow entailment".

IRENA BELLERT Montreal

A PROBLEM WITH SEMANTIC INTERPRETIVE RULES A PARTICULAR CASE OF RECIPROCALS

The purpose of this paper is to consider the follow­ ing question: (a) Should semantic interpretive rules for sentences of a given structure type be sensitive to idiosyncratic properties of the lexical items involved and the relevant knowledge of the world, or rather (b) should they be generally independent of the idio­ syncratic properties of the lexical items involved, which would then account for some specific additional restrictions imposed on general rules? I will consider this question by discussing one particular case concerning the interpretation of reci­ procal pronouns in single-clause sentences, but it seems to me that, in general, the principle (b), which I am opting for, is worth some consideration, and the question posed here is relevant for many other cases as well, I will first propose two rules for interpreting simple one-clause sentences consisting of a two place predicate with two arguments: the antecedent NP and a reciprocal pronoun each other, I will use the following notation. The antecedent NP will be symbolized by

44

IRENA BELLERT

QT, where Q is a linguistic quantifier (a determiner, specifier, numeral, etc.) and T is a set expression (the NP without its quantifier). I IT I I symbolizes the denotation class of T I IQTI I is a family of all those subclasses of I IT I I whose cardinality of relative 'size' is indicat­ ed by the quantifier Q R is the reference class of the NP-antecedent, which belongs to I IQTI I I IPl I is the denotation class of the predicate P

Interpretive

Rule

1:

A sentence of the type QT P each

other

is true in a

context C if - є I IP I I)) in C

Interpretive

Rule

2:

A sentence of the type QT

P each

other

is true in

a context C only if

Rule 1 is a sufficient condition, while Rule 2 is a necessary condition. There is no way for having both a necessary and sufficient condition which would account for all the possible predicates including those which impose idiosyncratic restrictions on the interpretation of sentences with reciprocal pronouns. If we accept Rule 2, it would hold true of every sent­ ence with a reciprocal pronoun, but only as a necess­ ary condition; it would leave other conditiors unspecif­ ied and dependent on the particular predicate involved. Rule 1, however, could be accepted as sufficient

45

SEMANTIC INTERPRETIVE RULES

and necessary, provided that it is restricted to only those predicates which do not impose idiosyncratic restrictions. Hence it could be taken as a general rule for all unmarked (non-idiosyncratic) cases, which should then be accounted for by additional rules pertaining to the particular, idiosyncratic verbs which may occur as a predicate. Let us then accept provisionally Rule 1 in the form of a necessary and sufficient condition:

Interpretive

Rule 1 ':

A sentence of the type

QT

P

each

other

is

true in a context C if and only if

Consider first sentences for which the reference class contains just two members

The interpretive rule l' is obviously valid for all these sentences independently of the verb which may occur as a predicate. Consider now examples in which the reference class of the antecedent NP consists of more than two entities. We will examine the following sentences one after another, in order to realize what kind of restrict­ ions may be imposed on the interpretation of sentences

46

IRENA BELLERT

with reciprocal pronouns. (2) (3) (4) (5) (6) (7) (8) (9)

(The) (The) (All) (All)

three girls kissed each other three boys washed each other the children liked each other the children washed each other are married to each other (All) those young P e o P l e { m a r r i e d e a c h o t h e r last ) year (All) the guests have conversed with each other during the party (All) the guests are conversing with each other

(All) these youngsters { love ) each are in love with other (10) (All) these youngsters are engaged with each other (11) (All) the students were sitting next to each other (2) is obviously interpreted according to the conditions of Interpretive Rule 1': each of the girls kissed the other two, (3) however, although it contains a similar verb which belongs together with the verb to kiss to the class of action verbs, it is not necessarily interpreted according to these conditions. Its more natural interpretation is: each individual washed one of the other two. In other words, (3) allows an interpretation in which each entity is related by the predicate with one of the remaining entities in the reference class. The difference in the interpretation of (2) and (3) comes from our factual knowledge of the world. If it is quite common for three people to

47

SEMANTIC INTERPRETIVE RULES

kiss each other, for instance, on the occasion of a family visit, it is not common to be washed twice on the same occasion. There is a similar difference between (4) and (5). The relation of 'liking' does not impose any restrictions on the conditions of Rule 1' (as each person in the reference class can like all the others, at the same time) whereas the relation of 'washing' does, as argued above, especially if the class is numerous. In (6) the relation of 'marrying' imposes a restriction which depends, in addition, on the type of society such a statement pertains to. In our society it will be interpreted with the restrictions that the particular individuals be related with each other pairwise, and that pairs do not overlap. The relation of 'conversing' imposes a restriction on a large class of entities, as it is not possible to converse with a large number of people at the same time. Therefore, (7) may be interpreted according to the general conditions on reciprocals, but not (8), where the above mentioned restriction in conjunction with the progressive aspect make it impossible: each guest cannot be conversing with all the other guests at the same time, (9) represents two sentences with verbs belonging to the same class, but the verb to love imposes no

restrictions, whereas to be in love with

suggests a

restriction according to which the sentence will be interpreted differently: each individual is in love with some other individual in the same class.

48

IRENA BELLERT

As it appears by virtue of these examples, the re­ strictions on the general conditions for sentences with reciprocals depend on specific lexical items. When sent­ ences with reciprocals contain verbs which impose no restrictions, they are interpreted according to the conditions defined in the Interpretive Rule 1'. Let me now consider the definition concerning reci­ procal sentences in the form of a truth condition for elementary reciprocal sentences proposed by Fiengo and Lasnik (1973). This condition is supposed to overcome some of the problems due to lexical differences. It has been also accepted in a recent paper by Higginbotham (198 2), Let me quote the proposal: "I, II.

R is eaoh-the-other in S if , x ≠ y > xRy (where 2) R is reciprocal in S if 3 a partition of S into S, ...

is eaoh-the-other

in S. This formalism implies that when S has less than four members, n must equal 1, since we have required that the subsets be disjoint (by the definition of partition), that each subset have at least two members, and that x be distinct from y. When n = 1, I and II are synonymous. II defines a relationship in a set. This set is the referent of the NP antecedent of the reciprocal pro­ noun each other," (Fiengo & Lasnik, 1973, p. 450, where the two conditions are numbered as (15) and (16), respectively). Condition I (which corresponds to my Interpretive Rule 1, except for some details irrelevant for this discussion) is assumed to hold true in the cases when the reference set of the antecedent has less than four

49

SEMANTIC INTERPRETIVE RULES

members, Condition II is assumed to hold true in the remaining cases. It will be shown right below that Condition II should be dispensed with altogether. Condition II has the form of a sufficient condition only, it is thus not assumed to be a necessary condition. I will argue, however, that it is neither necessary nor sufficient, but it holds true only in some particular cases in which the lexical material and the factual knowledge of the world restrict the general condition by imposing some specific

interpretat-

ion. The authors provide themselves some counterexamples, where there is a partition into subsets which

satisfies

condition II, and a sentence with a reciprocal pronoun would not be true. But a general lexical rule is proposed to the effect that stative verbs

(as opposed to

active verbs) do not allow a partition into subsets, and therefore in the case of stative verbs n = 1 in Condition II (which amounts to making Condition II synonymous with Condition I ) . The following

examples

are provided for stative verbs: (a)

The men in the room are the same height as each other.

(b)

The men in the room know each other.

(c)

The men in the room see each other. All these sentences are true only if the relation-

ship expressed by the verb holds for all possible pairs in the room

(which means that Condition I is

valid for such c a s e s ) . On the other hand, the following examples are provided for active verbs:

50

IRENA BELLERT

(c) (d)

The men in the room are hitting each other. The men in the room are looking at each other.

(e)

The men in the room are conversing with each other.

The authors argue that all these sentences can (the stress is mine) be true if there exists a partition of the men such that within each subset the each-the-other relationship holds (that is, Condition II holds true). In fact - as I argue - such an interpretation which restricts this relation to some subsets of the set of the men, in the case of hitting or c o n v e r s i n g , or just to two-element subsets in the case of looking at> is clearly due to the semantic properties of the respective verbs and the use of progressive, which results in a reading incompatible with the general knowledge of the possible situations (it is not possible for each individual to converse with a large number of other people at the same time, nor is it possible for more than two individuals to look or stare at each other at the same time). This point can be supported by the fact that if some other active verbs are used in sentences which are not in the progressive, there will be no such restrictions. For instance (e)

The men kissed each other when leaving the party.

(f)

The students corresponded with each other for

(g)

some time. The children have sent X-mas cards to each other.

And as a matter of fact, not all stative verbs disallow the partitioning of the set of referents. For instance, my examples (6) and (10) impose an interpret-

SEMANTIC INTERPRETIVE RULES

51

ation restricted to a pairwise relationship on disjoint subsets. (Moreover, the interpretation of (6) depends on the extralinguistic context, that is, on whether the sentence is uttered in a monogamic society or not). The authors themselves remark in a footnote that "There is one general class of exceptions. When the antecedent of the reciprocal defines natural pairwise relationships, the sentence will be true if the each-the-other relationship holds of each pair. For example: 'The husbands and wives in the room are similar to each other' (Fiengo & Lasnik, 1973, footnote 4, p. 454). Such pairs are clearly disjoint. Finally, the authors provide some counterexamples to their conditions with verbs which they call 'linear configurationals': (h) (i)

The trays are stacked on top of each other. The boxes are nested in each other.

Such verbs impose an interpretation in which each entity is related by the verb with just one other entity in the set in a linear way. The authors seem to consider these two verbs as exceptional. However, there are several other verbs (or contexts) which impose specific restrictions on the general conditions concerning reciprocals, so that these two verbs are by no means exceptional. Consider, for instance, my example (11) ('All the students were sitting next to each other'), or some other sentences containing other verbs:

lying,

standing

etc.

(j)

or kneeling

next to,

holding

The children were .in a row {in a circle}

hands with,

each other

52

IRENA BELLERT

Neither of the two conditions defined by the authors holds in the case of such sentences. The interpretation is clearly restricted by the specific semantic properties of these verbs together with the prepositions, and it also depends on the particular prepositional phrase which may cooccur in the sentence. Such sentences are true only if the particular subsets have just two members, and each subset overlaps with two other subsets (in a circle), except for two subsets each of which overlaps with just one other subset (in a row). As we have seen, the particular restrictions may correspond to relations defined on various possible subsets of the reference class of the antecedent NP, which depending on the specific verbs, are: conversing)

(1)

partitioned

(disjoint) subsets (e.g.

(2)

partitioned

(disjoint) subsets consisting of two

members each (e.g. being

married

to,

engaged

to) (3)

touching),

possibly overlapping subsets (e.g.

possible with an asymmetric relation (e.g.

washing, (4)

hitting)

subsets consisting of two members, each subset necessarily overlapping with two others

(e.g. sitting (5)

next

to in a

circle)

subsets consisting of two members, one member of each subset being related in an antisymmetric way to the other one, and each subset necessarily overlapping with two others, except for two subsets each of which overlaps with one other subset

(e.g. nested

in,

standing

SEMANTIC INTERPRETIVE RULES

53

behind in a row, etc.). These cases do not exhaust all the possible restrictions due to the semantic properties of specific verbs or contexts. As we see, Condition II holds for the cases (1) and (2) above, and only when the reference set contains more than three members. There is no justification for accepting then this condition as a general condition on sentences with reciprocal pronouns. It is applicable only to the cases in which the verb imposes some specific restrictions. It seems thus that the general conditions may be confined to Condition I alone, which corresponds, roughly, to my Interpretive Rule l', and which could also be accepted as a necessary and sufficient condition for all unmarked cases, that is, for all the cases which are not restricted by any idiosyncratic semantic properties of the verb or some other contextual elements (e.g. prepositional phrases). The necessary condition in all the cases (non-restricted or restricted by the semantic properties of the verb) is expressed in my Interpretive Rule 2, which says that each member of the reference class is related by the verb with at least one other member of that class.

54

IRENA BELLERT

NOTES

1 The proposed symbolism and interpretive rules make part of a unified description of a quantified NP-arguments in a predicate-argument structure, includ­ ing semantic interpretive rules in Bellert 1983 a and b. 2 For instance, for a sentence "Three boys washed each other" Q would stand for three, S would stand for boys, IITII would be the denotation class of boys I IQTI I would stand for all the subsets of I IT I I which contain just two members and the reference class R would be one of the subsets of I IQTI I 3 Higginbotham (1982) observed that this condit­ ion expresses only a sufficient condition and added that it is too liberal in some contexts by giving an example to this effect (They love each other, in a context of 20 people each of who loves and is loved by exactly one of the o t h e r s ) . But if the condition is too liberal, this means that it is not sufficient. Higgenbotham's remarks are then consonant with what I will argue for, namely, that this condition is neither necessary nor sufficient. Higginbotham nevertheless accepts this condition in his paper, in which he obtains further results. The examples he uses could, however, be interpreted in terms of the Interpretive Rule 1', which I propose here (and which is roughly equivalent to this first condition of Fiengo & Lasnik, 1973) where no partition is assumed. REFERENCES Bellert, Irena. 1983a. "Interpretive model for ling­ uistic quantifiers". Paper delivered during the 7th International Congress of Logic, Methodology and Philosophy of Science, Salzburg, 1983. To appear in a selection of contributions to the Congress. London & New York: Plenum Co. Bellert, Irena, 1983b. "Model for quantificational structure". Mimeographed, 80 p. McGill Univer­ sity . Higginbotham, James. 1982. "Reciprocal interpretat­

ion". Journal

of

Linguistic

Research

1. 97-116.

SEMANTIC INTERPRETIVE RULES Fiengo, Robert & Howard Lasnik. 1973. "The Logical Structure of Reciprocal Sentences in English". Foundations of Language 9. 447-468.

55

GERT JÄGER Leipzig

ZUM VERHALTNIS VON LEXEM UND PARAPHRASE

0. Die nachstehenden Uberlegungen zum Verhaltnis von Lexem und Paraphrase reduzieren die umfangreiche und vielschichtige linguistische Problematik urn die Paraphra­ se in mehrfacher Hinsicht: Zum einen geht es uns nur urn die Paraphrase als Gegenstand der linguistischen Beschreibung und nicht auch urn die Paraphrase als Instru­ ment der linguistischen Analyse (vgl. E. Lang 1977:99f) und zum anderen - innerhalb ersterer - nur urn die lexikalische Parahprase (Benennungsparaphrase), wahrend Satzparaphrasen, soweit sie nicht das Ergebnis des Vorkommens lexikalischer Paraphrasen sind, auβerhalb der Betrachtung bleiben. Weitere Einschrankungen ergeben sich schlieβ1ich daraus, daβ wir nur einige Aspekte des Verhaltnisses von substantivischem Lexem (=S-Lexem) und Pa­ raphrase berüihren werden und dieses Verhaltnis auch nur fur solche Paraphrasen diskutieren wollen, die als "definitorische Paraphrasen" (vgl. Viehweger u.a. 1977:266) oder als "hinreichende Beschreibungen" charakterisiert werden konnen. Dabei sollen unter "hinreichender Beschreibung" solche Paraphrasen verstanden werden, die im Unterschied zu den definitorischen Paraphrasen nicht auf alle bzw. nicht auf alle wesentlichen Eigenschaften des vom jeweiliqen S-Lexem bezeichneten Denotats Bezug neh-

58

GERT JAGER

men, wohl aber auf solche Eigenschaften, daβ eine kontextfreie Identifizierung dieses Denotats gegeben ist. Als Beispiel fur eine definitorische Paraphrase (im Folgenden P d ) und ihr Bezugs-S-Lexem verweisen wir auf (1). Unter (2) und (3) sind Beispiele fur die zweite Art der uns hier interessierenden Paraphrasen (im folgendenP,d) und ihre Bezugs-S-Lexeme angegeben. (1)(a) Planet (b) Himmelskörper, der keine eigene Strahlung aussendet und der sich auf einer ellipsenförmigen Bahn um die Sonne bewegt (2) (a) Mensch (b) Lebewesen mit der Fahigkeit zum abstrakt-begrifflichen Denken (3)(a) Trier (b) Geburtsort von Karl Marx Die besondere Aufmerksamkeit, die wir gerade dem Verhaltnis von S-Lexemen und Paraphrasen, die unabhängig vom Kontext dieselbe denotative Bedeutung haben, d.h. sich auf denselben Ausschnitt aus dem gesamtgesellschaftlichen Wissen beziehen, wie die jeweiligen S-Lexeme (insgesamt oder zumindest in jeweils einer Bedeutung), widmen wollen, erklart sich daraus, daβ diese Paraphra­ sen und dieses Verhaltnis unter ubersetzungswissenschaftlichem Gesichtspunkt von besonderem Interesse sind (vgl. Jager (1976)). Wenn wir auch die Paraphrasenproblematik im folgenden nur innerhalb einer Sprache betrachten, so ist doch die Konzentration auf das gegenseitige Ver­ hältnis von S-Lexem und P, bzw. Pb im Hinblick auf ihre d b signifikative Bedeutung (sprachliche Bedeutung) fur uns vor allem mit dem Ziel verbunden, die Translation zu erklaren (vgl. Jäger (1983)).

LEXEM UND PARAPHRASE

59

1. Zumindest für P, wird gewohnlich angenommen, daβ zwischen ihnen und den jeweiligen Lexemen Synonymie besteht und sie sich von Synonymen lediglich darin unterscheiden, daβ sie syntaktisch-semantisch gegliedert und von gröβerem Umfang als ein Semem sind (vgl. Viehweger u.a. (1977:257)). Wir haben an anderer Stelle (Jager (1976:49 ff)) auf dem Hintergrund der Annahme, daβ Paraphrasen und die jeweiligen Lexeme in ihrer denotativen Bedeutung ubereinstimmen (konnen), als wesentlichen Unterschied zwischen beiden auf die Okkasionalitat (Nichtkonventionalitat) der Paraphrasen gegenuber der Konventionalitat (Konventionalisiertheit) der Lexeme hingewiesen, den wir als einen Unterschied unter pragmatischen Gesichtpunkt bzw. in der intralingualen pragmatischen Bedeutung gewertet haben. Beide Annahmen lassen allerdings offen, inwieweit sich Paraphrasen in ihrer signifikativen Bedeutung von den jeweiligen Lexe­ men unterscheiden bzw. ob der Unterschied zwischen bei­ den Arten der Benennung im Hinblick auf den signifikativen Aspekt der Bedeutung dem Unterschied zwischen echten Synonymen entspricht. Letzteres wird wohl worausgesetzt, wenn P gegenüber den jeweiligen Lexemen gröβtmögliche Merkmalkorrespondenz (semische Identitat) zuerkannt wird (vgl. Viehweger u.a. (1977:266)). Wir wollen zunachst festhalten, was wir zu Konventionalitat/Nichtkonventionalitat von Lexem und Para­ phrase gesagt haben. Doch stellt sich dieser Unterschied zugleich auch als eine semantische Verschiedenheit von Lexem und Paraphrase dar: Wie B. Koenitz (1983) gezeigt hat, geht in die signifikative Bedeutung einer konventionellen Benennung eine Komponente ein, die als mini-

GERT JAGER

60

males Sprachwissen charakterisiert werden kann und als "x, das in L. konventionell ' e x p r ' heiβt" zu fassen ist, so da3 die entsprechende Bedeutungskomponente fur als "Individuum, das auf deutsch konventionell

(la)

"Planet"

heiβt" anzunehmen wäre. Eine derartige Bedeutungskomponente kommt weder Pd noch Pb. zu, d.h. auf deutsch

(lb) bedeutet nicht "Individuum, das

"Himmelskorper, der keine eigene Strahlung

aussendet ..." hei|βt" (daran wiirde auch nichts ändern, wenn man anstelle des ausgelassenen

"konventionell" etwa

"okkasionell" einsetzte). Wir miissen vielmehr davon ausgehen, daβ Pd, und

Pb

(wie alle Paraphrasen)

hinsichtlich

des minimalen Sprachwissens so beschaffen sind, daβ jedem der sie konstituierenden Lexeme eine solche Bedeu­ tungskomponente zugeordnet ist, so da3 sich fur

(lb)

vereinfacht ergibt "Individuum, das auf deutsch konven­ tionell

"Himmelskörper'

heiβt, nicht die Eigenschaft

aufweist, die auf deutsch konventionell

"aussenden"

hei(3t und etwas betrifft, was auf deutsch konventionell "Strahlung" hei(3t, ...". In diesem Unterschied

zwischen

Lexem und Paraphrasen liegt die Ursache dafüir, da(3 Pd oder Pb, zu S-Lexemen in 'heiβen'-Sätzen nur als Subjekt vorkommen konnen: (4) Himmelskorper, die keine eigene Strahlung

aussenden

... , heiβen Planeten (5)

Derartige Objekte hei(3en Himmelskorper, die keine eigene Strahlung aussenden und sich auf einer ellipsenförmigen Bahn urn die Sonne bewegen.

und da3 Sätze wie

(6)

(6) Himmelskorper, die keine eigene Strahlung aussenden ... , sind Planeten.

LEXEM UND PARAPHRASE

61

eine Interpretation als "... heiβen PLANETEN" finden können (vgl. Koenitz (1983)), aber Sätze wie (7) (7) Planeten sind Himmelskorper, die... nicht analog als "... heiβen HIMMELSKORPER, .. ." interpretiert werden konnen (die Schreibung mit Groβbuchstaben soll hier den Nicht-Subjektcharakter der Konstituente symbolisieren). Eine Interpretation von "sein" als "heiβen" setzt in Sätzen wie (6) und (7) immer voraus, die Paraphrase als Subjekt aufzufassen. Der Unterschied zwischen S-Lexem und P d bzw. Pb hinslichtlich der Bedeutungskomponente "minimales Sprachwissen" schlie3t ein, da3 die Synonymiebeziehung zwischen einem S-Lexem Lx1 und einer "dazugehorigen" P, oder P. anders ist als die Synonymiebeziehung zwischen Lx1. und einem ihm echt Synonymen Lxj., da beiden Ausdrucken hier jeweils als Ganzen eine Bedeutungskomponente "x, das in L. konventionell "expr" heiβt" zugeordnet ist, die je nach Belegung von 'expr" einen spezifischen Wert hat. Die Tatasche, da3 bei Synonymen jedes eine Bedeu­ tungskomponente "minimales Sprachwissen" hat, ist die Ursache dafiir, da3 Satze wie (8) . (8)(a) Linguisten sind Sprachwissenschaftier. (b) Sprachwissenschaftier sind Linguisten. (unter der Voraussetzung, da3 "Sprachwissenschaftier' nicht als okkasionelle Bildung verstanden wird, deren Bedeutung sich vollstandig aus der ihrer Teile ergibt, was ein bei motivierten Benennungen durchaus mägliches Verstehen ist) keine Interpretation als "hei3en"-Sätze erhalten konnen und "heiβen"-Sätze mit Synonymen nur in einer Form wie (9) auftreten konnen:

GERT JAGER

62

(9) (a) Linguisten heiβen auch Sprachwissenschaftler. (b) Sprachwissenschaftier hei3en auch Linguisten. Die Satze unter

(8) und

(9) lassen zugleich erkennen,

daβ die Bedeutungskomponente

"minimales Sprachwissen"

bei S-Lexemen grundsatzlich immer anwesend ist und

daβ

eine solche Komponente Pd und Pb als Ganzen nicht zukommt, weil sonst Satze wie

(4) bzw. Satze wie

(6) mit

"heiβen"-Interpretation nicht moglich waren. Die Abwesenheit einer der Pd bzw. Pb. als Ganzem zukommenden Komponente "minimales Sprachwissen" unterscheidet sie bedeutungsma(3ig auch von

lexikalisierten

(konventionellen) Mehrwortbenennungen, bei denen diese Bedeutungskomponente

zugleich als eine Instruktion auf-

zufassen ist, daβ weitere Bedeutungskomponenten der Mehrwortbenennung als Ganzem zuzuordnen sind. So ware "Schwarze W i t w e ' als lexikalisierte

Mehrwortbenennung

für eine bestimmte Art von Spinnen nicht eine Bedeutung zuzuordnen wie (10), (10) Individuum, das auf deutsch konventionell

"Witwe"

heiβt, . . . , und auf das die Eigenschaft zutrifft, die auf deutsch konventionell

'schwarz' heiβt, ...

sondern eine Bedeutung wie (11): (11) Individuum, das auf deutsch konventionell

'Schwar­

ze W i t w e ' heiβt, ... Ein Verstehen von lexikalisierten Mehrwortbenennungen nach ihrer Motivation

(wie nah oder fern es einem

"richtigen" Verstehen auch sein mag) ist auf alle Fälle schon insofern ein unvollstandiges Verstehen, als immer eine Unkenntnis der Bedeutungskomponente Sprachwissen" vorliegt.

"minimales

LEXEM UND PARAPHRASE

63

2. Neben dem mit der Anwesenheit/Abwesenheit einer Bedeutungskomponente "minimales Sprachwissen" gegebenen signifikativ-semantisch relevanten Unterschied zwischen

S-Lexemen und Pd bzw. P b verdient ein zweiter besondere d Aufmerksamkeit: Im Zusammenhang mit der Mehrdeutigkeitsproblematik (vgl. Jäger,Koenitz (1980: 40)) wurde die Annahme getroffen, die von B. Koenitz am Beispiel definitorischer Satze (vgl. Koenitz (1983)) erhartet wurde, daβ das gegenseitige Verhaltnis der Bedeutungskomponenten (semantischen Merkmale) eines Semems mit Ausnahme der Bedeutungskomponente "minimales Sprach­ wissen" durch die Alternative zu erfassen ist, so daβ ein Semem durch recht unterschiedliche Merkmalsequenzen (im Prinzip soviele, wie durch die alternative Verknüpfung beschrieben werden), die man Alloseme nennen konnte, realisiert werden kann. Das dem S-Lexem 'Planet 'zugehorige Semem konnte dementsprechend (wenn wir davon absehen, daβ bestimmte Merkmale die Anwesenheit anderer Merkmale voraussetzen oder nach sich Ziehen) u.a. durch folgende Alloseme realisiert werden: (12)(a) Individuum, das auf deutsch konventionell 'Pla­ net' heiβt (b) Individuum, das auf deutsch konventionell

'Pla­

n e t ' heiβt und ein Himmelskorper ist (c) Individuum, das auf deutsch konventionell

'Pla­

n e t ' heiβt und sich um die Sonne bewegt

Es ist offensichtlich, daβ (lb) - unter der Voraussetzung, daβ ein Satz mit

(la) als Subjekt und

(lb) als

Pradikatsnomen tatsachlich eine Definition ist und das Semem eines S-Lexems wie

daβ

'Planet' nur Merkmale um-

64

GERT JAGER

faβt, die sich auf def initorische ("wesentliche") Eigenschaften des Denotats beziehen - zwar auf dieselben Ei­ genschaften des Denotats wie (la) Bezug nimmt, aber in seiner signifikativen Bedeutung nicht mit dem Semem "Pla­ net" identisch sein kann: Dem steht - abgesehen davon, daβ bei (lb) die Bedeutungskomponente "x, das konventionell 'expr' hei(3t" fehlt - entgegen, daβ das gegenseitige Verhaltnis der Bedeutungskomponenten bei (lb) als Konjunktion zu fassen ist. Die Bedeutung von (lb) entspricht somit - wiederum abgesehen von der Bedeutungskomponente "minimales Sprachwissen" - nur einem der unter (12) moglichen Alloseme, d.h. dem maximalen Allosem des Semems "Planet". Ebenso entspricht (2b) in seiner signifikativen Bedeutung nicht einem Semem, das dem S-Lexem (2a) zugehorig ist, sondern nur einem bestimmten Allosem eines solchen Semems, wobei dieses Allosem nicht das maximale Allosem des entsprechenden Semems ist. Es erscheint hervorhebenswert, daβ das bisher Gesagte auch fur das Verhaltnis eines Eigennamens wie (3a) zu einer seiner Paraphrasen zutrifft, allerdings mit der Besonderheit, da(3 ein entsprechendes Denotat nur eine Beschreibung, aber keine Definition erfahren kann, so da(3 keine Pd wohl aber Pb moglich sind. Ob man Paraphrasen annehmen kann, die als vollstandige Beschreibung zu betrachten sind, hangt davon ab, ob man an derartigen Individuen wesentliche und unwesentliche Eigenschaften unterschieden kann und eine Beschreibung dann auf die Erfassung der wesentlichen Eigenschaften beschranken kann oder will.

LEXEM UND PARAPHRASE

65

3. Es ensteht nun die Frage, wie die oben getroffene Feststellung, daβ eine Pd bzw. eine Pb, hinsichtlich ihrer Bedeutung einem Allosem eines Semems, das einem (natiirlich bestimmten S-Lexem zugehorig ist, entspvicht wieder unter Vernachlassigung der Bedeutungskomponente "minimales Sprachwissen"), zu verstehen ist. Sind Pd bzw. Pb nicht nur in ihrer denotativen Bedeutung mit den jeweiligen S-Lexemen identisch, sondern ist ihre signifikative Bedeutung auch mit demjenigen Teil des betreffenden Allosems identisch, der mit Komponenten der denota­ tiven Bedeutung korrespondiert? Mit anderen Worten: Ist die signifikative Bedeutung der Pd oder Pb in dem jeweiligen Allosem (das ja ebenso wie sein Semem eine Einheit der signifikativen Bedeutung ist) enhalten? Diese Frage kann noch erweitert werden: Nehmen wir an, bei einem S-Lexem Lx1, handele es sich urn einen streng definierten wissenschaftlichen Terminus. Dann kann die signifikative Bedeutung mit der denotativen Be­ deutung in dem Sinne korrespondieren, da(3 jeder gesamtgesellschaftlichen Wissenstatsache, die eine Komponente der denotativen Bedeutung von Lx1. ist, ein semantisches Merkmal des Lx1. zugehorigen Semems entspricht. Trotz dieser Korrespondenz waren denotative und signifikative Bedeutung von Lxi. zweifellos insofern voneinander verschieden, als letztere die Komponente "x, das in L. konventionell 'Lx1. " heiβt" umfaβt und das Verhaltnis zwischen den Komponenten der denotativen Bedeutung wohl als Konjunktion gefa3t werden muβ, wahrend die entsprechenden Merkmale des Semems alternativ verknüpft waren. Allerdings ware der zweite Unterschied bei dem maximalen Allosem des Lx1. zugehorigen Semems aufgehoben, so da(3 die oben gestellte Frage in die Frage iiberfuhrt werden

66

GERT JĂGER

kann, ob in dem angenommenen Fall denotative Bedeutung und maximales Allosem (ohne Bedeutungskomponente "minimales Sprachwissen") sowie signifikative Bedeutung der dem maximalen Allosem entsprechenden Pd identisch sind. Die Beantwortung dieser Frage ist bereits bei G. W. Leibniz angelegt: :'. . . nicht nur die Dinge selbst, sondern auch die Vorstellungen der Dinge konnen und sollen vom Geist nicht immer deutlich wahrgenommen werden; und deshalb werden zur Abkurzung statt ihrer die Zeichen angewandt. Wenn der Geometer, sooft er eine Hyperbel oder eine Spirale ... in einem Beweise nennt, stets gezwungen ware, sich deren Definitionen oder Konstruktionen und weiterhin die Definitionen der in die­ se eingehenden Ausdrücke genau vorzustellen, gelangt er nur sehr langsam zu neuen Entdeckungen." (1960:110). Wir konnen diese Sachlage vielleicht so ausdrucken, daβ ein S-Lexem (auch bei einem Vorkommen, fur das die signifikative Bedeutung als maximales Allosem angenommen werden kann) und seine P., auf unterschiedliche Weise auf die die denotative Bedeutung konstituierenden Wissenstatsachen Bezug nimmt. Die "Abkürzung" ist dabei nicht nur und nicht so sehr als die Moglichkeit zu sehen, daβ S-Lexeme mit sehr "kurzen" Allosemen, d.h. mit Merkmalsequenzen, die nur wenige Glieder des Semems umfassen, vorkommen konnen, sondern vor allem als das zu verstehen, was in der Psychologie Verdichtung genannt wird (vgl. F. Klix (1977: 597 ff)).Wir konnten also sagen, daβ ein S-Lexem verdichtet auf einen bestimmten Ausschnitt aus dem gesamtgesellschaftlichen Wissen Be­ zug nimmt, wahrend eine Pd dieses S-Lexems in nicht verdichteter oder besser: in weniger verdichteter Form auf diesen Ausschnitt bezogen ist.

LEXEM UND PARAPHRASE

67

Es fragt sich nun, ob der unterschiedliche Verdichtungsgrad von S-Lexem (speziell in Vorkommen mit dem maximalen Allosem) und Pd linguistisch relevant ist, d.h. die signifikative Bedeutung berührt. Wir meinen, daβ diese Frage zu bejahen ist. So kann z.B. eine Pd und ebenso eine Pb - jederzeit in Verbindung mit dem jeweiligen S-Lexem als eine (explikative) Spezifizierungsangabe vorkommen: (13) ein Planet als ein Himmelskörper, der keine eigene Strahlung aussendet und ... (14) ein Mensch als ein Lebewesen mit der Fähigkeit zum abstrak-fe-bagrifflichen Denken (15) Trier als Geburtsort von Karl Marx Demgegenüber ist ein S-Lexem nicht als eine (explikati­ ve) Spezifizierungsangabe zu einer seiner Paraphrasen möglich: (16)

ein Himmelskorper, der keine eigene Strahlung aussendet und ..., als ein Planet

(17) (18)

ein Lebewesen mit der Fahigkeit zum abstrakt-begrifflichen Denken als Mensch der Geburtsort von Karl Marx als Trier

Die Unmöglichkeit von (16) konnte evtl. noch so erklart werden, da(3 ein S-Lexem in unmittelbarer Verkniipfung mit seiner P, immer nur mit seinem minimalen Allosem auftritt, so daβ 'Planet' in (16) nicht verwendet werden kann. Diese Erklarung würde allerdings bei (17) und (18) versagen, denn es gibt keinen vernünftigen Grund zu der Annahme, daβ ein S-Lexem, wenn es zusammen mit einer seiner Pb vorkommt, auch nur seine minimale signifikative Bedeutung (="minimales Sprachwissen")

68

GERT JĂGER

haben kann. Wenn dies aber so ist, dann kann die Ursache für die Unfähigkeit von S-Lexemen, als (explikative) Spezifizierungsangabe fur ihre Paraphrasen zu fungieren, wohl nur darin liegen, daβ die signifikative Bedeutung von S-Lexemen weniger spezifiziert ist, d.h. weniger Merkmale umfaβt, als die ihrer Paraphrasen. Da dieses Weniger aber nicht den Umfang (die Zahl) der bezogenen Komplexe der denotativen Bedeutung betreffen kann, kann es nur die geringere bzw. gröβere "Tiefe" des Bezugs auf die denotative Bedeutung sein, die die signifikative Be­ deutung von S-Lexemen von der ihrer Paraphrasen unterschiedliche Verdichtungsgrad, der eine geringere oder groβere Zahl von Merkmalen mit sich bringt. Wir wollen diesen Unterschied in der signifikativen Bedeutung von S-Lexemen und ihren Paraphrasen - im Anschluβ an eine bei W. Lorenz/G. Wotjak (1977:289 ff) erorterte Hypothese - so fassen, da|3 sich ein Semem (bzw. seine Alloseme) als die signifikative Bedeutung eines S-Lexems auf die Weise auf die denotative Bedeutung die­ ses S-Lexems bezieht, daβ sie diese denotative Bedeutung in Komplexe zerlegt, von denen jeder einem semantischen Merkmal zugeordnet ist, wobei diese semantischen Merkma­ le selbst nicht weiter zerlegbar sind und damit auch die Komplexe der denotativen Bedeutung durch das Semem nicht weiter auf gegliedert werden. Mit einer P, bzw. Pb erfolgt demgegenüber eine zweifache Zerlegung der denotativen Bedeutung: Sie gliedert zunachst dieselben Komplexe aus bzw. zerlegt in dieselben Komplexe wie das ihr entsprechende Allosem. Doch wird diesen Komplexen jeweils ein Semem zugeordnet, das seinerseits den ihm entsprechenden Komplex in Teilkomplexe gliedert, von denen jeder einem semantischen Merkmal dieses'Semems zugeordnet ist.

LEXEM UND PARAPHRASE

69

Pd und Pb umfassen in ihrer signifikativen Bedeutung somit tatsächlich mehr Merkmale als das ihnen entsprechende Allosem und haben gegenüber dem S-Lexem eine insgesamt gröβere (signifikative) "Tiefenschärfe". So umfaβt das Semem "Planet" u.a. das semantische Merkmal "himmelskorper", womit auf einen bestimmten Komplex von Wissenstatsachen als ungegliedertes Ganzes Bezug genoramen wird. Die Pd (lb) umfaβt in ihrer signifikativen Bedeutung demgegenüber das Semem "Himmelskorper", womit auf denselben Komplex von Wissenstatsachen Bezug genommen wird, aber zugleich eine bestimmte Zerlegung dieses Komplex erfolgt. Die so entstehenden Teilkomplexe bzw. die entsprechenden semantischen Merkmale des Semems "Himmelskorper" konnen etwa als "korper", "kosmisch", "untersuchungsgegenstand der beobachtenden astronomie" gefaβt werden, so daβ ein Allosem "Individuum, das auf deutsch konventionell "Himmelskorper " heiβt und ein kosmischer Korper ist und Untersuchungsgegenstand der beo­ bachtenden Astronomie ist" existiert. Fur dieses Allo­ sem gibt es eine Paraphrase, die eine Pd des S-Lexems 'Himmelskörper' ist und lautet "kosmischer Korper, der von der beobachtenden Astronomie untersucht wird" bzw. "kosmischer Korper, der Untersuchungsgegenstand der beo­ bachtenden Astronomie ist", deren signifikative Bedeu­ tung die Sememe "kosmisch", "Korper" usw. umfaβt. Wenn man in die Paraphrase (lb) anstelle von "Him­ melskorper" die eben angefüihrte Paraphrase einsetzt, entsteht eine Pd zum S-Lexem "Planet", die als eine Padraphrase hoherer Ordnung bzw. als eine gegenuber "Pla­ net" nicht unmittelbare (vermittelte) Paraphrase betrachtet werden kann:

GERT JAGER

70

(lc) kosmischer, von der beobachtenden Astronomie untersuchter Korper, der keine eigene Strahlung aussendet und sich ... bewegt Da (lc) mit (la) und (lb) denotativ identisch ist und theoretisch die Menge der Paraphrasen mit jeweils gröβerer "Tiefenschärfe" unbegrenzt ist, ist die am Anfang dieses Abschnitts gestellte Frage nach der möglichen Identitat von signifikativer und denotativer Bedeutung bei Pd negativ zu beantworten: Die denotative Bedeutung ergibt sich als ein bestimmter Ausschnitt aus dem gesamtgesellschaftlichen Wissen mit einer latenten, theoretisch unbegrenzten internen Gliederung, von denen eine bestimmte signifikative Bedeutung immer nur eine betrifft bzw. ausmacht. Praktisch ist die Moglichkeit der Paraphrasierung von S-Lexemen durch Paraphrasen hoherer Ordnung zweifellos stark begrenzt. Da der tibergang zu einer gröβeren "Tiefenschärfe" jeweils einen tibergang von Simultanem zu Konsekutivem darstellt und damit die mit einem S-Lexem gegebene Moglichkeit zur ganzheitlichen Erfassung der denotativen Bedeutung mit zunehmender Komplexitat der Paraphrase abnimmt bzs. verschwindet. 4. Die fur die Erfassung einiger Aspekte des gegenseitigen Verhaltnisses von S-Lexemen und ihren Pd bzw. Pb unter 1. - 3. getroffenen Feststellungen und Annahmen beruhren grundsatzliche Fragen der Bedeutungskonzeption und werfen eine Reihe weiterer Probleme auf, beispielsweise die Fragen, wie das Verhaltnis zwischen anderen Lexemen und ihren Paraphrasen ist, ob jedem semantischen Merkmal ein Semem mit gleichem Bezug entspricht und urn-

LEXEM UND PARAPHRASE

71

gekehrt (und damit auch die Frage nach der Endlichkeit des Merkmalinventars) oder welcher Zusammenhang zwischen der "Tiefenschärfe"-Annahme und dem Konzept der semantischen Netze besteht. Die Beriicksichtigung dieser und anderer Fragen iibersteigt die Moglichkeiten des vorliegenden Beitrags, der nicht eine Bedeutungskonzeption vorstellen soll, sondern ein wesentlich bescheideneres Ziel verfolgt: das gegenseitige Verhaltnis von S-Lexemen und ihren Paraphrasen in einigen Aspekten zu erhellen, wobei sich die allgemeinen Annahmen nur als fur diesen Zweck geeignet erweisen sollten. Aus unseren sehr gedrangten und fragmentarischen Überlegungen ergibt sich, daβ S-Lexeme und ihnen denotativ aquivalente Paraphrasen (Pd und Pb ) sich hinsichtlich ihrer signifikativen Bedeutung in mindestens drei Aspekten unterscheiden: Anwesenheit/Abwesenheit einer den gesamten Ausdruck betreffenden Bedeutungskomponente "x, das in L. konventionell "expr" h e i β " , Korrespondenz der Sememsequenz einer P. oder Pb mit nur einem Allosem (und nicht dem Semem) des S-Lexems und Grad der "Tiefenscharfe". Daraus lassen sich zweifellos noch weitere Unterschiede ableiten, beispielsweise der, daβ S-Lexeme Einheiten der Sprache sind und Paraphrasen dieser Status nicht zukommt. Eine Folgerung erscheint uns allerdings noch besonders hervorhebenswert: Die vom Jubilar, dem der vorliegende Beitrag gewidmet ist, mehrfach betonte Notwendigkeit, denotative Bedeutung (factual knowledge) und signifikative Bedeutung (linguistic meaning) zu un­ terscheiden (vgl. z.B. P. Sgall (1981)), gilt grundsatzlich auch fur das Lexikon.

72

GERT JAGER

LITERATUR Jager, G. 1976. "Zum Problem von 'Lücken' und 'Umschreibung' bei der Translation". Beiträge zur konfrontierenden Sprachwissenschaft Hrsg. von E. Eichler, J. Filipec, B. Havranek & R. Ruzicka, 42-57. Halle (Saale). Jäger, G. 1983. "Theorie der sprachlichen Bedeutungen und Transla­ tion". Semantik und Übersetzungswissenschaft (Ubersetzungswissenschaftliche Beiträge 6) Hrsg. von G. Jager & A. Neubert. Leipzig. Jäger, G. & B. Koenitz. 1980. "Semantische Eigenschaften sprachlicher Mittel". Linguistische Arbeitsberichte 28, 32-44. Leipzig: Sektion Theoretische und angewandte Sprachwissenschaft der Karl-Marx-Universitat Leipzig. Klix, F. 1971. Information und Verhalten. Berlin. Koenitz, B. 1983. "Grundsätzliche Aspekte der Bedeutungsstruktur definitorischer Sätze". Semantik und VJbersetzungswissenschaft (Übersetzungswissenschaftliche Beiträge 6) Hrsg. G. Jäger & A. Neubert. Leipzig. Lang, E. 1977. Paraphraseprobleme I. Ü b e r verschiedene Funktionen von Paraphrasen beim Ausfhren semantischer Analysen. Linguis­ tische Studien. Riehe A.Arbeitsberichte 42. Berlin: Akademie der Wissenschaften der DDR, Zentralinstitut für Sprachwissen­ schaft. 97-156. Berlin. Leibniz, G. W. 1960. Fragmente zur Logik. Berlin: von F. Schmidt. Lorenz, W. & G. Wotjak. 1977. Zum Verhältnis von Abbild und Bedeutung. Berlin. Sgall, P. 1981. "The Level of Linguistic Meaning". The Prague Bul­ letin of Mathematical Linguistics 35. 3-40. Praha: Univerzita Karlova. Viehweger, D. u. a. 1977. "Probleme der semantischen Analyse". Studia grammatica XV Hrsg. W. Motsch & J. Kunze. Berlin.

II. THE SENTENCE AND ITS STRUCTURE

OLGA MlŠESKA TOMlĆ Skopje

THE DEPTH OF DEEP STRUCTURE

Generative grammar has brought about a major change in the objectives of the study of language; it has opened the doors for theory building subject to conditions of explanat­ ory adequacy. Nevertheless, by many, it is most readily associated not with theory building itself but with one of its constructs - deep structure. As a feature of current linguistic analysis the concept of deep structure is so salient that people assume it to be a makeshift of the pres­ ent epoch, forgetting (or not wanting to know) that underly­ ing/surface structure distinctions have been made throughout history, that they are an integral part of the mentalist trend in language study, whose adherents have ever since antiquity maintained that universal categories of mind under­ lie all the languages of the world. In Cartesian Linguistics (Chomsky 1966b) and Language and Mind (Chomsky 1968) Chomsky suggests that the Port Royal Grammaire generate et raisonnee (Arnauld and Lancelot 1803 (1660)) can be viewed as a spiritual ancestor of his Syntact­ ic Structures (Chomsky 1957) and the deep/surface structure distinctions he and the linguists he has inspired have been making. This suggestion arose great interest in the seventeeth century French rationalists among historians of ling­ uistics, who have come to agree that the Port-Royal grammar­ ians have made a valuable contribution towards the develop­ ment of generativism (cf. Percival 1976; Salus 1976; Lakoff

76

OLGA MIŠESKA TOMIC

1976). Indeed, Arnauld and Lancelot (1803 (1660)) emphasize that the concept of "simple syntax" (i.e. deep structure) is necessary if one is to have the idea that language is basically rational; this concept, according to them, helps to resolve "irrationalities" in "figurative" (i.e. surface) constructions, which constitute the syntax people are conscious of and the one they try to account for.1 But it is doubtful that the simple/figurative distinction in Grammaire generate et vaisonnee was developed under the influence of Descartes, as Chomsky seems to assume when he calls the Port Royal linguistics "Cartesian". Robin Lakoff (1976) points out that Lancelot and Arnauld never mention Descartes but pledge indebted for their ideas to the Spanish Renaissance teacher of Latin, Sanctius, and to his concept of "logical language", that contains items not found in the specific, "illogical" languages which are 2 actually spoken . Nor was the Renaissance teacher of Latin ante quam nihil; theories do not arise in the vacuum. Sanct­ ius himself justifies his metaphysical principles of lang­ uage analysis by appealing to the authority of Aristotle and Plato , but variants of his theory of ellipsis and his belief that the meaning and the syntactic structure of the sentence are fully accounted for only if every missing element is restored, can be traced throughout the period intervening between the Classics and the Renaissance scholars (Percival 1976, Salus 1976, Trentman 1976). As a matter of fact, the general notion of understood elementsis part and parcel of the mentalist philosophical tradition, which has been thriving since antiquity with interchange­ able peaks and pitfalls. Next highest to the "Cartesian" was most probably the peak of the Modistae, the thirteenth century students of language who required that grammar be based on something beyond language, though within the real

THE DEPTH OF DEEP STRUCTURE

77

world; according to Salus (1976), Modistic grammatical theory depended upon relationship of the closest type between the operation of the mind and the structure of real­ ity, its central concerns being universals, adequacy, the incorporation of meaning into general grammatical statements

and deep

structure.

In the "post-Cartesian" period, mentalism receded. But the notion of deep structure was not wiped off the field of language study. Both in his Aspects

and in the Topics

(Chomsky 1965: 198-199)

in the Theory of Generative

Grammar (Chom-

sky 1966a: 7) Chomsky finds a very strong connection between his deep structure and Humboldt's innere

Sprachform.

Though

Coseriu (1970: 215) was most probably correct when he argued that the results of Chomsky's attempts to come to grips with Humboldt's very involved linguistic theory was not "der historische Wilhelm von Humboldt" but at best "ein hybrider

Noam von Humboldt", the connection of generative

grammar and Humboldtian linguistic though should not be overlooked. Present-day views of deep structure differ sub­ stantially from those of Humboldt: while the nineteenth century philosopher stayed within the boundaries of lang­ uage, the contemporary generativists explore the logical structure of the mind. In essence, what generative grammar regards as deep structure is referred to by the Port-Royal school as "logical structure". But the seventeenth century rationalists do not fathom the depths of the logical structures of the mind; they go only as deep as the surface structure of language allows. Thus, traditional mentalistic deep structures differ distinctly from most contemporary deep structures, which endeavour to represent what is going on in the depth of the speaker's mind. The ideas that there are principles that are universal in human language and that one must, consequently, distinguish surface from deep

OLGA MIŠESKA TOMlĆ

78

structure, have been held by grammarians of the mentalist 4 trend through ages. But, until recently, they were clear­ ly and unmistakably characterised as ideas that are the concern of logic. Generative deep structure, or rather structures, might be considered logical; they may, however, be claimed by psychology or else be defined as strictly linguistic. In the first generative model, that of Syntactic Structures

(Chomsky 1957), the deep structure was very

shallow. Deep/surface structure distinction was there established between the basic sentence patterns, "kernel strings", and the sentences which were in some way related to them.5 The whole set-up was strictly formal: Related syntactic structures were represented through the simplest among them - the structures of the simple declarative indicative sentences. The other structures were derived from the latter through formally specified operations transformations. In the Standard Generative Model (Chomsky 1965) deep structure came to be defined as a formal object specified by the application of three sets of rules: phrase structure rules, subcategorization rules and rules of lexical insert­ ion. The phrase structure rules generated phrase markers, the subcategorization rules specified the lexical items, the rules of lexical insertion substituted the complex symbols containing these specifications, provided their syntactic features were indistinct from those of the latter. While the subcategorization and lexical insertion rules produced elements which were associated with corresp­ onding

lexical material, the phrase structure rules gener­

ated not only phrase and word class constituents but also symbols whose single role was to trigger obligatorily the transformations which were originally formulated as mean-

THE DEPTH OF DEEP STRUCTURE

79

ing changing optional rules. Thus, in the Standard Model, the deep structures became more distant from the respect­ ive surface structures than it was the case in the Syntactic Structures model; they became really deep. The freedom to set up hypothetical underlying elements which might desappear without a trace in the surface struct­ ures appealed to many linguists, not only because it appear­ ed to be a truer recognition of the complex relationship between deep and surface structure but also since it re­ cognized frankly the abstract nature of all linguistic elements. For the generative semanticians this appeal was so strong that adequacy of linguistic description came to be measured by degree of abstractness, which, in its turn, was identified with degree of distance of deep structure from surface structure or degree of closeness of deep 7 structure to semantic representation. McCawley, Lakoff, Ross, Postal... claimed that the more abstract (i.e. more deep) the underlying representation the larger the number of significant generalizations it can capture (cf. McCawley 1968a, 1968b; Lakoff 1971; Lakoff and Ross 1976 (1967)). In line with this claim, they derived the more complex lexemes from their respective primitive predicates, which lead not only to a minimisation of initial input but also to a drastic reduction in the inventory of grammatical classes. Thus, the generative semantics' deep structure was successful both in justifying the raison d 'etre of a twolevel syntactic description - generalizations - and in satisfying the basic principle of scientific theories simplicity. Nevertheless, there had to be a compensation for this success - a very complicated transformational component, which, to boot, was too powerful (in spite of all the constraints imposed upon the derivation). Fighting overgeneration, the group of linguists who

80

OLGA MIŠESKA TOMIĆ

subsequently got to be known as lexicalists, twisted the derivations of the generative semanticians off the smooth path, which progressed directly from semantic representat­ ion to surface structure, by pulling deep structure upwards and introducing a very rich interpretive component. The strict lexicalists unburdened the former transformational component almost completely.8 Their interpretive and redundancy rules, however, mirror in crucial ways the distinguishing features of the transformational component of early generative semantics; while the former rules are varients of transformational deletions, the latter relate structures with different order of basically same constit­ uents. The original deep structure was gradually being made more shallow, to the point of overlapping with surface structure. But the underlying/surface structure distinction has not disappeared: the lexicon, where the computing of functional and semantic selections from structures which in earlier models would have been considered transformat­ ionally derived takes place, has actually been turned into a new site for deep structure - a really deep one, too. Chomsky himself currently allows and works with two types of transformations: (a) syntactic a-movement rules (applying to noun phrases and question words) which map deep structures (D-structures) onto abstract surface structures (referred to as S-structures), which contain indexed empty categories and (b) a variety of stylistic and clean-up rules, mapping the latter structure onto sur­ face forms (Chomsky 1981). Thus, instead of one, he now has two deep structures, differing in degree of depth but both of them providing input for the logical form, which takes care of the semantic interpretation. In addition, in his lexicon there are predicate-argument, "thematic" structures, which are in one-to-one correspondence with

THE DEPTH OF DEEP STRUCTURE

81

the D-structures. The whole approach is highly modular, in the sense that the full complexity of observed phenomena is traced to the interaction of partially independent subtheories of a set of primitives, axioms and rules of infer­ ence which make up universal grammar. Like deep structure, universal grammar has its roots in Classical philosophical tradition. Actually, Plato's notion of "ideal form", effectuated "in various (imperfect) ways in this transient world" (Salus 1976: 85) has given rise to both deep/surface and universal/individual distinct­ ions in linguistic theory. Nevertheless, the interrelation­ ships of these two distinctions have hardly ever been explicitly discussed. Even Chomsky, one of the few linguists who up to date make extensive use of both notions, doesn't speak about their relative coverage. For him universal grammar is a theory developed through "abstracting from 9 successful grammars and successful theories of grammar those more general principles that account for their success" (Chomsky 1981:2); when applied to individual languages, these principles are to be assisted by principles of "peripheral grammar", which are theory internal in the sense of being structured by "markedness theory", itself part of universal grammar. Deep structure, in its turn, is a component of "core grammar", a mediator between the representations of form and those of meaning. Since core grammar, according to Chomsky (1981), is one of the success­ ful grammars over which universal grammar abstracts, it follows that the notion of deep structure is subordinate to that of universal grammar. Originally, deep structure was a component of syntax, so much so that when the generative semanticians tried to make it coextensive with semantic representation, the lexicalists pleaded for "autonomous syntax". Subsequently, both

82

OLGA MIŠESKA TOMIĆ

lexicalists and generative semanticists have changed attitude. Thus, Chomsky extends syntax to cover many semant­ ic and even cognitive phenomena, defining it as "the computational aspects of language, meaning the system of rules that give certain representations and derivations" (Chomsky 198 2: 114). Kis deep structure is in one-to-one relationship with "thematic" relationships, highly remin­ iscent of Fillmore's (1968) deep cases. Fillmore, in his turn, is reexamining his cases; in his more recent papers, he expresses the view that the relationships which he used to express as deep cases actually fall without the language system in the strict sense (they belong to the realm of conceptualization) and assumes that the deep structure of language proper is set up by creating conditions under which a speaker choses to draw some of these relationships into perspective (cf. Fillmore 1977). In his more recent writings, Sgall also distinguishes meaning, taken as the linguistic counterpart of intentional structure of sense, from cognitive content, i.e. from the language independent patternings of factual knowledge (cf. Sgall 1980a, 1980b). He claims that, in his multi-level functional generative model, the predicates and particip­ ants of the deep, "tectogrammatical" structure, which can be considered as "an explicit representation of the mean­ ing of the sentence" (Sgall 1978: 45), should be set off the cognitive level "case roles" which designate (cyclo­ paedic) knowledge about the world. The deep (tectogrammat­ ical) structure itself is conceived as a dependency graph featuring a predicate with its arguments (Tesnière's aotants) and free adverbials (Tesnière's oiroonstants) . The relationships of the arguments to the predicates are being defined as the intentions of the respective predicat­ es. Sgall (1978) maintains that their meanings are determ-

THE DEPTH OF DEEP STRUCTURE

83

ined through translation into a "semantic language" some device more or less close to Carnap's intentional structure, which relates deep structure to cognitive content. Panevova' (1980) , in her turn, speaks of a natur­ al ordering of types of arguments, which requires shifting 12 of possible argument candidates. The relationship of this shifting to semantic interpretation is, however, not discussed. Unlike Chomsky, who assumes that the topic/focus distinctions have to be handled by surface rules, Sgall believes that the deep structure level should be concerned not only with semantic but also with pragmatic issues (in terms of Morris's and Carnap's tripartition), In line with this belief, the deep (tectogrammatical) structures of the functional generative model accounts for communicative sentence perspective distinctions and that not only in terms of a dichotomy that accounts for topic/focus struct­ ure, but also through two communicative scales: the scale of communicative dynamism and the scale of communicative importance. While the former scale is denoted by the linear ordering of elements, the latter is marked by super­ scripts depicting the respective elements as contextually bound or contextually unbound. The focus of the sentence is defined as that part of the respective deep structure "which includes all and only the contextually non-bound elements of its main bundle (consisting in the main verb and its immediate modifications in terms of dependency syntax) and all nodes governed by (i.e. directly or in­ directly dependent on) these elements" while the topic is "identical with the rest of the sentence, i.e. with the contextually bound elements of the main bundle and all nodes governed by them" (Sgall & Hajičová 1978: 36). One might say that Sgall's deep structure is as deep

OLGA MIŠESKA TOMIĆ

84

as that of the Standard Model; like the latter, it is the level which, on one hand, offers information for semantic interpretation, while, on the other, it is the starting point for changes leading to surface phonetic forms (via a set of transductions). Yet, it deals with phenomena which Chomsky treats as surface ones! But is the handling of questions concerning topic/focus distinctions by styl­ istic rules unquestionable? In English, the communicative sentence perspective may not be as perspicuous, but to the analyst of a Slavonic language a deep structure that serves as a basis for setting up possible representations in which the semantic aspects of topic/focus distinctions are accounted for, stands to reason. The research undertaken in connection with the Project for Automatic Generation of Serbocroation 13 has indicated that it would be necessary to work with a syntax that makes possible the computation of the semantic differences between sentences which differ with respect to their communicative perspectives. One has also been lead to conclude that, if native speaker's reactions are to be taken into considerat­ ion, an overt distinction between communicatively unmarked sentences, on one hand, and communicatively marked ones, on the other, would have to be made. Consider the Serbo-Croatian sentences: (1) (2) (3)

Petar je ponudio Nadi č a j . Petar je ponudio caj Nadi. čaj

je ponudio Nadi Petar. je Petar ponudio Nadi.

(4)

čaj

(5)

*čaj je ponudio Petar Nadi.

(6)

Nadi je ponudio caj Petar.

(7)

Nadi je Petar ponudio č a j .

(8)

*Nadi je ponudio Petar č a j .

THE DEPTH OF DEEP STRUCTURE

85

These sentences consist of same number of same lexemes with same syntactic functions; in each of them the proper names Petav and Nadi (dat. of Nada) are the subject and indirect object, respectively, the common noun čaj, "tea", is the direct object, while the auxiliary + participle structure je ponudio (3rd. p.s.m. non-durative perfect of ponuditi, "offer") features the predicate. The topics and foci (which like in all the Slavonic languages, are denot­ ed by initial and final sentential positions, respective­ ly) are, however, not represented by same lexemes. In the communicatively unmarked sentence (1) the topic coincides with the subject, the focus - with the direct object. In the marked sentences, either the topic is not coextensive with subject or the focus is not represented by the noun representing the direct object or else both topic and focus are featured by nouns differing from those featuring the subject and the direct object, respectively. The overlappings and discrepancies between the syntactic and commun­ icative functions can be depicted if we assume a semantically interpretable deep structure nucleus consisting of a predicate and up to three nominal arguments, in which both predicate/argument and communicative relationships are signalled. The nominal arguments reflect the features of the lexical frames of the respective predicates. In a three-argument deep structure nucleus the first argument, Nomin­ al One, would be the nominal from which the action or the state denoted by the predicate originates, the second argument, Nominal Two - the nominal towards which the act­ ion or state featured by the predicate is directed, while the third nominal, Nominal Three, would be any third part14 icipant in the action/state of the predicate. The nominals would be ordered: Nominal One would precede the predic­ ate while Nominal Three and Nominal Two, in this order,

86

OLGA MISESKA TOMIC

would follow it. When Nominal One and Nominal Three carry respective topic and focus signals the deep structure ordering of nominal elements corresponds to the underlying one; otherwise, that is not the case. We tend to assume that communicatively marked sentences are derived through topicalization and focusing of deep structure element carrying respective topic and focus signals. Thus, (2) is derived by focusing of Nominal Three, (3) - by topicalizat­ ion of Nominal Two and focusing of Nominal One, (4) - by topicalization of Nominal Tv/o and focusing of Nominal Three, (6) - by topicalization of Nominal Three and focusing of Nominal One and (7) - by topicalization of Nominal Three. Sentences (5) and (8), in which Nominal One is moved to a position other than the focus, are not acceptable. By featuring communicative sentence distinctions in semantically interpretable deep structures, by setting up a basic word order and deriving communicatively marked sentences through movement rules, we have reacted positive­ ly to native speaker's feelings: When asked to make a sent­ ence out of the lexical items in (1) - (8), speakers of Serbo-Croatian unfailingly come up with (1). The other sentences are judged as formed through movement of nominals in order to emphasize one thing or another. If native speakers judgements are criteria for adequacy of descript­ ion (as they have been admitted to be) our deep structure is adequate. What is more, topicalization and focusing have a "theory internal" justification, too; the derivation of Serbo-Croatian sentences requires other movement rules, one of them being the Clitic Placement Rule, which moves clitics to a position immediately following the topic. Having set the general goal of motivating linguistic theory and its formalization by means of descriptions subject to conditions of explanatory adequacy, Chomsky engag-

THE DEPTH OF DEEP STRUCTURE

87

ed in demonstrating the qualities of the grammar which meets these conditions. All his arguments are, however, theory bound. He defines his universal rules as properties of some physical system of the brain and builds up his grammar on the basis of them, relying heavily on simplicity as a criterion of adequacy. In reaction to Bresnan's (1978) contention that the adequacy (reality!) of grammars should be evaluated through their ability to be used in a compet­ ence-based model of language aquisition, comprehension and production, Chomsky takes the view that psychological real­ ity is no more than whatever linguistic theory is about (Chomsky 1982). But how can a theory be evaluated by a metric such as simplicity, which is the product of that theory. Although contemporary linguistic theory develops under the strong influence of Chomsky, it often goes off the lines along which he himself has been progressing. At the present, the field of grammatical research is approached by quite a number of grammarians following the generative trend and proposing generative models of description. We cannot but go along with Edith Moravcsik's (1980) statement that these models can and often do differ with regard to (a) the kinds of facts they account for, (b) the goals that these facts are to achieve and (c) the range of means employed in the pursuit of these goals in relation to the relevant facts. Some models are models of linguistic competence, while others are models of performance and take into considerat­ ion psycholinguistic and sociolinguistic evidence. There are models which assume that communicative perspective dist­ inctions are semantically relevant, and other that treat these distinctions as stylistic variations or do not cover them at all. Certain models work within a limited domain of language, believing that one creates grammar through con-

88

OLGA MISESKA TOMIC

structing fragments (cf. Partee 1978; Cooper 1980), whereas others tend to be universal. Some models provide for the construction of statements which are tested through matchings with structures in reality, whereas others start from a putative reality, to be evaluated against a decription. Which is the "right" model? Or, as a matter of fact, is there any? In spite of all its scientific methods, linguistics still is a humanitarian discipline, results of which cannot be experimentally tested (by making cuts into the brain, for example). In deciding what is and what isn't linguistically significant generalization one has, as far as we can see, three possibilities. One possibility is to rely heavily on native speaker's intuition and aim at grammars which stand the test of models of language acquisition, comprehens­ ion and production. Another possibility is to seek for some external justification of explanations, by fitting linguist­ ic theory into more general theories of human behaviour or looking into its relationship with other aspects of lang­ uage. A third possibility, probably the most eligible one, is to combine the first two. It is highly improbable that any model of grammar would ever be in one to one correspond­ ence with the grammar itself (the one human beings internal­ ize) . But the evaluations of the success of work since the publication of Syntactic Structures indicates that linguist­ ic description is on the track that leads to a good approx­ imation of linguistic reality. The variability of the deapth of deep structure is a testimony of the efforts for achiev­ ing a better approximation, rather than a sign of indeterm­ inacy.

THE DEPTH OF DEEP STRUCTURE

89

NOTES 1 The Port-Royal grammarians assume that syntactic relationships between the languages are to be ascribed to the logic underlying a language. While all languages may have an identical logical basis, the superficial forms are different in each of them; the order of the words may often not be logical unless some elements are supplied by the hearer or reader. According to Robin Lakoff (1976), while in Grammaire generale et raisonnee the logical basis is only referred to, in Lancelot's Nouvelle method pour apprendre facilement et en peu de temps la langue latine (first published in 1664) one finds a great many examples of underlying forms, some of them very close to the underlying structures which the generativists set up today. 2 In his discussion with Riny Huybregts and Henk van Riemsdijk, Chomsky (1982: 39) refutes the accusation that he has overlooked the Renaissance origins of Port-Royal Grammar. He points out that he did mention this the very time he introduced the Port-Royal Grammar in Cartesian Linguistics (Chomsky 1966b). This mention must have, how­ ever, been drowned in the overwhelming attention paid to the French rationalists. 3 According to Percival (1976), in Minerva: seu de causis linguae Latinae, Salmanticae: opud Ioannem and Andream Ronault fratres, 1587, Sanctius quotes Plato's Cratylus as a support for his argument that language is rational. The argument goes like this: "The universe is an ordered cosmos and language was deliberately created by man to enable him to refer to the world around him. The relation between word and referent is, therefore, a natural one (a natura) , not an arbitrary one (ab a r t e ) " (Percival 1976: 241). 4 They have had reflexes on the thinking of some non-mentalists, as well. Thus, Hockett (1958: 294) in one of the many passages in which he reached inspired conclusions, never to be developed into an actual analysis, refers to a distinction between "surface and deep grammar of a sentence". Chomsky has, most probably, appropriated (with slight modification) Hockett's terminology. 5 In Syntactic implicit.

Structures

the concept of deep structure was only

6 The subcategorization rules could be of three general types: context free rules - subcategorizing the word classes, strict subcateg­ orization rules - providing the syntactic frames of the nominals and selectional rules - specifying the predicates in terms of the syntactic features of their respective subjects and objects. 7 The generative semanticists actually insisted on the complete elimination of the level of deep structure, which they considered to be a superfluous interlevel between the levels of semantic representat­ ion and surface structure (Lakoff and Ross 1968). Justification for

90

OLGA MIŠESKA TOMIĆ

deepening the deep structure was found in Chomsky's statement that the formalism for capturing strict subcategorization and selectional restrictions in deep structure contributes to simplicity of grammar (Chomsky 1965), as well as in Katz - Postal's hypothesis that trans­ formations do not change meaning (Katz & Postal 1964). 8 Bresnan (1982) allows for two "transformational relations" to remain: a class of unbounded movement or deletion rules and a class of bounded rules that change the order but not the function of constit­ uents (scrambling rules, according to her). The Lexical-Functional Grammar she proposes, purports to provide a coherent explanation of how surface structures are related to representations of meaningful grammatical relations. Each sentence is provided with dual represent­ ations, consisting of a constituent structure and a functional struct­ ure. The constituent structure represents the superficial constituency of the sentence (which is phonologically interpreted) while the funct­ ional structure, computed in the lexicon, is the representation of its meaningful grammatical relations (which are semantically interpret­ ed) . 9 Chomsky maintains that the theory of grammar is fundamental since there is something in the brain that corresponds to it. Language itself is epiphenomenal - it is one among millions of things generated by grammar. "There is no obvious reason why any of the systems gener­ ated by this grammar should be regarded as more real or significant than any other. In fact, the notion of language might turn out just to be useless notion" (!) (Chomsky 1982: 107). 10 "The distinction (between universal and peripheral grammar) is in part theory-internal but that is unavoidable and quite reason­ able. I am sure that the periphemes and the theory of markedness have structure, but I just do not have any good ideas about what it should be. I suggested in the Pisa paper (cf. Chomsky 1981) which I do not think is correct, viz. relaxing some of the conditions of core grammar. May be that is somewhat the right idea but I do not really feel that there is any evidence. I do not even think it is clear whether we should make a sharp distinction between core and periphery. May be these are more closely related notions of some sort. But, whatever the answer to this is, it is obviously going to be the case that learning of exceptions is a highly structured matter. You cannot learn just any class of exceptions." (Chomsky 1982: 108). 11 According to Sgall (1978: 27) semantics for a logician is an attribute referring to interpretation whereas for the linguist it is an attribute of that level of description of natural language which can be a starting point for semantic interpretation in the logical sense of the word. 12 Panevova (1980) works on the assumption that if a predicate has only one argument it should be an agent, if two - agent and

THE DEPTH OF DEEP STRUCTURE

91

patient. When the dictionary entry of the respective predicate does not have an agent and/or a patient, there comes to a shift according to the following scheme:

(Panevova 1980: 45)

13 The Project is undertaken by the Institute of Mathematics, Belgrade. 14 In surface structure, Nominal One is featured by subjects, Nominal Two - by direct objects or subjective complements, Nominal Three - by indirect objects or objective complements. In Serbo-Croat­ ian, these are realized by specific inflections or preposition + inflection formemes. 15 Note, though, that when the predicate itself is the topic, Nominal One comes right behind it. Thus we have: Ponudio je Petar Nadi čaj. 16

See (1) - (8) and the example in note 15.

REFERENCES Arnault & Lancelot. 1803 (1660). Grammaire generale et raisonnee de Port-Royal Precedee d'un Essay sur 1'origine et les progrès de la Langue francaise, par M. Petitot, et suivie du Commentaire de M. Duclos, auquel on a ajoute des Notes. Paris. Bresnan, J. 1978. "A Realistic Transformational Grammar". Linguistic Theory and Psychological Reality ed. by M. Halle, J. Bresnan & G. Miller. 1-59. Cambridge, Mass.: MIT Press. Bresnan, J. 1982. "The Passive in Lexical Theory". The Mental Repres­ entation in Grammatical Relations. ed. by J. Bresnan. 3-85. Cambridge, Mass.: MIT Press. Bresnan, J. & R.M.Kaplan. 1982. "The Problem of Psychological Reality of Grammars". The Mental Representation in Grammatical Relations ed. by J. Bresnan. xvii -iii. Cambridge, Mass.: MIT Press. Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press. Chomsky, N. 1966a. Topics in the Theory of Grammar. The Hague: Mouton. Chomsky, N. 1966b. Cartesian Linguistics. New York: Harper & Row. Chomsky, N. 1968. Language and Mind. New York: Harcourt, Brace & Jovanovitch. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris Publications.

92

OLGA MIŠESKA TOMIĆ

Chomsky, N. 1982. Noam Chomsky on the Generative Enterprise, a discussion with Riny Huybregts and Henk van Riemsdijk. Dordrecht: Foris Publications. Cooper, R. 1980. "Montague Syntax". Syntax and Semantics 13: Current Approaches to Syntax ed. by E.A.Moravcsik & J.A. Wirth. 19-44. New York: Academic Press. Coseriu, E. 1970. Sprache: Strukturen und Funktionen. Tübingen Beiträge zur Linguistik. Fillmore, C.J. 1968. "The Case for Case". Universals in Linguistic Theory ed. by E. Bach & R. T. Harms. 1-88. New York: Holt, Rinehart and Winston. Fillmore, C. J. 1977. "The Case for Case Reopened". Klassifikation und semantische Interpretation ed. K. Heger & J. Petöfi. 3-26. Ham­ burg : Buske. Hockett, C. F. 1958. A Course in Modern Linguistics. New York: Macmillan. Katz, J. J. & P. M. Postal. 1964. An Integrated Theory of Linguistic Description. (Research Monograph 26). Cambridge, Mass.: MIT Press. Lakoff, G. 1968. "Deep and Surface Grammar". Indiana University Club Publication. Lakoff, G. 1971. "On Generative Semantics". Semantics ed. by D. Stein­ berg & L. Jakobovitz. 232-296. Cambridge University Press. Lakoff, G. & J. R. Ross. 1976 (1967). "Is Deep Structure Necessary?". Syntax and Semantics Vol. 7 ed. by J. McCawley. 159-164. New York: Academic Press. Lakoff, R. 1976. "La Grammaire generale et raisonnee ou la grammaire de Port-Royal". History of Linguistic Thought and Contemporary Linguistics ed. by H. Parret. 348-372. Berlin: De Gruyter. McCawley, J. 1968a. "The Role of Semantics in Grammar". Universals in Linguistic Theory ed. by E. Bach & R. Harms. 125-170. New York: Holt, Rinehart and Winston. McCawley, J. 1968b. "Concerning the Base Component of a Transformat­ ional Grammar". Foundations of Language 4. 243-269. Panevova, J. 1980. Formy a funkce ve stavbe ceske vety. Praha: Academia. Partee, B. Hall. 1978. "Logic, Language and the Mind", special lecture at the University of Massachusetts. Amherst, Mass. Percival, W. K. 1976. "Deep and Surface Structure Concepts in Renais­ sance and Mediaeval Syntactic Theory". History of Linguistic Thought and Contemporary Linguistics ed. by H. Parret. 238-253. Berlin: de Gruyter. Prager Autorengruppe. 1973. Functional Generative Grammar in Prague. Kronberg/Taunus: Scriptor Verlag GmbH. Salus, P. 1976. "Universal Grammar 1000 - 1850". History of Linguistic Thought and Contemporary Linguistics ed. by H. Parret. 85-101. Berlin: de Gruyter. Sgall, P. 1978a. "Towards a Pragmatically Based Theory of Meaning". The Prague Bulletin of Mathematical Linguistics 30. 43-60.

THE DEPTH OF DEEP STRUCTURE

93

Sgall, P. 1978b. "Synchronni srovnavanf jazyku a jeho sémantické vychodisko". Slovo a slovesnost 39. 25-32. Sgall, P. 1980a. "Case and Meaning". Journal of Pragmatics 4. 525-536. Sgall, P. 1980b. "Meaning, Context and Pragmatics". The Prague Bulle­ tin of Mathematical Linguistics 6. 216-225. Sgall, P. & E. Hajičovǎ. 1978. "Focus on Focus". The Prague Bulletin of Mathematical Linguistics 23.- 41. Sgall, P., Hajičovǎ, E. & E. Benesova. 1973. Topic, Focus and Gener­ ative Semantics. Kronberg/Taunus: Scriptir Verlag GmbH. Trentman, J. 1976. "Speculative Grammar and Transformational Grammar: A Comparison of Philosophical Presuppositions". History of Linguistic Thought and Contemporary Linguistics ed. by H. Parret. 279-301. Berlin: de Gruyter.

V. ROSENTSVEIG Moscow

LA VALENCE: SÉMANTIQUE ET SYNTAXE

Sollicitant la lexicographie autant que la grammaire, 1'étude des relations predicat-actants eclaire la connexion de faits de langue apparemment fort ecartes les uns des autres. C "est dans ce cadre que nous voudrions presenter en hommage á P. Sgall quelques notes sur la syntaxe et la semantique des verbes francais transitifs et á la fois intransitifs, plus particulièrement le verbe c u i v e 1 . 1. La première concerne la derivation semantique: laquelle des deux formes des verbes en question, la transiti­ ve ou 1 "intransitive, doit servir de base á la description léxicographique? Le calcul des solutions possibles est sim­ ple: si la structure, et sémantique et formelle, du verbe transitif est plus complexe que celle de 1 "intransitif, c'est cette derniére forme qui sera prise pour base de la derivation semantique; si le transitif est plus complexe par sa structure semantique mais plus simple par sa forme, la derivation est bidirectionnelle: pour le sens, c'est le verbe in transitif qui est la base, pour la forme, le verbe tran2 sitif . Il est entendu que la comphéxite formelle sera definie non seulement en termes morphématiques, mais aussi operationnels: le conversif sera considere formellement plus complexe que le verbe de base precisément parce qu"il est obtenu á partir de ce dernier par 1"operation de conversion. Ainsi, cuit en (b) est formellement plus complexe qu "en (a)

96

VICTOR YU. ROSENTSVEIG (1) (a) (b)

La ménagére cuit la soupe La soupe cuit

par analogie avec (il) se l a v e , formellement plus complexe que (il) lave. Reste á definir la relation semantique entre c u i t en (a) et en (b). Il est communément admis que la va­ lence du causatif augmentant celles du non causatif correspondant d "une unite, la definition lexicographique du pre­ mier est egale a celle du second plus le sens de causation. C"est cette régie qu "enonce L. Tesniére: "Si le nombre des actants est augmente d "une unite, on dit que le nouveau verbe est causatif par rapport á l"ancien. Ainsi nous pouvons dire que pour le sens, renvevsev est le causatif de tomber /.../ et montrer le causatif de voir /.../." (1959, p. 260). Pour Tesnière, les causatifs consideres etaient plus complexes aussi formellement: dans les langues germaniques, ils appartiennent a la conjugaison faible, de formation plus tardive que la conjugaison forte, qui est celle des non causatifs; dans les langues qui marquent 1 'opposition causatif- non causatif par un morphème, particulièrement en francais, les causatifs du type avancer, desoendre, vec u l e v , sortir comportent, estimait-il, une flexion zero. On le voit, á la question sur la relation entre le causatif et le non causatif Tesnière, en bonne logique, donnait une reponse univoque: plus simple á la fois pour le sens et pour la forme, le non causatif doit etre considéré 1 "elément de base, le causatif, 1 "element secondaire. Soujacente a cette solution, la supposition que le causa­ tif implique le non causatif, sans que 1 "inverse soit vrai, ne prend pas en compte la distinction entre causation volontaire et involontaire. Or le non causatif d "un causatif volontaire signifiant une action productive implique le causatif. A 1 "oppose des verbes non causatifs signifiant

LA VALENCE: SEMANTIQUE ET SYNTAXE

97

une phase d'un procés involontaire (du type commenser, cesser, finiv) un etat (du type tomber, sortir) et de beaucoup d"autres, les membres non causatifs des paires de verbes d'action productive ne peuvent recevoir une definition lexicographique sans reference aux causatifs respectifs. C'est d"ailleurs, fait éclairant, la pratique des dictionnaires francais: les non causatifs du type cuire y sont d"ordinaire definis par reference á I'action qui est cause du processus declenche. Par ailleurs, sont suggestives aussi les donnees etymologiques: au contraire des verbes non causatifs de phase, d"etat, de perception, entres dans la langue avant leurs causatifs, les non causatifs signifiant 1'effet d'une action de production (au sens large de ce terme) sont attestes postérieurement aux causatifs respectifs. Nous sommes conduits á conclure, revenant a la ques­ tion exemplifiée plus haut, que cuit en (b), plus simple pour le sens, est cependant plus complexe formellement que cuit en (a) puisqu "il en est le conversif. Il en sera donc considere le derive semantique et defini comme un decausatif. A la difference des non causatifs, les decausatifs seront ranges dans le dictionnaire á la suite des causatifs respectifs. Le decausatif ne peut signifier ni le virtuel, ni le passif potentiel, ni 1 "habituel: ces valeurs sont signifiees par le passif pronominal (comp. La viande cuit et La viande se cuit á l ea.u salée) . Il va sans dire que 1 "information pertinente concernant le decausatif sera reprise en grammaire. 2. Restitution idiomatique de la situation designee, la valence semantique du lexéme verbal trouve sa contrepartie en syntaxe. C "est ainsi que le causatif cuire comporte

98

VICTOR YU. ROSENTSVEIG

six valences semantiques et, respectivement, six relations syntaxiques profondes: agentive, objective interieure (la matière de la cuisson), objective exterieure (le produit de la cuisson), instrumentale (l"outil de la cuisson), mèdiale (le corps moyennant lequel se fait la cuisson - eau, huile, etc.), locative-instrumentale (1 'ustensile qui sert á recueillir la matiére en cuisson). Cumulant deux significa­ tions: traitement (de la matiére) et production (d"un ali­ ment) , le causatif c u i r e se classe parmi les lexémes reguliérement polysémiques (Apresjan, 1974). Le comportement syntaxique des verbes francais appartenant á ce groupe, y inclus les "verbes de cuisine", est soumis á diverses contraintes: ainsi, en tant que predicats verbaux, ils ne peuvent regir dans une seule et meme proposition á la fois les objets interieur et exterieur. Pour filtrer de telles constructions le dictionnaire doit dissocier chaque lexéme verbal de ce groupe en deux vocables á modéles de rection differents, 1 'un signifiant "traiter (une matière) ", 1 "aut­ re "produire (un aliment) ". Cette information alimentera en grammaire une règie syntagmatique formulant la distinction traditionnelle entre l"objet efficie et l"objet officiè. 3. Les valences semantiques permettent de distinguer aussi le causatif distant de celui contigu: le premier admet une forme synthetique aussi bien qu "analytique, le second une forme synthetique seulement, la valence agentive ne pouvant etre saturee dans ce dernier cas. Ainsi, on dira, pour le causatif distant, La mènagère cuit /fait cuive/ la soupe, á 1 'exclusion de *Le fourneau (l'huile, la poele) fait bien ouive la poisson et de *Le poisson est bien cuit par le four (1'huile, la poele) . Notons d "autre part qu "un predicatif ne peut exercer la fonction de sujet grammatical qu"aupres un causatif analytique (comp. Une simple pression

99

LA VALENCE: SEMANTIQUE ET SYNTAXE

de la touche pvession

feva

gvillev

de la touohe

votve

gvilleva

poission, votve

et

poisson).

*Une

simple

Ce qu on

appelle un causatif contigu, s'agissant d'une construction avec pour

sujet grammatical un nom qui ne signifie pas la

valence agentive, est, á proprement parler, non pas un cau­ satif mais un operatif.

NOTES

1 Ce verbe a fourni la matière á de multiples analyses. On ne mentionnera ici que celle des constructions factitives par Nicolas Ruwet, 1972, et celle des voix par J.-C. Chevalier, 1979. Pour une de­ scription detaillee des "verbes de cuisine" francais v. Rothemberg, 1974. 2 Pour la definition et le calcul combinatoire des voix on consultera 1'etude de I. A. Mel'čuk et A. A.Cholodovič, 1970.

BIBLIOGRAPHIE Apresian, Ju. D. 1974. Leksičeskaja semantika, noscou. Chevalier, J.-C. 1979. "Le concept de voix dans les langues romanes".

Relations

predicat-actant(s)

dans les langues de types divers

I,,

ed. by Catherne Paris. Paris: SELAF. Mel'cuk, I. A. & Cholodovic, A. A. 1970. "K teorii grammatičeskogo za-

loga". Narody Azii i Afriki 4. 111-124. Rothemberg, M. 1974. Les verbes transitifs et á la fois

intransitifs.

Paris: The Hague.

Ruwet, Nicolas. 1972. Theorie syntaxique et syntaxe du francais, Paris. Tesnière, Lucien. 1959. Elements de syntaxe structurale. Paris.

RUDOLF Leipzig

CONTROL IN COMPETING FRAMEWORKS

If "the central new idea that generative grammar brought to linguistics was that the study of precisely formalized systems of grammatical rules would yield interesting empirical results about the nature of human language" (Pullum 1983: 447), Petr Sgall's work must be counted among the earliest pioneering contribut­ ions to this momentous theoretical approach (cf. Sgall e.g. 1964, 1966, 1967). It has been detrimental to the development of syntactic and semantic metatheories that "linguistic ideas cross the ocean" (Sgall 1982: 399) with great difficulty. Living an age of healthy diversity in the field of generative grammar, linguists on both sides of the ocean would, of course, have preferred a lively exchange of ideas on the nature of the differences and the degree of agreement among their competing theoretical approaches. The problem with unexplored and unanalyzed differ­ ences among various frameworks brought forward within some common understanding of basic substantial concepts of generative grammar lies in the absence of a clear assessment of their relative values, or at least in the absence of a clear understanding of how exactly

102

RUDOL F

they differ. As a consequence, linguists may forfeit the chance of relating metatheories to each other when they are compatible and/or complementary in their data, goals or means, (cf. Moravscik & Wirth 1980). A wide range of grammatical work undertaken within different frameworks is available now for a comparative study and assessment. For example, in recent work on grammat­ ical theory, phenomena of control have been coming into focus. In particular, studies on control have been published among others by Chomsky (1980, 1981), Bresnan (1982), (1983a, 1983b), Abraham (1982) and Manzini (19 83). Control phenomena have turned out to be a point of intersection where rather different approaches meet and may be related to each other with a view to their comparative assessment. The first part (section 1) of this paper will be given over to a generalizing survey of the approach to control presented in R 1983a, 1983b. The sur­ vey will be accompanied by a comparative judgement on alternative frameworks of control offered by Chomsky (1980, 1981) and Bresnan (1982). In section 2, I will extend and in some respects refine my approach trying to capture some puzzles of control left without satisfying solutions. 1.1. In R (1983a, 1983b), I argued against Chomsky's (1980, 1981) theory of control putting for­ ward an alternative framework that relies crucially on thematic relations (theta roles). The main consider­ ations that have led me to prefer this domain of explication and explanation to others, e.g. a configurational account, may be summed up as follows:

CONTROL IN COMPETING FRAMEWORKS

103

In infinitive control clauses, if such a clause is a propositional argument of the matrix verb and does not appear in subject position, the choice of the control­ ler NP is determined by the thematic relation that the surface PRO subject of the infinitive clause, the controllee, bears, that selection being interpretable in terms and on the strength of the predicate argument structure of the matrix verb. In other words: the choice of the controller is directly or indirectly deducible from, or mediated by, the semantic structure of the matrix verb. This seems to be quite a natural consequence of the relations obtaining between two predicate argument structures if one is embedded as a propositional argument in the other. Two interconnect­ ed objections might be raised and should be dealt with. First, the embedded sentential argument of a matrix verb, e.g. promise or persuade, has an independent thematic structure of its own, but the controllee is invariably determined on syntactic grounds, by the grammatical function of subject, not by a thematic role. If, however, the thematic structure which a verb, say leave, produces in its own right, as a whole bears a thematic role within the thematic struct­ ure of the matrix verb, each individual theta role figuring in it may be sensitive to the matrix predicate in a sort of second order dependence. The global depend­ ence of the propositional argument is projected onto its interior arguments. In other words: The PRO subject in the control clause bearing a certain thematic role must select a controller carrying a role in the higher thematic structure that is compatible with its own, the

104

RUDOLF

precise sense of compatibility being determined or mediated by the matrix verb. In the simplest case, the agent (and subject) NP in the theta structure of the embedded predicate leave selects as its coreferential controller the NP bearing the theta role Agent, if the matrix predicate is e.g. promise or try. Second, the surface subject NP of the infinitive control clause being the only possible controllee, change of voice will change that controllee replacing it by an NP with a different thematic role. In accordance with our assumptions, this change must bring about a different selection of the controller NP. That is why diathesis variations in the control clause play a vital role in control relations, culminating in "Visser's generalization" (see below p.110). But the varying theta roles which the PRO subject of the control clause can bear are no obstacle to our theory: On the contrary, it has been so constructed as to provide for, and capture those cases. As the surface subject NP of the infinitive control clause is the only possible, a "given" controllee, selection is uni­ directional. It starts from the theta role carried by the PRO subject of the infinitive clause. Thus, control is realized by matching the theta roles carried by the NP's in the matrix clause with the theta role of the PRO-surface subject of the embedded clause, where matching is interpreted in terms of the semantic structure of the matrix predicate. It would appear then that the outcome of such matching operations is some systematic subtraction from the Cartesian product of pairing of the members of a given

CONTROL IN COMPETING FRAMEWORKS

105

set of thematic relations. The fact of the matter, however, is roughly this: The bulk of all obligatory control phenomena in infinitive control clauses can be captured by two complementary constraints that generalize on the individual compatibility restrict­ ions obtaining between controllee and controller. In the format of a feature, either constraint is assigned in the lexicon to the class of verbs which induce obligatory control. Two disjunct subclasses of those verbs are constituted by the respective features assigned to them. The two features (conditions) are the Thematic Identity Condition [+ TIC] and the Thema­ tic Distinctness Condition [ + TDC]. Besides those two major classes, there are at least two minor classes characterized by marked behaviour regarding control relations (cf. 1983b). 1.2, Before demonstrating some crucial and critical cases of control captured by TIC and TDC respectively, I should mention a natural and expected limitation narrowing the operational range of the two constraints As noted above (p. 103), control relations are depend­ ent on both the control clause and possible controller being arguments in the predicate argument structure of the matrix verb. I have left open which syntactic position the control clause must occupy or which grammatical function it must fulfil for the control mechanism to observe the two conditions stated above. The control clause must not appear in subject posit­ ion: "A PRO in a subject sentence (co)refers freely" (Manzini 1983: 424). When the control clause is an

106

RUDOLF

infinitive subject sentence bearing a thematic relation to the matrix verb, specific dependencies of control on semantic configurations disappear. Trying to account for this fact we have two options. We might refer to "... certain semantic asymmetr­ ies between subjects and nonsubjects (that) justify the suppresion of the 'logical' subject from lexical predicate argument structure" (Bresnan 1982: 349). Chomsky (1981: 37) takes a similar stance: "This account presupposes that the -role of a subject (where it has one) is determined by the VP of S rather than by the verbal head of this VP... I will assume this convention throughout". I think this option can be abandoned. Bresnan (1982: 349/390) argues convinc­ ingly against that asymmetry. The second option rests on the assumption of a selectional distinction between sentential (infinitive) and nonsentential subjects. In particular: two interconnected aspects of their dist­ inction regarding selectional sensitivity can be point­ ed out. (a) The bulk of the verbs involved in semantic and pragmatic constraints on control does not subcategorize for sentential subjects, e.g. p r o m i s e , try, per­ suade, (b) The rest of the verbs involved in control that is subcategorizable for sentential (infinitive) subjects have much less or no power of constraining and mediating control relations between the subject of the subject infinitive clause and a controller NP in the matrix clause. Characteristically, the matrix pre­ dicate VP very often is a copula with predicative adjectival or noun phrase in which no controller is available at all. Thus, the PRO in subject infinitive

CONTROL IN COMPETING FRAMEWORKS

107

clauses can have arbitrary reference or it can corefer into the matrix sentence (object), into a phrase superordinate to the infinitive S, or into a phrase sub­ ordinate to the infinitive S. (cf. Manzini 1983: 424). (28)

Mary knows that public]

(29)

[PRO to behave herself in

would help Bill

[PRO to behave himself in public]

would help

Bill's development (Manzini's (1983: 424) examples and numbering) 1.3. I will return to obligatory semantically bound control involving infinitive complement (object) clauses. The control relations obtaining in (2) - (3) should suffice to refute Chomsky's theory of control (Chomsky, 1980, 1981). (1) Johni promised Bill

[PRO. to shave himself]

(2) Johni was promised [PRO. to be allowed to shave himself] (3)

John promised Bill. [PRO. to be allowed to shave himself]

(1) is correctly accounted for in Chomsky's framework (1980), but neither (2) nor (3) can be captur­ ed by it: In (2), promise (d), a verb of subject control marked [+ SC], has no subject in accordance with Chom­ sky's assumptions. In (3) we find complement control with a verb of subject control. All cases (1) - (3) are captured straightforwardly by the Thematic Identity Condition [+ TIC] assigned to promise in the lexicon. In (1) - (3), the controller selected by the controllee

RUDOLF

108

PRO bears the same theta role or at least a cognate one, Agent

in (1), Addressee

in (2) and (3). The

slight strangeness or, with some speakers perhaps, acceptability of (3) is dealt with below (p. 114) (For a detailed discussion of examples analogous to (1) - (3) cf. Ruzicka 1983a, 1983b). 1.4. (1), (2) and e.g. Bresnan's (1982: 403) example (87) are, of course, perfectly normal cases of obligat­ ory control. As opposed to the present theory, in Bresnan's (1982: 303) they are handled as marked pheno­ mena: "In contrast, the lexical form for promise in (87) has an open XCOMP, which must be grammatically controlled. However, this is a marked control relat­ ion, for even though promise has an object, it evinces subj ect control". (87)

John promised Mary to be on time (Bresnan's numbering)

The lexical form for promise "(89) 'PROMISE

is

'" (Bresnan 1982: 403)

In Bresnan's framework, open grammatical functions (XCOMP and XADJ) are closed under functional (grammatic­ al) control. In my approach, neither the preference of the object to the subject as the controller nor even a choice between them is related to control rules and, in particular, to lexical properties of promise. The general scale of preference defining "lexically induced functional control" (Bresnan 1982: 376) must associate marked control properties with promise, because in

CONTROL IN COMPETING FRAMEWORKS

109

structures with promise as matrix verb, the subject is preferred to the object as the controller, the object ranging higher in the mentioned preference scale. The lexical properties of promise do not seem to induce, much less to imply preferential issues between grammat­ ical functions. Eresnan is forced to have recourse to markedness of control in (1), (87) , as she tries to construct an account of control unifying cases like (1) , (87) and (21a) : (21a)

John seems sick to Mary (Bresnan's numbering)

The demand put on the Lexical Rule of Functional Control is too strong, when it is required to capture (1), (87) and the fundamentally different case of (21a). In (21a) the subject is bound to be the "controller" by independent syntactic constraints on Raising. Then again and by the same token Bresnan is forced to turn to markedness in face of a rather transparent if complex case of control: "While functionally control­ led XADJS generally have a range of possible controllers, including OBI

..., there is one construction in

English in which functional control of the XADJ is re­ stricted. This is the clause-initial position of the adjectival phrase adjuncts shown in (32) and (33). (32a)

Sure of winning, Mary entered the competition yesterday.

(33a) *Sure of winning, the competition excited Mary yesterday. (33b)

Sure of winning, Mary was excited by the compettion yesterday.

110

RUDOLF

In all of (32a...) and (33a,b), sure of winning is controlled by the SUEJ of the sentence;... Thus, we will assume that SUBJ-control is a marked property of the "clause-initial XADJ construction..." (Bresnan 1982: 379). Constructions like (32a) and (33b) are widespread cross-linguistically. Clause-initial posit­ ion of the control clause signals a certain parameter or range of relations between the two propositions. The parameter is assigned values fixed by implicatures relying on the semantic and pragmatic setting of the sentence. SUBJECT control is a constant of the para­ meter by which all possible specifying implicatures are conditioned. By distinguishing and characterizing SUBJECT control as marked on grounds of the exclusion of, or preference to OBJECT control, Bresnan's propos­ al misses the relevant aspects of control, clinging to a rule (Lexical Rule of Functional Control) that on the one hand encroaches on the independent domain of Raising (Example 21a) and, on the other hand, tries to handle by (markedness) force connections between control restrictions and principles of condens­ ed encoding as in (32a). 1.5.

