Cooperating with Written Texts: The Pragmatics and Comprehension of Written Texts [Reprint 2011 ed.] 9783110881196, 9783110127232

Citation preview

Cooperating with Written Texts

Studies in Anthropological Linguistics 5

Editors

Florian Coulmas Jacob L. Mey

Mouton de Gruyter Berlin · New York

Cooperating with Written Texts The Pragmatics and Comprehension of Written Texts

edited by

Dieter Stein

Mouton de Gruyter Berlin · New York

1992

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

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

Library of Congress Cataloging in Publication Data Cooperating with written texts : the pragmatics and comprehension of written texts / edited by Dieter Stein. p. cm. — (Studies in anthropological linguistics : 5) Includes bibliographical references and index. ISBN 3-11-012723-7 1. Written communication. 2. Discourse analysis — Psychological aspects. 3. Comprehension. 4. Pragmatics. I. Stein, Dieter, 1946. II. Series. P211.C62 1992 302.2'244—dc20 91-38037 CIP

Die Deutsche Bibliothek — Cataloging in Publication Data Cooperating with written texts : the pragmatics and comprehension of written texts / ed. by Dieter Stein. — Berlin : Mouton de Gruyter, 1992 (Studies in anthropological linguistics ; 5) ISBN 3-11-012723-7 NE: Stein, Dieter [Hrsg.]; GT

© Copyright 1992 by Walter de Gruyter & Co., D-1000 Berlin 30 All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Typesetting and Printing: Arthur Collignon GmbH, Berlin. Binding: Lüderitz & Bauer, Berlin. Printed in Germany.

Table of contents

Cooperating with written texts: Towards a science of the text. Issues and overview Dieter Stein

1

Part I Writing, literacy, and comprehension: psycholinguistic aspects

13

On the relationship between writing system, written language, and text processing Florian Coulmas

15

Pause and intonation contours in written and oral discourse Gisbert Keseling

31

Aspects of writing development in argumentative texts Gerhard Augst

67

Psycholinguistic processes in the comprehension of written texts Simon Garrod

83

Coherence and coordination in written text: Reading time studies Gert Rickheit, Udo Günther, Lorenz Sichelschmidt

103

Levels of language comprehension and systems of informationprocessing Johannes Engelkamp 129 Part II Developmental aspects: the evolution of written texts

161

The pragmatics of medieval texts Heinz Bergner

163

The impact of sudden literacy on text comprehensibility: Mohawk Marianne Mithun

179

Scientific texts and deictic structures Konrad Ehlich

201

vi

Table of contents

Immediacy and displacement in consciousness and language Wallace Chafe

231

Part III Literary texts: inferring meaning(s)

257

Fictional conversation and its pragmatic status LuMing R. Mao

259

Presupposition and pragmatic inference in a literary text Helen Aristar-Dry

277

How to cope with dramatic texts including avant-garde playscripts Herbert Grabes 295 The pragmatics of literary texts Lothar Bredeila

313

"Deep structure signals" in fiction Volker Schulz

335

The pragmatics of poetic discourse Janet Byron Anderson

361

A pragmatic role for inserted clauses in literary texts Danielle Forget

373

Part IV Pragmatics and comprehension of individual text types

395

Teaching conscientious resistance to cooperation with text: The role of pragmatics in critical thinking Susan Meredith Burt 397 From private writing to public oration: The case of Puritan wills. Cognitive discourse analysis applied to the study of genre change Ulrich Bach 417 Parallelism in advertising copy Wolfram Bublitz

437

Notice is hereby given to hide ulterior motives Leo Hickey

459

Table of contents

vii

Renarration: Oral LI and L2 narrations as bases of written narratives Rüdiger Zimmermann

467

Strategies in text production and text comprehension: A new perspective Ruth Wodak

493

The axiological structure of discourse Tomasz P. Krzeszowski

529

PartV At the interface of linguistics and pragmatics: individual linguistic structures

547

On the (In)dependence of syntax and pragmatics: Evidence from the postposition -rä in Persian Mohammad Dabir-Moghaddam

549

VP inversion and aspect in written texts Gregory L. Ward, Betty J. Birner

575

Scope in discourse: Pragmatics or syntax? Richard A. Rhodes

589

Part VI Computational modeling of text comprehension

607

'Parsing' procedures for stories Yeshayahu Shen

609

Convergent evidence for a set of coherence relations Kathleen Dahlgren

631

Text pragmatics and computational modeling Annely Rothkegel

665

Integrating knowledge sources for the generation of referential expressions Russell Block

685

Subject index

699

Cooperating with written texts: Towards a science of the text. Issues and overview Dieter Stein

This volume contains the body, and a few more, of the papers from the conference on "Cooperating with written texts: the pragmatics and comprehension of written texts", held at Rauischholzhausen castle, the conference site of Justus-Liebig-University Giessen, West Germany, September 5 - 9 , 1989. There are two connected motivations for having a conference and a volume of papers with such a topic. The first is a unanimous tenor in the current research on written language that there is a fundamental lack of basic research in this field. The reason for this is that the extension from morpheme and sentence into discourse meant primarily spoken natural discourse. This is where the bulk of discourse analysis was concentrated as a reaction to both philology based on written texts and to language theories often drawing on "unnatural", made-up examples. Primarily two factors have conspired to make the lack of research in written discourse keenly felt. The first was brought up in discussions of problems of language standardization. What emerged was a close connection between standardization and written language. This factor and others, such as the discussion centred on the question of a cognitive divide, and educational problems associated with the acquisition of literacy in Western countries under the impact of other media, moved written language into focus. This lack of research meant that urgent problems on the applied side (language planning, educational and psychological) could not be tackled in a promising way. The second motivation is connected to both the theoretical and the applicational deficit. One of the courses offered at the university of Giessen, where the conference was held, focuses on the study of modern written language, since students taking that course are taking jobs in business and economy. The study of the communicational mechanisms associated with these texts is therefore a central component of this course. It was in the context of developing course contents and carrying out related research that the deficit in research in this field was keenly felt. This is the more surprising since the Western industrialized societies depend vitally for their func-

2

Dieter

Stein

tioning on this type of communication, from the exigencies of functionally differentiated standard languages to needs and constraints imposed by modern media and the computer. It is a fact that developments in, e. g., artificial intelligence and text representation by the computer are being severely hampered by deficits in knowledge about the structure and communicational mechanisms involved in texts ( = written discourse). It is clear that this type of research can only be of an interdisciplinary nature. Correspondingly, the papers in this volume represent very different foci on the subject, ranging from problems in the analysis of texts by the computer via the analysis of individual grammatical structures in written texts, psycholinguistic and developmental aspects to the analysis of literary texts. In a culture where literature implies writtenness, the analysis of literary texts is actually the discipline with the longest history of dealing with written texts. Any attempt to establish a text science on a broad basis is ill-advised to ignore the store of knowledge accumulated in that tradition of dealing with texts. It is the characteristic of this volume that it does not focus on any particular subject worthy of a study in its own right, such as educational issues connected to the acquisition of written language (Whiteman, 1981). The scope and range of content and contributions is rather determined by what could be considered the prototypical course content of a university course in text science, such as developed in Glessen. The important element of the scientific underpinning of such a science is the bringing (back) together what has long drifted apart, linguistics and literature. Under this general perspective of a text science, the literary text — provocative as this may sound to more conservative ears — is simply a special case of a text type under the more general aegis of a matrix science of (discourse) linguistics. Because of its pragmatically deviant nature — in many ways highlighted in the respective contributions in the present volume — it catalyzes insight into the workings of a "normal" text, the same type of methodological contribution that the study of other pragmatically deviant texts such as pathological texts or advertising texts with their intricate breaking of speech act rules have. But it remains to be seen what "normal" texts are. The more text types are closely scrutinized as to their constituent pragmatic rules of cooperation, the more it may well become questionable where there are such pragmatically "unmarked" text types, a point driven home by Hickey's paper. The call for an interdisciplinary approach to a text science is not new (Dressier— Beaugrande, 1981; van Dijk-Kintsch, 1983; van Dijk, 1985). Where the present volume differs from this work is that it is not a survey of the

issues

3

field, but exclusively contains concrete empirical case studies. It is also broadly focused on written language, rather than on both spoken and written language. Finally, as indicated, it tries to represent what could be the paradigmatic content of an academic course with the aim described above. If the volume contains a number of contributions dealing with literary texts, it is clear that the focus on literary — and other — texts must include the notion of "co-operating with written texts". Traditional analyses of texts, particulary literary texts, have focused on the linguistic sign itself. The tendency has been one of omnia in verbo, whereas an analysis of literary and other texts must relate the information contained in linguistic signs to their interaction with elements of a structured notion of context. Important elements of such an approach are the notion of the reader constructing the meaning of the text on the basis of instructions contained in the sign materiality of the text, or the notion of global text schema or text type as processing patterns and strategies, such as must be postulated as triggered by pre-signals and as a type of context built into and determining the structure of the linguistic signs in the text. A number of contributions addresses one or other aspect of this relationship between the built-in assumptions of reader co-operation implicitly or explicitly. This reader activity takes place on various levels. Such a hierarchy of reader activity takes place at the very elementary and atomistic level of recognition of the signs (letters, formal aspects of text arrangement like paragraph divisions and arrangement in acts, scenes), via the extraction of propositional meaning, if any, to highest-level instructions for perlocutionary effects or "messages" in literary analyses. It seems to be the case that there has traditionally been a neat division of labor between the job assigned to low levels — linguistics to the extent it has dealt with meaning at all — and very high levels, literature and conversational analysis, with discourse analysis rather more in between. What has become clear over recent years, however, is that even at the very lowest level, pattern recognition of letters and morphemes, there are top-down processes at work which are ultimately the result of a "set" — in the psychological sense — created or "switched on" by some pre-signal. What is equally clear, however, is that these level interactions can only be captured by an integrative text science, such as has been variously postulated, at the methodologically heavy price of relative abstractness of the concepts used. The opposite is an incommunicability of approaches. It is appropriate that a collection of articles with this aim should first deal with "external", medium-related constraints on written language.

4

Dieter

Stein

For a collection of papers which focus on language use in written text, it is fit to start with a paper by Coulmas ("On the relationship between writing system, written language, and text processing") that looks at the very lowest level of reader activity by dealing with what constraints on processing are imposed by the very nature of the script system used. Coulmas reminds us that very notion of writing requires a lot of clarification and differentiation, in that processing by human (and machine) cognition is very relative to the type of writing system. Based on experimental studies Augst looks at the development of written discourse production skills and identifies significant patternings in terms of the classical differences between spoken and written language (such as desubjectivization or decontextualization). One of the points emphasized is that the notion of an ontogenetic language development and ideas about when it ends depends, not only on the underlying theoretical framework, but on the medium studied. A lot seems to be happening between the ages 13 and 23. Keseling ("Pause and intonation contours in written and oral discourse") takes up a question that is central to the psycholinguistic representation of written discourse, and the relationship to spoken discourse. Are the modalities dependent on or independent of each other? Keseling uses evidence from an empirical study of prosodic features in different types of written and spoken discourse (mainly oral reproduction of written texts) to draw conclusions as to whether or not patterning features in text types remain stable across different medial representations and productions and, more generally, about the psycholinguistic relationship between spoken and written discourse. The next group of papers goes even deeper into the psycholinguistics of text comprehension. Garrod ("Psycholinguistic processes in the comprehension of written texts") deals with another set of general constraints on reading comprehension. The input for comprehension at any given moment in reading is taken in during eye fixations over a small "window" of 20 to 30 character spaces. Garrod shows that this communicative pragmatic processing limitation in the medium coincides with a restriction of cognitively present knowledge at the interpretation of referential and role expressions to enable efficient processing in comprehension. The notion of a "focus" as a highly restricted set of referents or topic entities at any given point in a text makes the identification of a referent (e. g., in cases of lexically empty pronouns like they with a multitude of potential candidates or roles) possible or at least facilitates it. This is a central problem in both human and machine processing and will turn up again

Issues

5

in the last papers of this collection. The more or less skilled handling of the linguistic means of the steering of this focusing is an important criterion in defining a psycholinguistically based notion of "well writtenness". Rickheit — Günther — Sichelschmidt ("Coherence and coordination in written text: Reading time studies") take a slightly different tack in studying the process of reading as a cognitive activity. They deliberately concentrate their perspective on the comparatively low-level process of how readers process input taken in during eye fixations to construct mental models. They deal with which contributions different types of expressions make towards the construction and modification of a coherent model as it is built up in word-incremental reading. They find, for instance, that in German an early position of the verb makes for easier processing than late processing. The authors examine a number of variant structures as to their facilitating or inhibiting effect in the task of constructing a mental model. By now it should have become clear that text comprehension is a process that takes place simultaneously and on-line on several levels. Engelkamp ("Levels of discourse processing") looks at the intricate interaction of these levels of processing and the relationship of information flows in top-down and bottom-up processing. The next four papers represent "developmental" aspects. In a case study, Mithun ("The impact of literacy on text comprehensibility: Mohawk") discusses the impact on a previously unwritten language of contact with a modern language in which culturally important functions are performed in writing. Mithun follows the evolution of written Mohawk under the influence of English. She notes several suggestive linguistic changes in the emerging written medium under the constraints of Mohawk, English, and the new medium, among them changes in word order, in particle use, in nominals and incorporation. Bergner's paper ("The pragmatics of medieval texts") serves an important function in the volume in that it provides a relativizing perspective especially for the papers dealing with modern texts. The paper shows how very radically different medieval texts were in their built-in expectations as to reader co-operation. Fundamental differences arose from the different intellectual and religious background in self-experience and in experiencing the world. Notions like "openness" of the text, "author", and "text structure" had fundamentally different interpretations, other concepts like "text type", and "stability of the text" simply do not seem applicable to the medieval text.

6

Dieter Stein

The last two papers in this group are developmental in a very basic sense. Both deal with the definitional characteristic of the written medium of representing experience or knowledge, and handling such knowledge, not simultaneously present at the time of speaking, i. e., with artificial deixis divorced from physically present and immediately perceivable experience. In a surprisingly parallel way both papers (Ehlich, "Deictic structures in scientific texts"; and Chafe, "Immediacy and displacement in consciousness and language") discuss culturally vital text types, literature and scientific texts. Both papers, amongst other things, make the point that it is possible to distinguish several stages in a developmental succession of "displacedness" of knowledge. While Ehlich's point is taken up at several points in the volume, Chafe's paper forms a convenient transition point to papers dealing with pragmatic analyses of literary texts. Most of the following papers make use of analytical tools supplied by pragmatics, such as speech act theory. Their common focus is on an inferencing process by the reader, which is necessary to arrive at, or is even itself part of, the meaning or message of the text. This process is "pragmatic" in its nature under several points of view. Firstly, there is a fundamental difference to meaning construction in non-literary texts. Generally, the result of meaning construction is "safer" the more the text offers surface-linguistic instructions to guide the processing of the signs the text contains. To a certain extent, the processing of any discourse, written or spoken, requires inferencing. This is a central issue for machine processing of non-literary texts. The type of inferencing at issue here is on a higher level, often concerned with what linguists called "implicatures". It is in the nature of those implicatures that they are calculable, defeasible, and not very watertight. More generally, the authors of the literary papers postulate a range of possibilities from correct "end-results" of various degrees, via general indeterminacy of literary texts to an invitation to compare the fictional world to the respective own world of any particular reader. For the latter view there can be no "correct" interpretation since the extremely context-guided nature of these inferences and the resulting meanings depend on the personal context supplied by each individual reader. In whichever version, the assumption of this strongly context-guided inferencing seems to make for a view, although not always explicitly so stated, that a defining property of the literary text is this specific type of inferencing, the direction of which is only weakly guided by explicit verbal signals, and which is very close to the concept of conversational implicature.

Issues

7

The second pragmatic aspect of this group of papers concerns not so much which or wheter such a "correct" end-result is postulated, but how these inferences are triggered. Mao ("Fictional conversation and its pragmatic status") uses the Gricean maxims, predominantly the maxim of quality, or rather its violation, to lacate the trigger of inferencing. AristarDry ("Presupposition and pragmatic inference in a literary text") sees the source of the inferencing process in a misfit between where new content is normally or canonically introduced in focused syntactic positions and where it is introduced in Henry James's The Turn of the Screw, namely, tucked away in the presuppositions of particular syntactic structures. While in these two cases the search for meaning is triggered by pragmatically deviating ways of presenting content, the case discussed by Grabes ("How to cope with dramatic texts, including avant-garde playscripts") involves a deviation on a higher level, namely violating the rules of genre. Here it is the clash between the expected constituent formal aspects of a text type and the disintegration of that form which triggers inferencing on the side of the reader. Grabes correctly points out that developing coping strategies for this type of texts involves different reading and processing habits different from both those felicitously applied to earlier literary, and to non-literary texts. This point is elaborated by Bredella ("The pragmatics of literary texts"), who — in a wide-scope paper — discusses several modern speech-act-theory based approaches in identifying the specifics of reading a literary text: his main point is that constitutive for a definition of processing literary text is not the presence of a moral perspective as a message, but the inferencing caused during the reading process itself, by a clash between the fictional world and the normality expectation brought to bear in the interpretation of the state of the world reported. This view could be taken even further to state that the inferencing process is a general tendency of human minds not to stop attempting a resolution of the clash until a synthesis is found with knowledge and presuppositions that will resolve the clash and is able to integrate the fictional facts of the world, i. e., until "sense constancy" in Hörmann's (1976) sense is reached. All these analyses provide a novel perspective on classic problems of interpretation. But the central — and third pragmatic aspect — of these analyses lies in the fact that readers must be willing to accept infringements of textpragmatic normalities on whatever level and to engage in inferencing at all. Some of the reactions by readers (and critics) quoted in the contributions clearly show that this willingness to let oneself in with the text in the way described cannot be taken for granted. This

8

Dieter Stein

implies a notion of "adequate" reading, an essential component of which is the readiness to abandon "normal" reading habits with no pragmatic deviations and no hard, painful, inferencing work to be done. This, then, would amount to a pragmatic definition of the literary text, a definition based on a willingness of the reader to cooperate with the text by taking pragmatic infringements, not as a sign of artistic deficiency, but as an invitation for inferencing. There is, then, a dimension of adequacy of response from the side of an adequate reader — an aspect which has been emphasized in recent argumentation against what has been characterized as the autonomous, de-contextualized nature of written language. Nystrand (1986) argues that written texts have their context built into them. They are therefore, as a rule, stronger conventionalized than spoken discourse. Part of that built-in context of written discourse are certainly genre-inherent (in this case literature-inherent) legitimacy expectations about what communicative, cooperational conventions to waive with what text type if a felicitous cooperation with that text is to come about. A more deterministic view is taken by Schulz ("Deep structure signals in fiction"), who metaphorizes more linguistic terminology in a "stratificational-functional" approach to the literary text in assuming a multilayered hierarchical underlying structure of a text, much the same vein as in generative or process-oriented grammars — and which reflect the logical if not real-time direction of information flow in production and comprehension. The central aspect of Schulz' contribution is the analysis of the signals contained on lower levels as text-given instruction for the construction by the reader of deepest level, the "thematic idea", i. e., the message. In text production, the information flow is, correspondingly, the other (i. e., vertical) way round. Schulz' idea of the logical layering of the literary text seems a slightly more deterministic one in that he assumes an ultimate message of the literary text more than the other contributors in the literary field in the volume would be willing to subscribe to. The existence of such signals as described by him is, however, good evidence for his case, and a counterpoint to the contributions emphasizing more the interpretive openness of literary texts. A last group of papers still deals with literary texts, Anderson ("The pragmatics of poetic discourse"); and Forget ("A pragmatic role for inserted clauses in literary texts"), but focuses not on indeterminacy but on how specific structures within typical pragmatic function are made the vehicle of expression. Anderson looks at the role of the presupposition

Issues

9

and deictic movement in the expression of poetic meaning. Forget analyzes the metadiscoursive role of parenthetical clauses in narrative texts. Another group of papers topicalizes text-types as genres, particularly the relationship between the form of the text, the dialectic between form, inherent presuppositions, and the sociocultural context. Co-operating means switching on expectations and interpretive contexts. Burt ("Teaching conscientious resistance to cooperation with text: The role of pragmatics in critical thinking") describes in terms of the Gricean maxims how an application to the teaching of composition can be used to make fallacies explicit. She shows that fallacies can be reduced to versions of maxim violations. The conclusion of her argument is aptly highlighted by the title of her last section (§ 5); "Grice for Undergraduates." Bach ("From private writing to public oration: The case of Puritan wills. Cognitive discourse analysis applied to the study of genre change") supplies a case study in a very special type of written texts: wills. He shows changes in the form of wills to be a consequence of a change in processing strategies concomitant with a change from a text read silently to one read out aloud publicly. Like Bach, Bublitz ("Parallelism in advertising copy") relates the linguistic form of a text to specific processing conditions as a function of the purpose of the text type. The specific linguistic structure Bublitz looks at, parallelisms, are not normally thought of as typical of written language. However, in the type of text Bublitz looks at, advertising copy, they do have a function related to the specifics of that text type. Still within the compass of legal texts Hickey ("Notice is hereby given to hide ulterior motives") represents a field that is underrepresented in this volume in view of its importance to everyday communication. He analyzes the status, in terms of speech act theory, of legal notices and finds that they represent a case for which no provision has been made in that theory, especially with reference to the schizophrenic perlocutionary effect in a case of a solicitor including on his invoice "a printed note informing his client that they may challenge his fee by writing to the Law Society." Surely the solicitor does not anticipate anybody performing the action indicated in the proposition. Zimmermann ("Renarration: Oral LI and L2 narrations as bases of written narratives") exploits the very specific pragmatic-communicative conditions of an experimental set-up to get at factors determining text production and comprehension. He varies the sequencing of oral and written renderings, LI (German) and L2 (English), of the Pear Stories film (e. g., oral LI followed by written L2, written L2 by oral L2) to

10

Dieter Stein

obtain a number of experimental groups of different text production and processing constellation. As an interim report of an ongoing research project, Zimmermann turns up a number of interesting results concerning such questions as language-specific de-oralization, the relative strength of different factors, the relative strength of individual strategies vis-a-vis general factors. In a theoretically wide-ranging paper, Wodak ("Strategies in text production and text comprehension: a new perspective") goes beyond the level of frames and schemata as cognitive determinants of text comprehension (and production) and identifies deepest social-cognitive and emotional layers of knowledge that determine differential comprehension strategies and how the comprehender makes the text — a theme very reminiscent of approaches in the literature section. Like Wodak, Krzeszowski ("The axiological structure of discourse") addresses deep and abstract layers of knowledge underlying text production and comprehension. He looks at texts as metaphorical, preconceptual images schema in terms of which all surface units can be interpreted. The movement of the text is then understood in his "axiological" approach as movement, journey, clash and balance between axiological poles. This provides a matrix for the well-formedness of text. At another remove from literary texts, but still within the analysis of the function of structures in written texts, is another group of papers. Unsurprisingly, this is arguably the most inhomogeneous group of papers. Dabir-Moghaddam ("On the independence of syntax and pragmatics: evidence from the postpositition -ra in Persian") takes up a classical problem of Persian and a notorious stumbling block for the teaching of Persian. He analyzes a Persian particle and shows that neither a purely syntactic nor a purely discourse-functional approach to the description of that particle will do, but that there is a complex interaction between the two. Ward —Birner ("VP inversion in written texts") look at the linguistic states of a structure typical for newspaper texts; and Rhodes ("Scope in discourse: pragmatics or syntax?") looks at a phenomenon that contributes significantly to textual coherence, namely different versions and scopes of scope. Shen's contribution ("Parsing procedures for stories") forms a bridge between the papers explicating the construction of the global text meaning by applying cognitive knowledge to it (e. g., Wodak) and the papers concerned with algorithmic comprehension in that it represents an attempt to formalize the notion of a story grammar by using x-bar theory as a matrix concept.

Issues

11

The final section of papers, then, deals with problems of text comprehension by the machine proper. Though at first glance it may seem farfetched to bring together papers from so heterogeneous-sounding fields like literature and the computer, there are several good reasons for doing so. The most important one is the aim and philosophy of the volume in its role as a representative collection of papers for a text science. N o such university course can dispense with computational linguistics and discourse representation, as the fields with the most intense cohabitation of linguistic theory and practical application, such as increasingly required as integral part of the job descriptions of the positions students of this type of course will take. The theoretical connection between computational discourse linguistics and the rest of the volume is also established by the tremendous thrust computational analysis of texts has brought for the linguistic analysis of texts at large. The necessity to formulate algorithmically to the smallest detail processes involved in comprehension has catalyzed research in these processes in human comprehension. To the extent that the problems involved are mostly concerned with indeterminacy of meaning and inferencing the content of the volume has at this point gone full circle. The difference is, however, that in computational linguistics inferencing is a problem, and in literary texts it is intentional. It is, of course, clear that computational text comprehension is concerned only with low-level aspects of comprehension, i. e., discourse representation at the level of individual propositions and their linear coherence. Hierarchical structurings seem yet out of reach. Higher level aspects of comprehension are, however, necessarily involved as top-down processes in low-level processing. One of the central problems is how the machine can know, or how to let it know, what the topic is and which knowledge frames are to be selected, and — more importantly — which ones not, which is central for the resolution of polysemy, homonymy, and pronoun reference. It would seem that there is an important role for research in text typology such as local text conventions, to be incorporated in computational discourse representation. These questions all surface in Dahlgren's ("Convergent evidence for a set of coherence relations") paper which proposes a solution to the problem as to how to algorithmically account for coherence. Where computational linguists have often stopped at a purely formal analysis of discourse, Dahlgren extends the domain of discourse representation to a "genuine" text level, i. e., more than local or propositional, by including global coherence relations. Coherence is established in a module separate from semantics and contains common sense world knowledge ("naive semantic theory") which is used to create

12

Dieter

Stein

a finite set of links between events in a model of the cognitive world built up during processing. The algorithm is explicitly demonstrated with a text. Rothkegel ("Text pragmatics and computational modeling") gives a survey of approaches to the computational analysis of texts and the various types of relationships between linguistic and extralinguistic world and text type knowledge necessary in such an analysis. She goes on to demonstrate how one particular version of the combination of these knowledge types is implemented in the practical analysis of a text. The final paper in the section discusses the solution to the compartmentalization and interaction of these different knowledge types in an example taken from a natural language dialogue project in German. Block's paper ("Integrating knowledge sources for the generation of referential expressions") is set in the context of a discussion of the relationship between linguistics pure and computational linguistics and explicitly highlights the epistemological and heuristic value of the latter for the former discipline.

References Dijk, Teun A. van and Kintsch, Walter. 1983 Strategies of Discourse Comprehension. New York: Academic Press. Dijk, Teun A. van (ed.). 1985 Handbook of Discourse Analysis. London: Academic Press. Dressier, Wolfgang—de Beaugrande, Robert 1981 Introduction to Text Linguistics. New York: Longman. Hörmann, Hans. 1976 Meinen und Verstehen: Grundzüge einer psychologischen Semantik. Frankfurt am Main: Suhrkamp. Nystrand, Martin. 1986 The Structure of Communication: Studies in Reciprocity between Writers and Readers. Orlando, Florida: Academic Press, Inc. Whiteman, Marcia F. (ed.). 1981 Writing: The Nature, Development, and Teaching of Written Communication. Hilldale, New Jersey: Erlbaum.

Part I Writing, literacy, and comprehension: psycholinguistic aspects

On the relationship between writing system, written language, and text processing Florian Coulmas

"Written Text" is a very general notion. It is important, therefore, to keep in mind that there is no specific object corresponding to this notion; for notions of this kind lend themselves easily to making very general statements. Most of the statements which have been made in recent years about written texts 1 are, of course, about very particular texts, for example, scientific texts or legal texts or texts from the French Revolution. Sometimes the specificity of the texts under discussion is made explicit, sometimes not, and this is where the danger lurks, because sometimes what are statements about particular kinds of texts are put forth with much more general claims. Even though a great deal of knowledge about writing and written language has been amassed in recent years, except for the most obvious things, most of what we know in this regard is very particularistic. Thus, when we talk about texts, we always talk about texts written in a particular language with a particular literary history by means of a particular writing system. At the present state of knowledge we are not in a position to appreciate these variables very well. Therefore, we cannot afford to ignore them. We should not allow ourselves to be overwhelmed by what has been called, quite abstractly, "the technology of the intellect" (Goody 1977), because this technology, writing, comes in many different forms serving different purposes. Every writing system and every written language and every piece of written text is part of a literacy. Literacies differ in various ways, and what I propose to do in this paper is to look at some of the factors influencing a literacy. The most fundamental prerequisite of any literacy is a writing system. I will focus, therefore, on both systematic and social properties of writing systems which, I hope it will become evident, have important repercussions for literacies and hence for the constitution and processing of texts. For brevity's sake I will refer to the Chinese and English writing systems as my main examples, although I will occasionally refer to others as well. The factors to be discussed are the following:

16

Florian Coulmas

(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)

the building blocks of writing systems; economy; learnability; transferability; retrievability; directionality; self-monitoring ability; dissemination; impact on language; meta-linguistic organization potential.

The building blocks of writing systems Every writing system, in one way or another, relates to a language, which implies that written text can be assigned to, and interpreted in terms of, a given language. Typically, texts can be interpreted in terms of one and only one language. The relations between writing systems and language are of different kinds. On the most general level just two types of writing systems can be distinguished, those making use of building blocks which denote linguistic meaning and form, on the one hand, and those making use of building blocks which denote linguistic form only, on the other. In glossematic terms they can be called "pleremic systems" and "cenemic systems," respectively (cf. Haas 1976). The systematic difference between pleremic and cenemic systems is obvious enough, but its implications for processing texts of either kind are not. This is so because text processing does not necessarily operate on the level of building blocks, but often relies on larger or smaller units. Let us consider an example. The Chinese writing system is the most widely used system of the pleremic type. It is a morpho-syllabic system whose building blocks are characters each denoting a morpheme plus syllable. The English writing system, on the other hand, is a system of the cenemic type whose building blocks are letters of the Roman alphabet each denoting one or more phonemes. The structural nature of the value of the building blocks of a writing system allows only partial conclusions about how the system operates and what it takes to process texts written with it. The many thousand Chinese characters are all different, but they can be grouped for certain graphical similarities. Most characters are composed of repetitive configurations of strokes one of which is a semantic classifier and another a phonetic. This means that the processing of

Writing system

17

Chinese characters is a parallel operation where both semantic and phonetic information is made use of. This can also be said of English where the cenemic principle of its elementary signs is supplemented at a higher level by pleremic applications. For example, the letter sequence < r-e-a-d > has two different phonetic values, [ri: d] and [red]; and the phonetic form [ri: d] has at least two graphic representations, < read > and < reed > . This is much like Chinese where, for instance, the graphic element ^ can represent both [cen] and [tag], as in Κ 'to stare' and i t 'mantis', respectively; and where the phonetic form [tag] has at least two graphic representations, ^ and if . Thus, in both systems graphic and phonetic elements are multi-valued with respect to each other. This is not to say that the Chinese and English writing systems are basically the same or even similar, as has been suggested occasionally, but that systematically different building blocks may allow for the application of similar principles, and that the interaction of cenemic and pleremic principles on all relevant levels must be studied for every writing system. Although writing systems can be classified as pleremic or cenemic on the basis of the nature of their building blocks, most systems are mixed systems if we look at their mode of operation. Nevertheless, the systematic nature of the building blocks is important in several respects, as will become clear in what follows.

Economy The systematic status of the building blocks of a writing system is directly reflected in the economy of its inventory of signs. There are more words than morphemes, more morphemes than syllables, and more syllables than phonemes. If a purely logographic writing system existed, it would have the most numerous inventory of signs. Since there is no such system, the morpho-syllabic system of Chinese has the most numerous inventory of all existing writing systems and, as far as we know today, of all writing systems that have ever existed. The number of Chinese characters that have ever been created may be as high as 80,000 (Du Ponceau 1883), and that of characters in common use at any one time ranges between 2000 and 5000, depending on the kind of text. As compared to the twenty odd letters of the alphabet, this is extravagant. However, the above reference to semantic classifiers and phonetics raises the question whether these rather than the characters should be considered the building blocks of the system. There are only 214 classifiers. Such an order of magnitude

18

Florian

Coulmas

would make Chinese look less exotic and more on a scale with other writing systems, such as the Ethiopic with its 182 letters or the two Japanese Kana syllabaries consisting of 48 signs each. But this would be like taking straight and slanted strokes, loops and half circles as the building blocks of alphabetic writing systems. We would be talking then about its graphic make-up rather than the way it visually encodes information. Although Chinese characters are composed of recurrent elements with relatively stable values, there are no rules for combining them, or, more accurately, while certain configurations can be ruled out on the basis of some general composition principles, there are no rules for composing characters. In alphabetic systems visual information coding is basically linear with graphotactic rules determining the order of elements. The arrangement of character components, by contrast, is in two dimensions covering a field where a semantic classifier can appear on either side; in the top half or third; in the center; or in the bottom half or third; to mention only the most common options (Figure 1). Where it appears is in no way systematically

• Figure 1. The most common options for arranging Chinese character components

determined. Thus, knowing the meaning of a morpheme, the speaker of Chinese cannot deduce how to write it. Therefore, characters rather than classifiers or any other recurrent components have to be considered as the basic units of the Chinese writing system.

Learnability Although the size of the inventory of basic signs is a factor influencing the learnability of a system, both are not co-variant. Clearly, the many thousand characters of Chinese present a formidable learning task compared to which the learning of any alphabetic system is relatively easy. But it must be noted that alphabetic systems vary greatly with respect to ease of learning (Scheerer 1986). The difference between mastering a "shallow" system such as the Spanish and a "deep" system such as the English is said to be as much as a year of instruction in elementary school.

Writing system

19

Another point to be taken into consideration with regard to the economy of a writing system is that the size of the inventory of signs and the ease of learning do not determine ease and efficiency of use. Rather, the proficient reader of Chinese and, say, English is said to be able to process Chinese texts more quickly, hence efficiently than English texts (Wang 1980). The economy of the inventory of writing systems has often been taken as a measure of its quality. However, if we recall Zipf s (1949) analysis of the "principle of least effort" and take a person's average rate of work-expenditure over time as a measure, that is, the effort necessary and using a system, then it becomes more difficult to justify the superiority of alphabetic systems in economic terms.

Transferability Transferability marks the property of a writing system to be applicable to languages other than the one for which it was invented. It has often been pointed out that the alphabet is the most flexible system which can be easily transferred to unwritten languages or languages written with other systems. It should be noted, however, that the underrepresentation of vowels in virtually all alphabetic writing systems still bears witness to the Semitic origin of the alphabet. For most languages the alphabet has to be expanded in some way. Most systems use digraphs, multi-valence of letters, diacritics and/or additional letters. Transfer problems are especially noticeable in the media of literacy technology. Because of the international dominance of English in communications and technology, many international communications media support only the 26 letters used by English. In telex or telegrams accented letters are, therefore, transmitted stripped of their diacritics. As will be pointed out in greater detail below, the fact that communication is not usually severely hampered by these omissions does not mean that accented letters can be dispensed with. A distinction should be made, then, between using the building blocks of a given system for creating another, on the one hand, and using the system of one language for writing another language. W h a t this implies is obvious when we consider, for example, proper names. In alphabetic literacies foreign proper names tend to be pronounced according to the pronunciation rules of the recipient language; in literacies based on Chinese characters, on the other hand, the situation is more complicated, since the countries of the universe of the Chinese script have adopted

20

Florian

Coulmas

Chinese characters for most of their proper names which, however, are not necessarily pronounced in a sinicized way. In Japan, for example, Chinese characters are used for personal names which are, however, Japanese words. At the same time, Chinese proper names are pronounced with Sinojapanese pronunciation, for example, Deng Xiaoping's name is read [to: so: he:]. W h a t this also shows is that cenemic and pleremic writing systems impose different restrictions on lexical borrowing and hence on the interaction between different literacies. Pleremic signs tend to be semantically interpretable across languages, whereas cenemic elements tend to be phonetically interpretable across languages. Certain distortions have to be reckoned with, however, in both cases.

Retrievability If a message that has been put down in writing cannot be retrieved later it will lose its use. Undeciphered documents have no utility. Writing systems and conventions differ on a large scale with respect to the retrievability of the messages encoded by means of them. Of Chinese it can be said that "anyone who can read the language has relatively little difficulty in understanding a text of any age, whether it was written now or a couple of thousand years ago" (Needham 1978: 14). 2 The Chinese can communicate, if not co-operate, with written texts across the millennia more easily, it would seem, than users of alphabetic writing systems. The main reason for this is that the written language is much less susceptible to changes in the spoken language than is usually the case in alphabetically written languages.

Directionality Directionality has to do with the flow of written documents inside the human group which makes use of them. Writing technologies set different conditions for the direction of this flow. In the era of hand writing technology was equally distributed among all participants. However, not everyone who knows how to write can make use of printing. Rather, with printing where possession of the means of production is limited, documents have become directed from those who had access to a printing press to those who were allowed only to read the printed documents.

Writing system

21

That property has had an influence on the development of society, literacy, and language (Eisenstein 1979). In some measure directionality is influenced by certain features of writing systems. The Gutenberg revolution could not affect Chinese literacy as much as European literacy. As a matter of fact, moveable type had been invented in China earlier than in Europe, but with a system where several thousand such types were needed printing remained cumbersome and slow. Therefore, handwriting has continued to play a much more important role in East Asia until the present than in other parts of the world. The new media have brought about significant changes in directionality. On the one hand, word-processing equipment and desk-top printing have greatly expanded access to printing, and on the other hand, tele-fax has somewhat unexpectedly revitalized handwriting. This is particularly obvious in Japan where fax machines have quickly become consumer articles. Because of the intricacies of the Japanese writing system, the typewriter never was a consumer article there. The writing system had also been a major obstacle for the development of word processing software and hardware. The Japan Industrial Standard stipulates that some 6,800 characters, including Chinese characters, the two Kana syllabaries, Chinese and Arabic numerals, as well as Roman letters are to be used for communicating visually in Japanese. To accommodate so many different characters on a manageable keyboard was a formidable task which could only be solved by equipping a microchip with the means of carrying out parsing operations and lexical identifications on the basis of which an input with alphabetic or syllabic script could be transformed into a proper Japanese text. These problems have been solved, and word processors have now passed the "one per office" stage and reached the "one per family" stage.

Self-monitoring ability It has often been remarked that writing is stable while speech is ephemeral. Since writing consists of producing a visible trace, it can be monitored more easily than speech, and deviations can be checked more easily. In handwriting the ability to limit deviations from a pattern depends on the writer alone. Printing and typing make possible the reproduction of more or less identical signs independent of the writer's skills.

22

Florian

Coulmas

Computer technology has broken new ground in offering the possibility of self-monitoring more complex signs, such as words. It is interesting to note that different writing systems and different languages define quite different conditions in this regard. Computer programs designed for word checking are called "spellcheckers". These programs are so devised that they detect every sequence of letters which is not stored in their dictionaries. The fact that the first spellcheckers were designed for English is not only a result of the advanced development of computer research in English-speaking countries. It was because English has a very simple morphology that the developers of these programs were able to put every word form into the dictionary, that is, to treat every inflected form as a separate entry. Attempts to develop spellcheckers for highly inflecting languages with a complicated word formation morphology, such as German, have generally failed. There are no spellcheckers for Chinese, although there is word processing software for Chinese. The reason for this is that no "misspelt" Chinese character could ever appear on the screen, since the characters are stored whole, that is, they are the elementary units. One could imagine, of course, a program for checking compounds, that is, sequences of characters. But since ad hoc compounds are extremely frequent and hard to distinguish from syntactic groups, automatic self-monitoring and checking for Chinese would really only make sense with a parsing program, that is, on the syntactic level.

Dissemination It has been argued that systematic properties of writing systems have an influence on their dissemination. In particular, the alphabet has been termed a "democratic" system (Diringer 1968) which makes universal literacy possible. In view of the fact that very simple, alphabet-like systems in India have never led to universal literacy, while the extremely complex system of Japanese has been disseminated to an extent which already in the 19th century was on a par with literacy rates in Europe, I never found this line of argument very convincing. Instead of considering the systematic conditions of the dissemination of different writing systems, I want to raise the question of what effects variable dissemination levels are likely to produce. A typical feature of societies with very limited dissemination of writing is the emergence of a diglossic split in the language, that is, a divergence

Writing system

23

between spoken and written language. The intensity of this effect is likely to vary with the dissemination of literacy: The smaller the number of those who can read and write, the more restricted the variety of the language which is written, and the smaller the chance that knowledge of the written language will influence the spoken language. Differences between spoken and written language are a general phenomenon; however, diglossia, that is, the deep rift between the two which makes the latter inaccessible without deliberate instruction, is typical only of speech communities where literacy has been low and restricted to a small group for a long time. Arabic, Tamil, and Sinhalese are standard examples (Britto 1986; Haas 1982). A question worth asking is whether the coming into existence of diglossia has anything to do with the systematic properties of writing systems. De Silva (1976) has argued that the Chinese writing system does not motivate the creation of a diglossic situation, because its loose connection with speech sounds makes written Chinese less likely to be affected by changes in the spoken language than alphabetically written languages. The gist of this argument is that only phonetic writing systems are likely to produce diglossia. De Silva maintains that "from the point of view of the relationship between speech and writing this kind of writing (i. e. Chinese) appears to be an ideal one, for it can always be said that the writing system is, in a sense, a true representation of each person's verbal behavior" (De Silva 1976: 36). However, while it is true, as De Silva (1976: 35) points out, that "different persons may read (Chinese characters) each in his own pronunciation", this does not mean that the Chinese written language is a "true representation" of anyone's speech behavior. Rather, it has been typical of written Chinese throughout its history that it was a fairly independent code far removed from speech. Mastery of this code was the prerogative of a small elite of literati who used it for purposes quite different from the reproduction of speech. It is a "book language" with many features of the " H " variety of a diglossic pair of codes. Thus it is for social reasons, it would seem, rather than for reasons having to do with the systematic make-up of the Chinese writing system that China has developed a literary culture where the distance between spoken and written language is as wide as in other diglossia situations. Appealing as it is, De Silva's argument is problematic because it suggests that the relation of written characters to speech sounds is the only or most significant aspect of the relationship between spoken and

24

Florian

Coulmas

written language. But writing can influence language in many other respects and, accordingly, spoken and written language can differ in many other ways.

Impact on language Writing a language that has never before been written has been labeled and conceived of as a "reduction to writing", because writing systems are generally seen as secondary systems unable to reproduce all of the distinctions of speech. The question then is whether and how different writing systems affect this reduction and through it the development of a language. That the writing system, conservative and fixed in nature, typically does not evolve in synchrony with the spoken language is a general phenomenon, but the extent to which this and the ensuing need to spend much time for teaching the mother tongue in literate societies have an effect on the spoken language varies with the writing system and the social conditions of its use. In Chinese, for instance, there can be no spelling pronunciation; and in languages with cenemic writing systems the conditions that favor spelling pronunciation depend on the abstractness and the accuracy of the fit between spelling and pronunciation, as well as on the perceived relationship between spoken and written language. In diglossia situations the written variety is not perceived as mapping the spoken variety or as a reduction thereof. Reduction in the sense of leaving some features of speech such as prosody, volume, or exact pronunciation unrepresented depends, partly at least, on the specific nature of the writing system in question. Writing should, however, not be seen solely as bringing about a reduction of speech. Through writing speech is not only reduced, but also expanded. There are two aspects to the expansion of speech through writing: (1) the size of the vocabulary, especially terminology; and (2) the size of the community making use of the code. Standardization on a large scale becomes possible only through writing. Written languages making use of pleremic writing systems can contribute less to the standardization of the spoken language than those making use of cenemic systems, but the written language can nevertheless, and maybe more easily than under conditions of a cenemic system, be devel-

Writing system

25

oped into a standard variety transcending the variability of the spoken language. The primary function of the Chinese writing system has never been to serve as an instruction for pronunciation. More than any other written language literary Chinese unfolds its peculiar qualities in the visual mode. For an administration which placed high value on keeping retrievable records and reaching the remotest parts of an immense empire this was, perhaps, more important than implementing a standardized pronunciation. As an administrative language it served functions similar to those of medieval Latin. One reason that Latin was abandoned earlier than classical Chinese is that its increasing distance from the spoken language was seen as a distortion and violation of the principle underlying its cenemic writing system. Hence, the vernacular varieties of the language were reduced to writing, a process which resulted in the emergence and standardization of the Romance languages. Turning to the other aspect of writing as an agent of standardization, it is clear that writing is an important aid, if not an indispensible prerequisite of, creating a controlled terminology. For example, there are some eight million chemical compounds that have been registered under unique structural formulae and given terms which are, ideally, also unique. No speaker can handle such a terminological system just by keeping it in his or her head. The terms must be registered in a data base which makes use of written signs. For co-operating with scientific texts it is essential that the participants can refer to a standardized terminology. There are no apparent properties which make one writing system more suitable for creating terminologies than another, although cenemic systems lend themselves more easily to the adaptation of words from other languages. It should be noted, however, that pleremic systems, too, can serve as a means of creating terminologies which are comprehensible across languages. For example, until the nineteenth century, many technical terms were borrowed from Chinese into Japanese. Ever since Japan outstripped China in science and technology, the information flow is in the opposite direction. In this way, Chinese characters are the basis of what might be described as a written Sprachbund. However, with respect to the systematic ordering of terminologies, different writing systems have very different properties which are discussed in the next section.

26

Florian Coulmas

Meta-linguistic organization potential Writing systems are based on, and hence incorporate a particular level of,linguistic analysis. Accordingly, writing systems can be viewed as schemes of meta-linguistic organization. More than that, they embody a view on language. This becomes clearly visible when we consider the structural patterns imposed on dictionaries by different writing systems, and their consequences for the electronic processing of language data. Word lists of alphabetically written languages are typically presented in alphabetical order, that is, for spelling rather than pronunciation. This implies among other things that homophones are not found in adjacent positions unless they happen to be homographs as well (e. g., phrase, frays [freiz]), and that adjacent words are often pronounced quite differently despite similarity in spelling. More generally speaking, for languages with complicated grapheme-phoneme-correspondence rules it may be difficult to find a word in a dictionary if the spelling is unknown. Clearly, this is also a problem in the pursuit of the ambitious goal of connecting voice recognition systems with computers in order to make a vocal language input possible. In Chinese, too, word lists are ordered on the basis of graphic rather than phonetic properties. It is even more difficult, therefore, to find a word of which the spelling is unknown, A rather special case is a particular kind of Japanese dictionary which owes its existence exclusively to the peculiar adaptation of Chinese characters to the writing of Japanese. Most Chinese characters are used to write both words of Chinese origin and Japanese words. In so-called Kanwa or 'Chinese-Japanese' dictionaries words of these two classes are related to each other by means of the characters with which they are written. To the Japanese it is more intuitive, therefore, to say that a character has two different pronunciations than that two different words are written by means of the same character. In this sense the character is the word. Yet another example testifying to the influence of a writing system on dictionary organization are dictionaries of Semitic languages which are ordered for consonant roots. Dictionaries are meta-linguistic devices on which co-operation with written texts is based, implicitly or explicitly, in many respects. It is important, therefore, to be aware of the different classifications they impose on a language by virtue of the writing system used.

Writing system

27

For further illustration it is useful in this connection to examine some of the problems of managing word processing and information processing at the United Nations in the six official languages: Arabic, Chinese, English, French, Russian, and Spanish. English, French, and Spanish make use of the Roman alphabet, but French and Spanish have accented letters. Although this seems to be a minor difference, it creates significant problems for automatic word processing. For instance, alphabetic sorting operations can be difficult. Standard word processing software treats accented letters as if the accent marks had been stripped of. However, according to French and Spanish lexicographical practice, unaccented letters are sorted before accented letters. Simple as this may seem, it is too demanding for a machine not specially equipped to perform this operation. A similar problem which machines cannot handle is that in Spanish the digraph < ch > should come after all < c's > and before the < d's > , just as the digraph < 11 > should come after all single < l's > . The Roman alphabet has been imposed as a lexicographical ordering code on languages with non-alphabetic literacy traditions, such as Chinese, in order to make them more manageable for telecommunication and electronic processing. The adaptation is more difficult here than in the cases previously mentioned. To call attention to just two points, in traditional Chinese dictionaries the lemmata are characters rather than words. In running text, word boundaries are not marked. Alphabetic listing, however, presupposes unequivocal segmentation. Also, tone is phonemic in Chinese and should, therefore, be marked when Chinese texts are written with the alphabetic system Hanyu Pinyin. This implies either diacritics or a non-segmental use of letters. Both options create problems for sorting and listing. Russian and Arabic pose altogether different problems having to do with competing transliteration conventions, which in turn are a consequence of the fact that the letters of the Roman alphabet are assigned different values in different languages. These examples suffice to show that a writing system is more than a transparent mapping device which does not affect the object it is used to represent. As a meta-linguistic organization principle every writing system imposes as much structure as it reflects. Therefore, the seeming triviality of orthographic conventions can present a range of problems for cooperating with written texts, especially regarding telecommunications, electronic data processing, and communication at the interface of different languages.

28

Florian

Coulmas

Conclusion In the preceding pages I have addressed some of the preconditions of cooperating with written texts, although I have said little about how people use texts to cooperate, because I have concentrated on the means of creating written texts rather than texts themselves. W h a t should have become obvious is that, in some measure, different writing systems favor different ways of cooperating with written texts. Without invoking the usual comparison of the human brain with a computer, we can learn much about the interaction of structural and functional properties of different writing systems and the languages they are used to write by studying the problems arising in conjunction with attempts to enable machines to process texts. Although the alphabet has been regarded as a universal writing system, its implementation on a machine reveals that, as a means of constituting written texts, it is always part of a particular literary tradition which defines particular conditions for cooperating with written texts.

Notes 1. Cf. for example, works by Chafe (1979); Goody (1977); Günther (1988); Feldbusch (1985); Olson (1977); a review of the literature is provided in Coulmas (1989). 2. Needham's statement should be understood to mean that anyone who has mastered the classical Chinese written language can read Chinese texts of any age. Because of the post-World War II reforms of the Chinese written language students who have learned the modern written language only are not necessarily able to comprehend classical texts (cf. Coulmas 1983).

References Britto, Francis 1986 Chafe, Wallace 1979

Diglossia: A Study of the Theory with Application Georgetown University Press.

to Tamil. Washington:

"The Flow of Thought and the Flow of Language", in: T. Givon, (ed.) Syntax and Semantics. Vol 12, Discourse and Syntax. New York: Academic Press, 1 5 9 - 1 8 1 . Coulmas, Florian 1983 "Writing and Literacy in China", in: F. Coulmas —Κ. Ehlich, (eds.) Writing in Focus. Berlin, Amsterdam, New York: Mouton, 239 — 254.

Writing system

29

Coulmas, Florian 1989 The Writing Systems of the World. Oxford: Blackwell. De Silva, M.W. Sugathapala 1976 Diglossia and Literacy. Manasagangotri, Mysore: Central Institute of Indian Languages. Diringer, David 1968 The Alphabet: A Key to the History of Mankind. (Third revised edition.) New York: Funk — Wagnalls. Du Ponceau, P. S. 1883 A Dissertation on the Nature and Character of the Chinese System of Writing. Philadelphia: American Philosophical Society. Eisenstein, Elizabeth 1979 The Printing Press as an Agent of Change. London: Cambridge University Press. Feldbusch, Elisabeth 1985 Geschriebene Sprache. Untersuchungen zu ihrer Herausbildung und Grundlegung ihrer Theorie. Berlin, New York: De Gruyter. Goody, Jack 1977 The Domestication of the Savage Mind. Cambridge, London: Cambridge University Press. Günther, Hartmut 1988 Schriftliche Sprache. Strukturen geschriebener Wörter und ihrer Verarbeitung beim Lesen. Tübingen: Niemeyer. Haas, W. 1976 "Writing: The Basic Options", in: W. Haas, (ed.) Writing without Letters. Manchester: Manchester University Press, 131—208. Haas, W. (ed.) 1982 Standard Languages, Spoken and Written. Manchester: Manchester University Press. Needham, Joseph 1978 The Shorter Science and Civilization of China. Abridged edition by C. A. Ronan. Cambridge, London: Cambridge University Press. Olson, David 1977 "From Utterance to Text: The Bias of Language in Speech and Writing", Harvard Educational Review 47: 257 — 281. Scheerer, Eckart 1986 "Orthography and Lexical Access", in: Gerhard Augst, (ed.) New Trends in Graphemics and Orthography. Berlin: De Gruyter, 262 — 286. Wang, William S.-Y. 1980 Speech and Script in some Asian Languages. Symposium on Bilingual Research — Asian/American Perspectives, September 3 — 5, Los Alamitos. Zipf, George K. 1949 Human Behavior and the Principle of Least Effort. New York: Hafner (facsimile reprint 1972).

Pause and intonation contours in written and oral discourse1 Gisbert Keseling

0. Introduction Recent research on written communication is characterized by a controversy about the question whether written and spoken discourse is based on the same abstract structure from which both oral and written discourse are derived, or whether there are two more or less separate and autonomous systems (e. g., Nystrand contra Olson; or Augst contra Feldbusch). Let me first explain what I understand by the dependency hypothesis and the independency hypothesis. Suggesting that written communication functions in a definable sense independently from oral (or conversational) communication means postulating a modular organization of our knowledge of discourse, i.e., to hypothesize that our knowledge of discourse consists of three systems, one in which our knowledge of grammar is organized, another one in which our knowledge of conversation is organized and a third one in which our knowledge of texts is organized. Figure (1) illustrates this assumption. speech

conversation

c,

c2

texts

c„

t,

t2

t„

Figure I. Components of linguistic knowledge I C|, c 2 , . . . , c„ = types of conversation t,, t2, . . . , tn = text types

This means that the two components "knowledge of conversation" and "knowledge of texts" are independent from each other, such that the former has to be subdivided into conversational subtypes and the latter into textual subtypes with no or only few connections between the conversational and the textual subtypes. To give an example: if c t stands

32

Gisbert

Keseling

for knowledge of personal conversation between family members and t, for written correspondence conducted between family members, there would be no or only little direct connection between these two subtypes of knowledge. On the other hand, to hypothesize direct connections between types of conversation and corresponding text types we would obtain Figure 2. speech

di

c,

d2

t,

c2

dn

t2

c„

d„

Figure 2. Components of linguistic knowledge II di, d 2 , . . . , d„ = types of discourse

This dependency hypothesis thus means that the knowledge of skilled writers and/or readers is organized in such a way that it contains pairs of knowledge types, one knowledge type which enables the individual to produce and to understand the oral form, and the other one which enables the individual to produce and to understand the written form. The dependency hypothesis thus means that the corresponding types of knowledge and skills are related to each other in a definable way, for each pair. The decision for Figure 1 or 2 depends, as I believe, on the answer to the (empirical) question, in what respect our knowledge of written and oral discourse differs and in what respect these two types of knowledge are identical. Since most of the contributions to this problem are based on facts derived from grammar, semantics, pragmatics etc. while prosody is widely neglected, I restrict myself to what I want to call the auditive component of conversation and texts; and from this auditive component I extract the systems of pauses and, partially, of intonation. My question, then, is: do conversation and written texts differ in this respect from each other, and if this is the case, how may these differences be described, explained and, possibly, integrated into the components of linguistic description? Before I begin with the empirical part of my investigation it is necessary to explain what I mean by a text, and by conversation. Following Ehlich

Pause and intonation

33

(1984) I understand by a text an utterance or a sequence of utterances that can be preserved either in our memory (oral texts), or on paper, etc. (written texts), and that therefore can be produced literally or, as for oral texts, literally and/or in a modified literal way. By semi-texts I understand a sequence of oral utterances spoken by one person, e. g., a story, a joke or an argumentation, which are embedded in a turn-by-turn-talk. Conversation, in my terminology, is every portion of oral speech which is organized by the turn-taking mechanism (Sacks —Schegloff—Jefferson 1974). In this paper I am mainly concerned with conversation and with written texts, but it is important to keep in mind that there is an intermediate zone between conversation and written texts as in Figure 1 and that the whole line from c t and t n may, possibly, be regarded as linearly arrayed, in a similar sense to Sacks —Schegloff—Jefferson's (1974: 729) linearly arrayed turn-taking systems. Let me first summarize some points of earlier research on pauses. There are, in my view, five main lines of approach (cf. also the chronological review of O'Connell - Kowal 1983). (1) The psychological aspects of pauses in e.g. personality, anxiety and related emotions (e.g., Mahl 1963). (2) Pauses as a hesitation phenomenon (e. g., Goldman-Eisler 1968; Maclay—Osgood 1959; and recently Chafe 1980 a, b). Main results or hypotheses were: Pauses indicate processes of programming and planning. In spontaneous speech, pauses are mainly, but not regularly, located at clause- and sentence boundaries and before words or phrases with the highest amount of information, i.e., before lexical words rather than before grammatical words (Goldman-Eisler 1968; Maclay —Osgood 1959); fluency and hesitation correlate with a lower or higher amount of information in the words related (Goldman-Eisler 1968). The length of hesitation depends on the "amount of reorientation the speaker must undergo in his or her own consciousness and must transfer to the consciousness of the listener" (Chafe 1980 a: 49). (3) Pauses linked with breathing (e.g., Henderson —Goldman-Eisler — Skarbek 1965 a, b, 1966; Goldman-Eisler 1968). Main hypothesis: frequency and location of breath pauses largely depends on the grammatical structure of sentences. (4) Pauses as realizations of juncture (e.g., Crystal 1969: 206). (5) Pauses at possible transition points (e.g., Duncan 1972) pauses are, together with intonational features, interpreted as turn yielding signals. There are only a few investigations concerning the differences between written and oral speech (apart from the trivial fact that the medium for

34

Gisbert

Keseling

written communication is written paper, etc., while oral speech is transmitted by sound-waves etc.). Experiments by Henderson — GoldmanEisler —Skarbek 1965 a, b) showed that in reading aloud gaps are normally accompanied by breath and that the breath time is 77 per cent of the pause time. They compared this with the pauses in spontaneous speech and found that in spontaneous speech only 34 per cent of the pause time is filled by breath. They also discovered that in reading aloud pauses occur exclusively at grammatical juncture, again in contrast to spontaneous speech. Thus, following Goldman-Eisler, Henderson and others, one of the main differences between reading aloud and spontaneous speech is the frequency and location of pauses. In conversation they are not only more frequent but they also show a lower degree of regularity with respect to their location. In reading aloud, we might conclude, the task of planning is lower than in spontaneous speech. From a linguistic point of view these results and hypotheses raise the question of what pauses are like, which part or level, component, etc., of linguistic description pauses are a part of. Do they, for example, belong to the langue or competence or are they, like errors, false starts and the like a phenomenon of parole, performance, use of language, as, possibly Goldman-Eisler and her colleagues would argue. Or are they, in a sense that has still to be defined, both? The main point of the empirical part of my lecture, with which I am now going to start, is, that there are reasons to assume that pauses play an important role in speech production as well as in speech perception. Because of this and other reasons that become clear later, it is reasonable to assume that there exists an abstract structure that underlies most — not all — of the pauses occurring in conversation as well as in reading texts aloud and that parts of these abstract features are inherent to the structure of conversation and texts. And, secondly, I want to argue that there exist two kinds of pause sub-systems, one for conversation and the other one for texts, the former being linked with a sub-set of intonational features of spontaneous speech and the latter being linked with a sub-set of intonational features of written texts. I shall now proceed in four steps: In the first step I shall summarize briefly the results of an earlier investigation of fluent writing processes, in which we analyzed the length and distribution of pauses in composing directions and personal letters (section 1).2 In the second step I shall outline some preliminary results of a new project, in which we — inter alia — compare writing processes with reading aloud processes. In my paper I shall, especially, analyze pause and intonational patterns of fairy

Pause and intonation

35

tales and directions read aloud, in order to compare the reading pauses with the writing pauses (section 2.1 —2.2). In the third step I shall analyze some features of pause and intonation patterns in conversation and oral directions, in order to compare these structures with the pause and intonation patterns of written communication (section 3). Based on the prosodic features that I have so far identified I shall analyze the question of what role these features play in the production and reception of text (section 4), in order to establish a hypothesis concerning the dependency and independency hypotheses (section 5). We have, thus, pause analyses of (1) writing processes, (2) reading aloud processes and (3) of conversation and oral directions, i.e., of written speech production, written speech reproduction (and possibly reception) and of spontaneous speech.

1. Pauses in written directions and personal letters We selected these text types, in an earlier investigation, in order to get fluently written texts. The results of the pause analyses which I am now going to report (cf. Keseling 1987 a, b, 1988 b; Keseling — Wrobel — Rau 1987; Wrobel, in preparation) are only pertinent to fluent production, i. e., texts (or passages of a text) which are written from beginning to end as one piece, the pauses of which do not exceed a certain length and in which no or only few corrections and crossings out occur. (For the difference between fluency and non-fluency cf. Keseling (1989 b) and Wrobel (forthcoming.) By taking directions it was possible to establish that there is a distinction between longer and shorter pauses. Longer pauses generally occur at sentence or paragraph boundaries and shorter ones often but not regularly occur (a) at clause boundaries (or after the first word of a clause), i. e., after conjunctions or pronouns etc. and (b) before descriptions of streets, places, directions, etc., i. e., before the word (or phrase) which is the most important with respect to the informational structure of the unit in question. (By a sentence I understand a segment that can be articulated with falling pitch contour, in yes —no questions with rising intonation.) For the sake of simplicity I call the pauses at sentence boundaries "sentence pauses", pauses at clause boundaries "clause pauses" and pauses before street names, etc., decription pauses. Pauses at places other that the ones mentioned are rare. An example is given in (1)·

36 (1)

Gisbert Keseling Pause transcript of writing directions (fluently written) a: German text 1. Wenn Du aus dem Bahnhof kommst 10,8 2. siehst Du vor dir eine 1,0 Straße, 0,8 3. die du immer geradeaus lang gehst 0,9 (du 4. überquerst die Lahn 5,3 dabeil) 5,1. 5. Am Ende 4,4 macht die Straße einen 6. 90° knick nach 1,1 links 1,4 dem 1. du 3,7 folgst, 4,3 . 1,9 8. links siehst du dann die Elisabethkirche 10,5 9. Du überquerst 2,8 die Ampel und 10. gehst weiter geradeaus 3,0 (6,9 11. du kommst dann auf den „Pilgrimstein" 1,2). 23,8 12. Am besten gehst Du auf der rechten 13. Straßen sei 1,1 te 7,1. 14. Wenn Du dann wieder auf eine 3,5 Ampel 15. triffst, bleibst Du auf dem Bürgersteig 16. und gehst an der Ampel vorbei . 3,1 17. (Du kommst dann an der alten Universität vorbeil) 5,3. 18. Dann überquerst Du eine Straße 0,7, 19. die rechts steil nach oben geht 6,4 . 5,1 20. Du gehst an der Unterführung vorbei 1,0 21. und biegst re 2,4 chts 4,7 ab 1,7. 11,2 22. Die Straße geht dann 1,1 einbißchen 23. bergan 2,4 ... b: Literal translation (fluently written) 1. When you come out of the station 10.8 2. in front of you you'll see a 1.0 street 0.8 3. here you go straight ahead 0.9 (you'll 4. cross over the Lahn 5.3 along the way!) 5.1 5. At the end 4.4 the street takes a 6. 90° turn to the 1.1 left 1.4 which 7. you 3.7 follow 4.3. 1.9 8. On the left you'll then see the Elisabeth church 10.5 9. You cross at the 2.8 traffic light and 10. go straight ahead 3.0 (6.9 11. you then come on to the "Pilgrimstein" 1.2). 23.8

Pause and intonation

37

12. It's best to go along the right 13. si 1.1 de of the street 7.1. 14. When you come to the next 3.5 traffic light 15. you stay on the sidewalk 16. and go past the light. 3.1 17. (Then you'll go past the old university!) 5.3 18. Then you cross a street 0.7, 19. that goes steeply up on the left 6.4 . 5.1 20. You go by the underpass 1.0 21. and take a ri 2.4 ght 4.7 turn 1.7 . 11.2 22. The street then goes 1.1 a little bit 23. uphill 2.4 ... The numbers indicate writing pauses in seconds. As the transcript shows the sentence pauses are often considerably longer than the clause and description pauses. (For further detail, statistics and discussion of the exceptions cf. Keseling 1987 b.) Example (2) is a pause transcript of the beginning of a personal letter. (2)

Pause transcript of writing a personal letter 1. Hallo mein Schatz! 1,7 2. Schön, daß Du 0,9 endlich Zeit gefunden hast, 3. 1,2 mich in Marburg zu besuchen. 4,9 Wird auch 4. langsam Zeit 1,4 , denn da ich 1,6 seit fast 5. 1,1 vier (!) Wochen fertig bin, weiß ich nicht, 6. 0.6 wie lange ich 1,2 noch hier 0,8 bleiben werde. 13,0 Du 7. hast am Telefon gesagt, 1,3 daß Du mit dem 8. Zug kommen wirst 1,5 . 2,6 Abgesehen davon, daß 9. 0,7 die Bahnverbindungen 5,5 von der Schweiz aus 10. 1,2 nach Marburg miserabel sind 2,1 , finde ich 1,8 es 11. schade 18,3 , 29,5 denn ich hätte 0,8 gerne mal wieder 12. im amerikanischen Schlitten gesessen. 4,3 Golfchen

It shows, roughly, the same pause structure as the directions, with the only difference that, instead of the pauses before descriptions of streets etc. we would have pauses here before, e.g., vier Wochen ('four weeks'); die Bahnverbindungen ('train connections'); schade ('too bad'), which appear to be important words. We thus may generalize that, in fluent writing, shorter pauses normally occur at clause boundaries and before words with a high informational value, while longer pauses, normally,

38

Gisbert

Keseling

occur at sentence boundaries. These observations seem to coincide with the observations which have been made by Goldman-Eisler, Maclay — Osgood, and others, for the production of spontaneous speech, though, with the important difference that, in spontaneous speech pauses are considerably shorter on the average and that there is no significant length difference between word pauses and sentence pauses, an observation that might be explained by the fact that there is no risk of writers losing the floor, in contrast to speakers (Maclay —Osgood 1959). This raises the question whether some equivalent to the longer sentence pauses in conversation might be found, i.e., filled pauses or the like or whether oral and written discourse differs in this respect. A second question which could not be answered in our earlier research is the problem of into which category the pauses of media length are to be placed.

2. International and pause patterns in fairy tales and directions The pauses treated in section 1 are pauses of speech production. It might now be interesting to compare these patterns with the first results of a new project, in which we started to analyze the pause and intonation patterns in the reproduction of written texts. The questions this project seeks to answer are inter alia: How do adult and skilled readers read these text types aloud? Where do they pause? Are there differences in the length of their reading pauses? Do the pauses interact with intonation and if that is the case, in which way? Ten test persons were asked to read aloud fairy tales, directions, scientific reviews and a newspaper article. The readings were tape-recorded. Two independently working and trained persons are now occupied analyzing the recorded readings with respect to: (a) the duration and location of pauses; (b) tone units and their boundaries; (c) the nuclear patterns. In this article I restrict myself to the problem of the relation between pauses and intonation. Since we have just started with these analyses there are only a few results now and they are of a preliminary nature. In what follows I shall first analyze the pause and intonation structure of

Pause and intonation

39

fairy tales and directions and, based on my results, propose a notion of "prosodic structure of texts" which is independent from individual readings (section 2.1.) and, second, compare the results of my analysis with the pause structure of the writing processes.

2.1. Analysis In extract (3) you may look at the first lines of a transcript of one reading of the German fairy tale "Der Froschkönig oder der eiserne Heinrich", written down by the Brüder Grimm with the original text of the Brüder Grimm (edited by Friedrich Panzer). (3)

Pause and intonation transcript of reading aloud the "Froschkönig"'', by text person 3 (beginning of the text). a: German text

b: Literal translation the "Froschkönig"

1. Es war einmal eine Königstochter,

0,32

Once upon a time there was a princess.

2. die ging hinaus in den Wald 0,56

She went out into the forest

3. und setzte sich an einen kühlen Brunnen. 0,70

and sat down by a cool spring.

4. Sie hatte eine goldene Kugel, 0,5

She had a golden ball

5. die war ihr liebstes Spielwerk,

which was her favourite toy.

0,36

6. die warf sie in die Höhe 0,43

She threw it up high,

7. und fing sie wieder in der Luft 0,59

and caught it in the air,

8. und hatte ihre Lust daran. 1,16

amusing herself.

9. Einmal 0,2 war die Kugel gar hoch

of

geflogen,

One time the ball went really high.

0,28 10. sie hatte die Hand schon ausgestreckt

0,21

She had already stretched out her hand,

11. und die Finger gekrümmt,

0,65

12. und [?] sie war [?] wieder zu fangen,'

and bent her fingers 0,88

13. da schlug sie neben 0,5 vorbei 1,4 auf die Erde, 0,44 14. rollte und rollte 0,37

in order the catch it, but it fell down onto the ground beside her and rolled and rolled

40

Gisbert

Keseling

15. und geradezu in das Wasser hinein} 1,65

right into the water.

16. Die

Shocked, the princess gazed after the

Königstochter

blickte

ihr erschrocken

nach, 0,75

ball.

17. der Brunnen war aber so tief, 0,34

The spring was, however, so deep

18. daß kein Grund zu sehen war. 1,0

that the bottom couldn't be seen.

19. Da fing sie an jämmerlich

So, she started to whine and cry mis-

zu weinen 0,33 und

zu klagen·. 1,2

erable,

3

20. „äcA ! 0,95

"Oh!

21. wenn ich meine Kugel wieder hätte, 0,24

If only I had my ball again.

22. da wollt ich alles darum geben. 0,37

I'd give everything to get it back;

23. meine Kleider 0,24

my clothes,

24. meine

my precious jewels,

Edelsteine0,49

25. meine Perlen 0,42

my pearls,

26. und was es auf der Welt nur wär'." 1,57

and whatever else there is in the world."

27. Wie sie so klagte,

0,67

28. steckte ein Frosch

...

1. 2. 3. 4.

As she was thus lamenting, a frog stuck ...

text of the Brüder Grimm: „um sie wieder zu fangen" paragraph in the text of the Brüder Grimm level tone, but distinct from the progressive intonation as, e.g., in line 2 (cf. Ehlich 1986) Brüder Grimm „Edelgesteine" (Tippfehler in dem Lesetext)

(the superscripts are placed on the first syllable of the nucleus. means falling pitch movement; " " means level pitch; and means falling movement, but without the terminal function of

—-J'

I adopted the intonational system from Pheby (1974, 1981) and Fox (1984) with the three main types of nuclear tones falling, rising and level, including compound tones with two instead of one nuclear tone (cf. Halliday 1970; Crystal 1969; Couper-Kuhlen 1986). Each line represents a tone unit, containing one and only one nuclear tone or a compound nuclear tone as, e. g., in line 2 (hinaus in den Wald). To identify boundaries between tone-units we applied the two criteria, proposed by Crystal (1968: 205 — 206): (a) a "perceivable pitch change, either stepping up or stepping down, depending on the direction of nuclear tone movement — if falling,

Pause and intonation

41

then step up; if rising, then step-down"; (b) "the presence of junctural features at the end of every tone-unit", i. e., usually a slight pause with "accompanying segmental phonetic modifications (variations in length, aspiration, etc.)". In what follows I restrict myself to those aspects of the results which are connected with the relationship between pauses and intonation and the question whether the pause structure of the readings is related to the pause structure of producing written texts. The main results of the analyses of the ten readings are as follows: (1) The readings coincide to a high degree with respect to the location of boundaries between the tone-units (approximately 5 per cent deviations). (2) As for the pauses nearly all of them are located after tone units, and in terms of syntax, after clause boundaries. But notice that there are some tone units which are smaller than a clause, as e. g., lines 23, 24 and 25. Since the reading of all of our ten test persons coincide in this respect, we may conclude that there exists some kind of rule that in reading aloud pauses normally occur at tone-unit boundaries. All of them seem to be breath pauses too. There are only a few instances of pauses at places other than at tone unit boundaries, in the transcript, e.g., in line 9 after Einmal·, or in line 13 between neben and vorbei, or, in another reading after die Königstochter in line 16; or between rollte and und rollte in line 14. (3) The pauses vary in their duration from very short pauses, which could not be measured by a stop-watch, to pauses of approximately 2 seconds. Again, all of the ten transcripts coincide in this respect, as Table 1 demonstrates. Tone units terminating with shorter pauses and tone units terminating with longer pauses seem to be ordered as follows: A set of several tone units, each terminating with a shorter pause, is followed by one tone unit with a longer pause. And this one tone unit is followed, again, by a set of several tone units each ending with a smaller pause etc. In other words: The longer pauses normally seem to indicate the end of a somewhat longer passage of text with an uncertain but not too high number of tone units, each followed by a shorter pause. I call a text passage of this kind a "prosodic paragraph". Notice that the ten readings only roughly coincide with respect to the paragraph boundaries, but that there are differences too. You may see that in Table 1. In the columns after the lines you find the pause lengths in the reading of each test person. We may thus compare where the different readers have paragraph boundaries. I have marked these boundaries by a thick

Gisbert Keseling

>+200 α

ω W "(Ο § ο

0

proper name 200

Figure 3

J

noun

pr. n. / pronoun

anaphor/antecedent

noun / pronoun

Psycholinguistic

processes

95

tained an apparently character neutral statement such as "the atmosphere was hot and sticky that night" readers would typically only ascribe knowledge of this to the focussed thematic subject in the story. So if they had encountered a short text like the following: Mary hated going to the dentist and felt anxious when the receptionist called out her name. The atmosphere was hot and sticky and they were then asked to say whether the dentist or receptionist found it hot and sticky or whether Mary found it hot and sticky, only half were prepared to attribute the judgement to the dentist or the receptionist, while nearly all readers attributed the judgement to the focused, named character Mary. Additional experiments with materials like these indicated that the focussing inference effect occurs at the time of reading the critical sentence and can have immediate consequences for interpreting subsequent material in the narrative (Garrod — Sanford 1988). So these various studies would tend to implicate a limited capacity memory system which is responsible for holding a very small number of focussed entities to serve as preferred referents for pronouns and focal points from which to draw certain types of textual inference. But do similar constraints apply in the case of referential role assignments of the sort discussed earlier?

Focussing on situations and the assignment of discourse roles The prior discussion tends to suggest that the ease of interpreting definite descriptions referring to antecedent roles is in no way affected by attentional focus but is only subject to constraints arising from the match between the description and the specification of the antecedent. However the situation is clearly more complicated. Consider for instance the following text fragment: Mary usually lunches at Valentino's because she fancies the waiter there. For dinner she prefers to go to La Grande Bouffe where the food is better but the waiter/*he is not nearly so handsome. In this text, it is clear that the pronoun does not substitute for the underlined definite description since it can only identify the antecedent waiter (the one at Valentino's), and this does not lead to a plausible interpretation. The full description on the other hand does not attach to any antecedent discourse referent even though there is a perfectly match-

96

Simon

Garrod

ing one. Instead it identifies a currently relevant discourse role, "the waiter at La Grande Bouffe". So what is striking about this example is that the role identifying function of the description so easily overrides the more standard anaphoric reference function. Even though there is a perfectly satisfactory antecedent waiter sitting in the context, this does not seem to interfere with the situational role interpretation in this case. But clearly such a non-anaphoric interpretation will depend upon the extent to which the prior text establishes a new situational role to which the description can attach. The empirical issue is to determine under what conditions such situational role attachments will occur, in the absence of explicitly introduced antecedents. As a move in this direction, Garrod — Sanford (1983, 1982) carried out a number of self-paced reading time studies to establish the ease of such role identifications under different circumstances. In the first of these studies titles were used to suggest different situations and hence establish different roles. For instance, in the materials shown in Figure 4 (Garrod — Sanford 1983), the same basic passage could be entitled either "Telling a lie" or "In court". In the latter case, it might be expected that the reader could identify a number of situational roles, including "the judge", "the lawyers", etc., not associated with the former situation of telling a lie. So the experiment contrasted reading time under different antecedent conditions for a critical sentence containing a reference to one of these characters. The text could either have an appropriate or inappropriate title with respect to the role and either contain an explicit antecedent mention of the character or leave it implied. As can be seen in the reading time data in Figure 4, readers experience no extra difficulty interpreting sentences containing a reference to a situational role in the absence of an explit antecedent when the title is appropriate for that role, but do have trouble otherwise (i. e., in the no antecedent — inappropriate title condition). So this study confirms the idea that immediate role identification can occur with a definite description so long as there is a clear indication of the situational setting behind the story fragment. It does not however, establish the range of such roles afforded by any situation. The other study reported by Garrod — Sanford (1982) gives a clearer picture of how texts may serve to identify situations which enable a reader to focus on a limited number of roles. In this study verbs were used as a means of establishing the roles. Consider for instance the verb 'to drive' in the sense of "to direct a vehicle". When such a verb occurs in sentences like the following:

Psycholinguistic

Garrod-Sanford

Appropriate Title:

processes

(1983)

Topic

Passage

Context S.

In court Fred was being questioned (by a

Filler S.

He had been accused of murder

Target S.

The lawyer

lawyer)

w a s trying to prove his

innocence Inappropriate Title:

Telling

Topic

a lie

Context S.

Fred was being questioned (by a

Filler S.

He couldn't tell the truth

Target S.

The lawyer

Ι inappropriate

CD

Ε

1500

1450 Topic

σ> c Ό CO CD

oc

1400

Ο-

1350

Appropriate

— ι

1

Stated Antecedent

Figure 4

lawyer)

w a s trying to prove his

innocence

ο

.01) or word incremental reading plus lexical decision (r = .605; ρ >.01). However, for sentence incremental reading (r = .817; ρ < .01) and for word incremental reading (r = .848; ρ < .01), processing time totals positively correlated with those for standard reading. In the present study, phrase incremental display led to major deviations from standard reading. At first sight, this seems counterintuitive as

Reading time

109

phrases are known to be relevant processing units. However, any facilitating effects of phrase-by-phrase segmentation have been observed with poor readers but not with skilled readers (Beveridge — Edmundson 1989). To the skilled readers who participated in our study, phrase incremental display even may have been an obstacle. As phrase sections had beed defined according to surface sequence, the segments on display may not always have been in keeping with functionally relevant segments. Thus, in the case of phrase incremental reading, readers' attempts to reconcile display units and functional units may well have contributed to the deviation from standard reading. Except for word incremental reading plus lexical decision, absolute reading times were comparatively close to standard reading latencies. In fact, some of the reading time totals were even less for incremental reading than for unsegmented reading — despite the additional keystrokes. These cases show that an additional motor task does not significantly obstruct cognitive performance. Introducing a "cognitive" secondary task such as lexical decision, however, may result in dramatic deviations from normal text comprehension processes. This is hardly surprising because during everyday reading, people do not permanently ponder over orthography. In all, the findings indicate that the processes underlying word and sentence incremental reading are similar to those underlying standard reading. Because of the high correlation between word incremental reading and unsegmented reading, and because data yield was higher for word-by-word segmentation than for sentence-by-sentence segmentation, we chose word incremental reading as the method in further experimentation.

4. Stage 2: Predicting reading time Mark Twain (1879: 268) on "the awful German language": "Begin at the wrong end ... That is the German idea."

In the second stage of our research, we brought to bear the "German idea" by making reading time data our starting point. The objective of this stage was to uncover systematic relationship between the time needed to comprehend a word and specific structural characteristics associated with it. Why do some words take longer to read and comprehend than others? To what extent can word processing times be accounted for by

110

Gert Rickheit,

Udo Günther, Lorenz

Sichelschmidt

structural variables? The statistical technique used to evaluate the extent to which a criterion variable can be traced back to a number of predictive variables is multiple linear regression analysis (Graesser — Riha 1984; Haberlandt 1984). In our case, reading time for any word was estimated by a weighted linear combination of structural characteristics associated with that word. In a regression study, 45 volunteer college students read ten expository texts, each about 100 words in length. Texts were displayed incrementally on a monochrome monitor (ITT 3030). For every word, the average processing time was computed as a criterion variable. In addition, every word was analyzed with regard to 30 structural characteristics. These structural variables were selected so as to cover a variety of aspects relevant to text processing. The list of structural predictors included: 1. Formal variables. For example: Does the word in question coincide with a line break (CRLF)? How many characters is it in length (CHAR)? 2. Syntactic variables. For example: Does the word in question coincide with the beginning (SBEG) or with the end of a sentence (SEND)? Is it an adjective or adverb (WADJ)? 3. Semantic variables. For example: Does the word in question constitute a case of coreference (CORE)? Does it occur at the end of an elliptical cataphor (CEND)? 4. Discourse variables. For example: Does the word in question serve to introduce a novel concept (INTC)? Does it serve to integrate several phrases (INTP)? Structural variables were coded independently by three expert raters on the basis of unanimous judgment. In accordance with the immediacy assumption, codes were assigned to the earliest possible verbal element. This coding principle shall be illustrated by means of the N E N D (end of a noun phrase) variable: As a rule, a N E N D score was assigned to the word immediately following the head noun unless there was no earlier phrase end indicator such as punctuation. If, for instance, the head noun was followed by a period, a N E N D score — and a SEND score, of course — was assigned to that particular noun. Nine of the 30 predictive variables proved to exert a significant influence on word reading time. Altogether, these nine variables accounted for 27 per cent of total variance; a value that is quite in line with the data reported by comparable studies (Kliegl — Olson — Davidson 1982; Haberlandt 1984). An optimal estimation of word reading time that includes all significant variables is given in the regression equation:

Reading

TIME e m p % TIME est = 0.37 + + + +

time

111

0.01 C H A R + 0.06 C R L F 0.09 SBEG + 0.27 SEND 0.40 C E N D + 0.04 WADJ 0.03 INTP - 0.06 INTC - 0.07 CORE

In summary, this equation says that word processing time in incremental reading increases when the word: 1. is composed of many characters (CHAR); 2. is the first one in a new line (CRLF); 3. is the first one of a new sentence (SBEG); 4. is the last one of a sentence (SEND); 5. occurs at the end of a coordinative cataphor (CEND); 6. is an adjective or an adverb (WADJ); and/or 7. serves to integrate many phrases (INTP). On the other hand, processing time decreases when the word: 8. serves to introduce a novel concept (INTC); and/or 9. refers to a concept mentioned earlier in the text (CORE). Disregarding trivialities like word length (CHAR) or line break (CRLF) effects which simply reflect the oculomotor activity during reading, we would like to point out two details. First, the most influential factors are not lexical ones but rather variables that imply some integrative operation. The end of a sentence, the end of coordinative cataphor and the number of phrases to be closed all go along with a delay in processing. The delay at the end of each sentence — aptly termed "sentence wrap-up" (Just — Carpenter 1980) — indicates that readers tend to take occasion to review or to rearrange the mental representation developed so far. The decelerating effect of the number of phrases to be integrated can be interpreted in a similar way. For the reader, working memory load increases with the number of open phrases, or, in a more mentalistic way of speaking, with the number of object nodes that stand alone, waiting to be integrated into a coherent conceptual structure. As this will often take place at the end of a sentence, this explanation is substantiated by the oberservation that the sentence wrap-up delay increases along with the number of concepts introduced in the sentence (Haberlandt et al. 1986). Second, there are two observations that appear to be incompatible: The regression equation suggests that both the introduction of a new concept (INTC) and reference to a known concept (CORE) go along with a decrease in processing time. In particular, the INTC effect runs contrary to intuition and to other empirical evidence (Just — Carpenter — Woolley 1982; Haberlandt 1984). A provisional explanation lies in the nature of our

112

Gert Rickheit,

Udo Günther, Lorenz

Sichelschmidt

materials. Texts described a walk across the University campus, a blocks array, or a pattern of colored spots. Separate analyses for each sort of text showed that the decrease in reading time associated with the first mention of a concept was an effect specific to the "campus walk" texts. As the participants in our study were well acquainted with the environment that was described, we cannot exclude that they may have anticipated some of the entities introduced in the text. Thus, what had been regarded as the introduction of a novel concept (INTC) may sometimes have been a case of reference to a concept that had already been activated (CORE). As to that, a decrease in reading time is consistent with the intuition that activation of a concept may facilitate subsequent referential access. On the other hand, a decrease in reading time is inconsistent with the intuition that referential access involves time-consuming cognitive operations. In face of this, coreferential processes in reading constitute an issue to be kept in mind for further investigation (Murphy 1984; G a r r o d - S a n f o r d 1985). By and large, the findings agreed with the assumptions outlined above; however, details will have to be worked out yet. Integrative effects at the phrase level, at the clause level, and at the sentence level indicate that readers strive for coherence in a variety of ways. Dynamic maintenance of the mental model, then, seems to be the process at the core of reading comprehension.

5. Stage 3: Reading-time profiles Mark Twain (1879: 277) on "the awful German language": "Some German words are so long that they have a perspective ... These things are not words, they are alphabetical processions."

From the results of the regression study it has become evident that word reading times to some extent reflect cognitive processes in reading. But though we now knew some of the influential variables, their psychological substrate was anything but clear. Therefore, in the third stage of our research program, we studied the latency profiles of selected verbal expressions. To begin with, we took a closer look at the reading time profiles within noun phrases. The texts that subjects had read in the regression study contained noun phrases with different degrees of complexity:

Reading time

113

1. Ν (... color ...); 2. DET-N (... the pattern ...); 3. DET-ADJ-N (... a yellow spot ...); 4. DET-ADJ-ADJ-N (... the red wooden block ...). We now compared the reading time profiles of these four noun phrase variants. To compensate for different word length, that part of variance which was due to word length was removed — or "partialed out", as we say. Technically speaking, the number of characters in a word was treated as a covariate. For illustrative purposes, average latencies have been adjusted to a standard word length of ten characters. Reading Time (s)

Syntactic Structure Figure 2. Reading-time profiles by noun-phrase structure

Reading time profiles show that readers allocated a comparatively large amount of processing capacity to the head of a noun phrase. Noticeably, this held only for complex noun phrases but not for simple ones. A significant intraphrase increase in reading time was observed with noun phrases of the DET-ADJ-N type (i (48) = 2.96; ρ < .05) and of the DET-ADJ-ADJ-N type (t( 12) = 4.49; ρ < .05). No increase in reading time, however, was observed with noun phrases of the DET-N type (t(79) = 1.83; ρ > .05). Besides, nouns took longer to read when they were constituents of a complex noun phrase than when they were constituents

114

Gert Rickheit,

Udo Günther, Lorenz

Sichelschmidt

of a simple noun phrase such as DET-N (/(57) = 2.29; ρ < .05) or Ν (?(71) = 2.22; ρ < .05). Obviously, the cognitive processes that are effective in the comprehension of complex noun phrases differ from those in the comprehension of simple noun phrases. This finding neatly fits our conception of human text processing: A simple noun phrase induces a reader to establish or to access a specific object node in his or her mental model. Processing of a complex noun phrase is more difficult because final interpretation of any prenominal modifiers has to be postponed until encountering the noun (Rommetveit 1974; Sichelschmidt 1989). Thus, when reading an adjective, the reader will mentally establish a relation link that must remain standing alone for the time being. Only upon reading the noun can the appropriate object node be identified and the stand-alone relation be linked to it. The noun, in this case, is the verbal element that serves the integrative function. Differences in cognitive processes between the comprehension of premodified noun phrases and the comprehension of unmodified noun phrases therefore should become manifest in reading times for the noun — and this was precisely what has been observed.

6. Noun phrase coordination Mark Twain (1879: 270) on "the awful German language": "Of course, then, the reader is left in a very exhausted and ignorant state."

Integrative processes in the comprehension of complex noun phrases can be made even more transparent by investigating reading time profiles for coordinated noun phrases. Texts used in the regression study contained several cases of coordinative expressions. These expressions were constructed from two complex nominal terms connected by the conjunction und (and). First and second nominal term always shared a common head noun. The head, however, was explicit in one of the terms only, so that either the head of the first term or the head of the second term was "empty". The coordinative expressions thus could be regarded as instances of nominal ellipsis (Klein 1981). We have examined two types of noun phrase coordination that were slightly different in surface structure: 1. Anaphoric expressions, with the elliptical term preceded by the explicit one (... ein roter Punkt und ein grüner ...). When analyzing the English

Reading time

115

translation (...a red spot and green [one] ...), it should be noted that the G e r m a n original does without an anaphoric one. 2. Cataphoric expressions, with the elliptical term preceding the explicit one (... ein roter und ein grüner Punkt...). In this case, the English translation (...a red [ ] and a green spot...) is literally equivalent to the G e r m a n original. We had learned f r o m the regression study that elliptical coordination exerts a significant influence on word reading times: The end-of-cataphor ( C E N D ) effect suggests that c a t a p h o r is a linguistic device that is particularly hard to comprehend. This was corroborated by significant differences in latency profiles between simple and complex n o u n phrases: Inevitable postponation of integrative processes until encountering the n o u n led to an increase in reading time. Effects due to postponation of integrative processes should be even more pronounced in the case of cataphoric coordination. When finally encountering the noun, the reader has to integrate not just one but two stand-alone relation links into a coherent mental representation. F r o m this, one would expect to find longer n o u n reading times with cataphoric expressions than with anaphoric expressions. Reading Time (s)

Syntactic Structure Figure 3. Reading-time profiles for coordinated noun phrases

This presumption was confirmed in the analyses of the latency profiles. There were no differences between anaphoric and cataphoric expressions

116

Gert Rickheit,

Udo Günther, Lorenz

Sichelschmidt

except one: Readers spent about half a second longer viewing the noun when it occurred in the second term than when it occurred in the first term of a coordinative expression (t( 12) = 2.77; ρ < .05). Obviously, readers have no trouble at all processing an anaphoric expression. It seems that a concept, once activated, remains so for some length of time. If an appropriate object node has been established during reading the first term of a coordinative expression, it can be easily accessed again when reading the second term. In the cataphoric case, however, no referent is specified in the first of two coordinated terms. So readers have to establish the appropriate conceptual relations subsequently, during processing of the second term. Prolongation of noun reading times in the cataphoric case thus may be attributed to the complex cognitive processes required to resolve that type of coordinative expressions (Dell — McKoon — Ratcliff 1983; Aaronson — Ferres 1984). We must be cautious, though. In the texts, the end of cataphoric coordination often coincided with the end of a sentence. A decrease in reading speed thus can be attributed to either. What is required, therefore, is experimentation with stricter control. We shall return to this issue after a brief look at the processing of verb phrases.

7. Verb position, or: The making of a proposition Mark Twain (1879: 282) on "the awful German language": "This important part of speech should be brought forward to a position where it may easily be seen with the naked eye."

Unlike English, German offers a variety of options for arranging a sentence. Aside from subject — verb — object expressions, readers of German will frequently encounter verb — subject — object and subject —object—verb structures (Uszkoreit 1987). At the expense of using materials which do not readily translate into English, we have taken advantage of the flexibility of German word order to investigate the role of the predicate in the constitution of a proposition. Bearing in mind Mark Twain's recommendation as well as the results we had obtained with noun phrases, we expected verb fronting to faciliate comprehension. Theoretical with arguments are furnished by verb valency and case role conceptions: A three-place verb, for instance, provides three case-role slots reserved for, say, an agent, an experiencer, and a destination

Reading

time

117

entity. In case of a verb-first expression, the verb will induce the reader to mentally establish a relation link between three objects yet to be specified. On reading subsequent noun phrases, the appropriate object nodes can be tied to that relation link right away. In case of a verb-last expression, though, the object nodes will have to remain standing alone because a relation link is not available yet. The information necessary to integrate the object nodes into a coherent conceptual structure is supplied only when reading the verb. Hence, verb reading times should be much shorter with verb-first expressions than with verb-end expressions, whereas noun reading times should be slightly prolonged. This hypothesis was tested in an experiment on clause processing. Forty-eight college students who were German native speakers volunteered for the experiment. They were given 16 three-sentence texts, each about 40 words in length. Here is a sample text, together with an approximate English translation: Inzwischen ist der März fast vorüber. Es ist wärmer geworden, und so treiben die Bauern das Vieh auf die Weide und pflügen die Äcker. Bald werden die Obstbäume zu blühen beginnen. March is almost past now. There has been a rise in temperature, and so the farmers take the cattle to the pastures and begin plowing. Soon the fruit trees will begin to bloom. From each text, we constructed two versions that were virtually identical except for the position of the predicate of the second sentence: 1. Verb-first in German (... und so treiben die Bauern das Vieh auf die Weide und ...); 2. Verb-last in German (... so daß die Bauern das Vieh auf die Weide treiben und ...J. Unfortunately, when translated into English (... and so the farmers take the cattle to the pastures and ...J, the syntactic difference between German verb-first and verb-last versions is lost. Texts were displayed on a monochrome computer screen (Eizo 4051). Each subject was confronted with all experimental texts and with both treatment conditions. The attachment of treatments to texts was rotated between groups. Thus, each subject read eight verb-first texts and eight verb-last texts, together with several distractor texts of a different structure. To ensure adequate comprehension of each text, subjects had to answer a question on its contents before proceeding to the next one. For every text and every subject, word reading times were recorded with

118

Gert Rickheit,

Udo Günther, Lorenz

Sichelschmidt

millisecond accuracy (Compaq Deskpro 386/20). Reading times were statistically evaluated for a fragment of the second sentence. Any latencies beyond the limit of two standard deviations above or below the cell mean were considered as outliers; they were replaced with the upper or lower cell limit, respectively. Data were subjected to two analyses of variance, the first one (Fs) with subjects and the second one (Fm) with materials as a random factor. R e a d i n g T i m e (s)

Syntactic Structure Figure 4. Reading-time profiles for different verb positions

The average reading time for the critical sentence fragment was 0.506 seconds per word. Analyses of variance showed that verb-first and verb-last sentences differed with regard to reading times for the verb (V) and for the last-mentioned noun (N3). The verb took longer to read when occurring in a verb-last version than when occurring in a verb-first version (F s (l,47) = 12.08; F m (l,14) = 17.80; ρ < .005). On the other hand, the final noun took longer to read when occurring in a verb-first expression than when occurring in a verb-last expression (Fs(1,47) = 19.00; F m (l,14) = 17.28; ρ < .005). No other effects were significant.

Reading

time

119

By and large, these findings are consistent with the results obtained in earlier experiments on noun phrase processing. As predicted, verb reading times were substantially longer with verb-last expressions than with verbfirst expressions. Supplying readers with the predicate at an early point during reading, then, may indeed facilitate comprehension. However, it would be premature to conclude that clause integration processes entirely depend on the predicate. -Rather, latency profiles indicate that reading the arguments prior to the predicate increases working memory load: The reader has to keep in mind several stand-alone object nodes until finally reading the verb that permits integration. So far, the expectations have been substantiated empirically. Contrary to the predictions, however, verb fronting did not lead to a general increase in noun reading times. So there was no empirical support for the idea that tying an object node to a relation link takes longer than merely installing it as a stand-alone. Still, the latency differences — though insignificant in size — were always in the predicted direction. It remains a question, therefore, whether the assignment of object nodes to caserole slots is too subtle a cognitive process to be caught by word incremental reading, or whether the effects of constituent order have been obscured by individual differences in the strategies people employ in comprehending clauses. In any case, further research is warranted in the face of the present state of findings on the mental integration of concepts during reading. Nevertheless, a significant decelerating effect of verb fronting has been observed at the last-mentioned noun (N3). This effect should perhaps be attributed to the fact that in the case of verb fronting, the last-mentioned noun occurred at the end of the clause. This suggests another mechanism to be operative in clause comprehension. To a certain extent, a wait-and-see strategy might be appropriate with languages as flexible in word order as German. Readers may want to postpone integrative processes until attaining some semantic equilibrium. Semantic proportions are saturated if there are no longer any object nodes or relation links left standing alone, that is, if nodes and links are entirely interconnected. Only then can readers reasonably carry out any sentence wrap-up operations. In consequence, integrative processes in comprehension are contingent on reading the last element of a clause or sentence. This should become manifest in prolonged reading times for the final element of an expression — which was exactly the pattern of results found in the experiment.

120

Gert Rickheit,

Udo Günther, Lorenz

Sichelschmidt

8. Verb phrase coordination Mark Twain (1879: 272) on "the awful German language": "The inventor of the language seems to have taken pleasure in complicating it in every way he could think of."

In the last experiment of this series we returned to coordinative expressions, studying the processing of a slightly different type of ellipsis. Again, the expressions under study were constructed from two terms conjoined by und (and). This time, however, it was not the head noun but the verb that was explicit in one term but "empty" in the other. According to linguistic theory (Sag — Hankamer 1984), ellipsis is a surface phenomenon: An elliptical gap must always have some specific verbal referent that must be syntactically consistent with the "empty" portion of the elliptical term. This is to enable the reader to mentally substitute the "empty" portion with a copy of the referent from the explicit term. A reader encountering a gap in place of a verb, for instance, is thought to search the accompanying text for a suitable predicate to fill that gap (Murphy 1985). In analogy to the findings on the processing of nominal ellipses, we expected the verbs in an elliptical cataphor to be particularly difficult. In case of cataphor, final processing of the first term has to be postponed until reading the verb in the second term. As this implies an increase in working memory load, reading times for second term verbs should be longer than reading times for first term verbs. Besides, we expected verb reading time to be influenced by syntactic-semantic parallelism of terms. If the second term of a coordinative expression does not match the first term with regard to syntactic-semantic structure, the consistency constraint is not met. Resolution of elliptical expressions thus should be rendered difficult or even impossible because the "empty" verb cannot be substituted with the corresponding constituent from the explicit term. For an empirical approach to the processing of verb ellipsis, we used three-sentence texts of about 40 words in length. A sample text and a rough English translation: Dagmar macht Pläne für ihren Sommerurlaub im Süden. Wenn Ute in die Alpen fährt und Anne an die Ostsee fährt wie im letzten Jahr, würde sie sich eine Eisenbahnfahrkarte kaufen. 'Roma Termini' würde darauf stehen. Sheila is planning her summer vacations trip to the sough. If Rita went to the Alps and Anne went to the Baltic Sea like the year before, Sheila would buy a railway ticket. 'Roma Termini' would be the destination.

Reading time

121

Two factors were varied orthogonally to produce six variants of the second sentence of each text. The first variade was the type of elliptical construction: 1.1. Unelliptical: The verb was explicit in the first and in the second term (... Wenn Ute in die Alpen fährt und Anne an die Ostsee fährt wie ...). The English translation (... If Rita went to the Alps and Anne went to the Baltic Sea like ...) preserves the structure of the German original. 1.2. Elliptical anaphor: The verb was explicit in the first term but "empty" in the second term (... Wenn Ute in die Alpen fährt und Anne an die Ostsee wie ...). The English translation (...If Rita went to the Alps and Anne f J to the Baltic Sea like ...) contains an anaphoric gap similar to the German original. 1.3. Elliptical cataphor: The verb was "empty" in the first term but explicit in the second term (... Wenn Ute in die Alpen und Anne an die Ostsee fährt wie ...). Though a literal English translation is ungrammatical (... If Rita [J to the Alps and Anne went to the Baltic Sea like ...), the German original is not. In fact, elliptical expressions of this sort are by no means uncommon in German. The second variable which we manipulated was the parallelism of terms: 2.1. Case match: First and second term did match with regard to cases (... Wenn Ute in die Alpen fährt und Anne an die Ostsee fährt wie ...). This is also true for the English translation (If Rita went to the Alps and Anne went to the Baltic Sea like ...). 2.2. Case mismatch: First and second term did not match with regard to cases (... Wenn Ute mit dem Auto fährt und Anne an die Ostsee fährt wie ...). This is also true for the English translation (... If Rita went by car and Anne went to the Baltic Sea like ...). Twenty-three female and twenty-five male German native speakers participated in the experiment. Aside from several distractor texts, 48 experimental texts had to be read incrementally, word by word. Experimental variation was "within subjects", so every subject was given eight texts each under any of the six treatment conditions. Assignment of treatments to texts was controlled by rotation. After each text, subjects were asked a question about its contents to ensure adequate comprehension. For every text and every subject, word reading latencies for a fragment of the second sentence were evaluated (Compaq Deskpro 386/20 with Eizo 4051). Latency data were analyzed in the same way as described above.

122

Gert Rickheit, Udo Günther, Lorenz

Sichelschmidt

Verb Reading Time (β) 1

0,9 -

0,8

T y p e of Ellipsis •

Unelliptlcal

V

Anaphoric

Δ

Cataphoric

Δ

Δ

0,7

0.6 0,5

Terml Term2 Case Match

Terml Term2 Case Mismatch

Figure 5. Verb reading times by term parallelism and ellipsis type

The average word reading time was 0.555 seconds; a value close to the one found in earlier experimentation. Analyses of variance yielded significant effects for the verb (V) and for the preposition (Prep). As hypotheses specifically pertain to verb reading times, we will only report these latencies here. First, verbs were generally read faster when occurring in an unelliptical structure than when occuring in an elliptical anaphor or cataphor (F s (l,47) = 5.54; F m (l,47) = 5.83; ρ < .05); on closer inspection, however, this was true of second term verbs only (7^(1,47) = 5.52; Fm (1,47) = 6.51; ρ < .05). Second, verbs were generally read faster at case match than at case mismatch (F s (l,47) = 6.16; Fm{ 1,47) = 6.14; ρ < .05); again, this applied to second term verbs only (F s (l,47) = 8.51; .Fm(l,47) = 10.18; ρ < .01). Third, verbs were read faster when occurring in the first term than when occurring in the second term (F s (l,47) = 28.09; Fm (1,47) = 120.02; ρ < .001). Fourth, though this was true at case match (F,(l,47) = 15.01; Fm{1,47) = 49.75; ρ < .001) as well as at case mismatch (Fs (1,47) = 29.01; Fm (1,47) = 78.62; ρ < .001), the effect was more pronounced in the case mismatch condition than in the case match condition (Fs(1,47) = 9.80; Fm{\,Al) = 13.72; ρ < .005). Fifth, the finding that reading times were generally faster for first term verbs

Reading time

123

than for second term verbs was true for elliptical expressions (i^ (1,47) = 23.56; Fm (1,47) = 73.88; ρ < .001) as well as for unelliptical coordination (F s (l,47) = 26.12; Fm(\,Al) = 68.27; ρ < .001); however, the effect was more pronounced in the comparison of elliptical anaphor to elliptical cataphor (F s (l,47) = 4.79; Fm (1,47) = 5.49; ρ < .05). No other effects were significant in the analyses of verb reading times. The major result of the present experiment corresponds to that obtained with noun phrase ellipses. People take longer to process the verb when it is part of an elliptical cataphor than when it is part of an anaphor. Once again, reading times have shown the cognitive resolution of a cataphoric expression to be a troublesome business (Dell — McKoon — Ratcliff 1983; Aaronson - Ferres 1984). Elliptical cataphor requires the reader to keep in mind several object nodes or relation links that cannot be integrated into a coherent conceptual structure until the appropriate verbal information is encountered. So retroactive integration processes are required, which cause a delay in reading time. Moreover, latency data have shown parallelism of first and second term to influence reading times for second term verbs. Partly, this is a truism, because readers have to advance to the second term in order to detect any similarities to the first one at all. However, it is also of consequence because reading time averages were as predicted: For unelliptical coordination, comprehension was rendered more difficult by a mismatch in case roles between first and second term. At that, case role mismatch was no less an obstacle to comprehension than was cataphor in case of ellipsis. With this finding, some recent approaches emphasizing the role of coordination in the development of a coherent mental representation have received further empirical support (Frazier et al. 1984; Gernsbacher — Hargreaves — Beeman 1989). These approaches assume that during the reading of the first of two parallel terms, a structural foundation is laid for the processing of the second term. In the present experiment, this could have facilitated comprehension in the case of case match. In the case of case mismatch, however, the structural groundwork developed during the reading of the first term would be dismantled and replaced by an independent structure. After all, pointing at the relevance of coordination is perhaps the most obvious way to summarize the outcome of the present experiment. Second-term verbs were read faster than first-term verbs. For elliptical expressions, this replicates the findings of earlier experiments: Elliptical cataphors are more complicated than elliptical anaphors because the integrative element — the verb, in this case — is introduced late. For

124

Gert Rickheit, Udo Günther, Lorenz

Sichelschmidt

unelliptical expressions, this emphasizes an aspect of coherence that has received little attention to date: the role of conjunctions and other coordinating devices in the development and maintenance of a reader's conceptual representation (Kindt 1985; Gernsbacher — Hargreaves — Beeman 1989).

9. Concluding remarks Mark Twain (1879: 267) on "the awful German language": "A person who has not studied German can form no idea of what a perplexing language it is."

Despite the perplexing peculiarities of German, word incremental reading has turned out to be a convenient and effective method for investigating the cognitive processes that take place when reading and comprehending written text. The analysis of reading time profiles for simple and complex expressions like noun phrases, noun phrase coordination, clauses, and clause coordination has furnished empirical support for the psycholinguistic processing principle behind the modeling assumption, the progressive assumption, the latency assumption, and the immediacy assumption (Johnson-Laird 1983; Hörmann 1986): To read and to comprehend written text means to construct and to maintain a coherent mental representation of the state of affairs conveyed by that text. With that, word incremental reading opens perspectives for further experimental research. As emphasized above, the investigation of those structural characteristics of a text that enable or induce the reader to perform conceptual integration processes appears a promising enterprise. Comprehension of conjunction or disjunction, of singular or plural pronouns, and of negatives or quantification are among the issues that deserve special consideration (Müsseier — Rickheit 1990; Sichelschmidt — Günther 1990). Investigating subtleties like these will be a first step toward modeling reading time procedurally — on the basis of cognitive operations rather than on the basis of structural properties. In effect, German may prove a language less perplexing, less irrational, and less slipshod than pictured by the witty "tramp abroad". Rather, with methods like word incremental reading, the investigation of human language processing in general — of German, in particular — may turn out to be one of the great scientific adventures of our time.

Reading time

125

Acknowledgments The research reported here was supported by grants from the German Science Foundation (DFG RI-314-8). Thanks are due to KoLiBri, the Bielefeld research group on coherence, for many intense discussions of comprehension and coordination issues, and to Jonathan Harrow and Hans Strohner for helpful comments on an earlier draft of this paper. Requests for reprints should be addressed to any author at the Department of Linguistics, University of Bielefeld, D-4800 Bielefeld 1, FRG.

References Aaronson Doris — Steven Ferres 1984 "The word-by-word reading paradigm: An experimental approach", in: David E. Kieras — Marcel A. Just (eds.), New methods in reading comprehension research. Hillsdale: Erlbaum. 31—68. Askwall Susanne 1985 "Computer supported reading versus reading text on paper: A comparison of two reading situations", International Journal of Man-Machine Studies, 22: 4 2 5 - 4 3 9 . Beveridge Michael — Susan Edmundson 1989 "Reading strategies of good and poor readers in word and phrase presentation", Journal of Research in Reading, 12: 1 — 12. Dell Gary S . - G a i l M c K o o n - Roger Ratcliff 1983 "The activation of antecedent information during the processing of anaphoric reference in reading", Journal of Verbal Learning and Verbal Behavior, 22: 7 5 - 8 7 . Frazier Lyn — Lori Taft —Tom Roeper—Charles Clifton —Kate Ehrlich 1984 "Parallel structure: A source of facilitation in sentence comprehension", Memory and Cognition, 12: 421 —430. Garnham Alan 1987 Mental models as representations of discourse and text. Chichester: Horwood. Garrod Simon C. —Anthony J. —Sanford 1985 "On the real-time character of interpretation during reading", Language and Cognitive Processes, 1: 43 — 59. Gernsbacher Morton Α. —David J. Hargreaves — Mark Beeman 1989 "Building and accessing clausal representations: The advantage of first mention versus the advantage of clause recency", Journal of Memory and Language, 28: 735 — 755. Graesser Arthur C. — James R. Riha 1984 "An application of multiple regression techniques to sentence reading times", in: David E. Kieras — Marcel A. Just (eds.), New methods in reading comprehension research. Hillsdale: Erlbaum. 183 — 218. Günther Udo 1989 "Lesen im Experiment " [Experimentation in Reading], Linguistische Berichte, 122: 2 8 3 - 3 2 0 .