Another serious consequence of the dividing

line Bresnan draws among the phenomena subsumed by her under control emerges in connection with "Visser's generalization": "... the observation that verbs whose complements are predicated of their subject do not passivize" (quoted after Bresnan 1982: 402). Visser's generalization excludes sentences as (4) and (90a):

CONTROL IN COMPETING FRAMEWORKS

(4)

John was promised to leave

111

(cf. '.

1983b:

309) (90a)

Mary was never promised to be allowed to leave (Bresnan's example and numbering) "(90a) appears to contradict Visser's generaliz­

ation, in that a subject control verb passivizes..."

{promise)

(Bresnan 1982: 4 0 4 ) . In

(1983b), sentences like

(90a) and (5), which are "un­

expected" and "curious" to Chomsky straightforwardly

(1980), are

handled by the Thematic

Identity

Condition. John was promised to be allowed to leave.

(5)

Thus, the TIC captures

(1) - (5) on a unifying

and uniform principle. Trying to explain

(90a) Bresnan

has recourse to anaphoric control, an essentially different type of control in her framework:

"Certain

apparent exceptions to the conditions on grammatical control relations are actually instances of anaphoric control".

(Bresnan 1982: 4 0 2 ) . T h u s , control in (90a)

is accounted for in a crucially differing way. Accord­ ing to Bresnan's theory, in case of anaphoric control there can be counterparts with lexical subjects and there need not be an antecedent. Bresnan thinks that "... to

be allowed

to

leave

(1982: 404) behaves like

a closed COMP, which cannot be functionally control­ led. Unlike XCOMPS, closed COMPS may undergo It position,,, clauses

Extra­

and may appear as the subjects of finite (90c)...

112 (90c)

RUDOLF

To be allowed to leave was never promised to Mary"

Certainly, (90c) is a case of control essentially like (90a). The subject infinitive sentence bears a thematic role by virtue of the fact that it is assign­ ed this role in object position in the VP headed by promise. The identity of the thematic role which the infinitive subject sentence bears in relation to pro­ mise in (90c) with the thematic role carried by the infinitive object sentence in (2), (90a) entails the identity of the thematic roles of the PRO-Controllees within the respective sentences. It seems natural then to account for control in (90a) and (90c) as well as in (1) - (5) by a single unifying and principled constrairt or condition. Incidentally, the links connect­ ing the interrelated syntactic phenomena of It-Extra­ position and subject position on the one hand, and the widely different accounts for control in (90a), (90c) and (1) on the other, remain somewhat obscure. 2. EXTENSIONS AND REFINEMENT OF TIC AND TDC Let us now consider some complications and poss­ ible extensions within my framework of TIC and TDC conditions. 2.1.

Thematic Identity Condition Some problems are posed by sentences like (6),

which should be captured by the TIC. (6)

?

Hans wurde von Georg versprochen

Hans (dat.) was by Georg promised

CONTROL IN COMPETING FRAMEWORKS [PRO

113

das Haus zu verlassen] the house to leave (Beisp. von W. Abraham 1983)

(6) does not seem acceptable. The Thematic Identity Condition predicts that Georg in the PP [pp von Georg] can be controller of PRO, as it bears the thematic role Agent, which is identical with the role that PRO bears within the thematic structure of the control clause. Certainly, NPs within preposition­ al phrases and/or "oblique functions" (Bresnan 1982: 376) cannot be excluded from being controllers on principle, at least in German. (7)

Er verlangte von ihmi.

[PROi. das Haus zu verlas­ sen]

he demandes from him (8)

ihm war versprochen worden him

(dat) had been promised

the house to leave (von PRO.) (by PRO)

[PRO. Hans in die Auswahlmannschaft Hans

(acc.) into the select team

aufzunehmen] to include The controller of the PRO subject is an empty N P , "covertly" bearing the theta role Agent in the matrix sentence and undergoing arbitrary interpretation, that is contextually and/or situationally determined inter­ pretation. When I started to collect judgements on sentences similar to

( 6 ) , (8), I realized that there is a lot

of variation among speakers. Slight modifications may

114

RUDOLF

change judgements: (9)

Hans war von niemandemi. zugesagt worden,

[PRO.

Hans (dat.) had by nobody been promised das Haus zu kaufen] the house to buy

(10)

Von niemandem war uns versprochen worden, [PRO. By nobody (it) has us (dat.) been promised zur Festschrift beizutragen] to the Festschrift to contribute

(9) and (10) seem better than (6), and (10) better than (9), to many speakers. In any case, the TIC pre­ dicts that the matrix subject cannot be the controller in (6). Whether, in sentences of type (6) - (10), a NP that bears the theta role Agent, but is not in subject position, can be the controller of PRO. bearing Agent role in the infinitive complement, seems to de­ pend on conditions of a different order, presumably of information structure. Changing word order we can arrange (10) in a more acceptable way than (9). In (6), considerations of information structure seem to resist control relations that are predicted by TIC. Generalizing over (6) - (10), we might tentative­ ly state principle (11): (11)

If the controllee denotes the person committing himself (herself) to do what he (she) promised

CONTROL IN COMPETING FRAMEWORKS

115

to do, the controller must be given salience or remain covert. In other words, control as predicted TIC, is thwarted by clash arising between the PRO Agent, natur­ ally foregrounded in subject position though without phonological content, and the peripheral backgrounding of the Agent NP, its virtual controller "supporting" the PRO-NP. The clash does not arise in case the potent­ ial controller has no phonological content itself, cf. (8) . 2.2. The Thematic Distinctness Condition applied to ASK Consider (8) and (9) taken from Chomsky (1981: 76), in his numbering (8)

John asked the teacher to leave early

(9)

John asked (begged, pleaded with...) the teacher to be allowed to leave early

"Thus (8) is ambiguous as to controller, and in (9), subject control is preferred to object control..." (Chomsky 1981: 76) Chomsky is unable to cope with the problems raised by the options of control in (8) and (9), because he has never taken up the sound idea interposed in his discussion of control: "A natural suggestion is that choice of controller is determined by -roles or other semantic properties of the verb, or perhaps pragmatic conditions of some sort" (Chomsky 1981: 76; cf. 1983a, 1983b). Preference of subject to object control in (9) is straightforwardly predicted by the TDC. The two control relations possible

116

RUDOL F

in (8) are not on a par with each other: ASK (in the meaning MAKE REQUEST) is assigned [+ TD] (Thematic Distinctness Condition) in my framework. The TDC pre­ dicts that the object NP the teacher must be the controller: it bears a theta role distinct from the Agent role that PRO carries. John is excluded by [+ TD] on grounds of the nondistinctness of its theta role from that of the controllee. In (1983b), I gave an explanation for the second possible control relation in (8). The TDC accounts for the second as well as for the first one. Note that ASK is a verb that has developed conventionally a strongly marked scheme of syntactic and semantic reduction in its use. The crucial point in this reduction is that ASK can idiomatically absorb the semantic content of PERMIT (give PERMISSION). Thus (8) can be construed as (8)': (8')

Johni asked the teacher. [PRO. to leave early]]

[PRO. to PERMIT HIM.

Characteristically, in German, the exact equi­ valent to (8) in its second interpretation, is (12): (12)

John hat den Lehrer zeitig weggehen zu dürfen John asked the teacher early to leave to be permitted

Clearly, the TDC is fully satisfied in (8) if it is construed - as it must be in its second understand­ ing - in the way of (8)'. Bresnan is puzzled about a similar problem arising in her theory of control: "Although the passives Mary was never promised permission to leave and Mary was never promised to be allowed to leave are both judged

117

CONTROL IN COMPETING FRAMEWORKS grammatical by many speakers, the actives Fred

ed Mary permission

to leave

and Fred promised

promis­

Mary to

be allowed to leave differ, the latter being less acceptable for many. It is unclear exactly why this should be so", (Bresnan 1982: 40., fn. 12). The second passive sentence is accounted for in my approach as above (p.107 ) , while in the first the control relation cannot be captured without construing and explicitly representing the thematic structure induced

by (13)

permission: Mary, was never promised

[PRO. to GET permission

[PRO. to leave]] l

Thematic identity ([ + TI]), or, at least, similarity, cf.

1983b, 315, regarding a scale of similarity,

is satisfied. In (14) (added by me, R.R.), on the other hand, quite naturally, a different thematic structure surrounding and including PERMISSION must be developed: (14)

Mary was never asked permission to leave

(14') Maryi, was never asked [PRO. to GIVE permission to N P j [ P R j . to leave]] ASK is assigned [+ T D ] , which is satisfied in (14). Concerning (15) (15)

Fred promised Mary permission to leave,

quite an analogous construal is possible, but there are two "equivalent" options: (16)

Fred, promised Mary. [PRO. to GIVE Mary'j per­ mission [PROj. to leave]]

118 (17)

RUDOLF

Fred promised Mary. [PRO. to GET permission [PRO! to leave]] In (16), Fred,

and PRO. satisfy the TIC, assigned

to p r o m i s e , MARY' and PRO. satisfy the TDC, assigned to PERMIT. In (17), MARYi. and PRO. satisfy the TIC and PROi. with PROi! the TDC.

Now consider the last of Bres-

nan's sentences, which is "less acceptable for many": (18)

Fred promised Mary to be allowed to leave

(18) is similar to (3). The oddity of (18) and (3), or the reluctance of many speakers to accept them, seems to arise from two interconnected conflicts: First, somebody (Fred) commits himself by a promise to do something, but the choice of the passive voice implicates his restraint in not including himself explicitly among those responsible to keep the promise he has given. Second, (18) may be considered the reverse counterpart of (6) (s.p. 112), the clash arising between the salient subject NP, the Agent of p r o m i s e , , and its exclusion from control. Control is inaccessible to the foregrounded Agent of promise, because it would violate the TIC. Observing the TIC, however, by coindexing Mary and PRO is opposed again to the pragmatic implicature of promise. This contrast is weaker in (17). As shown by (14) and (15), nominal deverbals may complicate control relations. Consider the Russian sentence (19): (19)

v

pros "be

sobrat'sja nam

(in) the request to gather

v odnu

us (dat.) in one

CONTROL IN COMPETING FRAMEWORKS

119

kameru, čtoby provesti poslednie časy vmeste, room to spend the last hours together otkazano ne bylo (it) was not refused (19) is virtually ambiguous like Chomsky's example (8) on p. 115. The overt encoding possible in Russian by inserting the dative subject nam (us) clearly indicates an interpretation which induces a construal similar to (8') on p. 116 . 3. Control, even if investigated only in the major domain of infinitive control clauses, proves to be a phenomenon, where syntactic, semantic and pragmatic determinants may interact in a nonstandard modular way. Their interplay should not be conceived and modell­ ed as a successive participation of modules, where return to the preceding one is impossible once the following has been reached. Thus, for example, pragmat­ ic factors may affect the structures determined by a lexical item, which in turn are referred to by grammat­ ical principles. Certainly, there are some "grey zones" in which judgements of speakers are elusive. This does not mean, of course, that control is an amorphous syndrome evading systematic and formalized description. I hope to have shown in the complex but I think persuasive course of argument that besides the overwhelming majority of clear cases of "unanimous" acceptance, hierarchies of acceptability can be made predictable by the two conditions (TIC and TDC) and

RUDOLF

120

pragmatic factors in connection with considerations of information structure and language dependent convention­ al semantic developments (e.g. ASK). The investigation into control phenomena may yield highly insightful results about encoding principles in natural languages underlying their extended manipulat­ ion with empty categories and condensed structures. FOOTNOTE 1 "(20) Lexical Rule of Functional Control Let L be a lexical form and F its grammatical function assignment. If XCOMP € FL, add to the lexical entry of L: ( ↑OBJ2) = (↑XCOMP SUBJ) if 0BJ2 G F ; otherwise: (↑ OBJ) = ( t XCOMP SUBJ) if OBJ G F ; otherwise (t SUBJ) = (↑ XCOMP S U B J ) "

(Bresnan 1982: 376)

REFERENCES Abraham, W. 1983. "Zur Kontrollbeziehung im Deutschen". Sprache, Diskurs und Text ed. by R. Jongen et al. 41-59. Tübingen: Niemeyer Verlag. Bresnan, J. 1982. "Control and Complementation". Ling­

uistic

Inquiry

13. 343-434.

Chomsky, N. 1980. "On Binding". Linguistic 1-46.

Chomsky, N. 1981. Lectures

on Government

Inquiry

and

11.

Binding.

Dordrecht: Foris. Manzini, M, R. 1983, "On Control and Control Theory".

Linguistic

Inquiry

14. 421-446.

Moravcsik, E. A. & J. V?irth, eds . 1980. Syntax

Semantics

13.

and

Academic Press.

Pullum, G. K. 1983. "How Many Possible Human Languages are there?". Linguistic Inquiry 14. 447-467.

CONTROL IN COMPETING FRAMEWORKS

121

, R. 1982. "Kontrollprinzipien infiniter Satzformen: Infinitiv und Gerundium (deepričastie) im Russichen und in anderen slavischen Sprachen". Zeitschrift für Slawistik 27. 373-411. , R. 1983a. "Autonomie und Interaktion von Syn­ tax und Semantik". Untersuchungen zur Semantik ed. by R. & W. Motsch. 15-59. Studia Grammatica XXII. , R. 1983b, "Remarks on Control". Linguistic Inquiry 1 4 . 309-324. Sgall, P. 1964. "Generative Beschreibung und die Ebenen des Sprachsystems". Zeichen und System der Sprache ed. by G. F. Meier. 225-239. III. Sgall, P. 1966. "Ein mehrstufiges generatives System". Kybernetika 2. 181-190. Sgall, P. 1967. Generativni popis jazyka a ceska deklinace. Praha: Academia. Sgall, P. 1982. "Can Linguistic Ideas Cross the Ocean". Folia Linguistica XVI. 399-410.

JARMILA PANEVOVA Prague

THE CZECH INFINITIVE IN THE FUNCTIONS OF OBJECTIVE AND THE RULES OF COREFERENCE

1.1. The functional generative description (FGD) proposed by Sgall in the early sixties works with the sentence as the highest unit of the language system. This does not mean, however, that those subscribing to FGD disregard the issues connected with the structure of text; Sgall (1979) supports that direct­ ion in the contemporary text linguistics that proceeds from what is known about the structure of the sentence to the study of the occurrences of sentences in dis­ course. Also in the domain of coreference Sgall dis­ tinguishes intrasentential and intersentential coreference, the former being the objective of the study of the grammatical system, while the latter belongs to the domain of the use of language. In this paper we would like to contribute to the study of the former type of coreference and to broaden the consider­ ations of Hajičová, Panevova and Sgall (in press) by taking into account Czech empirical material; our investigations point to the fact that the borderline between grammatical (intrasentential) and textual (intersentential) coreference is not always easy to

JARMILA PANEVOVA

124

find. 1.2. We focus on the class of embedded predicat­ ions in the function of Objective (and, due to a close relationship between the two, we discuss also the pre­ dications in the function of Intention) having the temporal grammateme of posteriority (of the dependent event after the governing event, cf. Panevová, Benešova" and Sgall 1971) and fulfilling the conditions for being "nominalized" by means of the infinitive. The issue under consideration is not a new issue from the point of view of FGD. We start from the assumption that under certain conditions an embedded predication in the given functions may be rendered on the surface level by a dependent clause, by an infinitive construct­ ion or by a verbal noun, without a change of the truth conditions; thus, according to Sgall's (1983) criteria, such surface sentences are synonymous. This criterion of mutual interchangeability, as well as other features of the verbs that provide for infinitive control in the given domain will be also taken into account in our discussion. 1.3.

The issues characterized above are being

paid a great attention in the contemporary, esp. generativist linguistics, first of all in connection with Chomsky's (1980, 1981) principle of binding and of 2 control. These studies were analyzed in detail else­ where, esp. by (1983; this volume). After having criticized Chomsky's (1980) account of infinitive control in terms of syntactic functions and having modified this account in terms of thematic (case) functions, R concludes that the rules of infinitive control are

THE CZECH INFINITIVE

125

governed by a non-standard interplay of syntactic, semantic and pragmatic factors. The core of the matter can be, in his opinion, accounted for by means of an assignment of one of two main features and of two variants of them that are attached in the lexicon to the verb in the function of the controller: the first feature [+TI] denotes the identity of thematic roles under control, the second feature [+TD] denotes the difference of these thematic roles. The remaining two features are marked variants of the main ones. In the majority of relevant cases one cannot but agree with 's analysis of the examples given by Chomsky, Bresnan and others. It seems, however, that the class­ ification of verbs according to the identity and difference of the controller and the controllee is not sufficient, since the principle of control often involves more than the two main features, that is to say, the relations of coreference may be more complicat­ ed. It should be noticed, however, that traditional syntactic writings, though mentioning sometimes a bearer (actor, subject) of the infinitive (cf. e.g. Peskovskij 1956), do not specify the relationships of the "bearer" (or, as the case may be, the subject) to other participants of the infinitive phrase or of the superordinated main clause. Svoboda (1962) claims that "the actor (bearer) of the infinitive action is being understood from a certain sentence part independent on the infinitive (italics K. S.). This claim is not quite exact as for the formulation of explicit condit­ ions, but it indicates to a certain extent a direction

126

JARMILA PANEVOVA

of further investigations. Šmilauer (1966: 237ff) and after him also, Sgall (1967: 163 in the frame of functional generative description), Bauer and Grepl (1972: 117f) speak about two types of object infinitive: (a) a subject type (with the initiator of the action expressed by the subject) - Šmilauer, or subject ident­ ity type - Bauer and Grepl: Snazil se ode jit (He at­ tempted to g o ) , (b) an object type (the initiator of the action being expressed by the object) - Šmilauer, or subject difference type - Bauer and Grepl: Zakdzal odohdzet (He forbade to leave). However, even this solut­ ion does not suffice, because not only examples such as Obdval se byt zahrnut mezi č l e n y tohoto spolku (He was frightened to be included among the members of this society), but also such as e.g. Bylo zakdzdno zaradit o toto misto (It was forbidden to Jana mezi uohazeče place John among the applicants for this position) are to be covered. The quoted writings do not also speak in exact terms about the coreference of the members of the governing and the dependent clause. 2, We have analyzed about 65 Czech verbs that may govern a dependent predication in the function of Object­ ive rendered by an infinitive

and about 15 verbs that

may govern a dependent predication in the function of intention.4 The function of Intention is differentiat­ ed by Poldauf (19 59) from Purpose as a phenomenon typic­ al for "the concentration on the realization of the action" after the verbs of movement (jdu se umyt - I go to wash myself); Poldauf thus denies the existence of the infinitive of purpose in Czech.

THE CZECH INFINITIVE

127

2.1. According to the type of coreference they display the governing verbs are divided into two main and two marginal classes. The types of coreference (control) are distinguished on the basis of the follow­ ing notions: the actor of the governing verb (as the antecedent of the infinitive - the controller) and the surface subject of the dependent verb, or, as the case may be, such a participant of the dependent pre­ dication (infinitive) that may become the surface subject (the controllee). The rules of coreference, which make it possible to render a dependent predication by means of an in­ finitive construction, assume that the following two conditions formulated in Hajičová, Panevová and Sgall (in press) are fulfilled: (i) in the frame of the respective (governing) verb there is a marker for that participant which is the antecedent and which controls the surface subject of the dependent predic­ ation; this is made (a) either by means of a marker of an obligatory coreference (an asterisk), (b) or by means of a marker of optional coreference (a cross). E.g. the frame of the verb minit (to mean) is Act* Obj, which is interpreted in the following way: in the dependent predication filling the slot of the Objective necessarily one of the elements which may become the surface subject is coreferential with the antecedent, which is the Actor of the governing pre­ dication (Act*); possible surface shapes are Jan minil odjet na hovy (John meant to leave for the mount­ ains) . Any other structures are blocked for this and similar verbs by the generative mechanism.

128

JARMILA PANEVOVA

The verb snazit se (to try) has the frame Ag Obj which admits the possibility of coreference (Snazil se včas - He tried to come in time; Snazil se byt přijit zavazen do družstva - He tried to be included into the team), but does not exclude other possibilities as well (Jan se snažil, aby se přztel zúčastnil konfevenoe - John tried to that his friend would take part at the conference - lit.). (ii) The coreferential element (COR) is in the dependent predication in the function of the element that plays the role of the controllee (it need not necessarily be the Actor or the Objective of the depend­ ent predication, but e.g. its Addressee, e.g. tie-it nekoho (Addr) něčemu (Obj) - to teach someone (Addr) something (Obj): Přál si být učen.,. (He wished to be taught...); however, always a semantic correlate of a deleted surface subject of infinitive is concerned. When the conditions (i) and (ii) ensuring the corresponding coreference are fulfilled, the following groups of verbs can be distinguished: 2.11. The group V with the coreference ActG : SbD (where the subscripts G and D denote the governing and dependent clause, respectively; Sb stands for sur­ face subject, Act for Actor); this group splits into two subgroups, one with Act*, the other with Act . Among the verbs with Act* there belong the following Czech verbs: bát se (to be afraid)7 , přát si (to wish), potřebovat1(to need), ostýchat se1 (to be coy), minit1(to mean), stydét se1 (to be ashamed), odvdzit se (to dare)1., opoazit se (to dare), osmélit se (to venture), stihnout (to make to do s.t.), odmitnout

THE CZECH INFINITIVE

129

(to refuse), zapomenout 1 (to forget), opomenout 1 (to neglect), začit7 (to start), p ř e s t a t 1 (to stop). Among the verbs with Act (optional coreference) there belong zamýšlet (to intend), usmyslet si (to de­ termine) , bat se2 (to fear), zasluhovat (to zaslouzit si (to deserve), strachovat se (to dread), zvyknout si (to get used to), nauoit se (to learn), c h y s t a t se (to prepare), snazit se (to endeavour), touzit (to yearn), troufat si (to dare), rozhodnout se (to decide), pokusit se (to attempt), souhlasit (to agree), zvolit si (to choose), vybrat si (to choose), nendvidet (to hate), zapomenout (to forget), opomenout 2 (to omit) , nabidnout se (to offer) , prihldsit se (to apply o.s.), vyźadovat - źádat 1 (to require), zavázat se (to bind o.s.), rozhodnout (to decide). While a subclass of V admits only an active in­ finitive, other elements of V admit - besides active infinitive (and besides concurrent means such as a dependent clause and a verbal noun, which will be ment­ ioned below) - under certain conditions also a passive infinitive. The passive infinitive is used, when the function in the dependent clause filled by the symbol COR is not an Actor (see (ii) in Sect. 2.1), e.g. (1)

Jan si zvykl být zařazován

mezi outsidery

John got used to be classed among the outsiders. (2)

Jan touzil byt jmenován ředitelem. John desired to be appointed the director.

The verbs přát si

si

(to wish), zasluhovat

(to deserve) and potrebovat

-

zaslouzit

(to need) behave in an a-

typical way; active infinitive with them can be synonym-

130

JARMILA PANEVOVA

ous with a passive one, or even it replaces it more or less obligatorily: (3)

Jan si přál zařadit do prvni skupiny. John wished to class into the first group

(lit).

(3') Jan si přál být zařazen do prvni skupiny. John wished to be classed into the first group. (4)

Ten váš syn potřebuje nařezat. That son of yours needs to give a thrashing (lit.)

(4')*Ten váš syn potřebuje byt nařezán. That son of yours needs to be given a thrashing. (4") Ten váš syn potřebuje naářez. That son of yours needs a thrashing. In other contexts, however, these verbs have a coreference typical for V

, where the index COR is

attached to the Actor: (5)

Jan si opravdu zaslouzil jet na ten vylet. John really deserved to go for that trip.

The idiosyncratic behaviour of these verbs was observ­ ed also by Koenitz

(1972), who quite correctly assigns

to the pairs of the type difference. As

(3) and

(3') a stylistic

(4") shows, there is also a possibility

of a phraseological paraphrase, which

indicates the

marginality of the behaviour of these verbs as for the rules of coreference. A s for the verbs connected with the function of Intention, such as j i t

(to g o ) , ode j i t

(to l e a v e ) ,

pospichat (to hurry), zaskočit si (to drop in), odklusat (to trot away), odebrat se (to start off), these verbs behave in the same way as the verbs of the first

THE CZECH INFINITIVE

subclass of V

131

(with obligatory coreference); other

verbs, such as smĕřovat arrive), přicházet

(to aim at) , přijzdèt

(to come)

(to

behave similarly, but

the presence of a Directional is obligatory here. 2.12. The group V2 with the coreference AddrG : Sb D ; here belong the Czech verbs zabrdnit (to prevent), zakdzat (to forbid) , usnadnit (to make easier) , zddat (to ask), predepsat (to prescribe), pozadovat - vyzad o v a t 2 (to claim), prosit (to beg), pomoci (to help), rozkdzat (to command), prikdzat (to order), ulozit (to enjoin), radit (to counsel), naridit (to impose), naučit (to teach), odnaucit (to unteach), presvedcit (to convince), Iákat (to allure), vyzvat (to invite), umluvit (to get a p. to do), primet (to compel), doporučit (to recommend), navrhnout (to propose). In a regular case, this type of coreference is rendered (besides other means of expression of the dependent predication) by an active infinitive: (6)

Otec zakázal synovi psát verše. Father forbid his son to write poetry. The Addressee of the governing predication (the 9 then

controller), however, may be a general one;

also the subject of the infinitive is general. (7)

Doporucil zařadit Smithův článek do sborniku. He recommended

to include Smith's paper into the

volume. In

(7) there is a general Addressee

(as the ante­

cedent) and a general Actor of the dependent

predicat-

JARMILA PANEVOVA

132

ation (as the controllee). The verb nutit (to compel) represents the only example of an obligatory coreference of the Addressee with the respective element of the dependent clause. This restriction is to be taken into account when form­ ulating the generative component of the description, similarly as with the first subclass of the verbs V . Its behaviour on the surface level, however, is differ­ ent: it does not exclude a dependent clause as a means

of expression {Jan nut it

,

přitele

aby dokončil

vyohle

svou knihu - John compelled his friend so that he finished his book quickly - lit.) or a nominalization

{Jan nutil

přitele

k rychlèmu

dokonćeni

jeho

knihy

-

John compelled his friend to a quick finishing of his book). With the corresponding subclass of the V group there are surface restrictions that admit only the infinitive (see above, Sect. 2,11). In case that the symbol COR in the dependent pre­ dication fills the slot of the Objective of that pre­ dication, passive infinitive is used: (8)

Doporućil Jirkovi byt zařazen do vyšši kategorie. He recommended to John to be included in a higher category. The regularities of this type of coreference are

broken when two cases of coreference get into conflict: (9)

Šef

přikázal byt jmenován členem nove'" komise.

The chief ordered to be nominated a member of a new committee. (9')

Šèf

přikázal G e n A d d r ' abY Gen

šèfa členem novè" komise.

Act

jmenovali

THE CZECH INFINITIVE

133

The representation (9') is a verbal paraphrase of the tectogrammatical (semantic) representation of (9). This combination of coreference shows that it is not always easy to draw a boundary line between the grammat­ ical and the textual coreference. 2.13.

The verbs of the group V , which is a

small group, have two possible coreference assignments, namely (a) A c t G : SbD_, or (b) AddrG_ : SbD_. There belong the Czech verbs slibit (to promise) and odepřit (to refuse). With these two verbs we are concerned with an ambiguous construction; which of the two possibilities is concerned, can be determined only according to the factual relations; the type (b) is possible only with some specific lexical cast: (10)

(11)

(12)

Jan slibil matce vrátit se domů včas. - (a) pre­ ferred John promised his mother to return home in time. Otec odepřel synovi jit s nim do kina. - (b) pre­ ferred Father refused his son to go with him to the cinema. Otec odepřel synovi jit do kina. - both (a) and (b) interpretations possible, without preference Father refused his son to go to the cinema.

(13)

Slibili Janovi stàt se předsedou. - (b) preferred They promised John to become the chairman.

(14)

Slibili vitezi dostat medaiii. - (b) preferred They promised the winner to get a medal.

JARMILA PANEVOVA

134

2.14, The verbs of the group V ded

govern the embed­

predication with the function of Intention and

have the coreference of the type ObjG : Sb D . There belong the Czech verbs poslat expel), zahnat vézt (15)

(to send), vyhnat

(to drive away), nest

(to

(to carry),

(to carry, to drive). Poslali Jana zatopit v krbu. They sent John to make a fire in the fireplace.

(16)

Otec vyhnal syna pást dobytek. Father sent his son to keep the cattle out at grass. 3. Also some other features are relevant for the 1 4

behaviour of the verbs of the groups V through V in the generative procedure and in the transductive components: (a) Some verbs admit in the function of Objective only a dependent predication ( z a m ý š l e t - to intend, usmyslet si - to determine), while other admit also an original, non-derived noun (snazit se o vitězstvi - to attempt at a victory, touzit po penëzïch - to long for money). (b) After some verbs the dependent predication can be generated with the temporal grammateme of simultaneity, anteriority and posteriority, e.g. straehoval se, že došlo  havdrii / ze docházi. . . / že dojde  havárii (he was afraid that a breakdown had occurred / occurred / would occur). Most verbs of the class V admit only a posterior dependent pre­ dication {touzit - to long, troufat si - to dare, pokusit se - to try, souhlasit - to agree); the verbs

THE CZECH INFINITIVE 2 of

the

3

classes V , V

(Intention)

dependent

135

4 and V

admit

predication

only

a

(beside

posterior a

non-derived

noun).

The features described in the points (a) and (b) concern a compatibility of certain grammatico-lexical units, which must be included in the generative apparatus, as well as the restrictions concerning morphological grammatemes, of. above Sect, 1.2. (c) Infinitive constructions governed by the verbs 1 4 of the groups V through V - with the exception of the cases of obligatory coreference - coexist with a depend­ ent clause (Objective with an active form of the verb, This function of Subjective with its passive variant). the subordinate clause is sometimes rendered by the con­ junction ze (that), or by the conjunction aby (so that); the choice between the two conjunctions is given by the "rection" given in the lexicon with each of the verbs (along with its surface features, e.g. snazil se, aby.,. - he tried so that... - lit.; pokusil se, aby... he attempted so that... - lit.; zvykl si, ze... he got used that...; souhlasilj že... - he agreed that... (etc.). Sometimes there exist two synonymous variants, one with an explicit rendering of a weak hortative modality, the other with the conjunction aby (cf. already Panevová, 1930, Chapter 5.2.1): (17) Lèkař předepsal pacientovi, aby bral kapky dvakrat denne, The doctor prescribed to the patient so that he may take the drops twice a day. - lit. (17') Lèkaf předepsal pacientovi, ze ma* brat kapky dvakrat denne. The doctor prescribed to the patient that he should

136

JARMILA PANEVOVA take the drops twice a day.

4 (d) Except for the verbs of the group V , also the forms with verbal nouns are to be considered as synonymous with the infinitive constructions, whenever the verb in the dependent predication forms a verbal noun and if also other conditions on nominalization are fulfilled (Cf. Panevovâ 1978), which we will not analyze here in detail; we also do not deal here with the morphemic form of the noun (the case), which is given by the rection of the governing verb (specified in the lexicon). The conditions described in the points (c) and (d) concern the transductive components of the description, namely those parts that translate a tectogrammatical repres­ entation to the surface syntactic form(s) and the latter to the morphemic representation(s)(we leave aside the details specified in the apparatus of the transductive components). 4. In conclusion we may summarize our observations on the coreference of participants of the governing and the dependent objective and subjective embedded clause in the form of a table, which is based on possible generated structures and takes into account the requirements discuss­ ed in Sect. 2, also selecting possible surface represent­ ations according to certain restrictions. The headings of the columns of the table correspond to the governing verbs and indicate which element of their frame fulfils the role of the controller (of the antecedent) the rows correspond to the verbs of the dependent predicat­ ion with the function of Objective/Intention and specify 12 the conditions that were discussed in Sect. 2. The abbrev­ iation DC1 stands for a dependent verbal clause, Nv for a verbal noun (for the choice of the Nv it is necessary to

V-Intention

V-Obj-poster

COR=obj

V-obj-poster-D

Nv

Table 1

Nv

Nv

DCl

V-inf.act.

V 3 -Act + ,Addr + v4-obj+

v V-inf.act. NV-inf.act. V-inf.act. DC1 DCl DCl

DC1

V-inf .act

V-inf.pass

COR=obj

V-Obj-poster-D

V -Addr+* '

v-inf .pss DC1 Nv

V-inf.act. V-inf.pass.

|V 1 -Act +

obj v- Int- pster-D COR=Act

V 1 -Act*

THE CZECH INFINITIVE 137

JARMILA PANEVOVA

138

specify further conditions, e.g. the aspect in which the verb forms a verbal noun, etc.). Also the choice of the passive form of the infinitive depends on many other fact­ ors: e.g. if the verb in its Czech surface form forms a compound passive form (such verbs as mit - to have, bát se - to be afraid etc. thus do not belong here). If the table reflected all these features (which are given for each verb in the lexicon), it would be too complicated and the main problems under discussion would step into the background. The specification of forms that are selected in Table 1 calls for a discussion of the mutual relationships of passivization and nominalization or condensation (one of the degrees of nominalization being the infinitive constr­ uction) . The rules for the selection of the form of the infinitive (active or passive) are included into the rules of coreference (special cases inside the same group of V correspond to a different assignment of the index COR to an element of the dependent predication, i.e. if the index COR is not assigned to the Actor, the passive infinitive is chosen). If a possible form is a dependent verbal clause (DC1), it undergoes a procedure in which it is decided about the active or passive form of its predicate: (18)

Snazil se, aby napsal knihu stručně a jasně. He attempted, so that he wrote the book in a concise and clear manner. - lit.

(18')

Snažil se, aby jim byla kniha napsána stručně a jasně. He attempted so that book were written by him in a

concise and clear manner. - lit. In the case of (18'), the rule selecting the passive form has a very low probability. Thus it may be summed up that

THE CZECH INFINITIVE the

139

processes of passivization and condensation in the

transductive components of functional generative descript­ ion are mutually interconnected.

140

JARMILA PANEVOVA

NOTES 1 In his paper delivered at the Prague Conf. on Automatic Processing of Texts, Oct. 1984; cf. Hajicova, Panevova & Sgall (in press). 2 For a more recent literature on this topic, see Hajicova, Panevova & Sgall (in press). 3 These verbs were excerpted from Svoboda (1962) and excerpted and translated from Russian from Conrad (1969); in his book, Conrad analyzes infinitive constructions in terms of Saumjan's applicative model. 4 The fact that the verbs governing an "intention" can be enumerated indicates that Intention stands close to the participants of verbs, i.e. to the elements of the verb frames. A semantic closeness of Intention to the modification "where to", which with the verbs směrovat (aim at), poslat (send) is obligagory, offers a possibility to interpret Intention as a syntactic grammateme (for this notion, see Platek et al. 1984) within the complementation of the "where to" type. Such a solution, however, does not offer an intuitive interpretation of such sentences as Smeroval k babicce najist se k o l a c u (He aimed at his grandmother's to eat the cakes), or Poslali Jirku do pokoje zatopit (They sent George to the room to make fire). For the purpose of this paper, we will leave aside the questions concerning the tectogrammatical structure of such constructions; only for the reasons of a certain analogy between the rules of coreference for the infinitive construction in the function of "intention" and of Objective we discuss the two kinds at the same place. 5 We leave aside examples of the so-called Slavonic accusative with infinitive {Videla chlapce vchazet - She saw the boy enter; divku zpivat - She heard the girl sing), since they are connect­ Slyšela ed with some unclear questions concerning the synonymy between Videla, jak chlapec vchazi (She saw as the boy was entering), Videla chlapce, jak vchazi (She saw the boy.as he was entering), or also Videla chlap­ ce vchazet (She saw - the boy - to enter - lit.), Videla, ze chlapec vchazi (She saw that the boy was entering), etc. We also left aside the constructions of the type Nechal si neco udelat (He had something done), Dal si neco opravit (He had something repaired), because they have a character of phrasal expressions. 6 There will be also other restrictions on verbal grammatemes, e.g. the requirement on indicative modality. 7 With some verbs it is necessary to take into account polysemy: bat se, ze se neco stane (to fear that something may happen) has a different meaning than bal se opustit matku (he was afraid to leave his mother).

THE CZECH INFINITIVE

141

8 Another idiosyncracy can be seen with the verb rozhodnout (to decide), where the construction with active infinitive (e.g. Š ē f r e d a k t 5 r rozhodl vydat sebrane spisy X.Y. - The editor-in-chief decided to publish the collected works of X.Y.) does not express the identity of Actor and the Subject (equal here to the Actor of the dependent predication), but corresponds to the tectogrammatical re­ presentation with a general Actor of the objective clause, which does not comply with the rules of coreference with the verbs of the class v1. 9 In this case it is necessary to modify some of the frames of verbs that were proposed in Panevova (1980). 10 From the point of view of the questions discussed here we focus our attention only on posterior predication, see above, Sect. 1.2. 11 Cf. Note 4 above for an unclear boundary line between Purpose, Direction and Intention, which comes to the foreground esp. in the cases governed by the verbs of the class V 4 . 12 There is, of course, a difference in the mode in which these conditions are discussed: in Sect. 2.11 we discussed all possible variants in the frame of coreference allowed by the verbs of the class V . In the table the verbs of this class are divided according to their behaviour: the first line of the table corresponds to the subclass of V 1 with obligatory coreference, the second line to a special case of coreference of the verbs V 1 when the controllee is a participant other than the Actor, and only the combination of the second column and the fourth line corresponds to the most common (unmarked) type of coreference of V 1 (Act : Sb). A similar division of special and unmarked cases can be seen in the table with the verbs of the class V .

REFERENCES Bauer, J. & M. Grepe. 1972. Skladba s p i s o v n ē c e š t i n y . [syntax of Standard Czech]. Prague. Bresnan, J. 1982. "Control and Complementation". Ling. Inquiry 13. 343-434. Conrad, R. 1969. Transformationsanalyse russischer Infinitivkonstruktionen. Halle. Hajičova, E., Panevova, J. & P. Sgall. In press. "A remark on Control", to appear in Prague Bull, of Mathemaqtical Linguistics. Chomsky, N. 1980. "On Binding". Ling. Inquiry 11. 1-46. Chomsky, N. 1981. Lectures on Government and Binding. Dordrecht: Foris. Koenitz, B. 1972. "Genus verbi in Infinitivkonstruktionen der tschechischen Sprache den Gegenwart". Zeitschrift fur Slavistik XVII. 636-646.

142

JARMILA PANEVOVA

Panevova, J. 1978. "Fragen der Nominalisierung zeitlicher Prǎdikation". Prague Bulletin of Mathematical Linguistics 30. 9-42. Panevova, J. 1980. Formy a funkce ve s t a v b ē česk è vety. [Forms and Functions in the Structure of Czech Sentences] . Prague. Panevova, J., Benesova, E. & P. Sgall. 1971. Cas a modalita v češti­ ne [Tense and Modality in Czech]. Prague: AUC, Phil. Mohographia 34. Peskovskij, A. M. 1956. Russkij sintaksis v naučnom osveščeniji. 119ff. Moskva. Platek, M., Sgall, J. & P. Sgall. 1984. "A dependency base for a linguistic description". Contributions to Functional Syntax, Semantics and Language Comprehension. 63-97. Prague. Poldauf, I. 1959. "Dej v infinitivu". [The Action in Infinitive.] Slovo a slovesnost 20. 183-202. Růžička, R. 1983. "Remarks and Replies". Ling. Inquiry 14. 309-324. Růžička, R. (this volume). "Control in Competing Frameworks". Sgall, P. 1967. Generativni popis jazyka a česka deklinace [Generative Description of Language and the Czech Declension.]. Praha. Sgall, P. 1979. "Remarks on text, language and communication". Text vs. sentences ed. by J. S. PetÖfi. 89-100. Namburg: Buste. Svoboda, K. 1962. Infinitiv v soucasne spisovne cestine [infinitive in Contemporary Standard Czech]. Prague. Šmilauer, V. 1966. Novočeska skladba [syntax of Modern Czech]. Prague.

MILAN BfLY and THORE PETTERSSON Uppsala

ERGATIVITY IN DYIRBAL

The purpose of this paper is to explain a number of phenomena associated with the ergative case as well as to reconsider the Australian language Dyirbal with respect to its alleged property of being an ergative language. More specifically, we intend to question the validity of Dixon's (1972, 1979) claim that Dyirbal is a genuine ergative language, which consistently and without exception exhibits ergativity in morphology and/or syntax. Since the paper is part of a research project in progress, we will first make a few remarks on the theoretical foundation of our analysis of Dyirbal. An outline of our model for the description of morpholog­ ical case systems has been presented elsewhere (Bily & Pettersson 1983, 1984); consequently, we will merely sketch out the main lines of the theory here. First, we stipulatively define morphological case markers as bound morphemes attached to and de­ termining noun stems via inflection. This definition automatically excludes languages such as English from the range of case languages. The few residuals of historical case marking in English do not qualify as case markers from a synchronic point of view, since

MILAN BÍLÝ AND THORE PETTERSSON

144

the so-called genitive 's is a NP determiner and case marking is otherwise restricted to pronouns, which, as the term implies, are not nouns proper but words used to replace full nouns or NPs. Our theory includes four components. The first views a case system as a genuine three-dimensional model of privative opposit­ ions between particular cases of a given language. Figure 1 shows the case space of Russian, where the letters signify the usually recognized six cases of Russian and the lines between the case nodes signify privative opposit­ ions holding between the cases in question. The second component Figure 1

consists of a set of explic­ it definitions of case

meanings, held to be universally valid. The definitions of the cases relevant for the discussion in the present paper are the following: Nominative: Something is said about x, denoted by X, where x is the kernel of a junction and X is the head of a syntactically independent constructions. Instrumental: Something is said about x, denoted by X, in order to say something about something else

ERGATIVITY IN DYIRBAL

145

(normally a non-x but also occasionally to qualify the predication itself). Ergative: Something is said about x, denoted by the first argument X. In the definitions, x stand for referent(s) and X for noun(s) or NP kernels. A junction is taken to be a one-sided relation between entities, where one entity determines the other, while the inverse relation is impossible. The term 'argument1 is used in the sense of Carnap (1921). The third component of our theory consists of a set of distinctive features based on the semantic interpretation of the case definitions and on possible three-dimensional case models. The feature system has no immediate relevance for the argumentation in this paper. The fourth component consists of hierarchies of cases with regard to the ability of particular cases in a given language to function as first or second argument markers, respectively. Since the first argum­ ent hierarchy is of some importance for the evaluation of the Dyirbal ergative, it is exemplified here (Fig­ ure 2( for the six syntactic cases of Czech (the vocat­ ive is excluded) and for the corresponding six central cases of Russian. The figure (where the arrow shows the ranking order) should be read as follows. Nominatives, being unmarked semantically, come highest and before other cases. They are followed by datives, genitives, accus­ atives, locatives together with all other preposition­ al phrases (neither Russian nor Czech has pure prepos-

146

MILAN BÍLÝ AND THORE PETTERSSON

Figure 2 ition-less locatives) and instrumentals in that order. Any case can function as the first argument of a pre­ dication, provided no other case coming earlier in the hierarchy is present in the predication. Since, however, the hierarchy forms a circle, the instrument­ al in languages such as Russian or Czech could be interpreted as coming before the nominative in spite of the fact that the nominative stands highest. Actual­ ly, they are in complementary distribution and the instrumental can be used in 'ergative' function under certain conditions, especially in Russian: (1)

Bratom napisano pis'mo a letter (N).

This is also true of PPs:

'Brother (I) has written

ERGATIVITY IN DYIRBAL (2)

147

U menja den'gi 'I (PP) have got money (N)' (lit. 'with me money')

Thus the hierarchy accounts very conveniently for why a true ergative, being the first argument case par preference, so often develops from an original instrumental or locative (cf. Trask 1979). The typological distinction between nominative-accusative and ergative languages can be drawn on the basis of several structural properties. Traditionally the nominative is the case of subjects in nominativeaccusative languages, regardless of whether the clause is transitive or intransitive, whereas the accusative is prototypically the case of the direct object (the patient). In ergative languages, the ergative is the case of the agent in transitive clauses, whereas the nominative is used as the case of the patient in trans­ itive clauses and of the subject in intransitive clauses, regardless of the semantic nature of the subject. The last decade has presented us with an addit­ ional distinction between syntactically, as opposed to morphologically, nominative-accusative and ergative languages. This distinction is based upon restrictions on certain grammatical rules, e.g. transformations in various transformational frameworks. Since Dixon (1972), most languages have been usually considered nominative-accusative from a syntactic point of view, while Dyirbal has been claimed to be one of the few syntactically ergative languages. There have been several other diagnostics present­ ed for determining the distinction between the two

148

MILAN BÍLÝ AND THORE PETTERSSON

types of language, taking into account voice and cross-reference marking in the verb among other things, but we will concentrate on the two referred to above. In our view, they are far from being satisfactory. The morphological definition no doubt comes about via an ethnocentric approach based on translations from exot­ ic languages into one of the "white" European languages. Martinet (1979: 39) is certainly right in his comment on such an approach: "The conclusion, normally implicit, is that 'ergative languages' illustrate a strange aberration probably representative of what people used to call the 'primitive mentality'." In order to avoid the ethnocentric view of ergative languages, the defin­ ition has to be completed with language specific or universal definitions of more terms than the definit­ ion itself defines. The problem is that there is no generally accept­ ed definition of subject. Rather there are a number of properties typical of the never properly defined notion of subject (cf. Keenan 1976: 303-333). There exists no satisfactory definition of transitivity either. Without trying to resuscitate the 19th cent­ ury view that the ergative equals the passive (e.g. Schuchardt 1896), it seems reasonable to admit that ergative constructions are in some sense related to the passive constructions of the traditionally studied European languages (cf. Wierzbicka 1980: 140ff). How­ ever, passive constructions differ from corresponding active ones by being intransitive, while only active constructions can be transitive. This, however, does not make the notion of transitivity in ergative lang-

ERGATIVITY IN DYIRBAL

149

uages any more transparent. The same criticism can be directed against the attempts to define ergativity at a syntactic level. Once again, one takes undefined concepts such as 'subject' and 'tr-ansitivity' as a point of departure. The notions, however, do not become more comprehens­ ible just because they are described in terms of deep syntax or surface constraints on coordination. More­ over, just as with the morphological definition, the argumentation is based on analogies with nominative-accusative languages. Anderson (1976: 11-21) claims that transformational rules such as Equi-NP Deletion, Subject Raising and Reflexivization apply in many ergative languages, just as they do in nominative-accusative languages, i.e. English, because English is claimed to exhibit nominative-accusative syntax! However, even if we were to accept the analogy reason­ ing as legitimate, it still remains to be proved what has been merely claimed, viz. that it is the subjecthood of a given NP that is the ultimate cause of the NP's involvement in these transformational rules and that it is not a consequence of some other factor. As for Dyirbal, Anderson asserts that the transform­ ational tests give the opposite result, which means that the Dyirbal nominative is also the sole subject case. He concludes that "Dyirbal is really ergative in a fundamental syntactic sense, while most other morphologically ergative languages are ergative only superficially: in syntactic terms, they are accusat­ ive" (Anderson 1976: 18). For languages similar to Dyirbal, therefore, "something like the 'underlying

150

MILAN BÍLÝ AND THORE PETTERSSON

passive' theory appears to be correct"

(Anderson

1976: 1 7 ) . This may very well be so and it is not even implausible that the ergative construction of Dyirbal may be a diachronically passive construction. But all this leaves us

with the main questions unanswered:

what is 'subject' and what is

'transitivity'?

With respect to the latter question we shall not be able to deliver a universally valid answer. But as far as genuine case languages are concerned we can give a water-proof stipulative definition: a transit­ ive predicate is a two-place predicate the first argument of which is in the nominative and the second argument of which is in the accusative. Optionally we could restrict the definition to apply to only the case marking of the second argument and define transit­ ivity accordingly: a transitive predicate is a two-place predicate the second argument of which is in the accusative. The latter option would allow us to accept Russian sentences such as (3)

Pulej ubilo soldata

'The bullet

(I) killed the

soldier (A)' (4)

Mne žal' ego

'I (D) pity him ( A ) '

as true transitive sentences. In our opinition, this opinion is to be preferred, since it is more econom­ ical. Otherwise, any of the definitions of transitiv­ ity suffices for our purpose, which is to claim that the typical ergative construction of an ergative language is never to be regarded as a transitive clause. The second argument of such a clause will never be in the accusative, no matter which argument

ERGATIVITY IN DYIRBAL

151

is taken as the subject. Our definition, accordingly, allows us to consider the ergative construction as intransitive without reference to some underlying passivity or passive origin. This is important, be­ cause there is a noticeable difference in the promin­ ence of the agentive NP in ergative constructions compared to the agent of passive constructions. We are now also able to clarify the notion of subject to some extent. Note that in the definition of the nominative we include the requirement that a noun in the nominative should be the head of a synt­ actically independent construction. The syntactic independence of nominatives is most clearly demonstr­ ated by the fact that they do not need to be part of any syntactic construction in order to function linguistically. Typically, nominatives are the lexic­ al forms of nouns not only in nominative-accusative but also in ergative languages. A nominative can also be used linguistically as a pure denomination, i.e. in Russian: (5) On čital knigu "Anna Karenina" book (A) "Anna Karenina" (N)'

'He has read the

One of Keenan's basic subject properties is the following: The entity that a basic subject refers to exists independently of the action of property expres­ sed by the predicate (Keenan 1976: 311). In fact, this is nothing but a description of the property "syntactic independence". No wonder then that the nominative is better suited to be the first argument and nominal predicate than other cases. In non-ergative languages, it is typically the first argument of

152

MILAN BÌLÝ AND THORE PETTERSSON

a predication that is treated as syntactically in­ dependent, though far from always. In Slavic languages, for example, there is a split between first argument and grammatical subject in sentences with demonstrat­ ive pronouns as subjects: (6) Eto byl ego brat

'That

(7) Eto byla ego sestra his sister

(neutre) was

(masculine)

(masculine)'

his brother

'That

(neutre) was

(feminine)

(feminine)'

The same holds true of sentences with the often disput­ ed oblique case subjects in Russian, Finnish, Latin and other languages: (8) Berega ne pokazyvalos'

'No shore

(G) could be

seen' (9) Menja tošnit

'I (A) feel sick'

The subjecthood of such oblique NPs is due to the fact tnat they

are the first and only argument of their

predications. Therefore, one has either to accept the status of first argument as the criterion for subjecthood or else take the morphological nominative mark­ ing as the determining factor. Now consider the translation of "The man killed the friend" in Basque and Georgian, respectively: (10)

Gizonak adiskidea

(era)hil du 'Man (E) friend

(LSI) kill AUX' (11)

kacma mokla megobari

'Man

(E) killed

(aor)

friend ( N ) ' To ask whether the E or the N noun in such sent-

ERGATIVITY IN DYIRBAL

153

ences is the subject is of course meaningless, as has already been pointed out by Blake (1976: 281-300), Verb agreement is of no help here and the split between the two subject properties is divided between the two arguments, gizonak/kacma being marked as first argument and adiskidea/megobari being marked as syntactically independent. The same thing could be said of the Russian sentence (1) above, but observe that this does not hold true of (12)

Pis'mo napisano bratom brother (I) '

'Letter (N) written (by)

(13)

Pis'mo napisano karandasom 'Letter (N) written (with) pen (I)'

In both sentences the noun pis 'mo is a syntactically independent first argument and the existence of the letter is asserted as dependent on the I argument as a means for the fulfilment of the action expressed by the predicate word. In (1) the I noun happens to be the first argument, but this fact does not make the I noun an ergative. Rather, the ergative function is implied by the definition of the instrumental, while the opposite does not hold. Therefore, we have to conclude that a prerequisite for distinguishing a separate ergative case in a particular language is that it be morphologically distinct from other oblique cases of the language in question. This is the situation in Basque and Georgian but not in Russian. As a matter of fact, Dyirbal is not very different from Russian in this respect. It has an instrumental that can be used with nouns denoting both living beings and things, but the fact that the instrumental freely takes an agentive function is, of course, no reason to postulate a separate ergative case. Dyirbal has three ways of render­ ing "The man is beating the woman", viz.

MILAN BÌLÝ AND THORE PETTERSSON

154

(14)

balan d'ugumbil bangul jarangu balgan DET

woman (N)

DET

man (I)

hit

(15)

baji DET

jara bangun d'ugumbiru balgalnanu man (N) DET woman (I) hit (antipass.)

(16)

baji DET

jara bagun man (N) DET

d'ugumbilgu woman (D)

balgalnanu hit (antipass.)

(16) implies that the woman did something to deserve being beaten. (14) and (15), on the other hand, have the same cognitive meaning. If we understand Dixon correctly the fine distinction between the two sentences can be paraphrased accordingly: (14) N is such that I had to beat vs. (15) N is such that I (undeservedly) was beaten. In both instances the interpretations follow immediately from our definition of the instrumental: something is said about x in order to say something about something else. Observe furthermore, although an agentive reading of the I noun of (14) is natural, this is quite inappropriate with respect to the I noun of (15), the referent of which is the semant­ ic patient. For this reason, among other things, it is clear that any reference to 'agenthood' is insufficient for defining the ergative case. In fact, the first argument criterion is the only functional definition. Observe furthermore that there is no possibility to take the N noun in any of the sentences to be the first argument; both Anderson (1976) and Dixon (1972) give ample evidence that the N noun is the only possible subject, which is to say that it is simultaneously first argument and syntactically independent. Thus both bangul jarangu of (14) and bangun d'ugumbiru of (15) are true instrumentals. Why then does Dixon insist in claiming Dyirbal to be an ergative lang­ uage?

155

ERGATIVITY IN DYIRBAL

The reason is presumably psychological. If a certain linguistic expression is found to contain more than one distinct function or meaning, it is quite natural to want to recognize separate lexical entities corresponding to the functions in question. This kind of reasoning is not different from that used to distinguish two lexical items can other

in English, the one meaning 'is able to' and the 'vessel, container'. What is therefore at stake is

the eternal question of how to distinguish polysemy from homophony. The task is not always easy to settle. Often it is solved according to the linguist's own preferences. Dixon is obviously very homophonic by nature, while we are more ready to make use of Occam's razor. Given our own prejudices, we will now scrutinize Dixon's argumentation. Dixon (1972: 62 and 94 ff.) disting­ uishes two functions of the instrumental, the "ergative" and the instrumental proper, on the basis of their differ­ ent syntactical behaviour. Only nouns in the postulated ergative case become nominatives after the antipassive transformation has applied: (19)

balam d'ugur DET

yam (N)

bangugaragu

bangu

two people (I) DET

gad'indu

bagan

yamstick (I) dig

'The two people (I) are/were digging yams (N) with a yamstick' (2)

balagara bagum d'ugurgu bangu gad'indu bagalnanu 'The two people (N) are/were digging yams (I) with a yamstick (I)' (antipassive form of the verb)

This is of course no valid argument. We could apply the same kind of argumentation to Russian and postulate a "temporal" and an accusative proper on the basis of the

MILAN BÌLÝ AND THORE PETTERSSON

156

fact that accusatives of time are not influenced by passivization as opposed to accusative objects, i.e. they remain in the accusative, while objects become nominative subjects. But an analysis based on such argumentation would lead to a disintegration of the linguistic description into un­ systematic fragments and must therefore be rejected. There are furthermore constructions, considered by Dixon to be intransitive, where the human agent is in the nominative and the patient is in the instrumental or the dative: (19)

baji jara bangu d'alguru rubin'u 'Man (N) meat (I) ate'

(20)

baji jara bagul barangu

'Man (N) (is going out to)

concern himself (with) wallaby (D)' An interesting phenomenon of Dyirbal grammar involves so-called implicated verbal complexes. From (21) and (22) one can form: (21)

baji jara walman'u

'man got up'

(22) (23)

baji jara wajn'd'in 'man went uphill' baji jara walman'u wajn'd'in 'man got (and then) went uphill'

The action referred to by an implicated VP is only possible by virtue of another event having taken place. Either the implicating event is performed as a necessary preliminary to the implicated action, or the implicated action is a natural consequence of the former event. The latter alter­ native may be illustrated by (14) and (24)

balan d'ugumbil bad'in'u 'woman (N) falls down'

These two sentences will give rise to (25)

balan d'ugumbil bangul jarangu balgan bad'igu

ERGATIVITY IN DYIRBAL 'man

(I)

157

hits woman (N) (causing her to) fall

down' bad ' i g u 'fall down' receiving the 'purposive' inflectional marker -gu instead of the ordinary tense marker. Such co­ ordination reduction is not restricted to non-agentive nouns in the nominative. In fact, Dyirbal has syntactic means to link any two sentences with one NP in common, whatever the function and the case in the two sentences. Thus two sentences with an identical instrumental may be conjoined as well: (26)

bala jugu bangul jararŋgu nudin baji n'alŋga bund'un

'man (I) cut tree (N) and spanked child

(N) ' Dixon (1972: 134) claims that such examples involve two separate sentences where the actor is left unspecified in the second clause, paraphrased as "child was spanked". On the other hand he does not want to apply the same sort of analysis to pronominal agents: (27)

ŋad'a

bala jugu juban balan d'ugumbil d'ilwan

'I (N) put down the stick (and I) kicked the woman (N)' Dixon takes such examples to be conjunctions of two clauses with an identical agent. The problem is that personal pronouns of the first and second persons have different case marking from full nouns. These pronouns have the same case when functioning as 'subjects' of both one and two-place predications, while a special accusative form is used for 'patients' in the case of two-place predicates. In other words, the pro­ nominal system has the nominative and the accusative instead of the nominative and instrumental of full nouns.

158

MILAN BÌLÝ AND THORE PETTERSSON

The solution chosen by Dixon is to analyse the pronominal 'transitive subject' as a 'deep ergative'. This is due to the syntactic behaviour of pronominal nominatives, which, in one-place predications, conjoin clauses in the same way as full noun nominatives do. That is to say, they are supposed to stand for 'deep nominatives', while the pro­ nominal 'transitive subjects' in the nominative behave as full nouns in the postulated ergative case. (28) ŋad'a banin'u baŋgun d'ugumbiru balgan

'I (N) came

here (and I) was hit by woman (I)' Accordingly, Dixon himself would have to admit that his deep case analysis leads to the conclusion that "I put down the stick (and I) hit the woman" is an indisputable instance of conjoining two clauses with the same I noun. There are two ways of conjoining and I and a N noun referring to the same entity: one can either use the above-mentioned antipassive construction, which turns the matter into a case of identity between two nominatives, or one can resort to another construction, formed by conjoining two clauses regardless of the case of the referentially identical nouns. By this process (29) and (22) give (30): (29)

bala jugu baŋgul jaraŋgu madan

'man (I)

threw stick

(N) ' (30)

bala jugu baŋgul jaraŋgu madan wajn'd'inura

'man

(I) threw stick (N) (and man) went uphill' Consequently the subjecthood of N cannot be established on the basis of restrictions on conjoining and Dixon's deep case analysis thus becomes completely vacuous. It turns out that many facts of Dyirbal can be re­ interpreted in the non-ergative way. For example, the preferred word order is initial N noun, regardless of

ERGATIVITY IN DYIRBAL

159

whether it is derived transformationally or not. This is reminiscent of the situation in our familiar European nominative-accusative languages. It iw worth noticing that even the word order of the Dyirbal pronouns identifies them as true nominatives and accusatives. Cf. (14) and !

(31) ŋad'a baji jara balgan (32)

ŋd'a

ninuna balgan

(33) ŋinda ŋajguna balgan

I (N) man (N) hit'

'I (N) you (A) hit' 'You (N) me (A) hit'

What Dixon (1979: 62ff) calls

"ergative syntax" may

therefore just as well be called "nominative syntax", since it involves the same treatment of the N noun, the syntact­ ically independent noun, as in ordinary nominative-accusat­ ive languages. In fact, Dyirbal is even more nominative accusative what is isually acknowledged..It is a rarely mentioned fact - since it apparently disturbs the idea of Dyirbal being an ergative language - that the nominative of socalled transitive sentences can be optionally substituted by a clear-cut accusative. This is said to be restricted to proper nouns, other nouns referring to human beings, and, occasionally, to some inanimate nouns (Dixon 1979: 88). To say that the accusative is optional is, of course, just a way of justifying why this phenomenon is ignored in grammatical descriptions. Actually, this type of construct­ ion is reminiscent of the Russian so-called semi-passive construction exemplified above in (3). However, since this construction has not been seriously analyzed

with respect

to Dyirbal, and we have no other data available, we refrain from further comments. Be this as it may be, the only reasonable conclusion to be drawn from Dixon's data is that Dyirbal is no ergat­ ive language. It has, however, one property in common with

160

MILAN BÌLÝ AND THORE PETTERSSON

true ergative languages in that it has a predominance of intransitive clauses. This can be seen from the following table, demonstrating our reinterpretation of Dyirbal clauses with NP f s referring to 'agents' and 'patients'.

Construction type

Case of 'agent'

Case of 'patient'

Remarks

active intransitive

N

-

I active intransitive

N

D

active intransitive

N

I

active intransitive

I

N

So-called ergative

active transitive

I

A

Patient NP taking A referring to humans or inanim­ ates

active transitive

N

A

Patient as above; personal pronoun of 1st or 2nd person

active intransitive

N

N

Agent as above

N first argument

"

All above-mention­ ed constructions unmarked as to morphological marking of verb antipassive intransitive

N

N

antipassive intransitive

N

I

Table 1

Verb morphologic­ ally marked

"

ERGATIVITY IN DYIRBAL

161

Owing to its hypertrophy of intransitive sentences at the cost of transitive ones, Dyirbal is similar to ergative languages, which by definition, provided they lack an accusative, have intransitive sentences only. Similarity, however, is not enough to establish a proper taxonomy for any kind of typology. Man, for example, is in many respects similar to the ape, but this fact does not per se allow us to consider man as really being an ape. At least, we would not hope so.

REFERENCES Anderson, S. 1976. "On the Notion of Subject in Ergative Languages". Subject and Topic, ed. by C. N. Li. 1-23. New York. BìilÝ, M. & Pettersson, T. 1983. "Towards a Theory of Morphological Case". Papers from the Seventh Scandinavian Conference of Ling­ uistics, ed. by F. Karlsson. 568-582. Helsinki. Blake, B. J. 1976. "On Ergativity and the Notion of Subject in Some Australian Cases". Lingua 39. 281-300. Carnap, R. 1921. Introduction to Symbolic Logic and Its Applications. New York. Dixon, R.M.W. 1972. The Dyirbal Language of the North Queensland. Cambridge Studies in Linguistics 9. Cambridge. Dixon, R.M.W. "Ergativity". Language 55. 59-138. Keenan, L.E. 1976. "Towards a Universal Definition of 'Subject'". Subject and Topic, ed. by C. N. Li. 303-334. New York. Ergativ­ Martinet, A. 1979. "Shunting on to Ergative or Accusative". ity. Towards a Theory of Grammatical Relationsf ed. by F. Plank. 39-43. London. Schuchardt, H. 1896. "Ueber den passiven Charakter des Transitivs in den kaukasischen Sprachen". Sitzungsberichte der philosophischhistorischen Classe der kaiserlichen Akademie der Wissenschaften in Wien 133. 1-90. Trask, R.L. 1979. "On the origins of Ergativity". Ergativity. Towards a Theory of Grammatical Relationsf ed. by F. Plank. 385-404. London. Wierzbicka, A. 1980. The Case for Surface Case. Ann Arbor.

PETR PIŤHA Prague

FOUR (SIMPLE) REMARKS ON COORDINATION

Observing Sgall's comprehensive work well rooted in European linguistic tradition and developing the ideas of Prague school one can find (among others) four prominent points of his approach to natural language. Each of them will be illustrated by a small example showing advantages of such an approach. (Readers are asked to divide the simplicity of their presentation between the author's capacities and the space limits given by the editors.) (a) Following the Prague school Sgall insists on distinguishing two distinct levels of syntax. This broadly accepted idea is however not used by all linguists. Accordingly there are in this description two distinct notions of coordination, that of a meaning (tectogrammatical, deep) level and that of a syntax (surface) level. Both levels are bound together by the functional relation of form and function, which may account for homonymy and synonymy. There are e.g. instances where syntactic coordination is just a form of quite another meaning relat­ ion. Two different modifications in certain verbal frames might be overtly realized as coordination, e.g.

odlisit A od B - odlisit A a B distinguish A from B - distinguish

A and B

164

PETR PITHA přirovnat liken

A k B - přirovnat A to B - liken

spojit A s B - spojit A a B connect A with B - connect The mentioned type type, where on the syntactic level is Mother with father have already come.

A a B

A and B

A and B

should be distinguished from another other hand a with-construction on the a realization of a deep coordination: have already come = Mother and father

It has been shown in Pitha and Sgall (1972) that it is not possible to account for the difference between sentential and phrasal coordination by identifying phrasal coordination with some kind of deep WITH-relations. There is also a meaning of accompaniment at the meaning level distinct from coordination, which is syntactically realiz­ ed by with-constructions. One might argue that the sent­ ence Mother came with father is somehow derived from the sentences Mother came and father came. The negative counterpart Mother did not come with father shows evident­ ly, that the scope of negation contains just with father (and on one reading also the verb) as its focus, totally independent from mother, which is a part of topic. There are of course plenty of details we leave aside, because what we want to show is that the mentioned ambiguous sentences are describable only if we work (i) with a level of meaning, on which we distinguish three items? sentent­ ial (loose) coordination, phrasal (close) coordination and accompaniment, and (ii) with a syntactic level, on which two ambiguous constructions - conjunctional coordin­ ation and with-construction represent these three mean­ ings .

REMARKS ON COORDINATION

165

(b) In the broad context of modern semantics Sgall argues for a precise distinction between meaning and content, the former being a part of language and consequent­ ly an object of linguistics, the latter being an extralinguistic phenomenon and consequently in the scope of other sciences, as logic, psychology, cognitive science etc. Mutual enrichment and fruitful cooperation of these sciences must be based on a well defined division of labour. The requirements Sgall formulates are not small. His definition of linguistic meaning combines semantic and pragmatic features and finds a link between the struct­ ures of language and truth-functional semantics. However, he insists on admitting as distinctions in meaning only such phenomena that are detectable in the language system. The difference that cannot be identified on the basis of grammatical structure and has to be characterized using our knowledge of the universe or logical rules is not a difference (opposition) from the level of meaning. The analysis of with-constructions shows that fairly well. In cases where the with-construction is a form expressing accompaniment we might in some instances according to the rules of functional sentence perspective (FSP) state the meaning and the presupposition (as an additional informat­

ion). E.g. the father in the mother is not supposed that

sentence S otcem v divadle není matka (with theatre is not mother lit.) means that with father in the theatre and it is pre­ father is in the theatre. Linguists might be

tempted namely because of some applications in which inferences play an important role to divide the meaning of the mentioned sentence into two elementary pieces

Mother is not in the theatre

and Mother is not with

father

and to try then to describe also with-constructions express-

166

PETR PIHA

ing coordination in terms of these pieces (three in the given case) of elementary information. But the sentence Mother with father are not in the theatre simply means these two are not in the building. It is purely logical, extralinguistic matter that in case M is with F, F must not be in the theatre; that in case M is not with F, F might be in the theatre; that in case both are out of the building we know nothing about their being together and that some combinations of this elementary information lead to a logical contradiction. On the other hand, however, it is not always enough for a description of meaning to use a mere extensional point of view. Czech sentences with the conjunction i (roughly and also) show this very well. The main difference of a {and) and i {and also) is that the latter has only the loose meaning while the former is ambiguous (close or loose). The sentence K. a J. maji knihu {K, and J. have a book) is homonymous, they each may have a book or they may have just one book together. The sentence K. i J. maj i knihu (K. and also J, have a book) has just the former meaning. This feature of i should be of course described and it serves in fact as an argument for distinguishing two types of coordination on the level of meaning namely the loose and close ones. When the sentences Večer prijde oteo a matka {Tonight will come father and mother) and Večer přijde oteo i matka {Tonight will oome father and also mother) are translated into the predicate calculus, then they both express the same logical conjunct­ ion and share the same truth conditions. However, linguist­ ic meaning differs, because the conjunction i brings an additional information that the appeareance of mother is either more important or less expected than that of father. In this sense linguistic meaning is more subtle

167

REMARKS ON COORDINATION

than the distinctions covered by predicate calculus. this difference should be respected in the study of natural language semantics. (c) Sgall has always worked with dependency syntax, in which of course the prominent role of verb and it's valency is very transparent. It is obvious that coordinat­ ion of verbs is conditioned by their valency frames. Two verbs are conjoinable when they share the same frame, e.g. Father

has shown and given

the

book to Paul,

Two verbs

are not conjoinable when their frames differ remarkably. Sentences My father

bought a house and My father

looks

well might be conjoined only as two sentences. It is a question of further empirical investigation to show which frames are compatible (similar enough) in order to enable verbs to be conjoined, e.g. The lost ring was found and given

to the

owner by a

policeman.

It looks like that the process of communication is based on verbal frames patterns. Namely, the hearer seems to make predictions about what should follow according to the verb used by the speaker. Consider the following examples: (1)

(2)

(3)

Otec

postavil

sold

a house)

Oteo postavil

sold)

Oteo loni

postavil

last

a prodal

loni-a

year and sold loni

dum,

a prodal,

and sold

Oteo postavil built

(6)

built

Oteo postavil last

(5)

dum

house and father (4)

a prodal

a

built

and

{Father

built

a

dum, dum,

dum a prodal

a loni

year

(Father

built

house.)

year a house and sold

Oteo postavil

(Last

house,)

prodal a

{Father

prodal

dum,

ho,

{Father

it,) {Father

built

168

PETR PITHA

and last

year sold

a

house,)

Sentence (3) in comparison with (4) shows, that in Czech the frame of the first verb must be completed if there is some modification after the verb. In such a case the possibility of coordination of verbs is blocked and coord­ ination must take the form of coordination of two sentences as in (5). Sentences (1) and (3) show that a conjunction immediately after the verb serves as a sign of a new start and only after the second verb the frame is completed. Sentence (6) shows even that the new start might be broader than the original one. Also the phenomenon of so-called gapp­ ing is well explicable in these terms. In the first part of sentence (7) there is a fully developed frame. In its second part there is just a lexical material, which re­ mains structured for the hearer because the preceeding frame is not replaced by a pattern of another verb.

(7) Otec dal Marii obrázek (Father gave a picture to Charles)

a strýe Karlovi knihu. to Mary and uncle a book

(d) In Sgall's description of language special care is given to the description of FSP. At first glance this point of view seems to be of little importance for coordination, since every coordinated structure is a separate field of FSP. However, the role of FSP is very important in the study of coordination because the usual requirement of the same syntactic role (sometimes follow­ ed by sameness of morphological form) might be suspended when two foci proper are conjoined e.g.

Don "t forget, that he has done it yesterday for Eva we will not do it today and without a reason.

and such

169

REMARKS ON COORDINATION Also coordination of wh-words of completely different syntactic roles is based on the same status in FSP and then supported by morphological likeness e.g. I do know,

whom and when to answer.

not

On the other hand, the

requirement of the same role in FSP is an absolute rule. Two sentences being same as to syntactic. structure and lexical cast but different in FSP cannot lead to a phras­ al coordination and must be coordinated as two sentences, perhaps with pronominalizations, but with no deletions in the second of them, e.g. Oteo Karlovi Otec

knihu

ukdzal Karlovi

Father

showed

book Father

gave

knihu. dal.

Charles the

book

a to

=> =>

Charles

Otec

Karlovi

dal Otec

knihu. Karlovi

ukdzal

a

ukdzal

knihu a dal mu j i . F showed and gave a book to Ch F showed the and gave

book

it

to

to Ch

him

Also the examples presented by Dik (19 81) as contrastive focus e.g. Charles

broke

not

a mirror

but

a

window

show that FSP is important in description of coordinat­ ion and vice versa.

REFERENCES Dik,

S. X. et al. 1981. "On the Typology of Focus Phenomena". Perspectives on Functional Grammar ed. by Hoekstra et al., 41-74. Dordrecht. Pitha, P. & Sgall, P. 1972. "Coordination in a Generative Descript­

ion". Prague Bulletin

of Mathematical Linguistics

18.

III. BELOW THE SENTENCE STRUCTURE

HARTMUT HABERLAND Roskilde

A NOTE ON THE 'AORIST'

1. In a scholium to Dionysius Thrax (Hilgard ed. (1901: 250)), the commentator compares the Greek aorist to the future tense by pointing out two parallels: first, the forms of the aorist and the future contain the same con­ sonants, second, they also share the property of being in­ determinate,

"since if you were to say 'I struck', you would not determine the point of time [of the action], only that you have performed it; and again, whenever you say 'I will strike', you have not signified the point of time, except only that you will strike; but you do not make it clear, when." Similar statements are found in two other scholia one which has:

174

HARTMUT HABERLAND "the aorist does not indicate anything with regard to any particular time" (Hilgard ed. (1901: 249)),

and another, attributed to Stephanos (Hilgard ed. (1901: 251)), where it is said that the aorist got its name be­ cause, in contradistinction to the perfect and pluperfect, it does not limit the time segment it refers to. These statements are difficult to interpret for several reasons: First, it is hard to see exactly what is the semantic status of the different terms in the scholia for (roughly speaking) 'to mean'. Second, these scholia are texts commenting on another text, viz., Dionysius Thrax' which in its turn makes statements about the grammar of a different set of texts, to wit, those of classical Greek literature. We do not know exactly when these scholia were written (6th century AD would be a good guess). Nor is it established clearly when Dionysius' text (or possibly, the text attributed to him) was composed (for this problem, see Pinborg 1975). But at least it is quite clear that the Scholiast wrote at a time which is considerably later than Dionysius', who again wrote at a time considerably later than the literary texts which were the object of the philological interests of himself and his like. Therefore, the Greek language, at the time when the scholia were written, was markedly different from both the Greek of Dionysius (and his contemporaries), and that of Homer or even the classical Attic authors. The same applies to the use of the tenses, especially that of the perfect (see Chantraine (1927) and Mihevc (1959)). Of course, neither the author(s) of the scholia, nor Dionysius Thrax himself were necessarily aware of this - actually, living in a diglossic situation, they might have believed that there

A NOTE ON THE 'AORIST'

175

was no difference between 'their' Greek and that of the classical Attic or earlier literature - but certainly we can assume that the language each of them spoke was different, both with respect to the language of the others, and to that of the texts on which they commented in their philological studies. Third, the scholia reflect - as is generally agreed upon - the theory of tenses of the Stoa. But it is not clear to which extent the Scholiasts still actually under­ stood Stoic grammar, or whether they just were rationaliz­ ing the latter's terminology. A rationalization of this kind (which may have been necessitated by the change in meaning of the tenses between the time of the Scholiast writing and the 'golden' period of Greek literature) is undoubtedly at work in another comment, to be found in the same context as the re­ marks quoted above. Here, the aorist is explained as being equivalent to a perfect when combined with a time adverb­ ial like 'just now', while with a time adverbial like 'long ago', it is said to be equivalent to a pluperfect. This latter theory has had repercussions up to, and including, the German grammarians of the 17th Century, who define the German perfect as the "jüngst" "vergangene Zeit" (like Schottel). In a certain way, this doctrine echoes Apollonios Dyskolos' discussion of tense meanings. Not only did the latter maintain that the perfect was a present tense, but he also investigated the tenses as to their possible cooccurrences with time adverbials - a theory which again is reminiscent of more modern ideas, such as developed by the Kahanes (1954), Crystal (1966), and Wunderlich (1970). Still, there is the specific claim about the aorist of not indicating any particular time. On a certain read-

176

HARTMUT HABERLAND

ing, this claim makes little sense, and hence lays itself wide open to ridicule. Thus, Max Polenz says: "Die Späteren halfen sich damit, der Aorist bezeichne keinen festen Zeitpunkt, da erst durch einen Zusatz wie oder der Abstand von der Gegen­ wart klar würde - als ob das bei anders wäre!" (Pohlenz 1939: 184) On the one hand, it is true that, taken by themselves, neither the aorist nor the imperfect indicate the distance of an event from the present time (or the point of speech, as we would say) for the simple reason that this is not what tenses are for. Rather, such a function properly belongs in the domain of the time adverbials. Therefore, Pohlenz is right in remarking that the difference between , and the imperfect cannot be in the aorist that the first does not indicate the time of event (and therefore is indefinite), whereas the latter does. On the other hand, I do not think that the matter is as easy as Pohlenz seems to believe. In fact, there is an additional catch here, and it is one which will turn out to be rather significant. 2. What the Stoics must have meant is something like the following: Taken by itself, the aorist (or the aorists, since for the Stoics, the future was as much of an aorist as the aorist of the past, which we call the aorist simplic­ ite?) do not contribute anything to time reference. Now it is not quite the case that there are no tenses which locate an event in time: the (simple) present tense does, in that it marks an event as occurring at the speaking point, or during a time span overlapping with the speaking point. We could say that in a tensed sentence, there are two empty 'slots' for point-of-time references (we may call them

A NOTE ON THE 'AORIST1

177

point of speech and point of event), and that for the (simple) present tense, these 'slots' are bound by the same λ-operator: λt o (...t o ...t o ...) In this sense, one could say that the meaning of the (simple) present tense is simply contextually bound; although one has to distinguish both points of time, a single variable and a single λ-operator will do the job of binding, since both points of time are identical anyway. For the other present tense, the perfect, the second vari­ able does noc necessarily have the same value as the first (actually, in post-Homeric Greek, the former's value must antecede the latter's on the time axis); however, it is bound by an existential operator, which means that also here, the whole formula is only open at one point: λt o 3t'(...t o ...t'...), where t' < to For the aorist, we get a different picture. Here, the value of the second variable is neither identical to the first nor bound by an existential quantifier. It has to be bound by a second λ-operator, which gives us the following:2 λtoλt'(...to...t'...) Thus, the aorist could be called 'indefinite' because the formula representing it is more 'open' than those for the present and the perfect. It has one more open 'slot* and is therefore less determined by one degree: it is 'indefinite'. Plausible as this line of thought might appear, it will bring us into trouble with the other past tenses of Classical Greek, the preterite and the pluperfect. Since the pluperfect is not just a past of the perfect, but actual-

178

HARTMUT HABERLAND

ly its [imperfect] (at least according to Stoic doctri­ ne) , any solution for the imperfect will carry over to the pluperfect. If we maintain the idea that the Classical Greek tenses differ by their way of binding variables for points of time, then there is virtually only one possibility to bind the second time variable for the imperfect, namely by universal quantification. In this case, we get for the imperfect: where t' < to and accordingly for the pluperfect: where t" < t'

(and

probably, t' < t , ever after Homer). Of course, this would mean that meant something like 'I always struck'. This is hardly the last word on the meaning of yet, this explanation is not altogether unreasonable if we consider that the Greek imperfect, among its other uses, has that of referring to habitual actions and to repeated events which actually can be interpreted as occurring 'always', at least within a certain universe of discourse. This attempt to do justice to the Scholiast's remark about the 'indefiniteness' of the aorist is of course rather mechanical (in addition to being speculative): moreover, with regard to the meaning of the tenses involved, our anal­ ysis is not necessarily more correct. What it does show, however, is that it makes some sense to talk about the aorists as 'more indefinite' and 'more contextually depend­ ent', compared to the other tenses: they are the only tenses with two λ-operators and therefore take two variables, not just one, as all other tenses do.

A NOTE ON THE 'AORIST'

179

Be this as it may, the Scholiast's other explanation according to which the perfect should contain the adverb is a rationalization of the original Stoic idea; a rationalization which imposed itself once the know­ ledge that the perfect in Classical Greek was a present tense had been lost. If the perfect was to be considered a past tense, it had to be a special kind of past - that is, the past of recent events. For Homeric Greek, and even for Classical Attic, this simply will not wash; but what else can one expect in a period when the classical perfect tense no longer existed in the spoken language, at least not with a meaning different from that of the aorist? 3. It is always dangerous to project modern theoretic­ al distinctions onto theories of the past (even though it has been done many times, also in recent linguistic histo­ ry) , Even so, it does not seem too daring to invoke the distinction here between 'sign meaning' and 'cognitive content' (Sgall 1977). After all, the Stoics, whose theory of the tenses our Scholiast obviously had some knowledge of (even if its actual content has been reduced to a term­ inology which was not too well understood any more), distinguished between a relationship of signifying (between linguistic form, and linguistic meaning, the significatium,

, and one of denotation

between meaning and the event respectively the object (Christensen 1962: 45) . Correspondingly, we distinguish between sentences (as linguistic forms), meanings of sentences (that is, functions from points of reference to propositions), and the content of a sentence (that is, a proposition: a function from points of refer­ ences to truth values). The schemata I have suggested earlier represent funct-

130

HARTMUT HABERLAND

ions; but what kind of functions are they? And what are the domains and the ranges of these functions? If we look at the schema for sentences (or clauses) having a present tense, we can see that it contains only one argument, and that this argument has to be a point of time, namely the point of speech. So we can take this schema as a schema for propositional functions which take points of speechs as arguments (which have the set of (some or all) points of speech as their domain) and yield as their values propositions. The meaning of 'I strike1 (or rather 'I am striking') would then be a function that takes a point of time and maps it onto a proposition, which in its turn is a function that takes a state of affairs (a possible world) and maps it onto either one of the truth values T or F. The same happens with all the other non-aorist tenses: Even though a second (or even third) point-of-time variable may occur, such a variable is bound by either an existential or a universal quantifier; hence this variable (or the two of them in the case of the pluperfect) does (do) not correspond to an argument of the propositional function in question. Stated informally: the event referred to by the verb still plays a role in the make-up of the world which is taken as an argument by the proposition, but its event time does not contribute to the truth value rendered by the proposition, since the formula merely specifies 'some time' (in the case of the perfect), or 'always' (in the case of the imperfect and pluperfect). The aorists are different in this respect. True, the meaning of a clause containing an aorist tense is properly a function from a point of time to a proposition, but the proposition still contains a λ-bound variable for the point of time at which the event denoted by the verb occurred. So here the proposition does not just take a state of

A NOTE ON THE 'AORIST'

181

affairs and maps it onto the set {T, F } : it takes a pair of a state of affairs and a designated point of time, that is, the point of event of the event referred to by the verb. In this sense, again, the aorists are 'indefinite', or rather, 'indetermined'3 , since their interpretation draws more heavily on the context than does the inter­ pretation of the other verbal tenses. This is exactly what makes the aorist the most obvious narrative tense: here, the context is given by the chain of narrated events of the story. The formula for the aorist given above applies both to the 'past' and the 'future' aorist in the Stoic system. The difference between the two of them can be stated now. The two points of time involved have a different status. In the past aorist (or simply: aorist), the point of event is earlier than the point of speech; in the future aorist (or simply: future), it is later. But since we have decid­ ed that the 'outer' λ-operator marks the argument of the meaning of the sentence, whereas the 'inner' λ-operator marks an argument of the proposition involved, they have a clearly different status as well, hence it makes sense to write λtQλ.t' (. . .to . . .t' . . .) , where t' < to for the (past) aorist, t' > to for the future (aorist). This analysis, of course, does not say anything about time adverbials such as

or

. The second part of

the Scholiasts' explanations (that involving time adverb­ ials) is clearly not compatible with the first (concern­ ing the indefiniteness of the aorist). It would also create problems - albeit not unsurmountable ones - for our sketchy analysis, since

different types of time adverb-

182

HARTMUT HABERLAND

ials would imply different relationships between that part of their meaning which is relevant for the meaning of the whole sentence, and that which is relevant for the cognit­ ive content of the utterance. 4. There is some reason to assume that in the Stoics' elaboration of the doctrine of tenses, time reference was a concept of subordinate relevance only. A much more centr­ al concept for them seems to have been that of aspect: that is, aspect as distinguishing between completed vs. 4 non-completed action. There are reasons to assume that for the Stoics, the aorist simply meant: the aspectually unmarked tense, i.e. the one that neither was completed (like the perfect) nor 'ongoing' (like the present and the imperfect). But this is not the gist of my discussion here. Rather, it is the following: to the early Byzantine scholars commenting on Dionysius Thrax, the Stoic doctrine of tenses was hardly more than bits and pieces of a terminology which in the meantime had been molded into a system of 'tines', viz. past, present, and future: still, it contained an occasion­ al, left-over stumblingblock (viz. the concept of 'aorist') which of course had to be eliminated. Since the Scholiasts felt that tenses somehow had to have something to do with the notion of the they had to make up an account explaining where the aorist got its name from. I hope to have shown, however sketchily, that there is a plausible formal explication of the idea behind this account, and that it should not be dismissed as a complet­ ely helpless (and perspectiveless) enterprise of rational­ ization, as Pohlenz has suggested.

A NOTE ON THE

'AORIST'

183

FOOTNOTES

"[the aoristos does not indicate any particular time] like the hypokeimenoi, but with 'recently' it is equivalent to the parakeimenos [perfect] like in 'I have struck' 'I struck recently', where­ as with 'a long time ago' it is equivalent to the hypersyntelikos [pluperfect], like 'I had struck' 'I struck a long time ago'." 2 This leaves open the question of how to distinguish between the past and the future aorist. (More on this below.) 3 'Indefinite' is perhaps not too felicitous a term, since it can be associated with 'indefinite' like in the 'indefinite article'. Now what I'm saying about the aorist here comes very close to attributing an anaphorical function to it; and anaphora is rather associated with definiteness than with indefiniteness. Yet, in a way the definite article creates the same kind of 'indefiniteness' for an NP, as does the aorist for a sentence of the type discussed here. If I say, I met the woman, the reference of the woman is indefinite, precisely because it contains the definite article: unless the referent of the NP is identified in the context, it is not clear who the woman is I am talking about. On the other hand, I met a woman is not indefinite in the same sense, precisely because it contains the indefinite article and therefore is contextually more independent. It is important to note that the same problem does not apply to the first person pronoun 'I' which in spite of its contextual dependency, does not make the sentence more 'indefinite' in the sense discussed here. This, again, can be explained by pointing out that 'I' con­ tributes to the meaning of the sentence, whereas the refer­ ence of 'the woman' is only relevant for the cognitive content of the sentence. 4 This may have something to do with the fact that at least some of the most prominent representatives of the Stoic school had a Semitic language as their mother tongue.

HARTMUT HABERLAND

184

REFERENCES Chantraine, Pierre. 1927. Histoire du parfait grec. (Collec­ tion linguistique publiee par la Societe linguistique de Paris, Vol. 2 1 ) . Paris: Champion.

Christensen, Johnny. 1962. An essay

on the

unity

of

Stoic

philosophy. Copenhagen: Munksgaard. Crystal, David. 1966, "Specification and English tenses".

Journal

of

Linguistics

2.

1-35.

Hilgard, Alfred, ed. 1901. "Scholia in Dionysii artem grammaticam". Grammatici Graeci I:3. Teubner. Kahane, Henry & Renee Kahane, 1954. "Review of Seiler, L'aspect et le temps dans le verbe

Language

Thracis Leipzig: Hansjakob neo-grec".

30. 115-123.

Mihevc, E. 1959. "La disparaison du parfait dans le grec de la basse epoque." Slovenska Akademija Znanosti in Umenosti. Razred za filoloske in literarne vede. Razprave 5. 91-154. Pinborg, Jan. 1975. "Classical antiquity: Greece". Current Trends in Linguistics 13,1. 69-126. The Hague: Mouton. Pohlenz , Max. 1939. "Die Begriindung der abendlandischen Sprachlehre durch die Stoa". Nachrichten von der Gesellschaft der Wissenschaften zu Gottingen. Philologisch-historische Klasse, Fachgruppe 1 Altertumswissenschaft. Neue Folge 3,6. 151-198. Sgall, Petr. 1977. "Sign meaning, cognitive content, and

pragmatics". Journal of Pragmatics 1. 269-282. Wunderlich, Dieter. 1970. Tempus und Zeitreferenz im schen.

Miinchen : Hueber .

Deut-

HERBERT E. BREKLE Regensburg

BEDINGUNGEN FUR DIE AKTUALGENESE DEUTSCHER NOMINALKOMPOSITA

In meinem Vortrag anlöβlich des 12. Internationalen Linguistenkongresses 1977 in Wien hatte ich einige Überlegungen, die Bedingungen des Zustandekommens, des Gebrauchs und des Verstehens von Nominalkomposita betreffend, mitgeteilt. Aus diesem Vortrag werde ich nur eini­ ge leitende bzw. programmatische Thesen übernehmen, urn dann das Programm eines Forschungsprojekts kurz vorzustellen, in dem die Thematik, die hier im Titel meiner Ausfiihrungen angesprochen ist, systematisch bearbeitet werden soll. Abschlie3end werden einige vorlaufige Ansätze und Losungen von Teilproblemen vorgestellt. Ausgehen möchte ich von der seit langem anerkannten Unterscheidung zwischen morphologisch komplexen Einheiten des Lexikons (einer Gesellschaft oder eines Individuums) einerseits, die von der Wortbildungsforschung in jiingerer Zeit iiberwiegend als Gegenstand ihrer Bemühungen angesehen wurden (philologische Richtung, textlich belegbare Daten; vgl. z. B. Marchands groβes Handbuch zur englischen Wortbildung oder Fleischers Deutsche V7ortbildung); diese Richtung untersucht iiberwiegend die "Wortgebildetheit" einer Sprache und andererseits "Wort­ bildung" verstanden als Prozeβ - ahnlich der Satzbildung

186

HERBERT E. BREKLE

- der zur Herstellung von immer wieder neuen Komposita und Ableitungen führt. Diese Wortbildungsprozesse werden von Sprechern/Schreibern ad hoc bei der Ausführung von Sprechakten zur Konstitution von Wortmaterial fur ihre Äuβerungen in Gang gesetzt. Zu unterscheiden ist hierbei zunachst zwischen Neologismen, die sich sozusagen auf der Eintrittsschwelle in das Wortschatzgebaude einer Sprache befinden und den echten ad hoc-Bildungen, die einmal gepragt, gebraucht und verstanden entweder wieder vergessen werden (wie Satze) oder unter bestimmten Bedingungen als Neologismen Aufnahme für kürzere oder langere Zeit in den Wortschatz eine Gruppe oder einer Gesellschaft finden konnen. Überlegungen dieser Art finden wir z.B. in einem objektiv-idealistischen Gewande schon in Schmitthenners Ursprachlehre § 79 (1826, Neudruck 1976) sehr schon ausgeführt. "Wenn aber die Sprache, theils einer unendlichen Ausbildung fähig ist, theils diese noch nicht ge­ funden hat; so muß in derselben außer den Wörtern und Formen, die das eben lebende Geschlecht in sei­ nem Besitze hat, noch eine unbegränzte Menge von Wörtern und Formen vorhanden sein, in denen dieje­ nigen, die noch nicht sind, ihre Vorstellungen derlegen werden. Nach dieser Entwicklung

nie­

ist uns

eine für die Kritik der Sprache sehr wichtige Enthellung gegeben in: a] möglich n e

[potentialiter)

v o r h a n d

Wörter, oder solche, deren Bedingmasse

e-

durch

die Elemente und Ableitungsgesetze der Sprache ge­ geben sind, und in b)

wirklich

[acta]

v o r h a n d e n e ,

oder

DIE AKTUALGENESE DEUTSCHER NOMINALKOMPOSITA

187

solche, die zu den cursierenden Schatzen der Sprache gehören." Wir schlagen nun vor, den Bedingungskomplex fur das Zustandekommen von ad ,hoc-Komposita verstarkt zum Gegenstand der Wortbildungsforschung zu machen. In einem gewissen Sinne konnen sogar die Bedingungen und Prozesse, die zum Entstehen von Neubildungen und speziell von adhoc-Komposita fiihren, als der Kernbereich der Wortbildungskomponente einer Sprache angesehen werden. Wir schlagen also eine Unterscheidung vor zwischen solchen lexematischen Konstituenten eines Satzes, die bei der Realisierung eines Satzes von der Liste der im Lexikon gespeicherten Worter - seien sie morphologisch einfach oder komplex - abgerufen werden und dem lexematischen Material, das beim Hervorbringen einer Auperung durch die Anwendung von Wortbildungsregeln auf Einheiten des Lexikons erst zustande kommt. Die Aufgabe, den zuerst genannten Bereich zu untersuchen, würden dann der Lexikographie zufalien, die jedoch dabei zweckma(3igerweise von den Methoden und Ergebnissen der Wortbildungsforschung im engeren Sinne Gebrauch machen sollte. Im Vergleich mit der traditionellen Wortbildungsforschung mu(3 sich die hier skizzierte Forschungsrichtung nich nur mit den abstrakten, systemlinguistisch erfaβbaren Gesetzma|3igkeiten befassen, z.B. die Morphonologie und Semantik einzelner Wortbildungstypen, sondern im Sin­ ne einer realistischen - evtl. naherungsweise sogar psychologisch realen - Grammatik auch die pragmatischen, kon- und kotextuellen Faktoren beriicksichtigen, die bei der tatsachlichen Produktion von ad hoc-Bildungen wirksam sind. Dabei sollte grundsatzlich die Sprecher-und die

188

HERBERT E. BREKLE

Hörerseite - also die Produktions- und die Interpretationsdimension - sprachlicher Äuβerungen in einem theoretischen Ansatz erfaβt werden; es wird also eine Theorie in diesem Bereich angestrebt, die die Sprachkompetenz von Sprechern und Horern möglichst umfassend, d.h. der kommunikativen Realitat angenahert, erfassen soll. Diese Forderung wird z.B. auch von Caroll/Tanenhaus (1975:47) aufgestellt: "[...] our interest in performan­ ce theory suggests that the theory of word-formation may be integrated within a general theory of language use not solely a theory of language structure". Die beiden Autoren nehmen an, daβ Wortbildungsprozesse nicht im Rahmen der üblichen Grammatik- oder Syntaxregeln beschrieben werden können; sie schlagen statt dessen vor, Wortbildungen durch besondere "Wortbildungsregelschemata" zur erfassen, die die morphonologische Form sowie eine schematische Beschreibung der semantischen Struktur der Wortbildungen angeben sollen. Als "input" für solche Regelschemata nehmen die Autoren satzahnliche Strukturen an? die Schemata selbst sind formal und funktionell ahnlich den Topikalisierungsregeln, wie sie in Brekle (1970:128ff) beschrieben wurden. Über die Bedingungen des Entstehens und die kommunikative Leistung neologistischer und ad hoc-Komposita weiβ man heute noch nicht sehr viel. Einige Autoren haben implizit oder explizit auf eine Art "Miminax-Prinzip" hingewiesen, das als die Grundlage einer allgemeinen Sprecherstrategie bei der Verwendung von Wortbildun­ gen angesehen werden kann. Eine frühe Formulierung finden wir bei von der Gabelentz (1901:466f.):

DIE AKTUALGENESE DEUTSCHER NOMINALKOMPOSITA

189

"An die Stelle des Satzbs: 'Er ging daran zugrunde, daB er seine Krafte zersplitterte', mag es dann heiBen: 'Er ging an Kräftezersplitterung zugrunde',... Wie muB der Geist seinen Stoff verdichten, ehe er ihn so zusammenzwangen kann! Und wie muβ der Hörer seinen Geist anspannen, damit er den so verdichteten Stoff richtig auflose ". Vendler (1967) und Brekle (1970) bieten ebenfalls Formulierungen dieses Prinzips. Caroll/Tanenhaus (1975: 51) liefern eine sehr konzise Fassung, die das Prinzip auf die Hone von Griceschen Konversationspostulaten hebt: "The speaker always tries to optimally minimize the surface complexity of his utterances while maximi­ zing the amount of information (underlying structure) he effectively communicates to the listener." So wie es dasteht, ist das Prinzip sicherlich zu stark formuliert; anzugeben wären verschiedene einschränkende Faktoren und insbesondere Kriterien dafiir, unter welchen Randbedingungen jeweils ein Optimum oder eine stilistisch noch vertretbare Grenze fur die gehaufte Verwendung von Komposita, Derivationen etc. erreicht wird. Das Wirken eines weiteren Prinzips - nennen wir es "Lückenbüβer-Prinzip" - wurde in der Literatur ebenfalls schon vor geraumer Zeit beobachtet. Insbesondere in der Kindersprache - bei Kindern im Alter zwischen drei und sechs Jahren - finden wir einen erstaunlichen Reichtum an ad hoc-Bildungen. Panagl (1976) zitiert etwa Bleistiftbvett (="Lineal"); aus eigener Beobachtung kann ich die Klassifikation von Bergen durch meinen Sohn (4;8) mitteilen:

Schneeberge,

Felsenbevge

und Wollbevge

( = "Berge, die mit niedrigen Biischen und Gras bedeckt

190

HERBERT E. BREKLE

sind, was ihnen ein "wollichtes" (vgl. J. G. Herder) Aussehen gibt"). Erscheinungen dieser Art lassen den Schluβ zu, daβ Kinder in dem genannten Alter über eine voll ausgebaute Wortbildungskompetenz verfiigen und diese - neben rein sprachspielerischen Bedürfnissen - haufig und erfolgreich zur Ausfüllung von Lexicon- und evtl. syntaktischen Strukturliicken einsetzen (dies ist natürlich vom Standpunkt des Erwachsenen aus gesehen!). Durch die mit zunehmendem Alter fortschreitenden sozialen Zwänge (Schule!) wird die Ausnutzung dieser sprachlichen Teilkompetenz stark eingeschränkt; häufig verkümmert sie sogar vollig. Wir finden jedoch auch in der Erwachsenensprache Anwendungen dieses Luckenbüβer-Prinzips. Wenn ein Sprecher nicht liber das "mot juste" verfiigt, kommt es vor, daβ er zu ad hoc-Wortbildungen Zuflucht nimmt. Bei bestimmten Aphasien wenden Patienten dieses Prinzip extensiv an. Eine weitere Bedingung fur die Verwendung von Neologismen und ad hoc-Bildungen kann in dem Hypostasierungseffekt - speziell von Nominalkomposita - gesehen werden. Konfigurationen von wahrgenommenen oder vorgestellten Qualitäten, die bisher nicht als reifiziert angesehen und deshalb auch nicht nominal ausgedruckt wurden, konnen durch eine Nominalisierung zu einem Quasi-Objekt hypostasiert werden. Damit können Wortbildungen auch für den Aufbau und die Terminologisierung wissenschaftlicher Theorien, aber auch für Ideologien, hoch rele­ vant werden. Bildlich ausgedrückt wird dabei je nach der Interessen- oder Bedürfnislage von Sprechern ein Stuck 'gefrorene Wirklichkeit" geschaffen, vorzugsweise durch Nominalbildungen ausgedruckt. Nur nebenbei sei darauf hingewiesen, daβ dieser Hypostasierungseffekt ge-

DIE AKTUALGENESE DEUTSCHER NOMINALKOMPOSITA

191

rade auch im politischen Bereich fiir Propagandazwecke ge- bzw. miβbraucht wird. Ich komme nun zu einer kurzen Darstellung eines von der DFG geforderten Forschungsprojekts, das zur Zeit an meniem Lehrstuhl durchgeführt wird: Kommunikative und pragmatisch-semantische Bedingun gen der Aktualgenese, der Verwerdung und des Verstehens von Nominalkomposita. Insgesamt setzt sich das Vorhaben zum Ziel, die komplexen Bedingungsfaktoren fiir Neologismen und ad hocBildungen im Bereich der Nominalkomposition des Deutschen empirisch zu ermitteln und theoretisch einzuordnen. Das Auftreten solcher Bildungen wird sowohl in der gesprochenen als auch in der geschriebenen Sprache untersucht. Im Falle der gesprochenen Sprache wird das thematisierte Phanomen direkt in der Entstehungssituation erfaβt, so daβ viele Informationen noch zuganglich sind, die bei einer nachtraglichen Analyse eines Korpus solcher Bildun­ gen nicht mehr ohne weiteres erfaβbar sind. Wir erhoffen uns auβerdem neue Einblicke in die Struktur des Lexikons (unter Einbeziehung psycholinguistischer Fragestellungen) sowie in die Semantik und Pragmatik solcher Bildun­ gen. Die Untersuchung erfolgt experimentell-empirisch und zwar anhand eines Sprechersamples: erwachsene Sprecher mit voller Sprachbeherrschung. Ihre Praduktionen lassen Rückschlüsse auf die in einer realistischen Grammatik des Sprachstandards zu berucksichtigenben Lexikonstrunkturen und Wortbildungsprozesse zu. In der geschriebenen Sprache sind die Bedingungsfaktoren fiir Neubildungen weniger gut zugänglich; insge-

192

HERBERT E. BREKLE

samt hat die Untersuchung von Neubildungen in der Schriftsprache im Projekt jedoch ungefahr den gleichen Stellenwert wie jene in der gesprochenen Sprache. Analysiert werden Zeitungs- und Zeitschriftentexte, die Analyse läuft in vier Hauptphasen ab: 1. Auswahl von ad hoe-Bildungen im Bereich der Nominalkomposition und deren syntaktische, semantische und textuelle Klassifikation. 2. Aufbau einer Datenbank, welche eine Sortierung des umfangreichen Datenmaterials nach einem zu entwickelnden Klassif ikationschema erlaubt, so daβ ein gezielter Zugriff zu den Daten in der Analyse- und Theoriebildungsphase möglich ist. 3. Systematische (evtl. auch quantitative) Überprüfung der Arbeitshypothesen zur Aktualgenese, zum Gebrauch und zum Verstehen von Nominalkomposita. 4. Gewinnung von Testmaterialien aus dem analysierten Korpus und damit Anschluβ an die experimentellen Untersuchungen. Neben diesen empirischen Zielen werden auch einige theoretische Ziele verfolgt. Diese sind zunachst allerdings der empirischen Forschung untergeordnet. 1. u.a. sollen die vorhandenen wortbildungstheoretischen Ansatze in bezug auf die hier verfolgte Zielsetzung auf ihre observationelle, deskriptive und explanative Adaquatheit hin untersucht werden. Insbesondere sollen jene Modelle, die eine psycholinguistische Relevanz beanspruchen einer kritischen Beurteilung aus der Sicht unseres Forschungsvorhabens unterzogen werden. 2. Es soll zumindest ansatzweise ein Modell formuliert werden, in dem die gefundenen Faktoren- und Situa-

DIE AKTUALGENESE DEUTSCHER NOMINALKOMPOSITA

193

tionskomplexe, welche spontane Wortbildungen beeinflussen, integriert werden konnen. 3. In einer weiteren Perspektive wird die unter 2. gennante theoretische Zielsetzung dahin führen, daβ die Dynamik der Sprecher- Hörerinteraktion in Abhängigkeit von Situationen in einem Kommunikationsmodell zu erfassen sein wird. Ohne jetzt weiter in eine detaillierte Beschreibung des Projektansatzes zu gehen, sollen abschlieβend zwei Experimente und einige vorlaufige Ergebnisse von theoretischen Detailuntersuchungen mitgeteilt und zur Diskussion gestellt werden. Während der Vorbereitungsphase für das Projekt wur­ den im Rahmen von Lehrveranstaltungen u.a. zwei Experi­ mente entworfen und durchgeführt; das erste mit dem Ziel, durch Darbietung relativ unüblicher Bildinhalte (Traumbilder, phantastischer Realismus) die Vpn (Kinder und Erwachsene) zur Produktion von Nominalkomposita anzuregen. Dabei wurde auch untersucht, inwieweit die Typik der Bilder zu bestimmten semantischen Strukturen bei den ad hoc gebildeten Nominalkomposita fiihrt. Dazu wurde eine grobe Bildanalyse in der Weise vorgenommen, daβ markante Details, Konfigurationen und Felder in den Bildern festgestellt wurden. Anhand der erzeugten Komposita wurde dann untersucht, welche Bildelemente, und Konfigurationen wie und wie oft im Determinans oder im Determinatum des Nominalkompositums abgebildet wurden. Insgesamt wurden zu sechs ausgewählten Bildern (drei von Margritte, drei von Schlote) von drei Sprechergruppen 237 Nominalkomposita gebildet. So wurden zu einem Bild von Schlote, das einen Elefanten zeigt, der auf

194

HERBERT E. BREKLE

dem Rüssel steht von der Kindergruppe (Alter sieben bis zehn Jahre) die folgenden Komposita gebildet: 1.

Elefantenbaum

2.

Baumelefant

3.

Weiβfuβelefant

4.

Rüsselelefant

5.

Hornelefant

6.

Rüsselstand

7.

Schwebeelefant

8.

Erdbohrer

9. Wassersucher 10. Erdschnüffler Die deverbalen Bestandteile: - s t a n d , S c h w e b e - , -bohrer-, -sucker, -schnüffler haben teilweise den Elefanten als implizites bzw. explizites Agens, teilweise konnen sie auch auf den Riissel bezogen werden, der dem Elefanten als Instrument seiner Tätigkeit dient (bohven, suchen, schnüffeln). Die verbalen Bestandteile haben kein direktes Korrelat in den Bildern, sie interpretieren aber eine im Bild durchaus vorhandene lokale Beziehungsstruktur zwischen dem Boden und dem Elefanten. Zusatzlich werden mögliche Funktionen, z.B. des Rüssels, mit verbalisiert. Dadurch, daβ die Bilder einen objektiven Stimulus darstellen, der fiir alle Vpn identische externe Ausgangsbedingungen schuf, läβt sich anhand der Komposita feststellen, welche kognitiven Tätigkeiten mit welchen Ergebnissen bei der Bildung von Komposita aktiviert wurden. Im weiteren Verlaufe des Projekts werden ähnliche, allerdings theoretisch und metodisch besser reflektierte experimentelle Arbeiten noch durchzufuhren sein.

DIE AKTUALGENESE DEUTSCHER NOMINALKOMPOSITA

195

In einem weiteren Experiment würden zunächst 100 deutsche Substantive mit den Merkmalen[konkret] und [zahlbar] mittels des Computers in der Weise kombiniert, daβ unter Ausschluβ der Kombination gleicher lexikalischer Elemente und unter Einfügung der üblichen Fugenelemente 9 900 Nominalkomposita ausgedruckt wurden. Erwartungsgemäβ fanden sich natiirlich in dieser Liste eine groβe Anzahl von uniiblichen deutschen Nominalkomposita. Eine Stichprobe aus dieser Menge wurde Versuchspersonen (Studenten) zur Beurteilung vorgelegt. Die Fragestellung war, ob die Versuchspersonen unabhängig von kontextuellen Informationen diesen Bildungen eine semantische Deutung, d.h. eine ihnen sinnvoll erscheinende Paraphrase zuordnen konnten. Bei denjenigen Bildungen, bei denen die Versuchspersonen auch nach einigem Nachdenken keine Interpretation zu geben imstande waren, wurden sie gebeten, sich auch u.U. phantastische Geschichten auszudenken in denen diese Bildungen dann doch verwendet werden konnten. Als Ergebnis zeigte sich, daβ sämtlichen unübli­ chen Bildungen schlieβ1ich eine Interpretation zugewiesen werden konnte. Dabei waren die Vpn auch oft bereit, wegen der beschrankten Anzahl von Verstehenstypen, nach denen sie die Komposita zu interpretieren suchten, gewisse Annahmen uber die Welt und Modifikationen in ihrem Weltwissen voriibergehend aufzugeben bzw. vorzunehmen. Unter den Verstehens- bzw- Interpretationstypen zeigten sich die folgenden prominent: funktionale Bestimmung, Form, lokale Beziehung und Beschaffenheit. Bei einigen Bildungen zeigte es sich, daβ von den Versuchspersonen das "Luckenbuβer-Prinzip" angewendet wurde {Baummaschi­ ne - "Sage"). Hier deutet sich auch die Moglichkeit an,

196

HERBERT E. BREKLE

über solche "Ersatzbildungen" einen empirischen Zugang zur Semantik von Simplizia zu gewinnen. Innerhalb der Verstehenstypen konnten noch spezielle "Verstehensmodi" isoliert werden. So tritt z.B. für Bildungen der Art 'X-Maschine' der Verstehensmodus 'Maschine zur Herstellung/Bearbeitung von X 'gehäuft auf. Das Auftreten die­ ses Verstehensmodus hangt dabei ab von sogenannten Verstehensbedingungen, die von der Bedeutung von X zu erfüllen sind, etwa, da(3 es sich urn einen Artefakt handelt, der eine gewisse Strukturkomplexitat nicht uberschreiten soll. Ferner scheint es gewisse Beziehungen zu geben zwischen den Interpretationsmodi und Bedingungen, die auftreten, wenn A Determinants ist bzw. wenn A Determinatum ist. Einzelheiten müssen hier noch naher untersucht werden. Weiterhin wird versucht, die Verstehensmo­ di mit konzeptuellen Dimensionen wie Form, Funktion und Beschaffenheit bei der Bedeutung der Simplizia zu korrelieren. Es wird angenommen, daβ jedes Simplizium eine inharente Rangordnung der mit ihm assoziierten Dimensio­ nen hat. Aus derartigen Rangordnungen der Simplizia ist dann eine Rangordnung der Kompositainterpretationen erschlieβbar. Auch dieses Experiment ist nur als ein vorlaufiger Ansatz zu verstehen, mit dem Zugange zu den eigentlichen Fragestellungen des Projekts gefunden werden konnen. Das hier kurz vorgestellte Projekt setzt sich ja u.a. zum Ziel, die Produktion, Verwendung und Interpre­ tation von ad hoc-Komposita möglichst nahe der Realitat von tatsachlichen sprachlichen Kommunikationen zu beschreiben. Einige Autoren (z.B. Downing (1977) und Bauer (1978)) haben mit einer ahnlichen Zielsetzung die Wichtigkeit des weiteren Kontextes (incl. Situation, Intention,

DIE AKTUALGENESE DEUTSCHER NOMINALKOMPOSITA

197

vom Sprecher fur den Hörer vorausgesetztes Wissen u.a.) und des Kotextes, in dem die Wortneubildung vorkommt, betont. Die Ergebnisse der bisherigen sprachsystemorientierten Wortbildungsforschung tragen zur Losung der eben genannten Fragestellungen nur wenig bei. Einer meiner Mitarbeiter, Gisbert Fanselow (jetzt publiziert als Fanselow 1981) , hat einige Thesen zu ei­ ner kontexttheoretischen Semantik und linguistischen Pragmatik im Hinblick auf eine sprachrealitätsnahe Beschreibung deutscher Nominalkomposita aufgestellt, die im folgenden auszugsweise mitgeteilt werden sollen. Unter Kontexttheorie wird hier eine um die Interpretation kontextabhängige Ausdriicke erweiterte Sematiktheorie (= logische Pragmatik im Sinne von Montague) verstanden; Aufgabe einer linguistischen Pragmatik wäre es, d5ie Mechanismen der Wertezuweisung fur die von der Kontexttheo­ rie erarbeiteten Parameter herauszufinden, vor allem aber deren psychologische und empirische Realitat nachzuprüfen. 1. In der Satzsyntax und Satzsemantik fiihrt heute an einer kontexttheoretisch erweiterten formalen Seman­ tik kein Weg mehr vorbei (Montague-Grammatik). Fur den Bereich der Nominalkomposition fehlt bisher eine solche Erganzung. Die kontexttheoretische Komponente einer Wortbildungssemantik wird eine Reihe von Parametern liefern, von deren kontextuell abhangiger Belegung die In­ terpretation, das Verstehen von Wortneubildungen, abhangt. 2. In praktisch alien bisherigen Beschreibungsansatzen für den Kompositionsbereich ist das Postulat enthalten, daβ die Menge der x, auf die AB zutrifft, enthalten sein soll in der Menge der x, auf die B zutrifft.

198

HERBERT E. BREKLE

In dieser Allgemeinheit ist das Postulat nicht haltbar. Man kann verschiedene Arten der Abweichung von der eben erwähnten Regel feststellen: a) die sogenannten exozentrischen oder bahuvrihi-Kompo­ sita, deren Sonderstatus aber auch schon von der traditionellen Wortbildungsforschung erkannt worden war. b) Es gibt Komposita und Ableitungen, bei denen die Determinans-Konstituente mit einem sogenannten NegaScheingefecht,

tionsfunktor besetzt ist krat;

Seheinarchitektur,

Pseudopolitiker,

Scheinblüte,

Pseudopatriot,

Psuedoö'kologe . . . ; Antinazi, fahver,

Antitechnokrat,

Unperson,

Unkultur,

,

Antiauto-

. . ;

Bildun-

Exb üvgemeister, Expräsident

...; Bildungen mit Nicht-: Nichtlinguist...;

Pseudorelativsatz Antifaschist,

Antialkoholiker.

gen mit dem Präfix Ex-: lev,

SoheindemoScheinkunst. . . ;

Nichtexistenz,

Nichtwäh-

Bildungen mit dem Prafix Un-: Unwahrheit... Bei alien diesen

Bildungen bewirkt eine Art Negationsfunktor, daβ die Formel AB ist enthalten in B nicht mehr stimmt. c) Es gibt eine betrachtliche Anzahl an Komposita, bei denen die Semantik des Determinans-Gliedes die Semantik des Determinatum-Gliedes beeinflussten: Bronzegott,

Steinlöwe,

Stoffhund,

Kunstkopf,

Spielzeugauto

usw. Hier scheint der Fall vorzuliegen, daβ die semantischen Merkmale von A Veranderungen im semantischen Merkmalskomplex von B hervorrufen; man kann sich dies etwa so vorstellen, daβ ausgelöst von bestimmten semantischen Merkmalen in A, diejenigen semantischen Merkmale in B "neutralisiert" werden die kombiniert mit jenen von A einen Widerspruch hervorrufen wurden. Dieses Phänomen tritt nicht nur bei Komposita auf, sondern auch bei freien syntaktischen Gruppen und

DIE AKTUALGENESE DEUTSCHER NOMINALKOMPOSITA

199

auch bei Simplizia im Satzkontext (z.B. steinevnev Lowe, oder die Frage eines Verkaufers: "Sind die Lowen schon ins Regal gestellt?"). Das hei|3t also, wir benotigen generell einen Mechanismus, der für die kontextabhangige Denotatszuweisung sorgt. In der Lexikographie hat man sich bislang damit beholfen, daß man viele Lexikoneintrage stark polysem machte; diese Losung führt aber zu einer Aufblähung des Lexikons, au3erdem wiirden wichtige Generalisierungen dadurch verloren gehen. Ein Vorschlag zur Losung dieser Frage geht nun dahin, daß man zwischen Bedeutungen im engeren Sinne und Bedeutungen im weiteren Sinne unterscheidet. Bedeutungen im weiteren Sinne sollen dann Funktionen von Kontexten bzw. Kontextparametern in Be­ deutungen im engeren Sinne sein. Anders ausgedrückt, die Obersetzung von Lowe soll also kontextabhangig sein, u.U. mu3 die genaue Denotatszuweisung auch noch probabilistisch gewichtet werden. Macht man die Semantik von Komposita, aber auch diejenige von Substantiven prinzipiell kontextabhangig, dann konnen sowohl die 'normalen' Komposita als auch die 'schwierigen" Komposita gemeinsam beschrieben werden und das Postulat, daß alle AB eine Teilklasse von B seien, gilt dann im jeweiligen Kontext wieder. Etwas formaler ausgedrückt: fur alle Kontextparameter k und fur alle x, wenn AB auf x in k zutrifft, dann trifft auch B auf x bezuglich k zu. Damit ware wenigstens ansatzweise erklart, wieso wir in gewisser Weise sagen kon­ nen, da(3 Fahrradbabys auch Babys sind (nämlich 'Baby' verständen als unerfahrenes menschliches Wesen) und Lesepistolen ( "Gerate zum automatischen Lesen irgendwelcher Daten') auch Pistolen sind.

200

HERBERT E. BREKLE

Aufgabe einer linguistischen Pragmatik ware es dann, herauszufinden, von welchen Faktoren die Kontextparameter abhangen und wie die Parameterbelegungen bei gegebenem Kontext bestimmt werden. Fur Komposita hieße dies, die Bedingungen anzugeben, bei denen bei gegebenem Kontext die Wahl des ersten Gliedes das Denotat des Zweitgliedes beeinflußt und umgekehrt; außerdem sind die außersprachlichen Faktoren anzugeben, die die Denotatszuweisung fur A oder B beeinflussen. 3. Jede Wortbildungstheorie, speziell jede Theorie über Nominalkomposita, muß den Unterschied zwischen Kom­ posita, die wir ohne explizite Kontextvorgabe verstehen und jenen, die wir nur durch Kontextinformationen ver­ stehen, erklaren können. Natürlich sind nicht die Kompo­ sita selbst kontextabhängig, sondern ihre Interpreta­ tion, oder anders formuliert, die tokens, nicht die ty­ pes der Komposita konnen kontextunabhangig interpretiert werden. Eisentopf in der Lesart "Topf aus Eisen" wird kontextunabhangig verstanden, aber die Interpretation "Topf in dem Eisen gekocht wird" ist nur in einem speziellen Kontext wahrscheinlich. Um zu einer Semantik für die kontextunabhangige Interpretation von Komposita zu kommen, kann man nun versuchen, iiber der Bedeutung der Teilausdriicke Beschrankungen iiber das Auftreten der Relationen zwischen diesen zu formulieren. Angaben dieser Art müßten in ei­ nem entsprechend ausgebauten Lexikon untergebracht wer­ den. Fur Topf etwa haben wir "besteht aus" und "kochen/ aufbewahren in" und unser semantisches Wissen von Topf und Milch sagt uns, daß Töpfe in der Regel kaum aus Milch bestehen, weiterhin, da(3 es relativ selten ge-

DIE AKTUALGENESE DEUTSCHER NOMINALKOMPOSITA

201

schieht, daß wir Eisen in Töpfen kochen. Eine solche Wissensstruktur kann man in der"artificial intelligence" simulieren. Derartige Beschrankungen führen dann zu einer Rangordnung der Lesarten. Kontextunabhangig waren dann die Lesarten, die in der natürlichen Rangordnung an erster Stelle stehen. Kontextuelle Informationen sind jedoch in der Lage, diese Rangordnung umzustoßen. Noch etwas anders formuliert: kontextunabhangige Lesarten ergeben sich aufgrund unserer Sprachkompetenz und aufgrund unseres mehr oder weniger variablen Weltwissen. Es ist deutlich geworden, daß hier ein strikt horerbezogener Standpunkt eingenommen wird. H hort S AB äußern und muß nun Annahmen machen über die in AB nicht explizierte Relation; der Horer darf auch annehmen, daß der Sprecher das weiß und daher (unter der Voraussetzung, S ist kooperativ und rational) nur solche Relationen implizit mitgibt, die fur den Horer auch erschließbar sind. Ist nun im unmittelbaren Kontext nichts besonderes erwahnt, so sind fur H nur solche Relationen erschließbar, von denen es zum gemeinsamen Glauben von S und H gegort, daß sie normalerweise zwischen Dingen, auf die A bzw. B zutrifft, bestehen. Wir wissen z.B., da(3 zwischen 'Fabrik' und 'Nagel' eben normalerweise die Relation des Produzievtwevdens in besteht, weil Fabriken eben norma­ lerweise etwas produzieren; zwischen 'Vogel' und 'Wald' die des 'Lebens-in' oder des 'sich-Ernahrens-in' usw. Die Aufgabe einer Kontexttheorie besteht also darin, die Abhangigkeit von Kontextparametern und der semantischen Interpretation des Kompositums zu formulieren. Die Aufgabe der linguistischen Pragmatik besteht darin, Bedingungen darüber zu formulieren, wann von der normalen Interpretation abgewichen werden darf, Fakto-

202

HERBERT E. BREKLE

ren zu finden, die erkennen lassen, daß eine solche Si­ tuation vorliegt und Regeln anzugeben, wie wir aus dem Text oder dem außersprachlichen Kontext eine verniinftige Interpretation ableiten können. Abschließend sei noch darauf hingewiesen, daß es auch eine quasi-pronominale Verwendungsweise von Komposita gibt, bei denen Sprecher und/oder Horer sich über die genaue Art der Relation zwischen den Konstituenten eines Kompositums im Unklaren sein konnen. Komposita-Verwendungen dieser Art treten auf, wenn in einem Text die Verwendung von normalen Pronomina nicht ausreicht, urn Referenzidentitaten zwischen einzelnen Ausdriicken in einem Text herzustellen. Prinzipiell handelt es sich dabei aber urn den Fall einer kotextabhangigen Interpretation von Komposita. 4. Die Probleme, die sich in der Kompositionsforschung beziiglich der Kontextabhangigkeit stellen, sind keineswegs kompositionsspezifisch, ahnliche Kontextabhangigkeiten zeigen mehrere syntaktische Konstruktionen, insbesondere die den Komposita verwandten Nomen-Prapositionen-Nomen-Konstruktionen. Damit wird deutlich, daß die Wortbildungsforschung im Sinne der Untersuchung ei­ nes Teilbereichs realistischer Kommunikationen auch einiges zur Losung allgemeinerer Fragestellungen, die sich mit der Beschreibung von Satzen und Texten in kommunikativen Situationen befassen, beitragen kann. (Erste Ergebnisse des hier in Umrissen vorgestellten Forschungsprojekts liegen jetzt in Form einer Microfische-Publikation (Kommunikative und pragmatisch-semantische Bedingungen der Aktualgenese, der Verwendung und des Verstehens von Nominalkomposita (im Deutschen) vor, vgl. Brekle (ed.) 1983).

DIE AKTUALGENESE DEUTSCHER NOMINALKOMPOSITA

203

BIBLIOGRAPHIE. Bauer, Laurie. 1978. . The Grammar of Nominal Compounding. With Special Reference to Danish, English and French. Odense: University Press. Brekle, Herbert E. 1970 ( 1976). Generative Satzsemantik (und transformationelle Syntax) im System der englischen Nominalkomposition. München: Fink. Brekle, Herbert E., ed. 1983. Kommunikative und pragmatisch-semantische Bedingungen der Aktualgenese, der Verwendung und des Verstehens von Nominalkomposita (im Deutschen). Regensburger Mikrofiche Materialien (RMM). Caroll, John M. & Tanenhaus, Michael K. 1975. "Prolegomena to a Functional Theory of Word Formation". Papers from the Parasession on Functionalism, April 17, 1975. Ed. by Robin E. Grossman, L. James San & Timothy J. Vance. Chicago, 111.: Chicago Linguistic Society. Downing, Pamela. 1977. "On the Creation and Use of English Compound Nouns". Language 53.810-842. Fanselow, Gisbert. 1981. Zur Syntax und Semantik der Nominalkomposition. Linguistische Arbeiten Bd. 107. Tübingen: Niemeyer. Fleischer, Wolfgang. 1969. Wortbildung der deutschen Gegenwartssprache. Leipzig: Bibliogr. Inst. Gabelentz, Georg von der. 1901. Die Sprachwissenschaft. Ihre Aufgaben, Methoden und bisherigen Ergebnisse. Hrsg. von Albrecht Graf von der Schulenburg. 2., verm. und verb. Aufl. Leipzig: Tauchnitz. Marchand, Hans. 1960. The Categories and Types of Present-Day English Word-Formation. A Synchronic-Diachronic Approach. Wiesbaden: Harrasowitz. Panagl, Oswald. 1976. Wortbildung diachron-synchron. Osterreichisches Linguistisches Programm. Sektion für diachrone Sprach­ wissenschaft. Schmitthenner, Friedrich. 1826. Ursprachlehre. Entwurf zu einem System der Grammatik. Mit einer Einleitung von Herbert E. Brekle. Stuttgart: Frommann, 1976. Vendler, Zeno. 1967. Linguistics and Philosophy. Ithaca, N.Y.: Cornell University Press.

JOSEF VACHEK Prague

A CURIOUS CASE OF PHONEMIC SUBSTITUTION

A language taking over into its word-stock a larger number of foreign loanwords is often faced in this take­ over with the necessity of adapting such loans by sub­ stituting its own, 'domestic' phonic elements for those which are common in the system of the donor language but are unknown in that of the recipient language. In view of the immense number of foreign loans that have found their way into English in the course of its historical development one cannot be surprised to find in it

a

fairly large number of such substitutions. They can be observed both on the phonemic and on the supraphonemic levels. As for the latter, plentiful instances of them are supplied by well known accentual shifts in loans from Anglo-Norman (and later on, from Central French), as in na

'nation', and many others of the

kind. More interesting, however, are cases of substitut­ ion on the phonemic level some of which we want to comment in our following brief remarks. Cases of phonemic substitution are again very frequent in English loans taken from Anglo-Norman (and Central French) sources. Instances of the replacement of AN /ü/, /ü/ by Middle English /u/, /iu/ are well

206

JOSEF VACHEK

known to be dealt with here in detail. Still, at least one interesting feature of the substitution of AN / / by ME /iu/ deserves to be singled out: the ME diphthong /iu/ remarkably managed to preserve the two characterist­ ic distinctive features of the AN / /, i.e. both its frontness and its labialization - the substitution only changed the original simultaneous implementation of the two features into a consecutive one. It should be added that this instance of substitution in the vocalic sub­ system of ME phonemes has a no less interesting parallel, also well known, in the consonantal subsystem. Quite a number of French loans taken into English contained two palatalized sonant phonemes /1' / or /n/, which, however, were unknown to the phonemic system of the recipient language. Here again the ME speakers had to take recourse to a process of substitution resulting, respectively, in the phonemic sequences of /jl/ or /jn/. (As concrete instances of the words in which this substitution has been effected may be mentioned, e.g., those which in Modern English figure as fail, veign - in the Modern English forms of the words, of course, the original / j / had coalesced with the vowel preceding it into a diph­ thong.) One notices, here again, that in Middle English were to be implemented consecutively the palatal and the sonantal phonemic elements which in the donor lang­ uage had been implemented simultaneously. The examples of phonemic substitutions adduced here so far were illustrated with the help of present-day Standard English lexical items still reflecting them. However, if one transcends the limits of the present-day standard language one may come across other cases of

PHONEMIC SUBSTITUTION

207

substitution, somewhat different from those discussed above but no less interesting. In the Northern English dialects one finds substitutions of /l'/ and / / ana­ logous to those found in the southern area (and in the standard language) but the sequence of the two elements of the phonemic sequence is here reversed, i.e. / l j / , /nj/, respectively (see K. Luick 1914: § 414). In some cases this reversed sequence was to penetrate also into the southern standard - see, e.g., instances like ModE onion / / o r Spaniard /sp nj d/ as opposed to the basis Spain ,/spein/ < Fr. Espagne (on this point, see E. J. Dobson 1968; § 441). However, in some exceptional cases the French palatal phoneme / / appears to have been subjected to another, different process of substitution. An interest­ ing instance of such exceptional process can be found in the Scottish dialect of English, spoken to the north of the Anglo-Scottish border and some nine years ago very ably analysed by the young Swiss anglicist B. Glauser 1974. Although in his comparison of northern and southern Scottish dialects he is mainly interested in the lexical differences of the two areas rather than in the phonological, his list of words is able to reveal some basic data concerning the phonological aspects of the examined materials. Here we will confine our attent­ ion to just one of such aspects, connected exactly with the above-mentioned word onion (Glauser's remarks on that item may be found on pp. 153-154 of his quoted monograph). In the Northern English materials which the Swiss scholar had elicited from his local informants the form

208

JOSEF VACHEK

of the word is virtually uniform, being of the type written onion and phonically differing from the southern standard, in most cases, in the initial [y] going back to ME u- (which in the southern standard was to be shift­ ed still further to [^] - it should be pointed out that one of the strongest sides of Glauser's book is his very careful phonetic transcription of each of the local forms, so that his lists of word forms can be evaluated as a reliable and welcome basis of theoretical conclus­ ions ). On the other hand, the speakers of the southern Scottish dialects adduced as their own forms of the same lexical item a number of varieties of the form i n g a n , where of course the digraph phonetically re­ presents the velar nasal sound [n]; among the phonetic shapes of the form, as registered by Glauser, the one most frequently met with is [ ]. The history of this word-shape can be traced from as early as the 16th cent­ ury thanks to the materials collected in the well-known Dictionary of the Older Scottish Tongue (now edited by A. J, Aitken), s.v. ingan. It is interesting to find that in the earlier centuries the spelling appears to furnish evidence also of the presence of [j] after [n] - cf., e.g., i n g 3 o ( u ) n , i n g e o ( u ) n , i n g y o n e , etc., while the present-day Scottish dialectal forms (at least those adduced by Galuser) do not indicate any trace of [j] at all. As to the presence of the phoneme (n) in the above adduced Scottish dialectical word-forms, it appears rather surprising, especially if one compares them with those used in the Northern English dialectal area. More light can be thrown on the striking differences of the

PHONEMIC SUBSTITUTION

209

types found in the two areas if one confronts both of them with their common ancestor (which in Central French has the form oignon, going back to Vulgar Latine o n i o n e , corresponding, in this turn, to Classical Latin uni on em). All these facts are carefully registered in Glauser's comment on the word, although he obviously does not draw from them all the consequences that appear to offer themselves. Let us, then, attempt a closer anal­ ysis of the two above-noted dialectal types. First of all, it should not be overlooked that the ultimate source of both forms, the Anglo-Norman word-form u i g n o n , contained not only one but two signals of foreign status, if analysed from the ME viewpoint. (On the problem of synchronistic foreignisms see especially V. Mathesius 1935.) One of them was, of course, the already mentioned palatal nasal /n/-phoneme; the other one, less conspicuous but no less important, is the diphthong /ui/, which also had been quite unknown to the native English word-stock at the beginning of Middle English. As is well known (see, e.g., K. Luick 1914: §§ 541 - 545), for a couple of the following centuries the first component of /ui/ was to develop in a way quite parallel to that of the short /u/; however, in Early ModE it was to undergo unexpected development - a change from / i/ to /ci/ (thus, e.g., ME j u i n t , puint > EModE jeint, peint > ModE [d'oint, point], The last-mentioned change of / i/ to / i/ has been attributed, as a rule, to the influence of spelling. True enough, AN /ui/ had been written, conformably to Norman scribal practice, as , since its first emergence in Early ModE words.

210

JOSEF VACHEK

However, the real cause of the change of /3i/ > /oi/, as discussed by J. Vachek 1965, appears to have been the introduction of /oi/ into the given word-type for the purpose of underlining its synchronistically foreign status. One thing deserves to be pointed out in this connection: since the ModE form of the word is /Anjen/, it is obvious that the initial diphthong /ui/ both in the Scottish and in the English dialectal contexts must have been simplified to /u/ and that this must have happened before the period of time at which the EModE continuation of ME /ui/, i.e. /ei/ was to be replaced by /oi/. Only this hypothesis can account for the initial /A/ found in this word in the southern standard. At the same time, this hypothesis goes very well with the sub­ stitution of /n/ by / n j / in Norhtern (and partly also Southern) England. Obviously, both the said changes contributed to the process of domestication to which the examined word was subjected. Such process of domest­ ication was probably felt the more desirable that the examined word denoted a piece of everyday extralingual reality. Let us now turn our attention again to those Scot­ tish dialectal word-forms which may be referred to as the ingan-type. As already mentioned, such forms are mainly characterized by the presence of the velar nasal phoneme /n/. This feature, by which all Scottish forms so strikingly differ from their English equivalents, is coupled by another one, also characteristically contrast­ ing with the contemporary English equivalents of the word, i.e., by the absence of / j / which appears to have existed in the earliest Scottish forms registered in the

PHONEMIC SUBSTITUTION

211

Dictionary of the Older Scottish Tongue and was already referred to here above. How, then, can these two feat­ ures of the Scottish dialectal form be accounted for? To deal with the disappearance of / j / first, it clearly appears to have been again a step towards a consistent domestication of this originally foreign but very commonly used word. Since the phonemic cluster /nj/ was wholly unknown to Scottish (as well as to English) words, it was to become simplified by the omission of its second component. It will be noted that a related cluster / n j / must have been just as familiar to Scottish English at it was to Standard English, at least since the Early Modern English period (see ModE instances like new , knew, etc.), and yet it was not to be chosen as a potential substitute for the unacceptable /nj/-group. Obviously, then, the most important issue to be tackled in the Scottish examples is that of the phoneme /n/, forming the first part of the original cluster /nj/. How, then, can one account for its continued presence in the Scottish varieties of the word? It appears, again, that this question, too, can be satisfactorily answered if one approaches it from the functionalist angle. The palatal /n/-phoneme, which was obviously evaluated as a signal of the foreign status of the word containing it, was again replaced here by a domestic phonic value, this time by the velar nasal phoneme /n/. One can certainly take for granted the presence of the /n/-phoneme in northern dialects of the Late Middle English period. As is commonly admitted, in Southern English dialects its emergence as a phoneme was conditioned by the ultimate loss of the unstressed /e/-vowel in suffixal and desinential syllables - this

212

JOSEF VACHEK

loss, as is well known, must have been compoleted by the end of Late Middle English (i.e. by about the year 1400). In the northern part of England (and the more so in Scotland) the loss of this unstressed /e/ must have taken place still earlier than in the south, so that the existence of /n/ in the northern dialects of England as well as in the southern Scottish dialect must be taken for granted for the Middle English period. The presence of /n/ in the system is, besides, also evidenced by the very spellings of the word in earlier Scottish texts where the graphemic groups like can only be interpreted as reflecting the phonic value [r|], because no explosive voiced [g] can have existed in the given word. Thus it appears obvious that the introduction of the /n/-phoneme in the Scottish dialects functioned as the substitution for the foreign /n/ just as /jn/ or /nj/ did in Northern England. Admittedly, even if we accept the indicated solut­ ion of the given problems, some of its aspects still remain which at the moment cannot be definitely elucid­ ated. One of them is certainly the exceptional character of the substitution of /n/ for /n/, inasmuch as it is not evidenced by other loans from French which have an analogous phonological structure (such as veign, sign, and the like), in which /n/ is regularly substituted by /jn/ or / n j / (cf. K. Luick 1914: § 414.1). Since the word-type ingan was hardly regarded, from the synchron­ istic viewpoint, as allied to any Latin word (of course it goes back, etymologically, to Latin unionem but from the synchronistic viewpoint neither the formal nor the semantic link with it can be observed). Thus one cannot

PHONEMIC SUBSTITUTION

213

account for its /n/ by using the explanation suggested for words of such a category by E. J. Dobson 1968, § 441, Note 2. He tries to explain the [n]-sound in words of this category (spelled with ) by one of the earlier types of the English pronunciation

of the Latin gn-group.

After all, he himself registers such pronunciations exclusively in words "of more learned type" (which is a status that definitely cannot be attributed to the ME word ingan)

. Very obviously, the given problem will still

call for further detailed research both by Scottish dialectologists and by historians of English. In concluding these remarks on a curious case of phonemic substitution we want to mention here, at least in passing, another rather curious instance of a similar (though reversely orientated) case of phonemic substitut­ ion. It was pointed out some three decades ago by the Czech phonetician M. Romportl 1954. He discussed a case of substitution which had taken place in Colloquial Czech where the /n/-sound found in loanwords

taken over

from German was to be replaced by the palatal nasal /n/ - see, e.g., CollCz

fedvunk,

verbunk

(from German För-

derung, Werbung). Romportl himself accounted for the /n/-quality in such substandard words by his theory that such German words had been taken over by Czech (and are still found in it, although not in the standard language) before the period in which the alveolar Czech [n] was assimilated to the following [k]. Though this may be so, one should not besides overlook the structural differ­ ence existing at the critical time between German as the donor language and Czech as the recipient language: the German consonant group [nk], then according to Romportl

214

JOSEF VACHEK

unusual in Czech, may have been evaluated as a signal of foreign status of the words containing it. The replace­ ment of [n] in such groups by [n], common to Czech at that time, may thus have been evaluated as a step lead­ ing somewhat closer to the domestication of the concerned category of words within the Czech lexical system. The modest aim of these few notes has been to point out an unsolved little problem of English phonological development the solution of which may be somewhat facil­ itated by the application of functionalist and structural­ ist methods of the Prague linguistic school. To those readers who will find the theme of these notes rather incongruous with the activity of the scholar to whom the present volume has been dedicated we feel sure to justify our choice of the theme by a recent statement of an Anglicist scholar pronounced on a representative linguist­ ic gathering. The said scholar (I. Poldauf 1983) evaluat­ ed very highly the activity of Peter Sgall's Department of Mathematical Linguistics in the Prague Faculty of Mathematics and Physics pointing out, very justly, that Sgall's team "has developed their own version of function­ al structuralism" in which "the Chomskyan rigor of approach is accepted and the generative procedure combin­ ed with profound interest in syntactic semantics associat­ ed with the ideas of the Prague School" (p. 185). Thus the gap existing between Sgall's and the present writer's conceptions may be much narrower than might seem at the first sight.

PHONEMIC SUBSTITUTION

215

REFERENCES Dobson, E. J, 1967-8. English Pronunciation 1500-1700. I, II, 2nd edition. Oxford: Clarendon Press. Glauser, B, 1974. The Scottish-English Linguistics Border. Lexical Aspects. Bern: Francke Vg. Luick, K. 1914-40. Historische Grammatik der englischen Sprache. I, II, Leipzig: Tauchnitz. Mathesius, V. 1935. "Zur synchronischen Analyse fremdem Sprachguts". Englische Studien 70. 21-35. Poldauf, I. 1983. "The rise and development of English Studies in the country of the Prague School, Towards a History of Engli sh Studies in Europe, ed. by T. Finckenstaedt & G, Scholtes. 177-187. Augsburg: universität. Romportl, M. 1 9 5 4 , "K artikulacni asimilaci nosovych souhlasek v cestine". Slovo a slovesnost 15. 10-19. Vachek, J. 1965, "The place of [ c1 ] in the phonic pattern of Southern British English". Linguistics 1 4 . 52-59.

WOJCIECH BUSZKOWSKI and PHILIP LUELSDORFF Regensburg

A FORMAL APPROACH TO ERROR TAXONOMY

1. INTRODUCTION Four major anthologies of error analyses have appear­ ed over the past twelve years (Fromkin 1973; 1980; Coltheart, Patterson & Marshall 1980; Cutler 1982) which are a veritable treasure-trove of data and theory in the domain of linguistic error. The time is right for a major theoretical synthesis (cf, Luelsdorff 1985a, where such a synthesis is essayed) based on sound empirical footing. Viewed within the context of knowledge representation and use erroneous patterns are all pervasive; consequent­ ly, linguistic error must be accounted for by an empiric­ ally adequate theory of linguistic structure and process. We attempt to refine the conceptual apparatus employ­ ed in linguistic error analyses and error analyses in general. Our main objective is to lay the foundation for the formal theory of linguistic deviation as a necessary complement to current theories of the linguistic norm. Thus conceived, such a theory should materially contribute to our understanding of human linguistic competence, our understanding of the interplay between linguistic compet­ ence and performance, language teaching strategies, etc. What follows is nothing

more than a first foray

218

WOJCIECH BUSZKOWSKI AND PHILIP LUELSDORFF

into the projected field of research. Some quite fund­ amental notions relevant to error analyses, especially those of a taxonomic character, will be defined in as precise a manner as we are able, and several of their principal properties will be discussed. We abstain from any detailed linguistic illustration (cf. Luelsdorff 1985a for thousands of examples) and thorough­ going empirical attestation (for which, cf. the items cited above) in the hope that the considerations artic­ ulated are sufficiently self-explanatory in order to evoke thought about the burning issues at stake. In section 2 some basic types of competence and performance error are introduced and briefly examined. Following the error-analytic tradition we study such error types on the level of segments and segment sequences, identifying various error-types with binary relations between them. Certain relevant interconnect­ ions between error types are defined and discussed within the framework of relational algebra. In section 3 we take constituent-structure into account (cf. Gladkij 1970) instead of segments and segment sequences. Beyond the shadow of a doubt const­ ituent structure is the major model of syntactic struct­ ure from the standpoint of any current theory of synt­ actic structure within formal grammar, including Trans­ formational Generative Grammar (TGG), Categorial Gram­ mar (CG), or Dependency Grammar (DG). In the tradition of CG constituent structure is usually related to a functional representation of semantics ((Fregean Semant­ ics) Buszkowski 1985). Though less developed, there have also been interesting attempts to employ constit-

ERROR TAXONOMY

219

uent structure in morphology (cf. Selkirk 198 2; 198 4; Toman 1984), phonology (cf. Booij 1984), and orthography (cf. Luelsdorff 198 5a). The theory of error becomes increasingly more subtile when one considers these levels of representation (cf. Luelsdorff 1985a, 1985b). Section 4 is given over to certain formal constraints on the structure of error. From the observation that it seems obvious that not every deviation from a norm (sub­ jective or objective) is performed by the language knower-user it follows that the theory of linguistic error should enable us to distinguish between extant and non-extant and possible and impossible linguistic errors and estimate their relative probabilities. We attempt to achieve this by employing the notion of rank (for the transposition-displacement error) involving a distance function on the constituent structure. Our notion of distance function in fact generalizes the one introduced by Luszczewska-Romahnowa (1961; cf. also Batog and Steffan-Batogowa (1980) for an application to phonetics) . Section 5 contains some summary remarks and suggest­ ions for further research. 2. ERROR TAXONOMY The letter 'V' will denote a non-empty set whose members are to be thought of as linguistic atoms (graph­ emes, phonemes, morphemes, words, depending upon the level of grammar under investigation). By a segment over V we mean a finite string of atoms over V. The empty segment is denoted by '^' S(V) denotes the set of all segments over V. We reserve the variables a, b, c for atoms, and A,

WOJCIECH BUSZKOWSKI AND PHILIP LUELSDORFF

220

B, C for segments (the scope of the latter will be adjust­ ed in the next section). AB denotes the concatenation of A and B in that order. We write A n for A... A (n-times). The position (0), and by means

five basic error types (mechanisms) of Trans­ (T), Displacement (D), Addition (A), Omission Substitution (S) can be formally characterized of the following schemata:

(T) AXBYC - AYBXC (X,B,Y ≠ A^ , X ≠ Y) , (D) AXYB - AYXB (X,Y ≠ ^, X ≠ Y) , (A) AB - AXB

(AB≠^

A, X≠ A ) ,

(0) AXB - AB (AB≠^ A, X≠ A), (S) AXB - AYB (AB ≠^ A, X,Y≠ ^ A, X ≠ Y) . where the left-hand segment string represents the correct form (target) and the right-hand segment string stands for the erroneous form (attempt). For the case of (T) and (D) we assume that the attempt differs from the target (this obviously holds for (A), (0) and (S)). Quite formally, an error of type T, D, & c , is to be defined as a pair (A1, A2 ) , where A, , A2 € S (V) , and the above-mentioned conditions are met. For example, (A,, A 2 ) is an error of type 0 if there are A, B, X 6 S(V) , such that AB ≠ ^, X ≠ ^ , A1 = AXB and A2 = AB. The letters T, D &c. will be used to denote the set of all errors of type T, D, &c. As errors are pairs of segment strings, these sets are in fact binary relat­ ions between segment strings. Accordingly, 'type of error' means some binary relation between segments (segment strings). We exemplify with a few errors found in English dictations administered to a 12-year-old pupil in a German Hauptschule (cf. Luelsdorff 19 8 5a). The pair

ERROR TAXONOMY



as well as of type D

. To type D we relegate the errors < a l a r r m arlam> and . The pairs ,

are examples of errors of type A, 3 hiest> of type 0, and , results from the combin­ ation of an omission and a displacement . The relations corresponding to these complex errors can be easily defined from relations T S by means of composition (relative product). Recall that if R, , R 2 are binary relations, then R1,. R2 (the composition of R1, and R 2 ) is a binary relation, satisfying the equi­ valence (3.) xR,.R2y iff there is an x such that xR 1 x and xR 2 y. We often write R1R2 instead of R1 .R2 . Now, double dis­ placement is the type DD; hence, (A1.., A 2 ) is an error of type DD iff there is a segment B such that (A1 , B) and (B, A 2 ) are errors of type D. In an analogous fashion we understand errors of type OD (e.g. ( e i g h t , eght)

r

and DS (e.g. twelve,

twelef),

&c.

As error types are relations, relational algebra seems to be the proper framework to express and examine their essential properties and interconnections. To this end we need the usual set-theoretic operations: U (union), n (intersection), the inclusion predicate c , and the empty set symbol 0. W

denotes the set of

all n-tuples of elements of the set W. By R

we de-

note the converse of the relation R. Thus, xR y iff yRx. R

denotes the transitive closure of R, i.e. the

224

WOJCIECH BUSZKOWSKI AND PHILIP LUELSDORFF

least transitive relation containing R. Thus, xR y iff there are x1,•..,xn (n > 1 ) , such that x and x.R

, for all

= x, x

= y

1 ≤ i ≤ n.

Fact 1. The following formulas are true: (4.) Tc = T, D C = D, S C = S, A C = 0, 0c = A, (5.) A C H A = ø, 0c n 0 = ø, (6.) T n A

= Tno

= DnA

= Dno

= ø.

In the light of (4.) T, D, and S are symmetric relations while A and 0 are mutually converse, and, by (5.), antisymmetric. It follows from (5.) and (6.) that relations T, A, 0 as well as D, A, 0 are pairwise dis­ joint (hence, for instance, no error of type A may be at the same time D, T, or 0 ) . We can derive similar proper­ ties for complex types. Using the law (R1 R2 = R2cRc1? we easily obtain: (7.) (DD)c = DD, (A0) C = AO, (0A) C = OA, (DA) C = OD and many other interesting dependencies. As T n D ≠ ø (recall (pupil, pilpu) , and also T n S ≠ ø, D n S ≠ ø, we are faced with the phenomenon of type ambiguity of error. To embrace this phenomenon in its full generality we must take complex error types into account. The simple calculation: (8.) AXBYC - AYXBC - AYBXC shows that any error of type T is also of type DD, hence T ≤ DD. We say that a type R is reducible (in a set U ≤ S(V), respectively) to type R + if R ≤ R + 2 + (R n U ≤ R ) . Type R is said to be weakly reducible

ERROR TAXONOMY

225

(in a set U ≤ S (V)) to R + if R ≤ (R+) (R n U 2 ≤ R + ) + ) . Of course reducibility entails weak reducibility but the converse need not hold. For, if R1 , R2 are defined as follows: (9.) a m R 1 a n iff n = m+1, a m R 2 a n iff m ≤ n, then R2 is weakly reducible but not reducible to R,. Fact 2: Each of the types T

S is reducible to AO.

T, D, S are reducible to OA. A and 0 are reducible to OA in the set S(V) - V. Fact 3: T is reducible to DD, hence weakly reduc­ ible to D. In the set n ≤ U,v D is weakly reducible to T. Proof: It is well-known from school algebra that any permutation results from the composition of displace­ ments of adjacent atoms. Accordingly, to verify the second statement, it suffices to profe that in a four-element segment (sequence) any displacement of the latter sort can be obtained by the iteration of trans­ positions. In fact, (10,) abed → cdba → bdea → bacd, dually: abcd → ... abdc (11.) abed → ... (by (10)) → abdc → aedb → ... (by (10)) → acbd Let us mention other cases of reducibility. In S(V) = V types A and 0 are reducible to S. T and D are reducible to SS. Some general rules may also be obtain­ ed, for instance:

WOJCIECH BUSZKOWSKI AND PHILIP LUELSDORFF

226 (12.)

i f R 1 ,, . . . ,Rn a r e r e d u c i b l e t o R 1 ' , . . . , R ' n , r e s p e c t i v e l y , then R1,..,Rn i s reducible to R1'

R'N'

(13.) if R1,, . . . ,Rn are weakly reducible to R'1,...,R'n, respectively, then R1,,, ,Rn is weakly reducible to R'1...R'n. The simple reducibility properties we have discussed above are intended to demonstrate the framework of relational algebra as applied to error theory. It seems evident that a mature theory should provide a much more subtle error taxonomy, more reducibility condit­ ions, &c. A few strides will be made in this direction in the following.

3. STRUCTURAL ERRORS. In this section we treat error types as binary relations on the set of constituent structures of expressions. We shall see that such a perspective often leads to a different classification of an error than in the preceding section. Moreover, one attains a less ambiguous error taxonomy. Finally, this approach enables us to interpret erroneous attempts as result­ ing from a deviation from the rules of grammar. For a set V of atoms, the set C(V) of constituent structures (c-structures) over V is defined by the following induction: (14.) V ≠ C(V) , (15.) if A 1 ,...,A n € C(V), n ≥ 2, then { A 1 - . A n }€ C(V).

227

ERROR TAXONOMY We assume, of course, that the symbols ' { ' and ' }' do not belong to V, A1,...,An are called "the immediate constituents of c-structure {A1...An }". The constituents of A G C(V) are A, its immediate constit­ uents, the immediate constituents of them, &c. By supp(A), A G C(V), we denote the support (frontier) of A, i.e. the segment over V that results after one has deleted all the symbols { and } from A. For example, the sentence The sun

shines

all

day

is the support of c-structure: (16.) {{The sun}{shines {all {day}}}}, whose immediate constituents are {The sun} and {shines {all {day}}}. The logical formula p A (q v r) is usually considered as being given in the form of c-structure. This obtains when we respect all the parentheses necessary for the full constituent analys­ is (except for the outermost ones) of any formula. In case one avails oneself of some conventions for the omission of parentheses, however, it seems more correct to regard the parentheses as constituents of the form­ ula under consideration. Accordingly, we may assign the latter formula the following c-structure: (17.) {p A {{q v r}}}. For A, B G C(V), a G V, by A(B/a) we denote the result of the substitution of B for every occurrence of a in A. We write oc1(a,A) if a has exactly one occurrence in A. We also write C(V, a1..,...,an) instead of C (VI) , where W = V u {a.,...,an }. (It is always assum­ ed that a.,,,.,an are pairwise distinct and that they do not belong to V ) .

WOJCIECH BUSZKOWSKI AND PHILIP LUELSDORFF

228

We now f o r m u l a t e p r e c i s e d e f i n i t i o n s o f types

Ť

Š

, t h e a n a l o g u e s of T

S,

but with

the

error

c-struct-

ures:

(T) A 1 Ť A 2 iff there are A G C(V,a,b), X,Y G C (V) , such that oc1.. (a,A), oc1.(b,A), a,b, are not adjacent in supp(A), and A1 = A(X/a)(Y/b), A 2 = A(Y/a)(X/b) (also X ≠ Y ) . (5) differs from (T) in stipulating: a, b are adjacent in supp(A). (A) A 1 ĀA 2 iff there are A G C(V,a), B1 ,...,Bn , X G C (V) , and 1 ≤ i ≤ n, such that oc, (a,A) and there hold: A1 = A((B1..,Bn )/a) (A1 = A(B1 / a ) , respectively), and either A2 = A((B1.....B1. XBi+1. ,...B n )/a) or A 2 = A((XB1...Bn)/a) (A2 = A((XB 1 /a) or A 2 = A ( ( B 1 / a , respectively)). (5) A 1 ōA 2 iff A 2 ĀA 1, (S) A1SA2 iff there are A € C(V.a), A non-atomic, and X,Y G C(V), X ≠ Y, such that oc1 (a,A), A1 = = A(X/a), and A 2 = (Y/a). If A,RA , we say that (A1 A 2 ) (also: A1.. → A2 ) is an error of type R (where R = T,...,S or R arises as a combination of them). For example, the pair: {{{The sun} {shines {all {day}}}} → {{all {day}} {shines {the {sun}}} is an error of type T. Let R denote the c-structure analogue of a type R, being defined for strings. We obtain the simple propos-

ERROR TAXONOMY ition: Fact 4. For all A, B € C (V) , if ARB then supp(A)R supp(B). Furthermore, if R1, is (weakly, respectively) reducible to R2 then R. is also (weakly, respectively) reducible to R2. The converse conditionals do not hold, however. Observe that: (18,)

{Joan {smiles charmingly}} → {Charmingly {Joan smiles}}

is an error of type 5 but not T though, of course, it is of type D if flattened to segments. Similarly, T is reducible to DD, but not to 55. For in, (19.) {Joan {smiles charmingly}} → {Charmingly {smiles Joan}}. to transform the attempt into the target one needs at least three displacements (notice that T remains weak­ ly reducible to 5 ) . It follows from these remarks that the constituent analysis of error structure yields, as a rule, finer stratification and less ambiguity. The more important advantage, however, consists in its direct relation to elementary transformations of grammatical rules. We have already mentioned that the constituent structure model is a fundamental tool for syntactic analysis within prominent trends in formal linguistics, such as TGG, CG, or DG. As a matter of fact, each of these theories proposes a different categorization stemming from TGG, viz. the phrase-structure grammar

229

230

WOJCIECH BUSZKOWSKI AND P H I L I P LUELSDORFF

framework ( c f . Chomsky 19 5 7 ) . Given a s e t L ≤ C(V) t h e

intersubstitutability

r e l a t i o n w i t h r e s p e c t t o L, d e n o t e d by I n t ( L ) ,

is a bin­

ary r e l a t i o n between c - s t r u c t u r e s o v e r V d e f i n e d

as

follows: ( 2 0 , ) AInt(L)B i f f , iff

f o r any C € C ( V , a ) , C(A/a) G L

C(B/a) € L.

Int(L) is a congruence on C(V), compatible with L, meaning: (21,) if A i Int(L)B i , 1 ≤ i ≤ n, then {A1...An }Int(L) {Bl..Bn}, (22.) if AInt(l)B, then A € L iff B G L, and, moreover, it is an equivalence relation. The equi­ valence classes if Int(L) are to be called (syntactic) categories of L. For A G C(V), by A/L we denote the only category of L containing A. Any formula of the form: (23.) {A 1 .,.A 2 }/ L → {A1/L...An/L} is referred to as a basic (grammatical) rule of L. The set of all basic rules of L will be symbolized by BRul(L). The set Rul(D) of all rules of L is defined by the following induction: (24.) BRul(L) c Rul(L) , (25.) if K → {K1,...Kn } and K1. → {Kl' ..Kn'} belong to Rul(L), then K → {Kl . ..K1-1{K'1. . -K' m }k 1 + 1 . Kn} also belongs to Rul(L) where Kj., K'j, K stand for (the names of) categories of L. For L c C(V) , by CON (L) we denote the set of all

231

ERROR TAXONOMY

constituents of c-structures from L. Observe that, for any A G CON(L), A/L = C(V)-CON/L which means the c-struct­ ures A G CON(L) constitute a single category of L to be termed the nonsence category. The remaining categories of L are said to be relevant to L. A rule of L is said to be relevant to L if its left-most category is relev­ ant to L (hence, all the categories appearing in this rule are relevant). The set of all the basic rules relev­ ant to L will be referred to as the grammar of L, and denoted by G (L) . L is said to be finitary if G(L) is finite. Fact 5. If L c C(V) is finitary, then supp(L) = {supp(A): A G L} is a context-free language; in fact G(L) is a context-free grammar generating supp(L) which yields exactly the c-structures from L(rules must be added for substituting atoms for categories). The converse is also true, namely, any context-free grammar can be reduced to a grammar G(L), for some finitary set L. For example, consider the set L consisting of the following c-structures: (26,) {Joan (27.) {Joan (28,) {Joan

smiles}, {smiles charmingly}} {smiles charmingly}

charmingly},

&c. One may easily verify that the categories of L are: (29.) N =

{Joan},

(30.) V = {smiles, {smiles charmingly}, ingly, charmingly}...}, (31.) A = {charmingly},

{{smiles

charm­

232

WOJCIECH BUSZKOWSKI AND PHILIP LUELSDORFF

(32.) S = L, and the nonsense category. The basic relevant rules are: (33.) S → {NV}, V → {VA}, as a non-basic rule of L we introduce: (34.) S → {N {VA}}. Let us consider the error: (35.) {Joan {smiles charmingly)} → {Joan {charmingly smiles}}, Obviously, this error can be analyzed as the result of displacement within the basic rules of L, viz. (36.) V → {AV} is used instead of V → {VA}. On the other hand, the error (18.) results by no iteration of displacements within the basic rules. In fact the only attempts available in this way from the left-hand side of (18 .) are: (37.) {Joan {charmingly smiles},} (38.) {{Smiles charmingly} Joan}} (39.) {{Charmingly smiles} Joan}, Clearly, we can treat (18.) as resulting from permut­ ation within (34.), which admits a reduction to double displacement. Nov;, if we assume that the performance of language proceeds by applying basic rather than non-basic relevant rules, then the latter analysis seems unsatisfact­ ory as an explanation of (18.). More precisely, the above assumption forces one to dismiss errors like (18.) from the domain of possible deviations from grammar or, to say the least, consider them to be much less probable than, e.g. (35.). In a natural way we define error types for the rules of L (since their right-hand sides are just some c-structures, one simply adapts the clauses (T) (S) . An error (A1,, A 2 ) is said to be of primary (secondary, respective­ ly) character if A2 can be derived by means of deviation

ERROR TAXONOMY

233

(of type (T) (S)) from the basic (non-basic, respective­ ly) rules of L (here L stands for a fixed set of targets). Thus, for instance, (35.) is of primary character, while (18.) is of secondary character. In the above definition we have not referred in an essential manner to A,. Observe, however, that according to our approach the grammar G (L) is unambiguous (in respect of c-structures), hence it yields exactly one derivation of A,. Therefore, we considconsider only those rules of L that participate in the only derivation of A1 with respect to G(L). In the light of the above remarks it should be evident that we regard errors of primary character as the most expected in real linguistic experience. In the next section the thesis will be refined by intro­ ducing a hierarchy of errors of secondary character. Prior to this, however, we mention the possibility of analyzing the relation between errors and grammatic­ al rules in a different, perhaps more stimulating, manner. The idea is to describe global effects of en­ riching the set of targets by an erroneous attempt. As a result, one obtains a larger set of targets with a modified grammar. Thus, the kind of error can be expressed in terms of the relation between the new grammar ("the grammar of that error") and the original one. Let us present one of several possible explicat­ ions of this idea. For L c C(V), A,B € C(V), the relation ADom(L)B (A dominates B in L) is defined as follows: (40.) ADom(L)B iff, for any C e C(A,a), if C(A/a) G L then C (B/b) £ L. Clearly, Dom(L) is a reflexive and transitive

234

WOJCIECH BUSZKOWSKI AND PHILIP LUELSDORFF

relation on C(V). Moreover, (41.) AInt(L)B iff both ADom(L)B and BDom(L)A. Notice also that Dom(L) satisfies: (42.) if A.Dom(L)B., 1 ≤ i ≤ n, then {A1 An }Dom(L) (B l ...B n }. Let L C C(V) be fixed, and let A,B C. C (V) . By Dom(L,A,B) we denote the least reflexive and transitive relation on C (V), such that: (43.) Dom(L) ≤ Dom(L,A,B), (44.) A Dom(L,A,B)B. and (42.) holds with Dom(L) replaced by (Dom(L,A,B). By L(B/A) we denote the set of all C e C(V), such that C'Dom(L,A,B)C, for some C € L. Fact 6. (i)

Dom(L(B/A) ) = Dom(L,A,B),

(ii) L € L(B/A), (iii) Int(L) € Int(L(B/A)), (iv) If L is finitary, then L(B/A) is also. We now explain how this construction relates to error analysis. Given an erroneous attempt (A,B), where A G L, B € L, L being a set of targets, one may assume that the grammar-user is inclined to perform B instead of A in any context. The set L(B/A) consists of all possible performances of that sort. In the light of Fact 6 (iv) L(B/A) is given by means of finitely many grammatical rules (provided L is). Furthermore, by Fact 6 (iii), the categories of L(B/A) can be identif­ ied with those of L (though different categories of L may fall into one category of L(B/A)). Accordingly,

ERROR TAXONOMY

G(L(B!A»

235

admits the same category symbols as C(L)

and, possibly, some new ones. One can analyze G(L(B!A»

as a transformation of G(L) and classify

the error (A,B) with respect to the form of this transformation. Let us turn back to the example (26)-(28). It is easy to verify that for A charmingly}}, B

=

=

{Joan {smiles

{Joan {charmingly smiles},

L(B!A) yields the following relevant rules: (45.) S

~

{NV ' }, V'

~

{AV} , V

~

{VA},

s

~

{NV},

where V'is the category of c-structures: (46.)

{charmingly smiles},

(47.) {charmingly {smiles charmingly}} &c. Thus, G(L(B!A»

arises from G(L) in effect

introducing a new relevant category, just V', and two new rules. As the first of the rules (45) can be regarded as the results of substitution within the last one, the error under consideration seems to represent an error of substitution on the level of grammar. Notice that our former approach has suggested to us assigning this error to the type displacement. Of course, the above analysis strongly depends on the assumption that B is to be substituted for A in all targets. Moreover, one regards the modified grammar as containing all the rules of the original one. Clearly, each of these assumptions seems questionable. Our aim has been only to outline a framework to deal with such problems; undoubtedly, the very framework admits of various

236

WOJCIECH BUSZKOWSKI AND PHILIP LUELSDORFF

modifications and refinements, A more thorough presentat­ ion of the topic dealt with in this section must be de­ ferred to a later study, 4. RANK HIERARCHY In the preceding section we distinguished between errors of primary and errors of secondary character, the former interpreted as resulting from deviations from basic grammatical rules, the latter from non-basic grammatical rules. Now, the distinction will be made more subtle in the corpus of errors of type T u D, which we refer to as transpositions (in a generalized sense). A sequence A 0 , A1,,...,A of constituents of A € C(V) is called a path of length n in A (leading from A0 to An ) , if A.i+1 is an immediate constituent of A , for all 0 ≤ 1 ≤ n. By the depth of a constituent A of B e C(V) (dB (A) ) we mean the length of the only path from B to A. If A, B are constituents of C E C (V), then supp(A,B) de­ notes the only constituent C' of C, such that there are paths from C' to A as well as to B and, for any C" having this property, d (C") ≤ dc (C') . We define dc1(A,B) ((1 (A,B), respectively) as the length of the path from supp(A,B) to A (to B, respectively). Finally, we put: (48.) dS(A,B) = dc1(A,B) +d c 2 (A,B), where A,B are constituents of C. The number d (A,B) is called the distance between A and B within C. C

Fact 7. For any C E C(V), d

is a distance function

on a set of constituents of C. This means, for all constituents A , A , A3 of C, the following conditions hold true:

ERROR TAXONOMY

237

(49.) d C (A 1 , A2 ) = 0 iff A1 = A 2 (they are equal as constituents!), (50.) d C (A 1 , A 2 ) = d C (A 2 , A 1 ) , (51.)

d C (A 1

A 3 ) ≤ d C (A l

A 2 ) + d C (A 2 , A 3 ) .

A c-structure is said to be regular if all its atomic constituents have the same depth. It is readily seen that if C is regular and a,b are atomic constituents of C, then: (52.) dc1(a, b) = dc2(a, b ) . According to Luszczewska-Romahnowa (1961), the distance between two atomic constituents of C is to be measured by dc1 rather than dc , provided that C is regular (strict­ ly speaking, the author was concerned with classificat­ ions, not c-structures, but this distinction is immaterial here). So, dc = 2dci. in the scope of atomic constit­ uents of a regular C; since the constant factor does not affect the mathematically essential properties of distance functions, we may regard our notion of distance as generalizing that given by Luszczewska-Romahnova (1961). Let us turn back to the clauses (T) (5) of the definition of (structural) error types. A pair (A,, A ) is a transposition (i.e. a T U 5 error) if there are A € C(V,a,b), X,Y C(V), X ≠ Y, such that oc 1 (a, A ) , oa 1 (a, A) , and A1 = A(X/a) (Y/b) , A 2 = A(Y/a) (X/b) . In what follows we additionally assume X, Y are minimal (i.e. the least complex) c-structures satisfying those equalities, and consequently, there is only one A ful­ filling them. By the rank of a transposition (A1,A2)

238

WOJCIECH BUSZKOWSKI AND PHILIP LUELSDORFF

we simply mean the distance between a and b or, equivalently, between X and Y within A1 (or A2 ) . Fact 8. The transpositions of rank 2 are exactly those of primary character. More exactly, (A1,, A 2 ) , where A1.. € L, A2 G L, is of rank 2 iff it can be derived by applying a single transposition of a basic rule of L. Consequently, transpositions of rank greater than 2 correspond to deviations from non-basic rules. The higher the rank, the more complex must be the non-basic rules under deviation. Hence, the appearance of such transpositions in linguistic data seems less probable. If p represented the frequency of the transpositions, one would expect the formula: (53.) p n = l/2 n _ 1 . Of course, any serious prediction of this sort should be based on empirical investigations. Errors of type (T u 5) are called iterated trans­ positions. Thus, (A,B) is an iterated transposition iff there is a sequence ( A , A . ) , 1 ≤ i ≤ n, such that A1 = A.An+1 = B, and (A , A.i+1 ) is a transposit­ ion, for all 1 ≤ i ≤ n. For any such sequence, the maximal rank of its terms is referred to as the rank of the sequence. Finally, by the rank of iterated trans­ position (A,B) we mean the minimal rank of sequence (Ai,Ai+1) fulfilling the above condition. Fact 9, The iterated transpositions of rank 2 are exactly those of primary character. More

ERROR TAXONOMY

239

precisely,

(A,, A2), where A1,€ E L, A2 . L is an iterated transposition of rank 2 iff it can be derived by applying basic trans­ positions of some basic rules of L. Clearly the remarks we have formulated above for the case of transposition probability pertain to the case of iterated transposition. Our final task consists of discussing the somewhat sophisticated matter of gener­ ating errors of an a priori bounded rank. Notice, first, that by iterated transposition of atoms one is able to obtain an arbitrary permutation of a given segment. On the level of c-structures, however, this usually involves transposition of arbitrarily high rank. We know that context-free languages are not closed under permutation (cf. Gladkij 1970). That is, there exists a set C ≤. S (V) which is generated by a context-free grammar, such that the set of all permutations of the segments from C is not context-free (take C = ((a, b) n C n : n ≤ 1). On the other hand, we obtain: Fact 10. Let L ≤ C(V) be finitary. For every n ≥ a, the set L , consisting of all c-structures derivable from those in L by iterated transposition of rank not greater than n, is also finitary, hence supp(Ln ) is a context-free language. Accordingly, restricting the rank of possible error one gains an effective procedure for generating the expected erroneous performances which still dwells within the realm of context-free grammar.

This strong­

ly motivates our interest in the rank hierarchy as

240

WOJCIECH BUSZKOWSKI AND PHILIP LUELSDORFF

yielding some natural constraints necessary for a practically manageable error description. 5. FINAL REMARKS. In section 2: Error taxonomy we define the five major linguistic error types: Transposition (T), Dis­ placement (D), Addition (A), Omission (0), and Substit­ ution (S). By "type of error" we mean a certain binary relation among segments or segment sequences. Simplex errors, the products of just one error operation, are distinguished from complex errors, the products of several error operations. We introduce the notion of the converse of a relation, thereby establishing T, D, and S as symmetric, A and 0 as mutually converse and anti­ symmetric, and T, A, 0 and D, A, 0 as pairwise disjoint. Finally, the notions reducibility and weak reducibility are introduced and the distinction between error mechan­ ism and error cause drawn and exemplified. Section 3: Structural errors treats error types as binary relations on the set of constituent structures of expressions. We present an error classification alternative to section 2, a less ambiguous error taxon­ omy, and the interpretation of erroneous attempts as resulting from a deviation from the rules of grammar. In a treatment more comprehensive than the one here possible, the crucial psyoholinguistio (as opposed to logical) distinction between competence and perform­ ance errors (cf. Luelsdorff 1985a) should be made and exemplified. In competence errors the left coordinate of a pair of speaker trees is not necessarily identified with the community norm.

ERROR TAXONOMY

241

Crucial to the argument is the notion that ortho­ graphic structure is hierarchical, the orthographic representation inheriting the c-structure of the phono­ logical representation (cf. Luelsdorff 1985b). The inherited c-structure is then used as a basis for pairs of trees whose structure reflects the constraints on error variables (cf. Luelsdorff 1985a, Chapters I-VIII). Precise definitions of the error types T

S, the

analogues of T S, are formulated and it is then shown that the constituent structure of errors yields, as a rule, finer stratification and less ambiguity. Further­ more, it is demonstrated that constituent analysis bears a direct relation to elementary transformations of grammatical rules. A distinction between errors of prim­ ary and errors of secondary character is drawn, the former resulting from a deviation from elementary grammatical rules, the latter from non-basic grammatic­ al rules. In Section 4: Rank hierarchy we introduce the notion of rank hierarchy, where by the rank of a trans­ position (A1 , A ? ) we mean the distance between a and b within A. Iterated transpositions of rank 2 are just those of a primary character. This distinction between error types and distances should be supplemented by the development of the modus errans

et c o r r i g e n s , i.e.

the theory of linguistic error and repair (cf. Luels­ dorff 1985a),

242

WOJCIECH BUSZKOWSKI AND PHILIP LUELSDORFF

REFERENCES Aho, A. V. & Ullman, J. D. 1972. The theory of parsing, translation and compiling. Vol. I. Parsing.New Jersey: Prentice-Hall, Batog, T. & Steffen-Batogowa. 1980. "A distance funct­ ion in phonetics", Lingua Posnaniensis 23. 47-58. B o o i j , Geert E. 1984. "Principles and parameters in prosodic phonology". Linguistics. 249-280. Coltheart, Max Karalyn Patterson & Marshall, John C., eds. 1980. Deep dyslexia. London: Routledge and Keagan Paul , Cutler, Anne, ed. 1982. Slips of the tongue and lang­ uage production. Linguistics 19; 7/8. Berlin: Mouton. Fromkin, Victoria, ed, 1973. Speech errors as linguistic evidence. The Hague: Mouton. Fromkin, Victoria, ed. 1980. Errors in linguistic per­ formance. Slips of the tongue, ear, pen, and hand. The Hague, Mouton. Kelley, J. L. 1955. General Typology. New York. on error Luelsdorff, Philip A. 1985a. Constraints variables in grammar: Bilingual misspelling orthotraphies. Amsterdam: John Benjamins. Luelsdorff, Philip A. 1985b. Orthographic representat­ ion. Unpublished m s , Universität Regensburg. Luszczewska-Romahnowa, S. 1961. "Classification as a kind of distance function". Studia logica 12. 41-81. Reprinted in: Twenty-five years of logical methodology in Poland, ed. by M. Przejecki, & R. Wojcicki. Warsawa-Dordrecht. 1977. Selkirk, Elizabeth 0. 1982. The syntax of words. Cam­ bridge, Mass.: MIT Press. Selkirk, Elizabeth O. 1983. Phonology and syntax: The relation between sound and structure. Cambridge, Mass.: MIT Press. Toman, Jindrich. 1983. Wortsyntax: Eine Diskussion ausgewZhlter Probleme deutscher Wortbildung. Linguistische Arbeiten 137. Tübingen: May Niemeyer Verlag.

IV. TOPIC AND FOCUS

NILS ERIK ENKVIST Abo

LINEARIZATION, TEXT TYPE, AND PARAMETER WEIGHTING

1 Let me begin my argument inductively, by citing some examples from an interesting type of text, namely the cook­ ery-book. The connoisseur of cookery-book styles will soon learn to place his texts on a scale between the extremely terse and the extremely chatty. The terse style is marked by short sentences which begin with verbs in the imperative. Object deletion is common: there is a default rule that, unless stated otherwise, sentences with deleted objects

hot;

turn;

keep covered;

roll

in powdered

(stir

chocolate)

while

refer

to the food and not to the cook. (In texts on soldiering, on the contrary, turn

and keep

covered

would be taken to

refer to what soldiers do to themselves.) A typical terse text would first list ingredients and implements and then give the instructions: (1) Preheat the oven, mark 5.375 F. Mince together the meat and onions. Mix in1/2ozof the flour, tomato puree, stock, salt and pepper. Turn into a pie dish. Rub the butter into the remaining flour until it resembles fine breadcrumbs, then stir in the grated cheese, herbs, salt and pepper. Spoon the crumble over the meat. Bake for 45 minutes to one hour. (Walker 1973:40.) Chatty texts typically add various comments to their instructions proper, which are usually given in the terse

NILS ERIK ENKVIST

246

style. A n example of such chatty additions is this: (2) Old Polish cuisine was very inventive and at the same time numerous garlands of fine dried mushrooms hung in pantries. Therefore, a way was found to prepare dried mushrooms that imitated the fresh variety. This dish was very popular on cold winter days. It is rather costly, but nevertheless worth trying out. Wash briefly 8 o z . not too large dried mushroom caps. ... Before serving check once more that the mushrooms are salted enough. Mushrooms prepared in this manner may be a Lent dish or a very attractive addition to fried and roasted meats. They may also be used as filling for omelets. (Lemnis and Vitry 1981:89-90.) Another characteristics of cookery-book language, not­ ably of the texts or passages in the terse style, is the fronting of valency-bound locatives. By valency-bound locat­ ives I understand locative adverbials when they occur together with a verb of whose semantic specification they are part, and which would be listed as obligatory for that verb in a valency grammar. Put

for instance implies locat­

ion: when you put something you must put it somewhere. And with

put,

locatives are obligatory: sentences such as

(3) Sussie put it. Put it! would have a very low acceptability in most contexts one can think of. To support my claim I shall cite a few examples of fronted locatives with put interestingly) with melt.

of the delightful Alice

and place

,

and one

(perhaps less

All come from a thirty-page sample

B, Toklas

Cook

Book:

(4) (a) In a huge copper pan put quantities of granulat­ ed sugar, moisten with cream, turn constantly with a copper spoon until it is done. Then pour into glasses. (Toklas 1961:164.) (b) Remove scales and fins of a 3 lb. shad. Wash thoroughly, but do not allow to remain in water. Drain and dry. Cut in slices. In a generously buttered fireproof dish, place a thin sprinkl-

LINEARIZATION, TEXT TYPE, AND PARAMETER WEIGHTING

247

ing of 4 finely chopped shallots... (Ibid. 176.) (c) If the mackerel are small take 4; if large, 4 slices. Clean, remove fins, wash, and dry. In a stew pan melt 6 tablespoons butter over low heat, add 4 medium-sized onions cut in fine slices. Cover and simmer ... (Ibid. 178.) (d) Ask the butcher to cut 2 lb. upper shoulder of beef in square pieces, two pieces for one serv­ ing. Marinate in 6 tablespoons of brandy for 6 hours, turning the meat frequently. In an iron pot over high flame put 4 tablespoons butter. While it is very hot place the meat in it in a single layer - no piece should cover another. Brown them on all sides, remove, and put aside. Put into the pot 1 cup diced side fat of pork, brown, and remove. Put into pot 12 small onions, brown, and remove. From the pot remove about two-thirds of the fat... (Ibid. 187.) (e) For each serving take 1 leaf of lettuce; on this place 3 tablespoons diced potato mixed with 1 teaspoon mayonnaise. On this place 3 poached oysters drained and placed on linen cloth to dry, then mixed with o i l , lemon juice, and pepper. On the oysters place a thin slice of truffle ... (Ibid. 193.) It seems to be an empirical fact that such fronting of valency-bound

locatives with verbs like put and place is

particularly common in cookery-books. A count of such frontings in the Brown and LOB corpora, for which 1 am indebted to my student Ann Westerlund, showed among other things that out of a total of 439 instances where the verb

-put

occurred with valency-bound locatives, only two such locat­ ives were fronted

(which makes 0.49 per c e n t ) . And both

these instances came from cookery books. It is of course possible to fabricate examples in which the needs for marked focus or phonological iconocity

(cf. Enkvist 1 9 8 0 a ) ,

syntactic

(to which I shall return b e l o w ) ,

and perhaps other forces might motivate locative fronting, particularly if the locative contains old information, as in

248

NILS ERIK ENKVIST (5) (a) On the shelf John put his hat; on the floor he dumped his suitcase.

or, chiastically, (5) (b) John dumped his suitcase on the floor; on the shelf he put his hat. As usual the acceptability of such fabrications improves with increased end weight, as in (5) (c) On the shelf John put the two books which he had inherited from his grandfather and carri­ ed with him for two months when he went on his disastrous safari in the footsteps of Hemingway. Such structures are likely to be rare in actual corpora, however. Once we disregard the artefacts that originate in the linguist's laboratory, we can take it as an empirical fact that, of all text types regarded as representative enough to be included in a standard corpus, the cookery-book, is one in which valency-bound locatives are actual­ ly fronted. Such an empirical, corpus-based observation raises various questions. I shall briefly discuss two of them. The first is, why should cookery-books be particularly hospitable to the fronting of valency-bound locatives? And the second is, what does the relatively high incidence of fronted valency-bound locatives in cookery-books imply for a total view (or, in more ambitious terms, theory) of word order (or, with a better term, linearization)? 2 In text typology we place texts into a multidimensional grid whose co-ordinates can reflect any text-internal or text-external (situational, contextual, functional) char­ acteristics that discriminate one text type from another. One dimension along which we might place pieces of discourse, and sometimes whole texts, is what I have called the scale

LINEARIZATION, TEXT TYPE, AND PARAMETER WEIGHTING

249

between the extremely operational and the extremely evocat­ ive. Let me explain what I mean. Texts such as (6) A: B: A: B:

"Now lift." "Can't. It's stuck." "Try moving it first. Two inches this way." "Right. Here she rises."

immediately suggest a certain type of situation, namely one in which A and B work together, for instance to repair or instal a machine. They obviously share a situation and therefore know what the deictics, shifters such as pronouns or pronominal adverbs, the deleted elements etc. refer to. References are situationally recoverable. Such a text I have called operational. On the contrary, the extremely evocative

text is the kind of narrative or statute text in

which the universe of discourse must be evoked, unambiguous­ ly described and defined and explicated, by the text itself and through verbal means. The text producer cannot take for granted that the receptor will capture references from a shared environment. In statutes and laws for instance it is vitally important that definitions are precise enough to preclude misunderstanding or even wilful misinterpretation. We all know what a car is, but there are statutes that begin by telling us what a motor vehicle is, at considerable length if need be. Cookery-books are one type of operational text, though one in which the situation and environment are shared by proxy as it were, rather than literally. Cookery-book writers assume that their readers are in a kitchen with standard equipment. They assume that the person cooking knows his pots and pans and has a store of common implem­ ents and spices at hand. These can be referred to without lengthy definition unless they are somehow unusual or exot­ ic. (Whether they should be regarded as old or new inform-

250

NILS ERIK ENKVIST

ation, as contextually bound or contextually free, is a question I shall not go into here.) Also, in cookery-books as in operational texts in general, brevity is a virtue. Operational texts differ from evocative ones in attempting a different balance between economy and verbal explicitness. They can profit from a shared situational environment. The corresponding balance struck by evocative texts must favour verbal explicitness without relying on a shared environment or on the receptor's good will. One way in which operational texts economise is through what I have called 'experiential iconicity' (cf. Enkvist 1981a). Experiential iconicity is definable as the principle which turns a text into a picture, an icon, of its universe of discourse by arranging textual elements into the same order which they have in the world of concepts or the world of things. Experiential iconicity thus strives for an isomorphy or homomorphy between a text and its universe of dis­ course. This is why gentlemen and ladies at the beginning of a speach would raise eyebrows. Traditional politeness suggests that ladies ought to come first. And (7)(a) Peter and Susie got married and had a baby does not mean the same as (7)(b) Peter and Susie had a baby and got married. If juxtaposed propositions might plausibly have a temporal or causal relation, we take their order of presentation to reflect the order of events. If not, the departure from experiential iconicity ought to be marked, as in (7)(c) Peter and Susie had a baby after they got married. Similarly, in an operational text such as a cookery-book, the fronting of a locative is an instance of experiential iconicity which contributes to the economy of the text. And this economy is so important, particularly in the terse style

LINEARIZATION, TEXT TYPE, AND PARAMETER WEIGHTING

251

that it is capable of stretching the valency bonds that usually tie the locative to a verb such as put.

Thus when

a cookery-book writer says (8)(a) Into a small frying-pan put two eggs. this is a terse-style equivalent of the more cumbersome (8)(b) First take a small frying-pan and then put into it two eggs. After these examples we are in a position to answer our first question. The reason why valency-bound locatives are fronted more often in cookery-books than in other text types included in the standard corpora is that cookery-books are representative of a text type we might call operational. In operational texts, the balance between economy and the necessary explicitness is struck so as to maximize the pro­ fit from context of situation. Such situational contexts can be shared literally, as when two people work on the same task, or by proxy, as when a cookery-book writer and his or her reader work in their respective kitchens but with simil­ ar equipment and ingredients. One way of economizing is to make use of experiential iconicity. And the fronting of locatives contributes to economizing through experiential iconicity, in examples such as (4) by implying the taking out of a utensil which is then used according to instruct­ ions given in the remaining part of the sentence. 3 Experiential iconicity is an important principle of linearization in texts. All the same it has received more attention from rhetoricians than from grammarians, perhaps because like Poe's purloined letter it has been too obvious and therefore trivial to be worth discussing. Here I have wanted to be explicit, even beyond the point of boredom, because the fronting of valency-bound locatives in operation-

252

NILS ERIK ENKVIST

al texts suggest wide and perhaps novel perspectives for the study of linearization (hinted at in Enkvist 1976:70-71) First a terminological note. I have preferred the term 'linearization' to 'word order', despite the traditionality of the latter, because 'word order' as we usually under­ stand it does not only involve words as separate units but also elements of various sizes, constituents, noun phrases and verb phrases, and clauses, both subordinate and main. The order of sentences and other text units in discourse is also linearization, and arrangements at such macro-levels are reflected in the linearization within the sentence because the information structure of the sentence is sub­ ordinated to the overall information strategies of the dis­ course. The linear arrangement of the sentence is one of the mechanisms that gives it the necessary textual fit (on textual fit, see further Enkvist 1977). In brief, lineariz­ ation in discourse should be seen as one totality governed by an overall text strategy. This overall strategy is reflected at all text levels from macrostructure to constit­ uent order within the sentence and the actual ordering of words within their various syntactic units. Hence 'linear­ ity' and not only 'word order'. Then to principles of linearization. Thanks not least to Petr Sgall and his collaborators, we know a great deal about the ways in which element order is affected by 'con­ textual boundness' or the distinction between given (or old) and new information. (I shall here refrain from using terms such as theme and rheme, topic and comment, or focus and presupposition because they have been used in confusingly many senses: cf. the discussion of their definition in En­ kvist 1981b). A general principle in many languages is that in the unmarked instance a sentence begins with information the speaker/writer assumes is old or given, or familiar to

LINEARIZATION, TEXT TYPE, AND PARAMETER WEIGHTING

253

the receptor, and somehow activated in his text-comprehend­ ing apparatus. To place references to old information in the beginning of the sentence we can use various devices. We can profit from converses in the lexis (A is B -s father / B is As daughter_, A bought the oar from B / B sold the oar to A) . We can choose among available syntactic structur­ es {A ate the apples / The apples were eaten by A) . And we can front, or topicalize, elements without further changes in the syntactic relations of the sentence {I have read this book / this book I have read). We also have a chance of departing from the unmarked pattern and place new informat­ ion at the beginning of the sentence. But if we do so we must signal what we have done by marking the new informat­ ion with cleft, with morphological markers such as the in­ definite article, or with focus-marking through the stressand-intonation complex (cf. Enkvist 1980a). Before linguists began seriously studying impromptu speech with all its peculiar characteristics that might even prompt it to deviate from 'normal' syntactic well-for­ medness, it was generally assumed that the information structure of a sentence could only work within the latitude allowed by canonical syntax. And this latitude differed in languages of different types. A language such as Latin or Finnish indicates syntactic functions such as subject funct­ ion or object function by morphological means. Linearity patterns can therefore be used to indicate, among other things, information structure, to a greater extent than in languages such as English, were linearization is more clos­ ely tied to the expression of syntactic function. The fail­ ure to recognize such typological differences has led many linguists to speak about the "free word order" of, say, Latin or Old English, without recognizing that the "free­ dom" was systematically used to expose information structure. It is of course not accidental that the pioneering studies

254 in

NILS ERIK ENKVIST information

as Henri Weil

structure and Vilèm

by

men such

Mathesius were

in fact

prompted by contrastive observations. Greek and Latin, said Weil, worked differently from French; English, said Mathe­ sius, works differently from Czech. More recently the differ­ ent relations between syntactic and information-dynamic rules and principles in the ordering of elements within the sentence have been suggested as a basis for a comprehensive language typology

(Li and Thompson 1976), But typological

differences appear even between closely related languages. Thus sentences such as (9)(a) Läraren ansåg eleverna vara obegåvad. (b) Förfalskarna arresterade polisen i går. (c) Sin flicka älskade sjömannen. have no exact linear equivalent in English. The word order must be changed in translation, or the Swedish

information

structure rendered by other means such as cleft . The latitude given to information structure by canonic­ al syntax thus differs from language type to language type and from language to language. But what the study of record­ ings of genuine improptu speech has shown is that there are next types in which canonical syntax itself may have to yield to information structure. In authentic improptu speech we find structures such as (10) North Wales - there is a strong antipathy between Catholics probably of [ai] of of Dublin extraction in Holyhead... North Wales hates there's a very very strong nasty ah an an a [n] natural antipathy ... a no popery feeling... (Svartvik and Quirk 1980:352-3; text simplified and diacritics removed by NEE.) One is tempted to interpret the dislocated and disjointed syntax by assuming that the speaker was at such great pains to mark his topic - North Wales - that his canonical syntax broke down. His processing load was so great that he could

LINEARIZATION, TEXT TYPE, AND PARAMETER WEIGHTING

255

not satisfy both canonical syntax and the desired informat­ ion structure at the same time. And what he jettisoned was canonical syntax rather than information structure: inform­ ation structure was at a priority and canonical syntax had to yield. If this interpretation is correct we may conclude that there are impromptu-speech situations in which inform­ ation structure is more important and weighted more heavily than is canonical syntax. If one of the two must be given up, the one to go is canonical syntax. Thus impromptu speech often comes to depart from the canons of syntactic well-formedness that are valid in careful writing. Another linearization principle is iconicity. In addit­ ion to the kind of experiential iconicity that we found in operational texts such as cookery-books, there are other kinds of iconicity. Syntactic iconicity occurs when syntact­ ic structures in a text have parallel structures and become icons of each other. Artistic prose, for instance of the kind culminating in euphuism, is a good source of examples: (11) But can Euphues convince me of fleeting, seeing for his sake I break my fidelity, Can he condemn me of disloyalty, when he is the only cause of my disliking? May he justly condemn me of treach­ ery, who hath this testimony as trial of my good will? Doth he not remember that the broken bone once set together is stronger than ever it was, That the greatest blot is taken off with the pumice? ... (From John Lyly, Euphues, 1578, quoted from Kermode et al. I: 1423). No less obvious, and perhaps even better known to all, is the iconicity that arises from metrical and phonological patterning in poetry. Thus Pope's couplet (12) Then case, bright nymph! to mourn thy ravished hair Which adds new glory to the shining sphere! (From The Rape of the Lock, V:141-2, quoted from Kermode et al. 1:1888.)