126

Gert Rickheit, Udo Günther, Lorenz Sichelschmidt

Haberlandt Karl F. 1984 "Components of sentence and word reading times", in: David E. Kieras— Marcel A. Just (eds.), New methods in reading comprehension research. Hillsdale: Erlbaum. 219-251. Haberlandt Karl F.—Arthur C. Graesser—Nancy J. Schneider—Judith Kiely 1986 "Effects of task and new arguments on word reading times", Journal of Memory and Language, 25: 314 — 322. Hörmann Hans 1986 Meaning and context: An introduction to the psychology of language. New York: Plenum Press. Johnson-Laird Philip N. 1983 Mental models. Toward a cognitive science of language, inference, and consciousness. Cambridge: Cambridge University Press. Juola James F. 1988 "The use of computer displays to improve reading comprehension", Applied Cognitive Psychology, 2: 87 — 95. Just Marcel A. — Patricia A. Carpenter 1980 "A theory of reading: From eye fixations to comprehension", Psychological Review, 87: 329-354. 1987 The psychology of reading and language comprehension. Boston: Allyn and Bacon. Just Marcel Α.—Patricia A. Carpenter—Jacqueline D. Woolley 1982 "Paradigms and processes in reading comprehension", Journal of Experimental Psychology (General), 111: 228-238. Kennedy Alan—Wayne S. Murray 1984 "Inspection times for words in syntactically ambiguous sentences under three presentation conditions", Journal of Experimental Psychology (Human Perception and Performance), 10: 833 — 849. Kindt Walther 1985 "Grammatische Prinzipien sogenannter Ellipsen und ein neues Syntaxmodell" [Grammatical principles of so-called ellipses, and a novel approach to syntax], in: Reinhard Meyer-Hermann — Hannes Rieser (eds.), Ellipsen und fragmentarische Ausdrücke (Band 1). Tübingen: Niemeyer. 161-290. Klein Wolfgang 1981 "Some rules of regular ellipsis in German", in: Wolfgang Klein —Willem J. M. Levelt (eds.), Crossing the boundaries in linguistics. Dordrecht: Reidel. 5 1 - 7 8 . Kliegl Reinhold—Richard J. Olson —Brian J. Davidson 1982 "Regression analysis as a tool for studying reading processes: A comment on Just and Carpenter's eye fixation theory", Memory and Cognition, 10: 287-296. Murphy Gregory L. 1984 "Establishing and accessing referents in discourse", Memory and Cognition, 12: 489-497. 1985 "Processes of understanding anaphora", Journal of Memory and Language, 24: 290-303.

Reading time

127

Müsseier Jochen —Gert Rickheit 1990 "Komplexbildung in der Textverarbeitung: Die kognitive Auflösung pluraler Pronomen" [Compound concepts in text processing: The cognitive resolution of plural pronouns], Zeitschrift fur Psychology, 198, 69 — 81 (in press). O'Regan J. Kevin —Ariane Levy-Schoen (eds.) 1987 Eye movements. From physiology to cognition. Amsterdam: North-Holland. Rayner Keith (ed.) 1983 Eye movements in reading. New York: Academic Press. Reinking David 1988 "Computer-mediated text and comprehension differences: The role of reading time, reader preference, and estimation of learning", Reading Research Quarterly, 23: 4 8 4 - 4 9 8 . Rickheit Gert — Hans Strohner 1985 "Psycholinguistik der Textverarbeitung,, [Psycholinguistic approaches to text processing], Studium Linguistik, 17/18: 1 —78. Rommetveit Ragnar 1974 On message structure: A framework for the study of language and communication. London: Wiley. Sag Ivan Α.—Jorge Hankamer 1984 "Toward a theory of anaphoric processing", Linguistics and Philosophy, 7: 3 2 5 - 3 4 5 . Sanford Anthony J. — Simon C. Garrod 1981 Understanding written language. Explorations of comprehension beyond the sentence. London: Wiley. Sichelschmidt Lorenz 1989 Adjektivfolgen. Eine Untersuchung zum Verstehen komplexer Nominalphrasen [Adjective order. Α study in the comprehension of complex noun phrases], Opladen: Westdeutscher Verlag. Sichelschmidt Lorenz —Udo Günther 1990 "Interpreting anaphoric relations during reading: Inspection time evidence". Journal of Semantics, 7: 321 — 345. Thibadeau Robert —Marcel A. Just — Patricia A. Carpenter 1982 "A model of the time course and content of reading", Cognitive Science, 6: 1 5 7 - 2 0 3 . Twain, Mark 1879 "The awful German language", in: M.Twain, A tramp abroad. Volume 2 (Stormfield Edition, 1929). New York: Harper & Brothers. 2 6 7 - 2 8 4 . Uszkoreit Hans 1987 Word order and constituent structure in German. Chicago: Chicago University Press.

Levels of language comprehension and systems of information-processing Johannes Engelkamp

For a long time psychologists directed their attention almost exclusively to the sequence of steps in information-processing. The stages of information-processing were analyzed in their chronological succession, from input to the output of overt behavior. The individual stages of information-processing were largely analyzed independent from stimulus or task. This viewpoint is beginning to change. Increasingly more researchers are of the opinion that human information-processing takes different paths from input to behavioral output, and that information-processing is performed by different sub-systems, depending upon type of stimulus and goal. From this perspective there arises, first of all, the question what are the most important global sub-systems that form the basis of human information-processing, and secondly, two questions especially crucial for linguistics, of whether there is a specific system for the processing of verbal information, and if so, how this system is structured. I shall deal here mainly with the over all framework of human information-processing and less with internal differentiation. In other words, it is not so much the question of which sub-systems play a role in the processing of verbal information, but rather primarily the question of whether a system for verbal information processing can be separated from other systems, for example, from those for the processing of nonverbal information or for the processing of meaning. To clarify this problem I shall first describe the past developments in psychological research on text comprehension. It will be seen that, regarding text comprehension, three levels are differentiated that can be considered the results of the processing of qualitatively different kinds of information. At the same time'it will become clear that the differentiation between these three levels remains vague, and their behavioral indicators are still unsatisfactory. For this reason, in the second part of my presentation I shall deal with research on verbal and imagery processing which shows that different systems, working partially independently of each other, are the basis for information-processing. As I shall show, this model permits not only a convincing separation into different systems, but also a more definite determination of the types of information

130

Johannes

Engelkamp

processed by the sub-systems. For this reason it is desirable to extend the methods of this perspective to the research on text comprehension. In the third part of my presentation I shall demonstrate the usefulness of this method for the internal differentiation of a verbal system, using the example of research on word-processing.

Language Comprehension I would like to begin with a short look at the psychologist's interest in recognition. A crucial element of this psychological interest is human behavior. This is usually understood as performance. Two kinds of performance that interest us here are language comprehension and memory. From an applied view, the primary interest is in how such performance can be improved. From a basic view, the main interest is rather in how this performance is accomplished. How does one have to conceive of the system "human being" for it to be able to achieve such performance (or, of course, in order to display such failures as can be observed)? An important determinant of performance is usually considered to be the stimulus or, more generally, the situation leading to performance. One line of research is, for example, in how performance varies with changes in the situation. As regards performance, fundamentally the quality and the timing are examined. Concerning quality, the interest is in how suitable or correct a performance is. In principle, an explanation is sought for not only how correct or incorrect performance is produced, but also why certain behavior displays certain characteristics of timing (see Engelkamp, 1991). Consequently, in examining comprehension performance, the question arises for psychologists of how this performance is to be defined in order to be able to examine what it depends upon and how it is accomplished. A first answer to this question was offered in the early 1970s, as psychologists began to construct a model for human semantic knowledge. In constructing a model for this kind of knowledge, they confined themselves mainly to knowledge about the physical aspects of the empirical world, that is, knowledge usually expressed descriptively and verbally in non-fictional texts. The following depiction deals only with such types of knowledge and with the descriptive function of language. Such types of knowledge have been represented, among others, in the form of networks or propositions (Lindsay — Norman, 1977). What was more obvious, seen in this context, than to conceive of comprehension as the construction of segments of this networks (cf., Anderson, 1985) or as the construction of a series of propositions (cf.,

Levels of language comprehension

131

Kintsch, 1974). This idea won in plausibility through the self-observation that, having read a book or an article, we do not remember the text word-for-word, but only remember the sense of what we read, in short, that which was believed to be represented in networks or propositions. Since basically the same aspects of knowledge were represented in networks and in propositions, I shall hitherto restrict myself to the idea that comprehension means the construction of propositions. Taking as example an experiment by Kintsch — Keenan (1973), I shall try to show, paradigmatically, how it was attempted to verify this concept. Kintsch — Keenan constructed texts that were differentiated by the number of underlying propositions. They had subjects read these texts under a subjective criterion of comprehension, and measured their reading-time. They also had the subjects reproduce these texts, and counted the number of correctly-remembered propositions. Both modes of measurement — reading-time as well as memory performance — were dependent upon the number of propositions underlying a text. At the same time, the findings from other experiments gave the impression that surface information of texts was forgotten immediately after reading them. Thus, Fillenbaum (1966) and Sachs (1967), in their experiments on text comprehension, observed that syntactic-stylistic changes of sentences were no longer recognized only 30 seconds later, whereas semantic changes were perfectly recognized even after quite long intervals. These and similar findings led to the following interpretations: (a) Comprehension is the extraction of propositions from texts. (b) The text surface is only something like a vehicle for this process. (c) For this reason the text-surface is quickly forgotten, and only the propositional meaning is retained. It is clear that comprehension performance, understood here as a specific type of memory performance, is memory performance for propositions. These propositions are, simultaneously, the textual characteristics that determine comprehension performance. Other textual characteristics, such as text-length measured by words, were seen as non-determining for comprehension performance. This concept of language comprehension was modified in the 80s mainly on the basis of two points of criticism. One point of criticism concerned the assumption that surface information serves only as the vehicle for meaning and is immediately forgotten. It became evident that surface information of texts is also retained. It was shown that the findings implying an immediate forgetting of verbal

132

Johannes

Engelkamp

surface information was based upon a certain method of measuring memory. If one did not test recognition by presenting subjects with different alternatives simultaneously, but by giving them the original sentences and the variations successively, with the instructions to decide if the respective sentence occurred in the text or not, and if one also took false alarms into account in addition to the hits, then it could be observed that the exact wording of sentences was recognized better than chance, even days later (cf., Masson, 1984; Perrig, 1988). By hits were meant the correct identifications of "old" sentences, and false alarms were the incorrect acceptance of changed, that is, "new" sentences. This point of criticism led to the differentiation of two types of performance as indicators for comprehension: the performance of recognizing exact wording of text segments, and the performance of remembering propositions. A second point of criticism arose soon after. Johnson-Laird — Herrmann — Chafin (1984) criticized the idea that language comprehension consisted of transforming sentences into propositions; with this idea, they said, it was overlooked that utterances of language often referred to the empirical world or to fictive situations. The assumption of only a knowledge of concepts, as imbedded in propositions, ignored the fact that the listener or reader establishes a frame of reference to what is referred to in the text, and also ignored how he or she does this. Consequently, Johnson-Laird (1983) insisted that the psychological models of language comprehension be based upon the assumption that language can refer to the world only with the help of mental or situational models that we possess concerning the conditions in our world. This means that there is a requirement for situational models in addition to propositional meaning. A situational model for a text has to represent the segment of the world that the text refers to. It is decisive here that such a situational model represent the segment of the world directly or analogously. This means, that the same operations can be mentally applied directly to a situational model (that is, without transformation) that are applied to the real segment of the world. The assumption of operations that are equivalent in structure implies that the mental behavior of a person acting upon a situational model is expected to depend upon the same characteristics as does the physical behavior within the world segment. I wish to make these general considerations more concrete with the help of an example. A typical characteristic of the physical world-segment is the dimension of space. Objects are situated in space in the empirical

Levels of language comprehension

133

world, and between them is a spatial distance. Certain behavior patterns concerning these objects often depend directly upon this spatial distance. For example, the time it takes to go from one object to another or the time it takes to glance from one object to another is a function of spatial distance. Correspondingly, it is required in a situational model that certain behavior patterns concerning objects depend upon the spatial distance of the objects. One way among others that this is shown is through so-called "spatial priming". For example, if a person is shown a room containing various objects, and is asked later if object X was in the room, and, immediately following, if a second object Y was in the room, then it can be seen that the reaction for the second object is a function of the spatial distance from the first object. This dependence of reaction-time upon the spatial distance between the second object and the first object is called "spatial priming". If a text is also, by way of spatial relationships between objects, supposed to lead to an analogous representation of spatial relationships, then corresponding spatial priming-effects should be observable here also. This question was examined by Denis — Zimmer (1991). They constructed two systematic spatial arrangements of stimuli which they called "Tanaland" and "Ekland". Both arrangements consisted of 3 χ 7 cells in the form of a matrix. In both arrangements, fifteen of these cells contained objects. The two graphics were given directly to a group with the instructions to memorize the spatial distribution of the objects. A second group was to memorize the same arrangements of the basis of description. The description for Tanaland, for example, began as follows. In the extreme north-west there is a hunter's look-out. East of this look-out cattle are grazing. East of the cows is a lake. South of the lake stands a wind-mill. West of the wind-mill there is a vineyard.

After the subjects had memorized the spatial distribution of the objects in both countries, they were presented with the names of various objects, one after the other, on a terminal screen. The objects were either from Tanaland, from Ekland, or they were "new". The subjects were to indicate their decision of whether each object was familiar or new by pressing a terminal key. Their reaction times were recorded. Four relationships were possible between any two objects presented one after the other. First of all, the second object could be from another country than the first one. Second, both objects had been separated in the matrix by at least two rows. Third, the two objects had been next to each other in the matrix. This adjacency-condition had two variations for the group working from a text. Under adjacency-condition 1, the two

134

Johannes

Engelkamp

objects were adjacent and were mentioned in the same sentence (for example, "South of the lake stands a wind-mill"). Under adjacencycondition 2, the objects were spatially adjacent, but they were mentioned in the same sentence (for example, "cows" and "vineyard"). With the aid of the variations from adjacency-conditions 1 and 2, it was to be tested if, in addition to the spatial distance, the mentioning of two objects in one proposition would effect the reaction times. According to the results, spatial priming was observable not only for the persons who saw the graphics, but also for those who heard the descriptions. For both groups, objects under the adjacency-condition were decided upon faster than objects separated by two or more rows, and these were decided upon faster than objects from different countries. This leads to the conclusion that the text-group also had access to a spatial situational model, and that their decisions were based upon it. Consequently, the number has now increased for the different kinds of performance that must be seen as indicators for comprehension. At the same time, our concept of comprehension has become more tangible. Through the assumption that descriptive texts serve to refer to segments of the world, a clear goal has been set for comprehension: the construction of a situational model. It is plausible to assume that the representations of text-surfaces and propositional meaning are steps on the way towards construction of such a situational model. Corresponding to the same interpretation, Perrig (1988: 141) states: The comprehension of texts necessitates the involvement of a number of distinguishable processes. In addition to processes of word-identification and sentence-analysis, one must assume processing-mechanisms that serve the extraction of meaning ... and the mental representation of so-called situational models ... It can be assumed that they can be assigned to different processes of qualitatively distinguishable niveaus [and] that these processes leave behind distinguishable traces in memory.

In line with this conception, there is a recent tendency towards differentiating three levels of representation for comprehension (e. g., Van Dijk - Kintsch, 1983): 1. the representation of the surface of language, 2. the representation of the propositional basis of the text, and 3. the representation of the so-called situational model. It can be added that the recognition performance for the wording of the text can be seen as an indicator for the representation of the first level, and free recall as an indicator for the propositional basis of the text. For the situational model, different indicators are conceivable, depending upon the world-segment represented, and depending upon the information of the segment, that is being tested.

Levels of language comprehension

135

That the types of performance defined in this way are based upon different types of information, can be surmised from an experiment by Perrig — Kintsch (1985), who compared these types of performance on the basis of two texts that differed only slightly. The texts dealt with the description of a hypothetical town in the mid-west of the U. S. This town was described once by way of a route-text, and in a second version by way of a map-text. With the route-text, the town was described spatially from the perspective of a car-driver. The map-text gave a geographical description in terms of north, south, east and west. These two texts were presented for reading, sentence by sentence, at a set rate of presentation. The authors tested the recognition of the surface structure, the recall of the propositions underlying the text, and the hypothetical model. In order to test the representation relating to the situational model, they offered the subjects sentences with information concerning locality. These sentences contained inferred information about locality that were correct or incorrect. This information was formulated in either the route-text or the map-text (for example, "The church is to the left of the highway" [route-text] or "The church is north of the highway" [map-text]). It was assumed that the inferred information of locality, that was formulated congruent to the situational model, would be more easily accessible than that formulated incongruently. They observed no differences between the two texts for the recognition performance pertaining to surface information. However in free reproduction the route-text was recalled better than the map-text. For the route-text, around 10 per cent more propositions were correctly recalled. For the verification of correctly inferred locality-information, an interaction was observed between the text version and the congruency of the formulation by the test. Whereas congruently-formulated and incongruently-formulated inferences about locality were both equally well verified on the basis of the route-text, for the map-text the congruently-formulated information was verified distinctly better than incongruently-formulated information. This experiment verifies once more that the different types of comprehension-performance are based upon different types of information, since the performance for both text versions influenced in different ways, and that the differentiation between the three levels makes sense, because it leads one to look for certain types of performance and not to test for arbitrary behavioral aspects. That we are dealing with levels of comprehension that are progressed through one after the other is a further assumption that is associated with the concept of levels. I shall try to present evidence for this assumption, too.

136

Johannes

Engelkamp

When the comprehension of texts leaves the level of text surface to go on to the propositional basis of the text, it is to be expected that changes of meaning be noticed less in sentences whose propositional meaning is more difficult to extract than in sentences where it is easier to extract. It is to be expected that surface changes, however, can be recognized equally well for both types of sentences. It has long been recognized that abstract sentences are more difficult to understand than concrete ones. This means that propositional representation can be expected to be accomplished more quickly for concrete sentences than for abstract ones. Abstract sentences will tend to be represented only superficially when the processing-time is limited. If one tests recognition of sentences that have been changed in meaning, then the performance for concrete sentences should be better than for abstract sentences. The recognition of surface changes, however, should be as good for abstract sentences as for concrete sentences. Begg — Palvio (1969) investigated this hypothesis. They presented their subjects with concrete and abstract sentences, with the same length of reading-time for both. Later they gave the subjects changed sentences for recognition. For both concrete and abstract sentences, it was possible to have changes in surface structure as well as in propositional meaning, and the recognition of the sentence was tested. The sentence "The loving mother served an excellent familiy," for example, was changed to "The loving mother served an excellent houshold" (surface change), or to "The loving family served an excellent mother" (change in meaning). Surface changes were recognized better in abstract sentences than in concrete sentences, but semantic changes were recognized better in concrete sentences than in abstract sentences. Whereas the finding that semantic changes are recognized better in concrete sentences than in abstract sentences is in agreement with the expectations upon the model, the finding concerning structure-change does not fit into this picture. It could be explained if it was assumed that subjects who have the propositional meaning available test mainly whether the meaning was changed. I shall leave the question open for now of whether this assumption is valid, and come back to it later. If this explanation is accepted, then the experiment by Begg —Paivo supports the assumption that comprehension progresses from the surface to the propositional structure, and remains at the surface level longer with abstract sentences than with concrete sentences. An observation that we ourselves have made points in the same direction. It is plausible to assume that poor readers are slower at converting texts

Levels of language comprehension

137

into propositions than good readers, and that therefore the surface representation plays a more important role for poor readers than for good readers. Ute Biegelmann and I produced indications to support this assumption. If one assumes that slow readers build up propositional structures more slowly than fast readers, and thus remain longer at the level of text surface, then this should be discernible when subjects are to decide if certain words from the text appeared in the text or not. For those persons who stay closer to the surface in their comprehension, the decision times should depend more strongly upon the surface characteristics of texts than for persons who quickly extract propositional meaning from a text. In order to test this assumption, we gave subjects short texts to read in a limited amount of time. The sixth sentence was always the testsentence. It was this sentence that contained, or did not contain, the testword. The subjects had to decide if the test-word occurred in the text or not. The reaction times were recorded. The sentence that the test-word referred to could be active or passive. The test-word could be positioned towards the beginning or towards the end of the test-sentence. The text following is an example. Sentences (6a) —(6c) are the test sentences. The asterisk * in the test-sentences marks the positions at which the test-word could occur. (1) (2) (3) (4) (5) (6a) (6b) (6c)

The courtyard has been deserted since early morning. Someone comes by only occasionally. A dog chases a rat into a cellar. Beside the house-wall stand some full garbage-cans. Some of them have lids that don't close right. The dog (*) jumps up on a wall (*). The rat (*) disappears into a crack (*). A cardboard box (*) sticks out a bit (*).

The test-word could relate to the grammatical subject or the grammatical object of the given sentence, and it could be placed towards the beginning or towards the end of the sentence. In other words, the testword in the test-sentence was in an analogous poisition as in the sourcesentence. (For a more detailed description of the procedure, see Biegelmann — Engelkamp, 1989.) For the fast readers, we expected an important element for their decision concerning the test-words to be whether the word had been used as the grammatical subject or as the grammatical object of the sourcesentence. In any case, we expected the subject-words to be more quickly decided upon than the object-words, since the subject-words function as

138

Johannes Engelkamp

the thematic concept of the source-proposition. The position in the testsentence was expected to play no role for these persons. For the slow readers, however, we expected the decision-time to depend upon the positional congruency of the test-words in the source-sentence and the test-sentence. The decision-time for the subject-words was expected to be shorter when they appeared towards the beginning of the sentence than towards the end of the sentence, since subject-words usually occur towards the beginning of the sentence; for the object-words the decision time was expected to be shorter when they appeared towards the end, since that is where they usually occur in a sentence. The results for the fast readers are shown in Figure 1.

fast subjects

1100 * subject + object

η >§ 1050 + 4)

ε e

«υ

1000 +

u

950

beginning

end

active voice

beginning

end

passive voice

Figure I. Mean reaction times (in ras) of fast subjects as a function of the syntactic status of the tested word (subject/object) in the reference sentences, its position in the test sentence (towards the beginning/towards the end), and voice of the reference sentence

It was observed that, with both active and passive source-sentences, the fast readers made their decisions for subject-words more quickly than for object-words, independent of whether the words in the test-sentences were towards the beginning or towards the end o f the sentence.

Levels of language comprehension

139

The slow readers displayed a different pattern of results (see Figure 2). Their reaction times depended upon the position of the test-words. Their reaction times were shorter for the subject-words in a front position than in a back position, and faster for object-words in a back position than in a front position.

slow subjects

beginning

end

active voice

beginning

end

passive voice

Figure 2. Mean reaction times (in ms) of slow subjects as a function of the syntactic status of the test-word (subject/object) in the reference sentence, its position in the testsentence (towards the beginning/towards the end), and voice of the reference sentence

These results support the assumption that fast readers construct propositional representations while reading the texts, and that during a test to check whether a word occurred in the text or not, they base their decision upon these representations. The propositional representation contains the utilized terms and the information about which ones were thematized. This means that fast readers oriented themselves only upon the subject-role or object-role which the test-term had in the sourcesentence. Slow readers, in contrast, utilized the surface information for their decision about test-words. How they decided upon subject-words and object-words depended upon the position of the words in the sentence. Subject-words that occurred near the normal position for the subject,

140

Johannes

Engelkamp

that is, in a front position, were decided upon more quickly, and objectwords at the normal object-position, that is, in a back position, were decided upon more quickly. These results, also, are in accordance with the idea that text comprehension progresses from the surface to propositional meanings. The two last experiments allow us to conjecture that the subject often does not get around to constructing a situational model for a text because, for example, there is not enough time. On the basis of other experiments, it can also be assumed that there are personality-specific differences in the ability and inclination to construct situational models. Therefore I wish next to show that the construction of a situational model for texts does not inevitably take place, and that inter-individual differences can be detected in the spontaneous tendency to construct situational models for a text. For this reason, I shall report upon an experiment by Denis (1987). Denis investigated the time that readers take to read and to comprehend concrete texts. Each subject was to read two text-passages after another. The instructions were varied between the first and the second text. The first text was always self-paced (SP), that is, it was read at a pace chosen by the reader. The first group of subjects read the second text self-paced also. The second group read the second text with speed instructions, that is, as fast as possible (FR). The third group received imagining instructions for the second text (IM). The persons in all three groups were divided on the median into good and poor visualizers, utilizing the Vividness of Visual Imagery Questionnaire (VVIQ) developed by Marks (1973). On the average, the good visualizers displayed a reading-time of 71.5 seconds and the poor ones a reading-time of 61.8 seconds. If we examine the reading-times for the good and the poor visualizers in the three groups, we obtain the following results. First of all, it could be observed that good visualizers read concrete texts more slowy when self-paced than did the poor visualizers. Second, it could be observed that the differences in the reading-times disappeared under speed instructions. Third, the reading-times for good and poor visualizers were both slowed by the instructions to visualize, but the poor visualizers were effected more than the good visualizers. The pattern of results can be interpreted in line with the Levels Model of language comprehension as follows. The good visualizers, more than the poor ones, tended to construct spontaneous situational models during their reading, and this process of constructing took additional time. Since the good visualizers already tended towards spontaneous construction of

Levels of language comprehension

141

situational models, the instructions to visualize, that is, explicit instructions, did not show as great an effect as it did with poor visualizers; this means, their reading-time changed but little. When the subjects were instructed to read as quickly as possible, the differences in reading-time between good and poor visualizers disappeared because there was no more time for visualizing, that is, the readers no longer took this time. I shall summarize the points I have made until now. Experiments dealing with language comprehension of descriptive texts led us to differentiate between levels of language comprehension: (a) a level at which the text is represented superficially, (b) a level at which the propositional meaning of texts is represented, and (c) a level at which those situations are prepresented to which the texts refer. Further, it is assumed that levels are progressed through one after the other, even when over-lapping of processes and interactions between the levels are not excluded. Finally, it is assumed that the construction of a situational model is optional. With these points, a framework has been set up concerning the process of comprehension. The basic outline of the process seems clear enough. The problem lies — as usual — in filling out the details. Especially the division between surface representation and propositional representation appears unsatisfactory. The representation of the language surface is necessary because it is recognized better than chance even after a relatively long time. Recognition performance can only be investigated in the context of and in contrast to changes in stimuli, the so-called distractors. The distractors determine the differentiation that is necessary for recognition. If we look at the experiments done in this area, then we see that different types of distractors are used, and that the distractors differ in relation to what kind of differentiation is called for. For example, some of the distractors that are used are changes in word-order, some are changes in the syntactical sentence-structure (e. g., active/passive), some are changes of lexemes (e. g., dog/cur), and some are changes between affirmation and negation. All these changes are considered equivalent variations of language surface. However, we can surmise that the recognition performance varies over types of distractors, that is, that these changes are not equivalent for the processing system. If, on the other hand, we look at the propositional representation, this is concluded from the free reproduction of propositional information.

142

Johannes

Engelkamp

Here lies a flaw, since only the number of propositions is evaluated. However, other information is probably also remembered and reproduced, for instance, information about sentence-structure, about the choise of lexemes, etc. If these points are taken into account, then a division between surface representation and propositional representation by way of the two types of performance — recognition and free recall — is not very appropriate. Better indicators are called for. To establish these, it would be necessary to undertake a more differentiated categorization of those kinds of information that are conveyed by the texts, and to examine more exactly which type of information is expressed by which kind of performance, and for how long. On the basis of such findings, the division of levels would have to be evaluated anew. The question thus arises whether we can find more convincing arguments for the division of levels and information-type. Another point that causes dissatisfaction concerns the assumption that the construction of situational models is optional. Variable reading-times alone are not a convincing argument for this assumption. Is it possible to show more convincingly that the construction of situational models is optional? It is helpful here to take a look into another area of research, namely a look at the efforts to differentiate systems of information-processing.

Systems of information-processing Under consideration of the general idea that human information-processing takes place within different sub-systems, it is plausible to assign the three levels of comprehension to three systems of information-processing, that is, to situate the construction of different types of representation in different systems. In this section I shall examine what speaks in favor of this idea. An early suggestion towards a differentiation of two basic systems of information-processing came from Pavio (1971). He made a differentiation between a verbal and a non-verbal system. Although the non-verbal system in principle included all sense systems, Paivio presented it as a general system to be contrasted with the verbal system. Actually, it was the visual non-verbal system that was almost exclusively examined. I also will limit myself — pars pro toto — to the visual system.

Levels of language comprehension

143

A possibility combining Paivio's idea with the Levels Model would be to assign the surface structure information of texts to Paivio's verbal system, and the situational model to Paivio's non-verbal system. But then where would the propositional information fit in? Without saying so directly, Paivio assigns it to both systems. Thus, again, the differentiation between surface information and propositional information is cancelled. It seems more useful to employ a wide-spread, alternate concept, namely that of a semantic system in addition to a verbal and a visual system (cf., Snodgrass, 1984). Here it is plausible to assign the representation of language surface to the verbal system, the representation of propositional meaning to the semantic system, and the representation of the (visual) situational model to the visual system. The equivalency of the framework alone does not help much, of course. Otherwise it would be only a word-game of exchanging one set of terms for another. Connected to this concept of systems, however, are certain methods to verify a separation into systems. In order to clarify the logic behind these methods, I wish to say a few words about the term "system" as I understand it. In a unitary system there is no specialization. Each element of the system contributes basically as much as any other element to the functioning of the system. The system's performance does not depend upon which parts are defect, but only upon how many are defect. The more system-elements are defect, the more the system's performance is adversely affected. In a system consisting of several sub-systems that work partially independently of each other, there is specialization. In most cases, the system's performance is brought about by several sub-systems, but not by all the sub-systems. In borderline cases, one type of performance is based upon a single sub-system. In other words, the disturbance of a certain sub-system leads to a hindrance of that type of performance which it contributes, but the other types of performance remain unaffected by the disturbance. I would like to make this clearer with the use of an example. Let us consider the case of a translation office that produces translations for the languages German, English, and French. Suppose each translator is fluent in all three languages, and there are 15 translators. In this example, the absence of each translator would impair to an equal amount the performance as a whole. A conception of work in one area — for example, with a translation from German to French — would impair the capacity and performance in another area — for example, from English to French.

144

Johannes

Engelkamp

If, in this translation office, there were specialists for certain languages, the system would work differently. Let us say there are five translators each for German/English, German/French and English/French. In this case, the performance for translations between German and English would remain uninflenced by the illness of a German-French translator. At any rate, this would be so if the translators worked completely independently of each another. This is an extremely simple example. It can be easily imagined how the situation could rapidly become more complicated with an increasing number of sub-systems. A system built up of many different sub-systems has the advantage that most disturbances remain local and limited in their adverse effects. The system can function as normal over a wide range of area, even when some parts are defect. I want to continue with the example of the translation office in order to clarify how psychologists try to identify independently-functioning sub-systems. One wide-spread method is that of selective inference experiments: the idea involved here is as follows. If I demand two types of performance to be carried out by one system, then I end up with less performance than if I demanded only one type of performance. Taking our example of the translation office: when all the translators are able to work on one book, then the translation — taken all in all — progresses more rapidly than when the translators are split up to work on two books. If I demand two types of performance that are to be accomplished by two systems, the performance does not change in relation to the situation in which I demand only one type of performance. As an example: the translation of the book from German to English remains uninfluenced by whether or not another group is translating a book from English to French at the same time. If, however, a second book must be translated from German to English, then the performance sinks, because the same sub-system has twice as much work-load. To word this in terms of our theory: for the paradigm of selective interference, we can, with the aid of such double tasks, learn something about whether two types of performance are based upon one or upon two systems. Another method puts a stronger accent upon finding out if a certain system is involved in a type of performance. It is based upon similarity variations. This method of interference by similarity is built upon the assumption that the more similar the types of information are that originate from one system, the more they interfere with each other. If

Levels of language comprehension

145

such a similarity has no effect, then we may conclude that the system upon which the processes are based was not involved in this type of performance. The test for individual sub-systems is best attempted with types of performance that are as simple as possible, since complex types of performance are usually based upon many sub-systems. This would make the isolation of individual systems more difficult. For this reason, I shall deal in this section with experiments on the processing of words and the processing of pictures of objects to which the words refer. I shall work with the representational model already sketched out, with its differentiation between three sub-systems. At the same time I shall set up the assumption that the information concerning word-forms is represented in the verbal system, the information concerning the form of pictorial stimulus in the visual system, and the information concerning the meaning of both in the semantic system. In addition, I shall assume that words automatically activate the representations of their word-forms and meanings, that is, they activate the verbal and the semantic systems, and that for words, the activation of the image-representation, and thus the involvement of the visual system, is optional and requires special conditions. In a parallel manner, the pictures are assumed to lead automatically to the activation of image and meaning, but the activation of word-representation thereby is assumed to be optional. This means pictures are expected to automatically activate the visual and the semantic systems, but not the verbal system (cf., Engelkamp, 1990: Chapter 4). These assumptions are illustrated in Figure 3. verbal stimuli

non-verbal (visual) stimuli

automatic Figure 3. Simplified diagram of a multiple-code theory

146

Johannes

Engelkamp

What findings are there to support these assumptions? Baddeley — Lieberman (1980), in order to test the assumption that visual secondary tasks interfere with the learning of words when the visual system is actively employed, had subjects learn concrete words according to a verbal strategy or according to an imagining strategy. The instructions for imagining were expected to lead to the active use of the visual system. The learning task took place with no secondary task or visual interference. The results shown in Table 1 were obtained. Table 1. Percentages of memory performance after verbal learning and after learning by imagining, as a function of visual interference (after Baddeley — Lieberman, 1980)

verbal imagining

control

visual interference

50.0% 67.3%

49.5% 57.5%

It could be observed that concrete words which were learned verbally were not interfered with by a visual secondary task, whereas words learned by imagining were. Thus it can be taken as probable that the visual system does not have to be activated by the processing of words, but that it can be activated by imagining. At the same time, the findings support a differentiation into a visual system and a verbal system. To be exact, it was shown only that a separate visual system must be hypothesized. On the basis of Baddeley — Lieberman's data, we cannot say if a verbal system is to be differentiated from a propositional system. We can deduce more about this problem from an experiment by Logie (1986). Logie had subjects also learn word lists verbally or by imagining. Here also, the lists had to be learned once without interference, and once with visual interference. Like Baddeley — Lieberman, Logie could observe that visual interference during verbal learning did not effect memory, but it did effect memory for learning by imagining; at the same time he could observe that verbal interference hindered verbal learning, as expected, but did not hinder learning by imagining. Table 2. Percentages of memory performance after verbal learning and after learning by imagining as a function of interference (without, visual, verbal) (after Logie, 1986)

verbal imagining

without

interference visual

verbal

69.7% 68.0%

65.0% 48.0%

57.1% 63.3%

Levels of language comprehension

147

The finding that a visual secondary task interferes only with learning by imagining shows anew that we must separate a visual system from a verbal system and a propositional system, and that the visual system is not automatically involved in the processing of words. The finding that a verbal secondary task interferes only with verbal learning and not with learning by imagining does not fit so well into the picture. It is to be expected that word-representation also be activated for learning by imagining. A verbal secondary task was expected to interfere here, too, since word representations are automatically activated for word learning. How can we explain the finding that a verbal secondary task did not interfere with learning by imagining? A possible explanation — as with the parallel finding by Begg — Paivio (1969) — could be that the subjects who learned with the imagining instructions used only the propositional system and the visual system for memory, but not the verbal system. This explanation is also supported by the finding that when the subjects were forced to use the verbal system for memory, there was an effect of verbal interference. This is supported by findings by Nelson — Brooks (1973 a). Nelson based his experiments upon the well-known observation that phonemic word similarity interferes with word learning. In his experiments the subjects had to learn stimulus-response pairs, for example, word pairs. When the stimulus word was presented, the response word had to be recalled. With this method of cued recall, the subjects were forced to use the verbal information of the stimulus word for memory. What was counted in these experiments is the number of errors until the correct learning of the pairs. If the phonemic similarity of the stimulus words was varied, then it could be observed that more errors were made with phonemically similar words than with phonemically dissimilar words as stimuli. This supports the assumption of the use of the verbal system and the effect within this system of phonemic similarity. Nelson — Brooks (1973 a) could then show that this interfering influence of phonemic similarity also remained effective when the subjects learned the word pairs with imagining instructions. The subjects were forced, through the presentation of the stimulus words, to use the verbal system during recall. In another experiment, Nelson — Brooks (1973 b) could demonstrate that the phonemic similarity of the stimulus words had no effect when

148

Johannes

Engelkamp

the stimulus words were replaced by the pictorial representation of the denoted objects. Table 3. Number of errors for learning by picture—word pairs and by word —word pairs as a function of phonemic stimulus-similarity (after Nelson — Brooks, 1973 b) phonemic similarity

picture-word pairs

word-word pairs

large small

17.8 21.4

69.7 39.6

These results, together with the observation that pictures when explicitly labelled have a definite effect of phonemic similarity, demonstrate anew not only that it makes sense to differentiate between a verbal system and a visual system, but also that pictures are not always spontaneously and implicitly labelled, which means that the systems are activated and can be used independently of each other. Just as we can vary the phonemic similarity of the words, so we can also vary the visual elements of pictures. In this case the errors for picture-stimuli, but not for word stimuli, should depend upon visual similarity. This could be demonstrated by Nelson — Reed — Walling (1976). Their results also support the idea that it makes sense to differentiate between a verbal system and a visual system, and that it is not necessary for the visual system to be activated during word-processing. Table 4. Errors for learning picture—word pairs and word—word pairs as a function of the visual similarity of the pictures or of the objects denoted by the words (after Nelson - Reed - Walling, 1976) visual similarity

picture-word pairs

word-word pairs

large small

26.2 16.7

25.9 30.7

U p to this point, the findings concerning selective interference and interference of similarity-variations support the reasons for differentiating between a verbal system and a visual non-verbal system. The findings say less about the necessity of postulating a propositional system. Since the assumption of this system is directly plausible and is least questioned

Levels of language comprehension

149

by psychologists, there have been the least number of experiments undertaken to verify this system. Nelson also showed that a verbal system must be differentiated from a semantic system and also from a visual system (for example, 1979). When in pair-association learning the semantic similarity of the stimulus words was varied, there was the same rate of increase in the frequency of errors for increasing semantic similarity, both for picture —word pairs and for word —word pairs. This finding supports not only the differentiation of a semantic system from a verbal system and a visual system, but also supports the assumption that pictures and their labels tap the same meanings. To summarize: the experiments on selective interference and on interference by similarity — variations support the differentiation of three systems of information-processing — a verbal system, a propositional system and a non-verbal visual system. They also support the idea that the visual system is not necessarily involved in language-processing. Thus, assigning the three levels of language comprehension to the three systems of information-processing just named proves, on the whole, to be meaningful, and the experiments on system-differentiation support the basic idea that language comprehension progresses through three levels. The two models differ in the types of information that are assigned to the three levels. In the Levels Model for language comprehension, surface information of a text, for example, is assigned to the first level of comprehension. What this includes remains vague, however. In the Systems Model, the information that is assigned to the verbal system is defined more precisely as phonemic information, but also more narrowly. On the one hand, it is limited to the characteristics of the sounds of language independent from meaning, and on the other hand it excludes all types of information connected with word order, since it deals with the processing of isolated words. Semantic information is also considered in a more limited way, since it only refers to the meaning of individual words. However, here semantic information about words can be regarded as a component of propositional information, as is postulated for text comprehension. The information concerning situational models was already, in the context of text comprehension — as an example, not as a principle — limited to visual — spatial types of information. In the Systems Model, visual information is also limited to the visual system. However, the form —information and not the spatial information is more central here.

150

Johannes

Engelkamp

The neglect of spatial information is also a direct result of the limitation to isolated words. All this confirms the idea of differentiating between three sub-systems in the informational processing of language; but the idea, especially, of assigning types of information to the verbal systems and the visual system are not in line with the types of information of text-surface and situational models, as suggested by the Levels Model. We are left with the idea that first the physical language event is represented, then the information about meaning; last of all can the referential information play a role. It still remains to be clarified which additional types of information play a role in comparison to word-processing by text comprehension. Here the situation becomes complicated, not only through the fact that new stimulus types (different word-classes) and the aspect of word-order play their roles as additional aspects of the language surface, but also through the fact that various new aspects of information enter in that are conveyed by these additional aspects of surface. Here those types of information are included that refer to the relationships between the meanings of words, to difference of importance placed upon the various meanings of words, to pragmatic information (for example, to the differentiation between "old" and "new" information), and so on. To clarify these points, it would, according to the Systems Model, be important to systematically vary individual surface aspects, and to examine them within the paradigm of selective interference and interference of similarity. Such research could help not only to explain if the postulated system-structure is suffficiently differentiated, but could also help us come to a better understanding of the question how the different types of information are extracted from the text surface. Since such research is still very much lacking, I wish, during the last section of my talk, to at least make the principles of the procedure clear with the example of word-processing, and to show that further differentiations in the systems are becoming evident. The surface aspect that is to be varied is the modality of wordpresentation. In this last section, I also wish to return to the phenomenon we first encountered with Begg — Paivio (1969) and then with Logie (1986). It concerns the observation that for the same performance — for Begg — Paivio for the decision about surface changes; for Logie for the recall of words — evidently different types of information-systems can be used. This aspect is important, because it indicates a flexible behavior of the system that makes the research concerning the over-all system much more difficult.

Levels of language comprehension

151

Further system-differentiation U p to now the impression has been that it is unimportant if a text is read or heard. For example, we did not mention at all how, in Nelson's experiments, interference of phonetic word similarity could interfere with the performance for visual presentations of words. In order to provide an answer to the question of the difference between hearing and reading, allow me to take one more step back. At the beginning of my talk I said it had at first been assumed that the language surface served as a vehicle for meaning and that it was then forgotten. Alongside this assumption, a short-term memory was postulated for this conveyance. Short-term memory was assigned the function of retaining the information about language surface until the analysis of meaning was finished. The meaning itself was assigned to long-term memory. The capacity of this short-term memory was measured with an immediate recall test. For this test, increasingly longer sequences of numbers were spoken aloud for the subjects. These sequences had to be repeated in the original sequence. On the average, adults can repeat 7 + 2 such numbers correctly. The capacity of short-term memory was therefore set at around 7 units. This assumption about retention capacity was connected with the explicit idea that there was only one kind of shortterm memory. Implicitly, this kind was seen as verbal. This came from the fact that research was almost exclusively carried out for memory of verbal stimuli. With time, this idea of a single short-term memory had to be revised. I wish to show that this revision contributes to the solution of both the problems I mentioned previously. During the 70s several patients were observed who had average intelligence, but whose short-term memory, measured as mentioned, showed considerable defects. These patients could not repeat more than 2 to 3 numbers correctly (Shallice, 1988: 43). The objection could be raised that the patients' peripheral acoustic sense processes were defect and they could not recognize words from hearing them. This objection can be dismissed. If the patients heard a recording of a list of words with the instructions to press a key when a word from a certain category came, for example, an animal, there their performance was practically perfect. Another conjecture can also be dismissed, namely that the verbal output system was defect. This is evident from the fact that the patients were also unable to reproduce numbers by pointing to them.

152

Johannes

Engelkamp

It makes sense, therefore, to assume that the short-term memory of these persons was defect. Now, it could also be observed that the span for immediate recall of visually-presented number sequences was considerably higher than for acoustically-presented ones (Shallice, 1988: 56). This observation forces us to assume not one, but several short-term memories, and, still more, one short-term memory for acoustic language stimuli, and another for visual language stimuli. This, again, leads to the assumption that it might be the same for non-verbal stimuli, so that we would have to differentiate between four short-term memories: two verbal and two non-verbal. The differentiation between a verbal short-term memory and a nonverbal visual short-term memory was already suggested by experiments by Nelson (for example, 1979). It is a verification for the differentiation between a verbal and a non-verbal acoustic short-term memory that in the case of patients with a defect in their short-term memory for number sequences presented acoustically, their performance for acoustically presented non-verbal stimuli was no different from that of normal persons (Shallice - Warrington, 1974). Thus, the assumption about the systems were already considerably differentiated. If we assume that our hearing of language is based upon verbal-acoustic short-term memory, and our reading upon verbal-visual short-term memory, then we can expect a differentiation between the processes involved in hearing and in reading texts. Baddeley (e. g., 1986) developed several ideas on how acoustic language comprehension could function, on which function is assigned to the shortterm memory for acoustic language information, and on how acoustic language comprehension is to be differentiated from visual language comprehension. Baddeley differentiates between visual short-term memory and verbal acoustic short-term memory. However, he says hardly more about visual short-term memory than that it exists. He says more about verbal-acoustic short-term memory. According to Baddeley, this consists of a phonological short-term memory and an articulatory rehearsal-mechanism through which information can be stored anew in the phonological short-term memory. The information in the phonological short-term memory decays quickly. Because it is articulated internally and is brought anew into the phonological short-term memory and this process can be repeated again and again, acoustic language information can be retained for longer timespans. Looking at visually-perceived words, it is assumed that these come

Levels of language comprehension

153

into a corresponding visual short-term memory and from there into the articulatory rehearsal-mechanism. Figure 4 illustrates this model.

visual words

—

visual short-term store

rehearsal mechanism

J acoustic words

phonological short-term store

Figure 4. A model for word-processing (adapted from Baddeley, 1986)

Although this model is simple, it still provides an elegant explanation for several phenomena of language processing. Let us take as an example the effect of phonological similarity. It consists of phonemically similar words resulting in poorer retention than phonemically dissimilar words. Let me remind you of the experiments by Nelson (e. g., 1979). The phonological similarity effect has its origins, according to Baddeley, in the phonological short-term memory. Here phonologically similar types of information interfere with each other. According to Baddeley, this effect cannot be hindered by acousticallypresented words, since they go directly from the ear into short-term memory. The situation is different when words are presented visually. These words, according to Baddeley, are assumed to come into the phonological short-term memory by way of the articulatory rehearsalmechanism, and are then retained by rehearsal. If this rehearsal-mechanism is blocked by so-called articulatory suppression, then words that are seen are expected not to enter the phonological short-term memory. If articulatory suppression is introduced — for example, by always speaking the same syllable aloud — then a phonological similarity effect can be observed for words that are heard, but no longer for words that are seen (e. g., Levy, 1971). This model of Baddeley's is in line with the phenomenon that the patients with defective immediate recall for acous-

154

Johannes

Engelkamp

tically-presented words showed no phonological similarity effect at all (Vallar — Baddeley, 1982). Their errors pertained to visual similarity. Another effect that is well known is the word-length effect (e. g., Baddeley — Buchanan, 1975). In an immediate recall test, more short words are recalled than long ones. According to Baddeley, this effect originates in the fact that a short-term phonological trace is assumed to be vulnerable to rapid decay unless it is refreshed by rehearsal; thus, long words are more vulnerable to decay than short words. At the same time, the rehearsal of long words takes more time. Here also, according to Baddeley, the word-length effect is altered by articulatory suppression for visually-presented words, but hardly at all for acoustically-presented words. With the help of words of the same number of syllables that were spoken in a shorter or longer length of time (e. g., lemon versus demon), Baddeley was able to show that it is in fact the characteristics of time for the acoustic event that are at the basis of the word-length effect, and not the number of syllables (cf., also Ellis — Hennelley, 1980). This representation of Baddeley's model, and of several phenomena of word-processing, supports not only the assumption of an acoustic shortterm memory and a visual-verbal one, it also leads us to suppose that the phonological short-term memory, together with the articulatory rehearsal-mechanism, serves to keep acoustically-presented language information accessible for semantic analyses for a sufficient amount of time. The function of this short-term memory could, therefore, lie in enabling back-up and reanalyses by language comprehension. A so-called phonological recoding of language that is read could serve to utilize the characteristics of the acoustic system for reading also. This mechanism, for example, could have been used by the subjects in Nelson's experiment. At the same time, however, we should not forget that the visual system has access to other mechanisms that achieve a similar or even greater efficiency of processing in other ways (e. g., refixations). We can conclude from this that the processes for hearing language are different from those for reading. So-called "look-ahead" windows of parsers — if the parsers, as we like to say today, are to be psychological or cognitive — would be expected to take into account these differences between a visual-verbal short-term memory and an acoustic-verbal one. In conclusion I want to talk about the possibility of the same types of performance having different system-foundations. This aspect is not only a central aspect of the general manner of functioning by the human processing of information, it also makes clear that language comprehen-

Levels of language comprehension

155

sion and language-production are based in part upon completely different systems. U p till now we have looked at types of performance for which persons have to speak or write. In the next section I want to take a look at only the performance of speaking. For this kind of performance I shall refer to specific implications from Baddeley's model and point out consequences of these implications for models of speech comprehension and speech production. The articulatory rehearsal-mechanism is, it is true, not tied down to the overt speaking aloud of words, but can be well understood as a basis for repeating words after someone else and for reading words. If this is true, then speaking would be possible without the involvement of the semantic system. There would be quasi direct paths from acoustic-verbal and visual-verbal short-term memory to a verbal output system. If there are also paths from the verbal input systems to the semantic system, then the question arises of whether there is a direct path from here to the verbal output system, or whether the path for speech goes from the semantic system by way of the verbal input systems. Many neuropsychological findings support the idea that both paths exist (e. g., Ellis — Young, 1989). Such findings lead to the assumption that word-processing can go by way of the semantic system or directly to the verbal output system. At any rate, output systems are called for in addition to input systems. Riddoch et al., for example (1988), propose along these lines the systems, as seen in Figure 5. In this system there is a differentiation not only between acoustic and visual representation, but also, within both sense systems, between a verbal and a non-verbal one. In addition, an output system is introduced. The relationships between the systems are also more complex. The path does not lead directly from one input system to another, but only by way of the conceptual system. Also, the path can lead from verbal input to speech output, but from a non-verbal input, however, the path leads only by way of the conceptual system. Thus it is implied that not only are language comprehension and production different processes, but also that we must differentiate between different types of language production. If things are already so complicated in relation to the processing of word and picture, then we need only a little imagination to think how complex the system must be that performs the processing of whole texts or actual every-day situations. It becomes immediately clear that the Levels Model of language comprehension can only be regared as a first

156

Johannes

Engelkamp

visual input

acoustic input

speech Figure 5. Systems of representation (adapted from Riddoch — Humphreys — Coltheart - Funnell, 1988)

outline for the analysis of language comprehension that still requires considerable differentiation and elaboration. If we take the idea seriously that such complex types of performance such as language performance are based upon the cooperation of very many sub-systems, and also the idea that types of performance that look the same on the surface can take place only through the cooperation of completely different sub-systems, then it is clear that psychologists stand at the beginning of a long road of experimentation concerning the processes involved in language comprehension.

References Anderson, John R. 1985 Cognitive psychology and its implications. New York: Freeman. Baddeley, Alan D. 1986 Working memory. Oxford: University Press.

Levels of language comprehension

157

Baddeley, Alan D. — Steven Grant — Elizabeth Wight—Nick Thomson 1975 "Imagery and visual working memory", in: P. M. Rabbitt—S. Domic (eds.), Attention and Performance V. 205 — 217. London: Academic Press. Baddeley, Alan D. —Keith Lieberman 1980 "Spatial working memory", in: R. Nickerson (ed.), Attention and performance VIII. 521 — 540. Hillsdale, N. J.: Lawrence Erlbaum. Baddeley, Alan D. —Nick Thomson — Martin Buchanan 1975 "Word length and the structure of short-term memory", Journal of Verbal Learning and Verbal Behavior, 14: 575 — 589. Begg, Jan — Allan Paivio 1969 "Concreteness and imagery in sentence meaning", Journal of Verbal Learning and Verbal Behavior, 8: 821 - 8 2 7 . Biegelmann, Ute—Johannes Engelkamp 1989 Zur differentiellen Verfügbarkeit von Konzepten beim Textverstehen. Saarbrücken: Arbeiten der Fachrichtung Psychologie der Universität des Saarlandes, Nr. 139. Denis, Michel 1987 "Individual imagery differences and prose processing", in: M. A. McDaniel —M. Pressley (eds.), Imagery and related mnemonic processes: Theories, individual differences, and applications. 204 — 217. New York: Springer. Denis, Michel — Hubert D. Zimmer 1991 "Spatial qualitives of cognitive maps constructed from texts and physical maps", (in press). Ellis, Andrew—Andrew W.Young 1989 Human cognitive neuropsychology. Hillsdale, N. J.: Erlbaum. Ellis, Nick C . - R i t a A. Hennelley 1980 "A bilingual word-length effect: Implications for intelligence testing and the relative ease of mental calculation in Welsh and English", British Journal of Psychology, 71: 43 — 52. Engelkamp, Johannes 1991 Language and modes of thought, in Appel, 6. — Dechert, H. W., A case for psycholinguistic cases. Amsterdam: Benjamins (in Press). 1990 Das menschliche Gedächtnis. Göttingen: Hogrefe. Fillenbaum, Samuel 1966 "Memory for gist: Some relevant variables", Language and Speech, 9: 217-227. Johnson-Laird, Philip N. 1983 Mental Models. Cambridge: Cambridge University Press. Johnson-Laird, Philip N. —Douglas J. Herrmann — Robert Chaffin 1984 "Only connections: A critique of semantic networks", Psychological Bulletin, 96: 2 9 2 - 3 1 5 . Kintsch, Walter 1974 The representation of meaning in memory. New York: Wiley. Kintsch, Walter—Janice Keenan 1973 "Reading rate and retention as a function of the number of propositions in the base structure of sentences", Cognitive Psychology, 5: 257 — 274.

158

Johannes

Levy, Betty. A. 1971

Engelkamp

"The role of articulation in auditory and visual short-term memory", Journal of Verbal Learning and Verbal Behavior, 10: 123 — 132. Lindsay, Peter H. —Donald A. Norman 1977 Human information processing. New York: Academic Press. Logie, Robert H. 1986 "Visuo-spatial processing in working memory", Quarterly Journal of Experimental Psychology, 38 A: 229 — 247. Marks, D. F. 1973 "Visual imagery differences in the recall of pictures", British Journal of Psychology, 64: 17 — 24. Masson, Michael E. J. 1984 "Memory for the surface structure of sentences: Remembering with and without awareness", Journal of Verbal Learning and Verbal Behavior, 23: 579-592. Nelson, Douglas L. 1979 "Remembering pictures and words: Appearence, significance and name", in: L. Cermak —F. I. M. Craik, Levels of processing in human memory. 4 5 - 7 6 . Hillsdale, N. J.: Erlbaum. Nelson, Douglas L. — David H. Brooks 1973 a "Independence of phonetic and imaginal features", Journal of Experimental Psychology, 97: 1 - 7 . 1973 b "Functional independence of pictures and verbal memory codes?" Journal of Experimental Psychology, 98: 44 — 48. Nelson, Douglas L. — Victor S. Reed —John R. Walling 1976 "Pictorial superiority effect", Journal of Experimental Psychology: Human Learning and Memory, 2: 523 — 528. Paivio, Allan 1971 Imagery and verbal processes. New York: Holt, Rinehard and Winston. Perrig, Walter 1988 Vorstellungen und Gedächtnis. Berlin: Springer. Perrig, Walter — Walter Kintsch 1985 "Propositional and situational representations of text", Journal of Memory and Language, 24: 5 0 3 - 5 1 8 . Riddoch, Jane J. —Glyn W. Humphreys —Max Coltheart —Ernest Funnell 1988 "Semantic systems or system? Neuropsychological evidence re-examined", Cognitive Neuropsychology, 5: 3 — 25. Sachs, Jacqueline 1967 "Recognition of memory for syntactic and semantic aspects of connected speech", Perception and Psychophysics, 2: 437 — 442. Shallice, Tim 1988 From neuropsychology to mental structure. Cambridge: Cambridge University Press. Shallice, Tim — Elisabeth K.Warrington 1974 "The dissociation between short-term retention of meaningful sounds and verbal material", Neuropsychologia, 12: 553 — 555. Snodgrass, Joan G. 1984 "Concepts and their surface representation", Journal of Verbal Learning and Verbal Behavior, 23: 3 — 22.

Levels of language comprehension

159

Vallar, Giuseppe — Alan D. Baddeley 1982 "Short-term forgetting and the articulatory loop", Quarterly Journal of Experimental Psychology, 34: 53 — 60. van Dijk, Teus Α. —Walter Kintsch 1983 Strategies of discourse comprehension. New York: Academic Press.

Part II Developmental aspects: the evolution of written texts

The pragmatics of medieval texts Heinz Bergner

Especially since the publication of the extensive bibliography of Nuyts and Verschueren (1987) it has become apparent how pervasively the linguistic discipline of pragmatics has grown and developed. Widely used textbooks, for example that of Levinson (1983), have made the discipline wellnigh fashionable and even literary scholars are no longer prepared to do without it 1 . But in spite of its general acceptance and propagation, the specific synchronic application of the pragmatic approach to individual languages has failed to emerge f r o m its infancy and has rather been confined to theoretical discussions on matters of principle. This lack of a pragmatic dimension can also be felt to a considerable extent in accounts devoted to individual languages. A mere glance at the detailed, monumental g r a m m a r of Quirk et al. (1985), which in every other respect meets the reader's desire for comprehensiveness, gives sufficient evidence of this absence. This holds true to an even greater degree for the opening u p of older stages of language use, the analysis of which has for many decades been confined to the classic branches of linguistic scholarship: historical phonetics and morphology, syntax, semantics, and lexicology. Especially in the field of diachronic language development, however, it is pragmatics which could make important contributions to a linguistic solution of problems. This could be achieved by trying to illuminate the variety of relations between the respective linguistic signs, between linguistic sign and sign-user, as well as between the respective creator of the sign and its recipient. For historical texts this means specifically that although their fundamental grammatical construction has long since been investigated, there is another plane beyond the form and content of the pure linguistic sign. This plane of fixed textual forms is the place where conscious choice f r o m a stock of signs, expressive and conative function, the execution of an authorial act, an intention is realized. This is in principle controlled by the manifold extra- and intra-linguistic contexts of communication. The usage- and evaluation-correlations of earlier texts are to a broad extent based on the particular nature of how speaking and writing individuals experience reality and their social surroundings,

164

Heinz

Bergner

on the conditions and presuppositions of a historically oriented text-production and -reception, on the imaginative spheres of individual sectors of the speech community and their frames of conception, which eventually determine the precarious relation between norm and the departure from it, misunderstanding, conflicts of communication, and successful or unperformable communication. Social standards determine the functional orientation of texts, the nature of their basically dialogical situation, their co-operative principles, their communicative regularities. Here knowledge about the application bases for direct and indirect speech acts proves to be of importance. We will demonstrate that the rules of speaking are of particular relevance, because for texts stemming from a remote past they are the primary communicative stratum. Therefore in such texts the interaction of spoken and written language gains essential importance, even if the reconstruction of historical everyday language poses enormous difficulties. A closer definition and circumscription of medieval laymen's speech would at any rate give us a better understanding regarding language change and the origins and development of new registers and textual forms. The investigation of the field of problems and tasks just outlined would certainly be asking too much of pragmatics alone, if it could not rely on receiving help from literary scholarship, historiography, theology, and from folkloristic and cultural studies. All of them have produced fundamental contributions conducive to the solution of the clusters of questions delineated above. It is interesting to note that there have been interpreters of the New Testament, particularly from Western Germany (Egger 1987), who made use of modern linguistic methods in the explication of texts deriving from past epochs. Here, however, the pragmatic approach of Güttgemanns (1973, 1975, 1978) remains entirely biased in favour of the structuralist thinking of modern syntax-theories, thus generating the impression that conventional ideas are only being clad in modern linguistic garb. A different approach is offered by Berger (1984), who starts by pointing to the achievements of contemporary discourse analysis and who furthermore succeeds in using these to detect important linguistic means of sentence-linkage in the New Testament (pp. 11 ff.). This working procedure is not necessarily superior to traditional methods. When Berger finally emphasizes the importance of the factors of text-production and text-reception for understanding the New Testament, and of always interpreting it in the pragmatic global context of the historical situation (p. 265) this is certainly an important but also a trivial observation. Unfortunately his thoughts do not go beyond the theoretical and me-

Medieval

texts

165

thodical stage. This is particularly regrettable since traditional theological exegesis has already made available a large number of preparatory works which could be of value in the attempt to successfully shed light on biblical texts by pragmatic means. To date linguists have hardly dared to advocate a historically oriented type of pragmatics. Noteworthy among the few who have are Sitta et al. (1980) and Schlieben-Lange (1983). Their arguments combine broadly based programmatic comments with short practical applications. It is characteristic that both works are not listed in the bibliography of Nuyts and Verschueren, although they fall within the period under review. The latest attempt to approach the topic have been the studies presented by Schwarz et al. (1988). They, however, fail ever to refer to the debate in Sitta et al. and Schlieben-Lange and in their discussions furthermore devote far more space to general theory than to concrete comments on the pragmatics of old texts. In what now follows, we therefore attempt to obtain a general view of the pragmatic essentials of medieval texts. In so doing, we respectfully ask the reader to bear in mind the disparity between the confined space of this article and the bulk of a tradition that has accumulated through having been handed down to us over a whole millenium. In the Middle Ages pragmatic linguistic action springs from a conception of the world which — in comparison with our de-sacralized, secular, rationalized, and pluralistic modern ways of thinking — was based on an eternal order of phenomena which was not only of divine design, but was also deemed to be perceptible as such, and which was strictly centered on a conception of God and Christ. For medieval society this results in a tightly constructed system of human dependencies, of corporative organisation, of hierarchies clearly defined and hard to break into, of super- and subordinations, and of rules officially sanctioned and for the most part also accepted. The corporate order and structuring in the society of the Middle Ages pervades all areas and is perceptible everywhere, hence also in a field where one would hardly expect it to be found, viz. in the church and monastery, the vehicles of civilization, where the principle was observed even more closely. This conception of ordo — the traditional appropriate name for an idea taken inter alia from the Old and New Testament — result both in social privileges, and in the formation of social classes and groups and is firmly rooted in the manner in which medieval authors and audiences/readerships experience the world. Man and his world in the Middle Ages are thus based on a preexisting divine order. This transforms medieval man into an official being,

166

Heinz

Bergner

a creature with a part to play, but unable to make decisions about its own existence as it pleases. The conspicuous thing about this structure of the medieval world, which is linked in all essentials to system and hierarchy, is a certain measure of intransigence, as well as the fact that in this conception of the world the tendency towards balancing out and compromise are only rudimentarily developed and that division and incompatibility prevail (Borst 1973; Delort 1982; Goetz 1986; Boockmann 1988). This disposition, dependent on the ori/o-principle, is in medieval times matched up with a relatively uniform picture of the world, rooted in classical and Christian conceptions, and certainly known to most participants in the communicative process. It remained effective in Western Europe until early modern times. Its characteristics are a geocentric structure of the whole universe, a variety of stratifications and — as people thought — a wonderful order of degrees. All the elements of the inspired universe stand in close relation to each other through a huge system of analogies and correspondences. The order of the macrocosm that is, the universe and the world, corresponds to a chain of being in the microcosm of the earth, reaching from the anorganic, via the organic, to the human world. Furthermore, the macrocosm influences the microcosm in many ways, the result is a stable, predictable interplay of mind, body, terrestrial and celestial world, a factor that explains the high importance of astromedicine and astronomy, (which was synonymous with astrology). This well-known Greek tradition of thinking in cosmic and ordo-categories was in the Middle Ages summarized by abstracting it in the number or relations of numbers, which could be justified by a frequently quoted verse from the Old Testament Book of Wisdom (11,21). Medieval texts therefore make a myth of numbers, and composition by numbers, which, however, can easily result in mere naive playfulness, is a firmly established structural element in many medieval works. Since for the medieval author and reader all existing things are in terms of cause and ontology equally and determinately connected to each other, medieval texts contain a multitude of references, concatenations, and correspondences; this is what makes these texts so characteristic, setting them apart in their own right. The employment of medieval speech-signs is in line with this. Because in individual texts linguistic relationships are often generated by close cross-references between the semantics and acoustics of sentences, parts of sentences and words, repetitions become perceptible in many ways. In general all this reaches much further than the linguistic network of modern texts. The setting up of relationships

Medieval

texts

167

just delineated is based additionally on two phenomena which lend medieval texts a structure of close interdependence. The first feature to be mentioned is the typological view (Ohly 1977, 1979), which is so exceedingly popular in medieval texts that it used to be taught in handbooks, such as in Pictor in carmine (12th century) and the two manuals drawn up in the late Middle Ages, Biblia pauperum and Speculum humanae salvationis. Typology is a way of thinking, well-known since its introduction by the Fathers of the Christian Church, which entails that events and characters in the Old Testament are prefigurations, typological prototypes and analogies of events and characters of the New Testament. In the Middle Ages this conception was applied to all textual varieties and thus generates a rich network of semantic relations, which finally explains the considerable transcendental dimension of medieval works. Typology facilitates the connection of all textual segments with each other. The texture of medieval texts is furthermore kept close-knit by the pointing out of etymological relations between words (Schwarz et αι. 1988: 207 — 257). The vast amount of etymological works of the Middle Ages indicates 2 that etymologizing does not aim at finding out or at properly reconstructing the respective linguistic roots but makes apparent the multitude of the — reputedly — only thinly veiled connections between lexemes. According to this view, words which in a way sound similar 3 also denote similar things. By pursuing this method references to content, metaphor, and metonymy can become transparent, words can be brought into an antonymous relation to each other, proper names can be traced back to interesting origins, words can be analyzed in order to criticize facts of contemporary life, and even the gap to other languages can be bridged. This is also possible by means of the cryptogram and acronym. Medieval etymologizing is thus a heuristic procedure, governed by Christian-theological impulses, and its significance for the the user of linguistic signs is that it helps him gain a deeper understanding of the meaning of a word and its relationship to other elements of speech. Here the signifier (signifiant) develops an identifying function, as the phonemic form of the individual speech sign is seen in the correlations on higher planes, and thus the respective items of contents and referents of the sign are brought into a new and often striking relationship to the other speech signs. Such almost unlimited processes of text formation facilitate essential arguments and proofs and they become operative for allegorical textstructuring and exegesis. Therefore we must concede that the medieval word itself stands in opposition to all day-to-day classifications in ever

168

Heinz Bergner

new correlations, which to a certain extent eliminates its firm phonemic and signifying contours. So the medieval text, in particular the literary text, contains a network of references and interdependencies which give it a high degree of interwovenness and set it generally apart from linguistic fragmentariness and anacoluthonic disjunctiveness — dominant features of the modern text. But, furthermore, medieval texts are characterized also by the specific way of understanding in which medieval man in his world encountered the signs he found. Contrary to modern experience, their typical feature was that, by and large, they were always legible and comprehensible to their then audience. Nature — so they deemed — could be read like a book, as is manifest in the then often used metaphor of liber naturae, standing for world and being. Because of the close affinity between human cognitive capacity on the one side and the object itself on the other, medieval man felt sure of his faculty always properly to recognize and decipher things. His cognitive capacity here addresses a variety of objects which in principle — apart from the everyday meaning they are meant to carry — had a higher, spiritual significance, which, effected by a divine decision, was imprinted on them like a shelf mark and — as stated above — could be identified unequivocally. In the Middle Ages, the referential and symbolic meaning (the latter being, by the way, not always of universal significance) of an object — and this judgment has often been expressed — form an indivisible union. Yet, the world of meanings thus created, which expressed the inner truth content of existence and was, incidentally, easily accessible in the countless number of lapidaria, herbaria, and bestiaries had essentially always been present. True, it had to be discovered and named by careful study, but its constitution was independent of the perceiving eye. In this manner things correspond beyond the usual orders in a setting of higher interdependencies, and this in turn has in an analogous way consequences for the co-existence and interconnection of the individual speech signs. After all that has been said it is apparent that the linguistic structuring of medieval texts is controlled and channelled by traditions, social orientations, and clerical authorities. The margin thus left for creativity is exploited by the individual author just as much as his talents allow. In general, however, it is fair to say that for medieval texts this results in a mode of representation with a clear preference for norm and typicality. The texts are in principle in line with established categories, with transparent, clear dimensions, and a sense of purposefulness. Particular texts, which nowadays would be ascribed to the fictional type, contain a third-

Medieval

texts

169

person point of view and are marked by their closed form, i. e., beginning and ending are deliberately indicated. The characters in the narrative therefore usually have their whole life-story depicted, the depiction usually proceeding in a linear manner, with each episode more of less firmly linked with its successor, and the single parts appearing to fit well together. It is necessary to point this out because respective modern texts often have an open design and a variety of points of view and are dominated by a principle of discontinuity. What in a very special way is typical of medieval texts — and not only of those just mentioned — is their inherent paradigmatic element. It is common to present fictional characters in representative, exemplificatory roles, with which the reader/listener is called upon to identifity. This device serves deliberately to portray these characters in an elevated and non-individual style. Finally this results in a tendency to idealization and the official, and thus everything private and personal (so often postulated today) seems to be rather excluded. Very much on the contrary, medieval texts have a tendency towards the impersonal, the formal, evidence for such being by no means restricted to the countless exempla, fables, and legends where this might possibly be regarded as mere conformity to the generic concept. By the way, this is the reason why the high degree of instruction and convention in medieval texts is compatible with long digressions in catalogue and didactic fashion. Thus in medieval texts the moralistic approach takes a prominent place in a style of composition and writing concerned with validity and irrefutability. This explains the wide spread at the time of sententious and aphoristic literature by way of a wide variety of textual forms. In accordance with this is the choice taken by the authors of medieval texts from the stock of speech signs. The explicit, directly formulated illocutionary speech-act is often marked, and the performative verb is rather the rule than the exception. In view of the various regulating forces the medieval author was facing, his close obligations to a great number of genres are not really suprising. These systematic patterns for literary works, with which the author and the reader/listener were usually familiar, demonstrate, through their supply of fixed themes, motives, structures, and language constituents, that every medieval text must be understood intertextually (Wolfzettel 1990). Since, however, each kind of medieval text appears in a different guise in the various eras and vernaculars, usually with mixed and transitory forms, a simple system of pigeon-holing is certainly not advisable. Nevertheless, a marked genre designation is often striking in that the genre specified appears to be clearly fixed by a multiplication of explicit signals.

170

Heinz Bergner

At the same time its special function within the whole system of genres and within the social cosmos is being underscored. Since it was possible for medieval man, as has been shown above, to read and interpret the medieval world, the authors of medieval texts were compelled to closely follow this guide-line. In order to do so they took over the ancient traditions of rhetoric. Rhetoric provides a variety of techniques for the composition of texts as well as instructions for appropriate and effective writing and argumentation, thus improving the reader's/listener's understanding of the text. Initially, rhetoric had been bound to the Latin language and for all those producing texts it had been so important as to become an integral part in the linguistic training of clerics-to-be. Rhetorically wrought texts can be found in virtually all areas of medieval writing, in letters, sermons, and in pamphlets as well as in religious and secular literature in the vernacular. Of the five stages into which rhetoric is usually divided (Lausberg 1960), the third, the socalled elocutio, was of particular importance for the fashioning of texts. It is concerned with the adornment of written as well as spoken language and comprises the figures of speech, the flgurae, and the metaphorical phrases, the tropi, both of which were supposed to make a text more pleasant and easy to understand for the reader/listener. In addition there is the study of the topoi, which teaches how to use effectively the phrases, idioms, stereotypical themes and motives of a certain group of speakers. And finally, through the well-planned and skilled application of rhetorical rules, the stylistic level of texts is determined by binding it very closely to the kind of topic treated. Undoubtedly, rhetorical fashioning is a considerable contribution to the standardisation of medieval texts. One important feature of medieval texts — for all those involved in the process of communication — is an openness and flexibility which does not occur in modern times. For it is a feature of all medieval texts, even in the early period of printing, that they never kept any one immutable fixed appearance. It is true that the individual text was put down in writing at one point in time; in order to pass it on, however, it had in the Middle Ages to be copied by hand time and time again. In this process the original was handled very liberally and those who did the copying — the clerics — were able to insert their own ideas into the copy, changing it, correcting mistakes or adding new ones. Thus there may be good and bad copies of a medieval text, but each one really is an original — a fact that has often been overlooked by editors who are more concerned with archetypes and systematical filiation. Even the individual manuscript is not unequivocal, for medieval manuscripts do

Medieval

texts

171

not usually have structural aids like paragraphs and punctuation; and spelling mistakes in the texts were not just an exception. All this explains why there will always be differing philological interpretations of medieval manuscripts and why "final" editions of medieval literature are quite impossible. This is well illustrated by the differing application of the modern principles of editing to medieval texts. In any case, the medieval original is a much more open system of communicative signs than has been assumed so far. One must also keep in mind that wherever the ability to read and write may be suspected in the Middle Ages far more competence and skill was required than nowadays, e. g., the knowledge of different writing techniques and conventions. Thus a high degree of openness and indeterminacy is the major feature of the outward appearance of medieval texts. The pragmatic knowledge of contemporary readers and authors is likely to differentiate between texts for practical and everyday use and literature. Such a distinction cannot be expected from medieval readers/ listeners and authors. Literariness, aesthetic principles, versification, and the application of rhetoric prevail even in those kinds of medieval texts which nowadays would certainly not be counted among literature, let alone poetry. Keeping in mind the fact that a great deal of technical literature, for example, was metrical, the feature of fictionalization is also incapable of determining literariness proper in the Middle Ages. The same holds true for the wide area of everyday and utilitarian writing as well as for the moralistic, historical, and religious. It is, regarding texts, virtually impossible to propose a separation of the two areas of art and everday life for the Middle Ages, which implies that the positions discussed so far may be identified in both. It was only in the early modern period that a clear distinction in that respect was established. From what has been said so far the pragmatic framework for the medieval author can be determined, a framework not unlike that of the authors of classical antiquity. His achievement is to be seen entirely in the sphere of creative imitation. The medieval author could rely on motives, genres, ideas, and patterns of thought which were vested in tradition and passed down a wealth of information often conveniently made accessible in "florilegia" (collections of selected literary, philosophical, and theological treasures of the past). To use a picture created by John of Salisbury (c. 1115 — 1180) in his Metalogicon, the medieval author stood, as it were, on the shoulders of classical and Christian writers who, functioning as standard and paragon, allowed him a kind of bound liberty. So he is not the original creator of his works, since the act of

172

Heinz

Bergner

creation, according to prevailing doctrine, was a privilege of the power and providence of God only. The justification for the medieval author's productions lies in the authorities of the past, who are in possession of the truth — the result being tradition brought up to date. There is really nothing else to be expected, since the author as a private being did not amount to much. It was only through his close relation to tradition that he gained importance at all. In this context it is instructive to look at the beginnings of medieval works of literature, the introductory verses or phrases of individual texts, where the authors usually disclose their sources. These introductions are used to connect the present to the horizon of true knowledge from the past, thereby justifiying and elevating the present while apparently suspending time. In this process, assuredly, the century- old concepts of society are codified and stabilized while new pretensions are refuted. In comparison, the idea of an entirely autonomous and original author does not appear until well into modern times. Since the medieval author saw himself only as the mouthpiece of some objective truth — because for him the spiritual heritage of the past was of fundamental importance — it is easy to understand why medieval literature in particular keeps repeating the praise of things long past, the laus temporis acti. This point of view may explain why so many medieval works have come down to us anonymously: their authors must have felt they were only reviving things from the past. The principle of productive imitation tends to direct the author's activities towards phenomena like the variation of favourite patterns and modification of details, a new way of presenting things already read and known. Because of this, the Middle Ages are innocent of the concept of plagiarism as well as the modern concept of entirely subjective innovation. Another idea that is difficult to imagine in medieval literature is that of forgery. Medieval authors either reproduced their original in its entirety or made slight modifications. The same is true for historical documents, which were manipulated at will if this promised to help some supposed truth to come through. Although literary innovation in the Middle Ages generally occurred within a broad field of well-known conventions and only on a narrow basis, it must be considered a real advantage that there existed an allencompassing wealth of mutual knowledge in respect to the individual work of art between the author on the one hand and the listener or reader on the other. The communicative achievement of medieval texts must, therefore, generally be rated very highly because they are an expression of collective experience and behaviour.

Medieval

texts

173

If the literary tradition of the past must be considered an unquestioned obligation for the medieval author, other — equally axiomatic — restrictions become evident in the obligations and expectations tying his relationship to some patron, noble benefactor, or protector. Quite a number of medieval literary works were commissioned and even in the late Middle Ages the author's social dependence on his superiors is still relatively high. All this goes to show how deeply medieval texts were enmeshed in social and political affairs, a fact which must not be neglected when they are analysed and decoded by scholars today. The pragmatic framework for the small circle of those who were able to read in medieval times, moreover, was determined, as far as the understanding of texts is concerned, by the hermeneutical system of the "four-fold levels of meaning" (de Lubac 1959 — 1964; Brinkmann 1980). Initially, this system was applied to the proper interpretation of the Bible; very soon, however, to other religious and secular works, also. It postulates that in deciphering texts not only the literal meaning, the sensus litteralis, but also a sensus spiritualis or mysticus, a higher spiritual meaning should be observed. The last can be split into a sensus allegoricus, meaning according to the doctrine of salvation, a sensus tropologicus or moralis, a moral meaning and a sensus anagogicus, meaning as related to eschatology. This procedure, which seems to draw only inherent meanings from the medieval text but is not an individual interpretation, tries to achieve a universally guaranteed interpretation of the text; it proved a favourite among scholars and was applied frequently. There are, however, further pragmatical circumstances regarding the communicative environment of medieval texts, which, after all, the author produced with a certain audience in mind. In the Middle Ages only a small elite, whose international character is indicated by the common use of Latin, was privileged to benefit from education and a knowledge of books (Clanchy 1979; Mc Kitterick 1990). At that time the monopoly of education was in the hands of the Church, education generally being an instrument of power. The production of texts was thus limited to this small group of people, books and the written word were often enjoying an appreciation verging on the mythical. This circumstance explains why the ability to read and write was very unevenly spread throughout the Middle Ages and certainly never exceeded one per cent of the population. The vast masses of those living in the country were generally not able to read or to write, which partly explains the importance of images and pictures as substitutes for formal education. Within medieval nobility (often up to and including the king) such capabilities as reading and

174

Heinz Bergner

writing were often frowned upon but can be taken for granted to a certain degree for practical reasons among the limited number of the late medieval middle classes comprising the merchants and craftsmen living in the towns. It was only within the small group of clerics, particularly the old monastical orders, that the exclusive pursuit of reading and writing was widespread, although even there quite a number of exceptions to the rule can be found. The communicative environment of the medieval text is determined by forces which are difficult to understand for the modern reader, who is used to literacy and a flourishing book culture. Although medieval literature is fixed in written form and has been transmitted as such, the text at that time was mainly presented orally and received by ear rather than by eye (Scholz 1980; Zumthor 1984; Erzgräber — Volk 1988). The secluded reader, influenced by subjective hermeneutical impulses, does not appear until long into modern times. Depending on the genre and socio-cultural conditions the relationship between oral and written aspects of a medieval text may be quite irresolute and unhomogeneous. It is possible, however, to characterize the medieval text, which was generally meant to be heard rather than to be read, by a number of typical traits: the manifold addresses of the audience, the frequently rambling style, the profuseness of tags and formulaic phrases, the syntactic insecurity and clumsiness, the numerous and repetitive usage of figures of speech, and, finally, the ever-dominant verse with its inculcatory and mnemonic function. The medieval text as it has been passed down to us is, as it were, secondary since it must have sounded quite different when transmitted to an audience: it is quite likely that it was not really recited word by word, but that there must have been much use made of gestures and mime, its actual performance having a good deal of theatricality. What has been transmitted to us is certainly not identical with the text as it appeared when actually presented to a larger audience. The intensely auditory aspect of medieval texts also emphasizes the strong ties between literary works and medieval society. The audience must also have shown a considerable power of recollection, since, unlike nowadays, what had been said could not easily be called up again; the modern reader, in contrast, may simply read any given written text again and ponder it over. The pragmatic basis for medieval texts and the particular conditions governing communication between author and audience connected with it are fundamentally different from those prevailing in modern times. A considerable number of disciplines have been employed to illuminate those particular conditions, among them being linguistics, philology,

Medieval

texts

175

theology, archaeology, history, philosophy, art history, paleography, and editorial skills. An adequate comprehension of medieval texts is liable to be achieved in the light of the pragmatic insights sketched out above.

Notes 1. C o m p a r e , for example, volumes VII (Das Komische, 1976); X (Funktionen des Fiktiven, 1983); and XI (Das Gespräch, 1984) f r o m the series Poetik und Hermeneutik (München: Wilhelm Fink). 2. Especially Isidor of Sevilla (ca. 560 — 636) with his Etymologiae in twenty volumes. 3. E.g., Lat. malum 'disaster' and Lat. malum 'apple' (reference to the Fall of M a n in Genesis 1,3).

References Berger, Klaus 1984

Exegese des Neuen Testaments. Neue Wege vom Text zur Auslegung. (2nd edition.) Heidelberg: Quelle & Meyer. Boockmann, Hartmut 1988 Das Mittelalter. München: C. H. Beck. Borst, A r n o 1973 Lebensformen im Mittelalter. F r a n k f u r t : Ullstein. Brinkmann, Hennig 1980 Mittelalterliche Hermeneutik. Tübingen: Niemeyer. Chevalier, Jacques M . 1990 Semiotics, Romanticism and the Scriptures. Berlin: M o u t o n de Gruyter. Clanchy, M . T . 1979 From Memory to Written Record. England 1066—1307. L o n d o n : Arnold. Delort, Robert 1982 La vie au Moyen Age. (3rd edition.) Lausanne: Edita. Egger, Wilhelm 1987 Methodenlehre zum Neuen Testament. Einführung in linguistische und historisch-kritische Methoden. Freiburg: Herder. Erzgräber, Willi — Sabine Volk (eds.) 1988 Mündichkeit und Schriftlichkeit im englischen Mittelalter. Tübingen: G ü n ter Narr. Finke, Laurie Α. — Martin Β. Shichtman (eds.) 1987 Medieval Texts and Contemporary Readers. Ithaca: Cornell University Press. Glück, Helmut 1987

Schrift und Schriftlichkeit. Stuttgart: Metzler.

Eine sprach- und kulturwissenschaftliche

Studie.

176

Heinz Bergner

Goetz, Hans-Werner 1986 Leben im Mittelalter vom 7. bis zum 13. Jahrhundert. (2nd edition.) München: C. H. Beck. Güttgemanns, Erhardt 1973 Studia linguistica neotestamentica. Gesammelte Aufsätze zur linguistischen Grundlage einer Neutestamentlichen Theologie. München: Chr. Kaiser. 1975 "Die Bedeutung der Linguistik für die Religionspädagogik", Der evangelische Erzieher 5: 319 — 333. 1978 Einföhrung in die Linguistik fiir Textwissenschaftler. Kommunikations- und informationstheoretische Modelle. Bonn: Linguistica Biblica. Henry, Avril (ed.) 1986 The Mirour of Mans Saluacioun. Aldershot: Scolar Press. 1987 Biblia pauperum. Aldershot: Scolar Press. Jordan, Robert M. 1987 Chaucer's Poetics and the Modern Reader. Berkeley: University of California Press. Lausberg, Heinrich 1960 Handbuch der literarischen Rhetorik. Eine Grundlegung der Literaturwissenschaft. München: Max Hueber. Levinson, Stephen C. 1983 Pragmatics. Cambridge: Cambridge University Press. Lindsay, Wallace Martin (ed.) 1911 Isidori hispalensis episcopi etymologiarvm sive originvm, libri XX. 2 vols. Oxford: Clarendon. Lubac, Henry de 1959 — 64 Exegese medievale. Les quatre sens de l'ecriture. 4 vols. Paris: Aubier. McKitterick, Rosamond (ed.) 1990 The Uses of Literacy in Early Medieval Europe. Cambridge: Cambridge University Press. Miller, Jacqueline T. 1986 Poetic License. Authority and Authorship in Medieval and Renaissance Contexts. New York: Oxford University Press. Nuyts, Jan and Jef Verschueren (eds.) 1987 A Comprehensive Bibliography of Pragmatics. 4 vols. Amsterdam: John Benjamins. Ohly, Friedrich 1977 Schriften zur mittelalterlichen Bedeutungsforschung. Darmstadt: Wissenschaftliche Buchgesellschaft. 1979 "Skizzen zur Typologie im späten Mittelalter", Medium Aevum deutsch. Festschrift für Kurt Ruh zum 65. Geburtstag, in: D. Huschenbett (ed.) Tübingen: Niemeyer, 251—310. Ong, Walter J. 1982 Orality and Literacy. The Technologizing of the Word. London: Methuen. Quirk, Randolph — Sidney Greenbaum — Geoffrey Leech — Jan Svartvik 1985 A Comprehensive Grammar of the English Language. London: Longman. Romaine, Suzanne 1982 Socio-historical Linguistics. Its Status and Methodology. Cambridge: Cambridge University Press.

Medieval texts

177

Schlieben-Lange, Brigitte 1983 Traditionen des Sprechens. Elemente einer pragmatischen Sprachgeschichtsschreibung. Stuttgart: Kohlhammer. Scholz, Manfred Guenter 1980 Hören und Lesen. Studien zur primären Rezeption der Literatur im 12. und 13. Jahrhundert. Wiesbaden: Steiner. Schwarz, Alexander — Angelika Linke — Paul Michel — Gerhild Scholz William 1988 Alte Texte lesen. Bern: Paul Haupt. Sitta, Horst (ed.) 1980 Aufsätze zu einer pragmatischen Sprachgeschichte. Tübingen: Niemeyer. Vance, Eugene 1986 Mervelous Signals. Poetics and Sign Theory in the Middle Ages. Lincoln: University of Nebraska Press. Wienhold, Götz 1971 Formulierungstheorie, Poetik, strukturelle Literaturgeschichte, am Beispiel der altenglischen Dichtung. Frankfurt: Athenäum. 1972 a Semiotik der Literatur. Frankfurt: Athenäum. 1972 b " D E O R . Über Offenheit und Auffüllung von Texten", Sprachkunst 3: 285-297. Wirth, Karl-August (ed.) 1990 Pictor in carmine. Berlin: Gebr. Mann. Wolfzettel, Friedrich (ed.) 1990 Artusroman und Intertextualität. Gießen: Wilhelm Schmitz. Zumthor, Paul 1984 La poesie et la voix dans la civilisation medievale. Paris: Presses universit ä r e s de France.

The impact of sudden literacy on the text comprehensibility: Mohawk* Marianne Mithun

It has long been recognized that spoken and written language differ in important ways. Transcripts of spoken discourse are often more difficult to read than dialogue in a novel. Written texts can also be difficult to process when read aloud: few listeners absorb every point in a scholarly paper read to an audience. Do most of the differences between spoken and written language arise spontaneously with the shift to the written word, or are they primarily the product of a gradual adaptation of style to this medium? Do the changes actually improve the comprehensibility of written texts? One way to investigate such questions is to compare the structures of three varieties of texts in a single language: spontaneous spoken discourse, the first written texts, and the later texts of more experienced writers. Mohawk, an Iroquoian language indigenous to northeastern North America, provides an opportunity for such a comparison. Mohawk speakers have long been renowned for their rich oratorical traditions. Skillful use of language has always been appreciated and cultivated in a variety of contexts, formal and informal. This skill was described by the first Europeans to encounter the Mohawk, and is still very much in evidence today. Until recently, however, the language was not generally written by its speakers. Missionaries translated the Bible and other religious texts, and a few speakers wrote letters to relatives who

* The orthogoraphy used here is that now used in the schools at Kahnawa:ke. For the most part, symbols correspond to their IPA values. Obstruents are automatically voiced before sonorants. Mohawk contains two nasalized vowels, [ Λ ] written en, and [y] written on before consonants or word-finally. The apostrophe ' represents glottal stop. The vowel i represents the glide [y] before vowels and the usual high front unrounded vowel elsewhere. The acute accent ' represents high tone in short syllables and rising tone in long syllables, indicated by :. The grave accent, which occurs only in long syllables, represents falling tone. I am especially grateful to Margaret Edwards, of Ahkwesähsne, and Annette Jacobs, of Kahnawä:ke, for sharing their expertise about their language, and to the other speakers and writers whose work is cited in this text.

180

Marianne

Mithun

were away from the community, but there was otherwise little need for written language. More recently, people in various Mohawk communities have become alarmed to see that the younger generation is no longer learning the language. They realize that when a language is lost, a fundamental element of the culture it encodes is lost as well. Centuries-old ways of viewing the world disappear, as well as traditional styles of personal interaction. A number of people in these communities decided to develop language programs for their schools, to teach children as much as possible about their linguistic traditions. They began their work by devising a writing system. Their experiences with the early stages of literacy provide an opportunity to observe the immediate impact of a sudden shift to the written medium on the comprehensibility of texts. Their later work shows the kind of gradual refinement in style made possible by the new medium. In what follows, cues to text structure inherent in spoken Mohawk will be compared with those of the first written texts and then with those of later written texts.

1. Cues to textual structure in spoken Mohawk As in any spoken language, the shape of texts is signalled by a variety of devices, both phonological and grammatical.

1.1. Intonational cues Intonation is a powerful aid to text comprehension. Speech does not emerge in a continuous stream: it comes in bursts followed by rises or falls in pitch and separated by pauses. Each burst, or intonation unit in the sense of Chafe 1987, generally corresponds to a semantic unit: it usually presents a single significant new idea. The material in (1) comes from the opening of a legend told by Sonny Edwards of Akhwesahsne. Like the other spoken material cited here, it is arranged so that each line corresponds to an intonation unit. The first line introduces the old people, the second their storytelling, the third the protagonist. In the fourth we are given his name. (1)

Teharensähkhwa'*

Sonny Edwards

Literacy

181

Ne iä.ken' ne' ki: rotiksten'okonken.ha it is they say it is this they are old deceased 'It seems the old people ratikä.ratonhskwe' ne wahon.nise', they used to story tell the long ago used to tell stories in the old days, ne:'e ne iä.ken' ron.kwe about the they say man

ne.ne ohniare'ko.wa that great snake

roton.'on, he had become about the person who, they say, became a serpent. ne: nen Teharenhsähkhwa' ronwä.iatskwe'. this then he picks up beads they used to call him This person was called Teharenhsähkhwa' ('he picks up beads'). Commas in the transcription indicate a partial fall in intonation, and periods a full fall. The function of these comma and period pitch contours in communicating the structure of the text is as would be expected. Sets of lines are also characterized by intonational contours. Sets like the one in (1) above exhibit a kind of downdrift: successive lines are spoken on successively lower pitches. A new set of lines, comparable to a new paragraph in written language, usually begins on a new high pitch. It typically reflects a new direction of thought. The passage in (2) followed that in (1). (2)

Ni:io tsi rotikä:raton so it is that they have told 'The way they have told the story, iä.ken' ne:ne wahon.nise' ronhtenkie'skwe'. they say that long ago they used to go away it seems that would go away long ago. Ronnaterien.tare' tsi ηόη: nihonnehthahkwe' they knew to where there they used to go They knew where to go tsi ronto.ratskwe' so they used to hunt to hunt

182

Marianne

Mithun

a.ki.ron ronnitsiato:ratskwe' I should say they used to hunt fish I should say to fish, tanon' okario'tatshön:'a ni: o:ni'. and all kinds of game really too and get all kinds of game as well.' Intonation can structure texts more elaborately than punctuation of course, because of the continua of pitch, volume, speed, rhythm, and special phonation types available to speakers. The punctuation of written texts can provide only a weak imitation of a subset of them.

1.2. Grammatical cues Textual structure is not signalled by intonation alone; grammar plays a significant role as well. The grammar of Mohawk is in many ways radically different from that of more familiar European languages. A major difference is in the complex structure of words. 1.2.1. Lexical classes Three lexical classes can be discerned in Mohawk on the basis of internal morphological structure: particles, nouns, and verbs. Particles are by definition monomorphemic: they have no internal structure. Among the Mohawk particles are numbers, various deictics, conjunctions, adverbials, and numerous markers of the source and certainly of the information presented, as well as the speaker's personal involvement in the message. (3)

Particles 'this'; tsi: 'so, that'; tänon': 'and' ne\ 'the'; iä:ken'\ 'it is said'; d:ni': 'also' ki'•. 'just'; kwcik. 'so, really'; ähsen: 'three'

Nouns have slightly more complex internal structure. Most native nouns contain a pronominal prefix indicating the gender and number of their referent or of a possessor, and a nominal suffix. (4)

okä.ra' o-kar-a' NEUTER.PATIENT-story-NOMINAL.SUFFIX 'story'

Literacy

183

Derivational suffixes may also appear, such as the augmentative in (5). (5)

ohniare'ko.wa o-hniar- '-kowa NEUTER.PATIENT-snake-NOMIN AL.SUFFIX-AUGMENTATIVE 'serpent'

Verbs can be morphologically quite complex. There are no free roots: the root 'go' in the verb in (6) is -e-, but it would not be recognized by speakers in isolation. (6)

tentke' t-en-t-k-e-' DUALIC-FUTURE-DISLOCATIVE-l.AGENT-go-PUNCTUAL again-will-here-I-go-PUNCTUAL Ί will come back.'

In addition to the root, all Mohawk verbs contain a pronominal prefix referring to their agents and/or patients, like k- Ί ' in (6); or ronwa- 'they/ him' in (7). (7)

ronwä:iatskwe' ronw-iat-s-kwe' Μ ASCULINE.PLURAL. A G E N T / M ASCULINE.PATIENTcall-HABITUAL-FORMER.PAST 'They used to call him.'

Verbs usually contain an aspectual suffix, like the punctual -' in (6) or the habitual -s in (7). They typically contain additional affixes as well. In (6), the dualic prefix t- indicates 'back', the prefix en- marks future tense, and the cislocative t- indicates 'here'. In (7), the suffix -kwe' adds former past tense. Verbs may also contain incorporated noun roots, such as -kar- 'story' in (8). (8)

ratikä.ratonhskwe' ra-ti-kar-aton-s-kwe' Μ ASCULINE. AGENT-PLURAL-story-tell-H ABITUALFORMER.PAST 'they used to tell stories'

184

Marianne Mithun

1.2.2. Morphological class versus syntactic function The morphological classes of Mohawk words are well-defined and distinct from each other, but they are not isomorphic with syntactic categories. Morphological verbs in particular serve a wide variety of functions. They can of course serve as predicates, like rotoni'on '(he) had become'. They also frequently serve as syntactic nominals, like those in (9) from the Teharensähkhwa' text. (9)

ronatin.ron ron-at-enr-on M.PL-RECIPROCAL-friend-STATIVE 'They (m) are friends.' > 'his friend' rotiksten 'okonkenha' ro-ti-ksten- 'okon-kenha MASCULINE.PATIENT-PLURAL-be.old-PLURALDECESSIVE 'The deceased old people.'

Verbs can serve as modifiers, much like adjectives in some other languages. (10)

ni.ra ni-hr-a PARTITIVE-M-be.a.size 'He was big.'

They can function as adverbials. (11) teiotenonhianihton tsi nihahetken. te-io-te-nonhiani-ht-on tsi ni-ha-hetk-en DU-it-SRL-afraid-CAUS-STAT so so-he-ugly-STATIVE it was horrible so he was so ugly 'He was so horribly ugly.' All Mohawk verbs are finite: all contain aspect markers and pronouns referring to their core arguments. For this reason, any verb can stand alone as a grammatical clause in itself. Additional nominals are not necessary for grammaticality. In fact, typical discourse often consists primarily of verbs. The sentence in (12), for example, consists of three morphological verbs.

Literacy

(12)

185

V V V Teiotenonhianihton tahiaton'neke' tehottenion. it was spectacular they two were startled he had changed 'They were spectacularly startled at how he had changed.'

Particles are also frequent in spontaneous speech, but nouns are relatively rare. Many concepts that are expressed in European languages with nouns are expressed in Mohawk with verbs. For 'at dawn', for example, Mr. Edwards used a verb: wa'orhen'ne' 'it dawned'. Ideas expressed by separate nouns in European languages are often carried by noun roots incorporated in Mohawk verbs. Text counts of spontaneous English discourse typically show a verb:noun ratio of 1:2 and 1:2 (Chafe p.c.). Counts of spontaneous Mohawk discourse of similar genres often show a verb:noun ratio of as high as 5:1. Note the lexical categories in the continuation of the Teharensähkhwa' tale in (13). (V indicates morphological verbs, Ρ particles.) (13)

V Wa'orhen'ne' it dawned 'They started

Ρ V V ki' orhon'kehstsi tahontähsawen just it just dawned they started early in the morning,

Ρ Ρ Ρ iä:ken' ähsen niha.ti. they say three so they number so they say, the three men. V P V Tahontähsawen iä:ken' wahonnitsiato.rate' they started they say they fish hunted They started, it seems, to fish, Ρ Ρ Ρ Ρ V V ne id:ken' kwah iah thahd.ten tehotitsiaientd:'on the they say just not not any did they fish come to have but they did not catch a single fish Ρ Ρ V V thi tsi niiohniseres wahonnitsiato.rate'. that at so it is a long day they fish hunted all day long.

186

Marianne

Mithun

V Kon'täkie's ... it day goes along All day long Ρ tshi at they

Ρ V ken ratihonionkie's ... here they are boating along were trawling,

Ρ Ρ V tsik ηόη: ronnitsiatoräthons, so there they were fish hunting fishing, Ρ kwäh ... just but Ρ V iah tehonnatera'swiiö:hon. not had their luck become good they had no luck at all.'

1.3. Grammatical structure and textual structure Structural differences like these between Mohawk and more familiar European languages can have major ramifications for the ways in which texts are understood. 1.3.1. Particles Particles not only serve their literal semantic functions, they can also play significant roles in structuring discourse. Note the uses of the particle iä:ken' 'it is said' in (1). The literal meaning of the particle is that the source of the information is hearsay. If this were its sole function, however, one mention in the first line should be sufficient. Yet it appeared again in the third line and again three lines after that in (2), as well as several times in the passage that followed in (13). The particle partitions the text into meaningful segments that can be longer than intonation units:

Literacy

187

the old people used to tell stories there was a person who became a serpent people would go away to hunt and fish three men started out one morning they went fishing they didn't catch a single fish Mohawk particles also play a role in regulating the flow of information. At one point one of the fishermen saw a log. It would have been grammatical to present this event with just a verb and a nominal, the first and last words in (14) below. Instead, it was presented in three intonation units, as can be seen from the arrangement of the lines. The speaker introduced each significant new piece of information separately. The particle ki: 'this' functioned in the first line as a place holder, announcing that what was seen would be explained subsequently. (14)

Wahatkähtho' he saw 'He saw

ki:, this

kerhite', tree stands a tree, ioronkenen: 'en. tree has fallen a fallen log.' Deictic particles like ki: 'this' and thi: 'that' can serve a further orienting function, establishing perspective. The proximate particle ki: 'this' in (14) set the log at center stage, appropriate since it would play a central role in the ensuing discussion. Particles can also serve to link segments of text. A phrase like that in (15) 'and just at that moment' closely links the event that precedes to what follows. (15)

Ne ki' thi.ken the just that 'And then,

The Mohawk demonstratives differ from their English counterparts syntactically. English 'this' and 'that' typically serve as determiners, combining syntactically with nouns to form noun phrases. The Mohawk equivalents do not have the same structural function. They may appear alone or with nouns, but when they do occur with nouns, they function more as appositives than as determiners.

188

Marianne Mithun

1.3.2. The backgrounding effect of morphologization The elaborate morphological structure of Mohawk functions powerfully to add texture to texts. A major effect of morphologization is to background information. Bound morphemes are not processed as separate pieces of information, worthy of individual attention. Rather, they are integrated with the other elements of the word of which they form a part, processed as part of a conceptual unit. Any information that can be expressed in Mohawk by means of an affix or incorporated noun root can also be expressed by a separate lexical item. The two kinds of expressions have subtly different functions. A separate word for 'again' was used several times in the Teharensähkhwa' tale, for example. (16)

rohnekenta:'on khale' iä.keri sahonia'täthen'ne'. he has drunk again they say again his throat dry became Having just drunk, he was thirsty again.

(17)

Sok ä:re' atsiä:kta nionsä.re' so again shore along there back he walked So he walked back again along the shore.'

The main point of (16) and (17), was the fact that the protagonist was thirsty a second time. It is thus worthy of a separate word: khale' 'again'. Mohawk also contains a repetitive prefix s- with much the same meaning: 'again, back, re-'. In (18), the protagonist explained that he would go back to the water again. The fact that he had been there before was indicated by the repetitive prefix. It was not the main point of the utterance, however, so a separate word was not used. (18)

Awen.ke ni: nienhenske'. water in I there will I again go Ί will return to the water.'

A similar difference can be seen in the use of noun incorporation. The first time the fish was mentioned in (19), it was the focal point of the utterance, so it was expressed as a separate noun. The second time, it was already established, old information, so it was incorporated. The main point of that sentence was the fear. (19)

"Kentsion se'" wahen.ron "se' wähe'." fish only he said only isn't it ' "It's only fish, isn't it", he said.

Literacy

O.nen ki: ronaten.ron wahatshä:nike' now this they are friends he feared But his friend was afraid

189

ne the

a:hentsiake'. he would fish eat to eat the fish.' 1.3.3. Pragmatic word order Texture is provided in Mohawk discourse by another powerful device: word order. Because all Mohawk verbs contain pronominal prefixes referring to their core arguments, primary grammatical relations are specified within the verbs themselves. Word order is thus not used to express syntactic relations between predicates and arguments. Instead, it is used for purely pragmatic purposes. The most significant information appears early in clauses, followed by more predictable or peripheral information. Compare the orders of the noun for "fish" and the verb "be in" in the two sentences below, both from the same tale. In the first, the noun "fish" precedes the verb, but in the second, it follows. (20)

Khiahatkähtho' iä:ken' there he looked they say 'He looked, they say, NP VP kentsion ratiiä.ti ki:. fish they are in this and saw fish in there.'

(21)

wahsheia'tatshen:ri' he body found them 'He found VP NP karon.takon kontiiä.ti ketsion. it is log inside they are in fish the fish inside a log.'

In (20), the fish are introduced for the first time. They represent the most newsworthy information of that clause. (21) followed a passage in which we were told that the protagonist was looking for the fish. Their location was thus the most newsworthy element of that clause, followed by the

190

Marianne

Mithun

fact that the fish were actually there. The identity of the fish themselves had already been established. A similar difference in ordering can be seen by comparing (22) and (23). The function of (22) was to reintroduce the friends. The nominal referring to the friends thus appears early in the clause. (22)

N P

VP

ki: ronnaten.ro's tehniiähsen thonateharä.ton this they are friends they are two they were waiting 'Those two friends were waiting.' The main point of (23) was that the friends, already at center stage, noticed something. The verb 'notice' appears early, and the nominal referring to the friend, late. (23)

VP O.nen wahiättoke' kü: now they two noticed this ' N o w his two friends noticed

NP ronnatin.ro kio.ts ... they are friends gee (what was happening to him).'

Spoken Mohawk thus contains a variety of cues to the structure of texts, some, such as intonation, similar to those of more familiar languages, but others, such as the use of morphologization and pragmatic word order, quite different. W h a t happens when the medium changes from speech to writing?

2. Early written Mohawk A group of dedicated Mohawk speakers at Kahnawa:ke, Quebec, gathered nearly twenty years ago to develop a program for teaching the Mohawk language at school. A first step in the program was the establishment of a practical orthography, a task that was accomplished the first summer. Soon afterward, teachers began to write. Their early written texts differed surprisingly from their spoken discourse. The language was fully grammatical, but surprisingly un-Mohawk in many of the usual cues to textual organization. The shift in medium did indeed have an immediate effect. A sample of dialogue from one of these texts is in (24). (The original was written in paragraph form so the division into lines here is not significant.)

Literacy

(24)

191

Early written dialogue Haki wä:s seniion.ten sanön.warore' tanon satia.tawi OK go you hang your hat and your coat 'Wont't you go and hang your hat and coat tanon serihsi sä.wen and you take off your and take off your rubbers

teionrahtahkwaneta's it goes over shoes

thd.ne o.nen iahä:sko sä:wen ionrahsi'tohrokstha'. then now you get your it covers foot then get your slippers.' Tesarenia't ki.ken kahiatonhsera you pass this picture Pass out this picture of a horse.'

akohsä:tens horse

raiä:tare. his picture

A sample from a children's story is in (25). (25)

Children's story Thv.ken eksa:'a wa'i.ron, "Sarahthähsi enhen.ton that girl she said you shoe remove first 'The little girl said, "Take your shoes off first tsi niiö.re entehsatäweia'te'", raksä.'a ki: wahonwahro.ri. this before you will enter boy this she told him before you come in," she told the little boy.

Both passages differ in a number of ways from the usual spoken Mohawk of these writers.

2.1. Particles Many of the particles that are so pervasive in spoken Mohawk are conspicuously absent from the early written Mohawk texts. Evidentials like iä.keri 'it is said' seldom appear. Emphatic particles like ki' 'just'; kwäh 'just, so, very, really', etc. are also rare. Demonstrative particles like ki:(ken) 'this' and thi:(ken) 'that' do appear frequently in the written texts, but not to regulate the flow of information. In spoken Mohawk, nouns do not require determiners, although certain particles may appear with them, such as demonstratives, or a particle ne that indicates that a nominal refers to given information.

192

Marianne

Mithun

The early written texts, however, contain constructions parallel to English noun phrases. Note the use of ki.ken 'this', thhken 'that', and ne before nouns in the written texts in (24) and (25). In spoken Mohawk, possession is normally shown by a pronominal prefix, as in sandn:warore' 'your hat'. A somewhat unusual construction sometimes appears in the written texts, apparently modeled after the use of English possessive pronouns as determiners. Note the use of Mohawk sä:wen 'you have' for 'your' in (24). Some conjoining particles appear more regularly in the written texts than in spoken language. Overt marking of conjunction is not obligatory in spoken Mohawk. Coordinate constituents, especially conjoined predicates or clauses, are often simply juxtaposed, linked intonationally. The overt marking of conjunction is much more frequent in written texts, such as the use of tänon 'and' in (24). The early written texts thus contain particles, but they function differently from those in the spoken language. Rather than supplying emotive texture or structuring texts, they tend to appear in those contexts where their English counterparts would be required by English syntax.

2.2. Nominals Recall that spoken Mohawk is characterized by a high proportion of verbs to nouns. The written texts exhibit more nouns than their spoken counterparts. Nominals are often more complex. Not only do they often contain demonstratives, as noted above, but they are often compound. Note the conjoined nominal from (24): seniwn:ten sandn.warore tanon' you-hang your-hat and 'Hang up your hat and coat.'

satiä.tawi your-coat

A speaker might be more likely to convey the same message in a single verb: (26)

satstahsion.ko you-self-dress-un 'Take off your things.'

Literacy

193

2.3. Incorporation Recall that in Mohawk, nouns representing given or incidental information can be backgrounded by incorporation into the verb. Incorporation for such stylistic purposes is rare in the early written texts. The incorporation that does appear consists primarily of established lexical items: the standard idiomatic words used for certain recurring concepts. (27)

satiä:tawi sarahthähsi ronähskwaien

'your-self-body-encircle' 'you-shoe-on-un' 'he-pet-has'

> 'your coat' > 'take your shoes o f f > he has a pet'

2.4. Word order Recall that word order in spoken Mohawk is used for pragmatic purposes; elements are ordered within clauses according to their significance to the discourse at hand rather than their syntactic role. The early written texts exhibit a surprising tendency toward SVO order. Note the order of constituents in both (24) and (25). Each line of (24) contains a VO clause. The first line of (25) contains an SV clause, SVO if the quotation is counted as a complement. These orders are not ungrammatical, but pragmatically unusual.

2.5. Factors behind the differences The shift from the spoken to the written medium had an immediate effect on some of the cues to text comprehensibility normally found in spoken Mohawk. Most of the traditional cues disappeared. Not only was intonational structuring lost, but also the stylistic use of particles and the pragmatic use of noun incorporation and word order. Several factors probably contributed to these changes. One is the influence of the English literary tradition. Although Mohawk is the mother tongue of all of these writers, all are literate in English. Apparently the act of taking pen to paper somehow called forth the English written style they already knew. Their English literacy explains in part the rarity of particles without English counterparts, the increased use of determiner constructions, the higher proportion of free nouns, the reduction in noun incorporation, and the tendency toward SVO word order.

194

Marianne

Mithun

The nature of the medium itself also prompted some instantaneous changes. As pointed out by Chafe 1982; Ong 1982; and Tannen 1982 among others, speakers address a seen audience that shares their local and temporal context. Spoken language is typically characterized by greater personal involvement of the speaker, showing more personal reference, more emphatic particles, and reports of the speaker's mental processes. The early Mohawk writing is characterized by a sharp reduction in the particles that convey those meanings. Because these writers had had little exposure to written Mohawk, spelling was not automatic. The spelling system is a good one, but writers had not memorized the spelling of most words in the way that experienced writers of other languages do. The spelling of nearly every word had to be invented as it was written. The written texts were produced one word at a time. Writers had smaller chunks of language within their consciousness as they wrote than speakers have when they talk. It is not surprising that traditional cues to larger discourse structure were absent.