256

NILS ERIK ENKVIST

could not be changed into (13) Bright nymph! To mourn thy ravished hair now cease, Which adds new glory to the shining sphere. without destroying the pattern to which Pope was dedicated. In metrically regular poetry, phonological iconicity is thus at a premium. One interesting consideration is to what extent a poet succeeds in reconciling the requirem­ ents of metre with canonical syntax, information structure, and experiential iconicity. Pope's brilliance appears, among other things, from the way in which he obeys the rules of a strict metre without doing violence to canonic­ al syntax or information structure. Lesser poets may be less capable of satisfying the simultaneous demands of syntax and the other text-ordering principles. We need thus not deny Crabbe's merits even if we note how much more obtrusively he must manipulate his syntax to fit the metre. Just one example: (14) 1 View now the Winter-storm! above, one cloud, Black and unbroken, all the skies o'ershroud... ... the waves so swelled and steep, Breaking and sinking, and the sunken swells, 5 Nor one, one moment, in its station dwells: But nearer land you may the billows trace, As if contending in their watery chase; May watch the mightiest till the shoal they reach, 9 Then break and hurry to their utmost stretch... (Quoted from Peacock 1943:463.) Crabbe's embeddings and interpolations might perhaps be said to serve experiential iconicity in that they suggest the disorder of a violent storm. But still he topicalises neavev land in line 6 so that the line flavours of artifice, and inserts now into line 1, they into line 8, and then into line 9 to fill out his line scheme. Even in Crabbe, then, we see how the needs of metrical iconicity were

LINEARIZATION, TEXT TYPE, AND PARAMETER WEIGHTING

257

weighted at the expense of a natural and idiomatic express­ ion. And Crabbe is still a fine poet. Doggerel is often doggerel precisely because in it metre has taken over from good sense and fluent syntax.

I started out from a scrutiny of fronted valency-bound locatives in cookery-books. My argument has led to a view of linearization and linearization parameters which might be summarized as follows: The linear pattern of elements in discourse, such as it appears on the surface of the text, is a resultant of several forces. These forces or parameters include (i)

canonical syntax, in terms of basic constituent order within the sentence (SVO, VSO, SOV etc., with the concomitant typological patterns within constituents and with reservations for typological mixes), the marking of sentence function (statement versus quest­ ion) and clause function (the main clause and sub­ ordinate clause may have different linearization patterns as in German or Swedish);

(ii) information dynamics, governed by the information strategy of the text which in turn regulates the plac­ ing of given and new, contextually bound and contextually free information in text units and sentences, among other things by giving each sentence its textual fit (cf. Enkvist 1977) and thus subordinating the structure of the sentence to the strategy of the text; (iii) experiential iconicity which may make the text into an icon or picture of its universe of discourse, by making the arrangement of elements in the sentence and text isomorphic with the arrangement of their referents in the world (Enkvist 1980b, 1981a); and (iv) aesthetic, metrical and phonological iconicity which makes sentences into syntactic icons of each other, and cements metrically regular poetry by making feet, lines and stanzas into metrical and phonological icons of each other through metrical structure, rhyme, al­ literation, assonance, etc. In different types of text, these different parameters

258

NILS ERIK ENKVIST

acquire different weights. There are instances, for in­ stance in cookery books, where the requirements of operat­ ionally motivate stretching the usually tight valency bounds that obtain between verbs such as put and their obligatory locatives. Such stretching seems to be rare but can still be comfortably accommodated within the latit­ ude allowed by canonical syntax. Then there are instances of poetry where poets have had to adjust their syntax and pad their text with unnecessary elements in order to satisfy the metre. In such instances the canonical syntax must adjust to the metre and the needs of phonaesthetic iconicity. ( L i o e n t i a poetica is a classic term covering, among other things, the subordination of syntax to metre.) And then we have instances where information structure is emphasized at the expense of canonical syntax. This is presumably a frequent occurrence in impromptu speech and impromptu dialogue. Another type of text which I have not so far mentioned is the modern poem which wilfully distorts syntax for poetic effect. Advertisers too play for effects by twisting their canonical syntax. In such wilful depart­ ures from canonicity the motivation may be complex and therefore more difficult to define in terms of the simpler parameters that prevail in more spontaneous text types. One plausible basis for a text typology, then, is the difference in relative weighting that is given to the different parameters of linearization. Experiential icon­ icity is heavily weighted for instance in operational texts (as well as in certain types of literary text). Information structure is at a premium in impromptu speech, phonaesthetic iconicity in certain types of poetry and artistic prose. There are situations in which a speaker or writer can successfully reconcile the requirements of several parameter categories with each other and with

LINEARIZATION, TEXT TYPE, AND PARAMETER WEIGHTING

259

canonical syntax, producing a text that strikes the receptor as normal, smooth and attractive. There are also situations in which some parameter or parameters must be sacrificed because the speaker or writer cannot manage the processing load involved in satisfying them all at the same time. This is one reason why the patterns of canonic­ al syntax are often broken against in impromptu dialogue. There, information structure counts for more than does syntactic 'corectness'. There would be a lot to add about this view of linear­ ization in relation to more specific patterns of 'word order' in discourse, and in relation to grammaticality, acceptability, situational and stylistic appropriateness, and aesthetic evaluation and functionality. I must leave my reader with a sketch. Speaking metaphorically, my suggestions add up to a conception of discourse, not as a result of a single mechanical process of generation but rather as an outcome of conspiracies and conflicts between various forces, some antagonistic and some pulling in the same direction. To what extent a speaker or writer can re­ concile these forces and to what extent he must favour some of them at the expense of others is partly a matter of text type and situation, partly a matter of his own skill as a producer of discourse.

NOTE 1 Those innocent of Swedish may find comments helpful at this point. Sentence (9a) translates literally as Läraren ansåg eleverna vara obegavad. The teacher thought the pupils to be untalented. Obegåvad, however, is singular and must therefore refer to the teacher, which in turn identifies läraren as a fronted object. To refer to eleverna 'the pupils' the plural obega.va.de must be used. Sentence (9b) is syntactically ambiguous in Swedish too, but as Swedish topic-

260

NILS ERIK ENKVIST

alizes more readily than English, pragmatic plausibility suggests a preferred reading were policemen arrest forgers rather than vice versa. In (9c) the possessive sin is co-referent with the subject; otherwise the pronoun would have been hans.

REFERENCES Enkvist, Nils Erik. 1976. "Notes on Valency, Semantic Scope, and Thematic Perspective as Parameters of Adverbial Placement in English". Reports on Text Linguistics: Approaches to Word Order (= Meddelanden fran Stiftelsens för Åbo Akademi Forskningsinstitut 8) ed. by Nils E. Enkvist & Viljo Kohonen, 51-74. Åbo. Enkvist, Nils Erik. 1977. "Contextual Acceptability and Error Evaluat­ ion". Papers from the Conference on Contrastive Linguistics and Error Analysis, Stockholm and Abo, 7-8 February 1977 ed. by Rolf Palmberg & Hakan Ringbom.1-26. (=Meddelanden fran Stiftelsens för Abo Akademi Forskningsinstitut 19). Åbo. Enkvist, Nils Erik. 1980a. "Marked Focus: Functions ,and Constraints". Studies in English Linguistics for Randolph Quirk ed. by Sidney Greenbaum, Geoffrey Leech & Jan Svartvik, 134-152. London: Long­ man. Enkvist, Nils Erik. 1980b. "Motives for Topicalization". Linguistic Studies in Honour of Paul Christophersen ed. by Robin Thelwall, 1-15. (Occasional Papers in Linguistics and Language Learning No. 7). Coleraine: The New University of Ulster. Enkvist, Nils Erik. 1981a. "Experiential Iconicism". Text 1. 97-111. Enkvist, Nils Erik. 1981b. "Intervention à propos de la communication du president Seiler". Actes du Colloque i n t e r n a t i o n a l et multidisciplinaire sur la comprehension du langage, Creteil 25-27 septembre 1980, ed. by J. Barbizet, M. Pergnier & D. Seleskovitch, 59-61. Paris: Didier Erudition. Kermode, Frank et al., eds. 1973. The Oxford Anthology of English Li­ terature. Romantic Poetry and Prose ed. by Harold Bloom & Lionel Trilling. New York etc.: Oxford University Press. Lemnis, Maria & Henryk Vitry. 1981. Old Polish Traditions in the Kitchen and at the Table. Warsaw: Interpress Publishers. Li, Charles N. & Sandra A. Thompson. 1976. "Subject and Topic: A New Typology of Language". Subject and Topic ed. by Charles N. Li, 457-490. New York etc.: Academic Press. Peacock, W., ed. 1930. English Verse III, Dryden to Wordsworth. London etc.: Oxford University Press. Svartvik, Jan & Randolph Quirk. 1980. A Corpus of English Conversation (Lund Studies in English 56). Lund: C.W.K. Gleerup. Toklas, Alice B. 1961. The Alice B. Toklas Cook Book. Harmondsworth: Penguin Books Ltd. Walker, Lorna. 1973. The Pennywise Cookbook. London: Produced for the Dairy Industry.

FRANTIŠEK DANEŠ Prague

THE "QUESTION TEST" RE-EXAMINED

1. In the last fifteen years or so, the investigat­ ions of the phenomena referred to by such terms as topic - comment, functional sentence perspective (FSP) and many others have brought forth an amount of interesting new knowledge. In the light of these results, diverse as they may be, FSP appears as a complex or bundle of pheno­ mena, that, in principle, are different but mutually interwoven and to a high degree affined, which features put them together into a special domain; may be that the contents of the said complex has not yet been fully identified and differentiated. At the same time it appears that the domain of FSP shows many functional connections with other linguistic domains, these connect­ ions being mostly of a complementary type in that the functionning of FSP phenomena "counts upon" some pheno­ mena of other domains and vice versa. The discussion in following sections will be con­ centrated on the distinctions between the particular components of the FSP-complex rather than on their affinities (letting aside the various means signalling the FSP-structuring).

262

FRANTIŠEK DANEŠ

2. Very generally speaking, at least three differ­ ent aspects (or components) of FSP may be distinguished: (1) theme (topic) - rheme (comment) bipartition in the sense "what the speaker is talking about" and "what he is saying about this theme"; this aspect I will call here communicative articulation of utterance; (2) given/ known/old/contextually bound/context dependent informat­ ion/item - new/contextually non-bound/context independ­ ent information/item; the interpretation of this distinct­ ion vary considerably with different scholars, and it is just this "information-bearing structure" (Garvin) or

informational

bipartition

(in my wording) that I will

make the centre of my following exposition; (3) differ­ ent degrees of "communicative dynamism (hierarchy of salience)". - To be sure, the said three distinctions are not made by all investigators, very often they are mixed in this or that way, differently formulated and termed, so that the whole field of FSP is not easy to survey. 2.1. In principle, I am in accord with those who, like Mathesium, differentiate between the "theme" and the "known information". And I also agree with Halliday (1967: 17 and elsewhere) that "thematization is independ­ ent of what has gone before", that the theme (Th) rheme (Rh) distinction is context independent, its task being to turn a content (propositional, ideational) structure into a message, into a communication. The fact that any utterance has a linear character will be util­ ized for the psychological articulation (arrangement) of the process of thought: as a rule, the speaker begins with Th,

ie.

with the "item to be described" (Allerton

THE "QUESTION TEST" RE-EXAMINED

263

1978: 137). With Halliday and Allerton I agree also in that probably any utterance (clause) posseses its

Th -

Rh structure, without respect to the possible informat­ ional bipartition

of the same utterance.

Of course, I do not wish to deny that between Th Rh articulation and informational bipartition there exist conspicuous interconnections, correspondences, preferences; these mutual affinities have been happily treated by Halliday (1970) in terms of "good reason": in principle, there obtains a free choice; nevertheless, Th will be associated with "given" and Rh with "new" unless there will be a good reason for another arrange­ ment. (Similar affinities obtain also in relation of FSP to the grammatical and semantic structures of sent­ ence; systematically they were investigated by Halliday, Sgall, Firbas, Kuno

&

Yokoyama 19 80; cf. also Dokulil

& Danes 1958.) 2 Th

appears in relation to Rh as the basic compon­

ent; Th may be tested by means of relatively objective testing devices (see below), whereas Rh simply represents the rest of the utterance. In contradistinction to this state of affairs, in the sphere of informational bipart­ ition it is the new (context independent) piece of information that appears as basic (it can be directly tested especially by means of the question test), while the piece of given (known, context dependent) informat­ ion is that which remains (and may be identified in the testing question). Further, in those "normal" cases, where the two structurations appear parallel (correspond to each other), Th will mostly appear as component of the more extensive "given information" (in some cases they may fully coincide), whereas Rh (in comparison with Th

264

FRANTIŠEK DANEŠ

mostly the more extensive utterance portion) includes hew information" in it (the focus proper or theme proper of some authors).

;

From the tests proposed for the identification of Th the circumlocations, as for, as far as ... is concern­ ed seem to be convenient. There were suggested some other identification devices, as well, such as the formula "I say to you (now and here) about X that ..." ( i . e . the use of verba dicendi with two complements; cf. e.g. Sgall 1973), or Bogusławski's (1977: 183) question test "What else about the same thing?" (Besides, mostly, the possibility obtains to single out Th by means of a special intonation contour.) From the fact that the Th-Rh-articulation is con­ text free does not necessarily follow that it appears irrelevant to text constitution. On the contrary, the choice of Th will often be motivated by context: the communicative articulation operates on the isotopic (anaphoric) relations in the text and thus it appears as a very important factor of text coherence (see 3.1. below). 2.2.1. The informational partition of utterance into two portions (I call them here, following Mathesius, "starting point" (S) and "core of information" (C)) has from the beginning of the inquiries into FSP been connected with the notions of "known (old, given)" and "new" information, resp., i.e. with the fact that utter­ ances mostly appear in a verbal context and that their boundness to context will be reflected in their inform­ ational structure: those elements of it that have been named or mentioned or pointed to in a certain portion

THE "QUESTION TEST" RE-EXAMINED

265

of preceding context or that may be derived (recovered) from that portion, reduced from it, that are implied by it, briefly those elements that appear to be just "at hand" are regarded as "known", whereas the others are "new". The terms "context dependent/contextually bound" and "context independent/contextually non-bound" refer, in principle, to the same distinction (though the interpretation of them in works of different scholars vacilates, to be sure). It is evident that the notion "given (known)" is relative and very broad (if not vague). Some years ago (1974: 109 ff.) I expounded six different moments, which appear relevant in judging "knownness4" . I will here only briefly summarize the main points: (1) Besides the preceding verbal context (co-text), also speech situat­ ion and the "stock of shared knowledge" are relevant factors. There may

be differences between the speaker

and hearer in this respect, of course; what is in the end decisive is the hearer's evaluation (cf. Chafe's "egocentrismus"), nevertheless he might try to respect the hearer's position. (2) "Knownness" is graded proper­ ty, which different utterance elements reveal in differ­ ent measure (degrees), depending upon various factors (e,g,hierarchy of factors mentioned

in (1), the order

in which the particular elements have been mentioned in previous context (the principle of "distance"), differ­ ent types of semantic implicative ties between a postcedent and its antecedent; etc.). (3) "Knownness" (and its different degrees) depend on the length of the port­ ion of preceding context in relation to which the evaluation is being carried out. The upper limit of such

FRANTISEK DANES

266

portion is to be empirically ascertained (thus Chafe (1974) maintains that only between elements of utterances in the contact position the relation of "knownness" may obtain, while Svoboda (1981) claims that this relation obtains even between distant utterances, the upper limit 5 being five intermediate utterances). We may expect that the portions ("intervals") are in a way correlated with the hierarchy of text segmentation (paragraphs, sections, chapters, . . . ) , and this hierarchical stratification reflects also in the grading of "knownness". - (4) In the evaluation of "knownness" even some subjective factors are effective, namely the different sensitivity of people to these relations in text and the delicacy with which the speaker respects (and the hearer evaluat­ es) a given item as semantically implied in a preceding item in the text. Utterances of a coherent text undoubtedly show interconnectivity of the sort we have just discussed. Our question is, however, whether or to what extent the said interconnections do, in a regular way, reflect in the informational content of utterances, so that on the basis of them we could divide the content of the major­ ity of utterances into two parts, having different functions in the information-bearing structure, viz.S and C. For a long time I believed (and I was not alone) that the informational bipartition on the basis "known (S) - new (C)" is possible and I suggested (partly influenced by the writings of A. Hatcher

and P. Garvin)

to use for the identification of the two utterance portions heuristic wh-questions. Without doubt, this

THE "QUESTION TEST" RE-EXAMINED

267

relatively objective testing procedure enables us to divide the utterance content in such a way that one portion corresponds to the wh-element

of the question

(this is C) and the other corresponds to the rest of the question (this is S ) . Today, however, I am no more persuaded that by means of this procedure we really as­ certain the contextual boundness/non-boundness of the utterance content. Even in my older discussions of the question test (1968: 127 f. ; 1974: 114 f.) I made clear that the pair "question - answer" may not, in these cases, be regarded as dialogue proper,

but as a heuristic means only. I

assumed that it is possible to assign to any sentence (taken as a systemic grammatical unit) a set of wh-questions, representing all possible

t y p e s

of differ­

ent context in which the given sentence is applicable (in the utterance function) . and, consequently, reveal­ ing all possible kinds of FSP-structuring that it may acquire, when used on different speech occasions. (I also pointed to the fact that in the case of a yes-no question, is is the sentence as a whole that is question­ ed.) But at last I arrived to the conclusion that the interpretation of the question test needs to be princip­ ally re-examined. - Let us consider the following text fragment: (1) In

this

seven

hospital

Hemingway

stayed

for

full

weeks.

One of the possible testing questions could be (1?) "For how long did Hemingway stay in this hospital?", and according to it "for full seven weeks" represents

268

FRANTISEK DANES

C, whereas the rest of the utterance appears as S (within which "in this hospital" represents, on the level of communicative articulation, the thematic component). Let us now find out in the given actual text the veryutterance immediately preceding (1); it is the utterance (2) He could

not

and listened

sleep, to

so he only

read

a

little

radio.

Is it at all possible to put somehow together the test­ ing question (1?) with this actually preceding piece of context (2)? It is evident that with the exception of the hyper-theme ("Hemingway-he"), sentences (1?) and (2) share no common sememe(s), neither sememes that would be in a relevant and regular way related; and what looks worse, such sememes do not appear in the whole preceding portion (5 utterances) of the given paragraph. It is true that sememes contained in S-portion of (1) reveal, in the end, some connecting ties with the preceding broad context, though these ties are rather weak, too general if not vague, so that they are capable only to ensure the fitting of (1) into the text, nothing more. Moreover, such weak semantic connections with the preceding context can be found even in the C-portion of (1), which, however, should allegedly represent "contextually non-bound" elements. Another example: (3)(a) There is plenty of snow on the now. (b) Our children are looking forward

mountains to

skiing.

The testing question (3b?) "What are our children look-

THE "QUESTION TEST" RE-EXAMINED ing forward to?"

269

shows that C is "skiing"; but this fact

appears to be in variance with the above assumption, since "skiing" is the only piece of information that is directly derivable from the context (cf. snow -

skiing)

and as such it is expected to belong to S. - Similar results yields the analysis of following example (known from the writings of Sgall's group): (4)(a) On Christmas we weve expecting (b) Uncle John came (b') First

our

relatives.

first,

came uncle

John.

Both the component "uncle John" and the component "came first" easily

follow from (4a) (from the respective

components "relatives" and "to come", implying an order of appearance). Both components are thus clearly contextually bound, nevertheless, each of them may appear either as S, or as C, so that the informational bipartititon (as well as the speaker's choice between (b) and (b')) cannot be explained in terms of contextual boundness. 7a A similar situation obtains in following text fragment (adapted from Hajičová & Vrbová, 1981: 293) : (5)(a) The school (6) (a)

garden was full

They talked

(b) because they (7)

Outside

noisily were

parents

of

children.

, excited.

were

waiting.

Question (7?) "Who was waiting outside?" ascribes the characteristics C to the element

"parents", in spite of

the fact that in the semantic (lexical) system it is coupled with the sememe "children" in (5) (a kind of ko-hyponymy) and therefore obviously and easily derivable

FRANTIŠEK DANEŠ

270

from it. (Hajičová & Vrbová rightly comment that the object "parents" is not "completely new because it is given by the broader scene (of family relations children - parents)", and that this item "receives a very high degree of activation"). Moreover, the same text fragment (5) - (6b) might continue with (7') (7')

Parents

weve

waiting

instead of (7), outside

("What about

parents?"), where the "most activated item" appears as S, ie.

as

"contextually bound element", while C is represented by "were waiting outside". (But examining this latter component we discover that even it appears as "context­ ually bound" ("known", "activated"), since it is clearly derivable from the antecedent "the school garden" in (5). - To sum up: Since both "parents" and "school garden" in (7) and (7') appear in the context (5) - (6b) as contextually bound (known, activated), it is not possible to explain the difference between (7) and (7') in terms of contextual boundness (newness, activation) (abstracting from another question, namely that of the motivation of the speaker's choice between (7) and (7')). How to solve this problem? One solution has been offered by Sgall's group (whose systematic and well-founded inquiries into FSP deserve to be highly appre­ ciated) .

In (1980:'155) they say: "Our understanding

of contextually bound elements (...)

is broader than

usual in that 'contextual' is not restricted to 'co-textual' here; the speaker connects this feature with some of the elements the counterparts of which in the s t o c k

of

s h a r e d

k n o w l e d g e

have

THE "QUESTION TEST" RE-EXAMINED

271

been foregrounded. (...) The distinction between the stock of shared knowledge as a whole and its foreground (the set of the activated elements) might be useful...". With other words, the "contextual boundness" ("knownness") has been thus substituted by the psychological notion of "activated elements of the stock of shared knowledge" (instead of "activation", Hajičová & Vrbová (1981) used recently the term "salience", so that "activ­ ated", "foregrounded" and "salient" seem to be synonymus here). This solution is, however, connected with at least three rather problematic points. First, how to test objectively the "degrees of activation"? Hajičová & Vrbova (1982) duly claim that "this is a point where linguistic and psychological investigations should go hand in hand". Nevertheless, it is by far not clear how to manage such an empirical psychological research (and I feel sceptical to optimistic predictions about possible results of such a research and their linguistic relev­ ance) ; after all, even the said scholars have contented themselves so far with co-textual clues only. Second, our conclusion (cf. above) that "contextual boundness" has been substituted by "activation" is not in accord with some other formulations of the said scholars; cf. eg, "The expressions referring to activated elements may function as contextually bound (...) , while the non-activated ones are a l w a y s contextually non-bound (...)" [underlined by F. D.] (Hajičová & Vrbova (1981:. 292). With other words, if the activated elements may appear also as contextually n o n - b o u n d , then the notion of "contextual boundness" has been left

272

FRANTIŠEK DANEŠ

undefined. Third, even if the equivalence "activated" = "contextually bound" would be maintained then in terms of this notion, we could not explain the difference, e.g., between (3) and (3'), or (7) and (7'), so that search for another explanatory principle of this phenomenon would appear necessary. 2.2.2. I take for granted that the application of the question test reveals an undoubtedly important fact of informational structuring of utterances, obvious even o

to linguistically untrained language users. Neverthe­ less, it appears rather difficult to present an adequate linguistic interpretation of it. Following formulation of A. Marty (quoted also by Sgall et al. 1980: 34) seems to me revealing:"Nicht das wertvollere zu sein, sondern zu e r k a n n t zu w e r d e n ist das Wesen des psychologischen Pr&dikats - ein nicht weiter analysierbarer Vorgang unseres psychischen Lebens, der nicht anders als durch Hinweis auf die innere Anschaung klar zu machen ist." Marty rightly sees the main point of our problem in the feature 'to be r e c o g n i z e d (understood) by the hearer a s the informat­ ional core ("new" piece of information)'. In a similar direction (though, of course, without taking recourse to the problematic -inneve Anschaung of Brentano' s philosophy) goes an interesting formulation by Halliday (1967: 7f.); he claims that the new piece of information is "new" not in the sense that "it has not been previous­ ly mentioned, although it is often the case that it has not been, but in the sense that the speaker presents it as not being recoverable from the preceding discourse". Nevertheless, one has to ask how the speaker manages

THE "QUESTION TEST" RE-EXAMINED

273

the hearer to interpret something in fact recoverable from the context (known to the hearer) as not being recoverable. I think that this seemingly unresolvable problem follows from the circumstance that the notion of "new information" will often be understood in a too simple and inadequate sense; in fact, informational newness mostly reveals a

n e x a l ,

relational

character: the newness of a piece of information I. 3

rests in its application to another piece of informat­ ion I, . Only if correlated in this way, I. turns into the "information core" (and appears as "new" to the hearer, if the correlation of I. with I, has not yet been known to him). Thus in (4b') First came uncle John. , what is new to the hearer is apparently not "uncle John" as such, but the fact that it was just he who came first. The nexal character of "new information" has been point­ ed to many years ago by Jespersen (1925: 245); he said: "The 'new information1 is not always contained in the predicate, but it is always inherent in the connexion of the two elements [subject and predicate] - in the fact that these elements are put together, i.e. in the 'nexus' (...)". (Cf. also Benes 1968: 271; Danes 1974: 111.) My following attempt at the solution of the "known - new information" problem has a functional and informat­ ional character, i.e. it approaches this problem from the point of view of the sense or goal of communication (after all, the application of testing questions pre­ supposes the same approach, without taking into account the "activation of shared knowledge"). In utterances that do not express an appeal, the speaker intends to convey to the hearer something that will, assumingly,

274

FRANTISEK DANES

be new to him (to convey something known would, except in special cases, be in variance with one of the maxims of communication). We know that the speaker attempts to cause the hearer to become better informed, ie.

the

speaker's intention is to change or modify somewhat the hearer's knowledge of the world (as it is assumed by the speaker). For our further consideration there appears sub­ stantial the fact that the speaker starts from an assumed state of knowledge of the world (of its certain fragment) on the side of the hearer, and that he then enriches it with something new which he assumes to be, very probably, interesting for the hearer (this enrichment being identical with the content of speaker's instructing the hearer to modify his stock of world knowledge). Thus in (8)

Jiok

and Vicky

divorced

yesterday

"new" information rests in the speaker's instructing the hearer to substitute, in his stock of knowledge, the relation 'to be married' holding between Dick and Vicky by the relation 'they are no more married (they divorced)' The information-bearing structure of (8) might be explicitly

paraphrased as follows: "I assume that you

suppose that Dick and Vicky live together as husband and wife. Now I convey to you (since I think that it might (just) interest you) that this state of affairs changed yesterday, namely so that they divorced". It is clear that the informational value of (8) rests in the substitution of the proposition reflecting an older state of affairs (assumed by the speaker as known to the hearer) by another proposition, reflecting a later

THE "QUESTION TEST" RE-EXAMINED

275

(newer) state of affairs that is (according to the speaker's assumption) to the hearer so far unknown, "knew". - Similarly in (9)

Yesterday

's race

was won by

Robertson

(as an answer to the question "Who won yesterday's race?"), the speaker starts from the assumption that the hearer knows about the existence of yesterday's race but that it is unknown to him who gained the victory and that he will be interested

to learn who was the victor. (In

this case, the unknown item of a given event is to be specified, inserted.) Notice that I am not trying here to give the motivation why the speaker began at all to speak about, say, the race; I am working with a weak assumption (on the speaker's side) that the hearer might have interest (not necessarily just at the given moment) in the whole affair.

Of course, this interpretation does justice to

the stiuation of conversation or simple, natural narrat­ ion; in such cases, mostly "episodic memory" (i.e. storage of specific incidents in the person's own experience cf. de Beaugrande 1980: 71) plays a substantial role. In monological texts such as technical discussions or expos­ itions, the informational interest or relevance is to be seen from the point of view of the whole process of subject-matter presentation (and, indirectly, from the point of reader's knowledge and interest, as well); in these cases it is mostly the "conceptual memory (systemized knowledge)" that shows its relevance. Our interpretation is based on the persuasion that a happy performance of the process of informing somebody

276

FRANTIŠEK DANEŠ

by means of an utterance is conditioned by the fulfilment of two speaker's assumptions that may be formulated as follows: (a) "I assume that you have believed (up to this moment) that ... // that you (still) do not know that ..."; (b) "I assume that (at the moment) you have 9 interest to come to know that ..." . This interpretat­ ion appears confirmed by possible reactions of the hearer in cases when one of the assumptions (or both of them) has not been fulfilled. Cf.: (a) "Why do you tell me this? I know about it already (to be sure)" (ie. the "new information" appears actually as not new to the hearer); or: "Why, I did not even know that Dick and Vicky were married! // ... that the races took place yesterday!" {ie. the assumption that the "old informat­ ion" is known to the hearer did not prove to be valid); (b) "Why are you telling me about it (just now)?" (the assumption of interest proved to be invalid). Sometimes the speaker himself takes explicitly into account the uncertainty of his assumptions about the hearer, when saying e,g. : Maybe that you know already that ...", or "I don't know whether you haven't learned already or whether it will interest you that...". From the discussion it follows that in any act of communication, in fact, two propositions are contained: one reflecting the older state of affairs, and another, reflecting the changed, new state of affairs (plus, automatically implied, the speaker's instruction to the hearer to exchange the older state for the new one). Of course, mostly not both of the propositions are conveyed explicitly: since they overlap (namely in the scope of the old information), it appears sufficient to express

THE "QUESTION TEST" RE-EXAMINED

277

the new proposition only, in which the hearer is able to identify (recover) , as the common portion of both propos­ itions, the old information. He is able to perform the ident­ ification even without help of explicit signals, simply on the basis of his actual knowledge of the world, since he himself is best aware of what he knows already, and what, in turn, is new to him. It is just the agreement between this actual state of hearer's knowledge on the one hand, and the speaker's assumptions of it on the other that represents necessary foundation of any happy act of communication.10 It is not difficult to explain now the correlation between a heuristic wh-question and the corresponding "answer": In both of them a certain "old" state of affairs is expressed; as for the "new" ("unknown") state, it is "blocked in" with the wh-word in the question, so that the part of the answer corresponding to it represents "new" information (i.e. C) . To be sure, even wh-questions may be connected with certain assumptions of the discussed type and thus reveal a further, superimposed informational bipartition, differ­ ent from the basic one (just described, and implied in any wh-question) and overlaying it. Whereas the basic interrogative structure is signalized by means of the wh-word, the said superimposed bipartition needs to be signalized by some other means, namely by different shifts of the intonation centre from its basic placement on the wh-word (in the initial position) . (Cf. Danes, in print; cf. also Kiefer 1980; Hajičová 1983). In addition let us mention that - in distinction to the Th-Rh structuring - not all utterances necessarily

278

FRANTIŠEK DANEŠ

reveal an informational bipartition. That means that there exist utterances (or, more precisely, such uses of utterances), to which a yes-no question only (but no wh-question) is applicable. Cf. the following "all new utterance": (10)

Argentine

troops

occupied

the

Falkland

Islands ("What's new in the world?"). Notice that in (10) the component "Argentine troops" represents Th (the utter­ ance is about Argentine troops), while its variant (in similar context) (10')

The Falkland Argentine

Islands

were

occupied

by

troops

has "the Falkland Islands" as its Th. (Cf. Allerton's statement quoted in note 2.) One may also ask which of the two variants appears as "more natural" or "neutral". Probably people would mostly feel (10) as "more natural" and their judgement could be tentatively explained either in grammatical terms (active vs. passive construction) or in grammatico-semantic terms (an agentive subject is more frequented and appears as "basic") or in terms of degrees of "communicative importance" (the semantic role of agent occupies a relatively low position in the scale of "systemic ordering" - cf. note 10). In any case, how­ ever, the actual motivation of speaker's choice between the two possibilities in an particular case can hardly be fully explained in terms of objectively statable factors and conditions.

THE "QUESTION TEST" RE-EXAMINED

279

2.2.3. Thus we have arrived at a somewhat surprising conclusion that even the informational bipartition of utterance, as it is heuristically disclosed by means of application of the question test (and confirmed by possible responses of language users), appears, in principle, as context independent, though, certainly, there are considerable affinities with context even here (cf. Halliday's principle of "good reason", above).

3.1. Let us now turn to the notion of "known (given, old) piece of information" in the contextual sense proper. Without doubt, any coherent text reveals a complicated net-work of such interrelations between particular dis­ course subjects that constitute the relationship of "knownness". A semantic component (postcedent) C

. of

a text T. appears as known (to the hearer/reader) if in T. it is preceded by such a semantic component (ante­ cedent) C

that their mutual relation prompts C

. as

"already known". Thus we have to assume a set K of objectively statable semantic relations constituting the relationship of knownness. Without taking, for the moment, into account the means and ways of expression, we may divide the set K into three subsets: (1) relations of total (referential) identity, (2) relations of set (class) identity (i.e. set-theoretical relations), (3) relations of semantic affinity (having the character of analogy or of contiguity; under this affinity also configurations of knowledge such as predicate with its arguments and the so-called scenes or frames may be subsummed) ; cf. Danes, 1983. (By some scholars the said

280

FRANTIŠEK DANEŠ

relations will be called "isotopic" (cf. Greimas, Agricola) or "anaphoric" (eg. Paduceva 1970; Agricola 1975 speaks also about "semantic equivalence".) Now, we may say that the communicative articulation and the informational bipartition of utterance as well operate upon this net-work of "isotopic" relations. One of such operations underlies, e,g. , the so-called thematic progressions: Mostly the speaker chooses as the theme of an subsequent utterance one of such discourse subjects that appears as already "known" (in this or that way) in respect to the preceding context. Doing this, he relates the utterance to the context (contributing thus to text coherence) and, at the same time, he contributes to the development of the "thematic route (path)" in the text and of the isotopic net-work as well. (The motivation of speaker's choice of Theme may sometimes be, especially in poetry, of another kind, to be sure; cf. Kuno's "unpredictable themes".) - The operating of the informational bipartition may be seen again, e,g. , in such statements as "the expressions referring to activated elements (Le. known elements - F.D.) may (...) constitute the topic of the sentence (ie. B in our conception - F.D), while non-activated ones are always (...) included in the focus (ie. in C - F.D.) of the sentence"(cf. Hajičová & Vrbová, 1981: 292). 3.2. In terms of "knownness - newness" also the different degrees of communicative

dynamism

(and such

notions, connected with it, as "focus (or rheme) proper", "theme proper", "diatheme") are to be interpreted. The degree of communicative dynamism (CD) will be defin-

THE "QUESTION TEST" RE-EXAMINED

281

ed as the relative extent to which the sentence element contributes to the further development of the communicat­ ion (cf., eg. Firbas 1979: 31). It seems to me very difficult to find any other explanation of the crucial notion "further development of the communication" than to relate it to different degrees and kinds of "newness (N)". (When speaking about different kinds of N, I have in mind the difference between N in the sense just discussed in 3.1. (i.e. the introduction into the text of a new subject-matter or state of affairs) and N connected with such an instance where one and the same discourse subject will be reintroduced in such a way that some new properties of it will be revealed (cf. eg. following text sequence of names referring to an identic­ al discourse subject: a man - my uncle John - this well-known artist).) - As to the different degrees of N (or of "knownness", which looks as the same phenomenom from the opposite pole) and to various factors conditioning them, it appears that an interesting and promising field of investigations is opening here. Some empirical studies presented by Sgall's group in last years, examining contextual conditions of the rise and fading of known­ ness in a detailed and principled way, seem to be of great use here.

FOOTNOTES 1 Th will be identified even in those cases where an utter­ ance does not reveal its informational bipartion at all. Cf. Allerton's formulation: (...) presumably a sentence must have a theme and a rheme, whatever the distribution of giveness" (1978: 137). (See further below, § 2.2.2.)

FRANTIŠEK DANEŠ

282

2 It is just the stating of these affinities between the components of different levels of the structuring of sentential utterance that is the most serviceable means against the "mixing of levels", which used to be relatively frequent in linguistics (also under the labels of "psychological/logical/grammatical subject and predicate"). (Cf. the useful critical survey in Jespersen 1924, 145ff., and the exposition in Mathesius, 1907). In Czech grammatic­ al tradition, a sound differentiation of all the three types of subjects and predicates may be found in Ertl (1926, § 464), who inspired many years later Dokulil & Danes (1958) (whether Ertl followed Marty's (1897) suggestions (cf. below) is not clear). 3

Sgall et al. use the terms topic

and focus

in this sense.

4 On this exposition Carlson (1983) developed his approach to text grammar (cf. esp. 240f.). 5 The refined and penetrating analysis of differeng degrees of "salience" (activation) of utterance items presented by Hajičová & Vrbova (1981) bears also upon the problem of "knownness". See further below. 6 Curiously enough, the opposite standpoint has been taken by Carlson, 1983, but not by de Beaugrande (1980: 120f.). 7 In all following examples I assume the "normal", unmarked intonational form of the utterance. 7a Firbas (1982) - who treats the context-dependence narrow­ ly, in respect to "the immediately relevant context" only - tried to resolve this problem in a somewhat dark formulation: He main­ tains that though the said sentence components are, in a certain sense, context-dependent, "lastly they are context-independent, due to various specifications that are underivable from the immediat­ ely relevant context" (286). But in fact, it is evident that even any cross-referring naming-unit in text that is not fully synonymous with its antecedent, necessarily brings forth some new specificat­ ions of the common referent, so that the referent could not, in Firbas1 approach, appear as context-dependent! 8 The said question-answer correlation has, in fact, an old (though half-forgotten) scientific tradition. Cf. following, rather interesting passage from the psychologist Stout (1902: § 2.212): "The psychological predicate of a sentence is the determination of what was previously indeterminate. The subject is the previous qualification of the general topic to which the new qualification is attached. (...) The subject is that product of previous thinking which forms the immediate basis and starting point of further development. The further development is the predicate. (...) All

THE "QUESTION TEST" RE-EXAMINED

283

answers to questions are, as such, predicates, and all predicates may be regarded as answers to possible questions. If the statement "I am hungry" be a reply to the question "Who is hungry" then "I" is the predicate. If it be an answer to the question "Is there any­ thing amiss with you?" the "hungry" is the predicate. Every fresh step in a train of thought may be regarded as an answer to a quest­ ion. The subject, is, so to speak, the formulation of the question; the predicate is the answer." Let us add that the Stout's idea underlies, in principle, the persuasion of the renowned Soviet literary scholar Bachtin, recently accepted also by Wierzbicka, viz, that monological texts should be interpreted as deep dialogues (based on dialectical interaction between the addresser and the addressee). 9 Additionally I found out that a somewhat similar inter­ pretation, though in another frame of reference, has been suggested (but not further elaborated) by Bellert (1972: 66f., 85f.). She has hinted at the possibility to differentiate two sentence parts, one containing the presupposition and corresponding to the propositional attitude "the speaker behaves linguistically as if he believed that ...", and the other containing assertion and corresponding to the attitute "the speaker behaves linguistically as if he asserted/ wanted to inform that ... ". - On the other hand, Allerton (1978: 151) makes an interesting distinction between "assumptions" and "presuppositions". Thus "givenness concerns speaker's assumptions about the addressees awareness of things about "him ", while "pre­ suppositions concern the beliefs of the speaker and the addressee. The speaker may assert a particular belief. (...), or he may hypo­ thesize (...) it. On the other hand, he may presuppose it". 9a Cf. also following Bilỳ's statement (1981: 42): "The FSP organization of an utterance can be viewed as a favour the speaker grants the listener in the interest of promoting understanding." 10 The distinct notion of degrees of "communicative import­ ance" of sentence elements, systematically connected with different semantic types of constituents (participants) of the propositional sentence structure, appears as a word-order factor. This "systemic ordering" (the scales of which has been tentatively established by Firbas and by Sgall) works in cases where other word-order factors are not present.

FRANTIŠEK DANEŠ

284

REFERENCES

Agricola, E. 1975. Semantische

Relationen

im Text

und

im

System, Halle: Saale. Allerton, D. J. 1978. "The notion of 'giveness' and its relation to presupposition and the t h e m e " , Lingua 44. 133-168.

Bachtin, M. 1979. Estetika slovesnogo tvorčestva. Moskva. Bellert, I. 1972. On the Logico-Semantic Structure of Utterances. Wrocław. de Beaugrande, R. 1980. Text, Discourse and Process: Toward a Multidisciplinary Science of Text. Norwood, N. J . Benes, E. 1968. "On two aspects of functional perspective". TLP 3. 264-274.

sentence

Bìlý, M. 1981. Inter sentential Pronominalization and Functional Sentence Perspective (in Czech, Russian, and E n g l i s h ) , Lund.

Bogusławski, A. 1977. Problems of the Thematic-Rhematic Structure of Sentences. Warszawa. Carlson, L. 1983, Dialogue Games: An Approache to Discourse Analysis. Dordrecht. Chafe, W, L, 1974. "Language and consciousness". Lg 50. 111-132. Dahl, 0. 1976. "What is new information". Reports on Text-

-Linguistics:

Approaches

to

Word-Order.

37-50. Abo.

Danes, F. 1970. "One instance of Prague School methodology: Functional analysis of utterance and Text". Method and Theory in Linguistics ed. by P. Garwin. 132-146. Paris: The Hague. Danes, F. 1974. "Functional sentence perspective and the organization of the text". Papers on Functional

Sentence

Perspective.

Prague.

Danes, F. 1983. "On text-constituting semantic relations". Language in Function ed. by S. Rot. Budapest.

Danes, F. In print. Včta

a text

[Sentence

and

Text],

Praha. Dokulil, M. & Danes, F. 1958. "K tzv. vyznamovè a mluvnickè stavbe vèty" [On the so-called semantic and grammatic­ al structures of the sentence], o v e d e c k e m p o z n a n i

soudobych jazykü. 231-246. Praha. Ertl, V, 1926. Gebauerova Mluvnice ceska pro skoly stredni a ustavy ucitelskè. Novè zpracoval V. Ertl. II. Skladba [Gebauer's Czech Grammar II, a revised version by V. Ertl.] Praha. Firbas, J. 1971. "A functional view of 'ordo naturalis'". BSE 1 3 , Brno,

THE "QUESTION TEST" RE-EXAMINED

285

Firbas, J. 1982.""Aktualni cleneni vetne", ci"funkcni perspektiva vetna"?" [Is "functional sentence perspect­ ive" an equivalent of "aktualni cleneni vetne"?]

Slovo a slovesnost 43. 282-293. Garvin, P. 1963. "Czechoslovakia". Soviet

and East

Euro­

pean Linguistics ed. by T. S. Sebeok. 499-522. The Hague. Hajicova, E. & J. Vrbova. 1981. "On the saliance of the elements of the stock of shared knowledge. Fol 65. 291-303. Hajicova, E. & J. Vrbova. 1982. "On the role of the hier­ archy of activation in the process of natural language understanding". Coling 1982 ed. by J. Horecky. 107113. Amsterdam, Hajičová, E. 1983, "On some aspects of presuppositions of questions". Questions and Answers ed by F, Kiefer. 85-96. Dordrecht, Halliday, M, A, K. 1967. "Notes on transitivity and theme in English. Part I I " . JL 3. 199-247. Halliday, M, A, K. 1970. "Language structure and language function". New Horizons in Linguistics ed. by J. Lyons. 140-165. Penguin Books. Halliday, M, A. K. 1970. "Functional diversity in language as seen from a consideration of modality and mood in English", FL 6. 322-361. Jespersen, O, 1980. The Philosophy of Grammar. London. Kiefer, F. 1980. "Yes-no questions and wh-questions.

Speech

Act

Theory

and Pragmatics

ed. by J. Searle,

F. Kiefer & M. Bierwisch. 97-119. Dordrecht. Kuno, S, 1976. "Three perspectives in the functional approach to syntax", Sound, Sign and Meaning ed. by L. Matejka. Ann Arbor, Marty, A, 1897. "Über die Scheidung von grammatischem, logischem und psychologischem Subjekt, resp. Prädikat".

Archiv

für

systematische

Philosophie

3. 174-190;294-

333 . Mathesius, V. 1907. "Studie k dejinám anglickeho slovosledu" [studies contributing to the history of English

word-order], Vestnik

Ceske

akademie

ved

a umèni

16.

261-275. Praha. Paduceva, E. V. 1980. "Anaphoric relations and their re­ presentation in the deep structure of a text". Pro­ gress in Linguistics ed. by M. Bierwisch & K. E. Heidolph. 224-232. Paris: The Hague. Stout, G, F. 1902. Analytic Psychology. London. Sgall, P., Hajičová, E. & E. Benešová. 1973. Topic, Focus

and Generative

Semantics.

Kronberg.

FRANTIŠEK DANEŠ

286

Aktualni

Sgall, P. Hajicova, E. & E. Buranova. 1973.

cleneni

vety

v cestine

[Topic

and Focus

in

Czech],

Praha,

Svoboda, A. 1981. Diatheme: A Study in Thematic Elements, Their Contextual Ties, Thematic Progressions and Scene Progressions Based on a Text from Aelfric. Brno. Wierzbicka, A. (in p r i n t ) . "Semantic analysis of the xicon as a tool for comparing cultures".

le­

FERENC KIEFER Budapest

FOCUS AND MODALITY

1. In this paper I am going to show that there is an interesting relationship between focus and modality. I will draw my material from Hungarian but at least some of the claims I am going to make will have a more general validity. I will confine myself to epistemic modality. The main concern of this paper will be to demonstrate that

there are two types of epistemic modality which

are radically different from each other. These two types of epistemic modality can be distinguished from each other on the basis of focus. 2. Focus in Hungarian can be defined positionally (syntactically). For the present purpose we can take a somewhat simplicistic view of focus and neglect the irrelevant details. The focussed constituent bears main stress and it immediately precedes the verb. Consider (1)(a)

Peter talalkozott Évával. 'Peter has met Eve'

(b)

Peter Évával talalkozott. 'It is Eve whom Peter has met'

In (1) (a) the focus-position is empty, in (1)(b), how­ ever, it is filled with the constituent Évával 'with

288

FERENC KIEFER

Eve'.

In what follows we will mark the focussed constit­

uent by underlining.

Notice, incidentally, that (1)(a)

is the unmarked, neutral word order. On the other hand, (1)(b) has a marked work order. Let us look at some more examples. (2) (a)

Peter megirta a levelet Evanak. 'Peter has written the letter to Eve'

(b)

Peter

Évával

irta meg a levelet.

'It is Eve to whom Peter has written the letter' (c)

Peter a levelet

irta meg Evanak.

'It is the letter that Peter has written to Eve' (d)

Vetev

irta meg a levelet Evanak.

'It is Peter who has written the letter to Eve' The only sentence that exhibits unmarked, neutral word order is (2)(a). All other sentences are marked. The verb ir fix

meg.

'write' is prefixed here by the perfective pre­ Whenever a constituent is moved into the focus

position, the prefix must be moved after the verb. This can be observed in sentences

(2)(b)-(d).

There are cases where focus does not affect word order, i.e. neutral word order is not destroyed. Consider (3) (a)

Peter levelet ir. 'Peter is engaged in letter-writing'

(b)

Peter levelet

ir.

'Peter is engaged in letter-writting (and not in something else)' (4)(a)

Peter a szobaban dolgozik.

289

FOCUS AND MODALITY 'Peter is working in the room' Peter a szobdban dolgozik.

(b)

'Peter is working in the room (and not some­ where else) ! (5) (a)

Peter megirta a levelet. 'Peter has written the letter' Peter megirta.

(b)

a levelet.

'Peter did write the letter

(believe m e ) '

In the above sentences focussing has not affected neutral word order. Certain constituents immediately precede the verb even if they are not focussed. Such are

(i) article-less object nouns

constituents

(cf. (3)(a)-(b)), (ii)

certain locative and temporal adverbials with intransit2 ive verbs (cf. (4)(a)-(b)) and (iii) prefixes . All these seem to form a close semantic unit with the verb, they can be considered to be constituents of complex verb phrases rather than free arguments. They have often been 3 referred to as reduced complements . V e r b s , too, can occupy the focus position. For example, (6) (a)

Peter irta a levelet. 'Peter has been writing the letter'

(b)

Peter ivta

a levelet.

'Peter has

been writing the letter

(believe

me) ' The comparison of (5)(a) with

(3)(a) with

(5)(b) and

(3)(b),

(6)(a) with

(4)(a) with

(4)(b),

(6)(b) reveals a

serious defect in the first formulation of the notion of focus.

(6)(b) shows that verbs, too, can be focussed. In

addition, there is a marked difference between the stress carried by the verb ivta

'write' in

(6)(a) and in (6)(b) .

290

FERENC KIEFER

In the latter case, the verb gets an extra heavy stress. Similar differences can be observed in connection with the sentences (3)(a)-(b), (4)(a)-(b) and (5)(a)-(b). In (3)(a) the constituent levelet 'letter' carries main stress, in (3)(b) it receives extra heavy stress. Similar­ ly, in (4)(a) the constituent szobádban is the bearer of main stress which gets and extra heavy stress in (4)(b). The same holds true for the prefix meg in the sentences (5)(a)-(b). In view of the above facts focus can now be defined in the following fashion. Focus (=F) is the constituent that exhibits the following two properties: (i) it bears an extra heavy stress and (ii) it is either the verb or the constituent immediately preceding the verb. Semantically, F is defined by the property of exhaustive listing . If 'Eve' is the focus, as in (1)(b), then 'Eve' is the only person met by Peter. If 'the letter' is focussed, as in (2)(c), then 'the letter' is the only thing that Peter has written. In general terms, "The notion of 'exhaustive listing' relates to the fact that, under some conditions yet to be specified, the focus contains all items that can be reported to bear the given relationship (usually expressed by the verb) to the items referred to by the bound segment of the sentence." (Sgall-Hajičová-Benešová: 129) Or, to put it differently, the set described in the focus is exhaust­ ive in the sense that it contains all items for which the property predicated in the sentence holds. In the examples discussed above the set under consideration consisted of one element only. Consider, however, the following sentence

FOCUS AND MODALITY

(7)

Peter Evdval

es Jdnossal

291 talalkozott.

'It is Eve and John whom Peter has m e t !

Here 'Eve' and 'John' are the only persons (the only elements of the set described in the focus) of whom it can be predicated (in the given context) that they have been met by Peter. 3. In general, possibility is expressed in Hungarian by means of the suffix -hat/-het. The choice between the two variants is determined by vowel harmony. For example, the verb l at 'see' will receive the suffix -hat: lathat 'may see', the verb enekel 'sing' the suffix - h e t : enekelhet 'may sing'. The possibility suffix is followed by tense/mood and personal suffixes . Necessity is expressed by the impersonal verb kell 'must' which is, however, impersonal in its modal mean­ ing only. As a full verb meaning 'need', German 'brauchen', it has a full paradigm. The modal auxiliary kell requires a rather complicated syntactic structure. Consider (8)

Peternek meg kell irnia a levelet. lit. Peter-dat. pref. must write-Pers.suf. the letter-acc . 'Peter must write the letter'

The modal auxiliary is inserted between the prefix and the main verb, the infinitive, which is inflected ing to person in this case. In addition, the

accord­

'logical

subject', 'Peter', is put into the dative case. Semantically one way of looking at possibility and necessity is to take

'modal reasoning' to be the basic

notion. In the case of epistemic modality the speaker

292

FERENC KIEFER

draws certain conclusions with respect to the validity of a given statement on the basis of what he knows. Each statement activates a certain portion of the speaker's knowledge. Let us call the knowledge activated by the state-of-affairs expressed by the statement at hand the background. The background consists of propositions all of which are taken to be true by the speaker and all of which are relevant for the evaluation of the statement in question. In the simplest case the speaker may just try to determine whether the state-of-affairs expressed by the given statement is compatible with the background or whether it follows from it. If the state-of-affairs at hand is compatible with the background, it is a possible state-of-affairs, if it is not compatible, it is not possible, that is, the state-of-affairs in quest­ ion is exluded. If the state-of-affairs in question follows from the background, it is necessary, if it does not follow, it is not necessary. Let us denote the state­ ment (=proposition) to be modally qualified by p and the background by B. To paraphrase Angelika Kratzer the following definitions may be stipulated (Kratzer 1978): (9)(a)

It is possible that p iff p is compatible with B.

(b)

It is necessary that p iff p follows from B.

We have defined B as a set of propositions representing that knowledge that is relevant for the modal evaluation of p. B may thus be termed epistemic background. Evident­ ly, for other types of modalities other backgrounds are needed: there are deontic, dispositional, boulomaic, circumstantial, etc. backgrounds.

FOCUS AND MODALITY

293

Since the possibility or necessity of p hinges on the epistemic background B, (9)(a)-(b) can be made more precise in the following fashion. (10)(a)

p is epistemically possible iff p is compatible with the epistemic background B.

(b)

p is epistemically necessary iff p follows from the epistemic background B.

It is generally taken for granted that epistemic possib­ ility and epistemic necessity are interdefinable in the same way as alethic possibility and alethic necessity are. That i s , the following equivalences are considered to be valid: (11)(a) (b) In addition, given an epistemic B , the validity of the following implication is also taken for granted: (12)

[]

p c p

That i s , given B, if p is epistemically necessary, then it follows that p is the case. Let us now see how epistemic possibility is express­ ed in Hungarian. Consider the following set of examples: (13)(a)

Péter levelet

irhat.

lit. Peter letter write-may 'Peter must be writing a letter' (b)

Péter irhat

levelet.

'Peter may be writing a letter' (14)(a)

Péter a szobâban

dolgozhat.

lit. Peter the room-in work-may 'Peter must be working in the room'

FERENC KIEFER

294

Peter dolgozhat a szobaban. 'Peter may be working in the room' (15)(a) Peter nyelvesz lehet. lit. Peter linguist be-may 'Peter must be a linguist' (b) Peter lehet nyelvesz (b)

'Peter may be a linguist' Though all these sentences contain the possibility suffix -hat/-het, the interpretation of the (a) sentences is radically different from that of the (b)-sentences. Notice that in the (a)-sentences the focus position is occupied by a reduced complement of the verb whereas in the (b)-sentences the modalized verb stands in focus. The (a)-sentences can only be interpreted epistemically whereas the (b)-sentences may also have other modal readings. In this paper, however, only the epistemic reading will be considered. The question is, then, what is the difference between the two epistemic readings, the epistemic reading of the (a)-sentences and the epistemic reading of the (b)-sentences? Let us first examine the (b)-sentences. On the epistemic reading the sentence (13)(b) means that the proposition 'Peter is writing a letter' is compatible with what we know about the world, i.e. with the given epistemic background. In other words, the state-of-affairs described by the proposition 'Peter is writing a letter' is possible, it is not excluded by the given epistemic background. But other possibilities, too, exist. Peter may very well do something else. In general, there are several things which are compatible with a given epistemic background. Similar considerations hold

295

FOCUS AND MODALITY for the sentences (14) (b) and (15) (b) . (14) (b) means

that it is possible that Peter is working in the room, that is, it is not excluded by the given epistemic back­ ground. But the speaker does not have any special reasons to believe that the proposition 'Peter is work­ ing in the room' is more likely to be true than any other proposition compatible with the epistemic back­ ground. Finally, (15)(b) means that on the basis of what we know about the world Peter may very well be a linguist but he may also be something else, say, a doctor, a mathematician, etc. The interpretation of the (b)-sentences can thus be accounted for by means of the definition of epistemic possibility as given in (10)(a). The interpretation of the (a)-sentences is quite different. Notice that in the English equivalents of the Hungarian sentences the modal -hat/-het

has been

rendered by 'must' rather than by 'may'. The modal 'must', however, should be taken here to mean something like 'it is very likely the case that'; it is the infer­ ential 'must' and not the expression of epistemic necessity. The meaning of (13)(a) can be paraphrased in the following manner. The speaker has certain reasons to believe that the most likely thing which Peter may be doing just now is letter-writing. When uttering (13)(a) one cannot go on saying 'but he may very well do something else' since (13)(a) expresses the only propos­ ition for which the speaker has some evidence. The interpretations of the sentences (14)(a) and (15)(a) are quite similar. On the basis of what he knows about the world the speaker draws the conclusion that Peter

FERENC KIEFER

296

is working in the room and that Peter is a linguist. Notice, by the w a y , that the (a)-sentences have neutral word order whereas the word order of the (b)-sentences is marked. Next, consider the modalized versions of (2)(b)-(c): (16) (a)

Peter É v á n a k irhatta meg a levelet. lit. Peter Eve-dat. write-may-Past-Pers.suf. perf.pref. the letter-acc. 'Peter must have written the letter to Eve-

(b)

Peter a levelet

irhatta meg Evanak.

'Peter must have written the letter (c)

Peter 'Peter

to Eve'

irhatta meg a levelet Evanak. must have written the letter to Eve'

In these sentences the focus position is occupied by a constituent which normally does not stand there: it must be moved there from its neutral position, All these sentences have an interpretation which is quite parallel to the interpretation of the (a)-sentences above. The interpretation corresponding to the interpretation of the

(b)-sentences can only be rendered by a sentence

such as (17)

Peter megirhatta a levelet Evanak.

which reflects a neutral word order. For the simplicity of exposition let us call the epistemic interpretation of the (a)-sentences epistemic

possibility

objective

epistemic

and that of the possibility

subjective

(b)-sentences

. As we saw above,

objective epistemic possibility can be accounted for in terms of epistemic logic, (10)(a) is an adequate definit­ ion for this modality. But what about subjective epistemic

FOCUS AND MODALITY

297

possibility? The sentences expressing subjective epist­ emic possibility have two properties in common: (i) they all contain a focus with the property of exhaustive list­ ing and (ii) they all contain the suffix -hat/-het

which

expresses possibility, i.e. compatibility with a given epistemic background. In view of these properties one might be tempted to derive the meaning of the (a)-sent­ ences, i.e. to account for subjective epistemic possibil­ ity in the following manner. The proposition levelet

ir

Peter

'Peter is writing a letter' is compatible

with the given epistemic background because of the suffix -hat/-het

in (13)(a). On the other hand, because of the

property of exhaustive listing associated with the focus in (13)(a) the proposition Peter

levelet

ir

is the only

one that is compatible with the given epistemic back­ ground. In other words, the proposition Peter

levelet

ir

is not only compatible with the epistemic background but it also follows from it. To be sure, this account seems to be rather attractive for various reasons. First of all, it would explain the use of 'must' in the English equivalents of the Hungarian sentences expressing subject­ ive epistemic possibility. Furthermore, we would have a unified account of modality in terms of the notions of compatibility and consequence. Subjective epistemic possibility would differ from objective epistemic possib­ ility whereas it contributes to the meaning of sentences expressing subjective epistemic possibility. Unfortunate­ ly, however, this account has some serious flaws as we shall see presently. We have already noted that the 'must' in the English equivalents of the Hungarian sentences expressing subjective epistemic possibility is not the 'must' of

298

FERENC KIEFER

objective epistemic necessity. For this 'must' neither the equivalence (11)(b) nor the implication (12) holds. It seems to be stronger in some sense than objective epistemic 'may' but at the same time it is weaker than objective epistemic 'must'. The first solution to this problem which might come to one's mind is to define a notion of relative modality by refining the theoretical apparatus which we already have at our disposal. It goes without saying that possibility and necessity could also be a matter of degree. (Kratzer 19 82) However, although a notion of relative modality will certainly be needed in any adequate account of modality, it will not do in the case at hand. The difference between, say, the objective epistemic 'must' and the subjective epistemic 'must' is not just a matter of degree, as we shall see immediately. One might also think of weakening the background. Evidently, speakers draw their conclusions not only on the basis of firm knowledge but also on the basis of assumption, beliefs, etc. Let us call such a background weak epistemic background. One might now argue that the subjective epistemic 'must' expresses the fact that a proposition follows from a weak background. Notice, how­ ever, that the notion of consequence cannot be the same here as in the case of objective epistemic 'must'. What we need is a notion of 'practical inference' rather than that of logical consequence. It is quite clear that we often (if not mostly) draw conclusions on the basis of weak epistemic backgrounds, therefore any adequate account of modal reasoning has to take into consideration practical inferences as well. Unfortunately, however, the

FOCUS AND MODALITY

299

difference between the two 'must's' is not a question of background. Before further elaborating on this point, an important remark seems to be at place. Some sentences seem to be ambiguous between the subjective and the objective epistemic reading. Consider Peter találkozhatott

(18)(a)

Évával.

lit. Peter meet-poss.-Past-Pers.suf. Eve-with 'Peter must have met Eve' or 'Peter could meet Eve' Peter megirhatta a levelet.

(b)

lit. Peter perf.pref.-write-poss.-Past-Pers. suf. the letter-acc. 'Peter must have written the letter' or

'Peter could write the letter' Peter èbedelhet.

(c)

lit. Peter dine-poss. or

'Peter must be eating lunch' 'Peter may be eating lunch'

In these sentences the verb is focussed. In the case of prefixed verbs, stress will automatically be shifted on the prefix. As we saw above, in all such cases the modal sentences are interpreted as expressing objective epi­ stemic possibility. At the same time, however, these sentences may also have the other interpretation, though in this case there is a tendency to pronounce the sent­ ences with a different intonation pattern but this need not be so. In any case, the ambiguity exhibited by such sentences as (18)(a)-(c) calls for an explanation. It has been observed by several scholars that there is a marked tendency to interpret English 'must', if stressed, as expressing objective epistemic necessity.

300

FERENC KIEFER

Karin Aijmer, for example, puts this in the following way: "It always seems to be the case that the 'Grundbedeutung' can be preserved when must

is stressed, how­

ever." (Aijmer 1980:130) What she calls 'Grundbedeutung' is the objective epistemic reading of 'must'. Thus, there seems to be a tendency to interpret the sentence (19)(a) as expressing objective epistemic necessity, in contrast to (19)(b) where the subjective reading pre­ vails . (19)(a) (b)

Petr must

be in the kitchen,

Petr must be in the

kitchen.

Apparently, these sentences have two different intonat­ ion contours as well. The situation is similar in German: (20)(a) (b)

Petr muss

in der Küche sein.

Petr muss in der Kuche

sein.

Now, in Hungarian possibility is expressed by a suffix. Suffixes, normally, cannot be stressed: stress always falls on the first syllable of the word, however long it may be. One might therefore claim that the verb in (18)(a)-(c) gets stressed on two different accounts. One gets the first reading (=objective epistemic possibility) if one wants to put special emphasis on the expression of possibility and the second reading (=subjective epi­ stemic possibility) if the verb is focussed. This explanation seems to be fairly plausible and, I think, it is basically correct. The ambiguity discussed above appears only in cases when the sentence has neutral word order. In such cases the verb carries the main stress. In sentences (18)(a)(c) 'normal' main stress overlaps with 'focus' stress.

FOCUS AND MODALITY

301

Sentences such as (13)(b)-(15)(b) can never be ambiguous in this sense. In these sentences the 'normal' bearer of main stress is the reduced complement. Reduced com­ plements precede the verb in neutral word order, in (13) (b)-(15)(b), on the other hand they follow the verb. The structural differences between subjective and objective epistemic possibility should be fairly clear by now. Hungarian draws a systematic distinction between these two readings. However, subjective epistemic possibility is still an intuitive term which calls for explication. In particular, we have to show that the difference between the two modalities is not a differ­ ence in degree but that the two have radically different roles in the semantics of modalities. 4. Notice first that subjective and objective epistemic possibility have different paraphrases. For example, (13)(b), but not (13)(a), can be paraphrased by (21)(b) (for convenience's sake I repeat here the sentence (13)(a) under (21)(a)): (21)(a) (b)

Peter irhat levelet. 'Peter may be writing a letter' Lehetseges, hogy Peter levelet ir. 'It is possible that Peter is writing a letter'

Both (21)(a) and (b) are descriptions of reality and it is meaningful to ask whether such a description is true or false (given an appropriate epistemic logic). The sentence (13)(a), on the other hand, (repeated here under (22)(a)) can only be paraphrased by modal sentent­ ial adverbials.

FERENC KIEFER

302

Peter levelet

(22)(a)

irhat.

'Peter must be writing a letter' (b)

Peter talan levelet

(c)

Peter valosziniileg levelet

ir.

'Perhaps, Peter is writing a letter' ir.

'Peter is probably writing a letter' Peter biztosan levelet

(d)

ir.

'Peter is surely writing a letter' (22)(b)-(d) are all close paraphrases of (22)(a) but none of them is completely synonymous with it. This is very often the case with sentence adverbials. Adverbials such as talàn biztosan

'perhaps', valosziniileg

'probably'and

'surely, certainly' express speaker attitudes,

they do not belong to the propositional part of the sentence, that is, they are not part of the description of the world. They rather express the speaker's attitude towards a certain

state-of-affairs. 8

Speaker attitudes 9 are not asserted, they are indicated or expressed .

Thus, sentences content

(22)(b)-(d) consist of a propositional

('Peter is writing a letter') and an expression

of the speaker's attitude towards the state-of-affairs expressed by the propositional content at hand. Con­ sequently, the question whether such sentences represent a true or a false proposition cannot be asked. It is impossible to account for (22)(b)-(d) in purely truth-functional terms. Speaker attitudes, in general, can neither be questioned nor negated. The sentences

(23)(a)

-(c) are all bad. (23)(a)

+Nem igaz, hogy Peter talan levelet ir. 'It is not true that perhaps Peter is writ­ ing a letter'

303

FOCUS AND MODALITY

(b)

+Nem igaz, hogy Peter valosziniileg

levelet

ir.

' It is not true that Peter is probably writ­ ing a letter' (c)

+ Nem igaz, hogy Peter biztosan levelet ir. 'It is not true that Peter is surely writing a letter'

In these sentences negation has the widest scope, i.e. the modals are included in the scope of negation. On the other hand, it does not come as a surprise that sent­ ences in which negation is included in the scope of the modal operator are quite alright. Consider (24) (a)

Peter talàn nem levelet ir. 'Perhaps, Peter is not writing a letter'

(b)

Peter valosziniileg nem levelet ir.

(c)

Peter biztosan nem levelet ir.

'Peter is probably not writing a letter'

'Peter is surely not writing a letter' Similar considerations hold true for questions as wells modal adverbials cannot be included in the scope of the question operator b u t , of course, questions can be 'modalized'10. Let us now return to

(22)(a). Intuitively, as point­

ed out above, this sentence is related to the sentences (22)(b)-(d). One could thus claim that this sentence is not a descriptive statement about the world either. It expresses, similarly to the sentences which contain modal adverbials, the speaker's attitude. Consider (25)(a)

+Nem igaz, hogy Peter levelet

irhat.

'It is not true that Peter must be writing a letter'

304

FERENC KIEFER (b)

+Peter levelet irhat? 'Must Peter be writing a letter'

Neither (25)(a) nor (25)(b) are possible sentences in Hungarian. The same holds for all the other cases of subjective epistemic possibility discussed above. We may thus conclude that sentences expressing subjective epi­ stemic possibility can neither be negated nor questioned. And, again, at least negation is possible if it is in­ cluded in the scope of the modal: Peter nem levelet

(26)

irhat.

'Peter may not be writing a letter (he must be writing something else) ! Thus, the modal operator that brings about the subject­ ive epistemic reading must have the wider scope. Notice, incidentally, that modal adverbials can easily be combined with objective epistemic possibility (in the case of subjective epistemic possibility such combinations are not excluded either but the sentences sound slightly redundant). For example, (27) (a) (b)

Peter talàn irhat

levelet.

'Perhaps, Peter may be writing a letter' Peter bizonyara irhat levelet. 'No doubt, Peter may be writing a letter'

In these sentences we have to do with two modal operat­ ors. The operator that brings about the objective epist­ emic reading is included in the scope of the operator that expresses various (modal) speaker attitudes. Negation, question-formation, etc. are propositional operators, they map propositions onto propositions. The modal operators of modal logic (epistemic operator,

FOCUS AND MODALITY

305

deontic operator, etc.), too, are propositional operat­ ors. Modal adverbials and subjective epistemic possibil­ ity, on the other hand, express attitudinal operators. Such operators evaluate propositions. The propositions thus evaluated are no longer propositions, they are no longer descriptions of the world and they can no longer be accounted for in terms of truth conditions. It is generally true that propositional operators can easily stand in the scope of attitudinal operators but the reverse does not hold: attitudinal operators can never be included in the scope of a propositional operator. The above general rule explains why sentences containing an attitudinal operator (modal or other) can­ not be negated, questioned, focussed, contrasted, co­ ordinated, etc. That is, none of the propositional operators is applicable. In sum, then, the essential difference between objective epistemic possibility and subjective epistemic possibility cannot be a matter of degree. Objective epistemic possibility is a propositional operator. If such an operator is applied to a proposition we get another proposition which can be accounted for truth-functionally. On the other hand, subjective epistemic possibility is an attitudinal operator. It turns a pro­ position into a subjectively evaluated proposition. Such a proposition is no longer a description of the world, it does not make any sense to ask whether it is true or false. In some cases it is possible to get the objective epistemic reading by putting extra stress on the modal

306

FERENC KIEFER

auxiliary. For example, Peternek kellett

(30)

Évával

találkoznia.

'It was necessary that Peter met Eve' This sentence can no longer be interpreted as expressing subjective epistemic necessity. The speaker's commitment in the case of subjective epistemic kell

is only slightly stronger (if at all)

than in the case of subjective epistemic -hat/-het.

It

goes without saying that the relationship between sub­ jective epistemic kell -het

and subjective epistemic

-hat/

cannot be expressed in terms of the equivalences

(11)(a)-(b). This follows already from the impossibility of negation. Furthermore, the implication (12) does not hold either for the subjective epistemic kell.

There is

no 'epistemic space1 for which (31)(a) would imply the truth of (31)(b): (31)(a)

Peternek konyvet

kell olvasnia.

'Peter must be reading a book' (b)

Peter konyvet

olvas.

'Peter is reading a book' We saw above that the objective epistemic reading of kell

can be made apparent in two ways: either by put­

ting extra stress on kell(this is the device used in many unrelated languages) or by moving kell

into a higher

clause. Notice that the subjective epistemic reading of kell,

too, can be made explicit. In (32)(a)-(c) the

sentences (28)(a)-(c) are modified by a believe-clause. (32)(a)

Azt hiszem, Peternek a konyhdban

kell lennie.

'I think, Peter must be in the kitchen'

FOCUS AND MODALITY

(b)

Azt hiszem, Peternek konyvet

307 kell olvasnia.

'I think, Peter must be reading a book' (c)

Azt hiszem, Peternek

Évával

kellett talalkoz-

nia. 'I think, Peter must have met Eve' For quite obvious reasons these sentences can only be interpreted as expressing subjective epistemic necess-

lty.12 In sum, then, we may conclude that if kell is focussed (stressed), the sentence in question cannot be interpreted as expressing subjective epistemic necess­ ity. But focus on a constituent different from kell does not make the sentence disambiguous. 6. Though we may have a relatively clear picture of subjective epistemic possibility and necessity by now, we have not answered the question as yet how this modality can be accounted for. In this concluding sect­ ion I am going to sketch a possible solution to the problem. As we say, objective epistemic possibility and objective epistemic necessity can be accounted for in epistemic logic. The modal sentences in which they occur are descriptions of the world, hence it is reasonable to assign them truth-values. Thus, the pro­ position Peter may he reading a hook is true just in case the proposition Peter is reading a hook is compat­ ible with the background B and the proposition Peter must he reading a hook is true just in case the pro­ position Peter is reading a hook follows from the back­ ground B. In general

FERENC KIEFER

308 (33)(a) (b)

p is true iff p is compatible with B. p is true iff p follows from B.

So far so good. But how should modal attitudes be accounted for? It should be sufficiently clear by now that they are not truth-functional, consequently means other than truth-conditions are called for. Modal attitudes express qualifications in terms of the speaker's commitment to the possibility of a certain state-of-affairs. Apparently, a modal attitude need not express the real beliefs of the speaker: he may express a weaker commitment. This is something which can be accounted for by means of the sincerity condition (34) and the conversational postulate (35): (34)

Sincerity

Condition:

Do not express ATT

(modal) if you do not have sufficiently enough evidence for it. (35)

Conversational

Postulate:

Express ATT(modal)

that indicates the strongest commitment for which you have evidence. ATT(modal) denotes any given modal attitude. The semantic structure of sentences which contain linguistic expressions of ATT(modal) consists of two main parts: (i) the proposition p and (ii) ATT(modal). Schematically, (36)

sem = (ATT (modal) ,p)

In truth-functional semantics one assigns denotata to propositions. One way to account for (36) would be to do essentially the same thing, only the denotata would be different. Notice first that the modal attitudes dis­ cussed have all to do with possibility (including

kell

309

FOCUS AND MODALITY

'must'), none of them qualifies necessity. This seems to be the general rule: in Hungarian there is not a single expression of modal attitude which would be related to necessity. They all express degrees and kinds of possib­ ility. And the speaker would only use expressions of ATT(modal) if he knows that p is possible. It seems thus reasonable to assign p in (36) the state-of-affairs expressed by the proposition 'p is possible'. The first part of (36), ATT(modal), can be interpreted as the indication of the speaker's commitment to 'p is possible'. However, semantically, subjective epistemic possibility does not simply mean possibility, as we saw above. The semantic meaning of subjective epistemic sentences derives from the interplay of possibility and focus (i.e. exhaustive listing). In other cases of ATT(modal), of course, this need not be so. Thus, we may formulate for subjective epistemic -hat/-het 13 'denotation-conditions': (37)(a)

and kell

the following

p is compatible with B, (=possibility)

(b)

p is the only proposition compatible with B

(c)

the speaker indicates that he has a certain

(=exhaustive

listing)

evidence with respect to the possibility of P To indicate a commitment should be taken to mean

'to

express a commitment'. The commitment expressed need not correspond exactly to the evidence the speaker has. The denotatum of ATT(modal) is the corresponding modal attitude expressed. It should be made clear that -het

and kell

reflect

-hat/

(are expressions of) two slightly

different modal attitudes but this fact has not been

310

FERENC KIEFER

taken care of in the formulation of (37)(c). This condition is anyhow rather vague but I cannot offer anything more adequate at the moment. One may claim that the conditions (37)(a)-(c) must be fulfilled for (36) to have meaning. This brings us to the end of our discussion of the interrelationship between focus and modality in Hungar­ ian. I hope to have been able to show that focus plays a decisive role in distinguishing two radically differ­ ent epistemic readings: the reading in which a modal propositional operator is involved and the reading in which an attitudinal operator is at stake. This distinct­ ion is certainly not a privilege of Hungarian: it is drawn in many languages. What is special about Hungarian is the interplay between the positionally

(syntactically)

defined focus and modality. NOTES 1 There is a steadily growing literature on the topic-focus structure of Hungarian. In the present context, however, it is sufficient to refer to E. Kiss 1981. 2 I am speaking of constituents within the verb phrase, of course. The topic of the sentence, too, precedes the verb. 3

Cf., for example, E. Kiss 1981.

4 In general, sentence stress is carried by the main verb in Hungarian. This means that in (6)(a), for example, the verb irta will receive (noncontrastive) main stress whereas in (6)(b) it will receive contrastive (emphatic, focus) stress. 5 For a detailed discussion of the semantics of 'exhaustive listing' cf. Szabolcsi 1981. 6 I have discussed the semantics of the possibility suffix in more detail elsewhere. Cf. Kiefer 1982. 7 This terminology comes from John Lyons. He points out that "In principle, two kinds of epistemic modality can be distinguished:

FOCUS AND MODALITY

311

objective and subjective." (Lyons 1977:797) It should be made clear, however, that my use of the relevant terms is somewhat more specific since it refers to the two different uses of the possibility suffix ~hat/-het and the auxiliary kell 'must'only. 8 That sentence adverbials express speaker attitudes has been argued for quite convincingly in Lang 1979. 9 For a more detailed discussion of speaker attitudes cf. Lang 1983. 10 It is an interesting fact about Hungarian that there is a close correspondence between the order of operators and their respective scopes. Let be 0. and Oj. two operators. Furthermore, let us assume that both precede the verb and that 0i. precedes 0.. It will follow then that 0. has wider scope than 0j.. Cf. E. Kiss 1984 for the details. 11 Sentence such as (29)(a)-(b) can easily be negated, questioned, etc., i.e. they are propositions. 12 Notice that azt hiszem 'I think', too, is an attitudinal operator. Objective epistemic necessity (and, as a matter of fact, also objective epistemic necessity) is not compatible with beliefs since this modality is a matter of knowledge. In contrast to (32) (a)-(c) a sentence such as Tudom, hogy Peternek a konyhaban kell lennie. 'I know that Peter must be in the kitchen' can only express objective epistemic necessity. 13 In the case of statements denotation conditions are equivalent with truth-conditions. Something like denotation condit­ ions has also been discussed in Lang 1983:331 (Lang speaks of 'Erfüllungsbedingungen'). 14 This paper is part of a more comprehensive study on epistemic modality in natural language. Cf. Kiefer (forthcoming).

312

FERENC KIEFER

REFERENCES Aijmer, K. 1980. Evidence and the Declarative Sentence. Stockholm. Kiefer, P. 1980. "What is possible in Hungarian?". Acta Linguistica Hungarica. Kiefer, P. "Epistemic possibility in natural language". Forthcoming. E. Kiss, K. 1981. "Structural relations in Hungarian, a 'free' word order language". Linguistic Inquiry 12. 185-213. E. Kiss, K. 1984. Aspects of Hungarian Syntax. Dordrecht. Kratzer, A. 1978. Semantik der Rede. Kronberg. Kratzer, A. 1981. "The notional category of modality". Words, Worlds, and Contexts ed. by H. J. Eikmeyer & H. Rieser. 38-74. Berlin. Lang, E. 1979. "Zum Status der Satzadverbiale". Slovo a slovesnost 40. 200-213. Lang, E. 1983. "Einstellungsausdrücke und ausgedrückte Einstellungen". Studia Grammatica 22. 305-341. Lyons, J. 1977. Semantics. Cambridge. Sgall, P., Hajičovā, E. & E. Benesova. 1973. Topic, Focus and Generat­ ive Semantics. Kronberg. Szabolcsi, A. 1981. "Composionality in focus". Folia Linguistica 15. 1-2, 141-162.

EVA HAJlČOVĀ Prague

A NOTE ON THE ORDER OF CONSTITUENTS IN RELATION TO THE PRINCIPLES OF GB THEORY

1. The order of major constituents seems still to lie beyond the main concerns of most generative gram­ marians, though in a certain sense this issue played a most important role in the split of the transformat­ ional grammar in the late sixties into interpretative and generative semantics (cf. Chomsky 1968, Lakoff 1969) and though also in the most recent development of Chomskyan government-binding theory (GB) it cannot be left out without mentioning (cf. Chomsky 1982: 8, 27f, 31, 34, 39, 93ff, 121, 128, 133, to quote only the most relevant places). The present note is intend­ ed to propose a possible account of the collection of empirical phenomena closely related to the order of constituents within the rule system and the sub­ systems of principles as advocated recently by the GB. Not being an expert in GB (the obstacles being given not only by geographical distance, but mainly by the linguistic background), I cannot give any theory internal arguments and I do not want to make any claims of contributing to the theory as such. However, having the privilege to be a disciple of Petr Sgall, who was the first to introduce a systemat-

314

EVA HAJICOVA

ic account of these issues into a formal description of language without depriving it of their empirical complexity and semantic far-reachedness (cf. e.g. Sgall 1967), I feel obliged to examine the feasibility of conciliating his insights with other prominent generative theories. 2. Taking for granted for the purpose of our present discussion that (i) universal grammar consists of interacting subsyst­ ems, which, from one point of view, are the various components of the rule system of grammar, and, from another, the subsystems of principles (Chom­ sky 1982: 5) , (ii) the components of the rule system are the base, a transformational component (transforming D-structures into S-structures), a LF-component (deriving logical forms from S-structures) and a PF-component (transforming S-structures into phonetic forms of sentences), our task can be formulated as follows: (1) (2)

(3)

to summarize first the hypotheses that we feel to be relevant for the order of constituents; to propose at which level of the general scheme of the rule system the order of constituents should be assigned; to look for a possible "division of labour" between the rule system and the subsystems of principles to account in an adequate way for the issues connected with the order of constituents.

THE ORDER OF CONSTITUENTS

315

3. A long-term empirical investigation by ling­ uists of most different orientation in general and Sgall's efforts to arrive at a systematic account of the issues under discussion within a specific framework of generative description of language (functional gener­ ative description, FGD) in particular have led to the following substantial assumptions (see Sgall, Hajičová and Panevová, in press, esp. Chapter 3 ) : (a) There is an intrinsic (basic) ordering of the types of complementations of verbs (arguments in a broader sense of the term, or 8-roles, see below) given by the grammar of a particular language, which is observed in the unmarked cases, i.e. in case the given complementation convey contextually non-bound inform­ ation (they form the so-called focus of the sentence); in the sequel, we will refer to this basic ordering by 2 the term systemic ordering (SO). (b) For each sentence there is an order of its constituents that corresponds to the degrees of commun­ icative dynamism (CD); as mentioned in (a), CD coincides with SO in the focus part of the sentence, while in the topic part(contextually bound items; roughly speaking, that part that renders what is the sentence about), CD is given by the degrees of salience of the constituents concerned and by other factors concerning the structure of the discourse; CD is semantically relevant (cf. the distinction in truth conditions between (1) and (2)) and in the surface shape of the sentence the same CD may be rendered by various means, cf. (3)(a) through (c) for the CD Mary - flower - John, which differs in its presuppositions from e.g. John gave Mary a FLOWER in

EVA HAJICOVA

316 a similar way as One of from (It was)

the

boys

a BOY (who) came

came LATE differs late.

The CD in (4)(a)

through (c) is again identical, differing from that of (5) and (6) in an analogous way. Here, as well as in the sequel, the capitals denote the bearer of the intonation centre. (1)

Everybody in this room speaks two LANGUAGES.

(2)

Two languages are spoken by everybody in this ROOM.

(3)(a) Mary was given a flower by JOHN. (b) It was JOHN who gave Mary a flower. (c) JOHN gave Mary a flower. (4)(a) Last year John came to Stanford from CAMBRIDGE. (b) Last year John came from CAMBRIDGE to Stanford. (c) John came from CAMBRIDGE to Stanford last year. (5)

Last year John came from Cambridge to STANFORD.

(6)

John came from Cambridge to Stanford last YEAR. In addition to the difference in presuppositions

it should also to be noted that these sentences differ in their potential to answer particular questions; the three sentences in

(4) all can answer the question

From where did John come to Stanford last year? , where­ as (5) in rather a full (redundant) answer to Where did come from Cambridge last year? and (6) can answer When did John come from Cambridge to Stanford? (Simil­ arly, the there sentences in (3) all can answer a question for the Actor, in contrast to John gave Mary a FLOWER, which answers a question for the Objective). (c) Every sentence splits into its topic

(convey­

ing the contextually bound information, spedifying those items that the speaker mentions to bring them

THE ORDER OF CONSTITUENTS

317

into the foreground so that the hearer may identify them in his stock of knowledge to be able to modify them) and its focus; the dichotomy of topic and focus is decisive for the assignment of the semantic scope of negation. In other words, the dichotomy is respons­ ible for some of the presuppositions of the sentence 4 (in the given reading). 4. When we try to locate the account of the order of constituents in a specific place of the rule system and the system of principles, we must emphasize one point; Since the underlying (deep) order, which may differ from the order of constituents of the phonetic form, has its consequences for the semantic interpret­ ation of the sentence, it follows that this order should be at the disposal of the LF-component; at the same time, as we have illustrated by (3), the phonemic component (now called by Chomsky the phonetic form component) may assign to a given order of constituents different phonetic forms (differing in the surface order of constituents, or in a combination of some of these). Thus the points in the rule system that are possible candidates for the representation of the order of constituents are the D-structure and the S-structure. There are two alternatives: (i) Taking into account the assumptions (a) and (b) in Sect, 3, we can say that SO should be represent­ ed in the D-structure, while CD should belong to the S-structure. Since SO is specified in terms of types of complementations of verbs, which more or less correspond to '-roles, the D-structure seems to be an appropriate place for accounting for (a). D-structures

318

EVA HAJ1C0VA

are mapped to S-structures by the rule "Move-a"; perhaps this rule could be formulated in such a way that it would rearrange the constituents according to their degrees of CD; the traces left behind the shifted constituents would not then be assigned any position in the hierarchy of CD. (ii) An alternative solution would be to let the base generate directly the order of constituents co­ inciding with the CD of the constituents of the sent­ ence generated. When characterizing the D-structures, Chomsky (1982: 39) states that it is the representat­ ion of ®-roles assignment and has also the properties that follow from X-bar theory and from parameters of the base in a particular language among which he ment­ ions also the ordering of major constituents. Under this approach, the role of the "Move-a" rule would be considerably reduced in comparison with the role this rule would play in alternative (i). As example (4) above illustrates, the rule "Move-a" should be adjusted in order to be applicable to one or more elements of the topic (contextually bound items) to transfer them to the end of the sentence. While in (3)(a) this optional rule is not applied (CD coincides with the surface word order), in (b) it is applied to to Boston, and in (c) first to to Boston and then to last year. The moved items do not acquire additional stress, so that the intonation centre is assigned (by phonemic rules) to the item that is marked as most dynamic (i.e. occupies the right most position in the D-structure). It is an open question whether with this approach also the three variants of (3) can be handled by rules deriving S-structures from

THE ORDER OF CONSTITUENTS

319

D-structures. 5. If we accept the position (ii), taking D-struct­ ures as the direct representation of CD, it remains to examine the assumptions (a) to (c) from the viewpoint of the interplay of principles, rules, and lexical conditioning. It seems possible to add to the theory the follow­ ing subsystem of principles: (A) The basic (unmarked) ordering of 0-roles (SO) can be determined by lexical means, in a way similar to what Chomsky cales "projection principle": in the lexical entries of the base component not only the 0-roles possible (optional) and necessary (oblig­ atory) as complementations of the given lexical item are specified, but also their SO. If the hypo­ thesis that SO is identical for different "frames" (subcategorization properties of different lexical items as for their @-roles) concerning the same word class is found to be plausible for the given language, then the "frames" can contain just the numerical values which can be assigned to 0-roles according to their position under SO. This approach has been formally elaborated in Hajičová and Sgall (1980) and Platek, Sgall and Sgall (1984); in the latter paper a framework is presented that covers also the main features of the interplay of the 6-roles assignment and the syntactic relations of coordination and apposition. (B) A second principle can state that whenever a D-structure contains two items A and B, where

320

EVA HAJICOVA So(A,B), but CD(B,A), then B is contextually bound, i.e. belongs to the topic (here X(Y,Z) is read "Y precedes Z under X " ) .

Recalling our examples from Sect. 3,John in (3) (a) through (c) is assigned the 8-role Actor, Mary is assigned the role Addressee; in the SO of English Actor precedes Addressee, while in the CD in (3)(a) through (c) Addressee precedes Actor. Thus, the sentences say some­ thing about the Addressee as (a part of) the topic; Mary is a contextually bound element of the sentence referring to a person who belongs (at the time point of utterance of the sentence) to the activated part of the 5 knowledge shared by the speaker and by the hearer. (C) In a similar vein, on the basis of SO and CD one can specify what is the topic and what is the focus of the given D-structure; for the tectogrammatical (underlying) representations as defined in the function­ al generative description (oriented rooted trees) this was done by Sgall (1979). The corresponding principle can be so formulated that at least one element of the subset consisting of the verb and the constituents assigned the 0-roles determined by the verb according to (A) - including the optional adverbials of all kinds - is contextually non-bound; the contextually non-bound elements of the mentioned set constitute the focus of the sentence (together with the embedded items belong­ ing to them). Thus every sentence contains a focus, whereas the presence of a topic is optional, the thetic judgements being the prototypical cases of topicless D-structures.

THE ORDER OF CONSTITUENTS

321

5. The importance of the inclusion of a descript­ ion of such phenomena as listed above into general linguistic theory can be now illustrated more explicit­ ly, e.g. on the difference in meaning of (7)(a) and (7)(b), capitals again denoting the intonation centre. (7)(a) Staff is allowed behind this COUNTER. (7)(b) STAFF is allowed behind this counter. In (7)(a) one speaks about the staff, or, perhaps more probably, about the rights of the staff, and states that (one of) the right(s) is to be behind the particular counter; more technically speaking, staff, and, on a preferred reading also the verb be­ long to the topic, whereas the adverbial constitutes 7 the focus. In contrast, (7)(b) speaks about that particular counter (topic) and states that the persons who are allowed to step there are the staff. Note that the two sentences differ (at least on their preferred readings) in the truth conditions: If I am a member of the staff, I should be behind the counter (rather than somewhere else), if I receive and accept the message expressed by (7)(a), which is not the case with (7)(b). As for the @-roles in both (a) and (b), staff is assigned the role of Objective, behind the counter the role of Location (it is not decisive for our point whether there is no Actor role assigned, or whether the role of a General Actor is assumed to be present). In the SO of English, Objective precedes Locative; (7) (a) and '(b) differ in their CD, which in (a) is in accordance with SO, while in (b) the CD is Locative Objective. Thus for (7)(b), there is only one reading, namely that having the Locative and the verb as con-

322

EVA HAJICOVA

textually bound. In (7)(a), the Locative is the last (most dynamic) item of the D-structure and thus is contextually non-bound according to the principle (C) (i.e. under the natural assumption that there must be at least one non-bound item, if the sentence is to bring some "new" information); the Objective is contextually bound, and so is the verb, in one of the two D-structures; such an ambiguity of the appurt­ enance of the verb to the topic or to the focus is a common phenomenon.

NOTES 1 A more detailed and a more general account of some of these points is presented in Hajicova and Sgall (in press) , 2 The empirical research based on a contrastive study of Slavic languages, English and German has led to the following specification of SO for the main 8-roles in English: Actor - Addressee - Objective (Patient) - Instrument - From where - Direction - Locat ive (for a detailed discussion, see already Sgall et al. , 1 9 7 3 ) . The position of the verb is not discussed here; for the purposes of this paper, the verb can be assumed to precede its G-roles under SO. 3 We leave aside for the purpose of this paper the embedded elements, which can belong to the topic even if contextually non-bound, and to the focus, even if contextually bound, cf. e.g. best and your, respecti ly in The dress I like best was made by your mother. 4 As for the placement of the boundary between topic and focus, most of our examples are ambiguous. 5 For the notion of the stock of shared know­ ledge and the hierarchy of activation of its elements, see Hajicova and Vrbova (1982) and more recently Sgall et al. (in press) . 6 Auxiliary verbs as well as prepositions and conjunctions are assigned no @-roles and no positions

THE ORDER OF CONSTITUENTS

323

under SO or CD. 7 There is also a topicless D-structure underly­ ing (7)(a), which is rather marginal, since staff, as a definite NP in the subject position, not carrying the intonation centre, is understood, with a high preference, as contextually bound ("given"). REFERENCES Chomsky, N. 1968, "Deep Structure, Surface Structure, and Semantic Interpretation". Mimeo. Printed in Steinberg & Jakobovits, eds. 183-216. Chomsky, N. 1982, Lectures on Government and Binding. Dordrecht: Holland - Cinnaminson - U.S.A. Hajicova, E. & P. Sgall. 1980. "A dependency-based specif­ ication of topic and focus". SMIL 1-2, 93-140. Hajicova, E. & P. Sgall. In press. "The ordering princip­ le", Prague Bull, of Mathematical Linguistics. Hajicova, E. & J. Vrbova. 1982. "On the role of the hier­ archy of activation in the process of natural lang­ uage understanding", Coling 82, Proc. of the Ninth Int. Conf, on Computational Linguistics ed. by J. Horecky. 197-113. Amsterdam & New York & Oxford. Lakoff, G, 1969, "On generative semantics". Mimeo. Prin­ ted in Steinberg S Jakobovits, ed. 1981. 231-296. Platek, M., Sgall, J. & P. Sgall. 1984. "A dependency base for a linguistic description". Contribution to Functional Syntax, Semantics and Language Comprehens­ ion ed, by P. Sgall. 63-97, Prague. Sgall, P. 1 9 6 7 , "Functional sentence perspective in a generative description". Prague Studies in Mathemat­ ical Linguistics 2. 203-225. Sgall, P. 1979, "Towards a definition of focus and topic". Prague Bull, of Mathematical Linguistics 31. 3-25. 1980, 3 2 . 24-32; printed in Prague Studies in Mathem­ atical Linguistics 7 (1981). 173-198. Prague. Sgall, P., Hajicova, E. & E. Benesova. 1973. Topic, Focus and Generative Semantics, Kronberg/Taunus. Sgall, P., Hajicová, E. & J, Panevová. In press. The Structure of the Sentence in Its Semantic and Prag­ matic Aspects. Dordrecht & Prague: Reidel & Academia. Steinberg, D. D. & L. A. Jakobovits, eds. 1971. Semantics. An Interdisciplinary Reader in Philosophy, Linguist­ ics and P s y c h o l o g y . Cambridge (U.K.).

V.

TEXT AND CONTEXT

ANITA STEUBE Leipzig

KONTEXT UND MÖGLICHE WELT (EINE UNTERSUCHUNG DER INDIREKTEN REDE)

o. Es ist ein besonderes Verdienst des Jubilars, die linguistische Forschung durch die Pflege interdisziplinärer Zusammenarbeit zu bereichern. Petr Sgall hat ein Urteilsvermögen dafür, wo und wie weit echte Angebote anderer Disziplinen fur die Linguistik bereitstehen und weiβ, Forderungen an Zusammenarbeit zu stellen. EinThema sei nur die Beziehung zwischen schlagig für unser Semantik und Logik genannt, der er sich in den letzten Jahren mehrfach zugewandt hat. Wir legen hier ein Mo­ dell zugrunde, das zwar den Einflu|3 der Modelltheorie erkennen läβt, aber in Ubereinstimmung mit den Forderun­ gen des Jubilars davon ausgeht, da(3 sowohl die lexikalische Besetzung als auch die verwendete Konstruktion eines Satzes für die Satzbedeutung im linguistischen Sinn grundlegend sind und daβ darüber hinaus jedes Satztoken in seiner Bindung an eine konkrete kommunikative Situa­ tion gesehen werden muβ, in. der es eine bestimmte Funktion zu erfullen hat. Dieses Modell wurde am ausführlichsten in Bierwisch 1980 erlautert. Hier genü'gt eine schematische Verdichtung mit kurzen Erklarungen.

ct,

m)

ias, cs)2

Kommunikatiyer $inn

Interaktionszusammenhang

sem),

Äuβerungsbedeutung

t, (pt, syn,

Kontext d. Âu(βerung

syntakt. Struktur

phonet. Struktur

Zeitpunkt

Person

Inskription

ins,p,

semant. Struktur, Satzbedeutung

ANITA STEUBE

328

sprachliche Struktur ls Äu(3erung u sinnvolle Äuβerung mu Sprechakt sa Sprechakt sa ins ist eine lautliche oder graphische Folge, von einem Sprecher zu einem Zeitpunkt produziert, der dieser Sprecher eine sprachliche Struktur zuordnet. Die semantische Struktur als Teil der sprachlichen Struktur - durch komplizierte Regeln über die syntaktische Struktur auf die phonologische Struktur bezogen-stellt sich als Funktion dar, die möglichen Kontexten (= Argument) die Äuβerungsbedeutung als Wert zuordnet. Damit stellt sich Bierwisch in die Tradition der intensionalen Logik, die Freges grundlegende Unterscheidung von Sinn und Bedeutung mit

KONTEXT UND MOGLICHE WELT

329

der Unterscheidung von Intension und Extension weitergefuhrt hat, was hier in dem Paar 'semantische Struktur - Äuβerungsbedeutung' wieder auftaucht. Anders als in der intensionalen Logik wird 1. die Wortbedeutung nicht als elementar angesehen, sondern in ejne Menge von Grundelementen zerlegt, die durch die Regeln einer Kategoriengrammatik strukturiert sind. Zwischen den Wortbedeutungen ist durch Bedeutungspostulate ebenfalls eine Beziehung aufgebaut. Damit sind die Erfahrungen früherer semantischer Theorien einbezogen. 2. werden die Denotatbereiche nicht einfach als mengentheoretische Strukturen vorausgesetzt. Eine semantische Struktur identifiziert mit Bezug auf einen Kontext3 eine mentale Struk­ tur eines realen Menschen, in der sich die objektive Realitat repräsentiert, wie sie der Mensch im gesellschaftlich bedingten Widerspiegelungsproze(3 erfahren hat. Die Äuβerungsbedeutung ist ihrerseits eine Funktion, die möglichen Interaktionszusammenhängen (= Argument) den kommunikativen Sinn als Wert zuordnet. Bierwisch unterstreicht aber ausdrucklich, daβ mit dem Übergang von m zu cs die Grenzen der Linguistik überschritten werden; denn zum ias gehort eine Menge institutionalisierter Regeln und Verhaltenskonventionen, die au(3erlinguistischer Natur sind, so daβ der Linguist in der Sprechakttheorie nur als ein Kooperationspartner auftre4 ten kann. Nach Frege drückt ein Aussagesatz einen Gedanken aus. Der Gedanke ist der Teil des Sinns, der ü'ber einen Sachverhalt spricht und dem damit allein die Eigenschaft zukommt, wahr oder falsch zu sein. Das Urteil selbst aber (die Anerkennung der Wahrheit des Gedankens) ist

330

ANITA STEUBE

wohl Teil des Sinns der Aussage, gehort aber nicht zum Gedanken. Diese Trennung hat Ewald Lang

(1979) in sei­

ner Untersuchung der Satzadverbiale modelltheoretisch rekonstruiert: ein Gedanke kommt in einem Satz nie rein vor, er ist immer mit einer Einstellung des Sprechers 5 zum Sachverhalt verbunden, uber den der Gedanke spricht. Der Sinn ist also immer weiter als der Gedanke .

Man

kann z.B. den gleichen Gedanken als Aussage, Frage oder Befehl for5mulieren, man kann den gleichen Begriff (z.B. Personenkraftwagen) in einem Satz sprachlich neutral als 'Auto' realisieren, aber unter Beibehaltung des Wahrheitswertes abwertend auch als 'Karre' oder aufwertend als 'Wagen' bzw. in f achsprachlicher Ausdrucksweise als "Personenkraftwagen, PKW". Was zum Sinn gehort, aber uber den Gedanken ■ (die Proposition) hinausgeht und so den Wahrheitswert nicht beeinflu(3t, wird von Frege als Art des Gegebenseins des Gedankens bezeichnet. Ob nichtpropositionale Bedeutung ein einheitliches Phanomen ist, ist offen. Die indirekte Rede mit daβ-Satz ist in mehrfacher Hinsicht ein fur die semantische Analyse geeignetes heuristisches Mittel. Wenn das Zitat mittels daβ-Satz nur den Gedanken aus der ursprünglichen Äuβerung aufnehmen kann - Frege sagt, der Nebensatz sei als Eigenname jenes 7 Gedankens aufzufassen - so läβt sich aus jeder Äuβerung durch Umformen in einem daβ-Satz der

Teil der Äuβerungs-

bedeutung herausfiltern, der zum Gedanken gehört8. la) Ich habe Hunger. lb) X s a g t , daβ sie Hunger hat. Aber ich g l a u b e , daβ sie nur eine Ausrede braucht, weil sie sich fur solche S-portveranstaltungen nicht interessiert.

KONTEXT UND MOGLICHE WELT

331

Fur die Bedeutung der Reporterauβerung ist es egal, ob der zitierte Gedanke wahr oder falsch ist. Das Urteil gehort, wie oben bereits behauptet, nicht zum Gedanken. Da auβerdem jede Äuβerung (sowohl die ursprungliche wie die Reporteräuβerung) in eine kommunikative Situation eingebettet und damit kontextabhangig ist, zeigt sich beim Zitieren mit daβ-Satz die Kontextabhangigkeit der Rede besonders deutlich. Die Umformung zum indirekten Zitat dient damit gleichzeitig als heuristisches Mittel 9 festzustellen, was alles als Kontext gefaβt werden kann. 1. Wir wollen uns nun als erstes die Kontextbildung und Interpretation direktor Zitate ansehen. la) ist dann eine behauptende 0riginaläuβerung, wenn fur einen zeitlichen Referenzpunkt ein Sprecher identifiziert ist, fur den gilt, da(3 er dann Hunger hat. Im direkten Zitat lc) X s a g t e , ieh habe Hunger. zum Beispiel auf die Frage Id) Was hat Inge denn gestern als

erstes

gesagt,

als

sie

vom Training kam? andert sich die ursprungliche Kontextbindung nicht ( i c h , habe). Fur die Interpretation ist jedoch die Kenntnis des Kontextes der Originalaufterung (mindestens: Sprecher, zeitlicher Referenzpunkt) notwendig. Die Originalau$erung 2a) 2a)

Hunger

habe

ich.

ist von der 0riginaläuβerung la) sinnverschieden. Sie kann i) bedeuten, daβ der Sprecher zum Ref erenzpunkt groβen Hunger hat oder ii) - mit Fokus-Topikalisierung -

332

ANITA STEUBE

daβ der Vorgängerkontext oder das Sachwissen fur die Kommunikationsteilnehmer des Originalsprechersi eine Klasse relevanter Eigenschaften prasupponiert und da zum Referenzpunkt behauptet wird, daβ die Eigenschaft für den Sprecheri zutrifft, Hunger zu haben. In unkenntnis des Kontextes dieser Originaläuβerung liegt für die Kommunikationspartner des Reporters mit 2b) eine mehrdeutige Äuβerung vor. 2b)

X s a g t e , Hunger

habe

ich.

Sie kann mittels Zusatzfragen disambiguiert werden. Der Reporter kann aber auch Kontext nachliefern und so von vornherein fiir die Disambiguierung Sorge tragen. la) kann als Behauptung gemeint sein, wie oben behandelt, es kann aber auch z.B. der Wunsch gemeint sein, die Sportveranstaltung zu verlassen und ins Restaurant zu gehen oder z.B. der Wunsch, da(3 das Essen bald vorbereitet wird usw. Der kommunikative Sinn hat sich dann betrachtlich geandert und ist auf der Grundlage von la) nur aus dem Interaktionszusammenhang zu schluβfolgern. Das direkte Zitat lc) verpflichtet den Reporter, diesen Interaktionszusammenhang zu erlautern, urn den Konversationsimplikaturen zu genügen, Relevantes zu sagen und seinen Kommunikationspartnern die gleichen Schluβfolgerungen zu ermöglichen. Sonst kann es dazu kommen, daβ mit aus dem Zusammenhang gerissenen Zitaten anderes oder sogar das Gegenteil von dem gesagt wird, was mit der Originaläuβerung behauptet wurde. Die wenigen Beispiele genugen urn zu zeigen, daβ sich das direkte Zitat auf die gesamte Äuβerungsbedeutung (Sprechereinstellung und propositionaler Gehalt) bezieht und da3 den Kommunikationspartnern des Reporters

KONTEXT UND MOGLICHE WELT

333

der zur Originalau(3erung gehorende Kontext und Interaktionszusammenhang zuganglich sein miissen. Ewald Lang meint, da(3 sich das direkte Zitat auf die Satzbedeutung bezieht, da mit 3a)

X sagt:

ich

bin mu.de

i) sowohl eine Behauptung uber den Originalsprecher als auch ii) ein grammatischer Beispielsatz, den der Origi­ nalsprecher gebildet hat, zitiert werden kann. Wir wollen dem weiter nachgehen. 3a) kann iii) ebenso als Originalau(3erung Teil einer Rolle sein, die ein Schauspieler auf der Biihne gesprochen hat. Mit "ich' wird dann zum Referenzpunkt die Person identifiziert, fur die in der fiktiven Welt die behauptete Eigenschaft gilt. Das direkte Zitat ist in alien Fallen moglich, aber der Re­ porter kann nicht ungeklart lassen, da(3 es sich im zweiten Fall urn eine metasprachliche Verwendung, im dritten Fall urn die Bindung an eine fiktive Welt handelt. Das Zitat wiirde sonst vollig falschen Schlu(3folgerungen iiber ii) z.B. ungebuhrliches Verhalten in der Schule, iii) z.B. einen Gag auf der Biihne Vorschub leisten. Ein Unterschied zum indirekten Zitat mit da(3-Satz ist, da3 nur direkte Zitate von metasprachlich verwendeten Au(3erungen und von solchen, die in anderen moglichen Welten ihren Wahrheitswert haben, iiberhaupt moglich sind. Ihre Kon­ text- und Weltbindung ist also nicht fur die sprachliche Gestaltung, wohl aber fur die Interpretation zu beriicksichtigen: 3b)

X sagte mit

3c)

erstem

X sagte mit

auf auf

erstem

die

Frage

nach

Personalpronomen: die

Frage

nach

Personalpronomen,

einem Ich einem

Satz

des

bin

miide.

Satz

da$ sie

des miide

Deutschen Deutschen sei.

ANITA STEUBE

334

Das indirekte Zitat fallt - trotz der Situationsangabe auf die hier unangemessene objektsprachliche Ebene zuriick. 3d)

X sagte mit

auf die

erstem

deshalb

Frage nach einem Satz

Personalpronomen^

keine

Antwovt

da

sie

des

Deutschen

milde sei und

wisse.

3d) liegt eine 0riginaläuβerung mit Ich

bin müde

in

objektsprachlicher Verwendung zugrunde. Davon ist ein indirektes Zitat moglich. Diese Satze verdanke ich einem Hinweis von Peter Lutzeier. 4)

Was hat Peter

4a)

Er hat gesungen:

Schreier

gerade

4b)

Er hat gesungen, daβ sein

gesungen?

Mein Mädel hat

einen

Rosenmund.

Mädel einen Rosenmund

hat. Das indirekte Zitat müβte sich in der

sprachlichen

Gestaltung dem neuen Kontext anpassen ( s e i n M a d e l ) . E i ­ nem indirekten Zitat ware aber ein Lacherfolg sicher, denn Peter Schreier hat nur vorgetragen, keine Behauptung über sich

gemacht.

Man konnte aber fragen: 5)

Was hat Faust gerade gesagt? Faust

und nicht der Schauspieler XY -

und als Antwort bekommen: 5a)

Er hat gesagt, daβ er nun ach Philosophie, Juristerei und Medizin und leider auch Theologie durchaus studiert hat mit heiβem Bemühn.

Frege sagt, daβ der Schauspieler nur aussagt, er behauptet nicht. In der Auffassung der Semantik der möglichen Welten

KONTEXT UND MOGLICHE WELT

335

wäre zu präzisieren: der Schauspieler als Teil der realen Welt sagt nur aus (interpretiert nur), aber in der fiktiven Welt (als die interpretierte literarische Gestalt) behauptet er. Mit ausdrücklichen Verweis auf die fiktive Welt kann deshalb auch mit indirektem Zitat zitiert werden. Also behalt das direkte Zitat seine Kontext- und Weltbindung bei (die der Reporter seinen Kommunikationspartnern offenlegen muβ, das indirekte Zitat hat aber eine doppelte Kontextbindung. Die Weltbindung indirekter Zitate wird in Punkt 5 abschlieβend besprochen. Kutschera weist schlie(31ich noch auf die sprachliche Bindung direkten Zitierens hin: 6a)

Galilei

sagte:

"'Die

Evde bewegt

sich 1 1 ,

ist nicht wahr, denn Galilei benutzte nicht die deutsche Sprache. Es ist aber wahr, daβ 12 6b) Galilei s a g t e , daβ sich die Erde bewegt. So konnen wir schlieβ1ich schluβfolgern, daβ sich das direkte Zitat auf die Äuβerung mit ihrer gesamten Äu(3erungsbedeutung bezieht, daβ das indirekte Zitat mit daβ-Satz dagegen, wie schon ausgeführt, nur auf den propositionalen Gehalt der Äuβerungsbedeutung bezogen ist. 2. Wir kommen nun ausführlicher zum indirekten Zitat in daβ-Sätzen. In Steube (1983) ist ausführlicher auf die Deixis und auf Sprechereinstellungen eingegangen worden. Wir wollen deshalb hier nur Anreden und Referenzbeziehungen und Konnotationen behandeln und eine Reihe weiterer Erscheinungen in Tabellen zusammenstellen.

336

ANITA STEUBE

2.1. Anreden sind in indirekter Rede generell nicht möglich, weil vom Reporter nicht zu der durch die Originaläuβerung angesprochenen Person, sondern iiber sie gesprochen wird. Wie iiber die ehemals angesprochene Per­ son gesprochen werden kann, ist unterschiedlich. 7a) Herr Vorsitzender, die Punkte 5. u und 6. der Tagesordnung sind auf der gestrigen Vorbereitungssitzung gestrichen worden. 7b) X sagte zum Vorsitzenden, daβ . . . 'Herr' ist auf die Anrede beschränkt, das ist eine Konvention der deutschen Sprache. Vorsitzender zu sein, ist eine uberprüfbare Eigenschaft, die auch weithin gewuβt wird. Der Reporter kann die Bezeichnung benutzen, um damit fur seine Kommunikationspartner die angespro­ chene Person zu identifizieren, um dann iiber sie sprechen zu konnen. Falls der ursprünglich Angesprochene jedoch nicht Vorsitzender ist, sondern vom Originalsprecher nur so genannt wird, diirfte das Zitat erneut von der 0riginaläuβerung abweichen müssen. 7c)

X sagte

zu Z, den er mit

te,

...

daβ

Herr

Vorsitzender

anrede-

Die beziiglich Sprechereinstellungen neutralste Wiedergabe der 0riginaläuβerung wird erzielt, wenn der Re­ porter fur Z den Namen der urspriinglich angeredeten Per­ son setzen kann. Ist ihm der Name nicht bekannt oder seine Kommunikationspartner konnen damit nicht den referentiellen Bezug zur ursprunglich angeredeten Person herstellen, müβte eine Kennzeichnung fur Z eingesetzt wer­ den, die fur den Reporter und seine Kommunikationspart-

KONTEXT UND MOGLICHE WELT

337

ner den richtigen referentiellen Bezug herstellt, z.B. 7d)

X sagte zu dem ihm gegenüber sitzenden er mit Herr Vorsitzender anredete, daβ

Eerrn, ...

den

Das bedeutet nun nicht, da3 die Einsetzung einer Kennzei­ chnung von vornherein eine Losung brächte. Mit den Kennzeichnungen sind selbst auch schon semantische Probleme verbunden. Nach Stalnaker und Strawson13 ist die Referenz von Ausdrucken etwas anderes als ihre Bedeutung 14 ; die Referenz ist Teil der Ausdrucksverwendung. Sprecher verwenden Kennzeichnungen referentiell, urn die Kommunikationsteilnehmer wissen zu lassen, über welches Objekt gesprochen wird. Über dieses so kenntlich gemachte Ob­ jekt kann dann eine Aussage gemacht werden. Der Sprecher kann aber zu diesem Zweck keine beliebige Kennzeichnung verwenden. Er ist abhangig vom Wissen und von den Erwartungen der Kommunikationspartner in der Auswahl der Eigenschaften des Objekts, die er mittels seiner Kennzeich­ nung benennt, urn damit das gegebene Objekt fur die Kom­ munikationspartner als Einzelding auszusondern: 8)

(Ein Ehepaar hat seine Kinder zum ersten Mal mit

8a)

ins Konzert genommen. Der Vater sagt zum Sohn:) Wenn der erste Ge5wandhaus-Kapellmeister kommt, beginnt

das

Konzert.

Die Tochter sitzt neben der Mutter. Sie konnte die Äuβerung des Vaters akustisch ni cht verstehen. Da die Mutter auβerdem weiβ, der Kleinen ist nicht gelaufig, daβ der erste Kapellmeister in dem Fall der Diregent ist, muβ sie im indirekten Zitat eine andere Kennzeich­ nung auswahlen. Die 0riginaläuβerung kann bei Auswahl einer anderen Kennzeichnung selbstverstandlich nicht sinngleich wieder-

ANITA STEUBE

338

gegeben werden. (Vgl. 8a) mit 8b) und 8c) im Abschn. 5.) Anders ist es beim Austausch von Eigennamen. Nehmen wir als Beispiel das Zitat 26) aus der Tabelle mit Mark Twain alias Samuel Langhorne Clemens. Diese Namen sind sinngleich, wenn jeder fur jemanden nichts anderes ausdrückt als "derjenige, der die vorliegenden Bücher (vgl. die Situation in Beispiel 26)) geschrieben hat". Weiβ das Kind jedoch nicht, daβ Samuel Langhorne Clemens der burgerliche Name und Mark Twain nur ein Pseudonym dafur war, ist die gemeinte Person von ihm mit Samuel Lang­ horne Clemens nicht identifizierbar und kommt in dem Kontext als Name nicht in Betracht. Nehmen wir nun noch ein Beispiel mit einer ganz auf individueller Haltung und Einstellung beruhenden Anrede.

9a)

Liebling_, ich rufe

dich morgen wieder

an.

Der Reporter kann mit dieser Bezeichnung nicht iiber die angesprochene Person sprechen, weil i) die Kommunikationspartner nicht zu wissen brauchen, wer der Liebling des Originalprechers war oder ist; ii) die Anrede in der Originaläuβerung u.U. nicht wortlich, sondern sogar ironisch gemeint war, was an Sinn verlorenginge, wenn mit dieser Bezeichnung iiber die urspriinglich angeredete Person gesprochen würde; iii) der dritte und wichtigste Grund aber ergibt sich aus der Funktion von Kennzeichnungen bei referentieller Verwendung (und nur diese kommt hier in Frage) . Es wird prasupponiert, daβ der Sprecher glaubt und auch die Kommunikationsteilnehmer glauben machen will, daβ ein Objekt existiert, das diese Eigenschaften erfüllt - nämlich genau dasjenige Objekt, iiber das er redet. Es wird also als wahr vorausgesetzt, daβ eine identifizierbare Person die Eigenschaft

KONTEXT UND MOGLICHE WELT

339

aufweist, Liebling des Originalsprechers zu sein. Diese Voraussetzung ist überhaupt die einzige Basis fur die Referenz: mittels der ausgewählten Eigenschaft, die auf ein Objekt zutrifft, wird das Objekt ausgesondert. (Wird diese Voraussetzung fallen gelassen, muβ eine Ersatzvereinbarung der Kommunikationspartner dafiir in Kraft treten, etwa: statt 'der Lange' wollen wir an alien Stellen des Vorkommens "der Niedliche" sagen.) Diese Voraussetz­ ung ist aber in der originalau$erung nicht enthalten. Aus ihr folgt nicht, daβ es fur die Angeredete zutrifft, da sie der Liebling des Originalsprechers ist. Wann man in diesem subjektiven Bereich jemanden wie anredet, unterliegt auβerlinguistischen Normen, die nur moralisch einklagbar sind. Anreden konnen auch zur Konvention wer­ den oder man kann den Wunsch äuβern, auf bestimmte Art angeredet zu werden. In gesselschaftlich relevanten Bereichen und besonders im institutionellen Verkehr sind die Anreden normiert. Unter dem hier interessierenden linguistischen Gesichtspunkt bleiben sie alle Namen und wirken nur als Signal fur den Angesprochenen, daβ sich der Sprecher mit der Äuβerung an ihn richtet. Der Angesprochene mu(3 den Namen durch Kenntnis des Namensgebungsaktes, der Konvention oder Festlegung auf sich beziehen konnen. Aus der Äuβerungsbedeutung ergeben sich keine Prasuppositionen iiber das Zutreffen von Eigenschaften, die in der Anrede verbalisiert sind, auf die angeredete Person. Werden also Kennzeichnungen zur Anrede verwendet, sind damit die einschlagigen Prasuppositionen nicht ver­ bunden. Werden die Kennzeichnungen in der indirekten Re­ de - wo Anreden nicht möglich sind - in die Redeankündigung aufgenommen und referentiell verwendet, sind die Prasuppositionen damit verbunden. D.h. der Reporter darf

340

ANITA STEUBE

die Kennzeichnung nur dann referentiell einsetzen, wenn er die dann gemachten Prasuppositionen teilt: es gibt eine Person, die die in der Kennzeichnung verbalisierten Eigenschaften aufweist und zwar genau diejenige Per­ son, über die er sprechen will. Weiβ der Reporter, da(3 die Eigenschaf ten nicht zutreffen bzw. weiβ er nicht, ob die Eigenschaften zutreffen, weil sie - wie im diskutierten Fall - der Originalsprecher auf der Grundlage seiner Haltung und Einstellung zur angesprochenen Per­ son ausgewahlt hat und weil diese Haltung und Einstellungen dem Reporter nicht zuganglich sind, müβte er zuriickhaltened zitieren, etwa 9b)

X sagte

zu der

Frau

am

Telefon,

daβ

... Er

nannte

sie Liebling. Wäre die Kennzeichnung im Zitat attributiv verwendet, ware die Voraussetzung auch vorhanden, daβ eine Person existiert, die die in der Kennzeichnung verbalisierten Eigenschaften aufweist. Das konnte dann aber irgendeine Person sein. Es wird keine bestimmte Person ausgewählt, urn genau iiber sie zu sprechen. Wenn also der Reporter nicht weiβ oder den Kommunikationspartnern auf Grund ihrer Kenntnisse nicht verstandlich machen kann, wen der Originalsprecher angeredet hat, muβ er die Kenn­ zeichnung attributiv verwenden. 9c)

X sagte te

sie

zu evner

Frau

am

Telefon,

daβ

... Er

nann­

Liebling

Mit attributiver Verwendung der Kennzeichnung ware die Originaläuβerung nur defekt wiedergegeben, denn Anreden richten sich an ganz bestimmte Personen. Durch die vorhandene Prasupposition bleiben aber auch in attributi­ ver Verwendung diejenigen Kennzeichnungen, die subjektiv

KONTEXT UND MOGLICHE WELT

341

empfundene Eigenschaften verbalisieren, inadaquat. Zitat 9d) weicht dann in doppelter Hinsicht von der Originaläuβerung ab. 9d)

X sagte

zu seinem

Liebling

daβ...

(attributiv),

2.2. Jetzt wollen wir noch einen Exkurs zu Konnotationen machen. Die Heterogenitat des sprachlichen Systems beschreibt die Lexikologie meist so, daβ jegliche Markierung sprachlicher Einheiten im Wörterbuch unter dem Begriff 'Konnotation' gefaβt wird. Neben denotativer spricht man dann auch von konnotativer Bedeutung. Das Kriterium der Zitierbarkeit mittels indirekter Rede erlaubt, hier eine interne Differenzierung vorzunehmen. territoriale Variante 10a) Mutter hat mir zwei Pakete Stullen mitgegeben. zwei Pakete Schnitten 10b) X s a g t , daβ ihr ihre Mutter mitgegeben hat. funktionalstilistische Variante 11a) An den Schaltern zeichen statt. l1b) Das Sohild sagt Briefmarken

findet aus,

verkauft

ist

12b) Unser

daβ

se

sagt,

daβ

Verkauf

von

an den Schaltern

Postwertkeine

werden.

soziale Variante 12a) Die neue Lehrerin Junge

kein

urst. die

neue

Lehrerin

groβe

Klas-

ist.

Wenn der Reporter nicht zur gleichen territorialen oder sozialen Sprechergruppe gehort wie der Originalsprecher, benutzt er nicht die so konnotierten Äuβerun-

342

ANITA STEUBE

gen. Ebenso kann bzw. muβ er im Funktionalstil abweichen, wenn er als Individuum (statt der Institution Post wie in 11a)) spricht bzw. wenn die Kommunikationssituation als ganze oder der Zweck der Äuβerung wechseln. Umgekehrt kann aber auch die Originaläuβerung neutral und die Reporteräuβerung konnotiert sein. Was ist nun die neutrale gegenüber der konnotierten Variante semantisch? Kier sind die Meinungen polarisiert gewesen, auf der einen Seite sprach man von unterschiedlicher konnotativer Bedeutung , auf der anderen davon, da(3 kein Bedeutungsunterschied vorliege . In diesem Streit konnte sich jede Seite ins Recht setzen, und der Widerspruch blieb doch bestehen, was beweist, daβ die theoretische Argumentationsbasis nicht weit genug ausgearbeitet war. Die hier verwendeten Grundbegriffe sind tragfahig genug, den Widerspruch aufzulosen. Wir sprechen nicht mehr schlechthin von Bedeutung, sondern losen unter Berücksichtigung der Fregeschen Trennung von Sinn und Bedeutung die semantische Struktur auf in propositionale- und nicht-propositionale Bedeutung. Die propositionale ist allein das, was als wahr oder falsch beurteilt werden kann, was einen Sachverhalt (in einem bestimmten Kontext) identifiziert. Wenn die Semantik, ausgerustet mit dem Instrumentarium der Erkenntnistheorie, die Bedeutung als intersubjektives Abbild gekennzeichnet hat, das an einem sprachlichen Zeichenkorper geknüpft ist, dann hat sie damit, bei aller Bewu3theit der subjektiven Seite der Widerspiegelung, Bedeutung doch meist auf den Fregenschen Gedanken, sprich propositionalen Gehalt reduziert. In diesem eingeschrankten Verstandnis von Bedeutung liegt bei keiner der drei Varianten im Zitat (vgl. 10b) bis 12b)) eine Bedeutungsveranderung

KONTEXT UND MOGLICHE WELT

343

gegenüber dem Original vor. Der identif izierte Sachverhalt ist der gleiche. Zwischen 'Stulle' und 'Schnitte', Postwertzeichen' und 'Briefmarke', 'urst' und "groβe Klasse' besteht dann zunächst erst einmal nur der Unterschied, daβ je zwei unterschiedliche Zeichenkörner die gleiche Bedeutung (propositionaler Gehalt) repräsentieren. Die Zitierweise in indirekter Rede zeigt aber, daβ damit die Erklarung noch nicht abgeschlossen ist. Die Sprachgemeinschaft zerfallt nach territorialen, sozialen und funktionalen Gesichtspunkten in kleinere Sprachgemeinschaften, die sich vielfach iiberschneiden konnen. In diesen Sprachgemeinschaften besteht ein Solidarisierungsbestreben, und die jeweiligen Kommunikationspartner erwarten, daβ sich der jeweilige Sprecher konform ausdrückt: dem Funktionalstil, der Stilebene, der von ihnen sozial und territorial bevorzugten Variante usw. entsprechend. Auβerdem spielt das soziale Verhältnis zwischen dem Sprecher und seinen Partnern eine Rolle, inwieweit - bei heterogenen Gruppen - der Sprecher auf die von den Horern bevorzugte Variante eingehen muβ. In jeder Kommunikationsgemeinschaft besteht also ein Erwartungswert bezüglich der Zeichenkorper (das sind morphologische, phonologisch-phonetische, syntaktische und stilistische Varianten) als Vehikel fur die gedankliche Repräsentation von Sachverhalten, oder der Erwar­ tungswert baut sich im Laufe der Kommunikation auf oder aus. 10c)

X s a g t , daβ ihv ihve mitgegeben hat.

Mutter

zwei

5Bikete

Stullen

344

ANITA STEUBE

10c) wiirde den Erwartungswert, den Nichtberliner oder solche, die sich als Nichtberliner fiihlen, hinsichtlich territorialer sprachlicher Ausdrucksmittel haben, nicht treffen. Es ware nicht ihre Variante. Wenn ein Reporter sie dennoch benutzte, könnte es als Nachahmung des Originalsprechers oder sogar als Ironie gedeutet werden. Am starksten wiirde die Nachahmung phonologisch-phonetischer Mittel einer fremden Variante empfunden, weil es den allermeisten Menschen sehr schwer fallt, ihre Artikulation umzustellen und hier der geringste Erwar­ tungswert beziiglich des 'Umsteigens' auf eine fremde Variante besteht. Der Reporter zitiert also in der Va­ riante seiner Kommunikationsgemeinschaft. Der semantische Unterschied zur 0riginaläuβerung ist ein Sinnunterschied. Urn mit Frege zu sprechen, hat sich die Art des Gegebenseins des Gedankens geandert. Wir konnten es uns also einfach machen und sagen, 'konnotative Bedeutung' weise einen Sinnunterschied zu nichtmarkierten Verwendungsweisen auf. Hier zeigt sich aber,' daβ der Begriff der Konnotation mehr für lexikographische als fur semantische Zwecke geeignet ist, er ist semantisch uneinheitlich. Erstens besteht der Sin­ nunterschied nicht immer darin, da(3 unterschiedliche Erwartungswerte beziiglich sprachlicher Varianten ausgedrückt werden. Oben waren mit 'Karre' bzw. 'Wagen' als Sprechereinstellungen auch schon negative und po­ sitive Bewertungen angezeigt worden. Zweitens gehoren zur sozialen Varianz auch ideologietragende Lexikonbestandteile. Sie sind in Zitaten oft nicht austauschbar: 13a)

Die H o c h r ü s t u n g verschlingt das ka eigentl-tch füv seine sozialen

Geld, das Belange

Amevibvauchte.

KONTEXT UND MOGLICHE WELT

13b)

X s a g t , die N achrüstung

verschlingt

345 das

Geld,

das Amerika eigentlich für seine sozialen Belange brauoht. Das Zitat wiirde natürlich den Reporter sofort als Anhanger westlicher Riistungspropaganda ausweisen. Dariiber hinaus hatte er aber den propositionalen Gehalt der 0riginaläuβerung geändert, indem er behauptet, die amerikanische Rüstung sei Riistung im nachhinein. Ebenso ist die Bedeutung (propositionaler Gehalt) von "Kapitalist" anders als die von "Arbeitgeber', die Bedeu­ tung von "Demokratie' im biirgerlichen Sprachgebrauch anders als die von 'Demokratie" im sozialistischen Sprachgebrauch. Bei diesen ideologietragenden Äuβerungen besteht nicht nur die oben besprochene Erwartung hinsichtlich gruppenkonformer Ausdrucksweise; es sind Termini aus zwei auf entgegengesetzter Ideologie aufgebauten Begriffssystemen, von denen jeder Terminus durch die entsprechende Ideologie gesteuerte subjektive Widerspiegelung in so hohem Ma(3 enthalt, da3 ein solches Begriffspaar keinesfalls identische propositionale Bedeutung umfa(3t und deshalt nicht so ausgetauscht werden kann wie 'Schnitte - Stulle - Bemme' usw. Deshalb genügt es auch nicht, in ein Worterbuch einzutragen 'Arbeitgeber bürgerlich'. Die Markierung 'bürgerlich' leistet nämlich nur, was eine Markierung 'berlinisch' oder eine Markierung 'familiär' leistet: Zuordnung des Lexems zu einer Variante. Sie leistet aber keinesfalls Auskunft darüber zu geben, was der biirgerliche Sprecher mit 'Arbeitgeber' anderes ausdriickt als im sozialistischen Sprachgebrauch mit "Kapitalist" gesagt wird. Die Markierung sagt allenfalls: du findest die Definition biirgerlichen philosophischen

346

ANITA STEUBE

oder ökonomischen Wörterbuch. Die dort verzeichnete Definition baut auf einer Vielzahl anderer Begriffe aus dem gleichen ideologischen Bereich auf, ist in dieses Gedankengebaude eingebunden und geht selbst in andere Definitionen ein. Ideologietragender Wortschatz sollte also von sozialer Konnotation der oben besprochenen Art (alters- oder berufsbedingte Sprache) getrennt werden. Es ist immer erst zu prüfen, ob durch den Austausch solcher Worter nicht eine Bedeutungsveranderung vorgenommen würde. Auf diesem Hintergrund zeigt sich aber auch, daβ in den Funktionalstilen mit den Termini der Wissenschaften ebenfalls vorsichtiger umgegangen werden muβ: Wenn in einem Syntaxkonzept z.B. Zahlworter zur Kategorie des Artikels rechnen, kann in einem Zitat nicht ohne Explikation 'Artikel' verwendet werden, wenn die neue Kommunikationsgemeinschaft mit groβer Wahrscheinlichkeit gedanklich von einem Syntaxkonzept ausgeht, das unter Artikel nur den bestimmten und unbestimmten Artikel versteht. Hier differieren die Denotate im Urnfang. Form und Funktion waren verwischt, wiirde man aus einer Transformationsgrammatik chomskyscher Pragung zitieren Jeder Satz besteht aus einer Nominalphrase und einev Vevbalphvase. x 14b) Jeder Satz besteht aus einem Subjekt und einem Prädikatj sagt X. x 14c) X sagtj daβ jeder Satz aus einem Subjekt und einem Pradikat besteht. Direkte und indirekte Zitate stimmen darin überein, 14a)

da(3 bei Veranderung des propositionalen Gehalts kein faires Zitat entsteht.

KONTEXT UND MOGLICHE WELT

347

Natürlich kann nicht generalisiert werden, daβ Termini schlechthin bedeutungsverschieden seien, was schon die Reihe: 'Satzradikal, propositionale Bedeu­ tung, propositionaler Gehalt, Gedanke' zeigt, wo nur Sinnunterschiede bestehen. Es ist nur Vorsicht geboten. Ist der Austausch von Konnotationen mit einer Anderung der propositionalen Bedeutung verbunden, hat das Zitat einen anderen Wahrheitswert als die Originaläuβerung. Die Sinnanderung im indirekten Zitat wird aber durch die Kontextanpassung der indirekten Rede gerechtfertigt. Fehlende Kontextanpassung würde den Erwartungswert hinsichtlich kommunikativ adaquater Sprachverwendung verletzen. 2.3. In tabellarischer Zusammenstellung soll nun auf eine Reihe weiterer sprachlicher Fakten eingegangen werden, die in indirekter Rede mit daβ-Sätzen semantische Veranderungen hervorrufen. Zuerst (Tabelle 1) folgen die Beispiele a) als 0riginalau(3erung und b) als Reporteräuβerung. Der fur die Interpretation notwendige Kontext wird in Klammern angefugt. Die kontextbedingten sprachlichen Veranderungen sind unterstrichen. In Tabelle 2 werden fur die gleichen Beispiele die Kontextfaktoren genannt, was nicht-propositionale Bedeutung (Np) ist und die Angabe, ob sich die pro­ positionale Bedeutung verandert hat (+) oder ob sie gleich bleibt (-) . Anschlieβend werden die Tabellen ausgewertet.

ANITA STEUBE

348 TABELLE 1 : 15a) Ioh treffe

dioh übermorgen nooh einmal an

dieser

Stelle,

15b) X sagte,

daβ

nächsten Stelle

er sie/ihn/dioh/mioh

Tag noch einmal an

am iiber-

dieser/jener

treffe/trafe/treffen

würde.

Deixis

16a) Woher kommst du? 16b) X fragte 17a) Laβ dir

y,

Frage

woher er

die Haare sohneiden!_

17b) X bat Y, daβ er sich lassen 18a) Was für

käme. Befehl, Bitte

die Eaare

schneiden

solle. ein

Unsinn!_

Ausruf

18b) X sagte , daβ das Unsinn sei._ 19a) (Vater konnte

eine

Vertretung

organisieren.

Also werden wir morgen endlioh rien

in die

fahren.

)

Fe-

Kommentar

19b) (Die Kinder wissen, daβ wieder zu Groβmutter gefahren wird. Sie lebt allein, ist Kinder nicht gewöhnt und erzieht standig an ihnen herum. Die Eltern sind meist damit beschaftig zu reparieren und zu wirtschaften.) (Mutter sagte, daβ Vater eine organisieren

konnte.)

morgen_ also

in die

Vertretung

Sie meinge, daβ wir Ferien

fahren.

20a) (Ein Herr zu einer Gruppe von Jugendlichen in der Straβenbahn:) Jetzt

haben

ganz sohb'n getreten. 20b)

Sie

mich

Konnotation

Was hat der Mann zum Langen gesagt? ..., daβ er ihn tüchtig

gelatscht

21a) Ioh werde den Kollegen vorführen.

hat.

meinen neuen Wagen Wertung

KONTEXT UND MOGLICHE WELT

21b) X sagte, daβ er den Kollegen Auto vorführen lich

349

sein

wird - er sprach

neues eigent-

'wlagen ".

von seinem

22a) Bin ich mude!

Wertung

22b) X sagte, daβ er sehr mude 23a) Du erzahlst

einen

sei.

Quádtsch!

23b) X sagte, daβ Y völligen

Wertung

Quatsch

erzahlte.

24a) (Zwei Bekannte im Restaurant:) Ich mochte wieder

einmal

einen

(In der letzten

frischen

Zeit

Süβwasserfisch

bin ich auf meinen

mehrmals ganz schö'n

gern essen.

Reisen

reingefallen.) Topik-Fokus-Gliederg

24b) (Einer der Gaste zum Kellner:) Ein x

bitte.

Und

mal einen

der Herr frischen

Filetsteak

er mochte wieder

Süβwasserfisch

Mein Herr, wir sind

(Kellner: tes

sagt,

ein­

essen.

ein gut

gefiihr-

Restaurant.)

25a) (Ein Herr zu einer Gruppe von Jugendlichen, die in der Straβenbahn vor dem Entwerter stehen:) Könnten Sie

ein

Stuckchen

rücken?

beiseite

indir. Sprechakt

25b) Der Mann hat gerade gesagt, du sollst ein Stückchen beiseite rucken. 26a) (Groβvater löst seine Bibliothek auf. Er schickt die Biicher mit einem Begleitbrief an seinen Sohn.)

Samuel Langhorne Clemens habe ich

meiner Jugend verschlungen. sollten

den Jungens gleich

26b) Groβvater

schreibt,

daβ

gehoren.

hat.

diese

gehoren

Biicher gleich

Er meint,

fahren.

daβ

seiner

euch

sollten.

27a) Kinder, morgen werden wir endlich rien

Referenz

er Mark Twain in

Jugend verschlungen

in

Diese Biicher

in die

Fe-

Anrede

ANITA STEUBE

350 27b) Die Mutter

sagte

am folgenden 28a) Nun beruhigen 28b)

X sagte X sagte

zu ihren

Tag in die Sie

sich

Kindern,

Ferien doch,

daβ

fahren.

Kindchen!

Anrede

zu dem Kindchen, daβ ... zu der jugen

chen, daβ sie 29a) Indirekte

sich

Zitate

Frau,

er nannte

beruhigen

andern die

29b) Der Autor meint,

daβ

Kind­

SprechereinstelTermini

indirekte

Zitate

onalen Typ von Fragesatzen

30a) (Standardtheorie:) Jeder ner Nominalphrase

sie

solle.

lung von Fragesatzen. positi

sie

Satz

und einer

den pro­

andern.

besteht

aus 0

ei-

Verbalphrase. Termini

30b)

Chomsky sagt, daβ jeder und Präadikat

Satz

aus

Subjekt

besteht.

31a) Die Messestadt ist schon gut auf das Sportfest vorbereitet. 31b) Die Leipziger Volkszeitung schreibt, daβ die Bezirksstadt schon gut auf das Sportfest vorberei­ tet ist.

KONTEXT UND MOGLICHE WELT

351

TABELLE 2: Satz- sprachliche Nr. Mittel

Kontext

Np

Veränderung d.prop.Bedeutg.

15

Deixis

Sprecher; Angesprochener; Horer; Zeit; lok. Posi­ tion d. Sprechers; Objekte, auf die referiert wird

+

16

Fragesatz Wissen, Wunsch nach Spezifizierung

Frageeinstellung

17

Befehlssatz

soz. Beziehung zum Hörer, Erwartung

Befehlseinstellg.

18

Ausrufesatz

Einstellungen, Emotionen

Exklamation

19

Satzadverbien

Einstellungen

Kommentar

20

lexikal. Variante

Erwartung bezügl. Varianten; soz. Gruppierung

Stilschicht

21

lexikal. Variante

Haltg., Interesse, Normen als Basis fur Bewertung

Wertung

22

Satzstel- Emotionen, Interessen, lg. supra-Verhaltensnormen segment. Eigenschaft

Wertung

23

supraseg- Normen als Basis fiir Werment. Ei- tung genschaft

Wertung

24

supraseg- geteilte Prasuppositionen ment Eigenschaft

Kontrastakzent

andere eigentliche Bedeutung

25

kommunika-Interaktionszusammenhang tive Variante

indirekter Sprechakt

Propositionalisierung des es

26

Namen

Wissen

27

Anrede

Einstellg. zum Horer; Kenntnis der Eigenschaften des Horers, Existenz von Horereigenschaften

Anrede

352

ANITA STEUBE

Fortsetzung Tabelle 2: 28

Anrede

dto

Anrede

29

Termini

Wissen

30

Termini

Wissen

+

31

Begriffs- Wissen ausdruck

+

Bezeichnungswahl

In Auswertung der Tabelle 2 ist zu erkennen, daβ indirekte Zitate mit deiktischen Elementen (Beispiel 15 - in den spateren Beispielen kommen auch deiktische Elemente vor, sind dort aber aus Einfachheitsgriinden nicht berucksichtigt), Termini und ideologietragenden Wortern sowie bestimmten Kennzeichnungen die propositionale Bedeutung des Originalausdrucks andern. Bei­ spiel 30 gehort davon eigentlich nicht in die Tabelle. Termini und ideologietragender Wortschatz war nur behandelt worden um zu zeigen, da(3 Konnotationen semantisch uneinheitlich sind. Bei der Deixis wie bei Kenn­ zeichnungen es sich schlieβlich um Referenzprobleme. Die Äuβerungsbedeutung ist naturlich ebenso verandert wie die semantische Struktur, wenn im Zitat 'er' statt "ich", "jenes" statt 'dieses', 'am folgenden Tag" statt "übermorgen", "der Mann im Frack" statt "der Diregent" gesagt wird. Es wird aber mit 0riginaläuβerung wie mit Reporteräuβerung auf das gleiche Objekt referiert, und das ist hier für faires Zitieren entscheidend. Wenn Referenz von Bedeutung getrennt und eigenstandig beschrieben werden muβ, stören diese Pluszeichen in der Tabel­ le nicht mehr, und die eingangs getroffene, auf Frege zurückgehende Feststellung, daβ sich indirekte Zitate

KONTEXT UND MOGLICHE WELT

353

mit daβ-Sätzen auf die propositionale Bedeutung der Originaläuβerung beziehen, ist in alien Fallen bestatigt. Das gilt ebenso fur die Beispiele 22 und 23: Was dort in der 0riginaläuβerung an propositionaler Bedeutung enthalten ist, wird beibehalten. Die Wertungsbestandteile aber, die sich in der Originaläuβerung durch suprasegmentale Elemente ausdrükken, sind in der Reporteräuβerung in die propositionale Bedeutung integriert, denn der Reporter spricht über urspriingliche Einstellungen, er drückt sie nicht direkt aus. Ähnliches geschieht schon beim Zitieren von Fragen, Befehlen, Ausrufen (vgl. 16 - 18), wo im Redevorspann (X fragte/bat/vief aus) propositional iiber die ursprunglichen Einstellungen, die dort durch Satzstellung und suprasegmentale Mittel zum Ausdruck kommen, gesprochen wird. Sind die ursprunglichen Einstellungen subjektiv an den Sprecher gebunden (vgl. Beispiel 18, s. auch A. Steube 1983, 130ff), fallen sie im Zitat weg oder werden in Einstellungen des Reporters umgewandelt (vgl. 20, 24). Nach unserem derzeitigen Uberblick tritt ein Wechsel in der Topik-Fokus Gliederung in Abhangigkeit vom Kontext in der indirekten Rede nur bei Kontrastakzent ein. Die Ursache des Wechsels ist, daβ Präsuppositionen der ursprunglichen Kommunikationspartner von den neuen Kommunikationspartnern nicht geteilt zu werden brauchen (vgl. 24; s. auch Pt. 4 ) . Bei R. Pasch 1983 ist der Topik-Fokus-Wechsel eine Veranderung der eigentlichen Bedeutung, d.h. der Operand der Sprechereinstellung (in 24 Wunscheinstellung) andert sich kontextabhangig. Ein Sinnwechsel findet auch statt, wenn etwas durch direkten statt durch indirekten Sprechakt mitgeteilt wird. Der Reporter hat den indirekten Sprechakt als sol-

354

ANITA STEUBE

chen aus dem Kontext zu erschlieβen. Wenn er seinen Kommunikationspartnern den Kontext mitteilt und ihnen dadurch zu den gleichen Schluβfolgerungen verhilft wie den ursprünglichen Hörern, kann er den indirekten Sprechakt auch in indirekter Rede ubernehmen. Andernfalls mu(3 der kommunikative Sinn der 0riginalau3erung direkt ausgedrückt werden, d.h., der Reporter bringt den cs in eine Proposition ein, propositionalisiert inn. Das ist dann mit einer Sinnanderung verbunden. In beiden Fallen wird der gleiche Sachverhalt identifiziert; die Wahrheitsbewertung wird also gleich ausfalien. Wir konnen in diesem Fall nicht davon ausgehen, daβ die propositionale Bedeutung der 0riginaläuβerung beizubehalten sei, denn diese bildet nur eine Voraussetzung fur den kommunikativen Sinn. (Vgl. 2 5 ) 1 7 . 3. Wenn wir nun näher auf den Kontext eingehen, so weist die Literatur bereits eine breite Skala von Hinweisen auf, was zum Kontext zu rechnen ist. Lewis sagt in "Prinzipien der Semantik", daβ zum Kontext alles gehort, was neben der Satzbedeutung bei der Bestimmung der Extension 18 eine Rolle spielt. 1973 hat er explizit aufgezahlt: a) die mögliche Welt, b) die Kontextkoordinate mit Sprecher; Horer; Zeit; Ort; Objekten, auf die hingewiewiesen wird; Vorgangertext; die Zuordnungskoordinate fur Variable. Cresswell und Bierwisch halten vor allem die mogliche Welt fur wichtig und schlagen vor, den Kon­ text dort einzuordnen bzw. meinen, da3 er sich aus der 21 22 möglichen Welt ergibt. Montague/Schnelle geben als Kontext zusatzlich an: das tatsachliche oder angenommene

KONTEXT UND MOGL1CHE WELT

355

Vorwissen der Gesprachspartner, ihre Intention, ihr Sta­ tus und ihre Einstellung zueinander. Das Vorwissen deckt 23 auch das ab, was A. Kratzer mit Redehintergrund bezeichnet hat und soll das identifizierbare Wissen einschlie24 3en, das Kemmerling bei einer Sprachgemeinschaft als Voraussetzung fur erfolgreiches Referieren ansetzt. Wir wollen aus unseren Beispielen weiter vervollstandigen: die Interessen der Kommunikationspartner, in der Sprachgemeinschaft übliche Normen und Bedingungen fur Bewertungen, von den Kommunikationspartnern geteilte Prasuppositionen, die Zuordnung der Kommunikationspartner zur Kreuzklassifizierung von regional, funktional und sozial unterschiedlicher Verwendung von Sprache und die Einstellungen der Kommunikationspartner dazu. Zur Identifizierung der Objekte, iiber die gesprochen wird, sind die Vorschlage 25 von Fodor/Sag zu beriicksichtigen, enteweder diese Objek­ te selbst zum Kontext zu zahlen oder aber die Absicht des Sprechers, referieren zu wollen. Das letztere erweist sich als richtiger: Kontext ist weniger in den objektiven Gegebenheiten, in den Dingen, Eigenschaften, Relationen und Sachverhalten zu sehen, iiber die gesprochen wird. Diese gehen nur in kognitiv und emotional verarbeiteter Form in den Kontext ein, als Wissen iiber Dinge und Sachverhalte, als Interessen, Absichten, Verhaltensweisen und Haltungen etc. beziiglich dieser Dinge und Sachverhalte. Die zentrale Stelle hat der Mensch mit seiner Fahigkeit zu erkennen und zu Kommunizieren iune, wobei seine Fähigkeiten natürlich auch objektiven Steuergröβen unterliegen. Auch die Sprache oder eine Variante dieses heterogenen Systems gehoren nur insofern zum Kontext, als der Mensch an sie gebunden ist und Enwartungen beziiglich ihrer Verwendung hegt. Wenn David Lev/is meinte, Kontext sei alles, was neben der Bedeutung zur Bestimmung der Extension dient, so

356

ANITA STEUBE

la(3t sich das nun auf an den Menschen gebundene Faktoren festlegen und einschranken. Diese empirisch gestutze Behauptung gilt unabhangig von einem Modell. In Tabelle 2 sind die für den jeweiligen Fall von Kontextabhängigkeit gültigen Faktoren eingetragen. Auf die Beziehung von Kontext und mogliche Welt geht Abschnitt 5 noch einmal ein, wenn zuvor noch die komplexeren Kontextverhältnisse in konjunktivisch uneingeleiteter indirekter Rede besprochen wurde. 4. Im Deutschen gibt es neben den da$-Satzen auch die Moglichkeit, die indirekte Rede mit uneingeleitetem konjunktivischen Nebensatz auszudrücken. Wir wollen noch untersuchen, worin sich diese zweite Ausdrucksform vom daβ-Satz unterscheidet und ob sie für die Semantik vom gleichen heuristischen Wert sein kann wie da(3-Satze. 4.1. In den Beispielen in Tabelle 1 wurde die Aufmerksamkeit bischer nicht darauf gelenkt, daβ die Konstituentenreihenfolge im Hauptsatz und im da(3-Nebensatz unterschiedlich sein konnen (vgl. z.B. Sätze 19, 20, 22, 26, 27, 28). Dafür ist eine Begriindung zu finden. "Wir waren gestern in der Treptower Volkssternwarte. Dort konnten wir durch ein Fernrohr die Sterne und den Mond beobachten. Danach habe ich mir dieses popu26 larwissenschaftliche Buch über Astronomie gekauft." 'Dort', 'danach' sind topikalisierte Satzglieder, die den Textanschluβ herstellen und aus diesem Grunde im Hauptsatz vorangestellt werden. Im daβ-Satz ist die Vo-

KONTEXT UND MOGLICHE WELT

357

ranstellung in diesen Fallen nicht möglich: X sagte, da3 dort wir .,. X sagte, da(3 wir dort ... X sagte, da$ danach ich .. X sagte, daβ ich danach .. Das Nebensatzsubjekt muβ aber nicht immer die erste Stelle im daβ-Satz einnehmen: 32)

X sagte,

daβ

morgen eine

Postkutsohe

naoh Dresden

abgeht. 33)

X sagte, daβ tatsächlich

kein

34)

X sagte,

gestern

Apfel

Zug

fährt.

daβ

diese

keiner

kaufen

daβ

am Sonntag in dev Treptower

woll-

te. 35)

X sagte,

warte beinahe 36)

Volksstern-

ein Brand ausgebrochen ware,

X sagte, daβ nach der Restauration

alles

ganz an-

ders wurde.

Das Subjekt nimmt dann nicht die erste Stelle im Nebensatz ein, wenn es fokal ist, was in den falschen Satzen oben nicht der Fall war. Einem solchen fokalen Sub­ jekt konnen dann sogar mehrere topikalisierte Konsti­ tuenten vorausgehen. Zur Erklarung der folgenden Beispiele ist diese Regel aber noch nicht ausreichend. 37)

X sagte,

daβ

sich

seine

Aufmerksamkeit

einer

ande-

ren Frau zuwandte. 38)

X sagte,

daβ daβ

sie

der alte

gestern

sie

Fuchs wiederfinden

39)

X sagte,

40)

X sagte, daβ naoh dem Umzug sie/Frau

ihr

das mitgebracht

Sohnchen ein Himmelbett zureohtgemacht

Muller

wird. hat. ihrem

hat.

Sind neben dem Subjekt andere Konstituenten topikalisch, konnen diese vorangestellt werden, wenn sie pronominal sind und das Subjekt selbst kein Pronomen ist. Ist das Subjekt topikalisch, mussen sogar Konsti-

358

ANITA STEUBE

tuenten mit Kontrastakzent folgen. 41)

X sagte,

daβ

das Konzert

42)

X sagte, daβ der groβe Junge dem Peter bahn schenken

gestern

ausge fallen die

-ist. Eisen-

wollte.

Sind alle Satzglieder fokal, bleibt das Nebensatzsubjekt ebenfalls an erster Stelle. 43)

X sagte,

daβ

berschatz

Kinder einen mittelalterlichen

beim Spielen

in einem Steinbruch

Silfan-

den.

Diese an die Topik-Fokus-Gliederung gebundene Konstituentenreihenfolge in daβ-Satzen zeigt, daβ es in einschlägigen Fällen zu Abweichungen von der Hauptsatzreihenfolge kommen muβ. 44a) Danach habe ich mir dieses liche

Buch

popularwissenschaft-

gekauft.

44b) ..., daβ ich mir danach dieses sehaftliche 45a) Der alte

Buch gekauft Fuchs findet

45b) ..., daβ sie

der alte

46a) Dem Peter wollte

sie

popularwissen-

habe. wieder.

Fuchs

wiederfindet.

der gro$e Junge die

Eisenbahn

schenken. 46b) ..., daβ der groβe Junge dem Peter schenken

die

Eisenbahn

wollte.

Die Gesamtabfolge von Topik und Fokus wird damit in diesen Beispielen noch nicht beeinträchtigt, die Zuordnung zwichen textlich und situativ Bekannten und 28 Neuem wird nicht geandert.

Aus der Stellungsregel

im daβ-Satz, daβ vor topikalischem Subjekt keine fokale Konstituente stehen darf, folgt aber auch, daβ Fokus-Topikalisierung dort nicht mäglich ist:

KONTEXT UND MOGLICHE WELT 47)

Was ist

47a) Mude ist

mit

Petev

359

los?

er.

Die an die Fokus-Topikalisierung geknüpfte semantische Information, da|3 gerade 'müde sein' die Eigenschaft ist, die aus einer Klasse präsupponierter Eigenschaften per Behauptung selektiert wird, ist im daβ-Satz nicht so nachvollziehbar. 47b) . . . , daβ er (sehv) müde i s t .

Das gleiche gilt auch fur fokus-topikalisierte nominale Konstituenten: 48)

Wo hat sich

denn dev Junge die

ganze

Zeit

hevumge-

tvieben? 48a) In dev Fétdsoheune 48b) ..., daβ er in

hat ev

gehockt.

dev Feldscheune

gehockt

hat.

Konjunktivische uneingeleitete indirekte Rede unterscheidet sich jedoch in der Konstituentenreihenfolge nicht von der 0riginaläuβerung. Es wird lediglich der Indikativ

durch den Konjunktiv ersetzt, und die

Rede wird mit dem Vorspann "X sagte/dachte/,.." versehen. 33a) X sagte, es fuhve

tatsächlieh

kein Zug.

40a) X sagte, nach dem Umzug hatte ihvern Sohnchen 47b) X sagte,

ein

müde sei

48c) X sagte, in

sie/Fvau

Himmelbett

Müller

zuvechtgemaeht.

er.

dev Fetdsoheune

habe ev

gehookt,

Wird der gesamte Dialog in die indirekte Rede übertragen, so daβ die Prasuppositionen (es gibt eine Klasse relevanter Eigenschaften, Orte; vgl. 47b), 48c) ) von den neuen Kommunikationspartnern auch geteilt werden und die Topik-Fokus-Realisierung da-

360

ANITA STEUBE

durch fur sie die gleiche Berechtigung hat wie fur die ursprunglichen Kommunikationsteilnehmer (vgl. 47a), 48a)), ist die uneingeleitete konjunktivische indirekte Rede das geeignetste Übertragungsmittel fur den Reporter, denn der daβ-Satz kann die Kontrastakzente nicht βbernehmen. 4.2. In Steube 1983 war bereits festgestellt worden, daβ uneingeleitete konjunktivische indirekte Rede ein Zitieren im engeren Sinn des Wortes bedeutet, weil die fremden Gedanken mit den fremden Worten wiedergegeben werden mussen. Ich habe in diesem Monat noch keinen Lohn hekommen. X s a g t e , ev habe in diesem Monat seine Mopse noch nicht. Das Zitat mit Konjunktiv - wenn der Konjunktiv als Modusdifferenz zum Indikativ bewuβt und systematisch 29 49a) 49b)

verwendet wird

- unterstellt, daβ der Originalspre-

cher genau diese Jargonausdriikke verwendet hat. In der Terminologie von E. Lang 1983 ist es somit kein faires Zitat, weil dem Horer die Rekonstruktion der Äuβerungsbedeutung der Originaläuβerung verbaut wird. Dem Horer wird suggeriert, der Originalsprecher habe eine andere nichtpropositionale Bedeutung als in Wirklichkeit zum Ausdruck gebracht. 49c) X s a g t e , daβ ev diesen Monat seine Möpse noch nicht hat. 49c) ist dagegen ein faires Zitat durch einen Repor­ ter, der in der passenden Kommunikationssituation aus eigenem Antrieb oder in Anpassung an seine Kommunika-

KONTEXT UND MOGLICHE WELT

361

tionspartner Jarconausdrücke verwendet. Indikativisches Zitat mit daβ-Satz ist, wie oben festgestellt, durch doppelte Kontextanpassung gepragt. Es bezieht sich nur auf den propositionalen Gehalt der 0riginaläuβerung. Wir führen noch einige konjunktivische Zitate an. 50a) Ich werde mir einen Trabant zulegen. x 50b) X s a g t e , er würde sich eine Pappkiste zulegen. Der Reporter unterstellt, die abwertende Einstellung, die in 'Pappkiste' zum Ausdruck gebracht wird, sei die Einstellung des Originalsprechers, weil die Verwendung des Konjunktivs unterstellt, der Originalsprecher habe 'Pappkiste' gesagt. Mark Twain hatte ein gutes Gespur für die Wirkung regionaler und archaischer Sprachformen. 51b) X sagte, Samuel Langhorne Clemens habe ein gutes Gespur fur die Wirkung regionaler und archaischer Sprachformen gehabt. Der Reporter unterstellt, der Originalsprecher habe die Bezeichnung "Samuel Langhorne Clemens" benutzt und verdeckt so einen Namensuterschied zwischen Originalund Reporteräuβerung. Die Referenz wird nicht angetastet (vorausgesetzt, die Kommunikationspartner verfiigen über das notige Wissen) . 51a)

In konjunktivisch uneingeleiteter indirekter Rede geniigt es nicht, daft die Referenz gewahrt wird. Die Beispiele 49b), 50b) und 51b) sind keine fairen Zitate, weil diese Zitierform verlangt, daft die ursprüngliche Ausdrucksweise beibehalten wird. Ist für die neuen Kommunikationsteilnehmer mit der ursprünglichen Variante die Referenz nicht zu sichern, muβ eine Explikation hinzugefügt werden.

362

ANITA STEUBE

X s a g t e , Mark Twain habe ein gutes Gespür fur die Wirkung regionaler und arohaischer Spraohformen gehabt. Mark Twain ist der Künstlername von Samuel Langhovne Clemens. Die Behandlung konnotierter Varianten in konjunktivisch uneingeleiteter indirekter Rede läβt nicht die Verallgemeinerung zu, daβ die ursprungliche Kontextbindung im Zitat immer beibehalten werden musse. Die Zeitverschiebung und die Verschiebung deiktischer Pronomen läuft ebenso ab wie in daβ-Satzen. 51c)

50c) X sagte, er würde sich einen Trabant zulegen. Es ist erst recht nicht die Verallgemeinerung erlaubt, uneingeleitete konjunktivische indirekte Rede würde sich auf die gesamte Äuβerungsbedeutung beziehen: Anreden und Ausrufe werden ebenso wie in daβ-Satzen in Reporterbehauptungen über stattgehabtes Anreden und Exklamieren umgewandelt, Satzadverbien, die Sprechereinstellung ausdrücken, konnen konjunktivisch ebensowenig zitiert werden. Hier ist allerdings eine zusatzliche Beobachtung beachtenswert. Ist der Reporterkontext bezüglich Einstellung leer, d.h. der Reporter wird von seinen Partnern nur als 'Sprachrohr' des Originalsprechers verstanden, dann unterstützt der Konjunktiv die Tendenz, die im Zitat enthaltene Sprechereinstellung als Einstellung des Originalsprechers zu verstehen. 52a) (Gehoren Sie zur Gruppe 79-04? Ich bin die Bereichssekretarin.) Dr. Mutter hat soeben telefonisch d u r o h g e s a g t , daβ er leider erst in der kommenden Woche gesundgeschrieben werde, ... 52b) ... , er werde leider erst in der kommenden Woche gesundgeschrieben.

KONTEXT UND MOGLICHE WELT

363

Das intuitive Wissen daruber, da3 der Konjunktiv fremde Gedanken mit fremden Worten wiedergibt, scheint die Interpretation zu fordern, 'leider' sei vom Originalsprecher geäuβert worden. Diese Interpretation wird aber nicht vorgenommen, wenn diese Einstellung mit dem Reporterkontext kollidiert. Zitiertes 'leider' verstieβe dort gegen die Kommunikationsimplikatur, ehrlich und relevant zu sein: Ein Student, der aus personlichen oder objektiven Griinden den Ausfall der Lehrveranstaltung begrüβt und weiβ, daβ er die Zustimmung seiner Kommilitonen hat, würde zitieren

52c)

... , er wevde erst geschrieben.

in der kommenden Woche gesund-

Die Maxime, fremde Gedanken mit fremden Worden wiederzugeben, hebt auch nicht das Fairnessgebot auf, den kommunikativen Sinn der 0riginaläuβerung zu iibermitteln. Ein indirekter Sprechakt darf wortlich nur so zitiert werden, wenn der Reporterkontext - wie beim daβ-Satz - den kommunikativen Sinn der 0riginalau(3erung erschlieβen läβt.

25c)

Der Mann f r a g t , ob du e i n Stückchen b e i s e i t e r ü c -

ken kannst. Der Angeredete kann naturlich auch bei diesem Zitat den kommunikativen Sinn der 0riginaläβuerung mit Hilfe des Interaktionszusammenhangs erschlieβen, aber ihm ist eher als bei einem Aufforderungszitat (vgl. 25b)) die Moglichkeit gegeben, durch eine dumme Bemerkung, ein Miβverstandnis vorschützend, die Aufforderung zu umgehen. Worauf bezieht sich also konjunktivische uneingeleitete indirekte Rede? Sie steht zwischen direkter und indirekter Rede mit daβ-Satz. Konjunktivische uneingelei-

364

ANITA STEUBE

tete indirekte Rede bezieht sich im wesentlichen auf den propositionalen Gehalt der Äuβerungsbedeutung. Im Bereich sinnverschiedener sprachlicher Varianten und Benennungen werden aber durch den Akzent auf formaler Übereinstimmung mit dem Original die Sinnunterschiede beim Zitieren nicht so automatisch offenbart wie bei da(3-Satzen, wo der kontextbedingte Sinnwechsel eintreten muβ. In konjunktivischer Rede wird in diesem Bereich der Sinnwechsel nur sichtbar, wenn bei Interpretationsschwierigkeiten eine Explikation eingefiigt wird ähnlich 51c. Im Bereich von Konnotationen ist die konjunktivische uneingeleitete indirekte Rede also kein so konsequentes heuristisches Mittel fur die Semantik wie die daβ-Sätze. in der überwiegenden Mehrzahl der Falle verhalten sich jedoch beide Redeformen gleich. Abschlieβend muβ zu diesem Punkt noch vermerkt werden, daβ die in 4.1. behandelten Unterschiede, die konjunktivische uneingeleitete Zitate gegemüber daβSätzen bezüglich der Konstituentenreihenfolge aufweisen, ihre Ursache darin haben, daβ die Konjunktion 'daβ' in der uneingeleiteten Rede fehlt. Der Konjunktiv konnte dabei in der Umgangssprache ruhig durch den Indikativ ersetzt werden. Demgegenüber sind die Un­ terschiede, die in 4.2. zwischen da-β-Sätzen und konjunktivischen Zitaten beziiglich der Zitierbarkeit von Sprechereinstellungen aufgedeckt wurden, ausschlieβlich auf das Vorhandensein oder Nichtvorhandensein des Konjunktivs zuriickzufuhren. Beispiel 52 ist in zwei Varianten angegeben, a) mit konjunktivischem daβ-Satz, b) mit konjunktivischem eineingeleiteter Rede. Wir konnten diese Doppelung an alien einschlägigen Fällen

KONTEXT UND MOGLICHE WELT

365

durchprobieren und wiirden keinen Sinnunterschied feststellen. 5. Zum Abschluβ soll auf die Beziehung Kontext mogliche Welt eingegangen werden. Wir wollen prüfen, ob das heuristische Mittel der indirekten Rede mit daβSatz dafür spricht, Kontext als Teil der moglichen Welt zu betrachten oder auch umgekehrt, wie es Bierwisch, Cresswell, Lewis, Montague vorschlagen. Das Konzept der moglichen Welten wurde dazu benutzt, auch die Satze der Wahrheitswertsemantik zuganglich zu machen, die nicht uber Sachverhalte der realen Welt sprechen, sondern uber solche aus fiktiven, erdachten, angenommenen Welten. Bierwisch (1980a) versteht reale Welt als Struktur der internen kognitiven Reprasentationen von Erfahrungen realer Personen. Mogliche Welten sind dann durch kognitive Operationen von rea­ len Personen auf realen Welten entstanden. Bei diesen Operationen werden die Sachverhalte entsprechend der mentalen Strategie des Menschen, der diese Projektion vornimmt, verandert. War in der Literatur schon immer klar, daβ nichtreale mogliche Welten mentale Konstruktionen sind (verwendet zur Beschreibung der Glaubensund Wissenssatze, von Dispositionen, Modalitaten und hypothetischen Satzen), so ist mit dem Bierwisch-Modell der dort noch offene Übergang von realen Welten (sie wurden entweder unbesehen als mengentheoretische Strukturen oder als die Struktur der realen Welt interpretiert) zu den übrigen moglichen Welten erklart. Wenn hier nun auch experimentell nachgewiesen wurde, daβ Kontext - und zwar unabhängig vom Interpretations-

366

ANITA STEUBE

zusammenhang - immer mentale Strukturen von Menschen sind (die im Text teilweise sprachlich realisiert wer­ den, teilweise nicht), stellt sich die Frage des Zusammenhangs oder Unterschieds von Kontext und mogliche Welt automatisch erneut. 31 Stalnaker sieht die Unterschiede in der unterschiedlichen Funktion von Kontext und mogliche Welt: Ein interpretierter Satz (in der hier verwendeten Terminologie = Sinn, Satzbedeutung, semantische Struktur) ist eine Funktion von einem Kontext in eine Proposition (Äuβerungsbedeutung) ; eine Proposition ist eine Funk­ tion von einer moglichen Welt in einen Wahrheitswert. Erst im letzten Schritt wird also danach gefragt, wie die Welt beschaffen sein mu(3, urn mit den Wahrheitsbedingungen des Satzes ubereinzustimmen. Stalnaker hat an einer Reihe von Beispielen gezeigt, da(3 sprachliche Fakten nicht adaquat beschrieben werden konnen, wird in

einem

Schritt direkt von der semantischen Struktur 32

zum Wahrheitswert ubergegangen. Die Wahrheitsdeterminanten, die Teil des Kontextes sind, miissen folglich von denen unterschieden werden, die Teil der moglichen Welt sind, wenn sie auch aus dem gleichen kognitiven Bereich kommen. Eingangs haben wir die Weltbindung der direkten Rede erläutert mit dem Ergebnis, daβ diese im Zitat durchsichtig, d.h. erhalten bleiben mu(3 (vgl. 3a)). Satz 4b) zeigt, daβ bei einem indirekten Zitat einer realen Person keine Worte über sich selbst als Teil einer fiktiven Welt in den Mund gelegt werden konnen. Wir wollen jetzt noch testen, was geschieht, wenn Kontext und mog­ liche Welt in daβ-Sätzen unabhängig motiviert sind.

KONTEXT UND MOGLICHE WELT

8a) 8b)

367

Wenn dev evste Gewandhaus-Kapellmeistev kommt, beginnt das Konzevt. Vatev hat gesagt, daβ das Konzevt b e g i n n t , wenn dev Mann mit dem Taktstock kommt.

(Es tritt Professor Bosse auf und diregiert von der Geige aus.) Das Kind könnte nun feststellen, der Originalsprecher habe etwas falsches gesagt: es ist niemand mit Taktstock gekommen, und das Konzert hat doch begonnen. In Wirklichkeit ist die 0riginaläuβerung wahr, aber die im Zitat postulierte Bedingung stimmt nicht mit der Wirklichkeit überein. Es wurde eine dem Kenntnisstand des Kindes angemessene Kontextbindung fur die indirekte Rede gewahlt, die zu einer falschen Aussage führte. In diesem Fall ist nicht fair zitiert worden. Ein faires Zitat ware gewesen 8c)

Vatev hat g e s a g t , daβ das Konzevt beginnt, wenn dev Mann mit dem Fvack kommt (vorausgesetzt, es fehlt nur noch der Diregent).

Wenn wir nur den Kontext in Betracht Ziehen, sind beide Zitate austauschbar. Erst die zusatzliche Berücksichtigung der moglichen Welt schafft die Auswahl, indem sie die Frage der Wahrheit ins Spiel bringt. An dieser Stelle muβ die Charakterisierung fur faire Zitate durch Ewald Lang33 erganzt werden urn das Wahrheitskriterium. Indirektes Zitieren muβ wie direkts welterhaltend geschehen, wahren der Kontext in indirekten Zitaten wechselt. So muβ die indirekte Rede als ein sprachlicher Zusammenhang betrachtet werden, in dem Kontext und mog­ liche Welt unabhangigen Regeln folgen was für die funktionale Trennung bei Stalnaker spricht. Kontext und mogliche Welt sollten zunachst als getrennte Kategorien

368

ANITA STEUBE

behandelt und in einem zweiten Schritt in ihrer ge34 genseitigen Beeinflussung untersucht werden. In der indirekten Rede sind nur solche Kontexte zulassig, die mit der ausgewählten Welt übereinstimmen.

KONTEXT UND MOGLICHE WELT

369

ANMERKUNGEN 1

Vgl. u.a. P. Sgall 1978 und 1982

2 Die schematische Verdichtung ist entnommen aus E. Lang 1983, S. 321. 3 Kontext schlieβt hier die mögliche Welt ein oder auch umgekehrt. Vgl. aber dazu die Punkte 3 und 5 in diesem Beitrag. 4

vgl. dazu auch R. Conrad 1983

5 zu Sprechereinstellungen vgl. auch M. Bierwisch 1980. M. Doherty 1981, E. Lang 1983, I. Zimmermann 1982. Sprechereinstel­ lungen zum dargestellten Sachverhalt werden in einigen jüngeren semantischen Arbeiten auch als propositionaler Typ bezeichnet. 6 Für 'Gedanke' steht in jüngerer semantischer Terminologie auch 'Satzradikal, propositionale Bedeutung, propositionaler Gehalt'. 7 vgl. G. Frege, über Sinn und Bedeutung, in: Funktion, Begriff, Bedeutung, 5 logische Untersuchungen, S. 54. 8

vgl. E. Lang 1983, A. Steube 1983.

9

vgl. A, Steube 1983, S. 124-147.

10 vgl. aber die Komplizierung sowohl für den zeitlichen Referenzpunkt als auch für den Sprecher bei A. Kratzer 1978, S. 17ff. 11 12 Quine.

E. Lang 1983, S. 321. F. v. Kutschera 1971, S. 183, in Auseinandersetzung mit

13 vgl. R. C. Stalnaker in R. J. Schmidt 1974; P. F. Strawson 1950, dazu auch E. Kemmerling in E. v. Savigny 1976. 14 Bei Frege sind die Bezeichnungs- und die Bedeutungsrelation bei Namen gleich. Hinweise zum Problem der Eigennamen, die zur Veränderung einer früheren Fassung führten, verdanke ich Peter R. Lutzeier. 15 vgl. besonders die Hochschullehrbücher von Th. Schippan 1972; W. Fleischer und G. Michel 1975; B. Hansen, A. Neubert und M. Schentke 1982 (im letzteren wird konnotative Bedeutung aber z.T. auch als Sprechereinstellung bezeichnet, so S. 147).

322) .

16

M. Bierwisch 1975.

17

Die Alltagsrede kennt nur sagen-Zitate (vgl. Lang 1983,

370

ANITA STEUBE

18

D. Lewis 1970: General Semantics

19

M. J. Creswell 1973

20

M. Bierwisch 1980, 1980a

21

dazu wollen wir in Punkt 5 Stellung nehmen

22

H. Schnelle 1972

23

A. Kratzer 1978, S. 108ff.

24

A. Kemmerling in E. v. Savigny 1976, S. 39-71.

25 26 S. 739.

J. D. Fodor und J. A. Sag 1982. Text aus K. E. Heidolph, W. Flämig und W. Motsch 1981,

27 Der Satz ist in abgewandelter Form E. Pasch 1983, S. 289 entnommen. 28 Was verändert werden kann, sind die Grade des kommunikativen Dynamismus innerhalb des Themas, Wir schlieβen uns der Argumentation von R. Pasch 1983 an, daβ diese Graduierung auf unabhängigen Prinzipien beruht, die mit der Einteilung der Informa­ tion in Bekanntes und Neues in der Äuβerungsbedeutung nicht vermischt werden sollte. 29 vgl. aber die statistischen Angaben zur Modusverwendung im heutigen Deutsch in verschiedenen Funktionalstilen und bei unterschiedlichen Autoren in G. Starke 1980, S. 706. 30 Wenn der Konjunktiv durch den Indikativ ersetzt wird, was in der Alltagssprache sehr oft geschieht, verschwimmt die Grenze zwischen indirektem und direktem Zitat: 48b) X sagte, daβ er in der Feldscheune gehockt hat. 48c) X sagte, in der Feldscheune habe er gehockt. 48d) X sagte, in der Feldscheune hat er gehockt. X sagte: "In der Feldscheune hat er gehockt." 31

R. C. Stalnaker in S. J. Schmidt 1974.

32

vgl. auch A. Kratzer 1978.

33

E. Lang 1983, S. 318.

34

vgl. R. Grunig 1982.

KONTEXT UND MOGLICHE WELT

371

LITERATUR Bar-Hillel, Y. 1974. "Indexikalische Ausdrücke". Pragmatik , Hrsg S. J. Schmidt, 166-186. München. Berka, K. und L. Kreiser (Hrsg). 1971. Kommentierte Auswahl zur Geschichte der modernen Logik. Berlin. Bierwisch, M. 1975. "Struktur und Funktion von Varianten im Sprachsystem". Linguistische Studien 19. 65-137. Bierwisch, M. 1980. "Semantic Structure and Illocutionary Force". Speech Act Theory and Pragmatics , Hrsg. J. R. Searle, F. Kiefer & M. Bierwisch, 1-35. Dordrecht: Synthese libr. vol. 10. Bierwisch, M. 1980a. "Utterance Meaning and Mental States". Memory and Cognition, Proceedings of the 1978 Symposium.Hrsg. F. Klix. Berlin. Conrad, R. 1983. "Fragesätze als indirekte Sprechakte". Untersuchungen zur Semantik. Hrsg. R. Ruzicka & W. Motsch. 343-367. Berlin: Studia grammatica. Cresswell, M. J. 1973. Logics and Languages. London. Doherty, M. 1981. "Grundlagen einer Theorie über sprachliche Ausdrucksmittel epistemischer Einstellungen". Erscheint in stu­ dia grammatica Bd. 23. Berlin. Fodor, J. D. & Sag, J. A. 1982. "Referential and Quantificational Indefiniteness". Linguistics and Philosophy 5, 355 p. Fleischer, W. & Michel G. 1975. Stilistik der deutschen Gegenwartsprache. Leipzig. Frege, G. "Funktion, Begriff, Bedeutung". 5 logische Studien. Hrsg. G. Patzig. Göttingen 1966. Frege, G. 1973. "Logik". Schriften zur Logik (aus dem Nachlass), mit einer Einleitung von L. Kreiser. 37-74. Berlin. Grunig, R. 1982. "La sémantique des mondes possibles et ses limites". Documentation et Recherche en Linguistique Appliquee. 26. 63-89. Hansen, B., Hansen, K., Neubert, A. & Schentke, M. 1982. Englische Lexikologie - Einfuhrung in Wortbildung und lexikalische Semantik. Leipzig. Heidolph, K. E., Flämig, W. & Motsch W. 1981. Grundzuge einer deutschen Grammatik. Berlin. Kapitol 4. Kemmerling, A. 1976. "Probleme der Referenz". Probleme der sprachlichen Bedeutung. Hrsg. E. v. Savigny. 39-71. Kronberg/Ts. Kratzer, A. 1978. Semantik der Rede (Kontexttheorie, Modalwdrterf Konditionalästze). Kronberg/Ts. Kratzer, A. 1980. "Possible-Worlds Semantics and Psychology". Linguistische Berichte 66.1-14. Kreiser, L. 1975. "Sinn, Gedanke und Bedeutung". Wissenschaftliche Zeitschrift der Universität Halle. 28. 69-78. Kutschera, F. v. 1971. Sprachphilosophie. Mtlnchen.

372

ANITA STEUBE

Lang, E. 1979. "Zum Status der Satzadverbiale". Slovo a slovesnost. 40. 200-213. Lang, E. 1983. "Einstellungsausdrücke und ausgedrückte Einstellung". Untersuchungen zur Semantik. Hrsg. R. Rużička & W. Motsch. 305-341. Berlin: Studia grammatica. Lewis, D. 1970. "General Semantics". Synthese 22. 18-67. Lewis, D. 1974. "Prinzipien der Semantik". Studlen zur Semantik. Hrsg. S. Kanngieger & G. Lingrün. 133ff. Kronberg/Ts. Montague, R. 1974. "Pragmatik und intensionale Logik". Pragmatlk I. Hrsg. S. J. Schmidt. 187-211. München. Ruzicka, R. & Motsch W. (Hrsg.). 1983. Untersuchungen zur Seman­ tik (= studia grammatica 22). Berlin. Pasch, R. 1983. "Mechanismen der inhaltlichen Gliederung von Sätzen". Untersuchungen zur Semantik (= studia grammatica 22). R. Ruzicka & W. Motsch, Hrsg. 261-304. Berlin. Schippan, Th. 1972. Elnfiihrung In die Semaslologle. Leipzig. Sgall, P., Hajicova, E. & Benešová, E. 1973. Topics, Focus and Generative Semantics. Kronberg/Ts. Sgall, P. & Hajičová, E. 1977/1978. "Focus on Focus". The Prague Bulletin of Mathematical Linguistics 28/1977. 5-54; 29/1978. 23-41. Sgall, P. 1978. "Satzsemantik und Logik". Llngulstlsche Studlen 47, Reihe A. 97-112. Berlin. Sgall, P. 1982. "Semantik und Pragmatik". Ü b e r s e t z u n g s w i s s e n schaftllche Belträge 6. Berlin. Schmidt, S. J. (Hrsg.) 1974. Pragmatik I. München. Schnelle, H. 1972. "Montagues Grammatiktheorie - Einleitung und Kommentar zu R, Montagues Universeller Grammatik". Schrlften zur Llngulstlk 5. Braunschweig. Stalnaker, R. C. 1974. "Pragmatik". Pragmatik. Hrsg. S. J. Schmidt, 148ff. Starke, G. 1980. "Zur mittelbaren Weidergabe von Rede- und Reflexionsinhalten im Deutschen". Wlssenschaftllche Zeltschrlft der Pädagogischen Hochschule Potsdam 5. 695-709. Steube, A. 1983. "Indirekte Rede und Zeitverlauf". Untersuchungen zur Semantik (-studia grammatica 22). Hrsg. R. Ruzicka & W. Motsch. 121-167. Berlin. Strawson, P. F. 1980. "On Referring". Mind. Zimmermann, I. 1982. "Explizite und implizite Faktivität". Lln­ gulstlsche Studlen 92. Reihe A. 81-139. Zimmermann, I. 1983. "Untersuchungen zum Verhältnis von Substantivgruppe und Nebensatz". Untersuchungen zur Semantik (=studia grammatica 22). Hrsg. R. Ruzicka & W. Motsch. 201-242. Berlin.

ELENA V. PADUČEVA Moscow

QUESTION-ANSWER CORRESPONDENCE AND THE SEMANTICS OF QUESTIONS

When attempting at a semantic analysis of quest­ ions we must from the very beginning draw a strict distinction between an interrogative sentence as a syntactic category and a semantical (and partly pragmatic­ al) notion of a question as an utterance expressing a demand for information, cf. 0. Jespersen's opposition

of questions

(in a strict sense) and

interrogatives.

A question is an utterance with the illocutionary force of a question. And an interrogative is a sentence con­ taining interrogative elements in its structure but not necessarily expressing the demand for information. About interrogative sentences expressing question we can say that they have standard

semantics, in contra­

distinction to different kinds of interrogatives, having idiomatic

semantics. One and the same sentence can be

used in different utterances and in one utterance a sentence may have standard semantics and in another not. So what is studied is, generally, the utterance. Here are some examples of utterances with non­ ­standard semantics: (1) Rhetorical

questions:

Who knows what will become of us?

374

ELENA V. PADUČEVA

(2) Interrogative (3) Interrogative (4) Interrogative (5) Eeho-questions: (6) Interrogative

requests: W i l l you p a s s t h e s a l t ? propositions: D o n ' t you want t o h a v e a walk? reproaches: Who asked you to inter­ fere with my business? Who told me that? Well everybody knows it. wonders:

Has he really done it?

The non-standardness of the semantics of a quest­ ion explains some queer replies to it, such as the following: (7) - Why don't you go to bed? - Wait a minute. (8) - Why do you whistle? - Excuse me. Many interesting problems arise in the area of questions with non-standard semantics, but the main subject of what follows will be the question in a strict sense. One of the natural approaches to the semantics of questions is to formalize the question-answer correspond­ ence. Indeed, we can assume that the listener has under­ stood the question if he knows what kind of information must be given as an answer - though, perhaps, he has no such information at hand. In other words, the listener understands the question if he can characterize correct­ ly the semantical scheme of the answer. So at any event a description of the semantics of the question must provide a semantic characterization of the set of all its appropriate answers. Not for every question the set of all appropriate

QUESTION-ANSWER CORRESPONDENCE

375

answers can be positively characterized, so our analysis is satisfactory not for all types of questions. One of the most difficult problems is to characterize the question-answer correspondence for how-questions (cf. Sgall 1982) . Consider such a question: How a child acquires its language? There is no procedure which would be able to determine that one sentence or text can be regarded as an appropriate answer to that question and the other not. What follows is a series of delimitations. Every delimitation gives rise to a condition which is fulfilled otherwise. So these conditions in their integrity serve to characterize the set of all appropriate answers to a given question. answers 1. The first delimitation is between direct (DA) and non-direct answers (NDA): (9) Question: Will she come? DA: Yes. No. Only if her father allows her. NDA: A little bit later. Only because you asked her. The sentence A is an indirect answer to a question Q if it is not a direct answer itself but it has some consequence which is a direct answer. There are some queer cases. In the pair (10) the answer (A), in some respect, satisfies the speaker, though it is not clear why: (10) Q: Where is John? A: He left. 2. Full v_s_. non-full aanswer. This distinction can be illustrated with the following example:

376

ELENA V. PADUCEVA

(11) Q: Whom did Mary invite to the party? A: She invited John. If there is nothing special in the question-answer situation, then the question is usually understood as demanding a full answer, so that a non-full answer amounts to concealing the truth. Cf. an example of a situation in which a non-full answer is appropriate: (12) Q; Who wants to say something? A: I do. Outside such special contexts which make non-full answers normal it is preferable to interpret any quest­ ion as requiring a full answer. So if no special context is mentioned or implied, only full answer will be regard ed as appropriate. There are indeed questions which cont ain an inherent presupposition of uniqueness. Thus, question (13) can have two readings. Under one of the readings, namely, one which includes a presupposition of uniqueness, to name one person amounts to give a full answer: (13) Do you know who lives in the cottage? 3. Answers to common, alternative and questions.

special

According to their syntactic structure questions are divided into 3 groups - common, alternative and special. (14) Common Q: Have you solved the problem? Alternative Q: Did you solve the problem or did you cheat? Special Q: Who solved the problem? It is clear that the semantic scheme of an appropriate answer is different for different types of

QUESTION-ANSWER CORRESPONDENCE

377

questions, The distinction is not always as clear as it seems to be. Cf. the following examples: (15) When do you leave, to-day or to-morrow? Where did you see him, in London? What feelings filled Ibrahim's soul - jealousy, fury, dispair? Another type of deviation is illustrated by example (16), where the question belongs to the type of common questions according to its structure, though it is under­ stood as a special question: (16) Do you go far? Do they pay you much? Will you stay here long? 4. The next distinction is valid only for special questions - it is a distinction between answers follow­

ing vs violating

the domain of the

question:

(17) Q: What kind of fish do you prefer? A: I prefer meat. (18) Q: What day was the most memorable in your life? A: It was a night. These dialogues are queer because the answers viol­ ate the domain of the questions. Such answers are consid­ ered to be inappropriate. If the answer is appropriate, it follows the domain of the question; e.g., from (19) we can make an inference that Mary is John's friend: (19) Q: Whom of John's friends do you like most? A: A like Mary most of all. A question can be ambiguous as to its domain: (20) Q: What feelings filled Ibrahim's soul? a. Q,: [What feelings] filled Ibrahim's soul?

378

ELENA V. PADUČEVA A 1 : Jealousy, fury, dispair. b. Q 2 : [What] feelings filled Ibrahim's soul? A 2 : Honorable ones (or: Mean ones; or: Strong

ones). Misunderstandings often stem from the ambiguity of domains: (21) Q: What anecdote did he tell you? A: A silly one. The speaker, most probably, wanted to know the anecdote, and the listener gave only its characteristics. 5. A question can carry a probable assumption with it. Thus, we have a distinction between answers follow­ ing vs violating the probable assumption of the quest­ ion: (22) Q: Who solved the problem? A: Nobody. The notion of a probable assumption is due to F. Kiefer (1977). There is something unusual in (22) be­ cause the answer violates the probable assumption of the question - the assumption that there exists at least one person who did solve the problem. But answers violat­ ing probable assumptions of questions can be regarded as appropriate. Equally unusual is the following dialogue: (23) - Who solved the problem? - Everybody. Indeed, question (22) has another probable assumpt­ ion - that not all members of the group have solved the problem. It is sometimes claimed that there is no difference

379

QUESTION-ANSWER CORRESPONDENCE

between probable assumptions of questions and usual pre­ suppositions. But this claim cannot be maintained: probable assumptions are to be distinguished from real presuppositions of questions; cf. presupposition of question (24): (24) Q: Who was glad that John failed? A1..: Nobody was. A 2 : John hasn't failed. Answer A2 is inappropriate, because it violates the presupposition of Q. In questions with existential presuppositions it is usually the proposition that gives rise to a pre­ supposition, not the question constituent. The only exception is a question in which the question word is connected with a noun phrase by means of the preposit­ ion of, expressing the definiteness of that noun phrase. Such questions always contain a presupposition that the domain of the question is not empty; e.g., (25) carries a presupposition that Mary has colleagues: (25) Whom of her colleagues did Mary invite to the party? 6. The last distinction is between answers ative

vs_ non-informative

for the speaker; cf. the

examples: (26) Q: Who built this house? A: The richest man in the town. (27) Q: What did you say? A: I said what I said. (28) Q: Who told you that? A: One person. (29) Q: Whom did you see there? A: Somebody,

inform­

ELENA V. PADUCEVA

380

Semantical opposition arising from distinctions 1 - 6 serve to characterize the propositional content of the question. Now what about its illocutionary force? It must be borne in mind that the article is limited to question in a strict sense, i.e. to question, as a special kind of illocutionary act, so we put aside such illocutionary varietes as exam questions, rhetor­ ical questions etc. In what follows I make use of the analysis of the illocutionary force of questions proposed by L. Aqvist and J. Hintikka (cf. Hintikka 1974). This analysis is based on the epistemic predicate KNOW. My own proposal is that two different meanings of the verb KNOW must be distinguished, in order to get an appropriate semant­ ic decomposition of the illocutionary force of question. Two meanings of KNOW when used with sentential complement are the following. The 1-st arises in the context where KNOW is used with that-clause. The 2-nd arises when KNOW is used with an indirect question. The 1-st meaning is obviously more elementary. There is a 2 formula which eliminates KNOW , with indirect-question complement, reducing it to KNOW , with that-complement. The semantic decomposition of question proceeds in two steps. On the first step Imperative operator Imp is introduced and the independent question is transform­ ed into an indirect question used as an argument of the 2 2 predicate KNOW . On the second step KNOW is reduced to KNOW 1 . Step 1 (1) Common question:

Did he solve the problem?

QUESTION-ANSWER CORRESPONDENCE

381

2 Imp Ks (if he solved the problem) (2)

Special

question: Who solved the problem? 2 Imp Ks (who solved the problem)

(3)

Alternative question: Did he solve the problem or cheat? 2 Imp Ks (if he solved the problem or cheated) Step 2 2

(1') K s (if he solved the problem)

(21)

[ (He solved the problem) → K1s (he solved the 1 problem)] & [(He solved the problem)]- K s(He solved the problem)] 2 a. K s (who solved the problem) E! x[(x solved the problem) & K1s (x solved the problem)] 2 b. K2s (who solved the problem) Vx{[(x solved the problem) - K1s (x solved the 1 problem) ]& v x [ ] (x solved the problem)→K1s

(3')

7(x solved the problem)]} 2 K s (if he solved the problem or cheated) [ (He solved the problem) → K1s (He solved the problem)] & [(He cheated) → K1s (He cheated)]. Full decomposition for the question (2) (in one of

its meanings)is as follows: (2a) 3 ! x (x solved the problem) & Imp K1s (x solved the problem)< = = => 'For the one who solved the problem make it so that I know that he solved the problem'.

382

ELENA V. PADUČEVA

REFERENCES Hintikka, J. 1974. "Questions on questions". Semantics and Philo­ sophy ed. by M. Munitz & P. Unger. 103-158. New York: N. Y. Univ. Press. Kiefer, F. 1977. "Some semantic and pragmatic properties of WH-questions.". statistical Methods in Linguistics 3. 42-71. Sgall, P. "Natural language understanding and the perspectives of question answering". Coling 82. Proceedings of the IX-th International Conference on Computational Linguistics. 1982. 357-364'. Amsterdam: North-Holland.

YORICK WILKS & CHRIS CUNNINGHAM Essex

A PURPORTED THEORY OF RELEVANCE

INTRODUCTION In recent years Sperber & Wilson (SW for short, 1982) have set out what they call a "theory of relevance". Its starting point is Grice's four maxims of communication (1975): they argue that these four can be reduced to one, that of Relation or "Be relevant", and their theory is intended to give content to that rather bare injunction. They state the aims of a pragmatic theory thus: - "to explain how successful communication is possible; in part­ icular, how utterances are understood" (ibid: 61). SW share with Grice the assumption that at the heart of human communication lie inferences, of a classical sort, from what is said and from other, implicit, assumptions that must be brought into the inference process. Thus far, the assumption is a very general one and shared by much work on discourse and communication analysis of the last ten years within psychology, philosophy, artificial intelligence (AI) and even linguistically, though to a rather smaller degree in the last field, almost certainly because the assumption cannot be accomodated within the transformational generative paradigm. The generative semanticists tried to do that and failed, and those parts of linguistics that

share the

assumption, but are now outside that paradigm are well sur­ veyed in Levinson (198 3).