3. Later written Mohawk Even under these conditions, writing proved valuable. The ability to write Mohawk made it possible for teachers to see structures in the language they had not been aware of before. They became more conscious of the richness of the language they speak so effortlessly. Writing also made it possible for them to plan systematic language classes. As time passed, and writers became more proficient in the system, they developed a rich, specifically Mohawk literary style. Note the opening to the legend in (28) written by Josephine Home. Mrs. Home also wrote the translation. (28)

Iakotinenioia'ks

'The legend of the little people' (Josephine Home) O.nert

ki'

ni.'i

nakkä.ra'

now just mine my story 'Once upon a time Wahon.nise'

skahwatsi.ra,

long ago one family there was a very happy

onkwehon:we

real person

ron.ne'skwe,

they used to roam

Literacy

195

tiötkon ronatonnhähere'. always they were happy Indian family. Ki.ken ne ronwa'niha ratoratsheraweienhen this the their father he knows how to hunt The father was a good hunter iäh nowen:ton not the ever

tehontonhkäria'ks are they hungry

Ki.ken ne shakoti'nistinha this the their mother and the mother

ne raohwä.tsire'. the his family

ieweienni.io she knows well

iakokhonniähtskon she likes to cook worked very hard tanon tiötkon ionhkwenniön.ni raonawenhshön.'a. and always she clothes makes their various things making clothes Kanehon tanon öwhare' iöntsha skin and fur she uses and blankets from the animal skins ne o:ni' iakonniaAha ne raonahsire'shön.'a. the also she makes with the their blankets and using the mat for cooking nourishing meals.' This later style shares some of the features common to most written languages, but it also contains some distinctively Mohawk qualities.

3.1. Particles Most kinds of particles are still less abundant in this written style than in normal spoken Mohawk. They are seldom used to manipulate the rhythm of information flow. They seldom indicate intensity or emphasis. They do sometimes serve an orienting function. The proximate particle ki.ken 'this' in the third and fifth lines of (28) first places the father in center stage, then the mother.

196

Marianne

Mithun

As in the earlier written style, particles are still used to mark syntactic structure more frequently than in spoken language. Demonstratives appear more often in a determiner position before nouns. Conjunctions appear more often to link clauses. Note the overt traffic signals provided by the particles in (29) from the Iakotinenioia'ks story. (29)

Ne: tsi seksa'ti:io tanon iäh tesathrö.ri tsi because you are a good child and not did you tell that 'Because you have been so nice to us and wahskwä:ken, ne: käti' enkönnion ähsen you us saw the why will I you give three kept our secret, niwaskanektsherä.ke so one wishes number I will give you three wishes.'

3.2. Nominale This later written Mohawk still exhibits a somewhat higher proportion of nouns than is usual in spoken Mohawk. The nominals themselves are also often more complex. Note the conjoined nominal kanihon tanon owhare' 'skin and fur' in (28) for example.

3.3. Incorporation The later written Mohawk texts often show more morphological elaboration than earlier written material. Part of this increased morphological complexity was a return to incorporation for pragmatic purposes. The passage in (30) occurs later in the same story by Mrs. Home. In the first line, the soup is presented as a separate noun. In the last line, the soup, now an established part of the scene, is expressed by an incorporated noun root. (30)

karihstote' e' tho kana'tsähere' karihstä:ke on0n:tara. it metal stands there it pot stands it meat on soup O n the stove was a pot of soup.

Literacy Wa'ontathon.karon

197

Katsi'tsaro:roks

they her invited she flower gathers Gathering Flowers was invited to sit down tahon:

tonte'.

they would eat together and eat with them. Wa'enontararonnion

ne

iakon.kwe.

she soup served the she person The woman served the soup.'

3.4. Return to more pragmatically based word order Finally, the later polished written style generally displayed the pragmatic use of word order typical of spoken Mohawk. The point of the second line of (28), for example, is to introduce the family; the noun for the 'family' appears early, before the verb 'roam'. The point of the fourth line is to shift attention to the father; he is named at the beginning of the clause. In the fifth line, by contrast, the family is already an established part of the scene, but the newsworthy part of the line is their satisfaction; the predicate appears early and the noun for 'family' appears late.

3.5. Factors shaping later written style Some of the forces affecting the earlier written style continue to be felt in this later style. The inherent distance between writer and reader still results in a rarity of evidential and emphatic particles. The influence of written English probably still contributes to the use of some syntactic particles, the abundance of nouns and complex nominal constructions, and some of the more complex syntactic structures. Early in the project, when writers were struggling with the new orthography, attention was focused on one word at a time, so many of the traditional cues to Mohawk textual structure disappeared. As writers became more proficient in the new medium, however, they were able to take advantage of the extra time at their disposal. Since writers are not subject to the same constraints of attention and short term memory as speakers, information can be more tightly packed in written texts. More morphologically elaborate vocabulary can be selected, clauses can contain

198

Marianne

Mithun

more constituents, many of these complex themselves, and sentences can be more elaborate syntactically, containing more overt coordination and embedding. Of course since intonation is not available to guide the reader through this more elaborate structure, particles appear more often to mark these structures overtly. In addition to allowing experienced Mohawk writers to pack information more densely into texts, the luxury of time permitted them to recover many of the devices for structuring text inherent in the spoken language, in particular, morphological backgrounding of established or incidental information, noun incorporation, and pragmatic word order.

4. Conclusion Differences between spoken and written Mohawk are the result of several related factors. As with any language, the shift to the written medium produces certain effects immediately. The writer is further removed from the audience, so reference must be more explicit; pronouns and deictic particles must be replaced by full nominals. The less intense personal engagement results in fewer emphatic and evidential particles. Much as Latin literary traditions affected vernacular European writing, English literary traditions influenced Mohawk written style. Stylistic devices that had evolved over centuries were transferred immediately to the new written language. The Mohawk writers were accustomed to reading and writing English, and many aspects of this style appeared spontaneously as soon as they began writing Mohawk. This influence probably also contributed to the increased use of particles where English would show determiners or conjunctions, the loss of Mohawk discourse particles with no English counterparts, the higher proportion of nouns, perhaps the reduced use of incorporation, and the tendency toward SVO word order. Finally, early writers, struggling with the spelling of every word, could devote less attention to text structure than speakers. Probably for this reason as well, they made less use of discourse particles, of the backgrounding function of morphologization, and of pragmatic word order than they would in speech. The combination of these factors, along with the loss of intonation, resulted in the disappearance of many of the traditional cues to textual organization from early written texts. As writers became more experienced, however, some of these cues reappeared, and certain advantages

Literacy

199

of the written medium were exploited to develop a uniquely Mohawk written style. With the luxury of time, writers could pack information densely, but also select just the right words, often words of considerable morphological complexity, and take advantage of the special pragmatic effects of incorporation and word order offered by Mohawk. Processing written Mohawk texts will probably never be easy for most readers. Few people spend their days surrounded by the written Mohawk word. Breakfast tables may never sport cereal boxes covered with written Mohawk, and daily newspapers in Mohawk may never be delivered to the door. Many Mohawk words are long, difficult to take in at a glance. Because a single word can be so complex, many words recur somewhat less often than the shorter words of some other languages. Each of these factors plays a role in the relative ease of word recognition. Nonetheless, the written medium is in many cases providing the luxury of time to writers to develop a rich, distinctively Mohawk style, full of many of the special devices unique to that language. * The orthogoraphy used here is that now used in the schools at Kahnawa:ke. For the most part, symbols correspond to their IPA values. Obstruents are automatically voiced before sonorants. Mohawk contains two nasalized vowels, [ Λ ] written en, and [Y] written on before consonants or word-finally. The apostrophe ' represents glottal stop. The vowel i represents the glide [y] before vowels and the usual high front unrounded vowel elsewhere. The acute accent ' represents high tone in short syllables and rising tone in long syllables, indicated by :. The grave accent, which occurs only in long syllables, represents falling tone. I am especially grateful to Margaret Edwards, of Ahkwcsahsne, and Annette Jacobs, of Kahnawä:ke, for sharing their expertise about their language, and to the other speakers and writers whose work is cited in this text.

References Chafe, Wallace 1982 1987

"Integration and involvement in speaking, writing, and oral literature", in: D. Tannen, (ed.), Norwood. New Jersey: Ablex. "Cognitive constraints on information flow", in: Russell S. Tomlin, (ed.), 21-52.

Ong, Walter J. 1982 Orality and literacy. London: Methuen. Tannen, Deborah (ed.) 1982 Spoken and written language: exploring orality and literacy. Norwood, New Jersey: Ablex. Tomlin, Russell S. (ed.) 1987 Coherence and Grounding in Discourse. Amsterdam: John Benjamins Publishing Company.

Scientific texts and deictic structures* Konrad Ehlich

1.

Some characteristics of scientific texts

1.1. Scientific texts and scientific knowledge "Scientific texts" (ScTs) form a specific text type. They underlie pragmatic constraints due to the aims of production and reception. ScTs are dependent upon the linguistic activities of writers and readers who share a large amount of highly specified knowledge. Their common participation in a discourse community can be defined by: (a) their institutionally determined common knowledge and (b) a shared stock of methods of knowledge expansion. The institutionally determined common stock of knowledge (a) can be termed "material scientific knowledge"; the shared stock of methods of knowledge expansion (b) can be termed "accepted methodology". The acquisition of (a) and (b) is accomplished by a highly specified process of institutional socialization which covers years of the knowledge acquirers' biographies. The text type ScT comprises a variety of specific text sorts which can be characterized as subtypes of the text type ScT. Well-known examples of subtypes of the text type ScT are text books, readers, articles, papers, scripts, monographs, atlasses etc. Each of these text types have their own aims to fulfill in scientific communication. These tasks determine their specific features. Other features are shared by all of them, features characterizing the text type ScT in general. These features probably have to do with the above mentioned pragmatic constraints. The development of science has had direct influences on the features of ScTs, in their various aspects. To illustrate this with at least one example, regarding the aspect of formulation (cf. Antos 1982) or "wording" of scientific texts, let us take a closer look at quantifying expressions.

* The author thanks Bill Elmer (Basle), Stephan Schlickau (Dortmund) for their help with the English version of the paper.

202

Konrad

Ehlich

The use of quantifying expressions in ScTs will tend to be realized by some specific selection of elements of the set of quantifiers. The application of the quantifier "all" is favoured. Other quantifiers such as "some" or "many" are disregarded. The definite article will be applied in its generic sense. In some sorts of ScTs, the artificial logic sign 'V' will fully replace the everyday expressions. All of these forms of wording reflect properties of societal knowledge generation which is to be achieved by science. It has found its expression in the generation of specific linguistic forms of knowledge, such as "laws", "hypotheses", etc. The use of other parts of the semantic potentials of the linguistic expressions being used for ScTs is negatively marked or qualified as being "inappropriate" or "not scientific". The reflection of the pragmatic constraints in linguistic structures is part of what result^ in the scientific texts' "abstract" character or "abstractness". Such characteristics are not only dependent upon and a consequence of the structures of material scientific knowledge (a). They are also dependent upon and a consequence of the scientific ways of how knowledge "has to be" and, consequently, is expressed. These ways are part of the accepted methodology (b). To include them, the concept of accepted methodology, then, is to be understood in a broad sense, also referring tho ways of speaking, next to methodology in the narrow sense of the term. The ways of expressing knowledge are argumentatively elaborated in, generalized for, and shared by scientific communities such as "schools" or "disciplines". For example, in some such schools "formalisms" are regarded a necessary feature of a scientific way of expressing knowledge. For them the capability to express newly found knowledge by use of expressions from a well-established set of formulas, following rules for their combination, is a necessary prerequisite for the acceptance of a suggested piece of knowledge as knowledge. The ways of transferring knowledge in words (or in subsidiary types of expressions, as number, symbols, etc.) are the result of at least two determinants: a determinant of matter and a determinant of communicative habit. Both determinants change in the history of disciplines, especially in the context of paradigm changes (Kuhn 1963). The result of these two groups of influences is what might be termed a "generalized style of scientific texts". It can be analyzed by empirical descriptive stylistics (cf. Rehbein 1983; Ehlich 1983 a; Spillner 1974). In speaking about "science" and "scientific" in the context of ScTs the two terms are taken to cover not only "natural sciences", but also "the

Scientific

texts

203

humanities" or the social and historical disciplines. The terms are used in the sense of French sciences de la nature et sciences de l'homme, or in the sense of Wissenschaft and wissenschaftlich in G e r m a n .

1.2. The analysis of ScTs Empiric analysis of ScTs is an important demand on linguistics, in several respects. It is important for: — — — — —

the development of pragmatic text-linguistics; the development of stylistics; training learners of scientific writing; teaching foreign scientific languages; comparative analysis of scientific literature belonging to different scientific language cultures, such as the Anglo-Saxon, the French, the G e r m a n , the Russian ones, etc.

Whereas the importance of some of these tasks stands beyond doubt (such as training skills in mother tongue and in foreign languages) others are less well accepted (such as the comparative analysis of scientific texts from a crosscultural perspective (cf. Clyne 1980; 1987; Galtung 1982; Feigs 1990)). As I see it, the relevance of research into the characteristics of ScTs for analytical and applicational issues is of m a j o r importance; but the analytic means at the linguists' disposal are not very well developed so far.

1.3. Scientific texts and literacy Some further aspects of "text" with regard to the distinction of orality and literacy which need to be taken into consideration for understanding the ways of how deixis can function in texts are to be specified next. The text type " S c T " nowadays usually appears as written (or printed or computerized) text. Only in very early stages of knowledge formation precursors of today's sciences m a d e use of oral text forms for knowledge conservation and knowledge expansion. The Graeco-European tradition of the sciences is largely dependent upon the existence of written text forms, and the formation of these text forms for writing/reading is a fundamental part of the organization of knowledge in this tradition (cf.

204

Konrad

Ehlich

Goody 1983; Ehlich 1983 b). The literality of ScTs makes them share general properties of written texts. Written texts have a major impact on the development of linguistic form in general. The main characteristics of "text", as opposed to "discourse" or single speech actions, is its separation from the "speech situation" as a fundamental linguistic unit. The text's coming into existence is due to the communicators' need to overcome the boundness of linguistic activity to speech situations. The systematics of texts as specific forms of linguistic activity is their unbinding or dissolution from speech situation (Sprechsituationsentbindung). The result of the interactors' departure from the common interactional ground of speech situation is a "dissolved speech situation". The speech situation common to speaker and hearer is, as it were, split up into two halves, each of which is incomplete in itself. The text is the intermediation between these two halves of the split speech situation.

2.

Deixis: from speech situation to text

2.1. Deixis and the deictic field One of the linguistic devices which are of importance for linguistic actions are deictic expressions. Their characteristics have been a major topic of linguistic analysis during the last decades. The most promising way of analysis is Karl Bühler's distinction of a "deictic" and a "symbolic field" in language (Bühler 1934 [1965]: part II). 1 Bühler demonstrated that deictic expressions differ essentially from most other types of expressions and that their analysis necessarily has to take into consideration linguistic activity in order to understand how deictic expressions function. The analysis of the "meaning" of deictic expressions, Bühler showed, can only be achieved by referring to their use in speech situation.

2.2. Speech situation, sphere of perception, and deixis The speech situation as a sphere common to speaker and hearer, as it is characteristic for ordinary linguistic activity, is of great importance for the organization of the psychological activities of the interactors. The speech situation in itself is part of the common sphere of perception. This common sphere of perception makes coordinated perception of parts of

Scientific

texts

205

it possible to both parties. Both parties can focus their attention on perceivable objects, etc., lying within their common sphere of perception. Speakers exploit the common sphere of perception for their performance of linguistic action. They do so by using the linguistic sub-system of deictic expressions. The sub-system of deictic expressions makes direct use of the speech situation for the processing of linguistic activity. It thus transfers properties of the common sphere of perception into the system of language. It is an important interface between perceptional and linguistic activities. The linguistic activity is used to organize perceptional activities, which is a specific type of mental activity. By the use of deictic expressions the speaker induces the hearer's organization of his/her attention in their common sphere of perception. He/she orientates the hearer's attention to entities in the — mentally represented — sphere of perception or to categories inherent in it. The orientation of attention is a permanent, on-going activity which corresponds to and follows a variety of demands. Some of them are brought about by the linguistic activity as such. The linguistic activity is in itself a specific form of senseperceptual activity. It demands specific ways of organizing attention. Ongoing speech as a highly complex type of activity centers the individual hearer's attention onto itself. This fact has been clearly identified in Karl Bühler's concept of the speech activity as origo for the application of deictic expressions (Bühler [1965]: § 7). Deictic expressions are used in "deictic procedures" (Ehlich 1979). Procedures are subforms of linguistic activity most of which enter into larger forms of linguistic activities such as illocutionary acts or propositional acts (cf. Ehlich 1986). By the application of a deictic expression in a deictic procedure, being part of a linguistic act, the speaker who processes the deictic procedure influences the mental orientation processes of the hearer by "pointing" to some item in the perceptually accessible common sphere of speaker and hearer, thus identifying it and making it salient for the hearer so that the hearer focusses his/her attention to it. The use of deictic expressions in deictic procedures is a highly economic and a highly efficient way of communication. This holds true as long as the common ground of reference, the sphere of perception which is accessible to the speaker's and hearer's senses, is present. The use of deictic expressions, which seems to be universal, is as effective and economic as it is because of its boundness to the origo, i. e., its boundness to the single speech action and its "environment", its immediate " Umfeld", with speech production and reception as parts of the overall sense activities of the communicators. Deictic expressions and speech action thus

206

Konrad

Ehlich

have an intimate, systematic interrelationship. Should this interrelationship be interrupted, severe problems will arise. Exactly this is the case as soon as "text", in the theoretical sense of the term as it has been described in section 2.2., enters into the picture. For texts, a systematic separation of linguistic activity from the common perceptual ground of speaker and hearer takes place. The unity of the common speech situation is dissolved, a dissolved speech situation, with its two halves, and the text as the only link between them, comes into being. This separation of linguistic activity from the common perceptual ground of speaker and hearer consequently has a detrimental, if not fatal effect on the whole deictic sub-system. The efficiency of the sub-system as a means of intermediation between speaker and hearer is at stake if this common ground is destroyed. N o similar effects are to be expected for other subparts of language, such as the "symbolic field" of Bühler's (Bühler [1965]: part III), being a "perception-detached" system of independent quality with regard to the "origo" and the speech situation. The inherent features of texts themselves, then, lead to severe problems for the deictic sub-system of language. The sub-system's positive qualifications for single speech actions, making it crucial and indispensable for linguistic activity, is completely reversed as soon as texts are involved. For texts, the system of deictic expressions and their application in deictic procedures as it had been developed for efficient and economic linguistic interaction, becomes aimless, and consequently, useless — unless new aims and new uses can be established for the application of deixis in a context which differs fundamentally from the original one. The system is set free for new applications in the context of dissolved speech situations.

2.3. Attentional preciseness and vagueness in deixis Whereas deixis in speech situation and deixis in text seem to be juxtaposed to one another and thus seem to leave us with an aporetic situation, this situation is not quite as desolate if we look at it in some more detail. The development of a new framework for the application of deixis in texts can make use of some aspects of deixis in its speech-situational applications. The situation for deixis in texts is not totally without parallels from a systematic point of view. Whereas the ordinary line of analysis of deictic expressions in speech situations takes their boundness to the speaker's and hearer's common sphere of perception in a very strict sense, in an — as it were — "ontological" or objectivistic sense, a

Scientific

texts

207

more detailed analysis shows that even in speech-situational uses of deictic expression there are some — though sporadic — elements of dissolution regarding the relationship of perceptual and linguistic activities. The objectivistic interpretation of this relation is one which reduces the notion of perception to a simplistic mechanistic concept in which the part played by knowledge structures for perceptual and linguistic activities is severely underestimated. The application of deixis, however, in many cases makes use of the common sphere of perceptional access in a strongly conceptualized form. This becomes obvious from the language system itself as soon as we consider, e. g., the plural forms of the personal deixis. In the use of we, it is far from self-evident which persons are meant by the use of the expression, even if the possible referents are physically present in the speech situation. Specific knowledge structures are implied for the precise identification of the extension of the expression we on the part of the hearer, e. g., knowledge structures such as membership devices. Only by the application of such knowledge the hearer can correctly identify who is meant by we next to the speaker, if the speaker makes use of we in a deictic procedure. This type of deictic expression, then, involves vagueness, causing perceptual uncertainty instead of perceptual certainty on the part of the hearer. So the analysis of deixis seems to be self-contradictory for such cases. Instead of being an economic and efficient means of achieving the hearer's orientation in the common sphere of perception, the deictic expression seems to end up with the hearer's non-orientation. This, however, is not really the usual effect of the application of expressions like we in deictic procedures. On the contrary, also in these instances of application, the deixis remains economic and efficient in achieving the hearer's orientation. But it does so in a more complex way than we have learned so far. Additional activities are involved to achieve the same effect. These additional activities involve the actualization of knowledge on the hearer's, and, of course, on the speaker's part. Previously acquired knowledge complements what is not accessible to immediate sense perception. Previous knowledge transforms the perceptual uncertainty caused by the vagueness of these types of deictic expressions. This change in the characterization of deixis becomes still clearer when the speaker speaks as one member of a set of persons of whom he is the only person present in the actual speech situation. The physically absent other members of the group are pointed to and are thus considered to be "present" in a very peculiar way: they are present in the knowledge of the hearer, thus in a way which is not accessible to sense perception.

208

Konrad

Ehlich

They are "presented" to the hearer in and by the use of the deictic expression which elicits the hearer's previously acquired knowledge. The results of our discussion of the personal deixis in its plural form hold true in a similar way for other deictic expressions. But in their case they are less obvious and less perspicuous. Another example is the temporal deictic expression now. It refers to the time dimension of the actual speech situation. How "far" does this time dimension extend? Consider the following examples: (1) A speaker at the Rauisch-Holzhausen conference: "I am now speaking at Rauisch-Holzhausen." (2) A speaker after three days of heavy rain, and a period of sunshine of several hours: "The weather is fine now." (3) During the last years the wall was an obstacle to free traffic between the G D R and the F R G but it is gone now. (4) For centuries the supply of corn was problematic for large parts of Europe's population but it can be managed now, with modern agriculture techniques at our disposal. (5) The development of new tools took thousands of years in prehistoric time but it has increased steadily and considerably now. The extension of now differs from the speech situation of the utterance in case (1), to "today" (2), to "this year" (3), to "the last one and a half centuries" (4), to "the last thousand years" (5). The impact and implication of mental structures is shown in another respect still with the deictic expression now. The dimension of the speech situation which is put into the center of the hearer's attention in the case of now is time. Time in itself, however, is unaccessible to direct sense perception. This incompatability of mental and perceptual structure in the case of time has been demonstrated clearly by modern, enlightenment and post-enlightenment philosophy (especially by I. Kant in his "Critique of Pure Reason"). At the same time it has been a permanent challenge to craftmanship and industry which managed to overcome this difficulty by developing clocks and watches which make time "visible", thus opening a pathway for direct sense perception of time. All of these characteristics of deictic expressions and of their usage, which could be enlarged and developed further with regard to other subparts of the deictic system, show ways of how to dissolve the immediacy of speech situation dependency of such expressions.

Scientific

texts

209

2.4. Deictic transfer: f r o m speech situation to imagination There are other uses of deictic expressions which m a k e methods of transferring deixis f r o m the immediate speech situation into more complex types of application accessible to communicative interactors. Systematically speaking, the first such transfer, in the course of which a transcendency of the immediate speech situation, its accessibility to sense perception and of the boundaries involved by it, takes place, is the context of the application of deixis to speech itself and to parts of it. As soon as one speaker in a dialogue refers to a previous speech action of the other party by means of a deictic expression he/she refers to a speech action that is no longer "there"; or, in other words, he/she refers to a speech action in completed transgression. (6) gives an example: (6)

(a) A to B: " (b) Β to A: "You have stated that nicely."

"

In example (6), with its neat turn allocation, by the fact of change of the turn f r o m A to Β which takes place with B's very utterance, the utterance (a) of A is no longer "present" when Β speaks. It has come to an end at the very m o m e n t when Β starts. B's referring to A's utterance (a) by means of that in utterance (b), thus, focusses A's attention to something which is no longer present in the speech situation and no longer accessible to sense perception — though of course it had been present some seconds before B's utterance (b). The relation to A's speech action or, more precisely, to some part of it (its propositional act and the way of how A had put it) is a relationship to something the only presence of which in the present speech situation (the situation of B's uttering of (b)) can be a presence in memory. The deictic object is not present to physical, to sense activity — though it was present and accessible to them immediately before. Nevertheless, the deictic expression that is used. The immediacy of the immediate speech situation is transcended. The deictic procedure is transposed to a different type of space, a space which is constituted by discourse and its inherent structures. The deictic procedure becomes a procedure in the "space of discourse". The immediacy of the speech situation proper, as a sphere of perceptual access, has been transcended. W h a t is relevant now is just a previous speech product f r o m the previous speaker. This product is mirrored in the mental sphere of the previous hearer, and in his/her consequent speech action he/she makes use of what he/she has in mind, and of what he/she knows the previous speaker has in mind, and he/she no longer makes use of something which is the case

210

Konrad

Ehlich

in the present, perceptually accessible reality. In this instance we can clearly see a step of great systematic importance, the step of dissolution from the original qualifications of deixis. The second such transcending of the immediate speech situation occurs when a speaker deictically refers to entities, etc., which are neither part of the present speech situation as such nor of an immediately preceding speech situation, but which are "imported" into the present speech situation by the act of verbalization of one of the interactors in order to elicit, by means of the verbalized notion, a notion on the side of the hearer equivalent to that which the speaker has in mind. This is a case which Bühler termed Deixis am Phantasma ('deixis at the phantasma', Bühler [1965]: 123). For the sake of a consistent terminology, it is advisable to speak here of a "deixis in the space of imagination". Let us consider this case in some detail also. Here, the dissolution from the actual speech situation has a more principled status. The factual interaction is one between the speaker's imagination, knowledge etc., and the hearer's imagination, knowledge etc. It is mediated by the deictic expression. By means of the deictic procedure in the speech action an orientation of the focus of attention on the part of the hearer is affectuated in identity with a corresponding orientation of the focus of attention of the speaker. So these general characteristics of deixis and deictic procedure remain the same but the sphere of their application, the deictic space, differs fundamentally. The physical presence of the possible deictic objects and their perceptual accessibility are no longer part of the picture. In these two cases of dissolution, the linguistic interactors share a common transfer of the deictic system into different spheres of relation between deictic expression and deictic object. These spheres of relation are characterized by their mental qualities. The objects of deictic reference are units of the hearer's and the speaker's short time memory (in the case of (parts of) the previous speech actions), or of their imagination (in the second case). How is the transfer to different spheres of relation brought about? It is brought about by common mental activities of the two involved interactors. They construe a mental sphere for the application of deictic procedures. This sphere is called a "deictic space". The procedures themselves remain the same, but the spaces of their application change. In this process of change, the sense-perceptual access is substituted by a verbal and/or mental access. The process of transfer is not without considerable dangers of failure. The great advantage of deictic procedures in the speech situation proper,

Scientific

texts

211

in comparison with the use of symbolic (in the Bühlerian sense) procedures, is the immediate reliance on the speaker's and hearer's common ground, the common sphere of perception, which can permanently be actualized and which offers a source for solving communicative failure, in all cases of communicative misfit. Other than verbal means for clarification (such as gestural pointing) can be applied, and they are well accepted among the interactors for these purposes. They, together with a close interrelationship of the speaker's speech actions and the hearer's activities of understanding, ensure certainty in most practical contexts. This basis of reliance and of certainty is left behind in the process of transfer of deictic procedures into deictic spaces other than the one characteristic for the immediate speech action, i. e., for the speech situation itself as space of deictic procedures. At the same time, the constructive part of both parties increases considerably. The construction of an imaginative space has to be organized, mainly by symbolic expressions. They usually do not, however, suffice. Instead, a large amount of shared knowledge prior to the speech situation is indispensable in order to succeed in the construction of a deictic space of imagination. The dangers of breakdown of deictic procedures necessitate specific measures of caution on the part of the speaker and of the hearer. The aim of the processing of deictic procedures is the establishment of a focus of attention of the hearer with regard to the deictic space under consideration. Failure of the application of the deictic procedure may turn out to have fatal consequences because it causes disorientation in the very locus of the linguistic activity which is meant to achieve orientation. A failure in the application of deixis thus is directly counterproductive in the speaker's on-going linguistic activity, and not only a linguistic infelicity.

2.5. Deixis in texts The construction of deictic spaces beyond the limits of the speech situation has been analyzed in a systematic "step-by-step" procedure in the transition from immediate speech situation to the deictic space of discourse and to the deictic space of imagination. This process continues, as soon as "texts" are systematically taken into consideration. The text itself is taken as a specific deictic space. But, again, at the same time with the establishment of a new space for the application of deictic procedures

212

Konrad

Ehlich

the dangers for the success of the linguistic activities increase. Further measures of precaution are needed. This means that there is also an increase of abstractness in the construction of the deictic space. The demands on the hearer's mental activities become more and more intense. The "hearer", becoming a reader in the case of written texts (cf. section 1.3.), is faced with the need to do more and more on his/her part, in order to let the deictic procedure do its job. The increase of abstraction, then, necessitates an increase of mental activities on the part of the text recipients. In order to counterbalance the risks of this process, the application of deixis in the space of text is strictly limited to a small range of possible deictic procedures. They are highly standardized in nature. Their use is part of the professional training of the recipients, be it in the context of learning how to read, be it in their further socialization. The readers of scientific texts are included in this process. The training of how to make use of deictic expressions in a ScT is part of the scientific socialization they have to undergo. These restrictions are also part of the generalized style of scientific writing. The ordinary scientific writer has been a scientific reader before. The restrictions for the application of deictic procedures in the space of text are thus bound to a strict stylistic tradition leaving little alternatives to the individual writers in their ScT production. One last important general aspect of the application of deictic procedures in texts needs to be mentioned. The more complex forms of deictic procedures, related to the different types of deictic spaces, are combined structurally and individually in discourse and text. That means: deictic procedures in texts are not restricted to text deixis. On the contrary, the surface of texts, as the surface of discourse, shows all sorts of deictic procedures. It is one of the most difficult tasks for a reader of a text to identify the specific quality of a particular deictic procedure and to perform exactly those mental constructions of deictic spaces that are demanded by the author of the text in order to fix his/her focus of attention to exactly that object which the author wants it to be focussed upon. An important help for correct identification of the deictic procedure used by the author, for correct selection of the respective deictic space, and for the mental (re-)construction of the appropriate space of knowledge, imagination, etc., is the standardization of application of deixis in the more complex and more abstract deictic spaces.

Scientific

3.

texts

213

Deixis in scientific texts

3.1. Problems of application of deixis in ScTs ScTs in themselves are highly abstract in character, that means: they are detached not only from the specific situation of their production as all texts are but they are detached from situation as such. The knowledge incorporated in them is transsituational knowledge. Generalization is a process of de-situalization. This characterizes the problematic character of the applicability of deixis in ScTs. The response of speakers/writers and of hearers/readers to this problematic constellation can be characterized in three ways: 1. The application of deixis is strongly restricted to very specific cases. 2. The application of deixis is done in a formularized way. 3. The application of deixis is avoided according to a permanent calculation of the costs and benefits in the use of alternative procedures, such as symbolic procedures or semiotic (e. g. graphic) procedures. In the subsequent sections of my paper, I will discuss some examples for these three aspects of the applicability of deixis in ScTs under the headings of specification (3.2.), formularization (3.3.), and avoidance of deictic vagueness (3.4).

3.2. Examples for specification One of the major specifications in the application of deixis to ScTs is that deictic procedures in this context are restricted to uses of deixis of the type "in this paper ... ". They occur mainly in the beginning part of ScTs, or at the end, or in summaries and abstracts. This application of a deictic procedure makes direct use of the perceptual accessibility of the text itself. "This paper" is a new entity in the world of writer and reader. The writer constructs this reality, and the reader has it in his/her hands. The text itself is present and accessible to sense perception for the two of them. The use of deictic expressions referring to the text as perceptionally accessible object consequently is a combination of two types of deictic procedure: speech situational deixis and text deixis. A second specification can be analyzed by making reference to the application of deixis in complex oral texts in early societies. These texts are accessible for us in a literary form, e. g., in some parts of biblical texts. As an example I quote Genesis 36, 15 — 19, a combination of lists

214

Konrad

Ehlich

'ellä 'allupe b.ne-cesaw: these (are) the dukes of the sons of E.

b.ne 'alipaz b.kor cesaw: the sons of E. the firstborn of E. 'allup 'allup 'allup 'allup allup 'allup 'allup

Θ

temän 'omar $.po q.naz qorah gactäm camäleq

'ellä 'allupe 'alipaz b. 'äräs 'adom. these (are) the dukes of E. in the land of E.

- / t h e s e (are) the sons of A . / o ^ j

Ύ / / / / / / / / / / / / / / / f/ / / A w.'ellä b.ne r.cu'el bän-cesaw: a n d these (are) the sons of R., E's son 'allup 'allup 'allup 'allup

Φ

nahat zärah sammä mizza

'ellä 'allupe r.cu'el b.'äräs 'adom. these (are) the dukes of R. in the land of E.

w.'ellä b.ne 'oh0libämä 'esät cehäw: a n d these (are) the sons of O., E.'s wife

Φ

'allup y.cus 'allup yacläm 'allup q o r a h

'ellä allupe 'oh°libämä bat-canä 'esät cehaw. these (are) the dukes of O., the daughter of Α., E.'s wife

'ellä b.ne cesäw; these (are) the sons of E.

w.'ellä 'allupehäm. and these (are) their dukes

Figure I

f r a m e of the list of lists

,

f r a m e element without deixis

f r a m e of the list

Up

f r a m e element etc. of second structure

Scientific texts

215

of names of some of the dukes of the sons of Esau (Fig. 1) (cf. Ehlich 1979: § 6.4.2 and Ehlich (1982: 332)). The text consists of three lists l(v. 16), 2 (v. 17), 3 (v. 18). These three lists are integrated in to a larger list (v. 15 — 19). At the end of verse 16 and verse 17 we find two elements of a secondary structure ((e), (i)). The lists of names ((c), ((g), (k)) are the basic elements of content. They are organized in a complex structure of framing ((b)/(d); (f)/(h); (j)/(0· Also the larger list v. 15 — 19 has its framing (a)/(m). In all of the frame elements (except for (b)) we find deictic expressions. It is obvious how they function: they serve to set up frames. At the beginning of a list and at its end deictic expressions are used. In order to integrate lists into lists deixis again has a fundamental organizational function. If we take into consideration our analysis of the general function of deictic expressions in speech activity, we clearly see: deictic expressions can serve to establish lists by giving the hearer a help for the organization of pieces of knowledge, in setting his/her focus of attention to these elements. The deictic expressions reorientate the attention of the hearer to the textual structure itself. Two types of deixis are of importance here. The one is catadeictic in nature, the other one is anadeictic. By 'catadeictic" I mean a procedure which has as its aim the orientation of the attention of the reader to a subsequent part of the text in his/her way through the "space of text" as sphere of perception, and by 'anadeictic' I mean a procedure which has as its aim the orientation of the attention of the reader to a previous part of the text in his/her way through the "space of text" as sphere of perception. 2 Lists are specific forms of texts, and they are very early ones in the history of the development of text structures. The application of deixis for frame formation responds to obvious needs in the textual interaction, as long as the mode of transmission of texts is an oral one. Re-orientation of attention is essential to process correctly such complex types of information. One might expect these uses of textdeixis to occur also in written texts. This expectation however does not seem to be fulfilled in the data. 3 The application of deixis in present day German ScTs does no longer concentrate on major text forms. Instead, the (re-)organization of the attention of the reader is done with regard to much smaller units than the major or overall structuring units of texts. Example (7) reads: (7)

(a) Die sogenannten Funkengesetze beschreiben den zeitlichen Verlauf der elektrischen Größen bei Funkenentladungen, und zwar vorzugsweise im Bereich des raschen Spannungszusammenbruchs.

216

Konrad Ehlich

(b) Die Anwendbarkeit dieser Gesetze wird aber dadurch erschwert, daß in allen Konstanten vorkommen, deren Zahlenwert nicht aus den Eigenschaften des betreffenden Gases berechnet werden kann. (Pfeiffer (1971) / LIMAS 091.wis) (a) The so-called spark-laws describe the temporal course of electric units in the case of spark discharge, preferentially in the domain of a quick tension breakdown. (b) The applicability of these laws is made difficult however by the fact that there are constants in all of them the value of which cannot be calculated from the properties of the gas in question. Example (7) consists of two complex sentences (a) and (b). Sentence (a) has one basic verb (beschreiben / describe) and six noun-phrases which are integrated into the basic structure established by the verb as sentential predicate (cf. (8) and Fig. 2; I do not discuss the various structural components of the graph here. Straight lines mark the basic verb-noun relation reflecting the direct verb valency, double straight lines mark the "genitive"-relation, and bent lines mark the adverbial relation). Besides, there are two structuring elements {"und zwar" / " — "; "vorzugsweise" / 'preferentially'). (8) I Die sogenannten beschreiben

Funkengesetze

I den zeitlichen Verlauf I der elektrischen Größen I bei Funkenentladungen und zwar vorzugsweise | im Bereich | des raschen Spannungszusammenbruchs I # NP YPr OP :#

= = = =

NP1 VPr NP2 NP3 NP4 OP1 OP2

Noun Phrase (Prepositional phrases included) Verb as Predicate Operator Completion of sentence form

NP5 NP6

Scientific

texts

217

VPr

NPl

NP2

- NP3

NP5 = NP6

NP4 Figure

2

The deixis "dieser" / 'these' has its position in the second noun phrase of sentence (b): "die Anwendbarkeit" = N P l ; "dieser Gesetze" = NP2. Why, then, reorganize the attention of the reader at this point? The deixis "dieser" / 'these' is a clear instance of an anadeixis, as described above. It orientates the attention of the reader to a previous part of the text. How large the part is which is put into the focus of the reader is not obvious per se. The specification of extension of the deictic object basically has to rely on linguistic experience of the trained reader. Possible candidates for the reader's identification of the deictic object are — — — — —

parts of the previous sentence parts of the previous paragraph the previous paragraph parts of the previous text the previous text.

In order to help the reader to identify the deictic object the deixis is specified by additional information, in this case "Gesetze" / 'laws', combined with inflectional information. The grammatical form which is used for this combination is the attribution of the deixis to the noun "Gesetze" / 'laws'. The characteristics of this construction can be illustrated by way of opposition of a simple deixis at the position of the noun phrase "diese Gesetze" / 'these laws', as is shown in example (9): (9)

(a) Die sogenannten Funkengesetze beschreiben den zeitlichen Verlauf der elektrischen Größen bei Funkenentladungen, und zwar vorzugsweise im Bereich des raschen Spannungszusammenbruchs. (b) Dies ist seit langem bekannt. (a) The so-called spark-laws describe the temporal course of electric units in the case of spark discharge, preferentially in the domain of a quick tension breakdown. (b) This has been well known for a long time.

218

Konrad Ehlich

Here, "dies" / 'this' has as its deictic object the whole previous sentence (a). In (7), however, the deictic object is part of (a) — but which part? In order to specify which part of (a) is the deictic object it is useful to give a second analysis of (a), an analysis which "breaks up" (a) along the lines of information processing. That means: the "subject-predicate" structure is taken as basis of analysis (cf. Fig. 3).

Κ

Ν

/ NPl Key: IU = information unit IP = ongoing information process IC = information completion Κ = the known Ν = the new NP,VPr,OP : see (8) Figure 3

The subject position is interpreted as "the known", the predicate position as "the new". The predicate position is seen as comprising the verb plus its objects except for the subject, but also the operators. Other classifications are possible, especially with regard to the operators. The predication on the subject, or, in other words, the new information Ν which is added to information already in hands of the reader (K) is transformed into a new uniform information unit which is realized at the point where " # " stands. Information having been processed so far the previously new part has been integrated into the known parts of the reader's knowledge. The recall of what had been Κ in structure of Fig. 3 implies the actualization of what had been N. In other words, the actualization of "Funkengesetze" / 'spark laws' makes active different knowledge structures before "(a) # ", and after it. In order to make use of the result of the integration of Κ and Ν in (a) the deictic procedure is applied. The reader is asked to treat, what has been pronounced before in a series of

Scientific texts

219

verbalization steps, as one deictic object. By the further application of the combination of a deictic procedure plus a term from the previous subject position, namely "Gesetze" / 'laws', the writer asks the reader to take K, enriched by N, as his/her deictic object, or, in other words, to concentrate his/her attention on it. The reorganization of the attention of the reader helps him/her with the processing of complex knowledge structures which are not only introduced in the ongoing linguistic activity of the speaker but also made use of in its continuation. The situation could be totally different in case we would not have an anadeixis but an anaphor, cf. (10) (10) (a) Die sogenannten Funkengesetze beschreiben den zeitlichen Verlauf der elektrischen Größen bei Funkenentladungen, und zwar vorzugsweise im Bereich des raschen Spannungszusammenbruchs. (b) Ihre Anwendbarkeit wird aber dadurch erschwert, daß in allen Konstanten vorkommen, deren Zahlenwert nicht aus den Eigenschaften des betreffenden Gases berechnet werden kann. (a) The so-called spark-laws describe the temporal course of electric units in the case of spark discharge, preferentially in the domain of a quick tension breakdown. (b) Their applicability is made difficult however by the fact that there are constants in all of them the value of which cannot be calculated from the properties of the gas in question. In German it would be possible to have as antecedents of "ihre" / 'their' all nouns with the specification (i) { — plural/ + feminine) or (ii) ( + plural), i.e. NP1, NP3, or NP4 (all being of type (ii)). The application of the anaphoric procedure would mean a continuity of knowledge elements in the focus of attention of the reader which does not exist in his/her knowledge space. The reader would have three choices as to what the object might be to which the speaker refers by using "ihre" / 'their'. A new orientation of the reader, a new focus of attention is at its place here. By means of the application of the deictic procedure the speaker gives the reader the chance to do mentally what is needed. The reorientation in the processing of the text is an efficient means for text comprehension. I call this a microorganization of attention of the reader. The microorganization of attention differs from the type of organization of attention illustrated by example (Fig. 1). In the biblical text, being a text dependent on orality, macrostructures, namely the textform in itself, are actualized by means of deictic procedures. I call this the macroorganization of

220

Konrad

Ehlich

attention. In case the two examples (Fig. 1) vs. (7) are representatives of two stages of text formation we can hypothesize: There has been a shift from macroorganization to microorganization of attention in the process of de-oralization which is a process of growing complexity of knowledge. The type of application of deixis illustrated by examples (7) to (9) is very prominent and occurs very often in ScTs, in German. The repetitive application of the deictic procedure with reference to the space of text means that the reader has to perform a permanent reoorganization of his/her focus of attention at various instants of the continuing of the linguistic entities he/she is reading. The function of the deictic procedure is to help him/her manage this permament shift of focus. Smaller and smallest text units can be so complex that these processes of focus adjustment are necessary. This holds true especially when syntactic structures are used to express maxima of information. German is known (and hated) for the "complex sentences" it has, and "German scientific style" is experienced as "very complicated" and "difficult to process" (as certainly will be the present one, notwithstanding its English "disguise"). The application of deictic procedures to the microorganization of attention is a major resource for the reader to overcome the difficulties which lay at the basis of the (negative) evaluations of German ScTs.

3.3. The use of deictic procedures in standardized ways: formularization Next, I would like to discuss a second set of deictic procedures very common in German ScTs. Let us reconsider example (7) for an illustration. (7 (b)) reads: (7)

(7)

(b) Die Anwendbarkeit dieser Gesetze wird aber dadurch erschwert, daß in allen Konstanten vorkommen, deren Zahlenwert nicht aus den Eigenschaften des betreffenden Gases berechnet werden kann. (b) The applicability of these laws is made difficult however by the fact that there are constants in all of them the value of which cannot be calculated from the properties of the gas in question.

The expression "dadurch" has two components: (i) "da" and (ii) "durch". (ii) is a preposition which is semantically similar to and etymologically cognate with English "through", (i) is a deictic expression. Its semantic diversification in German is very large (cf. for detailed analysis Redder

Scientific

texts

221

1990). Let us consider shortly what the function of "dadurch" in example (7 (b)) is. In doing so we can refer to the results of 3.2. The structure of (7 (b)) is characterized by the fact that two positions of the basic sentence S (b) are filled up not with phrases but with sentences, cf. (11) and (12).

(11) I Die

NP1 NP2

Anwendbarkeit I dieser Gesetze

wird erschwert

VPr OP1

aber dadurch, daß (c) (c) I in allen kommen ... vor

NP3 VPr

| Konstanten (d)

NP4

(d) I deren Zahlenwert kann berechnet werden

NP5 VPr OP2

nicht aus den Eigenschaften | des betreffenden Gases

NP6 NP7

The integration of sentences into sentences, treated in linguistics under headings like "subordinated clauses" or "Nebensatz" ('by-sentence') constitutes a complex task for the recipient. In order to process these types of sentences the hearer/reader needs help from the speaker/writer. In case such help were lacking there would be a risk for the reader to get lost in structural complexity. The reader would find at his/her disposal a set of information pieces without indication of their interrelationship. The processing of the information pieces is largely organized by means of adequate adjustment of the focus of attention. In other words, deictic procedures can provide some of the help the reader needs. The application of deixis in these contexts is technical in a linguistic sense of the word. Deictic procedures which are used for these technical activities on the part of the reader are even more "specialized" in their function than those types of deictic procedures which were discussed in 3.1. This secondary type of specialization leads to a standardization of deictic uses which is realized by a formalization of the use of deixis. A

222

Konrad Ehlich

VPr

NPl

=NP2 VPr

NP3

NP4

NP5

=fdl

NP6 =

=

NP7

Figure 4

set of expressions is used in order to achieve the goal of proper and easy processing of complex syntactical structure. The compound "dadurch" is one of the elements of this set of deictic expressions in German. In (11 (b)) "dadurch" enters into the position of an adverbial phrase which specifies the cause or reason for the difficulties in the application of the "spark laws". This cause or reason is given in (11 (c)). "daß" / 'that' is the formal "complementizer". It also is deictic in nature, cf. below. This formal complementizer is not enough however for the reader to reconstruct the information structure which the writer transmits. Instead, the expression "dadurch" orientates the focus of attention of the reader to an issue to follow. "Dadurch" is used catadeictically in (11 (b)). The attention of the reader is not satisfied with sufficient information, but is kept active for further information supply. The deictic expression functions as a substitute, and it does so by attention management of the reader.

Scientific texts

223

In principle, the structure is similar to the one which was analyzed in 3.2. But the technical character of this type of deictic procedure means that the whole process of application in the writer's text and of actualization in the reading process tends to be "automatized". Languages provide different sets of deictic expressions which can be made use of for these technical purposes. The list in (12) contains German expressions of this type. (12)

(a) da (b) dabei dadurch dafür dagegen damit danach daneben daran darauf daraus darin darüber darum darunter davon davor dazu (c) daher dahin (d) derart dergestalt dermaßen deshalb deswegen (e) indem nachdem trotzdem zudem ( f ) hingegen (g) sofern sogar

224

Konrad Ehlich

somit sosehr soweit sowie (h) dahingegen daraufhin sodann sowieso These expressions all function in a formalized way for the organization for complex sentence and textual structures. Only (a) is a simple deictic expression, (b) to (h) are compounds. They are of different structural types, see (13). (13) (i) (ii) (iii) (iv) (iv') (iv") (iv'") (ν)

deixis + preposition : (b), (f) preposition + deixis : (e) deixis + inflectible other than preposition : (g) deixis + deixis : (c) deixis + preposition + deixis : (h): "daraufhin" deixis + inflectible ... + deixis (h): "sowieso" deixis + deixis + preposition : (h): "dahingegen" deixis + substantive (d)

As is shown by the lists (12) and (13) there is a rich typology of word formation structures and of expressions in German. Most of these expressions are very prominent in German ScTs. Some of these expressions change their categorial status. They become "conjunctions". Nevertheless their components remain present in how they function for structuring complex sentences and texts, i. e., for helping the reader to transfer complex relations of information pieces from the text into his/her mind. One obvious example for a categorial shift is the expression "daß" / 'that', the general complementizing expression in German. As a consequence the organization of complex and of abstract texts implies the use of a large variety of deictic expressions being used for text organizing deictic procedures. Deictic procedures with formulaic deictic expressions form a reservoir for the construction of means of expressions for the indication of complex relations. These expressions which are formed by use of deictic elements in combination with others are especially important for the representation of abstract categories such as cause, consequence, and reason. Many of these expressions and some of the formation structures have equivalents in English such as 'therefore' (deshalb), 'thereby' (dadurch),

Scientific

texts

225

'hence' (deshalb), etc. But the number seems to be considerably less than in German. It is likely that, besides the smaller number of these expressions, their application in ScTs is less frequent. English seems to have chosen different strategies for representing complexity and abstractness in ScTs. So English seems to be poor in deixis, German rich, and French even richer. These differences are of considerable impact for the translation of ScTs from one of these languages into another one, and they cause much trouble for non-natives when writing ScTs in a language other than their mother tongue. It would even be worth while to find out whether research in textual differences of ScTs in the different scientific cultures which make use of these different languages is influenced by subtle inferencing from the structural characteristics of the researcher's language into evaluation criteria for, e.g., perspicuity and well-organizedness of ScTs.

3.4. Avoidance of vagueness and crypto-deixis Deixis in its more complex uses can lead to considerable vagueness. This effect is contrarious to the genuine benefits of this type of linguistic expression, as has been pointed out above. ScTs are written with reference to clarity as one of the main maxims of text production. As far as clarity cannot be reached by means of the application of deixis, avoidance of deictic procedures follows as a submaxim to the maxim of clarity. There are various fields in which standardized forms of deixis avoidance are practised in the context of ScTs. One of them is the field of the expression of time. In concordance with the detemporalized character of text, temporal deixis is depreferred in ScTs. Instead, "objective" expressions such as expressions from the official calendar and from generalized systems of measurement take their place. Another field is the field of the expression of cause, consequence, and reason, where, again, non-deictic, descriptive expressions are preferred in many instances to deictic expressions. I don't go into this topic further, but I would finally like to discuss two aspects of the use of expressions in ScTs which I would like to call "pseudo-avoidance" of deictic expressions. The first example is the use of terms like "oben" / 'above' and " u n t e n " / 'below' in ScTs. These expressions are very common for reference to other parts of a ScT. In themselves they are not deictic expressions. The expressions " o b e n " and " u n t e n " are descriptive in nature. Their applica-

226

Konrad

Ehlich

tion is restricted to entities with a shape or structure which makes the identification of top and bottom possible. The qualification "oben" / "unten" is related to these "Gestalt" characteristics. Where the "upper" and the "lower" part of the Gestalt begins depends on the identification of sort of a Gestalt center. These properties are characteristic for, e. g., houses, hills, tables etc. The application of one of the expressions "oben" or "unten" to them is by no means deictic in the sense that it depends on the speech situation. Texts in themselves are not characterized by such a Gestalt quality. Written texts in the Phenician-Latin alphabet tradition have a top-bottom characteristic with regard to pages or similar semiotic units, but the single pages are not arranged in a top-bottom order. Nevertheless the expressions "oben" and "unten" are applied to texts. In doing so the writer constitutes, as it were, a "mental" Gestalt of the text transposing the factual order of the pages into an ideal one-page order. But still it is not clear how the application of the terms "oben" and "unten" works. If one tries to find an answer to this question one comes across an interesting phenomenon which makes understandable how the impression of a deictic quality of these expressions can arise: The qualifications "oben" and "unten" are, as had been said before, dependent upon the identification of a Gestalt center. Now, this Gestalt center in the case of text is constituted by making the act of writing, i.e. the specific utterance act in its written form, this center. In doing so, the writer interprets a linguistic and cognitive structure with relation to his/ her linguistic activity. In other words: the Gestalt center is interpreted as "origo" in the analytical sense of the term. In this way "oben" and "unten" acquire a linguistic property which genuinely is owned by deictic expressions. In order to characterize this specific transposition of an expression from one field to another, I use the Greek prefix "para-". "Oben" and "unten" in their use to refer to previous or to later parts of a text are "para-deictic expressions". Paradeictic expressions seem to make use of symbolic linguistic qualities, in the sense of Bühler's "symbolic field", thus avoiding the difficulties of deictic expressions in ScTs. In fact, however, the avoidance of deixis does not really occur. "Oben" and "unten" are not the only expressions which are used in this way. Other expressions are "folgend" / 'following', "später" / 'later', etc. Their transference to para-deictic use is similar to that of "oben" / "unten".

Scientific

texts

227

The second example is the use of expressions like "vorliegend" / 'present', as in (14): (14)

"der dem Leser vorliegende Text" "the text present to the reader"

Again there is no overt deixis in (14). Nevertheless there is a relation to an element of the "speech situation", namely the text as an object in the reader's perceptual space during his/her act of reading. This object is described. But the description is done in a way in which a hidden reference to the reader's reading activity is made. So, the description is not really dissolved from (parts of) speech situation. The same holds true for the expression "dem Leser" / 'to the reader'. He/she is not any reader but that reader who is the reader "now". These avoidances of deixis, then, refer to parts of the speech situation but do not do so overtly. Implicitly there is a deictic procedure involved. I would like to call this type of hidden deictic use "cryptodeixis". Crypto-deixis is another form of how writers try to overcome the difficulties of applications of deixis in texts, and especially in ScTs. It is a consequence of depreference of deixis for this text type — but in fact closer analysis shows that even writers and readers of ScTs cannot easily get rid of deictic procedures being an elementary part of the overall linguistic structure.

Notes 1. In the original German text, Bühler uses terms which differ semantically from one another, for the field and the expressions belonging to the field. So he speaks of the "Zeigfeld" ([1934]) for the "deictic field" and of "Deixis" for the expressions belonging to it. This is a useful distinction which is taken up as much as possible in the extended field theory (cf. Ehlich 1986). It seems to be difficult to simulate these distinctions in English, cf. the English translation 1990. 2. Catadeictic and anadeictic uses of deixis are to be differentiated from "anaphora" and "cataphora". By not making this distinction the discussion of phoric as well as of deictic expressions and their use is obscured in large parts of the literature. 3. Though there has not been a detailed quantitative analysis up to now, as far as I see. The existing corpora of ScTs, e . g . , L I M A S for German, are not tagged in a way which would enable detailed research regarding the use of deixis. A complexer type of corpus would be very valuable, and it would be worthful to have a large size corpus which might come up to some representativity for the text types involved. Setting up such a corpus would need major funding. A small scale corpus is put together at Dortmund University, "Dortmunder Korpus Wissenschaftlicher Texte ( D o K o W i T ) " .

228

Konrad Ehlich

References Antos, Gerd 1982 Bühler, Karl 1934 [1965] [1990] Clyne, Michael 1980 1987

Grundlagen einer Theorie des Formulierens. (Reihe Germanistische Linguistik 39) Tübingen: Niemeyer. Sprachtheorie. Jena: Fischer. [(2nd edition) Stuttgart: Fischer], [Theory of language. The representational function of language. Translated by Donald Fraser Goodwin. Amsterdam, Philadelphia: Benjamins].

"Writing, testing and culture", The Secondary Teacher 11: 1 3 - 1 6 . "Cultural differences in the organization of academic texts", Journal of Pragmatics 1 1 : 2 1 1 - 247. Coulmas, Florian — Konrad Ehlich (eds.) 1983 Writing in focus. (Trends in Linguistics. Studies and Monographs 24.) Berlin, New York, Amsterdam: Mouton. Ehlich, Konrad 1979 Verwendungen der Deixis beim sprachlichen Handeln. (2 vols.) Bern, Frankfurt, Las Vegas: Peter Lang. 1982 "Anaphora and deixis: Same, similar, or different?", in: Robert J. Jarvella-Wolfgang Klein (eds.), 3 1 5 - 3 3 8 . 1983 a "Denkweise und Schreibstil. Schwierigkeiten in Hegeischen Texten: Phorik", in: Barbara Sandig (ed.), 1 5 9 - 1 7 8 . 1983 b "Writing ancillary to telling", Journal of Pragmatics 7: 495 — 506. 1986 "Funktional-pragmatische Kommunikationsanalyse — Ziele und Verfahren", in: Wolfdietrich Härtung (ed.), 1 5 - 4 0 . [1990] [reprinted in: Dieter Flader (ed.), 127-143.] Flader, Dieter (ed.) 1990 Verbale Interaktion. Stuttgart: Metzler. Galtung, Johan 1982 Struktur, Kultur und intellektueller Stil. Ein vergleichender Esssay über sachsonische, teutonische, gallische und nipponische Wissenschaft. Berlin. Goody, Jack 1983 "Literacy and achievement in the Ancient World", in: Florian Coulmas — Konrad Ehlich (eds.), 8 3 - 9 7 . Härtung, Wolfdietrich (ed.) 1986 Untersuchungen zur Kommunikation — Ergebnisse und Perspektiven. (Linguistische Studien A 149.) Berlin: Akademie Verlag. Jarvella, Robert J.—Wolfgang Klein (eds.) 1982 Speech, place and action. Chichester: Wiley. Kuhn, Thomas S. 1963 The structure of scientific revolutions. (International Encyclopedia of Unified Science 2,2.) Chicago: University Press. Pfeiffer, W. 1971 "Der Spannungszusammenbruch an Funkenstrecken in komprimierten Gasen", Zeitschrift für angewandte Physik 32.4: 265 — 272.

Scientific texts

229

Redder, Angelika 1990 Grammatiktheorie und sprachliches Handeln: „denn" und „da". (Linguistische Arbeiten 239.) Tübingen: Niemeyer. Rehbein, Jochen 1983 "Zur pragmatischen Rolle des Stils", in: Barbara Sandig (ed.), 21 —48. Sandig, Barbara (ed.) 1983 Stilistik. Band 1: Probleme der Stilistik. (Germanistische Linguistik 3—4/ 81.) Hildesheim: Olms. Spillner, Bernd 1974 Linguistik und Literaturwissenschaft. Stilforschung, Rhetorik, Textlinguistik. Stuttgart: Kohlhammer.

Immediacy and displacement in consciousness and language Wallace Chafe

There are a number of important aspects of language which I believe can never be fully understood unless one recognizes that they are dependent on and shaped by human consciousness. Such a clearly linguistic phenomenon as the use of pronouns, for example, is dependent to a large extent on the flow of ideas into and out of the consciousness of language producers and receivers (Chafe 1974, 1976, 1980, 1987). Understanding extends in the other direction too. Not only is it necessary to take consciousness into account in understanding language, but language in turn provides important insights into the nature of consciousness. Thus, linguistics and psychology have the opportunity to support each other in developing a more comprehensive picture of the workings of the mind. 1 Information flow, the flow of ideas through consciousness as language is being created and comprehended, is one important part of this picture. But there is another set of phenomena that trace their origins to the fact that consciousness can alternate its focus between experience that is immediate and experience that is displaced. It is one of the most remarkable properties of the human mind that it is able to be conscious of experiences that are displaced from the immediate environment in which the possessor of the mind is located. This ability is so obvious, so basic a part of human experience, that it is seldom noticed or remarked upon, but its importance to human thought and language is profound. Without it, people would be condemned to experiencing only their immediate perceptions, actions, and feelings. They would be like a frog whose experimential repertoire was limited to noticing and happily devouring passing flies, its consciousness devoid of memories of the days when it was a prince, and incapable of imagining days when it might be united with a princess. The capacity for displaced consciousness enormously expands the range of human experience. Displaced consciousness is crucial to language. When Charles Hockett discussed what he called the design-features of human language — features differentiating it from the communication systems of other animals — he included displacement among them: "Man," he said, "is apparently almost unique in being able to talk about things that are remote in space of time (or bo'th) from where the talking goes on." (Hockett 1960: 90)

232

Wallace

Chafe

This insight can shed useful light, not only on what people talk about, but also on how they talk about it. What follows is an attempt to explore some of the ways immediate and displaced experiences interact with language in both speaking and writing, with special attention to ways in which writers of fiction manipulate these modes of consciousness to produce effects that are quite different from those encountered in ordinary speech. I will try to clarify the aims of this paper with a few brief examples. There may be times during a conversation when people talk about things that are present in the immediate environment. For example, during a dinner party the following exchange took place between speakers A and Β regarding some food that lay on the table in front of them. 2 (1)

(A) I think I should take this away. Are you guys still eating it? (B) Just hold it for just a moment, if you don't mind, and just slip another little bit in here.

But much more of this conversation dealt with things that were not immediately present. For example, one of the participants told the following brief anecdote to illustrate how pleasant it had been to move to a city with friendly inhabitants, in contrast to the inhabitants of the city in which he had lived before: (2)

I went out for a stroll, on my first time on Chestnut Street, and just was astounded at how pleasant things were. And as I was out for a stroll, a man watering his lawn turned to me, as I was walking past, and said, "Good evening."

It appears that the most common conversational situation is one in which the consciousness of the participants is focused most often on displaced events in the manner of (2), although there may be an occasional focus on immediate perceptions and actions as in (1). There are, of course, special circumstances under which a focus on immediate experience will predominate. Our particular interest here will not be conversational language per se, but certain usages in written fiction which, viewed from a conversational perspective, are unusual. I will suggest their nature with two brief examples, on which I will later expand. In a western novel by Walter Van Tilburg Clark called The Ox-Bow Incident (1940), the first paragraph begins as follows: (3)

Gil and I crossed the eastern divide about two by the sun. We pulled up for a look at the little town in the big valley and the

Immediacy and displacement

233

mountains on the other side, with the crest of the Sierra showing faintly beyond like the rim of a day moon. We didn't look as long as we do sometimes; after winter range, we were excited about getting back to town. When the horses had stopped trembling from the last climb, Gil took off his sombrero, pushed his sweaty hair back with the same hand, and returned the sombrero, the way he did when something was going to happen. We reined to the right and went slowly down the steep stage road. As in (2), this language recalls some events that were experienced at an earlier time by the person producing the language, the person who identifies himself as I, who in this case is fictional rather than real. So far as consciousness is concerned, however, there is a significant difference between (2) and (3). In (2) the language flows from a consciousness whose location in space and time is recognizably different from the location of the events being talked about, events that the speaker recalls as he is talking. In (3) there is a sense in which the language coincides with the events being talked about, in which the words are being produced as the narrator is experiencing the riding, the horses' trembling, and Gil's taking off his sombrero, rather than at some other spatiotemporal location which remains, nevertheless, recognizable as the vantage point of the narrator. How can language achieve this double effect, experiencing directly the events described and yet also emerging from a later recounting of those events? What can this use of language show us about the nature of consciousness? Why should it be especially characteristic of written fiction? These are the kinds of questions that will occupy us here. Perhaps even more striking is the effect illustrated by the opening of a short story by Ernest Hemingway called "Big Two-Hearted River" (Hemingway 1987: 163). (4)

The train went on up the track out of sight, around one of the hills of burnt timber. Nick sat down on the bundle of canvas and bedding the baggage man had pitched out of the door of the baggage car. There was no town, nothing but the rails and the burned-over country. The thirteen saloons that had lined the one street of Seney had not left a trace. The foundations of the Mansion House hotel stuck up above the ground. The stone was chipped and split by the fire. It was all that was left of the town of Seney. Even the surface had been burned off the ground.

234

Wallace

Chafe

Whereas in (3) the language seems to flow from the displaced consciousness of the person producing it, in (4) it seems to express the consciousness of some third person called Nick. The words: Nick sat down do not convey someone else's external observation of that act, but Nick's own awareness of what he did. The words: there was no town do not express someone else's, but Nick's own perception that the town was no longer there. It was Nick who remembered the Mansion House hotel. Beyond the spatiotemporally displaced immediacy illustrated in (3), the excerpt in (4) shows an ego-displaced immediacy. Thus, there appear to be two major types of displaced experience. The type mentioned in the quote from Hockett above was spatiotemporal displacement: the ability to think or talk about experiences distant in space and/or time from the location of the thinker or talker. The second type is ego displacement: the ability to think or talk about experiences that are or were present in the consciousness of someone else. It will be easiest to begin by focusing principally on displaced space and time; later we can turn more of our attention to a displaced ego. I will begin by constructing a tentative model of the basic structure of immediate and displaced experience, based partly on introspection (an indispensable guide), partly on evidence from conversational language. I will then point out some qualitative differences between these two kinds of experience. After that I will turn to the kind of first person fiction that exhibits spatiotemporally displaced immediacy, as in example (3), where some of the qualitative differences between immediate and displaced experience are overridden. Finally, I will extend the discussion to third person fiction that exhibits ego-displacement, as in example (4). I will deliberately limit this discussion to samples of writing in which language does not represent other language, whether that other language is overt speech or the covert language of silent thought. Although it is an important part of the total picture, the representation of language in written fiction is too complex, and too much has been written about it, to permit its adequate inclusion within the space available here. In a few brief remarks at the end I will suggest how it can be included within the framework of immediate and displaced conscious experience.

The basic structure of spatiotemporal immediacy and displacement The notion of displacement has to be understood with relation to a reference consciousness, which I will call proximal consciousness. It is the consciousness with respect to which displacement is defined. In

Immediacy

and displacement

235

conversations it is the consciousness of the current speaker. I will refer to the possessor of proximal consciousness as ego. Ego's identity and spatiotemporal location provide, among other things, the reference points for person and tense. In certain kinds of writing there may no longer be a recognizable ego, but even then the uses of person and tense usually provide evidence for a residual proximal consciousness which underlies the language. Proximal consciousness may be suppressed, but it is nearly always represented in language in some way. In what we may call its immediate mode, proximal consciousness interacts directly with the "outside world", ego's immediate environment. But of interest to us is the fact that it also has the ability to focus on experiences that were, or will be, or might be located at a different place and/or time. In that case we can say that proximal consciousness is in a displaced mode, where it focuses on another, distal consciousness. That distal consciousness may be one that was itself proximal at some earlier time in the life of ego, in which case it is reactivated or recalled into what is now ego's proximal consciousness. Or it may be a consciousness that ego imagines, perhaps in anticipation of some future immediacy, perhaps in a fantasy that is further removed from any past or future reality. Proximal consciousness alternates easily between the immediate and displaced modes. A person may think or talk about what is immediately present or what is not. However, there are reasons to believe that neither immediate nor displaced experience dominates proximal consciousness to the complete exclusion of the other. To say that proximal consciousness alternates between the immediate and displaced modes means only that focal, or fully active consciousness alternates in that way. When consciousness focuses on displaced experience, immediate experience is not absent altogether, but is present in a peripheral or semiactive state (for the distinction between active and semiactive consciousness, see Chafe 1987). Although it may focus on displaced experience, proximal consciousness remains peripherally aware of immediate space-time (and, when there is ego displacement, of its own ego). Conversely, when consciousness is focused on immediate experience, there may well be a peripheral awareness of some relevant displaced experience, which may have been previously focused on, or may subsequently be focused on. Thus, the alternation in proximal consciousness between the immediate and displaced modes takes place against a background in which experience in the opposite mode is peripherally present.

236

Wallace

Chafe

Qualitative differences between immediate and displaced experience Everything that follows will derive from the observation that, under ordinary circumstances, experiences in the immediate and displaced modes are qualitatively different. There is more to the alternation between these two modes than just replacing the spatiotemporal location of one experience with that of another. The experiences themselves are of a different kind. Why they are different can be traced to constraints that are imposed by recall and imagination, the two processes through which a distal consciousness can be imported into a proximal one. Neither of these processes reproduces a distal immediate experience with verisimilitude. A distal consciousness is reconstructed, not replicated, in proximal consciousness (Bartlett 1932). There are a number of ways in which a distal experience is changed when it is reconstructed in proximal consciousness. I will limit the discussion here to three properties of immediate experience, noting how each of them is modified by constraints imposed by recall. These properties can be labeled continuity, fine-grained resolution, and deixis.

Continuity Immediate experience is continuous and uninterrupted, flowing out of what happened just before and being expected to flow into what will happen next. The familiar metaphor of a stream of consciousness (James 1890: 224 — 290) captures this quality. In contrast, recall and imagination yield isolated chunks of displaced experience whose antecedents and consequences are usually inaccessible to consciousness. They appear in consciousness as experiential islands, disconnected from their surroundings, rising out of a dark sea of surrounding unawareness. It is because of the island-like nature of displaced experience that when people verbalize things they have recalled they typically provide at least an orientation in time. The consciousness of neither a speaker nor an addressee is able to function normally unless it has that kind of orientation. Thus, it is typical for a speaker to begin talking about a recalled experience by providing the kind of orientation that was illustrated in (2):

(5)

I went out for a stroll, on my first time on Chestnut Street,

These orientations, or settings, are necessary just because displaced experience is interrupted. Speakers know that they need to locate an ex-

Immediacy

and displacement

237

periential island in a way that is sufficient to orient the addressee's consciousness in space and time (on my first time on Chestnut Street), and typically with respect to some background activity (I went out for a stroll; Chafe 1980: 4 0 - 4 7 ) . Another indication of this concern for the addressee's orientation is the categorization of new referents in such a manner that addressees can locate them with relation to knowledge that is already shared. For example, the following speaker was introducing a story whose protagonist was the guy who was known to the speaker through the guy's sister, and who was second in command at Livermore Radiation Lab. (6)

You know, there's a Livermore Radiation Lab. That's where they manufacture Atron. And I knew the sister of the guy who was second in command.

The speaker made an obvious effort to place this person, who was previously unknown to the addressees, within their knowledge space. As they introduce new information, conversationalists take account of shared and unshared knowledge, orienting addressees to their islands of recall through categorizations as well as settings.

Fine-grained resolution Immediate experience is not only uninterrupted, but it also has access to a wealth of nonsalient, particular detail, all of which is potentially available for it to focus on. If I look at the vase of irises on the table beside me, I believe that I "see" everything that is there: the exact number of stems, leaves, and blossoms; their exact shapes and colors; the small unique markings on each; and so on. I need only to turn my head away to discover that what my consciousness retains is only a sparse interpretation of that richness. I can recall a general shape and texture, along with the few details on which I happened to focus while I was looking. As soon as I look back at the irises again, the availability of the myriad details on which I could focus my attention creates an immediate experience that is qualitatively very different from the displaced one. To the extent that recall is the principal source of the experiences on which it focuses, conversational language can be expected to manifest the coarse-grained resolution that is typical of displaced experience. That property is also evident in example (2), repeated here in its entirety.

238

(7)

Wallace Chafe

I went out for a stroll, on my first time on Chestnut Street, and just was astounded at how pleasant things were. And as I was out for a stroll, a man watering his lawn turned to me, as I was walking past, and said, "Good evening."

Mention was made of an orienting background situation, a general feeling of surprise at people's pleasantness, and a brief example that climaxed with the neighbor's greeting. These were the salient states and events, and no one would have expected this speaker to have talked about nonsalient details like the number and variety of trees along the street, the appearance of the neighbor's house, or what the neighbor was wearing. If he had noticed such details at the time, they would probably not have been recalled, and even those that were recalled would not have been judged interesting or relevant enough to be verbalized. The richness of detail available to his earlier immediate consciousness was thus modified at several points in the total process: first during perception, then during storage, then at the time the incident was recalled during the conversation, and finally through the selectivity that is always involved in putting ideas into words. It may be objected that conversational language often does include more detail than I have allowed for. Certainly conversationalists often mention objects, events, or states that seem trivial, such as the name Chestnut Street. Tannen (1989: 140, 144) mentions such details as particular dates or house numbers that conversationalists often include, but that seem not to matter to the addressee. But such isolated details are for one reason or another salient to the speaker, and are by no means equivalent to the comprehensive, fine-grained resolution that is unique to immediate experience. There is, however, one circumstance under which events are recalled with a degree of resolution which, though probably never equivalent to what is available to immediate experience, goes beyond the coarse-grained resolution illustrated in (7). Some experiences are remembered genetically: as events that were experienced in numerous distal consciousnesses and are later recalled as types rather than tokens. When storage and recall are aided by rehearsal, events repeatedly refreshed are more richly recalled. Furthermore, it is likely to be the relative salience of such events that motivates their rehearsal in the first place, so that when they are recalled they are likely to be judged more worthy of telling. The following example is taken from an account of the speaker's brush with death while he was diving off the coast of California. He was not

Immediacy

and displacement

239

only an expert diver but also an expert storyteller, and his account was a work of oral art. This segment of his story began with mention of his affective state: (8)

And after the long swim out, and my wrestling with the seaweed, I was pretty tired.

At this point, having created an expectation of something exciting to follow, he interrupted the flow of particular events with a suspenseful description of the diving environment, before returning to the main sequence with the word anyway: (9)

You don't feel that tired down below, you're down there, you're taking pictures, there're anemone down there, that're I would say a foot and a half in diameter, like huge, white, flowers. With green, thick, green, stems. Their stems aren't really green. When you take a picture of them, the strobe light shows that they're a bright red. Anyway,

Though it describes a displaced experience, this excerpt approaches a degree of resolution that would be characteristic of immediate experience. All of it, however, is generic. Linguistic evidence is provided by the use of the generic you throughout, as well as the generic uses of the present tense — you're down there — and the progressive aspect — you're taking pictures (see Chafe 1970: 168 — 178 for a discussion of generic tense and aspect). In conversational language this kind of resolution is largely restricted to generic recall.

Deixis Language provides various so-called deictic elements that serve to locate experience in space, time, and with respect to an ego. It will be important here to distinguish two major classes: (1)

(2)

Adverbial deixis: linguistic elements that refer to the space or time of immediate or displaced experience within proximal consciousness, and Tense and person deixis: linguistic elements that relate the time or ego of proximal consciousness to the time or ego of the experience being focused on.

240

Wallace Chafe

In the first class are spatiotemporal adverbs like here and now, which ordinarily refer to the space and time of immediate experience, and there and then, which refer to the space and time of displaced experience. In the second class are markers of tense and person. Tense has to do with the temporal relation between proximal consciousness and the experience being focused on. With immediate experience there is no difference between the time of proximal consciousness and the time of the experience being focused on, and the use of present tense conveys that fact. With displaced experience the time of proximal consciousness and the time of the experience being focused on are different. Past tense "means" that proximal consciousness is focused on experience from a temporally prior distal consciousness. Future tense "means" that proximal consciousness is focused on experience from a distal consciousness anticipated to follow it in time. Person deixis will be discussed below. In conversational language, the functions of adverbial and tense deixis are compatible. For example, when proximal consciousness is in the immediate mode, the adverb now refers to the time of the experience and present tense signals a coincidence in time between proximal consciousness and the experience being focused on. Thus, the use of now is compatible with the use of the present tense. The same can be said for the use of then and the past and future tenses. Then refers to the time of a displaced experience, and the past and future tenses express differences between the time of proximal consciousness and the time of the consciousness reflected in a displaced experience. These compatibilities of adverbial and tense deixis may change when language is used in other ways.

How writing may modify the basic structure of displacement The qualitative differences between immediate and displaced experience are differences observable in ordinary conversational speaking, and represent a baseline from which language may deviate when it is used in other ways. Our interest here is in writing, which can create an environment in which it is easy for displaced experience to share some of the characteristics of immediate experience. Since my objective is to discuss some of the ways in which writing may overcome the natural differences between displaced and immediate experience, and since the possibilities are especially great in written fiction, I will limit what I say here to

Immediacy

and displacement

241

samples from that area. First, however, it is necessary to set the stage by looking at situational properties that distinguish writing from conversing. There are two basic oppositions that are especially relevant: (1) the question of whether the producer and receivers of the language are copresent, and (2) the question of whether they interact. Conversational language typically involves participants who share the same location in space and time, and who alternate in their roles: the individual who is the speaker at one moment may be the addressee at the next. Writing is different in both respects. Typically the writer and readers occupy distinct locations. It is, of course, one of the chief advantages of writing that it allows this separation, though it does so at the sacrifice of advantages associated with copresence, above all the possibility for direct interaction. Writing is typically an isolated activity. In conversational consciousness, awareness of the immediate spatiotemporal environment is always present in the peripheral, if not the focal consciousness of the interlocutors. Similarly, conversational interlocutors are always peripherally if not focally aware of their own and each other's identities. Writing consciousness is quite different. Because of the absence of copresence and interaction, the location of the writer may be, and typically is, irrelevant to what is written. Even more irrelevant is the location and even the identity of the readers, who are typically unknown to the writer, though he or she may be aware of a class of people for whom the language is intended. The result of this desituatedness can be that the immediate experience of the writer has no direct bearing on the language produced, so that the writer deals exclusively with displaced experience. I should point out that this limitation to displaced experience is a separate matter from the alleged autonomy of at least some written language, as discussed most notably by Olson (1977). It is not that writing is devoid of a context, but that the immediate context in which it is performed is seldom the focus of attention.

Why writing is hospitable to fiction It is in written fiction that normal constraints on displaced experience are especially open to modification. There is a special compatibility between writing and fiction. I do not mean that oral traditions are impoverished when it comes to fiction, but rather that the deliberate composition of elaborate fictional works by unique individuals is something for which writing provides an especially happy environment. The

242

Wallace Chafe

constant proliferation of individually authored novels, short stories, and dramas of all sorts and sizes is an obvious feature of literate societies. There may be several reasons. Undoubtedly the desituatedness of writing has something to do with it. The writing situation is itself unreal in its detachment from the copresence and interaction which are the normal conditions of language; writing of any kind is in that respect intrinsically unreal. The further step to total fictionality is not a big one. Writers, too, are particularly blessed with the leisure that is conducive to imaginative creation. In many writing situations there is ample time to indulge oneself in contemplative imagining between and during the conversion of imagined experiences into visible words. Beyond that, face to face interaction tends to keep conversationalists in a factual mode, whereas a faceless, unknown audience eliminates that barrier to imaginative license. Removed from eye contact, writers can be less concerned with the need to adhere to what is taken to be the truth. It is not that writing encourages prevarication, though it may sometimes do that as well, but that it is especially hospitable to the spinning of carefully thought out fictions for both enjoyment and edification. Finally, once it has been entered into, fiction snowballs. The writer's imagination may acquire a momentum that leads from fantasy to fantasy. The basic unreality of the writing situation, the availability of contemplative leisure, the absence of face to face interaction, and the snowballing effect all support a condition than can be summarized in the aphorism, "Writing fosters fiction." In written fiction, the author's own consciousness cannot be the proximal consciousness expressed by the language, since the author is real and what the language expresses is not. The author's proximal consciousness and the proximal consciousness expressed by the language necessarily belong to different worlds. An author is thus free either to take on a fictional identity, or to relinquish any identity at all. The remainder of this discussion will focus on two specific realizations of these two options.

An example of written fiction with an invented ego One of the ways an author may deal with the split between his own proximal consciousness and the proximal consciousness expressed by the language is to pretend that the two are the same. Since they cannot really be the same, the solution is a fictional author. Hence the strategy of writing with a fictional first person narrator, into whose proximal con-

Immediacy

and displacement

243

sciousness distal experiences are recalled. This strategy bears a superficial resemblance to the situation in which someone narrates a personal experience in the midst of a conversation. But, aside from the fact that ego is fictional rather than real, the consciousness expressed by such language has a very different character from anything a conversationalist would produce. Here is the example introduced in (3), with the rest of the paragraph added: (10)

Gil and I crossed the eastern divide about two by the sun. We pulled up for a look at the little town in the big valley and the mountains on the other side, with the crest of the Sierra showing faintly beyond like the rim of a day moon. We didn't look as long as we do sometimes; after winter range, we were excited about getting back to town. When the horses had stopped trembling from the last climb, Gil took off his sombrero, pushed his sweaty hair back with the same hand, and returned the sombrero, the way he did when something was going to happen. We reined to the right and went slowly down the steep stage road. It was a switch-back road, gutted by the run-off of the winter storms, and with brush beginning to grow up in it again since the stage had stopped running. In the pockets under the red earth banks, where the wind was cut off, the spring sun was hot as summer, and the air was full of a hot, melting pine smell. Rivulets of water trickled down shining on the sides of the cuts. The jays screeched in the trees and flashed through the sunlight in the clearings in swift, long dips. Squirrels and chipmunks chittered in the brush and along the tops of snowsodden logs. On the outside turns, though, the wind got to us and dried the sweat under our shirts and brought up, instead of the hot resin, the smell of the marshy green valley. In the west the heads of a few clouds showed, the kind that come up with the early heat, but they were lying still, and over us the sky was clear and deep.

One of the ways this language departs from that of a conversational narrative is obvious: it would be hard to imagine a conversation in which one of the participants was allowed to hold the floor long enough to produce language that would fill several hundred pages. More interesting to the present discussion is the nature of the displaced experience expressed by such language. The normally constrained recall that mediates

244

Wallace

Chafe

between immediate and displaced experience is replaced by unconstrained recall. The author has invented a narrator who has unrealistic powers of reactivating past experience, powers manifested in significant changes in the quality of displaced experience. In effect, displaced experience is no longer qualitatively different from immediate experience. There is a pretense that proximal consciousness is focused on a displaced experience that is a replica, not a reconstruction, of an earlier immediate experience. With respect to continuity, fine-grained resolution, and deixis such a displaced experience is indistinguishable from an immediate one. Under such conditions we might say that language expresses a displaced immediacy. Let us look at each of the ways in which this displaced immediacy is manifested.

Continuity We saw above how conversationalists are motivated to provide an orientation for each new topic, typically through the provision of a temporal setting and the explicit introduction of any important referent whose identity is not already accessible. If (10) were translated into conversational language such a setting would normally be present. Perhaps a speaker would begin by saying something like: (11)

Last spring I was out riding with my friend Gil. Did I ever tell you about him?

(14), in contrast, begins in medias res. There is no introduction of Gil, and no explanation of what the two companions were doing. The fact that they were crossing the eastern divide, or that it was about two by the sun is of little help in providing the kind of spatiotemporal orientation that would satisfy a conversational interlocutor. The reader is given the impression of an immediate experience that has flowed without interruption out of whatever preceded it. This impression is reinforced by the pretense of identifiability for the eastern divide, the little town, the big valley, and so on. The definiteness of these noun phrases implies that their referents are already shared and identifiable. But here there is no identifiable addressee to do the sharing. The definiteness arises from the fact that these referents are identifiable to the narrator alone, in the midst of his ongoing, pseudo-immediate experience (cf. Fillmore 1974: V —17).

Immediacy

and displacement

245

Fine-grained resolution The most spectacular property of this language, that which differentiates it most obviously from something that might be told in a conversation, is its access to fine-grained detail. A fiction writer is able to supplement the coarse-grained detail associated with the recall of real experiences with a wealth of nonsalient information. The result is not that the audience is treated to details which, in themselves, might seem irrelevant. Language like this establishes the illusion that the experiences it conveys are immediate to the time and place of the events being verbalized; that the displaced consciousness, since it has access to such detail, is not displaced at all, but is an immediate consciousness that coincides with the reported events. It is this device more than any other that creates the impression that the proximal and distal consciousness expressed by the language are equivalent, the impression of displaced immediacy. It is interesting to speculate on how a writer is actually capable of providing such detail. If it is not available to ordinary constrained recall, how does a writer have access to it? Painters typically gain access to detail by locating themselves in the environment itself, painting from a model or landscape that provides them with constant perceptual access. While it is possible for a writer to do that also, one suspects that Clark did not sit on a horse on the eastern divide, observing the rivulets of water trickling down on the sides of the cuts and the jays screeching in the trees as he wrote about them. In practice such details usually arise from a writer's imagination. But is imagination not subject to constraints similar to those that affect recall? To imagine a vase of irises is more like remembering it than like seeing it. I suspect that the kind of detail authors are able to present in passages like (10) results from a blending of imagination with generic recall. Having seen rivulets of water, having heard jays screeching on repeated occasions, Clark was able to reconstruct a fine-grained experience of this kind.

Deixis Earlier we noted that words like now and today refer to the time of a proximal immediate experience. In writing such as we are now considering, since the distal immediate experience is replicated in proximal consciousness, these adverbs can function as additional indicators of unconstrained recall:

246

(12) (13) (14)

Wallace Chafe

Monty had been a kind of half-hearted rider once, but now he was the town bum, (p. 7) [The bar] was clean and dry now, (p. 8) I didn't see how anybody could find anything to laugh at today, (p. 203)

This combination of immediate deixis with the past tense has often seemed unusual. Ordinarily there is an incompatibility, since now, for example, refers to the time of a proximal immediate experience, while past tense signals a displaced focus on a distal consciousness that is temporally prior to proximal consciousness. Now indicates a focus on a proximal immediate experience, past tense on a displaced experience. But when properties of immediacy have been transferred to displaced experience, the use of immediate adverbs like now is another signal of this displaced immediacy. One may wonder at this point why such writing retains the past tense, and thus maintains a separation between proximal and distal consciousness, even though the latter has acquired properties of immediacy. I will return to this question below, when it can be discussed in the context of ego displacement in addition to the spatiotemporal displacement examined so far. Just now, however, we can note that this narrator occasionally exploits the distinction between proximal and distal consciousness which the present and past tenses serve to keep apart.

The occasional recognition of proximal immediate experience The kind of writing illustrated in (10) represents a proximal consciousness that has potential access to two kinds of immediate experience. During most of the novel the focus is on a displaced experience into which have been transferred properties of an earlier immediate experience, as just discussed. Proximal immediate experience is irrelevant. Occasionally, however, just as ordinary proximal consciousness may alternate between displaced and immediate experience, the focus here may also shift to the immediate environment of the narrator. The effect is to remind the reader that the language is being produced by a (fictional) proximal consciousness that is still separated in time from the distal events on which most of the novel is focused: (15)

Now I can see that he was perhaps still having a struggle with himself that he was here at all, but then it just angered me that

Immediacy

(16)

and displacement

247

one of us failed to be alert; then it just seemed to me that he was too scared to know what to do, and I got furious at him for a moment, the way you will when you think another man's carelessness is risking your neck. (p. 143) It was a heavy wind with a damp, chill feel to it, like comes before snow, and strong enough so it wuthered under the arcade and sometimes whistled, the kind of wind that even now makes me think of Nevada quicker than anything else I know. (p. 53)

It is interesting to compare the kind of immediacy expressed by the word now in (15) and (16) with the displaced immediacy expressed by the same word in (12) and (13). In (15) and (16) we recognize that the narrator is momentarily looking back on the events from the vantage point of the time at which he is writing. The effect, however, is a curious one. In (16), for example, it is the immediate quality of the experience conveyed by the beginning of this excerpt that reminds the narrator, as he is narrating, of his generic reaction to such an experience. This is no ordinary kind of reminding, where some immediate experience activates a temporally displaced one. In this case the protagonist's detailed experiencing of the earlier events has served to remind him of events he has experienced generically in the meantime.

Ego displacement U p to this point we have focused largely on spatiotemporal displacement: the ability of proximal consciousness to focus not just on its own immediate environment, but also on displaced experience imported from some other place and/or time. But proximal consciousness also has the ability to focus on displaced experience imported from a different ego, or alter, and in that sense we can speak of ego displacement. Whereas the abilities that underlie spatiotemporal displacement are recall and imagination, ego displacement is made possible by empathy. Empathy, however, is nothing more than a special use of imagination: the ability to imagine what it is like to experience a consciousness different from one's own. It is within the power of a fiction writer to pretend that proximal consciousness has unconstrained and direct access to alter's immediate experience. What is then verbalized is equivalent to the immediate experience of some other person at some time in the past. The language

248

Wallace Chafe

that flows from proximal consciousness expresses a replica of alter's immediate experience at an earlier time. With ego's own identity irrelevant because of the desituatedness of writing, there is no recognition of ego's own distal consciousness. Nevertheless, there remains a distinction between the proximal consciousness from which the language flows and the distal consciousness which it expresses, a distinction manifested in the use of past tense and third person. Why the distinction should be maintained under these circumstances is a matter to which I will return below.

An example of written fiction with ego displacement An excellent example of this situation is provided by Ernest Hemingway's short story "Big Two-Hearted River". It deals with a fishing trip by a lone fisherman. The entire reason for the story seems to be its attempt to put the reader inside the protagonist's consciousness so that he or she can experience what the protagonist was experiencing. The first paragraph, already quoted above in (4), reads as follows (Hemingway 1987: 163): (17)

The train went on up the track out of sight, around one of the hills of burnt timber. Nick sat down on the bundle of canvas and bedding the baggage man had pitched out of the door of the baggage car. There was no town, nothing but the rails and the burned-over country. The thirteen saloons that had lined the one street of Seney had not left a trace. The foundations of the Mansion House hotel stuck up above the ground. The stone was chipped and split by the fire. It was all that was left of the town of Seney. Even the surface had been burned off the ground.

As before, I will discuss the language of this story in terms of continuity, fine-grained resolution, and deixis.

Continuity Like (10), (17) begins in medias res. Certain referents and events are in Nick's consciousness at this moment. He is already thinking about a train that is moving away from a known location. As we go on, the baggage man and his act of pitching the bundle out of the baggage car is accessible

Immediacy

and displacement

249

information. The language reflects what is active or semiactive in Nick's mind from the beginning. All that is left to verbalize are Nick's ongoing new experiences as the story proceeds. The language also reflects Nick's background knowledge, as shown by the references to the train, the hills of burnt timber, the baggage man, the town of Seney, and the Mansion House hotel. The definiteness of these noun phrases implies that their referents are already known and identifiable. It is not that the knowledge is shared with an addressee, as is normally the case with identifiable referents. There is no one else to do the sharing. The language comes from Nick's mind, and it knows what he knows.

Fine-grained Resolution The effect of this story, the reliving of Nick's immediate experience, derives especially from the inclusion of detail that shows a degree of resolution not normally found in a conversation: (18)

Nick put the frying pan on the grill over the flames. He was hungrier. The beans and spaghetti warmed. Nick stirred them and mixed them together. They began to bubble, making little bubbles that rose with difficulty to the surface. There was a good smell. Nick got out a bottle of tomato catchup and cut four slices of bread. The little bubbles were coming faster now. Nick sat down beside the fire and lifted the frying pan off. He poured about half the contents out into the tin plate. It spread slowly on the plate. Nick knew it was too hot. He poured on some tomato catchup. He knew the beans and spaghetti were still too hot. He looked at the fire, then at the tent, he was not going to spoil it all by burning his tongue, (pp. 167 — 168)

One may recall the way bubbles rise to the surface as food is cooking over an open fire, but recall of that kind is ordinarily generic and not the recall of a particular event. The detail of (18), like that of (10), reflects an immediate, not a displaced experience.

Deixis The initial paragraph of the story, cited in (17), establishes the location of an ego-displaced consciousness in space and time. To read that the train went on up the track out of sight is to enter into the consciousness

250

Wallace

Chafe

of the person from whose deictic center it went on up, and from whose sight it disappeared. When Nick is introduced in the second sentence there is already an understanding that the language expresses the immediate experience of the person so named. The kind of use of now and today that appeared in examples (12) through (14) can be observed here as well, signaling again the equivalence of a proximal displaced experience to a distal immediate experience: (19)

(20)

Now as he looked down the river, the insects must be settling on the surface, for the trout were feeding steadily all down the stream, (p. 166) Nick did not want to go in there now ... He did not want to go down the stream any further today, (p. 180)

Why is the proximal—distal distinction maintained? One may ask why, if the author's intent is to allow the reader to experience alter's immediate experience, there remains a residual reflection of a proximal consciousness that is different from alter's. Why doesn't the author avoid displacement altogether, and simply express directly what passes through alter's mind? In that case alter would no longer be alter, for no other consciousness would be involved: the ultimate consequence of the desituatedness of writing. The residual presence of a proximal consciousness that is different from alter's is evident in the use of past tense and third person. If (18) were modified to replace past tense with present and third person with first, it would read as follows: (21)

I put the frying pan on the grill over the flames. I am hungrier. The beans and spaghetti warm. I stir them and mix them together. They begin to bubble, making little bubbles that rise with difficulty to the surface. There is a good smell. I get out a bottle of tomato catchup and cut four slices of bread. The little bubbles are coming faster now. I sit down beside the fire and lift the frying pan off. I pour about half the contents out into the tin plate. It spreads slowly on the plate. I know it is too hot. I pour on some tomato catchup. I know the beans and spaghetti are still too hot. I look at the fire, then at the tent, I am not going to spoil it all by burning my tongue.

One may speculate on why authors usually avoid this alternative, preferring to maintain the residual proximal-distal distinction that is

Immediacy

and displacement

251

signaled by the tenses and persons of (17). For one thing, (21) is farther removed from conversational language. As noted earlier, it is less common in ordinary speaking to talk about proximal immediate experience. Whereas the use of past tense, as in (10), or of past tense and third person, as in (17), retain basic properties of ordinary language, (21) is more deviant in that respect. Still, both (10) and (17) deviate from ordinary usage in their use of unconstrained recall and empathy, so one wonders why the additional deviation that eliminates a proximal consciousness altogether should be less attractive. It would seem that the effects produced by (10) and (17) depend crucially on the direct, unmodified importation of a distal immediate experience into a proximal consciousness. When the proximal-distal distinction is obliterated, as in (21), that effect is destroyed. Art depends fundamentally on devices that "hold the mirror up to nature." In (10) and (17) the mirror is proximal displaced consciousness and nature is a distal immediate consciousness. To remove that duality, to express only a single consciousness as in (21), is like replacing a statu with a living model, a portrait with its subject. The residual presence of a proximal consciousness in (10) and (17), a consciousness responsible for the language, stimulates our powers of interpretation and thereby our aesthetic appreciation, causing us to marvel at what can be done with language. (21) destroys that effect.

Summary and unfinished business I have tried to construct a coherent picture of consciousness and language within which various effects that may differentiate written fiction from spoken conversation can be naturally explained. Central to this picture is a proximal consciousness that alternates easily between an immediate and displaced mode. In the immediate mode it focuses on the immediate environment, in the displaced mode on experience that was, will be, or might be immediate to some other, distal consciousness. It is most often the case in conversational language that the focus is on displaced experience. Under ordinary circumstances immediate and displaced experience is qualitatively different. First, whereas immediate experience is part of an uninterrupted flow, displaced experience has an island-like quality that motivates the provision of a setting, as well as the inclusion of proper introductions for important unshared referents. Second, whereas imme-

252

Wallace Chafe

diate experience has access to a wealth of fine-grained nonsalient detail, displaced experience does not. Fine-grained resolution is approached only when the experience is a generic one that has been reinforced through rehearsal. Third, language provides various deictic elements, among which are (1) adverbs that index immediacy or displacement, referring to the space or time of proximal immediate or displaced experience, and (2) markers of tense and person, signaling the temporal and ego relation of proximal consciousness to the experience being focused on. Under ordinary circumstances these two kinds of deixis are compatible. In written fiction a writer may verbalize displaced experience with devices normally reserved for immediate experience, thus pretending that the language expresses a distal immediate experience directly, not being subject to the ordinary constraints of recall. First, by beginning an episode in medias res and by treating referents as already shared and identifiable, a writer can pretend that the episode is part of an uninterrupted flow and not an experiential island. Second, by providing a richness of finegrained detail a writer can create the illusion of an immediate consciousness that has access to such detail. Third, a writer may refer to displaced experience with deictic elements that normally refer to immediate experience. I ended by speculating on why, under these circumstances, a writer may nevertheless preserve the past tense, and sometimes third person, that maintain a separation of proximal from distal consciousness. Why not surrender entirely to a fictional immediate consciousness? I suggested two answers. One is that the kind of language that fully expresses immediate experience is unnatural and awkward when used for narrative purposes. We are accustomed to past tense narration. A more intriguing answer, perhaps, is that art depends on the mirroring of life, not life itself. Thus, the retention of a residual proximal consciousness through the use of past tense and third person retains the distance between representation and reality that is essential to art in all its forms. I warned that I would deliberately avoid an entire range of additional devices by which a fiction writer may import qualities of immediacy into displaced experience. I have in mind the fact that the language expressed by proximal consciousness is frequently used to convey other language that originated in some other, distal consciousness. That other language may have been overt, as in some earlier conversation, or it may have been covert, as in someone's inner speech, for the contents of consciousness may take either a nonverbal or a verbal form. Perhaps the most spectacular thing a fiction writer can do is to represent the language that

Immediacy

and displacement

253

is passing through the mind of another person. Under ordinary circumstances we may infer something of other people's consciousness from their appearance or behavior, but it is never possible to know the words within another consciousness unless those words are made overt. Fiction writers, however, have it in their power to overcome this constraint as well. Thus, passages like the following from Hemingway's story convey a particularly striking kind of displaced immediacy: (22)

He'd bet the trout was angry. Anything that size would be angry. That was a trout. He had been solidly hooked. Solid as a rock. He felt like a rock, too, before he started off. By God, he was a big one. By God, he was the biggest one I ever heard of.

This passage goes beyond Nick's perceptions, actions, and feelings as illustrated in (18), where they remained unverbalized, to express words that Nick was thinking to himself. That it represents verbatim language in Nick's mind is evident from the presence of wordings that can only be language itself, not indirect reports of language: wordings like He 'd bet, that was a trout, and the repeated exclamation by God. In the last sentence of (22) the word I, in place of Nick, comes as a surprise. For one brief moment the third person indicator of distance between proximal consciousness and the experience being verbalized is also eliminated. In this one sentence even proximal consciousness belongs almost completely to Nick. I say "almost" because the past tense remains to show the distinction. If Hemingway had written By God, he's the biggest one I ever heard of, the effect would coincide with that of (21). I have ended with this example to suggest that there is much more to take into account as soon as we examine language that represents language, whether it is the overt language of distal conversations or the covert language of distal thought. Although much has already been written on that subject (among works in the linguistic tradition are, for example, Fillmore 1974; Leech and Short 1981; and Ehrlich 1990), I believe that viewing it within the framework of displaced consciousness can add to our understanding of it, as I will try to show in a subsequent longer work (Chafe, forthcoming).

254

Wallace Chafe

Notes 1. This study, part of a larger study of speaking and writing, was supported by the Center for the Study of Writing at Berkeley, which is funded by the Office of Educational Research and Improvement of the U. S. Department of Education. 2. The data that underlie this discussion come from two separate studies. The first (Chafe and Danielewicz 1987) included twenty dinnertable conversations among academic people. The second, still in progress, includes five conversations among diverse people outside the academic world.

References Banfield, Ann 1982

Unspeakable sentences: Narration and representation in the language of fiction. Boston: Routledge and Kegan Paul. Bartlett, Frederic C. 1932 Remembering: A study in experimental and social psychology. Cambridge: Cambridge University Press. Chafe, Wallace 1970 Meaning and the structure of language. Chicago: University of Chicago Press. 1974 "Language and consciousness", Language 50: 111 — 133. 1976 "Givenness, contrastiveness, definiteness, subjects, topics, and point of view", in: Charles N. Li (ed.), 25 — 55. 1980 "The deployment of consciousness in the production of a narrative", in: Wallace Chafe (ed.), 9 - 5 0 . 1987 "Cognitive constraints on information flow", in: Russell Tomlin (ed.), 21-51. Discourse, consciousness, and time: The flow and displacement of conscious experience in speaking and writing, forthcoming Chafe, Wallace (ed.) 1980 The pear stories: Cognitive, cultural, and linguistic aspects of narrative production. Norwood, NJ: Ablex. Chafe, Wallace — Jane Danielewicz 1987 "Properties of spoken and written language", in: Rosalind Horowitz — S. Jay Samuels, 8 3 - 1 1 3 . Clark, Walter Van Tilburg 1940 The Oxbow Incident. New York: Random House. Ehrlich, Susan 1990 Point of view: A linguistic analysis of literary style. London: Routledge. Fillmore, Charles J. 1974 "Pragmatics and the description of discourse", Berkeley Studies in Syntax and Semantics, Volume I. Berkeley: Department of Linguistics, VI — V21. Hemingway, Ernest 1987 The complete short stories of Ernest Hemingway: The Finca Vigia Edition. New York: Charles Scribner's Sons.

Immediacy and displacement

255

Hockett, Charles F. 1960 "The origin of speech", Scientific American 203/3: 89 — 96. Horowitz, Rosalind — S. Jay Samuels (eds.) 1987 Comprehending oral and written language. New York: Academic Press. James, William 1989 The principles of psychology. New York: Henry Holt. Leech, Geoffrey N. — Michael H. Short 1981 Style in fiction: A linguistic introduction to English fictional prose. London: Longman. Li, Charles N. (ed.) 1976 Subject and topic. New York: Academic Press. Olson, David R. 1977 "From utterance to text: The bias of language in speech and writing", Harvard Educational Review 47: 257 — 281. Tannen, Deborah 1989 Talking voices: Repetition, dialogue, and imagery in conversational discourse. Cambridge University Press. Tomlin, Russell (ed.) 1987 Coherence and grounding in discourse. Amsterdam: John Benjamins.

Part III Literary texts: inferring meaning(s)

Fictional conversation and its pragmatic status1 LuMing R. Mao

John Searle, in his "The logical status of fictional discourse" (1979c: 58 — 75), argues that fictional discourse is constituted by illocutionary acts whose normal illocutionary commitments and "felicity conditions" are suspended, or rendered inconsequential by a set of horizontal conventions of fiction.2 It then follows that when Iris Murdoch starts her novel The red and the green with this passage: "Ten more glorious days without horses! So thought Second Lieutenant Andrew Chase-White recently commissioned in the distinguished regiment of King Edward's Horse, as he pottered contentedly in a garden on the outskirts of Dublin on a sunny Sunday afternoon in April nineteen-sixteen" (Searle 1979c: 61), she is not committed to the truth of the proposition asserted in the passage; nor is she obligated to provide evidence to make the truth stand. What she is doing, so the argument goes, is performing a series of speech acts which is pretended or "parasitic" (Austin 1975: 22) in such a way that it allows Murdoch, or any fictional writer for that matter, to convey some serious message(s). Miss Murdoch accomplishes all of these with the help of the set of horizontal conventions of fiction — one she invokes by performing utterance acts (i. e., writing her novel). As a result, fictional discourse becomes distinctively different from our normal discourse in spite of their ostensible resemblance to each other, since the normal rules relating illocutionary acts and our real world have been "impaired" or "incapacitated" in fiction.3 Definitely, Searle is not the only proponent of this theory. Richard Ohmann echoes the same line of argument (1971: 1 — 19). Specifically, he contends that the writer pretends to "report discourse" with the usual illocutionary rules suspended, but with its acts imitating a series of our real world speech acts which does not have any other ontological existence. Because of that, the reader, with the help of her/his real world knowledge, is able to construct a speaker, a situation and a set of other related events, and eventually to secure associated perlocutionary effects (14 —17).4 In a similar vein, Susan Lancer (1981: 2 8 3 - 2 9 4 ) , in light of Searle's five-category taxonomy of illocutionary acts (1979: 1—29), proposes that fictional discourse is a class of hypothetical speech acts that

260

LuMing R. Mao

behaves differently from the other illocutionary acts, because it "brings a hypothetical or alternative world into existence and commits him- or herself [the writer] to representing that world consistently" (290). But, as a matter of fact, fictional discourse, being one of many language games within one of many external worlds, consists of full-fledged speech acts hooked up only to one external, that is, fictional world, by the same semantic and pragmatic rules.5 These rules are not "obstructed" to be nonoperative by the set of horizontal conventions as so claimed. On the contrary, they become capable of relating fictional discourse to the fictional world just because of the existence of this set of horizontal conventions of fiction. Otherwise stated, the conventions of fiction actualize the fit between fictional discourse and the fictional world that would otherwise remain dormant, and subsequently connect the properties and meanings of fictional discourse to that fictional world. As regards the reader, it is also the conventions of fiction that alert or cue her/him to applying her/his pragmatic knowledge and knowledge about fiction so as to make appropriate connections between fictional discourse and the fictional world, or to "make sense" of their relationship. An intelligent application on the part of the reader further leads her/him to discover the illocutionary intention of the creator and the perlocutionary effect of her/his creation. Meanwhile, within the fictional world, both the narrator, as often distinct from the writer — a distinction that Searle fails to observe,6 and characters are equally subject to "felicity conditions" of speech acts and conversational maxims. Any violation of felicity conditions or maxims will subsequently give rise to various conversational implicatures if the participants in the fictional world have not indicated any sign of opting out of their cooperative interaction; or they will, with all the more importance, contribute to formulating some "meta-implicatures" or underlying themes on behalf of their creator if they ostentatiously opt out. Therefore, the narrator, rather than Miss Murdoch, is making a fullblooded assertion in The red and the green, and she will be held responsible in one form or another, if she or Second Lieutnant Andrew Chase-White later denies the fact that he was pottering in a garden on the outskirts of Dublin on a sunny Sunday afternoon in April nineteen-sixteen. The writer will be held accountable, too, if she/he appears to be noncooperative. The writer, by creating the narrator (first person or third person, "felicitous" or otherwise) and the characters, ("cooperative" or "sullen"), is expected to convey to the reader her/his illocutionary intent with the corresponding perlocutionary effect. And, in the case of the

Fictional

conversation

261

writer being considered identical with the narrator — such as the narrator in Anthony Trollope's novels being thought of by Henry James as Anthony Trollope himself (Hancher 1977: 1093), the writer or the narrator is communicating with both the fictional world and our real world. In turn, the reader attempts to recreate the total interpretive, pragmatic context by means of her/his pragmatic knowledge and of five dimensions — authorial, generic, collective, specific and textual (Harris 1988: 82 — 122). Briefly, these dimensions include our knowledge about the writer relevant to any specific pragmatic reconstruction, about the genre with which a particular discourse is expected to align itself, about the general background and intertextual information, about the awareness specific to the particular discourse, and about the total textual givens such as events, scenes, etc. Fictional conversation is no less liable to pragmatic principles and constraints, and all participants in fictional conversation are just as rational and cooperative in honoring Gricean conversational maxims 7 either by observing or flouting them to generate "standard" 8 or conversational implicatures. Furthermore, fictional conversation functions in a well-structured, hierarchical manner. It contains a series of local speech acts, which constitutes part and parcel of verbal interaction among fictional characters. It also features some global speech acts that, on a higher, more abstract level, organize and structure associated local speech acts. These two kinds of speech acts are often referred to as "micro-" and "macro-speech acts", respectively (van Dijk 1981: 195 — 209). Since micro- and macro-speech acts have such an intertwined relationship, they tend to be construed as resembling the relationship of part and whole, element and set or member and class (van Dijk 1980: 3 — 9; Ferrara 1985: 141 — 158). While it is important to understand their affinity to such other cognitive formulations, it is equally crucial to realize that micro- and macro-speech acts do not have to be susceptible to the paradox of the hermeneutic circle. To put it simply, although micro- and macrospeech acts are interdependent of each other in forming a discourse, or fictional conversation, each has its own autonomous property and each claims its "existence", so to speak, by its own right. For example, a micro-speech act, be it an assertive or a directive, has its own characteristics, or properties; therefore, it can be understood on its own terms without being "referred up" to any macro-speech act. Its essential dimensions, such as the purpose of the act and the direction of fit between words and the world, 9 will remain intact regardless of what kind of macro-speech act it belongs to.

262

LuMing R. Mao

On the other hand, although our conception and full understanding of any macro-speech act within a discourse does, in one way or another, trace back to or call upon the meaning of low-level micro-speech acts, it has its own cognitive "necessity" (van Dijk 1981: 210). In order to secure overall discourse coherence, to comprehend local speech acts, we have to mentally formulate some macro-speech acts relative to the topic discourse but not necessarily derivable from each and every micro-speech act. Let us briefly look at three short fictions — the first two are written by American writers, and the third one by a Chinese writer — just to see how fictional conversation evolves in the fictional world and how it plays an instrumental role in depicting fictional characters and in bringing about associated implicatures on different levels. It seems fitting to first consider Ernest Hemingway's A clean, welllighted place, as the story is virtually dominated by conversation. The story itself is a simple one. It is about the two waiters, old and young, who disagree as to how late they should keep the old man at their cafe. While the old waiter shows sympathy to the old man and to his desire to stay late in this clean well-lighted place, the young waiter is more concerned about keeping his wife waiting for him at home. As the story advances to its destined conclusion where the old waiter finally succumbs to an overwhelming sense of despair and nonexistence, the conversation between the old and young waiter evokes a rich pragmatic context in which each character interacts with the other, and each is committed to his own acts. Their speech acts are further "embedded" within one macrospeech act, which highlights the thematic implicatures and leads to the global coherence of the story. The overall conversation in the story can roughly be divided into five parts, each part being separated by a brief narrative. The first four parts mainly comprise the interactions between the old and young waiter, and the last part between the old waiter and the bartender. The conversations between the old and young waiter are centered on the old, deaf patron, and on their splitting opinions over how late the old man should be allowed to stay. Towards the final moment of their conversation, the old waiter summarizes their difference in this way: "We are of two different kinds. It is not only a question of youth and confidence although those things are very beautiful. Each night I am reluctant to close up because there may be some one who needs the cafe." (Hemingway 1981: 473) The statement, "We are of two different kinds" — an assertive itself — is the topical sentence in this passage. More importantly, by unequivocally enouncing the difference between these two characters, it serves as

Fictional

conversation

263

a distinctive macro-speech act that sheds a great deal of light on their individual speech acts. In return, each part of their conversation enriches and adds a different dimension to this macro-speech act while maintaining its own distinctive properties. For example, the first part of the conversation introduces us to a brief account of the old man's attempt to commit suicide. It offers us the first insight into their differences: it is the old waiter that is privy to the old man's life of despair. And, through the old waiter's penultimate response, it foreshadows indirectly his pessimistic, nihilistic vision of the world — a different kind, indeed. The second part, which begins with the old waiter's "The guard will pick him up," 1 0 further explains why the young waiter is eager to go home, while the old waiter, extending his sympathy towards the soldier on the street, is willing to let the old man stay. The third and fourth parts of their conversations, in increasing detail and intensity, delineate their differences and culminate in the old waiter's above-mentioned revealing, summary statement. This kind of pragmatic, speech act perspective, I suggest, offers an alternative approach to the controversy over the attribution of certain lines of the conversation between the old and young waiter in the story — since sometimes Hemingway himself does not make it quite clear. The main dispute in question originates from the following stretch of their conversation after the young waiter becomes increasingly impatient at the old man's reluctance to go home, and after the old waiter has demonstrated his sympathy towards him: "He's lonely. I'm not lonely. I have a wife waiting in bed for me." "He had wife once too." "A wife would be n o g o o d to him now." "You can't tell. He might be better with a wife." "His niece looks after him." "I know. You said she cut him down." "I wouldn't want to be that old. An old man is a nasty thing." (472)

U p to this juncture, Hemingway has been fairly consistent in complying with the conversation of having each waiter alternate in turn. If such a structure applies to this part of their interaction, the line "I know. You said she cut him down." has to be ascribed to the old waiter, which contradicts what has been firmly established: it is the old waiter, rather than the young waiter, that knows of the old man's aborted suicide attempt due to his niece's timely intervention. To resolve this apparent discrepancy, Hagopian (1964: 140 — 146) has argued that the discrepancy is caused not by the violation of convention

264

LuMing R. Mao

but by a typographical error, which apparently has the endorsement of Scribner's.11 So, the line "You said she cut him down," which is infused with "all the overtones of his [the young waiter's] sadistic irony" (146), is only ascribable to the young waiter. Hence, it should be transposed to the preceding line. Assuming that the discrepancy does result from a typographical error — though there is no compelling textual evidence for that, and Hemingway himself after all does not correct this "error" in his life time, Hagopian's solution, I believe, still fails to be in harmony with the intended meaning of the story. First, nowhere in the story does Hemingway indicate that the young waiter is aware of the old man's niece taking care of the old man. Therefore, it does not square well by suggesting that "His niece looks after him" belongs to the young waiter except that it has kept the convention intact. But, convention can just be honored by flouting it as by observing it. Second, in attributing "I know" to the old waiter, Hagopian argues that it befits the old waiter since "I know" has the connotation of "I am perfectly aware of what happened and why — much more than you, you young puppy." (146) This, to say the least, is stretching. All his preceding local speech acts have not pointed to such kind of connotation. So far, the old waiter has patiently been persuading the young waiter to let the old man stay, and trying to present a different perspective on why an old man needs a clean, well-lighted place. On the other hand, "I know," without the benefit of prosodic information but with the benefit of its context, at least has the appearance of impatience and youthful arrogance, one that is more in keeping with the young waiter. Therefore, instead of attributing the apparent "unexplainable" to some kind of unconfirmed error, and then coming up with some forced interpretation so as to buttress the "error" argument, I suggest that we keep the text intact, and approach the disputed lines from a pragmatic, in particular, a macro- and micro-perspective. In particular, I want to argue that the contradiction here is no more than apparent and largely a result of our expectation that the writer is complying with the literary convention. However, the true picture is perhaps the opposite. Here, Hemingway flouts the convention of sequence and has each waiter speak two lines in a row. Hence, both "You can't tell. He might be better with a wife" and "His niece looks after him" should be attributed to the old waiter. Likewise, "I know. You said she cut him down" and "I wouldn't want to be that old. An old man is a nasty thing" belong to the young waiter.

Fictional

conversation

265

My reason is two-fold. First and foremost, after telling the young waiter that the old man might be better off with a wife, the old waiter certainly has reason to supply some additional information about his status quo. In no way does this new attribution conflict with what has been established so far and with the rest of the story. Moreover, it conforms to and enhances the above-mentioned global speech act that they are of the two different kinds. Not only in telling the story does the old waiter unequivocally make known where his sympathy lies, but also he reveals his own vision of how an old man can survive with less agony or suffering — by way of a wife, a niece or staying at a clean, well-lighted cafe. Secondy, it is clear that the young waiter is by now getting increasingly impatient because there is no sign that the old man is leaving soon. The next two lines reflect this mood as well as his own cold, hostile attitude towards the old man. They seem to call upon the reader to deduce the following implication: "I do not want to hear any more of it since you have already told me she cut him down." Once again, the attribution here yields one more concrete dimension to the global speech act: "We are of two different kinds". The flouting of the convention in this particular juncture thus enables Hemingway to generate a set of implicatures crucial to the development of the story. For one thing, with two lines being uttered by one speaker at one time, this kind of violation indicates that the seriousness and intensity of their conversation have reached a new height, and that their differences have become all the more visible — and irreconcilable. In this regard, it paves the way for the young waiter's upcoming confrontation with both the old man and the old waiter. On the other hand, by allowing the old waiter to volunteer the additional information concerning the old man ("His niece looks after him"), the violation "gives away" the old waiter's ever-growing interest in the well-being of the old man, and his effort, perhaps subconscious, to seek some kind of vicarious comfort. It prefigures the old waiter's final revealing moment when he himself looks for a clean, well-lighted place, a respite from despair and from that ubiquitous presence of "nada", i. e., nothingness. One question perhaps still lingers for this kind of interpretation, that is, if an allowance for deviation is made in one instance, would it be possible not to allow for the rest of their conversation similar interpretations of conflating those apparently ambiguous lines and then introducing some corresponding plausible implicatures? The answer, I believe, is no.

266

LuMing R. Mao

If we look at the story more closely, it becomes obvious that there is one place that has the possibility of attributing two lines in sequence to one speaker. This is where the third part of their conversation begins after the young waiter comes back from serving the old man one more drink: 12 '"He's drunk now,' he said./'He's drunk every night.'" (471) Undoubtedly, the young waiter initiates this part of the conversation. It is in agreement with the story that the second line is a response from the old waiter. But, since Hemingway has violated the convention of sequence elsewhere, would it be possible for him to do the likewise here, and for us to make one more exception? After all, there is no clear attribution here after the utterance. The answer is no if we take into consideration the global speech act of "We are of two different kinds." To have the young waiter speak the second line would run counter to his overall attitude toward the old man. In the context of the preceding conversation, "He's drunk now" partakes of the young waiter's negative assessment of the old man. At the same time, it indirectly blames the old waiter for allowing him to stay late, which has led to his being drunk now. This may, among other things, result in his departing without paying for his drinks. However, if we attribute the second line to the young waiter again, it will then contradict his alreadly established characterization, for "He's drunk every night" amounts to the speaker's attempt to find an excuse for the old man's being drunk now, and has the standard implicature of "What is the big deal" — a gesture that is more characteristic of the old waiter than of the young waiter. John Cheever's much anthologized The swimmer shows a pragmatic structure that is strikingly similar to that of A clean, well-lighted place. The four major conversations in the latter part of the story between Neddy Merrill, the protagonist, who decides to go back home by swimming across the numerous swimming-pools, and his neighbors and former mistress are organized in such a way that suggests their respective pragmatic structures. For example, in the wake of suffering an unexpected exposure to physical elements and to human (motorists') ridicule on the highway, and of experiencing mystification as a result of crossing the Lindley's riding ring overgrown with grass and discovering the Welchers' dry pool, Neddy comes to the Hallorans' pool. He signals his intrusion by a brief greeting so as to warn them of his presence, since they are swimming without bathing suits. The following is what takes place between Neddy and Mrs. Holloran, who is "not surprised nor displeased to see him" (Cheever 1981: 547):

Fictional conversation

267

"I'm swimming across the county," Ned said. "Why, I didn't know one could," exclaimed Mrs. Halloran. "Well, I've made it from the Westerhazys'," Ned said. "That must be about four miles." He left his trunks at the deep end, walked to the shallow end, and swam this stretch. As he was pulling himself out of the water he heard Mrs. Halloran say: "We've been terribly sorry to hear about all your misfortunes, Neddy." "My misfortunes?" Ned asked. "I don't know what you mean." "Why, we heard that you'd sold the house and that your poor children ..." "I don't recall having sold the house," Ned said, "and the girls are at home." "Yes," Mrs. Halloran sighed. "Yes ..." Her voice filled the air with an unseasonable melancholy and Ned spoke briskly, "Thank you for the swim." "Well, have a nice trip," said Mrs. Halloran (547).

Neddy's opening remark neatly observes Grice's first maxim of Quantity since it contains enough information as "required by the current talk of exchange," and it serves as an answer to her possible question because of his inopportune intrusion. It also purports to indirectly remove her possible disbelief or doubt as regards his plan to swim across the county. In spite of his effort, his whole adventure does not make sense to her, and she therefore objects to it mildly and with hedged surprise in order to attend to his "positive face" — a desire to be compliment or "stroked" (Brown — Levinson [1987]: 61—74). As he is leaving after offering her some "hard fact," Mrs. Halloran introduces a new yet unpleasant topic, one about Neddy's family misfortunes as a way of showing her sympathy and understanding. Neddy is apparently struck by the unexpected news, and asks for more information. But no matter how his response is construed, it is not an illogical utterance of dreams (Bell 1987: 435); nor could it be considered as an evasion since he is devastated in discovering in the end that his house is deserted. Rather, it is an indirect speech act of a felicitous request in the form of a question. Surprised by his ignorance, she thus grudgingly provides the minimum information to back up her statement, but much less than what she knows. She violates the first maxim of Quantity by not being as informative as she could. But her implicature is obvious: it is just too inappropriate for her to talk in detail about his misfortunes — a threat to his "positive face" again. Thus far, both Neddy and Mrs. Halloran have honored, in one way or another, the pragmatic principles of conversation. Meanwhile, this stretch of conversation contributes significantly to the global, or macrospeech act of the story that "a long swim might enlarge and celebrate its [the day's] beauty" (542) — but only in a highly ironic fashion: a long

268

LuMing R. Mao

swim is certainly to enlarge his vision of the "true beauty" and to lead him to the abysmal, catastrophic existence in the midst of communal revelry a and bacchanalia. Similarly, the rest of their conversations are equally pragmatics-oriented. Each segment has its own distinctive topic, and each generates similar implicatures about Neddy's own rejection, isolation, and ultimate defeat. Each segment at the same time becomes progressively brief and vacuous in content, ranging from asking for a simple drink to again asking for a drink and yet being abjectly rejected at the hands of his former mistress — in such a way that complements the gradually empty, annihilating outcomes of Neddy's identity-searching journey. They together add the true color to the seemingly benign global speech act, whose complete significance or implicature is to be fully understood when in the end Neddy "shouted, pounded on the door, tried to force it with his shoulder, and then, looking in at the windows, saw that the place was empty" (551). Otherwise stated, the beauty on this summer day, or in this apparently convivial reality is superficial, deceptive and ephemeral. Underneath, it is the ugliness, alienation, rejection that hold sway. But this kind of pragmatically rich conversation has been ignored. The critical attention has disproportionately been focused on the story's supposed symbolic, allegorical significance. As a consequence, his journey turns into a midsummer's nightmare, simply "because Cheever's references to midsummer seem insistent" (Bell: 433) and "he has been led to the vision that his dream of wealth, status, and happiness is transitory, illusory, and fraught with perils" (Bell: 436). Or, Neddy's journey is interpreted as his twisted vision of three holy sacraments: the Eucharist, baptism, and marriage, which subsequently leads to his spiritual emptiness and isolation at the end of the story (Blythe 1984: 383-394). Or, Neddy's experience and alien existence are compared to a modern Rip Van Winkle, and his original or absurd swim signifies the central hollowness of his middle years, "that American emptiness between Pepsi-Cola and Geritol" (Slabey 1982: 183). It is, of course, not the aim of my paper to critically review these interpretations. But what surprises me is that not one of these criticisms has closely examined the four major conversations I have been talking about. This kind of neglect, I believe, has sacrificed an important aspect of the story, an aspect that opens a new window onto the protagonist's development as a character and offers an additional dimension to the progression of the theme. While it may be appropriate to examine the story from a symbolic perspective as a scathing indictment of American

Fictional conversation

269

values, it is doing a disservice to the story and its overall significance by disregarding its conversations. As a matter of fact, it is Neddy's action, both his swim and talk with people, that gives him a sense of identity, a glimpse of hope. It, in this sense, no longer matters whether or not his whole quest should be taken purely as symbolic, or whether or not the outcome is futile. The pragmatic, macro-/micro-speech act structure of fictional conversation, strikingly enough, applies as well to Chinese short fictions. The Austinian, Gricean principles of pragmatics, exclusively born of AngloAmerican culture, apparently have some ready "spokesmen" in some Chinese fictional world. Ts'ung-wen Shen (1902 — ), a well-noted Chinese writer whose productive writing career spans mainly from the 20s to the 40s, is just an example. During his literary creative career, Mr. Shen, combining his distinctive regional color with his pessimistic, nostalgic vision, creates a host of characters, mainly countrymen that owe their prototypes to a southwestern, rural part of China. These characters are at once trying to live harmoniously with nature and being inevitably alienated and subsequently crushed by the onslaught of civilization and progress. His countrymen, in this regard, bear a striking resemblance to Camus' Stranger, although they sharply differ in face of death — Shen's countrymen accept death with peace and resignation while the Stranger challenges and scorns it (Nien 1972: 67). At the same time, Shen's characters, apparently uneducated and less sophisticated, are pragmatically conscious, and readily committed to the consequences of their speech acts. Their verbal interactions manifest features that are explainable in terms of Grice's Cooperative Principle and its associated maxims. They help to illustrate and illuminate some global speech acts, and as a result, foreground the themes of his short fictions. Shen, by way of creating these characters, is able to convey his own message or implicature, one that dramatizes his sense of the absurdity of the world he inhabits and his deep longing for a primitive, tranquil land that is slipping away from him irrevocably. In one of his well-known short fictions, San ge nanzi he yi ge nüren [Three men and a girl], the narrator, a squad leader — though not the writer — and the bugler, who becomes crippled shortly after their Company comes to a small town, start frequenting a beancurd shop to watch its owner, a reticent young man, perform his work. But they are more interested in the pretty young girl with big white dogs who lives right across the street. Meanwhile, the shop owner remains a somewhat elusive

270

LuMing R. Mao

figure, who is hospitable to them, and yet only gives his usual silence with a mysterious smile to either their queries or taunting remarks. One day after six months or so, the news comes to them that the girl has killed herself by swallowing gold pieces. Soon, the bugler disappears only to come back and report to the narrator that the girl's body has been excavated from her grave — which, they later learn, is discovered in a stove cave. They also learn that the shop owner has vanished as well. The conversations between the shop owner and the two soldiers, or any other customers, are scarce, since his normal response to either conversational overtures or ghastly killing scenes 13 consists only of silence accompanied with a sincere, bashful smile, which he acts out while giving an extra push to the grinding stone. Because of that, and of his mysterious departure after the death of the girl, his silence and smile have been symbolically interpreted as "both weapon and sanctuary for the young Countryman who can look at death without fear" (Nieh: 71). He becomes synonymous with or a personification of death, and his spiritual link to the girl attests to his being the natural man crippled and crushed by the onslaught of modern civilization. Like other critics, by focusing exclusively on the symbolic significance of the character, his silence and his smile, Nieh has underestimated an important aspect of the story: the interaction, verbal or silent — silence after all is an integral part of human intercourse — between the young countryman and the two soldiers. Their interaction, in particular, demonstrates that Gricean conversational maxims are in force, and that the micro- and macro-speech act approach does seem to present a new vantage point to look at fictional conversations — or fictional silence, for that matter. The story portrays the shopkeeper as a sturdy and reticent young man who comes from the country. He converses with customers only when it is absolutely necessary; otherwise, he always reacts with silence and a bashful smile. Be friendly or nonhostile, his silence in face of the two soldiers' questions or provocative remarks violates Grice's maxim of Quantity: be as informative as is required for the current talk exchange. For he has his opinions, when asked by the two soldiers, about whether or not the girl is pretty (Shen: 232), or about whether or not he is scared by the dead bodies on the execution field. The persistent violation here and elsewhere is deliberate, and thus gives rise to an important implicature: he at this moment does not want to reveal himself and his secret, perhaps unattainable, love for the girl because he still entertains the optimistic prospect of being together with her. It is thereby quite likely

Fictional conversation

271

that his silence signifies his hope, though very unrealistic, for the future. It is by no means an omen for his later disappearance, or his spiritual death as Nieh has suggested. On a macro level, his silence does anticipate a verbal outburst, which Nieh has completely neglected. For if my analysis so far is correct, namely, his silence is tantamount to a glimmer of hope, the destruction of hope is to call for some kind of verbal outbreak. This is what exactly transpires towards the end of the story. The shopkeeper's verbal interaction with the two soldiers becomes a form of total resignation and his way of renouncing the modern world. It epitomizes the thematic implicature of the story and adds a dramatic, tragic touch to the whole fictional world. This entire event takes place when the two soldiers, after learning of the tragic death of the girl, have gone to the shop: He seems to be a bit angry when he sees us as if to conceal his wound — but still with the same smile. His smile indicates he is still healthy and kind. "Why? Have a headache?" "Buried, buried, buried at dawn." "Buried at dawn?" "It was carried away before dawn." "What's the matter with you, and why are you so unhappy?" "I do not feel nothing." (the italics being mine) (240)' 4

The young shopkeeper is finally speaking out. By volunteering the information ("Buried, buried, buried at dawn.") that is not quite what is sought after by the narrator, he once again flouts the first maxim of Quantity. In so doing, he contrasts with his previous reticent self, signaling a change in his personality. If his silence amounts to an indication of his hope, his "breaking-off' can well be perceived as a sign of despair and a gesture of farewell to a world that is violently alien to him. On the other hand, his final answer to the narrator's question is an evasion, an attempt to deny his intense emotion. As a result, it violates the first maxim of Quality: be truthful. But, its implicature is obvious: his sense of hopelessness, and his determination to embrace the inevitable have come to the fore. This denial can further be seen as a macro-speech act that serves as a thematic statement illustrating not only the young shopkeeper's tragic, vacuous existence but also the writer's own vision of our modern world and his futile longing for an idyllic, primitive existence. In a way, this macro-speech act shapes the entire interaction between the young shopkeeper and the two soldiers. The latent sense of "nothingness" has initially kept the shopkeeper silent and yet secretly hopeful of himself — a reason for his sincere, bashful smile. But it does not make

272

LuMing R. Mao

his violation of the conversational maxim of Quantity less effective. As his existence in this world becomes increasingly more precarious, culminating in the death of the girl, the sense of numbness and despair sinks in, and results in his only short, distinct verbal interaction with the narrator. Ironically, this speaking out, an "improvement" upon his unusually reticent behavior but a violation nonetheless of the conversational maxims of both Quantity and Quality, constitutes his final pathetic imprint on this strange land, and pragmatically, it encloses a well-structured, hierarchial construct for the story. It should by now become clear that this theme of a fading existence either against a chaotic, alienating reality or amidst an intoxicating atmosphere of "hustle and bustle" strikes a common chord among all three short fictions. If the old waiter in A clean, well-lighted place towards the end of the story still tries to ward off the pervasive sense of nada by attributing it to the effect of insomnia, the young shopkeeper accepts the inevitability without much fuss, or Neddy Merrill meets the catastrophe with resignation. If Hemingway still wants to close his story with a trace of hope, Shen makes it abundantly clear that both the shopkeeper and himself "do not feel nothing". On the other hand, the fictional conversations in these three short fictions, be they the bulk of the story in the case of A clean, well-lighted place, or a fraction of it in the case of Three men and a girl and less so in The swimmer, have all kept the participants to the respective illocutionary commitments. They have also testified that both Gricean conversational maxims and micro- and macro-speech act approach are powerful tools in "making sense" of different aspects of the fictional world. In turn, the reader, sharing "the purpose of the talk exchange" with the writer, draws implicatures both from the fictional conversation and from the fictional text — a "display text" that itself reaps rich implicatures because of its "display-producing relevance" or "tellability" (Pratt 1977: 136 —148).15 For my part, the "implicature" I draw from this short excursion into the fictional world is no less significant, to wit, fictional conversation has its own pragmatic status and it plays a central role in our understanding of a particular character and its story. Perhaps more importantly, the basic pragmatic principles originated from Anglo-American philosophy and culture may have universal significance, as they seem to have applied to fictional conversations of their birthplace and of an "exotic" land alike.

Fictional conversation

273

Notes 1. I am grateful to Professor Michael Hancher for his instructive comments and for his consistent support in the entire course of this project. The discussions I have had with Professor Robert Brown, Jr., and with my colleague Joe Moses, a professional writer himself, have been very beneficial to my constructing and writing this paper. I am solely responsible for any errors or "pragmatic infelicities" in the paper. 2. Searle refers to normal illocutionary rules that establish connections between language and reality as "vertical rules" (1979b: 66). 3. So is the case with fictional characters. According to Searle, Murdoch, by pretending that there is such a character as Andrew Chase-White and by pretending to refer to him, creates a fictional character — although Searle does admit that a work of fiction may sometimes contain real references as well (1979b: 72). The problem of reference in fiction is just as philosophical as it is pragmatic, the discussion of which is obviously beyond the scope of this paper. For a start, refer to Adams' Pragmatics and fiction (1985: 2 - 1 2 ) . 4. Since the essay in question was written almost twenty years ago, I doubt if Professor Ohmann still holds the same position. But, I am not so much interested in what Professor Ohmann believes now as in the point of view espoused there, which still seems to have influence at present. 5. For a similar point of view, refer to Harshaw (1984: 227 — 251). 6. Hancher makes a similar point and observes that Searle is mistaken in thinking that the author of a novel always speaks "in his own, unmediated voice" (1977: 1093). 7. They are: Quantity: (1) Make your contribution as informative as is required for the current purposes of the exchange; (2) Do not make your contribution more informative than is required. Quality: Try to make your contribution one that is true, specifically: (1) Do not say what you believe to be false; (2) Do not say that for which you lack adequate evidence. Relation: Be relevant. Manner: Be perspicuous, specifically: (1) Avoid obscurity of expression; (2) Avoid ambiguity; (3) Be brief (avoid unnecessary prolixity); (4) Be orderly (Grice 1975: 46). 8. According to Levinson (1985: 104), implicatures that derive from observing Grice's maxims are "standard implicatures", which Grice takes for granted in formulating his theory of conversational implicature. 9. Here I am following the characterization of illocutionary acts proposed by Searle (1979b: 2 - 8 ) . 10. There has been some controversy concerning this part of their conversation in the story as well; in particular one supposed controversy revolves around the first two lines: "'The guard will pick him up,' one waiter said./'What does it matter if he gets what he's after?'" (471) Some critics have argued that the first line should be attributed to the old waiter. But as long as we agree that it is the old waiter that is in general sympathetic to people who want to stay late outside of their home to seek relief and respite from "nothingness", be it a clean, well-lighted cafe or being in company with a streetwalker, the first line cannot be ascribed ot the old waiter, but should be attributed to the young waiter. For more about the dispute, refer to Hurley (1976: 81 —85). 11. According to Scribner's, the discrepancy was caused when a slug of type was misplaced in the first printing of the story in Scribner's magazine in 1933, and since the original manuscript is no longer available, reprint plates had been made from that printing until 1965 when Scribner's made an "editorial" correction (Bennett 1970: 70 — 71).

274

LuMing R. Mao

12. Joe Moses alerted me to this part of their conversation during one of our many discussions. 13. In the middle of the story, the two soldiers take him to an execution field where the dead, bloody bodies of bandits are lying everywhere (232 — 233). 14. Here is its original in Pinyin (Chinese phonetic alphabet): Jian women lai le, ta you le yi diandian bu gaoxing, haoxiang shi zheyan ziji shanghen, rengran dui women weixiao zhe. ta de xiao, shuoming ta hai yiran you ge jiankang de shenti he shanliang de renge. "weishenme? tou teng ma?" "mai le mai le, yizao jiu mai le!" "zaoshang jiu mai le?" "dan hai hu da Hang jiu chu men le de." "ni you le xie shenme shiqing, zheyang bu kuaile?" "wo shenme ye bu." 15. According to Pratt, it basically refers to the property of representing the state of affairs to be "unusual, contrary to expectations or otherwise problematic" and creating "an imaginative and affective involvement" in the hearer.

References Adams, Jon.-K. 1985 Pragmatics and fiction. Amsterdam: Benjamins. Austin, J. L. 1975 How to do things with words. (2nd edition). J. O. Urmson — Maria Sbisa (eds). Cambridge: Harvard UP. Bell, Loren G. 1987 " 'The swimmer': a midsummer's nightmare", Studies in Short Fiction 24: 433-436. Bennett, Warren 1970 "Character, irony, and resolution in 'a clean, well-lighted place'", American Literature 42: 7 0 - 7 9 . Blythe, Hal 1984 "Perverted sacraments in John Cheever's 'the swimmer'", Studies in Short Fiction 21: 3 9 3 - 3 9 4 . Brown, Penelope — Stephen Levinson [1987] Politeness: some universals in language usage. Cambridge: Cambridge UP. Byrne, Michael D. 1986 "The river of names in 'the swimmer'", Studies in Short Fiction 23: 326-327. Cheever, John 1981 The swimmer, in: Eugene Current-Garcia — Walton R.Patrick (eds.), 541-551. Cole, Peter —Jerry Morgan (eds.) 1975 Speech acts. Syntax and semantics 3. New York: Academic Press. Collins, R. G. (ed.) 1982 Critical essays on John Cheever. Boston: G. K. Hall and Co.

Fictional conversation

275

Current-Garcia, Eugene — Walton R. Patrick (eds.) 1981 American short stories. (4th edition). Glenview, IL: Scott, van Dijk, Teun A. 1980 Macrostructures: an interdisciplinary study of global structures in discourse, interaction, and cognition. Hillsdale, NJ: Erlbaum. 1981 Studies in the pragmatics of discourse. The Hague: Mouton. van Dijk, Teun A. (ed.) 1985 Discourse analysis in society. Vol. 2 of Handbook of Discourse Analysis. 4 vols. London: Academic Press. Ferrara, Alessandro 1985 "Pragmatics", in: Teun A. van Dijk (ed.), Vol. 2, 1 3 7 - 1 5 8 . Grice, Paul 1975 "Logic and conversation", in: P. Cole — J. Morgan (eds.), 41 —58. Hagopian, John V. 1964 "Tidying up Hemingway's clean well-lighted place", Studies in Short Fiction 1: 1 4 0 - 1 4 6 . Hancher, Michael 1977 "Beyond a speech-act theory of literary discourse", MLN 92: 1081 - 1 0 9 8 . Harris, Wendell V. 1988 Interpretive acts: in search of meaning. Oxford: Clarendon Press. Harshaw (Hrushovske), B. 1984 "Fictionality and fields of reference", Poetics Today 5: 227 — 251. Hemingway, Ernest 1981 A clean, well-lighted place, in: Eugene Current-Garcia — Walton R. Patrick (eds.), 4 7 0 - 4 7 4 . Hurley, C. H. 1976 "The attribution of the waiter's second speech in Hemingway's 'a clean, well-lighted place'", Studies in Short Fiction 13: 81—85. Lanser, Susan S. 1981 The narrative act: point of view in prose fiction. Princeton: Princeton UP. Levinson, Stephen C. 1983 Pragmatics. Cambridge: Cambridge UP. Ling, Yu (ed.) 1982 Shen Ts'ung-wen duanpian xiaoshuo ji [The complete short stories of Shen Ts'ung-wen]. Vol 1 of 4 vols. Beijing, China: The People's Publishing House of Literature. Nieh, Hua-Ling 1972 Shen Ts'ung-wen. New York: Twayne Publishers, Inc. Ohmann, Richard 1971 "Speech acts and the definition of literature", Philosophy and Rhetoric 4: 1-19. Pratt, Mary Louise 1977 Toward a speech act theory of literary discourse. Bloomington, IN: Indiana UP. Searle, John 1979a Expression and meaning: studies in the theory of speech acts. Cambridge: Cambridge UP. 1979b "A taxonomy of illocutionary acts", in: Searle (1979a), 1 - 2 9 . 1979c "The logical status of fictional discourse", in: Searle (1979a), 58 — 75.

276

LuMing R. Mao

Shen, Ts'ung-wen 1982 San ge nanzi he yi ge nüren [Three men and a girl], in: Yu Ling (ed.), 223-245. Slabey, Robert M. 1982 "John Cheever: 'The swimmer' of America", in: R . G . C o l l i n s (ed.), 180-191.

Presupposition and pragmatic inference in a literary text Helen Aristar-Dry

1. Introduction The following paper is an analysis of a single work of literature, Henry James's The Turn of the Screw. It is offered here as an example of the way that infringement of pragmatic norms can produce equivocal discourse effects in written texts. The Turn of the Screw may function as an appropriate example, since the equivocal nature of the narrative is not in doubt. The book has been the focus of controversy for over seventy years. And, as we might expect from an issue in literary criticism, the debate has been very well-documented. The following paper makes no attempt to resolve the controversy but only to suggest that pragmatic features of the text have helped to fuel it. Specifically it suggests that James's narrator frequently infringes discourse norms in the use of presuppositional constructions and that, as a result, the narrative loses some of its transparency as an account of events and becomes the focus of metalinguistic inquiry. 1 As most will recall, The Turn of the Screw is the story of a young governess striving to protect her charges from the ghosts of two corrupt former servants. The governess, an unnamed young woman whose manuscript is the ultimate source of the tale, is employed by a handsome man-about-town to superintend his young wards, Miles and Flora, at a country estate where her only adult companion is the uneducated housekeeper, Mrs. Grose. There she begins to glimpse two mysterious figures who, she comes to believe, are the ghosts of Peter Quint, the former valet, and Miss Jessel, the children's former governess. These two servants — now both dead — had been conducting an illicit affair while they had charge of the children. Now, their ghosts have apparently returned to complete the corruption of the children and possess their souls at last. The tale chronicles the governess's attempts to protect the children: a protection which ends with one child dead and the other sick and terrified, not of the ghosts, but of the governess. Published in 1898, The Turn of the Screw was apparently read as a "straight" fireside ghost story for the first twenty years after its appear-

278

Helen

Aristar-Dry

ance. But, in the twentieth century, the narrative has sparked a heated controversy, centering on the existence of the ghosts. As early as 1949, Evans could remark that "there are now almost as many interpretations of the story as there have been critics willing to venture them" (200). And the fancifulness of certain interpretations has prompted BrookeRose to remark that "the [hallucinatory] state of the governess must be contagious." (1976a: 268) However, within this flurry of critical discussion, three major positions may be discerned. The first is that, within the world of the novel, the ghosts are "real" apparitions and the governess is a reliable narrator. Taking this position are Reed (1949); Evans (1949); Vaid (1964); Allen (1979); and Sheppard (1974). The second position is that the ghosts are imagined and the governess is an unreliable narrator: a spinster, secretly in love with her employer, who has hallucinated the ghosts as projections of neurotic desire. Although the best known proponent of this position is Wilson (1934), it was previously voiced by Harold Goddard in 1920 and Ezra Pound in 1918. Later proponents include Cargill (1956); and Cranfill and Clark (1965); but variants of the position abound. And the third position is espoused by those who believe that the existential status of the ghosts is probably undecidable but that the real significance of the apparitions lies elsewhere. Early on, for example, Heilman (1948) and Hoffman (1953) asserted that the ghosts are best interpreted as symbolic representations of evil, whose character as "real" or "imagined" visitations is irrelevant to a proper reading of the text. An interesting post-modern variant of this position is that the ambiguity of the ghosts derives from the nature of human sign systems. Thus Cook and Corrigan remark that "the key to The Turn of the Screw is found not in one or the other of two possible meanings but in narrative ambiguity itself' (1980: 560), and Mansell asserts that the ghosts are "scared up into being by language" (1985: 53). 2 Assuming, like Feldman (1982) and Faulkner (1983), ambiguity rooted in the self-referentiality of the text, Mansell notes that the governess seems to be writing "not about her subject but to be writing the subject itself' (1985: 57). Citing Wittgenstein and Iser, 3 he attributes this partly to the ability of words to have meaning without extension, to "bring [referents] into existence" because of language's "inducement to realize the very thing meant not to be" (1985: 52). Mansell's formulation may prompt us to seek an even more precise description of the language which has "scared ghosts into being." For,

Presupposition

and inference

279

although all instances of denotative language may have the power to evoke referents, not all have the power to conjure "ghosts", i. e., discourse constructs that seem to flicker, not just before the eyes of the characters, but also before the eyes of the reader. That these ghosts are so ephemeral may result, in part, from the textual exploitation (whether designed or undesigned) of conventions of pragmatic inference. Specifically, the equivocal discourse status of the ghosts may be linked to their presentation in presuppositional constructions. By means of these constructions, information about the ghosts is consistently presented as though already assumed by both speaker and hearer. Thus, not only is important information offered in a way that bypasses normal questions of evidentiality, but also the eccentricity of the discourse raises metalinguistic questions about the competence of the narrator. In this way, a pragmatic device usually employed to structure mutual assumption comes to have the opposite effect: that is, it generates scepticism, rather than shared belief.

2. Presuppositional constructions Since 1892, when Frege first used the term to designate the assumptions surrounding proper names, there have been various definitions of "presupposition" current in philosophy and linguistics. For the present purpose, however, presuppositions may be identified as the pragmatic inferences triggered by a fairly well-known set of "presuppositional constructions" and isolated through linguistic tests discriminating between what is asserted and what is presupposed. Thus (la) below, "All of John's children are bald" presupposes (lb), "John has children." And this inference can be isolated by negating the verb of the main clause, as in (lc); questioning the assertion, as in (Id), or inserting it into an if ... then clause, as in (le). (1)

a. b. c. d. e.

All of John's children are bald. John has children. All of John's children are not bald. Are all of John's children bald? If all of John's children are bald, then John's wife must have bad genes.

These three tests explicitly call into question the assertion that John's children are bald, but the presupposition that John has children still holds. Although the three tests do not always pick out the same set of

280

Helen

Aristar-Dry

presuppositional triggers, certain constructions may be considered paradigm instances. Those relevant to The Turn of the Screw are reviewed in (2) below. (2)

a. definite descriptions: John saw/didn't see the man with two heads. » There exists a man with two heads. [ » = "presupposes"] b. factive verbs: Martha regrets/doesn't regret John's marrying Lisa. » John married Lisa. c. iteratives: The flying saucer came/didn't come again. » The flying saucer came once. d. temporal clauses: Before Strawson was born, Frege noticed/ didn't notice presuppositions. » Strawson was born. e. clefts: It wasI wasn't Harry that kissed Rosa. » Someone kissed Rosa. f. pseudo-clefts: What John lost/didn't lose was his wallet. » John lost something. g. comparative constructions: Jimmy is/isn't as stupid as Billy. » Billy is stupid. h. non-restrictive relatives: The Jets, who beat the Giants last year, are/are not conventional ball-players. » The Jets beat the Giants last year, (following Levinson, 1983: 181 — 84)

Presuppositional constructions introduce information into a discourse as assumed, not asserted. Thus they background information in the sense of Eco and Violi (1987: 6), who state that presupposed information establishes a background frame "which determines the point of view from which the discourse will be developed." It constitutes information which "both [Speaker and Addressee] should take for granted, while the asserted meaning constitutes the foreground information. 4 Presupposed information, then, constitutes part of the information given by a text, but only "as part of a mutual agreement between speaker and hearer." Presupposed information has been described as "shielded from challenge" in conversation because hearers cannot query it directly without conspicuous social and contextual consequences (Givon 1982). But, correspondingly, speakers cannot encode just any information in presuppositional constructions without, eventually, risking some discourse credibility. Although both given and new information may be communicated through presuppositional constructions, only certain types of new information are common, or unmarked, in presuppositional constructions.

Presupposition

and inference

281

Following Prince (1981), we may characterize these types as "inferrable" and "unused," as opposed to "brand-new". "Inferrable" information is that inferrable from old information or from the situation, whereas "unused" information is information that may not form part of the hearer's active, conscious knowledge but is consistent with his or her model of the world. So, for example, it is acceptable to say, "I'm sorry I'm late; my car broke down" even when the hearer has no definite knowledge that the speaker has a car (Levinson, 1983: 205), presumably because the possibility of car ownership is likely to be "in" most hearers' models of the world. But "brand-new" information is a less likely candidate for encoding in presuppositional constructions. And this may account for the well-known observation (cf. Prince 1978: 370; Levinson 1983: 205) that it is inappropriate to say, "I'm sorry I'm late. My fire-engine broke down," unless the hearer already knows that the speaker has a fire-engine. Given the "shielding" of background information mentioned above, an occasional presupposing of "brand-new" information is not likely to precipitate a breakdown of communication or even a challenge of the presupposition. But it will prompt a request for more information. Someone who hears "My fire-engine broke down" is not likely to challenge the content ("You don't have a fire-engine") or the presuppositional act ("What an odd way to put it!"). But he is likely to respond, "I didn't know you had a fire-engine. (Tell me more.)" That is, he will co-operatively assume the existence of the fire-engine but expect shortly to learn more about it. If a speaker were to thwart that expectation, or to persist in presupposing highly salient "brand-new" information, he or she might attract the kind of metalinguistic challenges mentioned above, as well as others ("Why should I know anything about your fire-engine?" or "Why do you keep making these strange assumptions?") which normal discourse co-operation renders infrequent. As Eco and Violi point out (1987: 13), such challenges always result in a change of topic. Thus they not only call into question the manner of the discourse, but they also compromise the content by distracting attention away from it. Conversational norms for presupposition are worth reviewing because they may offer an analogue for the reader's response to presupposition in written texts. In narrative texts a certain amount of "brand-new" information 5 is regularly introduced through presuppositional constructions — particularly in initial chapters, when presuppositions may be employed to arouse expectations and generate reader interest. 6 Existential

282

Helen

Aristar-Dry

presuppositions of elements of setting, moreover, are frequent throughout; and the reader simply interprets these as an instruction to instantiate the referents into his or her model of the scene. 7 However, in a narrative as in a conversation, it is unusual to find the pervasive presupposition of "brand-new" information. And it is unusual to find such information unreinforced by embellishing assertions. Faced with an occasional presupposition of "brand-new" information, the reader, like the auditor of the "fire-engine" sentence, will co-operatively assume the information but expect shortly to learn more about it. And a narrator who thwarts this expectation, or who presupposes "brandnew" information beyond some threshhold of tolerance, draws attention to the manner of his discourse in such a way as to court metalinguistic challenge.

3. Presupposition in The Turn of the Screw In The Turn of the Screw, presuppositional constructions are unusual in two ways: first, in the type of information backgrounded and, secondly, in the fact that this information is not unequivocally reinforced elsewhere in the text. Presuppositional constructions background almost all of the "brandnew" information about the ghosts — information vital to the external events of the plot. At the same time, the foreground of the narrative, as evidenced in the sentence assertions, is given over to the governess's inner life. This reversal of foreground and background may augment reader uncertainty about the governess's reliability by focussing attention on her perceptual processes, rather than the thing perceived. At the least, however, it marks her discourse style as eccentric. For not only is the reader asked to presuppose an unusual amount of "brand-new" information, but that information is not reinforced in the usual way. Since presuppositional constructions encode most of the information about the ghosts, the expected supporting assertions are not forthcoming. And, at the same time, confidence in the presupposed information is undermined by other indeterminacies in the text. These indeterminacies have been much analyzed in the critical literature. Here it may be sufficient to recall that the housekeeper, Mrs. Grose, never sees the ghosts, that Miles's delinquencies at school are left obscure, 8 and that every aspect of the childrens' behavior is open to more than one interpretation. 9 Furthermore, the initial impression of the governess's

Presupposition

and inference

283

stability is subtly undercut by her words and behavior: she has "family problems" and a romantic attachment to a relative stranger, her tone is often fevered, and she herself notes that her sanity might have been questioned. 10 In this context, in which both the governess's character and the nature of events are underdetermined, the marked reliance on presupposition serves to heighten the indeterminacy.

3.1. Presupposed ghosts Characteristic is the first ghost scene, given as (3) below, where the information that the governess is arrested by something startling is tucked into presuppositional constructions, while her thoughts are the topic of the sentence assertions. The first part of the passage is quoted below with "brand-new" presupposed information italicized. The quotation has been severely edited to focus on the governess and the ghost, but the omitted scenic description is relevant only insofar as it would exemplify a more normal distribution of assertion and presupposition. (3)

... it would be as charming as a charming story suddenly to meet someone . . . . That was exactly present to me ... when ... I stopped short on emerging from one of the plantations and coming into view of the house. What arrested me ... was the sense that my imagination had, in a flash, turned real. He did stand there! — but high up ... at the very top of the tower ... yet it was not at such an elevation that the figure I had so often invoked seemed most in place. It produced in me, ... I remember, two distinct gasps of emotion, which were the shock of my first and that of my second surprise. My second was a violent perception of the mistake of my first: the man who met my eyes was not the person I had ... supposed .... the figure that faced me was ...as little anyone else I knew as it was the image that had been in my mind ... . the man who looked at me over the battlements was as definite as a picture in a frame. That's how I thought, with extraordinary quickness, of each person that he might have been and that he was not. ( 3 1 0 - 3 1 1 )

To highlight the reversal of foreground and background exemplified in this passage, (4) below extracts from the Jamesian web of embedded clauses, the "brand-new" information communicated through presup-

284

Helen

Aristar-Dry

positional constructions. In (4), the presuppositional construction is noted in brackets: (4)

a. b. c. d. e. f. g. h. i. j. k.

a story is charming [comparative clause]; the governess stopped short [temporal clause]; she was arrested by something [pseudo-cleft]; she had a sense that her imagination had turned real [definite description]; the figure would seem in place elsewhere [cleft sentence]; the governess had two surprises [definite description]; she perceived that her initial recognition was a mistake [definite description]; the figure was facing her [definite description]; a picture in a frame is definite [comparative clause]; she thought of everyone the figure might have been [implicit cleft, 11 definite description]; the figure was not anyone she thought of [definite description].

Read in sequence, these observations constitute a precis of the scene that includes all its external action, including the vital point that the governess's initial identification of the figure is revealed to be a mistake. As (5), the continuation of the passage, demonstrates, presuppositional constructions also encode almost every detail we learn about the physical appearance and movements of the ghost. (5)

... The great question ... is ... with regard to certain matters, the question of how long they have lasted. Well, this matter of mine ... lasted while I caught at a dozen possibilities ... in there having been in the house ...a person of whom I was in ignorance. It lasted while I just bridled a little with the sense that my office demanded that there should be no such ignorance and no such person. It lasted while this visitant, at all events — and there was a touch of the strange freedom, as I remember, in the sign of familiarity of his wearing no hat — seemed to fix me ... with just the question, just the scrutiny through the fading light, that his own presence provoked. (311—312)

As in the first part of the scene, presuppositional constructions background "brand-new" information about the external action: e.g., that the governess caught at a dozen possibilities [temporal clause] and that she bridled a little [temporal clause]. In addition, they background "brand-new" information about the ghost:

Presupposition

and inference

285

a. he fixed her with the same scrutiny he provoked [temporal clause, definite description]; b. he displayed a strange freedom [definite description]; c. he was not wearing a hat [definite description]; If we add these details to the disclosures about the ghost backgrounded before (i. e., that he is facing her across the battlements and is unidentifiable), we see that the narrator asks the reader initially to assume not only the ghost's appearance and stance, but also her own external response to the apparition. Indeed, presuppositional constructions background much of what is new, and almost all of what is important, in the governess's first encounter with the ghost. By way of contrast, (6) extracts the sentence assertions, as reported in the non-embedded clauses of the passage. Listed are those which offer "brand-new" information: (6)

a. b. c. d. e. f. g. h. i. j. k. 1.

"it would be ... charming to meet someone"; "[the master's face] was present to me"; "What arrested me ... was the sense"; "it was not at such an elevation that the figure ... seemed most in place"; "It produced in me ... two distinct gasps of emotion"; "My second was a violent perception"; "the man ... was not the person"; "the figure ... was ... [not] anyone else I knew; "the man ... was ... definite"; "That's how I thought"; "The great question ... is ... the question of how long [such matters] have lasted"; "this matter of mine ... lasted".

Many of the governess's main clauses function only as syntactic matrices, encoding little information in themselves. Nevertheless, their tendency to report inner rather than outer experience is observable in the preceding list, where six of the twelve assertions concern the governess's thoughts or emotions. Thus, although the pseudo-cleft ("what arrested me was ...") presupposes that something arrested the governess, that "something" is asserted to be a "sense" that her "imagination" has turned real. Here, as elsewhere in the governess's assertions, contextually ambiguous diction contributes, like the backgrounding of vital information, to

286

Helen

Aristar-Dry

the half-deceptive quality of her discourse. 12 Although the phrase "the sense that my imagination had turned real" refers in this context to the apparition, the lexical items actually denote mental states. The description of her second surprise as a "violent perception" also exemplifies this tendency. And the assertion made by "It produced in me ... two distinct gasps of emotion" continues the focus on her consciousness. This focus is maintained throughout the ghost scenes. When the supposed ghost of Miss Jessel appears on the lake shore, for example, there is a similar contrast between what is asserted and what is presupposed. In passage (7), as in those that follow, italics indicate "brand-new" presupposed material, while holding and capitalization mark those clauses making assertions about the governess's inner life: 13 (7)

I BECAME AWARE that on the other side of the Sea of Azof, we had an interested spectator. T H E WAY this knowledge gathered in me WAS THE STRANGEST T H I N G IN T H E W O R L D . ... I BEGAN TO TAKE IN WITH CERTITUDE, and yet without direct vision, the presence, at a distance, of a third person. ... T H E R E WAS N O A M B I G U I T Y IN ANYT H I N G ; N O N E WHATEVER, AT LEAST, IN T H E CONVICTION I from one moment to another found myself forming as to what I should see ... as a consequence of raising my eyes . . . . I RECOLLECT ... R E M I N D I N G MYSELF that nothing was more natural, for instance, than the appearance of one of the men . . . . THAT R E M I N D E R H A D AS LITTLE E F F E C T U P O N MY PRACTICAL CERTITUDE as I was conscious still without looking — of its having upon the character and attitude of our visitor." (327)

In reading this passage, many readers have been struck by the extent of the governess's certainties, given that she never looks up. Only paragraphs later do we read, "Then I ... shifted my eyes — I faced what I had to face" [italics added]. At this point the chapter ends; what she confronts is not described. However, it is presupposed: "what I had to face" carries the inference that there was something to be faced. Thus the ending of the scene parallels the beginning, in which the ghost is introduced with a factive verb ("I became aware that ...") presupposing the truth of the complement clause, i.e., that the governess and Flora have "an interested spectator". This is "brand-new" information of a critical sort, as is the information that the ghost remained unchanging despite the governess's attempt at self-reminder.

Presupposition

and inference

287

At the same time as information about the visitor is presupposed, lexical choices in the assertions emphasize the governess's thoughts. Thus we find that what there is "no ambiguity in" is actually a "conviction" and that the reminder has as little effect on the governess's "certitude" as she was "conscious" of its having on the visitor. The foregrounding of the governess's consciousness, and the backgrounding of information about the ghosts, continues when the governess meets Quint on the stair. The encounter is described in passage (8) below: (8)

... I PRECIPITATELY F O U N D MYSELF AWARE O F T H R E E T H I N G S . . . . My candle ... went out, A N D I PERCEIVED ... that the yielding dusk of earliest morning rendered it unnecessary. Without it ... I SAW that there was someone on the stair. I SPEAK O F SEQUENCES, BUT I R E Q U I R E D N O LAPSE O F SECONDS TO STIFFEN MYSELF for a third encounter with Quint. The apparition had reached the landing halfway up . . . . He knew me as well as I knew him. He was absolutely ... a living, detestable, dangerous presence. BUT THAT WAS NOT T H E W O N D E R O F WONDERS; I RESERVE THIS DISTINCTION FOR ... the cicumstance that dread had unmistakably quitted me and that there was nothing in me there that didn't meet and measure him. I H A D PLENTY O F A N G U I S H A F T E R THAT EXT R A O R D I N A R Y MOMENT, BUT I HAD, T H A N K GOD, NO T E R R O R . And he knew I had not - I F O U N D MYSELF ... M A G N I F I C E N T L Y AWARE O F THIS. I FELT ... that if I stood my ground ... I should cease ... to have him to reckon with; and during the minute ... the thing was as human and hideous as a real interview. ... It was the dead silence of our long gaze at such close quarters that gave the whole horror, huge as it was, its only note of the unnatural. I CAN'T EXPRESS what followed it save by saying that the silence itself ... became the element into which I saw the figure disappear. (341 — 342)

As the italics above indicate, presuppositional constructions encode a substantial amount of "brand-new" information about the external events of the scene: (9)

a. The dusk rendered [the candle] unnecessary [factive verb]; b. There was someone on the stair [factive verb]; 14 c. An apparition existed [definite description];

288

Helen

d. e. f. g. h. i. j. k.

Aristar-Dry

She knew him well [comparative clause]; Dread quitted her [definite description]; All her faculties met and measured him [definite description]; They gazed a long time at close quarters in silence [definite description]; The horror was huge [definite description]; Something gave it a note of the unnatural [cleft sentence]; Something followed [definite description]; The figure disappeared [definite description];

Once again, a list of the presupposed information constitutes almost a complete summary of the scene. And, once again, many of the sentence assertions emphasize the governess's mental processes. She "finds herself aware of three things"; "perceives"; "sees"; "feels"; distinguishes "the wonder of wonders"; "has plenty of anguish"; but "has no terror"; "finds herself aware of this"; and "can't express" what she sees except by reporting another impression: that "the silence ... became the element" into which the ghost disappears. Only five of the eighteen assertions in the passage purport to inform us about the ghost. And the assertion that he knew she had no terror is later revealed to be something that she "found herself magnificently aware of ", i.e., it is subsequently presented as the implied complement of a factive verb and hence as presupposed. Although it is important that the pattern of assertion and presupposition continues throughout all the ghost scenes in the text, it is, of course, impossible to discuss them all here. For reference, the other appearances of the ghosts are given in Appendix A and marked, with italics and bolded caps, to discriminate presupposition from assertion. Certainly not all the assertions in The Turn of the Screw focus on the governess's consciousness; nor is all she tells us about the ghosts presupposed. Nevertheless, the ghost scenes reveal a distinct tendency to posit as background most of the information that would propel the plot — if the "plot" of this psychological thriller were, in fact, taken to inhere in the external events of the story.

3.2. 'Real-world' assertions Given the way that new information about the ghosts tends to be backgrounded, it might be suggested that this process merely reflects the singular Jamesian style, i.e., that the backgrounding of vital "brand-

Presupposition

and inference

289

new" information is ubiquitous throughout the text and does not, therefore, contribute significantly to the special ambiguity that surrounds the ghosts. Certainly, mental processes tend to be foregrounded throughout any Jamesian narration. But in The Turn of the Screw presuppositional constructions encoding "brand-new" information about external events are relatively infrequent before the governess falls prey to her delusions. Thus, the narrative of her arrival at the estate, passage (10) below, introduces Mrs. Grose and the children, as we would expect, primarily through assertion. (10)

In this state of mind I spent the long hours of swinging, bumping coach that carried me to the stopping place at which I was to be met by a vehicle from the house. This convenience, I was told, had been ordered, and I found ... a commodious fly in waiting for me. Driving at that hour, on a lovely day ... my fortitude mounted afresh and, as we turned into the avenue, encountered a reprieve . . . . There immediately appeared at the door, with a little girl in her hand, a civil person who dropped me as decent a curtsy as if I had been the mistress or a distinguished visitor . . . . The little girl ... appeared to me on the spot a creature so charming as to make it a great fortune to have to do with her. She was the most beautiful child I had ever seen, and I afterward wondered that my employer had not told me more of her. I slept little that night — I was too much excited; and this astonished me, too, I recollect ... adding to my sense of the liberality with which I was treated. The large, impressive room ... the great state bed, ... the full, figured draperies, the long glasses ... all struck me — like the extraordinary charm of my small charge — as so many things thrown in. It was thrown in as well, from the first moment, that I should get on with Mrs. Grose in a relation over which, on my way, in the coach, I fear I had rather brooded. The only thing indeed that in this early outlook might have made me shrink again was the clear circumstance of her being so glad to see me. I PERCEIVED within half an hour that she was so glad ... as to be positively on her guard against showing it too much. (298-99)

In this passage, almost all of the "brand-new" information about the governess's journey and arrival is asserted. And, in contrast to the way

290

Helen

Aristar-Dry

that physical descriptions of the ghosts are presupposed, it is also asserted that the little girl is charming and is "the most beautiful child" the governess has ever seen. Although definite descriptions (e.g., "The large, impressive room ... the great state bed, ... the full, figured draperies") encode "brand-new" descriptive detail, this is an unremarkable use of existential presupposition to posit information about setting. Only at the end, when something ominous at Bly is hinted at, does the governess slip into the discourse eccentricity of presupposing "brand-new" material salient enough to be startling; i.e., (1) that something might have made the governess shrink [definite description]; (2) that Mrs. Grose's gladness was a "clear circumstance" [definite description]; and (3) that Mrs. Grose was so glad as to be on her guard against showing it [factive verb: perceive]. These assumptions set the scene for what many would consider the governess's plunge into fantasy.

4. Conclusion Examination of the ghost scenes suggests that presuppositional constructions contribute in two ways to the equivocal discourse status of the ghosts. First, by positing the ghosts as assumed, not asserted, they establish the apparitions "in the background" — in a discourse arena which is "shielded from challenge" but which is also highly dependent for its definition on co-operation between speaker and hearer. At the same time; they compromise that co-operation, because eccentricities in their use help to call into question the discourse competence of the narrator. In The Turn of the Screw the narrator asks the reader to presuppose information that is not only "brand-new" and pervasive, but also unreinforced in the succeeding text. This has effects on the reading experience similar to those which would obtain in a conversation where pragmatic norms were violated. For some readers at least, a discourse "co-operation" that was automatic and unconscious becomes the focus of attention; and assumptions that were "shielded from challenge" attract challenge, even at the expense of a "change of topic." For these readers, the topic becomes, not the ghosts, but the governess. The ghosts cease to function as actants and become triggers for metalinguistic inquiry. That is, they are unveiled as

Presupposition and inference

291

discourse constructs whose credibility cannot be assessed apart from that of the speaker. Thus The Turn of the Screw may be said to exemplify one possible effect of the infringement of discourse norms in the realm of pragmatic inference. Such infringements will always invite metalinguistic inquiry, although the meanings finally attached to them may be dependent on other features of the text. Here, where neither the nature of the ghosts nor the character of the governess is fully determined, the unusual reliance on presupposition serves to heighten the indeterminacy — until the ghosts come to seem equivocal parts of the narrative, only half "there" in the text.

Appendix A italics bolded caps

= "brand-new" presupposed material = asserted material concerning the governess's inner life

(1) ... [THE A F T E R N O O N LIGHT] ENABLED ME ... TO BECOME AWARE O F A PERSON on the other side of the window and looking straight in ... . MY VISION WAS INSTANTANEOUS; ... HE A P P E A R E D T H U S AGAIN WITH ... A NEARNESS THAT ... M A D E ME, as I met him, CATCH MY BREATH A N D T U R N COLD. He was the same ... and seen, this time, as he had been seen before, from the waist up ... . His face was close to the glass, yet THE E F F E C T O F this better view WAS ... ONLY TO SHOW ME HOW INTENSE THE F O R M E R H A D BEEN His stare was as deep and hard as then, but IT Q U I T T E D ME FOR A M O M E N T D U R I N G WHICH I C O U L D STILL WATCH IT, SEE IT fix successively several other things. On the spot T H E R E CAME TO ME THE A D D E D SHOCK O F A CERTITUDE that it was not for me he had come there. (315 — 316) (2) I ... R E C O G N I Z E D the presence of a woman seated on one of the lower steps with her back presented to me, her body half-bowed and her head, in an attitude of woe, in her hands. I had been there but an instant ... when she vanished without looking around at me. I KNEW, nonetheless, what dreadful face she had to show." (345) (3) In the presence of what I saw I REELED STRAIGHT BACK UPON MY RESISTANCE. Seated at my own table in clear noonday light I SAW A PERSON whom, without my previous experience, I should have taken ... for some housemaid . . . . There was an effort in the way that, while her arms rested on the table, her hands with evident weariness supported her head; but ... at the moment I took this in I HAD ALREADY BECOME AWARE that, in spite of my entrance, her attitude strangely persisted. Then it was ... that her identity flared up in a change of posture. She rose ... and ... stood there as my vile predecessor. Dishonored and tragic, she was all before me; but even as I fixed and, for memory, secured it, the awful image passed away. ... I HAD THE EXTRAORDINARY CHILL O F F E E L I N G that it was I who was the intruder. (365)

292

Helen

Aristar-Dry

Notes 1. I would like to thank my former student, Susan Kucinkas, for her significant contributions to this paper and to note that an expanded, joint-authored version is currently in progress. 2. A few linguists have also taken this position, e.g., Eaton (1983), who points out the "layered ambiguity" of the lexical choices and speech acts in the text, and Weber (1982) who analyses frame construction. Weber also briefly notes the role of presupposition in constructing implicit ideological frames. 3. Mansell cites Wittgenstein's observation [in Tractus Logico-Philosophicus] that "the propositions ρ and ~ p, while they have opposite senses, have corresponding to the propositions one and the same state of affairs" and Iser's assertion [in The Implied Reader] that "language ... can never be about nothing" (Mansell 1985: 52). 4 Foreground here is not intended to refer to the discrimination of events from commentary and setting, a phenomenon that is usually dependent on tense and aspect (Dry 1983). 5. A number of linguists have shown how the use of presuppositional constructions for non-controversial new information helps to establish a community of belief between reader and writer (see, for example, Fillmore 1974; and Prince 1973). 6. Thus the first nine words of The Turn of the Screw, for example ("The story had held us, round the fire, sufficiently breathless. ..."), presuppose a story, a fire, and a set of hearers that the reader has yet to encounter. 7. Much of the information posited about setting through existential presuppositions (e. g., " The tall trees tossed in the wind") might also legitimately be considered "unused," or "inferrable," rather than "brand-new." 8. Even after he confesses to having "said things", it is unclear whether these (presumed) obscenities might have been learned from the living Quint. It will be remembered that Mrs. Grose, speaking of Flora's "appalling language", confesses that she has "heard some of it before" (388). 9 See Brooke-Rose (1976a and b; 1977) for a discussion of the way that two interpretations of the events are sustained throughout by the structural features of the text. 10. See, for example, the following: She herself had seen nothing, not the shadow of a shadow, and nobody in the house but the governess was in the governess's plight; yet she accepted without directly impugning my sanity the truth as I gave it to her. [italics added] (The Turn of the Screw, p. 322) Here the governess asserts her "truth" in a typically ambiguous way, giving us at once the information that the housekeeper saw nothing and the hint that her own sanity might have been — but was not — impugned. 11. I have called "That's how I thought of ..." an "implicit cleft" because it is equivalent in meaning to "It was thus that I thought of ...". What is important here, however, is that it meets the tests for a presuppositional construction. Note that (a), (b), and (c) below all presuppose (d): (a) That's not how she thought of everyone he might be. (b) Is that how she thought of everyone he might be? (c) If that's how she thought of everyone he might be, then we probably shouldn't believe her. (d) She thought of everyone he might be.

Presupposition and inference

293

(The sentence has been altered to third-person above to circumvent the oddity — irrelevant to presupposition — of a speaker's questioning her own thought process.) 12. Eaton (1983) makes a related point when she attributes the elusiveness of the text to a "layering" of speech act ambiguity with ambiguity of word-meaning. However, she does not discuss the particular kind of lexical ambiguity that characterizes references to the ghosts. 13. Certain constructions, most notably definite descriptions, participate in both; but, as my intent is to describe a tendency in the governess's narrative, rather than an absolute distinction between presupposed and asserted material, I have not attempted doublemarking and, in fact, have marked only assertions and presuppositions relevant to the discussion. 14. See ... that is usually considered factive, although see NP is not. Cf. "John didn't see that there was someone on the stair" presupposes "There was someone on the stair". The apparent failure of the negation test with first-person sentences (?"I didn't see that there was someone on the stair") may arise from the pragmatic impossibility of claiming not to know something presented as a fact, rather than from the non-factivity of see.

References Allen, John 1979

"The governess and the ghosts in The Turn of the Screw", The HenryJames Review 1.1: 73 — 80. Brooke-Rose, Christine 1976a "The squirm of the true: an essay in non-methodology", PTL: A Journal for Descriptive Poetics and Theory of Literature 1: 265 — 294. 1976b "The squirm of the true II: a structural analysis of Henry James' The Turn of the Screw", PTL: A Journal for Descriptive Poetics and Theory of Literature 1: 513 — 546. 1977 "Surface structure in narrative: the squirm of the true, part III", PTI: A Journal for Descriptive Poetics and Theory of Literature 3: 517 — 562. Cargill, Oscar 1956 "Henry James as Freudian pioneer", Chicago Review X: 13 — 29. Cook, David — Timothy Corrigan 1980 "Narrative Structure in The Turn of the Screw",Studies in Short Fiction 17: 5 5 - 6 5 . Cranfill, Thomas M. — Robert L. Clark, Jr. 1965 An Anatomy of The Turn of the Screw. Austin, Texas: Gordian Press, Inc. Eaton, Marcia 1983 "James's Turn of the Speech Act", British Journal of Aesthetics 23/24: 333-345. Eco, Umberto — Patrizia Violi 1987 "Instructional semantics for presuppositions", Semiotica 64—1/2, 1 —39. Evans, Oliver 1949 "James' Air of Evil: The Turn of the Screw," Partisan Review 16: 175 — 187. [Reprinted in Willen (1960): 200-212.]

294

Helen

Aristar-Dry

Faulkner, Howard 1983 "Text as Pretext in The Turn of the Screw", Studies in Short Fiction 20. 2 - 3 : 87-94. Felman, Shoshana 1982 "Turning the screw of interpretation", in: S. Felman, (ed.), Literature and Psychoanalysis. Baltimore, Md.: Johns Hopkins University Press, 94—200 Givon, Talmy 1982 "Logic vs. Pragmatics, with Human Language as the Referee: Toward an Empirically Viable Epistemology", Journal of Pragmatics 6.2: 81 — 133. Heilman, Robert 1948 " The Turn of the Screw as Poem", The University of Kansas City Review XIV: 2 2 7 - 2 8 9 . [Reprinted in Willen (1960): 174-189.] Hoffman, Charles 1953 "Innocence and evil in James' The Turn of the Screw", The University of Kansas City Review XX, 9 7 - 1 0 5 [Reprinted in Willen (1960): 2 1 2 - 2 2 3 . ] James, Henry 1898 [1962] The Turn of the Screw. Signet Classics Edition. New York, N.Y.: N A L Penguin. Levinson, Stephen C. 1983 Pragmatics. Cambridge: Cambridge University Press. Mansell, Darrel 1985 "The ghost of language in The Turn of the Screw", Modern Language Quarterly 46 (1): 4 8 - 6 3 . Prince, Ellen 1978 "On the function of existential presupposition in discourse", in: D. Farkas — M. Jacobson —K.Todrys, (eds.), Papers from the Fourteenth Regional Meeting of the Chicago Linguistic Society, 362 — 376. 1981 "Toward a taxonomy of given-new information", in: P. Cole (ed.), Radical Pragmatics. New York: Academic Press, 223 — 256. Reed, Glenn A. 1949 "Another turn on James' The Turn of the Screw", American Literature XX, 4 1 3 - 4 2 3 . [Reprinted in Willen (1960): 189 - 200.] Sheppard, E. A. 1974 Henry James and 'The Turn of the Screw'. Bungay, Suffolk: Richard Clay, Ltd. Vaid, Krishna B. 1964 Technique in the Tales of Henry James. Cambridge, Mass.: Harvard University Press. Weber, Jean Jacques 1982 "Frame Construction and Frame Accommodation in a Gricean Analysis of Narrative", Journal of Literary Semantics 11.2: 90 — 95. Willen, Gerald (ed.) 1960 A Casebook on Henry James's "The Turn of the Screw", (New York: Crowell. Wilson, Edmund 1934 "The ambiguity of Henry James", [in: Willen (ed.), 115 — 153]. [1960]

How to cope with dramatic texts including avant-garde playscripts Herbert Grabes

The reading of dramatic texts generally takes place under the assumption that plays are written to be performed, that the lines are to be spoken by actors, and that the so-called stage-directions are to be taken by a director as indications of the facial expressions, movements and gestures of the characters impersonated by the actors as well as of the kind of setting in which they are to be placed (including lightning and sound effects). This assumption that dramatic texts serve primarily as blueprints for a theatrical performance has not only determined the way they are written. It has not only given them a structure, indicating that some parts of them are to be transformed into spoken language and that the other parts are there to inform us by whom and in which surroundings this spoken language is to occur. What it has also done is to divert attention from what happens when such texts are being read — so much so, in fact, that within the broad scope of dramatic theory and criticism cognizance is hardly ever taken of the reading of plays as written texts. I must admit that I was not aware of this before I started preparing a paper for the Jerusalem Theatre Conference 1988 on the theme "To Read a Play" 1 and noticed that there was very little previous research to draw on. This is all the more astonishing in view of the fact that within the literary theory of the last decades the focus of attention has definitely shifted towards the reader and the reading process. But a closer look reveals that examples are taken almost exclusively from the domains of prose fiction and poetry, whilst the drama, at least in its more recent experimental forms, has been omitted from consideration. On the other hand, dramatic texts are read not only by the professionals of the theatre, who have to become acquainted with a playscript first before they can go and stage it. The very fact that — unlike most film scripts — theatrical plays are generally published and sold on the book market proves that there must be a wider readership, even if dramatic texts never become popular enough to make it onto a bestseller list. The reading of a dramatic text after one has seen it performed is quite different

296

Herbert

Grabes

from any other kind of reading of that text. In many ways such a reading is much easier, if more biased, because the memories of the non-verbal elements of the performance unavoidably become an important and sometimes decisive factor in the process of making sense of the text. Thus the following observations and assertions pertain only to a situation where a dramatic text is read before it has been seen in performance — that is, where the reader has just the written dramatic text to rely on. This does not mean, of course, that it would be wise to posit a totally naive reader — on the contrary. The following remarks concerning the general and specific problems one has to cope with when reading dramatic texts are meant to be valid for any kind of reader, although with increasing experience — also experience as a theatre-goer — the ease and the scope and degree of inventiveness with which these problems are solved quite naturally also increases. Let us first look at some of the general problems we encounter when trying to cope with written dramatic texts. It may be of some help here to venture a comparison with the reading of prose fiction, not least because we can draw on the results of a considerable amount of research in that field. And I hope it does not need stressing that I have to be highly selective in this comparison in order to emphasize the particular problems caused by dramatic texts — and only those. When we start trying to make sense of what we encounter in the process of reading, it seems to be of utmost importance to work with units of meaning small enough to be grasped in an instantaneous mental synthesis, and comprehensive enough to attain a degree of autonomy which makes them seem sufficiently reliable. And in narrative prose we find that such a reliable elementary unit of meaning is the sentence. Whilst reading novels and stories we take in strings of sentences, form a semantic synthesis for each sentence we encounter, connect it to the meaning of the previous sentences and build expectations concerning those which are to follow. We thus form larger units of meaning by what phenomenologists have called "retention" and "protention", a backward and forward movement of the synthesizing faculty of the mind bent on constructing more comprehensive clusters of meaning in order to finally integrate these in a over-all dynamic synthesis of human consciousness. What is important here is, first, that in all this flux we can always rely on the semantic unit of the sentence. But in narrative prose we usually also find some help when trying to integrate the meaning of a sentence we have just read into the provisional synthesis of all the earlier sentences and when making assumptions about the ones to come. What I am

Dramatic

texts

297

referring to is the predominance of a more or less identical perspective which is sustained beyond the unit of the sentence and proves to be the most effective instrument of narrative mediation. This stability of perspective is, of course, no more than an assumption, but it is an assumption which is successful enough with narrative texts to serve as a fairly reliable guide. How important this guidance is becomes quite clear when it fails, when the subsequent sentences in a narrative text cannot be linked up with an identical perspective, as for instance in the first chapter of Joyce's Ulysses or in much of postmodern narrative, for instance, John Barth's Lost in the Funhouse. A n d I am stressing this fact because we then approach a situation much closer to the one a reader of dramatic texts quite generally has to cope with. In order to explain why this is so I have to start from the surface structure of dramatic texts and therefore cannot avoid pointing out some facts that may seem all too obvious. Thus, for instance, it must be mentioned that the most striking feature of dramatic texts is their allocation of various textual quantities to individual or functional names which tend to recur. What we encounter there is indeed quite varied: it ranges from a single word or a short phrase to a long string of sentences which may even form a little story. Thus, when reading dramatic texts, though we will encounter sentences we cannot rely on the unit of the sentence in the same way we do when reading fictional prose. In reading what is called dramatic dialogue we quite often have to synthesize sequences of words or phrases allocated to various names in order to reach the degree of comprehensiveness and autonomy of meaning guaranteed by the sentence in narrative prose. A n d as long as we recognize a pattern in the sequence of words and phrases which we know quite well from speech acts (such as questions and answers), we do this quite easily. But, as we shall see, this is not always so — especially in some recent avantgarde playscripts — and in this case the demands made on the synthesizing faculty of the reader are extremely high. That they are quite generally higher as far as the formation of larger units of meaning is concerned has, however, above all to do with the lack of guidance through a predominant perspective. N o t that any perspective would be lacking — indeed the perspective for each quantity of text allocated to a name (or what we take to be an utterance) is by this very allocation more clearly marked than in narrative prose. The problem is, that none of these subsequently encountered perspectives can prima facie be considered as being predominant.

298

Herbert

Grabes

Thus with dramatic texts we have to form larger units of meaning in a different way. As already indicated, in his attempt to link each new allocation of a verbal unit to a name with the immediately preceding one or even earlier ones, and in his assumptions about what is to follow, the reader will use his knowledge of the various types of speech acts and more comprehensive patterns of verbal action. But at the same time he will relate each new allocation to the earlier verbal units allocated to the same name, thus creating textual strings with an identical perspective which are used to build the larger semantic units we call dramatic characters. And by taking both this actional and figural synthesis together the reader will be able to reach the more comprehensive synthesis we usually call plot. So far the lack of guidance by narrational mediation can be compensated for by the reader's knowledge of patterns of speech and verbal action as well as typical kinds of behaviour, character traits and types of character. Where this lack makes itself sadly felt, however, is on the higher level of thematic synthesis. Here the reader of dramatic texts is left much more to his own ingenuity, and I would venture the hypothesis that for this reason there is generally more critical dissent about the thematic meaning of plays than there is about that of works of prose fiction. Whilst we may eventually come to agree on what happens in Hamlet, the shelves full of interpretations bear ample proof that we may still disagree considerably on what the play means. Thus, all in all, the demand on the synthesizing faculty of the reader when coping with dramatic texts in not only of a different kind than in the case of fictional prose, but also a higher one. This may be one explanation why the reading of plays not only requires some practice but also tends to be slower. And though the so-called "stage directions" which are placed at the beginning of a play or its acts and scenes or interspersed in the main text provide the reader with information about the spatial and situational placing of the verbal exchanges, about the speakers signified by the recurring names, and about the way in which the verbal units signifying speech are taken to be uttered — and thus aid the building of more comprehensive units of meaning — they often enough interrupt the creation of an illusion of dialogue and are then rather so many stumbling blocks in the process of reading. In the written text these stage directions are mostly to be taken as equivalents to iconic signs in a performance, though especially since the advent of the realistic mode in the theater they also quite often contain information, for instance about the psyche of the characters, which cannot

Dramatic

texts

299

be directly transferred to the stage, but helps the sense-making of the reader. In order to be more specific we will now have a look at a number of examples — examples which are chosen because they more or less deviate from the dramatic texts of the so-called "well-made play" and thus create special problems for the reader and — what is more important here — allow us to see how these problems are connected with the absence of specific textual features. Let us assume that we have just started reading the Grove edition of David Mamet's Glengarry Glen Ross (1984). On the title page we find in addition to the title of the play and the name of the author the line "1984 Pulitzer Prize for Drama", an item of information which makes us expect a conservatively written play of superior quality. Then, after a list of other plays by Mamet published by Grove Press and a repetition of the title of the play and the name of its author, we encounter a dedication: "This play is dedicated to H A R O L D PINTER." For those who know Pinter's work this can serve as a suggestion concerning what they are to expect — although the hint is rather vague, of course: will we encounter a similar play or just a similar theme or similar language? Then, following another repetition of the title, we find a list of the casts of the 1983 London performance and the 1984 Chicago performance — a list which will prove useful insofar as it gives the full names of the characters of the play, although this only comes out when we look at the subsequent list of the characters, which only contains their surnames. In the Grove Press edition the characters on this list are grouped according to their age: four of the seven are "Men in their early forties" and three "Men in their fifties", and as this is the only additional information included in this list it might be assumed that age is more important for the theme of the play than other qualities of the characters. There is, however, some more advance information about the characters implied in the immediately following stage direction about "the scene": as Act Two takes place in a real estate office they must be somehow connected with that kind of business. And that we will have to do with business may, above all, be taken from the epigraph that we encounter next: "ALWAYS BE CLOSING. Practical Sales Maxim." The actual text then begins with the indication of the act and scene division which we normally expect in dramatic texts; this is followed by a further stage direction, pertaining to "Scene One" and placing one of the younger men, Williamson, and one of the older ones, Levene, in a booth at a Chinese restaurant. What then follows is a dialogue with a

300

Herbert

Grabes

very uneven distribution of textual quantities. It is obvious that Levene does almost all the talking, while Williamson hardly ever gets in more than a short sentence, mostly even less. On the level of purely figural synthesis we may infer from this that Levene is a talkative guy and Williamson is not; when, however, we consider what they are saying we soon come to the conclusion that it is rather the uneven distribution of power than character traits which make for the uneven distribution of speech. Williamson does not have to say much, because he has the say anyway, whereas the older Levene has to try all kinds of persuasive strategies, including bribery, to get the help from his colleague he depends on for his livelihood. This may serve as an example of how the various kinds of synthesis required for making sense can be differentiated in theory, but must interact very closely indeed in the process of reading. The main problem the reader is faced with in this case is that he has to keep trying to infer from the text not only the wider implications of the very specific business talk he is becoming acquainted with, but also the narrower context in which this talk is to be placed. He must even tolerate the fact that so much of this immediate context remains unknown during this whole scene that he cannot make a fair guess in which direction the plot may develop. Those who are well acquainted with more recent dramatic texts will not, however, be surprised at this kind of uncertainty, which may seriously disorientate a less experienced reader. As the following examples will show, Glengarry Glen Ross is still close enough to the convention of the well-made play to be in fact a very mild case as far as the probing of the reader's patience and perseverance is concerned — especially as later on in the play this patience will be rewarded. The best known example of a dramatic text in which this uncertainty is created right at the beginning, continued throughout the whole play and never resolved, is, of course, Beckett's Waiting for Godot (1956). Though all the conventional features of a dramatic text are there and it is not too difficult to combine the subsequent textual units of meaning in such a way that we can imagine a sequence of concrete speech situations and scenes, we continually encounter textual elements which do not seem to fit, elements which resist a synthesis of this kind and thereby undermine the creation of a consistent illusion. There are two examples right at the beginning of the play. Whilst we may still link Vladimir's reaction to Estragon's initial statement by assuming that he takes the "Nothing to be done" (Beckett 1956: 9) (clearly referring to the unsuccessful attempt to take off a boot) as a very general

Dramatic

texts

301

statement about his whole life, Estragon's "Am I" in sequence with Vladimir's "So there you are again" is definitely puzzling. A second instance can be found just a few lines further down: Vladimir: Estragon: Vladimir: Estragon:

And they didn't beat you? Beat me? Certainly they beat me. The same lot as usual? The same? I don't know. (Beckett 1956: 9)

Not only don't we know who "they" are and why they should have beaten Estragon: what is puzzling is the matter-of-fact way in which Estragon takes it for granted that he was being beaten, the fact that Vladimir seems to share this attitude, and especially that Estragon professes not to know whether it was the same lot that beat him, although if anyone knows it should be him. It is details of this kind which keep prompting the reader to ask himself whether the "characters" he is to synthesize from the sequence of such statements are in their right mind; and it is such details that have led critics to apply to this play the terms "absurd". But at least at this early stage of the reading process we may still think that the open questions will be answered — that, for instance, we will learn in good time who "they" are who do the beating and why they do it. Eventually, however, the impression becomes overwhelming that such uncertainties will never be resolved but just followed by more uncertainties. This is reinforced by the fact that the abundant stage directions are anything but supportive of a semantic synthesis. The gestures and actions they describe more often than not seem to be totally dissociated from the verbal exchanges, and must appear meaningless in themselves when those who act are convinced that nothing is to be done. Thus, in the process of reading, the reader's hope that he will soon be in a better position to make sense of what is taken in has to be continuously deferred until it finally dies. And only a reader who possesses enough perseverance to continue, a reader who takes this persistent frustration of sense-making as a challenge and asks himself what the meaning of a text might be which consistently negates the construction of meaning, will finally get a chance to solve the problem. For at least one way to solve it is to assume that Beckett has tried to write a text that will make the reader experience a kind of perseverance beyond hope which is similar to the one that governs the life of the protagonists of the play, so that what we finally get is a mimesis which strongly emphasizes the point that this kind of situation is not a special case with alienated tramps but a universal human predicament.

302

Herbert

Grabes

What we can learn from this example as far as our frame of reference is concerned, is that if the usual way of coping with a dramatic text should fail it is well worth trying to revert to a quite different approach. This we will definitely have to keep in mind when dealing with avantgarde playscripts, which pose even more serious problems, because they seem to defeat sense-making on much more basic levels of semantic synthesis. What I am referring to will soon become clear when we look at the beginning of Act I of Robert Wilson's I Was Sitting On My Patio this Guy Appeared I Thought I Was Hallucinating (Wilson 1979). For this purpose, and because this text is not so widely known, I shall have to quote from it more extensively:

ACT I ι

I was sitting on my patio this guy appeared I thought I was hallucinating I was walking in an alley you are beginning to look a little strange to me I'm going to meet them outside 5 have you been living here long NO just a few days would you like to come in sure would you like something to drink 10 nice place you've got don't shoot don't shoot and now will you tell me how we're going to find our agents might as well turn off the motor and save gas 15 don't just stand there go and get help I've never seen anything like it what are you running away from [you] you 20 has he gotten here yet has who gotten here yet NO

25

what would you say that was [what would you say that was] 1 2 5 [12 5] very well [very well] play opposum

Dramatic texts

303

30 [play opposum] open the doors [open the doors] one you all set [one you all set] 35 go behind the door [go behind the door] now is the time to get away [now is the time to get away] 1 and2 40 [1 and 2] I'll be with you in just a minute [I'll be with you in just a minute] I'll be with you in just a minute [I'll be with you in just a minute] 45 oh hello that's just the call I was waiting for [oh hello that's just the call I was waiting for] ready aim fire [ready aim fire] aim fire so [aim fire] where [where] you're here for ulterior motives I graduate with honors 55 you're O.K. watch out father [father] NO 60 NO NO 15 years ago I remember the address it was 7 Pearl Street I must keep that in my mind or was it 7 years ago the reindeer are getting restless 65 there's a mechanical drummer there's a mechanical soldier there's a mechanical bird girl she's all made of silver the snake was used to living in a warm climate 70 Codie (Wilson 1979: 2 0 1 - 2 1 8 )

What most probably strikes one first is that the names to which the verbal units in dramatic texts are usually allocated are lacking. On the other hand, the fact that we are faced with lines very much varying in length and not syntactically linked creates the impression that they have

304

Herbert

Grabes

to be allocated to various imaginary speakers. Thus the reader finds himself confronted with the question, which of the subsequent lines should be attributed to the same and which to a different speaker, without knowing how many speakers there will be. But this is not all. Though the graphemic, syntactic and semantic structure of the text insinuates that there must be several speakers, a conscientious reader who has taken in the "Prologue" will know that all the lines are meant to be spoken by one and the same actor, "punctuated by music played on an offstage piano." (Wilson 1979: 202) When, keeping this in mind, one starts looking for at least some semantic links between the subsequent lines, one will find them here and there, but then they are structured on the questionand-answer pattern, as are lines 5 to 9; or 20 to 22 — that is, they call for two speakers and not one, or even more often they consist simply of repetition (lines 11 and 12; 18 and 19; 25 and 26; 27 and 28; 29 and 30; 31 and 32 and so on). And only a reader who reaches the very last line of the "Post Script" of this play will at least know that those of the repeated lines that are put in brackets are "spoken by a disembodied male voice," (Wilson 1979: 218) that is, they are obviously to be taken as a kind of echo coming from a loudspeaker. All in all, the attempt to construct any situational, actional, or figural synthesis beyond the scope of a few lines at best, is frustrated very early on. On the other hand, except for the first two lines all statements are in the present tense and thus give the impression of being units of spoken language such as we find in dramatic texts. How are we to cope with this? We can go one step further by drawing a positive conclusion from the failure to combine the string of lines into the hypothetical synthesis of a single speech situation, and instead by considering it as a sequence of snatches from different speech situations. And we get even further if we connect this conclusion with the information that these snatches are rendered or quoted by one and the same speaker, who may in the first line be speaking of himself or, indeed, be hallucinating. This assumption would also be in tune with the title of the play. But it may also help to cope with a montage of fragments of speech to remember the stage instruction in the prologue that "the stage is set but dark except for a spotlighted telephone on a small aluminum table downstage left. The telephone begins ringing continuously 10 minutes before the curtain." (Wilson 1979: 202) What we will get when we follow the suggestion will be something like a continuous recording of one side of various conversations over the phone.

Dramatic

texts

305

This would also fit the surprising fact that Act II of this play consists of a text identical with that of Act I, only this time spoken by a woman, "her words punctuated by music played on an off-stage harpsichord." (Wilson 1979: 217) And when Wilson adds in the "Post Script" that "The first act could just as easily be played by a woman and the second act by a man, or the cast could very well consist of two men and two women" (Wilson 1979: 218), it becomes quite certain that the traditional link between speaker and spoken text is deliberately made arbitrary and thus broken. And as the spoken words should also be punctuated by music, their dissociation from the situations in which they have been used is likewise stressed, so that attention is focused on nothing but the bare phrases. Having come thus far by just relying on the text of the playscript, it may be permissible to introduce a remark by the author about the actual composition of this text: I wrote a few pages of dialog at a time in a large notebook with blank pages that I often write and draw in when working on a piece. The language I wrote was more a reflection of the way we think than of the way we normally speak. My head became like a TV, switching from thought to thought (and in writing from phrase to phrase) like flipping from channel to channel. (Wilson 1977: 76)

Here we finally have the missing pragmatics for a congenial reading of this playscript: it just takes a reader experienced enough in channel flipping to make sense of the seeming chaos — or at least feel quite at home with it. The second example of an avant-garde playscript I would like to present is Gertrude Stein's Counting Her Dresses (1968). Again I have to quote more amply so that you can get an impression of how the text is written and what kind of problems it poses for the reader: C O U N T I N G HER DRESSES (A PLAY) PART I. ACT I.

When they did not see me. I saw them again. I did not like it. ACT II.

I count her dresses again. ACT III.

Can you draw a dress.

306

Herbert Grabes ACT IV.

In a minute. PART II. ACT I.

Believe in your mistake. ACT II.

Act quickly. ACT III.

Do not mind the tooth. ACT IV.

Do not be careless. P A R T III. ACT I.

I am careful. ACT II.

Yes you are. ACT III.

And obedient. ACT IV.

Yes you are. ACT V.

And industrious. ACT VI.

Certainly. PART IV. ACT I.

Come to sing and sit. ACT II.

Repeat it. ACT III.

I repeat it. PART V. ACT I.

Can you speak quickly. ACT II.

Can you cough. ACT III.

Remember me to him. ACT VI:

Remember that I want a cloak. PART V I . ACT I.

I know what I want to say. How do you do I forgive you everything and there is nothing to forgive.

Dramatic texts

307

PART V I I . A C T I.

The dog. You mean pale. ACT II.

N o we want dark brown. ACT III.

I am tired of blue. PART V I I I . A C T I.

Shall I wear my blue. ACT II.

Do. PART I X . A C T I.

Thank you for the cow. Thank you for the cow. A C T II.

Thank you very much. PART X . A C T I.

Collecting her dresses. A C T II.

Shall you be annoyed. A C T III.

Not at all. PART X I . ACT I.

Can you be thankful. A C T II.

For what. A C T III.

For me. (Stein 1968: 2 7 5 - 7 7 )

First we may notice that, as in Wilson's play, there are no names allocated to the verbal units which are to be taken as utterances. Thus we will again have to make guesses about what is probably spoken by whom. Whilst attempting to do this we cannot fail to see that Stein uses the terms "Act" and "Part", which usually designate large textual and semantic units in dramatic texts, in a totally different way. Her play comprises no fewer than 41 "Parts" of between one and seven "Acts" which are never longer than three lines each. And it very soon becomes

308

Herbert Grabes

clear that the term " A c t " here has to be taken literally in the sense of a verbal act, a single utterance to be allocated to one speaker, so that the act divisions are indicative of a change of the speaker and thus at least partially make up for the lack of names which usually fulfil this function. It is not quite so easy to find out what the division into "Parts" designates. Though we can soon see that each " P a r t " possesses some kind of unity in terms of comprising one verbal exchange, it seems unclear whether we have to assume that at least one o f the speakers changes, or the theme, or both. Only after surveying a greater number of "Parts" does it become most likely that they are meant to contain one speech situation each, and that we get a new " P a r t " whenever one element of that situation changes. All this still sounds rather abstract and the lack of names and stage directions makes it difficult indeed to form the kind of actional or figural synthesis we need in order to get a proper illusion o f concrete situations, characters and plot. Here it makes a difference whether the reader has had prior experience of 20th century artistic texts, quite a few of which are underdetermined in this way and must be taken rather as being suggestive o f a multiplicity o f possible meanings to be supplied by the reader than as providing sufficient information to reconstrue a single determinate meaning. When one approaches the text in this way, it is not too difficult to invent story lines which easily accommodate the subsequent textual units. Let us start with the title: who would be counting her dresses? Perhaps a husband who thinks his wife spends too much money on dresses and already possesses more than enough? The drift o f the following snatches of conversation soon indicates that we would do much better to think of someone who is left with just the dresses of a woman that lived with him and left him, a woman he misses, and who therefore holds on to what is left o f her and keeps counting her dresses. We then get various snatches from conversations of that person — who may also be a woman, by the way — with friends or with him- or herself, referring to the inner state of the grief-afflicted person or some external banality or some further stage of their relationship with the lost lover (as in Part X when there is mention of "Collecting her dresses" and he or she is asked "Shall you be annoyed?"). There remains, of course, a high degree of uncertainty. In the last " P a r t " of the text, for instance, there is even talk of a "homecoming", and the whole parting of lovers may have been just a temporary falling-out. But it is surely possible to build a story of this kind from the text, and though plenty of variations may be possible, it is also possible

Dramatic

texts

309

to cope with an underdetermined text of this kind in a way which opens up a multiplicity of changes for the imagination of the reader without becoming totally arbitrary. Rather than being similar to a crossword puzzle to which there is only one proper solution, a text of this kind provides something like a game of chess at a particular stage, which can be continued and ended in various ways. My last example will be another play by the same author, Gertrude Stein's What Happened (1968). At least according to the subtitle this text is supposed to be "A Five Act Play", and there is indeed a proper division into five acts, even if these are at best of the length of a short scene in a traditional play, and though the names to which the verbal units making up the single utterances are lacking again, there are bracketed numbers between these units which may indicate various speakers. But in spite the fact that this makes the text look more regular than the last two I presented, the reader will find it at least as difficult to cope with. Here is "Act One": ACT ONE (One.) Loud and no cataract. N o t any nuisance is depressing. (Five.) A single sum f o u r and five together and one, not any sun a clear signal and an exchange. Silence is in blessing and chasing and coincidences being ripe. A simple melancholy clearly precious and on the surface and surrounded and mixed strangely. A vegetable window and clearly most clcarly an exchange in parts and complete. A tiger a rapt and surrounded overcoat securely arranged with spots old enough to be t h o u g h t useful and witty quite witty in a secret and in a blinding flurry. Length what is length when silence is so windowful. W h a t is the use of a sore if there is no joint and no today and no tag and not even an eraser. What is the commonest exchange between m o r e laughing and most. Carelessness is carelessness and a cake well a cake is a powder, it is very likely to be powder, it is very likely to be m u c h worse. A shutter and only shutter and Christmas, quite Christmas, an only shutter and a target a whole color in every center and shooting real shooting and what can hear, that can hear that which makes such an establishment provided with what is provisionary. (Two.) Urgent action is not in graciousness it is not in in clocks it is not in water wheels. It is the same so essentially, it is a worry a real worry. A silence a whole waste of a desert spoon, a whole waste of any little shaving, a whole waste altogether open. (Two.)

310

Herbert

Grabes

Paralysis why is paralysis a syllable why is it not more lively. A special sense a very special sense is ludicrous. (Three.) Suggesting a sage brush with a turkey and also something abominable is not the only pain there is in so much provoking. There is even more. To begin a lecture is a strange way of taking dirty apple blossoms and is there more use in water, certainly there is if there is going to be fishing, enough water would make desert and even prunes, it would make nothing throw any shade because after all is there not more practical humor in a series of photographs and also in a treacherous scupture. Any hurry any little hurry has so much subsistence, it has and choosing, it has. (Stein 1968: 2 0 5 - 2 0 6 )

If the fact that we have only numbers instead of names to suggest the speakers discourages any kind of figural synthesis, the strings of sentences and phrases allocated to these numbers defeat any attempt to construe semantic units in terms of a fictitious conversation, action or plot. The reason for this is not only that the sentences and phrases appear to be quite unconnected on that level and are at best linked up by the recurrence of syntactic patterns (as in paragraph five on page 205: "Length, what is length ..."; "What is the use of a sore ..."; "What is the commonest exchange ...") or the recurrence of single words over some distance (p. 205: "Silence is in blessing ..."; Length what is length when silence is so windowful ..."; p. 206 "A silence a whole waste of a desert spoon"). It is also quite difficult to make sense of the single sentences and phrases: A single sum four and five together and one, not any sun a clear signal and an exchange. Silence is in blessing and chasing and coincidences being ripe. A simple melancholy clearly precious and on the surface and surrounded and mixed strangely. A vegetable window and clearly most clearly an exchange in parts and complete. (Stein 1968: 205)

Not that it would be impossible to construe some sort of meaning on the basis of this kind of text. We must, however, be ready to accept quite unusual and sometimes bold metaphors to an extent we normally do not expect when approaching dramatic texts, but only when reading modern poetry. And a reader who knows that Gertrude Stein used to write poetic prose of this kind will be less surprised when this happens. But this does not invalidate the fact that we first have to change our strategy of reading from the one we are used to applying to written dramatic texts to another one required for reading modern poetry before we can cope with this kind of text. And from this we could draw the conclusion that we are not faced with a dramatic text at all, but a poetic one — notwithstanding the subtitle.

Dramatic

texts

311

On the other hand this does not say more than that our normal strategy of reading applied to dramatic texts is based on the expectation of what has been called by Aristotle an imitation of an action, even if it should be mere verbal action. We might also then take this text by Gertrude Stein as an attack on this very assumption, as an attempt to stretch the term "play" so that it may also accommodate a recital of poetry by several speakers, with no imitation of an action whatsoever. In this case the title would have to be seen as an ironic question — which is not so far-fetched, because Stein at least in this text consistently avoids question marks where we would expect them. "What happened" would turn out to be a text that places in question the traditional convention of reading dramatic texts, or at least makes clear how essentially this convention is governed by an interest in learning "what happened". This is, of course, only one more of the various possibilities of disappointing the readers' expectations we have found in dealing with avantgarde playscripts — though it may be a more fundamental one. As we have seen, none of the structural and semantic changes working to this end have resulted in a situation where an ingenious or at least flexible reader cannot finally cope with such texts. On the other hand it has to be stressed that the specific effect created by these avant-garde texts relies heavily on the existence not only of a convention of writing but also of a concomitant convention of coping with dramatic texts. The emphasis created by deviation presupposes the existence of a norm. In this way we can take the deliberate defeating of the readers' expectations as convincing proof that there exists not only a discrete range of written texts we call dramatic, but also a concomitant discrete range of strategies to cope with them. Coping with dramatic texts thus ultimately means the acquisition and application of this particular range of coping strategies. And what I think I have been able to show is that a familiarity with the languages of dramatic texts is quite sufficient for this purpose — though it certainly helps the mental staging of a play to be aware of ways in which texts behave or are made to behave in performance.

Note 1. The papers of this conference have been published: Reading Plays, ed. H. Scolnicov and P. Holland, Cambridge: Cambridge University Press 1991

312

Herbert Grabes

References Beckett, Samuel 1956 Mamet, David 1984 Stein Gertrude 1968 1968 Wilson, Robert 1977 1979

Waiting for Godot: A tragicomedy in two acts. London: Faber and Faber. Glengarry Glen Ross. New York: Grove Press. "Counting Her Dresses", Geography and Plays. New York: Something Else Press, 2 7 5 - 2 8 5 . "What Happened", Geography and Plays. New York: Something Else Press, 2 0 5 - 2 0 9 . "... I thought I was hallucinating", The Drama Review. (T 76). XXI, 4: 76. "I Was Sitting On My Patio this Guy Appeared I Thought I Was Hallucinating", Performing Arts Journal 11/12. IV: 201 —218.

The pragmatics of literary texts Lothar Bredella

Is it at all meaningful to speak of the pragmatics of literary texts? D o they actually have any pragmatic effects? But what is meant by pragmatics? In his essay "Pragmatics and Poetics" Teun van Dijk demands that we have to develop a theory of pragmatics before we can determine within this general theory the pragmatics of literary texts. What should such a theory deal with? Van Dijk suggests that the "relations between author, 'narrator' and the presented persons and events of a story" are perhaps the "most obvious subject for a systematic pragmatic account." (van Dijk 1976: 24) However, he also recognizes the "problems concerning the speech act status of literary utterances, viz. the set of conditions under which a literary utterance may be said to be 'appropriate' with respect to a given context." (1976: 25) It is van Dijk's intention to examine literary phenomena in terms of an explicit pragmatic theory of a natural language. Thus he formulates as the aim of his essay "to find out which specific requirements a more general pragmatic theory must fulfill in order to be successfully applied in the study of literature." (1976: 25) My approach to the question of the pragmatics of literary texts will be much more modest and will be inductive rather than deductive. I will start with a consideration of several literary texts, paying special attention to what the reader has to do in order to make sense of the literary text and how this process might affect him. I doubt, however, that it will be at all possible to determine the pragmatics of literary texts within a general theory of pragmatics. It is for me highly questionable whether a general theory of pragmatics can determine how we should conceive of the "relation between author, 'narrator', 'and the presented persons and events of a story.'" Even more problematic for me is that van Dijk limits his attention to the relations between author and text, without directing it to the relations between text and reader, which seem to me the decisive ones for the pragmatics of literary texts. I shall put this relation at the center of my considerations and investigate what it means to the reader to understand a literary text. From this perspective we can ask how the reader conceives of the relations between author, narrator, and the text's

314

Lothar Bredeila

characters and what these distinctions mean for the sense making process. But before we can explore this process we have to define what constitutes a literary text.

1. What kind of speech act is the literary text? In his essay "Pragmatics and Poetics" van Dijk discusses whether it would be appropriate to regard the literary text as a speech act and rejects this idea. He says the "drawback of such a treatment is of course that other speech acts, e. g. assertions, questions, commands, etc., also occur in literature, and it seems problematic to have more than one speech act at the same time." (1976: 36) He considers whether one could not solve this problem "by distinguishing between macro-speech acts and micro-speech acts, the first determining the whole discourse, the second merely characterizing the individual sentences of the discourse." (1976: 36) Thus the "'occurrence' of commands, questions, advices etc., may be seen as embedded, as speech acts of represented persons." (1976: 37) Yet, for van Dijk this is not a satisfactory solution because "it excludes the possibility that literary utterances have other pragmatic functions." (1976: 37) As he points out "a poem may function as a confession, a novel as an assertion or a piece of advice, etc., at least if some semantic conditions are satisfied." (1976: 37) But what seems unsatisfactory from the perspective of a general theory is appropriate for the reader who tries to make sense of the literary text. The reader has to deal with two kinds of speech acts, the individual speech act of the narrator or character in a literary text and the literary text as a whole, and it is exactly the relationship between these two kinds of speech acts that elicits the reader's interpretive efforts and enriches his reading experience. I would like to illustrate what this means by analyzing the understanding process of a simple poem by Langston Hughes: Mother to Son Well, son, I'll tell you: Life for me ain't been no crystal stair. It's had tacks in it, And splinters, And boards torn up, And places with no carpet on the floor — Bare. But all the time I'se been a-climbin' on And reachin' landin's

Literary

texts

315

And turnin' corners, And sometimes goin' in the dark Where there ain't been no light. So boy, don't turn back. Don't you set down on the steps 'Cause you finds it's kinder hard. Don't you fall now — For I'se still goin', honey, I'se still climbin', And life for me ain't been no crystal stair. (Hughes 1985: 1656)

Although the mother's words seem to express a simple and clear speech act I asked the students in one of my classes to describe it. For one student the mother's speech act is a command: "... don't turn back, don't you set down on the steps ... Don't you fall now." For another the poem expresses above all a mother's love and concern for her son. For yet another student the mother puts her son under pressure and imposes her concept of life on him. Another student adds that the mother wants to see her attitude and values confirmed because it would be painful to question her life full of privations. The more she has suffered, the more she wants to see her hard life confirmed. Another student introduces a different perspective. She states that the poem expresses the strength and pride of a black mother who encourages her son not to give in so that they both keep their self-respect and make the white people feel ashamed. For another student the poem expresses a mother's desperate attempt to reach out to her son, but the attempt will fail because the son will not accept her philosophy of hard work and privations. Another student does not place the emphasis on the fact that it is a black mother addressing her son, but stresses the notion of a mother who passes on her experience and wisdom to the younger generation. Another student stresses that the poem expresses an absurd situation: The mother is climbing but there is no goal, no reward, only hard work, privations and suffering. For her it's climbing for climbing's sake, like Sisyphos. Another student takes up this comment and says that this is an ironic comment on the American myth of success and the American Dream. The "crystal stair" reminds him of the "ladder of success" which guarantees wealth and prestige through hard work, industry, temperance and thriftiness. Yet, in the case of black people the reward is only the consciousness of their strength to endure hardship. These first responses indicate that it is possible to look at this apparently simple speech act from different perspectives and that an under-

316

Lothar

Bredeila

standing of the poem will be mediated through these responses. But before we pursue the understanding process further let us consider how these first responses come about. As we have seen, some of the students try to understand the mother's conscious intentions. Yet students who stress that the mother imposes her concept of life on her son seem to have gone beyond the mother's conscious intentions. While the first group of students tries to understand what the mother wants to express, the second group tries to understand what her behavior means to them, the students. Another form of understanding comes into play when a student argues that this poem is an ironic comment on the American concept of success and the American Dream. Here, the mother's words are not interpreted as signs of her conscious or unconscious intentions but related to concepts or myths of a culture. Another frame of reference or form of understanding comes into play when a student says that the mother's words express the strength and pride of black people or the wisdom of an older person addressing the younger generation. What do these responses and comments reveal about the understanding process? In order to understand the poem the reader has to consider several perspectives: He has to realize what the mother intends to say, what she perhaps unconsciously expresses and the author's intention in presenting the mother's speech act. Thus the meaning of the poem is mediated through an understanding of the mother's speech act. Therefore the literary text can be considered a speech act in which the "'occurrence' of commands, questions, advices, etc. may be seen as embedded, as speech acts of presented persons", but this does not exclude as van Dijk assumes that the poem as a whole "may not function as a confession, a novel as an assertion or a piece of advice etc." (van Dijk 1976: 37) A literary text can function as a confession or a piece of advice but on the other hand it is different from a real-life confession or piece of advice because its effect is mediated through the presented speech acts. This tension between the presented speech acts and the poem as speech act will be further explored in considering Ohmann's position in his essay "Speech Acts and the Definition of Literature." At the beginning of his essay Ohmann refers to literary theories which try to define literature through its relationship to the outside world. He examines the position which argues that language in a literary text does not refer in the usual way. As he points out, there is some truth in this position. "The name 'Mr. Pickwick' lacks a referent of the standard kind, as does 'Brobdingnag'." (1971: 4) Yet this feature, the lack of reference, is for Ohmann not sufficient for a clear cut distinction between literary texts and non-literary

Literary

texts

317

texts, because "many words in ordinary discourse lack referents of the sort postulated by this astringent notion of reference. I have in mind not only syncategorematic words like 'for' and not only oddities like 'chimera', but commonplace nouns like 'mailbox' in Ί can't find a mailbox anywhere in town'." (1971: 5) But there is another reason speaking against this position: "many words in literature do refer in the usual way." (1971: 5) Without reference the author would not be able to build up the fictional world of the literary text. The other position Ohmann considers says that literary texts cannot make assertions. One can develop this position, as he points out, in two opposite directions. One direction would stress that literature is lies. It may sometimes tell the truth, but as one does not know when this is the case, literature is unreliable. The other direction avoids this consequence by arguing that literary texts do not claim to tell the truth and "cannot therefore be held accountable to standards of verity." (1971: 5) Ohmann finds this position "intuitively attractive," but thinks that it cannot serve as a basis for a definition of literature. According to Ohmann, both criteria, lack of reference and lack of verity, do not define the relationship between the world of the text and the outside world adequately. Therefore, he tries a new approach and considers the nature of the literary text as a speech act. For him, the essential insight is that literature is not a speech act but an imitation of a speech act. What does this definition imply? According to Ohmann, an imitated speech act has lost one essential element, its illocutionary force: "A literary work is a discourse abstracted, or detached, from the circumstances and conditions which make illocutionary acts possible; hence it is discourse without illocutionary force." (1971: 13) Ohmann adds that this definition is in line with Austin's view that a poem uses language in a "parasitic" way, so that the illocutionary forces undergo "etiolation" (cf. 1971: 13). Ohmann's position does justice to the fact that a speech act in a poem is different from the same speech act in everyday life. The reader of "Mother to Son" is in a different position than a son being addressed by his mother in real life. But Ohmann goes too far when he says that the language in the poem has lost its illocutionary force. By denying the illocutionary force he also denies the perlocutionary force. But as we have seen, readers of "Mother to Son" may experience shame, may be impressed by the strength and pride of the black mother or may critically respond to the hard work ethic. When we read the poem as social criticism, it may even be absurd to say that the poem is not social

318

Lothar

Bredeila

criticism, but rather an imitation of it. Ohmann is right in stressing that the presented or imitated speech act is not identical with the effect of the poem, however we cannot infer from this that literary texts have no illocutionary and perlocutionary force. Although Ohmann's attempt to define the literary text seems problematic, the analysis of his argumentation can give us a better understanding of the interaction between the text and the reader. He points out that an illocutionary act demands that "I must speak within a framework of conventions and circumstances, and do so in prescribed ways." (1971: 10) For example, the sentence "I thereby dismiss the class" only has an illocutionary force if I have the power to dismiss the class, but Ohmann argues that most of these felicity conditions as they are described in detail by Austin are not applicable in the case of literary texts. In support of his argument he analyses two lines from a poem by Richard Eberhardt: In June, amid the golden fields I saw a groundhog lying dead.

The first felicity condition — "There is a conventional procedure for stating, and more particularly, for reporting one's experiences" (1971: 11) — gives Ohmann no trouble. Yet the second criterion for felicity, that "the particular persons and circumstances in a given case must be appropriate for the invocation of the particular procedure invoked" (1971: 11), presents a problem for Ohmann: this second criterion means we have to ask whether Eberhardt is entitled or disqualified to make the statement about the groundhog lying dead. Yet Ohmann remarks that this criterion is irrelevant if we deal with a poem: Presumably if he (Eberhardt, L. B.) were blind, or asleep, or a total amnesiac, or a city dweller who did not know a groundhog from a warthog, he would be disqualified. Yet in fact none of these disabilities would in any way impede the business of the poem. (1971: 12)

Ohmann ignores one essential distinction: the distinction between the author of the poem and the speaker of the poem. The poet does not need to have experienced what he presents. Yet, if we want to understand the poem we have to imagine a speaker who is qualified to make this statement. If we were to find out later in the poem that the speaker is blind or asleep, we would be confused, to say the least. But there are cases in which author and speaker are not completely indifferent to each other. For many readers the understanding of "Mother to Son" would change if the poet were not black. For them, the fact that Langston Hughes is black qualifies him to speak for the black experience.

Literary

texts

319

According to Ohmann, Austin's felicity conditions number three and four are also irrelevant for the two lines from Eberhardt's poem. They run: "3) The procedure must be executed by all participants both correctly and 4) completely." (1971: 11) As he points out, "the golden fields" are not specified. We do not know which fields are meant and where they are. Again Ohmann plays off the world of the poem against the world outside the poem when he says: "... if the reader misunderstands a declarative sentence, the statement fails; but no misunderstanding will alter the status of the poem as speech act (or quasi-speech act) as I shall contend." (1971: 12) But this argument overlooks the fact that we cannot decide about understanding or misunderstanding an utterance without taking the context into consideration. Ohmann is right. We do not have to find a referent for the "golden fields" in Eberhardt's poem nor do we have to identify the mother in Hughes's poem "Mother to Son" with a person the poet has known in order to understand the poems. Yet what is the consequence of this conclusion that the representational function of language in literary texts is weakened or obliterated? Does it imply that the world of the poem has no connection with the world outside? It is obvious from the students' first responses to "Mother to Son,, that they relate the world of the poem to their own world and the evocative language of the poem involves them in a process of interpretation in which their experiences, values, and attitudes towards reality are at stake. As Jan Mukarovsky has pointed out, the weakening or obliteration of the representational function of language in works of art is compensated by their capacity "to refer to reality as a whole and to express and evoke man's relation to the universe." (1978: 21) Therefore it would seem that Ohmann is not justified in asserting that the literary text as an imitated speech act uses language in a"parasitic" way with no illocutionary force. The literary text as an imitated speech act liberates the reader from practical concerns so that he or she can respond more freely and does indeed affect his or her values and attitudes towards reality as a whole. The fifth felicity condition demands that the person believes what he is saying is true. Again, Ohmann ignores the essential distinction between the poet and the speaker when he argues that it does not matter "if Eberhardt had seen the groundhog in September (and knew it), or if what he saw was really a rabbit, or if he had had no experience resembling the one his sentence describes." (1971: 12) But Ohmann suddenly admits that every reader makes the distinction between the author and his creations. He says that his arguments might

320

Lothar

Bredeila

"run quite counter to what every reader knows about the conventions for this sort of discourse." (1971: 13) Like van Dijk, Ohmann considers the possibility that "there are two speech acts to judge: Eberhardt's, of quoting (reporting discourse), and the other speaker's, of reporting his encounter with the groundhog." (1971: 13) And, like van Dijk, he rejects this possibility and feels entitled to return to his former position. His main argument for doing this is that the speaker is imaginary "and one cannot ask about his qualifications and intentions as if they had any existence outside the poem." (1971: 13) But as we have already seen, understanding a literary text implies that we have to ask whether a speaker in a poem or a character in a novel is qualified to say what he says. This will become even more obvious when we turn to a closer analysis of the understanding process.