384

YORICK WILKS AND CHRIS CUNNINGHAM The aim of SW's analysis can be stated as one of mak­

ing explicit the appropriate inferences so as to show, within a single logical space: what is said, by a speaker what additional implicit items of information must be brought to bear by a hearer what inferences follow from the above, including those Grice would have called implicatures. A further aim of SW's is to go beyond making all that explicit, in a way Grice did not, by placing it within a set of assumptions about "human information processing", rather in the way psychology and AI work does but, as we shall see, without making any strong commitments as to what is modelled by the hypothesised information processing. THE RELATION TO 'MUTUAL KNOWLEDGE' ACCOUNTS In the paper referred to, SW have chosen to contrast their ideas with work that makes use of terms such as "com­ mon knowledge" or "mutual knowledge" (Schiffer 1972, Clark & Carlson 1982) and which, in a philosophical-psychological manner, while also deriving from the work of Grice, attempts to characterise what an utterer can be assumed to believe about the knowledge and belief states of his interlocutor. We shall not discuss SW by contrast to that work, since we believe the two bodies of work to be more different than SW seem to: in particular because the recent work of Clark et al. has been concerned almost entirely with mutual knowledge in virtue of temporal or spatial co-presence of persons. Such a situation arises when two people go to the cinema together and both know the other knows he was there, and so on indefinitely. Such situations can only constitute a special subset of our knowledge and belief about the real world and nothing will be lost, initially, by examining SW

THEORY OF RELEVANCE

385

only in their own terms. CONTEXTUAL IMPLICATIONS: THE MAD-PASSER-BY EXAMPLE A key term for SW is "contextual implications": these are nontrivial

inferences that can be drawn from context

and utterance combined, for: "having contextual implications in a given context is a necessary and sufficient condition for relevance" (ibid: 73) and deriving the contextual implications is, in effect, the establishment of the relevance of an utterance. A key re­ quirement will be a procedure for establishing what the context (in the sense of a set of propositions as input to an inference procedure) is for a given utterance, for that was just the lacuna that Grice failed to fill. A main argument of this paper will be that SW do not provide one. Let us now set out, by fuller quotation, one of SW's main illustrative examples. They declare such examples to be merely illustrative, not conclusive, and note that fuller treatment will be available in a forthcoming book. A critic must therefore be sure that any criticisms made are of the principles involved rather than of details of exposition, and that we hope to observe. We shall refer to what follows as the mad-passer-by example (ibid: 73): (16) a. Flag-seller: Would you like to buy a flag for the Royal National Lifeboat Institution? b. Passer-by: No thanks, I always spend my holi­ days with my sister in Birmingham. Not everyone finds the response in (16) b. immediate­ ly comprehensible. In order to understand it fully, the hearer has to supply (at least) the premises in

386

YORICK WILKS AND CHRIS CUNNINGHAM (17) and derive the conclusion in (18): (17) a. Birmingham is inland b. The Royal National Lifeboat Institution is a charity. c. Buying a flag is one way of subscribing to a charity. d. Someone who spends his holidays inland has no need of the services of the Royal National Lifeboat Institution. e. Someone who has no need of the services of a charity, cannot be expected to subscribe to that charity. (18) The speaker of (16)b. cannot be expected to sub­ scribe to the Royal National Lifeboat Institut­ ion. In our terms, (18) is a contextual implication of (16) b. in a context which contains (17). It follows from (16)b. and (17) taken together, but from neither (16) nor (17) in isolation from each other. What is interesting about (16)b. from our point

of

view is the intuitive connection it reveals between being able to derive the contextual implications of an utterance and being able to see its relevance. Those who fail to see the relevance of (16)b. at first sight are precisely those who have failed to derive the contextual implication in (18), and any­ one who sees this implication will concede the relev­ ance of (16)b." One additional piece of cultural knowledge sometimes proves necessary for readers of the above: a flag-seller is a volunteer for a charity who stands on the pavement/side­ walk importuning passers-by to donate money and who then

THEORY OF RELEVANCE

387

receive a small paper flag in return. SW assume that the remark was understood by the flagseller, and claim that the example is an attested piece of dialogue. But setting doubts aside for a moment, the examp­ le should be able to tell us in what ways arriving at a contextual implication is different from doing a logical exercise; that is, how context, once found, is used together with the content of the utterance, to reach a contextual implication. This is the precise point at which any fail­ ures will constitute failure to construct an integrated, explanatory theory of pragmatics, which SW claim to have done. It must also be noted that we have to assume that (18) is

the intended interpretation, since SW say that without

(18) the remark is incomprehensible, and they are contend­ ing that it is not. So following SW, we have to take (18) as being what the speaker- the passer-by - in the example, thought to be "maximally relevant" to the hearer - the flag -seller - "the intended interpretation is generally the only

one that the speaker might have thought would be

maximally relevant to the hearer", (ibid: 81). One might reply initially to the above, not as fund­ amental criticism but as a form of mind-clearing, that what is interesting about it is the degree to which matters need not be at all as SW describe them. There are two related issues here: (i) the point of view problem, or "whose in­ ferences are the ones characterised above?", and (ii) the alternative hypotheses problem. ALTERNATIVE POINTS OF VIEW AND BELIEF SETS As to (i), the passage strongly suggests that in the above account it is a model of the hearer that is being offered ("the hearer has to supply..."), and that the

388

YORICK WILKS AND CHRIS CUNNINGHAM

information processing or inferencing is being done by the hearer. The additional premises in (18) are then those provided by the hearer, the ones he believes the speaker may believe (and, in something like the classic Gricean manner, believes that the speaker intends him to attribute to the speaker in the inference process). So, on that view, the model proposed (to use Al-psychology terminology with­ out apology) is of the hearer!s model of the speaker. This is most important, and we shall argue later that one can­ not analyse human dialogue without some clear notion of the "ownership" of beliefs and inferences. As to (ii), it is clear that the hearer may, in fact, attribute a set of beliefs to the speaker quite different from those in (17), yet still derive (18), and hence "see the relevance" of the mad-passer-by!s remark. An alternat­ ive set would be (17'): (17!)a. The Royal National Lifeboat Institution is a charity that provides cheap holidays for poor elderly people (cf. the lexical and real semantics of "The Salvat­ ion Army"). b. The speaker is a shabby elderly looking person. c. Someone who already has holiday provision, will not need the services of a charity providing it. d. Someone who has no need of the services of a charity cannot be expected to subscribe to that charity (the same as 17e.). It is of no importance that (17')a

is a false belief,

as was (17)e, for belief attribution in communication can­ not require that we attribute to others only beliefs we happen to hold. If that were needed communication would rapidly collapse, and we could not talk to those of polit­ ical or linguistic

beliefs opposed to our own. (17')b has

THEORY OF RELEVANCE

389

a special quality, in that it appears to be a belief of the hearer ABOUT the speaker, rather than about the speaker's own beliefs. This proposition might then be an inappropriate inferential construction by the hearer, if he believed the speaker was unaware of his own appearance. Decisions like this, as to what "logical space" or "mental environment" the inferences are to take place in (e.g. within the hear­ er's view of the speaker, the speaker's view of the hearer's view of the speaker, or...) are, in our view, vitally important, though they have no place in SW's scheme. We shall return to this example later. The issue for the moment is simply the wrongness of SW's "Those who fail to see the relevance of (16)b. are precisely those who have failed to derive the contextual implication in (18)..." If that means "derived by any pre­ mises at all" then it is not an interesting claim, and does not require the setting out of (17) to show it, for it could hardly be false. If, on the other hand, it means "derive by means of (17) as context" then it is quite false as we have shown (and that they intend the latter is shown by "the hearer has to supply (at least) the premises of (17)"). The difference here is much more than the lack of a footnote from SW noting the possibility of other derivat­ ions: it is that they do not really accept that such infer­ ences must be some particular individual's inferences, and so they feel free to opt for an "objective" set of hypo­ thesised premises, ones they believe correct. A moment's more ingenuity will provide yet further interpretations, which may have more contextual implicat­ ions than the use of (17) or (17'), a matter that will be important in the next section: (17'')a. People who spend holidays with relatives nor­ mally murder them sooner or later.

390

YORICK WILKS AND CHRIS CUNNINGHAM

b. The Royal National Lifeboat Institution is a charity (same as (17)b.) c. People who commit murder do not also sub­ scribe to charities which yields not only (18) but (18''). (18'') The speaker of (16)b. is likely to murder or has already murdered his sister. The premises must be derived from the "content of the utterance", since together, SW say, they form the context (i.e. 17a-e in the example). The utterance in question here, is (16)b :- No thanks, I always spend my holidays with my sister in Birmingham.

Clearly, that provides the

cue for premise (17)a. - Birmingham is inland. and (17)c

But (17)b

are derived from the previous utterance (16)a.

SW do not mention this difference.(17)a-c

are facts assoc­

iated with the utterance to be interpreted and the previous one. But where do they come from? If they are somehow held in the hearer's memory, why is it this set of facts which is retrieved, and not sundry others which are in some way related to the content of the utterance? We could have, for example, from among a large set of candidates:(17''')a. Birmingham has a high crime rate. b. Flags given out by charities have pins in them. c. Flags with pins in them could be used to hurt people. If we look closely at premises (17)d. and (17)e : d. Someone who spends his holidays inland has no need of the services of the Royal National Lifeboat Institution. e. Someone who has no need of the services of a charity cannot be expected to subscribe to that charity

THEORY OF RELEVANCE

391

we can see that they are quite different in kind from (17)a-c , for they are not facts but general principles. Premise (17)c cannot be derived from the content of the utterance in the rather straight-forward way that (17)a-c of SW's set can, nor does it follow directly from premises (17)a-c

either. Moreover, if (17)d-e are in the hearer!s

memory (as his own or anyone else's beliefs) they will have to be constructed in some way by the hearer. There are two problems for SW's theory here; while they claim that these premises have i)

to be supplied

(ibid: 73),

they give no explanation of how and why these

rather complex premises are constructed, ii) their system does not and cannot make allowance for the cost in time that must be expended on such construct­ ions. We shall return to such cases below under the title "ethynemic constructions". Also unfortunate for this example, and for SW's gener­ al theory, is the fact that for all practical purposes (and that is what SW have said they are concerned with) the hearer does not need fully to "understand" the utterance in question in order to get the pragmatic point of it, and would probably not bother deriving any contextual implicat­ ions at all. In the situation described, "No thanks, I al­ ways spend my holidays with my sister in Birmingham" simply constitutes a polite refusal, plus additional comment in­ cluded to avoid appearing too curt or to slide out from a slightly

embarrassing situation. Suppose that, for such

reasons, the comment had been: "Sorry, I've got no money on me." On SW's account the hearer would have to infer that, if the speaker had money, then he would subscribe. This is very probably false, and only the most gullible of flag-

392

YORICK WILKS AND CHRIS CUNNINGHAM

-sellers would draw such an inference. Thus it seems again that using the context of the utterance in the way describ­ ed by SW will not tell the whole story of comprehension. DEGREES OF RELEVANCE So far, the broad outlines of SW's world will be famil­ iar to anyone who has worked within what one might loosely call the "inference paradigm of understanding" within psychology, linguistics or AI since 1965: understanding is to be seen in terms of the ability to draw appropriate in­ ferences, and the coherence-cum-internal relevance of a dis­ course is to be seen as the applicability to it of whatever formal system the writer happens to be advocating. Things get much more interesting when claims move to a quantitative stage and specify the inferences appropriate for understanding in terms of processing resources avail­ able, or offer quantitative selection of the MOST appro­ priate inference or inferences. This has been done within AI/psychology under the term "resource limited processing" (e.g. Norman & Moore, 1975) and, as a special case within the field of natural language processing as "least effort" or "preference" theories (e.g. Bień, 1980, Wilks, 1975). SW's version starts with their principle of Relevance (ibid: 75), which is "the single principle governing every aspect of comprehension": "The speaker

tries to express the proposition which

is the most relevant one possible to the hearer" This is a counsel of perfection, of course, and may, like all such principles, not be adherred to by the speak­ er. Here we shall understand it in reverse, as it were, in keeping with the hearer-orientated aspect of SW, as a principle that the hearer is well advised to believe the speaker is observing. But the reader should note in passing

THEORY OF RELEVANCE

393

that this is not a trivial gloss since this principle, un­ like SW's treatment of the mad-passer-by example, refers only to the speaker's intentions. It is one of the continuining themes of this paper that such perspective switching by SW leads to muddle throughout. The content of the principle is on p. 7 5 and we shall call it the Claim: "Of two utterances that take the same amount of pro­ cessing, it is the one with the most contextual implic­ ations that will be the more relevant; and, of two utterances which have the same number of contextual implications, it is the one which takes the least amount of processing that will be the more relevant". Some care is needed now, in interpreting this as we declared we would (in terms of processing by a hearer), since "relevant" in the claim and the principle may not refer to the same items. This is a problem for SW, one which we will not attempt to solve comprehensively for them here. The real origin of the problem is the ultimate incompatib­ ility of a Gricean speaker's-intention approach and one based on hearer's-information-processing, let alone with an abstract non-directional model based on notions of Chomskyan competence, although SW preserve elements of all these. It is in fact possible to have a non-directional (as between analysis and generation) computational model of dialogue (e.g. Reichman 1978) but only on assumptions far from those of SW. The Claim will be our central text here, and we shall approach it a number of times from different directions. Let us first take it at face value, without worrying too much about the precise meaning of "amount of processing", and look back at our alternative treatments of the mad-passer-

394

YORICK WILKS AND CHRIS CUNNINGHAM

-by. Applying the SW claim context (17'') wins (as the one appropriate to utterance (16b)) since it gives rise to not one but two contextual implications. And that is not a conclusion SW should find congenial. This rapid application of the Claim to (17) etc. is not unfair: the Claim actually refers to the comparison of alternative utterances (not differing interpretations of the same utterance, nor differing contexts for the same utterance), and only covers cases where either the process­ ing effort is the same in the cases to be compared, or the number of contextual implications is the same. Whereas the comparison we just have made in applying the Claim to (17), (17'), (17'') is between differing contexts for a given utterance and therefore, in order to come within the Claim, we have had to assume, with SW at this point, that the con­ texts have all been established without processing costs. However, on p. 74 SW explicitly allow the Claim to cover the "same utterance in different contexts", and since they are at this stage of their exposition assuming cost-free context establishment, our application of the Claim cannot be attacked on the ground that different contexts might require differing amoungs of computational effort to establish them (though that is in fact the case). Much difficulty will arise with SW's central metaphor of "information processing", set in a form of research in which no real or hypothetical information processing is done. That is not, of itself, a fatal defect, since much AI work is not accompanied by real working programs but is only a form of procedural speculation. However, there are dues to pay if one is to work in that environment, and they require much detailed consideration of what real processes might come under such a Claim. The particular problem here concerns the establishment of the context itself e.g. the

THEORY OF RELEVANCE

395

processes needed for hearer to locate or construct the sets (17), (17'), (17'') etc. It is highly dangerous to speculate on what the "information processing costs" of such location would be in any model of the hearer, for they would depend very much on the memory organization or data-structure employed. But, as we shall see, we cannot avoid this issue entirely. THE THALASSEMIA EXAMPLE SW do not in fact assume throughout their paper that the context is given and, in the next section, we shall give their version of that point. But we shall first set out their principal illustrative example for their case on degree of relevance: the thalassemia example (p. 74-5): "... compare utterances (19)-(21) in a context consist­ ing of (22a-c): (19) Susan, who has thalassemia, is getting married to Bill. (20) Susan is getting married to Bill, who has thala semia. (21) Susan, who has thalassemia, is getting married to Bill, and 1967 was a very poor year for Bordeaux wine. (22)a. People who are getting married should consult a doctor about the possible hereditaryx risks to their children. b. Two people both of whom have thalassemia should be warned against having children. c. Susan has thalassemia. In this context both (19) and (20) carry the contextual implication that Susan and Bill should consult a doct­ or, but (20) also carried the implication that Susan

396

YORICK WILKS AND CHRIS CUNNINGHAM and Bill should be warned against having children. The sentences in (19) and (20) are almost identical in linguistic and lexical structure. Suppose that process­ ing involves identifying the propositions expressed by the utterance, computing its non-trivial implications, and matching each of these against the propositions in the context to see if further non-trivial implications can be derived. Then (19) and (20) should take roughly equal amounts of processing. In this context, since (20) yields more contextual implications than (19), with the same amount of processing, it should be more relevant than (19) and this seems intuitively correct. By contrast (19) and (21) have the single contextual implication that Susan and Bill should consult a doct­ or.

(21) is linguistically more complex than (19).

On the above assumptions about processing, (21) will thus require more processing and be predicted as less relevant in context; again, this prediction seems to be intuitively correct". First, let us remind ourselves of the essence of the Claim: for a given amount of processing (which so far does NOT include accessing the context), the most relevant utter­ ance is the one producing, from the context, the most con­ textual implications. Conversely, for the same number of contextual implications (from two "competing" utterances) the one requiring the least amount of processing is the more relevant. We have restated the Claim because its application to this example is quite different from its earlier applicat­ ion to the mad-passer-by example. A clear defect in the Claim, to be noticed in passing, is that it applies only when one of two equalities holds. Without descending to mere quantitative considerations, we suggest that it will

THEORY OF RELEVANCE

397

therefore not apply to most cases! Let us put four immediate considerations against this: a) what serious quantitative information processing comparison can be going on in which a hearer can be consid­ ered as comparing the relevance of two DIFFERENT UTTER­ ANCES (outside explicit psychological laboratory test, that is) as distinct from realistic situation where a hearer compares two alternative interpretations of a SINGLE utter­ ance so as to select the more relevant? A hearer is normal­ ly offered an utterance, not several between which to choose, so what consequences for the information processing of a real hearer could possibly follow from SW's supposit­ ions? Certainly, a speaker could compare possible outputs before speaking, and it is supposed to be a consequence of the Claim that a speaker should compare his own possible utterances against the context he intends the hearer to construct. That may have a certain plausibility, but re­ quires that what SW are up to is modelling speaker-informat­ ion-processing, contra hypothesis. b) Even on the fixed-context assumption, what clear characterisation of "non-triviality" do SW give that will allow instantiation, from (22)a, as a "non-trivial context­ ual implication", but not mere repetition (from p=> p) or arbitrary conjunction? Repetition will be especially tempt­ ing for a hearer seeking to maximise implications by in­ ferring the context itself and in toto. Defence against arbitrary conjunction is normally taken, as in the work of Belnap et al., to require a theory of relevance itself, so SW will not want to invoke that since it is just that they claim to be establishing! Gazdar and Good (1982) have al­ ready made a full criticism of SW along these lines and we will not duplicate it here.

398

YORICK WILKS AND CHRIS CUNNINGHAM

c) It is true, as they note, that (20) produces the, undoubtedly non-trivial, implication that the couple should consult a doctor, but surely that must have required a great deal of processing to obtain: the location and applic­ ation of an AND rule, and the location and application of some form of modus ponens to (20) + (22b) AND (22c)? Or do SW somehow imagine that the actual inference itself, normal­ ly set out as explicit steps, does not require processing effort? If they believe that, they should not use the meta­ phor at all, but leave it in the hands of others. If the processing required by inferencing is taken into account, as they seem to intend, then the assumption of equal processing effort required by (19) and (20) is plain­ ly ludicrous, since the accessing of a rule of conjunction and its application is a clear quantifiable cost. In gener­ al, the safe assumption, other matters being equal (which, of course, they are not), is that more implications will require more processing effort, exactly the opposite of what the Claim suggests. d) The real issue has been ignored till now: what basis can there possibly be for assuming, as SW still do at this point in their account, that (19),(20) and (21) all access the same context, and that that access will require the same effort in all three cases (for, if it does not, then the comparisons drawn so far fall to pieces)? Since (22c) is already present explicitly as part of (19), the context invoked by (19) cannot include (22c) as it does here (or, if it does, then other parts of utterances can occur explicitly in contexts, which will have other dis­ astrous consequences for SW's Claim). Hence the assumption that (19) and (20) "require the same processing effort" will be quite false if that effort includes context-location (and in the next section they concede that it does). If the

THEORY OF RELEVANCE

399

effort is not the same then we have another case where the Claim fails to apply, since neither equality is satisfied, although that, too, can give no comfort to SW. Again, (21) with its mention of Burgundy must draw into the context propositions about wine. They can only be kept out by the indefensible assumption that this simply IS the context, achieved cost free, and declining to dis­ cuss the matter further. If the mention of Burgundy did draw in that wider context, at correspondingly greater effort than those drawn in by (19) and (20), then again the whole comparative farrago would fall to bits, since the assumption that (21) will yield the same number of context­ ual implications as (19) may well turn out false. The same ingenuity that the flag-seller showed with what he heard could certainly produce a context for, and a reply to, (21). Gazdar and Good (198 0) have pointed out that if a hear­ er has additional or idiosyncratic information about, or interest in, a topic mentioned, then this may well give rise to a great number of non-trivial implications. Had the speaker said, in place of (21), "Susan who has thalessemia, is getting married to Bill, who is a wine expert", then a wine expert hearer could have correctly inferred a great deal about Bill. SW, lacking any clear notion of what a hearer or speak­ er believe, separately or about each other, have no defence to this and make none in their reply to Gazdar and Good in (Smith ed. 1982). In a properly founded theory, of course, it would be a requirement that the inferencing was constrain­ ed to a sub-space of assumptions that was the hearer's view of what the speaker believed. That would meet Gazdar and Good's point, for a hearer behaving appropriately would then not draw such "expertise" inferences if he believed the speaker did not know he was in possession of such inform-

400

YORICK WILKS AND CHRIS CUNNINGHAM

ation. Only in that way can context finding deal with apparently irrelevant input (and real errors by speakers and hearers can of course occur concerning such assumpt­ ions) . We return to this point under the heading Cognitive Solipsism below. SW want a single principle or mechanism to explain the "near identity" of interpretation between interlocut­ ors (ibid: 84), and at the heart of their "single princip­ le" is the idea of maximal relevance, on which the speaker tries to express the proposition which is maximally relev­ ant to the hearer. But as the above example has shown it is still quite unclear how it is that the speaker can select his utterance according to this principle, and exactly how the hearer comes to see which is the most re­ levant/intended interpretation. Remember that SW began by saying of the Gricean maxims "they are far too vague, and it is equally vague exactly how they are to be used" (ibid: 71) . LOCATING THE CONTEXT At this point in the exposition, SW introduce a prin­ ciple that draws the whole theory closer to reality, but has the effect of vitiating much of what has gone before. They now face up to the consequences of the fact that locating contexts is a matter of processing effort: "We want to argue ... that the search for the inter­ pretation on which an utterance will be most relevant involves

search for the context which will make this

interpretation possible. In other words, determination of the context is not a prerequisite to the comprehens­ ion process, but a part of it."

(ibid: 7 5)

In defence of this late realisation, they note

that

(p. 76) "Most pragmatic accounts assume that the context

THEORY OF RELEVANCE

401

for the comprehension of a given utterance is fixed in advance", which can only be understood if whole areas of research are just ignored. The paper by Joshi that accomp­ anies SW's in (Smith, ed. 1982) should help here. But are we faced at this point by the withdrawal of a minor simplifying assumption, one which can now be with­ drawn without ill-effect, or is it rather that the recogn­ ition that context finding costs processing effort (one accepted ab initio by all those in the AI-psychology trad­ ition who have discussed the issue) makes nonsense of the Claim and everything based on it? It may be worth reproducing here some of SW's propos­ al to meet the newly discovered need: "There is, to begin with, an initial context which consists of the interpretation of the immediately preceding utterance in the conversation or the text. The hearer attempts an interpretation in this context by looking at what contextual implications can be derived from it. If these are lacking or not consider­ ed sufficient to satisfy the principle of relevance, the context can be expanded several times, in three different directions. The hearer can add to the con­ text what he remembers of utterances further back in the conversation... he can add encyclopaedic knowled­ ge which is attached in his memory to the concepts present in the utterance or in the context... or he can add to the context information about whatever he is attending to at the same time as the conversation is taking place... each expansion of the context creates new possibilities of deriving contextual im­ plications. On the other hand, these extensions in­ volve an ever-increasing cost in amount of processing ..." . (ibid: 76)

402

YORICK WILKS AND CHRIS CUNNINGHAM

This is the sort of thing (we imagine) to make workers in the field of the psychology of memory weep with frustrat­ ion: how can anyone write of adding encyclopaedic informat­ ion in memory without any notion of what that involves in either computer or human memory? As Gazdar and Good (1982) have pointed out, SW cannot avoid such points by retreat to a classic competence account of these matters, for theirs is a full blown performance account if it is anything. If this latter quotation is important for SW, how can it be that the principle and Claim were set up without reference to it, given how devastating this new realism is for what has gone before? SW's answer is the notion of "accessibility of con­ text". This is expressed in two slightly different ways: (i) "The speaker must have grounds for thinking that the hearer has an easily accessible context in which a sufficient number of contextual implicat­ ions can be derived." (ibid: 77) (ii) "In order to be relevant to the hearer, the speak­ er must make more or less specific assumptions about the contextual propositions that the hearer has access to or can infer." (ibid: 80) SW assume that "people are fairly accurate in the assumptions they make about what others know" (ibid: 81). They also imply that a good assessment of the hearer's interests is a way of guaranteeing an easily accessible con­ text. In an informal sense, all these comments about the speaker-hearer situation are quite acceptable, the fact that conversations actually proceed is sufficient evidence to support these assumptions. They do not in themselves, however, constitute an adequate explanation of anything, and are no assistance in supporting SW's theory of utterance interpretation in preference to any other.

403

THEORY OF RELEVANCE At a later stage in their argument, SW give great

importance to the Accessibility' feature by using it in a redefinition of relevance: "relevance is a function of the amount of processing and hence of the accessibility of the context required to derive contextual implications"

(ibid:

83). It is not at all clear what sort of status we are supposed to assign to this statement. Their remarks on 'accessibility1 are informal, yet they also use this term in what appears to be a formal definition in their theory of relevance. Both the actual and assessed amounts of knowledge and interest pertinent to a given utterance (the factors affect ing accessibility of context according to SW) must affect the amount of processing involved. SW seem to want to have their cake and eat it by being unclear about whether these factors are part of their theory of relevance, or simply facts about conversation which are not contradicted by it. The status of their comments here are similar to those made on context expansions, and also have the effect of weaken­ ing rather than strengthening their claim for their "single principle governing all aspects of conversation". A shift from criticism to charity may be more appro­ priate here. The following definitions would at least re­ move the obvious absurdities in SW's position: New principles for hearers (and assumed by speakers to be in use by hearers): (1) MAXIMISE the number of contextual implications drawn for some total given processing effort (as to some arbitrary maximum per unit time, say, one to be empirically determined) for interpretation of input, location of con­ text, and drawing of implications; (2) MINIMISE amount of processing for context finding so as to leave more available for drawing contextual implic-

404

YORICK WILKS AND CHRIS CUNNINGHAM

ations under (1). These proposals, too, may well not stand up to any detailed examination, but they are at least procedurally plausible and not self-evidently self-contradictory (though they are not independent), whereas, on SW's account, the hearer is under an injunction simultaneously to maximise and minimise the same sort of thing, in that the hearer is to minimise effort OVERALL, while at the same time maximis­ ing the number of contextual implications, whose product­ ion must require effort. Our hunch, for what it is worth, is that SW should go simply for a least-processing-effort theory (as we have ourselves) for it is not clear what help having more contextual implications is. As we saw with the flag seller, the number can vary unpredictably with contexts, however chosen, and we believe their addiction to the notion comes from a false view of maximising information in commun­ ication (a point we return to at the end of this paper). In their reply to Gazdar and Good (1982), SW deny that they assume "processing speed is constant" (ibid: 106) and so, they argue, processing per unit time considerations are not relevant. They also claim that context-finding process­ ing is non-inferential, as distinct from the inferential processing that draws implications, and so no considerations drawn from summing processing capacities are appropriate in discussion of their system. This is, we believe, the merest obfuscation, and nothing SW write gives any support to the view that there are separate processing capacities that can­ not be added. It is certainly possible that the brain does have separate capacities for the two processes, ones that cannot be added, but claiming that would require some shred of physiological evidence, above and beyond the undoubted convenience to SW. Moreover, such a concession would be quite at variance with the discovery that "determination of

THEORY OF RELEVANCE

405

context is not a prerequisite to the comprehension process but a part of it (ibid: 7 6 ) " . Given (1) and (2) above, some version of SW's Claim could now be reinstated, and they might well feel that the­ se new principles are just what they intended and have expressed (cf. "... we would suggest that the amount of processing tends to remain roughly constant throughout a stretch of discourse" (p. 77), if that is taken to mean processing per unit time!), but they certainly have not, and indeed are unable to do so, because the Claim was stat­ ed on the basis of the false assumptions about cost-free context location, and cost-free implications. In fact, the only concession to genuine information processing in the whole of SW's paper is the assumption (p. 74) that the identity of processing effort for (19) and (20) is a funct­ ion only of their similar lexical and syntactic structure, but which is a matter of complete irrelevance to logical complexity and inferential effort. It would be hard to find in the recent linguistic literature a clearer example of the bad effects of the hangover of beliefs in the autonomy and primacy of syntax. COGNITIVE SOLIPSISM We drew attention earlier to the fact that SW have no clear or consistent appreciation of the fact that real in­ ference must go on somewhere and, in a model of human communication, that must be in a hearer model or a speaker model (where each may, and must, contain models of the other). This lack surfaces in their paper at intervals, as when they discuss the "common ground", which is their vers­ ion of "mutual knowledge": the set of facts, serving as potential contexts, that both conversational participants know. But that very conception is a static, abstract,

406

YORICK WILKS AND CHRIS CUNNINGHAM

Chomskyan one: a psychologist, or benevolent third party, might risk claims about the nature and extent of what two people both know: the intersection of their sets of facts, and known by both to be the intersection. But that would be a matter of research or revelation. When two people commun­ icate normally, there is no such objective common ground available ("Did you hear a sound, I did?". "Look at the candle set between us"). I speak to you on MY assumptions about what you believe or can construct, and they remain MY assumptions, often wildly, and undiscoverably, wrong as to YOUR actual beliefs. So, in the SW world, there is no method or theory to explain the difference between: (a) premises/beliefs believed by the hearer to be held by the speaker when speakingT (b) general knowledge known by the hearer and believed by him to be imputed to him by the speaker. (c) enthymemic beliefs, not retrieved by the hearer from anywhere, but constructed by him, as part of a context, and imputed by him to the speaker on the basis only of what is said (i.e. not believed previously by the hearer, nor previously believed by him to be held by the speaker). No theory that fails to reflect these distinctions in some well-motivated way can be taken seriously in this area. SW see the problems at intervals, as when they first declare that inference must take place within "common ground", but then, as is their habit, they withdraw the restriction in the face of examples (pp. 77-78) in which, quite clearly, a speaker can utter something requiring knowledge he does not have but believes the hearer to have (I do not know your phone number but can speak on the assumption that you know it) . Investigations that make the distinctions (a-c) above

THEORY OF RELEVANCE

407

are the very stuff of much recent work in AI, both logical -theoretical (e.g. Moore 1975) and programmed-applied (e.g. Perrault and Allen 1980). In the important latter series of papers the notions of belief, speech acts, plans, reference, inference, and models of the other, were ingeniously explor­ ed, and the above distinctions were fundamental. A crucial advantage of work like the latter type is that it can show the contiguity of relevance and inference with human plans and goals, with what it is someone is talking FOR. There is no way this can be done with SW's work, and that must be a serious short coming in a general theory of pragmatics. More relevant to the matter in hand is the series of papers by Wilks and Bien (1979, 1983) now being programmed, on the procedural location and manipulation of belief spaces or environments in which such inferences can go on, and which simultaneously constrain relevance, inference and the beliefs of the other. This work is very elementary as yet, and nothing what­ ever follows from it, but it makes none of the elementary errors of SW. In such a system, one can actually compute (albeit by naive algorithms) the appropriate environment for inference as when, for example, an expert in a subject (say a physician specializing in genetic disorders) listens to a layman who will not have that knowledge. So, one can show how such a physician, say, listens to a patient who does not know the consequences of being told he has thalas­ semia, so the physician must interpret what is said in an environment that is not at all his own belief set about disease and children, but his model of the speaker's, impo­ verished and probably false, belief set. Moreover, this work is set within a general claim about least-effort human processing of language, and goes right back to early work (Wilks 197 5) on inference chaining in

408

YORICK WILKS AND CHRIS CUNNINGHAM

utterance interpretation, in which programs interpreted utterances by establishing the shortest possible chain of inferences from context to utterance. This, as we said, was elementary stuff, but at least it was not incoherent: it was a wholly plausible assumption that locating and applying a shorter chain of inference would require less processing effort than the provision of any longer chain from context to utterance. The greatest lacuna in the list above is (c) and it is, at the moment, no more than an aspiration for all research workers, but SW do not see this because they have no proced­ ural grasp of what can and cannot be done, so for them the very difficult is all one with the well understood. That problem is the one of "enthymemic construction", where a hearer understands, as he does much of the time, not by retrieving knowledge/belief of his own, nor of be­ liefs he imputes to the speaker, but in virtue of a third category: beliefs (context, in SW's sense) that must be constructed and subsequently attributed to the speaker so as to make the implications follow from what is already in the inference space, e.g. the utterance. In writing this paper, we produced, out of the blue, more or less plausible sets (17') and (17'') to prove (18) but as to how this process can be modelled algorithmically no one has much to say. It is the great problem in SW's theory and all those working in this area. RELEVANCE AND CONVERSATION SW relax their definition of relevance later in the paper and begin to use the word in an everyday, non-theoret­ ical way, but without an acknowledgement of the fact or an explanation of the relationship between the theoretical and non-theoretical uses. They want to establish that the

THEORY OF RELEVANCE

409

assumption that a remark is relevant leads to the correct inference being made, as in the following example (ibid: 79) . Ann: Will you have a glass of brandy? Omar: You know I am a good Moslem. If we assume Omar's remark to be relevant in the general sense i.e. that it is the answer to the question, then we have to establish how it is relevant in the technic­ al sense. This, SW suggest, could lead us to a set of pre­ mises thus: Brandy is alcohol Moslems do not drink alcohol and an inference that Omar will not have a glass of brandy. Thus the process of finding out how an utterance is relev­ ant, built on the assumption that it is so, will lead to an inference which demonstrates the relevance of the re­ mark, and because it does this, it must be the inference intended by the speaker. The assumption that Omar's answer is relevant to the question is obviously an essential one, otherwise the process of interpreting his remark in terms of an answer would not begin. But if Omar was not certain that Ann knew good Moslems did not drink alcohol, he could not be confident that his answer would be interpreted as a refusal. If Ann thought that good Moslems always drank brandy she would misinterpret the remark. So, the everyday idea of relevance gets the interpretation process started, but in this example the crucial factor in arriving at the correct answer is individual belief sets. If Ann's beliefs about Moslems are wrong she will get the wrong interpretat­ ion, but may not assume the remark is irrelevant. If she has no. knowledge at all about Moslems she may do one of two things, either assume that the remark is irrelevant,

410

YORICK WILKS AND CHRIS CUNNINGHAM

or assume that the speaker is expecting her to know some­ thing she does not. SW discuss this, like all their examp­ les, as if the possibility of irrelevance never arises, and so have nothing to say about it in this particular example. But the other two possibilities (getting the wrong interpretation, or realising that there is a relev­ ant intended interpretation, but not knowing which it is) break the connection proposed by SW between relevance, in their everyday sense, and intended interpretation. Let us consider (ibid: 80) Ann: Did you like the book you were reading? Bob: I don't much like science fiction. Required context premises: The book Bob was reading is a book of science fiction. Unlike Grice, in his remarks on the maxim of relation, SW do not acknowledge the fact that there are changes of topic in conversation. When this does happen the first utterance on the new topic may have no relevance at all to the utterance it follows. Suppose, for example, that Bob's remark above was in fact not about the book he was reading, but that he was ignoring Ann's question and commenting on the science fiction programme on the television at the time. If Ann then interprets his remark as suggested by SW, she will misinterpret - and the full version of the princip­ le of relevance contains nothing which could provide a check on the misinterpretation. Equally, in the case of Bob's book and the TV programme both being science fiction, the theory predicts that he is referring to the book, while this may not necessarily be the case. Another related matter not discussed by SW is the exploitation of the same type of lexical cue, with the deliberate intention of being irrelevant: For example:

THEORY OF RELEVANCE

411

Do you like dogs? Well I live on the Isle of Dogs. If the hearer searches for the intended interpretation in the way suggested by SW - assuming the relevance of the reply as an answer to the question - but lacking knowledge about the Isle of Dogs, he may construct false premises about the place in order to establish relevance, e.g.: The Isle of Dogs must have a large population of animals, hence its name. The speaker lives there so he must like animals. Now, such plays on words are often used in conversation for a myriad of purposes. The above example could be used as a joke if the second speaker was relying on the hearer's knowledge that the Isle of Dogs had no significant connect­ ion with animals. If he was depending on a lack of such knowledge, the utterance could be made on amuse himself, or other listeners. The single principle of relevance takes no account of these facets of conversation. BACK TO THE FIRST PRINCIPLE: THE "FULL" PRINCIPLE OF RELEVANCE One of the odd features of the SW paper (one that may disappear in the fuller version to come) is its internal development: a manifestly unsatisfactory principle is stat­ ed, backed by an oversimplified

Claim. Later, more realist­

ic assumptions are added (e.g. the cost of locating con­ text) , which utterly disrupt and devalue the principle and Claim. Yet somehow SW never see this or recognise it by re­ stating more complex and adequate principles (we offered some in a spirit of helpfulness ourselves). To the very end of the paper, the fiction is kept up that the principle, properly understood, will deal with all difficulties (p. 82):

412

YORICK WILKS AND CHRIS CUNNINGHAM "however, the full principle of relevance provides a natural check on the conclusion that a certain back­ ground proposition was intended to be used. Suppose that to achieve a particular interpretation for an utterance a certain background premise would have to be used; and suppose that if this premise had been intended by the speaker, the fact that he intended it to be used would have been more relevant than the con­ tent of the utterance itself. This interpretation would automatically violate the principle of relevance be­ cause, by hypothesis, the proposition expressed by the utterance was not the most relevant proposition avai­ lable to the speaker in the circumstances. Hence. this interpretation could not have been intended by a speak­ er attempting to observe the principle of relevance... Or take the case where an utterance has two interpret­ ations, one with a normal degree of relevance and the other with considerably more than that. In the simpl­ ified model (i.e. where the speaker actually expresses the "most relevant" proposition YW) the latter should be chosen. However, the full principle of relevance will (correctly) generally select the former. For instance, imagine two mothers chatting,, and one say­ ing to the other: (36) My son has grown another foot. This can mean either that her son has grown bigger or that he has grown an extra limb. The second interpret­ ation is of course much more relevant. However, if this were the interpretation intended, then the princ­ iple of relevance would have been grossly violated. The speaker could indeed, if her son had become threefooted, produce a much more relevant utterance than a

THEORY OF RELEVANCE

415

mere statement of the fact... hence the full principle of relevance unquestionably selects the less dramatic and less relevant interpretations in such a case". It is very difficult to discuss this example in the light of SW's written statements of position. First, the exposition of the example involves radical speaker-hearer switch of a kind that makes it difficult to bring to bear the very considerations they have set out. It is not possib­ le to contrast, in a model of a hearer, what the speaker really intends with what the hearer takes him, on the basis of analysis and inference, to be intending (though there can be subsequent contradiction and explicit establishment of misunderstanding, of course). Such oppositions only make sense in the world of commonsense talk, or that of a God's-eye view of the intentions of others, but not in SW's model. Hence, the observation on (p. 82) "if this premise had been intended by the speaker, the fact that he had intended it to be used would be more relevant than the content of the utterance itself" can have no place in SW's hearer-modelling theory, nor can "the fact that" have its relevance assessed at all in the implication-metric they have set up at some length! If one were to accept the God's-eye view of intention for one moment, it is hard to see why a speaker conforming to SW's principle does not utter both the sentence he does AND the whole context in which he intends it to be taken! (in fact, we do sometimes do this in some degree when saying something that could be misunderstood). To do that in gener­ al would conform to SW's principle in a way nothing else would, simply by dramatically minimising the effort required from the hearer. How then do they explain that that is not normally done?

414

YORICK WILKS AND CHRIS CUNNINGHAM

Secondly, the repeated assertion that the unusual interpretation of (36) , the third foot, is "much more relev­ ant" can have no basis in either common sense nor, more importantly, in their computational notion of relevance. They simply assert it and the reader's 30b is to work out what they could possibly mean! The third foot interpretat­ ion would certainly be more unusual in the biological world we happen to live in. Perhaps they have in mind some version of an "information-redundancy theory" of meaning (e.g. Bar-Hillel and Carnap 1953), on which less probable messages carry more information, and that is broadly correct. But how can it be connected to their own theory? We can either assume that the two interpretations of (36) are assessed against different contexts, or that there is a single context against which both interpretations are to be matched. On either view, plausible candidates for the context would be: (36)a. Children change height (positively) with time. (36)b. The cardinality of the major human organs does not increase (except teeth). Whether we consider these a joint set or not, the "height interpretation" of (36) will produce no contextual implications from (36)a, but is confirming evidence for it, if one needed that. And that is almost certainly what does make it the correct interpretation, though SW's theory of relevance has no place for that being a ground of relevance, which is yet another material shortcoming. SW may be assuming that (36) follows from some appro­ priate form of (36)a (e.g. "If it is later, some child is taller"), but that is not so: If I know (x) (Px => Qx) and someone gives me Qa, nothing follows. If the woman had said "My son is older than he was" then an appropriate form of

THEORY OF RELEVANCE

415

(36)a would yield the (technically) non-trivial implication that her son was now taller, but people do not talk like that much and for good reason. But the "organ interpretation" of (36) actually contra­ dicts (36)b and so produces either no implications or all possible ones, depending on your logic. If it produces none, then there is nothing to choose between the interpretations, which makes it hard to reconcile what SW say with what their theory predicts. If (36) + (36)b. produce all possible implications then the organ interpretation should win out on their metric principle, but probably not quite in the way they had in mind, since obtaining all possible implicat­ ions tends to be thought a trivial matter, and if the number is infinite generating it would be a considerable informat­ ion processing task! All this, of course, is on the cost-free-context assumption, which SW abandoned but did not replace by anything more adequate, so the reader has no other way to examine this example in more detail. It is not possible to give any clearer sense to SW's claims about this example because they lapse back into the view that the hearer can know what the speaker intended in­ dependently of what he said. However, one can begin to put together what they might have meant: that if the child had grown another organ, then (36) is not the way to say it, because the normal hearer will take (36) the normal way. There are many theories of understanding that can show why the normal (height) interpretation would be made, but SW's does not seem to be among them, since it seems always to achieve the organ interpretation, or nothing. A theory that based relevance on the confirmation of existing generalisations held by the hearer could get the normal result; so could one based on the statistical like­ lihood of interpretations (though that would be a very bad

416

YORICK WILKS AND CHRIS CUNNINGHAM

general basis for a theory of understanding). On a histor­ ical note, it may be added that systems that attempt to identify correct interpretations with least-effort and least-information have had difficulty with this class of examples before. In Wilks (1975) there is discussion of the problem for a least processing effort theory that equates that notion to finding the appropriate interpretat­ ion as the one bearing the least information. The example used was: He was named after his father where, in our society, it is almost certain that a person will have been named at a point in time after the one at which his father was named, and so that interpretation carries virtually no information. Yet it is not the correct one, for a normal listener would opt for the "sameness

of

name" interpretation which is genuinely informative. Hence the problem for such a theory is that the wrong interpretat­ ion is selected, yet the point is exactly the one SW make with (36). The solution suggested by one author in 197 5 was that the principle of least inferential effort and least content should apply to utterance interpretation UNLESS that cont­ ent was zero or close to it. That was a suggestion at least in keeping with the general form of the theory, and did not require breaking the paradigm by claiming access to what the speaker should have said, had he meant the odd inter­ pretation. SW continue to reject any idea that their theory would survive better as an information minimising rather than maximising view (ibid: 121): "By contrast, our system expects the hearer to maximise new information of a certain type (the contextual implications of the utterance)". But this is a complete misunderstanding of information: anyone

THEORY OF RELEVANCE

417

who really wanted to maximise the information in what they hear would do so by interpreting it as a sequence of un­ connected sentences without mutual reference or coherence which would, in turn, lead to a wide (incoherent) context and more potential implications. It is a triviality that mutually irrelevant items contain the most information. CONCLUSION While the idea of a single principle in a field as difficult as pragmatics has obvious appeal to the theorist, and the notion of relevance certainly has power in shaping utterances and their interpretations, it is clear that SW's exposition of a theory of relevance does not yield a single principle. It contains a multiplicity of ideas, but the terms are still too ill-defined, and their application to everyday conversation has too loose and pre-theoretic a fit. Their claims to be offering more than Grice, more than psychologists and AI specialists, while claiming psycholog­ ical plausibility for their theory are not defensible. A great deal of thought, and dare one say, programming, would be required for any progress with these difficult and complex issues. But it is our contention that an account as muddled and flawed as SW's may not yet be a starting point.

REFERENCES Bar-Hillel, Y. & Carnap, R. 1953. "Semantic

Brit,

J. Philos.

of

Science

Information".

4. 147-157.

Bien, J. 1980. "Articles and Resource Control". Proc. IJCAI 83. 675-677. Clark, H. & Carlson, T. 1982. "Speech Acts and Hearer's Beliefs". Mutual Knowledge ed. by N. Smith. 1-37. Lon­ don : Academic. Gazdar, G. & Good, D. 1982. "On a Notion of Relevance". Mutual Knowledge ed. by N. Smith. 88-100. London: Academic.

418

YORICK WILKS AND CHRIS CUNNINGHAM

Grice, H. 1975. "Logic and Conversation". Syntax and Semant­ ics: Speech Acts, Vol. 3 ed. by Cole & Morgan. 41-58. London: Academic. Joshi, A. 1982. "Mutual Beliefs in Question-Answer Systems". Mutual Knowledge ed. by N. Smith. 181-199. London: Academic. Norman, D. & Bobrow, D. 1975. "On Data-Limited and Resource-Limited Processes". Cognitive Psychology 7. 44-64. Levinson, S. 1983. Pragmatics. Cambridge: CUP. Moore, R. 1975. "Reasoning from Incomplete Knowledge in a Procedural Deduction System". MIT-AI Lab., AI-TR-347. Perrault, R. & Allen, J. 1980. "A Plan-Based Analysis of Indirect Speech Acts". Amer. J. of Comput. Linguistics 6. 167-182. Reichman, R. 1978. "Conversational Coherency". Cognitive Science 3. 283-327. Schiffer, S. 1972. Meaning. Oxford: Clarendon Press. Smith, N., ed. 1982. Mutual Knowledge. London: Academic. Sperber, D. & Wilson, D. 1982. "Mutual Knowledge and Relev­ ance in Theories of Comprehension". Mutual Knowledge ed. by N. Smith. 61-87. London: Academic. Wilks, Y. 1975. "A preferential Pattern-Seeking Semantics for Natural Language Inference". Artificial Intellig­ ence 6. 88-111. Wilks, Y. & Bien, J. 1979. "Speech Acts and Multiple Environ­ ments". PROC.IJCAI 79. 451-455. Wilks, Y. & Bien, J. 1983. "Beliefs, Points of View and Multiple Environments". Cognitive Science 8. 120-146.

RUDI CONRAD Leipzig

BEDEUTUNG UND FUNKTION VON GEGENFRAGEN IM DIALOG

Das gewachsene Interesse an Struktur und Organisationsprinzipien von Dialogen, wie es sich z.B. in Untersuchungen zur Konversations- und Gesprächsanalyse dokumentiert, ist sicher nicht nur auf die Entwicklung der Textlinguistik schlechthin zuriickzufiihren, sondern hat wohl eine sehr wesentliche praktische Grundlage in den Forschungen zur künstlichen Intelligent, die sich mit der Schaffung natiirlichsprachiger Dialogsysteme MenschMaschine beschäftigen. Dass dabei Fragen als sequenzstiftende Sprechakte eine besondere Rolle spielen, liegt auf der Hand, entsprechen die durch sie geschaffenen Dialogstrukturen Frage: Antwort doch am klarsten dem Grundschema Sprecherinitiative: Replik. Gegenstand des vorliegenden Beitrages sollen jedoch nicht solche klassischen Frage-Antwort-Strukturen sein, sondern Dialogsequenzen des Typs Frage, : Frage2 d.h. Folgen von Satzen , die zumindest von der ausseren Erscheinungsform her nicht dem Schema Initiative: Replik entsprechen, sondern einem etwas anderen Grundschema, das man als Sprecherinitiati­ ve: Gegeninitiative des Horers darstellen könnte. Die Untersuchung solcher Gesprächs(teil)strukturen ist unter verschiedenen Gesichtspunkten interessant. Zum einen ist damit natiirlich eine tiefer gehende Analy-

420

RUDI CONRAD

se bestimmter Fragesatzstrukturen verbunden, bei der das Zusammenspiel zwischen syntaktischen, semantischen und pragmatischen Komponenten der linguistischen Beschreibung deutlich gemacht werden kann. Zum anderen scheinen damit aber auch einige grundsächtzliche Probleme des Aufbaus von Dialogen angesprochen zu sein. Versteht man unter einer Replik des Hörers auf eine verbale Initia­ tive des Sprechers eine vorwiegend passive Reaktion in dem Sinne, dass diese in wesentlichen Punkten durch den initiierenden Akt des Sprechers vorausbestimmt ist, wie dies z.B. bei Frage-Antwort-Paaren eindeutig der Fall ist, so liegt demgegenüiber bei Sequenzen des Typs Frage: Gegenfrage zunachst offenbar ein aktives Horerverhalten vor, das nicht oder wenigstens nicht in gleicher Weise durch den vom Sprecher geausserten Sprechakt prädeterminiert ist. Dieser Unterschied wird durch das in gesprachsanalytischen Untersuchungen gangige Prinzip des Rollenwechsels zwischen Sprecher und Horer (turn-taking) eigentlich nicht erfasst. Die meisten Gesprachsanalytiker beschaftigen sich vorwiegend und vordergründig mit sozialen und anderen ausseren gesprachsorganisatorischen Aspekten von Dialogen, die zweifellos wichtig sind. Dabei wird jedoch die Rolle der inneren logischen Struktur von Dialogen vernachlassigt. Urn Gesprache z.B. in Computerprogrammen simulieren zu konnen - aber auch fur andere Zwecke -, genügt es eben nicht, einfach den Sprecherwechsel als konstitutives Element von Dialogen zu konstatieren und eventuell noch weitere aussere Bedingungen fur die Übernahme eines Gesprachsschritts (wie z.B. freies turn-taking oder Erteilen des Wortes durch den Gesprachsleiter im organisierten Gesprach usw.) zu beschreiben. Es muss vielmehr ergründet werden, wieso ein

GEGENFRAGEN IM DIALOG

421

Gesprachspartner überhaupt die Absicht entwickelt, einen Gesprachsschritt zu machen, was ihn dazu veranlasst, sich zu Wort zu melden. Dabei ist die aussere Organisationsform des Dialogs nur eine Seite. Wie kommt es aber dazu, dass ein Partner etwas zu sagen hat, wenn er an der Reihe ist? Wie kommt es zur Entwicklung von Sprecherinitiativen? Die Teilnahme der Partner am Gesprach beschrankt sich ja nicht bloss auf Repliken, die durch vorangehende Sprechakte oder seine Stellung im Gesprach gefordert werden, sondern er bringt auch von sich etwas aktiv ins Ge­ sprach mit ein. Deshalb muss eine wichtige Aufgabe darin bestehen, die Genesis, den Mechanismus der Entstehung von Sprecherinitiativen im Dialog zu erforschen. Dariiber kann m.E. eine Analyse der logischen Beziehungen zwischen den Ausserungen der Partner im Dialog einigen Aufschluss geben; denn sie fiihrt zu inneren Zusammenhangen, die als Triebkraft fur die Entwicklung eines Dialogs angesehen werden konnen. Es scheint, dass logische Widersprüche eine wesentliche Grundlage fur den Aufbau und den Ablauf von Dialogen bilden, und zwar vor allem dadurch, dass sie die Grundlage fur die Motivation und Auslosung von Sprecherinitiativen darstellen. Dialogpartner vergleichen angebotene Äusserungen des anderen mit den eigenen Annahmen, Behauptungen, Bewertungen etc. in bezug auf Übereinstimmung/Nichtubereinstimmung, und dies lost zustimmende oder widersprechende Reaktionen, Argumentationen, Begründungen, Rechtfertigungen usw. aus bzw. kann sie auslosen. Eine Analyse verschiedener Arten von Gegenfragen in natiirlichen Dialogen zeigt, dass deren Funktionen im Dia­ log in der Regel recht vielschichtig sind. In den meisten

RUDI CONRAD

422

Fallen ist ihre interrogativische Funktion durch andere, z.T. indirekt ausgedruckte Funktionen überlagert, was vor allem auf dem Wege iiber bestimmte Annahmestrukturen (zumeist epistemische Sprechereinstellungen) erreicht wird, die mit der Semantik des Fragesatzes verbunden sind. Eigentlich gibt es nur sehr wenige Gegenfragen mit ausschliesslich interrogativischer Funktion im Dialog. Dazu gehoren als die wohl am meisten verbreitete Klasse die sogenannten Echofragen, wie jeweils F2 in den folgenden Beispielen: (1) F1 : F2

:

Wohin gehst du? Wohin ich gehe?

A2:

Ja.

A2 :

In die Stadt.

(2) F5: F2: A2:

Kommst du mit? Ob ich mitkomme? Ja.

A1 :

Nein.

Es handelt sich dabei urn Rückfragen, die die Struktur des Fragesatzes F, ganz oder teilweise wieder aufnehmen. Sie kommen dadurch

zustande, dass entweder die

Frage F1 insgesamt oder einzelne ihrer Bestandteile vom Hörer nicht oder nicht einwandfrei identifiziert werden konnten, so dass sich vor der Beantwortung der Frage F, fur ihn die Notwendigkeit ergibt, eine Zwischenfrage zu stellen, die den Zweck hat, sich Klarheit iiber F1 zu verschaffen. Echofragen sind in diesem Sinne als frageaktbezogene Reaktionen einzustufen oder, genauer gesagt, als Fragen

iiber

eine

Frage,

d.h. als metakommunikative

Sprechakte. Dass dem so ist, bezeugt u.a. auch die in

GEGENFRAGEN IM DIALOG

423

(1) belegte Beantwortung der Echofrage "Wohin ich gene?", also einer scheinbaren Erganzungsfrage, mit. "Ja." Sowohl diese für Ergänzungsfragen nicht zulassige Antwortform als auch die Struktur der Echofragen, die. z.B. auch bei (2) der syntaktischen Struktur eingebetteter Fragesatze entspricht (vgl. z.B.: Weisst du, wohin ich gehe? Weisst du, ob ich mitkomme?), deuten darauf hin, dass die Echofragen in (1) - (2) elliptische Verkiirzungen einer expliziten Frageform "Du fragst, wohin ich gehe?" bzw. "Du fragst, ob ich mitkomme?" sind, in der der metakommunikative Charakter dieser Fragen deutlich zum Ausdruck kommt. Wie auch der in (1) - (2) dargestellte (mogliche) Gesamtablauf des Dialogs zeigt, sind Echofra­ gen ein spezifischer Typ von Zwischenfragen, durch die die natiirliche Frage-Antwort-Struktur F1 : A1 nicht zerstort oder aufgehoben, aber auch nicht gesattigt, sondern durch eine dazwischengeschobene zweite FA-Struktur F2 : A 2 expandiert wird. Daraus ist auch die rein interrogativische Funktion der Echofragen klar zu erkennen, und es ware sicher abwegig, ihnen noch gewissermassen gewaltsam über mehrere Stationen eine zusatzliche Interpreta­ tion als "indirekte ausweichende Antwort" zuweisen zu wollen, d.h. sie ausserdem noch gleichzeitig als Kundgabe dessen zu bewerten, dass der Horer nicht in der Lage 2 oder nicht willens ist, auf F1 zu antworten . Erweiterte Dialogstrukturen des gleichen Typs, d.h. F1

: F2 : A2 A1 , können auch durch nicht metakommunika-

tive Zwischenfragen zustande kommen, z.B.: (3) F1 : Wann fahren wir auf den Flugplatz? F 2 : Habt ihr schon eure Sachen gepackt? A 2 : Ja.

RUDI CONRAD

424

A1, : N a , dann fahren wir urn 10 Uhr. Auch hier handelt es sich urn das Einholen einer (notwendigen) zusatzlichen Information, bevor F1 richtig beantwortet werden kann. Sie ist nur nicht metakommunikativer A r t , sondern betrifft einen

aussersprachlichen

Sachverhalt. Der Fragecharakter ist auch hier offensichtlich, und eine Interpretation als "Antwort" in dieser oder jener Weise würde der Tatsache, dass die Beantwortung von F1 nur verschoben wird, zuwiderlaufen. Im Unterschied

zu diesen beiden Arten von Zwischen-

fragen scheint bei alien anderen Gegenfragen eine Doppelinterpretation als Kombination einer Frage und einer damit direkt oder indirekt ausgedrückten Form der Antwort unvermeidlich. Eine relativ selten anzutreffende, aber deshalb nicht minder interessante Klasse von Gegenfragen, die in

(4) und

(5) reprasentiert ist, konnte man als "fra-

gende Antworten" bezeichnen: (4) F1 : Welches grosse Ereignis hat in letzter Zeit bei uns stattgefunden? F2: (5)

Die

Eishockey-Weltmeisterschaft?

F : Was soll ich tun? F 2 : Vielleicht solltest du das Buch noch einmal lesen? In beiden Fallen ist F2 zweifellos ein Fragesatz,

und dessen interrogativische Funktion diirfte unbestreitbar sein. Zugleich handelt es sich aber auch nicht einfach urn eine Frage, denn die Gegenfrage F2 enthalt in ihrem propositionalen Teil eine Struktur, die einer moglichen Antwort auf F, entspricht, und steht somit auf

GEGENFRAGEN IM DIALOG

425

dieser Ebene zu F1 ebenso unbestreitbar in einer Frage-Antwort-Beziehung. Der Unterschied zu "gewohnlichen" Antworten besteht lediglich darin, dass die Antwort A(F1 ) nicht als Behauptung gegeben, sondern zugleich ihrerseits in Frage gestellt ist. Diese Form der Antwort kommt dadurch zustande, dass sich der Horer seiner Ant­ wort nicht sicher ist, d.h. dass er nur über eine Annahme, eine Vermutung verfugt, und da diese Vermutung aus seiner Sicht in ziemlich hohem Grade unsicher ist, stellt er sie vorsichtshalber gleich selbst in Frage. Die semantische Struktur einer solchen fragenden Antwort muss al­ so offensichtlich aus mehreren Komponenten bestehen: ei­ ner Fragestruktur und einer damit verbundenen Annahmestruktur, die beide auf die gleiche Proposition bezogen sind. Die Annahmestruktur drückt eine epistemische Sprechereinstellung aus. Die allgemeine Form einer solchen Struktur fur (4) und (5) ist: (6)

ANNEHM(H, A ( F 1 ) ) A

?(H, A(F1))

(Lies: "Der Horer nimmt an, dass A(F,)) gilt, und fragt, ob A(F 1 ) zutrifft"). Da in solchen fragenden Antworten die propositionale Struktur des Antwortsatzes explizit enthalten ist, dürften sie als eine Sonderform der direkten Antworten anzusehen sein; durch die gleichzeitig ausgedruckte Einstellungskomponente können sie jedoch nur den Status ei3 ner einschrankenden (direkten) Antwort haben . Der weitaus überwiegende Teil von Gegenfragen, die in naturlichen Dialogen anzutreffen sind, hat wahrsheinlich den Status indirekter Sprechakte mit den vielfaltigsten Schattierungen bis hin zu rhetorischen Fragen. Thre zusatzliche Interpretation als verschiedene Arten von in-

426

RUDI CONRAD

direkten Antworten - neben der interrogativischen Funktion, die m.E. erhalten bleibt - erfolgt meist über eine spezifische Annahmekomponente, die in der Regel als ein Ergebnis von Schlussoperationen auf der Grundlage der gegebenen Semantik des Fragesatzes und anderer Ebenen des generellen und situativen Wissens zustandte kommt. Mit anderen Worten, fur die richtige Bewertung der Funktion der meisten Gegenfragen im Dialog ist ausser der sematischen Reprasentation von F2 selbst ein allgemeiner Wissenskontext und/oder ein aktueller Situationskontext notwendig, der bestimmte Prämissen fur die in Gang gesetzten Schlussoperationen liefert. Dies soll an einigen etwas komplizierteren Beispielen haufig vorkommender Ge­ genfragen des Typs WARUM p? / WARUM - p? demonstriert werden: (7) F1,: Warum fragst du immer? F 2 : Warum soll ich denn nicht fragen? (8) F1 : Was für einen Zug soll ich denn machen? F 2 : Warum ziehst du nicht mit dem Springer? Solche WARUM-Fragen werden häufig als rhetorische Fragen empfunden, dürften aber durchaus nicht immer bloss rhetorisch gemeint sein. Vor allem ist zu ergriinden, wodurch sie als Gegenfragen so besonders geeignet erscheinen. Aus der semantischen Struktur der WARUM-Fra­ gen einschliesslich der mit ihnen verbundenen Prasuppositionen allein kann diese Verwendung offenbar nicht erklart werden. Fur das normale Verstandnis wie auch fur die indirekte Interpretation von WARUM-Fragen scheinen drei Arten von Prasuppositionen und deren Relation zu einem Situations- bzw. Wissenskontext von Wichtigkeit:

GEGENFRAGEN IM DIALOG

427

(i) die Präsupposition der Gültigkeit von p: WARUM p → p', d.h. die Frage "Warum fragst du?" setzt voraus: "Du fragst." (ii) die Existenzprasupposition "Es gibt einen Grund q fur p", d.h. die Frage "Warum fragst du?" setzt vo­ raus, dass es einen Sachverhalt q gibt, der als Ursache von p (= "Du fragst") gelten kann. (iii) die pragmatische Präsupposition des Nichtwissens, d.h. "Der Sprecher weiss keinen Grund q fur p". Vergleichen wir nun die indirekte Interpretation von Fragen wie F2 in (7) und (8) mit diesen Prasupposi­ tionen, so ist festzustellen, dass diese eher einer Paraphrasierung entspricht, die die Prasuppositionen (i) und (ii) ausschliesst: "Ich sehe keinen Grund fur p, also nehme ich an, dass p nicht gilt". Wahrscheinlich liegen dieser Verwendungsweise unterschiedliche mögliche Verhaltensweisen in einer Situation des Nichtwissens zugrunde: Wer keinen Grund fur p kennt, kann entweder annehmen, dass es einen gibt, und dann eine reine Informationsfrage nach diesem Grund stellen, oder er kann annehmen, dass es keinen gibt und dass deshalb p auch nicht gelten kann. Fragt er im letzteren Falle trotzdem, muss er damit rechnen, dass seine Annahme korrigiert wird. Er kann aber auch dariiber hinaus annehmen oder gar wissen, dass der Horer auch keinen Grund kennt; dann darf er da­ mit rechnen, dass seine Frage rhetorisch verstanden wird, weil ein Widerspruch zwischen den Prasuppositionen und dem angenommenen Wissen von Sprecher und Horer besteht, so dass beide die Frage nur als "pro forma" gestellt auffassen. Schliesslich konnen beide auch von einem gesicherten Wissen dariiber ausgehen, dass p nicht gilt; dann ist

428

RUDI CONRAD

auch die Annahme eines Grundes fur die Gültigkeit von p überflüssig, und damit ist ein allgemeiner Wissenskontext gegeben, in dem die Frage selbst von beiden als uberfliissig empfunden wird: Prasuppositionen und Allgemeinwissen stehen im Widerspruch. Dies ist die klassische Situation einer rhetorischen Frage. Bei der nun folgenden Darstellung dieser Zusammenhange an den Beispielen (7) und (8) ist zu beachten, dass der zu beschreibende linguistische Sachverhalt durch zusatzliche Begleitumstände verschleiert bzw. weniger durchsichtig gemacht werden kann, die die Analyse erschweren. Bei WARUM-Fragen und einigen anderen, ahnlich gelagerten Gegenfragen wird haufig die zur Diskussion stehende Proposition in der Gegenfrage leicht verandert, so dass nur scheinbar auf die eigentliche Frage eingegangen wird. Dies liegt z.B. im Falle von (7) vor, wo mit F "Warum p?" gefragt wird, in F 2 aber nicht mit p operiert wird, sondern mit einem Sollsatz (=SOLL(p)), in dem p ausserdem durch das Weglassen von "immer" nicht unwesentlich modifiziert ist, so dass sich als Struktur der Gegen­ frage F 2 WARUM(SOLL(-p!))? ergibt. Diese Frage nun wird in einen allgemeinen Wissenskontext gestellt, der allgemeingültige Richtlinien enthalt, wie "Man kann doch fragen" (=MöGLICH(p')), "Es gilt nicht, dass man nicht fragen soll" (=~(SOLL(~p'))). Diesen ausserlinguistisch gegebenen Prämissen widerspricht die in F2 WARUM(SOLL(~ p'))? involvierte Prasupposition SOLL(~p'), so dass sich, da dies auf Grund des Allgemeinwissens nicht gelten kann, auch die Frage nach einem grund (verbunden mit der Annah­ me, dass es einen gibt) eigentlich erübrigt. Dies ist die Grundlage fur eine Uminterpretation der Frage F2

GEGENFRAGEN IM DIALOG

429

als eine rein rhetorische Frage. Die auf diese Weise indirekt ausgedrückte Feststellung "Ich darf doch fragen", die durch eine Gleichsetzung von ~(SOLL(~p')) und DARF(p') zustande kommt, erweist sich aber als keine echte Antwort auf F , da sie weder p (= Du fragst immer) enthalt noch auf den gefragten Grund eingeht. Es wird mit diesem indirekten Sprechakt also gar nicht auf die wortliche Bedeutung von F, reagiert, sondern der Horer postuliert seinerseits von vornherein eine indirekte Interpretation von F1 im Sinne einer generellen Vorschrift "Du sollst nicht immer fragen." Und genau dazu würde eine indirekte Bedeutung von F2 als ~(SOLL(~p)) die Negation, d.h. eine Zurückweisung, darstellen. Die Tauschung liegt dann nur noch darin, dass p (= Du fragst immer) stillschweigend durch p' (= Du fragst) ersetzt worden ist. Das bedeutet, dass der Horer mit F2 eigentlich ein Ausweichmanover vollzieht, statt auf die tatsachlich gestellte Frage sowohl in ihrer direkten als auch in ihrer indirekten Bedeutung zu antworten. Es bleibt nun dem Fragesteller von F1 iiberlassen, ob er dies erkennt und den Dialog ensprechend weiterfuhrt, oder ob er F2 wortlich nimmt und nun seinerseits auf F2 direkt antwortet. Auf analogen, aber von solchen Verdunkelungstaktiken freien und daher etwas einfacheren Wegen lässt sich fur F2 in (8) letztlich die Empfehlung "Zieh (doch) mit dem Springer" ableiten, wobei allerdings die Fragebedeutung von F2 starker im Vordergrund bleibt. Der Horer sieht keinen Grund fur ~p und nimmt daher an, dass p (= Du ziehst mit dem Springer) moglich ware. Er ist sich aber dessen nicht sicher und fragt deshalb vorsichtig,

430

RUDI CONRAD

womit er sich zwei mögliche weitere Ablaufe offen hält: Kann der Sprecher einen Grund fur ~p nennen, ist damit die Frage F2 direkt beantwortet, und sie erhalt den Sta­ tus einer Zwischenfrage, da die Antwort auf F, weiterhin offen bleibt. Kann der Sprecher jedoch auch keinen Grund fur ~p finden, so wird er zu der gleichen Annanme neigen, nämlich MÖGLICE(p), und die Frage als indirekten Rat interpretieren, mit dem Springer zu Ziehen. Die dabei zu vollziehende Kette praktischer Schlusse müsste ungefahr folgendes Aussehen haben: "Es gibt keinen Grund fur ~p, und p ist möglich, also ANNEHM(p), also !p". Dieser Fall ware dann gleichbedeutend mit einer Interpretation von F2 als indirekte echte Antwort "Du sollst mit dem Sprin­ ger Ziehen." Die hier meist nur skizzenhaft behandelten wenigen Beispiele, die das Spektrum der moglichen Arten und Funktionen von Gegenfragen bei weitem nicht ausschöpfen, gestatten dennoch einige grundsatzliche Schlussfolgerungen. Es muss zunachst unterschieden werden zwischen solchen Fragen, die als echte Gegeninitiative gedacht sind und durch ihre Funktion als Zwischenfragen eine zwischenzeitliche Erweiterung der Dialogstruktur zur Folge haben, d.h. zur Einbettung einer neuen FA-Struktur in eine bereits vorhandene FA-Struktur führen, und solchen Gegen­ fragen, die in dieser oder jener Weise eine Antwortreplik darstellen und damit eigentlich die mit der ersten Frage eingefiihrte FA-Struktur sattigen, zugleich aber potentiell eine neue, nachfolgende FA-Sequenz eroffnen oder als Antwortsatz den Ansatzpunkt für eine anderweitige Fortsetzung des Dialogs bilden. Wahrend sich die Funktion der ersten Gruppe von Gegenfragen eindeutig auf ihre Rolle als erotetischer Sprechakt konzentriert, sind

GEGENFRAGEN IM DIALOG

431

die Gegenfragen der zweiten Gruppe in der Regel durch eine überaus grosse Vielschichtigkeit der semantischen und pragmatischen Funktionen charakterisiert, die sie im Dialog zu erfüllen haben. Nur in relativ wenigen Fal­ len handelt es sich urn eine Doppelfunktion von divekter Antwort und Frage. In der Mehrzahl der Fälle ist die interrogativische Funktion der Gegenfrage mit zusatzlichen Interpretationsmöglichkeiten als irgendeine Form der indivekten Antwort verbunden. Dabei gibt es haufig unterschiedliche Auslegungsvarianten, die jeweils nur auf einem allgemeinen oder speziellen Wissenshintergrund bzw. auf dem Hintergrund aktueller Situationskenntnisse erschliessbar sind. Besondere Aufmerksamkeit dürfte da­ bei einem speziellen Teil sowohl des aktuellen situativen Wissens als auch des generellen Hintergrundwissens zu widmen sein, den man als "ungesichertes Wissen" bezeichnen konnte, weil er generelle oder subjektive Annahmen, Vermutungen usw. sowie subjektive Bewertungen von Sachverhalten und Tatsachen enthalt. Da diese Art von Wissen offenbar auch eine grosse Rolle beim Ablauf der praktischen Schlusse spielt, die zu Interpretationen von Satzen als indirekte Sprechakte führen, dürften auch verschiedene Arten von Wahrscheinlichkeitsfolgerungen und unsicheren Schliissen generell besonders zu beachten sein. Die Aufdecken solcher Zusammenhänge bis ins kleinste Detail, die hier nur angedeutet werden konnte, ist zugleich eine unerlassliche Vorbedingung fur tiefere Einsichten in die innere Struktur von Dialogen.

432

RUDI CONRAD

ANMERKUNGEN 1

Wobei Frage und Frage von unterschiedlichen Partnern im Rahmen des gleichen Dialogs geäussert werden.

2

Damit soil nicht in Abrede gestellt werden, dass mit Echofragen auch das Nichtverstehen einer Frage vorgetäuscht werden kann. In diesem Falle fällt die funktionelle Bewertung der Frage im Dialog naturlich anders aus; aber dazu ist auch ein verändertes Bedingungsgefüge notwendig, in dem z.B. das Aufrichtigkeitsprinzip der Kommunikation verletzt ist.

3

Bei der Klassifizierung der Antwortarten beziehe ich mich auf R. Conrad, Studien zur Syntax und Semantik von Frage und Antwort, Studia grammatica XIX, Berlin 1978, S. 66ff.

JANUSZ S. BIEN Warsaw

ARTICLES, WORD ORDER AND RESOURCE CONTROL HYPOTHESIS

ABSTRACT The paper elaborates the ideas presented in (Bień 1983). The definite and indefinite distinction is viewed as a manifestation of the variable depth of nominal phrase processings: indefinite phrases are represented by frame pointers, while definite ones by frame instances in­ corporating information found by memory search. In general, the depth of processing is determined by the availability of recources. Different word orders cause different distributions of the parser's processing load and therefore influence also the depth of processing. Articles and word order appear to be only some of several resource control devices available in natural languages.

ARTICLES VERSUS WORD ORDER In the note (Szwedek 1973) the problem was stated in the following way. "There would seem to be in Polish three possible ways of showing the definite/indefinite distinction: (a) pronouns, (b) stress and intonation,

JANUSZ S. BIEN

434

(c) word order. The first is obvious, the second has only recently been mentioned in the case of English (...), the third might be considered absurd." and the following conclusions were presented: "(a) contrary to the general belief word order in Polish is not free (b) it is used as one of the ways to express coreferentiality or noncoreferentiality which in English are indicated by the definite/indefinite distinction." Actually the conclusions are valid also for other Slavonic languages and were known earlier at least in the folklore of the Prague school of linguistics. Never­ theless, Szwedek seems to be the first one to analyse the phenomenon independently of any specific linguistic theory and to support his claims by sufficient empirical evidence in Polish (Szwedek 1976). The Szwedek's astonishment by his conclusions is, in my opinion, fully justified. It is really curious that so different linguistic mechanisms may be functionally equivalent, and the explanation of the phenomenon is a serious challenge to every linguistic theory claiming its generality. ARTICLES Teaching the proper use of articles to the students whose native language does not posses them is one of the most difficult tasks. Despite the effort of both the teachers of English who are and who are not native speakers, no satisfactory teaching methods have been developed. Usually the students are presented with some examples and told to acquire the proper usage by linguist-

ARTICLES, WORD ORDER AND RESOURCE CONTROL

435

ic experience: in many textbooks some typical uses are omitted, especially the "forward-pointing" use of definite article (Leech, Svartvik 1975) and some false generalisation suggested. In hope to find some hints useful for teaching English to Polish students, an experiment was performed by Smolska (19 76) . A short English text (a page from Brian Foster's "The Changing English Language", New York, Macmillan 196 8) was deprived of all articles, and the underscore character was inserted at all the potent­ ial places of their occurrence. Here is the sample of the text: Languages do not exist in _ vacuum but are power­ fully affected by _ social, political, economic and technical change. If _ great many more people are studying Russian today than was _ case only twenty years ago this is not due solely to _ intrinsic merits of that very interesting language but to dramatic rise in _ status and _ power of _ Soviet Union. _ similar tremendous increase ... Next, 4 educated native speakers of English willing to be the subjects of the experiments were found in Warsaw: the reader should appreciate that it was not a trivial task. The subjects were independently asked to fill in the articles when appropriate. The result was rather surprising: the subjects agreed with themselves and the author of the original text in 86 % cases. The Smolska's experiment strongly suggests that the semantic impact of articles is negligible, but of course it does not prove the case. Unfortunately, the native speakers of English are too scarce in Poland to

436

JANUSZ S. BIEŃ

allow an experiment on a larger scale. Therefore the onus lies on those opponents which have no difficulty to gain cooperation of sufficient number of native speakers. For the purpose of the present paper I simply assume the claim to be true. Let us notice now that articles are not just historical remnants, but emersed relatively lately as a new linguistic mechanism, so it is natural to assume that they serve some specific purpose. If they do not affect WHICH meaning is assigned to an utterance, they may influence HOW the meaning is assigned. In other words, it may be the case that articles control in some respect the process of computing the meaning. I think that the proper use of articles generally facilitates processing of an utterance; its impact may be of two kinds: the time needed to understand an utterance may be shorter or deeper understanding may be achieved in the same time. It seems rather easy to design an experiment testing the first hypothesis; it may be much more difficult to design an experiment testing the second one. WORD ORDER Roughly speaking, the syntactic structure of a sentence in Polish (and in other Slavonic languages) is determined mainly by inflectional properties of individual words; for more detailed discussion cf. (Bień, Szpakowicz 1982). If the inflectional properties are unambiguous, then changing the word order does not affect the syntactic structure. For example, all the 24 permutations of the words

437

ARTICLES, WORD ORDER AND RESOURCE CONTROL

(1)

dziewczynka

("girl", nom)

dała

("gave", fern)

chłopcu

("boy", dat)

piłke.

("ball", ace)

are valid Polish sentences Both in phylogenesis

(Szwedek 1976:

56).

(the emergence of language

in humans) and in ontogenesis

(language acquisition by

individuals), it is the spoken language which appears first. In consequence, the language understanding mechanisms are developed under the constraint of real-time processing. It means in particular that the words of spoken utterance are processed in the time sequence and that the written utterances are processed from left to right

(the claim is neither original o r , to the best

of my knowledge, controversial). It is therefore quite natural that the 24 permutations mentioned above differ in their properties despite the identical lexical content and syntactic structure. Nevertheless, it is not an easy task to reveal the nature of differences induced by different word orders. When a native speaker of Polish tries to pronounce the 24 permutations, he will notice immediately that some of them allow for a more or less restricted

choice

of the intonation pattern while others must be pronounc­ ed in a special way. For example, all the permutations ending with the verb, e.g. (2)

Piłke

dziewczynka chłopcu dała.

are best pronounced with a firm, impatient or angry voice; it is closely related to the fact that they sound natural only when used to confirm the state of affairs already known or considered. It may be the case

JANUSZ S. BIEN

438

that for these utterances the contextual information is necessary to guide the parsing process, deprived until the last moment of the important hints supplied by the verb. Another method to pin down the differences between the permutations of

(1) is the question test advocated

in (Sgall et al. 1 9 7 3 ) . For example, the question (3)

Komu dziewczynka dała piłke? ("whom" "girl" "gave" "ball")

is best answered by (4)

Dziewczynka dała piłke chłopcu.

while e.g. (5)

Chłopcu dziewczynka dała piłke.

requires a heavy stress on "CHłopcu", which in turn needs some justification in the context. The most intuitive method used e.g. by Szwedek consists in checking the coherence of the text construct­ ed by preceding the permutation in question by some other sentence. For example, (6)

(4) preceded by

Na podworzu bawił sie chłopiec piłka. ("on"

"yard" "played" "boy", nom

"ball", instr.)

is coherent, because no relation is seen either between "boy", nom of

(6) and "boy", dat of

"girl", nom of

(4) nor between

(4) and the situation described by (6).

Continuing our investigation we would find that the permutation (7)

Dziewczynka dała chłopcu piłke.

is the most universal: it may be used with many intonat-

ARTICLES, WORD ORDER AND RESOURCE CONTROL

439

ion patterns, it answers several different questions and can be coherently used in a variety of contexts. This permutation demonstrates the so called neutral word order of (1). According to (Sgall et al. 1973), the neutral word order is determined by the semantic roles of sentence elements. The existence of standard order of components, determined either by their semantic or syntactic features, would definitely make the job of parsing easier both for humans and computers. RESOURCE CONTROL HYPOTHESIS The resource control hypothesis has been formulated in (Bien 1980) as a further development of the multiple environments model of natural language advocated e.g. in (Bień 1975, 1976a,b). The essence of the hypothesis consists in the claim that some seemingly unrelated linguistic phenomena are manifestations of the sophist­ icated interaction of a few basic factors, one of them being the availability of resources. Below we discuss briefly those aspects of the hypothesis which are relev­ ant for the present paper. We accept the wide-spread assumption that informat­ ion in memory is clustered into frames. At every moment some frames are in the focus of attention; the saliency of other frames may be represented by an appropriately defined distance from the focus of attention. Changes in saliency are viewed as a suitable displacement; such a displacement requires always some time and other resources. For more details of the proposed Displacement Model of Memory see (Bien 1980) . A frame in the strict sense describes a specific object or notion, e.g. a specific copy of a specific

440

JANUSZ S. BIEN

book; more general information is stored in frame proto­ types. Instantiating a prototype usually requires memory search to locate the relevant data; in consequence, it is a process with substantial demands for resources. The frame prototypes are of course interrelated, e.g. the prototype for a book contains a slot for its author, the prototype for an author contains a slot for his/her books etc. The interrelation is represented by means of frame pointers. They are just data structures allowing to locate the appropriate prototype when needed. Constructing a frame pointer requires a negligible amount of resources. We assume that there is some limit for the human language processor capacity and that the linguistic processes must compete for resources. The lower level processes (acoustic, phonetic and syntactic analysis) are data-limited processes, while the higher level processes (semantic and pragmatic processes, including memory search and spontaneous inferences) are resource-limited (Norman, Bobrow 1975). Data-limited processes must have higher priority, otherwise the resource-limit­ ed processes would use up all the resources available: in consequence, the semantic and pragmatic processing uses only the resources left by the lower level processes. A nominal phrase in an utterance is always trans­ formed at first into a frame pointer. Depending on the availability of resources, the pointer can be developed into a prototype, accomodating some additional informat­ ion contained in the utterance. The prototype can be then instantiated by memory search. Finally, the frame instance can be displaced toward the focus of attention.

ARTICLES, WORD ORDER AND RESOURCE CONTROL

441

In other words, the availability of resources determines the depth of processing for a given nominal phrase. Usually the speaker wants the hearer to process the nominal phrase at some specific depth; we will see in the next section how the speaker can control the hearer's allocation of resources. CONTROLLING THE DEPTH OF PROCESSING Most devices for controlling the allocation of re­ sources for the resource-limited processes and, in con­ sequence, controlling the depth of processing seem to be language universals, but in different languages they are used to different extent. The most general way of allocating the resources consists in choosing for a phrase an appropriate distance from the beginning of the utterance; we will call it the time ordinate principle. Putting a phrase close to the beginning results in the early initiation of its process­ ing, and gives it more chances to collect the appropriate amount of resources. We assume that the scheduling of the resource-limited processes is governed by the prefer­ ence for deterministic processes (Kowalski 1979: 100, Bień 1980 : 4) . This strategy distributes the resources to the memory search, the inferences and the focusing in the proportions depending on the context of computation. If a phrase is placed at the beginning of a sentence, the global effect is determined by several factors. We will mention here only the extreme cases. The most typical one is when the phrase refers to some information known already, which is then retrieved and brought to the

JANUSZ S. BIEŃ

442

focus of attention: the information from memory becomes available soon enough to influence the procesing of the rest of the sentence. If the phrase introduces new information, as in (8)

(Bien 1980:15, Szwedek 1981: 74)

A man was coming.

the inferencing tries to establish its relation to the known information; on the other hand the new information becomes the focus of attention. Another case is represented by English cleft sentences of the type (9)

It was John who did it.

where memory search is usually not needed and all the resources are used for focusing and spontaneous inferenc­ ing, which results in the effect of strong emphasis. As it can be seen now, the claim that there is a universal tendency to put the given or definite information at the beginning of the utterance

(Clark, Clark 1977: 548) is

just a rough approximation of the reality. Another general way of controlling the resources is called here the failing expectations principle. We assume that the human parser operates on sequential in­ put according to the Marcus' wait-and-see

strategy

(1974, 1 9 8 0 ) . If the items encountered confirm the semantically or syntactically generated expectations, the parser's processing load is relatively constant, because it is busy with building the overall

structure

of the sentence. If a phrase is encountered which violates the expectations, the parser has to suspend temporarily its activities and release resources, which

ARTICLES, WORD ORDER AND RESOURCE CONTROL

443

allow for deeper processing of a given phrase. In the languages under consideration, the verb is the main supplier of the parser's ecpectations. This explains the fact, acknowledged e.g. by (Sgall et al. 1973: 49) and (Clark, Clark 1977: 548), that the position of a phrase relative to the verb is often crucial. Another empirical evidence for the principle is the impact of distorting the neutral word order, investigated in (Sgall et al. 1973: 59, 64); Sgall describes the effect of moving a phrase forward relative to its neutral position as an increase of the "communicative dynamism" but we view it as allocating additional resources, which may result in memory search or increasing saliency or both. It is worth stressing that the changes of word order discussed here are too subtle to attribute their effects to the time ordinate principle. For obvious reasons, the failing expectations principle is used in free word order languages to much larger extent than e.g. in English. The next way of controlling the resources consists in using demonstratives. We take for granted that the primary use of demonstratives is to draw the hearer's attention to physical objects in his environment. The hearer's strategy to process demonstratives is therefore to assign the top priority to the task (the object pointed should be located as soon as possible, because e.g. it can be moving and the speaker's description may soon become obsolete). Successful processing of demonstratives requires accomodation of the information perceived and not contained in the utterance, therefore the nominal phrase containing a demonstrative should be processed at least at the level of frame instances. The speaker's strategy for using demonstratives includes the

444

JANUSZ S. BIEŃ

requirement for the adequate description of the object, so the hearer may assume the description to be precise and not requiring any additional inferences- Now it should be noted that what we actually perceive e.g. visually is only a little spot (the data from the retinas of our eyes) in our mental image of the environ­ ment, so the hearer's strategy can be formulated simply as: locate the memory representation of the object, assuming that the description is precise, and store the retrieved information in a frame instance as quickly as possible. In consequence, both in English and Slavonic languages the nominal phrase containing demonstratives may refer to the information given earlier in the text; cf. e.g. (Hawkins 1978: 149). The next method of controlling the depth of process­ ing exploits some properties of numerals. The primary use of numerals expresses the cardinality, but the side-effect of it is the low level of processing: it is impossible to construct 1000 frame instances for the phrase "thousand soldiers", it would be useless to construct 10 frame instances for "ten sold,iers", so by analogy no separate instances are created (unless forced by other reasons) for "two soldiers" although it is feasible. We claim that the analogy holds also for the numeral meaning one, i.e. that a nominal phrase contain­ ing the numeral "one" is not processed deeper than to the level of a frame pointer or a single frame proto­ type. It explains why the numeral is occasionally used both in English and Slavonic languages to signal new information. The next very important method applies only to spoken language but had, at least for English, a far-

ARTICLES, WORD ORDER AND RESOURCE CONTROL

445

-reaching consequences also for written language: it is the sentence stress. When we assign the sentence stress to a phrase, we pronounce it more clearly and in a longer time span than usually (the claim is based only on' my intuition, but I am not aware of any evidence contradicting it). The clearer pronounciation makes the task of the acoustic analysis easier, so it consumes less resources; the larger time span generally increases the availability of resources. The global result is the substantial increase of resources available for the higher level processes. As usual, the resources can be allocated to them in several ways. Typically, the in­ ferences and focusing provide the effect of emphasis. On the other hand, unstressing a phrase or a word deprives it of resources. To the best of my knowledge, in all languages articles originated from unstressed numerals or demonstrative pronouns; articles have therfore much in common with their source. Our treatment of articles, which intends to incorporate the results of (Hawkins 1978) by interpreting them in the framework of (Bien 1980), will account for this fact. The definite article in English originated from demonstratives; depriving demonstratives of resources by unstressing them does not allow to assign the top priority, but still the task is the frame instance depth of processing. The process of establishing the frame instance takes now longer, and in the meantime some use­ ful results of spontaneous inferences may become avail­ able; therefore it would be unefficient to stick to the precision requirement. The frame instance may be needed to accomodate either the information retrieved from the

446

JANUSZ S. BIEŃ

memory (the anaphoric use of definite article) or for the elaborate description contained in the utterance (the forward-pointing use), or for other reasons (e.g. the generic use). The indefinite article originated from the numeral; its function is simply to signal that shallow processing is sufficient. In the next section we will try to demonstrate that the Hawkins' notions of inclusiveness and exclusiveness are just the manifestat­ ion of the different depth of processing. ENGLISH EXAMPLES As it is always risky to design sophisticated examples in a non-native language we take both the examples and their meanings from (Hawkins 1978); we preserve the original numbering of examples. At first we will illustrate the basic uses of articles. (3.02) Fred was discussing an interesting book in his class. I went to discuss the book with him afterwards. In the first sentence we have a typical use of indefinite article, which signals that the processing level of the frame pointer to the book prototype would be sufficient; the

attribute "interesting" may be attached to the

pointer or, if there are resources available, it may be accomodated in the instance of the book prototype locat­ ed by means of the pointer. In both cases the nominal phrase sets only shallow processing. In the second sentence we have an anaphoric use of definite article; the book prototype is used to create a frame instance, then the memory is searched in an attempt to further

ARTICLES, WORD ORDER AND RESOURCE CONTROL

447

instantiate it. The antecedent pointer or frame instance is easily found and merged with the instance which initiated the search. (3.03)

Fred was discussing an interesting book in his

class. He is friendly with the author. In the second sentence we have the associative anaphoric use of definite article. The frame instance for the author is created and the memory is searched as before. However, this time only one of its slots ("the author of what") will be successfully matched against the frame instance in the memory. (3.02') Fred was discussing an interesting book in his class. I .went to discuss this book with him afterwards. The example is fully analogous to the preceding one. (3.03') Fred was discussing an interesting book in his class. He is friendly with this author. In this case the adequacy condition of using demonstrat­ ives is not fulfilled, so the memory search misses the book frame instance and at best, when some author frame is sufficiently salient in the memory, the text sounds awkward. (3.02") Fred was discussing an interesting book in his class. I went to discuss a book with him after­ wards . In this text we have two separate pointers or frame prototypes (i.e. frame instances containing only the information stored in the prototype) for "book". Hawkins

448

JANUSZ S. BIEŃ

(1978: 87) is wrong claiming that the phrases cannot refer to the same book, as the continuation is possible "It appeared later to be the same book" which results in collapsing two frame instances into a single one. (3.03") Fred was discussing an interesting book in his class. He is friendly with an author. In. Hawkins1 terms, "an author" (and "a book" in the preceding example) is a non-located (in the shared know­ ledge of speaker and hearer) specific indefinite, which exclude (this is for Hawkins the essential property of indefinite reference) an infinite number of potential referents. This sophisticated mental construction describes, in our opinion, the simple fact that a shallowly processed nominal phrase represents an object unrelated to other instances of the same prototype stor­ ed in the memory, unless such a relation is established by some additional effort; it is not worth mentioning that an unlimited number of the new instances of the same prototype may be created later. The exclusiveness of indefinite reference is more intuitive in the follow­ ing example: (4.16)

I've just decided to inspect a house. I decided not to buy it. A roof was leaking.

Hawkins claims that "a roof" cannot refer to the roof of the house mentioned, because it has to exclude at least one other possible referent; as there is only one roof talked about, the exclusion is impossible and results in the unacceptability of the text. In our opinion, (4.16) is fully analogous to (3.02') and (3.03"), the difference consisting only in the fact

ARTICLES, WORD ORDER AND RESOURCE CONTROL

449

that even a temporary assumption of the non-coreferentially of "a roof" with the inferred roof of "a house" leads to the incoherence of the text under consideration. In consequence, we accept the exclusiveness rule only as an interplay of the shallow depth of processing with general coherence rules. Our position is additionally supported by the Hawkins1 observation that the exclusive­ ness condition disappears with the verbs "to have", "to be" and the set-existential verbs: (4.137)

I have a head.

(4.145)

There is a roof on my house.

(4.158)

Do you remember the other day I was talking about a student called Smith?

Such sentences may be called attention-shifting, because usually they do not introduce new information, but change the saliency of the already known facts. According to Hawkins, the essential property of definite reference is its inolusiveness

(which in case of singular count

nouns turns into uniqueness), e.g. (3.175)

Fred brought the wickets in after the game of cricket.

"is understood as making a claim about all the wickets in question". We think the property originates from the fact that the deep processing of a nominal phrase fully associates it with earlier mentions of the referent. It should be noted that we do not see any essential difference between a prototype and its copy, i.e. a frame instance containing only the information inherited from the prototype. In consequence, we attribute the

JANUSZ S. BIEŃ

450

difference between generic and non-generic to the context, not to the nominal phrase. In the sentences (4.104)

A lion is a noble beast.

(4.106)

Lions are noble beasts.

the nominal phrases are processed analogically as in sentence "I saw a lion", "1 saw some lions"; this view can be traced

back at least to Christopherson (Hawkins

1978: 214). As for definite generics, e.g. (4.132)

The hammer was elaborated very early.

(4.133)

The dog is a friend to man.

we again assume a standard definite reference mechanism, the difference consisting only in the nature of the frame retrieved, which happens to be an appropriate prototype frame. In both cases our analysis seems compatible with the spirit of (Hawkins 1978). In our future work we hope to reinterpret also more difficult cases of nominal phrase reference. POLISH EXAMPLES According to the hypothesis advocated here, the depth of processing in Polish is strongly related to the parser's operation. Until the parser will be de­ scribed with sufficient precision, it is practically impossible to verify the hypothesis. Therefore the examples discussed below have different purpose than those in the preceding section: their intent is to illustrate the problem, not the solution. Let us consider the Polish translations of (3.02). In my opinion, the first sentence is best translated as

ARTICLES, WORD ORDER AND RESOURCE CONTROL

(10)

451

Fred omawiał w klasie interesujaca ksiazke. ("Fred" "discussed" "in" "class" "interesting" "book") Some other possible translations are:

(11)

Fred omawiał interesujaca ksiazke v swojej klasie. ("Fred" "discussed" "interesting" "book" "in"""his" "class")

(12)

Interesujaca ksiazke omawiał Fred w swojej klasie. ("interesting? "book" "discussed" "Fred" "in" "his" "class") The second sentence of (3.02) may be translated, among others, in the following ways (instead of the literal translation of "discussed" we use "porozmawiac o ..." meaning "talk about"). (13)

(14) (15) (16)

(17)

Poszedłem pozniej porozmawiać z nim o tej ksiazce. ("want", 1 per sing "later" "talk" "with" "him" "about" "this" "book") Poszedłem pozniej porozmawiac o niej z Fredem. ( "about" "it", fern "with" "Fred") Poszedłem porozmawiac z nim pozniej o tej ksiazce. (... "talk" "with" "him" "later" ) Poszed^em pozniej z nim o niej porozmawiac. (... "later" "with" "him" ) Poszedłem pozniej o niej z nim porozmawiac. (

"about" "it", fern

)

We can make the following observations. First, definite phrases can be simply translated by pronouns; the grammatical gender of Polish makes them easier to resolve, so they seem to be used more often. Secondly, a corefer-

452

JANUSZ S. BIEŃ

ential phrase occurring at the end of a sentence is either a proper name (14) or is accompanied by a demonstrative (13); in (12) the possesive also seems obligatory. However, the basic question is how to translate (3.02) as a whole. In my opinion, (10) is preferrably followed by (13), (14) or (15), while (11) by (16) or (17). In other words, if the phrase "ksiazka" occures earlier in the first sentence, then the corefer­ ential phrase in the second sentence also tends to occur earlier. It suggests that investigating only the preced­ ing context may be an unjustified limitation: the speaker may decide on a specific word order taking into account not what he has already said, but what he intends to say. It is interesting that neither (10), (11) nor (12) can be naturally followed by (18)

0 ksiazce poszedłem pozniej z nim porozmawiac. ("about" "book"

)

but (19) must be used instead: (19)

0 tej ksiazce poszedłem pozniej z nim porozmawiac. ("about" "this", fern "book" )

This is one of several phenomena which still require substantial investigations. CONCLUSIONS I hope to demonstrate how the resource control hypothesis may provide a "common denominator" for the totally different but functionally equivalent mechanisms. I hope also to supply additional arguments that some

ARTICLES, WORD ORDER AND RESOURCE CONTROL

453

properties attributed to nominal phrase of their meanings (e.g. specific versus generic) may actually reflect the properties of the contexts of their use. REFERENCES Bien, J. S. 1975. "Toward a multiple environments model

of natural language". Advance International Joint Conference

Papers on

of the Fourth Artificial

Intelligence. 379-382. Tbilisi. Bien, J. S. 1976a. "Multiple environments approach to

natural language". American Journal al Linguistics Microfiche 54.

of

Computation­

Bien, J. S. 1976b. "Computational explication of inten-

sionality". International Conference al Linguistics, preprint no. 41.

of

Computation­

Bien, J. S. 1980. "A preliminary study on linguistics implications of resource control in natural language understanding. ISSCO Working Paper no. 44. Geneva. Bien, J. S. 1983. "Articles and resource control". Proc.

of

the

Eight

International

Conference

on

Artificial

Intelligence ed. by A. Bundy. 675-677 . Karlsruhe. Bien, J. S. & S. Szpakowicz. 1982. "Toward a parsing method for free word order languages". Coling 82 Abstracts ed. by E. Hajicova. 37-41. Praha: Univerzita Karlova. Clark, H. C. & E. V. Clark. 1977. Psychology and Language. New York.