2. The interaction between text and reader In his book Der Akt des Lesens Wolfgang Iser points out that understanding is the reduction of contingency and indeterminacy. This implies that if literary texts engage us in an understanding process they first have to create contingency and indeterminacy. According to Iser there are two basic strategies to reach this goal: negation and gaps. Here I will only refer to the second strategy and examine its function in the understanding process of James Joyce's short story "Eveline." Eveline has decided to leave Dublin with her friend Frank in order to live in Buenos Aires. At the beginning of the story she is sitting at the window remembering what her life in Dublin has been like and picturing her life to come. Suddenly an organ grinder reminds her of her mother's death and the promise she gave to her dying mother, namely to keep the family together. But she is also reminded of her mother's unhappy life. This confirms her decision to leave Dublin: She stood up in a sudden impulse of terror. Escape! She must escape! Frank would save her. He would give her life, perhaps love, too. But she wanted to live. Why should she be unhappy? She had a right to happiness. (1927: 41 —2)

These are the last words of the first part of the story. At the beginning of the second part we find Eveline on the pier ready to board the ship that will take her to Buenos Aires. So far the story is coherent and plausible. Eveline suffers in Dublin and therefore wants to leave. Yet the following events challenge our expectations. It is impossible for Eveline to go on board. And the short story ends with the following words:

Literary

texts

321

No! No! No! It was impossible. Her hands clutched the iron in frenzy. Amid the seas she sent a cry of anguish! 'Eveline! Evvy!' He rushed beyond the barrier and called to her to follow. He was shouted at to go on, but he still called to her. She set her white face to him, passive, like a helpless animal. Her eyes gave him no sign of love or farewell or recognition. (1927: 42 — 3)

According to Iser a gap comes about when two segments of a text border on each other in such a way that the expected order of the text is interrupted and the reader has to connect the contradictory elements (cf. Iser 1976: 302). This is the case with the two passages I have quoted. In the first one Eveline reinforces her decision to leave Dublin because she has a right to happiness. In the second one we learn that she cannot board the ship. Understanding the short story means that the reader is able to find the missing link that makes it possible to relate the two passages. Goethe once said that a text will be boring if the writer does not leave the reader something to think about. Here Joyce has left it up to the reader to find explanations for Eveline's behavior. These explanations are not in the text but the text guides the reader and directs his attention to some of the details that might serve as a clue. We will remember that she promised her mother "to keep the family together as long as she could." Thus we might infer that she felt obliged to stay. But her behavior at the pier speaks against a moral decision: "She set her white face to him, passive, like a helpless animal." This might pose the question whether Eveline has been conditioned to have no will of her own. Do others make decisions for her? "She had consented to go away, to leave her home. Was that wise? She tried to weigh each side of the question." She had not decided to go away but only consented to do so and she tried to weigh each side of the question. Does the question "Was that wise?" indicate that she is afraid of leaving and that she is looking for reasons to stay? The following sentence would confirm this: "It was hard work; — a hard life — but now that she was about to leave it she did not find it a wholly undesirable life." Yet she is also aware of her father's violence: "... lately he had begun to threaten her and say what he would do to her only for her dead mother's sake. And now she had nobody to protect her." The destructive atmosphere in which Eveline lives is further emphasized by the story's symbolism. Dublin here is closely associated with dust and its connotations of frustration and paralysis. This is reinforced by the contrast to Buenos Aires and its romantic connotations of freedom and happiness. It is not my intention to give an interpretation of "Eveline"; my hints should only indicate that the answer to the question "Why is Eveline

322

Lothar

Bredella

unable to leave Dublin?" is not formulated in the text, but has to be formulated by the reader. But this is not an arbitrary process. The more the text leaves for the reader to do the more it involves him and sharpens his sensitivity for what the text is saying. In Lector in fabula Umberto Eco even argues that each literary text creates its own model-reader. He points out that an open text such as Finnegan's Wake demands a reader who possesses an enormous cultural knowledge and is willing and able to pursue connotations and associations. The concept of the model-reader also includes that the reader tries to reconstruct the author's intention but this does not mean that the meaning of the text is identical with the author's intention (cf. Eco 1987: 72). As we have seen, for van Dijk the relationship between the author and the text is essential for the pragmatics of literary texts. Yet, from the perspective from which I approach the pragmatics of literary texts it is important to consider how the reader conceives of the relationship between author and text. For Wimsatt and Beardsley it is impossible to know the author's intention because the author's declarations about his intentions may not be the intentions which actually are effective. However, this impossibility is not a disadvantage for them because every human product is not to be evaluated by the producer's intention but by his achievements. They want to lay the foundation for an objective criticism. Therefore, they reject the question of the author's intention as mere speculation which will produce impressionistic and relativistic criticism (cf. Wimsatt — Beardsley 1954: 21). E. D. Hirsch agrees with Wimsatt and Beardsley in their endeavor to lay the foundation for an objective criticism. But he draws the opposite conclusion from their common goal. For Hirsch a literary text as a semantic object is open to many interpretations. If we want to have an objective criticism, then we have to concentrate on what the author intended in order to find out the correct interpretation. The author's intention is the "principle for judging whether various possible implications should or should not be admitted." (Hirsch 1967: 219) Thus the attempts to establish an objective criticism lead to opposite and contradicting strategies. But when we direct our attention to the interaction between text and reader we can develop a more flexible position. For Wimsatt and Beardsley as well as for Hirsch the reader plays no active role. They limit the function of the reader to registering what is in the text rather than contributing to the text's meaning. In Stories of Reading. Subjectivity and Literary Understanding Michael Steig remarks that he was educated in the era of the New Criticism and

Literary

texts

323

was convinced of the irrelevance of the author's intention. But he has found that this doctrine inhibits the reader's interaction with the text and that his students are relieved when they are told that it is not forbidden in his classes to consider the author's intention (cf. 1989: 20). According to Steig we cannot help but construct an author and attribute the literary text to him or her: A biographical construct is involved even in the simple attribution of some aspect of a work to an author. "Shakespeare shows," "Dickens says," or "Joyce alludes to" are rudimentary but widely used forms of constructing a phantasy author, just as Foucault insists, rather than statements about the 'real' author. The author we refer to is indeed a function of the process of interpreting, but Foucault's rejection of the "author function" as no longer necessary and restrictive to the full play of discourse is not self-evidently true. What may be restrictive for one reader may be liberating for another. (1989: 36)

Steig stresses that the question of the author's intention has to be seen from its function in the reading process. He makes clear that we construct the author's intention as we construct the text's meaning. But in this process information about the author's intention can play different roles. In many cases the knowledge about an author's intention will influence our understanding of the literary text. If we, for example, tend to take a racist statement ironically, but then learn that the author actually meant it seriously, we will have difficulties in keeping up our ironic reading. But in other cases we may flatly ignore the author's intention, as Charles Altieri points out: For example, many contemporary poets, like Allen Ginsberg, seek to break through the idea of persona and present their work as the author's direct personal speech addressing an audience. Yet we read such poems less as personal speech than as performance, to be understood and assessed in terms of the intellectual, moral, and emotional qualities they exhibit in response to what we take as a dramatic situation. (Altieri 1981: 189)

If understanding a literary text means becoming aware of what the author intended, then this concept can imply that we understand the author better than he understands himself, because we have to make conscious and explicit what the author might have done unconsciously. We might become aware of aspects of the literary text which the author was not aware of (cf. Altieri 1955: 169). In A Theory of Semiotics Eco remarks that understanding a literary text means that we as readers interpret what the author intended but at the same time we know that there is a surplus of meaning and that a text's meaning goes beyond the author's intention:

324

Lothar

Bredeila

So that in the interpretative reading a dialectic between fidelity and inventive freedom is established. On the one hand the addressee seeks to draw excitement from the ambiguity of the message and to fill out an ambiguous text with suitable codes; on the other, he is induced by contextual relationships to see the message exactly as it was intended, in an act of fidelity to the author and to the historical environment in which the message was emitted. (1977: 276)

Unfortunately at the universities this dialectic has not been developed. The fidelity is stressed but excitement "to fill out an ambiguous text with suitable codes" is not encouraged but frowned at. Yet, in order to understand the pragmatic effects of literary texts, we have to take this dialectic into account. While stressing that the reader has to supplement what the text left unsaid, however, we will also have to take into consideration that the reader is guided so that it is possible to say that the text creates the reader it needs. As Eco points out, we tend to conceptualize the reading process as a series of levels. On the one level we actualize our encyclopedic knowledge and carry out contextual and situational selections. On the next level we recognize the topic. Then we build up macro structures and reconstruct the fable and finally identify the ideological structures of the text. However, in the actual reading process, we do not proceed exactly like this. We jump from one level to the other in order to be able to make sense of what we are reading. When the sense making process breaks down on one level we switch to another (cf. Eco 1987: 83 — 106). This I would like to illustrate with the beginning of the short story I'm a Fool by Sherwood Anderson. The title I'm a Fool probably creates a certain expectation: We as readers might want to know why the narrator is a fool and wonder about the criteria for being a fool. Perhaps the narrator is a fool according to one set of values but he may be wise according to another. Our experience with literature might tell us that literary texts often play with these different perspectives. We may also expect a self-reflective narrator because most people who are fools do not confess that they are fools. The story begins with the following words: It was a hard jolt for me, one of the most bitterest I ever had to face. And it all came about through my own foolishness, too. Even yet sometimes, when I think of it, I want to cry or swear or kick myself. Perhaps, even now, after all this time, there will be a kind of satisfaction in making myself look cheap by telling of it. (1986: 1 0 5 - 6 )

The next two sentences seem to describe the place and time of his foolish act: "It began at three o'clock one October afternoon as I sat in the

Literary

texts

325

grandstand at the fall trotting and pacing meet at Sandusky, Ohio. To tell the truth, I felt a little foolish that I should be sitting in the grandstand at all." But then the narrator does not continue to report what happened, but reports what happened a year ago when he left home "with Harry Whitehead and, with a nigger named Burt." We learn that his mother cried and his sister Mildred, "who wanted to get a job as a school teacher ... stormed and scolded about the house all during the week" before he left, because: "They both thought it something disgraceful that one of our family should take a place as a swipe with race horses." (1986: 106) From the perspective of the fable the narrator does not come to the point. He jumps from one item to the next without telling us about the supposedly foolish act which made him so unhappy. Yet from an ideological point of view his ramblings are very interesting. While he describes his mother and sister as class-conscious, he seems to be free of these attitudes and enjoys his life as a swipe which seems to him the ideal life: Sometimes now I think that boys who are raised regular in houses, and never have a fine nigger like Burt for best friend, and go to high schools and college, and never steal anything, or get drunk a little, or learn to swear from fellows who know how, or come walking up in front of a grandstand in their shirt sleeves and with dirty horsey pants on when the races are going on and the grandstand is full of people all dressed up — What's the use of talking about it? Such fellows don't know nothing at all. They've never had no opportunity. (1986: 106)

The life of a swipe, as the narrator describes it, seems to be the American dream of the simple free life in contrast to the restricted life of civilization. Yet, if we remember that he sat in the grandstand we might get the idea that he is trying to explain away that he is also class-conscious. In this context the reader might take "grandstand" and "the stables" as symbols of two different worlds and develop the connotations of these two terms as the reader of Eveline develops the connotations of the terms "Dublin" and "Buenos Aires." The grandstand may stand for glamour, wealth, elegance, etc., while "the stables" may stand for poor people and hard work. But there are further connotations. From a Marxist point of view the grandstand may connote exploitation and the stables may connote the exploited class. From a Puritan point of view the grandstand may connote leisure, extravagance and sin. From the perspective of the American Dream the grandstand may connote success while the stables connote opportunity. Or "grandstand" and "stables" by their mere existence may indicate that there are different classes which might be a challenge to the American myth of equality. The narrator seems to be troubled by their existence but he makes the best of it. He says that he likes sitting in the grandstand as much as working in the stables as a swipe:

326

Lothar

Bredeila

I liked one thing about the same as the other, sitting up there and feeling grand and being down there and looking up at the yaps and feeling grander and more important, too. One thing's about as good as another, if you take it just right. I've often said that. (1986: 108)

The last sentence of the quotation "I've often said that" may make the reader wonder whether he said so and acted accordingly or whether he said so but acted differently. Again and again the narrator comes back to his beliefs. In another remark he stresses that for him all people, no matter whether they take care of horses or whether they own them, are of equal worth. But then, when he sees a beautiful girl in the grandstand he pretends to be the son of a rich man who owns race horses. Yet he does not consider that his lie betrays his belief in equality. On the contrary, for him the lie is not a moral but a practical issue: His lie was foolish because the beautiful girl will write letters to an address where he is not known. This story demands a model-reader who is able to see more in the story than the narrator and who develops interpretations different from those offered by the narrator. This is an irritating phenomenon for those who stress that the literary text is an autonomous object and to those like Ohmann who hold that it does not matter whether the narrator is qualified to make the utterances he makes.

3. When to practice suspension of disbelief? In an essay "Concerning what kind of speech act a poem is" Samuel R. Levin argues that a poet invites us to conceive of a world he creates. Therefore we cannot ask "whether what the poet (persona) says is true; one can only ask whether the speech act is felicitous." (1976: 151) This implies for Levin that we practise the willing suspension of disbelief: ... we have in the reader a tacit agreement to contemplate a world different from the actual world, a world of the poet's imagining in which novelties of reference and suspension of truth conditions will be tolerated. In short, if the illocutionary forces of the two sections of our higher sentence go through, then the perlocutionary effect on the reader is just what Coleridge called "the willing suspension of disbelief," the condition that constitutes poetic faith. (1976: 152)

Levin gives a sensitive description of the literary speech act. Yet I think it does not do justice to the reading and understanding process. The reader does not passively take in the meanings of the text but helps create them. In this complex interaction it is often difficult for him to decide

Literary

texts

327

when to practice suspension of disbelief, i. e. give up his values and conception of the world and when he has to activate them to make sense of the text. To a large extent the pragmatic effects of literary texts result from this complex relationship. The literary text demands of the reader that he is willing to conceive of a world that is different from his own. If we take a fairy tale such as "Little Red Riding Hood" it is clear that the world of the fairy tale is governed by conventions which are different from those of our own world. We have to accept that a wolf can speak and that characters can be swallowed by a wolf and survive. But even here, practicing the willing suspension of disbelief is a very complex process in which our prior knowledge plays an important role. Only against the background of our prior knowledge of the world can the rescue of Little Red Riding Hood and her grandmother be experienced as miraculous and bring us relief. This makes clear that the suspension of disbelief comprises only a small part of our prior knowledge. The fairy tale relies on our cultural knowledge that a wolf is a rapacious animal and is thus appropriate to take the role of the villain. If this role were taken on by a lamb, it would irritate the reader and thus elicit quite different responses from him. This indicates that no literary text can be understood without an interaction with the reader's prior knowledge. Little Red Riding Hood also relies on our prior knowledge about young girls. Such knowledge includes that young girls are credulous and can easily be deceived. We also put our knowledge about family relationships at the text's disposal so that such terms as mother and grandmother make sense. (Cf. Eco 1987: 162 — 65). This seems to be obvious and trivial. Yet literary theory, as we have seen, forgets that the literary text is not an autonomous self-contained object and that the reader's projections are an essential element of the reading process. When the understanding is running smoothly we are often unaware of how much it depends on our expectations and cooperation. We forget that meanings are mediated and get the impression that we merely read them off. But when we cannot make sense of the text our often aggressive responses indicate how we are involved. Shirley Jackson's famous short story "The Lottery", which was published in The New Yorker June 28, 1948, caused a storm of indignation. Hundreds of readers wrote letters to The New Yorker and the author. In her report "The Biography of a Story" Shirley Jackson quotes from these letters. Before I comment on some of them I will briefly refer to the short story. At its beginning we as readers are told that the morning of June 27th on which the lottery was to take place "was clear and sunny, with the

328

Lothar Bredeila

fresh warmth of a full summer day; the flowers were blossoming profusely and the grass was richly green." (1966: 556) Thus we might expect that a lottery will take place in an idyllic setting. Yet some of the events do not seem to fit into our frame of a lottery. For example: "Bobby Martin had already stuffed his pockets full of stones, and the other boys soon followed his example, selecting the smoothest and roundest stones." (1966: 557) But none of the readers will at this point of the story guess that these stones are collected to kill the person who will draw the slip of paper with the black spot on it. As we read on about the preparations for the lottery, we learn that Tessie Hutchinson "came hurriedly along the path to the square" saying: "Clean forgot what day it was." (1966: 559) Then follows the detailed description of how every family draws. We are also informed that one village is considering giving up the lottery but one member of the village, Old Man Warner, assures us that the lottery is identical with civilization and he rebukes those who want to give up the lottery: "Next thing you know, they'll be wanting to go back to living in caves, nobody work any more, ..." (1966: 561) Finally we learn: "Bill Hutchinson's got it." From Mrs. Hutchinson's reaction we can infer that "winning" this lottery is not good luck but bad luck: "You didn't give him time enough to take any paper he wanted. I saw it.. It wasn't fair." But Mrs. Delacroix seems to comfort her: "Be a good sport, Tessie." (1966: 562) Then members of the Hutchinson family are told to put back their slips of paper and draw again. They are informed by Mr. Summers: "'Remember, take the slips and keep them folded until each person has taken one. Harry, you help little Dave.'" (1966: 563) When they are told to open them, Mrs. Hutchinson refuses to do so and her husband forces the slip of paper out of her hand. It has a black spot on it. Then Mr. Summer says: "Let's finish quickly." And the story ends with the following sentences: Tessie Hutchinson was in the center of a cleared space by now, and she held her hands out desperately as the villagers moved in on her. "It isn't fair," she said. A stone hit her on the side of her head. Old Man Warner was saying, "Come on, come on, everyone." Steve Adams was in the front of the crowd of villagers, with Mrs. Graves beside him. "It isn't fair, it isn't right," Mrs. Hutchinson screamed, and then they were upon her. (1966: 564—565)

Let us now consider the readers' responses. One reader wrote: "Tell Mrs. Jackson to stay out of Canada." Another one said: "I think I had better switch to the Saturday Evening Post." Another reader wrote: "I expect a personal apology from the author." A reader from Massachusetts wrote: "I will never buy The New Yorker again. I resent being tricked into

Literary

texts

329

reading perverted stories like 'The Lottery'." (Jackson 1968: 220) And a reader from Minnesota expressed his frustration and disgust in the following words: "Never in the world did I think I'd protest a story in The New Yorker, but really, gentlemen, "The Lottery" seems to me to be in incredibly bad taste. I read it while soaking in the tub and was tempted to put my head under water and end it all." (1968: 221) One letter begins with the following words: "Never has it been my lot to read so cunningly vicious a story as that published in your last issue for June. I tremble to think of the fate of American letters if that piece indicated the taste of the editors of a magazine I had considered distinguished. It has made me wonder what you had in mind when accepting it for publication." (1968: 222) Our first reaction to these comments might be that these readers are unable to practice the willing suspension of disbelief. They do not seem able to accept the autonomous world of the literary text. But I think such a comment would even be more inadequate than the readers' responses I have quoted so far. If we accepted the arbitrary killing of a person at a lottery as a normal event, then we would probably not have understood the short story adequately. The short story wants us to read it against the background of our world view and expects us to be shocked. If we were to bracket our beliefs and practice the "willing suspension of disbelief' we would be like the persons in the short story. I think that the readers whose responses I have quoted fail because they are not willing to or are unable to consider that the world presented in the story sheds light on the world in which they live. They make the author responsible for what she has presented and demand an apology from her. Through their responses the readers indicate that they feel that "The Lottery" is not presenting an imagined or isolated world which has to be understood on its own terms, but a challenge to their own world view although they are not willing to consider it. What makes "The Lottery" so disturbing and disquieting is not so much the event itself as the way in which it is presented without any moral perspective. Not the slightest attempt is made by the narrator to explain the fact that people take part in a lottery in which they choose the victim from among themselves and carry out the stoning without any sign of guilt or any indication of justifying their deed. Therefore the readers whose reactions I have quoted are rightly shocked, but the shock does not lead to a self-reflective attitude in which they could learn something about human behavior. On the contrary, they reject the world presented in the story as the product of an insane and perverted mind

330

Lothar

Bredeila

and thus explain away the upsetting experience. One reader from New Jersey explicitly wants to be assured: "Surely it is only a bad dream the author had?" (1968: 217) Another reader senses that "The Lottery" may say something about his world but is deeply troubled that this may convey a negative image of the United States abroad. Therefore he is glad that The New Yorker does not have the wide circulation of Reader's Digest: I am glad that your magazine does not have the popular and foreign-language circulation of the Reader's Digest. Such a story might make German, Russian, and Japanese realists feel lily-white in comparison with the American. The old saying about washing dirty linen in public has gone out of fashion with us. (1968: 221)

One reader grasps the outrageous element in "The Lottery" very precisely, even if he rejects it as "false": (Illinois) If it is simply a fictitious example of man's innate cruelty; it isn't a very good one. Man, stupid and cruel as he is, has always had sense enough to imagine or invent a charge against the objects of his persecution: the Christian martyrs, the New England witches, the Jews and Negroes. But nobody had anything against Mrs. Hutchinson, and they only wanted to get through quickly so they could go home for lunch. (1968: 218)

This reader is aware of the shocking "naturalness" of the event. Everybody just wants to return to their daily routine. In this respect "The Lottery" resembles Erskine Caldwell's "Saturday Afternoon" in which a black man, Will Maxie, is lynched, but the lynching itself is given little attention. It is the reader who is trusted to bring a human perspective to bear on the world depicted in the story and to correct the distorted view of the narrator and of the characters for whom the lynching is only an exciting but brief interruption of their daily routines. Yet there is a group of readers for whom "The Lottery" is not outrageous at all. These readers are curious and want to know more about these strange customs. (Kansas) Will you please tell me the locale and the year of the custom? (Oregon) Where in heaven's name does there exist such barbarity as described in the story? (1968: 214) (Boston) Apparently this tale involves an English custom or tradition of which we in this country know nothing. (1968: 215)

This group of readers apparently practices "the willing suspension of disbelief." But they do not read "The Lottery" aesthetically — i.e. they do not explore how the short story challenges their views and what they experience responding to it — but read it as a documentary about strange

Literary

texts

331

customs they want to know more about. Thus the literary text has lost its challenging character. For Levin, as we have seen, we cannot ask whether it is true what the literary text presents but only whether it is a "felicitous" speech act. But what is the literary text as a felicitous speech act? For Levin the felicity conditions are fulfilled if "the reader is in the hands of the poet or, what would be better to say, in the world of the poem." (1976: 153 — 154) For Levin the poet transports the reader into the world of the poem so that he forgets the world he lives in and considers the fictional world to be the real one. As Levin points out conventions and metaphors achieve this transportation of the reader into the world of the poem and this immersion into the poem should be total: "... as long as we perceive the poet's descriptions as metaphors, our suspension of disbelief is not total and we do not share fully in the poet's vision. True poetic faith would consist in our perceiving, with the poet, his descriptions as literally true." (1976: 159) But I think that this is a one-sided description of the reading and understanding process. It does not do justice to the fact that the reader builds up the world of the poem and that his imaginative as well as his critical faculties are needed to do so. The world of the poem does not simply replace the reader's world, but both worlds interact and influence each other. I think that Levin as well as Ohmann are right in stressing that it is irrelevant whether the events described in a literary text have really happened or have been invented by the author. In this respect the author of a literary text cannot lie because he or she does not claim to be reporting facts. But it would be too rash to draw from this insight the conclusion that the world of the literary text is an autonomous world that can only be understood by practising suspension of disbelief. It is true that in understanding a literary text the relationship between the world of the literary text and the world it refers to recedes into the background but the relationship between the world of the literary text and the reader's world moves into the foreground. The effect of "The Lottery" is based on the discrepancy between the characters' and the readers' evaluation of the events. It is this disagreement between the two worlds that activates the reader's prior knowledge and experience, arouses his interpretative energy and forces him to find an acceptable perspective. The pragmatic effects of literary texts result from this intense interaction. This makes clear that pragmatic effects occur not so much after the literary text is understood but during the understanding process, which activates the reader's sedimented experi-

332

Lothar Bredeila

ence, concepts and values and which encourages him to make projections and to correct these projections. Aesthetic pleasure results from this activation of the reader who cannot subsume the particular under general concepts and is thus engaged in an interminable understanding process. From this perspective aesthetic pleasure and the pragmatics of literary texts can be seen as two sides of the same coin: the intensified interaction between text and reader.

References Altieri, Charles 1981

Act and Quality. A Theory of Literary Meaning and Humanistic Understanding. Brighton: The Harvester Press. Anderson, Sherwood 1922 "I'm a Fool", in: Lothar Bredella et al. (eds.), Confidence. Bochum: Verlag Ferdinand Kamp, 105—113. Dijk, Teun A. van 1976 "Pragmatics and Poetics", in: Teun van Dijk (ed.), Pragmatic of Language and Literature. Amsterdam —Oxford —New York: North Holland Publishing Company, 23 — 27. Eco, Umberto 1977 A Theory of Semiotics. London: The Macmillan Press Inc. 1987 Lector in fabula. München—Wien: Carl Hanser Verlag. Hirsch, Eric D. 1967 Validity in Interpretation. New Haven: Yale University Press. Hughes, Langston 1985 "Mother to Son", in: Nina Baym, et al. (eds.), The Norton Anthology of American Literature. Vol. 2 (2nd edition). London: W. W. Norton and Company, 1656. Iser, Wolfgang 1976 Der Akt des Lesens. München: Wilhelm Fink Verlag. Jackson, Shirley 1966 "The Lottery", in: J. Moffets - K. R. McElheny (eds.), Points of View. New York: The New American Library, 556 — 565. Jackson, Shirley 1968 "Biography of a Story", in: Stanley Edgar Hyman (ed.), Come along with me. New York: The Viking Press, 211 —214. Joyce, James 1927 "Eveline", in: Joyce, Dubliners. London: Jonathan Cape, 37—43. Levin, Samuel R. 1976 "Concerning what kind of speech act a poem is", in: Teun A. van Dijk (ed.), 1 4 1 - 1 6 0 . Mukarovsky, Jan 1978 Structure, Sign, And Function. New Haven: Yale University Press.

Literary texts

333

Ohmann, Richard 1971 "Speech Acts and the Definition of Literature", Philosophy & Rhetoric, 4: 1 - 1 9 . Steig, Michael 1989 Stories of Reading. Subjectivity and Literary understanding. London—Baltimore: The Johns Hopkins University Press. Wimsatt, William K. — Monroe C. Beardsley 1954 "The International Fallacy", in: William K. Wimsatt, The Verbal Icon. Studies in the Meaning of Poetry. Lexington: University of Kentucky Press, 3 - 1 8 .

"Deep structure signals" in fiction Volker Schulz

1. The historical and theoretical background of the concept of "deep structure signals" The " 'multi-layered character of literary a r t ' " had already been " 'noticed by Aristotle in his Poetics'' ' M — he enumerated and to some extent discussed six constitutive elements of tragedy, three of which make up the represented object (action, agents, and thought, i.e. the characters' mental make-up and ethical orientation), whereas two concern the means of mimesis (the songs and the direct speech of the characters) and one is characteristic of the specific medium of drama as play (scenery). 2 Aristotle's ingenious suggestion of a hierarchy of elements, however, was not taken up and systematically elaborated until our century. And even the first half of the 20th century was still characterized by the predominance of the notion developed in the 18th and dogmatized in the 19th century that a literary work consists of a one-dimensional bundle of distinct elements of "content" and "form", of "Gehalt" and "Gestalt".3 It is true, "structure" and "Gefüge" were key-words in the two probably most influential works of literary theory in Germany in the fifties and sixties, Wellek/Warren's Theory of Literature (1942 [1956]) and Kayser's Das sprachliche Kunstwerk (1948), but both words, in spite of the authors' occasional attempts at formal definition, were little more than vague labels which helped their users to avoid more old-fashioned metaphors like "organism" as well as the vague and simplistic "form" — "content" dichotomy of their predecessors. Neither Wellek/Warren nor Kayser, that is, were able to replace these obsolete concepts by a convincing new model. 4 This is particularly obvious in the case of Wellek/Warren, since the "system of strata" 5 postulated in one of their introductory chapters shrinks down to an amorphous, though superficially ordered, conglomerate of mere formal features when the "intrinsic study of literature" is discussed in some detail in chapters XIII to XV; in spite of the explicit assertion that " 'structure' is a concept including both content and form so far as they are organized for aesthetic purposes", 6 content elements

336

Volker Schulz

are only discussed in connection with "narrative fiction";7 and major elements of the literary work like composition/arrangement, tone, or theme(s) are not considered at all. Kayser's treatment of the constitutive elements of the literary work is less fragmentary and more clearly arranged but what he calls "die elementaren Sachverhalte";8 i. e. the element of content, on the one hand, and the three formal elements of 'verse' ("Vers"), 'style' ("sprachliche Formen") and 'composition' ("Aufbau"), on the other hand, are also seen as equally important and fairly independent 'qualities' ("Eigenschaften" )9 of the literary work, which in some mysterious way combine to bring about its individual shape. Both works, then, in spite of their explicit claims to systematical stringency, suffer from the lack of a coherent literary theory that is characteristic of the German school of the so-called werkimmanente Literaturbetrachtung and the American school of "New Criticism" to which they belong. The idea of a hierarchical order and a close interdependence of the constitutive elements of a literary work was first elaborated in a systematical way by the Polish phenomenological philosopher Roman Ingarden, who, in his book Das literarische Kunstwerk (1931), expounded "den mehrschichtigen Aufbau und die damit zusammenhängende Polyphonie"10 of the literary work. Unfortunately, however, Ingarden had a tendency towards vagueness and redundancy, his terminology, moreover, was so idiosyncratic and his style so involved and abstract — due to the ultimately philosophical, not literary-critical, motives of his undertaking 11 — that his stratificational model did not immediately find the wide acclaim it deserved and the clarification, modification, elaboration, and application it needed. In Western Europe and the United States the voice of dissatisfaction with the prevalent one-dimensional model of the literary work was first clearly to be heard in the publications of the "Chicago Critics", who were sensitive to the "unscholarly improvisation that characterized much of the literary theorizing" 12 of the New Critics and went back to Aristotle to find a reliable basis for their own work in this field. Although the Chicago Neo-Aristoteleans made a highly valuable contribution to literary theory, particularly in the realm of genre theory 13 , their general model of the literary work was a mere replica of Aristotle's model. This is particularly obvious when Olson defines the literary work as a "synthesis of differentiable parts" 1 4 — the very word "parts" being an Aristotelian echo 15 — or when he distinguishes between "what is imitated in tragedy

Deep structure

337

(object of imitation), in what (means of imitation), how (manner of imitation)". 16 A number of factors combined and mutually reinforced each other in the sixties to bring about a radically new orientation in literary theory, particularly in narrative theory, and finally replace the one-dimensional by a genuinely structural, i. e. a multi-dimensional, model of the literary work. Factor one was the belated but vigorous impact of Ingarden's ideas in the wake of the second and third editions of Das literarische Kunstwerk (1959 and 1965), increasing in the seventies partly due to Iser's repeated references to the Polish philosopher's innovative thinking. 17 Factor two was the rediscovery, propagation and modification/elaboration of Russian Formalist notions, most prominent among them the concepts of the shaping of non-aesthetic 'materials' ("Materialien") through aesthetic 'devices' (" 'Kunstmittel' oder 'Kunstgriffe'") in order to achieve a particular aesthetic intention or effect 18 and Propp's concept of 'function' ("Funktion")19 — this renaissance of Russian Formalist concepts was brought about by Slavist scholars like Erlich and Striedter 20 and by French theoreticians like Bremond, Greimas, Todorov, and Barthes. 21 Factor three was the growing awareness of far-reaching correspondences between the System of language and the 'secondary', semiotic, 'system' ("sekundäres System") of literature which uses the system of language as its medium or "Material". 2 2 This growing awareness resulted in a growing willingness to borrow concepts developed by structuralist and generative linguistics, e.g. "syntagmatic"/"paradigmatic", "signifiant"/ "signifie", "surface structure"/"deep structure", "transformation", or "semantic macrostructure", and to apply them to literary phenomena in a metaphorical way. On the basis of these three decisive factors and through the combined effort, and the mutual criticism, of the French theoreticians already mentioned and of other eminent scholars like Bradbury, Chatman, Culler, Dolezel, Fietz, Fowler, Genette, Hendricks, Holloway, Lodge, Lotman, Pfister, Prince, Ruthrof and Sceglov/Zolkovskij, 23 to name only some of the most important in alphabetical order, an understanding of the structure of the literary work has been emerging since the sixties which seems to me to be strikingly superior to that of former times. Some of these scholars have provided the basically structuralist framework with a markedly generative character (Dolezel, Fietz, Hendricks, and Sceglov/Zolkovskij) and I believe that their way of thinking in terms of an idealized process of the generation and, reversely, of the analysis of a literary work is particularly helpful. My own "generative model of narrative" (figure

338

Volker

Schulz

1) is a sort of synthesis and further development of a number of less comprehensive and less schematic models constructed by structuralist and structuralist-generative theoreticians.

Textual Surface

V Content

IV I

ψ

I ^

Content Surface Structure

Content D e e p Structure

Thematic Idea (Ultimate Meaning)

1

m - 4

-

! [ ! !

Hap !

1

^ ! ! H r t \ I

System Coordinate

Figure 1. Generative model of narrative

---—-!

v — Poison» _ _

m

\

—

lonenls Com!" of Fictitious V/olW

Deep structure

339

Although a detailed explanation of this model is not possible here 24 some words of comment are called for, I think. Firstly, each stratum or layer of the literary work which is "higher" than the basic "underlying" thematic idea is to be taken as the result of a particular phase in the series of transformations — indicated by arrows — leading from this thematic idea (or ultimate meaning) through various grades of specification, ordering and presentation of the fictitious world to the textual surface. Vice versa, the textual surface is the ultimate verbal manifestation of the whole of this generative process, its constituents acting as signifiants which stimulate and control the reader's construction of the various levels of "content" leading down to the work's ultimate meaning. 25 This means that strata II to IV consist of content as shaped by the artist so that the former distinction between content and formal elements no longer applies. Lotman is right in arguing that all the elements of literary works are "semantische Elemente".26 Although I definitely share Dolezel's opinion that " S T R A T I F I C A T I O N A L models" are "the most promising base of a complete text theory" 2 7 I am eager to stress the fact that considerable work has still to be done before we will arrive at a sufficiently sophisticated generative model of the narrative, let alone the literary work in general. Secondly, the author (ideally) first conceives an abstract "thematic idea" or "ultimate meaning" (I), which I take to include his basic attitude towards the abstract notions in question, his " Wertprogramm",28 in Fietz' terminology, or 'the intended reader's perspective', "die auktorial intendierte Rezeptionsperspektive",29 in Pfister's terminology. He then conveys and concretizes his thematic idea by way of embedding it in "content", i.e. in a fictitious world consisting, like the real world, of the three dimensions or components of happenings/acts, personae and space/time. By the procedures of concretization and specification, the author transforms the initial, rather abstract, content deep structure consisting of fable structure, structure of actants and spatio-temporal structure (II) into what I call "content surface structure" (III), i.e. the fable, as the sum total of the major events ("cardinal functions" or "nuclei" in the sense of Barthes), 30 the agents, i. e. the characters reduced to the features relevant to the fable, and the setting(s). Then, through artistic decisions concerning the order and the segmentation, the extensiveness and the ways of presentation, including point of view and tone, he transforms the content surface into the (still abstract) "content" (IV), i. e. the fictitious world as it is presented by the author in all its details. This content is finally transformed — through the author's stylistic decisions — into the

340

Volker Schulz

work's concrete "textual surface", its text (V), i. e. a syntagmatically and paradigmatically ordered set of signifiants which allows the reader to follow the reverse path in his attempt at first constructing the fictitious world in its various levels of abstraction and, via this procedure, at finally arriving at an understanding of the work's ultimate meaning. Thirdly, information on the fictitious world is given consecutively but not necessarily in accordance with the chronological order of the events: "fable" refers to the "natural" order of the major events, "plot" to the order, and the way of presentation, of the sum total of the events. Fourthly, the syntagmatic or process coordinate is equally important as the paradigmatic or system coordinate although it figures less prominently in the diagram to ensure its "readability": the narrative text is segmented into sentences, paragraphs, and often into chapters and "books", and the fictitious world is not "static" but "dynamic" since it consists of "events" as well as of "existents", to use Chatman's terminology, 3 ' so that changes in the components of happenings/acts, personae and space/time are possible in the course of the fable, which are indicated by a number of plural forms in the model (in most narratives you may, e.g., distinguish between an initial, one or several intermediate and a final structure of actants or constellation of agents). What I am going to do in the following — instead of discussing the particular status and function of the various components and layers — is to show that the level of "content deep structure" is more than an arbitrary, speculative notion. This will be done by demonstrating that in many cases the author of a narrative explicitly refers to this level in his text thus stimulating and controlling the reader's construction of this very level, i. e. ensuring his correct understanding of the thematic function of fable, agents and setting(s) by inserting "deep structure signals" into his text. 32 Deep structure signals are a special case of a more general phenomenon which has not yet found the amount of attention it deserves from literary theoreticians although it has always been intuitively made use of in practical criticism, i. e. in interpretations/close readings/analyses of individual texts, to a greater or lesser extent. The more general phenomenon consists in the fact that any literary work "may or may not contribute the terms of the analyst's code or metalanguage of description". 33 Words and sentences in the surface text, that is, are not necessarily confined to their primary task of gradually evoking the more or less numerous particles of which the "content" consists (stratum IV) so that the reader, by synthesizing these content particles, may continuously form, corrob-

Deep structure

341

orate, or falsify hypotheses about the nature of the deeper layers by way of abstraction until, at the end of his reading, he will make his final decision as to the ultimate meaning of the work in question. By words or sentences in the text the author can also provide the reader with "shortcuts", 34 can help him in his attempt at "making sense" of certain clusters of content particles, i.e. at constructing the content surface or the content deep structure or even the thematic idea itself before the cumulative process of reading the complete text is at an end. Often the first help of this kind already occurs at the beginning: the title may suggest parts of the content deep structure (e. g. in D. H. Lawrence's Sons and Lovers or Poe's The Fall of the House of Usher) or even the ultimate meaning (e. g. in Dickens' Hard Times or in K. Mansfield's short story Bliss). In connection with fiction, Hendricks has coined the term "metanarration" for this phenomenon. 3 5 "Deep structure signals", i.e. words, phrases, and sentences designating parts of the content deep structure, then, are a special case of meta-narration.

2. A tentative typology of "deep structure signals" Deep structure signals may be categorized, I think, according to the degree of indirectness of their reference to the level of content deep structure. The degree of indirectness mainly depends on the rhetorical status of the word or phrase used for meta-narration. Leaving special cases aside, there seem to exist four main types of deep structure signals. In the following, I will briefly discuss and fairly extensively demonstrate these types by drawing on a number of (arbitrarily) selected 19th and 20th century short stories, novellas and novels.

2.1. Critical terms The least indirect way of referring to the level of content deep structure of a narrative is the use of critical terms which designate constituents of the fable structure ("motifemes" 36 in Dolezel's terminology), or their beginning and ending, actantial roles of the agents ("Aktanten" 37 in Greimas' terminology), or spatio-temporal structures. In his entertaining as well as subtle novella May We Borrow Your Husband?, e. g. Graham Greene directs his readers' attention to the comedy-like, or rather tragicomedy-like, structure of the fable by making

342

Volker Schulz

his narrator use technical terms of literary criticism in phrases like "sad little comedy" (Greene 1967 [1969]: 9); "on the verge between comedy and tragedy" (Greene 1967 [1969]: 30); "the comedy was all the crueller" (Greene 1967 [1969]: 22); or "It was in the end a comedy and not a tragedy, a farce even." (Greene 1967 [1969]: 33) The narrator also explicitly announces the "climax" (Greene 1967 [1969]: 25) of the action proper. By deep structure signals Greene ensures the readers' realizing the complexity of the situations and events he presents: their impact depends on the degree of awareness and on the emotional detachment or involvement of the various characters. What Poopy mistakes as the characteristic comedy structure of serious complications being resolved in a "happy ending", her husband and his two new friends regard as an ingenious farce in which they not only play leading "parts" (Greene 1967 [1969]: 27) but which they have planned and directed from the first to the last minute. For William Harris, however, who is aware of what is going on around him and who has fallen in love with Poopy, her deception, which he is unable to put an end to, is tragic as well as comic. Whereas the author elucidates the ambiguous fable structure by using Aristotelian terms, he points to the structure of actants by occasionally employing words which are very similar to terms that were coined by 20th century structuralist theoreticians like Propp, Souriau and Greimas. Unwillingly, the middle-aged narrator-agent William Harris, "the observer, the subsidiary character" (Greene 1967 [1969]: 17), acts as the "ally" (Greene 1967 [1969]: 15) of the two antagonists, Tony and Stephen, who sense the newly-wed husband's latent homosexuality and successfully try to "take him over" (Greene 1967 [1969]: 15) by seducing (Greene 1967 [1969]: 21) him before his wife's eyes. The young husband, Peter, is the "desired object" (Greene 1967 [1969]: 15) of both the naive protagonist, his wife Poopy, and the two clever homosexuals. William Harris tries in vain to "interfere" (Greene 1967 [1969]: 12) but is reduced to the role of a "comforter" (Greene 1967 [1969]: 38) of the young woman with whom he has fallen in love but whose total lack of perceptiveness prevents her from realizing the narrator-agent's feelings as well as the nature of the "lovely plot" (Greene 1967 [1969]: 38) that the male trio has thought up. Because of her mixture of "innocence" (Greene 1967 [1969]: 35) and "ignorance" (Greene 1967 [1969]: 29), Poopy becomes the tragi-comic "victim" (Greene 1967 [1969]: 37) of her unscrupulous husband and his new friends, who are Poopy's "enemy" (Greene 1967 [1969]: 19). In Souriau's terms, as Scholes translates them, Peter is the "Desired Good"; Tony and Stephen the "Lion", whose "desire ... precipitates the action";

Deep structure

343

Poopy the "rival or Opponent"; and William Tony's and Stephen's as well as Poopy's "Helper". 3 8 In Angus Wilson's highly amusing novella More Friend than Lodger, the narrator points to the beginning of the second of the three motifemes in equally quasi-critical terms: "Now, we come to the most important point in this story: When Rodney Gait became our lodger" (Wilson 1957 [1968]: 63). The end of this second and, as we are correctly told by the narrator, main part of the novella is suggested in a similarly explicit way: "he went out of my life. It was all rather an anti-climax without Rodney ... I did miss the excitement of life with him." (Wilson 1957 [1968]: 81) 39 Note that words like "ally", "desired object", "victim", "most important point", "climax" or "anti-climax" are the very terms which the critic is in the habit of using when he tries to analyze the structure of actants or the fable structure of a work of fiction (or drama). In cases like these, the author, as it were, anticipates the job of the critic.

2.2. Abstract nouns Much more frequent are the cases in which "normal" abstractions help the reader to comprehend the content deep structure of stories and novels. In E. A. Poe's novella The Black Cat, e. g., the narrator-hero starts his narration by indicating not only its atmosphere and tone — he announces a "wild ... narrative" (Poe 1960: 214), "a series of mere household events" (Poe 1960: 214) which to the narrator-hero, however, "have presented little but horror" (Poe 1960: 214) - but also the fable structure, his gradual and painful destruction brought about by a mysterious black cat: "In their consequences, these events have terrified — have tortured — have destroyed me." (Poe 1960: 214) The final situation following the climax of the discovery of the dead body of the narrator-hero's wife, whom he had killed and walled up in the cellar of his house without, however, realizing that he had walled up the black cat as well, is also precisely designated in the introductory paragraph: the process of destruction will end with his death by hanging: "tomorrow I die." (Poe 1960: 214) In narratives written in the third person, it is often the hero or heroine who, in retrospect, puts his or her past experience in its true light, elucidates the particular nature of its various phases by giving them abstract labels. Thus, e.g., in John Wain's first novel Hurry on Down (1953 [I960]), Charles Lumley repeatedly tries to make sense of the

344

Volker Schulz

sequence of periods he has lived through. At one time he comes to the conclusion: "This must be his drifting period; he had finished with his period of intense effort, and his initial phase of revolt was a thing of the past." (Wain 1953 [I960]: 181) (Note the alternation between a normal abstraction, like "period", and a critical term like "phase".) A slightly different but equally fitting interpretation emerges in Charles' final talk with Blearney, the broadcast comedian who persuades him to enter the "entertainment business" (Wain 1953 [I960]: 248). The stages of Charles' persistent refusal "to take sides" (Wain 1953 [I960]: 248) reflected in his stubborn rejection of what everybody else calls a "decent" job, i. e. " 'the sort of thing you were brought up and educated to d o ' " (Wain 1953 [I960]: 174), as well as the ultimate goal of his "quest" (Wain 1953 [I960]: 174), social "neutrality" (Wain 1953 [I960]: 248), fall into place when he has finally found "his niche ... beyond the struggle" (Wain 1953 [I960]: 249): "So far, he had set himself target after target that had proved out of reach: economically, the quest for self-sufficient poverty; socially, for unmolested obscurity; emotionally, first for a grand passion and then for a limited and defined contentment. And now he valued his niche." (Wain 1953 [I960]: 2 4 8 - 2 4 9 ) This consists of a job in the media, i.e. outside the traditional English battlefield of establishment versus working-class. An important point which has to be made in connection with content deep structure in general but which has not found sufficient attention from pioneers of structuralist literary theory like Propp and Greimas is that, due to the differing perspectives of the agents (and, possibly, of the narrator as well), one and the same series of events may be experienced/ interpreted in different ways. The result is a complex or ambiguous fable structure and structure of actants in many narratives. What the narrator in Greene's May We Borrow Your Husband?, e. g., regards as the occasion of Peter's momentous "seduction", Poopy misinterprets as a harmless ' " d a y off alone'" (Greene 1967 [1969]: 38) that has helped her husband to manage the abrupt change from his bachelor's to his married life and Peter himself experiences as the welcome liberation from the painful " ' s t r a i n ' " (Greene 1967 [1969]: 24) caused by Poopy's "'hungry feminity' " (Greene 1967 [1969]: 23) and " 'unwanted attentions'" (Greene 1967 [1969]: 24). Similarly, Stephen and Tony justify their behaviour as " ' a very good action: You've never seen him so detendu\" (Greene 1967 [1969]: 36) They even boast: " 'We are saving him: " (Greene 1967 [1969]: 24) 40 A similar complexity is to be found in Elizabeth Bowen's short story Maria. Here, the divergence of evaluation is even reflected in verbal

Deep structure

345

ambiguity: In the eyes of Maria's aunt, the Doselys "keep" (Bowen 1934: 109) the child in their vicarage during the school holidays in the sense of providing board and lodgings and humane care; Maria, however, feels that she is "kept" there as in a prison; whereas the Doselys, as well as the curate Mr. Hammond, think this '"hell-cat"' (Bowen 1934: 109) of a little girl would deserve to be "kept" in " 'some kind of shed or loosebox at the bottom of the y a r d ' " (Bowen 1934: 109), like a wild animal. Accordingly, the complex fable structure of this short story consists in the failure of a sensible arrangement, in the successful escape of an imprisoned child, and in the futile attempt at taming a dangerous wild animal. That one and the same sequence of happenings/events may allow various complementary interpretations, i. e. that fable structures (as well as structures of actants) are often complex rather than simple is also corroborated by a passage of meta-narration to be found in Wain's novel. It illustrates the third, and least frequent, type of deep structure signals.

2.3. Allegory This consists of allegory reminiscent of the Bunyan model. In John Wain's novel Hurry On Down, e. g., the abstract names which are given to the agents, the settings and the motifs that make up the fictitious world of the allegorical story-within-the-story simultaneously designate the novel's content deep structure (cf. Wain 1953 [I960]: 2 3 3 - 2 3 5 ) . Thus, e.g., the final motifeme of Charles' entering the world of entertainment is visualized by way of the personifications of the "young man (Hopeless)" (Wain 1953 [I960]: 233) and "giant Racket" (Wain 1953 [I960]: 235), his taking up a job as gag-writer for radio programs is "translated" into the act of Hopeless entering the castle, his finally finding "neutrality" into Hopeless lying down to "sleep" (Wain 1953 [I960]: 235). As far as the degree of indirectness of deep structure signals is concerned, allegory can be seen as a step towards type 4, which is perhaps the most widely used type of meta-narration.

2.4. Imagery Imagery is a more indirect method than allegory since the abstract concept is not given, as it is in the case of a personification, but only suggested. The reader has to infere it from the image(s). Often he is helped by the

346

Volker Schulz

author's mixing of abstractions and imagery. Thus, e. g., Doris Lessing in her short story He visualizes the bitterness and aggressiveness of a middle-aged woman who has been left by her husband in the metaphors of her "armour" (Lessing 1957 [1966]: 186) and her resembling a statue ("plaster saint"; "stony" face; [Lessing 1957 [1966]: 188]) and suggests the fable structure of this story by repeating the deep structure signal "appeal" again and again (Lessing 1957 [1966]: 186, 188, 195). In combination with words of similar meaning, like "apology" (Lessing 1957 [1966]: 186), "conciliatory" (Lessing 1957 [1966]: 186), and "implore" (Lessing 1957 [1966]: 188), "appeal" suggests the common denominator of the various events which form the fable of this story: with the help of the wife's neighbour the husband tries to bring about a reconciliation, tries to change his embittered wife's militant attitude towards him, tries to pacify her. Similarly, the military metaphors in Graham Greene's May We Borrow Your Husband? suggest the aggressive nature of Stephen's and Tony's "scheme" (Greene 1967 [1969]: 17) in respect to the unsuspecting young wife. The narrator compares their attempt at "taking over" (Greene 1967 [1969]: 15) her husband with the "siege" (Greene 1957 [1969]: 16) of a "fortress" (Greene 1967 [1969]: 16): they were "closing in", they "dug their trenches" (Greene 1967 [1969]: 16), they "tunneled" (Greene 1967 [1969]: 17) and at last "the enemy are within the citadel." (Greene 1967 [1969]: 19) The extensive use of hunting imagery underlines the narrator's attributing the role of aggressors to the two homosexuals: they are described as a "hungry pair of hunters" (Greene 1967 [1969]: 22) who are patiently "lying in wait" (Greene 1967 [1969]: 13) until they smell their prey, are overcome by hunting fever — "the two flickered their eyes at each other like lizards' tongues" (Greene 1967 [1969]: 18) — and finally catch their prey. However, only one of the two is successful in this hunt: "can people ever hunt quite equally in couples or is there always a loser?" (Greene 1967 [1969]: 36) In Virginia Woolf s short story The Duchess and the Jeweller the image of the "peacock" (Woolf 1944 [1973]: 105) suggests the lady's aristocratic pomp and pride which the rich self-made man Oliver admires and envies although he realizes that the "truffle" (Woolf 1944 [1973]: 102) he has routed by making her and her beautiful daughter's acquaintance is "rotten at the core" (Woolf 1944 [1973]: 108). In spite of his intuitive knowledge that the pearls the duchess wants to sell to him for 20,000 pounds are false and the honour of a notorious gambler only a word he accepts the deal since his sexual desire and his social ambition are equally insatiable.

Deep structure

347

The image of the truffles implies Oliver's role of the greedy "hog" (Woolf 1944 [1973]: 102). Although the nature and function of imagery have been widely discussed in connection with poetry, imagery has been severely neglected by n a r c o l o g i s t s . This is obviously due to the erroneous notion that imagery in fiction is nothing but a stylistic, rhetorical device, a mere ornament. On the contrary, I would argue, it often fulfils the function of metanarration. Graham Greene is right in maintaining that "a simile is a form of reflection." 41

3. Two paradigmatic analyses: Graham Greene's novel Travels with My Aunt and James Joyce's short story The Boarding House The following paradigmatic analyses of two works of fiction are meant as a demonstration of the usefulness of the concept of "deep structure signals" for the interpretation of "non-experimental", "realistic" narratives. In each case the analytical procedure starts off with level III and ends at level II.

3.1. G. Greene: Travels with My Aunt Constellation of agents and settings Let us start by looking at my attempt at elucidating the novel's constellation of agents by way of a diagram. At the beginning of the action proper the agents in the left halves of the two large circles described round the two major characters, the respectable but dull bank clerk Henry Pulling and his disreputable and adventurous but fascinating "aunt" Augusta, are already dead (in the case of Monsieur Dambreuse it remains dubious to the very end of the novel whether he is still alive or not, whereas General Abdul dies in the course of the action proper). However, they definitely belong, as well as the living agents, to the diametrically opposed "worlds" of Aunt Augusta and Henry: 42 her exciting, exotic, "cosmopolitan" world (Greene 1969 [1971]: 12; the page numbers given in brackets in the following analysis refer to this edition), geographically concretized in terms of France, Italy, Turkey, and Latin America and socio-morally concretized in a world of

348

Volker Schulz

Figure 2. Constellation of agents in G. Greene's Travels with My Aunt

restless "extravagance" (73), of "pleasure" (72), of "adventures", with a definitely criminal touch (cf. 63, 69), is contrasted with his philistine world concretized in the small Southern English town of Southwood with its rows of small houses and gardens where everybody is successfully "conditioned" (16) to wanting "everything to be done or said at a suitable time" (11), to leading a "regular" (42), a "simple life" (49), to living "quietly" (9) and in "peace" (26), which means " Ά game of bridge once a week at the Conservative Club. And my garden, of course. My dahlias'" (42). The key-words "travel" and "dahlias" are used throughout the novel in the form of leitmotifs to suggest the contrast of these two worlds and world-views, which may be labeled as "dynamic" vs. "static". All the characters that appear either in the many anecdotes which Aunt Augusta tells her nephew or in the action proper belong either to Henry's narrow bourgeois world or to Aunt Augusta's much wider shady world, whose inner circle is made up of her main past and present lovers (all of whom, by the way, have been "cheats" 43 ). 44 Characteristically, in the course of his travels in the company of his aunt, Henry is only once "seriously tempted to leave her" (96): when they pass through Switzerland, the paragon of bourgeois respectability, order, and peace in Henry's eyes, of "dull" (98) philistinism in the eyes of his adventurous aunt. Switzerland is a sort of blown-up Southwood: "everything seemed clean and arranged and safe, as my life had been."

Deep structure

349

(96) Accordingly, Henry, the "respectable" (82) Englishman, when looking at the idyllic scenery from the train, thinks of his "garden": "I missed my dahlias." (96) Fable The first part of the fable of Travels with My Aunt is largely made up of "travelling" (41), i.e. of a series of journeys in which Henry accompanies his aunt and during which he comes to know her particular "world" by way of first-hand experience as well as via the many tales which Aunt Augusta tells him about certain "periods" (43) of her past. Important aspects of these "anecdotes" (261) about his aunt's colourful "history" (88), due to his naivety and narrow-mindedness as well as due to his aunt's rather cryptic style of narration, Henry only comprehends in parts or in retrospect. Fable structure and spatio-temporal structure On the level of fable structure, this multitude of experiences and information "makes sense" as constituting Henry's gradual initiation, accompanied by a mixture of "fear and exhilaration" (49), into the world of his aunt. 4 5 This global structure is signalled by a number of metaphors fittingly derived in part from the sphere of travelling. Henry is first persuaded by his aunt to enter her world as a "tourist" (249) and, for quite some time, is torn between fascination and exasperation until the fascination prevails, which is suggested by his regretting the fact that his aunt "had a world of her own to which I would never been admitted" (129), and he finally decides not to "catch the boat home" but to apply for a "visa" (168) to "the world of Wordsworth and Curran and Monsieur Dambreuse and Colonel Hakim and the mysterious Mr Visconti" (167 — 168) and resolutely passes "the border" (249). His plan of marrying the daughter of the Paraguayan chief customs officer whose good-will is important in respect of the huge smuggling racket that Henry, Aunt Augusta and Mr Visconti combine to set up at the end of the novel — in a country where smuggling is a sort of "national industry" (207) — confirms Henry's irrevocable passage into his aunt's "crooked world" (179), an amoral world of "romance" ("romantic" [263]), of "adventure" (41), of (anarchic) "life" (159), a "world of the unexpected character and the unforeseen event." (202) 46 Reversely, Henry comes to see his former world as a "prison" from which he has "escaped" or, rather, has been "snatched away" (202). The final revelation that Aunt Augusta is Henry's

350

Volker Schulz

own mother fits into this picture: like his father who hid his hedonistic nature, his being a "hound" (15), behind a fagade of middle class respectability, Henry, too, has been destined, by origin, to "immoral freedom" (203) and, once he has tasted its flavour, is eager to leave his and formerly his father's, "prisonhouse" of "home" (203) and join his mother: "My real mother had certainly not been imprisoned anywhere." (203) Structure of actants Accordingly, the structure of actants is made up of a prisoner who, because he does not know anything else, feels happy and at "home" in his "prison" until he is liberated and, by his saviour, taken into the world of "freedom" (203); after some difficulties of adjustment, he becomes a settler in "this new world" (116). The roles played by Henry and his aunt are the roles of prisoner and saviour, of "tourist" and guide, of a middleaged disciple marked by an extraordinary amount of "innocence" (100) and wondering whether it might not be "too late" for him "to begin" (73) a life dedicated to "pleasure" (72, 73) and a master who, instead of administering abstract lessons, teaches by making the disciple, directly and imaginatively, share her own kinds of "experience" (113) until he becomes "accustomed" (92) to them and realizes how "bored" (256) he had been in his former "uneventful ... life" (255).

3.2. J. Joyce: The Boarding House Constellation of agents The four main agents, i.e. the main characters reduced to those features which are relevant to the fable, live under one roof. There is Mrs. Mooney, a vulgar, but "shrewd" (Joyce 1916 [1968]: 63; the page numbers given in brackets in the following analysis refer to this edition) and "determined" (61) woman who, after obtaining a "separation" (61) from her good-for-nothing husband, has managed to make a living for herself and her two children as landlady of a boarding house, "governing" the house "cunningly and firmly" (62); there is her "lively" (63) and moderately pretty, though equally "vulgar" (66), nineteen-year-old daughter Polly whose girlish "innocence" (64) has already been so corrupted by the spirit of clever calculation determining her mother's behaviour and so confused by the contradictory demands of Christian bourgeois morality, the preservation of virginity, on the one hand, and the finding of a husband who

Deep structure

351

is willing to provide for her, on the other hand, 4 7 that she looks like "a little perverse madonna" (62 — 63) and is alarmingly "wise" (64) and "naughty" (62) for her age; there is Mrs. Mooney's son Jack tending towards "violence" (68) and following his father's example of being "disreputable" (63, 66) — "a hard case" (62) — easily provoked by any "free allusion" (68) to his sister without realizing the ludicrousness of his double standard of morality; finally, there is Bob Doran, a confirmed bachelor ("celibate" [67]) in his mid-thirties, who, after a short period of youthful rebellion ("free-thinking" [66]), has long since largely adapted himself to the prevailing social and moral norms. 48 By his "industry and diligence" (66), he has become quite well-off (he has "money enough to settle down" [66], or, in Mrs. Mooney's cruder style, "a good screw" and "a bit of stuff put by" [65]), and is leading a "regular life" (66) and attending to his "religious duties" (66). Fable The fable consists of the past history of Mrs. Mooney's wrecked marriage and successful new start as landlady of a boarding house and of the action proper of the "affair" (63, 65, 66) between Polly and Mr. Doran. After a period in which Mrs. Mooney, "a butcher's daughter" (61), confines herself to "watching" (63) the pair with "tolerance" she finally "intervenes" (63) in the most "decisive" (65) way possible: "She dealt with moral problems as a cleaver deals with meat" (63). As soon as Polly has admitted what her mother already "suspected" (64), namely that the affair has already progressed far beyond the stage of "casual caresses" (67) to what the young man has experienced as a short-termed "delirium" (67) but what in their society is generally regarded as the scandalous "loss" of a girl's "honour" (65), she sets about to demand the conventional "reparation" (64, 65), marriage, from the conscience-stricken young man. Although the inveterate bachelor's "instinct" (66, 67) urges him "to remain free, not to marry" (66), the combined "force" (68) of his Christian conviction that he has to make up for his "sin" (67), of Polly's pressure who blackmails him with suicide threats ("She would put an end to herself, she said" [66])49, and of his fear of having to face "publicity" (65), "the weight of social opinion" (64) directed against him, and as a consequence, of losing his job ("sit" [65]) in "a great Catholic winemerchant's office" (65) and having to leave Dublin altogether ("run away" [65]) proves to be stronger. He tries to talk himself into assuming that

352

Volker Schulz

"perhaps" Polly and he "could be happy together" (67) while Polly indulges in "a revery" (68), i. e. in vague "hopes and visions of the future" (68). Fable structure This summary of the fable of Joyce's story, however, only accounts for part of the complex chain of events which is presented. By way of a number of deep structure signals the author unmistakably suggests that he is not concerned with a story in which the driving forces solely consist of feelings of sexual attraction, sin, moral indignation, penitence, fear, and hope. Rather, for the "mercenary mind" 5 0 of Mrs. Mooney, whose very name indicates the money aspect, 51 individual feelings as well as social norms are nothing but "cards" (62) in a "game" (68) which she has started and which she is determined to "win" (64, 65). 52 Her trump card is her society's rigid sexual morality which interprets any sexual union between a man and an "innocent" girl as the girl's seduction by the man who, therefore, has to make reparation by marrying her. "Social opinion" is definitely "on her side" (64). The object of the game which Mrs. Mooney plays with a great deal of cunning and hypocrisy is to "get" her daughter " o f f ' her "hands" (65), i.e. to transfer her own (financial) responsibility for her daughter to an appropriate husband. Mr. Doran is appropriate since he is "serious" (65), i. e. a respectable citizen who has internalized his society's moral norms and cannot afford a scandal. Although Mr. Doran has "a notion" that he is "being had" (66), i.e. that he has been seduced by Polly rather than having seduced her and that he has thus been "trapped into marriage", 53 he is therefore unable to hinder Mrs. Mooney from winning the game but puts a good face upon it. Structure of actants In this game which has nothing playful about it but rather resembles a business transaction — tellingly, none of Mrs. Mooney's other lodgers "meant business" (63)54 — Mrs. Mooney has the initiative 55 and is, at the same time, one of the two persons to profit from it. As the information about her past life which we are given at the beginning of the story clearly shows "Mrs. Mooney is accustomed to dealing with men." 5 6 Throughout, she is completely in control. Alternately, she slips into the roles of unsuspecting or alarmed or "outraged mother" (64), 57 whereas in actual fact she has not only "connived" (64) at a crucial stage of her daughter's

Deep structure

353

affair but has planned it from the very beginning 58 and brings it to a successful end. She is an "imposing" (62) woman in more than one sense, commanding as well as deceitful. 59 Mr. Doran realizes only dimly, and too late, that Mrs. Mooney and her compliant daughter 6 0 have tricked him into the disagreeable role of a rake who, pretending to be a "man of honour" (64) but shamefully "abusing the hospitality" (64) granted to him, has "taken advantage of Polly's youth and inexperience" (64), i.e. has seduced an innocent girl. The rigidity of the social norms existing in his world allows him no escape. Mr. Doran is a "helpless" (66, 67) victim, the loser in a dirty game. Mrs. Mooney is helped by the latent threat 61 emanating from her son's well-known aggressiveness but above all by her daughter's intuitively acting according to her secret plans (Polly "had divined the intention behind her mother's tolerance" [64]). Polly has a triple role: Firstly, she is her mother's silent accomplice ("no open complicity" [63]) in the undertaking of securing a respectable and well-off husband for her, 62 using, like her, the weapon of blackmail if need be; secondly, she is the second person to profit from the game since in her society the status of a married woman is considerably higher than that of a girl; 63 and, thirdly, she is the object of the transaction, the commodity changing hands in the course of the action. By repeatedly drawing the reader's attention to the fact that Mrs. Mooney is being called and thought of as "The Madam" (62, 68), a term which strongly connotes a procuress, 64 by the resident young men and by presenting Polly as a singer of risque songs in a boarding house that "was beginning to get a certain fame" (66), Joyce even suggests the similarity between the way in which petty bourgeois marriages in Dublin are brought about and the prostitution trade, boarding house and brothel, Mrs. Mooney's and Mrs. Warren's profession. 65

Notes 1. Ingarden (1962: 24), quoted (and translated) by Fizer (1973: 33, n. 4). 2. The German equivalents in Gohlke's translation are: "Handlung"; "Handelnde"; "Denkweise und Gesinnung"; "Lied- und Sprachgestaltung"; "Ausstattung" (cf. Aristoteles [1959: 64]). 3. Cf. Ingarden's (1931 [1965]: XI) observation "daß man das literarische Werk immer fur ein einschichtiges Gebilde hielt, während es tatsächlich aus mehreren heterogenen Schichten besteht."

354

Volker Schulz

4. It is true that both Kayser and Wellek/Warren applaud Ingarden's concept of the network of strata, i. e. of the "structure", of the literary work. However, instead of actually making use of, and further developing, his insights they merely pay lip-service to him (cf. Kayser (1948: 17, 391)) or cut his model down to a mere framework for the accomodation of traditional categories like "meter", "rhythm", "style", "metaphor" or "symbol" in which they are primarily interested (cf. Wellek/Warren (1942 [1956]: 139 f.)). Accordingly, it is more than accidental that Wellek/Warren occasionally use the words "structure" and "organism" as synonyms (cf. Wellek — Warren (1942 [1956]: 203)). 5. Wellek/Warren (1942 [1956]: 16). 6. Wellek/Warren (1942 [1956]: 129). 7. Cf. Wellek/Warren (1942 [1956]: 206 ff.). 8. Kayser (1948: 54). 9. Kayser (1948: 82). 10. Ingarden (1931 [1965]: XI) 11. Cf. Ingarden (1931 [1965]: XII, XIX). 12. Crane (1957: v). 13. Cf. Olson (1961 [1966]), Olson (1968). 14. Olson (1961 [1966]: 10). 15. The word is (misleadingly) translated as "Stücke" by Gohlke (Aristoteles (1959: 64)). 16. Olson (1961 [1966]: 10). 17. Cf. Iser (1970: 36, n. 6), Iser (1972: 69, 216, 2 3 8 - 2 3 9 , 2 9 7 - 2 9 8 , 326, 340, 343, 371), Iser (1976: 10, 38, 87, 1 0 2 - 1 0 3 , 105, 1 6 7 - 1 6 8 , 180, 1 8 3 - 1 8 6 , 234, 2 6 7 - 2 8 4 ) . Iser himself, however, concentrated on those features of Ingarden's theory which were relevant for his own project of establishing reception aesthetics as a new literary critical method, e.g., the concepts of "Konkretisation" and "Unbestimmtheitsstellen". 18. Cf. Erlich (1955 [1964]: 208). 19. Cf. Propp (1928 [1972]: passim). 20. Cf. Erlich (1955 [1964]), Striedter (1969 [1971]). 21. Cf. Bremond (1964 [1972]); Bremond (1966); Greimas (1966 [1971]); Todorov (1966 [1972]); Todorov (1971 [1972]); Barthes (1971). 22. Lotman (1972: 22). 23. Cf. Bradbury (1966); Chatman (1969); Chatman (1978); Culler (1975); Dolezel (1971); Dolezel (1972); Fietz (1976); Fowler (1977); Genette (1972 [1980]); Haubrichs (1976); Holloway (1979); Lodge (1981); Lotman (1970 [1973]); Pfister (1977); Prince (1973); Prince (1982); Ruthrof (1981); Sceglov/Zolkovskij (1976). 24. For this, the reader is referred to Schulz (1987a: 3 - 2 0 ) and Schulz (1989). 25. Cf. Fowler (1977: 129): "The reader's encounter with the text consists basically of the release of knowledge through response to the patterns which the author has deployed in encoding the deep structure of the work. It is an act of discovery". Representatives of semiotics like Umberto Eco arrive at a similar conclusion although their startingpoint is different: "Ein Text ist ein syntaktisch-semantisch-pragmatisches Kunstwerk, an dessen generativer Planung die vorgesehene Interpretation bereits teilhat" (Eco 1979 [1987]: 83). 26. Lotman (1970 (1973]: 27). 27. Dolezel (1972: 58). 28. Fietz (1976: 44).

Deep structure

355

29. Pfister (1977: 97). Perhaps RuthroFs (1981: 49) "work ideology" is an even more felicitous term. 30. Cf. Barthes (1966 [1974/75]: 248). 31. Chatman (1978: 19). 32. Cf. Eco's (1979 [1987]: 114) observation that a text may point to its "Topic", i.e. its content deep structure and ultimate meaning, "explizit in Form von Topic-Kennzeichen, Titeln, Untertiteln oder Schlüsselausdrücken." 33. Chatman (1969: 18). 34. Although Eco dislikes the idea of a linear course of the (idealized) analytical procedure he, too, stresses the fact, "daß in dem konkreten Ablauf der Interpretation alle Ebenen und Unterebenen — in der Tat reine metatextuelle 'Schubfächer' — auch durch größere 'Sprünge' erreicht werden können" (1979 [1987]: 84). 35. Cf. Hendricks (1973: 191): "'meta-narration' ... may provide valuable clues about the organization and thematic significance of the story." 36. Cf. Dolezel (1972). 37. Cf. Greimas (1966 [1971]: 107 ff.). 38. Scholes (1974: 52). In Greimas' (1966 [1971]) terms, Peter is the "Object" (118), William Harris the "Adjuvant" and Tony and Stephen the "Opponent" (142). 39. For a full interpretation of this novella cf. Schulz (1987b). 40. For a full interpretation of May we Borrow Your Husband? cf. Schulz (1987a: 245 — 261). 41. Greene (1971 [1972]: 145). 42. Cf. Monod (1970: 454): "deux univers opposes". 43. Chaudhury (1972: 82). 44. The only exception seems to be Henry's father who, however, obviously led a double life: in Henry's memory he figures as " ' a very sleepy m a n ' " whereas Aunt Augusta knows from her own experience that he " 'needed bedrooms for more than sleep'." (11) 45. Cf. Chaudhury (1972: 79): Travels with My Aunt "is a study of Henry Pulling's initiation into a new view of life." 46. Cf. Chaudhury (1972: 81): Travels with My Aunt depicts "Henry Pulling's transformation from a toff into a tart." 47. Cf. Schneider (1980: 103). 48. Ganzmann (1986: 97) rightly stresses his "Neigung, seine Entscheidungen in Abhängigkeit von den Erwartungen seiner Umwelt zu treffen". 49. The author's change from direct to indirect speech with this very sentence subtly indicates his warning the reader not to take Polly's utterance at face value. Therefore, Kenner's (1965: X) thesis that "her passivity balances her mother's purposefulness" has obviously to be qualified. 50. Garrett (1968: 14). 51. Cf. Schneider (1980: 103). 52. Cf. Kenner (1965: IX): "it is with delicate, inarticulate feelings that Mrs. Mooney is trafficking so blandly." Ganzmann's (1986: 81) thesis that Mrs. Mooney works herself up into moral indignation ("moralische Entrüstung") is completely unwarranted. 53. Garrett (1969: 4). 54. Cf. Schneider (1980: 102): "Prägnanter könnte die Verflechtung von Heirat und Geschäft kaum zum Ausdruck kommen," 55. The fact that the story starts by providing information about her past history confirms the leading part she plays in the fable as a whole. 56. Kenner (1965: XIII).

356

Volker Schulz

57. The very phrasing suggests that Mrs. Mooney does not try to define her feelings but her social role: "she had all the weight of social opinion on her side: she was an outraged mother." (64) 58. Cf. Ganzmann (1986: 88). 59. Cf. Schneider (1980: 103). 60. According to Schneider (1980: 103), Polly's name, "der im Englischen ein Synonym fiir einen Papagei ist, suggeriert, daß sie gedankenlos nachplappert, was ihre Mutter ihr vorsagt." 61. Cf. Ganzmann (1986: 87). 62. Cf. Schneider (1980: 103): "das Zusammenspiel zwischen Mutter und Tochter ist so perfekt, daß es keiner Absprache zwischen ihnen bedarf." 63. Cf. e. g. the heroine's reflections in Joyce's Eveline: "Then she would be married — she, Eveline. People would treat her with respect then." (Joyce (1916 [1969]: 37) 64. Cf. Scholes/Litz (1969: 477). 65. Cf. Schneider (1980: 101): "In diesem Milieu wird auch Liebe zu einer Ware degradiert, die es möglichst teuer zu verkaufen gilt."

References I Texts Bowen, Elizabeth 1934 Maria, in: The Cat Jumps and Other Stories. London: Jonathan Cape. Greene, Graham [1969] 1967 May We Borrow Your Husband?, in: May We Borrow Your Husband? And Other Comedies of the Sexual Life. Harmondsworth: Penguin. Greene, Graham [1971] 1969 Travels with My Aunt: Harmondsworth: Penguin. Joyce, James [1968] 1916 The Boarding House, in: Dubliners. New York: The Viking Press. Lessing, Doris [1966] 1957 He, in: The Habit of Loving. London: Granada. Poe, Edgar Allan 1960 The Black Cat, in: Tales of Mystery and Imagination. London: Pan Books. Wain, John [1960] 1953 Hurry On Down. Harmondsworth: Penguin. Wilson, Angus [1968] 1957 More Friend than Lodger, in: A Bit Off the Map. Harmondsworth: Penguin. Woolf, Virginia [1973] 1944 The Duchess and the Jeweller, in: A Haunted House and Other Stories. Harmondsworth: Penguin.

Deep structure

357

II Secondary Literature 1. Literary Theory Aristoteles 1959

"Poetik", in: Aristoteles, Die Lehrschriften, hg., übertr. und in ihrer Entstehung erläutert v. Paul Gohlke. Paderborn: Ferdinand Schöningh.

Barthes, Roland 1966 [1974/75] "An introduction to the structural analysis of narrative", New Literary History 6: 2 3 7 - 2 7 2 . Bradbury, Malcolm 1967 "Towards a poetics of fiction. 1) An approach through structure", Novel 1:45-52. Bremond, Claude 1964 [1972] "Die Erzählnachricht", in: Jens Ihwe (ed.), Literaturwissenschaft und Linguistik. Ergebnisse und Perspektiven. Bd. 3. 177 — 217. Frankfurt: Athenäum. 1966 "La logique des possibles narratifs", Communications 8: 60 — 76. Chatman, Seymour 1969 "New ways of analyzing narrative structure, with an example from Joyce's Dubliners", Language and Style 2: 3 — 36. 1978 Story and discourse. Narrative structure in fiction and film. Ithaca: Cornell University Press. Crane, Ronald S. 1957 "Preface", in: Ronald S. Crane (ed.), Critics and criticism (abridged edition). iii —vi. Chicago: The University of Chicago Press. Culler, Jonathan 1975 "Defining narrative units", in: Roger Fowler (ed.), Style and structure in literature: Essays in the new stylistics. 123 —142. Ithaca: Cornell University Press. Dolezel, Lubomir 1971 "Toward a structural theory of content in prose fiction", in: Seymour Chatman (ed.), Literary style: A symposium. 95 — 110. London: Oxford University Press. 1972 "From motifemes to motifs", Poetics 4: 55 — 90. Eco, Umberto 1979 [1987] Lektor in Fabula. Die Mitarbeit der Interpretation in erzählenden Texten. (translated into German by Heinz-Georg Held.) München: Carl Hanser. Erlich, Victor 1955 [1964] Russischer Formalismus. München: Carl Hanser. Fietz, Lothar 1976 Funktionaler Strukturalismus. Grundlegung eines Modells zur Beschreibung von Text und Textfunktion. Tübingen: Max Niemeyer.

358

Volker Schulz

Fizer, John 1973

"The concept of strata and phases in Roman Ingarden's theory of literary structure", in: Joseph Strelka (ed.), The personality of the critic. 10 — 39. University Park.

Fowler, Roger 1977 Linguistics and the novel. London: Methuen. Genette, Gerard [1980] 1972 Narrative discourse. An essay in method. Ithaca: Cornell University Press. Greene, Graham [1972] 1971 A sort of life. Harmondsworth: Penguin. Greimas, Algirdas Julien [1971] 1966 Strukturale Semantik. Methodologische Untersuchungen. Braunschweig: Westermann. Haubrichs, Wolfgang 1976 "Einleitung: Für ein Zwei-Phasen-Modell der Erzählanalyse", in: Wolfgang Haubrichs (ed.), Erzählforschung 1. Theorien, Modelle und Methoden der Narrativik. 7—28. Göttingen: Vandenhoeck and Ruprecht. Hendricks, William O. 1973 "Methodology of narrative structural analysis", in: Essays on semiolinguistics and verbal art. 175—195. The Hague: Mouton. Holloway, John 1979 Narrative and structure. Exploratory essays. Cambridge: Cambridge University Press. Ingarden, Roman 1931 [1965] Das literarische Kunstwerk. Tübingen: Max Niemeyer. Ingarden, Roman 1962 Issledovaniia po estetike. Moscow. Iser, Wolfgang 1970 Die Appellstruktur der Texte. Unbestimmtheit als Wirkungsbedingung literarischer Prosa. Konstanz: Universitätsverlag. 1972 Der implizite Leser. Kommunikationsformen des Romans von Bunyan bis Beckett. München: Wilhelm Fink. 1976 Der Akt des Lesens. Theorie ästhetischer Wirkung. München: Wilhelm Fink. Kayser, Wolfgang 1048 Das sprachliche Kunstwerk. Eine Einfuhrung in die Literaturwissenschaft. Bern: Francke. Lodge, David 1981 Working with structuralism. Essays and reviews on nineteenth- and twentieth-century literature. Boston: Routledge. Lotman, Jurij M. [1973] 1970 Die Struktur des künstlerischen Textes. Frankfurt/M.: Suhrkamp. Olson, Elder [1966] 1961 Tragedy and the theory of drama. Detroit: Wayne State University Press. 1968 The theory of comedy. Bloomington: Indiana University Press. Pfister, Manfred 1977 Das Drama. Theorie und Analyse. München: Wilhelm Fink.

Deep structure Prince, Gerald 1973 1982 Propp, Vladimir [1972] 1928 Ruthrof, Horst 1981

359

A grammar of stories. The Hague: Mouton. Narratology. The form and functioning of narrative. The Hague: Mouton. J. Morphologie des Märchens, ed. Karl Eimermacher. München: Hanser.

The reader's construction of narrative. London: Routledge and Kegan Paul. Sceglov/Zolkovskij 1976 "Poetics as a theory of expressiveness: Towards a 'theme — expressiveness devices — text' model of literary structure", Poetics 5: 207 — 245. Schulz, Volker 1989 "Structuralist narrative theory and methodology: W. O. Hendricks' analysis of Ambrose Bierce's short story Oil of Dog' as testcase", Fu Jen Studies 22: 1 - 3 2 . Souriau, E. 1950 Les deux cent milles situations dramatiques. Paris. Striedter, Jurij (ed.) [1971] 1969 Russischer Formalismus. Texte zur allgemeinen Literaturtheorie und zur Theorie der Prosa. München: Wilhelm Fink. Todorov, Tzvetan [1972] 1966 "Die strukcturelle Analyse der Erzählung", in: Jens Ihwe (ed.), Literaturwissenschaft und Linguistik. Ergebnisse und Perspektiven, Bd. 3. 265 — 275. Frankfurt/M.: Athenäum. [1972] 1971 Poetik der Prosa. Frankfurt/M.: Athenäum. Wellek, Rene/Warren, Austin [1956] 1942 Theory of literature. New York: Harcourt, Brace and World.

2. Studies of individual works Chaudhury, Μ. Κ. 1972 "Graham Greene's 'Travels with My Aunt': A picaresque novel", Research Panjab University Bulletin (Arts) 3/2: 7 9 - 8 5 . Ganzmann, Jochen 1986 Vorbereitung der Moderne. Aspekte erzählerischer Gestaltung in den Kurzgeschichten von James Joyce and Kather ine Mansfield. Frankfurt/M.: Peter Lang. Garrett, Peter K. 1968 "Introduction", in: Peter K. Garrett (ed.), Twentieth century interpretations of 'Dubliners'. A collection of critical e s j i y j . l — 17. Englewood Cliffs, NJ: Prentice Hall. Kenner, Hugh 1965 Studies in change. A book of the short story. Englewood Cliffs, NJ: Prentice Hall. Monod, Sylvere 1970 "Graham Greene: 'Travels With My A u n t ' " (review), Etudes Anglaises 23/4: 4 5 3 - 4 5 4 .

360

Volker Schulz

Schneider, Ulrich 1980 "Eine irische Heirat: James Joyce's 'The Boarding House'", Englisch Amerikanische Studien 2: 99 — 105. Scholes, Robert/Litz, A. Walton (ed.) 1969 Joyce, 'Dubliners'. Text, criticism, and notes. New York: Viking Critical Library. Schulz, Volker 1987 a Das kurzepische Werk Graham Greenes. Gesamtdarstellung und Einzelinterpretationen. Trier: Wissenschaftlicher Verlag Trier. 1987 b "Angus Wilson's novella 'More Friend than Lodger': A structuralistgenerative interpretation", Fu Jen Studies 20: 29 — 53.

The pragmatics of poetic discourse Janet Byron Anderson

To speak of a pragmatics of poetic discourse is to imply that poetry may contain interactive elements which, if identified and understood, can enhance our delight in poetic texts. This is the claim of this article. 1 However pragmatics is defined by its practitioners — and definition of the field and its basic concepts is problematical in some respects (see Levinson 1983: 5 — 35) — there is agreement that pragmatics is concerned with aspects of the use of utterances between interlocutors, in specific speech situations (Leech 1983: 15). Among the elements of communication that fall under the pragmatic rubric are conversational structure, conversational implicature, presuppositions, deixis, and speech acts. Of these, my discussion will focus chiefly upon the last three. Analyzing Sonnet I, 3 of Die Sonette an Orpheus by Rainer Maria Rilke (1875 — 1926), I will show that a pragmatic approach to poetic discourse draws a reader into exciting depths of poetic strategy that would be difficult to reach or appreciate otherwise. 2 Such an approach must, however, take into account the unmarked rhetorical features of a poem; for these constitute the predictable literary structure within which the poet creates unexpected poetic meaning. Sonnet I, 3 is constructed as an Italian sonnet. This type of sonnet has a bipartite structure consisting of an octave and a sestet. From the pragmatic viewpoint, it is this stanzaic structure, rather than the poem's meter or rhyme, that is most significant. (See Brodsky 1988: 156 — 157 on the structural variety among Rilke's sonnets.) Although the poetic boundary between the octave and sestet is not always solid, the two divisions have traditionally functioned jointly to encode a general theme. For example, the octave might pose a problem, raise a question, or enunciate a proposition; while the sestet might refine what was introduced in the octave by, for instance, presenting a resolution of the problem, an answer to the question, or a telling comment upon the proposition. (On the properties of the Italian sonnet and its distinction from other sonnet forms, see H o l m a n - H a r m o n 1986: 265 and 4 7 6 - 4 7 7 ; and Preminger 1974: 781-784.) Sonnet I, 3 lays out, by means of contrast, two bases of poetic creativity — the spiritual and the corporeal; and it indicates unequivocally that

362

Janet Byron

Anderson

creativity which is spirit-based is superior to that which is rooted in physical desire: Ein Gott vermags. Wie aber, sag mir, soll ein Mann ihm folgen durch die schmale Leier? Sein Sinn ist Zwiespalt. An der Kreuzung zweier Herzwege steht kein Tempel für Apoll. Gesang, wie du ihn lehrst, ist nicht Begehr, nicht Werbung um ein endlich noch Erreichtes; Gesang ist Dasein. Für den Gott ein Leichtes. Wann aber sind wir? Und wann wendet er an unser Sein die Erde und die Sterne? Dies ists nicht, Jüngling, daß du liebst, wenn auch die Stimme dann den Mund dir aufstößt, — lerne vergessen, daß du aufsangst. Das verrinnt. In Wahrheit singen, ist ein andrer Hauch. Ein Hauch um nichts. Ein Wehn im Gott. Ein Wind. [(1) A god can do it. But, tell me, how is (2) a man to follow him through the narrow lyre? (3) His (the man's) mind/understanding is conflict/ disunion. At the crossroad of two (4) heartways stands no temple for Apollo. (5) Song, as you teach it, is not craving, (6) nor is it petitioning for/the pursuit of something which is to be reached only in the end; (7) Song is existence. An easy thing for the god. (8) But when are we to be/ exist? And when will he (the god) spend (9) the earth and stars upon our being? (10) Youth, your loving, — this is not it; even if (11) then your voice jolts your mouth open, — learn (12) to forget that you sang out impetuously. That will take flight/run out. (13) Authentic singing is another breath. (14) A breath for nothing. A waft in the god. A wind.] 3 For purposes of pragmatic analysis, I begin by viewing the speech situation of the poem as self-contained. Within the fictive world of this particular sonnet, we find three beings: a speaker ich, who articulates a certain relationship to two different receivers, each addressed as du. Since each du resides within the microcosmic world of the poem, the term micro-addressee would be a more accurate description of the two receivers. 4 From beginning to end, the poem is infused with a conative spirit; that is, the speaker's orientation is towards one or the other addressee: the speaker variously acknowledges the addressee's presence in face-toface discourse, orders him to do something or to forgo doing it, or

Poetic discourse

363

prescribes what the addressee should think. In most conative utterances, then, the speaker is intimating a desire for a specific action or change on the part of the addressee. Hence conative utterances bear an emotive load as well, for they suggest that the speaker has a stake in a given issue or outcome. 5 The identity of each du in I, 3, the distinction between them, and finally the precise conative ties linking the speaker to each of his addressees — all these interactive elements are borne stanzaically, through the Sonnet's bipartite division: the du of the octave is distinct from the du of the sestet. The octave's du is a god, the inspirer or teacher of song — in short, Apollo, as muse. In relation to the god the speaker is a student-poet, whom the god has illumined to distinguish spurious from genuine poetic creation. While the former is rooted in the corporeal, the latter has its origin in spiritual impulsion. But to tap this source of authentic creativity is difficult, for the endeavor compels the poet to transcend the ephemeral and to overcome the fragmentation that typifies human thinking. We gather these messages from several presuppositions within the second quatrain. For example, the lines Gesang, wie du ihn lehrst, ist nicht Begehr,j nicht Werbung um ein endlich noch Erreichtes (lines 5 — 6) presuppose several things. The verb lehrst '(you) teach' of the embedded clause wie du ihn lehrst 'as you teach it' presupposes the obvious fact that du is a teacher. But inasmuch as teaching is semantically a member of the converse pair teaching/learning, or, agentively, teacher/learner, we accordingly infer from this embedded clause that the speaker ich stands in a learner relationship to du. This relationship is depicted with convincing explicitness in the main clause of lines 5 and 6, where we discover precisely what the speaker has learned: it is that Gesang ... ist nicht Begehr,/nicht Werbung um ein endlich noch Erreichtes. In other words, what has been taught is not merely 'song' (Gesang) in its broadest poetic sense; but more significantly the existence of two antithetical views about the basis of song, or poetic creativity. Furthermore, one of these views is false, and is to be rejected: the other is true, and to be embraced. The nature of the former (the false) teaching is that poetic creativity is founded upon human craving. The specific identification of this craving as human love, with its potential for inner conflict, is foreshadowed as early as lines 3 and 4 of the first quatrain: 'At the crossroad of two heartways stands no temple for Apollo.' One coherent theme of the octave is now clearer: the idea of undesirable fragmentation pervades it, although the linguistic expression of this idea grows subtler as the octave develops. But the entire body of thought wherein human affectional craving and human

364

Janet Byron

Anderson

fragmentation are believed to constitute the basis of poetic creativity is rejected. Its dismissal is signalled strongly in the negative adverb nicht, which occurs twice in the main clause of lines 5 and 6: ... nicht Begehr,/ nicht Werbung ... The nature of authentic song is finally enunciated in line 7: Gesang ist Dasein, 'Song is existence'. (The two references to Hauch 'breath' in the sestet reiterate this view, for breath is associated with life; i.e., 'sung existence', in Rilke's thinking.) In other words, the genuine poet creates not through craving or loving, but through being. But this is difficult to do. It appears that only a god can achieve it! Now we can begin to understand the referent of the abbreviated es 'it' in Ein Gott vermags, Ά god can do it' (line 1). This reference is of course cataphoric, for the referent of es follows rather than precedes the pronoun. The clue to the pronoun's referent occurs in the infinitive folgen 'follow' in the question, Wie aber, sag mir, solllein Mann ihm folgen durch die schmale Leier?, 'But, tell me, how is a man to follow him through the narrow lyre?' (lines 1 — 2). Like lehren (earlier commented upon), the verb folgen is a member of a converse pair; for 'following' presupposes that someone is leading, or, agentively, that the speech situation includes a leader-follower relationship. The embedded clause, ihm folgen durch die schmale Leier, accordingly presupposes that a god can go; i.e., lead, into the narrow lyre (a metaphor that appears to mean to create deeply, from within; i.e., from one's being). The referent of es is then precisely this presupposition: the "it" that a god can do is lead into the narrow lyre. The speaker takes this presupposition for granted. However, through abruptly — though obliquely, by means of a presupposition — introducing this idea, while suspending any pre-established contextualizing or clarifying details, the speaker compels, rather than invites, his audience to entertain the same presupposition: hence the peremptory tone of the sonnet's onset, Ein Gott vermags. But more is involved in the cataphoric strategy of this initial line than solely the conveyance of a tone of certainty. This type of syntax produces two other effects. First, by occurring at the earliest point in the sonnet, the cataphoric line arouses the curiosity of the reader; for one is impelled to wonder what it is that a god can do. One's curiosity is also heightened because the construction intimates first, that in progress is a dyadic dialogue which the reader has unexpectedly found himself overhearing; and second, that within this dialogic situation someone, besides the speaker, also knows what it is the god can do. This turns out to be true; for the initial utterance foreshadows the presence within the poem of an addressee. In fact, as we have seen, two addressees appear: the god and

Poetic discourse

365

later the Jüngling. Ironically, as will become clear, the only one of these who knows what the god can do is the god himself; i.e., the very god mentioned as a third person referent in Ein Gott (line 1), ihm (line 2), and allusively in Apoll (line 4). The syntactic strategy in force here entails the first of two poetic moves wherein Rilke shifts the speaker's position, vis-ä-vis the god, between second person and third person reference. The god's presence hence becomes overriding throughout the poem because of the pragmatic liberty Rilke is exercising here. (I will return to this point momentarily.) 6 The second effect of the initial cataphora is that by drawing the reader into a seemingly narrative and dialogic situation, the poet creates an appearance of solidarity between the poem's persona, on the one hand, and the reader, on the other. Through this means, the reader becomes circuitously implicated in a common experience with the poet/speaker, and thus impelled to share his assumptions and presuppositions. Cataphoric beginnings of the type found in Sonnet I, 3 are common as openings in short stories and other narratives, where they produce similar effects — such as the establishment of tone, the arousal of the reader's curiosity, and the appearance of shared experience (see Halliday —Hasan 1976: 298 on cataphora in narrative fiction). To appreciate the narrative-like qualities and effects of the opening of I, 3, one need only see and read the first quatrain with its lineal poetic boundaries removed: Ein Gott vermags. Wie aber, sag mir, soll ein M a n n ihm folgen durch die schmale Leier? Sein Sinn ist Zwiespalt. A n der Kreuzung zweier Herzwege steht kein Tempel für Apoll.

Having introduced his central presupposition in the leading utterance of this quatrain (i. e., in Ein Gott vermags), the speaker thereupon adopts the role of inquirer, appearing (in lines 1—2) to question a matter that depends upon this presupposition; and he indirectly invites the reader to question the same matter: How, asks the speaker, can a man follow the god through the narrow lyre? That is, how can a man create deeply, from within his being? How can he be "song"? The interrogative utterance beginning Wie ...?, together with its parenthetical directive sag mir, 'tell me', predisposes a reading according to which the speaker is truly requesting information from an addressee. However, this utterance is only thinly interrogative and conative (i. e., addressee-oriented). Its force is more clearly emotive, suggesting the speaker's exasperation, born of personal trials, over his own endeavor to follow the god. 7 For indeed, as

366

Janet Byron

Anderson

lines 3 — 6 of the octave reveal, the speaker is knowledgeable about the very situation which his question appears to be seeking clarification of: lines 3 — 6 articulate the obstacles to a poet-aspirant's following the god; and the sestet will address the issue more explicitly. So the interrogative beginning Wie ...? (lines 1—2) thus impels a declarative reading: A man can follow the god through the narrow lyre, but it is exceedingly difficult for him to do so. Thus besides being conative and emotive, then, the interrogative locution is also rhetorical; because its answer is obvious. Obvious to the speaker, that is. For the details contained in lines 3 — 4 (directly following the question), which appear to be autonomous statements of fact (in speech act terms, representatives), are actually verdictive in intent: they convey the speaker's prompt assessment of the dilemmas the poet must overcome; and as such, lines 3 — 4 imply an answer to the question in lines 1—2. The only way a poet can overcome the dilemmas is to stay behind the god, figuratively speaking. This image is reinforced stanzaically; for the two key verbs earlier indicated (folgen and lehren) occur in both quatrains — folgen in the first, and lehren in the second. As noted, both verbs presuppose an identical converse relationship. But as is now clear, each quatrain foregrounds one respective half of it. The first foregrounds the poet as follower/learner (who is also, grammatically, the subject offolgen), by making explicit reference to his untutored state of mind: its diffuseness, etc. The second quatrain, in contrast, foregrounds the god as leader/ teacher (who is also, grammatically, the subject of lehren), by underscoring the unity of his being and his creating. The conclusion of the octave flows into the sestet, where the speaker now addresses a Jüngling, imaging an aspirant poet who erroneously believes that inspiration derives from the physical exaltation of human love. The speaker's tone now becomes didactic and hortatory: the Jüngling is directed to observe that the merely affectional basis of 'song' is insubstantial. The speaker's negative judgment of this Jüngling-thinking is conveyed in his words, Das verrinnt, 'That will run out' (line 12). The referent of das is teasingly ambiguous. What will run out? Either the specious poetic power itself, or the actual productions deriving from it (presupposed in the clause, daß du auf sangst), or both. Thus in contrast to the octave, wherein the speaker enacts the role of student of the god, in the sestet the speaker emerges as himself a teacher, an authority, visä-vis the Jüngling. This posture of authority is effected through the same syntactic strategy as was employed in the first quatrain of the octave; namely, through cataphoric reference. The dies (the leading word of line

Poetic discourse

367

10) that the Jüngling is warned to disavow is designated only after the demonstrative, specifically in the two embedded clauses which follow it; i. e., in ... daß du liebst, wenn auch/die Stimme dann den Mund dir aufstößt (lines 10 — 11). Once again, as in the octave, through suspension of contextualizing details, the speaker compels the addressee (here the Jüngling), as well as the reader, to agree with beliefs whose validity is presented as a given. A key verb in the sestet — a verb whose meaning and modality foreground the "teacher" role of the speaker in this concluding part of the sonnet — is the directive lerne (line 11), although here the meaning of the verb is probably broader than simply "learn" in its basic literal sense. Nevertheless, in light of the type of converse relationship that predominates both in the octave and sestet, the occurrence of lerne here cannot be altogether incidental. Because of the dual roles which the speaker enacts in the octave and in the sestet respectively, the identity of ich now arises as an important issue. In relation to the Gott, he is a student; while in relation to the Jüngling, he appears to be himself an authority, almost an intermediary between the god and other poets. However, as the octave moves more closely towards its conclusion, we discover that the relationship between the speaker and the two other referents is more complex; and within this complexity we find more information about ich's identity. Beginning with line 5, we notice that the god gradually ceases to be addressed as second-person du; instead, he reverts to the third-person god introduced in the first quatrain. (The du of line 5 is accordingly pivotal to two shifts in personal deixis: it simultaneously ends one of them and initiates the other, which is a reversal of the first.) The reversal moves from du 'you' (line 5), to den Gott 'the god' (line 7), and finally er 'he' (line 8). This latter shift in reference suggests a subtle distancing between the speaker and the god. This distancing occurs simultaneously with the speaker's incorporation into a 'we' (note wir, line 8; and unser Sein, line 9), that excludes the god (the god's exclusion being foregrounded through placement of a strongly emphasized er as the last word of line 8), but that appears to include the speaker along with, cataphorically, the Jüngling who emerges explicitly in line 10. Starting with line 8, then, the dyadic teacher-student roles between the speaker and the Jüngling, respectively, now appear paradoxically to have merged. For inasmuch as this wir stands collectively in need of a common thing from the god, the speaker/learner implied in lines 1 — 7 no longer appears to enjoy a superordinate position. The lines at this point (beginning with line 8) now

368

Janet Byron

Anderson

invite us to view the Jüngling not as representing a separate being, but rather as representing an aspect of the speaker himself; in particular, that aspect of his being that remains as yet weakly taught, or as only barely touched, by the god; and that is waiting for the god to 'spend' upon it 'the earth and stars' (lines 8 — 9). The apparent merger of the roles, then, suggests that the speaker is himself ambivalent about where he thinks poetic power should reside. That he is not mildly, but on the contrary exceedingly, conflicted about this is reflected in the irony of his expecting the god to bestow upon the human spirit the earthy along with the celestial! Nevertheless, that the god must bestow something (lines 8 — 9) is an important note in the pragmatic harmony of this sonnet: these lines mirror the same converse relationship that has hitherto predominated in the poem; for the kernel of the predication, namely wenden, implies a giver and a receiver. But these lines develop yet another aspect of the converse relationship; specifically that until the god actually pours his bounty upon the human spirit, and the human spirit absorbs it, the living and breathing of song will remain, for the aspirant poet, merely an intellectualized intention. This point requires elaboration. The overriding message of Die Sonette an Orpheus is that poetic creation is a mission, one that demands the poet's total devotion if his poetry (his 'song') is to be authentic. What genuine poetry is, is known only to the gods; e. g., to the Gott of Sonnet I, 3 (cf. Brodsky 1988: 103, 1 5 7 - 1 5 8 ; also Prater 1986: 3 5 0 - 3 5 1 ) . It is the poet's duty to tap the divine source of creativity for the nourishment needed to reach a level of singing that can pass muster before the gods, so to speak. For Rilke, the archetype of the divinely eloquent poet was Orpheus, whose name graces the title of Rilke's sonnet sequence. In I, 3 Orpheus also appears metonymically in the image of the Leier 'lyre' (line 2) and semantically (in yet another, presupposed, converse relationship), in the allusion to his father Apollo, the god of music and poetry (line 4). Rilke had become familiar with the story of Orpheus at least through having read Ovid's Metamorphoses (Prater 1986: 350 — 351). Just as Orpheus acquired his musical gifts from his father, so must the mortal poet evidence filiation by sanctifying his being, or by having it sanctified by the divine. The poet can promote this end only through self-denial. Orpheus, we know, evidenced his poetic sonship by charming the gods in heaven; captivating the birds, beasts, trees, and rocks upon earth; and finally by persuasively touching even the "bloodless ghosts" of hell (see Ovid, Metamorphoses, Book X). It is important to underscore that Or-

Poetic discourse

369

pheus's songs produced mighty effects on the sentient and on the nonsentient: his songs transformed things. (For example, in Book X of Metamorphoses Ovid reports that trees literally moved their position when the strains of Orpheus's music traveled through the air. And even after he died, the universe of things, into which Orpheus's songs had breathed animation, mourned his death: the trees that had formerly stirred now "shed their leaves", as reported in Book XI.) Just as the idea of transformation (metamorphosis) was important to Ovid, so it was to Rilke too: many of his sonnets develop or allude to this theme. (See analytical comments by Brodsky 1988: 159 — 160 and passim.) While a sonnet such as II, 29 (which I will not discuss here) is more clearly and symbolically Orphic, in that the poet is actually enjoined to address, and through addressing, to transform the 'quiet earth' and the 'rushing water', in I, 3 the being for whom transformation is desired is the poet himself: he constitutes the exclusive topic of I, 3. He must transform himself by quelling egocentric and other poetically obstructive tendencies, and by submitting to the god's instruction. These motivic ends are all subserved through the pragmatic strategies discussed earlier. The poet's novitiate lived in obedience to, and under the tutelage and patronage of the god is conveyed through various images of converse relationships, each of which foregrounds a particular aspect of that relationship. The first quatrain begins appropriately with allusions to the divine, in the father-son relationship between Apollo and Orpheus. In this quatrain Apollo is simultaneously depicted, quite subtly, as the preeminent manifestation of poetic creativity and as the guide of poets. In legend, Orpheus found it easy to follow his father. But as I, 3 underscores, in contrast to Orpheus, the mortal poet-aspirant can only with great difficultly follow the god; for the mortal has to overcome desire and longing; in sum, a diffuse consciousness. The second quatrain portrays the god more explicitly as a teacher in relation to the poetaspirant. The god's lesson is that the poet must live 'song' (Gesang ist Dasein). But the poet cannot live song solely through his own effort. In order to attain Orphic heights, he must also be showered with the god's bounty (symbolized by die Erde und die Sterne), which the god gives of his own accord, and which the poet must await: hence the significance of Wann, in Wann wendet er/an unser Sein die Erde und die Sterne? (lines 8 — 9). The poet is here depicted as a passive recipient, for he cannot summon this divine bounty through his own will. In this instance, the god is depicted as the benefactor of poets, who are in a sense his heirs. Understanding all this, the poet/learner now tries to teach the Jüngling,

370

Janet Byron Anderson

who symbolizes untutored modes of thought and expression based not on 'sung' being, but on the emotional exaltation that stems from feelings, especially those engendered by human love. This viewpoint is one which the poet's enlightened side wishes to quell, or at least tame. All of these converse relationships image different aspects of poetic life which is informed by divine inspiration. But this kind of life represents an ideal; for the speaker is ambivalent about it in actual experience — as we discover from the shifts in personal deixis. Nevertheless, the voice which speaks in the poem makes it clear that this ideal is the only authentic model for the poet. Although, moreover, two important issues are expressed questioningly, namely: (rhetorically), How can a man follow the god? — the active side of creativity; and When will the god bestow his bounty? — the passive side of creativity, even these questions accept as a given the indisputable role of the divine in poetic endeavor. The speaker's clarity on this point, as well as his permitting no gainsaying of his views, is conveyed in part through strong affirmations (e. g., In Wahrheit singen ist ein andrer Hauch: line 13). Such utterances, by their very explicitness, invite disagreement. But the speaker is not so much interested in laying out potentially arguable views, as in impelling the addressee to accept his (the speaker's) views. The two techniques that subserve this latter and are chiefly cataphoric reference and presupposition. By postponing the revelation of important elucidating details, cataphoric reference elicits curiosity, and by so doing, draws the addressee into a situation-in-progress, which thereupon becomes a part of his emotional experience. On the other hand, presuppositions — particularly the recurring ones involving the converse relations I pointed out earlier — forestall the addressee's opportunity to recognize a particular view, let alone to question it. In either case, the impulsion for the addressee to agree with the speaker's position becomes ineluctable. These techniques all bear an axiomatic stamp: jointly, they heighten the tone of authority in Rilke's articulation of the message that poetic life is not a profession but a mission.

Notes 1. This article is a revised version of the original talk that I delivered at the International Conference on "Cooperating with Written Text: The Pragmatics and Comprehension of Written Texts", at the University of Giessen. Federal Republic of Germany, September 5 - 9 , 1989.

Poetic discourse

371

2. In my original presentation, I also spoke on Sonnet II, 29 of Die Sonette an Orpheus; but because of limitations of time, I was unable to go into a comprehensive analysis of this sonnet during the conference. In this article, my comments on II, 29 are also brief. A fuller pragmatic interpretation of it occurs in a chapter of my research project (in process), The application of sociolinguistics to literary analysis. I wish to express my gratitude to Professor Patricia Pollock Brodsky and Professor Peter Hinze for their helpful evaluation of my interpretation of II, 29 in the first draft of that chapter. Because of the stature of his works, Rilke's poems have elicited much critical attention. But pragmatic study of his poetry has understandably been rare. The only such study that has come to my attention is found in York (1986: 93 — 108). York's discussion covers much of Rilke's poetry, but it is brief. Consequently it does not offer a detailed pragmatic reading of any single poem of Rilke's. Instead, we are given essentially a pragmatic overview of Rilke's poetry. But because of the complexity of Rilke's poems, an overview cannot display the full power of pragmatics as an instrument of poetic analysis. Nevertheless, I regard York's commentary as a valuable contribution to a pragmatic understanding of Rilke's poetry; I am indebted to it in my research. 3. All quotations from the Sonnets are from Volume 1 (Gedichte), 1955, of Rainer Maria Rilke, Sämtliche Werke, the standard collected works, published 1955 — 1966 and are quoted by permission of the publisher, Insel Verlag. Sonnet I, 3 appears on page 732 of Volume 1 and Sonnet II, 29 on pages 770—771. These sonnets belong respectively to Part I and Part II of Die Sonette an Orpheus. Part I contains twenty-six sonnets and Part II, twenty-nine; none of the sonnets are individually titled. The entire set of fiftyfive constitutes a sonnet sequence, directed to Orpheus, the poet-musician of Greek mythology; hence the phrase an Orpheus 'to Orpheus' of the title. The prose translations (which are fairly literal) are mine. Numbers in the translation correspond to successive lines in the poem. Alternative readings are indicated by slant lines; romanized words in the original lines correspond to constituents that are italicized in the original. 4. Separation of the speaker in a work from the writer of the work, and the fictive audience from the real-world audience of the work, commonly occurs as an analytical tool of fiction, see, e.g., Leech — Short's discussion (1981: 259 — 269), with respect to English fiction. The distinction is comparable to Burton's (1980) differentiation between the micro-addressee (the audience within the work) and the macro-addressee (the audience outside the work), except that Burton draws this distinction with respect to drama. Such distinctions as these can, I believe, be extended to poetry; for to the extent that poetry can be viewed from the pragmatic perspective of interaction, it is not surprising that analytical tools employed in the linguistic study of genres in which interaction is known to predominate might also be used to enhance our understanding of poetry. 5. For a description of the conative function of speech, see the classic article by Jakobson (1960: 353 — 357). Within speech act theory, the effect that the speaker's utterance produces is known as the utterance's perlocutionary effect (Searle 1970: 25). 6. Since in the first quatrain the speaker is in effect simultaneously talking to the god and about the god, technically the lines evidence paradox — but paradox operating at the pragmatic (specifically, in this case, deictic) level. Like every paradox, the quatrain therefore produces emphasis and attracts attention (see comments on literary paradox in Holman — Harmon 1986: 357). The paradox in these lines accordingly reinforces the emphasis contributed by cataphoric reference, and vice versa. Objectively speaking, however, no paradox occurs; for inasmuch as Rilke is depicting a ubiquitous god, it would be impossible for discourse about this god (third person reference) to escape his

372

Janet Byron Anderson

ears (second person address). For another perspective on the shift in personal deixis, see pages 367 — 368. 7. For an earlier comment on the correlation between conative and emotive utterances, see pages 362 — 363.

References Brodsky, Patricia Pollock 1988 Rainer Maria Rilke. Boston: Twayne. Burton, Deirdre 1980 Dialogue and discourse: A sociolinguistic approach to modern drama dialogue and naturally occurring conversation. London: Routledge and Kegan Paul. Halliday, Μ. A. K. — Ruqaiya Hasan 1976 Cohesion in English. London: Longman. Holman, C. Hugh — William Harmon 1986 A handbook to literature. (5th edition). New York: Macmillan, and London: Collier Macmillan. Jakobson, Roman 1960 "Closing statement: Linguistics and poetics", in: Thomas A. Sebeok (ed.), 350-377. Leech, Geoffrey N. 1983 Principles of pragmatics. London and New York: Longman. Leech, Geoffrey N. — Michael H. Short 1981 Style in fiction: A linguistic introduction to English fictional prose. London and New York: Longman. Levinson, Stephen C. 1983 Pragmatics. Cambridge University Press. Ovid [Publius Ovidius Naso] 1955 Metamorphoses. [Translated by Mary M. Innes.] Harmondsworth: Penguin Books. Prater, Donald 1986 A ringing glass: The life of Rainer Maria Rilke. Oxford: Clarendon Press. Preminger, Alex, (ed.) 1974 Princeton encyclopedia of poetry and poetics. Enlarged edition. Princeton, New Jersey: Princeton University Press. Rilke, Rainer Maria 1955 — 1966 Sämtliche Werke. 6 Volumes. Edited by Ruth Sieber-Rilke and Ernst Zinn. Frankfurt-am-Main: Insel Verlag. Searle, John R. 1970 Speech acts: An essay in the philosophy of language. London: Cambridge University Press. 1979 Expression and meaning: Studies in the theory of speech acts. Cambridge: Cambridge University Press. Sebeok, Thomas A. (ed.) 1960 Style in language. Cambridge, Mass.: The MIT Press. York, R. A. 1986 The poem as utterance. London and New York: Methuen.