Hawkins, J. A. 1978. Definiteness and A Study in Reference and Grammaticality

Indefiniteness. Prediction.

London: Croom Helm. Ltd. Kowalski, R. A. 1979. Logic for Problem Solving. Holland. Leech, G. & J. Svartvik. 1975. A Communicative

of

English.

North Grammar

Longman.

Marcus, M. P. 1974. "Wait-and-see strategies for parsing

natural language". MIT AI Lab Working Marcus, M. P. 1980. A Theory of Syntactic for Natural Language. MIT Press.

Paper 75. Recognition

Norman, D. A. & D. G. Bobrow. 1975. "On data-limited

resource-limited processes". Cognitive 7. 44-64. Sgall, P., Hajicova, E. & E. Benešova. 1973.

Focus

and Generative

Kronberg/Taunus .

Semantics.

and

Psychology Topic,

Scriptor Verlag:

JANUSZ S. BIEŃ

454

Smólska, J. 1976. Unpublished note. Szwedek, A. 1973. "Some aspects of definiteness and undefiniteness of nouns in Polish". Papers and

Studies

in

Contrastive

Linguistics

II. ed. by J.

Fisiak. 203-212.

Szwedek, A. 1981. Word Order, Sentence Stress and ence in English and Polish, WSP Bydgoszcz.

Refer­

LOUISE B. HAMMER Bloomington

CODE-SWITCHING IN COLLOQUIAL CZECH

The term code-switching is used to describe the alternations which take place between one code and another within discourse. The subject of this paper is Prague colloquial Czech ( hereafter PBMČ). In PBMČ the two basic codes are liter­ ary Czech (spisovnd ĉ e ŝ t i n a , hereafter SČ), and common Czech (obecnâ c e s t i n a , hereafter O Č ) . They blend into a unique, unstable conversational style which is used in many semiformal and informal situations. This con­ versational switching differs from the situational type characterized as diglossia, in which distinct codes are employed in particular settings (Ferguson 1964). In conversational code-switching items from the two codes are found in the same sentence (intra-sentential code-switching). The switches are metaphorical or functional in the sense that they may indicate irony, in-group membership, intimacy/formality, or provide other clues to the speaker's attitude towards the topic or other participant/s in the conversation (Blom and Gumperz 1972:425). During the nineteen thirties and forties members

456

LOUISE B. HAMMER

of the Prague Linguistic Circle, in particular Havránek, investigated the relationship between SČ, its spoken form which he called h o v o r o v á č e š t i n a (hereafter H Č ) and oč. Havránek claimed that oč was one of several Czech interdialects and that H £ was a functional stratum of s£, the national language. This spoken variant (H£) , allowed a few of the morphological colloquialisms of OC into the grammar of s£, but was less tolerant of the phonological features, e.g. dekuju/dekujou was admitted as H £ (1936: 119). These views were formulated from the standpoint of language culture (jazyková kultura). Havránek further asserted that the folk language was not the only spoken form of Czech: that literary Czech had, since the turn of the century, once again become a liv­ ing language (1942: 414-416). He was referring to the fact that after the Battle of White Mountain (1620), Czech was gradually supplanted by German in official use and to a great extent ceased being the conversation­ al language of the upper classes and nobility, particul­ arly in the cities (1979: 72). The SČ which had been reconstructed by Dobrovsky became a spoken language during the period of the National Revival: however, its use was more or less limited to intellectual circles. During the latter half of the nineteenth century, SČ was reintroduced as the language of instruction in the schools. Its use in journals, newspapers, and later in the oral media of communication, became widespread. As SČ became more ordinary in public life, it increasing­ ly confronted OČ By the fifties Havránek had somewhat modified his views on the relationship between HČ and OČ. He saw each of the two forms as losing its more marked charact-

CODE-SWITCHING IN COLLOQUIAL CZECH

457

eristic (regional forms in OČ such as -ej-; such forms as the bookish infinitive -ti in sČ) (1951: 45-46). He repeated this observation in the sixties when he perceiv­ ed the situation as open-ended, but continued to view the situation through the framework of language culture (1963: 259). Bělič addressed the problem from a dialectological point of view. His complex of views were in close concordance with those of Havránek, as were those of Jedlicka and TráVnicek, when this question was discuss­ ed in the Journal Slovo a slovesnost in the early nine­ teen sixties. Sgall had given the impulse to this discussion when he claimed that OČ was developing .into the spoken form of the national language and that it was necessary to interpolate elements from OČ into sČ, particularly lexic­ al and phonetic items that were neutral, i.e. not marked­ ly OČ, in order to bring codification and the spoken norm into a more realistic relationship. He pointed out that the concept of a spoken form of the literary language (HČ) was controversial and that in fact, the two codes involved were sČ and OČ and speakers vacillated between them (1960: 16-20). In later articles he asserted that OČ was the main form used in ordinary communication and that the particularity of Czech was not only the limited usage of sČ in ordinary speech, but the very extensiveness of one of its non-literary forms (1962: 38; 1963: 249; 1981: 299) . All of the scholars who engaged in this discussion were united in the opinion that more research was needed on the non-sČ variants of the language and that the term­ inology needed to be made more precise. Although the literary language had been the subject of numerous stud-

458

LOUISE B. HAMMER

ies, relatively few works existed on OČ before Hronek published his description (1972) . It is the form de­ scribed by him that is designated OČ in this paper. PBMČ refers to the speech of the group of educated inform­ ants recorded by me. Some studies which appeared in the seventies adopted the view that colloquial speech or beznĕ m l u v e n ý jazyk was a mixture of OČ and HČ with variabil­ ity extending in both directions (Šmilauer 1972: 21? Jedlicka 1974: 38-44). Examples of these HČ forms are: kupuju (1st per. sg.), nesem (1st per. p i . ) , ŕict (inf.), ŕek sem (masc. sg. past), mliko (Šmilauer 1972: 23; Havra"nek, Jedlicka 1981: 4-8). Since the speakers in this study did use forms which belong to the written or book­ ish form of sČ, e.g. řici; lépe, it was necessary to make the distinction between sČ and OČ. However, it is important to point out that the forms designated as HČ do indeed enjoy a wide acceptance, in contrast to other features of oČ origin. Additionally, they appear in text­ books of the Czech language (Šára, Šárová, Bytel 1969) . This paper is based on data collected in 1981 during seven months of field research in Czechoslovakia. The project was funded by a grant from the International Research and Exchanges Board. Based on census data it was determined that a group of 25 to 30 speakers would provide a representative sample for the study. The quote sample technique was used for their selection. The norm­ ally occurring speech of twenty-eight native speakers was recorded on a Sony 150 cassette recorder using an electret condenser microphone (ECM-16). The recordings were transcribed with the help of native Czech phonetic­ ians and teachers of the Czech language. The informants were limited to persons who satisfied the following

CODE-SWITCHING IN COLLOQUIAL CZECH

459

criteria: 1. born in Prague or within a fifty-kilometer radius of the city. (Two informants were from the central Bohemian region outside this radius, but had lived in Prague for twenty-five years.) 2. be between the ages of thirty and fifty-five. 3. have a university degree or equivalent degree (e.g. engineer) or profession (e.g. writer). 4. be currently employed. Educated speakers were selected in order to insure that the informants not be ignorant of the SC code: how­ ever, as Blom and Gumperz have pointed out, it is possible that intrasentential code-switching is related to university education (1972: 428-431). The participants spoke on two different topics: everyday and cultural, in order to determine the effect of topic on code selection. Some speakers shared a common profession, and for those groups a third, professional topic was elicited. Chloupek has pointed out the importance of theme or topic in conversation (1980: 180). In order that the cultural data not elecit sČ, such as would be heard in a univers­ ity lecture for example, the topics were limited to plays currently running in Prague, e.g. Tři v torn 'Three In That Condition'; artistic films, e.g. Postřiřiny 'Cuttings'; currently popular books etc. As Poplack has pointed out, balanced bilinguals tend to switch more often than non-fluent bilinguals and many of those switches occur intrasententially (1980: 581). Although the term bilingual may not be exact when referring to the Czech situation, which concerns two varieties and not two separate languages: the fact that the majority

460

LOUISE B. HAMMER

of the switches in this data also occured intrasententially indicates that the speakers had a good command of both codes and were able to switch freely from one to the other. The two codes have many features in common. In contrast to the kind of code-switching that involves two different languages, some utterances in colloquial Czech are neutral in that they contain no phonological, morpho­ logical or syntactic features which would identify them as belonging to either of the codes, e.g. Jeho jazyk je modern' 'His language is modern". Among the most strik­ ing differences between the two codes are the pnohological divergencies found in the vowels.

1. o- (in initial position), okno 'window', ona "she" (on a morpheme bound­ ary) naopak "0on the contrary" týden

week , dobry

-i- after c,s,z cizi "foreign" 3

good

vo- (prothetic v) vokno "window", vona "she" navopak "on the contrary' tejden

week , dobrej "good "

cizej "foreign"

é~i~ mléko 'milk', dobré 'good' mliko 'milk', dobri "good" nést "to carry" nist "to carry"

A fourth feature, initial ú-being manifested as the diphthong ou-, is generally listed with the preceding three, but was not in my data. Hronek comments on its rarity and describes it as being in retreat (1972: 27). Since Czech has quantity, it is necessary to mention that in rapid speech, which is common for colloquial

CODE-SWITCHING IN COLLOQUIAL CZECH

461

Czech, changes may occur in vowel length. The vowels which are long in the SC genitive and dative plural end­ ings are frequently short in OČ, e.g. mužu, mužum 'men'. Hronek (1972: 30) has pointed out that this shortening is generally found in the last syllable of the word for the vowels u and i, e.g. domu 'homeward', neni 'it's not', jeji 'her'. He points out that Í is shortened in other positions in certain lexical items, e.g. ŕikala "she said". My data provide the example of u shortening in other positions also, e.g. zustane "it"ll remain". The vowels a and e tend to shorten in root positions in part­ icular lexical items, e.g. pomahat "to help", lekarna "pharmacy". The example mam "I have" frequently occurred with a short or half-long vowel, as did these other high frequency 1st person singular verbal forms: znàm, znam "I"m acquainted with"; vim, vim "I know". Additionally, the -ý- in the oČ adjectival declension was frequently shortened, e.g. dobryho "of the good". As Kucera has pointed out, this shortening occurs in non-contrastive environments where the quantitative distinction is redundant (1961: 92). Vowel lengthening may denote emphasis or irony. Consonantal differences between the two systems are of two types. Syllable structure processes in OČ and PBMČ such as consonant cluster simplification may like vowel length reduction be related to rapid speech. TYPE I: A. Initial clusters, který > kerý "which" kdepak [gdepak] > depak "where ever" kdyz [gdiz] > dyz "when" kdyby [gdibi] > dyby "if, when, in case"

462

LOUISE B. HAMMER vzdycki [vzdicki] > dycky 'always' chces [xces] > ceš 'you want'

B. Clusters in medial position. jestli > jesli 'if' prazdniny > pràzniny ',holidays' vezmu > vemu 'i take' C. Final clusters. It is characteristic of oČ(although not of eastern Moravia) and PBMČ that the final -1 of the masc. sg. past tense of the verb is lost following a consonant, řekl > řek 'he said'. D. The loss of j occurs before a consonant in initial and medial position, e.g. jdu > du 'I go'; pUjdu > pudu 'i will go'. Prothetic j is lost, e.g. ještě > este 'still'. The loss of initial j has become common to all of spoken Czech, but has been codified only in the present tense of the verb 'to be', e.g. jsem > sem 'I am' (Havránek, Jedlička 1981: 47). TYPE II. The second type of change has to do with high frequency lexical items and may be due to voicing assimil­ ation in some cases. A. Nasal assimilation. m → n/_D

sem slisela > sen slisela 'I heard'

n → m/_L

hanba > hamba

'shame'

B. Fricative + stop. s → c/k-t zenská > žencká woman ; vest > vect

'to take'

CODE-SWITCHING IN COLLOQUIAL CZECH

463

C. Contraction. poněvadž > pač 'since' nějaký > ňáký - ňákej 'some'; nějak > ňák 'in some way ' The impact of phonology on morphology can be de­ monstrated by a comparison of the adjectival declensions. The /é/ in the Czech adjectival declension functions as an indicator of bookishness and elsewhere in the language it may signal an archaism (Vachek 1968: 32). An addition­ al morphological feature which is doubtless due to analogy, is the -y ending in the nominative and accusative plural of the neuter form, e.g. SČ malá města; 0Č malý města. SČ SING. M.a. ina. Nom. -y -y Voc. Acc. -ého -y Gen. -ého Dat. -ému Loc. -ém Instr. -ým PLURAL N V A G D L I

-f

-é

-é

-é -ýcn -ým -ých -ými

oČ N. -é

F. -à*

-é

-ou -é -é -é -ou

-á -à

-é -é

-ej -yho

-ej

-y

-ej -yho -ýmu -ým -ým

-y

-a -ou -ý/-ej -ý/-ej -ý/-ej -ou

-ý -y -ejch -ejm -ejeh -ejma (Hronek 1972: 47)

As can be seen from the preceding table the only utterances containing a hard adjective which have the possibility of being neutral in regard to code selection

LOUISE B. HAMMER

464

are: 1. 2. 3. 4.

fern. nominative singular fern. accusative singular fern. instrumental singular masc./neuter instrumental singular

In the nominal declension, the most striking oČ feature is the -ma ending for all genders in the instrum­ ental plural of nouns and adjectives, e.g. mezi dobrejma/dobrýma klukama, ženama, městama. Another position of interest is the masculine locative plural. In the sČ form, the velar undergoes mutation and the ending -ich is added; whereas, the oČ form evidences no mutation and adds the ending -ich. ch /x/ > šǐ Češǐch Čechách 'Czech' h rozǐch rohách 'horn' g geolozich geologách 'geologist' k > c klucich klukách 'boy' In conjugation the alternative, -m (1st p.pi.) ending is no longer regarded as a markedly oč form, nor, as stated earlier, are the -u (1st sg.), -ou (3rd pi.) forms, e.g. dekuju/dekujou. The markedly oč form is the/ 1st pers. pi. conditional bysme 'we could', while the sč bychom is somewhat bookish. The masc. sg. past tense forms ending in a consonant such as řek 'he said', moh 'he could', are neutral to speakers in the western part of the Czech lands, but may be marked to those coming from eastern Moravia (Jedlicka 1974: 107). The colloq­ uial forms mủžu 'I can' (1st sg.), mủžu 'they can' (3rd pi.) are also neutral. Class III verbs such as prosit 'to request', trpet 'to suffer', sázet 'to plant', umĕt 'to be able' are divided into two categories in sč according to the 3rd person plural endings, e.g. prosi and trpi but sázeji and umĕji. The solution in the

CODE-SWITCHING IN COLLOQUIAL CZECH

465

dialects and in Oč was to assign all of these verbs to either one or the other type, e.g. proseji (ej), trpeji, sázeji, umeji or prosi, trpi, sázi, umi (Dokulil 1960: 199). In PBMC these verbs are in a state of flux with usage varying from speaker to speaker, but there appears to be a preference for the -eji form as being natural. One of the most characteristic syntactic features of OČ and PBMC is the use of oo instead of ktery to introduce a relative clause. Ten chlap, co tarn podpiral satnu, tak byl lákán. 'That fellow who was leaning against the cloak­ room was really propositioned.' To bila taková" ta korba, co je celá" proste... 'It was that kind of body (of a truck) , which is completely, simply...' Ty, co jako uměji dobře Česky..., ty na to budou chtit jit. 'Those, who, uh, know Czech well..., those will want to go to it (a play). Another feature is the use of the demonstrative almost as if it were an article. A on de teda na tu přednášku, kde ten přednáši, žejo... 'And so he goes to the lecture, where that one is lecturing, right...' Co ten Mencl s nima na ty scěně udělěl 'What that Mencl does with them on the stage! ' Hrajou teda v torn Rokoko a pak hraje Zábradli (v divadle Na Zábradli) 'They are playing in the Rokoko, and afterwards the Zábradli will do it...' (in the theatre Na Zbradli) In addition to the lexical fillers, e.g. teda 'then', proste 'simply', jako 'like', which are characteristic of fluent speech, other words are found in PBMČ which, while bookish in themselves, are used in an ironic way,

466

LOUISE B. HAMMER

e.g. pravil 'he said," kdežto "whereas", posléze 'ultimat­ ely". They are counted as code switches in the data. ... tady nekolikrát mél kafe (OC) a pravil, že von zezere vsecky kostky,... "he had coffee here several times and said that he ate up (used with animals) all the sugar cubes... (It is interesting to note that there are two words for coffee in Czech, sČ káVa and OČ kafe.) ... tak sme šli na činu (čiénske' jidlo) no a pak sme skoncili na posléze na Václaváku (na Václavském námésti)... "and so we went for some Chinese food and after­ wards we ended up, ultimately, on Wenseslas Square..." ... ze proste zencky si na sebe proste jako dbaji, dešto (kdežto) muskej proste dyš... "that women simply take care of themselves, whereas, a man simply, when..." Other lexical items, many of German origin, e.g. furt, jo "always", "right", are classified as oč forms. In the literature on code-switching between differ­ ent languages, early studies tended to investigate fact­ ors such as setting, topic and participants in order to define the sociolinguistic parameters of the selection. On the basis of a preliminary investigation of this data, it appears that topic is significant for most of the speakers, but not for all of them. (See my Ph.D. dissertat­ ion, Indiana University, for statistical data on topic.) That is to say, the percentage of oČ forms is greater in conversations of an everyday or informal character and the percentage of SČ forms is greater when the topic is of a professional or cultural nature. Not all categories of words participate in the switching to the same degree. For example, substantives are most likely to lose the prothetic v- in the cultural data: whereas, the pronouns

CODE-SWITCHING IN COLLOQUIAL CZECH

467

von 'he', vono 'it', vona 'she', voni 'they', are more likely to retain it to the same degree in both sets of data. The sČ -y- is more likely to replace -ej- in the nominative and accusative cases in the cultural data, although in the oblique cases it is less likely to be selected. The occurrence of i < e remains more stabile in both sets of data. This may be explained by the fact that (i) is common to the phonologies of both sČ and OČ and is therefore not perceived by the speakers as marked­ ly non-literary. In addition, this acceptance of the -1 ending is reinforced by the soft adjectival declension. Investigators in the area of Spanish-English code-switching such as Gingras (1974) and Lipski (1978) have made assumptions about the grammars which would generate conversational code-switching. Kučera has addressed this question in regard to the Czech data (1973: 499-521). Although he claims that 'certain gener­ alizations about the structure of the colloquial strings would be missed' in a two-grammar approach, and that a two-grammar approach would require the duplication of a vast number of rules (501): he also states that 'multi-grammar approach may be feasible' (520). In actuality, either approach is possible in the Czech situation; this is largely due to the fact that the two codes are struct­ urally very similar and especially due to the fact that colloquial Czech has no features that are not found either in sč or oč. However, this question is complicated by the fact that not all speakers, not even all educated speakers, have sč as their underlying code (Vachek: 1977) . Indeed, in everyday conversations, the underlying code is more often oč. Therefore, at least some of the rules would need to be bidirectional (input switch rules), and not unidirectional (from sč to Oč) as Kučera has

468

LOUISE B. HAMMER

them. Additionally, the phonological lenition processes of casual speech produce intermediate forms, e.g. nějakej, ńákej, ńráaky, which would add to the complexity of the description. Several co-occurrence constraints described by Kučera are confirmed by this study: 1. A combination of the oč -ej- and sč -mi never occurs, e.g. *dobrejmi (1961: 102). 2. When there is more than one modifier in the NP, the modifiers must agree in code (1973: 519). a. takovi dobri student b. takovej dobrej student but not c. *takovi dobrej student d. *takovej dobri student Many of the switches can be attributed to the intent of the speaker to foreground or emphasize some element in the sentence. Pro mou matku Malostransk ý (Oč) povidky nejsou nic zastaralého (Sč). "For my mother the Tales of the Malá Strana are not anything old fashioned. The switch to the sč form emphasizes the world zastaralého which is also foregrounded syntactically by its posit­ ion in the sentence. An additional incentive for the switch to sč may be related to the fact that zastaraly vyraz "obsolete expression" is commonly found in a diction­ ary key and might bring these associations into the dialog as well. Another example of the sč form being used for emphasis is the following: To bylo vubec (oč) , teda (Oč) , to byl výlet (sč) ! "That was on the whole, then, that was an excursion! " Intonation is very important in these switches and freq­ uently gives the clue to the speaker's intent. One speaker often switched to the oč code when describing something comforting and pleasurable.

CODE-SWITCHING IN COLLOQUIAL CZECH

469

,,. piju káVu (Sč dost... Ale pravda je, ze jednou za čas mam (oč) hrozne (oč) rád kdyz mi někdo udělá čaj , třeba ńákej (OČ) cejlonskej (OČ) ńákej (OČ) - viš... "I drink a fair amount of coffee... But the truth is that every once in a while., I really like it when somebody makes me some tea,maybe some sort of Ceylon - some - you know..." The same speaker used the sč code to add importance to an event or to mark it as formal. A vstoupili (sč) sme jako do to o vestibule (sč), a on (Sč) uz tarn čekal! 'And we entered into that vestibule, and He was already waiting there." In other switches, the speaker uses the participants' common knowledge of situational norms to communicate meta­ phoric information. Jako majitel vozu este by te tarn žencky (oč) obtezovali (Sč). "As the owner of a car, moreover, the women would bother you there.' žencky is slightly pejorative although commonly used in speech even by women. The use of the sč form of the verb ohtezovali heightens the sense of just how offensive such a situation would be to him. Switches also occur between sentences, although this is uncommon. Haha to je hezky jak my teda nespraVne mluvime česky. (Oč) Měli bychom tedy spravne mluvit Česky - vsichni mluvi naprosto spisovnou cestinu. (sČ) "Haha that's nice how we are speaking Czech incorrect­ ly. We should now speak Czech correctly - we are all speaking absolutely correct Czech." (Said with great irony for the comic effect.) When the first word of the title of a book or film begins with an o, it is almost always found without the prothetic v-, even when the elements preceding and follow-

470

LOUISE B. HAMMER

ing are in the OČ code. No já jako (oČ) - mně se libily — z Hrabala Ostre (SČ) sledovany (OČ) vlaky a to teda...

jenom

"Well^I, somehow, - I liked - of Hrabal"s (films), only 'Closely Watched Trains', and that, then... Numbers beginning with an o-, are frequently found without the prothetic v-, e.g. osm, osmnáct, osmdesát, Sgall has suggested that this may be due to the influence of the schools where these numbers are learned, e.g. pet plus tři je osum "five plus three are eight" (Private communicat­ ion) . Since switches also occur between a lexical form and a bound morpheme, Czech violates the "free morpheme constraint" proposed by Poplack (1980) for code switching between different languages. One encounters examples such as velkýma 'great'(instr. pl. adj.) which combine sč -ýwith oč -ma, and opuštěny "destitute" (neut. nom. pi. adj.) which evidences sč o- and oč -ý. In conclusion it must be stressed that the Prague colloquial Czech studied in this paper is the speech of intellectuals and as such, has characteristics peculiar to that group. In particular, switching to the sč code for purposes of emphasis, irony or in order to validate their status as intellectuals, whose tool is after all language, may be more frequent than would be found in a more diversified group. However, it is true for Czech as it is for other languages that two types of sociolinguistic rules are functioning, alternation rules and co-occur­ rence rules. Alternation deals with choice among the alternative ways of speaking, and corresponds to the para­ digmatic axis, whereas cooccurrence concerns interdepend­ ence on the syntagmatic axis.

CODE-SWITCHING IN COLLOQUIAL CZECH

471

NOTES

1 M. Vey published his Morphologie du tchěque p a r l ě in 1946. It was based on data which he collected in the twenties. Although it deals primarily with morphology, this study also includes a brief chapter on phonetics and contains a few pertinent sociolinguistic comments on the use of this variant. See also: A. G. Sirokova, 1954. K voprosu o razlicii meždu češskim literaturnym jazykom i narodnorazgovornoj rečju. Slavjanskaja filologija 2; 1955. Iz istorii razvitija literaturnogo cesskogo jazyka. Voprosy jazykoznanija 4. See also H. Kucera 1955; 1961; 1973.

REFERENCES

Baker, 0. R. 1980. Categories of Code-Switching in Hispanic Commun­ ities: Untangling the Terminology. Sociolinguistic Working Paper No. 76. Austin, Tex.: Southwest Educational Development Laboratory. Barnet, V. 1971. "Vztah komunikativni sfery a ruznotvaru jazyka v slovanskych jazycich". Slavia 46. 337-347. Belič, J. 1955. Sedm kapitol o cestine. Praha: Statni pedagogicke nakladatelstvi. Belič, J. 1972. Nastin č e s k e dialektologie. Praha: Statni pedagogic­ ke nakladatelstvi. Belie, J., Havranek, B., Jedlicka, A. & Travnicek, F. 1961. "K otázce obecne cestiny a jejiho pomeru k cestine spisovne". Slovo a slovesnost 22.98-107. Belie, J., Havranek, B. & A. Jedlicka. 1962."Problematika obecne cestiny a jejiho pomeru k jazyku spisovnému". Slovo a sloves­ nost 23. 108-126. Blom, J. & J.J. Gumperz. 1972. "Social meaning in linguistic struct­ ures: code-switching in Norway. Directions in sociolinguistics ed. by Gumperz & Hymes, 407-434. New York: Holt, Rinehart and Winston. Chloupek, J. 1980. "35 let noveho vyvoje cestiny". Slovo a sloves­ nost 41. 178-184. Danes, F., Dokulil, M., Hausenblas, K., Helel, M., Jedlička, A., Kuchař, J., šmilauer, V. & F-. Vahala. 1960. O č e s t i n e pro Cechy. Praha: Orbis. Dressier, W. & R. Wodak. 1982. "Sociophonological methods in the study of sociolinguistic variation in Viennese German". Lang­ uage in Society 11. 339-370.

472

LOUISE B. HAMMER

Ferguson, C. A. 1964. "Diglossia". Language in culture and society ed. by Dell Hymes, 429-439. New York: Harper & Row. Gingras, R. 1974. "Problems in the Description of Spanish-Eng. Intra-Sentential Code-Switching". Southwest Areal Linguistics ed. by Bills. San Diego: Institute for Cultural Pluralism. Gumperz, J. J. & E. Hernandez-Chavez. 1972. "Bilingualism, bidialectism and classroom interaction". Language in Social Groups ed. by Dil, 84-108. Stanford Univ. Press, Gumperz, J. J. & E. Hernandez-Chavez. 1976. The Sociolinguistic Significance of Conversational Code-Switching. Working Papers of the Language Behavior Research Laboratory No. 46. Berkeley, Cal.: Univ. of California. Hausenblas, K. 1962. "0 studiu syntaxe bezne mluvených projevu". Otázky slovanske syntaxe, 313-323. Praha" Statni pedagogicke nakladatelstvi. Havranek, B. 1936. Ceskoslovenska vlastiveda II. Spisovny jazyk česky a slovensky. Praha: SFJNX. Havranek, B. 1942. "K funkcnimu rozvrstveni spisovneho jazyka". č a s o pis pro modernl filologii 28. 409-416. Havranek, B. 1951. Stalinovy price o jazyku a jazyk literarniho dila a prekladu. Praha: československy spisovatel, Havranek, B. 1963. "Na zaver dvoulete diskuse o obecne a hovorove cestine". Slovo a slovesnost 24. 254-261. spisovneho jazyka česHavranek, B. S. J. Bělič. 1960-1971. Slovnik keho. Praha: Academia. Havranek, B. 1979. Vyvoj ceskeho spisovneho jazyka. Praha: Statni pedagogicke nakladatelstvi. Havranek, B. & A. Jedlicka. 1981. č e s k ā mluvnice. Praha: Statni pedagogicke nakladatelstvi. Hronek, J. 1972. Obecnā č e š t i n a . Praha: Univerzita Karlova. Jedlicka, A. 1974. Spisovny jazyk v soucasne komunikaci. Praha: Univerzita Karlova. Jedlicka, A. 1979. "0 typologii spisovneho jazyka a jeho teorii". Z teorie spisovneho jazyka 23-31. Bratislava: Veda. Jelinek, M. 1963. "K poměru mezi hovorovou cestinou a spisovným jazykem". Slovo a slovesnost 24, 47-54. Kucera, H. 1955. "Phonemic Variations of Spoken Czech". Word 11. 575-602. Kucera, H. 1961. The Phonology of Czech. The Hague: Mouton. Kucera, H. 1973. "Language variability, rule interdependency, and the grammar of Czech". Linguistic Inquiry 4. 499-521. Lipski, J. M. 1978. "Code-switching and the problem of bilingual competence". Aspects of Bilingualism ed by Paradis. Columbia, S.C.: Hornbeam Press, Inc. Mickelsen, L. 1978. "Czech sociolinguistic problems". Folia slavica 1. 437-455.

CODE-SWITCHING IN COLLOQUIAL CZECH

473

Mullerova, 0. 1976. "K tematicke vystavbe nepřipravenych mluvenych dialogickych projevu'. Slovo a slovesnost 37. 308-317. Novák, P. 1962. "0 smysl diskuse o mluvene cestine". Slovo a slo­ vesnost. 23. 266-272. Poplack, S. 1982. "Bilingualism and the vernacular". Issues in International Bilingual Education ed. by B. Hartford, A. Valdman & C. Foster. New York: Plenum Press. Sgall, P. 1960. "Obichodno-razgovornyj cesskij jazyk". Voprosy jazykoznanija 9. 11-20. Sgall, P. 1961. "Nespisovne tvary v nasi literature". Plamen 12. 97-101. Sgall, P. 1962. "Znovu o obecne cestine". Slovo a slovesnost 23. 31-46. Sgall, P. 1963. "K diskusi o spisovne a obecne cestine". Slovo a slovesnost 24. 244-254. Sgall, P. & A. Trnkova. 1963. "K metodám zkoumáni bezne mluvene cestiny". Naše řeč 46. 28. Sgall, P. 1981. "K nekterým otazkám naši jazykove kultury". Slovo a slovesnost 42. 299-306. Sara, M., šarová, J.& A. Bytel. 1970. č e š t i n a pro cizince: Czech for English-speaking students. Praha: Statni pedagogicke nakladatelstvi. Smilauer, V. 1972. Nauka o českēm jazyku. Praha: Statni pedagogic­ ke nakladatelstvi, Townsend, C. 1977. Czech through Russian. Columbus, Ohio: Slavica. Vachek, J. 1968. Dynamika fonologickeho systemu souēasne spisovne č e š t i n y . Praha: Academia. Vachek, J. 1977. "Poznámky k fonologicke stylistice jazykovych variet". Slovo a slovesnost 38. 81-89. Vey, M. 1946. Morphologie du tcheque parle. Paris: Librairie C. Klincksieck.

VI. FORMAL AND COMPUTATIONAL METHODS

HELMUT SCHNELLE Bochum

ARRAY LOGIC FOR SYNTACTIC PRODUCTION PROCESSORS An exercise in structured net-linguistics

INTRODUCTION The wish has often been expressed that linguistics should come into closer contact with the detailed analysis of the human ability of language use. K. Buhler and E. Zwirner emphasized this idea in the early thirt­ ies (cp. K. Buhler, 1933, p. XXVI, E. Zwirner, 1936) and Jakobson took over at the end of the thirties (Jakobson, 196 2) and tried to investigate "the interconnect­ ion between the topography of the brain and the struct­ ure of language" (Jakobson, 1981, p. 19) in the early sixties (cp. Jakobson, 1971, p. 289 f., p. 307 f.). At the time attention was mainly directed towards inter­ preting the functions of anterior and posterior centers for speech and understanding. Today we must go into further detail and try to relate language processing to nerve nets. Let us thus assume that the interconnection of nerve-cells in certain sections of the brain determine

HELMUT SCHNELLE

478

the process of articulation and understanding of utter­ ances in a language. Differences of connectivity account for the differences in language use. The development of the specific form of connectivity cor­ responding to a particular idiolect is determined in two steps: By genetically controlled development of a universal net in the embryo and during a child's first month; and by selective suppression of connectivities and build-up of additional connectivities during lang­ uage

acquisition.

How should we proceed to come to an understanding of the system of connectivities and its build-up? There is no doubt that the system is extremely complicated. Its function cannot be understood on the basis of a few regularities, in contrast to explanations in physics (e.g. classical particle mechanics, electrodynamic field of empty space etc.). Some theoreticians propose the method of abstraction: Define an abstract level of description underlying the concrete processes such that a restricted class of regularities (such as some types of movement transformations, in the case of Chomsky's proposals) could provide an understanding of the processes at least for an underlying structure. Indeed, the mechanisms of string manipulations derived from calculi and formal languages and applied to the descript­ ion of ordinary languages are well-understood. But I think that the specification of the relation between the abstract level and the concrete biological process­ es is too complicated or, in any case more complicated than more direct approaches. The more direct approach I have in mind has first been attacked by analyses in

SYNTACTIC PRODUCTION PROCESSORS

479

terms of nerve-nets (cp. Schnelle, 1981; Schnelle/Job, 1983; Schnelle/Rothacker, 1984). These studies showed how grammars can be realized by signal flow in nets without any manipulation of auxiliary symbols. In view of a more general definition it is necessary to apply the methods of structured design of information process­ ing, in particular the methods of programmable array logic developed for very large scale integrated hardware design. It is certain that these methods will have to be tested extensively and appropriately adapted to the specific problems of language processing before they become a powerful tool of linguistic analysis. The present article introduces the available means for such an analysis and shows how the production of strings defined by arbitrary CF-grammars can be realized in hardware. It will be shown that the resulting system is an orderly arrangement of encoders and decoders combin­ ed with pushdown stores. 1. CF-GRAMMARS AS NETWORKS OF SEQUENTIAL ENCODERS AND DECODERS Let us first recall some facts about simple (bin­ ary) encoders and decoders in the sense of computer architecture. According to Hayes (1978, p. 94) a decoder is a combinational circuit with n input data lines and 2

output data lines (all for binary signal values 0,1)

such that each of the 2

possible input combinationx x.

activates (sets to 1) exactly one of the output lines z. One, therefore, speaks of a l-out-of-2

or 1/2

de­

coder. A decoder has an additional input line, called the ENABLE line e; the zi.-th output is determined from

480

HELMUT SCHNELLE

the x. input when the enable line is on (=1). We shall present a 1/4 decoder as in figure la (to give an example). An encoder as we conceive it is the inverse of a decoder; it has 2

input data lines and k output lines.

An appropriate encoder of this type must assure that only one input line is active at a time or, if more may be active, that a priority measure assures which line is encoded, i.e. determines the output. The encoder has an additional output line which indicates whether an input line is active or none active; it is the INPUT-ACTIVE line (see Fig. lb for a 4 bit encoder).

SYNTACTIC PRODUCTION PROCESSORS

481

Let us add a remark on the realization of decoders and encoders. In programmable logic arrays as in VLSI design, a decoder corresponds to the AND-plane and an encoder to the OR-plane realizing the combinational parts of sequential circuits (cp. Mead/Conway, 1980, pp. 80/1, plate 8 and pp. 161/2). AND- and OR-planes are in turn usually realized in NOR form (ibid, plates 7 and 8, figure 3.20) by appropriately redefining the input and output values (0 and 1 ) ; the combinational circuits thereby assume a very regular structure. We therefore marked the decoder black-box by A (for AND) and the encoder black-box by 0 (for OR). As a next step of complications we introduce a sequence of an encoder, a delay-type flip-flop and a decoder, coupled by the INPUT-ACTIVE- and the ENABLElines, and combine a pair of inversely coupled sequences to a unit, as exemplified in figure 2. This is the basic unit of the networks representing CF-grammars. As we shall see we shall have exactly one such unit for each non-terminal symbol occurrence to the right of a rule arrow in a CF-grammar. The number of input resp. output lines from the top of the unit representation depends on the number of different symbol occurrences from which a rule headed by that non-terminal is "addressed" and the number of lines towards the bottom is determined by the number of rules which are headed by the non-terminal, i.e. the number of alternatives for the non-terminal's expansion. We shall specify these conditions in detail below. When these numbers are different from some 2 2

or

(for some natural numbers n, k) we leave some lines

unused.

482

HELMUT SCHNELLE

Figure 2 We compose a network representing the CF-grammar by building a sequence of such units. We may follow the sequence in which the non-terminals occur at the left of the arrows in the rule system and we may label the units by the symbols of the non-terminals. The topmost unit is to be a special unit for the symbol S and the bottom most unit is one for the set T of terminals; it is the output unit. An arrangement such as in figure 3 is

SYNTACTIC PRODUCTION PROCESSORS

Figure 3

483

484

HELMUT SCHNELLE

obtained for n non-terminal symbol occurrences differ­ ent from "S". In order to obtain the generative capacity of a CF-grammar the upper input-output lines for each N.-unit have been connected to the same storage unit which in some cases must function as a pushdown store. The essential information to be added to this arrange­ ment is the appropriate connection between the units in this sequence. This will be determined from the rules of the grammar realized by the sequence using the methods explained below. The encoders, decoders and storage units will be realized as programmed array logic circuits. It will be necessary to explain the essentials of such circuits before discussing their programming from grammars. 2. PROGRAMMABLE ARRAY LOGIC The programming of CF-grammars will be determined by the principles characteristic for programmable array logic (PAL) (cp. e.g. PAL-Programmable Array Logic Handbook, Monolithic Memories Inc. 1981). Logic arrays can be programmed either from an algorithmic represent­ ation of a process given in the form of a state diagram, from the corresponding transition table, or alternative­ ly from a definition in terms of a set of Boolean logic equations. The former can easily be translated into the latter but the programming can also be done directly from the algorithm, as we shall see. The basic combinational elements of a logic array are AND-gates and OR-gates. PAL uses a specific convent­ ion for representing the inputs to these gates which is

SYNTACTIC PRODUCTION PROCESSORS

485

particularly convenient in complex arrays. The ordinary representation of a three-input AND-gate and of a corresponding OR-gate and the PAL notation are represent­ ed in figure 4. As usual, a three place AND-gate and a three place OR-gate realize the input-output tables (Tables Ia and b resp.) i.e. the AND output is on if, and only if, all inputs are on. The OR output is out if, and only if, all inputs are out. ('On' means: high voltage 'H', 'out' means low voltage ' L ' ) .

486

HELMUT SCHNELLE

SYNTACTIC PRODUCTION PROCESSORS

487

The table must be understood as specifying the result of a transition of the voltage at the output of the gate. In synchronized circuits, the voltages are only used temporarily at equidistant clock signals whose period is long in comparison with the time necessary for the changes at the gates. It is because of this, that the exact temporal processes of change need not be considered. Only the initial states and the resulting states are relevant. In addition to AND and OR we have NEGation, repres­ ented by a bar in Boolean formulas. We have also an in­ put-operation IN which has two output lines and one in­ put line. The two elements are represented as in figure 5. Their operation is defined by the Boolean tables (Table I, c and d ) . The principles of programmable logic arrays can now be understood from the following example. Let us assume that the Boolean function

o = Ī1 . Ī2 + Ī1 . Ī2 should be realized, i. e. the function defined by table I e. The manufacturer provides an array like that in figure 6. The points mark fixed connections of horizontal wires with vertical ones. At the positions F1 to F8 it contains fusible links. In order to realize the Boolean function the links F 2 , F 3 , F 5 , F8 must be fused (i.e. interrupted). The remaining links will then realize the function defined above. The resulting array can be re-

488

HELMUT SCHNELLE

Figure 6

Figure

Figure 8

Figure 9

7

SYNTACTIC PRODUCTION PROCESSORS

489

presented in PAL notation as in figure 7. Programming a general array provided by the manufact­ urers ( such as MMI, Signetics, NSC etc.) consists in indicating the links that must be fused by specifying the Boolean function. Readers familiar with ordinary combinational logic diagrams will verify that figure 7 is equivalent to figure 8. In complex systems, inputs and outputs could occur in an array section such as in figure 9. In the examples of figure 4, only the inputs to AND-gates have been programmed. Thus, the array of fusible links represents an

AND-plane or decoder as

explained above. OR gates can also be programmed in field programmable logic-arrays (FPLA's), i.e. also inputs to OR-gates are fusible (whereas in PAL's only the inputs to AND gates are programmed). The correspond­ ing array connected with the OR-gates is an OR-plane or an encoder. The system of fusible connections is thus partitioned into two subsystems, the AND-plane and the OR-plane. We get the general form of a PLA-implement­ ation for a finite state machine as in figure 10 if the outputs of the OR-plane are fed onto storage registers such as a sequence of flip-flops (cp. for systematic considerations Mead/Conway, 1930, p. 84 and § 6.2). We shall make use of a D-type output flip-flop, depending on a clock-signal. The D-type flip-flop realizes a function

0(t+l) = I(t), i.e. its output at the next

clock signal corresponds its present input. The essential aspect of the use of flip-flops is that their outputs are fed back into the array as in­ puts. By this feed-back, the array can be programmed into a sequential network. In principle, finite state

490

HELMUT SCHNELLE

F i g u r e 10

F i g u r e 11

SYNTACTIC PRODUCTION PROCESSORS

491

machines can also be programmed in a PAL-network (i.e. fixed OR-plane). In this case the typical use of a D-type flip-flop is represented in the form of figure 11). 3. AN EXAMPLE: THE PROGRAMMING OF AN ARRAY FROM THE TRANSITION TABLE OF A SEQUENTIAL FINITE STATE MACHINE. A processor can be programmed by two different but interrelated methods. We may set up the Boolean equat­ ions, one for each output- and next-state-line, specify­ ing how their activity is determined from the input and previous state lines. Alternatively, we may set up a table of transitions and program the PAL (or PLA) net­ works immediately from the transitions. We shall make use of this method by translating CF-grammar rules into transitions which are used in turn for programming a PLA network. This is, by the way, the usual method in VLSI design (cp. Mead/Conway, 1980, p. 195 and pp. 8588). Let us briefly recall the meaning of a transition table. The method of programming a processor by transition tables stems from Turing. He showed in 19 36 that sequent­ ial automatic machines and processes can be defined in terms of transition tables (cp. Turing 1965, p. 126). A transition table consists of a list specifying five points of information: 1. The present state of the machine, 2. The symbol presently read at the input (-tape) of the machine, 3. The symbol presently printed at the output (-tape) of the machine,

492

HELMUT SCHNELLE

4. A move of the stored input and output (-tape) information providing the input information for the next state and the position to print on the next output information at the next state, 5. The next state of the machine. The points of information 1. and 2. are called the in-put term, and those of 3., 4., and 5. the output term of a transition. Turing assumed that the machine had just one tape to read from and to print on and that after each reading and printing it could move either one position to the left, one position to the right, or not move at all. If we allow only for moves in one direction we obtain a machine which is much less powerful than a Turing machine: it is a sequential finite state transducer. Usually the move is identified with the temporal flow of information at the input port or the output port of the machine. In any case, a sequential finite state transducer can be defined by a list of transitions containing the information points 1., 2., 3., and 5. above (since 4. is taken account of by natural informat­ ion flow). The list of transitions can also be represented by a finite graph - a state

graph

- whose nodes are the

possible states of the machine and whose arcs are labell­ ed by the pair of input information and output informat­ ion occurring at a transition from one state to another. It is now possible and even straightforward to program a PLA network from a transition table or state graph defining a finite state machine. Let us discuss a

SYNTACTIC PRODUCTION PROCESSORS

493

simple example presented by the transition table IIa, Its corresponding state graph can be represented in two versions as in figure 12. The system generates either sequences of points or expressions of the form [A*C B*] in cycles starting from and returning to S . (A* is a finite string consisting uniquely of occurrences of A s ) . In view of programming the state graph we shall have to encode the three states and the six output symbols. The state will be represented by the states of two CDtype) flip-flops F , F as follows:S° ~ 00, s1 ~ 01, 2 S ~ 10. The outputs will be coded by the states of 12 three parallel output lines 0 , 0 , 0 3 as follows.

HELMUT SCHNELLE

494

INPUT - TERM Present state

OUTPUT - TERM

Input variable

Output variable

Next state

0

a) b)

1

[

0

A

d)

1

B

e)

0

C

f)

1

]

c)

Table H a

F1

INPUT - TERM 2 F

1

I

o

OUTPUT - TERM 2 o3 0

2

F1

F

a)

0

0

0

0

0

0

0

0

b)

0

0

0

1

1

0

0

1

c)

0

1

0

0

0

1

0

1

d)

0

1

1

0

1

0

1

0

e)

1

0

1

0

1

1

0

0

Table IIb

SYNTACTIC PRODUCTION PROCESSORS

495

The programming is now straightforward: Each line of the transition table is represented by a vertical line and identified by the crosses in the AND plane corresponding to the input terms. The reader should verify this correspondence in figure 13. The OR-plane shows, how the next states and outputs depend on the input terms. Since we assume that inputs and outputs are "read" at clock times, a position on the OR-plane which is not crossed for an input term marks the fact that the corresponding output or next state is 0 for that input term. In view of the programming of more complicated problems, it is not advisable to encode the states and variables completely instead of realizing them separate­ ly. It is true that the number of flip-flops and gates becomes minimized in this way. On the other hand the organization becomes less perspicuous and regular. It has become clear in VLSI design that, due to the fact that the cost of transistors is no longer an essential factor, perspicuity of construction and local organizat­ ion is of predominant importance. We shall, therefore, reorganize the system in the following way: We shall represent each state by a (D-type) flip-flop, each in­ put subspecifying a state by a specific input line and we shall add output lines for organizational purposes. The reorganization is represented by Table IIIa. It contains a number of special features which need to be explained. In order to show that the system may operate without producing an output symbol at each move, we eliminated the symbols (.} , ([) and ([) and encoded

HELMUT SCHNELLE

496

a)

b)

c)

d)

e)

Figure

13

f)

SYNTACTIC PRODUCTION PROCESSORS

497

the remaining vocabulary. Since now, however, the output is only meaningful when there is indeed a symbol that must be sent to the printer (or to the output tape), these output states must be marked by a "1" on an additional line, the PUSH T line, i.e. the outputs are only read from our system when the PUSH T line is on. The output symbols are correspondingly encoded by three parallel bits, the first being the PUSH signal: A = 110, B = 101, C = 111 (see figure 14 and table IIIb). As a next feature we introduce additional input and output lines: (a) one input line for each flip-flop in order to specify the branching of signals, when the system is in the state represented by that flip-flop, (b) for each input line a corresponding output line, which returns the information after the system has used it. In view of their use in CF-grammar programming the input and output lines do not all serve the same purpose. Some will be used later to communicate with the same store (e.g. push down-store) and will secure appropriate coordination. These lines will be termed Q lines for outputs to the stores and P lines for inputs from them. In our system we shall have the input lines I0,I1, and P and the output lines 0 , 0 , and Q . As in the case of the proper output lines, the information at the lines 0 , 0

and Q

is meaningful

only when the system is in an appropriate state, namely when its next

state is S01, S10, and S11 resp. We therefore

utilize these next states, i.e. the input signals for the D-type flip-flops as PUSH signals on output lines: PUSH 0°, PUSH 0 1 , PUSH O1 (see figure 14).

498

HELMUT SCHNELLE

Figure

14

SYNTACTIC PRODUCTION PROCESSORS

INPUT - TERM

OUTPUT - TERM

Input variable

Present state

499

Output variable

Next state

a) b) c)

A

d)

C

e)

B

f)

Table IIIa

INPUT - TERM Present state

Next state Input variables Output variables

I° a) 1 b) 1 c) d) e) f) -

OUTPUT - TERM

1 P

0° Q1 0

-

-

1

1

0

-

- 1

-

0

-

-

1

-

1

-

-

0

1

-

1

-

1

-

-

-

0

-

-

1

1

-

1

-

o1

I1

Table IIIb

Proper outputs PUSH T

TO

TO

1

-

-

0

-

-

-

0

-

-

-

-

1 1

-

1

1

1

1

-

1

1

1

1

0

-

-

1

1

1

-

0

-

-

500

HELMUT SCHNELLE

Finally, an input symbol from the different lines I , I and P is only needed at times when the present states are S01, S11, and S10, i.e. when outputs of the flip-flops F , F2, F are on. We utilize these outputs as signals triggering the input from input lines: POP I , POP I , and POP P . The terms PUSH and POP indicate that we assume communication with shift registers serving as input- and output buffers (see below). A final remark on the crossing notations in figure 14: In order to make the reading of the array more per­ spicuous, we used three indicators for crossings: points identify horizontal input term lines with the correspond­ ing vertical lines. Half crosses indicate the crossing with the next state line and proper crosses indicate the outputs. The differences are only notational. For PLA programming each mark is a simple cross i.e. a connect­ ion not to be fused in the realized array. 4. CF-GRAMMARS AND PUSHDOWN STORE AUTOMATA We may now turn to the central feature distinguish­ ing the programming of CF-grammars. It is well known that those formally definable languages which are gener­ ated by CF-grammars can also be determined by pushdown store automata. Our proposal for the programming of CFgrammars will show in which sense CF-grammars may be understood to be notations for pushdown store automata. Let us recall some well-known facts about CF-grammars. CF-grammars are quadruples where T and N are sets of mutualy exclusive sets - the sets of the terminal

symbols

and of the non-terminal

symbols

501

SYNTACTIC PRODUCTION PROCESSORS respectively - and p - the set of rules

- is a subset

of N x (T ((set-theor. union))N)*, where (T((set-theor. union))N)* denotes the set of strings

of elements from

T or N. Elements of P are usually denoted by expressions of the form X → ω where X e N and ω ε (T ( (set-theor. union)*. S is an element of N. It is obvious that the structurally essential part of a CF-grammar is P, the set of rules. The ordinary interpretation of a CF-grammar assigns to each rule a rewrite operation over strings of symbols: Its arguments are strings containing the non-terminal occurring to the left of the arrow and the correspond­ ing values for these arguments are those strings with one occurrence of the non-terminal replaced by the string to the right of the arrow. In other words, the rewrite operation replaces a non-terminal symbol X in its context by the string ω. This will not be our inter­ pretation of CF-rules. Our interpretation takes them as indicators of networks for appropriate signal flow, i.e. as specifiers of circuits. Let us consider our example. Readers of Schnelle (19 81) may have remarked that figure 12b looks like the operational diagram of the CF-grammar

This grammar generates the language It differs from the language

discussed above

only insofar as in the former each string must contain as many

as

for the latter.

whereas this restriction does not hold

HELMUT SCHNELLE

502

Since the PAL network described above determines only the latter we must add something to it in order to account for the restriction. In considering the graph in figure 12b the restriction is as follows: The operat­ ional signal circulating over the graph must return left 1

at S 1(generating thereby a B) as often as it arrived 1 before at S from the left (i.e. as often as it had o generated an A ) . In terms of the equivalent graph of figure 12a, this means that the two cycles must be used the same number of times, an idea underlying my proposal of CC-automata (cycle coordinated automata) in Schnelle (1964). This condition can be easily implemented by lett­ ing the two transitions mentioned - i.e. the two transit­ ions marked by Q

pushdown

and P

- communicate with the same

store.

We must now come to explain the communication with the pushdown stores. They will be implemented by shift registers (such as those commercially available by PAL manufactures; SN54/74 LS 498 from MMI is an example). For the proper inputs and outputs to our system, i.e. the I and the 0, and for the terminal symbols, the shift registers serve merely as buffers; for the P and Q, they serve the purpose of coordinating corresponding cycles in the system of transitions. Let us assume that we use a shift register from the left, i.e. its left-most register element bears the topmost symbol of the pushdown store implemented by the shift register. In this mode of use it is sufficient to consider the follow­ ing signal ports only: SHIFT LEFT (=P0P), SHIFT RIGHT (=PUSH), Rilo (Right in, Left out) and Q7 the output state of the topmost register element (=0 or Q ) . For

SYNTACTIC PRODUCTION PROCESSORS

503

our example represented in figure 14, the communication pattern with shift registers takes the form represented in figure 15. 5. TRANSITION TABLES FROM CF-GRAMMARS Let us now turn to the discussion of the general method for programming CF-grammar production in terms of PLA-networks. As is usual in VLSI design the problem should first be specified by a transition table. The programming from a transition table is an automatic procedure. What we must specify is a method for trans­ lating systems of CF rules into transitions. Similar to the basic idea of ATNs a rule will be translated into a sequence of transitions but our coding is quite differ­ ent: Each non-terminal symbol will be coded by a pair of flip-flop name's. If the non-terminal is N then the corresponding names are Nio and Ni1. The different occurr­ ences of the non-terminal symbols to the right of an arrow are differentiated by different input-output informations, i.e. Ps and Qs in our example above. The upper index is to correspond to the index of the non-terminal it is assigned to whereas the lower index varies according to the occurrence to be differentiated by the Ps and Qs with the same upper index. Occurrences to the left of the arrow in a CF-rule get differentiated by I and 0 infromations. Terminal symbols need not be recoded. Let us now consider an example. As we saw, the CF-grammar generating

is

504

HELMUT SCHNELLE

POP I

o PUSH

-0

S

POP I

0 PUSH

POP P Q

0°

pl/Ql

PUSH -1

S

POP I 0

I

1

PUSH

POP I

o PUSH

POP I

o PUSH T

POP I

o PUSH

Figure 15

T02

SYNTACTIC PRODUCTION PROCESSORS

505

The operational diagram of this grammar (in the sense of Schnelle, 1981) has already been given in figure 12b. Let us consider the coding of rule R1. We 1 start in state S . R1 can only be applied when the in1 put store I is O. Should input-store be 1, the transit­ ion determined by rule R2 would be selected. During the same transition we must store output symbols and go to the next state. A is an output symbol to be stored to the output tape T 0. We should go back to state S10 as indicated by the rule, but we should remember that we enter in this case from rule R1 itself instead of coming from RO. We do this by storing information Q11 to the 1 1 P /Q store. This completes the first transition in rule R1. There is another transition contained in R1 which starts when the system is in S11 and reads from store 1 1 P /Q that the last Q-symbol stored there has been 1, i.e. it reads P11. When this is the case it stores a B to the T 0 tape and returns to S11 in sending an output signal O10 to the output tape 0 . This description gives a first indication of how the coding of R1 in terms of two transitions is to be under­ stood. The two transitions RO and the single transition in R2 can be understood in a similar way. Our CF-grammar is then coded as follows

506

HELMUT SCHNELLE

The reader should verify that R

is easily interpreted

as just described. We obtain the five transitions b) - f) in our trans­ ition table above (Table IIIa). Transition a) may be trivially added. We may now state the rules for the derivation of the transition table from CF-rules for the general 1. Each occurrence of the i-th

case:

non-terminal N

occurring in the rule system is substituted by an express­ ion of the form

if it occurs to the right of the arrow or of the form

if it occurs to the left of the arrow, except for the initial rule which has 2. The substitutes for different occurrences of the same non-terminal N

on the right of an arrow - such as

e.g. in the same rule

or in

different rules - will differ in their a indices for the Q and P accompanying the different occurrences of pairs For example we could have

SYNTACTIC PRODUCTION PROCESSORS

507

3. The substitutes for different occurrences of the same N

to the left of different rules differ in the

indices 3 for the I and 0. For example we should have something like

4. The terminals of the rules need not be substitut­ ed; they are simply kept in their positions. Let us apply these conventions to a slightly more complicated CF-grammar

We obtain

508

HELMUT SCHNELLE

OUTPUT TERM

INPUT TERM PRESENT STATE

INPUT VARIABLES

OUTPUT VARIABLES

a) b) c) d) e) f) g) h) i) k) 1) m)

T a b l e IV

NEXT STATE

SYNTACTIC PRODUCTION PROCESSORS

509

The reader may now read off the transitions: 2 in RO, 3 in Rl, 2 in R3, 2 in R4 and 1 in R5. The transitions are those in table IV. Let us now discuss the derivation of the correspond­ ing PLA network. Just as in our example discussed above, we shall have one vertical line for each line in the transition table. The crossing of the input-term line with its transition line will be marked by a point. More­ over, each next state of the transition line will be marked at the appropriate position of the array by a half-cross (-arrow), and the output variables are marked by crosses at their positions. The reader will easily verify that figure 16 corresponds to transition table IV. The connections with an appropriate sequence of shift registers will be left for the reader as an exercise; this is not a difficult task when taking account of the example of figure 15. It should be obvious that the result fits the general form postulated above and represented in figure 3. Its internal structure as indicated in figure 2 should be verified. The input de2 1 2 coders for I , P , and P initiate the operation of the next flip-flop reached by the sequence of a point and a half-cross along a vertical line. The input lines for the flip-flops and the output lines assigned to it form together the encoders represented in figure 2. This completes our discussion of the principles for programming CF-grammars into PLA arrays. I think that it adds to the structural and procedural understanding of

510

HELMUT SCHNELLE

F i g u r e 16

SYNTACTIC PRODUCTION PROCESSORS

511

CF-grammars. But let me emphasize that the real value of this kind of programming for CF-grammars will only turn up in the programming of parsers which realize a flow of parallel information for all possible predictive paths during the parsing processes. But this problem must be left for another publication.

REFERENCES 2 Bühler, K. 1933. Sprachtheorie. Stuttgart: Fischer 1965 . Hayes, J. P. 1978. Computer Architecture and O r g a n i z a t i o n . New York: McGraw Hill. Jakobson, R. 1962. "Kindersprache, Aphasie und allgemeine Lautgesetze". Selected Writings I., by R. Jakobson. 328 ff. S'Gravenlage:Mouton. Jakobson, R. 1971. "Toward a linguistic classification of aphasie Writtings impairments" & "Linguistic types of aphasie". Selected II by R. Jakobson. S'Gravenlage: Mouton. Jakobson, R. "Gehirn und Sprache". Sprache und Gehirn ed. by H. Schnelle. Frankfurt: Suhrkamp (cf. also Brain and Language, New York University, Slavic Papers Vol. IV, 1980). Mead, C. & L. Conway. 1980. Introduction to VLSI-Systems Reading. Mass.: Addison-Wesely. MMI - 1981. PAL-Programmable Array Logic Handbook, Sunnyvale: Mono­ lithic Memories. Schnelle, H. 1964. "CC-Automata". Lecture presented at the Conference on Formal Grammars and Automata. Jerusalem 1964. Schnelle, H. 1981. "Elements of theoretical net-linguistics". Theoret­ ical Linguistics 8. Schnelle, H. & D. M. Job. 1983. "Elements of theoretical net-linguist­ ics, Part II". Theoretical Linguistics 10. Schnelle, H. & E. Rothacker. 1984. "Elements of theoretical netlinguistics, Part III". Theoretical Linguistics 11. Valvo-Signetics. 1981. Signetics - Bipolare Speicher - MOS Speicher. Valvo Handbuch. Sunnyvale: Signetics. der Phonometrie. Berlin. Zwirner, E. 1936. Grundfragen

JÜRGEN KUNZE Berlin

MATHEMATISCHE BETRACHTUNGEN ZUR ZUVERLÄSSIGKEIT VON VERFAHREN

1. VORBEMERKUNGEN

Es gibt eine Reihe von Aufgaben in der automatischen Sprachverarbeitung, bei denen das Verhältnis zwischen dem Aufwand und der Qualitat eines entsprechenden Verfahrens einerseits eine gravierende Bedeutung besitzt und andererseits in ausreichendem Maβe ermittelt werden kann. Dies gilt in erster Linie für relativ einfache Aufgaben auf der Wortformenebene, z. B. fur die automatische Worttrennung innerhalb eines Programms fur den automatisierten Satz. Bei komplizierteren Problemen wie der automatischen syntaktischen Analyse oder erst recht der automatischen Übersetzung ist das Geflecht von Fehlern, Unvollstandigkeiten und ihren Ursachen oft so undurchschaubar, daβ vorerst an eine einfache quanti­ tative Fehlerbewertung nicht gedacht werden kann. Beschranken wir uns also auf solche Probleme, bei denen die Begriffe "Fehler" und "korrekte und vollstandige Bearbeitung des Inputs" einigermaβen klar und einer quantitativen statistischen Bewertung zuganglich sind. In solchen Fallen hat es Sinn, von einer 90%-Lösung

514

JÜRGEN KUNZE

zu sprechen, und die einzige offene Frage ist dabei, ob eine 100%-Lösung durch gewisse Strategien iiberhaupt erreichbar ist. Dies ist fur die automatische Worttrennung aufgrund der Existenz von Homographen bei isolierter Bearbeitung der V7ortformen sicher nicht der Fall. Bei praktischen Anwendungen entsteht fast immer das Problem, ein ausgewogenes Verhaltnis zwischen der (möglichst hohen) Qualitat und dem (möglichst niedrigen) Aufwand herzustellen. Dabei ist der Begriff "Aufwand" eine im konkreten Fall festzulegende Mischung aus dem Umfang des Verfahrens (Speicherplatz), der Verarbeitungsgeschwindigkeit, der intellektuellen Investition (Erarbeitungskosten) und anderen Faktoren. Unsere folgenden Betrachtungen beziehen sich im Grunde nur auf den ersten Faktor, und dies auch noch stark vereinfacht, wie in 2. dargelegt. Es ist nun eine Erf ahrungstatsache, daβ das Erreichen der qualitativ optimalen Losung, d.h. die Erarbeitung eines Verfahrens mit der maximal erreichbaren Zuverlässigkeit von (100-ε)% mit ε ≥ 0 auf ungeahnte Schwierigkeiten stoβen kann. Die Verbesserung von 98% auf 99% (falls erreichbar) la(3t den Umfang des Verfah­ rens, gemessen durch die Anzahl von "Regeln", oft unzumutbar in die Hone schnellen. Keinesfalls ist der Zusammenhang linear, d. h. um von 98% auf 99% zu kommen, braucht man i. a. wesentlich mehr zusatzliche Regeln als um von 94% auf 95% zu kommen. Hierbei hat man zu unterscheiden, ob man falsch bearbeitete Inputs als solche zählt oder die einzelnen Fehler in den Inputs (also z. B. an wenigstens einer Stelle falsch getrennte Wortformen oder falsche Trennungen fur sich). Dieser Unterschied kann das o. g. Verhaltnis verschieben, ins-

ZUVERLÄSSIGKEIT VON VERFAHREN

515

besondere wenn man die Unvollstandigkeit des Verfahrens (also z. B. nicht ermittelte Trennungen, d. h. die Trennquote) mit in Betracht zieht. Unsere eigenen Erfahrungen gehen soweit, daβ z. B. die Verbesserung der Trennquote um einige Prozent oberhalb 90% und die Erniedrigung der Fehlerquote um einige Promille unterhalb 10‰ bei der Worttrennung des Deutschen zusammen eine Verdopplung der Regelmenge erfordern konnen. (De facto lief der Erkenntnisprozeβ natürlich so, da3 fur bereits erarbeitete Verfahren sehr unterschiedlichen Umfangs nachträglich durch groβe Tests diese Zusammenhänge ermittelt wurden.) Dennoch war eine 98%-Losung durchaus erwiinscht: Dies hangt mit den Kosten zusammen, die ein Fehler verursachen kann (beim automatisierten Lichtsatz sind diese Kosten sehr hoch!). Diese recht ungunstige Relation zwischen Aufwand und Qualität läβt sich aus verschiedenen Gründen nicht verallgemeinern, u. a. weil die be­ reits genannten Veranderungen von Trenn- und Fehlerquo­ te mit einer wesentlich verbesserten Klassifizierung der Trennstellen verbunden waren (Abschiedsa-bend ist viel schlechter als Abschieds-abend, und erst ein Verfahren, das auch dies hinreichend zuverlassig unterscheiden kann, findet den Beifall der Polygraphen.) Wollte man dies auch noch mathematisch fassen, so müβte man sprachlich richtige aber falsch klassifizierte Trennungen dem Ver­ fahren ebenfalls als Fehler anlasten. Dann allerdings sind die Veranderungen bei der Zuverlassigkeit nicht mehr nur einige Prozent, sondern so groβ, daβ die Verdopplung der Regelmenge plausibel ist. Wie man sieht, bringt bereits die Diskussion derart einfacher Falle eine Reihe von Nebenfragen und Unbestimmtheiten zutage. Will man zu mathematisch beweisbaren Zu-

516

JÜRGEN KUNZE

sammenhängen gelangen, so muβ man folglich recht rigorose Vereinfachungen vornehmen. Wir werden im folgenden davon ausgehen, daβ die Begriffe "Verfahren (mit einer Regelmenge)" und "vollständig und korrekt bearbeiteter Input" definiert sind und daβ weitere feinere Gesichtspunkte wie die folgenden ausgeklammert werden: Abstufung der Fehler in schlimme und weniger schlimme: Ein Input kann nur entweder vollstandig und korrekt oder falsch bearbeitet werden. Erkennung falscher Inputs: Fur sie bleibt die Bearbeitung irrelevant, obgleich es fur das Verfahren spricht, wenn es solche erkennt (Untrennbarkeit von Hochhcus impliziert orthographischen Fehler ). Unterschiedlicher Status von Regeln: Alle Regeln des Verfahrens bilden eine Menge. Nicht-Existenz einer 100%-Lösung: Hier kann man sich, vereinfacht gesagt, so helfen, daβ man gewisse Werte mit (100 - ε)% multipliziert. 2. DEFINITIONEN UND EINIGE ZUSAMMENHANGE Nach den einleitenden Betrachtungen wenden wir uns nun der Aufgabe zu, die skizzierten Zusammenhange mathematisch zu erfassen. Wir nehmen folgende Grundvoraussetzungen (A) bis (F) an. (A) Gegeben sei eine (eventuell unendliche) Menge X von Objekten (Inputs) x. Fur unsere Belange kann X° als die Menge aller Wortformen (oder eine Teilmenge davon) einer natürlichen Sprache, als die Menge aller akzeptablen Satzen (oder eine Teilmenge davon) aufgefaβt werden usw.

ZUVERLÄSSIGKEIT VON VERFAHREN

517

(B) Gegeben sei ein Verfahren V, das dazu geeignet ist, Objekte x zu verarbeiten. Das Verfahren V kann also eine Wortformenanalyse, ein Verfahren zur Worttrennung oder zur Graphem-Phonem-Umsetzung sein, falls X ein

aus Wortformen besteht, es kann

Parser sein, falls X

Satze enthält usw.

(C) Das Verfahren V sei so aufgebaut, daβ es einen festen prozeduralen

( ="logischen") Teil besitzt und

auβerdem einen variablen Regelteil

("Informations-

teil") R, der Regeln r enthalt. Dieser Regelteil sei bei jeder Variante von V endlich. (D) Es gibt eine maximale endliche Regelmenge Ro für V, die es erlaubt, durch V alle Objekte x aus Xo vollstandig und korrekt zu beschreiben. Die verschiedenen Varianten von V entstehen also durch ?Vuswechslung von R, und bei R c R

hat man damit zu rech-

nen, da3 nicht alle Objekte x aus X

vollstandig und

korrekt bearbeitet werden. (E) Über Xo sei eine Häufigkeit h definiert. h(x) kann z. B. als die "relative Haufigkeit des Auftretens von x in einem gewissen Text trachten Sprache

(oder in Texten der b e -

schlechthin)"

im Falle Xo = Wort-

formenmenge interpretiert werden. Ist Xo eine Satzmenge, so korreliert h(x) mit der Haufigkeit "Gebrauchlichkeit") der

(z. B. syntaktischen) Struk-

tur von x. Es gelte (1) h(x) ≥ 0 (2)

fur alle x є X ° ,

E h(x) = 1 X6X°

(oder

518

JÜRGEN KUNZE

definition

1:

Es sei (3) Offenbar gilt fur beliebige Teilmengen X von X : (4) (0 bezeichnet die leere Menge). Definition

2:

Es sei (5)

M(R) = Def

Menge aller

die durch die Variante

V[R] mit der Regelmenge R vollständig und korrekt verarbeitet werden. Aus (D) folgt (6) ferner gilt fur beliebige Teilmengen R' und R" von Ro: (7) Wenn

so M(R')

M(R").

Mit V[R] bezeichnen wir die Variante von V, die sich bei R als Regelmenge ergibt. Es sei V

Definition

= V[R ] .

3:

Es sei (8)

Z(R) =

H(M(R))

Def Z(R) ist also die "Qualitat oder Zuverlassigkeit von V [ R ] " , gemessen durch die Summe aller h(x) fiir genau diejenigen x, die durch V[R] vollstândig und korrekt arbeiteten werden. Aus (9)

( 3 ) , (4)

und

(6) folgt

be-

ZUVERLÄSSIGKEIT VON VERFAHREN

519

(F) Fur jedes x є Xo ist die (nicht-leere) Menge der Regeln r aus R , die bei der Bearbeitung von x durch V anzuwenden sind sowie die Anzahl dieser Anwendungen eindeutig bestimmt. Definition

4:

Es sei (10) e(r,x) =

(11) E(r,x) =

Anzahl der Anwendungen der Regel r bei der Bearbeitung von x durch V . E e(r,x).

Def rGR E(R,x) ist also die Gesamtzahl von Anwendungen von Re­ geln aus R bei der Bearbeitung von x durch V . Definition

5:

Es sei x G X , r G R

und

Est gilt fur alle r є R°, R', (14) (15)

E({r}, x) = e(r,x), D({r},x) = d(r,x) E(R'

U R",x) = E(R',x) + E ( R " , x ) ,

falls R' ∩ R" = ø. Definition

6:

Es sei x G X .

520

JÜRGEN KUNZE

(16)

= Σ Def rєRo

n(x)

e(r,x)

n(x) ist die Gesamtzahl von Regelanwendungen bei der Bearbeitung von x durch V .

Definition

7:

Ss sei x e X

und i ≥ 1.

(17)

R.(x)

= Menge der Regeln r G R°, die bei BearDef o beitung von x durch V genau i Mal angewendet werden. Offenbar gilt für alle r є R°: (18)

r G Ri(x) genau dann, wenn i = e(r,x).

Definition

8:

Es sei x G X o . (19) (x) = U Ri (x) Defi≥ Offenbar gilt: (20) (21)

R.(x) = ø für i > n(x)

(22)

R.(x) n R.(x) = ø für i ≠ j ,

(23)

x є M(R) genau dann, wenn R(x)

(24)

r

є

(x) genau

R

dann, wenn d(r,x) = 1.

R(x) ist die Menge derjenigen Regeln aus R , die bei einer vollständigen und korrekten Bearbeitung von x durch V

tastsachlich benötigt werden. Aus der Funktion h(x) wird nun eine Funktion w(r)

ZUVERLÄSSIGKEIT VON VERFAHREN abgeleitet. w(r) beschreibt r bei der Bearbeitung aller nehmen wir an, daβ gewisse der Bearbeitung durch V in (Tabelle 1 ) :

X

r

1

X

521

den "absoluten Anteil" von x G X . Zur Illustration Objekte x und Regeln r bei folgendem Verhältnis stehen

2

X

3

X

4

0

2

1

r

1

0

3

0

r3

0

1

1

2

r, 4

0

2

0

0

n(x)

8

7

12

6

Tabelle

0 ...

...

1

Die Tabelle ist zo zu interpretieren, daβ bei der Be­ arbeitung von x2 insgesamt sieben Regelanwendungen stattfinden, dabei wird u. a. r1 zweimal, r3 einmal und r4 zweimal angewendet. Es ist also n(x 2 ) = 7, r1 є R2 (x2), r2 ∉ R(x2), r3 є R1 (x2), r4 є R2(x2).

Definition

9:

Es sei r G R (25)

(26)

w(r)

= Def

und R = R°. Σ

d(r,x)h(x), d. h. w(r) xєxo

= Σ h(x) rєR(x)

W(R) = Σ h(x) Def R∩R(x)≠ø w(r) ist also der "absolute Wert einer Regel r "bei der Bearbeitung der Objekte x: Dabei spielt es keine

JÜRGEN KUNZE

522

Rolle, wie oft r. und r. bei den einzelnen x angewendet werden. In Tabelle 1 ergibt sich w(r1)=

h(x 2 ) + h(x 3 ) + ...,

W({r2,r3})

= h(x ) + h(x2) + h(x3) + h(x4) + ... .

Es gilt fur beliebige r G R°, R', R" = R°: (27)

0 ≦ w(r) ≦ 1,

≦ Σ w(r) = E |R(x)|h(x) ≦ |R°| rєR° xєx° (|... | = Kardinalzahl von .. .) (28)

1

Die Gleichung in (28) folgt daraus, daβ h(x) genau so oft als Summand bei irgendwelchen w(r) auftritt, wie verschiedene Regeln bei der Bearbeitung von x durch V angewendet werden. (29)

W({r}) = w(r)

(30)

W(R' U R") ≦ W(R') + W(R"), wobei " < " auch dann gelten kann, wenn R' ∩ R" ≠ 0 ist.

W(R) ist ein Maβ für den absoluten Anteil der Regelmenge R bei der Bearbeitung von Objekten x durch V (vgl. (52)) . Auβer w(r) und W(R) kann man noch einen "relativen Anteil" fiir einzelne Regeln r und Regelmengen R definieren. Hierbei ist (im Gegensatz zu w und W) die Anzahl der Anwendung pro Objekt x in Betracht zu Ziehen. Definition

10:

Es sei r є R° und R = R°.

ZUVERLÄSSIGKEIT VON VERFAHREN

(32)

Q(R)

523

= Σ q(r) Def rєR

Aus Tabelle 1 ergibt sich z. B.

q(r) ensteht also so, da3 von jedem x, an dessen Bearbeitung r beteiligt ist, ein Bruchteil von h(x) zu q(r) zugeschlagen wird, der durch den Quotienten aus der Anzahl der Anwendungen von r und der Gesamtzahl von Regelanwendungen bei x gegeben ist. Der Ausdruck (31) kann umgeformt werden in

524

JÜRGEN KUNZE

wobei (34)

ai

(r)

=

Σ h(x) e(r,x)=l n(x)=i

+ 2

Σ h(x) e(r,x)=2 n(x)=i

+ i

Σ e(r,x)=i n(x)=i

+ ...

-!-

h(x) .

a.(r) ist die mit h(x) gewichtete Haufigkeit der Anwendung von r bei der vollständigen und korrekten Bearbeitung von Objekten x mit n(x) = i, wobei mehrfache Anwendungen mehrfach gezahlt werden. Die Funktion q erfüllt die charakteristischen Bedingungen (vgl. (1), (2)): (35) q(r) ≧ 0 fur alle r є R°, (36)

Σ q(r) = 1. rєR° Letzteres ergibt sich daraus, daβ die Summe aller q(r) nur eine Umverteilung der Summe aller h(x) ist. Ferner gilt (37) O = Q(0) ≦ Q(R) ≦ Q(R°) = 1 Schlie31ich kann durch h die Haufigkeit g(r) der einzelnen Regeln r (und entsprechend G(R) fur R = R°) definiert werden. Dazu benotigt man den Erwartungswert n der Regelanwendungen.

ZUVERLÄSSIGKEIT VON VERFAHREN Definition

11

(38) n

= E n(x)h(x) Def xєXO

525

:

n ist die mit h(x) gewichtete mittlere Anzahl von RegeLanwendungen pro Objekt x, also die mittlere Lange von x, gemessen in Regeln. Es ist n > 0. Definition

12:

Es sei r є R°, R

R° und n < ∞.

(39) g(r) = Def

Σ

(40) G(R) = Def

Σ xєX0

e(r,x)h(x) xєXO

=

E rєR

=

Σ g(r) rєR

E(R,x)h(x)

Σ xєXo

e(r,x)h(x)

(vgl. 11))

g(r) kann in folgender Weise als relative Häufigkeit interpretiert werden: Zu jedem x є X° wird ermittelt, welche Regeln aus R bei der Bearbeitung von x durch V° angewendet werden. Die Menge dieser Regeln ist gemäβ (17) bis (19) die Menge R(x) = R1 (x) U R2(x) U ... . Die Zahl n(x) sowie die Indizes i = e(r,x) bei R.(x) sind so festzulegen, daβ eine Regel r genau dann i Mal angewen­ det wird. Dieser Zusatz ist dann von Bedeutung, wenn x eine Ambiguität aufweist, die z. B. bei einer Analyseprozedur V zu verschiedenen Analysewegen führt, bei de-

526

JÜRGEN KUNZE

nen durchaus gewisse parallele Regelanwendungen auftreten können (Anwendung gleicher Regeln an gleichen Stellen). Ist z. B. Xo eine Wortformenmenge und gibt es in R genau eine Regel r zur Behandlung eines bestimmten Morphems (z. B. Präfix), so ist g(r) genau die relative Häufigkeit eben dieses Präfixes auf der Basis der Wortformenhäufigkeiten h(x). Gibt es mehrere derartige Re­ geln (z. B. r 1 , r 2 , r3,), so stellt die Summe (g(r1) + + g(r2) + g(r3)) die relative Häufigkeit des Präfixes dar. Entsprechendes ergibt sich für Syntagmen usw. Wir zeigen zunachst, daβ naheliegende Beziehungen zwischen q(r) und g(r) nicht bestehen:

Satz

1:

Es gilt im allgemeinen Fall weder q(r) ≧ g(r) noch q(r) ≦ g(r). Beweis: ergibt sich später aus Tab. 2. Wir können also unterscheiden: (a) die relative Häufigkeit einer Regel: g(r) (b) den relativen Anteil einer Regel: q(r) (c) den absoluten Anteil einer Regel: w(r) Dabei gilt (41) (42) (43)

0 ≦ g(r)

,

E g(r) = 1 rєRO 0 ≦ q(r) , E q(r) = 1 rGRO 0 ≦ w(r) ≦ 1 , Σ w(r) ≦ |R°| rєRO

Ferner foLgt aus (12), (25), (31): (44)

q(r) ≦ w(r) für alle r є R°.

ZUVERLÄSSIGKEIT VON VERFAHREN

527

(45) Wenn w(r) > 0, so q(r) > 0 für alle r є R°. Auβerdem gilt: (46) Weder g(r) ≧ w(r) noch g(r) ≦ w(r) im allgemeinen Fall. (47) Wenn w(r) > 0, so g(r) > 0 für alle r e R°. Zum Nachweis von (46) dient das folgende Beispiel: X = {X1, x2}, R = { r 1 , r2}.

r1

x

l

X

3

0

2

r2

0

1

n(x)

3

1

h(x)

½

½

e(r1,x])=3,

e ( r 2 , x 1 ) = 0 , e(r2,x2)=l

Tabelle

Man erhält:

e(r1,x2)=0

2

528

JÜRGEN KUNZE Somit gilt g(r2) < w(r2) und g(r ) > w(r ) , ferner

gilt g(r 2 ) < q(r2) und g(r ) > q(r )

(Beweis von Satz 1) .

Der Nachweis von (47) ist so zu fürhen: Wenn w(r) > 0, so gibt es ein x e X mit h(x) > 0 und d(r,x) > 0, also e(r,x) > 0, also tritt bei g(r) wenigstens ein positi­ ves Glied auf. (45) zeigt man ebenso. Die einzige noch interessante Beziehung ist zwischen q(r) und w(r) zu vermuten. Hier hat man das "statistische Gefuhl", daβ es bestimmte Zusammenhänge zwischen den Werten von q(r) und w(r) gibt. Das dies jedoch nicht im exakten Sinne gilt, soll zunächst durch den Satz 2 klargestellt werden, der seiner Natur nach wieder eine "Negativaussage" ist. Satz

2:

Es seien q und w beliebige reelle Zahlen mit 0 < q ≦ w ≦ 1. Dann gibt es ein Paar X und R (das Verfahren V spielt hierbei keine Rolle), eine Funktion h(x) und ein r є Ro mit q(r) = q und w(r) = w. Beweis: Es sei R

= {r,, r2}, X° = {x, x2, . . . } .

Wir legen fest:

ZUVERLÄSSIGKEIT VON VERFAHREN

X

h(x)

e(r1

1-w

m = 0 o

o Xl

m1

x2

m2

X. 1

mi

,x)

529

e(r2,x)

n(x)

n - m > 0 o o n1- m1

n

n2- m2

ni -mi

Tabelle

o

n1 n

2

ni

3

Ovenbar gilt, falls ni+i > ni > 0 für alle i > 0 ist, (48) w(r1) = Ferner ist nach (31) (49) q(r 1 ) = Die Folgen (mi) und (ni) wählen wir dabei so, daβ mi > 0 für alle i > 0 gilt und daβ ferner, falls wir (50) setzen, die Summe q1 + ... + q. die ten approximiert, d. h. da(3 (51)

q

-

2-1

≦

q1

+

•••

+

Zahl

q von un-

qi

gilt. Dies ist möglich, da jede reelle Zahl beliebig gut durch einen Bruch mit beliebig groβem Nenner approximiert werden kann. Ist rational, d. h. so kann

530

JÜRGEN KUNZE

man z. B. mi = i . m und ni = i . n wählen. Ist dies nicht der Fall, so schwankt der Wert von . Aus (51) und (49) folgt unmittelbar

urn den Wert q(r ) = q.

Bevor wir mit eingehenderen statistischen Betrachtungen beginnen, sei noch darauf hingewiesen, daβ die gegebenen Definitionen auch für andere Anwendungen geeignet sind. Man kann z. B. R als ein Alphabet, X als eine Wortmenge (etwa das Vokabular einer natürlichen Sprache) fiber R und h(x) als Frequenz von x (etwa wie in einem Häufigkeitswörterbuch) wählen. g(r) ist dann die Häufigkeit eines Buchstabens r aus R , w(r) die Summe der Häufigkeiten derjenigen Wörter, in denen ein Buchstabe r vorkommt. Die untersuchte Frage wandelt sich dann in folgende: Wie viele Wörter (prozentual) eines fortlaufenden Textes kann man mit einer Schreibmaschine schreiben, auf der die Typen für einige bestimmte Buchstaben (nämlich die aus R \ R) nicht funktionieren? Intuitiv ist klar, daβ das Fehlen von "e" (im Deutschen g("e")~ 0,2) die Schreibmaschine unbrauchbar macht, das Fehlen von "x" (g("x") ~ 0,001) dagegen kaum auffäit. 3. EINIGE STATISTISCHE BERECHNUNGEN Nachdem nun keinerlei Hoffnung mehr besteht, das gestellte Problem arithmetisch-algebraisch behandeln zu können, hat man sich nach statistischen Hilfsmitteln umzusehen. Genauer gesagt, es kommt darauf an, h(x) durch eine einigermaβen plausible wahrscheinlichkeitstheoretische Verteilung p(x) zu ersetzen, die h(x) hinreichend gut approximiert. Etwas Analoges mu3 man auch für n(x) tun. Wir gehen im folgenden von der Grundvorstellung aus,

ZUVERLÄSSIGKEIT VON VERFAHREN

531

daβ Ro in zwei Mengen R' und R" zerlegt wird. Dabei sind die r aus R" die Weglaβkandidaten, während R' diejenige Regelmenge ist, mit der V[R'] auf seine Zuverlässigkeit beurteilt werden soll. Setzt man R' = R, R'"" = R° \ R, so hat man nach (3) , (5) , (8) , (26) : (52)

Z(R') = H(M(R'))

= Satz

1 - W(R").

3:

Ist R' U R" = R° und R' ∩ R" = ø, so gilt Z(R') = 1 - W(R"). Der Verlust an Zuverlässigkeit beträgt W(R"). Es kommt also darauf an, W(R) aus 'bekannten" Gröβen zu berechnen. W(R) ist nicht additiv (vgl. (30)), was man auch in Tabelle 1 leicht verifiziert: W({r1}) + W({r 2 }) = W({r1,r2}) + h(x3) + ... . Wir können hier nicht alle denkbaren Approximationen für h(x) und n(x) betrachten und müssen uns auf eine Auswahl beschränken. Definition

13:

Es sei (53)

hi

= Σ h(x) Def n(x)=i

532

JÜRGEN KUNZE

hi ist also die relative Häufigkeit von Objekten x mit der Länge i (gemessen in Regeln r ) . Annahme

1:

Für h. liegt eine Poissonsche Verteilung vor, d. h. (54) hi =

e-n

Dabei ist e = 2,718 ... und n der Erwartungswert von n(x) (vgl. (38)). Annahme

2:

Bei fester Länge n(x) sind die Elemente r aus R

bino­

mial verteilt. In unserem Fall bedeutet dies, daβ bei Zugrundelegung von g(r) die Wahrscheinlichkeit für das j-malige Auftreten von r (nur darauf kommt es an, nicht auf die Stellen, wo r auftritt) in einem x der Länge i durch (55) gegeben ist. Insbesondere ist die Wahrscheinlichkeit des nullmaligen Auftretens durch (56)

= (l-g(r))1

gegeben, während das wenigstens einmalige Auftreten die Wahrscheinlichkeit (57) Pi = 1 - (1- h(r))i hat. Geht man zu den üblichen Wahrscheinlichkeitsmodellen über und betrachtet die r aus R' als rote und die r

533

ZUVERLÄSSIGKEIT VON VERFAHREN aus R" als grüne

Kugeln, die sich in einer Urne in sehr

gro3er Anzahl im Verhaltnis G(R') : G(R") (vgl. (40)) befinden, so hat man entsprechend (58)

=

G(R))i

(1 -

(59) Pi = 1 - (1- G(R))i Dabei ist P die Wahrscheinlichkeit dafür, daβ bei den Objekten x mit n(x) = i keine Regel aus R angewendet wird. Demzufolge kann man bei festem i die Wahrschein­ lichkeit daftir, daβ D(R,x) = 0 ist, (vgl. (13)), durch (1 - G(R)) approximieren. Dies ergibt insgesamt zunächst ftir R = {r}: w(r) =

Σ d(r,x)h(x) xєXo

=

E

ΣΣ

(vgl. (25):

d(r,x)h(x)

i=0 xєX° n(x)=i = E hi Σ d(r,x) i=0 XєX° n(x)=i =

Σ i=0

=

= e-n

(l-g(r))i)

hi(1-

e-n Σ i=0

(1 -

Σ i=0

=e-n . = 1 -

en e-n.

e-n

-

-e-n.en(1-g(r)) g

(r)

(1- g(r))i)

E i=0

(Summenglied für i=0 verschwindet) (vgl. (53))

(vgl.

(57))

(vgl.

(54))

534

JÜRGEN KUNZE

E n t s p r e c h e n d e r h ä l t man d i e Satz

Verallgemeinerung

4:

Ist R - R , so gilt unter den angegebenen Annahmen W(R) = 1 -

e-nG(R).

Zusammen mit Satz 3 ergibt sich (60)

Z(R') =

e-nG(R").

Das bedeutet, je kleiner das Produkt n.G(R") ist, desto geringer ist der Verlust an Zuverlässigkeit. Ist z. B. n = 5 (ein für Wortformen realistischer Wert, wenn man wieder an das Schreibmaschinenbeispiel und Buchstaben denkt) und G(R") = 0,05 so ist Z(R') = 0.779. Dieses Ergebnis ist äuβerst plausibel: Bei durchschnittlich 5 Buchstaben pro Wort wird ein Buchstabe aus R" durch­ schnittlich in 5.0,05 = 25 % der Wörter auftreten, also Z(R') ~ 0,75. Da3 Z(R') etwas gröβer ausfällt, liegt daran, daβ in einigen Wörtern mehrere Buchstaben aus R" auftreten, dafür in zusätzlichen anderen keiner. Wegen der Additivität von G(R) (vgl. (40)) kann man Satz 4 auch so ausdrücken: Nimmt man aus R die Regeln 1 2 r , r , ... heraus, so erniedrigt man die Zuverlässig1 2 keit des Verfahrens urn den Faktor exp(-n(g(r )+g(r )+ ...)). Damit verfügen wir über einen Zusammenhang zwischen G(R") und Z(R'). Dasselbe kann man nun auch für Q(R") und Z(R') versuchen. Dabei ergibt sich jedoch nichts Neues; denn es gilt: Satz

5:

Für alle r є R° ist q(r) = g(r), falls Annahme 2 gilt.

ZUVERLÄSSIGKEIT VON VERFAHREN

535

Beweis: Nach (33), (34) ist

Um komplizierte Rechnungen zu umgehen, betrachten wir z. B. i = 4. Es ist

a4(r) =

E h(x) -!- 2 E +3 E h(x) + e(r,x)=l e(r,x)=2 e(r,x)=3

Σ

+ 4

h(x) ,

e(r,x)=4 wobei nur über solche x zu summieren ist, für die n(x) = 4 gilt. Die Annahme 2 ergibt, falls wir g(r) = g und 1 - g(r) =  setzen: a4(r) = = h4[1 .4g 3 +2.6g 2  2 +3.4g 3 +4. 1 .g 4 ] = 4 g h4[3 + 3

= 4 g

h4

+ g)

= g

allgemein

g2

=4

+3g2

+g3]

h4g,

also

also

h4,

= g(r)hi

536

JÜRGEN KUNZE

∞ q(r) = g(r) Σhi = g(r). i=1 Der durch Satz 4 gegebene Zusammenhang zwischen Z(R') und G(R") ist praktisch noch nicht von groβem Wert, da die Häufigkeit g(r) oft nicht bekannt ist. Interessanter ist die Frage, wieviele Regeln aus R man weglassen kann, ohne mit der Zuverlässigkeit einen gewissen Wert zu unterschreiten. Dazu denken wir uns Ro = {r 1 , ..., r k } nach fallenden Werten g(r) angeordnet, d. h. g(r 1 ) ≧ g(r2) ≧ ... ≧ g(rk). Für diese Anordnung nach Rängen gilt das sog. Zipfsche Gesetz als eine Approximation: Annahme

3:

Die Werte g(r ) entsprechen der "harmonischen" Verteilung, d. h. es gilt (61)

g(r x )

wobei c eine Konstante ist. Weqen

können wir c = ansetzen. Wir wollen nun weiter annehmen, daβ die Anordnung r 1 , ..., rk wenigstens einigerma3en bekannt ist, d. h., daβ wir bei der Zerlegung von R in R' und R" in R' diejenigen rx mit niedrigeren Rän­ gen unterbringen, genauer, daβ R' = {r1, ..., rj} und R" = {r. , ..., rk} gilt. Dann ist die Frage zu beantworten: Wie kann bei gegebenem k die Zahl j gewählt wer-

537

ZUVERLÄSSIGKEIT VON VERFAHREN den, ohne daβ G(R") einen bestimmten Wert übersteigt? Wegen G(R') =

gilt Satz

5:

(bei Gültigkeit von Annahme 3 ) . Dieser Zusammenhang wird exemplarisch in der folgenden Tabelle 4 dargestellt. Dabei wenden wir zunächst nur Satz 5 an. Danach wird Satz 4 herangezogen, die Ergebnisse finden sich in Tabelle 5. Tabelle 4 ist so zu lesen, daβ z. B. G(R") = 0,05 ist, falls bei =0,25 3.10 10

1

2

0,40

0,5

0,8

0,9

0,95

0,98

0,99

0,20

0,07

0,03

0,015

0,006

0,003

0,30

0,15

0,05

0,02

0,011

0,004

0,002

0.24

0.12

0.04

0.02

0,009

0,004

0,002

3

0,20

0,10

0,03

0,02

0,007

0,003

0,001

3.103

0,17

0,09

0,03

0,01

0,006

0,003

0,001

10 4

0,15

0,07

0,02

0,01

0,006

0,002

0,001

3.102

10

Tabelle 4

538

JÜRGEN KUNZE

k = 100 Regeln in R° die Menge R' 80 (= 0,8k) und die Menge R" 20 Regeln enthält, d. h. die letzten 20 weggelassen werden. Tabelle 5 ist so zu lesen, daβ z. B. Z(R') = 0,61 ist, falls n = 10 und G(R") = 0,05 ist. Die unausgefüllten Stellen entsprechen Werten, die für Z(R') völlig undiskutabel sind (z. B. n = 50, G(R") = 0 , 2 : Z(R') = 4,5 . 10" 5 .

n = G(R") 0,2 0,1 0,05 0,02 0,01 0,005 0,002 0.001

2

5

0,67 0,82

0,37

.

0.61 0,78

0,37

0,90 0,96 0,98 0,99 1,00 1,00

0,90 0,95 0,98

10

0,61 0,82 0,90

0,99

0,95 0,98

0,99

0,99

20

50

. .

. .