A pragmatic role for inserted clauses in literary texts Danielle Forget

1. Introduction The inserted clause has long been neglected by linguistic studies. They have a lot in common with what is usually called parentheticals and have the particularity of being inserted in a longer string. Note the use of la bramante in the following: (1)

Certes il avail entendu parier au Chili d'une colline appelee El Bramador — la bramante — parce que du sable mis en mouvement par les pas d'un marcheur imane comme un grondement caverneux. (Tournier, Vendredi p. 83) ['It was true that he seemed to have heard of a hill in Chile called El Bramador (a word which meant the bellowing of a stag) because when anyone walked on it the shifting of the sand set up a subterranean rumble.' (p. 80)]

Many problems arise from trying to integrate this type of construction — la bramante — the inserted clause, into a sentence grammatical description. In the present study it will be demonstrated that it is more adequate to consider them at a textual level. Their contribution to the pragmatic orientation of the discourse will be illustrated by analysing their role in a literary text. More specifically, this study will show that the narrative exploitation of inserted clauses in literary texts follows directly from the formal properties of these clauses. Examples will be drawn mainly from written literary excerpts, though comparisons will be made with examples from an oral corpus, which will be referred to as the Montreal corpus '84. 1

2. General presentation of the approach In the following, the inserted clause will be discussed using the general schema: Enunciation-Inserted Clause-Enunciation' (E-IC-E') where the

374

Danielle

Forget

IC is the inserted clause and the E-E', the clause in which it is inserted. To be considered an inserted clause, the clause has to a) interrupt the EE'; b) but, at the same time, be in cohesion (in the sense of "underlying semantic relation") with the E-E' and/or the linguistic context in which it is inserted. It follows from these conditions that simple cases of inversion, as well as cases of accidental interruption of the E-E' due to extralinguistic factors, will be ruled out. The present discussion will be limited to the case of insertion into the main clause although it can precede or follow it. Traditional grammar has characterized the inserted clause very loosely, usually limiting itself to the descriptive aspect: its position in the linear order, usually in the middle of a main segment or postposed to it. A semantic characterization is also present in regard to those inserted clauses which play the role of introducing the direct speech. A sentence containing this type of inserted clauses indicates the existence of two speakers. But no formal description has been provided for the main clause and its relation to the inserted clause. 2 More recent studies in syntax have dealt with parentheticals which have a lot in common with the present construction under discussion. Even though the class of parentheticals has never been clearly delimited, it usually includes entities such as adverbials, adverbial clauses, infinitival clauses, -ing clauses, nominal relative clauses and appositions. 3 Several authors like Nemni (1981); Ziv (1985); Delomier and Morel (1986) have noted the intonational particularities of the inserted clauses. Parentheticals are said to be marked by a particular tone unit, separated or not by a pause. Among the few attempts that have been made to stress its pragmatical aspect, the property of expressing a comment is usually retained (Ziv 1985). This last pragmatic characterization seems to fit the type of sentences to be examined in the present study, but it has the drawback of being so general that it may include many other linguistic elements, like interjections, that may be considered as comments; moreover, it doesn't take into account the structural specificity of such a clause. The pragmatic role of the inserted clauses (from now on, IC) will be specified relying on the intonational and syntactic properties that have been pointed out in recent studies, such as Ziv (1985) and Delomier —Morel (1986). 4 Among these properties, the most important are its mobility in the sentence and, more significantly, its autonomy. "Accordingly, parentheticals are regarded as constructions which show no dependence on other elements in the sentence and which do not partake in any sentential syntactic process." (Ziv 1985: 1 8 2 - 1 8 3 )

Inserted clauses

375

This last aspect is essential for the characterization and the delimitation of the IC and will be developed in the next section.

3. The autonomy of the inserted clause Among the properties attributed to the inserted clause, its autonomy is the most important one. The autonomy of the IC can be demonstrated by the following criteria: the E-E' counts in itself as a grammatical sentence; as a consequence, the IC can be deleted without affecting the E-E'. The particular intonational pattern of the IC reinforces this assumption. But looking at the internal configuration of inserted clauses, the conclusion should not be so clear. Some show great dependency to the string E, and in fact count as a constituent of the main predication in the EE', or as a clause with a structural link, an explicit connective to E. (2)

II comprit que I'ile administree demeurait son seul salut aussi longtemps qu'une autre forme de vie — qu'il n'imaginait meme pas, mais qui se cherchait vaguement en lui — ne serait pas prete ά se substituer au comportement tout humain auquel il etait reste fidele depuis le naufrage. (Tournier, Vendredi p. 125) ['He realized that in the ordering of the island lay his only salvation, until such time as another kind of life — impossible to imagine, but already struggling into being within him — was ready to take the place of the very civilized behavior he had clung to since the shipwreck.' (p. 119)]

Many show such an autonomy that they are not so much in relation with the segment Ε as to the whole sentence E-E' (3)

Par exemple sur les quelques milliers de personnes qui se font chaque annee ecraser par des voitures, un bon nombre — c'est prouve — se sont precipitees inconsciemment sous une voiture pour repondre ά une condamnation prononcee par le surmoi. (Tournier, Coq p. 226)

Moreover, instead of being inserted in the middle of the E-E' they could have followed it. However besides these examples in which there exist explicit, structural links between the two sets of clauses, there are others cases, by far the more frequent in the oral corpus, which don't show any formal link with

376

Danielle

Forget

and, furthermore, stand at an extraneous level to the sentence E-E'. Here is an example from a written corpus. (4)

Enfin le soir du deuxieme jour, alors que la baronne et ait sortie — il fallait bien que quelqu'un fasse les courses — on sonna. (Tournier, Coq p. 205)

This last type of structure gives a better example of the autonomy of the IC. The IC can easily be erased without structurally affecting the sentence E-E' while, as we will see, at the semantic level, its contribution is important. So the notion of autonomy has to be further specified in terms of structural autonomy. It is nevertheless, possible to conclude from the variety of these examples that inserted clauses have no fixed internal configuration and that they entertain a particular link with the linguistic context. 5 The indetermination of the IC between autonomy and dependency can also be measured by looking at their position on the linear string. They have a certain degree of mobility in attaching to the E-E', when the inserted clause is a relative clause or an apposition, its position is structurally ruled, but it also faces some constraints of a different nature. Example (5) tolerates mobility in b) but not in c) which would change the meaning, the IC being associated with "degre de profondeur" rather than "amour": (5)

a. Et pourtant un sentiment comme I'amour se mesure bien mieux il me semble — si tant est qu'il se mesure — ά I'importance de sa superficie qua son degre de profondeur. (Tournier, Vendredi p. 69)] ['Yet it seems to me that a feeling such as love is better measured, if it can be measured at all, by the extent of its surface than by its degree of depth.' (p. 67)] b. Et pourtant un sentiment comme I'amour — si tant est qu'il se mesure — se mesure bien mieux il me semble... c. Et pourtant un sentiment comme I'amour se mesure bien mieux il me semble ä l'importance de sa superficie qua son degre de profondeur — si tant est qu'il se mesure.

These examples illustrate the dependency of the IC on the E-E'. Further investigation of anaphora and connectives will help to specify this characteristic of the IC. Although the IC displays no fixed internal structure, the study on the oral corpus (1989) has revealed that there exists a particular structural

Inserted

clauses

377

pattern of the relation between the IC and the E-E'. 6 By the use of pronouns, tenses of verbs, the presence or absence of temporal and locative devices, and speech acts selected inside the IC, its construction uses more subjective devices (cf. Benveniste 1966). The presence of particles (or mots du discours according to Ducrot et al. 1980) frequently initiates the IC and marks a break in the continuity of E. (6)

... parce qu'un ordinateur personnel, je veux attendre dans quatre cinq ans pour I'acheter quand mes enfants vont ... etre habilites, quand Mathieu bon je dis quatre cinq ans, mettons dans quatre ans c'est Qa quand Mathieu va etre capable de travailler dessus, la j'en acheterai un. ( # 13,27: number in the oral corpus)

The IC may also express a speech act different from the one in the dominant clause. For example, it may carry an interrogative form no matter what the form of the E-E' is: (7)

J'aime un restaurant ordinaire lä comme je sais pas si tu connais ςα? le Jardin la au coin de Hochelaga puis... ( # 2,85)

The impact of the interruption on Ε may be marked by the presence of such an expression as "qu'est-ce que je disais" in E', which serves to identify the performance problem of retracing the main topic of conversation. But sometimes the contrast comes from the absence of such enunciative marks and the use of impersonal style. Delomier and Morel (1986) have focused on the elements of the combination E-IC-E' that mark the IC as a separate element: E' usually presents a grammatical structure; when there is a connective in E', it is connected to the Ε and not to the IC; Ε is incomplete in itself and has to be combined with E'. Of course the particular intonational pattern, the presence of a pause in spoken form and a punctuation in written form reinforces this separation. In brief, many linguistic devices at the intonational and syntactic level show the autonomy of the IC. But nevertheless, as example (5) suggests, the ICs entertain some particular and constrained relations with the EE'. In order to determine the nature of that link it is necessary to investigate the anaphoric and thematic relations between the IC and the E-E'.

378

Danielle

Forget

4. Anaphoric relations The most common link that the IC entertain with the E-E' is anaphoric. But the difficulty of applying this notion strictly appears with example (3) (3)

Par exemple sur les quelques milliers de personnes qui se font chaque annie ecraser par des voitures, un bon nombre — c'est prouve — se sont precipitees inconsciemment sous une voiture pour repondre a une condamnation prononcee par le surmoi. (Tournier, Coq p. 226)

or with: (8)

Parce que c'etait mardi — ainsi le voulait son emploi du temps —, Robinson ce matin-lä glanait, sur la greve fraichement decouverte par le jusant, des especes de clams... (Tournier, Vendredi p. 55) ['Because it was a Tuesday morning, Robinson, in accordance with his work schedule, was on a strip of beach newly uncovered by the tide, collecting a particular kind of clam.' (p. 56)]

In (3), the whole predication "un bon nombre de personnes..." is concerned with c' of c'est prouve, which plays the role of a cataphore. In (8), le cannot be associated with any linguistic segment of the E-E': it is rather a causal relation indicated by parce que that is referred to. The distinction between anaphoric and deixis might be useful. Levinson (1983), following Lyons (1977), has stressed the importance of distinguishing anaphora from discourse deixis. While "anaphora concerns the use of (usually) a pronoun to refer to the same referent as some prior term" as in: "(90) Harry's a sweetheart; he's so considerate." (Levinson 1983: 86) discourse deixis instigates a reference to the linguistic expression itself or to a chunk of discourse. So the link, is made with a portion of the discourse, be it the preceeding utterance or any utterance in the surrounding text, as in: "(89) That was the funniest story I've ever heard." (Levinson, 1983: 85) The referential link between the inserted clause and the E-E' is better explained by discourse deixis than by anaphora, even though both types may be present.

Inserted clauses

379

Lyons (1977: 670) has also introduced an in-between category: the "impure textual deixis", where the reference is made not so much to a sentence but to the statement behind it. "(94) A: I've never seen him. B: That's a lie." (Levinson 1983: 87) In the following example, (9)

" . . . car la langue majestueuse et maternelle — il faut bien le dire et le constater — a un Statut de langue morte!" (Aquin, Trou

p. 95) the inserted clause announces a portion of discourse (what is called a cataphoric expression), and seems to combine these last two categories of links-discourse deixis and impure textual deixis. The verb dire makes us interpret Ε as a sentence, while the use of constater makes us look as Ε as a statement. The presence of anaphora in inserted clauses does not frequently show a simple relation with a referent by the intermediary of a linguistic segment in E-E'. The links between the IC and the E-E' are very loose. As it is the case in the present corpus, Delomier and Morel (1986) outlined in their oral corpus, the selection of c'est, ςα, ily a which all have in common greater autonomy than simple linguistic anaphora and that usually don't refer back to a single linguistic segment of the preceeding occurrence. 7 When the referential link is chosen, it has more to do with discourse deixis and even when the connectives are chosen, it is also the discursive level that is primarely involved, as illustrated by this example: (10)

II etait parvenu ä un equilibre — ou a une serie d'equilibres — oil Speranza...(Tournier, Vendredi p. 236) ['He had achieved a state of stability, or series of states, in which he and speranza...' (p. 218)]

N o relation of equivalence is present in this type of us of ou; it is more two expressions that are juxtaposed. The same happens with et and mais: their occurrence has to be related to an "enunciative move", that marks the relinking of E' to E. 8 The expression "enonciation" is used here as in Ducrot (1984: 69): an act of enunciation is an individual event whose instigator is a speaker in a particular utterance situation. Even if discourse deixis is extended to include reference not only to the linguistic expression itself but also the the whole utterance context, it remains that the IC refers back to the enunciative event that is taking

380

Danielle

Forget

place. The inserted clause makes use of deixis because of the fact that it refers to the discourse "hie et nunc" that is performed. The most evident uses of discourse deixis concern what Authier-Revuz (1987) has studied under the term pseuso-anaphoric deixis ("pseudo-anaphore deictique"). "... par lequel I'enonciation dun element X quelconque se double d'une representation d'elle-meme, de type je dis X"' (Authier 1987: 20), as in: (11)

"Depuis quelque temps en effet je m'exerce ά cette operation qui consiste ά arracher de moi successivement les uns apres les autres tous mes attributs — je dis bien tous — comme les pelures successives d'un oignon." (Tournier, Vendredi p. 88) ['For some time indeed I have been performing an act of surgery on myself which consists of stripping away in turn all my attributes — I say all — like peeling an onion.' (p. 85)]

So the word X' (tous) refers back in the E-E' to the use of the word and not only to its content. It is both reflexive and opaque, and as the author stresses, it refers to the conjunct signifie/signifiant of a determinate segment. 9 This particularity of the inserted clause, of referring back to the E-E' in the discourse taking place, has to be related to a more general process: the metadiscursive role of the inserted clause.

5. The metadiscursive role One of the characteristics of the IC is its structural autonomy from the E-E'. It is precisely because of its structural autonomy that it can best perform what we will call a metadiscursive role. Several attempts have been made to treat this type of clause by staying very close to the sentence analysis (J. Emonds 1976; McCawley 1982): as adverbials, or a type of coordinate or subordinate. However, it is rather the textual dimension of these inserted clauses which gives them a multiplicity of forms, a controled position on linear order but which constrains their role and insertion in the text. Without pretending to give any formal description of it, it will be shown that the specificity of the inserted clause justifies its role in the argumentation and narration of the literary text. It is commonly pointed out that this type of clauses usually have the function of adding some details to an object just introduced in the discourse, of explaining its meaning. The use of the IC to add a detail

Inserted clauses

381

creates a referential illusion and the argument that the inserted clause contains information that may be substituted for that object reinforces this view. But such a perspective on this ICs is too restrictive. Firstly, in cases where the IC is presented as some kind of paraphrase of an antecedent χ in the E-E', it is rarely true that the substitution of x' for χ is adequate. Let's look at: (12)

"Joan" — je veux dire: celle que j'ai aimee et dont la realite m'obsede — est la specialiste du trompe-l'oeil au theätre; ... (Aquin, Trou p. 129)

This main clause comes from a paragraph characterized by an objective point of view of Joan's work at the theater, so the insertion of the segment between dashes, in spite of being presented as "another way of saying the same thing", prevails by its contrast with the main clause. It is the relation between the IC and the E-E' that is important and has to be preserved, as illustrated by an example such as (11), where the reiteration is important. Secondly, if it specifies the members of a class of objects, the presentation in the IC is not exhaustive, indicating that its role is not primarily informative. It serves rather as an example to ensure better comprehension on the part of the listener/reader. Too much emphasis has been put on the informational power of the IC and yet such a construction produces much more than a mere ajustment of the content of the message. Specifying the semantic link of the IC can better be accomplished by looking at its participation in the thematic structure. Since the IC cannot be accidental, its occurrence acts as an anchorage point in the E-E' or in the general discursive context of the E-E'; using Lyons' definition of theme as "the expression used by the speaker for what he announces as the topic of his utterance" (1977: 507), it borrows its theme from outside. Let χ be the anchorage point in the E-E' that motivates the occurrence of the IC. It may be said that χ counts as a theme of the IC, no matter what its status is in the E-E', be it a theme or a rheme. There seems to be another argument in favour of the autonomy of the IC: it is not part of the thematic structure of the E-E', although it stresses an element of this clause by retaking it as a theme for the IC. Under the pretext of reestablishing the proper meaning or delimiting the meaning of a word as a second try for marking reference, the speaker informs his audience of the way his message should be interpreted. Not only does he add some information but he also presents himself as

382

Danielle

Forget

intervening on the sense of the act of enunciation he is making. It is its function in discourse that is essential, the strategy of the speaker in using it that has to be taken into account. That is meant by saying that its role is metadiscursive. Many occurrences of inserted clauses relate metalinguistically to the E-E'. Jakobson's work has made the metalinguistic function of language well-known and associated with a return to the linguistic code. For the purposes of this discussion the role of the IC to refer back to the particular use of χ in the E-E' and its linguistic context, will be described as a metadiscursive function. The most common use of the metadiscursive function is when χ is considered as a metalinguistic sign, in the sense of Jakobson: the IC makes a comment on χ as a linguistic code unit. A particular case is when the narrator wants to provide his listener/reader an equivalent word in another language, as in the French translation in (1), la bramante. But reference back to the χ of the E-E' also occurs when χ is a statement, an interrogative, a referential act etc. The metadiscursive role appears when the IC refers back to a discourse utterance which is being produced, to add some other clues for the identification, make a comment, express an attitude, etc. In the following examples, the IC refers to the statement, the connotation, more than the use of words. (13) (14)

(15)

Pierre XMagnant n'estpas — euphemisme! — un homme comme les autres. (Aquin, Trou p. 109) Contre ce chaos, Vile administree — de plus en plus administree, car en cette matiere on ne reste debout qu'en αναηςαηί — est mon seul refuge, ma seule sauvegarde. (Tournier, Vendredi p. 117) ['Against this chaos the ordered island — ordered more and more rigorously because in this matter one cannot hold one's ground except by going forward — is my only safeguard, my only refuge.' (p. 112)] Mais tandis que la souille me faisait hanter principalement ma sceur Lucy, etre ephemere et tendre — morbide en un mot —, c 'est a la haute et severe figure de ma mere que me voue la grotte. (Tournier, Vendredi p. I l l ) ['But whereas the mire principally evoked the memory of my sister Lucy, a passing, tender fancy — morbid, in a word — the cave brings back the lofty and stern figure of my mother.' (p. 106)]

Inserted clauses

(16)

383

... personnage imaginaire, certes, mais non pas — comme il I'avait cru — inexistant. (Coq p. 136)

The pronoun Ϊ announces the E', so its role is cataphoric. The link is not so much with the occurrence itself as with the statement. Another element confirms the importance of looking back at the discourse choices that have been made in the E-E' or in its linguistic context: it is the fact that in most examples no linguistic segment of the preceeding occurrence can be associated with the IC, in spite of the cohesion that it manifests. (17)

II a finalement accepte — un tiens vaut mieux que deux tu 1'auras — que je lui remette la somme que j'avais sur moi.

So what is aimed at by the IC may be the conventional but also the communicational content of the utterance: a)-inferences from the E-E': (18)

Autrefois on y allait, bien autrefois je dis ςα je suis pas vieille lä mais tu-sais: dans ma: mon adolescence j'ai bien aime ςα il y a des belles plages I'autre bord on allait lä en bateau lä. ( # 85,8)

b)-presuppositions; ^-referential act; and d)-attitudes, as in: (19)

Elle m'a tellement supplie de lui faire mon propre recit de l'evenement, de lui raconter — oui, moi! — comment cela s'est passe ... (Aquin, Trou p. 179)

Let's represent the link of the IC with the E-E' by the following schema: E-E' χ·

IC x' (about "x")

where χ is any constituent of the enunciative act: inferences, possible conclusions, attitudes, etc. Even when the IC are not strictly metalinguistic, they instigate an adjustment in the course of producing the message, towards the enunciation. When a speaker uses an inserted clause, it is not the content of the message that he is aiming at (not primarily) but to control the comprehension and the adhesion of his addressee. So it works at a different level than the one present in the E-E', and the structural break (by metalinguistics means, as noted by Delomier —Morel 1986) precisely marks this change of level.

384

Danielle

Forget

The metadiscursive interruption that the inserted clause initiates is not a strange phenomena and corresponds in fact to transitions present in dialogue. Examine for instance a dialogue between A and B, where A asserts (20): (20)

A: John said that he saw a unicorn

and Β reacts, requesting more information. By responding: B: A unicorn? 10 (assuming that the interrogative or exclamative intonation does not clearly differentiate the two possible senses). As such B's question could be interpreted as a request for explanation of the code itself (what is a unicorn?); the use of the code (do you really mean a unicorn?), or the assertion linked to the situation described (how can he have said that? how can he have seen a unicorn?). It is common to use language both to aim at reality and language itself, as pointed out by Berrendonner: "Dire

qu'un enonce

realite",

"est en lui-meme

c'est bien faire

allusion,

Γenonce ne se mue en fait que lorsqu'il l'enonce,

mais I'evenement

un fait",

il me semble,

d'enonciation

ou bien qu'il est "considers

non ä l'enonce,

a ete actualise,

mais ά son

et le fait,

comme

enonciation:

c'est alors non plus

qui a eu lieu." (1981: 131)

The change of level noticed between the inserted clause and the main clause doesn't mean that the IC is not concerned with the rules of expansion of a text. If so, how could we explain the ackwardness of this combination: (21)

II a finalement accepte — de ma poche j'ai sorti dix dollars — que je lui remette la somme que j'avais sur moi.

Here the insertion interferes with E' (the more precise information being presented before the general) because it is not possible to see the progression from the IC to the E ' . n The fact that the insertion of the IC is not exempt of this discourse constraint gives us a hint of its textual participation.

6. The role of the inserted clause in narration and argumentation It is tempting, at the oral, to associate the interruptions created by the inserted clauses with hesitations, memory problems, in other words discontinuity, having to do with the difficulty of constructing a perfect

Inserted clauses

385

message given the constraints of time, and interference of the situational context inherent to performance. 12 But the presence of inserted clauses in written texts and particularly in literature, throws a doubt on this type of hypotheses. 13 6.1. Let's have a look at the participation of the inserted clause to the argumentative movement of a text. Let's consider this excerpt from Tournier's book: Je ne me risquerai jamais a lui dire "aime-moi", parce que je sais trop que pour la premiere fois je ne serais pas obei (PI). Pourtant iln'a aucune raison de ne pas m'aimer (P2). Je lui ai sauve la vie (P3) — involontairement il est vrai, mais comment s'en douterait-il (P4)? Je lui ai tout appris, ά commencer par le travail qui est le bien supreme (P5). Certes.je le bats (P6), mais comment ne comprendrait-il pas que c'est pour son bien (P7)? (Tournier, Vendredi, p. 154) [Ί would never venture to say to him "Love me," because I know all too well that for the first time I should not be obeyed. Yet he has no reason not to love me. I saved his life — unintentionally, it is true, but how can he know that? I have taught him everything, and above all to work, which is our greatest wealth. Certainly I beat him from time to time, but he must surely know that this is for his own good.' (p. 145)]

The first two sentences initiates the argumentation in this portion of text. The first sentence, let's call it PI, aims at the conclusion "he would refuse to love me" (or "he doesn't love me"), at which the second one marks an opposition "he doesn't have any reason not to love me" by the intermediary of "pourtant". The role of "pourtant" is well-known since Anscombre's study (1983). Given two propositions PI and P2 linked by pourtant, a speaker may use this connective to argue in favour of P2 by denying directly the argument in PI, or to defend a conclusion different for the one to which the argument in PI leads. It is this last interpretation, called "denegation" that is relevant in the excerpt. Schematically these propositions are present in the debate: PI Conclusion r: "he doesn't love me" P2 Conclusion r': "he doesn't have any reasons not to love me", where r' = non-r. The continuity of the text is based on the reinforcement of the conclusion brought by "pourtant" (with a change: instead of presenting possible reasons not to love him and refuting them, Robinson presents reasons that Vendredi should have to love him). From P3 to P7, arguments are

386

Danielle

Forget

presented that serve alternatively the conclusion r or r'; they are balanced to finally reject the conclusion r to the benefit of r', and to convince the listener/reader of the love that Vendredi should have towards his master. Cond r

Concl r'

P4 " involontairement il est vrai ...

P3 "je lui ai sauve la vie" mais comment s'en douterait-il? P5 "je lui ai tout appris."

P6 "certes je le bats." P7 "... c'est pour son bien?" In order to better understand the role of each proposition in the argument, it is necessary to examine them closely, starting with P5. The only proposition isolated in a sentence, P5, supports clearly the conclusion r', the conclusion to which Robinson subscribes. Propositions P6 and P7 are a pair. P6 acts as a counterargument to the conclusion r', a counterargument which is immediately rejected in P7. The expression certes ... mais, typical of the double movement of the concession, is used. Going back to P3-P4, it is possible to observe a similar relation between the arguments that are expressed, leading to opposed conclusions. P3 is first presented as support for Robinson's conclusion r', and P4 is introduced second in order to restrain the persuasive impact of P3: the use of "il est vrai" transforms the proposition into a concession by its acknowledgement of others' points of view. So the two pairs P3-P4 and P6-P7 share many similarities, as shown by the equivalence of certes and il est vrai. Following Ducrot et al. (1980) and using Mikhael Bakhtine's concept of dialogism on the study of such linguistic entities, we might say that one member expresses an enunciator's voice which is different from the one in the other member and which the speaker supports. For a better understanding of the functioning of certes ... mais, let's recall some key elements of Ducrot's hypothesis of polyphony. The speaker, who is responsible for the enunciative act taking place, might express in his utterance different points of view, which have to be attributed to abstract participants, the enunciators e1, e2, etc ("enonciateurs"). So the speaker may choose to take a certain point of view: if he does, he expresses adherence to a particular enunciator. The distinction between speaker and enunciator, and the possibility for an utterance to express many points of view, allows pragmatics studies to take into account many expressions based on a double movement, like negation, irony and concession.

Inserted clauses

387

—,mais

Certes 1

Arg. of e for concl. r

arg. of e2 + speaker for concl. r'

As shown in the schema, the construction presents enunciators' e' and e 2 points of view; it further establishes a relation of argumentation such that, though the speaker doesn't reject e''s argument, e 2 's is determinant. In other words, certain entities have the property of contributing to the main argumentation by revealing others' argument. In the following, the property of the IC to express a contrastive point of view will be explored. Let's return to the two pairs P3-P4 and P6-P7: their difference relies on the status of P4 as an inserted clause. P4 expresses an argument against r' which is then rejected, in a reasoning that is marginalized, because of its special status as an inserted clause, while P6-P7 participate equally in the argumentative movement (even though P7 expresses the determinant argument). U p to that point the inserted clause follows the general pattern but where the specificity of the literary text intervenes, is in the correspondance of the formulation of the IC with P7, the last argument, a decisive one in favour of r': P4 ... mais comment s 'en douterait-il? P7 ... mais comment ne comprendrait-il pas que c'est pour son bien? It compensates for the inherent marginalization of the inserted clause. Even if in relation to P3, P4 presents an argument that is immediatly rejected and it participates in the argumentation; the parallelism of the construction reinforces its role on a longer string by bringing an echo of it at the end of the demonstration, which contributes directly to its impact. Indeed, the structural separation of the inserted clause corresponds to the marginalization of a proposition from the general argumentative movement. Even though an argument is presented in the IC, the fact that the E' does not continue the IC but the E, gives it a special status in the text as a whole. It allows the speaker to eliminate the intersentential links that would have to be present if the IC were related to the E-E' as a whole. In consequence, it provides the text with a direct flow. The speaker talks "about" some elements of the discourse as in annotation and incidentally, as also characteristic of footnotes: some are in syntactical continuity, others are not.

388

Danielle

Forget

The narrator may benefit from the marginalization of the information contained in the IC, as in this example: (22)

Le soir tombait, et ils venaient enfin de trouver un objet intact — la longue-vue — lorsqu'ils decouvrirent le cadavre de Tenrt au pied d'un arbre. (Tournier, Vendredi p. 187) ['It was growing dark, and they had found one object that was still intact, namely the spyglass, when they came upon the body of Tenn stretched under a tree.' (p. 178)]

The specificity of the object "longue-vue" is given as, not irrelevant, but superfluous. Indeed, at this point, what is important to know is that after discovering all sorts of broken objects caused by the explosion, the protagonists finally could find one in one piece. But in fact, for us to know that it is the "longue-vue" will be very important because of its role later on in the text. The identification made by way of the inserted clause plays a role on the whole narrative level, even though the text is used in can't expand directly from it, because of the inserted clauses' properties. Also, notice the use of the definite article in the inserted phrase: it accentuates the complicity of the narrator with the listener/reader, with whom he shares the knowledge of the existence of this object. Another way for the literary text to distort the required exclusion of the IC is by building a text with inserted clauses in correspondance: (23)

II y exposa sur une sorte d'autel — comme autant d'idoles — et contre les murs — comme les armes de la panoplie de la raison — les etalons du pouce, du pied, du yard... (Tournier, Vendredi p. 70) ['In it he displayed, on a sort of altar, as though they were idols, and on the walls — like a panoply of the weapons of reason — the measures of an inch, a foot, a yard...' (p. 68)]

The narrator delivers us the keys to understanding Robinson's actions; the contrast between abstract words and the concrete ones in the main narration turns the IC into an interpretation of x, while the use of "comme" suggests that he is making a comparison, an analogy with other mentions of the contrast idole/raison in the text. The reiteration of the form contributes to specify the role of the narrator. 6.2. Because the occurrence of the inserted clause is maintained separated from the main textual movement, it is preferably used to express another

Inserted clauses

389

point of view in the text. In Vendredi ou les limbes du Pacifique, there are explicit interventions of the narrator: (24)

Mais il soupgonnait ce dernier propos-qui repondait si peu en verite ά 1'esprit de l'ile administree — de lui avoir ete sourdement inspire par "I'autre He"... (Tournier, Vendredi, p. 153) ['But he suspected that his last idea, not really compatible with the true spirit of the cultivated island, was subconsciously inspired by his longing for the "other island" ...' (p. 144)]

In doing so, the inserted clause contributes not only to the utterance EE' or the paragraph but to the text as a whole, in creating the fictional world. The omniscient narrator delivers information to us and the inserted clauses play an important role in this process. They mostly express a supposedly objective information a b o u t Robinson and his world: it makes contrast with the non-inserted clauses where the world is seen through the eyes of Robinson, exposing his thoughts, his feelings, most of the time confused, a b o u t the world. Other inserted clauses relate to other events or themes previously exploited, as if the narrator wanted to make the listener/reader understand the character better. By specifying a few details concerning time, place and characters, the IC Orientes the reader towards the context and more precisely towards the fictional world. It transforms the usual plot exposition by restating details that the characters might already know and that the narrator can transmit directly and almost secretly to the reader, expressing an evaluative point of view of the characters' actions, in other words, establishing a distance f r o m the narrative. Distance doesn't necessarily mean that the n a r r a t o r doesn't take in charge the orientation, but that the IC allows him, besides other means, to consider the fiction f r o m the outside. He may adopt the character's point of view and reinforce it, but he always does it with a critical perspective, taking the role of the judge w h o has access to truth, to objectivity (see example 24). A parallel narration is present in Robinson's diary, the log-book. When Robinson is in the role of the narrator, the IC allows the character to explain himself, to m a k e explicit the principles, the basis of his reasoning. Considering Robinson's situation, it would seem abusive to interpret these writings as directed towards an eventual reader: let's say that as a mean to clarify his own thoughts, they present the same characteristics as if they were directed towards a reader. The character — narrator stands back in an analytic role.

390

Danielle

Forget

In general, although the inserted clauses usually establish a formal link with a particular passage of the text, at the whole textual level, they contribute to the construct for both the role of the narrator, as was illustrated in example (24), and the presence of a listener/reader. Many passages are for the listener/reader's benefit and the internal construction also signals it, be it by the negative form or by the use of connectives referring to the possible interpretations that are to be rejected. The distance conveyed by the IC in general reaches its highest point when the narrator controls the listener/reader's possible interpretation of the discourse and prevents erroneous conclusions.

7. Conclusion The inserted clause is characterized by the separation from the E-E' that it creates, which corresponds, in pragmatic terms, to the distance that the speaker/narrator establishes from the current discourse. This distance may be explicitly signaled by the presence of metadiscursive expressions or deduced from the linguistic context. Even when the IC is interpreted as apparently specifying or completing the context of χ in the E-E', the fact that it relates to the strategies the speaker uses in delimiting the range of conclusions, attitudes and reactions to which his message must be given priority. Its secondary role — that of reformulating x, giving the justification of the arguments that prevailed to arrive at a certain effect, adding superflous details and presenting an attitude towards χ is not to be considered as more significant than its primary role — that of controlling the pragmatic orientation of the message. Since the IC makes reference to x, the tendency may be to assume that there is paraphrastic relation between the elements of the IC and the EE', that would be linked to a stylistic effect of insistance. But very few examples present themselves under this relation of identity and it was seen that usually the substitution of one element with another is impossible. The function of the IC to contribute to the development of the message, providing the addressee information that he might not have, has been noted several times but it has frequently been associated with an "imperfect message", the speaker neglecting to take into account informational elements and having to make some sort of "second try". In this paper the inserted clause has been presented in a different way: as a strategy, serving both the purpose of argumentation and narration,

Inserted clauses

391

for the speaker to prevent some selection of meaning, inferences, conclusions to take place. It guides the interpretation and expresses a norm that the text itself sets up. In agreement with the metadiscursive role that is attributed to it, the IC acts at a different level from the E-E'. Taking χ as a theme, on the mode "when I say χ, I mean x"', it refers to the E-E' as an enunciative event that is taking place, whence its deictic aspect. Of course in doing so, the speaker seems to specify what he has in mind, but what is more important is that he anticipates the possible interpretation of the message as a whole by excluding, directly or indirectly, some orientations. The role of the addressee is dominant and always present because it is to him that the IC are directed. While in the correction, the act is composed of instructions to substitute one member for another, the IC requires that the addressee takes into account the two propositions, because it is the relation between the two that is essential. The contrast (between χ and "about x") is better measured by the particularity of the point of view that it allows in the literary text. Vendredi et les limbes du Pacifique was used to illustrate the difference of level between the IC and the E-E'. This difference is developed by the confrontation of consciousness, not between two protagonists but in one protagonist where the IC corresponds to an analytic role that Robinson is involved in, or, for the narrator, the task of presenting in a convincing and intelligible way for the reader Robinson's world. In both cases, the distance from the narration created by the use of inserted clauses produces a direct contact with the reader. With the narrator, this contact is even more evident because the text creates the image of a listener/reader that has to be informed of the rules in Robinson's world, of details — temporal, locative — of the fictional world; the narrator acts as an intermediary — creating the illusion that he reports an event which he has witnessed. The inserted clauses have a great responsibility in clarifying the role of the narrator: a role of authority linked to the knowledge of Robinson's internal and external universe, of judgments, of commentaries.

Notes 1. The literary text I am using is Vendredi ou les limbes du Pacifique, by Michel Tournier, translated in 1985 by N. Denny; I will add examples from Tournier's Le Coq de Bruyere and Trou de memoire, of Hubert Aquin. My research on the oral corpus has been done on a selection of examples provided by Carole Laurin (U. de Montreal) on the Montreal corpus '84, by P. Thibeault; D. Vincent; W. Kemp; D. Sankoff. I wish to thank Carole

392

2.

3. 4.

5.

6. 7. 8. 9. 10. 11.

12. 13.

Danielle Forget Laurin for having made these examples available to me. I wish also to thank L. Butters, G. Losier, L. Travis and 0 . Ducrot. An example of this type of definition is in Grevisse (1969): "La proposition incise ou intercalee est une proposition generalement courte, tantdt inseree dans le corps de la phrase, tantdt rejetee ä la fin de la phrase, pour indiquer qu'on rapporte les paroles de quelqu'un ou pour exprimer une sorte de parenthese..." (p. 127). Another problem comes from the terminology used. Bonnard, for example, refers to these two aspects just mentioned, by respectively using "incidentes" and "incises" but he finally seems to interchange the terms. Cf. Bonnard 1985: 2 9 3 - 2 9 4 . Cf. Ziv 1985: 181. Though I follow these authors in delimiting the IC using both the intonational and the syntactical criteria, the literary texts with which I am concerned contain a superimposed mark, the punctuation, more specifically, the dashes. It is important to note that, since we do not want to generalize and say that all relative and appositive clauses count as inserted clauses, it has the further condition of being marked intonationally. This also seriously compromises the possibility of characterizing the inserted clause structurally. Forget (1989) "Les incises: un point de rencontre entre la syntaxe et la pragmatique", paper presented at the Canadian Learned Society Congress, Quebec, May 1989. Cf. D e l o m i e r - M o r e l 1986: 146. See Delomier —Morel 1986: 151 for an analysis of these connectives in their corpus. Cf. Tamba (1987) for a specific study of ou. Cf. Authier 1987: 35. In a similar way, "parce que" may introduce "an answer to an implicit question" and give an explanation of a previous assertion; cf. Ducrot 1972: 125. Ducrot (1972) states two conditions for texts: one of coherence and the other, of progression: "II est interdit de se repeter: chaque enonce est cense apporter une information nouvelle, sinon il y α rabächage". (p. 87) Note the correspondance of such a view with the one on phatic elements that refuses them any impact on the semantic content of the discourse. Some of the inserted clauses may even be typical of the written level: "Je η'attends pas beaucoup de raison d'un homme de couleur — de couleurs, devrais-je dire, puisqu'il y a en lui de l'Indien et du negre." (Vendredi p. 154)

References Anscombre, Jean-Claude 1983 "Pour autant, pourtant (et comment): ä petites causes, grands effets", Cahiers de linguistique franfaise 5: 37 — 85. Aquin, Hubert 1968 Trou de memoire. Ottawa: Le Cercle du livre de France. Authier-Revuz, Jacqueline 1987 "Modalite autonymique et pseudo-anaphore deictique", Cahiers de lexicologie vol. 51, II: 1 9 - 3 7 . 1987 "L'auto-representation opacifiante du dire dans certaines formes de 'couplage'", DRLAV 3 6 - 3 7 : 5 5 - 1 0 3 .

Inserted clauses

393

Benveniste, Emile 1966 Problemes de linguistique generale. Paris: Gallimard. Berrendonner, Alain 1981 Elements de pragmatique linguistique. Paris: Minuit. Bonnard, Henri 1985 Code du frangais courant. Paris: Magnard. Delomier, Dominique 1989 "Le role des incises dans l'interaction", Interaction, Association des sciences du langage, BUSCILA, 5 6 - 6 6 . Delomier, Dominique — Marie-Annick Morel 1986 "Caracteristiques intonatives et syntaxiques des incises", DRLAV34—35: 141-160. Dessaintes, Maurice 1960 La construction par insertion incidente. Paris: Editions d'Artrey. Dik, Simon C. 1981 Functional Grammar. Publications in Language sciences, 7. Dordrecht: Foris publications. Ducrot, Oswald 1972 Dire et ne pas dire. Paris: Hermann. 1984 Le dire et le dit. Paris: Minuit. Ducrot, Oswald — et als. 1980 Les mots du discours. Paris: Minuit. Emonds, Joseph 1976 "Appositive relatives have no properties", Linguistic Inquiry 10: 211—243. Fayol, Michel 1985 Le ricit et sa construction. Paris: Delachaux and Niestle. Grevisse, Maurice 1969 Le bon usage. Paris: Hatier. Kleiber, Georges 1986 "Deictiques, embrayeurs, 'token-reflexives', symboles indexicaux, et comment les definir?" L'information grammatical, 30: 3 — 22. Levinson, Stephen C. 1983 Pragmatics. Cambridge: Cambridge University Press. Lyons, John 1977 Semantics. Vol. I and 2. Cambridge: Cambridge University Press. McCawley, James D. 1982 "Parentheticals and discontinuous constituent structure", Linguistic Inquiry 13, 1: 9 1 - 1 0 6 . Murat, M. Cartier — B. Bresson 1987 "C'est-ä-dire ou la reprise interpretative", Langue franQaise 73: 5 — 16. Nemni, Monique 1981 "L'identification de l'incise par l'intonation", Studia Phonetica 18: 103 — 112.

Rioul, Roger 1983 Tamba, Irene 1987

"Les appositives dans la grammaire franpaise", L'information cale 18: 2 1 - 2 9 .

grammati-

"Ou dans les tours du type: un bienfaiteur public ou evergete", Langue franfaise 73: 1 6 - 2 9 .

394

Danielle Forget

Tournier, Michel 1972 Vendredi ou les limbes du Pacifique. Paris: Gallimard. [Translated by Norman Denny, 1985] New York: Pantheons Books. 1978 Le Coq de Bruyere. Paris: Gallimard. Ziv, Yael 1985 "Parentheticals and functional grammar", in: A . M . Bolkenstein — C. de Groot — J. L. Mackenzie (eds.), Syntax and Pragmatics in functional grammar. Dordrecht: Foris publications.

Part IV Pragmatics and comprehension of individual text types

Teaching conscientious resistance to cooperation with text: The role of pragmatics in critical thinking* Susan Meredith Burt

1. Introduction The "pragmatics of text" may sound to people who are not linguists or literary theorists like a rather arcane and obscure field of inquiry, but "cooperating with text", describes a problem confronted, I think, by all readers. A reader wants to cooperate with a text enough so as to extract the writer's intended meaning successfully, or some approximation thereof; however, a reader does not want to cooperate with text so much as to be manipulated by the writer. The purpose of this paper is to explain the rationale behind a proposal for teaching college students to walk the fine line between these two types of cooperation with text. I shall propose that raising students' consciousness about the expectations for clarity expressed in the Cooperative Principle and maxims of Grice (1975) can improve their critical abilities in reading and writing texts. American teachers of freshman composition, that is, of courses in essay writing, are of necessity concerned with text, and with students' interaction with text, either in the consumption or the production thereof. Part of the assignment in teaching composition is to teach students to look out for logical flaws in texts that they read, and to avoid logical flaws in texts that they produce. However, one can quickly become dissatisfied with the brief exposure to logic that composition textbooks give. Though most textbooks contain a discussion of the basics of logic, including a look at syllogisms such as the old chestnut in (1): (1)

All men are mortal. Socrates is a man. Therefore, Socrates is mortal.

* Georgia Green, Kent Machina, and Betty MacMahan provided comments on an earlier version of this paper. They are not responsible, however, for any failure on the part of the author to fulfill reader expectations of quality, quantity, relevance or manner.

398

Susan Meredith

Burt

the baptism in logic consists of a sprinkling on the forehead rather than an immersion. Supplementing the glance at logic-as-syllogism, most textbooks also include lists of logical fallacies, that is, various ways in which an argument can go wrong — from perhaps four to up to seventeen or more, with examples of varying degrees of clarity, to guide the student. Some of the most able students study this list, successfully identify fallacies in texts, and write reasonably logical critiques of assigned readings such as the excerpt from a student essay in (2). (2)

Teller's facts are wonderful, but he gives no source for them. He also uses two fallacies in the essay; the either-or-fallacy, and the appeal to ignorance. On the surface, Teller's essay is informative and has good intentions, but upon closer examination, the reader will note the flaws in his argument and how his intentions become tainted.

But most students do not do so well; I suspect that they find the fallacy lists both formidable and unlearnable. This paper will not discuss alternatives to the teaching of logic and syllogisms; Fulkerson (1988) has already done so. Furthermore, composition teachers have a great deal to keep track of, so I do not propose to increase the classtime devoted to logic, but rather the effectiveness of the time so devoted. My proposal is that explicit teaching of the principle and maxims of Grice (1975) can aid in the understanding and learning of logical fallacies — indeed, that fallacies can be seen as intentional and manipulative violations of the maxims. This approach to the fallacies will be discussed in the first half of the paper. Second, I hope to show that learning Gricean maxims will also aid students in learning to discern other types of manipulative passages in argumentative texts, items that are not covered in the lists of fallacies that students are now taught. While I will discuss this problem in the context of teaching American college students to read and write critically, the remedy I propose is applicable for other readers who are confronted with manipulative text.

2. Grice's model of conversation One obvious question that might arise is whether Grice's model, a model for conversation, is appropriate for analyzing text. In light of the differences between text and talk discussed, for example, in the papers in

Critical thinking

399

Tannen (1982), it might be argued that a conversational model is inappropriate as an approach to reading and writing. However, there are plenty of reasons to believe that Grice's model can provide a fruitful approach to argumentative text. Grice's concern is to provide a model of cooperative activity in general, including discourse, of which conversation is only one type. Whether it is conversational or written, argumentation usually presupposes cooperation with an audience (even if the audience is oneself)· Evidence from Grice himself supports my claim that the conversational model is applicable to an enterprise like argumentation. Two of Grice's "features that jointly distinguish cooperative transactions" (Grice 1975: 48) characterize the acts of writing and reading text. In the writing/ reading transaction, "the participants have some common immediate aim [although] their ultimate aims may, of course, be independent and even in conflict." (Grice 1975: 48) Thus, just as a writer has the aim of imparting his views to others, those others, as readers, have an interest in having those views imparted to them, although these readers and writers may ultimately have different aims. Similarly, in writing and reading, "the contributions of the participants should be dovetailed, mutually dependent." (Grice 1975: 48) Writers must cooperate with readers, for example, by choosing a register or style to write in that meshes with what the writer knows or guesses about audience background knowledge and expectations. Readers in turn must cooperate with writers in a variety of ways, such as suspending disbelief for the sake of a story, or assuming the truth of a proposition for the sake of argument. Writers' and readers' functions are clearly mutually dependent. Grice himself notes that writing is an exception to his third feature, the "understanding that the transaction will continue until both parties agree it will terminate." (Grice 1975: 48) Both his noting of the exception and his use of an example of writing to illustrate one of his maxims make clear that Grice, at least, saw his model as applicable to writing. Writing argumentative prose would therefore clearly fall under the rubric of purposive activity whose felicitous accomplishment is characterized by Grice's Cooperative Principle and maxims, which are summarized in (3): (3)

The Cooperative Principle: Make your conversational contribution such as is required, at the stage which it occurs, by the accepted purpose or direction of the talk exchange in which you are engaged.

400

Susan Meredith

Burt

Quantity 1. Make your contribution as informative as is required (for the current purposes of the exchange). 2. Do not make your contribution more informative than is required. Quality 1. Do not say what you believe to be false. 2. Do not say that for which you lack adequate evidence. Relation Be relevant. Manner 1. Avoid obscurity of expression. 2. Avoid ambiguity. 3. Be brief. 4. Be orderly. (Grice 1975: 5 6 - 6 ) Grice's claim is that these maxims shape our expectations in conversion or in other types of discourse; when these expectations are not fulfilled (and they frequently are not), listeners or readers try to figure out rational explanations for the gaps between expectation and reality thus produced — these explanations give rise to implicatures of various types, according to Grice. Grice describes four ways in which speakers can fail to fulfill the expectations described by the maxims: (1) they can violate a maxim; (2) opt out of it; (3) fail to fulfill it because it clashes with another maxim; or (4) they can flout or exploit it for purposes of making an implicature. While implicatures arise because a maxim is flouted, most of the traditional fallacies can be seen as violations of the maxims; that is, the speaker or writer "may quietly and unostentatiously VIOLATE a maxim; if so, in some cases he will be liable to mislead." (Grice 1975: 49) Writers can attempt to mislead readers with fallacies of quality, of quantity, of relation, and of manner. These are described next.

3. Fallacies as violations of the maxims Green (1989: 88) notes that the maxims may sound like "composition teachers' futile rules." As a composition teacher, I propose in this section to use them not as rules, and hopefully, not futilely, but showing what can happen in argumentative writing when the maxims are violated. I

Critical

thinking

401

think this use of the maxims may be justified because the maxims have been called "maxims of clarity" (Lakoff 1977: 87), and clarity, of course, remains one major aim of composition instruction. This section will outline how the traditional fallacies can be classified as violations of the four Gricean maxims and submaxims. Machina (personal communication) has pointed out that all fallacies can be seen as violations of the second maxim of Quality — in other words, an argument which is fallacious is ipso facto one with inadequate evidence; however, as some examples will show, a fallacy may violate more than one maxim. Characterizing this additional violation can help us fit the fallacies into a classification scheme based on the maxims. While this scheme may not necessarily replace the old fallacy names, I hope it can serve to organize the fallacies into more easily learned superordinate categories, and thus address the pedagogical problem described above.

3.1. A fallacy of Quantity One fallacy, circular reasoning (or begging the question), can be seen as a violation of the first maxim of Quantity. In reasoning circularly, the arguer fails to "make her contribution as informative as is required." The student example in (4) illustrates: (4)

It is my opinion that since it is the woman's right to have an abortion, the decision must be entirely up to her. Even if everyone does not agree with the woman's choice, it is illegal to prohibit any woman from exercising her rights. The government must protect these rights because they are guaranteed to everyone by law.

The student argues, essentially, that abortion is lawful because it is now legal. She has failed to address the question of whether abortion ought to be legal, which is, of course, the issue in the on-going dispute. Failing to address that issue can be seen as a failure of Quantity. In using circular reasoning, she has failed to be adequately informative. At the same time, she has repeated herself — which one does when arguing circularly — which looks like a case of violating the second Quantity maxim. This example may be a case where the traditional fallacy name is quite descriptive and should still be used, but the inadequately informative quality of the argument that results should be pointed out to the student as a violation of the Quantity maxim.

402

Susan Meredith

Burt

3.2. Fallacies of Quality Fallacies of Quality abound. In reading a text, it may not always be possible to say whether lapses in Quality arise from a deliberate attempt by the author to mislead — if they do, they are violations of the first maxim of Quality, "Do not say that which you believe to be false." One example of a deliberate violation of the first Quality maxim is a case of what is traditionally called false analogy, a fallacy, from a Sears and Roebuck advertising circular: (5)

Unless you've used your Sears Charge account recently, you could have as much as $ 1 300 or more to spend at Sears on anything we sell. You could almost think of it as found money to do with as you please.

Only the hedge almost, probably inserted by the legal department, keeps this from being a true example of violation of the first Quality maxim; using a credit card is not like finding money, because credit card purchases, unlike "found money", must be paid back. The comparison looks like a deliberate attempt to mislead, i. e., to write that which the writer believes — indeed, knows — to be false; such an attempt is a violation of the first Quality maxim. Similarly, many examples of violations of the second Quality maxim, "Do not say that for which you lack adequate evidence," can be found, not suprisingly, in student papers. Example (6) illustrates what would traditionally be called "the slippery slope fallacy," which can also be seen as a violation of the second Quality maxim — the student writer lacks evidence to support the evils that he predicts would follow from single teen parenthood: (6)

Take for example, young girls that are still in college and have no source of income that get pregnant ... Since she does not have a steady income to support the baby, it would be kind of selfish of her to have the baby and no way to support it. Let's say she decides to have the baby. There will be great pressure put on her. The pressure will come from still being in college, no income, not being married, and the many responsibilities of having a baby. From all of the pressures, the mother could come to the point of abandoning the baby. As it is now, there are too many discarded children in institutions that the state takes care of because their parents couldn't handle the pressures of raising them.

Critical thinking

403

While this student's argument has the virtue of being original, it lacks any evidence to back up the rather far-fetched claim that childbirth out of wedlock will inevitably lead to overcrowded state orphanages. The lack of sufficient evidence involved in a slippery slope case, such as this one, justifies seeing this as a violation of the second Quality maxim. Another violation of the second Quality maxim is the either-or fallacy, in which the arguer claims that one of only two possible events could take place. The student who wrote (7) succumbed to this fallacy: (7)

If we didn't use the bomb on Hiroshima, we would have used it on someone else.

Of course, use of the bomb was not inevitable, but the result of conscious decision-making; no evidence is really available for the claim made in (7), and the student does not provide any. Machina (personal communication) pointed out that we should assume another maxim to be at work in argumentation, to the effect that one should offer the best evidence one has. In example (7), the student has either violated this maxim or the second maxim of Quality — or both. Besides false analogy, slippery slope and the either-or fallacy, other fallacies of Quality include use of a false premise, post hoc ergo propter hoc, "many questions" (as in "Have you stopped beating your wife?"), protecting the hypothesis, ill-founded generalizations, and the appeal to ignorance. All of these present or rely on statements which are either false or for which evidence of truth is lacking. The author who uses these fallacies has either written something which he knows to be false, violating the first Quality maxim, or something for which he lacks adequate evidence, violating the second Quality maxim.

3.3. Fallacies of Relevance Fogelin (1987: 86) and Walton (1987: 219) have noted that there are a number of related fallacies, all of which fail in a similar way — they fail to be relevant. Relevance, of course, is both audience — dependent (see, for extensive discussion of relevance, Sperber and Wilson 1986), and topic - dependent (Dascal 1977: 315). Nevertheless, as Dascal (1977: 309) points out, people trust their intuitions about what is relevant to a topic, and composition teachers have their share of such intuitions. 1 One obvious failure to be relevant involves the fallacy traditionally called distraction, in which the arguer tries to convince the audience on

404

Susan Meredith

Burt

points not really pertinent to the issue under discussion — in composition, this point is probably the issue on which an essay was assigned. The student who wrote (8) was assigned to write a critical response to an essay by Ronald Reagan on the topic of abortion, but the student attempted to distract the audience from the assigned topic by writing about the Hiroshima bombing. (8)

When Reagan brings in the start of the Declaration of Independence and starts to talk about us being a nation that does not "want to play God with the value of human life. It is not for us to decide who is worthy to live and who is not," he must have forgot about the time we dropped the nuclear bomb on Japan. Who was playing God then? We destroyed two major cities and wiped out almost all life around the area where the bomb had hit. If that is not taking control of life into someone's hand then I don't know what is. As it says in the Declaration of Independence: "We hold these truths to be self-evident, that all men are created equal, that they are endowed by their Creator with certain unalienable rights, that among these are life, liberty, and the pursuit of happiness." As you can read above that all men are created equal not just American men but men all over the world, then who gave the United States the right to kill all those Japanese? With abortion, the women are trying to do the right thing.

One can see what the student is trying to do here: he is trying to discredit Reagan's claim of being pro-life, by calling attention to Reagan's support of the arms race. But the student misses in this attempt; the Hiroshima bombing was not during Reagan's administration, and is consequently quite an excursus away from the assigned topic, responding to Reagan's view on abortion. The student compounds this violation of relevance throughout this part of the text, so that his ultimate return to the topic looks bizarrely out of place. For the whole passage, the main problem is that he has steered away from the topic and failed to be relevant. Other fallacies which violate relevance are the non sequitur; poisoning the well; the ad populum; ad misericordiam; ad hominem; name-calling; the genetic fallacy (though see section 4.3., below); and appeal to the wrong authority.

Critical

thinking

405

3.4. Fallacies of Manner Fallacies of Manner are a bit difficult to find; the so-called fallacy of amphiboly would certainly seem to be a violation of the maxim, "Avoid ambiguity." An amphiboly is produced when a writer is unaware of scope, syntactic, or lexical ambiguity or even vagueness within a sentence, as in the instructions in the laundromat recorded by Walton (1987: 213), "When the machines stop, customers should remove their clothes." However, it is not at all clear that cases of amphiboly actually reflect or produce lapses in reasoning, so they probably play little role in argument. However, it is possible to see cases of the fallacy of personified abstraction as violating the maxim, "Avoid obscurity of expression." The student who wrote (9) indulged in personified abstraction: (9)

America has always been fighting for equal rights, such as equal rights for women and blacks ... What about the rights of the mother? Rights, such as freedom of choice, property, and her right to liberty are guaranteed to all mothers as well.

In this passage, America evokes an idea of the full force of the United States government, with the whole-hearted backing of its people, engaged in the struggle for equal rights, including the right to choose whether or not to have children. Given that the struggle has been, and remains, an internal one, with some opposition within the society, using a phrase like America to describe the protagonist in the struggle for equal rights results in a passage that is at least misleading. To the extent that the personified abstraction, "America", is used intentionally to gloss over the internal nature of the struggle for equal rights, as I think it is here, it qualifies as a violation of the first maxim of Manner. Similarly, the fallacy called equivocation, shifting the definition of words mid-argument to suit the arguer's purposes, clearly violates the maxim, "Avoid ambiguity." One example of equivocation is given in (10): (10)

At least Molly Yard, president of the National Organization for Women, regrets the switch to single-issue voting, but she says, "I don't think we have a choice." It is ironic to hear "pro-choice" people say they are "no-choice" single-issue voters. This approach is a mistake that could be costly to a wellfunctioning government. ("Don't let abortion be controlling issue": 1989: A6)

406

Susan Meredith

Burt

The writer here seems to be calling into question the consistency of the National Organization for Women's position, with her use of pro-choice and no-choice; while the writer would like to lead readers to believe that the same kind of choice is at stake, choices in voting are clearly different from choices in child-bearing. The writer of (10) is playing on the use of choice in both cases to try to lead the reader to infer that the position of the National Organization for Women is inconsistent; this looks like an intentional violation of the maxim against ambiguity. While I am not aware of other fallacies of manner, composition teachers are well aware that frequent textual faults result from failures to be brief and orderly.

3.5. Summary Since one of the things students need to learn in writing courses is the standard of quality and quantity needed for successful argumentation, it is not surprising that violations of the maxims, such as most of the examples so far, occur in their papers. We can assume, if we wish to be charitable, that these violations are due simply to students' ignorance. More sinister than the student examples, of course, is the example from the Sears circular. The false analogy of example (5) is not due to ignorance on the part of advertising writers, but is indeed deliberate; the presence of the hedge almost in the text shows the writer's (or legal department's) awareness of the questionable quality of the analogy. Writers who use fallacies on purpose are deliberately not following the Cooperative Principle. Unsophisticated writers who fall into fallacies need to be aware that their adherence to the Cooperative Principle will be questioned. What I have claimed so far, that fallacies are failures to meet expectations for communication such as those outlined by Grice, is not surprising. Explicit teaching of those expectations to composition students and to other arguers, however, remains a pedagogical problem. Fulkerson (1988: 439), for example, has proposed a metric for evaluating the suitability of support in argumentative prose; his STAR method encourages students to ask whether evidence is Sufficient, Typical, Accurate, and Relevant. This looks a bit like what I have been proposing so far. However, the Gricean perspective I propose offers students more refined tools for critical reading as well, because the Gricean perspective allows readers to see, besides fallacies, three further kinds of manipulation in text: a) cases where a maxim is intentionally violated, although no traditional fallacy is overtly committed; b) cases in which a writer opts

Critical thinking

407

out of a maxim with the intent to mislead — a type of rhetorical move not on the fallacy lists; and c) cases in which the writer flouts or exploits a maxim, giving rise to an implicature, which, if made explicit, would be fallacious in the argument it appears in. These kinds of textual problems, with examples from more sophisticated writers than my students, will be discussed next.

4. Non-fallacy violations of the maxims, opt-outs, and flouts While the purpose of teaching students of composition about fallacies is to innoculate them against manipulation by writers and to encourage them to write in a non-manipulative way, clever writers can succeed in being manipulative without being so obvious as to commit a fallacy. One way to do this is to opt out of one of the maxims; another is to violate a maxim without committing a fallacy; a third way is to flout a maxim, giving rise to an implicature. While there are innocent opt-outs and flouts, there are manipulative ones as well, which students — and other readers — need to be aware of.

4.1. Non-fallacy maxim violations One way for a writer to fail to fulfill a maxim is to violate the maxim but, in contrast to the examples in section 3, without committing an overt fallacy. One very common way to do this, in talk or in text, is to ask a rhetorical question, and to use that question in lieu of providing adequate evidence for the proposition the question purports to question. The reader or hearer is led to infer a proposition which is not argued for. This results, I think, in a violation of the first maxim of Quantity, and of the second maxim of Quality, as shown by example (11): (11)

We have negotiated for 25 years, and the results are readily visible. Why would a totalitarian empire that depends on military force to maintain its power voluntarily disarm itself? (Teller 1981: 216)

The writer, Edward Teller, here overtly commits the fallacy of "many questions," in that he presupposes a proposition that ought to be argued for — in other words, he asks a loaded question. But he also substitutes a rhetorical question for actual argument for the main proposition; the

408

Susan Meredith

Burt

reader is invited to infer that the totalitarian empire in question would not indeed disarm itself, although this is also a proposition that ought to be argued for. Teller fails to be as informative as required, and also fails to give adequate evidence for what he intends to lead readers to believe. Use of rhetorical questions is not usually described as fallacious in textbooks addressed to writing students, and indeed is not necessarily fallacious in itself. However, using a rhetorical question in place of evidence, as Teller does, can be seen as an intentional violation of the first maxim of Quality (i. e., as being inadequately informative) and of the second maxim of Quality (i. e., as having inadequate evidence). The example illustrates that a clever writer can use maxim violations — which are usually innocent — as well as traditional fallacies to mislead readers. Alerting readers to the possibility of manipulative non-fallacy violations of maxims in text will further encourage their resistance to cooperation with manipulative text.

4.2. Opting out of a maxim Grice notes that speakers occasionally opt out of maxims, and Green (1989: 90) has noted that there are conventional signals for doing so: for example, "by the way" signals that the speaker is about to opt out of the Relation maxim. This kind of signaling can occur in text as well as in talk. A footnote, for example, is an innocent and conventional way of opting out of the Relation maxim in text; placing the footnote's contents at a remove from the rest of the text signals that the writer considers those contents as somewhat irrelevant to the main flow of the argument. However, there are also more sinister ways of opting out of a maxim, as in example (12): (12)

Neither the United States nor the Soviet Union publishes information on its current arsenals, and secrecy laws prevent me from discussing even the available estimates. There is, however, an officially released fact: between 1966 and 1981 the total megatonnage of the American nuclear arsenal was reduced to less than one-half its former size. The Soviet arsenal has rapidly increased in yield, accuracy and diversity during the same period and currently includes a total nuclear explosive power in excess of what the United States ever had. (Teller 1981: 215)

By focussing on the "officially released fact" of megatonnage, Teller here tries to lead the reader to think that this is all that the public is permitted

Critical

thinking

409

to know about arsenal comparisons. However, by 1982, a great deal was being published about relative missile counts, launchers, MIRVed warheads, and so on; I suspect that the information was available in 1981, when Teller wrote this. Teller has opted out of the first Quantity maxim; with his reference to "secrecy laws," he has signaled readers that he is not going to be as informative as is required. Of course, this kind of reference to national security is an old ploy, but unsuspecting readers may cooperate with this text to the point of believing it. Not all opting out is manipulative — in fact, most is not. Writers and speakers frequently opt out of Quantity because of time and space constraints. But dishonest opt-outs, which is what I think Teller is up to, manipulate readers and hearers. Educating readers to awareness of this kind of opting out of the CP and maxims may raise their resistance to cooperating with manipulative text by teaching them to recognize precisely what a writer like Teller is doing when he signals that he is opting out of a maxim.

4.3. The flouting of maxims The flouting, or exploiting of maxims, of course, gives rise to "classic" Gricean conversational implicature. The speaker appears to violate a maxim of conversation, but in fact, the hearer can derive further meaning from the utterance by calculating to fill in the "gaps" between expectation and the reality of the utterance — and the speaker knows that the hearer can do this. One of Grice's classic examples illustrates this in (13): (13)

A: Smith doesn't seem to have a girlfriend these days. B: He's been going to New York every weekend.

While Β seems to have violated the Relevance maxim, A can use this apparent violation to derive further meaning — that Smith has a girlfriend in New York, for example — and Β knows that A has the capability to do the necessary calculating to figure this out. Β can be said to have implicated conversationally that Smith has a girlfriend in New York. Green (1989: 101 ff.) has shown that this kind of inference is also done by readers of texts and that writers rely on readers' abilities to make these inferences (Green claims that what produces coherence in a text is a successful match between writers' setting up and fulfilling of expectations, and readers' abilities to draw appropriate inferences to produce those expectations). In argument, one of the traditional fallacies relies on

410

Susan Meredith

Burt

readers' abilities to make connections that are not overtly stated. The genetic fallacy, also called "guilt by association," illustrated by the example in (14), relies on the reader to connect apparently irrelevant facts by means of an implicature. (14)

The Committee in Solidarity with the People of El Salvador, or CISPES, has been turning up the volume since last autumn. CISPES was formed in 1980 after its founding members spent time working with Farid Handal, the brother of Salvadoran Communist Party Chief Shafik Handal. ("The Harkin Doctrine" 1989: np)

In (14), the reader is led to connect CISPES with a prominent Salvadoran Communist, and is invited to infer that CISPES is a communist organization, because the remark about the Handal brothers' family relationship, as it stands, otherwise would seem to be irrelevant. Thus, the genetic fallacy flouts, and by doing so, exploits readers' expectations that Grice's Relevance maxim will be followed in a text. The genetic fallacy uses readers' inference abilities in a manipulative way. Not all uses of these abilities are manipulative — most are not. However, a clever writer can further exploit reader expectations that the maxims will be followed without committing a traditional fallacy, but by leading the reader to infer a fallacy, as Caldicott does in example (15): (15)

[The Manhattan project scientists] were not sure before they blew up the bomb that the whole atmosphere would not go critical. They were worried about this probability, so they redid their calculations and the probability remained the same. It was not extremely small. One technician was upset to hear Enrico Fermi taking side bets as the "gadget" was hoisted to the tower, that New Mexico would be incinerated. After the explosition, the radioactive cloud hovered overhead for some time, worrying the scientists, because if it did not blow away in the direction they had prescribed, it could have killed them or injured them severely. However, it did eventually blow away. That night the scientists had a party (Caldicott 1981: 227).

Caldicott's comment about the party does not constitute a fallacy by itself but the sentence presents an apparent violation of the maxim of Relevance. Consequently, the reader can be expected to enter into a calculation such as in (16)

Critical thinking

(16)

411

"I assume that Caldicott is adhering to the Cooperative Principle and the maxims. However, the comment about the scientists having a party would seem to violate the Maxim of Relevance. People have parties because they want to celebrate, so perhaps the scientists were celebrating something. The explosion of a nuclear weapon is not usually considered an event to be celebrated, at least not by people who dislike death and destruction. Yet the scientists celebrated, so perhaps they were unaware of the horrible implications of exploding a nuclear weapon. Caldicott knows that I am capable of working out this last step, so she probably meant to implicate that the scientists were unaware of the consequences of their actions."

Later in the text, Caldicott tells of J. Robert Oppenheimer's later regrets, giving support to this interpretation. Caldicott sets up an expectation by means of an implicature, an expectation which she later fulfills. The calculation in (16), like other Gricean implicatures, is worked out by the reader quickly and without conscious effort. What is interesting here is that the result of this calculation — that the scientists were insensitive — is what traditionally would be called an ad hominem remark — an attempt to discredit the creators of nuclear weapons, whom we can presume to have been proponents of nuclear weapons, at least at the time. Now, whether the result of this calculation itself violates one or more of the Gricean maxims depends on the role that the reader sees this proposition (that the scientists were insensitive) playing in the flow of argument as a whole. If one sees Caldicott's purpose as condemning the nuclear arms race, the implicature derived from (15) seems relevant. An audience unfriendly to her purpose, however, might well be inclined to interpret the implicature of (15) as leading to an irrelevant side-issue — such an audience might well be offended at the last sentence of (15) and the implicature derived from it. The relevance of any proposition, asserted or implicated, depends on various aspects of context, including the purpose of the argument and beliefs of the audience. What is interesting about (15), however, is that a seeming violation of the maxim of relevance is actually a flout or exploitation of that maxim which gives rise to an implicature which is itself subject to evaluation in terms of its adherence to the Cooperative Principle. Implicatures can themselves be judged to be fallacious, or in terms preferred here, can themselves be violations of maxims.

412

Susan Meredith

Burt

4.4. Summary Writers, like speakers, can use flouting and opting out of maxims in argument. While there are innocent uses of these devices, I have shown that there are manipulative uses as well. The important point is that fallacy lists alone are not adequate to describe the types of manipulative devices available to and used by clever writers. A Gricean perspective, on the other hand, can be used to describe most of the traditional fallacies as violations, and can describe manipulative uses of other Gricean options — non-fallacy violations, opt-outs and flouts. If a reader understands Grice's principles, she can resist a great many more manipulative passages than she can if she is equipped only with a list of seventeen unconnected traditional fallacies. An understanding of Grice's principles can yield readers and writers improved critical abilities in their approaches to text. The Gricean perspective also explains why the use of fallacies and manipulative language is so profoundly uncooperative: the manipulative arguer unilaterally substitutes his own aim (to change the hearer's or reader's beliefs) for the ostensibly mutually accepted aim of the interaction. Use of manipulative language thus flies in the face of the Cooperative Principle.

5. Grice for Undergraduates Can a Gricean perspective be integrated into composition classes? Just a few remarks will indicate how I think a Gricean approach is both appropriate and accessible for composition students. Basic to the statement of the Cooperative Principle is the idea that contributions, whether in actions, talk or text, are evaluated for their acceptability in terms of the mutually understood purpose of the undertaking at hand. In evaluating a passage in a text, readers can best understand the logic of the passage if they have a reasonable understanding of the author's purpose: learning to find a purpose for the text, in both reading and writing, is one of the basic aims of composition classes. Consequently, learning to evaluate a portion of text in terms of that purpose — with Gricean principles as tools — can sensibly be integrated into composition courses. The learning of fallacies, however we may name them, is a case in point. Examples of fallacies in the infamous lists are seldom tied into larger stretches of text or discourse, and are thus difficult to evaluate in terms of purpose. Learning to look for violations,

Critical

thinking

413

opt-outs and flouts of maxims, however, necessarily requires attention to the purpose of the discourse, because it is in terms of that purpose that quality, quantity, relevance and manner can be evaluated. What about the accessiblity of Grice's ideas to undergraduates? While it is true that, for example, the difference between a violation and a flout is difficult to understand, I nevertheless think Grice's basic ideas can be conveyed to composition students. A conversation I had with my nineyear old son will illustrate why I think so. I was in one room and could hear the noise of radio in another, but I could not tell exactly what Andrew was listening to; the dialogue of (17) took place; (17)

Mommy: Are you listening to the ball game? Andrew: No. Mommy: Are you listening to "Problems and Solutions?" Andrew: No. Mommy: Well, what are you listening to? Andrew: The radio.

After this final answer, Andrew giggled. About a week or two after this conversation, I reminded him of it, and asked why he had laughed, why this had been funny. He replied: "It's funny because you were asking me what program, [emphasis his] and I just said, 'the radio'." Although Andrew couldn't put it in these terms, he knew that he was failing to be cooperative — he was violating the first maxim of Quantity, in fact — and the thwarting of his mother's attempt to get information was amusing to him. The giggle demonstrates that Grice's Cooperative Principle exists as a reality in conversation, and that even young speakers have access to the working of that reality. Surely the reality accessible to Andrew is also accessible to college students. Grice's Cooperative Principle and maxims and his description of violations, clashes, opt-outs, and flouts characterize a reality which we are aware of as speakers. My proposal is that we find ways to implement our knowledge of that reality as readers, as well, and that we teach college students to do so too. A heightened awareness of this reality can improve students' critical abilities, including the ability to resist cooperation with manipulative text.

414

Susan Meredith Burt

Notes 1. What is relevant depends on one's assumptions, in ways parallel to the way that grammatically was claimed by the generative semanticists to depend on beliefs about the world. The following dialogue might appear to an outsider to be incoherent: (i) Larry: Andrew has wet his bed again. Susie: That's OK. I have to make bread today, anyway. Susie's response looks irrelevant, but a complex set of assumptions renders it relevant: when a child wets his bed, laundry must be washed and dried. In this proces, the dryer must be run. The dryer produces excess heat when it runs. The excess heat can be used to make bread rise. If bread needs to be made, it is useful to have laundry to dry. Only by understanding these domestic peculiarities does the dialogue become coherent. Yet there is no doubt that it does so.

References Barnet, Sylvan — Hugo Bedau (eds.) 1987 Current Issues and Enduring Questions. New York: St. Martin's Press. Caldicott, Helen 1981 "This Beautiful Planet", in: S. B a r n e t - H . Bedau (eds.), 2 2 6 - 2 3 4 . Cole, Peter — Jerry Morgan (eds.) 1975 Syntax and Semantics 3: Speech Acts. New York Academic Press. Dascal, Marcelo 1977 "Conversational Relevance", Journal of Pragmatics 1, 4: 309 — 28. "Don't Let Abortion Be Controlling Issue", unsigned editional, The Pantograph Bloomington, Illinois. July 9, 1989 A6. Fogelin, Robert J. 1982 Understanding Argument, (third edition.) New York: Harcourt, Brace, Jovanovich. Fulkerson, Richard 1988 "Technical Logic, Comp-Logic and the Teaching of Writing", College Composition and Communication 39, 4: 436 — 452. Green, Georgia M. 1989 Pragmatic and Natural Language Understanding. Hillsdale, New Jersey: Lawrence Earlbaum Associates. Grice, H. Paul 1975 "Logic and conversation", in: Peter Cole—Jerry Morgan (eds.), 41 —58. "The Harkin Doctrine", unsigned editional, The Wall Street Journal New York, New York, February 7, 1989. Lakoff, Robin 1977 "What you can do with words: Politeness, pragmatics, and performatives", in: A. R o g e r s - B . W a l l - J . P. Murphy (eds.), 7 9 - 1 0 5 .

Critical thinking

415

Rogers, Andy —Bob Wall,—John P. Murphy (eds.) 1977 Proceedings of the Texas Conference on performatives, presuppositions and implicatures. Arlington, Virginia: Center for Applied Linguistics. Sperber, Dan — Deidre Wilson 1986 Relevance. Cambridge: Harvard University Press. Tannen, Deborah (ed.) 1982 Spoken and written language: Exploring orality and literacy. Norwood NJ.: Ablex. Teller, Edward 1981 "Dangerous myths About nuclear Arms", in: S. Barnet —H. Bedau, 215-219. Walton, Douglas 1987 Informal fallacies: Towards a theory of argument criticisms. Amsterdam: John Benjamins.

From private writing to public oration: The case of Puritan wills. Cognitive discourse analysis applied to the study of genre change Ulrich Bach

0. Prologue The last will made in 1566 by Henry Cockrofte, who was a fellow of Trinity College (Cambridge) and a Marian exile, contains 25 manuscript lines in the volume of registered copies of the Vice-Chancellor's Court. Only three and a half lines contain legacies proper of his worldly goods. The rest, after the opening formula that the testator is making his will, is devoted to religious considerations. It is not the fact that there are only few legacies but the length of the religious part which at a glance marks this will as different from the traditional, pre-Reformational will. What is the reason for this change, which can also be noted in the wills of his fellow Puritans? We may readily answer this question by pointing out that the insertion of religious passages in these wills is not really surprising; after all, the testators were ardent Protestants, advocates of beginning Puritanism. Professing one's absolute belief in the certainty of salvation, quoting the Bible and praising the Lord for wordly goods are demonstrably all elements of the Puritan code of conduct which is known from other sources, 1 and this fully explains their occurrence in these wills. But this answer does not satisfy the student of discourse who looks pragmatically for the recipients and the functions of a will. A will's natural recipient is the executor. It is his particular task of carrying out the testator's last instructions meticulously and to the best of his understanding which is met by the traditional will's characteristic structural feature — the simple itemization of the series of bequests. Religious catechizing, confessions and polemic discussions of controversial issues of faith are at best irrelevant if not interfering with this task. Doubtless we may assume that Puritan testators, too, had a genuine interest in providing for surviving dependants. To achieve this aim, however, clear comprehensibility and successful execution of the will are required. So the real problem lies in

418

Ulrich Bach

the pragmatic conflict between this aim and the Puritan change of the testament as exemplified by the will of Henry Cockcrofte. The general question of the reason of this change becomes the pragmatic one of "who is to benefit by it?"

1.

Introduction

1.1. Theoretical background It is my thesis that the Puritan changes of the will are in the interest of a public audience which is to be edified, catechized and convinced by a reading out of the will's religious parts. I want to prove this by showing that, and how, the linguistic changes meet the particular comprehensional requirements of this type of audience, in an oral situation of communication. In other words, I shall ask: "Does a listening comprehender benefit from the changes?" The more far-reaching implication of a positive answer is the discovery of a functional change of the genre of last will following the upheavals of Reformation thinking in England, a change that went unnoticed by diachronic linguistics. The analysis of strategies of specific comprehension management in Puritan wills implies a cognitive approach. Before applying it to the linguistic phenomena under consideration I want to answer two preliminary questions of theoretical and historical interest respectively. First, what do I mean by "cognitive approach"? Second, what is the pragmatic background of will-making in the 16th century? By "cognitive approach" I mean in an ad hoc way the analysis and explanation of linguistic phenomena with a view to describe their effects on comprehension and retention of the comprehender, using recent findings in cognitive discourse research. Put differently, this approach aims at a methodology for interpreting the occurrence of specific linguistic phenomena in historical discourse as part of a comprehension-guiding strategy. Let me illustrate this point by a brief example. The inclusion of a prayer expressing the absolute certainty of salvation in a will is a fact which can be accounted for by tracing its source to a Puritan doctrine of belief and also to Puritan readings of the Bible, 2 that is, by comparing it with contemporary writings and ideas — what I would call a philological approach. But the same fact also represents an instance of the modification of the will's superstructure which has consequences for the comprehender. Now if from the cognitive point of view we find that the

Puritan wills

419

original unmodified superstructure complies with the particular task of the will's "natural" recipient in the typical situation of communication and thus recognizably fulfills a cognitive function, then the modification of that superstructure raises the problem of how it affects that function. In principle, then, I propose to make use of concepts and findings of discourse comprehension research as a means for identifying the function of particular linguistic phenomena and explaining their occurrence in a particular historical discourse. Discourse comprehension research as a subfield of cognitive science has since its emergence in the early 1970s restricted itself to synchronic analyses of present-day occurrences of discourse, as any look at journals such as Text and Discourse Processes, or series like Advances in Discourse Processes will quickly show. One of the rare examples of historical pragmatics employing a mentalistic point of view is, within the field of French Studies, the work of Schlieben-Lange (1983), who analyses traditions of talking and concentrates on the cross-relationship between the French Revolution and contemporary linguistic consciousness. From the point of view of the historian, Spufford (1974: 320 — 334) has explored the mentality of 16th and 17th century Cambridgeshire villagers, using their last wills as evidence. Discourse-level analysis of legal texts including last wills "has hardly begun" (Danet 1985: 285); it is non-existent with regard to historical last wills. The little research there is, following Danet (1985) — e. g., Fisher —Todd (eds. 1986) — does not cover the diachronic perspective, neither do so the short studies of wills by Finegan (1982) and Kurzon (1984, 1986: 3 9 - 4 7 ) . Doubtless the reason for this obvious reluctance towards a diachronic application of pragmatic and cognitive discourse analysis lies in the fact that any attempt to reconstruct historical instances of discourse comprehension is hampered by serious difficulties. The biggest of these is that, with psychological experiments being impossible, we cannot hope to obtain certain knowledge of causes and effects where mental processes are concerned. But what the cognitive approach legitimately can do is to use present-day knowledge of discourse comprehension as a heuristic which enables us to recognize the potential explanatory power of the occurrence of linguistic phenomena which other approaches either fail to appreciate or whose occurrence is taken for granted (as in the case of the religious passages inserted in Puritan wills).

420

Ulrich Bach

1.2. Historical background What are the general aspects of willmaking in the 16th century? The last will is a genre which was widely familiar and closely connected with everyday life. Model wills, 3 sermons and the appointment as a testamentary witness were means by which knowledge of the genre and its functions was spread. 4 In addition, the street and, from the last quarter of the century onwards, the stage function as further sources of information. 5 Probably the most important factor of all was the Church which, as guardian of a dying man's last words, expected a will with only few exceptions to be forthcoming from everybody. It is estimated that in the 16th century, up to 10% of the adult population did make a will. This may appear as a relatively small percentage. But even taking further into account that a number of these wills were never submitted to the ecclesiastical courts for official grant of probate (and thereby risked getting lost in the course of time), 6 that still leaves hundreds of thousands of these written utility texts in archives ready to be exploited as linguistic source material of nearly all social groups. A will was usually made on the deathbed. There it offered to people the last if not the only chance for expressing in a permanent medium one's views on death and salvation as one immediately concerned. It is here that the essentially twofold function of the last will as a legal and as a religious instrument becomes evident. In the context of dying, so we may assume, a testator was serious in what he had to say, for he acted on God's explicit command "put thine house in order." 7 But to do this by means of a testament, i. e., verbally, can only be successfully achieved if the will is understood by others. Any misinterpretation of the will's terms by the executor carries the risk of serious consequences for the surviving dependants. Here, the particular pragmatic status of this texttype becomes a relevant issue: A last will represents a verbal act with delayed and with delegated performance. Because it may be revoked subsequently to signing, it does only become legally effective as a last will after the testator's death, and then it is submitted for probate, interpreted and executed by the executor as the persona of the testator. Therefore, when the act of bequeathing is finally performed, there is no chance for explanatory comments by the testator himself anymore. This fact makes the executor and his ability to comprehend written discourse most important factors of the pragmatics of last wills. But with the will being a legal text formal defects, too, apart from misinterpretations and incomprehensibility, can have negative conse-

Puritan wills

421

quences for the surviving dependants and for the weal of the testator's soul if they forestall the execution of the goods and alms bequests. Presumably for this reason the intended universality of the will led to the cognitively relevant legal provision of minimalized formal requirements, making the will an "exceedingly formless instrument" (Pollock — Maitland 1952, II: 319). The individual testator was in principle free to adapt both language and structure of the will to his private ideas of comprehensibility: "Wördes and sentences are not required for the forme of a testament", as the Elizabethan lawyer Henry Swinburne (1590: f. 190v) put it. Paradoxically it was this desire of the testator for ready comprehensibility of his will that favoured the strategy of following a well-tried text structure and using traditional formulae. Substantial changes like those made by Puritan testators do therefore deserve our special attention. For the changes to stand out more clearly I shall first briefly describe the most characteristic linguistic property of the non-Puritan, traditional will.

2.

Analysis

2.1. The traditional will In the traditional will, explicitly religious passages are rare outside its religious preamble. In the preamble, the testator declares that he is making his will in the name of God and the holy company in heaven, that he is of sound mind thanks to God and that he commends his soul to God and his body to the sacred ground in the hope of salvation. This short religious passage is made shorter still and marked as irrelevant for executorial purposes by substituting "etc." for parts of it in the courtcopy which was given to the executor. Those religiously motivated passages that remain — the bequest of the soul and the body, the mortuary, and gifts for prayers for the testator's soul — have the same simple and expedient itemstructure as the worldly bequests of goods and chattels: First I bequeth my soule to almighty God my maker etc. And my body tobe buried in christione buriell ... Item I bequeth to the fryers obseruantes of Estgreynwyche in countie Kent to pray for my soule xl s ... Item I bequeth to Elizabeth Bac my god daughter to pray for my soule xls. Item I bequeth to Sir petir de Lira iii'1 in bookes. (John Boiedens, University stationer, 1502. Wills II: f. 3 v - 4 r ) 8

The sequential arrangement of items turns out to be the dominating structural principle of the traditional will, which thus subordinates lin-

422

Ulrich Bach

guistically the religious function to its legal function. The application of this principle therefore accounts for the traditional will's essentially simple formal superstructure. Except for the preamble's introductory declaration which identifies the testator and states his legal authority ("being of sound mind") as well as the time and the place of making the will, and also excepting the final attestation that the words are the testator's, the will's superstructure does only contain the single functional category of "bequest". The cognitive relevance of this structure for the recipient and the typical situation of communication in which he makes use of the will is revealed by its essentially list-type character. If we follow Ong (1982: 40) and Goody (1977: 17, 53, 80 ff.), lists are a child of literacy because of their demands of the limited human memory capacity. Therefore the contents of lists are processed easier by reading them than by listening to them. For, principally, only a reader is in a position to re-read at will what is still unclear to him or what he has forgotten. The dominance of the simple list-structure in wills is therefore explained by the lack of any need to make the will's contents memorable. This makes superfluous additional linguistic means such as mnemonic aids. The sequence of itemized bequests follows no principle of cognitive structuring: The bequests are ordered neither according to their nature, nor according to their value, nor according to the degree of relationship of the legatees. Instead, the structure of the traditional will is determined solely by the specific task of the executor, thus complying with his particular demands on comprehensibility as a reader of the last will.

2.2. The Puritan will Let us now turn to the Puritan will. What are the inserted passages in detail? Consistently recurring textual elements are (1) the profession of faith: I beleeve in god the father, god the sone, and god the holie ghost three persons but one eternall and ever-lyvyng god and I do fullie looke to be saved by thys my beleiff (Thomas Merburie, fellow-commoner, 1571. Wills II: f. 62'),

(2) the confession of having been a sinner: the damnable pyt of Idolatry wherin I was plonged ... I was most vnworthie yea which well deserved to have been 10000 tymes vtterly rejected from him ... all my stynkynge synnes (Robert Beaumont, 1567, Master of Trinity, Vice-Chancellor, Wills II: f. 45'),

Puritan wills

423

(3) theological catechizing: everye man is of thees two partes soule and bodye the soulle being from heaven hevenly/the bodie made of the earthe and is earthlie (Thomas Merburie 1571. Wills II: f. 62v),

(4) bible quotations: When I shall heare that Joyfull voyce/come the blessed of my father inherite yee the kyngdome prepared for yow before the begynnynge of the worlde ... according to that saying of the profet David what man is he that lyveth and shall not see deathe ... there is nothyng more vncertayne than the vncertayn howr of deathe, for whoes cause we are admonished in the 24 of matthew continuallie to watche. (Thomas Merburie 1571. Wills II: f. 6 2 r - 6 3 r )

and (5) polemical controversy: withowt all vayne opinion of any mans merites which I do vtterly reiecte, deteste and abhorre as mervellous Injurious to the blude of my savior Jesus ... (Thomas Merburie 1571. Wills II: f. 62v).

The effect of passages such as these on the comprehension of the will by the executor is briefly stated. The mixing of burial instructions and alms bequests with religious considerations increases the processing load of the executor by forcing him frequently to decide between what is relevant and what is irrelevant for his task. In the will of Robert Beaumont, the relevant instruction concerning the burial is embedded in religious utterances which are superfluous for executing the instruction: And for my body I desyer that it may be buried semely in the churche Chappell or churcheyarde where I departe and leave from this sinfull lyffe Provyded that my buryall nor after there be no vayne Jangelynge of belles nor anye other popishe ceremonyes or mystrustfull prayers as though my happye state with God were doubtefull. (Robert Beaumont 1567. Wills II: f. 4 5 r - 4 5 v )

In order to understand what the relevant instruction is, the executor, after semantically decoding this passage, first has to recognize that the contents of "and leave from this sinfull lyffe"; "vayne"; "mystrustfull"; and "as though my happye state with God were doubtefull" do not represent anything which he can "execute", and then he has to process them further, i.e., delete them mentally. In the following extract from the will of Henry Cockrofte, the clause "wherof it cam" and the whole passage extending from "vntyll" to "attayne vnto" are irrelevant for the task of arranging the burial: My bodye also I do commende vnto the earth wherof it cam there for to rest in Christen mans buriell vntyll the seconde commynge of our savior Christe when bothe bodye and soolle shalbe ioyned togyther agayne and receave that heavenly

424

Ulrich Bach

blysse and Joye which we all most certaynelye loke for to attayne vnto. (Henry Cockcrofte 1566. Wills II: f. 46v)

Instead of complying with the particular comprehensional situation of the executor, the religious textelements make it more difficult for him to identify that part of the information which is both stimulus and matter for his activity. On the other hand, these very insertions of professions of faith, bits of theological thinking and polemical rejections of ceremonies and tenets fulfill the functions of persuading, asserting and attacking which ask for the public in order to achieve a maximum effect. The profession of faith and publicity, for instance, are explicitly correlated with each other: "Yt shalbe evident to all that duryng my lieffe I held the profession and beleiff of a trew Christian man ... And herin doe I openlie deteste the fond opinion of the papistes" (Thomas Merburie 1571. Wills II: f. 62").

Whom does this "all" represent? What kind of audience are we to regard as the public in the case of testaments? Considering the fact alone that a valid last will usually exists or is available for the executor in a single manuscript copy only, it would have to be a listening audience. (In Cambridge, after 1540 the Vice-Chancellor's Court generally kept the original and handed out to the executor an official courtcopy.) The following analysis will show how the specific conditions of text comprehension for a public listening audience are met by the linguistic phenomena pertaining to the Puritan changes of the will. In detail these phenomena are: the occurrence of negations, the occurrence of religious binomials and of paraphrases, of phonological and rhythmic structures, and finally the occurrence of elliptic pro-forms as means of inter-sentential coherence, mone of which are typical features of the traditional testament. They will be analysed from the cognitive point of view under the headings of "schema modification", "mnenonic aids" and "semantic coherence". 2.2.1. Schema modification "Schema" in the sense of Graesser (1981: 29 — 35) refers to an ordered cluster of knowledge which is activated in discourse comprehension as the pre-knowledge of the recipient and which may be modified by the knowledge manifested linguistically in the text. As pre-knowledge, it provides the relevant knowledge frame for comprehending the text. It directs the comprehender's attention, guides his expectations and enables him to bridge gaps in the text by inferencing — gaps, that is, which stand for knowledge presupposed by the speaker/writer as known to the com-

Puritan wills

425

prehender. The knowledge of text-types and their functions, too, is assumed to be organized in the form of schemata. Schema modification is a relevant issue here because the schema of the texttype last will is changed by the religious parts inserted into Puritan wills while at the same time the traditional label of last will and testament is retained. In this context, the use of negatives, which will be discussed presently, fulfills an essentially comprehension-guiding function. By integrating religious textual elements of the sort described above the superstructure of the testament is made more complex. Prayers, admonitions, catechizings and confessions represent textual elements which explicitly fulfill functions different from that of a testamentary bequest or instruction: "This I do take to be the sure and perfyt confessyon of my fayght" (John Scarlett, stationer, 1551. Wills I:f. 89r). Therefore, the function of the traditional label last will and testament as schemaactivating cue will generate wrong expectations in an audience whose knowledge of the texttype last will is determined by the traditional, preReformation will. For we must not forget that Puritanism was still an emerging phenomenon; the group-identifying term Puritan itself is not recorded prior to 1566. 9 Even if comprehension of the changed testament is made easier by the fact that its religious textual elements were familiar from their original context of worship — they represent four of the six ordinances of the radical Protestant worship — 10 negation still fulfills its part in the strategy of comprehension management by the testator. Its preference over other means of linguistic encoding is strikingly demonstrated by the will of Thomas Fletcher, where "and solely hopinge" is crossed out and substituted by "no thinge waveringe nor doubtinge" in the original manuscript. 11 The following examples demonstrate the range of expressing negation in the Cambridge wills: Provyded that at my buryall nor after there be no vayne Jangelynge of belles nor anye other popishe ceremonyes or mystrustfull prayers" (Robert Beaumont 1567. Wills II: f. 45"), Fyrst I geve not as a recompens for my negligent dewits but as a playne and sure commaundement of god sett furthe in his holy worde in dyuers places as exodus leviticus and deuteronomium ... to the poore mens boxe (John Scarlett 1551. Wills I: f.89 r ), My bodie also I commende vnto the earth to be buried ... withowt all pompe veyne glorie or superstitious facion eyther in mournyng apparell or chauntage at my buriall or other wyse not onlie vnprofitable to me but also to the hearer (Thomas Merburie 1571. Wills II: f. 62 v ),

426

Ulrich Bach

Being assuredly perswaded of my vndowghted salvation by no other meanes but by the onlye deathe and passion of Christ Jesus (William Burwell, vintner and University appraiser, 1587. Wills II: f. 105r).

What precisely is the cognitive effect of these occurrences of negation in the context of schema modification? The explicit quoting of what the testator either does not believe to be true or does not want to be done guides comprehension in that it puts right an anticipated "wrong" expectation of the recipient. In the examples quoted, these are expectations connected with knowledge schemata which are signalled by the terms "burial", "salvation", "rest" and "negligent dewties". They represent ordered and conventionalized sequences of actions or clusters of concepts from the religious domain which the testator presupposes to be known to recipients but which he, being a Puritan, wants to change. Again we must not forget that from the 1560s Puritanism was associated with innovation and subversion (Porter 1970: 5). The cognitive effects of negations include more than just putting right "wrong" expectations of recipients. The quoting of individual rejected elements facilitates and makes faster the access to the appropriate knowledge frame, to a specific schema or sub-schema. This, too, is valid only under the condition that the negated contents of an utterance represent knowledge which the recipients actually possess. The rejected elements of jangling bells and praying for the soul of the dead in the examples just quoted are fixed and to-be-expected elements of the traditional burial schema. Thus, it is traditional knowledge which is negated by no, by no other means, and not ... but (and, wherever applicable, substituted by "new knowledge") in the wills of Puritan testators who are committed to the idea to bring about a change. The precise locating of single to-bedeleted as well as to-be-substituted knowledge elements in the recipients' relevant schemata is guided by the negation of explicitly or even implicitly ("and by no other means") stated elements. By this means, the new is more easily anchored in the relevant schema and the schema itself is thus successfully modified. As cognitive effects of linguistic means on the comprehension of the recipient may be a two-sided matter, "trade-off' becomes a relevant issue in evaluating them. We shall therefore also have to consider briefly the unfavorable effects of negations on comprehension processes. From experiments it is known that negative constructions require more time for decoding than assertive constructions — which in itself is a sure sign of an increased processing load. However, in the context of a public reading of a testament for the purposes of edification, catechization and persua-

Puritan wills

427

sion, the positive effect of facilitating the integration of new knowledge elements is to be rated higher than the unfavorable effect of an increased processing time. Moreover, we shall presently see that there are other linguistic features of Puritan wills which counter the negative effect of an increased processing time. 2.2.2. Mnemonic aids New information is not easy to retain under the conditions of a single, non-repeated presentation. This is a common experience in situations where texts are read out to a listening audience. It is still less easy in the linguistic context of negative constructions just discussed. But among the linguistic phenomena typical of Puritan wills there are also such as have a decidedly positive mnemonic effect. Because of the limited capacity of the human memory these are especially relevant in situations of oral communication which normally do not admit of repetitions at will. Mnemonic aid is provided by phonological and rhythmic structures known from oral poetry. In the wills of Robert Beaumont and Richard Streat we find: "The damnable pyt of Idolatry ... his glorious gospel ... my stynkynge synnes" (Robert Beaumont 1567. Wills II: f. 45r); "the mercies and merites" (Richard Streat, fellow of Pembroke, 1588. Wills II: f. 110r). These examples point distinctly towards dactylic metre, alliteration and assonance — means which, being phonic and rhythmic, achieve their full mnemonic impact when read aloud to a listening audience. This function of their occurrence in these wills is made more evident still by the fact that they occur with expressions which refer to central religious tenets either new or reinterpreted: the rejection of images within the religious domain, the authority of the Bible, sin and grace. Apart from these structures, the occurrence of binomials is mnemonically hepful, too, because of the repetitive nature of these parallel expressions. A binomial is a pair or sequence of syntactically coordinated and semantically related, often near-synonymous words of the same formclass. They have a long tradition in legal English (Mellinkoff 1963: 121 — 122), being obviously intended as a safeguarnd against legal contesting of provisions. Their high frequency makes them a style-marker in law language (Gustafsson 1984: 123). As such they occur in the traditional as well as in the Puritan will, often in combination with multinomials: "I ...do make and ordayne this my last wyll and testament in maner and forme Folowing; hereby revokyng and disanullyng all former wylles and testamentes by me made ... First and principallye I give and bequeath ...

428

Ulrich Bach

my lease/right/title/terme and interest." (William Bosome, apothecary, 1582. Wills II: f. 87v). Puritan testators significantly transfer the use of binomials to the domain of religious concepts, where the argument of "legal safeguarding" is no longer a convincing explanation for their occurrence: "I most humbely beseche hym so to washe and cleanse me with the bloude of his great and only redeemer" (Robert Beaumont 1567. Wills II: f. 45r); "trusting throughe ... Jesus Christ my onlie savior and redeemer to have Free pardon and forgyvenesse of all synnes and iniquities" (William Bosome 1582. Wills II: f. 87 v -88 r ); "where I hope to have rest by the mercies and merites of my savior Jesus Christ" (Richard Streat 1588. Wills II: f 110r); "no thinge waverynge nor doubtinge" (Thomas Fletcher 1582); "I beleeve most steadfastlie to obtayne free forgyuenes and remission of by synnes ... duryng my lieffe I held the profession and beleiff of a trew Christian man" (Thomas Merburie 1571. Wills II: f 62v). Emphasizing the concepts of forgiveness, of the certainly of salvation, of justification, of grace and of the savior by formulaic combinations of synonymous terms makes these hotly debated issues more memorable for a listener. Moreover, the fact that the pairs of terms do not strictly represent cases of repetition but of paraphrase lends them a further property. They are alternatives offered to a heterogeneous audience of recipients whith different lexical knowledge. As pairs of lexemes of different etymological origins, profession-belief, savior-redeemer and forgiveness-remission, to name but a few, function as mutually explaining translations of each other. A repetition-induced mnemonic effect on the sentence-level is offered by the Puritan "thanks-and-praise" formula which precedes the bequests of worldly possessions: "Concerninge my earthly goodes wherwith God hath blessed mee I giue all to ..." (William Fulke, Master of Pembroke, 1589. Wills I:f. 112r); similarly: "desirous to have in Redines my last disposition concerninge the worldlye goodes wherwith yt hathe pleased god to endewe me ..." (Thomas Hoddilowe, brewer, 1595. Wills II: f. 135r)· It takes up the similar thanks-and-praise-formula for being of sound mind, which is common in the religious preamble of both traditional and Puritan testaments: "I Oswalde Archar of Clare hall in Cambryge of hole mynde and in good and perfyte remembraunce lawde and prayse be vnto Allmygthye god" (Oswald Archar, butler, master's servant, Clare, 1547. Wills I:f. 80r). Thanksgiving and praise, too, are concepts emphasized by the Puritans, especially when it is a matter of expressing gratitude and praise towards God for property acquired through one's

Puritan wills

429

calling in life.12 It renders this property a visible sign of God's grace shown to the individual testator, who thereby proves himself to belong to His elect. Puritans therefore had a strong interest in the lasting dissemination of these concepts in the minds of people. Their partial recurrence in the formula for wordly possessions thus becomes a cognitively relevant mnemonic means in Puritan wills. A further linguistic phenomenon typical of the Puritan changes of the traditional will is the use of proforms. They are discussed here under the heading of "mnemonic aids" because of their implications for memory, remembering and the retention of verbal material, being elliptic forms which serve as a means of syntactic cohesion. In the sentence "now this I do take to be the sure and perfyt confessyon of my fayght" (John Scarlett 1551. Wills I:f. 89r), this refers anaphorically to the contents of the confession of faith expressed earlier in the will: "whome he hathe redemyd by his only deathe and passyon" (and which itself contains proforms referring back to the testator "John Scarlett" and to "God" respectively). The elliptic this requires far less processing time than the clause it stands for, and it enables the comprehender to keep the clause's "access point" to its contents in his limited working memory in order to establish coherence with the proposition processed after it (cf. van Dijk 1980a: 165). Because of its reduced processing time and its storage-saving shortness, the elliptic proform this has a beneficial processing effect especially on listeners who have no means of rereading the text and who don't have to execute a sequence of instructions item by item, but instead are expected to follow a train of connected religious discourse or reasoning. Traditional wills (and the legal parts of Puritan wills) tend to avoid proforms as a means of inter-sentential cohesion and instead repeat proper names, genre names and other full lexical expressions: My wretched body tobe buried in the southe yle in the churche of oure Lady beside the markett in Camebrige. Item I bequeth to the hye awtere in the sayd churche of oure Lady ... Item I wille that imediatly aftere the southe yle of the said churche of oure Lady is new made ... Item I gyff and bequeath to John Veyse my sonne alle my londes and tenementes ... the sayd John Veysee to have and to hold the sayd londes and tenementes ... Item yff it fortune the sayd John Veesey to discease withoute heyers ... (Henry Veesy, apothecary, 1503. Wills I:f. 4V —6r)

Here, twice the sayd churche is repeated where syntax, in the first instance, would have allowed its substitution by the shorter there, and in the second by it. Similarly, the name of the beneficiary of the testator's lands and tenements, John Veesy, is repeated in full in two successive bequests after

430

Ulrich Bach

its first occurrence instead of being substituted by the elliptic pronouns he and him. The executor as reader of a legal instrument does not profit by the use of proforms as a listener does. On the contrary, proforms make the execution of his task more difficult because they principally reduce the degree of a text's determination, making it semantically less unambiguous than a full repetition of expressions, of proper names and generic terms does. Because proforms cannot be interpreted without the coreferent expression which they substitute, an executor trying to comprehend a specific instruction would have to turn to an earlier instruction as well, whereas the repetition of expressions reduces that necessity. Repetition supports full determination and unambiguity of single instructions: by repeating the full name of John Veesy, it is clear in each of the three bequests independently of the other two who the beneficiary is. The testator's strong interest in legal safeguarding meets the specific recipient's interest of guided comprehension. Repetition avoids the risk that, from a purely syntactic point of view, the occurrence of a proform may refer to more than one expression in a given preceding context. The avoidance of proforms therefore is a typical feature of legal language which aims at full linguistic determination to minimize the chances of legal contestation of its terms. This fact that proforms, being weakly determined forms, cannot always be matched unambiguously with only one — the appropriate — substituted full expression increases the amount of processing work. This may be further increased by a representational problem when it is a question of "which of several syntactically possible expressions or phrases is the one substituted or represented by the form this?" Even if syntactically there is no problem in identifying the coreferent expression, a long interval between an expression and its proform may still present the mnemonic and representational difficulty of remembering it, necessitating timeconsuming search strategies. These negative implications of the use of proforms principally hold for any comprehender. Does this not contradict our claim to interpret the occurrence of proforms in Puritan wills in terms of comprehension guiding? Here again the concept of trade-off is the answer. Gains and losses with regard to comprehension management by a particular linguistic phenomenon have to be weighed differently, dependent on the situation of communication and the specific needs of a given comprehender. For a listener who (as one to be edified) is a spiritual beneficiary of the will only, there is less need of full linguistic determination ( = avoidance of

Puritan

wills

431

proforms) and legal safeguarding than there is need of a reduced processing load ( = use of elliptic proforms). For this recipient, any loss in determination is more than made up for by the gain in processing time on account of elliptic proforms which enables him to follow the speed of reading aloud more closely. Similarly, any possible loss from the occurrence of proforms (forgetting the substituted coreferent expression or having difficulty in identifying the appropriate one of several potentially coreferent terms) weighs less heavy for him who doesn't have to fear legal consequences from misinterpretations than the just described gain of an easier adaption to the speed of reading aloud and of an easier connecting of new propositions to old ones. On the other hand, for the executor as reader, this gain does not count much because he may re-read prior propositions at will. Instead, for him the gain in determination and disambiguation of the testator's instructions brought about by repetition ( = avoidance of proforms) outweighs any gain in processing speed by elliptic proforms. Finally, the load for the listener of having to remember the substituted expression to be able to interpret its proform is mitigated by other linguistic features of Puritan wills — the mnemonic aids discussed earlier, including the paraphrases by synonymous terms. These do not help the executor to perform his legal function, because "synonyms" are open to contestation. The occurrence of proforms therefore is evidence for the guided, specific nature of the comprehension management in Puritan wills. 2.2.3. Semantic coherence The analysis of the propositional content of the property-related thanksand-praise-formula and its hierarchical rank within the overall propositional structure of the Puritan will discloses a further instance of comprehension management. Its cognitive effect rests on the fact that the underlying proposition of this formula contributes to the semantic coherence of the will as a whole. The construction of semantic coherence is part of the recipient's interaction with the text and as such it is part of the comprehension process. Here, the occurrence of the thanks-and-praise-proposition enables the recipient to establish an explicit connection between the religious part of the testament and the individual wordly bequests: "My body I comitt vnto the earth from whence it was taken in stedfaste hope of a glorious resurrection vnto lyfe euerlastinge through the mercy and merittes of the

432

Ulrich Bach

same our Lord Jesus Christe. Concerninge my earthly goodes wherwith God hath blessed mee I giue all to ..." (William Fulke 1589. Wills II: f. 112r). The reference to the linguistic centre God sets forth a semantic property which both parts of the testament have in common and which therefore enables the comprehender to construct the global semantic coherence of the will. Moreover, the scope of this proposition extends over all those propositions which have, as one of their arguments, the concept of a particular piece of goods and chattels, of a fixed sum of money, or of a building or piece of land. These individual legacies can be described by bundles of semantic features: in the case of household goods, for instance, these would be < + concrete), EO 1 —• EW1 Birnenpflücker —> farmer —• farmer/pear-picker

3.2. Hesitations, new starts and the like in oral narratives Here are a few passages for illustration: (1)

im Hintergrund hat en andrer Farmer — oder Bauer Farmer halt hm — ne Ziege an ner Leine geführt ... und — ja der Fa Birnenpflücker der hat die Birnen ganz bedächtig ... (43.GO 1: 16 ff.) 'in the background another farmer — or peasant well farmer — led a goat on a lead ... yea — the fa pearpicker he put the pears really carefully ...'

3.3. "Self-negotiations" of alternatives or clarifications in TAPs during written narrations (2)

(3)

the boy continues — his ride on the bicycle — and the other boys — ja other is necessary — go — in the opposite direction (45.TAP: 760) the three boys discover the: — first boy's — oh nee no the: — the lost hat ... that the boy on the bicycle lost when he passed the girl ja (41 .TAP: 465)

3.4. Both 3.2. and 3.3. contain traces of "latent cognitive bases" which have not yet been optimally verbalized: (4)

and — uh /where/ the boys whistled — after him — and uh — brought his hat — and then he gave him uh three of his fruits (41.EO 2: 150ff.)

Renarration

471

Although this narrator is first unclear about the fact that it was not all the three boys (the boys) that whistled after the pear-thief (him) the formulation of the last event corrects this: The pear-thief (he) gives the fruit to the one of three boys (him) who had actually whistled.

4. Quantitative aspects Among the first impressions from our data was the extent of individual variation between informants. The lengths of the versions alone convey a good idea of this. Table 1. Quantitative Analysis Length rankings: individual subjects 40:

GOl 382

>

EO 2 292 >

EW1 av. 443 388 E02 391

av. 337

41:

GOl 537

>

E0 2 404

42:

EW1 434

>

GOl 407

>

43:

GOl 701

>

E02 532

>

EW1 415 av. 549

44:

EW1 296

>

EOl 294

>

GO 2 264 av. 285

45:

EOl 352

>

GO 2 339

46:

EOl 643

>

GO 2 590

47:

EW1 578

>

EOl 482

48:

EOl 743

>

GO 2 683

50:

EOl 358

>

GO 2 315

>

EW1 311 av. 328

51:

EOl 378

>

EW1 348

>

GO 2 345 av. 357

52:

GO 2 258

>

EOl 243

>

EW1 219 av. 240

53:

EOl 419

>

EW1 354

>

GO 2 329 av. 367

av. 411

av. 346 av. 617 >

GO 2 476 av. 512 av. 713

472

Rüdiger Zimmermann

Table 1. (Continued) Length rankings: individual subjects 54:

EW1 292

>

EOl 246

>

GO 2 230 av. 256

55:

GO 1 > 870

EO 2 658

>

EW1 av. 682 429

56:

E0 2 648

>

GOl 638

>

EW1 534 av. 607

57:

EW1 435

>

GOl 372

>

EO 2 248

58:

GO 1 > 554

E02 432

>

EW1 396 av. 461

59:

EO 2 343

>

GOl 320

>

EW1 av. 325 311

60:

GO 2 > 493

EOl 437

>

EW1 av. 419 326

61:

E02 386

GOl 353

>

EW1 281 av. 340

av.

GO 1 478

E02 473

GO O l 436 447

Ο 440

EOl 416

GO 2 EO 0 2 394 445 434

EW 388

>

av. 352

4.1. First and second oral versions Some overall tendencies are noteworthy. After the first showing of the film most 01-versions are longer than Ο 2 (16), only 5 Ο 2 are longer than 0 1 . This effect is underlined when we look at the individual length differences as percentages of the first versions: Whereas longer second versions are mostly just a little bit longer than the corresponding 0 1 (between 1.5 and less than 10 per cent, with one exceptionally high value of 11 per cent), longer first versions are as a rule much longer than the corresponding 0 2 narratives (between 1.2 and 33.3 per cent, but often more than 20 or even 25 per cent longer). The average 0 1 length is also about the average for Ο 2 (447 vs. 434). There are two opposing tendencies which might explain this: — On the one hand narrators will as a rule have a better grasp of their story during the second narration rejecting awkward passages and maybe omitting irrelevant details.

Renarration

473

— On the other hand, the better handling of some aspects may in some cases enable them to add further details which they were unable to narrate during their first version.

4.2. English and German oral versions But there is a contradictory tendency. Slightly more English narratives are longer than the German versions told by the same informants (12, whereas only 9 German narratives are longer than those in English). This corresponds to a high average length in EO narrations (445 words vs. 436 in the German versions. And note that these averages are almost exactly the same as the Ο I/O 2 values.) How does this compare to Wolffs (1989) results? In his study of written L 2 and L1 Pear Story narratives he found a much bigger majority of longer English versions (25 as opposed to 8 longer German ones). But Wolff had all the English versions narrated before the German ones. In the light of our two conflicting lines of evidence his results could display the language effect he sees in his data, but it might equally be a sequential effect. For our oral data the sequential effect can perhaps override the language factor if we consider the 0 1 — Ο 2 sequences more closely: Although the English length average is a bit higher than the German one the sequential effect seems a bit more relevant when we look at our two sets of data separately: GO 1 478 EO 1 416

> >

E 0 2 av. 473 GO 2 av. 394

In short, the average Ο 2 story seems to be shorter (or at best stable) irrespective of the language. A glance at the individual narratives underlines this tendency. GO 1 > EO2 : 7, EO 1 > GO 2 : 9,

E02>G01:3; GO 2 > EO 1 : 2.

Thus, of the 21 oral versions 12 EO are longer than GO, and not many less, namely 9, are shorter. But comparing successive versions, only 5 0 1 are shorter than Ο 2, but 16 0 1 are longer than Ο 2. The language effect seems to be overridden by the sequential one.

474

Rüdiger

Zimmermann

If this generalization should hold for oral narratives it might still not be valid for written ones owing to the different conditions under which the narrative process takes place. I think this question must remain unanswered until a study of written narratives with an L1 — L 2 sequence has been performed.

4.3. Oral and written versions After the second showing of the film 10 EW versions were shorter than the oral ones, 5 were longer, two had middle length (4 EW narratives from the first set were unusable for technical reasons). We attribute these differences to a stronger writing or speaking orientation in our subjects. This seems to be the foremost factor since the length differences between several oral and written versions are too big to be attributable to "trimming" or elaboration after the second showing alone. 7

5. The effect of renarration on text structure A further observation relates to the development of the narrative through the successive versions: As can be expected, later versions seem to be better organized. Written versions may be more complete, but some omit secondary details contained in the earlier oral narratives. (This is not surprising after a second showing of the film.) This "trimming" effect is illustrated by the following passages. (5)

a. a little boy comes along he's riding a bike — a very big bike for such a little boy — and uh — it's red — actually — a red bike ok \ - and he's riding by (46.EO 1: 23 ff.) b. während er das tut — kommt ein kleiner Junge auf einem riesengroßen roten — Fahrrad darhergefahren (46.GO 2: 13 ff.) 'while he does that — comes a little boy on a very big red — bicycle driving by'

Compare this set of passages from another informant: (6)

a. jedenfalls — uh fährt der Junge dann weiter — aber we als sie ihm aufgeholfen haben — aber dann finden die anderen die gehen

Renarration

475

halt — also er fährt in — Richtung des Weges — wo er sowieso langgefahren ist — weiter — aber die anderen — uh finden noch seinen Hut — pfeifen dann daß er zurückkommen soll — uh nee — einer läuft hin — pfeift ihm nach und gibt ihm seinen Hut. (55.GO 1: 62 ff.) 'anyway — ah the boy rides on — but after they helped him to get on his feet — but then the others find they go — so he rides in — the direction of the path — where he came along anyway — further — but the others — uh find his hat — whistle for him to come back — uh no — one runs there — whistles after him and gives him his hat' b. and so — he rides along — again — and — the guys — uh — walk the other way — but then they — whistle after him — because he — uh — lost his hat — uh — when he saw the girl and — they gave — gave him back his — uhm — his hat (55.E0 2: 46 ff.) Despite its better overall organization 55.0E2, as other second English versions, has comparatively many short hesitations; no doubt an L 2 effect. Apart from the hesitation phenomena both second oral versions could almost be read as written passages, whereas the first versions could not. 8 (7) (8)

Während er das tut, kommt ein kleiner Junge auf einem riesengroßen roten Fahrrad daher gefahren. And so he rides along again, and the guys walk the other way. But then they whistle after him because he lost his hat when he saw the girl. And they gave him back his hat.

In EO 2 only the two repetitions gave and his interfere with this written impression. On the other hand EO 1 and even more so GO 1 would still strike us as oral even if presented in standard punctuation. (9)

(10)

A little boy comes along. He is riding a bike, a very big bike for such a little boy. And it's red, actually a red bike. Ok. And he is riding by. Jedenfalls fährt der Junge dann weiter. Aber als sie ihm aufgeholfen haben ... aber dann finden die anderen ... die gehen halt ... also er fährt in Richtung des Weges, wo er sowieso langgefahren ist, weiter. Aber die anderen finden noch seinen Hut, pfeifen dann, daß er zurückkommen soll. Nee — einer läuft hin, pfeift ihm nach und gibt ihm seinen Hut.

476

Rüdiger Zimmermann

Not all first oral narratives are so loosely structured. Some narrators have a better grasp of an episode right away. But even their second versions of the passage show an improvement over the first one more often than not. (11)

(12)

ah ja dann kommen se wieder an dem Birnbaum vorbei da gehen se nämlich lang und der Mann steht also immer noch auf dem Birnbaum — pßückt Birnen — kommt dann aber gerade runter — guckt sich seine Körbe an und da fehlt einer na natürlich wundert er sich ziemlich weil einer der Körbe weg ist (58.GO 1: 40 ff.) 'ah yes then they pass the pear tree again because they go by there and the man stands well still in the pear tree picks pears but is just coming down looks into his baskets and one is missing well naturally he's quite astonished because one of the baskets is gone' — then they walk by the /pia/ tree where the man is still picking the Ipiars/ but in the meantime he'd come down the ladder and see — that one of the baskets is missing — so — of course he's quite surprised (58.E02: 28 ff.)

6. Successive narrations on an oral-written scale? Do such observations possibly imply that second versions can be seen as less oral in general? Are they in a way steps in the development of the written narratives, half oral and half written? Or is the L 1 / L 2 effect stronger? As a first test of these hypotheses we analyzed our texts for some major dimensions of orality and literacy as proposed by Chafe — Danielewicz (1987). We focused on several aspects of sentence coordination and subordination, on lexical hedges, nominal word-formation and narrator involvement. 9 Table 2 gives the percentages for several features. Some of the most interesting ones are combined in charts 1 and 2. I think the following conclusions can be drawn from this preliminary quantitative analysis.

Renarration Table 2. Some grammatical and lexical features (%)

and then so but other sub. cl. rel.cl. when as because other that Part I Part II Inf. PPs. mult. hedges WF/N you I, we

and then so but other sub. cl. rel.cl. when as because other that Part I Part II Inf. PPs. mult. hedges WF/N you I, we

50 : EO 1

50 : G O 2

50 : EW

51 : EO 1

51 : G O 2

51 : EW

6.6 2.0 1.4 0.3 0.8 -

7.3 3.5

2.6 1.0

10.6 1.8

10.7 3.2

0.6 0.3 1.0 -

1.0 1.0 2.6 1.3 1.6

0.3

0.3 0.6 1.2 0.9 0.3

4.0 0.6 0.3 0.6

_

_

1.0 0.8 0.3 0.3 _

0.3 0.3 0.3 5.4 1.0 0.6 0.6 — —

52 : EO 1

52 : GO 2

52 : EW

11.6 0.4 0.8 0.4

8.1 0.4 1.2 0.8 0.4 0.8 -

4.6 0.5 0.5 2.3 0.5

1.4 1.4 0.2

0.5 0.9 2.3

0.2 0.2 0.7 0.2

2.1 0.8 0.4 0.4 0.4 0.4 _

0.3 0.6 2.6 0.6 6.1 0.6 0.3 1.0 — —

_

0.3 2.5 5.9 0.6 0.3 0.3 — —

_

0.5 1.0

0.4 5.8

1.4 6.8

0.4 0.4 — 1.7

1.6 — 0.4

0.5 — -

0.9 -

0.5 4.9 0.3 0.3 0.3 0.3 —

2.0 4.6 1.2 0.6 0.9 — —

0.9 — —

53 : EO 1

53 : G O 2

53 : EW

7.9 1.6 0.7 0.4

6.3 1.2

3.9 1.1

1.2 1.5 1.5 1.5 0.6

1.1 0.8 2.2 1.4 0.3

0.6 0.6 _ 0.6 6.6 0.3 0.3 2.1 — -

0.3 1.1 0.6 0.8 _ 0.8 5.1 0.3 0.3 —

_

0.4 5.8

_ -

1.7 0.9 0.9 0.3 _ 0.3 0.3 2.3

_ 0.4 3.9 0.2 0.4 0.2 0.4 0.7

1.4 6.3 0.9

477

478

Rüdiger

Zimmermann

Table 2. (Continued)

and then so but other sub. cl. rel. cl. when as because other that Part I Part II Inf. PPs. mult. hedges WF/N you I, we

and then so but other sub. cl. rel. cl. when as because other that Part I Part II Inf. PPs. mult. hedges WF/N you / , we

54 : EO 1

54 : GO 2

54 : EW

55 : GO 1

55 : E 0 2

55 : EW

7.0 1.6 0.4

5.2 2.6

4.5 0.3

7.7 3.1 0.1 0.8 0.8 1.0 0.2 0.2

7.4 0.3 0.9 1.1 0.5 1.8 0.6 0.2 0.2 0.2 0.8 0.6 1.7 0.2 0.6 4.6

4.2 0.9 0.5 1.2 0.9 1.2 0.7

—

— —

16 1.6

0.4 —

2.6 1.3 1.3

—

—

0.3 2.1 2.4 —

—

1.0

—

—

—

—

—

0.3 0.7 0.7

1.2 2.0

1.3

—

—

—

1.3 5.7

1.0 5.5

0.8 6.5

—

—

-

-

-

0.5 0.3 —

0.3 0.6 3.0 —

1.5 0.8

-

— — —

0.5 0.5 2.1 —

1.4 6.3 0.2 0.2 0.7 0.9

2.0

0.4

0.3

0.2

0.5 0.2 1.2 0.8

56 : Go 1

56 : E 0 2

56 : EW

57 : GO 1

57 : E 0 2

57 :

6.6 3.0

6.4 0.9

3.6 0.2 0.2 0.2 0.4 2.6 0.6 0.2 0.4 0.2 1.1 0.9 3.4 0.4 2.1 7.1 0.4 0.2 1.7 0.6

6.9 5.1

5.6 0.7 0.3

2.1

-

—

0.3 0.8 1.9 0.8 —

0.2 0.3 1.1 0.3 —

—

0.3 4.4 0.3 4.1 1.4 -

1.7

0.4 -

—

0.5 0.2 1.7 0.6 0.2 —

0.2 0.9 0.5 2.4 —

0.8 5.3 0.5 1.2 0.6 0.6 1.4

0.7 -

-

-

—

0.2 1.0 3.2 0.2 1.3 — —

0.2 1.6 0.2 —

0.2 6.5 0.8 0.2 1.3

— -

2.6 —

1.3 -

0.3 0.3 0.7 1.7 —

0.3 7.9 0.7 -

-

-

-

-

-

—

0.2 —

1.1 2.5 0.7 0.9 —

0.2 0.5 1.1 2.1 —

1.1 9.4 1.1 -

1.1 0.2 0.2

Renarration Table 2. (Continued)

and then so but other sub. cl. rel. cl. when as because other that Part I Part II Inf. PPs. mult. hedges WF/N you I, we

and then so but other sub. cl. rel. cl. when as because other that Part I Part II Inf. PPs. mult. hedges WF/N you I, we

58 : G O 1

58 : E 0 2

58 : EW

59 : GO 1

59 : E 0 2

59 : EW

3.6 1.6

6.9 1.6 1.6 0.7 0.9 2.1 0.9

2.5 0.3 0.3 0.3 1.8 3.0 1.5 0.3 0.8

3.4 0.6

8.5 1.2

4.2

-

0.4 1.6 1.8 1.3 0.2 —

0.7 0.4 0.4 — —

0.2 5.6 —

1.4 0.5 -

0.4

—

0.9 —

0.7 0.5 —

0.5 5.6 0.2 0.5 0.2 -

—

1.5 0.5 0.8 —

0.5 6.1 0.3

—

0.3 2.2 1.3 0.6 0.6 -

—

0.6 0.3 0.3 0.3 1.3 7.8 0.9

—

0.6 0.3 1.5 0.6 — -

—

1.2 0.3 0.6 0.3 0.6 7.0 0.6

—

0.3 —

0.6 2.3 0.3 1.0 0.3 —

1.0 —

1.9 —

0.6 7.1 0.3 0.3

-

-

-

-

1.6

-

-

-

-

-

-

0.3

-

-

-

60 : E O l

60 : G O 2

60 : EW

61 : G O l

61: E 0 2

61 : EW

4.8 0.9 0.2 0.2 0.9 1.8 0.5

6.5 2.4

2.5 0.6 0.3 0.6 0.9 1.8 1.2 0.6 0.3 0.6 0.3

5.1 2.3

6.7 1.5

3.2

— —

0.9 0.5 0.5 1.8 —

0.5 5.7 0.9 0.9 -

1.6

— —

1.4 1.2 0.4 0.4 — —

0.4 0.4 —

0.6 0.2 5.9 0.4 1.2 1.2 -

0.4

—

1.2 —

2.6 6.4 0.9 —

0.3 0.3 0.3

—

—

—

0.4

1.1 3.1 1.4

0.3 0.8 2.1 1.5

—

—

—

—

—

—

—

—

0.8 1.7 —

1.1 0.8 6.2 0.7 0.3 2.5 (0.3) 0.8

0.3 1.3 0.5 1.5 0.8 0.5 2.8 0.5 — -

2.1 1.3

—

0.4 2.1 0.7

—

1.4 0.7 3.2 1.1 1.1 5.3 — —

0.4 0.7 0.7

479

480

Rüdiger

Zimmermann

Chart 1

6.1. Syntactic aspects 6.1.1. Most figures for written and oral averages are in line with differences known from the literature. 10 In particular, the relation between oral and written English versions suggests that our film data can be considered

Renarration

481

as fairly natural everyday narratives: and/und, then/dann, all dependent clauses (including relative clauses), infinitives, involvement all correspond more or less to well-known oral-written differences. The only exception is nominal word-formation (cf. 6.1.4.). 6.1.2. For some features there is a development from first to second oral versions in the direction of written values. Involvement and lexical hedges fall from 0 1 through Ο 2 to EW; on the other hand there is a mild increase in relative clauses and a clear one in infinitives. 6.1.3. There also seems to be some striking counterevidence, namely a slight increase in and/und values and a very strong one in then/dann. There is at least one possible explanation for this phenomenon: First versions are often very poorly structured in terms of temporal (and other) connectors. A first improvement in sequential organization leads from zero or and/und to then/dann etc.; a further step from then/dann to when/ als etc. is only taken from Ο 2 to EW. 6.1.4. There seem to exist some L I - and L2-specific phenomena: Whereas averages for und and dann are distributed similarly in German, the English of our informants shows a clear preference for and over then in oral as well as written versions. There is an even stronger L 1 effect in nominal word-formation: Whereas word-formation is lower in EO than EW, it is so much more frequent in G O that this figure raises the non-language specific Ο average above the EW level. Further language-specific effects occur with respect to involvement as expressed in first person proforms, and in lexical hedges. The strong involvement in EO is largely connected with L 2 deficit statements, especially with respect to unknown or inaccessible lexical items. (13)

and there is a big tree — where — okay that is the first problem — I don't know what the fruit is called — I mean I forgot it — it's not peaches — it's like apples — you know (55.EO 2: 3 ff.)

The much higher frequency of lexical hedges in the G O average is a result of the individual fad for the pseudo-hedge (especially in subject 56 : 4,1 per cent!), even if there is no insecurity about the following noun at all. 11 (14)

also es war irgendwie so'n Film — in so'ner Landschaft — mit uh — Obstbäumen — also Birnbäumen hauptsächlich — oder

482

Rüdiger

Zimmermann

wenigstens den Baum den man genauer gesehen hat war so'n Birnbaum — η da war — so'n — ja — so'n mittelalterlicher Mann der da auf so'ner Leiter stand. (56.GO 1: 1 ff.) 'well somehow it was this kind of film in this landscape with pear trees well mainly pear trees or at least the tree which could be seen more closely was this sort of pear tree and there was this — yes this middle-aged man who was standing on this ladder.'

6.2. Lexical aspects Chart 2 contains four averages of features relating to lexical density and variation. We are considering the possibility of establishing lexical density via the ratio of content words to all words of a text. It goes without saying that a relatively high frequency of content words can easily result in overexplicitness, since often a proform will do. So it is not necessarily a sign of narrative quality. Nevertheless, many — of course, adequate — content words establish identifiability better than underexplicit reference mainly through proforms. We are experimenting with the ratio of content words (CR) as one variable to be compared to others (and to an intuitive notion of which of our narratives are "good").

45 r

Lexical density and variation TTR (N compl.)

entails a number of schemata, such as SOURCEPATH-GOAL, FRONT-BACK, STRAIGHT-CROOKED, BALANCE, etc. The SOURCE-PATH-GOAL schema is instantiated in every journey in as much as every journey has a beginning (SOURCE), proceeds along some real or imaginary PATH and ideally leads to some destination (GOAL). The moving object, be it a traveller or a vehicle, typically moves F R O N T W A R D maintaining BALANCE. Loosing balance constitutes an impediment and permanently or temporarily makes progress towards the goal impossible.

5. The fundamental axiology of the balance schema Elsewhere, I suggest that all preconceptual image schemata are endowed with positive-negative vectors, which ultimately provide basis for various axiological distinctions along the good-bad scales at various levels of the axiological hierarchy (Krzeszowski 1990). Consequently, the SOURCEPATH-GOAL schema, on which every discourse is founded, contains the pole plus at its GOAL end and the pole minus at its SOURCE end. (Hence movement forward towards a goal is conventionally experienced as something good, while movement backwards and everything which

536

Tomasz P.

Krzeszowski

retards the progress forward are conventionally experienced as bad.) Consistent with this fundamental axiology is the axiology associated with the BALANCE schema, which also provides the foundation for the coherence of discourses. Thus, balance is experienced as positive and ultimately good, while lack of balance is experienced as negative and ultimately bad.

6. The axiological balance, axiological clashes and meta-axiological interludes In Krzeszowski (1990) I claim that all words and all linguistic symbolic units are assessable on some axiological scale. Even if some words seem to be axiologically neutral, they are still prone to axiological distinctions given appropriate contexts (accrued meanings, metaphorical extensions, etc.). The axiological balance (henceforth AB) of a discourse consists in maintaining a consistent positive axiological load of all the linguistic forms (symbolic units) at the semantic pole of every unit. This is connected with the fact that every discourse is ideally oriented towards a goal (realized through various literary and rhetorical means) which is of necessity positive (relative to some specific hierarchy of values subscribed to by the speaker or writer). Everything which interferes with reaching that goal must be evaluated as negative and provides experiential foundations for an axiological clash (henceforth AC). Every incident of this sort is also connected with the loss of BALANCE and is consistently evaluated as negative. AC can be horizontal or vertical. In a horizontal AC only values at one level of the axiological hierarchy are involved, for example at the level of sensory values. An example of a horizontal AC can be seen in the following dialogue: (1)

A: These apples are too sweet. Β: I don't think so, I like sweet apples.

In (1) the AC is caused by A's and B's different ideas about the degree of sweetness of apples relative to which they can be evaluated as good, A and Β derive different degrees of pleasure from experiencing apples of a given degree of sweetness which inevitably leads to AC's whenever their respective preferences are confronted in a conversation. Let us note that masochism seems to be the limiting case of AC generating conditions in as much as what a masochist considers to be pleasant most people

Axiology

537

consider to be utterly unpleasant. By contrast, the AC in (2) involves values at two levels of the axiological hierarchy: (2)

Child (trying to swallow some medicine): I hate it, it's bitter] Mother: Swallow it. It's good for your tummy.

In (2) the vertical AC is caused by the fact that the child subscribes to the concept of as defined relative to the lowest level of values, identifying what is good with what is pleasant, in this case with what tastes good. Mother, on the other hand, identifies good at a higher level of the axiological hierarchy, i.e. the level of vital values by identifying what is good with what is beneficial for the child's health. Axiological balance can be restored in the course of a meta-axiological interlude (henceforth MAIN), which may follow an AC, A MAIN, whose linguistic exponents will be briefly discussed in section 12, is successful if AB is restored, otherwise it is unsuccessful. Intuitive evaluations of discourses and texts as being well-written or poorly written can be made more explicit in terms of the concepts just introduced. A text is well written if it is balanced throughout, or if there are ACs, as is typically the case in all argumentative texts and most fiction, leading to MAINs which restore balance. In a well-written text the goal is always reached, often with the assistance of various "roadsigns", such as the point of the text being clearly stated at the outset or made evident in the appropriate place for the reader not to get lost in the course of reading. On the other hand, poorly written texts will contain unexpected interruptions, deviations from the path, unresolved ACs, and unresolved MAINs. Unfortunately, in the present paper we have no room to elaborate this particular aspect of textual axiology.

7. The axiological typology of discourses The concepts of AB, AC and MAIN provide grounds for the following most general typology of discourses: I. II. III. IV.

AB# AB AC # AB AC, (MAINj, (... AC n MAIN m ) # AB AC; (MAINj, (B) (... AC n M A I N J #

where # stand for the termination of the discourse.

538

Tomasz P.

Krzeszowski

Type I represents discourses without ACs, and consequently without MAINs. They are balanced throughout and the path present in the backbone leads directly to the goal. Type II represents discourses containing an AC which does not yield a MAIN (an unresolved AC), which directly leads to the termination of the discourse. Type III represents discourses with one or more ACs, all or some of which yield unsuccessful MAINs, as a result of which balance is not restored before the termination of such discourses. Finally, Type IV represents discourses with one or more AC, all or some of which yield MAINs, some of which are successful, the final M A I N being obligatorily successful, so that the particular discourse terminates only after balance has been restored.

8. Some illustrations In order to illustrate this typology with examples, let us analyze some appropriate discourses. Type I can be illustrated with the following brief story: (3)

"He poured her a large double brandy and a smaller one for himself. She took it from him self-consciously. 'Cheers!' he said. 'Cheers!' They both drank."

(3) contains no ACs and consequently no MAINS, since its development at no point deviates from the straight path leading to the GOAL viz. having a drink. The successive sentences each bring us closer to that particular goal through the successive actions: pouring out, cheering and, finally, drinking. Thus the aim is reached without any interruptions. At no point is there the basically positive load of the discourse interrupted or interfered with. Hence, there is no need for MAIN 2 . Let us note with considerable emphasis that this axiological observation is consistent with the total absence in (3) of any words with the negative load, which could signal a possible AC, i. e. a distortion of the BALANCE initially underlying every discourse. By way of contrast let us consider the following discourse in which an AC and the resulting M A I N do occur, hence (4) represents Type III: (the AC is placed in braces while the M A I N is placed in square brackets):

Axiology

(4)

539

"I hope I haven't interrupted ..." "Of course not. Have a drink?" "Thanks. Beer?" "{Sorry, We haven't a frig — we send out for ice} [What about a Scotch?" "A small one, if you don't mind. I'm not very keen on hard liquor."] "On the rocks?" "[Plenty of soda] — if you aren't short."

The AC in (4) consists in the incompatibility of preferences of various kinds of drinks: one of the speakers prefers beer but is forced to drink Scotch for the lack of ice. Still, in the M A I N that follows, he tries to abide by his values, asking for only a small drink and explicitly expressing his limited liking of the proposed drink. The M A I N ends with a compromise: the guest accepts a small Scotch with a lot of soda. In contrast with the previous discourse, the goal, viz. having a drink is not reached directly but is preceded by the AC (preference for beer over Scotch) and the consequent MAIN (settling down on a small Scotch). It must also be noted that neither the successful outcome of the MAIN nor reaching the goal, i. e. having a drink are expressed overtly. Thus, though BALANCE is not overtly restored since the MAIN does not continue, one can safely assume that agreement was arrived at, and the goal was reached. Type II can be illustrated with the following paraphrase of a fable by Aesop "The Donkey and the Lion's Skin": (5)

A donkey found a lion's skin in the forest one afternoon. He put it on and went to the barnyard to frighten the other animals. "Hee-haw, I'm a lion," the donkey brayed at a rooster. {"That's a pretty stupid joke," the rooster said. "Even though you look like a lion, anyone can tell you're a donkey as soon as you open your mouth."}

The AC in (5) is caused by the fact that the rooster does not share the donkey's ICM of what constitutes a good joke. What the donkey considers to be a good joke the rooster considers to be stupid. The AC leads to no MAIN and the discourse terminates without any resolution of the AC. (6), another fable by Aesop, "The Dove and the Hawks", is a good illustration of type III, which was also illustrated by (4) above:

540

(6)

Tomasz P. Krzeszowski

A chicken grower set a trap for hawks that had been attacking his flock. One afternoon he found a dove in his nest. {"Let me go. I'm not a hawk,"} the dove begged. The chicken grower agreed, ["Maybe not. But I'm not going to let you go. You're as bad as they are, if you're with them. After all birds of a feather flock together."]

In (6) the AC consists in the fact the chicken grower, in contrast to the dove, does not consider not being a hawk to be a sufficient condition of being good. He expresses his view in the MAIN that follows. The MAIN remeins unchalleneged by the dove, and thus the discourse terminates, while balance is not restored. Type IV can be illustrated by yet another paraphrase of one of Aesop's fables "The Ant and the Grasshopper"; (7)

{While a grasshopper was taking it easy in the shade on a hot summer's day}, an ant struggled in the sun with a grain of rice that he was carrying out to his nest. "Hey, Mister Ant," the grasshopper said. ["Why don't you take it easy, like me? You can work tomorrow".] The ant paused. ["I'm saving up food now for the cold winter ahead, and if you know what's good for you, you'll do the same,"] he said. Three or four months later, winter came and it was very cold. While the ant was snug in his nest, the starving grasshopper shivered under a pile of dead leaves and wished that he'd paid attention to the ant's advice.

In (7) the AC is caused by the fact that the grasshopper thinks that work consisting in saving up food for the cold winter has no bearing on his well-being while the ant thinks that saving up food is good. In the MAIN that follows he tries to make the grasshopper adopt this view. Though at first the grasshopper does not agree, he finally acknowledges the ant's view on the matter, and thus the axiological balance is restored. The fable ends with the expression of the two creatures' sharing the same values concerning the matter.

9. Discourse Coherence The concept of AB is crucial in establishing the axiological coherence of discourses. It consists in maintaining the axiological balance which is realized by the consistency of axiological polarity of the lexical items

Axiology

541

(and all other linguistic expressions) throughout the discourse. Such consistency is evident in (3) but lacking in (4). Therefore, (3) contains no AC and, consequently, no MAINs. By contrast (4) contains an AC and a MAIN. The AC in (4) is linguistically expressed by the expression of apology ("Sorry") and the hidden negation implied by this apology: the impossibility of satisfying the guest's request. Because of the AC the ensuing MAIN is absolutely necessary to keep the dialogue going. At the point at which the AC occurs the axiological positive balance is lost. Being unable to satisfy the guest's request, the host tries to make up for the lack of beer by suggesting something else and thus initiating the MAIN. Feeling guilty of the shortage, he is also being polite, providing some reason for being unable to serve beer. All this serves the purpose of restoring the lost balance to its original positive value. Without the MAIN initiated by the host the dialogue would either end without having reached the originally intended goal (the guest's having a drink) or would become incoherent. In the former case (4) would look as follows: (4)

a "I hope I haven't interrupted ..." "Of course not. Have a drink?" "Thanks, Beer?" '(Sorry.) We have no beer". ("Too bad".)

In the latter case (4) might look as follows: (4)

b "I hope I haven't interrupted ..." "Of course not. Have a drink?" "Thanks. Beer?" "Sorry. We haven't a frig — we send for ice." "On the rocks?" "Plenty of soda — if you aren't short."

10. Multiple ACs and MAINs Some MAINs do not directly follow the ACs by which they are evoked. Nor do all mains lead to a successful restoration of the axiological balance. (8) represents a longer discourse with several AC's and MAINs of which only the final one is successful in as much as it restores the balance. Like in the previous examples, in (6) the ACs are placed in braces, while the MAINs are placed in square brackets:

542 (8)

Tomasz P. Krzeszowski Dinner began in silence; the women facing one another, and the men. In silence the soup was finished — excellent, if a little thick; and fish was brought. In silence it was handed. The fish was taken away, a fine fresh sole from Dover. And Bilson brought champagne, a bottle swathed around the neck with white. Soames said: "You'll find it dry!" Cutlets were handed, each pink-frilled about the legs. ACl{They were refused by June, and silence fell.} Soames said: MAINl["You'd better take a cutlet, June; there's nothing coming." But June again refused], so they were borne away. "Salad, sir?" Spring chicken was removed. But Soames was speaking: AC2{The asparagus is very poor, Bosinney,} a glass of sherry with your sweet? AC3{June, you're drinking nothing!}" June said: MAIN3["You know I never do, Wine's such horrid stuff."] An apple charlotte came upon a silver dish. "Sugar, please, Bilson." Sugar was handed her, and Soames remarked: MAIN2["This charlotte's good!"] The charlotte was removed. Olives from France, with Russian caviare, were placed on little plates. The olives were removed. Lifting her tumbler June demanded: "Give me some water, please." Water was given her. A silver tray was brought, with German plums. In perfect harmony, all were eating them. (John Galsworthy).

The axiological structure of (8) is more complex than that of (3) and (4). Before the first AC ("refused") the discourse is axiologically balanced in the sense that it contains words and linguistic expressions which are either neutral or positive (dinner, begin, silence, the soup, excellent, fish, fine, fresh, etc.). The discourse develops along the AB expressing the canonical, purposeful succession of events, i. e. serving and eating the meal. Dishes

Axiology

543

and drinks are brought and served. The participants accept them and subsequently what is left is taken away. The axiological balance is lost when the first AC (AC1) occurs. It is expressed by the definitely negative word refuse, which expresses June's different values associated with the course in question (the cutlets). AC1 evokes MAIN1, in which Soames attempts to negotiate the issue: ("You'd better take a cutlet; there's nothing coming"). Yet, June refuses again. The next AC (AC2) occurs when Soames negatively evaluates the asparagus: ("The asparagus is very poor"). AC2 does not immediately evoke any MAIN since Soames' comment about the poor quality of the asparagus is not taken up by either Soames himself or by anyone else present. However, at this point AC3 occurs when Soames realizes that June is not drinking anything: ("June, you're drinking nothing"). AC3 does evoke a MAIN (MAIN3), in which June overtly expresses her negative evaluation of wine: ("You know I never do. Wine's such horrid stuff."). Thus, the two MAINs differ in that whereas MAIN1 is negotiative, MAIN3 is declarative. In the course of negotiative MAINs an attempt is made on the part of one or more of discourse participants to modify the system of values of other participants. N o such attempts are undertaken in declarative MAINs, whose only purpose is to make certain evaluative statements. MAIN2 ("This charlotte's good!"), which is a delayed reaction to AC2 is also declarative. MAIN2 restores the initially positive load necessary in obtaining the successful completion of the event (the family dinner). What follows MAIN2 is not interrupted by any further ACs and the event ends following the BALANCE schema, which is overtly expressed by the sentence "In perfect harmony, all were eating them".

11. Some linguistic exponents of ACs Both AC's and MAINs may have various overt linguistic exponents. ACs are prototypically associated with negation, either overtly expressed or implied. The negation may concern individual lexical items or propositions. Thus, in (4) the negation constituting the AC is overtly expressed by the words "We haven't a frig", which are accompanied by the expression of apology "Sorry". The pragmatic impact of this negative proposition is equivalent to "You cannot have beer" of some other negative expression, referring to the ««availability of beer. The AC consists in the fact that the unavailability of beer constitutes an obstacle on the path to carrying out the goal of the intercourse, viz. having beer. The ensuing

544

Tomasz P. Krzeszowski

MAIN restores the balance and renders it possible to reach the goal. In (7) the negation inherent in the AC is not overtly expressed but it is quite clear that the grasshopper does not share the ant's attitude to the relevance of work. The ant's positive effort towards a goal is contrasted with the grasshopper's idleness. The initial balance (positive effort) is thus ruined by the grasshopper's negative attitude to work. Examples of ACs expressed by individual words can be found in (5) and in (8). In (5) the AC is marked by the word stupid, expressing the rooster's negative appreciation of the donkey's joke, and in (8) the word refuse is an indirect rendition of an explicit denial "I do not want any cutlets".

12. Some linguistic exponents of MAINs MAINs are expressed through a variety of linguistic means, all of which involve, among other things overt exponents of some values. Thus, for example, in (6) and in (7) the respective MAINs contain explicit evaluations expressed by the most prototypically evaluative words (Krzeszowski 1990) bad/good: "You're as bad as they are, if you're with them", "if you know what's good". Various other explicit evaluations appear in the MAINs of the quoted discourses (eg. better, poor, horrid, perfect). Instead of being expressed directly, evaluations can be implied. In (4) I'm not very keen on hard liquor, from the axiological point of view, is roughly equivalent to I consider hard liquor to be bad.

13. The Structure of MAINs It is perhaps interesting to observe that MAINs have their own structure which we have no room to discuss in this paper. Let me only note one aspect of the structure connected with the coherence of MAINs based on a specific kind of what might be called axiological logic. In (4) asking for small Scotch rather than large Scotch is coherent with the negative evaluation of Scotch {I'm not keen on hard liquor), which automatically reorientates the values in the well known metaphor MORE IS BETTER (Lakoff and Johnson 1980: 22 ff.). The metaphor is cognitively valid if and only if the target domain is initially evaluated as positive. If the target domain is initially evaluated negatively, the metaphor LESS IS BETTER becomes cognitively valid. Therefore, asking for small Scotch with a lot of soda (the more soda one has in one's drink the less whisky

Axiology

545

one is likely to have) is coherent with the system of values subscribed to by the relevant person. Observe the bizzare effect of a sentence which would not be coherent with the system summarized by the metaphor LESS IS BETTER: A big one, if you don't mind, I'm not keen on hard liquor, uttered in reply to What about a Scotch?

14. Conclusion I have tried to show that the axiological parameter, obligatorily present in the semantic description of words and other linguistic expressions, has a direct impact on well-formedness of discourses and texts. Specifically, it manifests itself in a rich axiological structure, heretofore unnoticed, which constitutes one of the factors accounting for discourse coherence. The axiological parameter makes it possible to formulate some generalizations about coherence of discourses which cannot be formulated in traditional terms. Among them are discontinuities of certain stretches of discourses, such as the one encountered in (4b), incoherences or interruptions which inevitably result if ACs remain unresolved. It goes without saying that the present paper shows only a tip of an iceberg which must be explored in the course of future investigations.

Notes 1. Danielle Forget (personal communication) notes that in Canadian French the particle Ιά is used in a conversation to mark breaks (stops!) in order to make sure that the listener is following (!) the speaker. The pragmatic content of the particle is clearly coherent with the J O U R N E Y metaphor and amounts to the enquiry "You there, are y o u still with me?" 2. It is conceivable that, as Wallace Chafe (personal communication) has suggested, (3) may entail some negative axiological aspects, for example ensuing from the fact that the female discourse participant received a large double brandy while the male participant poured himself a smaller one, which may induce one to speculating about possible evil reasons of trying to get the female participant drunk. However, (3) does not contain any overtly expressed A C , since regardless of its motives, the goal inherent in this dialogue is having a drink, and this goal is reached without any interruptions. A n A C could occur if the discourse participants made the issue explicit, for example by disagreeing over the amount of alcohol that can be safely consumed.

546

Tomasz P. Krzeszowski

References Brown, Gillian and Yule, George 1983 Discourse analysis. Cambridge, London, New York, New Rochelle, Melbourne, Sydney: Cambridge University Press. Chesterman, Andrew 1980 "Contrastive generative grammar and the psycholinguistic fallacy". Papers and Studies in Contrastive Linguistics 11: 17 — 24. Halliday, Michael and Hassan, Ruqaia 1976 Cohesion in English. London: Longman. Krzeszowski, Tomasz P. 1989 "The Axiological Parameter in Preconceptual Image Schemata". Duisburg: L. A. U. D. A 254. 1990 "The axiological aspect of idealized cognitive models", 1991 In: Jerzy Tomaszczyk and Barbara Lewandowska-Tomaszczyk (eds.) Meaning and Lexicography. 135 — 165 John Benjamins Β. V.: Amsterdam. Lakoff, George Women, Fire, and Dangerous Things. Chicago and London: The University 1987 of Chicago Press. Lakoff, George "The Invariance Hypothesis: Do Metaphors Preserve Cognitive Typol1989 ogy?" Duisburg: L. A. U. D. A 266. Lakoff, George and Johnson, Mark 1980 Metaphors we live by. Chicago and London: The University of Chicago Press. Leech, Geoffrey 1974 Semantics. Harmondsworth: Penguin Books. Lyons, John Semantics. Cambridge, London, New York, Melbourne: Cambridge Uni1981 versity Press. Reddy Michael "The Conduit Metaphor", in: A. Ortony (ed.) Metaphor and Thought, 1979 284—324. Cambridge: Cambridge University Press. Scheler, Max 1916 Der Formalismus in der Ethic und die materiale Wertethik. 1973 Formalism in Ethics and Non-Formal Ethics of Values. Translated by Manfred S. Frings and Robert L. Funk, Evanston: Northwestern University Press. Tischner, Jözef "Etyka wartosci i nadziei" [The ethics of values and hope], in: Wobec 1982 wartosci [Facing values], Poznan: Wdrodze, Wydawnictwo Polskiej Prowincji Dominikanow. Widdowson, Henry 1979 Explorations in Applied Linguistics. Oxford: Oxford University Press.

Partv At the interface of linguistics and pragmatics: individual linguistic structures

On the (In)dependence of syntax and pragmatics: Evidence from the postposition -rä in Persian* Mohammad Dabir-Moghaddam

1. Introduction This paper intends to discuss and explain the distribution of the postposition -rä in Persian — an enclitic which has evaded satisfactory description despite the considerable attention it has received from both Iranian and non-Iranian linguists and grammarians (see Section 2 for a brief review of previous scholarship on -rä). Teachers of Persian to nonnatives and text editors have also voiced the problems involved in teaching, learning (Phillott 1919: 451 and Windfuhr 1979: 47), and correct usage of this postposition. This paper (Section 3) will show that the pragmatic function of -rä is a natural projection of the syntactic role of this postposition. It will also show that despite this dependence of syntax and pragmatics, there are clear differences between the syntactic and pragmatic behaviour of this postposition. Finally, (Section 4), implications of this study for general and text linguistics will be elaborated on.

* In this article the following notations and abbreviations are used: [a] low front vowel [ä] low back vowel (However the phonetic symbols used in the quoted examples will be reported as they appear in the original texts.) PS past Ε Ezafe/genitive marker R restrictivity marker (marking the head N P in restrictive relative clauses) I N D indicative mood In the Persian examples which I have quoted from different sources morphemic segmentations, glosses, and translations, if not given in the original sources, are provided by me. As function of -rä is the topic of this article, no gloss is provided for this postposition. A number enclosed in brackets to the right of the translation of a Persian example refers to the number of the quoted example in the original text.

550

Mohammad

Dabir-Moghaddam

2. Review of Previous Scholarship Traditional grammars of Persian have generally described -rä as a marker of (definite) direct object (Khanlari 1976: 3 1 9 - 3 2 0 ) and Lambton 1953 [1984]: 4). Sentence (1), taken from Lambton, exemplifies such a usage. (1)

ketab-ra be man dad-0 book to I gave-s/he 'He gave the book to me'. [Section 6]

In a squib Browne (1970: 359 — 363) argues that -rä is a marker "... of [+specific] objects, not of [ +definite] ones" (362). (Lambton 1953 [1984]: 131—132 presents similar examples to hint at the same conclusion). In example (2), taken from Browne, -rä is used with the N P having the indefinite marker -/, or yek 'one, a' or both together. (2)

man (yek) ketäb-(i) -rä did-am I one book -asaw-I Ί saw a, some book'. [13]

Peterson (1974: 71 —76) claims (a) that -rä is a marker of topicalization and (b) that "... in Persian specificity is a necessary condition for topicalization" (71). As specificity as a feature of the NPs followed by -rä was mentioned in the previous paragraph, here I will only focus on Peterson's first claim. He argues that this claim is based on the following "... facts about the distribution of -rä ... It can be attached to virtually any N P in a sentence, with two notable exceptions: subjects and objects of prepositions. It can occur only once in a simplex sentence." (72) To substantiate his assumption of -rä as a topic marker he reports example (3) below from Lazard (1957) "... in which an N P is placed at the beginning of a sentence and pronominalized in the main clause." (73) (3)

äqä Kucik-rä ceqadr xarj-e tahsil-es kard-am Mr. K. how much expenditure-Ε education-his made-I Lit. 'As for the little man [ = Mr. K.] how much expenditure for his education I have made!' [33]

In a footnote, Peterson (1974: 79, fn. 14) mentions "the rule that -rä cannot be attached to a subject has one fairly common exception: if a subject N P is the head noun of a relative clause and if its counterpart in the clause is an object, this subject can be marked with -rä." Lambton (1953 [1984]: 76) reports the same observation and cites sentence (4) as an illustrative example.

Persian postposition

(4)

551

Zan-i-ra ke did-id inja-st-0 woman-R- that saw-you here-is-she 'The woman whom you saw is here', [section 7]

Phillott (1919: 144) and Khanlari (1976: 319) have also discussed similar examples. They have labelled these cases "... erroneous but now common ..." and "... erroneous ...", respectively. Lazard (1982: 177 — 207) argues that the presence of -rä gears to a number of heterogeneous factors such as "... degre de «definitude» et d'« humanitude » de I'object, degre de plenitude du verbe, distance semantique du verbe et de l'objet, poids relatif des groupes syntaxiques, structure de la « visee communicative»" (177) [... the degree of the definiteness and humanness of the object, degree of the fullness of the verb, the semantic distance between the verb and the object, the length of the object, communicative goals]. In his article Lazard cites an example in which two occurrences of -rä appear in a simplex clause contradicting Peterson's claim quoted above that "It [-rä] can occur once in a simplex sentence" (Peterson 1974: 72). This example is quoted in (5) below, -o is the spoken form of -rä when the noun to which it is attached ends in a consonant; when this noun ends in a vowel the spoken form will be -ro. (5)

portoqäl-o bäyad puss-es-o kand bad xord orange must skin-it peel then eat 'Les oranges, il faut les eplucher avant de les manger (literally: il faut enlever leur pelure, puis mangery [As for orange, one must peel it before one can eat it]. [92]

He calls the first instance of -o in (5) "quasi objet polarise" [polarized quasi-object] and its second instance "objet polarise" [polarized object] (203). Windfuhr (1979: 47 — 57) has reviewed previous studies on -rä which ends with Peterson's analysis sketched above. In the closing paragraph on -rä he mentions without any discussion "... as yet undiscussed instances where rä follows a relative clause ..." (57). The only example given by Windfuhr is reported in (6). (6)

Towzih: än ke explanation that that ketäb-e 'dar läbelä-ye book-Ε inner folds-E [yek nosxe az ketäb-e one copy of book-Ε

Mahmud Μ. äsär-e works-Ε xod va his and

Haqiqi, nevisande-ye H. author-E Jamälzäde', J. boride-ye excerpt-E

552

Mohammad

Dabir-Moghaddam

bahsha-i [ke dar bäre-ye Jamälzäde discussions-R that about-E J. dar keyhän cäp sode bud]] rä hamräh bä in K. publish become was together with näme Ϊ baräye ostäd ersäl däst. (Keyhän, 12 letter a for master sent Amordad 1354, P. 12) 'The explanation being the fact that Mahmud Haqiqi, the author of the book' the inner folds of Jamalzade's work' sent [a copy of his book and excerpts of discussions [which had been published about Jamalzade in Keyhan together with a letter]] (-rä) to the master'. [There is no sentence number in the original text.] (The direct object phrase and the relative clause embedded therein are marked by brackets.) In her dissertation Karimi (1989) has adopted Government — Binding as her theoretical framework (p. 9). In her long discussion of -rä (pp. 53 — 124) she suggests conditions in (7) below for the occurrence of this spostposition. These conditions simply guarantee that subjects and objects of prepositions do not take -rä. (7)

"Specific Oblique Case rä indicates oblique ( # [ + NOM]) case on an NP if the latter is a. [+specific] b. not in the minimal government-projection of a, a = N, A, or P." (p. 86, no. 74)

In a footnote (p. 117, fn. 27) she mentions that she has used the term oblique in its classical meaning. That is "... it is a collective term for all cases other than nominative case". Conditions in (7) easily explain the presence of -rä in sentences like (2) above. Karimi also discusses constructions in which a clitic pronoun is coreferential with an NP appearing in an Α-bar (topic) position in the same construction (p. 85). She calls these constructions "clitic binders" (p. 87) which she claims to be base-generated (ibid). According to Karimi (p. 86), in clitic binder constructions, the NP coreferential with the clitic pronoun inherits the case of the clitic. Furthermore, she adds a principle, which she calls Extended Locality principle (p. 85), which specifies S' as the minimal government projection between a clitic pronoun and the NP coreferential with that pronoun.

Persian postposition

553

Thus, according to Karimi (p. 86) in sentence (8), though the clitic -es is oblique it does not take -rä because it is preceded by the preposition be 'to'. The proper noun Sepide, on the other hand, meets all the conditions specified in (7), hence it inherits the case of its clitic pronoun -es, i.e., oblique, and is obligatorily followed by -rä (ibid.). Karimi uses -rä as a gloss for -o or -ro. (8)

sepida-ro be-h-es goft-am Sepide-ra to-her told-I 'As for Sepide, I told her.' [72]

As another instance of clitic binder construction, Karimi (p. 87) presents example (9) in which two occurrences of -rä appear in a single S' (cf. sentence (5) above for a similar example). (9)

otäq-o dar-es-o bast-am room-ra door-its-ra closed-I 'As for the room, I closed its door.' [75]

As Karimi (p. 87) puts it "the specific direct object dar-es 'its door' receives [-NOM] case from the verb closed. The noun phrase otäq, being coreferential with es, inherits specificity and [-NOM] case from the clitic pronoun." Furthermore, none of these NPs is preceded by a preposition. "Consequently, rä shows up in both cases, marking them for specificity and oblique case." (ibid.) Finally example (10 a) from Karimi (p. 95) shows clitic binder constructions in which the N P and its coreferential clitic pronoun are not in the same simplex S'. The underlying structure of (10 a) as suggested by Karimi is given in (10 b) below. (10)

a. ki-ro goft-i pedar-es raft orupä who-rä said-you father-his went Europe 'Who did you say his father went to Europe?' [91c] b. s [ki-ro V p[goft-i s [pedar-es vp[raft orupä]]]] [93]

Karimi (p. 96) argues that though ki and -es in pedar-es, which is the subject of the embedded clause, are coreferential ki does not inherit the case of -es because "they are not in the same minimal governmentprojection", i. e., they are not contained in the same simplex S'. Hence, ki and pedar-es will receive their cases independently. According to Karimi (p. 97) as ki meets all the conditions specified in (7) above, it will be marked with -rä. Before closing this section, it should be noticed that items such as (4) above which are labelled "... common exception ..."; "... erroneous but

554

Mohammad

Dabir-Moghaddam

now common ..."; and "... erroneous ..." as well as instances where -rä follows a relative clause as exemplified in (6) remain as unanswered questions. My analysis of rä in the next section will account for these cases.

3. A new proposal for the treatment of -rä This section consists of two parts (3.1 and 3.2). In (3.1) I will present a diachronic sketch of the syntactic changes in -rä in four stages in the history of Persian. This discussion, I believe, will shed light on the present synchronic behaviour of this postposition which is the topic of part 3.2.

3.1. Diachronic evidence According to Kent (1950: 205), the New Persian -rä appears as the postposition rädiy which means 'on account of in Old Persian (6c B. C. — 3c B.C.). Item (11), taken from Kent (1950: 116-117), exemplifies its usage in an Old Persian text. The translation is from Kent (1950: 119, Part 3). Colon shows word-boundary. (11)

avahyarädiy: vayam: Haxämanisiyä: Oahyämahy 'For this reason we are called Achaemenians'.

Middle Persian, more specifically Pahlavi (225 A . D . —651 A.D.), grammars have mentioned a variety of syntactic functions for the postposition räy/räd in this language. According to Brunner (1977: 149) "The designation of cause was the basic use of OP [ = old Persian] rädiy and remained a major use of räy." Brunner (p. 171) claims that "Indication of Indirect Object" is another use of this postposition. He adds " R ä y is abundantly used to indicate a beneficiary." (ibid.) The expression in (12), among others, is presented in Brunner (p. 172) to substantiate this usage. (12)

xwästag rawän räy paydäg kardan 'To declare property for the benefit of one's soul'.

Brunner (p. 173 and 174) finally claims that "use of the postposition [i. e., räy] with the direct object of a verb is a late development found ... in Phi [ = Pahlavi] texts." Heston (1976: 215 and 216) suggests that "in Pahlavi, clauses of possession usually have a copula ... a post-position

Persian postposition

555

(ray) is used to indicate the possessor." Example (13) is one of the (transliterated) examples provided by Heston (p. 216). (13)

LK WM I 'd BRE-l A YT' 'You have a son (lit., of you (pi.) a-son is)'.

In classical Persian (around 1000 A.D.), equally called Early New Persian, texts -rä marks indirect objects, direct objects, and possessors in constructions with a copula indicating possession. In a quantitative study Lazard (1970: 381—388) has analyzed the frequency of the occurrence of -rä in a number of texts belonging to classical Persian Period and a few other texts written in the present century representing contemporary Persian. In this study, the occurrences of -rä have been arranged in two groups. The first group includes examples which can undergo Passivization. Example (14), taken from Lazard (p. 383), belongs to this group. In (14a) the NP followed by -rä, i.e., kudakän 'children', is a direct object. (14 b) is the passive counterpart of (14 a). (14)

a. fer'own kudakän rä mikost 'Pharaon tuait les enfants'. [Pharaoh used to kill children.] b. kudakän koste misodand 'Les enf ants etaient tues.' [Children used to be killed.]

The second group contains examples which cannot be passivized. That is the NP which takes -rä in these examples is either indirect object or possessor in constructions with copula indicating possession. (15) and (16), from Lazard (p. 383), exemplify the former and the latter usage, respectively. (15) (16)

firuz rä goftand On dit ä Firuz' [PRO (pi.) told Firuz.] Samarqand rä cahär dar-bud 'Samarkand avait quatre portes'. [Lit. 'Of S. four door was.'; S. had four gates.]

In a table Lazard (p. 384) shows that out of 642 occurrences of -rä in his selection of classical texts 329 of them belong to the first group and 292 of them to the second group and the remaining 21 have been listed as ambiguous cases. On the other hand, out of 317 occurrences of -rä in contemporary Persian texts, 261 of them belong to the first group and 20 of them to the second group and the remaining 36 have been considered

556

Mohammad

Dabir-Moghaddam

as ambiguous cases. It appears to me that those figures clearly show a change in the syntactic function of -rä in the direction of its stabilization as a direct object marker in contemporary Persian. Moreover, the classical Persian constructions with a copula indicating possession have been reanalyzed in contemporary Persian as constructions in which the possessor (formerly followed by -rä) now appears as the subject of the main verb dästan 'have'. Accordingly, sentence (17) would be a contemporary Persian version of (16) which belongs to classical Persian. (17)

Samarqand cähär dar däst S. four door had 'S. had four gates'.

-0 -it

On the basis of the observations and discussions in this part, the syntactic functions of -rä in Old, Middle, Classical, and Contemporary Persian may be summarized in (18). (18)

Old Persian Middle Persian Oblique > Oblique, IO, Possessor, D O > Classical Persian Contemporary Persian IO, Possessor, D O > D O

These changes and developments clearly show the Accessibility Hierarchy (cf. Keenan and Comrie 1977) being played backward.

3.2. Synchronic Behaviour of -ra In this paper I assume a close link between syntax and pragmatics. Thus a sentence is considered as a pragma-syntactic unit. In the unmarked situations one of the correlations shown in (19) and (20) could be postulated between the syntactic module and the pragmatic module of a two-place predicate in Persian. (19)

Primary Topic t Subject

Secondary Topic t Direct object

(20)

Primary Topic I Subject

->i, mo'allem-e

s[ke

man be-s, näme nevest-am]] mo'allem-e I to he letter wrote-I 'The man whom I wrote a letter to is a teacher.'

b.

HP[mard,-i-ro

a.

N P [mard r i

$[ke man be-s, näme nevest-am]],

X, mo'allem-e

s[ke 0, bä man sohbat kard-0]] mo'allem-e with talk did-he 'The man who talked to me is a teacher.'

b. *NP[mardri-ro

s[ke

0, bä män sohbat kard-0]],

χ

mo'allem-e

It may be noted that sentence (44 b) is similar to sentence (4) reported above which is described by Peterson (1974: 79 fn. 14) as "common exception" to "the rule that -rä cannot be attached to a subject". (Other scholars have labeled these cases" ... erroneous but now common ..."; or "... erroneous ..." see Section (2)). In my analysis -rä is not attached to the subject; the subject is the resumptive pronoun which is deleted by the nonemphatic subject pronoun drop process. A particular class of examples clearly support my analysis. They are genitive constructions whose modifier is a relative construction. When the relative construction in those examples is left dislocated, a resumptive pronoun shows up in

Persian postposition

567

the original place of the moved construction and the left dislocated relative construction is marked with -rä. Sentence (47 b) below, whose addressee was myself, supports this observation. (47 a) is the reconstructed deep structure of this sentence and (47 c) provides another left dislocated possibility from (47 a) in which the whole genitive construction is fronted hence the resumptive pronoun has no surface manifestation. (47)

a. tah-E -is[ke to tu-Sj qazä NP\päkatj bottom bag- R that you in-it food mi- gozär-i]] päre ast IND.-put-you torn is 'The bottom of the bag that you put food in is torn.' b.

NP [päkat-i-ro

s[ke to

tu-Sj

qazä mi-gozär-i\] tah-E-sit pare ast

c. tah-E MP|päkat-i-ro s[ke to tu-s qazä mi-gozär-i]], öü päre ast Examples (48) —(50) illustrate the application of LD to relative constructions originally in indirect object position. In (48) the head NP is coreferential with the direct object in the relative clause, in (49) it is coreferential with the indirect object in that clause, and in (50) LD is blocked 1 because the head NP is coreferential with the subject of the relative clause. After LD, the resumptive pronoun along with the indirect object marker be 'to' surface. (48)

a. (man) peyqäm-E somä-rä be NP[avvalin äsnä-i I message you to first friend s[ke 0j bebin- am]\ xäh-am dad see I will-I give Ί will give your message to the first acquaintance that I will see.' b. w[avvalin äsnä,-i-rä s[ke 0, bebin-am]], (man) peyqäm-E somä-rä be w, xäh-am däd

(49)

a. (man) peyqäm-E somä-rä be NP[avvalin äsnä,-i s[ke bäs, sohbat kon-am]] xäh-am däd Ί will give your message to the first acquaintance that I will talk to.' b.

NP [avvalin

äsna,-i-rä s[ke bas, sohbat kon-am]], (man) peyqäm-E somä-rä be u, xäh-am däd

568

(50)

Mohammad

Dabir-Moghaddam

a. (man) peyqäm-E somä-rä be m.[mardri s[ke 0j dar xäne bud-0]] däd-am in house was-he gave-I I gave your message to the man who was in the house.' b. * NP [mard-i-rä s[ke0i dar xäne bud-0]], (man) peyqäm-E somä-rä be Ui däd-am.

The examples in (44) —(50) further substantiate the dependence of the pragmatic function of -rä and its syntactic role — that relative constructions will be affected by LD and will be marked with -rä if the head N P in these constructions is coreferential with any NP in the relative clause except the subject. So far I have dealt with part of the title of this paper — that is I have presented evidence which show a close dependence between the syntactic role and the pragmatic function of -rä. In what follows I will provide three pieces of evidence in support of the independence of the two. While NPs such as ketäb in (21), repeated in (51 a) below, in which the syntactic role and the pragmatic function of -rä are merged may be affected by the passive rule 2 as shown in (51 b), left dislocated (or topicalized) NPs marked with -rä can not be passivized. 3 The ungrammatical sentence (52 b) which is the result of the application of passivization to (52a), i.e., (35b), supports this observation. (51)

(52)

a. man ketäb-rä xar-id-am I book buy-PS-I Ί bought the book.' b. Ketäb (tavassot-E man) xar-id-e by PS-participle sod-0. became-It 'The book was bought by me.' a. (man) hamsäyerro be mehmän-eSj xand-id-am I neighbour to guest-his laugh-PS-I b. *hamsäyej be mehmän-es, (tavassot-E man) xand-id-e sod-0

It was argued above that LD of a relative construction in which the head NP is coreferential with the subject of the relative clause produces ungrammatical sentences e.g., (46 b) and (50 b). However, as (53) shows when a relative construction is in direct object position its head NP may be coreferential with any NP, including the subject, in the relative clause.

Persian postposition

(53)

569

barädar-am NP[mardri-rä &[ke 0, be xäne-E mä brother-my that to house our ämad-0]] ne-mi-senäs-ad came-he not-IND-know-he 'My brother does not know the man who came to our house.'

A comparison of (46 b) and (50 b) on the one hand, and (53) on the other clearly shows the different behaviour of the pragmatic -rä and the syntactic -rä. As another piece of evidence in this regard it may be noted that while syntactic -rä in sentences such as (53) may surface after the head N P or equally at the end of the whole relative construction as supported by (54), such a movement in (44 b), (45 b), (48 b), (49 b) and indeed in all similar relative constructions where -rä is used in its pragmatic function is impossible. The ungrammatically of (55), which is based on (44 b), supports this observation. (54) (55)

barädar-am senäs-ad *^p[mard-i

NP[mardri

s[ke 0, be xäne mä ämad-0]] -rä ne-mi-

™ 0 , diruz did-im]] -ro, mo'allem-e

The movability of -rä in relative constructions provided that they appear in direct object position explains Windfuhr's observation exemplified in (6) above.

4. Implications In Section 3.1 of this paper it was argued that -rä is diachronically stabilized as a marker of direct object in contemporary Persian. In Section 3.2 it was shown that this syntactic role is projected to mark secondary topicalization; hence a close dependence between the syntactic role and the pragmatic function of -rä was suggested. It was also shown that despite this dependence there is evidence in favour of the independence of the syntactic role and the pragmatic function of this postposition. These observations imply that neither form-based explanations, which account for the data presented in this paper in the syntactic component, cf. Karimi (1989) which is based on Government — Binding model, nor the function-based explanations, which claim that "... a fully coherent theory of language must begin at (and not merely include) the level of discourse MOTIVATION for individual sentences" (Hopper and Thomp-

570

Mohammad

Dabir-Moghaddam

son 1980: 295) are adequate linguistic models. The paper suggests that though syntactic and pragmatic principles constitute independent modules, there is a great deal of dependence between the two, hence it recommends a parallel and mutual study of the two. Another implication of the discussions in this paper is based on the diachronic developments of -rä. These developments which were discussed in Section 3.1 clearly hint at (a) the teleological nature of the linguistic changes and (b) the diachronic verification of the Accessibility Hierarchy. Still another implication of this study is that all instances of A-raising and B-raising which have been suggested in the linguistic literature of Persian, e. g., Soheili-Isfahani (1976: chapter 6) and Hajati (1977: chapter 3), are inaccurate characterizations of the phenomenon. Raising does not exist in Persian; all cases which resemble raising are indeed instances of L D (e.g., examples (42) and (43)). Finally text editors and teachers of Persian to natives and nonnatives should also be warned of the existence of a syntactic -rä and a pragmatic -rä. In unmarked two-place predicates represented in (19) above the two functions merge.

Notes 1. The claim that L D will be blocked in case the head N P and the subject of the relative clause are coreferential may eventually turn out to be too strong at least for cases when LD moves a relative construction from an oblique object position. That is to say, in those cases L D applies but -rä does not appear with the head noun. A comparison of items (i) a.—c. below supports this observation. (0

a. man I

soräq-E

somä-rä

az -

inquiring

you

from

s[/ce

0; ämad-0]]

that

came-he

NP[har

any

kas,-i body-R

gereft-am took-I

Ί inquired about you from anybody who came.' b. *Ne[har kas,-i-rä s[ke 0, ämad-0)], man soräq-E somä-rä az u, gereft-am c.

Nl.[/iar

kasri s]ke 0, ämad-0]], man soräq-E somä-rä az μ, gereft-am

At this stage I do not intend to generalize the possibility exemplified in (i) c. to examples (46) and (50) in the text because in the former after L D (and the subsequent deletion of the nonemphatic subject pronoun) the resulting construction will be exactly like (46 a), hence there is no clear-cut indication that LD has actually taken place. In the latter after the application of L D a construction like (50 b) but without -rä will be produced. This construction still sounds ungrammatical.

Persian postposition

571

2. The presence of -rä with a direct object is not a necessary condition for passivization. The only restriction in this regard is that the NP to be affected by the passive rule must be a direct object. The direct object in example (22) in the text, repeated in (i) a. below, is not marked with -rä but as (i) b. shows it is passivized. (i)

a. man ketäb-i xar-id-am I book-a buy-PS-I Ί bought a/some book.' b. ketäb-i

(tavassot-E man) xar-id-e sod-0 by -participle became-it Ά/some book was bought by me.' xn For a diachronic and synchronic description of passive in Persian see Dabir-Moghaddam (1982 b). 3. An apparent exception to this general rule is dative movement in Persian which I consider to be an instance of left dislocation (and compound verb formation). After dative movement, the syntactic role and the pragmatic function of -rä merge again. In example (i) below, c. is the result of dative movement and b. is the output of the b. application of passivization to b. (i) a. mädar qazä-rä be bacce däd-0 mother food to child gave-she 'The mother gave the food to the child.' b. mädar bacce,-rä be -ί, qazä däd-0 'The mother gave the child the food.' c. bacce (tavassot-E mädar) qazä däd-e sod-0 'The child was given food by the mother.'

References Brown, Gillian — George Yule 1983 Discourse analysis. Cambridge: Cambridge University Press. Browne, Wayles 1970 "More on definiteness markers: Interrogatives in Persian", Linguistic Inquiry 1.3: 3 5 9 - 3 6 3 . Brunner, Christopher 1977 A syntax of Western Middle Iranian. Delmar, New York. Caravan Books. Chomsky, Noam 1981 Lectures on Government and Binding. Dordrecht: Foris publications. 1985 Knowledge of language: Its nature, origin, and use. New York: Praeger. Dabir-Moghaddam, Mohammad 1982 a Syntax and Semantics of Causative Constructions in Persian. [Unpublished Ph. D. dissertation. University of Illinois, Urbana.] 1982 b "Passive in Persian", Studies in the Linguistic Sciences, 12.1: 63 — 90. 1987 "Morphological Causatives in Persian", Proceedings of the 11th Annual Meeting of the Kansai Linguistic Society of Japan, 60 — 70.

572

Mohammad

Dabir-Moghaddam

Hajati, Abdol-Khalil 1977 Ke Constructions in Persian: Descriptive and Theoretical Aspects. [Unpublished Ph. D. dissertation, University of Illinois, Urbana.] Halliday, M . A . K . 1985 An Introduction to Functional Grammar. London: Edward Arnold. Hedayat, Sadegh 1977 Buf-E Kur, new edition, Javidan Publishing Company. Heston, Wilma 1976 Selected Problems in Fifth to Tenth Century Iranian Syntax. [Unpublished Ph. D. dissertation, University of Pennsylvania.] Hopper, Paul — Sandra Thompson 1980 "Transitivity in Grammar and Discourse", Language. 56.2: 251 —299. Karimi, Simin 1989 Aspects of Persian Syntax, Specificity and the Theory of Grammar. [Unpublished Ph. D. dissertation, University of Washington.] Keenan, Edward — Bernard Comrie 1977 "Noun Phrase accessibility and universal grammar", Linguistic Inquiry 8: 63-99. Kent, Roland 1950 Old Persian, New Haven, CT, American Oriental Society. Khanlari, Parviz 1976 Dastur-E Zabän-E Färsi [Persian Grammar] published by bonyäd-E. Farhang-E irän. Lambton, Ann K. S. 1984 Persian grammar. Cambridge: Cambridge University Press. Lazard, Gilbert 1957 Grammaire du Person Contemporain. Paris: Klincksieck. 1970 "Etude quantitative de revolution d'un morpheme: la postposition rä en Persan", in: David Cohen (ed.), Melanges Marcel Cohen. 381—388. The Hague/Paris: Mouton. 1982 "Le morpheme rä en Persan et les relations actancielles", Bulletin de la Societe de Linguistique de Paris 73.1: 177 — 208. Levinson, Stephen 1983 Pragmatics. Cambridge: Cambridge University Press. Li, Charles (ed.) 1976 Subject and topic. New York: Academic Press. Peterson, David 1974 Noun phrase specificity. [Unpublished Ph. D. dissertation, University of Michigan.] Phillott, D. C. 1919 Higher Persian grammar. Calcutta: Calcutta University. Plank, Frans (ed.) 1984 Objects: Towards a theory of grammatical relations. London: Academic Press. Riemsdijk, Henk — Edwin Williams 1986 Introduction to the Theory of Grammar. Cambridge: The MIT Press. Soheili-Isfahani, Abolghasem 1976 Noun Phrase Complementation in Persian. [Unpublished P h . D . dissertation. University of Illinois, Urbana.]

Persian postposition

573

Windfuhr, L., Gemot 1979 Persian Grammar: History and State of its Study. (Trends in Linguistics.) State-of-the Art Reports 12, The Hague: Mouton Publishers. 1987 "Persian", in: Bernard Comrie (ed.), The World's Major Languages. London: 523 — 546. Croom Helm.

VP inversion and aspect in written texts* Gregory L.Ward —Betty J. Birner

Introduction While the syntax of English inversion constructions has been the recent subject of considerable debate (e.g., Levin 1985; Green 1985; Bresnan —Kanerva 1989; Levine 1989; Culicover —Rochemont [in press], inter alia), far less attention has been paid to the pragmatics of these constructions. Notable exceptions include Green (1980, 1982) and Hartvigson —Jakobsen (1974), who posit a number of discourse functions for the various types of inversion. C o m m o n to all work on inversion with which we are familiar, however, is the assumption that inverted sentences and their canonical word order counterparts are semantically equivalent. That is, as Green (1980: 582) notes, there seems to be no truth-conditional difference between an inversion and its canonical word order variant. Nonetheless, we have discovered a non-truth-conditional asymmetry between certain types of inversion and their subject-verb-object (SVO) counterparts. 1 These inversions, in which a VP appears in sentence-initial position, are literary in nature and particularly characteristic of journalistic prose (Green 1982: 134 — 139). In certain contexts, these inversions permit an interpretation not shared by the corresponding SVO variants. Consider (la) and (lb) in the following context: (1)

In the last few weeks before yesterday's final vote in Washington, many Senators changed their positions on the new income tax proposal. a. Senator Jones of Ohio was voting in favor of the bill. b. Voting in favor of the bill was Senator Jones of Ohio. c. 'Senator Jones of Ohio voted in favor of the bill.'

* We would like to thank David Dowty for many useful discussions of the issues raised in this paper and, in particular, for suggesting the Gricean approach adopted herein. We are also indebted to Beth Levin, Bradley Pritchett, and Richard Sproat for helpful comments. Finally, we would like to thank Georgia Green for her much-appreciated insights into the syntax, semantics and pragmatics of inversion constructions, which have inspired this and related work. As always, all errors are ours alone.

576

Gregory L. Ward— Betty J. Birner

From the SVO variant (la) one could plausibly infer that the completed, or perfective, interpretation in lc does not hold (e. g., that Senator Jones changed his mind and ultimately voted against the bill). From the inversion in lb, on the other hand, one could reasonably infer that Senator Jones did in fact vote for the bill in question. In this paper we investigate the nature of this asymmetry. We argue that the aspectual asymmetry illustrated in (1) is the result of a syntactic constraint on inversion interacting with neo-Gricean pragmatic principles.

Previous studies As noted by Bennett - Partee (1972); Dowty (1979); and Stump (1985), inter alia, a sentence in the past progressive, unlike the preterit, does not necessarily entail perfectiveness. 2 Consider, for example, the accomplishment predicate in (2): (2)

a. Sixteen-year-old Johnny Smith swam across Lake Michigan yesterday. b. Sixteen-year-old Johnny Smith was swimming across Lake Michigan yesterday.

If it is true that Johnny Smith swam across Lake Michigan, then it follows that at some interval he was swimming across Lake Michigan. However, the entailment is unidirectional in that (2b) does not entail (2a): Unlike (2a), (2b) can receive an imperfective interpretation, i. e., that Johnny did not swim completely across the lake. That non-completion of the swim in question contradicts (2a) but is consistent with (2b) can be seen in (3): (3)

a. Sixteen-year-old Johnny Smith swam across Lake Michigan yesterday. # Unfortunately, he got caught in a bad storm and never made it to the other side, b. Sixteen-year-old Johnny Smith was swimming across Lake Michigan yesterday. Unfortunately, he got caught in a bad storm and never made it to the other side.

The oddness of (3a) results from the perfectiveness of the preterit contradicting the subsequent assertion of imperfectiveness; with the past progressive in (3b) there is no entailment of perfectiveness and thus no contradiction of the subsequent assertion. Correspondingly, a subsequent assertion of perfectiveness is redundant only in the case of the preterit, as illustrated in (4):

Inversion

(4)

577

a. Sixteen-year-old Johnny Smith swam across Lake Michigan yesterday. # In fact he made it all the way to the other side, b. Sixteen-year-old Johnny Smith was swimming across Lake Michigan yesterday. In fact he made it all the way to the other side.

In (4a), the assertion of completion in the second clause is redundant given the preceding preterit. In (4b), however, completion is not entailed by the past progressive, hence no redundancy. In fact, use of the past progressive may license an inference of now-completion, as in (la) above and the following exchange: (5)

A: Lately, many high school students have been attempting ridiculous stunts to impress their friends. Yesterday, sixteen-yearold Johnny Smith of Martinsville was swimming across Lake Michigan. B: How far did he get before he gave up?

In (5), Β infers that Johnny did not in fact complete his swim across the lake. Thus, we can conclude that the progressive and preterit sentences in the above examples stand in a unidirectional entailment relation. Given this relation, the imperfective interpretation associated with the past progressive can be accounted for pragmatically, as a case of conversational implicature involving Grice's (1975,1978) maxim of Quantity. Under this view, a speaker's use of a "weaker" or less informative form can implicate that (the proposition associated with) a "stronger" or more informative form is false, unknown, or otherwise inappropriate (cf. Horn 1972, 1984; McCawley 1978; Gazdar 1979; Dowty 1979; Hirschberg 1985; inter alia). Such an approach is taken by Stump (1985) in his analysis of the perfect tense. Stump argues that a relation of mutual entailment holds between the perfect tense, illustrated in (6a), and the preterit tense, illustrated in (6b): (6)

a. John has seen Mary, b. John saw Mary.

In Stump's theory, the two tenses have identical truth conditions and differ only pragmatically, with the additional interpretation associated with the perfect tense being analyzed as a case of generalized conversational implicature. Citing McCawley's (1978) "principle of less linguistic effort", Stump claims that a speaker's use of the "more effortful" (6a) will implicate that the "less effortful" (but equally truthful) (6b) does not

578

Gregory L. Ward—Betty

J. Birner

apply. 3 Given the greater structural and semantic complexity of the perfect tense, it is argued that its use will implicate inapplicability of the preterit. As stump (1985: 233) puts it, "someone using a present perfect sentence like [6a] conversationally implicates that s/he has a reason for not using the corresponding preterit [6b], namely that John's seeing Mary happened during some extended now which is especially salient." The pragmatic approach to tense and aspect outlined in Stump 1985 can be straightforwardly extended to cover the past progressive/preterit alternation discussed earlier. However, unlike the mutual entailment relation which holds between the perfect and the preterit, we have seen that the entailment relation between the past progressive and the preterit is unidirectional. Given this relation, the use of the past progressive can generate a conversational implicature involving Horn's (1984) lowerbounding "Q-Principle": "Make your contribution sufficient; say as much as you can." From this principle, it follows that the use of the relatively marked, morphologically more complex progressive Q-implicates that the less marked and more informative preterit does not hold. Thus, a speaker's use of the past progressive — in an environment where the preterit form is structurally possible — can implicate that the use of the less marked, i.e., preterit, form is inappropriate. In the case of (2b) above, use of the past progressive can thus induce an imperfective interpretation, given the existence of the "competing" preterit form in (2a).

Aspectual asymmetry and inversion Now we are in a position to address the question posed at the outset regarding the aspectual asymmetry illustrated in connection with (1) above. That is, why does the inverted past progressive in (lb) seem to allow a perfective interpretation, while the corresponding SVO progressive in l a does not? The answer lies in the fact that, unlike the SVO progressive, the inverted variant has no preterit form with which it can "compete", and therefore lacks the implicature of imperfectiveness identifiable with the SVO progressive. Consider (7) —(8): (7)

a.

Pennsylvania was among six states to boycott Soviet-made vodka in protest of the Soviet attack, Liquor Control Board officials said yesterday ... Joining Pennsylvania in the boycott of the Soviet-made vodka in state liquor stores were Alabama, Iowa, Ohio, New Hampshire and West Virginia. [Philadelphia Inquirer, p. 1 - A, 9/8/83]

Inversion

579

b. *Join(ed) Pennsylvania in the boycott of the Soviet-made vodka in state liquor stores (did) Alabama, Iowa, Ohio ... (8)

a.

b.

Alabama, Iowa, Ohio, New Hampshire were joining Pennsylvania in the boycott vodka in state liquor stores, Alabama, Iowa, Ohio, New Hampshire joined Pennsylvania in the boycott of the in state liquor stores.

and West Virginia of the Soviet-made and West Virginia Soviet-made vodka

Here we see that while the past progressive and preterit forms compete in the SVO examples in (8), such competition is ruled out on syntactic grounds in the case of inversion, since the preterit does not undergo inversion (cf. [7b]).4 Therefore, no implicature of imperfectiveness is generated with the participial inversion, since there is no "less marked" perfective inverted form with which the progressive could compete. 5 As a result, the progressive inversion is free to receive either a perfective or an imperfective interpretation, as evidenced by (9): (9)

Joining Pennsylvania in the boycott of the Soviet-made vodka in state liquor stores were Alabama, Iowa, Ohio, New Hampshire and West Virginia ... a. ... until the national liquor lobby intervened, preventing them from actually initiating any boycott. b. ... for a total of six states involved in the boycott.

The felicitous final phrases in (9a) —(9b) demonstrate that the inverted participle is "perfectively neutral" in the sense that neither a perfective nor an imperfective interpretation is induced by virtue of the progressive's form or (semantic) meaning alone. This observation is consistent with Stump's (1985: 261) analysis of the progressive as simply a predicative construction consisting of a copula and a present participle phrase. 6 Interestingly, we find that even when the progressive is anomalous in canonical word order, the corresponding inverted progressive is nonetheless possible. Consider the SVO past progressive in (10): (10)

There were only classes of people in 18th century France: the very rich and the very poor. # O v e r 97% of the population was falling into the latter category.

As Dowty (1979) and Stump (1985), inter alia, have argued, the anomaly of participles like that in (10) can be attributed to semantico-pragmatic factors. Given the complete temporal overlap of the progressive and the

580

Gregory L. Ward-Betty

J. Birner

preterit for stative predicates, there is nothing that one form conveys which is not also conveyed by the other. In Hornian terms, then, there is nothing for the more complex and marked progressive to convey which the preterit doesn't convey as well, hence the infelicity of employing the more marked form. However, consider the inverted stative progressive in (11): (11)

a. Falling somewhere in a category between Einstein's theory and sand fleas — difficult to see but undeniably there, nevertheless — is the tropical 'city' of Islandia, a string of offshore islands that has almost no residents, limited access and an unlimited future. (Brown Corpus 47 1744) b. 'The tropical "city" of Islandia ... falls somewhere in a category between Einstein's theory and sand fleas ...'

Here, the stative idiom fall into a category can felicitously appear as an inverted progressive, as in (11a), with the interpretation of ( l i b ) . Such an interpretation arises in the absence of an inverted simple form with which the inverted participle in (11a) could compete. Again, we see that inverted progressives correspond to both the SVO simple present and the SVO progressive forms; inverted stative predicates are different only in that one of the two corresponding SVO forms is ruled out on independent pragmatic grounds. 7 Similarly, when the simple form is deviant in SVO order, as in (12a), the corresponding inversion may nonetheless be well-formed (12b) and receive a progressive interpretation (12c): (12)

a. # Many Chicago businessmen learn Japanese this summer. b. Learning Japanese this summe are many Chicago businessmen. c. 'Many Chicago businessmen are learning Japanese this summer.'

In fact, it is consistently the case that given a well-formed SVO simple present, preterit, or progressive predicate, there exists a corresponding and semantically equivalent inverted progressive. Thus, we conclude that the absence of a Hornian Q-implicature associated with inverted progressives arises from the absence of a competing inverted simple form. This, in turn, follows from a syntactic constraint in English on inversion, to which we now turn.

Inversion

581

The syntax of VP inversion We have seen that the interpretation of inverted progressive VPs follows naturally from independently motivated pragmatic principles, given that progressives may undergo inversion while verbs in the simple present and preterit tenses may not. The question remaining, then, is why the simple forms do not occur in inversions. We claim that this restriction follows from a general syntactic constraint allowing VP inversion only when auxiliary be is present. 8 Consider the paradigm in (13) —(15): (13)

a. John Smith swam across the Atlantic. b. John Smith was swimming across the Atlantic. c. *Swam/swim across the Atlantic (did) John Smith. d. Swimming across the Atlantic was John Smith.

(14)

a. John Smith runs a mile every morning. b. John Smith is running a mile every morning. c. *Run(s) a mile every morning (does) John Smith. d. Running a mile every morning is John Smith.

(15)

a. John with will win the diving competition. b. John Smith will be winning the diving competition. c. *Win the diving competition will John Smith. d. Winning the diving competition will be John Smith.

Here, we see that the preterit in (13c), the present in (14c), and the future in (15c) all lack auxiliary be, and are thus excluded from inversion contexts. All of the corresponding progressive sentences (13d) —(15d), on the other hand, do contain auxiliary be and are fully grammatical. However, it is not the case that all auxiliary verbs allow inversion. Consider the examples of auxiliary have in (16): (16)

a. John Smith has run a mile every morning. b. John Smith has been running a mile every morning. c. *Run a mile every morning has John Smith. d. Running a mile every morning has been John Smith.

Here we see that the auxiliary verb have does not license the inversion of the VP in the perfect tense (16c). Thus, on the basis of such examples, we conclude that the presence of auxiliary be is both a necessary and a sufficient condition for grammatical VP inversion. 9 An apparent counterexample to our claim that the presence of auxiliary be is sufficient to allow VP inversion is provided in (17):

582

(17)

Gregory L. Ward—Betty

J. Birner

a. # Being over 6 feet tall was John Smith, b. # J o h n Smith was being over 6 feet tall.

Given the presence of auxiliary be in (17a), why is it nonetheless illformed? We maintain that the source of the deviance of (17a) is pragmatic rather than syntactic. Consider again the inverted stative present participle in (18): (18)

a.

Falling somewhere in a category between Einstein's theory and sand fleas — difficult to see but undeniably there, nevertheless — is the tropical 'city' of Islandia, a string of offshore islands that has almost no residents, limited access and an unlimited future. [ = 12a] b. # T h e tropical 'city' of Islandia, a string of offshore islands that has almost no residents, limited access and an unlimited future, is falling somewhere in category between Einstein's theory and sand fleas — difficult to see but undeniably there, nevertheless.

Recall that Stump (1985) and Dowty (1979) analyze SVO stative predicates such as (17b) and (18b) as pragmatically deviant. However, the inverted VP in (17a) is deviant while that in (18a) is not. Again, the difference can be attributed to the presence of a "competing form" for the inversion in (17a), namely the AdjP inversion in (19): (19)

Over 6 feet tall was John Smith.

That is, given the complete temporal overlap of the progressive in (17a) and the preterit in (19), there is no reason for a speaker to use the more complex and marked progressive form in (17a) given the existence of the simple preterit in (19). However, in the case of (18a), there exists no less marked or less complex inverted form for the inversion to compete with, hence no infelicity arises from its use. 10 Thus, we conclude that the infelicity of (17a) does not counter our claim that auxiliary be is a necessary and sufficient condition for grammatical VP inversion. We have argued that the aspectual asymmetry found in connection with inverted progressives is the result of there being no inverted simple form with which the inverted progressive can compete. When competing inverted forms are possible, then aspectual symmetry obtains between SVO and inversion. Consider the passives in (20). (20)

a. John Smith was driven out of office. b. John Smith was being driven out of office.

Inversion

583

c. Driven out of office was John Smith. d. Being driven out of office was John Smith. Here, the presence of auxiliary be licenses the inversion of both the simple passive VP (as in [20]) and the passive progressive VP (as in [20d]). As a result, these two inverted forms compete in the same way as do the corresponding SVO passives (20a) — (20b), and we therefore find the same implicature of imperfectiveness in the inverted progressive in (20d) that we found in connection with the SVO progressive in (5). As an example of this implicature in the case of inverted passives, consider the discourses in (21): (21)

a. Last year's crop failure bankrupted a number of farmers in California's Central Valley. Being forced to sell his farm was 64-year-old Sam Gustafson of Bakersfield. Fortunately, he inherited a great deal of money and didn't have to sell after all. b. Last year's crop failure bankrupted a number of farmers in California's Central Valley. Forced to sell his farm was 64-yearold Sam Gustafson of Bakersfield. /Fortunately, he inherited a great deal of money and didn't have to sell after all.

As was the case for SVO word order, the competition between the inverted progressive and simple forms in (21) licenses the inference in (21a) that Farmer Gustafson did not in fact end up selling the farm. An imperfective interpretation in (21b), however, is infelicitous in the context of the inverted (non-progressive) passive. Thus what would otherwise be a puzzling exception to the asymmetry displayed by inverted progressives is straightforwardly accounted for.

Conclusion In this paper we have argued that the aspectual asymmetry between inverted and SVO progressives is the result of an interaction between a syntactic constraint and neo-Gricean pragmatic principles. Specifically, we have shown that VPs can invert only around auxiliary be. Given this syntactic restriction, the asymmetrical temporal and aspectual interpretations associated with inversion and canonical word order can be straightforwardly accounted for by appeal to general pragmatic principles. Use of the SVO past progressive, for example, can implicate that a perfective interpretation is inappropriate, given the existence of another

584

Gregory L. Ward- Betty J. Birner

(less marked) SVO form — the preterit — which could also be used to convey this interpretation. Since the simple form is disallowed in inversion on syntactic grounds, there is no alternative form with which the inversion competes; thus the implicature associated with the SVO. progressive does not obtain. Where there is a competition of forms, as in the case of inverted passives, the aspectual interpretation of inverted and non-inverted forms is entirely symmetrical.

Notes 1. Our use of the term inversion is not meant to imply the acceptance of a multi-stratal syntactic analysis which involves movement of the VP. In this paper, we remain neutral with respect to the correct syntactic analysis of inversion; however, see Birner (in prep.) for a review and discussion of the syntactic issues related to inversion. 2. By preterit, we mean the simple past tense, as in John left; by progressive, we mean the progressive aspect, regardless of tense, i.e., a copula and a participial phrase, as in John isjwas leaving; by perfect, we mean the perfect tense, as in John has/had left; and by perfective, we mean "completed", as in Laid off from his job, John applied for unemployment, where the past participial free adjunct is interpreted perfectively. See Stump (1985) for discussion. 3. See Horn (1984) for a critique of McCawley's "principle of less effort". 4. In the following section, we propose a syntactic principle to account for this restriction. 5. An interesting implication of our analysis is that inverted and non-inverted sentences are not "competing alternatives" in the sense of Horn (1984). That is, the use of a marked syntactic construction (inversion) does not seem to implicate that the unmarked (SVO) word order variant "does not obtain". This suggests that differences between related syntactic constructions do not give rise to Hornian Q-implicatures. 6. Stump provides several other examples of "perfectively neutral" participial phrases, including adnominal participles, augmented adjuncts, and augmented absolutes. For each of these participial constructions, Stump observes, the phrase in question is vague with respect to perfectivity; only the context — and not the linguistic form — will supply the perfective or imperfective interpretation. On the basis of this evidence, Stump (1985: 256) concludes that "the progressive aspect in fact has no independent semantic status in English — that its truthconditions are entirely determined by general semantic properties of present participles." Of course, what these constructions have in common with inverted present participles is that they all occur in environments in which the preterit is not possible, hence there is no form with which the participial phrase can compete. 7. David Dowty has pointed out to us that when fall is used non-idiomatically (and nonstatively), the SVO progressive is fully felicitous: (i)

a. Johnny Smith was falling into the lake, b. Falling into the lake was Johnny Smith.

As before, the inverted variant in (ib) corresponds to two SVO forms: the preterit and the progressive. However, with non-stative fall, both forms (and interpretations) are

Inversion

585

possible with SVO word order, given the temporal distinctness of the two forms in question. Dowty has also provided us with two further examples of Stative fall, in contexts other than inversion: (ii)

a. Any pesticides falling into this category should be labeled. b. Falling into the category of class 1 hypertensives, John is a candidate for a severe heart attack.

In these examples of present participle phrases (see Stump 1985), no infelicity results, despite the Stative aspect of the predicate. Again, we can attribute the well-formedness of these examples to the lack of a simpler competing form which could serve to block the participial forms. 8. There is little agreement in the syntactic literature as to the correct constituent analysis of the English auxiliary; therefore, we shall simply assume the existence of some AUXlike constituent for the auxiliary elements under discussion. However, nothing in the following discussion depends on this assumption. 9. While we are concerned here with VP inversion in particular, the be requirement appears to be more general. That is, with the exception of locative inversion (i), all other types of inversion require be: (i) (ii)

In the garden stands a scarecrow. a. Injured by the gunman was Sandy Thomas, (past participle inversion) b. "Injured by the gunman {appeared/lay} Sandy Thomas. (iii) a. Also angry were three Republican congressmen. (AdjP inversion) b. *Also angry {grew/complained} three Republican congressmen. For evidence that locative inversion is distinct from VP inversion on syntactic, semantic, and pragmatic grounds, see Ward —Birner (1989) 10. The existence of (i) is not a counterexample to our claim regarding the absence of a competing inverted form for (18a), since it is an instance of locative inversion, argued above to be a distinct construction. (i)

Somewhere in a category between Einstein's theory and sand fleas — difficult to see but undeniably there, nevertheless — falls the tropical 'city' oflslandia, a string of offshore islands that has almost no residents, limited access and an unlimited future.

As discussed above (footnote 5), a speaker's choice of one syntactic construction over another (truth-conditionally equivalent) one appears not to give rise to Hornian Qimplicatures; the existence of (18a) and (i) as non-competing forms for expressing the same propositional content supports this conclusion.

References Bennett, Michael — Barbara Partee 1972 "Toward the logic of tense and aspect in English", distributed by the Indiana University Linguistics Club. Birner, Betty i. p. "The discourse function of English inversion".

586

Gregory L. Ward— Betty J. Birner

Bresnan, Joan — Jonni Kanerva 1989 "Locative inversion in Chichewa: A case study of factorization in grammar", Linguistic Inquiry 20. 1: 1—50. Culicover, Peter — Michael Rochemont 1989 English focus constructions and the theory of grammar. Cambridge, U. K.: Cambridge University Press. Dowty, David 1979 Word meaning and Montague grammar. Dordrecht: Reidel. Gazdar, Gerald 1979 Pragmatics: Implicature, presupposition, and logical form. NY: Academic Press. Green, Georgia 1980 "Some wherefores of English inversions", Language 56: 582 — 601. 1982 "Colloquial and literary uses of inversions", in: Deborah Tannen (ed.), Spoken and written language: Exploring orality and literacy. 119 — 53. Norwood, NJ: Ablex. 1985 "The description of inversions in Generalized Phrase Structure Grammar", in: M. Niepokuj et al. (eds.), Proceedings of the 11th Annual Meeting of the Berkely Linguistics Society. 117—146. Berkely: Berkely Linguistic Society. Grice, H. Paul 1975 "Logic and conversation", in: Peter Cole—Jerry Morgan (eds.), Syntax and semantics. 3. Speech acts. 41 — 58. NY: Academic Press. 1978 "Further notes on logic and conversation", in: Peter Cole, (ed.), Syntax and semantics. 9. Pragmatics. 113—28. NY: Academic Press. Hartvigson, Hans — Leif Jakobsen 1974 Inversion in present-day English. Odense: Odense University Press. Hirschberg, Julia 1985 A theory of scalar implicature. [Ph.D. dissertation, University of Pennsylvania.) Horn, Laurence R. 1972 On the semantic properties of logical operators in English. [Ph. D. dissertation UCLA. Also 1976, IULC.] 1984 "Toward a new taxonomy for pragmatic inference: Q-based and R-based implicature", in: Deborah Schiffrin (ed.), GURT 84. Meaning, form, and use in context: linguistic applications. 11—42. Washington, D. C.: Georgetown University Press. Levin, Lori 1985 Operations on lexical forms: Unaccusative rules in Germanic languages. [Ph. D. dissertation, MIT.] Levine, Robert 1989 On focus inversion constructions and the role of a SUBCAT stack in phrase structure grammar. [Ms.] McCawley, James 1978 "Conversational implicature and the lexicon", in: Peter Cole (ed.), Syntax and semantics. 9. Pragmatics. 245 — 59. NY: Academic Press. Stump, Gregory 1985 The semantic variability of absolute constructions. Dordrecht: Reidel.

Inversion

587

Ward, Greogry — Betty Birner 1989 "A semantico-pragmatic taxonomy of English inversion", paper presented at the LSA Annual Meeting. Washington, D. C. Ward, Gregory — Betty Birner In prep. "A semantico-pragmatic taxonomy of English inversion".

Scope in discourse: Pragmatics or syntax? Richard A. Rhodes

In this paper I will address several scope phenomena in written English text. 1 Contrary to currently accepted wisdom, I will argue that most of the phenomena I will present are syntactic rather than pragmatic. And although I will draw my examples from written English narrative, the points I will make are general properties of text structure. But I could perfectly well make the same points using as data Ojibwa oral narratives. This said, let us turn to the matter at hand. Over the past decade there has been a shift of thinking in the minds of many of us about what constitutes the proper domain of linguistic study. We have expanded our horizons beyond the assumption that there is only coherent structure in linguistic units at or below the sentence level. The work resulting from abandoning this assumption has brought to light new aspects of sentential syntax and has uncovered evidence pointing to the existence of textual units larger than sentences. But, although we do not lack proposals regarding the organization of multisentential stretches of text, the questions these proposals raise are still quite open. This is largely because we do not have much in the way of reliable syntactic tools for working above the sentence level. Most of the means of garden variety syntax fail us. Attempts to provide traditional generative style treatments of multisentential structures, e. g., van Dijk (1972), don't seem to work. For this reason, we have assumed that the lack of syntactic tools and the absence of satisfying syntactic analyses means that there is no syntax above the sentence level — that pragmatic means are the way to handle suprasentential textual organization. In this paper I want to argue that such a conclusion is premature. I will make some proposals regarding a tool for exploring text level syntactic structure. This tool is based on the grammatical property of scope.2 I will show that there are two types of scope phenomena in text and argue that one type is syntactic and the other pragmatic. Before starting I want to lay out three crucial working assumptions on my part. This is important to do because these assumptions are not widely shared. In fact, I could argue at length for them, but to do so at this point would detract from the point of the paper and I therefore present

590

Richard A. Rhodes

them with only a brief apology and leave it for the reader to decide about them for him or herself. The first assumption is that if a phenomenon can be given a coherent syntactic analysis then it should be given a syntactic analysis as opposed to a pragmatic analysis. For example, the German use of subjunctive in marking extended indirect discourse should be treated as a syntactic phenomenon. The reason for this assumption arises from the fact that language change continuously and uniformly turns pragmatics into semantics into syntax, indicating a strong tendency in the human speech capacity to make syntactic analyses where possible. The second assumption is that closely related phenomena need not be given the same analysis. One may be pragmatic and the other syntactic. The distinction in types of analysis does not vitiate the relatedness. An example of this sort can be seen in L a k o f f s (1987: 462 ff.) discussion of the family of ί/zere-insertion constructions in English. The third assumption is that, relative to the question of whether a given suprasentential phenomenon is syntactic or pragmatic, there is no null hypothesis. Put another way, because of the deep rooted assumption that syntax stops at sentence level, there is a tendency to believe that all suprasentential coherence is calculated, sometimes even in the face of evidence to the contrary. But I contend that there are no grounds for believing this, either in the data itself, or in linguistic theory —apart from the a priori assumption about the primacy of sentence level syntax. Now since it is that very assumption which is being called into question by this line of inquiry, there can be no null hypothesis. Thus even where only suggestive evidence exists, that evidence carries significant weight. We are now ready to move on. In any given text or a portion of text, it is common to find certain words, phrases, or sentences that are understood to have scope over some other nearby stretches of that text and not over others. This scope does not depend on superficial adjacency or even necessarily on superficial proximity. For example, in (1) there is a reading in which the not in the first clause scopes the complement of expect, even though they are not adjacent, and does not scope the expect even though they are adjacent, nor does it matter in (2) that the not and the clause it scopes are widely separated. (1)

I don't expect he'll be coming in today.

(2)

I wouldn't suggest, knowing her tendency to get mightily angry when she's in one of these moods, that you go in right now.

Scope

591

Examples of truly non-adjacent scope such as that in (1) and (2) are relatively rare. It is unimportant for the moment that there are analyses of such cases which rectify the adjacency problem, but it is important that the scope of the negation in sentence level examples like those in (1) and (2) always correlates with the independently establishable unit nature of the stretches of text scoped and not scoped. 3 Given that scope at sentence level always respects the unit structure of constituents, it is of particular interest that there are scope relations above the level of the sentence. Thus examples like that in (3) are very common, in which the temporal adverbial in the first sentence has scope over several contiguous sentences. (3)

A couple of days later Jane was standing at the window of her house in North Town Creek. She saw some smoke rise above the pines from over in the direction of the flight line. Just that, a column of smoke; no explosion or sirens of any other sound. She went to another room, so as not to have think about it but there was no explanation for the smoke. She went back to the window . . . . (The Right Stuff: 14)

The phrase or clause which scopes a section of text does not have to be at the beginning of the section. For example, in (4) the reading is that the I said has scope over the entire excerpt, even though it appears in the middle of it. (4)

"I was there," I said. "Gwalchmai did not kill her. And I'm certain she would have killed him — or killed Medraut, for that matter, if she thought it would help." (Kingdom of Summer: 239) 4

At this point I am not interested in what we call the group of sentences which fall in the same scope. If one wants to think of them as paragraphs, subparagraphs, sentence clusters, or whatever, it doesn't matter. What is important is the fact that they fall within the scope of some surface constituent, and that this constitutes evidence that together this group of sentences has the status of a grammatical unit. A preliminary classification yields at least four kinds of wide scope readings at text level. They are: complement scope, anaphor scope, adverbial scope, and relational scope. I will exemplify each type in turn. Complement scope which is found with verbs of speaking or thinking is very common in narrative. In (5a) there is an example in which seven

592

Richard A. Rhodes

sentences are in the scope of my father said. In (5b) there is a passage in which the quote starting on page 22 and continuing through page 26 is in the scope of [he] went on, which appears at the end of the first sentence in the passage. (5)

a. "They will fail," my father said, his voice shaking, but still quiet, angry. "They are trying to fight Darkness when they have too much Darkness in themselves. Not so much Gwalchmai, nor the Pendragon either, but do you think the Family is made up of such men? Warbands in general care nothing for the fight to preserve civilization or for the Light — oh yes, I know what you're after, I've felt the pull myself — but warbands care for plunder and for glory. Civilization is here, in the order and peace of the householding, and not at Camlann. Look at Gwalchmai. He is a fine man, sensitive and honorable, and even in so pure a warband as Arthur's he's dragged into a crime and made to wear himself away with suspicion and doubt. If he had given himself to land and a clan, and kept them in order, and if the Pendragon had done that, they could have a place where the crime could never have occurred." (Kingdom of Summer: 54) (b) Gwalchmai gave my father a straight, fierce look, then unclenched his hands, rubbed them together and went on. "I went and talked to Bran of Llys Ebrauc. (22)

" 'Yes,' she said, Ί came.' She turned to the servant and said, 'Hywel, could you stay here and watch the horses?' "The old man nodded unhappily, and we walked off together into the green silence of the forest." (Kingdom of Summer: 26) Complement scope has been classified in other ways. Traditional descriptive grammarians recognize the special status of this construction by providing special terminology for the clause containing the verb of speaking or thinking, viz. the quote formula, a term I will freely borrow. In another tradition, Longacre (1976), whose general concern is with the relationships between clauses and clause complexes, proposes a special relation, attribution, to cover just these cases. This special attention seems appropriate enough. There are a number of features of complement scope which demand explanation. Most notably there needs to be an explana-

Scope

593

tion for the normal pattern of complement scope in English in which the quote formula appears, not at the beginning of the quote, but at the end of the first sentence or in a syntactic niche in the middle of it. That is, the sentence syntax of the first sentence in an extended direct quote contains a construction that Ross (1973) analyzed as Slifting. We will have more to say about this later. The second type of scope is anaphor scope.5 Anaphor scope constructions involve demonstratives or other pro-forms, most commonly, this or that referring to clauses, clause complexes, or sentences. Consider the example in (6). (6)

Above all, of course, he would be wired with biosensors and a microphone to see how a human being responded to the stress of the flight. That would be his main function. (The Right Stuff: 180)

In (6) the that in the second sentence scope the purpose clause to see how a human being responded to the stress of the flight, in the first sentence. The most common examples of this construction scope are units at or below sentence level. However there do exist example of wide anaphoric scope like that in (7) and the very striking example in the first sentence of the prologue of Follett's Triple in (8) in which the this scopes the entire novel from chapter one to the end, not including the prologue or the epilogue. (7)

(8)

"Now, you women," he'd say. "I don't want you women to be macking with the brothers if they ain't tending to business. You women make your men get out of the house and get to work for the Youth Coalition. Don't you be macking around with nobody who ain't out working for the Youth Coalition. If he ain't man enough to be out on the street working for the people, then he ain't man enough for you to be macking around with." This worked like a charm with the women and the men, too. (Radical Chic & Mau-mauing the Flak Catchers: 122) They met many years ago, when they were young, before all this [emphasis Follett's] happened; but the meeting cast shadows far across the decades. (Triple: 3)

This this scopes the entire novel from chapter one to the end, not including the prologue or the epilogue. It is important to distinguish the cases of pronominals referring to clauses, sentences, and larger units, which we are here calling anaphor scope, from cases of ordinary pronominalization,

594

Richard A.

Rhodes

in which pronominals are bound to nominally labelable referents in the world of the text. There is a potential for confusion because the binding of referential pronominals over long stretches of text is normally discussed under the rubric of scope. But, I would argue that, in spite of the use of the same terminology, such binding is a very different kind of phenomenon. It does not respect structural units and is of no interest here. The third type of wide scope found in text is adverb scope. It is exemplified in (9). (9)

The next day we rode on to Camlann, arriving at the fortress just after noon. At midmorning we turned from the Roman road onto the raised track that goes through Ynys Witrin and the marshes, and made good speed toward the irregular hills that fill the horizon. (77)

By the time Gwalchmai arrived, at twilight, I had managed to put everything in a place. ... (Kingdom of Summer: 88) In this example, even after 11 pages it is still completely clear when one comes to the phrase by the time Gwalchmai arrived, at twilight that at twilight refers to the twilight of the day that was referred to 11 pages back in the phrase the next day. Again the phrase or sentence need not appear initially in the section scoped, e. g., it was nearly mid-morning in (10), the scope of which extends to the final sentence on the next page. (10)

When we reached the main road it was nearly mid-morning, and the sun had dried the dew. We found a hard bank and led the pony across it and into a break, hidden from the road. Eivlin took some bracken and swept the bank, unnecessarily, I thought, so that anyone following would not be able to tell that we had left the road. (161)

I held her and she put her head down on my shoulder and sobbed loudly while I stroked her hair and made soothing noises. For all the danger and weariness, for all my aches and sickness, that was one of the fullest moments of my life. (Kingdom of Summer: 162)

Scope

595

Examples of this sort of scope are limited to temporal and locative adverbials. Other sorts of adverbials do not have wide scope readings, as is illustrated in (11). None of the adverbials there, for the most part, technically, and politically, can be read as having scope over any more than the sentence they appear in. (11)

For the most part, the men involved in the X-15 program were realistic about the situation. Technically there was no reason why the X-15 should not lead to the X-15B or the X-20 or some other aerodynamic spaceship. Politically, however, the chances were not good and hadn't been good since October 1957, when Sputnik I went up. The politics of the space race demanded a small manned vehicle that could be launched as soon as possible with existing rocket power. And as the Edwards brethern knew, there was no use trying to wish the politics of the situation away. (The Right Stuff: 205)

The final type of scope, relational scope, is somewhat subtler than the first three. In this type stretches of text are understood as relating to one another in ways that some analysts, like Longacre (1976), account for by a propositional calculus. Consider the excerpt in (12). (12)

Gwalchmai looked at him a moment steadily, then sighed. The whole journey suddenly made sense to me. Caer Gloeu stands between Powys, Dumnonia, and the southern wilds of Elmet. Anyone who traveled to any of these lands was likely to stop for the night at Caer Gloeu, and a woman traveling alone, with only a few servants, would probably have been remembered. Our host did not seem to be lying, so it was plain that Elidan had not come to Cear Gloeu, and we could turn back to Camlann in the morning. I was relieved; Gwalchmai was plainly disappointed. He took another drink of his hot ale, then set the mug down. (Kingdom of Summer: 63 — 64)

In this example, the sentence, the whole journey made sense to me, stands in a type of appositive relationship to the stretch of text from the immediately following sentence, Caer Gloeu stands between Powys, Dumnonia, and the southern wilds of Elmet, through Elidan had not come to Caer Gloeu. Because this type of scope is more subtle, I have only included it for completeness sake. I will make no further argument from relational scope.

596

Richard A.

Rhodes

The existence of wide scope readings in the examples above all stand as prima facie evidence bearing on the question of suprasentential syntactic units. But it remains to be explored which of these wide scope readings are to be accounted for syntactically, i.e., by positing multisentence units, and which are only an epiphenomental product of pragmatics. In this paper I can't hope to answer this question definitively. In fact, the best we will be able to do at this time it to put forward some suggestive lines of argument. On the one hand the syntactic interpretation of a phenomenon is based on the view that if one can define a syntactic process, then there must be syntactic units and syntactic constructions. 6 On the other hand, where a formal specification of a construction is impossible, the analysis must be pragmatic. Given that the class of syntactic processes is taken to include substitutions, deletions, and adjunctions, and that substitutions and deletions are often subject to interpretative treatment, let us concentrate on adjunctions which are, in general, not. Thus the most compelling arguments for the existence of a syntactic process come in the form of adjunctions. I will therefore concentrate on arguing that various examples crucially involve adjunction. Let me discuss each of the phenomena mentioned above in turn. First let us consider the class of examples like that illustrated in (5a). The first sentence of that example involves a well-studied rule called Slifting (Ross, 1973), short for Sentence Lifting. Slifting pulls certain types of matrix verbs down into an adverbial niche in the logically embedded sentence. Thus sentences like that (13a) are related to more basic sentences like that (13b) by Slifting. (13)

(a) He's sick, I think. (b) I think (that) he's sick.

On the basis of examples like that (5a) want to argue that Slifting is not an operation on sentences, but is an operation on, for want of a better general term for sentence complexes, paragraphs. Hence I will rename the process of Slifting (Sentence Lifting), Plifting (Paragraph Lifting), and treat sentence level examples as simply involving minimal sentence complexes. For example, the excerpt in (14) has as its normal reading one in which the he said scopes the entire passage. This means that the underlying structure must have the quoted parts of the passage as a unitary complement of say. (14)

"I was concerned the evening when you first disappeared," he said at last. "Degannwy is a hard place, and many things could

Scope

597

have happened. I asked Rhuawn whether he had seen you, and he was affronted and uneasy. 'He was insolent to me,' he said, 'so I gave him a blow. Probably he has run of because of that.' And he told me that I should thrash you." (Kingdom of Summer: 2 0 7 - 8 ) This analysis of direct quotes composed of more than one sentence as syntactically unitary complements of verbs of speaking also has other support. For example, such verbs can passivize and extrapose paragraphs, as in (15). 7 (15)

As it is written: "There is no one righteous, not even one; there is no one who understands, no one who seeks God. All have turned away, they have together become worthless; there is no one who does good, not even one." "Their throats are open graves; their tongues practice deceit." "The poison of vipers is on their lips." "Their mouths are full of cursing and bitterness." "Their feet are swift to shed blood; ruin and misery mark their ways, and the way of peace they do not know." "There is no fear of God before their eyes." (Rom 3: 10, 18, NIV)

If in spite of this evidence one refuses to believe that direct quotes are unitary syntactic objects which participate in a syntactic pattern lowering quote formulae into complement clauses, then one is forced into the only other possibility for analyzing such constructions, which is to treat each sentence as being underlyingly embedded under parallel, and presumably concatenated, verbs of speaking which are deleted by a gapping type of rule, with any internal logic pragmatically determined. Let me argue against this possibility. The logic of (14) is given in (16) in a semi-iconic layout. In this format, the brackets indicate the extent of the scoped sentential complexes, which, incidentally, frequently overlap. 8 The brackets are labelled and the labels index the sentential complexes to the interpropositional relations which govern the scoped stretches of text. The interpropositional relations that

598

Richard A.

Rhodes

appear in caps do not appear on the surface. Also the nearer the left margin a relation is, the wider its scope. 9,10 (16)

he said at last, [,· [j Degannwy is a hard place and T H E R E F O R E , [k many things could have happened to youj] THEREFORE* [ζ I was concerned about you the evening when you first disappeared*] THEREFORE/ [„ [m I asked Rhuawn whether he had seen y o u j and IN VIEW O F THIS m he was affronted and uneasy«,] THEN« [r he said 0 [o

[q

[/> [q

He was insolent to me so THEREFORE,, I gave him a blow p ] Probably because of that (=

I gave

him

a

blow)q

he has run off q ] 0 ]„] and T H E N , he told me, [, [, that I should thrash you BECAUSE, you

ran

away,

]s ]r ],· ]

As a close inspection of (16) will reveal, the logic of (14) differs significantly from its surface form. And while there are any number of points about (16) worthy of further discussion, only one is of importance at this point. Suppose we wanted to analyze (14) as consisting of a series of conjoined instances of say each with one of the sentences of the quote as its complement instead of treating it abstractly as in (16). The fact that the logical relations are not vague, in the technical sense, and the fact that the instances of logical relations among the clauses of this passage cannot also be construed as relations between instances of speech acts, indicates that they are not pragmatically calculated. Therefore the only way to obtain the correct logical relations outlined in (16) under a gapping analysis would be anaphorically. Thus the section labeled k in (16) would have to be as in (17) underlying.