0,37 0,67 0,82

0,37

0,90 0,96 0,98

0,79 0,90 0,95

0,61

Tabelle 5 Selbstverständlich hängen die bewiesenen Zusammenhänge (und damit auch die Tabellen) von den Annahmen 1 bis 3 ab, deren Approximationsgüte ohne Festlegung auf ein bestimmtes praktisches Problem nicht beurteilt werden kann. Gegebenenfalls hat man empirische Verteilungen zugrunde zu legen. Man kann auch andere mathematische Verteilungen annehmen, so z. B. für die Wortlänge in natürlichen Sprachen (sowohl in Buchstaben als auch Phonemen gezählt) eine solche, wo h. eine Normal-

ZUVERLÄSSIGKEIT VON VERFAHREN

539

verteilung über 1n i ist (vgl. Annahme 1) . Eine Stützung dessen findet sich in Herdans Type-Token Mathematics. Die Annahme 3 über die Rang-Häufigkeitsrelation muβ im konkreten Fall modifiziert werden. Wie wir oben schon dargelegt hatten, ist der Zusammenhang zwischen der Häufigkeit einer Regelanwendung und der Häufigkeit einer bestimmten linguistischen Einheit (z. B. Affix oder Syn­ tagma) möglicherweise recht kompliziert. Die Eliminierung der Regeln aus R" bedeutet ja nicht, daβ die zugrundeliegenden Einheiten ebenfalls ignoriert werden, sie können in R' noch eine Rolle spielen, aber ihre Behandlung ist dann teilweise fehlerhaft gegenüber der durch R . Dadurch ergibt sich eine weitere statistische Dimension, da die Einheiten dann in Abhängigkeit von Kontext oder Verwendung aufgeteilt werden müssen, was die ursprüngliche Rangordung zerstören kann. Auf der anderen Seite ist natürlich klar, daβ man bei praktischen Fragen der hier behandelten Art mit Faustregeln auskommt, und diese können aus den dargelegten statistischen Zusammenhängen hergeleitet werden. Durch die im vorangehenden Abschnitt gegebenen Gegenbeispiele darf man sich von der Anwendung statistischer Methoden nicht abschrecken lassen: Sie waren höchst unstatistisch und hatten nichts mit natürlichen Sprachen gemeinsam.

ILPO TAPANI PIIRAINEN Universität Münster

COMPUTERUNTERSTUTZTE UNTERSUCHUNG DER GRAPHEM- UND PHONEMEBENE

Seit F. de Saussure gilt das primäre Interesse der Linguisten der gesprochenen Sprache. Diese Tendenz wurde durch die Konstituierung der Phonemtheorie in der Prager Schule verstarkt und im amerikanischen Strukturalismus als die einzige Betrachtungsweise angesehen. Die Entwicklung der generativen Grammatik und der Valenztheorie rückten die gesprochene Sprache als ein Produkt des sprachlichen Erzeugungsprozesses erneut in den Vordergrund. Die Prager Schule entwickelte jedoch Theorien über die Literatursprache und die Sprachkultur, die eine vielseitige Forschung auch im Bereich der geschrie2 benen Sprache vorangetrieben haben. Aber auch bei anderen Forschern zeichnet sich ab, daβ die geschriebene Sprache auf mehreren Analyseebenen einen autonomen Charakter besitzt und nicht nur als Reflektierung der ge­ sprochenen Sprache anzusehen ist. So hat 1977 der Fennist P. Saukkonen anhand eines umfangreichen Textcorpus gezeigt, daβ die geschriebene Sprache von der informa­ tions- und kommunikationstheoretischen Sicht gedrangter und weniger redudant als die gesprochene Sprache ist. Im Vergleich mit der geschriebenen Sprache hat die ge-

542

ILPO TAPANI PIIRAINEN

sprochene Sprache einen kleineren Informationswert für eine konstante Menge von Oberflächenstruktureinheiten. Diese Feststellungen lassen die Schluβfolgerung zu, daβ sowohl die gesprochene als auch die geschriebene Spra­ che einen Eigenwert haben; sie konnen jedoch auch paral­ lel zueinander erforscht werden, urn die strukturellen Gemeinskamkeiten und Unterschiede der beiden Sprachformen aufzuzeigen. Gemeinsame Ansichten finnischer und tschechoslowakischer Forscher sind nicht nur in der neueren informa4 tionstheoretischen Linguistik festzustellen. Bereits in den 30er Jahren definierten gleichzeitig und unabhangig voneinander A. Penttilä in Helsinki sowie A. Artymovyč und J. Vachek in Prag das Graphem als eine eigenstandige Einheit in der Struktur der Sprache.5 A. Pent­ tila und J. Vachek sind spater auf die Graphemik zurückgekommen und haben beide noch in den 70er Jahren Monographien vorgelegt, mit denen sie die Entwicklung der Graphemtheorie wesentlich beeinfluβt haben.6 In den 60er und 70er Jahren ist die Graphemik als ein autonomes Forschungsgebiet weitgehend akzeptiert worden, und zahlreiche Untersuchungen vor allem aus dem deutschsprachigen Raum zeigen, daβ die Graphemik nicht nur theoretisch, sondern auch in ihrer praktischen Anwendung in der Beschreibung mehrerer Sprachen und Sprachstufen ein 7 zunehmendes Interesse findet. Dabei ist in Übereinstimmung mit J. Vachek zu betonen, daβ es sich urn die Erforschung von Graphemsystemen und nicht um die Aufstellung der Inventare von Graphen handelt, obwohl die Analysen bis in die Graphemik der Handschriften eindrin3 gen.

GRAPHEM- UND PHONEMEBENE

543

Bei der Definition des Graphems spielen die Beziehungen zwischen der graphemischen und der phonemischen Ebene eine wesentliche Rolle. Darauf ist zuruckzuführen, daβ sich in der heutigen Forschungslage vier verschiedene Graphembegriffe feststellen lassen, bei denen die Interdependenz der Graphem- und der Phonemebene unterschiedlich beurteilt wird. Die Graphemmodelle werden im 9 folgenden kurz angefuhrt. Am weitesten auf die funktionale Theorie der Prager Schule geht der Graphembegriff zurück, den der Verfasser dieses Beitrags 1968 in einer mit dem Computer durchgeführten Analyse des Frühneuhochdeutschen vorgelegt hat. Dabei wird im Anschluβ an J. Vachek die Autonomie der graphemischen Ebene betont und das Graphem als die kleinste funktionale Einheit auf der Ebene der Geschriebenen Sprache postuliert. Das Graphem ist eine Klasse von Graphen, die distributionell als eine sprachzeichenunterscheidende graphische Einheit bestimmt wird. Die Graphemanalyse wird aufgrund der Distribution (Verteilung und Kontext der Zeichen) und der statistischen Signifikanz der Graphe durchgefuhrt; eine Phonemkomponente wird bei der Bestimmung des Graphems nicht berücksichtigt. Da die Distribution und die statistische Sig­ nifikanz entscheiden, welche Zeichen funktional als Grapheme und welche als Varianten des Graphems anzusehen sind, ist eine weiterführende phonemische Interpre­ tation der Graphemebene jedoch möglich. Die Beriicksichtigung der Varianten ist besonders fur die Beschreibung von Sprachstufen aufschluβreich, in denen noch keine einheitliche Norm gilt, hat aber auch eine theoretische Bedeutung. So hat E. Stra(3ner in seiner ebenfalls compu-

544

ILPO TAPANI PIIRAINEN

terunterstiitzten Analyse von friihneuhochdeutschen Texten diesen Graphembegriff weiter ausgebaut und bewiesen, da3 die Distribution von einzelnen Graphemen strukturelle Merkmale aufweist; die Vokalgrapheme , , und im Friihneuhochdeutschen haben ahnliche graphemische Kontexte.11 Der zweite Graphembegriff ist 1966 von W. Fleischer fur die Erforschung des Friihneuhochdeutschen entwickelt worden. Nach der Definition W. Fleischers ist das Gra­ phem die kleinste distinktive Einheit der geschriebenen Sprache, die ein Phonem reprasentiert. Bei dieser Berücksichtigung der Phonemkomponente werden auch solche Zeichenverbindungen als Grapheme gewertet, die mit einem Einzelzeichen als einem Graphem nicht austauschbar sind (z.B. und , un im Friihneuhochdeut­ schen) oder sich auf ein einziges Phonem beziehen (z.B. die digraphische Schreibweise wie bei den Konsonanten oder bei Vokalen). Grapheme sind auch Zei­ chenverbindungen oder Einzelzeichen, die Diphthonge auf der phonemischen Ebene reprasentieren. 12 Diese Beispiele deuten darauf hin, als sei die Graphemdefinition W. Fleischers wegen des Heranziehens der Phonemwerte in der Theoriebildung widersprüchlich. Bei der Untersuchung der zahlreichen Graphemvarianten im Friihneuhochdeutsch­ en zeigt sich jedoch, daβ der Bezug des Graphems zum (hypothetischen) Phonem für sprachhistorische Untersuchungen manche Vorteile bietet; die auf der phonemisch­ en Ebene stattfindenden Anderungen konnen durch die Aufstellung des Graphem- und Variantensystems recht eindeutig erlautert werden. Für die Effektivitat der Vorgehensweise W. Fleischers sprechen auch die Arbeiten

GRAPHEM- UND PHONEMEBENE

545

über das Frühneuhochdeutsche, die seinen Graphembegrif f anwenden. Der dritte Graphembegriff verbindet die relative Autonomie des graphischen Bereichs mit der relativen Abhangigkeit der Graphemik von der Phonemik. Diese De­ finition, die als Graphonemtheorie bezeichnet werden kann, ist von J. C. McLaughlin fur die Beschreibung des Mittelenglischen aufgestellt worden. Danach sind Graphe­ me Gruppen von Schriftzeichen, die in Opposition zu anderen Schriftzeichen oder Schriftzeichengruppen stehen. Wenn eine graphische Einheit (ein Einzelzeichen oder eine Zeichenkette) ein Phonem reprasentiert, wird diese 14 ein Graphonem genannt. Diese Auffassung, die sowohl fur die alteren Sprachstufen als auch fur die Gegenwarts sprachen anwendbar ist, ist von R. Harweg weitergefiihrt worden.15 Die vierte Graphembegriff geht auf H. P. Althaus zurück und ist fur die Beschreibung altjiddischer Texte entwickelt worden. Darin werden ein Subgraphem, ein Graphem und ein Diagraphem unterschieden. Graphe bilden ein Subgraphem als eine Klasse von graphischen Einheiten. Das Graphem ist eine Klasse von Subgraphemen oder ein Subgraphem, das semantisch distinguierbar ist. Ein Diagraphem ist wiederum eine graphische Einheit, die einen komplexen graphemischen Charakter zeigt und auch mehrere Grapheme reprasentieren kann.16 Die Differenzierung von Subgraphemen und Graphemen schlieβt eine Reihe von phonemischen Erkenntnissen (vor allem im Bereich der Silbenstruktur) ein, so daβ dieser Graphembegriff durchaus Ähnlichkeiten mit den Graphemdefinitionen von W. Fleischer und J. C. McLaughlin erkennen läβt.

546

ILPO TAPANI PIIRAINEN

Bezeichnend fur die angeführten Graphemmodelle ist, da sie die Beziehungen der Graphemebene zur Phonemik nicht als strukturelle Äquivalenz betrachten, sondern nur in isolierten Fallen Phonemwerte hinzuziehen. Es ziegt sich, daβ über die Graphemik und Phonemik der Gegenwartssprachen kaum Arbeiten vorliegen, die etwa die Distribution der Grapheme und Phoneme zusammenhangend beschreiben.17 Erst eine Beschreibung der Grapheme mit ihren jeweiligen Kontexten und die ihres tatsächlichen und voraussagbaren Vorkommens in Texten würde es ermöglichen, die Strukturen der Graphem- und Phonemsysteme einander gegenüberzustellen. Auch im Finnischen, in dem eine groβe Ähnlichkeit zwischen dem Graphem- und dem Phonemsystem in der Gegenwartssprache existiert, zeigt ein Blick ins altere Schriftfinnische, daβ das Graphemsystem in alteren Texten uneinheitlich ist, zahlreiche Varianten aufweist und somit keineswegs dem Pho­ nemsystem entspricht; die Annaherung des Graphemsystems an das Phonemsystem ist ein sprachhistorischer Prozeβ, der bisher nich beschrieben worden ist.18 Die Erklarung von Haas, daβ die Graphemik und die Phonemik in einer Art Übersetzungsrelation zueinander stehen, berücksichtigt nur solche funktionale Aspekte des Graphem-Phonem-Verhältnisses, die nicht auf die Distribution zurück19 gehen. Das Beispiel des Finnischen zeigt auβerdem, daβ u.U. auch die historischen Aspekte bei der Beurteilung des strukturellen Graphem-Phonem-Verhaltnisses be-

r ü c k s i c h t i g t werden müssen. Zu einer theoretischen Klarung des Verhaltnisses zwischen der Graphem- und der Phonemebene wird hier eine quantitative Losungsmoglichkeit vorgeschlagen. Untersuchungen über die Phonotaktik des Deutschen zeigen, daβ

GRAPHEM- UND PHONEMEBENE

547

im Neuhochdeutschen bei den Grundmorphemen nur 15 Strukturtypen in der Reihenfolge der Vokale und Konsonanten vorkommen; ahnliche Analysen iiber die Graphemik liegen jedoch noch nicht vor. Es ist daher notwendig, die Distribution der Grapheme und Phoneme in gleicher Art zu ermitteln und die Frequenzen der jeweiligen Einheiten durch die Untersuchung eines statistisch reprasentativen Textcorpus zu klaren. Bis jetzt vorliegende, die Graphemik und die Phonemik vergleichende Untersuchungen aus dem Tschechischen zeigen, daβ die Graphemik eine 21 hohere Redundanz als die Phonemik aufweist. Dieses Ergebnis steht im Gegensatz zu den Informationstheoretischen Untersuchungen von P. Saukkonen, die am Anfang 22 dieses Beitrags angefuhrt wurden. Die Redundanz in der Graphemik spricht jedoch fur die Autonomie der Gra­ phemik und fur die Wichtigkeit der Untersuchung iiber die Interdependenz von Graphemik und Phonemik in weiteren Sprachen. Zur Klärung des Verhaltnisses zwischen der Graphe­ mik und der Phonemik treten neuerdings Arbeiten hinzu, die in mehreren Sprachen die automatische Umsetzung von Graphem in Phonem durchfiihren. Die Untersuchung W. Kastners iiber das Deutsche konzentriert sich auf die Umset­ zung von einzelnen Graphemen und Graphemfolgen (Klustern) und führt in Annaherungswerten auch die Frequenzen ein23 zelner Einheiten an. Diese Arbeit berucksichtigt auch die graphisch identischen Graphemfolgen, die unterschiedliche Phonemwerte haben (z.B. /kst/ in w ä c h s t , /Xst/ in suchst) und die Differenzierungsschwierigkeiten infolge der Wortbildung (Eid, aber Eidotter). Insgesamt läβt sich die Frage nach Umsetzung der graphemischen Texte in phonemische Texte als gelost ansehen, so daβ

548

ILPO TAPANI PIIRAINEN

auch wertvolle theoretische Erkenntnisse über das graphemisch-phonemische Verhältnis vorliegen. Für die slawischen Sprachen liegen erste graphe­ misch-phonemische Umsetzungsverfahren iiber das Tsche­ chische und das Polnische vor. Während H. Kučera die automatische Phonemisierung als Hilfsmittel für die Ermittlung der quantitativen Strukturen in der Phonemik 24 des Tschechischen ansieht, geht es bei M. Steffen-Batogowa urn ein moglichst vollstandiges Umsetzungsver­ fahren. Sie entwickelt ein Programm aufgrund von sieben Prinzipien, zu denen auch die Berücksichtigung der Wortbildung gehort. Sie gesteht aber auch die Schwierigkeiten des Verfahrens ein; durch die gro3e Anzahl der "orthographischen Buchstaben" und durch das Vorkommen der Grapheme und Graphemverbindungen in unterschiedlichen Kontexten sind die Regeln sehr zahlreich und kompliziert. Da das Umsetzungsverfahren erst in einem Systementwurf vorliegt, läβt es sich nicht beurteilen, inwieweit graphemische und phonemische Strukturen aufge25 deckt werden konnen. Die beiden angefuhrten Ansatze konnten jedoch dazu führen, daβ strukturell sehr ahnliche Sprachen wie das Tschechische und das Polnische im graphemisch-phonemischen Bereich auch in bezug auf die Distribution der Elemente einander gegenübergestellt werden. Die Arbeit E. Wolfs iiber die automatische Umsetzung des Schriftenglischen ins phonemische Englisch versucht ein Entdeckungsverfahren zu entwickeln. Dieses vermag einen regelmäβigen Umsetzungszusammenhang zu beschreiben, der zwischen den Komponenten von Paaren aus graphemischen und phonemischen Wörtern vorgefunden wird.26 Die Vorgehensweise schlie|3t die Eingabe iiber die graphe-

GRAPHEM- UND PHONEMEBENE

549

misch-phonemischen Entsprechungen ein, die nicht maschinell festgestellt worden sind. Es ist fur das Entdeckungsverfahren charakteristisch, daβ das entwickelte Umsetzungsverfahren auch hinsichtlich dessen Umsetzungsqualitat kontrolliert und bewertet. Die Ergebnisse dieser theoretisch interessanten Arbeit lieβen sich durch eine quantitative grapho- und phonotaktische Analyse fur praktische Zwecke relativ einfach erganzen. In der Untersuchung J. Rolshovens iiber das Franzosische werden ein graphemisches und vier verschiedene phonische Alphabete definiert. Dies ist notwendig, weil die strukturelle Phonemik mit dem Kriterium der Funktio27 nalitat zu unterschiedlichen Phoneminventaren kommt. In der ersten Phase der automatischen Bearbeitung wer­ den Texte mit Ziffernketten in graphische Texte umgewandelt. Im Hauptprogramm werden aus rein graphischen Eingabesatzen phonemische Ausgabesatze produziert. Es zeigt sich, daβ im Franzosischen z.B. im Falle der Liai­ son die Beschreibung von ganzen Sätzen notwendig ist; damit ist das Umsetzungsverfahren umfangreicher als im Englischen, in dem einzelne lexikalische Einheiten korrekt umgeschrieben werden können, und anspruchsvoller als im Deutschen, in dem eine Wort-für-Wort-Umsetzung, z.T. auch schon die Umsetzung von einzelnen Zeichen geniigt, urn phonemische Texte zu produzieren. Insofern geht das von J. Rolshoven entwickelte Umsetzungsverfah­ ren iiber das Graphem-Phonem-Verhaltnis hinaus und liefert auch Teile einer syntaktischen Analyse. Die angefuhrten Beispiele zeigen, daβ die automatische phonemische Transkription Methoden voraussetzt, die die Interdependenz der Graphem- und der Phonemebene in einem weiteren, morphemischen und syntaktischen Rah-

550

ILPO TAPANI PIIRAINEN

men behandeln und auch die quantitativen Aspekte berücksichtigen. Die Umsetzungsverfahren sind sprachenspezifisch, lassen jedoch eine Reihe von Parallelen aufzeigen. Die Typologie der Graphemik und der Phonemik, die als ein erstes erstrebenswertes Analyseziel gilt, läβt sich anhand der Umsetzungsmethoden auf die Ebene der graphemisch-phonemischen Beziehungen struktureller Art 28 erweiteren. Die Erkenntnisse über die Struktur der Graphem- und Phonemsysteme, die auch die Distribution und die Frequenz beriicksichtigen, sind in der automatischen Textverarbeitung z.B. für die automatische Doku29 Die praktische

mentation von Texten von Wichtigkeit.

Anwendung der Analyseergebnisse reicht jedoch in weitere Bereiche hinein, z.B. in die Untersuchungen über die Lesbarkeit der Texte sowie über die Regeln und Schwierigkeiten der Orthographie von Einzelsprachen.

GRAPHEM- UND PHONEMEBENE

551

ANMERKUNGEN 1

2

Piirainen, I. T. 1968. Graphematische Untersuchungen zum Frühneuhochdeutschen. Berlin. 12-19.

Vachek, J. 1973. "The Present State of Research in Written Language". Folia Linguistica 6. 47-60. Havranek, B. 1976. "Die Aufgaben der Literatursprache und die Sprachkultur". Grundlagen der Sprachkultur. Beiträge Prager Linguistik zur Sprachtheorie und Sprachpflege.

der Teil 1.

Hrsg. von J. Scharnhorst und E. Ising. 103-141. Berlin. Jedlicka, A. 1966. "Zur Prager Theorie der Schriftsprache". Travaux Linguistiques

de Prague 1. 47-58.

Jedlicka, A. 1982. "Theorie der Literatursprache".

Grundlagen

der Sprachkultur. Beiträge der Prager Linguistik zur Sprach­ theorie und Sprachpflege. Teil 2, hrsg. von J. Scharnhorst

3

und E. Ising. 40-91. Berlin. Saukkonen, P. 1977. "Spoken and Written Language". Folia guistica. 11. 207-215.

4

Vgl. Sammelband Teorie

5

Artymovyc, A. 1932. "Fremdwort und Schrift". Charisteria Mathesio. 114-117. Prag. Penttilä, A. 1932. "Grafeema- ja foneemasuomen suhteesta (Über das Verältnis des Graphem- und Phonemfinnischen) ". Virittäjä. 36. 26-26. Vachek, J. 1939. "Zum Problem der geschriebenen Sprache".

6

Penttilä, A. 1970. "Zur Grundlagenforschung der geschriebenen

Travaux du Cercle

informace

Linguistique

a jazykověda.

Lin­

1964. Praha.

de Prague 8. 94-104. Prague.

Sprache". Acta Societatis Linguistica Upsaliensis, ries. 2.2. Uppsala. 21-55. Vachek, J. 1973. General Problems and Problems of

Nova Se­ English.

Paris: The Hague. Eine deutsche Übersetzung dieses Buches liegt vor in: "Geschriebene Sprache. Allgemeine Probleme und Probleme des Englischen". Grundlangen der Sprachkultur. Beiträge der Prager Linguistik zur Sprachtheorie und Sprachpfle­ ge. Teil l, hrsg. v. J. Scharnhorst und E. Ising. Berlin

7

1976. 240-295. Brekle, H. E. 1971. "Einige Bemerkungen zur Graphematik - Diskussion". Linguistische Berichte. 16. 53-59. Piirainen, I. T. 1983. "Entwicklung der Graphemtheorie". All­ gemeine Sprachwissenschaft, Sprachtypologie und Textlinguis-

tik. Festschrift für Peter Hartmann, hrsg. von M. Faust, R. Harweg, W. Lehfeldt und G. Wienold. 249-257. Tübingen. 8

Allen, S. 1965. Grafematisk analys som grundval för textedering med särskild hänsyn t i l l Johan Ekeblads brev t i l l bro-

dern Claes Ekeblad 1639-1655 (Die graphematische Analyse als Grundlage für die Textedition mit besonderer Berücksichtigung zu Johan Ekeblads Briefen zum Bruder Claes Ekeblad 1639-1655). Göteborg.

ILPO TAPANI PIIRAINEN

552

Feigs, W. 1979. Deskriptive Edition auf Allograph-, Wort- und Satzniveau, demonstriert an handschriftlich überlieferten, deutschsprachigen Briefen von H. Steffens. Bern & Frankfurt

a.M. & Las Vegas. Grothausmann, K.-H. 1977. Das Stadtbuch von Karpfen (Krupina). Edition, Darstellung der Graphien, Glossar. Bern & Frankfurt

a.M. & Las Vegas. Hakkarainen, H. J. 1971. "Graphemik und Philologie". für Literaturwissenschaft

9

10

und Linguistik.

Zeitschrift

1. H.1-2. 191-204.

Singer, H. 1971. "Der Graphembegriff bei der Analyse altdeutscher Handschriften". Linguistische Berichte. 13. 83-85. Althaus, H.P. 1980. "Graphemik". Lexikon der germanistischen Linguistik. 2. Auflage, hrsg. von H. P. Althaus, H. Henne und H. E. Wiegand. 142-151. Tübingen. Piirainen, I. T. 1968. Graphematische Untersuchungen zum Frühneuhochdeutschen. Berlin.

Piirainen, I. T. 1971. "Grapheme als quantitative Gröβen". Linguistische Berichte. 13. 81-82. 11 Straβner, E. 1971. Graphemsystem und Wortkonstituenz. Schreibsprachliche Entwicklungstendenzen vom Friihneuhochdeutschen zum Neuhochdeutschen untersucht an Nürnberger Chroniktexten.

Tübingen. 12 13 14

Fleischer, W. 1966. Strukturelle Untersuchungen zur Geschichte des Neuhochdeutschen. Berlin. Fleischer, W. 1970. Untersuchungen zur Geschäftssprache des 16. Jahrhunderts in Dresden. Berlin. McLaughlin, J. C. 1963. A Graphemic-Phonemic Study of a Middle English Manuscript. The Hague.

15

Harweg, R. 1966. "Das Phänomen der Schrift als Problem der historisch-vergleichenden Sprachforschung". Kratylos. 11. 33-48.

16

Althaus, H. P. 1972. Die Cambridger Ldwenfabel von 1382. Untersuchung und Edition eines defektiven Textes. Berlin & New

17

York. Ludvikovā, M. und Königovā, M. 1967. "Quantitative Research of Graphemes and Phonemes in Czech". The Prague Bulletin of Mathematical

Linguistics.

7. 15-29.

Piirainen, I. T. 1969. "Quantitative Analyse der Graphem- und Phonemebene". The Prague Bulletin

18

of Mathematical

11. 3-15. Ojansuu, H. 1909. Mikael Agricolan k i e l e s t ä von Mikael Agricola). Helsinki. Rapola, M. 1965. Suomen kirjakielen

historia

Linguistics.

(Über die Sprache pἄἄpiirteittἄin.

I.

(Die Geschichte der finnischen Schriftsprache in ihren Grundzügen.I.). Helsinki. 19

Haas, W. 1970. Phono-graphic

20

Die deutsche Sprache, 1970. Hrsg. von E. Agricola, W. Fleischer & H. Protze. Band 2. 807-810. Leipzig.

Translation.

Manchester.

553

GRAPHEM- UND PHONEMEBENE

Kučera, H. und Monroe, G. K. 1968. A Comparative Phonology of Russian,

21

23

24

Quantitative

New York.

Vgl. auch Roceric-Alexandrescu, A. 1968. Fonostatistica limbil române. Bucuresti. Dolezel, L. und Prucha, J. 1966. "A Statistical Law of Graphe­ me Combinations". Prague Studies in Mathematical Linguistics. 1. 33-43. Ludvíkovā, M. und Königovā, M. 1967. "Quantitative Research of of Graphemes and Phonemes in Czech". The Prague Bulletin Mathematical

22

Czech and German.

Linguistics.

7. 15-29.

Saukkonen, P. 1977. "Spoken and Written Language". Folia guistica. 11. 207-215. Kἄstner, W. 197 2. Automatische Texte im Deutschen. Hamburg.

Phonemisierung

Lin-

orthographischer

Kučera, H. 1963. "Mechanical phonemic transcription and phoneme frequency count of Czech". International Linguistics and Poetics. 6. 36-50.

Journal

of

Slavic

25

Steffen-Batogowa, M. 1974. "Prinzipien der automatischen Pho­ nemisierung orthographischer Texte im Polnischen". Hambur­ 13. 117-161. ger Phonetische Beitrἄge.

26

Wolf, E. 1977. Vom Buchstaben zum Laut. Maschinelle Erzeugung und Erprobung von Umsetzautomaten am Beispiel SchriftenglischPhonologisches Englisch. Braunschweig. Rolshoven, J. 1978. Automatische Transkription franzősischer

27

Texte.

Tübingen.

Fonematycznej

Tekstów Polskiego.

1975. Warszawa.

28

Skalička, V. 1966. "Konsonantenkombinationen und linguistische

29

Piirainen, I. T. und Wenzel, G. 1971. Contribution

Typologie". Travaux Linguistiques Foundation

of Automatic

de Prague.

Text Processing

1. 111-114.

to the

(Documentation).

Heidelberg. Wenzel, G. 1967. Textverarbeitung

auf der

Graphemebene.

Stuttgart. Wothke, K. 1983. Automatisierung analysen. Tübingen.

strukturalistischer

Phonem-

PAUL L. GARVIN Buffalo, New York

SEMIOTIC ASPECTS OF MACHINE TRANSLATION

There is a trivial way of looking at this question: machine translation belongs into semiotics because semiotics deals with sign systems, language is a sign system, and machine translation deals with language. Once this has been said, what can be said next? In other words, are there nontrivial connections between machine translat­ ion and semiotics? In my view, there are at least two such connections; they are connections between the theoretical foundations of machine translation and aspects of semiotic theory. One of these has to do with general sign theory and how it relates to and can be applied in machine translation; the second concerns the possibility of relating the problems of esthetic translation by machine (albeit only experimentally) to some theoretical principles of esthetic semiotics. 1. GENERAL SIGN THEORY From the beginning of my interest in machine trans­ lation (see Garvin 1956: 182-4) I have always felt that one important area of semiotic theory had direct bearing on machine translation, namely, Karl Bühler's doctrine of the two fields (Zweifelderlehre, 1934: 119). Bühler

556

PAUL L. GARVIN

divides the general environment in which signs function, their surrounding field (Umfeld, 1934: 52), into a symbolic field (Symbolfeld, 1934: 149-154) and a deictic field (Zeigfield, 1934: 79-82). The symbolic field consists of those other signs of the same system to which the sign under consideration is directly related - in the case of language, the strictly linguistic context. The deictic field includes all the remaining context in the case of language, what might be considered the extralinguistic setting. The question which then arises and which is the most important from the standpoint of machine translation concerns the boundary between the two fields. The importance of this boundary derives from the fact that the relations which a sign has to other signs of the same system with which it is connected are of a different kind than the relations that a sign has to its wider environment - that is, relations within the symbolic field are of a different kind from those within the deictic field. The significance of this difference for machine translation is that this differ­ ence in the relations requires a different kind of processing for the two kinds of fields. The question of the boundary between the two fields can for machine translation purposes be resolved by assuming that the symbolic field is limited to the particular sentence (in the broad sense, including compound sentences) in which a given linguistic unit is contained. This is based on the underlying further assumption that properly linguistic relations in the strict sense are limited to the sentence: relations between sentences (in spite of such well-known phenomena as anaphora) are not of the same strictly linguistic

SEMIOTIC ASPECTS OF MACHINE TRANSLATION

557

sort (e.g., the antecedent of an anaphoric pronoun may be a picture). From a machine translation standpoint it can further be said that relations within the sentence are deterministic in nature, i.e., capable of being processed by more or less an algorithmic process based on parsing. Relations beyond sentence boundaries, on the other hand, tend to be probabilistic. This means that algorithmic processing will not suffice and other kinds of processing (such as possibly heuristic) will have to be considered; much less is known about the latter than the former. This means that the resolution of ambiguities based on conditions that transcend sent­ ence boundaries will be based on different and less clearcut principles that that for which conditions can be ascertained within the sentence. The latter can to a large extent be resolved on the basis of an adequate parse of the sentence, while the former will require more complex and less clearcut probabilistic solutions. This can be illustrated by the lexical translation ambiguity presented by the Russian word pa6oTa which is rendered in English as either "work" (work accomplished, as on the job) or "paper" (a paper written, as in a scholarly journal). In example (1), the linguistic context (following (1) PaoTa B. . BaHOBa Ha paper/work (by) V.I. Ivanov on

Ty TeMy ... this topic ...

genetival modifier consisting of a nominal structure identifiable as a proper name by virtue of its graphic shape) serves to resolve the translation ambiguity and to select "paper" as the appropriate translation. In

PAUL L. GARVIN

558

example (2), on the other hand, the translation (2) B.

.

BaHOB

V. I. Ivanov Ta

Haca

CTaTbIO

wrote (an) article

Ha

Ty

on this

TeMy. topic.

paoTa ...

This paper/work... ambiguity is not resolvable by just parsing either of the two sentences, but only on the basis of drawing upon the relation between the two sentences: the word CTaTbIO in the first of the two sentences of which the example consists constitutes an antecedent of sorts of the word paoTa contained in the second sentence. This relation is not identifiable on the basis of parses but might be dealt with by some probabilistic mechanism such as, perhaps, an "antecedent storage" - a running store of the semantic categorization labels of all the relev­ ant terms (however 'relevant1 is defined for a given translation system) encountered in the sentences already processed, on the basis of which a resolution of the translation ambiguity encountered in the current sent­ ence can be attempted. Thus, in the case of example (2), terms such as the already noted CTaTbio,

as well as

HanHcaji, encountered in the immediately preceding sent­ ence (and presumably other terms in earlier sentences) will carry semantic categorization labels favoring the choice of a translation such as "paper", thereby resol­ ving the ambiguity. A similar ambiguity resolution, but with choice of the other translation alternative, can be suggested for examples (3) and (4).

SEMIOTIC ASPECTS OF MACHINE TRANSLATION

(3)

Pa6oTa

B

paper/work

in this

(4)

p

during

TO

apKe

pa

bI

times

was

In example

...

factory...

CTpOKe

MeTpO

construction

559

(of the)

epeBbIoHeH overfulfilled

HeCKObKO

metro

several

aH.

PaoTa. . .

(the) plan. Paper/work...

( 3 ) , the linguistic context

(following

prepositional modifier consisting of a prepositional structure attachable to the governing noun through an appropriate government code) serves, as did the linguist­ ic context in example

( 1 ) , to resolve the ambiguity and

to select the appropriate translation - this time, "work" and not "paper". In example

( 4 ) , as in example

( 2 ) , the

ambiguity again is not resolvable on the basis of pars­ ing either of the two sentences but only by drawing upon the relation between them: the words CTpoHKe aH

and

in the first of the two sentences constitute ante­

cedents of sorts of the word paoTa with which the second sentence begins. Once again, this relation will require something like an "antecedent storage" for its identification, in which terms such as CTpoKe or

aH

will carry the semantic categorization labels allowing the selection of the translation alternative "work" rather than "paper", thus serving to resolve the ambiguity. The antecedent storage could be combined with some wider-reaching technique such as

information-retrieval

methods serving to establish the general character of a text

semantic

(e.g., some form of key-word-based

560

PAUL L. GARVIN

processing), especially for those cases of ambiguity where the immediately adjacent sentences do not provide the needed information. The link between this problem area and some of the current arguments about the semantic aspect of linguistic competence versus "real world knowledge" is evident. 2. ESTHETIC SEMIOTICS The consideration of machine translation of esthetic text (such as literary passages) will to many appear as a contradiction in terms: how can a machine program be expected to perform (or simulate) an essentially creat­ ive activity? Thus, in order to consider the possibility of machine translation - or, more realistically - mach­ ine experimentation - in this area, it will therefore be necessary first of all to find an approach to the study of esthetic text which does not posit creativity (or beauty, or some other noncomputable property) as the primary defining criterion of esthetic phenomena. Such an approach could well be based on Jan Mukařovský's concept of foregrounding as the basic explan­ atory principle of esthetics. This concept is based on the notions of the expected as over the unexpected - a conception which does not require recourse to the non-computable properties noted above. Foregrounding is the unexpected (unexpected behavior, unexpected features of objects), highlighted against a background of automat­ izations which are the expected (expected behavior, expected features of objects; cf. Mukařovský 1932: 19). Foregrounding may occur under ordinary, everyday circum­ stances in an unstructured manner (as in spontaneous joking and punning); in esthetic texts and other

SEMIOTIC ASPECTS OF MACHINE TRANSLATION

561

esthetic objects, it occurs in a highly structured and systematic manner (such as the highlighting of certain linguistic features, for instance rhyming phonological elements, in a text). In Mukařovský's terms this constitutes the difference between the structured and the unstructured esthetic (see 1948: 31-32). In the processing of text, the difference between the expected and the unexpected seems to me to be computable, at least in principle, since under the expected one can include the kinds of features, terms, expressions and constructions most commonly found under certain definable conditions. One can then assume that anything other than the latter would be unexpected and hence foregrounded. This is particularly applicable to the structured esthetic, since the unexpected phenomena, the foreground­ ings, will here occur systematically. Recognition rout­ ines can then be envisioned that will identify such structured foregroundings, bysed on both formal linguist­ ic properties, and (though undoubtedly less easily) on semantic ones. An example of the machine recognition of formal properties might be the identification of alliteration or rhyme. Such poetic devices could conceivably lend themselves to identification through some phon signed simply to "spot" foregrounded passages, that is, a recognition routine such as that just discussed, without a command routine for producing translated equivalents. One way of designing such a "foregrounding spotter" would be to design a routine for identifying expected structures and the assuming that foregrounding is present whenever the routine fails because of the unexpectedness

562

PAUL L. GARVIN

of the structures encountered. This could be combined with an interface capability allowing a human investigator to examine the text and the results of its machine processing, and to give his interpretation. Another point made by Mukařovský is relevant here, namely, that foregrounding occurs against two automatized backgrounds. In the case of esthetic text, one of these backgrounds is ordinary language usage, the other is the esthetic canon of the period (1932: 22). I am inclined to make his statement more specific and characterize the second background as the esthetic canon of a given readership or audience. The machine analysis of both kinds of backgrounds is conceivable (although, need I say, not immediately implementable): a known esthetic canon can, in principle, be incorporated in a recognition routine just as can a known linguistic usage. Thus, the role of machine translation research here will clearly not be one of contributing directly to the achievement of literary translation, but - as I have already noted - one of serving as a testing ground for crucial esthetic semiotic notions. REFERENCES Bühler, K. 1934. Sprachtheor ie. Jena: Gustav Fischer. Garvin, P. L. 1956. "Some linguistic problems in Machine Translation. For Roman Jacobson ed. by II. G. Lund et al. The Hague: Mouton. Garvin, P. L., ed. & transl. 1964. A Prague School Reader

on

Esthetics,

Literatury

Structure

and

Style.

Washington, D.C.: Georgetown University Press. Mukařovský, J. 1932."Jazyk spisovny a jazyk bāsnický".

Spisovna

čeština

a jazykova

kultura

ed. by B. Havra-

nek & M. Weingart. 123-149. Prague; quoted from Engl. transl. Standard Language and Poetic Language,

SEMIOTIC ASPECTS OF MACHINE TRANSLATION

563

in Garvin 1964. 17-30. Mukařovský, J. 1948. "Estetika jazyka". Kapitoly z ceske poetiky, 2nd ed., vol. I, 41-77. Prague. Quoted from Engl, transl. The Esthetics of Language, in Garvin 1964. 31-69. Mukarovsky, J. 1972. "Vančurovskā prolegomena". Cestami

poetiky

a estetiky.

242-245. Prague.

WALTER V. HAHN Hamburg

SOME IDEAS ABOUT DENSITY IN KNOWLEDGE BASES

I.INTRODUCTION The more Artificial-Intelligence systems meet realistic applications, the more they must cope with the vast amount of details contained in their knowledge bases. Two questions which arise are: - how to avoid communicative inadequacy resulting from too specific or too general natural language utterances of the system, and, from a more technical point of view, - how to control the access to and the inferencing on extrem­ ely dense knowledge bases, which by definition must be as coherent as possible. We are confronted with - the knowledge base paradox: the more knowledge a system has, the more stupid it will appear because of its inabil­ ity to cope with large amounts of detailed knowledge. The underlying assumptions of this paper can be summar­ ized as follows: (a) Knowledge bases have hierarchies, e.g. more general and more specific parts, (b) experts in the subject field can make judgments about the density of a knowledge base or parts of it, (c) experts have specific techniques and can recognize natural language indicators for jumps from a more detailed part of their

(natural) knowledge base to

a less detailed area and vice versa,

WALTER V. HAHN

566

(d) density or detailedness belong to a quite different dimension than preciseness or exactness. In this paper I give an incomplete and in most parts very associative overview of the correlated problems and present some initial heuristics for coping with density problems. Starting point and permanent source of encourage­ ment was Hayes (1979). Let me in a first sketch take aim at the density problem by separating it for the moment from two closely related problems, that of very complex knowledge bases and that of very large knowledge bases. Hayes characterizes density as follows: 'A dense formalization has many links between the separate concepts expressed by tokens in the formalizat­ ion. Density is clearly a matter of degree ' (Hayes 1979: 245) . Dense knowledge bases thus have elements, with a lot of parts, subparts, attributes as values of many differ­ ent dimensions, etc. The first problem is: How to find the adequate level of detail, how to control, e.g., inheritance and how to connect the proper parts in a situation of defined abstr­ action. Let's abbreviate this first meaning of 'density1 as the 'kitchen-type problem'. Complex

knowledge bases consist, in contrast to this

type of numerous heterogeneous parts including the cor­ responding very complex inference component. These know­ ledge bases are not necessarily dense. On the contrary, most of the theoretically important implications, e.g., of semantic networks have been demonstrated with complex but small experimental nets, a fact which was already critisized by Hayes (1979). Let's call this the 'broker-type problem'.

DENSITY IN KNOWLEDGE BASES

567

The other 'near miss' is the mass-problem in know­ ledge bases, which in most cases can be understood as how to represent and use a huge body of uniform or unified knowledge. Let us call it the 'stadium-type problem' of knowledge bases. Certainly in reality all three problem types occur more or less mixed. For purposes of demonstration, I will take the envir­ onment of the system HAM-ANS (see Hoeppner, Christaller et al. 1983). My examples of a 'desk-world' can be looked upon as a fragment of the hotel manager situation of the HAM-ANS applications, though none of the example sentences are real utterances of the system. The main hypothesis of this paper is that density problems cannot be solved by more elaborated types of net formalisms or more complicated algorithms for the infer­ ence machinery, nor by a better flow of control, but can only be tackeled by a group of techniques simulating natur­ al (language) management of dense knowledge. As a consequence of the experience with user's difficulties in building up and handling dense knowledge bases one might be tempted to formulate the problem in a very provocative fashion: If a user (e.g., of an expert system) creates knowledge representations (or at least tries to do so), which are not consistent in terms of syntax or logic, one might be able to show him, that the inferences will run into difficulties, but it is impossible to prevent the naive user from understanding all the utterances of the system according to his own 'faulty' taxonomy. So, if the naive user is a reliable expert in the domain, he is not responsible for his unsuccessful attempt, but rather the designer of the formalism or the guardian of the metarules.

WALTER V. HAHN

568

We will pursue this line in a more moderate way by gathering arguments, examples and counterexamples from the fields of natural language semantics, representation, ontology, and pragmatics, 2. SOME EXAMPLES As an answer to 'What sort of things do you offer in your hotel?' it is certainly inadequate to reply *'bedside tables, plugs and faucets' *'137 rooms' or

or

*'things and living beings' (asterisk marks inadequateness). You would rather expect answers like 'rooms, conference facilities, party service, travel agency . . . ' If you ask 'What is on your desk?' you would not expect *'a writing-pad' as the first thing to be mentioned, but rather 'a mess of books and papers'. As a definition of 'chair' in most pragmatic settings, it is much more adequate to use 'a seat, which..,' in comparison to *'a piece of furniture, solid, with four legs, consists of back and seat, normally wooden'. If you are looking for a certain letter 'Where is the letter from Plietsch & Fix?' it is useless to learn that the letter is *'in the office'

or

DENSITY IN KNOWLEDGE BASES

569

*'near the letter from Ratsch & Rammel'. Most likely you would expect a sentence like 'In the file on the right corner of the desk in my office'. A last example: 'What's in this book?' *'pages'

or *'characters'

'fiction'

but

or

'the well known story of John and Mary retold for all AI-researchers'. Though the stared sentences seem to be very odd in most contexts, even in LSP settings (language for spec­ ial purpose, sublanguages) one is frequently confronted with this type of communicative inadequacy (see Grice's maxims (1975) ) . How do natural speakers avoid or repair these communicative inadequacies? First of all with natural language comments. As cooccurrent comments for adjust­ ment one often hears 'roughly speaking...' 'in general... ' 'in specific...'

or

'in a very specific way...'

or as responses: 'no, I meant that very generally...' 'ah, I see, let me think a moment...' Many other adverbs like 'normally', 'naturally', 'certainly' or certain verbal constructions with 'use', 'get' are also vague indicators of changes in the 're­ solving power' of the applied inference mechanism.

WALTER V. HAHN

570

3,

SEMANTICS

In this section I will discuss some semantic difficult­ ies of density. Semantic nets are in most cases labeled with a natur­ al language designator the purpose of which is to show the plausibility of the existence of a node and to cover the 'cognitive resolution' of the domain (this seems to be an unspoken and newly introduced similarity to an old theorem of J. Trier's 'Wortfeld'). The use of the nets for a valid inferencing requires that the cognitive depth of all the branches of the conceptual hierarchy be at least similar. From the linguist's point of view, three properties of natural language will complicate the construction of homo­ geneously dense semantic nets and cause difficulties in their use. (1) It is disputable that concepts and the correspond­ ing designators show as their most obvious structure, a hierarchical one: ',.. it is as yet impossible to evaluate, even for well studied and easily accessible European languages, the hypothesis that the vocabulary is hierarchically order­ ed, as a whole, in terms of hyponymy and quasihyponymy; the theoretical semanticist should be correspondingly cautious about putting forward general hypotheses of this kind'. (Lyons (1977:301)). Experience with general classifications such as those of Hallig and Wartburg (19 52) show that the more one moves upwards in the hierarchy the more questionable is the system's plausibility. Lyons admits, that there are, especial­ ly in technical fields, partially (e.g. hierarchically)

DENSITY IN KNOWLEDGE BASES

571

ordered regions of the vocabulary, but at the same time he shows that 'indeed any particular part of it, may be structured hierarchically from a point which itself is not associated with an actual lexeme' (Lyons (1977: 297)). Another difficulty arises from the internal struct­ ure of the vocabulary: the parts of speech cannot be handled in a uniform way. Even if one tries to order them in one hierarchical structure each, the links between the elements of the hierarchy are of different types and belonging to different inference and metarule (e.g., transitivity) classes. (2) Contrastive linguistics has spent a great deal of time on the topic of lexical gaps and codability. Even though AI research in most applications is not concerned with interlingual processes, knowledge eng­ ineers are sometimes informed by native speakers, that there is no simple designator in the language for a specific concept of a field. Hence, lexical gaps coincid­ ing with conceptual nodes cannot be labeled with a simple word or cannot be labeled at all. Codability, certainly, is a consequence of the cultural role which the object plays in normal life, but lexical gaps prevent homogene­ ous density of semantic networks. On the other hand, it seems to be very difficult to prove the existence of a concept or even the usefulness of a hypothetical concept, if you don't have the evidence of a natural language word, (3) Given a hierarchical skeleton of a specific field of knowledge resp. vocabulary of a natural lang­ uage, in the process of filling up this schema with highly detailed items many words are found which fit into several hyponyms, e.g. a 'piano' (piece of furnit-

WALTER V. HAHN

572

ure, musical instrument, e t c . ) . If multiple set member­ ship is allowed, difficulties arise with cyclic structures, a property which restricts the inference techniques very persistently. (4) There are many natural language words, which are famous among linguists for their specific referential attitudes, e.g., mass terms words

('gold', 'wood') or partitive

('border', 'contents', 'element'). The evaluation

of expressions with those words is extremely difficult in all networks connecting conceptual data with referent­ ial data

(see Bunt 1 9 7 6 ) . Similarly the evaluation of

natural language quantifiers in a semantic network

(e.g.,

as a specification for the part-of-relation) is at least extremely time consuming in very dense knowledge bases, because mappings of quantifiers onto numerical values must be controlled by utilizing sort restrictions and/or prototype knowledge

(Wahlster 1981: 1 2 5 ) .

(5) Semantic decomposition is a linguistic method of reducing the set of semantic descriptions by use of semanticly primitive elements

(Sampson 1 9 7 9 ) . In AI this techn­

ique is widespread in very different representation lang­ uages. With dense representations, at least four difficult­ ies are encountered: Lexical decompositions have, partly depending on the parts of speech, involved in them, an internal syntax which can be linear or hierarchical: 'kill' =

(cause

(become

(not

(alive))))

'run' = Move ∩ upright ∩ on ground ∩ fast This example shows that in contrast to net hierarch­ ies lexical decompositions are not limited to one class of words. If the representation of the presumed dense knowledge base provides decomposition techniques, the interaction of both levels of detail, the net and the

DENSITY IN KNOWLEDGE BASES

573

decomposition, must be controlled. The control is much more complicated if the primitives as well as the compos­ ed words are elements of the net. This might be necessary, because, for example, not all the elements of 'fast ∩ move' are contained in decompositions so that semantic processing needs direct access to a net entry. In any case, the semantic handling of decomposed vocabulary and the interaction with other levels of detail (and other 'syntaxes' of detail) again show that dense knowledge bases require very complex processing. (6) The complexity of processing needed to handle specific links in semantic networks, e.g., the part-of-relation, has often been noted (Hoeppner 198 0; Cruse 1979). But research on part-of-relations has proven that this specific relation actually consists of two variants (part-of and attachment). In addition, inference algo­ rithms have to deal with the decreasing plausibility of the transitivity rule in case of multiple application (see Bever & Rosenbaum 'the bigger the jump, the greater the oddness', cited in Cruse 1979). Those relationships, which consist of one single link in experimental repres­ entations, are stretched to a hierarchy of several steps in dense knowledge bases. So plausibility of the infer­ ence rules is highly dependent on the 'resolving power' of the knowledge base. Moreover, the inferences must be sensitive to a varying degree of detail. 4, REPRESENTATION McKeown (1982) solves the density as well as the partitioning problem by defining three formal degrees of similarity as 'very close', 'class difference' and 'very different'. By use of these inference restrictions

WALTER V. HAHN

574

she avoids a more static concept of traditional partit­ ions or frames while avoiding the difficulties posed by the much more complex construction of dynamic context spaces (or related macro-focus-mechanisms). However, the attempt to apply these degrees of similarity to other, very dense knowledge bases, shows very clearly, that her ONR data base and the corresponding background knowledge is of the 'stadium-type', with very strong terminologically bound words, with unique representation of objects and without cycles. By virtue of the data base the knowledge is homogeneous and there is no difficulty in interpreting the length of the branches as cognitive depth or, in other words, in interpreting branches of the same length as comparably cognitively deep. Thus inferences which work over several steps will deliver results of a similar degree of cognitive plausibility in every part of the knowledge base. I have chosen McKeown's TEXT-system as an example because of its highly elaborated facilities of plausible text generation. However, it seems nearly impossible to apply the methods of TEXT to dense knowledge bases. It can be argued that, for example, the part-of-hierarchy is in most cases only ontologically valid if parts are smaller than the composed concept and if higher part-of links are more general, i.e., high order parts:

sitting-room

bathroom

ceiling

p-o shower But Bever & Rosenbaum (1971) quote the example of

DENSITY IN KNOWLEDGE BASES

575

electrons, which everything consists of:

unlimited inheritance of the high part-of links leads to very inadequate answers, but no powerful algorithms are known for the control of the specific range of inherit­ ance, A last example of those difficulties which arise in dense knowledge bases in spite of well-known and well-tested techniques: case slots. Consider the sentence 'I wrote a letter to my friend1 It's quite clear that 'I' is the AGENT. But is it me as a whole with e.g., my toenails, who writes the letter, or only my head, my hand or my fingers (are the fingers INSTRUMENT, or the pencil?). Even in the best case, where the case frames of the verbs have access to the knowledge base (as in HAM-ANS), the parser must decide, which inferences are allowed to identify the object to fit into the restrictions. The case frame of 'write' in the HAM-ANS- formalism (where case frames are formulated in the semantic languages

576

WALTER V. HAHN

DEEP/SURF which are used throughout the evaluation, see Hoeppner (1982)) has the optional case LOCATIVE: (r1-s: Locative (d-l:rolemarker:F restrictions: (d-o:OR (lambda x1 (af-a:ISA x1 Schreibtisch) )) (lambda xl (af-a:ISA xl Tisch))))) But strictly speaking you write on the table-top or on the writing-pad, which may also occur in actual sentences in which you have to detect that the writing-pad is supported by the table-top, which is part of the desk and that the writing-pad is a filler for LOCATIVE. After hav­ ing built up several very complex inferences in this way, it becomes apparent, that deep-cases have, as a general quality, a very low resolution of reality. They look at the world with half-closed eyes and can only make out the outlines of what is going on. But experts and expert-systems must handle all the tiny and ugly details of a specific task. 5. ONTOLOGY Meanwhile it seems no longer controversial, that the traditional classification approach to the object of a realistic field of application is not sufficient even in thesaurus based work. This is not only true because every thesaurus' aspect is a snapshot in the research history, but it is also a result of the dynamics of multilevel dis­ course about a subject area and divergent individual know­ ledge about a field (see Minski 1981). Thus it is generally conceded that any given object field is not only structured but that it also has several alternative overall structures or systematizations. More­ over, within each systematization there are a number of

DENSITY IN KNOWLEDGE BASES

577

descriptive levels. In some fields the technical reality suggests a well-defined layer-structure, as in computer science: user level program language level operating system level . machine level microprogram level logical level gate level circuit level In other fields (e.g., in the humanities) the structures are somehow orthogonal or only 'somehow...' pragmatics semantics } within language 'as a system' syntax These layer-structures are, however, not the magic remedy for difficulties with details, because the levels cannot be strictly partitioned. As in the paradox of the small number it would be inappropriate in natural discuss­ ions to activate another level of description if only one step has been made into its field. The borderlines of each level of description must be very elastic. The main obstacle, as Hayes (1979) had shown, is the interference or penetration of all natural knowledge (on all levels) with naive systematizations or folk taxonomies (see Lyons 1977: 297). In most cases, these taxonomies handle rather the rough blocks of reality and are some­ times abandoned when confronted with details. Thus, in my 'desk-fragment' we must foresee, e.g., a representat­ ion like

front back in which the two parts can be used independently. Now

578

WALTER V. HAHN

let's go into more detail: On the front of my paper there is a text consisting of sentences, words, morphemes and graphemes, respective­ ly. The graphemes are represented by lines and points. Given an inference module which does not allow the deriv­ ation 'a line is a part of a paper', it is still necess­ ary to specify where the 'information' is located, because the net of actions may need the access to PHYSOBJ's in an inference chain. One certainly will not solve this problem by intro­ ducing a 'meaning'-node into the semantic net. It is not without good reason that generations of semanticists have taken pains to avoid just such a 'solution*. Object­ ions to the effect that this domain of signs and symbols is too complex are not realistic: A desk is a very small domain and any expert system must cope with the non-physical motives in its domain and thus with symbolic interaction. Mechanizing this interaction to MTRANS's and SURPRISES will not do in realistic applications. Under the general heading of homogeneity we must also include the effects of spatial representation, because the conceptual entities described so far are linked to real objects, which often can only be discrimin­ ated unambiguously by spatial attributes. Again an object dependent grid on the desk has to represent the proper place of every small object, as a sort of objective physical order, whereas a 'cognitive geography' (Downs & Stea 1976) of my desk handles only 5 positions from left to right in two rows in front of me. The corresponding difficulties of mapping vague spatial descriptions of natural language onto pixel stuff are a matter for research on image understanding, but natural language is able to zoom up and down using linguistic hedges:

DENSITY IN KNOWLEDGE BASES

579

'Nearly in front of the two X's left of Y', using a general and a specific spatial description. The following fictive data base application of HAM-ANS exhibits spatial and temporal descriptions which must be handled in a manner sensitive to detail (see Wahlster et al. (1983))- When did the Valdivia cruise in the region 530000 - 632000 / 483000 - 300100? - From 10.28.80 to 2.15.81 - How many catches did the Valdivia make on 12.11.80? - Five - when did they take place? - at 8:00, 10:00, 12:00, 18:00, 20:00 - where were they carried out? - at 623012' / 334000 In our small desk-world there exists a meaningful temporal structure as well, which can be utilized to describe details, since the temporal order of receipt of papers can easily be reconstructed by describing the sedimentary structure of a pile. There are corres­ ponding natural techniques for structuring future act­ ions: Very urgent papers are placed very near to the desk chair and are not covered. Incidentally the mass of papers is structured with folders, paper clips, files, colours or by position each of which represents a different systematization, to group objects for a less detailed handling. As the variety of structures and the amount of detail demonstrate, after having collected all the concepts and objects, we must explore the natural techniques for structuring a field of application and try to build up action dependent structures, the basis

580

WALTER V. HAHN

of which will be the subject of the next section. 6, PRAGMATICS One of the most promising ways of eliminating the difficulties with details is an action dependent mask on the knowledge base. This technique reduces the search space drastically to the (conceptual or real) objects around a current 'focus of attention'. Transitions are defined by action trees containing (or generating) the steps involved in any action and their alternatives. Transitions are triggered either by the action schema or word clues. These action-based structurings are in fact a very power­ ful and 'natural' means of restricting inferences and of matching the user's desires or expectations to a particular level of precision or detail. But experience with users shows that (1) at any point users may break out of the action plan and initiate general orientation procedures, e.g., requests for definitions, examples, or metainformation about the organization of the knowledge base. (2) the alternatives are innumerable if real tasks are modelled in great detail corresponding to an extremely dense object world. Thus an action structure split into numerous substeps and sub-substeps including free facilities for POPs to very general levels of orientation is unrealistic and cannot handle alone the vast amount of material which must be present at each step An additional

of

discourse.

fact which can be used is, that in

the flow of a discourse (resp. the performing of an act­ ion) the degree of detail varies in the following way:

DENSITY IN KNOWLEDGE BASES

581

This means, that, in the starting phase, the partners tend to be more general, to switch to the meta-level, to define basic knowledge or to activate each other's memories. In a working phase, the degree of detail is relatively high, whereas in the final steps (esp. in teaching or advisory discours) more general statements are plausible. In the same way, explanations (as microstructures in discourse) start from general knowledge ('genus proximum') and zoom to the point of explanation ('differentia specifica'). These techniques are biased by a partner model. As McKeown (198 3) pointed out, the fact that the status of dialogue partner (novice, casual user or specialist) is known, may be of great importance for the design of answers (except definitions). Moreover, a coreference network like the one contained in the HAM-ANS system) may be used to adjust preciseness and the degree of detail to what is known to the partner. On the other hand, the partner's knowledge, like the knowledge of every natural problem-solver, can be seen as tripartite. He knows, that a specific amount of know­ ledge is always present. For other knowledge elements he knows the internal 'address' of his long term memory, which contains either the item or the derivational algorithm for it. After thinking about it for a time, he will remember this detail. The third part of knowledge is specified only

WALTER V. HAHN

582

by external addresses: here the natural technique is to examine the state of affairs visually or to look in papers, files or books. So not every detail of knowledge has the same state of memory. It seems natural for a system to proceed in a similar way and this points the way for processing in the discourse. 7. SOME HEURISTICS In this section, I will summarize some initial considerations about what seems necessary as a conseq­ uence of dense knowledge base. (1) Every technique to be incorporated in an AI-system must be proved as fully acceptable under high density conditions. This quality must be entered in the list of preconditions for every realistic module, (2) the knowledge base must be partitioned or structured according to - different systematizations of the field, - several levels of description within every systematization, identified by empirical methods or the literature of the area, (3) it is necessary to support the system additionally with dynamic and dialogue dependent memories such as - a partner model containing the partner's status, a coreference file, the partners (users) presuppos­ ed view of the field of application. - a dialogue memory containing information about syntax, semantics and inferences (4) Every more within the systematization and every change in the level of description as well as every bigger jump in a net must be specified by natural language comments, e.g., 'naturally', 'strictly',

DENSITY IN KNOWLEDGE BASES

(5)

583

'in general', 'roughly viewed', 'in a certain way'. The user must be guided very conscientiously to assure him of the cooperativeness of the system and of the proper mutual understanding. The method which is widely used in natural dialogue to safeguard against misunderstanding and commun­ icative inadequacy is the clarification dialogue and the metadialogue. In realistic dialogues in­ volving a mass of details, these natural techniques must be provided for orientation. Therefore the system must be able to verbalize at least the schema and the basic notions of the different systematic views and the 'headlines' of the levels of descript­ ion.

8. ACKNOWLEDGEMENT The author is very grateful to Thomas Christaller for useful comments to an earlier version of the paper.

584

WALTER V. HAHN

REFERENCES Bever, T. G. & Rosenbaum, P. S. 1971. "Some lexical structures and' their empirical validity". Semantics ed. by Steinberg & Jacobovits. Cambridge. Cruse, D. A. 1979. "On the transitivity of the part-whole relation". J. of Linguistics 15. Downs, Roger & Stea, David,eds. 1976 . Image and Environment. Cognit­ ive Mapping and Spatial Behaviour. Chicago. Doyle, Jon. 1980. A Model for Deliberation, Action, and Introspection. REPORT MIT AI-TR-581. Hallig, Rudolf & v. Wartburg, Walter. 1952. Begriffssystem als Grundlage für die Lexikographie. Berlin. Hayes, Patrick. 1979. "The naive physics manifesto". Expert Systems in the Electronic Age ed. by D. Mitchie. Edinburgh. Hoeppner, Wolfgang. 1980. "Reprἄsentationsstrukturen und Inferenzen für zusammengesetzte Objekte". Inferenzen in natürlichsprl. Systemen der KI ed. by Rollinger & Schneider. Berlin. Hoeppner, Wolfgang, 1982. "ATN-Steuerung durch Kasusrahmen". GWAI-82, 6th German Workshop on AI ed. by w. Wahlster. Heidelberg. Hoeppner, Wolfgang & Christaller, Thomas et al. 1983. "Beyond domain-independence: Experience with the development of a German Lan­ guage access system to highly diverse background systems". IJCAI 1983. Karlsruhe. Lyons, John. 1977. Semantics. Cambridge. McKeown, Kathleen. 1982. Generating Natural Language Text in Response to Questions about Database Structure. Philadelphia. Ph.D.Th. Minski, Marvin. 1981. "A framework for representing knowledge". Mind Design ed. by J. Haugeland. Cambridge. Sampson, Geoffrey. 1979. "The indivisibility of words". J. of Ling­ uistics 15. Wahlster, Wolfgang. 1981. Natürlichsprachliche Argumentation in Dialogsystemen. Berlin-Heidelberg. Wahlster, Wolfgang et al. 1983. "Overanswering Yes-No-Questions: Extended Responses in a NL Interface to a Vision System". IJCAI '83. Karlsruhe.

MAGDALENA ZOEPPRITZ Heidelberg

INVESTIGATING HUMAN FACTORS IN NATURAL LANGUAGE DATA BASE

1

QUERY

INTRODUCTION

For many years now, natural language access to a com­ puter has been an area of research. In the last few years, first systems have come on the market that permit natural language interaction with data bases for data query and analysis and it can be assumed that more will come. A common property of existing systems is now, and will remain for a long time to come, that they cannot cope with all of natural language. Compared to human language understanding, the language facilities offered by such systems are very much limited. Nevertheless, they are coming into use and it is time to ask about their properties and the effect of these properties on the people using them, in short, about their human factors. This raises the question of adequate methods for finding the relevant facts. In this paper I shall outline methods that have been proposed for investigating natural language communication with a machine, focussing on database interaction, and discuss some of the problems regarding design of experiments and interpretation of results where natural language is involved. 2

METHODS Before going into the methods themselves, it is necess-

586

MAGDALENA ZOEPPRITZ

ary to briefly describe the two approaches to experimentat­ ion that are to different degrees implied by the methods below. The approaches can be characterized as more laborat­ ory oriented experimentation in controlled environments, and more field oriented studies. Controlled experiments, that are designed so as to test variables in relative isol­ ation and permit statistical analysis of the results, make it possible to test hypotheses about factors of interest and to distinguish between accidental phenomena and pheno­ mena that are systematic, i.e. inherent in the object of the investigation. Field oriented studies, where behavior in natural or natural-like situations is investigated, make it possible for the range of phenomena belonging to a given context to appear and be observed. Such studies can be done either to explore a new area, to get data on what to expect in such context and what factors might be important, or to validate experimental results in real situations. 2.1 SIMULATION One way of investigating natural language communication with a machine is by simulation, i.e. by simulating for the user the situation of communicating with a computer in natural language. The user enters his questions into a system that mediates between the user and a human experiment­ er. The system transmits the queries to the experimenter, who sends back the answers to the user, again via the system. Simulation experiments were done in controlled environ­ ments to test the effects of various constraints (c.f. Chapians, 1973), independently of a specific implementation. Simulation was also used (since Malhotra, 1975) to find out about what a natural language system should be able to handle (sentence types, vocabulary, explanations, etc.) for a given application.

HUMAN FACTORS IN DATA BASE QUERY

587

Where machine responses are simulated, the language behavior of the user is not changed over time by experience with (or fear of) what the system can or cannot handle. But it may be difficult to judge what language facilities implem­ enting the simulated function would entail. Because human beings are normally not able to react as consistently as a machine does, semantic differences can easily be overlooked that may not become apparent until after implementation. Another difficulty is that of estimating the extent to which the sample of language obtained through simulation stands for itself or is representative of a class of utter­ ances that could have occurred just as well in the situat­ ion. If, for instance, only those temporal phrases are implemented that are found in the sample, there will be omissions (cf. Kelley, 1984). If all members of the class of temporals in the language are implemented, many of them will not be used. What is needed for a rounded system is a class of temporals delimited by the situational context, and there is as yet little empirical evidence as to what this entails. An interesting method, combining simulation and test­ ing with users, has been proposed by Kelley (1984). He developed a natural language system for managing an appoint­ ment book (CAL for Calender Access Language). He begins with simulation for obtaining initial language requirements. The resulting first version of the system is then tested and extended with users. In this phase, only missing elements are simulated. As the system develops, the need for simulat­ ion decreases. The resulting version is then validated with additional subjects. This is clearly better than just simul­ ation, but consistency and carry over (also to other applic­ ations) remain a problem.

588

MAGDALENA ZOEPPRITZ

2.2 COMPARISON Experiments comparing one language against another, with paper and pencil (e.g. Shneiderman, 1978) or with systems that present tasks and evaluate the results (e.g. Ogden/Brooks, 1983) in general have as their goal to find out, not whether a language is good or bad in an absolute sense, but whether one language is better or worse than another with respect to e.g. successful queries, number of errors, learning time, retention, etc. Comparative exper­ iments are conducted in controlled environments, to make it possible to filter out personal differences between users, preferences for one or the other language, differences in the applications and tasks, and the effect of learning sequence. It is also possible to present the tasks in such a way that expected difficulties will occur, and are not just observed sporadically. The latter can easily happen where user interest and the properties of the application determine the queries formulated. Controlling for all these factors is needed to make sure that observed differences in the results depend on the languages compared - here natural language versus a formal query language - and on nothing else. A thorough discussion of experiments with formal query languages, their experimental design and the interpretation of their results is found in Reisner (1981). Many of the difficulties pointed out there with respect to studying formal query languages also apply to experiments involving natural language. Designing comparative experiments in controlled environ­ ments involves, among other things, knowing what to control for, choosing tasks for the experiment that are relevant to the problem under study, and - where natural language is involved - presenting the task in such way as to prescribe the contents but not the form of the query.

HUMAN FACTORS IN DATA BASE QUERY

589

Knowing what to control for is a variant of a well-known problem in linguistics: what is 'same' and what is 'different'. Ideally, all conditions in comparative exper­ iments should be the same, except for the things compared. Where languages are involved, distinctions made in linguist­ ics can help to isolate language from other factors and to differentiate between levels within language. The situational contexts of the experiments comparing data base query languages are often designed to approach, but cannot actually produce, a working situation. In the experimental situation, the interest of the subject is given by the proposed task, as opposed to a situation in which users try to solve problems with the aid of their own data. This requires a high degree of sensitivity when proposing tasks (sensible questions and question sequences, that are interesting to the user/subject under otherwise equal condit­ ions for both languages tested). The difference in query interest can affect the subjects' behavior in error situat­ ions (provided subjects are notified of errors). Problem solving strategies, that emerge where users follow their own query interest, must be purposely elicited in an exper­ imental situation. If the comparison is not between two formal languages, the form of presenting tasks becomes a problem. While the content of a query that shall be posed by the subject in a formal language can easily be given in natural language, this is not advisable where natural language is part of the comparison. In the latter case, ways have to be found for eliciting natural language queries in such way as not to prejudice the form they will take. 2.2.1

Language: Ogden's tables

An interesting way to overcome the language problem when eliciting queries has been proposed in Ogden/Brooks

590

MAGDALENA ZOEPPRITZ

(1983, comparing various restrictions on English). Ogden presents the tasks in the form of data base tables. Informat­ ion that will fill the gaps in these tables - indicated by question marks - are to be obtained by a query. Aside from not prejudicing query formulation, this method succeeds in separating knowledge of the language from knowledge of the data base (two variables that are treated as one in Shneiderman 1978): Both the subjects using formal language and the subjects using natural language see the same tables and in this way have the same amount of information about contents and naming conventions in the data base. However, this method does not seem to lend itself easily to presenting other than relatively simple tasks. It is difficult to see as yet how questions involving negation, quantification, or subtotalling can be elicited with tables. Since such questions occur in working situations, and can be avoided, if at all, only with difficulty, it would be desirable to extend the method in this direction. This would make it possible to compare the languages in question at a more realistic level. 2.2.2 Language and task: Chauffeurs A combination of user study (see below) and controlled comparison was attempted at New York University (Turner et al. 1982): Two groups of so-called chauffeurs were introduc­ ed as mediators between users and the system. The users described a problem in natural language. The chauffeurs all worked on the problem by posing the necessary set of quer­ ies to the system, each group using the language assigned to them, either natural language or the formal query lang­ uage. The experiment was done in three phases. Each group was assigned first to one of the languages then to the other, so that both groups had used both languages, but in a differ­ ent sequence. In the third phase, the chauffeurs could

HUMAN FACTORS IN DATA BASE QUERY

591

decide individually on which language to use for a given problem. This method provides that the tasks correspond to realistic query interests, while the formulation of the individual question is not prejudiced for either language. Also the query and error strategies become observable in both languages under comparison. By keeping the overall tasks the same for both languages, comparison is possible, though not at the level of individual questions. The combination of realistic tasks on the one hand and experimental conditions on the other, that is provided by using 'chauffeurs', is an interesting solution because it makes comparison possible without reducing the complexity of the problems to be solved. However, compared to situations of actual use, there are differences in communication and motivation: Chauffeurs are not the owners of the data, the meaning of the data and the purpose of a given task must be explained to them, and the interest in having a specific question answered does not originate with them. All this could make a difference in how they go about their tasks. However, a situation where the questions that are put to a system are formulated by a different person than the one to whom the question occurred, is not really unrealistic. Even with natural language for interaction with the computer, it can be observed that a group may delegate actual question asking to that member of the group who is most ready to use the machine (Kettler at al. 1981). 2.2.3

Paper and pencil studies

If there is no system available, or if one wants to avoid the effects of a specific system, comparative exper­ iments can be done with paper and pencil. Subjects are given a number of tasks, they write out the answers and the results

592

MAGDALENA ZOEPPRITZ

are inspected by the experimenter. Using this method, response time, minor syntactic errors, or typing errors do not detract subjects from the problem in hand. Badly formul­ ated error messages cannot mislead subjects into further error. But since there is no response, learning from error cannot be observed. For realism, they depend on the design­ er's sensitivity in formulating tasks and on the imaginat­ ion of the subjects in carrying them out. An added problem is that of maintaining consistency in grading the results, less severe if more than one person does the grading. Also, where errors are rated as more or less severe, there is the question of what system characteristics are assumed as the basis for such judgments. 2.3 USER STUDIES User studies, or field tests, depend on the availabil­ ity of a natural language system (Damerau 1979 and Krause 1982, the user studies by Tennant 1979 differ from these in that the users did not work on their own data). User studies are done to evaluate how a system meets the requir­ ements of the users and serve to find out more specifically what these requirements are. In these studies, users employ the system for investigations involving analysis of their own data. With their consent, queries, commentaries, etc. are protocolled and analyzed. In such studies users work with their own data on selfimposed tasks. They ask what they want to know and if the system does not respond satisfactorily at the first attempt, pursue the question further by trying different strategies. This makes it possible to observe how users cope with restrictions, what rules they abstract from the system's behavior, and what language and function they expect from a machine.

HUMAN FACTORS IN DATA BASE QUERY

593

An argument against testing with implemented systems, rather than simulating an ideal system, is that the lang­ uage behavior of the users is modified by their experience with the system. In simulation experiments, users can form­ ulate naturally. If they go beyond the limitations of the system, there is no break in communication, because the experimenter will take over. When working with a system, users may learn what the system can or cannot do and re­ strict themselves accordingly. Such an effect can be observ­ ed (see Krause 1982 for detail and qualifications), but the effect is not as bad as it may at first seem, because adaptation to the restrictions of a system will be the realistic situation for a long time to come. So rather than testing for the unrestricted case, one will need data to be able to differentiate between more or less tolerable kinds of restrictions. The advantage of user studies, that type and content of the interaction is structured by the user, and not by experimental design, is at the same time their greatest problem in that there is no control. There is no control for the type of questions asked: One may have to wait a long time before a given event occurs, because user charact­ eristics or the nature of the application do not make it happen. Looking at different users and a variety of applic­ ations can help to counterbalance this effect. Since there is no control over the tasks, the source of difficulties (other than bugs) cannot easily be attributed to a specific factor or group of factors. Since the users work on differ­ ent tasks, problem situations may be similar for several users, but never really the same for all users. There may not even be control of the system: While it is acceptable not to make changes to the system for the duration of an experiment, users who want to get on with their work will

594

MAGDALENA ZOEPPRITZ

find it unfair if gaps and errors in the system that they have discovered (often painfully) are not corrected quick­ ly, where possible. The problems with user studies need to be taken into account when interpreting the results: if certain functions and language facilities are used, this is evidence that they are needed. If problems occur, this is evidence that they exist. However, the reverse need not hold and the importance of a problem can often not be characterized by number of occurrences with a number of different people. User studies permit to observe the language behavior and strategies of users in working situations. Apart from providing data on the merits and defects of the system be­ ing tested, observations from user studies can yield valu­ able hypotheses for testing in controlled environments. 3. QUESTIONS In all experimentation, there is the question of what to look for in the results and what aspects of language to compare. I shall discuss some of these questions against the background of observations from user studies. Results of controlled experiments are generally discus­ sed with respect to the initial hypothesis that motivated the study, its design, and the variables measured. This is often complemented by observations, that may be extraneous to the purpose of the experiment at hand, but may have a bearing on the subject studied and then become a source of hypotheses for further experimentation. The question of what is relevant and what is merely accidental is part of the initial hypothesis, but may be reconsidered in view of the results. In the user studies conducted so far, as generally in exploratory studies, there was no initial hypothesis in the

HUMAN FACTORS IN DATA BASE QUERY

595

strict sense, but rather a set of ideas and assumptions, subject to revision or specification depending on the data obtained. In the studies done with the User Speciality Languages System (Kettler et al. 1981, Krause 1982), there were assumptions about what users would need, (as implement­ ed in the system or planned for implementation), a hope that restricted natural language would be usable at least for some user groups, and the conviction that we did not really know what we would see, other than detect bugs. As the studies continued, and the system developed, language and behavioral aspects emerged, that attracted our attention. Of these I shall discuss language used, behavior in error situations, and the complexity of the questions. Experiments are done on the assumption that the design brings out relevant aspects of the subject studied and that the results are valid for that subject. Findings from study­ ing the language behavior of users at work can shed light on the extent to which design and results of a given exper­ iment can be expected to carry over into working situations (cf. Krause 1982, Zoeppritz 1983, 1984). 3.1

LANGUAGE USE Beyond investigating whether users can work within the

restrictions imposed by the system, and statistics about percentage of successful queries, there is the question of the types of queries that occur and the style that users find appropriate to use with a machine. The former sheds light on needed function, the latter on variants that need to be added (or that could be deleted if necessary). Quest­ ions put in natural language can be expected to reflect more closely what users want to know, and the corresponding function needed, than is the case where facility in using the formal language is an added, possibly constrainiixg,

596

MAGDALENA ZOEPPRITZ

factor. Information on needed function is not only useful for natural language systems, but also for query language design. Looking at language use, together with user comments and frequency or urgency of requests for certain functions, can help to establish priorities, to distinguish between what is needed and expected, what would be nice-to-have, and what could perhaps be left out (language use can also be compared with functional requirements elicited by question­ naire, as in Morik 1983). To mention just one instance: We did expect that our users would need personal pronouns, which we had not implemented. However, much more than the absence of personal pronouns, they minded the absence of decent formatting routines, the availability of which they had expected as a matter of course when interfacing with a computer. 3.2 ERRORS The kinds of errors that occur indicate where there are problems. There are things that must be learned, some with more, some with less difficulty. In a working situation, errors interrupt the user's thinking and detract from the task at hand. When learning a formal language, users expect this. If there are differences between the user's understand­ ing of the language and the implementation in the system, users of a formal language assume responsibility for them. Unless they are under pressure of time, they may not even be aware of the actual distraction so much as of the chall­ enge of finally mastering the system. With natural language, users' attitudes are different. Users know their language better than the system does, so in case of error it is clear that the system, not the user, is at fault. Errors are per­ ceived not as a challenge, but merely as disruptive.

HUMAN FACTORS IN DATA BASE QUERY

597

If breaks in the communication are less acceptable when using natural language, the error rates which one has come to consider as indicating success with formal languages will have to be lowered. Several studies report around 7 0% correct­ ly formulated queries (references in Reisner 1981) , between 80% and 90% correct questions and an average of 1.4 attempts to formulate a question is reported in Ogden/Brooks (1983), to support the usability of the language (or restrictions) studied, while Shneiderman (1981) rates 90% as not yet good enough to be acceptable. However, as Reisner (1981) has pointed out, there is lack of empirical evidence for what error rates are acceptable in working situations. Where errors, or better, breaks in the communication, are reported back to the user, error handling and error circumvention strategies can be observed, in addition to error types. Apart from indicating bugs, gaps, and faulty implementation, error types can point to restrictions that users find difficult to understand or to follow. Error handl­ ing strategies show that users make assumptions about what could have caused the error. The kinds of strategies that can be found say something about the direction of their thought (cf. also Lewis/Mack 19 82). This is important for at least formulating less disruptive and misleading messages, where the corresponding restrictions cannot be removed. In addition, sequences of unsuccessful questions show something about user's expectations and perceptions of the capabilit­ ies of the machine: what properties go together (if it can do more than 500, it should be able to do more than Smith) , and what is easy or difficult for the machine (if it can't do Who is S m i t h , maybe it can do List Smith) . This offers some insight into consistency from the user's point of view.

598

3. 3

MAGDALENA ZOEPPRITZ

COMPLEXITY

From the user studies with the natural language, it appears that users ask relatively complex questions right from the start. This is not surprising if one considers that even rather innocent looking natural language quest­ ions can lead to quite complex operations in terms of a relational data base, and thus to complex expressions in the formal data base query language, involving operators that are considered difficult to handle and are therefore not taught to beginners. So if one assumes that users might accept a system that can handle only simple questions, provided they can get the information they need, it may be misleading to think of simplicity in terms of the data base operations involved. Rather, one would need a measure that reflects user percept­ ion of 'easy' or 'hard' requests. In the absence of such a measure, the (in relational terms) often complex queries put by natural language users seem to indicate that, from the user's point of view, the available database operations have the same status, in as much as they, alone or in combination, are needed to produce the required information. For the purpose of comparing query languages at an interesting level, it is then useful to consider all available function when studying the ease or difficulty of expressing requests for information in the languages under comparison. 4. INTERPRETING THE RESULTS The design of experiments, as well as the interpretat­ ion of the experimental and observational data depend fund­ amentally on the underlying assumptions about the subject

HUMAN FACTORS IN DATA BASE QUERY

599

under study. Where languages are involved, this refers to assumptions about language, what language is and what aspects need to be differentiated. For the majority of experiments I know, such underlying assumptions have not been explicitly stated, notable exception Shneiderman/Mayer (1979, summary in Shneiderman 1981) or justified. Shneiderman/Mayer's model is applied to database query (Shneiderman 1981), recognizes syntax and semantics, whereby the term syntax covers both syntax and semantics of the query language, while semantics covers the applicat­ ion domain as well as structure and content of the database. For programming languages (Shneiderman/Mayer 1979) the model distinguishes between higher and lower level semantic knowledge, a cline on which the following items would appear in descending order: the purpose of a group of state­ ments in terms of the application domain (binary search for student with highest grade), the purpose of such a group independently of the domain (search for the largest value) and what a statement does (adds 1 to value of I ) . Shneiderman/Mayer's model, as well as the model that can be reconstructed from statements about language in the discussion of query language experiments, are too weak to deal with natural language. A more sophisticated model is needed, and can be derived from theoretical linguistics (so from Sgall et al. 1969), that would not only be more adequate for interpreting natural language phenomena, but also serve to understand the behavioral aspects of programm­ ing and query languages (as evidenced by the work of Car­ roll on naming, cf. Carroll 1982 and references there). Two examples shall illustrate the need for more differentiation. 4.1

SEMANTICS VS. CONTENT In a frequently cited paper and pencil experiment

600

MAGDALENA ZOEPPRITZ

comparing natural language and a formal query language (Shneiderman 1978) the subjects were told about a depart­ ment store employee data base and instructed to put quest­ ions to it that would help them decide on the department in which they would want to work. One group were to write their questions in natural language, another in a formal data base query language, then the groups switched languages. For using the formal language, subjects had to know about the structure, content, and naming conventions in the data base they were to query. The subjects using natural lang­ uage first did not have this information, those using natural language second had only what they recalled from the earlier phase. Success was measured in terms of the number of quest­ ions that could have been answered from the database present­ ed. It was found that subjects tended to ask more questions that went beyond the data base (termed 'semantic overshoot' after Codd) when they used natural language than when using the formal language. Most of these questions were asked by the subjects who used natural language first, less by those subjects who used natural language second. Shneiderman attributes this difference to the disciplining effect that learning and using the formal language has had on subjects and uses this to support his doubts as to natural language for communication with the machine. According to Shneider­ man (1981), familiarity with the data base might have had an effect, but not enough to change the results. This view abstracts from the differences between know­ ing the semantics of a language (language meaning in terms of Sgall 1981) , knowing about a subject domain (content in terms of Sgall 19 81), knowing what is represented in the database (subset of content), and knowing how this is represented (database structure: conceptualization of cont-

HUMAN FACTORS IN DATA BASE QUERY

601

ent in terms of the database model). Rowe (1982 on Shneiderman 1981) argues that, if using natural language leads to questions that are not answerable from a database, this does not prove the inadequacy of natural language for data base query, but points out in­ adequacies of today's database systems. This addresses the question of how far and how well concepts with which people are familiar, and which they can express, i.e. that are representable in terms of natural language meaning (here English), are representable in terms of the conceptualizat­ ions underlying database models and expressible using the language meanings of the corresponding query languages. 4.2

SYNTAX VS. SEMANTICS

The focus on difficulties of syntax when discussing the behavioral aspects of query languages can lead to over­ looking the semantic side of language, i.e. language mean­ ing, also in natural language. Thus Ogden/Brooks (1983) looks for syntactic restrict­ ions in his experiments to find a subset of natural lang­ uage that is both implementable and usable. His goal, of finding a set of rules that will make the limitations of the system predictable for the user, is shared, though not necessarily met, by natural language systems in general. However, restrictions in natural language systems are rarely due to structures that cannot be analyzed syntactically. In the majority of cases, things are left out because it is not clear how they can be interpreted in a consistent and general way. Furthermore, according to Krause (19 82), it was easier for users to stay within functional restrict­ ions (reference by personal pronoun is not included) than to follow the purely syntactic restrictions (adverbials only before the adjective but not after).

602

MAGDALENA ZOEPPRITZ

The predictability of a system as to its limitations remains an important goal. But this goal needs to be address­ ed in terms of rounding out a system in accordance with user expectations as to what capabilities follow from one another (which may involve dropping structures that can be interpreted only in special cases), rather than in terms of syntactic constraints on evoking available function. 5. SUMMARY Natural language systems must be built to fit the requirements of their users in working situations, and this raises the question of how to elicit these requirements. Natural language systems must be compared against formal query languages: Since no system can do all of natur­ al language, how does restricted natural language compare with formal query languages as regards function, expressive­ ness, ease of learning and recall, precision, conciseness, etc. Lastly, system must be tested in actual applications, to see whether predicted requirements are met, what addit­ ional requirements there are, and to what extent the find­ ings from experimental or study environments carry over into a working situation. There is no ideal method that will serve all purposes. While user studies certainly provide the most realistic data, they are the most difficult to interpret. Simulation has yielded interesting results in comparative experiments (for a brief overview, see Zoltan et al. 1982), and is certainly useful in an iterative development process (Kelley 1984), where unconscious inconsistencies in the reactions of the person simulating the computer can be detected after implementation of the corresponding function. Controlled experimentation is needed to bring out differences and

HUMAN FACTORS IN DATA BASE QUERY

603

separate relevant parameters, but there are external constr­ aints to such experiments - subjects need to be taught the languages in question, learn about the database, and per­ form the assigned tasks, all at reasonable cost in time and money. These constraints have often led to reducing the complexity of language, database structure, and tasks for the experiment. If complexity is a relevant factor, as evidence from user studies seems to indicate, then care must be taken not to make the tasks in controlled experim­ ents too simple. Much will depend on an adequate model of language in communication, derivable from existing linguistic theories, that includes programming and query languages and that allows for sufficient differentiation when designing exper­ iments or interpreting experimental and observational data. The number of problems shows that much remains to be done before it will be possible to separate and assess adequately the factors that play a role in the human factors of natural language as an interaction language with comput­ ers. There do not seem to be easy answers or quick roads to them. Despite its history, research in this area is really just beginning to take shape. Acknowledgement. I wish to thank Michael Holz for his interesting and thoughful comments on this paper.

604

MAGDALENA ZOEPPRITZ

REFERENCES Carroll, J. M. 1982. "Creative Names for Personal Files in an Interact­ ive Computing Environment". International Journal of Man-Machine Studies 16. 405-438. Chapanis, A. 1973. "The Communication of Factual Information Through Various Channels". Information Storage and Retrieval 9. 215-231. Damerau, F. J. 1979. "The Transformational Query Answering System (TQA) Operational Statistics", IBM Research Report RC 7739. New York: Yorktown Heights. Kelley, J. F. 1984. "An Iterative Design Methodology for User-Friendly Natural Language Office Information Applications". ACM Transactions on Office Information Systems 2,1 . 26-41. Kettler, W., Schmidt, A. & M. Zoeppritz. 1981. Erfahrungen mit zwei natiirlich-sprachlichen Abfragesystemen. Heidelberg Scientific Center TR 81.01.001. Krause, J. 1982. Mensch-Maschine-Interaktion in natiirlicher Sprache. Tübingen: Niemeyer. Lewis, Clayton & Robert Mack. 1982. "The Role of Abduction in Learning to Use a Computer System". IBM Research Report RC 9433. NY: Yorktown Heights. IBM. Malhotra, Ashok. 1975. "Design Criteria for a Knowledge-Based English Language System for Management: An Experimental Analysis". Report MAC TR-146, based on the author's PhD thesis. Cambridge, Mass.: MIT. Morik, K. 1983. "Marktstudie zu natürlichsprachlichen Zugangssysteme. Forschungsstelle für Informationswissenschaft und Künstliche Intelligenz. Report ANS-14. Hamburg University. Ogden, William C. & Susan R. Brooks. 1983. "Query Languages for the Casual User: Exploring the Middle Ground between Formal and Natur­ al Languages". Human Factors in Computing Systems, Proceedings CHI'83 ed. by Ann Janda. SIGCHI Bulletin special issue. 161-165. New York: ACM. Reisner, P. 1981. "Human Factors Studies of Database Query Languages: a Survey and Assessment". ACM Computing Surveys 13. 13-31. Rowe, N. C. 1982. "On Some Arguable Claims in B. Shneiderman's Evaluat­ ion of Natural Language Interaction with Data Base Systems". ACM SIGMOD RECORDS 13.1. 92-97. Followed by a response by Shneiderman. Sgall, P., Nebesky, L., Goralčikovā, A. & E. Hajicova. 1969. A

Function­

al Approach to Syntax in Generative Description of Language. New York: Elsevier. Sgall, P. 1981. "The Level of Linguistic Meaning". The Prague Bulletin of Mathematical Linguistics 35. 3-40. Shneiderman, B. 1978. "Improving the Human Factors Aspect of Data Base Interaction". ACM Transactions on Database Systems 3. 417-439. Shneiderman, B. & R. Mayer. 1979. "Syntactic/Semantic Interactions in Programmer Behavior: A Model and Experimental Results. Internation­ al Journal of Computers and Information Science 7. 219-239.

HUMAN FACTORS IN DATA BASE QUERY

605

Shneiderman, B. 1981."A Note on Human Factors Issues of Natural Language Interaction with Database Systems". Information Systems 6. 126-129. Tennant, H. 1979. Experience with the Evaluation of Natural Language Question Answerers. Working Paper 18. Advanced Automation Group, Coordinated Science Laboratory, University of Illinois at UrbanaChampaign, Urbana, 111. Turner, J., Jarke, M., Stohr, E., Vassiliou, Y. & N. White. 1982. "Using Restricted Natural Language for Data Retrieval - a Field Evaluat­ ion". Proceedings NYU Symposium on User Interfaces, May 26-28, 1982. Zoeppritz, M. 1983. "Human Factors of a 'Natural Language' Enduser Syst­ em". Enduser Systems and Their Human Factors. Eds. by A. Blaser & M. Zoeppritz. 62-93. Heidelberg: Springer, Lecture Notes in Computer Science 150. Zoeppritz, M. 1984. "Datenbankabfrage in natürlicher Sprache: Diskussion von Eigenschaften der natürlichen Sprache aufgrund von Beobachtungsdaten". ONLINE 84 Proceddings. Ed. by Ockl. 3:3M-1 - 3M-16. Velbert: ONLINE84. Zoltan, E., Weeks, G. D. & W. R. Ford. 1982. "Natural Language Communic­ ation with Computers: A Comparison of Voice and Keyboard Input". Analysis, Design, an Evaluation of Man-Machine Systems, IFAC/IFIP/ IFORS/IEA Conference. Ed. by G. Johannsen, J. E. Rijsdorp. BadenBaden, FRG. Sept. 27-28, 1982.

POSTFACE

PATRICE POGNAN Paris

15 ANNÉES DE COOPÉRATION AVEC LE GROUPE DE LINGUISTIQUE ALGÉBRIQUE

C"est sans aucun doute la lecture de 1 ouvrage "Generativní popis českého jazyka a ceska deklinace" de Petr Sgall qui m á porte à demander une bourse d "étude pour la Tchecoslovaquie de décembre 1970 à fevrier 1973. Ma première rencontre avec Petr Sgall, dans le cadre grandiose de Prague, fut chaleureuse et, en aucun cas, depaysante: à cette epoque, Petr avait adopte... le beret basque! Cette première période de contact avec le Groupe de linguistique algebrique a ete très riche: la seule decouverte de la bibliothèque du groupe a dure plusieurs mois. En effet, peu de bibliothèques de laboratoire presentent une telle exhaustivité, tant en ce qui concerne les ouvrages d "Europe centrale et orientale que les travaux d "autres pays. Grace aux conseils prodigués, au soutien moral et financier de l"equipe, ce séjour a débouché sur la réali­ sation d'une thèse de è me cycle sur 1 "extraction automatique du verbe dans des textes medicaux tchéques. (Les calculs ont ete executes au Laboratoř počítacích strojů du Vysoke Uceni Technicke de Brno). Une seconde periode de cooperation de 1973 à 1981 a vu, de notre côté, au sein du Centre de recherche Jean Fa-

610

PATRICE POGNAN

vard, la realisation d'une analyse "processuelle" du tchèque. Pour le Groupe de linguistique algebrique de l'Universite de Prague, c "est sur la base d'etudes theoriques très remarquées, le début du développement rationnel d "un bon nombre d "applications dans de domaines divers: - traduction automatique en particulier d'anglais vers le tchèque (dont la tradition remonte déjà à 1957) Signalons à ce propos les études réalisees à l'aide des langages Q par Kirschner et Vrbova - indexation automatique . par la methode MOZAIKA d'analyse des textes d "electrotechnique (à laquelle nous avons eu la joie de collaborer en particulier par 1 "intermédiaire du protocole CNRS - Académie des Sciences Tchecoslovaque). . fabrication automatique d'index, qui presuppo­ se une lemmatisation generale du tchèque et oriente déjà vers la comprehension automati­ que des textes. - dialogues homme - machine . accès en langage naturel à des bases de donnees . système TIBAQ (Text-and-Inference Based An­ swering of Questions). sans oublier 1 "analyse et la synthèse automatiques du tch­­ que, specialite de l'équipe, qui sont sous-jacentes dans bien des systèmes cités plus haut. Depuis 1982, date de la creation de notre nouvelle equipe, le CERTAL (Centre d'Etudes et de Recherches en Traitement Automatique des Langues) , c "est déjà une nou-

15 ANNÉES DE COOPÉRATION

611

velle generation de bohémisants francais qui utilise la cooperation avec le Groupe de linguistique algebrique par l'intermédiaire du ministère des relations exterieures fran­ cais et le ministère de l'éducation tchèque. C"est ainsi que Jean-Francois Blanc qui s"interesse aux problèmes d'analyse automatique du francais et du tchèque, a debute des etudes sur la lemmatisation gènérale du tchèque où les nombreux travaux du Groupe de linguistique algebrique servent de reference. Ainsi, depuis près de 15 ans, par le jeu des protocoles d'accords culturels entre la France et la Tchecoslovaquie, une cooperation très efficace nous a permis d "assurer nos travaux sur 1 "analyse automatique du tchèque, de prevoir 1 "e"change de textes d "essai par 1 "adoption du meme code de transcription pour le tchèque et le russe, de nous affermir, au vu de la majorité des travaux du Groupe de linguistique algebrique, dans l'opinion qu 'il est possible d "analyser un texte sans typologie prealable. Par l'intermediaire de cette équipe de renom, universellement reconnue et qui avait magistralement organisé le COLING de 1982, nous avons nous-memes élargi notre champ de relations à d'autres equipes europeennes. Ce n"est certes pas un mince mérite de Petr Sgall et de ses proches collaborateurs que d'avoir pu maintenir 1'efficacité d "un très petite equipe malgré les inevitab­ les fluctuations de personnel. Particulièrement remarquable est 1 "attention portee aux jeunes par des séminaires à la faculté de mathématique et physique et par des stages dans le laboratoire. Esperons qu'ainsi sera assuree la perennite de cette equipe parti­ culièrement appreciee par ses pairs ... souvent très desireux de maintenir un contact permanent avec elle.