Scope

(17)

599

he said, at last [, I was concerned the evening when you first disappeared BECAUSE O F k,] A N D HE SAID, [, Degannwy is a hard place,] and HE SAID* [k many things could have happened TO YOU BECAUSE O F j k ] u

At first, such a structure looks like a promising way to avoid the syntactic consequences of having a single quote formula commanding both parts of the quote underlyingly. And it also reflects the fact that say with a direct quote complement is an achievement verb in Dowty's (1972) sense, something which is at odds with the non-instantaneous character of speech, 12 and breaking speech acts up in some way might help reduce this discrepancy. However, this is not a viable option. Based on an argument first articulated in Lakoff (1970), we know that this sort of sentential anaphora is interpreted, not indexical. Therefore even under a gapping analysis (17) must be derived from a structure similar to (18), in which the full structure of the argument is in the complement of say, anyway. (18)

he said, at last [, I was concerned the evening when you first disappeared BECAUSE many things could have happened to you BECAUSE Degannwy is a hard place,] A N D he saidj \j Degannwy is a hard place,] and he saidk [k many things could have happened to you BECAUSE Degannwy is a hard place*]13

Thus the complex complement structures we had hoped to avoid by attempting a gapping analysis are forced right back on us by that analysis. 14 Given this result, the only way to save a gapping analysis would be to invoke an ad hoc principle that certain verbs, including verbs of speaking, would be optionally transparent to certain types of logical relations. This principle would have to be optional because in (21) the gapped says must be transparent to both THEN„ and THEN,, but the expressed verbs of speaking in the scope of THEN„ and THEN,, must not be, and we would be faced with the problem of defining the conditions for the transparency.

600

Richard A. Rhodes

Moreover when different speech acts are involved separate constructions often appear — employing explicit verbs for each speech act. For example in (14) there is a clause complex labelled ο under say and a second clause labelled s under tell. This section is repeated here as (19). The clause complex ο is an assertion while s is a suggestion. (19)

I

he said 0 [o [q [p

He was insolent to me so T H E R E F O R E , [q I gave him a b l o w j Probably because of that (= I gave him a blow)q he has run off J J „] and T H E N , he told me, [s [, that I should thrash you BECAUSE, you ran away,] J r] ,]

These examples are not isolated. Many quotes have an internal structure of similar logical complexity. A gapping analysis will have the same set of problems that we have just seen. In the face of such evidence I maintain that a gapping analysis is simply not a viable option to account for the scope of quote formulae, let alone to explain it. This leaves us with the necessity of positing an analysis which involves a syntactic pattern of lowering a commanding S into a niche in the first sentence of a quote. This is clearly an adjunction involving a suprasentential structure. 15 If there are still skeptics, there are plenty of variations on the kind of data we have just seen. All of which can be exploited to make further arguments favoring the syntactic approach to this phenomenon. For example, there are other niches, besides in the middle and at the end of the first sentence. Presumably there is a niche after every sentence. On occasion, one can find examples in which the quote formula in a Plifted construction ends up after the second sentence, as in (20). (20)

"Our prime suspect is Hugh Tiff. He's been trying to buy the diamond for years," said the inspector. "I'm having him picked up." (The Case of the Arrowless Bow, in Sobol 1975)

Most of our examples to this point, show the adjunction of only simple clauses. Many examples of Plifting show the movement of bigger chunks of text. Two examples are given in (21).

Scope

(21)

601

a. "I do not understand," said Rhuawn at last, speaking in Latin so that Maelgwn's men would not understand. "That woman is your mother, the queen of the Orcades, the daughter of the emperor Uther?" (Kingdom of Summer: 107) b. "You won't die," I said, and hauled myself up on one elbow. "As God in heaven is just, you will not die. Believe me . . . . " (Kingdom of Summer: 162)

Cases like (21a) show that multiple clause complexes can be lowered. Cases like (21b) show that conjoined sentences can be lowered. Even more complicated cases can be found. In (22) a sentence that does not contain a verb of speaking is lowered into a quote. The reading is very clearly that the action referred to in the lowered sentence happened during the time of speaking. (22)

"My lord" — I drew a deep breath — "in due respect, I do." (Kingdom of Summer: 131)

Finally, as in (23), even as large a unit as a sentence complex describing an action, series of actions, or states temporally overlapping a speech act can appear lowered into a quote. (23)

"Your lord Arthur." Lot rested his chin on his hand, looking to Gwalchmai; his face was also turned toward me as he did so, and I could see that his eyes were narrow and fierce. "Your lord Arthur. It is true that you have sworn that bastard war leader the Threefold Oath of allegiance?" (Kingdom of Summer: 111)

In all these more complicated cases the quote constitutes a single unit from the point of view of scope, with the intervening sentences related to the expressed or unexpressed verb of speaking by the temporal relation of simultaneity. The second half of the argument for the unit status of multisentential complements of verbs of communicating is that there is no ambiguity about where quotes begin and end. Of course, one might argue that this lack of ambiguity comes from the appearance of quotation marks. In fact, I would argue the reverse, viz. that the unambiguous stretch of text scoped by the quote formula, gives rise to the punctuation, not vice versa. In contrast to the syntactic nature of complement scope and Plifting, adverbial scope, I would argue, is calculated pragmatically. There is basically one argument for such an approach. A pragmatic analysis of

602

Richard A. Rhodes

adverbial scope phenomena would explain why they are limited to temporal and locative adverbials of setting. Every event happens at a time and in a place. It is not logically necessary that events happen in a certain manner, in a certain logical connection, or even for a certain reason. Thus pragmatically we have expectations about the existence of temporal and locative settings for events, and in the absence of explicit information to the contrary carry that information over from the last mention. A pragmatic analysis also explains the fact that there can be vagueness, in the technical sense, associated with wide scope readings of adverbials, as is often the case. At this point I feel it necessary to emphasize that I do not want to argue for the existence of such things as paragraphs, sections, episodes, and the like, per se. Rather I only want to take the position that there exist suprasentential groupings of various sizes. I don't think it is particularly helpful, given our current meagre state of understanding, to try to define levels of groupings. In fact, I would argue that the principles governing the groupings of sentences into units and the groupings of suprasentential units into even more comprehensive units are both complex and recursive and yield many more sizes of groupings than our editorial devices — indentation, line skipping, headings, and so on — suggest, so it is quite premature to attempt a categorization of such groupings (as did Longacre, 1979). In particular the structures involved look to be the product of recursive processes, as in (24) which shows complement scope, but appears in the middle of the section cited in example (5b) which is itself an example of complement scope. (24)

" 'Enough!' Arthur said. 'As before when you set your men to quarrel with mine, you dredge up excuses, and fasten upon a private grievance as the pretext for rebellion.'" (Kingdom of Summer: 37)

Both (5b) and (24) contain Slifted sentences, so the operation giving rise to such constructions must also be recursive. There is further support for the position that the units revealed by scope phenomena are more complex than previous attempts to characterize suprasentential structure will allow for. There is no lack of examples in which the suprasentential grouping demanded by the scope relations does not even correlate completely with the grouping of clauses into superficial sentences, as in (25) in which the anaphoric scope of that in the final sentence has a reading including the unexpressed causal relation between the clauses you said you would find me another lord at Camlann and you didn't want a servant,16

Scope

(25)

603

"No. But you said you would find me another lord at Camlann." Gwalchmai was silent. "You told me you didn't want a servant." "Oh. So I did." He tossed the taper in the fire, stood and leaned against the wall, watching it burn. "I'd forgotten that." (Kingdom of Summer: 90)

Examples such as this suggest that the basic units of discourse level syntax are clause-sized, and that, in addition to other sorts of operations, there are principles governing the grouping of clauses into superficial sentences (cf. Hollenbach, 1974). However, the class of operations involved is quite complex, and the assumption that the sentence is the basic unit of syntactic organization is so fundamental to our conception of syntax that the crucial questions have yet to be asked by most of linguistic community, let alone answered, so I cannot affort the time to lay the groundwork for saying more about this problem here. In summary, I have shown that there are several classes of wide scope phenomena found in written English text. I have examined two of these cases in some detail and found that for one the wide scope arises from the application of a syntactic operation and for the other a pragmatic analysis more adequately accounts for the pattern observed.

Notes 1. This paper draws heavily on ideas f r o m a paper presented at the University of Michigan and at the University of Oregon. The development of these ideas has benefitted greatly f r o m comments and criticisms of b o t h audiences, especially those of Fred Lupke, Pete Becker, and Paul Hopper. 2. I will grant that m a n y syntacticians find semantically based arguments less convincing than those arguments whose semantic basis can be hidden behind paradigms. However, such luxuries are not available to the text analyst. 3. Scope can also operate at word and subword levels, but in no case does it fail to coincide with unit boundaries by applying to only part of one unit and part of another. 4. The page numbers cited in examples refer to the editions listed under Sources. 5. In fact, there are b o t h anaphoric and cataphoric examples. But following the widespread practice I will use the term anaphor to refer to both. 6. This transformational language is a little misleading. W h a t I am really concerned with are metaconstructions. The existence of formalizable metaconstructions guarantees the existence of constructions and, by extension, the existence of syntax. 7. Granted that (18) is, in point of fact, a translation, it is nonetheless flawless English. 8. This is to say that (19) represents a multi-rooted network, not a tree. Only such a network offers a way to account for the fact that m a n y interpropositional relations

604

9.

10.

11.

12.

13.

14.

15.

16.

Richard A. Rhodes have scope over only part of the structures that are defined by the scope of other interpropositional relations. This argument is parallel to the argument made in Morin and O'Malley (1969). This distinction between single-rooted structures and multi-rooted structures could well provide the basis for an explanation for why sentences appear to enjoy a privileged status. I will use a modification of Longacre's system of interpropositional relations here, the listing of which can be found in McCune (1983: 54). For some of the relations, like temporal sequence, the English lexeme which is used as the systematic representation of a particular relation, if taken literally, is at worst awkward, but for other relations, like that of simultaneity, the systematic representation, if taken as a simple English lexeme, can be misleading. The form of (18) is really only schematic. For example, the clauses are written out in surface form. At the level of abstraction which (18) is intended to represent, the clauses themselves would be quite abstract, with indices rather than nouns or pronouns and adverbials separate from the clauses they scope and so on. One concession to abstraction is the use of italics to supply deleted or pragmatically implied material. The reason for presenting a mixed structure like that in (18) comes from the significant difference in readability. (18) represents in comparison to the fully abstract form for which (18) is the shorthand. There are a number of other similar possibilities for analyzing this excerpt. For example, one might not split up the j and k clauses. For the purposes of the point being made here, the differences among such possible analyses are irrelevant. Some of the complexity of the readings of quotes is due to this lexical property of say. Some further complications arise in examples of the sort we are considering when the quote formula has a manner of speaking verb (Zwicky, 1971), but we cannot afford the time to discuss such verbs here. There are a number of other similar possibilities for analyzing this excerpt. For example, one might not split up the j and k clauses. For the purposes of the point being made here, the differences among such possible analyses are irrelevant. Actually the argument is more complex than this sketch of it might indicate. Most types of sentential anaphora must be done by actually connecting the references in a network, otherwise the nasty cases like the case from Triple cited as example (12) above would involve repetitions of immense stretches of text underlyingly, an analysis which I believe is untenable. This leads to the conclusion regarding the underlying structure of texts mentioned in note 8: texts are networks of a type more complex than trees. Lakoff (1974) argued that instances of Slifting are syntactic amalgams, i.e., sentences containing material not in their logical structure but coming from other derivation. None of Lakoff s examples involve direct quotes, and in particular, all his arguments for amalgamation crucially depend on the indirectness of complement. There are a number of possibilities, including the possibility of there being several Slifting patterns with a single target structure. But I should point out that even taking an amalgamation mechanism, the scope relations would still require that in some form there be a structure in which the entire quote is the complement of the verb of speaking, and the argument that I am making will still go through. This passage is ambiguous with respect to the scope of the that. All that is necessary to make this argument is that this passage has the intended reading. In fact, the intended reading is favored by the preceding context of the passage.

Scope

605

Sources Bradshaw, Gillian 1981

Kingdom of Summer. New York: The New American Library. [Signet (451AE1550)].

Follett, Ken 1979 Triple. New York: Arbor House. The Holy Bible 1978 New International Version. Grand Rapids: Zondervan. Sobol, Donald J. 1975 Still More Two-minute Mysteries. New York: Scholastic Book Services. Wolfe, Tom 1971 Radical Chic & Mau-mauing the Flak Catchers. New York: Bantam Books. 1979 The Right Stuff. New York: Farrar, Straus, Giroux.

References Dowty, David 1972

Studies in the logic of verb aspect and time reference in English. [Unpublished Ph. D. dissertation, University of Texas Austin.] Fillmore, Charles J. — D. Terence Langendoen 1971 Studies in Linguistic Semantics. New York: Holt, Rinehart, and Winston. Givon, Talmy (ed.) 1979 Syntax, and Semantics, Vol. 12, Discourse and Syntax. New York: Academic Press. Gordon, David — George Lakoff 1971 "Conversational Postulates." Papers of the Seventh Annual Meeting of the Chicago Linguistic Society: 63 — 84. Grimes, Joseph 1975 The Thread of Discourse. Berlin: Mouton. Gross, Maurice — Morris Halle —Marcel Schutzenberger (eds.) 1973 The Formal Analysis of Natural Languages. Hague: Mouton. Hollenbach, Bruce 1974 Discourse structure, interpropositional relations, and translation. [Unpublished MS.] Jacobs, Roderick A. — Peter S. Rosenbaum (eds.) 1970 Readings in English Transformational Grammar. Waltham, Massachusetts: Ginn and Company. Jones, Linda K. 1977 Theme in English expository discourse. (Edward Sapir monograph Series in Language, Culture and Cognition: 2.) Lake Bluff, Illinois: Jupiter Press. Lakoff, George 1970 "Pronominalization, negation, and the analysis of adverbs", in Jacobs and Rosenbaum (eds.), 145—165. 1971 "The role of deduction in grammar", in: Fillmore and Langendoen (eds.), 63-72.

606 1974

Richard A. Rhodes

"Syntactic amalgams." Papers of the Tenth Annual Meeting of the Chicago Linguistic Society: 321 — 344. 1987 Women, Fire, and Dangerous Things. Chicago: University of Chicago Press. Longacre, Robert An Anatomy of Speech Notions. Lisse: Peter de Ridder. 1976 1979 "The paragraph as a grammatical unit", in: Talmy Givon, (ed.). McCune, Grace 1983 On themantic structure in English exposition: a study of news commentary texts. [Unpublished Ph. D. dissertation, University of Michigan.] Morin, Yves C. — Michael O'Malley 1969 "Multi-rooted vines in semantic representation." Papers of the Fifth Annual Meeting of the Chicago Linguistic Society: 178 — 185. Rosenbaum, Peter S. 1967 The Grammar of English Predicate Complement Constructions. Cambridge: MIT Press. Ross, John R. "Slifting", in: Maurice Gross —Morris Halle —Marcel Schutzenberger 1973 (eds.), 1 3 3 - 1 6 9 Sadock, Jerrold 1977 "Modus Brevis: truncated argument", Papers of the Thirteenth Anmual Meeting of the Chicago Linguistic Society: 545 — 554. van Dijk, Teun 1972 Some Aspects of Text Grammars. Hague: Mouton. Zwicky, Arnold 1971 "In a manner of speaking", Linguistic Inquiry 2: 223 — 232. "Arguing for constituents." Papers of the Fourteenth Annual Meeting of 1978 the Chicago Linguistic Society: 503 — 512.

Part VI Computational modeling of text comprehension

'Parsing' procedures for stories

Yeshayahu Shen

Introduction One of the most influential approaches within the study of story comprehension has been the Story Grammar approach (cf. Rumelhart 1975, 1977; Mandler —Johnson 1977; inter alia). The main idea underlying the various story grammars proposed was that stories are as structured as sentences and can be generated by a generative grammar, that is, by a set of generative rules along the lines proposed by sentence-level grammars. Such a model takes the following form: a story structure is generated by a set of context-free phrase-structure rewrite rules, which correspond to the "syntactic" rules of sentence level grammars; now, as stories are essentially action descriptions, these rules use categories adopted from action theory such as EPISODE, PROBLEM, TRY, and O U T C O M E (cf. Rumelhart 1977). This grammar, namely, the set of syntactic rewrite rules and the action categories they use, operates on propositions as the minimal units of analysis at the story level, corresponding to words (or morphemes) as the minimal units in the analysis of sentences. Such a grammar, then, ends up with a tree diagram representation of stories, in which the terminal nodes are propositions while the structural nodes represent action categories. Clearly, a most crucial question that such a framework raises, is parsing. The problem may be formulated as follows. Given that the very idea of any story grammar is based on assigning a structural description to the propositions of a given story, or (to put it differently) on the mapping of the propositions which constitute the "actual story text" onto its underlying structure, it is evident that an indispensable component of such a theory must be a parser of some sort, i.e., a set of procedures which will assign the required structural description to the propositions of the story. Crucial as it is, however, this problem is perhaps the most neglected area among workers within the domain of story grammar. Thus, virtually no attempt has been made within this tradition to accommodate the system with such parsing procedures, which is perhaps one

610

Yeshayahu Shen

of the main reasons for the rejection of this idea by large portions of the linguistic community. In what follows I would like to present portions of a larger framework which I have developed elsewhere (cf. Shen 1988; 1989) which attempts to take a first step to remedy this lack of parsing procedures, shared by all current story grammars, by proposing a framework as well as actual algorithm for parsing stories. The parsing procedures which will be proposed, have been developed within a version of the story grammar paradigm which I have developed elsewhere (Shen 1986; 1988), and which has been dubbed "the X-Bar story grammar". Let me, then, briefly summarize the main properties of the X-Bar story grammar and then move on to the main part of the paper and introduce the parsing procedure within such a system.

Section 1: The X-Bar story grammar The main difference between the standard story grammars (cf. Rumelhart 1975, 1977; M a n d l e r - J o h n s o n 1977; inter alia) and the XBSG is the incorporation of a central formal property borrowed from the X-Bar theory (developed in sentence level syntactic theory by Jackendoff 1977; Chomsky 1986; and so forth) which has to do with the notion of "A projection category". Consider, for example, the N O U N PHRASE "little boys" as represented in figure 2. Within the X-Bar framework, the dominating N P is a projection category of the corresponding daughter node, the N O U N , as it is of the same category type, namely, the N O U N type. In this notation the dominating NP is called an N ' (where the ""' applies to the number of bars) — namely, a N O U N with one bar expansion (similarly, N" = N O U N with 2 bar expansions etc.). The corresponding dominated category which is of the same type is called the H E A D of the phrase, while the other category dominated by N ' (in our case the ADJECTIVE) is the M O D I F I E R of this HEAD. In this framework there are two types of MODIFIERS, namely, SPECIFIERS and COMPLEMENTS. For now it will be sufficient to point out that in English, and in many other languages, SPECIFIERS precede their HEADs and COMPLEMENTS follow it. In those terms a generalization over the internal structure of all phrasal categories used by such a grammar is proposed, according to which, all the latter conform to the template presented in (1) below: Ο) XP = X-Phrase Μ = MODIFIER

, I Χ

XP

1 I Μ

Stories

611

Note, further that a given lexical category may have more than one bar projection. Thus, the preceding phrase ("little boys") can be extended by adding the D E T E R M I N E R "the", as in: "The little boys" which is (in X-BAR terms) a N" (i.e., a N O U N with two bars) and so on. I will not address here the issue of the constraints of the number of projections a given category may get. The above characteristics of X-Bar grammar result in a restriction form of grammar which imposes the constraint that all structures generated by the grammar must be of the following form as in Table 1. Table 1. The X-Bar rule-format X n —> (SPECIFIER) - X (n " - (COMPLEMENT). (""" = the number of bars.) ("X" = each category in the grammar).

This basic scheme (or rule format) of the X-BAR model which reflects the formal property of "projection" is to be read as follows: The immediate constituents of X" (i.e. category of the type X with Ν bars) are X ( n _ , ) (i.e.), the same category type — X — with ( n _ 1 ) bars), which is an obligatory constituent, and a SPECIFIER and/or COMPLEMENT, which are optional. Having explained the above formal property of an X-Bar system in general, we may now turn to the story level and briefly describe the main characteristics of the XBSG. Roughly speaking, the main difference between sentence and story grammars bears on the type of categories those two systems use: whereas sentence grammars use syntactic categories such as Noun and Verb, story grammars use ACTION categories, such as PROBLEM, TRY and O U T C O M E which are derived from action theory. Let me briefly illustrate the action structure underlying a portion of the Farmer story to be analyzed in subsequent sections. The story propositions are represented in Table 2, while its tree diagram according to the XBSG is to be found in Figure 1. Let us consider, then, the first Episode which consists of propositions 1 —4 a. This episode consists of a causing event (namely, the kid's trodding upon the serpent's tail), which creates an implicit problem: the serpent wants to take revenge of the kid by killing him. The serpent then tries to solve this problem (namely, to accomplish his goal) by biting him, and the O U T C O M E is that the kid indeed dies. Underlying this sequence of

612

Yeshayahu Shen (STORY) I ρ O Τ SPE COM EP

= = = = = =

PROBLEM OUCOME TRY SPECIFIER COMPLEMENT EPISODE

|-Q'-j SPE

Ο

SPE

I

— τ SPE

Ί

I

P'Γ SPE I - Ο '

I Ο

SPE I - T ' SPE

ΪΕΡ3)

Ί

I

P'Γ SPE O ' l f E P 2) I Ο

Γ SPE

Γ SPE

η Τ

I

ΓΤ'~Ί

SPE

r SPE

I

O"

p

η SPE

Γ SPE

-

1 (EP i )

Γ°Ί

rSPE

Τ

o

COM

Τ

1 Τ

I ο I

2

3

4

4a

5

6 7

Figure 1. The tree diagram of the Farmer story

9

10

11

12

13 14 15

(EP4)

Stories

613

Table 2: The farmer and its serpent 1. 2. 3. 4. 4 a. 5. 6. 7. 8. 9. 10. 11. 12. 13. 14. 15.

A countryman's son, by accident, trod upon a serpent's tail. The serpent turned and bit him, so that the died. The serpent left the place. The father, in revenge, got his axe, pursued the serpent with that axe, and cut off part of his tail So the serpent, in revenge, began stinging several of the farmer's cattle This caused the farmer severe loss. Well, the farmer thought it best to make it up with the serpent. So he brought food and honey to the mouth of its lair. and asked for reconciliation. But the serpent refused.

events, then, is a basic action structure which consists of the above three action categories. Combining the aforementioned formal property of the X-Bar system and the Action categories, as well as assuming that the terminal nodes in such a grammar are represented by propositions rather than lexical items, yield the type of story grammar which I dubbed "The X-Bar Story Grammar". Thus, the (formal properties of the) X-Bar system provides the compositional framework for the representation of the Action categories. NP' N P = Noun-Phrase ADJ = Adjective

| ADJ

Ν

little

boys

Figure 2. The phrase structure of the phrase "little boys"

In order to give the reader an idea of what such a grammar might look like, let be briefly introduce the rules of the XBSG and illustrate them by analyzing the Farmer story, presented in Table 2. Figure 1 presents the tree diagram of the structure of this story as generated by the XBSG syntactic rules. Prior to this presentation a key feature of the present grammar should be emphasized. The action categories introduced above, when linearly

614

Yeshayahu Shen

combined, constitute a higher order category, namely — what is traditionally called — the Episode. Thus, the structural unit consisting on a Problem, Try and Outcome constitute an Episode as previously explained. Now, it is crucial to understand clearly how the Episode is conceived of within the XBSG. Based on observations made by various story grammarians it is assumed here that the EPISODE category is represented as a bar projection of the O U T C O M E category; this implies that the H E A D of each EPISODE is its OUTCOME, while the EPISODE'S other constituents (the TRY and the PROBLEM) are conceived of as MODIFIERS of the HEAD. Since the "story" consists of a (usually complex) EPISODE, the above implies that the H E A D proposition(s) of the entire "story" (the SETTING excluded) is the proposition dominated by the O U T C O M E node whose highest projection category is the Highest OUTCOME. The Farmer story, for example, consists of a complex episode dominating a series of four linearly ordered "simple" Episodes. Propositions 1 — 4 a constitute the first Episode, which describes how the serpent killed the farmer's son; the second episode, consisting of propositions 5 — 8, the farmer's actions of revenge against the serpent are described; Episode 3 consists of prop. 9 — 11 and relate the severe loss caused by the serpent to the farmer's cattle, while prop. 12 — 15 depict the farmer's attempt and failure to make it up with the serpent. Each of these Episodes is viewed here as a projection category of its Outcome. Thus, for example, the entire first Episode (propositions 1—4 a) is represented as a projection category of proposition 4, which represents the Outcome. Note further that these four "simple" Episodes together comprise a complex Episode, namely, a projection of an Outcome category whose Head is proposition 15. Having introduced this observation we can move on to the introduction of the generative rules. The grammar is based on three such rules which are presented in Table 3. Table 3. The syntactic rules of the X-Bar Story Grammar 1. X N > (SPECIFIER) - X N (COMPLEMENT) 2. SPECIFIER • X N or a proposition. 3. C O M P L E M E N T > O U T C O M E N or a proposition. Notes: 1. X stands for the categories used by the XBSG, namely, PROBLEM, TRY and OUTCOME. 2. the N stands for the number of bars (namely phrasal extensions) of a given category X.

Stories

615

Rule 1 is the central rule of the XBSG model, since it presents the basic scheme of the X-bar model. As this rule has already been discussed, the following two brief illustrations may suffice. Consider first the highest EPISODE, i. e., the O' dominating the whole story, namely, propositions 1 to 15. The internal structure of this O', as generated by rule 1, consists of a Specifier consisting of propositions 1 — 14 and an Outcome (i.e., on Ο with n _ 1 extensions) comprising of proposition 15 describing the serpent's refusal to make it up with the farmer. Moving down to the lowest O' in the tree, namely, the O' dominating props. 1 to 4 we may now consider the internal structure of that O'. According to rule 1, this node dominates a SPECIFIER (props. 1—3) which describes the events leading up to the boy's death, an Ο (dominating proposition 4) which describes the boy's death. This structure is clearly generated by Rule 1. Rules 2 and 3 generate the internal structure of the MODIFIERS — the SPECIFIER and the COMPLEMENT. Rule 2 states simply that the SPECIFIER of a given phrasal category can dominate either another phrasal category or a proposition(s) (i.e., a terminal node). The O" which dominates propositions 1—4 a clearly illustrates this rule, in that it is a daughter node dominated by a SPECIFIER node. Rule 3 states that the C O M P L E M E N T of a given phrasal category can dominate either an O U T C O M E PHRASE, or a proposition(s) (i.e., a terminal node). Due to time limitations I will not bother to explain the reason for that. Proposition 4 a illustrates the application of this rule. These three rules constitute the syntactic component of the XBSG model and are responsible for the generation of such structures as the one representing the Farmer story. Now, one might reasonably ask what is that grammar all about, and what is it needed for. Part of the answer is this. On top of being a formalism for representing the structure of stories, the XBSG model also claims to be an adequate formalism for representing (aspects of) the processing and storage of narrative text. Elsewhere (Shen 1989) I reported some summary and recall experiments which provide some empirical support for the model. Here are some notes regarding the psychological validation of the model. Two principles render the model into a psychological model of processing a story: firstly, the principle that the basic processing unit of stories is the Episode, which is a projection of the Outcome category and,

616

Yeshayahu Shen

secondly, the principle that the head proposition of any given unit is the most important proposition of that unit. From these principles one can derive a grouping of the story propositions into three levels (or degrees) of "Importance": 1. The highest level of "Importance" consists of the H E A D proposition of the story category (which is a projection of the O')· In the farmer story it is proposition 15 (which describes the failure of the farmer's attempt to make peace with the serpent). 2. The second level of "Importance" consists of the Head propositions of each of the linear Episodes comprising the story and, in addition, the Head proposition of the Try unit of the highest (complex) Episode. In our story this level consists of props. 1, 3, 4, 8, 10, 11 and 14. 3. The lowest level of "Importance" consists of the rest of the propositions, namely, propos. 2, 5, 6, 7, 9, 12, and 13. In summary and recall experiments I have conducted (see Shen 1989) it was found, as indeed predicted by the model, that the probability of occurrence of a given proposition in subject test protocols correlates with its level in the above "Importance" scale, across stories and across subjects. Thus, propositions of the first degree have the highest probability of occurrence, followed by those of the second degree, while the third group has the lowest probability. Thus, the XBSG was found to be a psychologically adequate model for representing aspects of the processing of stories in memory (for a fuller description see Shen 1989).

Section 2: the parser 2.1. Introduction The representation of stories within the XBSG, then, has gained some psychological validity. The tree diagram generated by the XBSG rules can be viewed both as a formalism for representing some structural regularities underlying stories, as well as a model for representing stories in memory. However, the description still leaves a basic issue, unaddressed, which is the main issue of the present paper: How, according to this grammar, are the propositions of a given story to be mapped onto its underlying structure? In other words, how does a reader of a given story assign a structural description to (i.e., identifies the category structure of) the propositions of that story? The problem is this. Unlike sentence grammars which specifies for each lexical item their syntactic

Stories

617

category, any story grammar cannot in principle contain a similar predetermined and fixed lexicon of propositions. Propositions which can neither be listed in such a lexicon nor be determined by any straightforward and automatic procedure mainly because identifying the structural category of a given proposition is context dependent: the same proposition can be identified as a TRY in one context and as an O U T C O M E in another. (This was elaborated elsewhere, see Shen 1986.) In spite of the centrality of this problem, however, neither theories working within the Story Grammar paradigm nor related theories of narrative processing have suggested a "parser" or procedures for it. The proposed parsing procedures aim to solving both of the above problems. Corresponding to these two problems two types of procedures are distinguished: 1. Category identification procedures; these are responsible for the identification of the narrative categories which the propositions constitute. 2. Node construction procedures; these are responsible for the construction of the tree. Before formulating these procedures, a key feature of the present model should be emphasized, concerning the processing assumptions which the procedures must meet. Processing a (narrative) discourse is a linear process in which adjacent units (in our case the propositions) are comprehended step by step, on the basis of their local relations, i.e., the causal relations. Thus it is assumed that processing a story takes the following form: The parser conceives of proposition a', and then proposition b'. He then examines which causal relation holds between these propositions, if any. Only at this stage is he able to identify a's category (on the basis of the Category identification procedures), and to "insert" it under its node in the tree (on the basis of the Node Construction procedures). The importance of this characterization is that it enables the parser to meet a fundamental requirement of any theory which aims at psychological reality, i.e., the requirement that only a limited number of information units be held in Short Term Memory ( = STM) at any stage of the parsing of a given unit. Note that the two procedure-types parse a given story by operating in tandem on each successive proposition. That is, it is not the case that the category identification procedures first assign all category nodes in the entire story, then the node construction procedures build the tree; rather, both procedures operate together to parse first one proposition, then the

618

Yeshayahu

Shen

next, etc. Thus the tree is built up stage-wise, node by node. However, it is not clear at this point whether the procedures are being applied simultaneously or in sequence.

2.2. The parsing procedures Let us, then, introduce the procedures for parsing stories. I would like to emphasize right at the outset that the procedures to be presented represent the (central) part of the entire parser which has been developed elsewhere (see Shen 1989). Due to space limitations, I have limited myself only to those procedures which are sufficient for the parsing of the Farmer story. The fuller parser, however, consists of an additional category identification procedure as well as an additional node construction procedure. The reader interested in the full parser is referred to Shen (1989). As previously explained, a distinction is made between two types of procedures, i.e., the category identification and the node construction procedures (which are presented in Tables 4 and 5, respectively; these procedures will be discussed, shortly). In order to illustrate how this parser analyzes a given story, I will again use the Farmer story which was presented in Table 2; its tree diagram according to the X-Bar Story Grammar was presented in Figure 1 while Table 6 presents the various parsing stages of the entire story, according to the XSBG's parser. Table 4. The Category Identification Procedures Procedure I. Given: proposition a' encauses proposition b' Rule 1: If a' is an intentional mental state, identify a' as the PROBLEM (namely write all projections of a' as PROBLEM) Rule 2: a' is the TRY if a' is an intentional action and — 2.A: a' is performed intentionally in order to accomplish b' or, 2.B: a' is intended to create the necessary preconditions for solving a PROBLEM that was identified in an earlier proposition (x) in the text (where χ can be inferred or explicitly stated). Rule 3: If b' is neither a PROBLEM nor a TRY, identify it as the O U T C O M E . Procedure II. If a' does not encause b', move to c' and Rule 4: If only b' encauses c', than b' starts a new Category in the diagram, to which Rules 1 — 3 above apply (and a' is conjoined to the preceding category).

Stories

619

Table 5. The Node Construction (NC) procedures NCI: If a given node X encauses Y and Y is not a Dead End, attach X to (a projection category) Y + 1 (or: insert X to the tree as a SPECIFIER of a projection ofY. NC2: If X encauses Y and Y is a Dead End, attach Y to = encauses); see parsing rules 1—3 below. II. Procedures which apply to a', b' if a' does not encause b'; see parsing rules 4 — 5 below. Let us briefly discuss each rules separately. Illustrations for the rules will be taken from the parsing analysis of the Farmer story as presented in Table 2. Rule 1 is used in order to identify the PROBLEM. The rule imposes two conditions that a given proposition(s) must meet in order to be identified as a PROBLEM: The first condition might be called the "contextual condition", that is, the requirement (shared by all three parsing rules 1 — 3 comprising Pro-

620

Yeshayahu Shen

Table 6. The Processing of the Farmer story Category Identification Pr. processed & relations —» = encause ' = does not encause

The Construction of the Tree

Rules used R = rule pr. = proced lire

Result

1 —»2

R3

1=0

2—3

R2

2=Τ

3 —> 4

4 —4a

R2

R3

3=Τ

4=0

Result

NC (Node Construction) Rules

0 I 1 NCI

NC1

NCI

ι SPE 1 οΙ 1

T

1 SPE 1 T' 1 -12

T

•

•

1 SPE 1 T' 1 -13

O'

O" -1 COM 1 ο 1 4a

(

)

•

4a, 5

pr. II

move to 6

5 —»6

R3&4

4a = 0

NC2

1 SPE 1 Ο' 1 1-4

5 —»6

R1

5=Ρ

NCI

ι — p' SPE I O' 1 1 —4a

6—7

R2

6=Τ

NCI

ι— SPE 1 P' 1 1-5

T

'

—

—

Stories Table 6. continued Category Identification Pr. processed & relations 7—8

Rules used R2

The Construction of the Tree Result 7=Τ

Node Construction rules NCI

Result ι—T' SPE 1 τ-

•

Ι

1-6 8—9

R3

8=0

NCI

ι — — SPE () 1 T' 1 1-7

9— 10

R1

9=Ρ

NCI

ι— SPE 1 O' 1 1-8

10-.11

R2

10 = Τ

NC1

p

'"—

ι—T

SPE

1 P' 1 -19 11 -> 12

12—13

R3

R1

11 = 0

12 = Ρ

NC1

NCI

1)

ι — —

SPE

C)

1. τΙ 1-10

11

1

P

'

SPE

1 O'

1

1-11 13 — 14

R2

13 = Τ

NCI

1

SPE

12 T

' '

I

P'

1

1-12

13

622

Yeshayahu

Table 6.

Shen

continued

Category Identification Pr. processed & relations 14 — 15

The Construction of the Tree

Rules used

Result

R2

14 = Τ

Result

Node Construction rules NCI SPE I τ I 1-13

R3

15 = 0

Τ

14

NC1 SPE i T' Γ 1-14

Ο

15

cedure I) that the proposition which is identified as a PROBLEM should encause the following propositions. This condition reflects the assumption, previously explained, that the identification of the category of a given proposition is context dependent. Thus the context here is the (immediately) subsequent proposition(s). The second condition imposes a constraint on the semantic content of the proposition in question, namely that it represents an intentional mental state, (e. g. "want", "decide" etc.). This semantic constraint distinguishes the PROBLEM category from the TRY and OUTCOME. Consider, for example, the parsing of proposition 12, which illustrates the above Rule. 12. Well, the farmer thought it best to make it up with the serpent. 13. So he brought food and honey to the mouth of its lair. Proposition 12 meets the above two conditions and therefore is identified as a PROBLEM: 1) It encauses proposition 13 (the farmer's bringing the food and honey to the mouth of the serpent is motivated by his decision to make it up with his opponent); 2) It represents a mental state. According to rule 1, then, proposition 9 is identified as a PROBLEM. The second rule, i. e., the TRY identification rule, also imposes both "contextual" and "semantic" constraints on the proposition to be identified as a TRY. Two points are worth emphasizing with respect to this rule. 1. The "constant" contextual condition (shared by all the rules in procedure I.), requires that the proposition to be identified as belonging to

Stories

623

the TRY category should encause the following proposition(s). Over and above this condition, Rule 2.1 requires that the preceding proposition(s) must also be taken into account (provided that Rule 2.1 does not hold); that is, the parser is assumed to identify an earlier (explicitly stated or inferred) PROBLEM. 2. As for the "semantic" constraint, note that what distinguishes propositions to be identified as a TRY from those to be identified as a PROBLEM is that the former represent an intentional action whereas the latter represent an intentional state. Let us consider the parsing of proposition 3; the relevant row of Table 6 is presented below in Table 7, represents the parsing of proposition 3 (which is also presented below in addition to proposition 4). 3. and (the serpent) bit him, 4. so that he died. Table 7. The category identification of proposition 3 3 — 4

I R2 I

3 = Τ

According to the present proposal the parser first examines proposition 3 to determine whether it encauses the next proposition (i. e. proposition 4), and finds that it does; this is represented by an arrow in the second column connecting propositions 11 and 12. The next step is to identify which rule (from among rules 1 —3) is to be applied to this proposition. Rule 1 is rejected by the parser as it requires the proposition to be a mental state. Rule 2.II is then selected, as this proposition is indeed an intentional action intended to create the necessary preconditions for solving the problem that was identified earlier, namely, the goal to get revenge of the boy. Since Rule 2 applies to proposition 3 it is identified as a Try as can be seen in the tree. Rule 3 is straightforward and needs no further comment. An illustration of this rule is to be found in the parsing of proposition 4. 4. so that he died. 4 a. The serpent then left the place. This proposition encauses the next proposition (proposition 4 a) and thus falls within the domain of rules 1 — 3. After checking, whether the first two rules can be applied to this proposition, the parser rejects both. Proposition 4 does not represent a mental state (and therefore rule 1 is

624

Yeshayahu Shen

rejected) nor does it meet the conditions of either Rule 2.1 or 2.II: it is not performed in order to accomplish the following proposition which describes its leaving the place, and it is not intended to create the necessary conditions for solving the current PROBLEM, since, in fact, it solves this very PROBLEM (that is, it obtains this very goal). Hence, neither rule 1 nor rule 2 applies, resulting in the application of rule 3, according to which proposition 4 is identified as an OUTCOME. Rule 4 is used throughout for parsing propositions which are not causally related. That is, it applies to those cases where there is a "gap" in the continuity of the causal chain of events. As the rule states, given that two propositions a' and b' are not causally related, where only b' encauses the following proposition(s), then a new unit started by b' is constructed, while a' is conjoined to the preceding category. A case in point is propositions 4 and 4 a to which rule 4 applies. The relevant row of Table 6 is, presented below in Table 8 (which represents the parsing of propositions 4, 4 a and 5). 4. 4 a. 5. 6.

so that he died. The serpent went on his way. The father, in revenge, got his axe,

First the parser determines that propositions 4 a and 5 are not causally related, since the fact that the father wanted to get revenge of the serpent (proposition 5) is not encaused by the serpent's leaving the place (prop. 4 a), but rather from the death of his son. This is reflected in Table 8, where 4 a and 5 are connected by the sign "," which means that they are not causally related. Hence the parser must invoke Procedure II, according to which it moves to proposition 6 and tries to determine which of the two rules subsumed under procedure II, (parsing rule 4), can be applied. Since it is only proposition 5 (i. e., the fact that the father wanted to get revenge of the serpent) that causes him to get his axe, (i.e., proposition 6) and not the conjunction of propositions 4 a and 5, rule 4 applies (see the third column of the relevant row of Table 6). According to this rule proposition 5 starts a new EPISODE, whereas proposition 4 a is conjoined to the preceding category as its complement. 2.2.2. Node construction (NC) procedures The stage-by-stage construction of the tree is presented in the right half of Table 6, and consists of two main parts represented in the fifth and

Stories

625

Table 8: The parsing of propositions 4, 4a, and 5 4 —4a

R3

4= 0

NC1

ι—

1

SPE

Ο

τΙ

1-3 4a, 5

pr. II

5—6 •

R3&4

move to 6 5 4a = 0

NC2

I

SPE

O- -,

COM

οΙ

Ο I

4a

1-4 5—6

R1

5=Ρ

NC1 SPE

I O' I 6—7

R2

6= Τ

NCI

1 —4a- T SPE

P' I I

1-5

sixth columns respectively: the node construction ( = NC) procedures presented below in Table 5 which apply to each proposition at a given parsing stage, and the result of applying the procedures, i. e., (sub) tree that is constructed. Node Construction Rule 1 If a given node X encauses Y and Y is not a dead end, attach X and Y to (a projection category) Y + 1 (or: insert X in the tree as a SPECIFIER or a projection o f Y ) . This is the most significant Node construction rule as it is responsible for most node constructions throughout the parsing of simple stories. What this rule comes down to is the idea that for each pair of nodes (save for the case of Dead end which will be addressed shortly) which are causally related, the "encaused node" is the higher in its importance than its encausing one. This observation is based on data obtained in various summary and recall experiments (Rumelhart 1975; Carmeli 1983; Shen 1986; 1988; Forthcoming; inter alia) in which it was found that

626

Yeshayahu Shen

encaused nodes tend, generally, to be retained in memory, while "encausing nodes" tend (fully or partly) to be deleted (all other things being equal) (for a fuller discussion on this topic, see Shen 1988; Forthcoming). This observation, then, is represented in NCI rule in that the encaused node is always (save for the Dead End case) identified as the Head node to which the encausing node is attached as its Specifier. Let us turn now to the other significant notion which appeared in our formulation of the rule, i. e., the Dead End notion. This notion represents a central phenomenon in narrative texts. Its definition is given in (2). (2): Dead End: A proposition b' in a sequence a', b', c', is a dead-end if b' is encaused by a' and b' does not encause c'. This notion applies to a case where a given proposition is encaused by the preceding proposition but in itself does not encause the following proposition(s). The concept is similar to what researchers like Schank (1975) have likewise called a "dead end", that is, a unit (in our case a proposition) which marks a dead end in the narrative sequence of events in that it does not encause the following events. An example case in point is proposition 4 a: 4. so that he died. 4 a. The serpent went on his way. 5. The father, in revenge, Proposition 4 a is encaused by proposition 4 (having got revenge of the boy the serpent could leave the place with full satisfaction); however, leaving the place does not encause proposition 5 since it is neither a necessary nor sufficient condition for the father's revenge plans. Proposition 4 a, then, counts as a dead-end. With the above two definitions out of the way, we may turn to an illustration of the way the parsing rule NCI is applied. Consider, for example, the parsing of propositions 3 — 4 — 4 a of our story, as presented in the relevant rows of Table 6. 3. and bit him, 4. so that he died. 4 a. The serpent went on his way. Once proposition 3 is identified as belonging to the ("basic") category TRY, and proposition 4 as a ("basic") OUTCOME (by using the category

Stories

627

identifying procedures), the next parsing stage is the construction of a node to dominate both these nodes. Using N C I , the parser begins by determining that these two nodes are indeed causally related (the biting of the boy caused his death). It then determines that the O U T C O M E node (dominating proposition 4) is not a dead-end (since it encauses proposition 4 a). Given the above, it attaches the TRY node to a (projection category) OUTCOME'; that is, the node dominating Τ and Ο is O'. The above illustration reflects a major advantage of the proposed parser. In order to parse two given nodes X and Y, the user of this parser must hold in his STM buffer only the nodes in question and the following proposition. (The need to hold the following proposition is implied by the [subjcondition according to which Y is not a dead end, since in order to determine whether a given proposition is a dead end the parser must check whether or not it encauses the following proposition.) Thus, in parsing any two adjacent nodes, the STM only has to hold on to a (relatively) small amount of information. This feature of the parser meets a crucial requirement for any psychologically valid theory, that is, given that STM has a limited capacity, the parser requires the STM to hold only a limited number of information units at any given stage of the on-line processing of the unit. As was previously explained, this crucial requirement is met by both the Node construction procedures and the Category identification procedures. Node Construction Rule 2 If X encauses Y and Y is a Dead End, attach X and Y to (a projection category) X + 1. The second N C rule applies to the case where the two nodes to be parsed, X and Y, are causally related but Y is a dead end, that is, Y does not encause the following proposition. In this case the application of NC2 results in attaching Y to a projection category of X plus one bar (in the former case where Y was not a dead end, it was X that was attached to Y + 1). An example is the parsing of proposition 4 a. 4. 4 a. 5. 6.

so that he died. The serpent went on his way. The father, in revenge, got his axe,

628

Yeshayahu Sheti

At this point the parser has already identified the unit dominating propositions 1 - 4 as an OUTCOME' and 4 a as an OUTCOME (see the relevant row in Table 6). Moving towards the construction of the node dominating the above two nodes, proposition 4 a, which describes the serpent's leaving the place, it identifies 4 a as a dead end, since it does not encause proposition 5 (as explained previously). Given that proposition 4 a is a dead end, the parser applies NC2, which results in attaching the OUTCOME node dominating proposition 4 a to a higher level node — O" which has previously been constructed (the latter being a projection category of the preceding O' dominating 1 — 4), as its COMPLEMENT. As previously explained, it should be emphasized that during the node construction procedures, the parser is assumed to hold in STM the subtree (that is, the root node of that subtree) which has already been processed at the preceding stage of the tree construction. Consider, for example, the processing of propositions 3 and 4. The relevant row of Table 6 were previously presented in Table 7. 2. The serpent turned 3. and bit him, 4. so that he died. When the parser reaches these two propositions he is assumed to be holding in STM the (previously) inferred PROBLEM (which is: "The serpents seeks for revenge") as well as the previous proposition. He then identifies the category of proposition 3 on the basis of its causal relations with proposition 4. At this stage, proposition 2 has been released from the STM buffer and transferred to Long Term Memory, where it represents the node dominating whatever portion of the story that has been processed so far; note, however, that it is only after the "construction of the tree" stage that proposition 2 is finally released from STM, since only at this stage is it assigned on the basis of its causal relations with proposition 3 to the structural position of a SPECIFIER of the OUTCOME dominating this latter proposition. (Note that what is actually maintained in STM is not merely a proposition [in the above example, proposition 2] but rather the last node that has been processed up to the given stage). In sum, the construction of the tree always presupposes that the parser holds in STM the previously processed node (in addition to the proposition representing the current PROBLEM).

Stories

629

This sketchy discussion may give the reader an idea of the way stories are to be parsed within this framework. The proposed parser has been used to parse 10 short stories with reasonable results (for some details, see Shen 1986; 1989). Needless to say, however, that the above theory represents an initial and partial attempt to address the issue of parsing stories, rather than a complete theory.

References Carmeli, Sara 1983

The hierarchy of action and of thematic structure in the narrative text. [M. A. Thesis. Tel-Aviv Univ. Mimeo].

Jackendorff, Ray 1977 X-bar syntax. A study of phrase structure. Cambridge: MIT Press. Mandler, Jean — Nancy J. Johnson 1977 "Remembrance of things parsed: story structure and recall"; Cognitive Psychology 9: 1 1 1 - 1 5 1 . Rumelhart, David 1975 "Notes on a schema for stories", in: Bobrow Daniel — A. C. Collins (eds.) Representation and understanding: Studies in cognitive science. New York: Academic Press. 1977 "Understanding and summary of brief stories", in: David Laberge-Samuals Jay (eds). Basic processes in reading: perception and comprehension. Hillsdale, N.J.: Erlbaum. Schank, Roger 1975 "The structure of episodes in memory", in: Bobrow Daniel — Allan Collins (eds.). Representation and understanding: Studies in cognitive science. New York: Academic. Shen, Yeshayahu 1986 The structure of action in the short narrative text. [Unpublished P h . D . Dissertation, Tel Aviv Univ., Tel Aviv, (in Hebrew). 1988 "Schema theory and the processing of narrative texts. The X-Bar story grammar and the notion of discourse topic". Journal of Pragmatics Vol. 12, No. 5/6: 6 3 9 - 6 7 6 . [1989] [Reprinted in: A. Kasher (ed.), 199. Cognitive Aspects of Language Use. North Holland, 135-172.] 1989 in press. "The X-Bar grammar for stories: Story grammar revisited", to appear in Text, 9/4. (Forthcoming) "Centrality and causal relations in narrative texts", to appear in: Journal of Literary Semantics.

Convergent evidence for a set of coherence relations1 Kathleen Dahlgren

This paper develops a theory of discourse structure based upon the notion of coherence. We examine the question of the locus of coherence — in the text, in the world, or in the speakerhearer's cognitive model? We argue that coherence relations are elements of naive theories of causal and other structure of the actual world applied to the events and entities mentioned in a discourse, rather than belonging to separate linguistic or rhetorical knowledge. Coherence theory is placed within the theory of cognitive models. An allied question is that of the architecture of grammar. Is coherence part of formal linguistics, so that discourse structure is a type of grammar? We reject this hypothesis because we locate coherence inferencing in broader causal reasoning about the world. Coherence inferences are of the same type as those required for the interpretation of observed events as they unfold. Thus we propose an architecture in which syntax, formal semantics and world knowledge are separate modules which interact in parallel tp produce a model of discourse content. Felicitous discourse enables the construction of a cognitive model which coheres. Each event or state is connected by some coherence relation to some other event introduced into the discourse. In this paper a short finite set of coherence relations is proposed and justified on the basis of psychological and linguistic evidence and by philosophical argument. A computational discourse relation assignment algorithm is proposed which incorporates input from syntax, formal semantics and world knowledge in the form of naive semantic lexical representations. Its application is illustrated with an example. Computational and psychological approaches to discourse structure have included top-down approaches which try to predict in advance what the structure will be, as in story grammars and scripts (Schank and Abelson 1977). Despairing the brittleness of top down methods, others have attempted to discover the structure bottom up. The present work is part of the latter trend, and tries to move beyond the emphasis upon structural cues found in the work of Cohen (1984), Reichman (1985), Grosz and Sidner (1986). Our empirical studies show that cue phrases are very infrequent, and that in order to find the discourse structure, a deep

632

Kathleen

Dahlgren

analysis of content is required. We build upon earlier content-oriented research in discourse structure, such as Van Dijk and Kintsch (1983), Hobbs (1985), Mann and Thompson (1987) and Polanyi (1988). Our contribution is the integration of syntactic, formal semantic and naive semantic components into coherence theory.

Coherence The notion of coherence is based upon the intuition that the discourse in (1) seems to "hang together", while that in (2) does not. (1) (2)

John invested heavily. He profited handsomely. John invested heavily. He ate pizza.

The coherence of [1] is said to be explained by the reader's inference of a goal relationship between the investing and the profiting (Hirst 1981, Hobbs 1985, Mann and Thompson 1987, Van Dijk and Kintsch 1983). Notice that simple coreference is insufficient to explain text acceptability, because (2) has terms which are coreferential with terms in (1). Nor is plausibility sufficient, as Johnson-Laird (1983) suggests, for we can know that both statements in (2) are correct, and nevertheless find them unacceptable as a discourse. Beyond this intuitions there is little agreement as to definition, as shown by Noel 1989. We attempt to establish in this paper that a coherent discourse is one for which the hearer can build a cognitive representation such that the relations among individuals, events, states and other types in the representation correspond with her/his understanding (theory) of the way actual world individuals and events relate. 2 This is our fundamental point. It is extensible to account for hearer theories concerning mental states of interlocutors, coherent sets of propositions, logical arguments and so on. Although the representation may contain a variety of types of sensory images, in general, the hearer's "understanding" amounts to a naive theory (in the sense of Hayes (1985) (Graesser 1981, van Dijk and Kintsch 1983)) of the causal and other structure of objects and events (Brewer, 1977). For (1), the hearer brings the discourse into accord with his/her theory about investing. We say "theory" because very often people, and cultures, are quite mistaken in such causal inferences. Similarly, the world coheres under many conditions, but we humans may not understand. We may see some situation as chaotic, and only later discover its structure. Despite their incorrectness, people do use such structuring theories to manage the environment

Coherence relations

633

and to communicate via language. Because members of a subculture SHARE naive theories, the speaker can juxtapose just the two sentences in (1), and know that the hearer will guess that John's goal had been profit. Recent cognitive psychological research in reading has established the psychological reality of coherence inference. First, the intuition that coherent texts are more acceptable is supported in several studies (Kieras 1978, Pohl 1982). Graesser and Clark (1985) show that subjects reliably draw large numbers of inferences about texts when questioned after reading. Especially frequent are cause and goal inferences. These inferences connect the events and individuals in the story to each other. Thirdly, those events which are causally or superordinately related to others in a text are more quickly interpreted, better recalled, and better recognized than unrelated events (Townsend et al 1987, Graesser et al 1980, Black and Bower 1980, Black and Bern 1981, Trabasso and Sperry 1985, Foss and Bower 1986, Nicholas and Trabasso 1980). Texts which are organized for ready identification of event hierarchy (topicality) are more quickly read and better understood [50], This and other work (Tanenhaus et al 1987, van Dijk 1987), suggest that discourse interpretation involves the construction of an event model of the emerging situation. This is based both upon world knowledge and details of the discourse structure (Townsend et al 1987). The model involves causally related events, a hierarchy of events and the goals of a protagonist or main actors. Apparently, more of inherences are drawn while reading earlier portions of the text, and later text is interpreted in terms of the constructed context (Millis et al 1990). What is missing from these studies is a theory of where coherence lies in cognitive architecture, what the stock of coherence relations is, and what coheres, pieces of text, semantic interpretations, events in a cognitive model, or events in the world. Our view of coherence is that speakers in a given genre make a discourse (and thereby their reporting of events) understandable by choosing to report events in a certain sequence. This choice in a well-structured discourse makes it maximally possible for the hearer to build a cognitive picture of these events which coheres. It will cohere to the extent that he/she can bring it into accord with her/his theories about the way the world works. The relationships in the naive theory of the world are causal, intentional, comparative, part-whole, etc. Thus in our view, coherence operates over events in a cognitive model, and the relationships are explained by naive theories of causal structures in the world. Coherence is a gradient phenomenon in that the better-structured the discourse, the more readily and reliably a hearer will make coherence inferences.

634

Kathleen

Dahlgren

Similarly, the more knowledgeable and tuned in the hearer, the more accurately he/she will recover the speaker's intended cognitive representation.

Studies of coherence in commentary text This work is part of an ongoing project (Dahlgren 1988a, Dahlgren, McDowell and Stabler 1989), for computational text understanding and precise text selection. Our initial target was Wall Steet Journal (WSJ) commentaries. "Commentary" is a genre in which a single event (real or hypothetical) is reported, and the remainder of the discourse consists of 1) the importance of that event for the reader from the author's point of view, 2) comments from experts and interested parties, and 3) background events. We have carried out three studies of coherence in an expanding corpus of WSJ commentary texts. Study 1 (Dahlgren 1988a) of 8,000 words in six articles sought to examine what information is used in coherence relation assignment, and to determine whether syntactic markers and cue phrases were sufficient information for coherence relation assignment. The coherence relation literature was reviewed (Cohen 1984, Fox 1984, Hirst 1981, Hobbs 1985, Mann and Thompson 1987), and coherence relations (defined below and displayed in Table 1) were identified. A coherence relation was assigned to each clause in the corpus by two judges. The syntactic and semantic properties of each clause were encoded. These properties included clause type, voice, mood, presence of negation, agentiveness of subject, type of subject and object, and aspectual class of the verb. The correlations between a clause's coherence relation and its syntactic/semantic properties form the basis of the algorithm described below. We found that the information used in coherence relation assignment includes: 1) syntax, 2) cue phrases; 3) lexical items; 4) tense; 5) aspect; and 6) world knowledge. The order in this list corresponds to informationbearingness, in that the few syntactic indicators of coherence, where present, were decisive, and the same held for cue phrases. However, the most frequent indicators of coherence are temporal order, aspect and world knowledge. World knowledge involves inference from the directly stated content of the discourse to the typical coherence structure of events in the naive theory of events English speakers agree upon. In other words, not just any events can be connected causally or in any other way in discourse, but only those which English speakers believe belong in struc-

Coherence relations

635

tures of events or event chains. The same type of inferencing is used to connect other abstract types such as states. Table J. Coherence Relations and Cognitive Capacities Coherence Relation

Cognitive Capacity

Background Goal Cause Enablement Constituency Contrast Elaboration Evaluation

Perception of figure/ground and salience Naive Theories of Intentionality Naive Theories of Causation Naive Theories of Causation Recognition of Part/Whole Recognition of Similarity Perception of spatio-temporal contiguity Preference (goodness ratings)

Study 2, of the same 8,000 words and 8,000 additional words of text, examined discourse segmentation to determine which factors influenced it. For each new S, the possibility of a new sister- or sub-segment arises. We found that change of coherence relation was the most reliable indicator of new segment and change of subject next most reliable. Other factors were paragraph indentation, length of segment with the same coherence relation to some other, cue phrases, and event anaphors. Significantly, there was a segmenting cue phrase such as, "turning to ..." in only 16 per cent of the cases of a new sister segment. Clearly any theory which includes only direct cues will miss most of the structure. Substantive coherence relations, if they can be extracted, are powerful indicators of structure. Study 3 (Lord and Dahlgren, manuscript) of anaphor resolution in a corpus of commentary genre text drawn from the Wall Street Journal is a further source of evidence for the psychological reality of discourse coherence. We found that pronominal anaphors cannot find their antecedents across segment boundaries. Segment boundaries are most often signalled by change in coherence relation and change in sentence subject, not by cue phrases. Therefore constraints on anaphor resolution provide evidence that coherence relations must be assigned during discourse interpretations. The phenomenon is intuitively illustrated by the example in (3). The cue phrase "turning to" forces a change of segment. It is impossible to assign Jones as antecedent of Smith across that segment boundary.

636

(3)

Kathleen

Dahlgren

Jones invested heavily in the stock market, and profited handsomely. But he got greedy. He played fast and rules with stock broking rules. Now he is under indictment. Turning to other insider trading cases, Smith is now serving a term in Federal prison. He had invested based upon insider tips.

If the last S in (3) is replaced by "He had invested more than Smith", where "he" is coerced to corefer with Jones, the text seems incoherent. The same constraint holds with an implied segment boundary, as when the last two sentences in (3) are replaced with (4). (4)

Similarly, Smith is now serving a term in Federal prison. He had invested based upon insider tips.

Why assign coherence? Some have argued that although coherence is psychologically real, it cannot be in a text understanding system (Grosz and Sidner 1986). They consider parsing and formal semantic translation challenging enough tasks at the present stage of linguistic theory. Coherence is far more challenging, because it relies upon world knowledge. They see world knowledge as impossible to encode. We disagree because without coherence, segmentation of text cannot be computed, and without segmentation anaphors cannot be resolved. That text understanding requires segmentation has been established in several major studies (Grosz and Sidner 1986, Reichman 1985). In order to segment coherence must be computed, because these segments (or focus spaces) can only be located by noticing change in coherence relation, as demonstrated in our Study 2. Second, without coherence and thus without segments, a text understanding system cannot resolve anaphors, and therefore cannot get even the truth conditional semantic translation (Cohen 1984). Third, because of point two, a system without coherence answers some questions about what a text has said incorrectly (Lockman and Klappholz 1980, Hoppen and Thompson 1980). A computational system with coherence can answer many more queries accurately. Given a text such as (1), a system with coherence can answer as in (5) — (7). (5) (6) (7)

Why did John make a profit? — Because he invested. Why did John invest? — In order to make a profit. What did John invest in? — Typically, stocks or bonds.

Coherence relations

637

Fourth, with coherence, anaphora resolution can be significantly constrained. In general, antecedents of personal pronouns are found in the same segment or the global topic (see Study 3 and below). Only certain types of anaphors can have antecedents outside the segment (event anaphors, heavy definite NP's, those with antecedents in the topic event or topic event participant). Fifth, more intelligent text understanding is made possible by coherence inferences. Text is telescopic, and the reader fills in the gaps. A computational system which models such inferences will be able to reflect much more accurately the human understanding of text. Sixth, when considering whether one text is relevant to another, the more naive inferencing computed, the more accurate the relevance reasoning will be. The alternative to coherence is brute force, which leads to a combinatorial explosion, and still indeterminate results, or cues which are insufficiently information bearing. Cue phrases are present at only 16 per cent of the segment boundaries. Fortunately, coherence isn't as costly as it seems in a system with commonsense knowledge. This knowledge is needed anyway for syntactic and lexical disambiguation, and we have shown that only a shallow layer of it is sufficient to account for over 97 per cent of coherence inferences.

Problems with coherence theory One of the reasons for the emphasis on overt elements such as syntax and cue phrases, is that unsolved problems have plagued the coherence approach and made it a dubious notion. The unresolved issues in coherence theory have been: 1) What is a coherent discourse? We try to rectify this by giving a philosophically based definition. 2) Which entities cohere: Sentences, clauses, propositions, the content of clauses, the real events denoted? 3 3) How many coherence relations are there and which are they? 4) Informality. Much of the study of coherence has been descriptive and has not attempted to provide a direct link between coherence theory and formal semantic theories. (Hobbs 1985, Mann and Thompson 1987) Recent developments in formal semantics provide a framework and ongoing research devoted to giving truth conditions for entire discourses as well as single sentences (Kamp 1981, Asher 1987). It is now possible to integrate coherence structure with formal semantic representation (especially of temporal order and aspectual class which our Study 2 found are particularly significant indicators of coherence relations). Further, we can integrate an account of discourse segmentation based upon coherence

638

Kathleen

Dahlgren

can be integrated with semantic representation. 5) No-one has proposed an algorithm for extracting coherence relations. This is because to do so, by all accounts, requires world knowledge, and it has heretofore seemed impossible to encode world knowledge in a non-ad hoc, transportable and practicable day. Our theory of lexical representation, Naive Semantics (NS), offers a solution (Dahlgren 1988a). In the next sections we examine psycholinguistic, morphological and philosophical bases of coherence. In what follows we address the psychological, morphological and philosophical bases of the coherence relations proposed in Table 1.

Psychological bases of coherence Each of the relations in Table 1 has been substantiated in psycholinguistic research as being constructed during text interpretation. The relation BACKGROUND affects anaphor resolution. People prefer to resolve anaphors to the protagonist of a story, and when the protagonist is in a clearly backgrounded clause, they shift to resolving the anaphor to a non-protagonist. Clause order doesn't affect this. In contrast, places are always resolved according to recency (Morrow 1985). The relation CAUSE is amply supported by studies of causal connectivity in discourse events. (Black and Bower 1980, Black and Bern 1981, Trabasso and Sperry 1985, Graesser and Clark 1985a, Millis et al 1990, Nicholas and Trabasso 1980). That the goals of main characters are inferred is shown in many studies. (Morrow 1985, Seifert et al 1985 Black and Bower 1980) The interpretation of discourse events as constituents of each other is shown by Black and Bower's (1980) event hierarchy, as well as studies of van Dijk and Kintsch (1983). CONTRAST in story events is part of many analysis of text, including van Dijk and Kintsch. Sequences of events (ELABORATION) are justified by the work of Graesser and Clark (1985), where subjects reliably agree upon the answers to "What happens next?" questions. The importance of the temporal connectivity of events is also indicated by Black and Bower (1980) and Trabasso and Sperry (1985). EVALUATION has'nt been studied, yet without it, some sentences cannot be related to the rest of the discourse. To justify the exclusion of relations from the set, we consider those proposed by Mann and Thompson (1987) which we rejected in part. Mann & Thompson propose a larger set of coherence relations. Each of the relations proposed in their work subsumed under one of the relations in Table 1 in combination with formal semantic representations. As we

Coherence relations

639

showed above, in a system where verbs are classified as goal and nongoal, there is no need to distinguish volitional from non-volitional CAUSE. Their distinction between CAUSE and RESULT can be indicated by the order of arguments in the coherence predicate: cause (el, e2) = result (e2, el). Mann and Thompson's MOTIVATION, EVIDENCE, JUSTIFY and CONCESSION all fall under a theory of speech acts. Their RESTATEMENT and SUMMARY are indicated by our CONSTITUENCY relation. Their ANTITHESIS is our CONTRAST. Each case of their S O L U T I O N H O O D can be reanalyzed as CAUSE and negation.

Morphological bases of coherence As for the morphological basis of coherence, inspection of Cohen's (1984) survey of English connectives shows that our set of coherence relations takes into account all of the connectives except those involving textual structure ("turning to", "and now", ...), and those indicating stages of arguments ("that implies, ipso facto"), which belong in the theory of speech acts. Table 2 illustrates the connectives expressing each of the coherence relations. The fact that they find expression in morphological forms substantiates their psychological reality. Table 2. Connectives expressing Coherence Relations Coherence Relation

Connective

Background Goal Cause Constituency Contrast Elaboration Evaluation

"when, while" "in order to, so that" "because, as a result, consequently" "in summary, for example, first, second" "similarly, likewise, in contrast" "then next, that is to say" "evidently, that means", modality, negation

Philosophical bases of coherence Over and above these empirical bases of coherence, we need to locate coherence in epistemology and psychology, so that the terms of the theory have the potential of being explained in the end by psychological, soci-

640

Kathleen

Dahlgren

ological and other human processes. Furthermore, without a firm philosophical basis, we have no theory at all, but rather, a set of researcher intuitions. A coherence relation such as CAUSE has been located in several underlying theories: presentation style (Hobbs), pragmatics (Mann and Thompson, 1987), and causal inferencing (van Dijk and Kintsch, 1983). The first question is, what coheres? Does discourse itself cohere as Grosz and Sidner (1986) claim, or the discourse situation, as in Mann and Thompson (1987), or a mental model of the content of the discourse as in Johnson-Laird (1983)? We will argue that what coheres is a mental model of the situation a discourse describes, and that the way the mental model coheres is explained by naive theories of causal and other structure in the actual world (for many of the coherence relations). CAUSE: What is the coherence relation CAUSE? Formally, we identify it as a predication cause (el, e2) (stated or implied in a text), which represents a speaker/hearer theory about the causal connections between events. So it is a predication over events. But what is an event (denotationally)? One standard view, with many problems, is that an event is a spatially and temporally located occurrence in the real world. A more sophisticated view of events is as concrete particulars individuated by their causes and consequences (Davidson 1987). However, our concern is cognitive representations of events and their structure in discourse. We must explain human interpretation of the actual world (naive metaphysics), not the world itself. Even the direct observation of some real event involves observer interpretation, minimally, of the idea that each frame he sees is part of the same real event and not a series of different events. The observer view that a certain thing is happening, e.g., that car is falling off that cliff, is a theory — often a conscious verbal theory. "Oh, that car is falling off of that cliff." When the observer thinks, "Oh, an accident", we have even more theorizing and interpretation as actor goals (or lack of them) are inferred. So the cognitive representation of events is epistemological, heuristic, and retractable. Now consider observer interpretation of events. When a person has seen a car fall over a cliff, he/she may interpret it as a whole with two causally related subparts, as in (8). (8)

(el) The car came to the edge. (e2) The car fell over the edge.

spl. The car came too close to the edge, a critical point was reached in which the car was no longer balanced on the edge, and it fell over. How shall we analyze observer construction of causal inferences? In the cognitive interpretation of events (and of texts reporting events),

Coherence relations

641

people actively theorize and hypothesize about the causal structure of events as they unfold. (Graesser 1981) People are tentative about such inferences, but in order to function, they must guess. Such guesses are naive theories about the causal structure of events. Furthermore, the inferred causal structure is an important correlate of memory for the events. (Graesser 1981) In example (8), such a naive hypothesis would be formed when the observer uses naive physics to infer that (el) and (el) are causally related. In understanding, interpreting and labelling observations of real world events like (8), members of a Western culture (and probably of any culture) infer that the critical point was reached, and that the coming close to the edge (in the end) caused the falling over the edge. The same holds for discourse and text understanding. As they read, people infer causal structure.

Abstract types Events are harder to individuate than are individual objects such as chairs or people. Imagine a situation in which a boy is hitting a girl. Someone points to the boy and says (9). Someone points to the pair and says (10). Finding the antecedent of "he" for the hearer of (9) is less difficult than finding the antecedent of "that" is for the hearer of (10). (9) (10)

He is a bully. That isn't nice.

In building a theory of discourse coherence we must take this difference into account. Discourse coherence explains the coherence of events (and other abstract types) introduced into a discourse. While such types are inherently more difficult to individuate than are concrete physical objects, they are not totally observer or speaker relative either. Such an extreme position would imply that there is no basis for believing that any of us is referring to the same events ever. So we assume that there IS a significant level of agreement concerning the individuation of events, and it is at this level that coherence theory explains plausible coherence among events. Though the individuation of events is difficult, people do agree upon the salience of relations in a given situation, and in interpreting discourse about it. Event predication itself (e. g., "John is hitting Mary"), involves naive theorizing about the structure of events and how that structure relates to what the speaker is seeing. We assume that this layer of reasoning is on relatively solid ground, and that people in a given

642

Kathleen

Dahlgren

culture agree significantly upon what they are seeing, for at least a large enough number of situations to make human conversation and interaction possible (Bürge 1979). Returning to our original point, then, what is the coherence relation CAUSE? It is an element of naive metaphysics, of naive theories about the relations between events. Next, in terms of semantic theory, how shall we handle the logical form of a coherence relation such as CAUSE. Davidson [21] argues against treating it as a logical connective like the material condition, and in favor of treating it as a predicate. He considers sentences such as (11) and argues for a logical form like (12) (11) (12)

Jack fell down and broke his crown. Ε el Ee2 (el = fall_down (ja) & e2 = b r e a k h i s c r o w n (j) & cause (el, e2)

In Davidson's analysis, a cause statement introduces a causal predication over events. He points out that the truth conditions of causal statements require that very many of the conditions present in a causal situation are part of the evaluation of truth, not just the descriptive predications for the events. We must distinguish firmly between causes and the features we hit on for describing them, and hence between the question whether a statement says truly that one event caused another and the further question whether the events are characterized in such a way that we can deduce, or otherwise infer, from laws or other causal lore, that the relation was causal. "The cause of this match's lighting is that it was struck. — Yes, but that was only 'part' of the cause; it had to be a dry match, there had to be adequate oxygen, and the striking was hard enough, etc." We accept Davidson's analysis, and interpret CAUSE relations as predications of events added to the semantic representation for a discourse, either because of direct surface forms like "cause" or "because", or by hearer inference. Turning to other relations, GOAL is explained by naive theories of intentionality. Basic to human thinking is that a distinction between persons and all other objects (Strawson 1953, Gelman and Spelke 1981). This distinction arises early (Carey 1985). Persons are conceived as having goals and as agents of actions. The agentive distinction (conscious, deliberate action as opposed to actions which just happens) is gramma-

Coherence

relations

643

ticized in many languages (Fillmore 1968). In reasoning about situations, one of the most important psychological distinctions is goal vs. non-goal (Trabasso and Sperry 1985, Graesser and Clark 1985a). So the coherence relation GOAL in discourse understanding is explained by naive theories of intentionality. CONTRAST. Just as people look for causes and goals to make sense of situations, they also look for similarity of difference in things and events. That is what literary analogy and metaphor play upon. Event predication itself is a form of similarity inference (Partee 1984). Furthermore, similarity and difference can be evaluated for truth. "These two chairs are similar" is a testable, though gradient, assertion. Syntactic and other devices used to highlight coherence relations are more or less isomorphic to relations in the world (e.g., nouns for things, verbs for actions). The coherence relation CONTRAST is derivative from real similarity and real difference. To the extent that these may be abstract, they be more theoretical, less reliable across people, and harder to interpret when reading a text. CONSTITUENCY. Generalization is reflected in the binary coherence relation CONSTITUENCY. Like CAUSE and CONTRAST, generalization is explained in terms of naive theories of events. In particular, generalization is just naive event summation (Asher 1989). All but instantaneous events have the potential of being broken into parts and stages. Event perception involves placing a spatial and temporal bound in a situation, and of selecting salient actors, objects and relations. This has led some to claim all event perception is subjective in the sense that it is observer-relative. Others uphold the mind-independent reality of events. Nevertheless, all agree that events (whether only in our heads or not) are inherently divisible. This means that in naive metaphysics events are divisible, and on the other hand, can be summed. However, not any set of events can be summed, only those which accord with some plausible event chain (Dahlgren 1988a). Further, other types can be summed. Not only can we speak of World War II (a protracted series of events), we can speak of a sum of states, as in (14), where the last state "happy" states, sums the previous two. (14)

a. John felt proud of his Β average. b. He loved his new car. c. For once he was happy.

Likewise, we can sum propositions as in Anna Karenina tells more about Levin than Anna, where the entire novel is viewed as one summed

644

Kathleen

Dahlgren

proposition. Thus generalizations and elaborations are just processes of event summation. Event summation is just as truth conditional as CAUSE. In other words, (15) is true in the worlds in which (14 a and b) are but (16) is not. (15) (16)

In sum, he was happy. In sum, he was unhappy (tired, ...)

All of these types of summing can be called "abstract type summation". Thus each of the coherence relations listed in Table 1 derives from a linguistically-independent cognitive strategy employed in the interpretation of events and situation. The strategies are listed next to the relations in the table.

Naive semantics Turning to the third problem with coherence theory, we briefly describe our solution, a way of representing commonsense knowledge. Naive Semantics (NS) (Dahlgren 1988) is a theory of word sense meaning representation in which associated with each sense of each content word (noun, verb, adjective) is a naive theory of the sort of object, action or property named by the word. NS rejects a theory in which word meanings are broken up into atomic primitives which directly play a truth conditional role in sentence and discourse interpretation. Instead, word meanings these are supposed to contribute non-monotonically to the formal representation, and further, they involve many unfired inferences. These rich naive theories are generalizations. Though not "true", they are and must be close enough to true, enough of the time, for people to refer correctly to real objects and events, and to communicate using language. NS representations are rich and open-ended. Anything at all can be there. The number of feature values could be as large as the number of words in English (Schubert et al 1979). Word meanings are seen as the names of concepts, and concepts are mental representations which may take many forms — visual, motor, tactile or verbal. Obviously, present computational systems are limited to the verbal aspects. NS representations of noun concepts come from the results of psycholinguistic studies of object concepts in the prototype theory (Rosch et al 1976, Dahlgren 1985). In principle, such representations could be quite extended. In practice, we use the first 1.5 minutes of subjects' freelisting of properties.

Coherence relations

645

These are typed (e. g., color(red)). An example is banker, shown below in English translation. Typically, a banker is a well-dressed educated male who works in an office in a bank. He is trained in mathematics. He is dedicated, civic-minded, and has high status. He functions as a financier who lends money. Inherently a banker is a person in authority whose function is to engage in business and handle money. For verbs, we use the approach of Graesser and Clark (1985 b), who show that people conceive of actions in terms of their implications, such as cause, goal, result, location, manner and so on. We also classify verbs aspectually after Vendler (1967). The verb entry is based upon typical and inherent implications. An example is invest: Typically, investing is done with capital in the form of money or other asset. A sentient invests in stock, commodities, or real estate. Later, one may sell it or use it as collateral. Inherently, a sentient invests with the goal of making a profit. NS provides a means of representing the world knowledge attached to English words in a general, non-ad hoc way, resulting in transportable representations. These representations are in fact feasible, and they go a long way toward providing the information necessary for syntactic disambiguation (Dahlgren and McDiwell 1986b), word sense disambiguation (Dahlgren, 1988b), relevance reasoning and coherence. Fortunately, it is not necessary to encode all of commonsense knowledge in order to achieve significant and useful results in text understanding. Without independently derived representations, we were able to make all of the assignments in Study 1 of coherence. Thus only a very SHALLOW level of naive semantics is required. This fact answers Fodor's claim that semantics is impossible (1987).

Cognitive semantic architecture We claim that coherence belongs in a cognitive inference module, not in syntax or semantics. We assume a modular linguistic model with separate modules for phonology, syntax, formal semantics (which represents the contribution of the syntactic elements to the conditions of the discourse), and naive semantics (which contains lexical representations of naive

646

Kathleen Dahlgren

theories about objects and events). In addition, departing from others such as Fodor, we assume that each component communicates with all of the others whenever necessary during processing. Our assumptions are consistent with the psycholinguistic model suggested by Tanenhaus, Dell and Carlson (1987), and that of Marslen-Wilson and Tyler (1987). The view holds not only for processing, but for the formal models of syntax, semantics, discourse and naive semantics. Each of these components requires information from each of the others in order to generate structure, not just from the level "below" it. Recent developments in formal semantic theory make possible the integration of formal and naive semantic theory. Formerly, semantic theory was concerned with truth conditions for individual sentences. Recently semantic theory has broadened. Discourse Representation Theory aims to account for the compositional semantic interpretation of entire discourses and to unify its account of truth conditions for discourses with an account of the mental representation of the discourse built by the speaker/hearer (Kamp, 1981). The cognitive architecture suggested by D R T is schematized in Figure 1. Suppose that the real past world consisted of only four people, (John, Mary, Bill and Martha), and the relations that John loved Mary and kissed her, and that Bill loved Martha and kissed her (as described in the right-most column of Figure 1). Suppose that the John-Mary situation is reported in the discourse: John loved Mary. He kissed her. The discourse representation structure (DRS) of that discourse is shown in the second column of the Figure 1. The DRS is a partial model of the world, partial because it mentions only John and Mary and the relations between them, not Bill and Martha. The insight is that the model of a discourse contains only those individuals and events mentioned in the discourse, rather than all of the individuals in the world model. The truth conditions of the discourse are defined in terms of a standard intensional model (column 3 of the schematic). If the individuals with names John and Mary were in the set of sets related by "love" at time t, then "John loved Mary" is true. In the model Μ the real John and Mary are denoted by model elements al and a2. An embedding function is defined to relate the DRS to the world model M.

Coherence relations Text

John loved Mary. He kissed her.

Representation

Full Model -

M =

ul,u2,si,now, rl,u3,u4,el

U = (al,a2,a3,a4)

john(ul) mary(u2) si love(ul,u2) rl before now si incl rl he(u3) her(u4) u3 = ul u4 = u2 el incl rl

F(al) = F(a2) = F(a3) = F(a4) =

John Mary Bill Martha

F(man) = (al,a3) F( woman) = (a2,14) F(love) = ((al,a2),(a3,a4)) F(kiss) =

647

World

al denotes john, ... john loved and kissed mary, bill loved and kissed martha at some time before now

((al,a2),(a3,a4)) Figure 1

To summarize, in D R T meaning representations are partial models defined in a language L with individuals, relations between them, properties, propositions, and mental states. The partial models (DRS's) are defined as m = , which consist of a universe of (U) of entities, and a set of conditions which are said to hold of those entities. A DRS is intended to be a representation of the meaning of a discourse which is both truth conditional and reflects the mental representation of the discourse built by the language-user. The partial model is embedded in a full model Μ = which consists of a set U of (non-linguistic) entities and an interpretation function F which assigns an element of U to each name in the language, a subset of U to each common noun and intransitive verb phrase, and a set of pairs of elements of U to each transitive verb. Truth in D R T is not defined relative to the discourse model m directly, but relative to M, the "big" model of all objects, and relations among them at a time in a world. The discourse is true if there is a proper embedding of m into M. Thus D R T links up a mental model of a discourse with a world model and truth conditions. This is an important advantage where computational theories are concerned. Alternative approaches which pay little attention to logic and truth are inadequate for computation. Further, partial models eliminate some of the spurious ambiguity implicit in earlier intensional theories. A discourse is not evaluated relative to all objects and relations, but only relative to those objects and relations which are contextually salient.

648

Kathleen

Dahlgren

D R T provides the truth conditional semantics of a discourse, reflecting only the objects and events explicitly mentioned in the discourse, and the inferences required to connect up the sentences in the discourse, such as that "he" in the second S of the discourse has the same referent as "John", and "her" Mary, and that the kissing probably overlapped the loving in time. The temporal equations in the DRS in Figure 1 show that the loving and kissing are both included in a time interval rl before now, after Partee (1984). Such semantic inferences are tied to morphological forms such as pronouns and tense markers. We claim that there is another, separate component in the cognitive architecture, the naive semantic component, in which the hearer draws inferences for the interpretation of the discourse based upon what the hearer knows about the world. Such inferences correspond to the speaker's mental picture of the situation when the speaker is trying of the world knowledge in the naive semantic component is represented as naive theories about objects and events connected to (labelled by) lexical items (Dahlgren 1988a). We must place coherence in a cognitive inference module, not in syntax or semantics, because coherence relation assignment requires cognitive reasoning which goes beyond what the text says directly. Furthermore, the forms of representation and the mechanisms of reasoning may be different. Formal semantics comes straight from the syntax, while naive inference is generalized reasoning, which may not even be verbal. A person sees a bee move from flower to flower, and infers that it is collecting nectar. Is that inference verbally represented? Couldn't it be visually represented, as an extrapolation from experience, or some picture in a science class showing a magnified bee collecting nectar? And even if it IS verbally represented, IS the inference is probably not based upon linguistic rules. Suppose a text says "The bee is moving from flower to flower." The reader infers he is collecting nectar. Is that inference based directly upon linguistic expressions in the text, or does the inference employ world knowledge and general cognitive strategies. Returning to the car falling off the cliff, suppose a person sees the car coming very close to the edge. Isn't it possible for the person to experience a fight/ flight response based upon an inference that the car is going to fall over the cliff, without ever verbally representing that inference? Don't other animals draw such inferences all of the time? However, for our purposes it is not necessary to speculate upon the verbal/non-verbal question in reasoning. An assumption that all causal reasoning is verbally represented does not alter the fact that we need a separate cognitive component for coherence reasoning. The difference

Coherence relations

649

between formal semantics and coherence is that with the former, there is a formulaic mapping f r o m structure to the logical form of interpretation. Given the rules and the discourse, a set of logical forms is generated. In DRT, the construction algorithm maps f r o m syntax to compositional semantics via relatively straightforward rules off the syntax. (This is true in other semantic theories as well). The indicators of the semantics are directly expressed in the sentence. For example, a tense inflection on a telic verb requires that an event reference marker be introduced into the DRS. A definite article requires one set of rules of quantification, the indefinite another set. A p r o n o u n or definite N P invokes a n a p h o r a resolution rules. In contrast, the indicators of the coherence inferences intended by the writer are not always (or even usually) in the text as content or morphology. In (1), the hearer infers goal (el,e2) from his knowledge of investing named with his word "invest")· Furthermore, discourse segmentation is inferred using knowledge of the genre and knowledge of the world. In most cases, discourse segmentation is not directly signalled in the discourse as revealed in our Study 2. That formal semantics is formulaic is a claim which holds even though these rules sometimes tap world knowledge, as with the assignment of temporal order, clause aspect, and a n a p h o r a resolution. A sequence of two simple past tense telics strongly suggest temporal order of the two discourse events, not overlap. The rules produce this preferred inference, which has to be checked against world knowledge. Similarly, pronouns demand a search for antecedents. The salience of entities in the stack of discourse entities in the focus space (the current segment) and the discourse structure (the tree built up so far), is partially determined by morphology, discourse properties such as proximity, and world knowledge. The fact that world knowledge is an input to a n a p h o r a resolution doesn't change the form of the semantic rules. There is a rule-governed mapping from the form of words to the preferred inference. If coherence belongs in the cognitive inference engine, when and how is gathered the information contributed by syntactic coherence cues in the syntax, such as participial and causal connectors, and lexical cues such as "mean, differ", and coherence cues in the semantics such as temporal order, conditionals, aspect and so on? Syntax, formal semantic and naive semantic components provide input to the cognitive component for coherence relation assignment, event summation and discourse segmentation. In a processing model, this would imply that as the sentence is parsed, coherence and segmentation hypotheses are generated. Syntax contributes some coherence information, semantics some, but coherence

650

Kathleen

Dahlgren

relations are assigned in a separate component which incorporates world knowledge, including information which is not in the discourse, to construct an interpretation of the causal structures of the events. Sometimes there is a causal connector ("because"); other times reasoning is applied to infer that the events are causally connected. This theory assumes that whether coherence relations are directly signalled or not, speakers usually intend them in many genres, and they are a necessary part of the constructed mental representation of the discourse. These relations are assigned in a different module than the formal semantics. On the other hand, although coherence belongs in generalized cognitive inferencing rather than formal semantics, coherence relations, once assigned, are truth conditional, in the sense that what is inferred (or asserted, when there are explicit connectors like "because" or "in order to"), is that the situation described hung together in that way. Consider the text (13). (13)

John pushed Mary. She fell down.

Hearers infer that the pushing caused the falling. If we add this inference to the semantic interpretation the Davidsonian manner, we could say that the discourse is represented as: Ε el Ε e2 (el =push(j,m) & e2 = fall_down(m) & cause(el,e2)) So, once the coherence relation is established, it affects the truth conditions of the discourse, even if retractably. If later fact-finding showed that the falling hadn't been caused by the pushing, but instead by a big truck which hit Mary, the hearer would think it infelicitous for the speaker to have said (11). He would think, he should have been told that she fell down because a big truck hit her.

Other approaches to coherence In comparison with other work, ours is similar to van Dijk and Kintsch (1982) in that they define coherence by whether sentences in a discourse describe related facts in some possible world, and they assume that large amounts of world knowledge are employed in building a cognitive model of a discourse. We differ in defining coherence as relating discourse events, rather than as relating propositions. Furthermore, we clarify the question of truth conditions as opposed to naive (or heuristic) inference regarding discourse interpretation. And we provide an algorithm. We draw upon

Coherence relations

651

Hobbs (1985) and Mann and Thompson (1987) for coherence relations. However, we define them as relating discourse events (and other abstract types) and structures of these in a cognitive event model, rather than as relating utterances, clauses, or spans of discourse, as they do. For them, coherence is essentially a property of presentation style, of the speaker's intended effect on the hearer. In contrast, for us, coherence is essentially a property of mental models which finds its origin in beliefs about relationships among real events Johnson-Laird. Our approach appeals to cognitive strategies and beliefs people use all of the time, whether thinking verbally or not. An approach which is very similar to ours is that of Trabasso and Sperry (1985) and Trabasso and van den Broek (1983). Their theory of causal chains as accounting for discourse coherence is essentially the same as ours. The differences are that they do not integrate their theory with a formal semantic theory, and do not place coherence in a cognitive architecture relative either to semantics or syntax. At some points in their analyses the coherence relations they propose belong in the theory of intrasentential thematic roles. On the other end of the spectrum, Grosz and Sidner (1986) deny that coherence is a useful notion, but in actuality they use as coherence relations "dominance" and "precedence". Clearly a far larger number of substantive coherence relations than that is needed in order to capture the facts of discourse structure. Grosz and Sidner (1986) define discourse structure in terms of the speaker's goals. We agree with Polanyi (1988) that this aspect of coherence belongs in a level of theory above that of cognitive models of interpretation of discourse.

Local vs global coherence Many genres require global topics, including the commentary genre, where a single event or state is reported, and the remainder of the article consists of commentary by the author, interested parties and other observers. Most often these viewpoints concern the importance of the reported event for the readers. Each of the articles in our WSJ corpus has an easily identifiable global topic. In the commentary genre, the journalist must directly express the topic, is taught to do so in journalism school, and is required to do so by editorial policy. Thus the global topic is more readily identifiable in the commentary than in dialogue. In general, it seems clear that the directness of expression of global topic will vary from genre to genre. We should expect written text to directly

652

Kathleen

Dahlgren

express global topic more often than oral. Thus the topic of a dialogue in general should be more difficult to extract and may often have to be inferred. However, in certain oral genres, such as task-oriented dialogue (as in the Grosz p u m p case) we also expect directly expressed topics. In that example of task-oriented dialogue, the parts of the object to be constructed were topics of segments. A m o n g written genres such as commentary, research reports in medicine and psychology, and so on, topics are more often directly expressed than in narratives. Some narratives and dialogues may have not global topic at all. The main point here is that we have found that in computational text understanding for the commentary genre (and very probably extensible to all types of technical writing including law and medicine), there is a global topic, and once extracted, it can play a very important role in simplifying the text understanding task. People use certain predictable strategies for making what they write comprehensible, and we must take advantage of that in building computational algorithms. In van Dijk and Kintsch (1983), a distinction is drawn between global coherence (the overall structure of a text), and local coherence (the relationship between the content of individual clauses in a text), where the relationship is the assignment of some label such as C A U S E or G O A L to the two clauses. We tried to simplify our theory by eliminating that distinction. We found in Studies 1 — 3 that the content of local and global coherence predicates was the same. This led us to attempt to build a theory of discourse structure in which local and global coherence relations are the same set of relations, and label recursively generated segments of text up to the top (or global) segment. Coherence hypotheses concerning coherence between individual clauses and coherence between chunks are generated by readers. These hypotheses are essentially the same, and take the form, "This individual (clause, chunk of clauses) is related to this other in one of a number of ways which is ordinarily found in this genre of discourse." For example: This event is a C A U S E of that event. This structure of events is a C A U S E of that event, or This structure of events is a C A U S E of that structure of events

Local coherence algorithm In the context of a computational text understanding system, it was necessary to formulate explicit rules of C A R . O u r local C R A algorithm assigns a coherence relation to the content of two clauses, where clause

Coherence relations

653

means any underlying S node, including derived nominals. We ignored a number of complex embedding problems for two reasons. First, our corpus did not contain such structures, and second, we believe, with K a m p (1981) that the contributions of such structures should be incorporated in the process of formal semantic translation. In building the algorithm, we used the data f r o m Study 1, described above. Each of the clauses had been assigned a coherence relation to some other (s), and each clause was coded for a number of factors which had been suggested in the literature as signals of coherence relation (or discourse function) (see above, Section "Studies of Coherence in Commentary Text", for the list). The coherence relation assignment algorithm was developed by examining the information-bearingness of each of these factors as signals of coherence relation assignment. Information-bearingness was determined by tabulating the coherence relation assigned against the factors. Those factors which strongly correlated with a certain coherence relation, and were not evident for others, or which in combination with one other factor were decisive, had high information-bearingness. Those with a high informational load were including as factors to be considered during processing. In addition, we considered the most efficient ordering of tests for the factors. Information-bearingness results follow. Cue phrases such as "in order t o " for G O A L and "because" for C A U S E (Cohen 1984) are decisive for coherence relation assignment where present, but are only present in 9 per cent of local coherence and 16 per cent of global. Similarly, certain specific lexical items such as the verbs "contrast" or "oppose" for the C O N T R A S T relation are highly indicative, but rare in the data. As for syntax, most constructions are merely suggestive of coherence relations. For example, a main clause is more likely to introduce an argument of certain coherence relations. However, such tendencies are not helpful in building an algorithm. A small number of syntactic structures, on the other hand, are decisive. These are the comparative for C O N T R A S T , a generic sentence for a C O N S T I T U ENCY, verb ellipsis for a C O N T R A S T , relative clause, participial and appositive for B A C K G R O U N D . Turning to formal semantics, tense alone is not informative for C R A , because clauses in the simple past tense introduce events which can bear any coherence relation to other discourse events. Clauses in the simple present introduce events which bear all of the coherence relations to other discourse events. However, temporal order, that is, whether or not the events or states introduced in two clauses overlapped in time, is informative. CAUSE, G O A L , E L A B O R A T I O N and EVALUATION

654

Kathleen Dahlgren

require temporal precedence, while CONTRAST, CONSTITUENCY, B A C K G R O U N D , and others can relate fully overlapping events or states. Thus lack of temporal order can be used to exclude the possibility of certain coherence relations. Aspect is more decisive than temporal order in coherence relation assignment. Aspect refers to the temporal perspective, the continuity and completion, of a clause. One of the two clauses introducing discourse events must be telic for certain causal coherence relations to hold between the events. For other relations to hold, clauses must be clause activity or clause Stative. In NS, the aspect of a verb is listed as part of its lexical entry. But context affects aspect, so the aspect of the entire clause must be computed (Moems and Steedman, 1987), taking into account factors such as progressive verb marking and quantified or unspecified subject or object. An algorithm for clause aspect assignment is under development. A final factor which influences discourse coherence inferences is commonsense knowledge, which is required when a pair of clauses provide no or insufficient cue phrases, syntactic properties, temporal order or aspect information for CRA. An example is (1), which has two simple past tense clauses, both clause-telic, both main clauses with no discourse cues. These properties are consistent with CAUSE, GOAL, ENABLEMENT, ELABORATION, EVALUATION. Here naive semantics can be used. The naive semantic representation of the verb invest is powerful enough to drive the inference that goal (el,e2). In the corpus, independently derived naive semantic representations are sufficient for coherence relation assignment in all of the cases where it is needed. Coherence relation assignment algorithm. The local algorithm, shown in Table 3, considers each clause (Source) in relation to the others in a segment one at a time (Target). Another (global) algorithm builds the segment tree. The information the local algorithm uses are syntactic properties of the source clause, connectives in either the source or target, the temporal order of the events in the source and target, naive semantic information associated with the verbs, semantic information such as types of adverbials, mood, and agentiveness in the clauses. The algorithm was hand tested in the original corpus with 97.5 per cent accuracy. A test on an additional corpus of 8,000 words is in progress.

Coherence relations

655

Table 3. Discourse coherence algorithm

Source and Target cohere under Relation if syntactic tests return Relation, or connectives indicate Relation, or Relation = EVALUATION if import, evaluation or comment tests succeed or causal tests return Relation and not both Source and Target are Stative and source is temporally before target or Relation = B A C K G R O U N D if activity tests succeed, or Relation = ELABORATION if sequence tests succeed and Source and Target are telic and source is temporally before target or Relation = C O N S I T U T E N C Y if implication or instrumental tests succeed.

Applying the algorithm Finally, we step through the algorithm on a complex sample text, which is a paraphrase of one of the articles in our WSJ corpus. Levine, (el) charged with SEC violations last May, (e2) was convicted (e3) and sentenced here yesterday. Levine (e4) had engaged in extensive insider trading. He (s5) was greedy and (s6) wanted more money. Levine's light sentence (s7) reflects (e8) an attempt by the court to (e9) reward cooperation in such cases. The judge (elO) said that Levine's ( e l l ) cooperation (el 2) had influenced him in his favor. Critics (el3) argued that light sentences (el4) will result in more violations. The coherence relations in this text that we explain are ELABORATION^^), BACKGROUND^!,e2), BACKGROUND^,e2), CAUSE(s5,e4), GOAL(s6,e4), EVALUATION^,e2), EVALUATION(elO,e2), and EVALUATION(el3,e2). We use the notation Ci to denote the clause which introduces an event ei or a state si. The first clause to be considered is C2 in relation to the clause C3. Referring to the algorithm in Table 3, we see that the main clause C2 will designate a E L A B O R A T I O N because C2 will fail the syntactic tests (it is not a relative clause, appositive, nor any of the syntactic structures the algo-

656

Kathleen

Dahlgren

rithm looks for). There is no connective, no verb of saying for the comment test, and no modal, conditional, interrogative or import verb for the import lest. When the algorithm tries elaboration tests to prove that C2 expresses an E L A B O R A T I O N , it will succeed. Next, the algorithm considers CI in relation to C2. Here tests succeed on the source clause C I . Since CI is a participal it must be B A C K G R O U N D . N o t e that the time adverbial in the main clause C2 did not result in the same assignment. Next the algorithm considers C4 in relation to C2. Considering C4 in relation to C2, syntactic tests, connectives, comment tests and import tests all fail. There is only an indirect relation between breaking the law and being convicted, so causal tests fail. Now the algorithm tries B A C K G R O U N D , and succeeds because C4 is in the perfect. Next the algorithm tries C5 in relation to C2. All tests up through import fail. Now the algorithm tries causal tests and finds in the NS representations that greed can cause people to break the law, so it assigns cause(s5,e2). Similarly, it finds that a typical goal of breaking the law is making money, so it assigns goal(s2,e4). Turning to C7 in relation to C2, the syntactic and connective tests fail. The import test succeeds because (s7) overlaps (e2) in time, and reflect is an import verb. Finally, the two comment clauses CIO and C I 3 are discovered because they fail the syntactic and connective tests, and they contain nonperformative verbs of saying. Thus C7, CIO, and C I 3 are all assigned the relation EVALUATION.

Interactions of levels of grammar The complex interactions of these factors in determining coherence are illustrated by the connective "when". A "when"-clause can relate to the main clause as B A C K G R O U N D , E N A B L E M E N T , C A U S E or ELABO R A T I O N . If the "when"-clause has aspect clause-activity or clausestative, "when" means "at the same time as". The "when"-clause is B A C K G R O U N D relative to the main clause as in (17), but only if it can be interpreted as temporally overlapping the main clause. (17)

When John was (*had been, *will be) a stockbroker, he invested heavily in wheat. When John was (had been, *will be) working on Wall Street, he invested heavily in wheat.

If the when clause has telic aspect, clausetelic, " w h e n " can mean "before", "at the same time as" or "after". Because the clauses are telic,

Coherence relations

657

CAUSE. G O A L , E N A B L E M E N T and E L A B O R A T I O N are favored relations. In this case, the temporal relationships between the clauses and naive semantic information are used to assign coherence. The three examples below are taken from Moens and Steedman (1987). In (18), the "when"-clause event occurs after the main clause event, narrowing the possible relationships. Naive semantic information concerning "build" and "architect" are used to infer an E N A B L E M E N T relation. (18)

(el) When they built the 39th St. bridge, (e2) a local architect (and drawn) drew up the plans. — enablement(e2,el)

In (19), the two clauses overlap in time (by naive semantic knowledge of "build", which implies using materials). Of the preferred relations for telic clauses with "when", the only one with overlapping time is ELABORATION. (19)

(el) When they built the 39th St. bridge, (e2) they used the best materials. — elaboration(e2,el)

In (20), naive semantic knowledge is used to infer that the "when"-clause event occurs before the main-clause event, and that a bridge could solve traffic problems, so that a C A U S E relation between the clauses is plausible. (20)

(el) When they built the 39th St. bridge, (e2) they solved most of their traffic problems. — cause(el,el)

Naive semantic information must always be consulted for "when"-clauses. When there is no causal relation between the events, the effect of the "when"-clause is determined by temporal and aspectual relations between the clauses. Temporal relations of past-perfect/past, past/past, or past/ future select a reading for "when" of "at the same time as" or "just after", as in (21). (21)

When John had eaten dinner, he listened to the news. When John ate dinner, he listened to the news. When John graduates, his father will die.

A past/past-perfect sequence results in a "just before" reading for "when". In all cases, the relation of the "when"-clause to the main clause is BACKGROUND. (22)

When John graduated, his father had died.

658

Kathleen Dahlgren

Conclusion In conclusion, we have shown that coherence relations are justified on psychological, morphological and philosophical grounds. Coherence belongs cognitively as part of language-independent naive reasoning about the relations among events in the world, applied to a cognitive model of discourse content. We have integrated coherence theory with syntax and formal semantics. A formal coherence relation assignment algorithm is feasible and highly predictive.

Notes 1. Whatever is valuable herein is the result of discussions with N. Asher, C. Lord, J. P. McDowell, B. Partee and E. P. Stabler, Jr. and to the stimulation of the Symposium on Discourse Coherence and Segmentation, University of Texas, 1989. 2. We would claim that this is true even in the interpretation of metaphors and fiction. 3. We mean by clause any deep S, including those for derived nominals.

References Asher, Nicholas 1987 "A typology for attitude verbs and their anaphoric properties", Linguistics and Philosophy 10: 1 2 5 - 1 9 8 . 1989 Abstract objects and anaphora in semantics. [MS.] Black, John B. — Hyman Bern 1981 "Causal coherence and memory of events in narrative", JVLVB 20: 267 — 275. Black, J. B. - Gordon H. Bower 1980 "Story understanding as problem-solving", Poetics 9: 223 — 250. Brewer, William F. 1977 "Memory for Pragmatic implications of sentences". Memory and Cognition 5: 6 7 3 - 6 7 8 . Brachman, Ronald J. — V. P. Gilbert — H.J. Levesque. 1985 "An essential hybrid reasoning system", Proc. IJCAI. 1: 532 — 539. Bürge, Tyler 1979 "Individualism and the mental", Midwest Studies in Philosophy 4: 73 — 121.

Carey, Susan 1985 Cohen, Robin 1984

Conceptual change in childhood. Cambridge: MIT Press. "A computational theory of the function of clue words in argument understanding", COLING 2 5 1 - 2 5 8 .

Coherence relations

659

Dahlgren, Kathleen 1985 "The cognitive structure of social categories", Cognitive science 9: 379 — 398. 1985b "Social terms and social reality", Folia Unguistica Historica 6: 107 — 126. 1988 Naive semantics for natural language understanding. Boston: Kluwer Academic Press. 1988 b "Using commonsense knowledge to disambiguate word senses", in: F. St. Dizier — V. Dahl (eds.) Natural language understanding and logic programming 2, 255 — 275, Amsterdam: North Holland. 1988 c Coherence relation assignment. [Proceedings of West Coast Conference of Linguistics, Fresno, California.] 1989 Coherence relations and naive semantics. [Paper given at the Symposium on Modelling Discourse Structure: Discourse Segments and Discourse Relations, University of Texas, Austin.] Dahlgren, Kathleen and Joyce P. McDowell 1986 a "Kind types in knowledge representation", Proceedings C O L I N G 86. 1986 b "Using commonnsense knowledge to disambiguate prepositional phrase modifiers", Proc. AAAI 86. Dahlgren, Kathleen — Joyce P. McDowell — Edward P. Stabler, Jr. 1989 "Knowledge representation for commonsense reasoning with text." Forthcoming in Computational Linguistics. Davidson, Donald 1967 "Causal relations", J. Phil. 64: 6 9 2 - 7 0 3 . Decker, Nan 1985 "The use of syntactic clues in discourse processing", Proceedings Association of Computational Linguistics. Dowty, David R. 1979 Word meaning and montague grammar. Dordrecht, Holland: D. Reidel Publishing Company. Fillmore, Charles 1968 "The case for case", in: Bach, E. — R. T. Harms (eds.) Universals in linguistic theory. 1—90, NY: Holt, Rinehart and Winston. Fodor, Jerry A. 1987 "Modules, frames, fridgeons, sleeping dogs, and the music of spheres". In ed. Jay Garfield, Modularity in knowledge representation and naturallanguage understanding. Cambridge, MA: MIT Press. Foss, D . J . 1988 "Experimental psycholinguistics", Annual review of psychology 30: 301 — 348. Foss, Carolyn L. — Gordon H. Bower 1986 "Understanding action in relation to goals", in: N.E. Sharkey (ed.) Advances in cognitive science Chichester, England: Horwood 95 — 157. Fox, Barbara 1984 Discourse structure and anaphora in written and conversational English. [UCLA Dissertation.] Garfield, Jay L. 1987 Modularity in knowledge representation and natural-language understanding. Cambridge, MA: MIT Press.

660

Kathleen

Dahlgren

Gclman, Rochellc — Elizabeth Spclkc 1981 " T h o u g h t s about animate and inanimate objects", in: J. H. Flavell — L. Ross (eds.) Social Cognitive Development, 43 — 66, Cambridge, England: Cambridge University Press. Graesser, Arthur 1981 Prose comprehension beyond the word. New York: Springer-Verlag. Graesser, Arthur — Leslie Clark 1985 a Structure and procedures of implicit knowledge. N o r w o o d , NJ: Ablex. 1985 b "The generation of knowledge-based inferences during narrative comprehension", in: G . Riekheit — H. Strohner (eds.) Inferences in Text Processing 53 — 94, Amsterdam: N o r t h - H o l l a n d . Graesser, A r t h u r C . — Scott P. Robertson — Edward R. Lovelace — Daniel M. Swinehart 1980 "Answers to why-question expose the organization of story plot and predict recall of actions". Journal of Verbal Learning and Verbal Behavior 19: 1 1 0 - 1 1 9 . Grosz, Barbara — C a n d a c e Sidner 1986 "Attention, intensions and the structure of discourse. A review." Computational Linguistics 7: 8 5 - 9 8 , 12: 1 7 5 - 2 0 4 . Hayes, Patrick J. 1985 " T h e second naive physics manifesto", in: J. R. H o b b s — R . C . Moore (eds.) Formal Theories of the Commonsense World. N o r w o o d , NJ: Ablex. Hinds, John 1977 " P a r a g r a p h structure and pronominalization". Papers in linguistics 10: 1 - 2 , 77-99. Hirst, G r a e m e 1981 "Discourse-oriented a n a p h o r a resolution: A review", Computational linguistics 7: 8 5 - 9 8 . H o b b s , Jerry A. 1979 " W h y is discourse coherent?", SRI Tech. N o t e # 1 7 6 . 1985 " O n the coherencc and structure of discourse", CSLI Report # CSL1-8537. Holland, Dorothy — N a o m i Q u i n n (eds.) 1987 Cultural models in language and thought. Cambridge, England: Cambridge University Press. Hopper, Paul — Sandra T h o m p s o n 1980 "Transitivity in g r a m m a r and discourse", Language, Vol. 56: 251 —299. Jackendoff, Ray 1985 Semantics and cognition. Cambridge, MA: M I T Press. Johnson-Laird, Phillip N. 1983 Mental models. H a r v a r d University Press. Kamp, Hans 1981

Keil, Frank C. 1979

"A theory of truth and semantic representation", in: J. Groenendijk — Th. Janssen — Μ. Stokhof (eds.) Formal methods in the study of language. Amsterdam: Mathematisch C e n t r u m , 277 — 322. Semantic and conceptual sity Press.

development.

Cambridge, MA: H a r v a r d Univer-

Coherence relations Kicras. D. E. 1978

661

" G o o d and bad structure in simple paragraphs: F.ficcts on apparent theme. reading time, and recall", Journal of verbal learning and verbal behavior 17: 1 3 - 2 8 .

Lockman, Abe — David Klappholz 1980 "'Toward a procedural model of contextual reference resolution". Discourse processes 3: 25 — 71. Lakoff, George 1985 Women, fire and other dangerous things. Chicago. 1L: University of Chicago Press. Lord, Carol — Kathleen Dahlgrcn 1989 Tracking participants and events in newspaper articles. In preparation. McDowell, Joyce — Kathleen Dahlgren 1987 " C o m m o n s c n s e reasoning with verbs", Proc. IJCAI. M a n n . William — Sandra T h o m p s o n 1987 Rhetorical structure theory: A theory of text organization. IS1 Reprint Series: IS1-RS-87-190. Marslcn-Wilson. William (cd.) 1989 Lexical representation and lexical process. Cambridge: M I T Press. Millis. Keith K. — David C. Morgan — A r t h u r C . Graesser 1990 "The influence of knowledge-based inferences on reading time of expository text", to appear in: Graesser and Bower (cds.). Moens. M a r c — Mark Stcedman 1987 "Temporal ontology in natural language". Proc. Association Computational Linguistics 1—7. Morrow, Daniel G. 1985 "Prominent Characters and Events Organize Narrative Understanding", Journal of Memory and Language 24: 304 — 319. Morrow. Daniel B. — S. L. Greenspan — G . H. Bower 1987 "Accessibility and Situation Models in Narrative Comprehension", Journal of Memory and Language 26: 165—187. Murphy, Gregory L. and Douglas L. Medin 1985 "The Role of theories in conceptual cohcrcnce". Psychological Review 92: 289-316. Nicholas, D. W. — T. Trabasso 1980 "Toward a Taxonomy of Inferences", in: F. Wilkening — J. Becker — Γ. Trabasso (eds.) Information baum. Noel, Dirk 1989

Integration

by Children.

Hillsdale, N . J . : Erl-

"The study of coherence relations: What is wrong with it and how it can be improved u p o n " . Journal of Applied Linguistics 3: 137 — 143.

Partee, Barbara 1984 " N o m i n a l and temporal a n a p h o r a " , Linguistics and Philosophy 7: 243 — 286. Pohl, R. F. 1982 "Acceptability of story continuations", in: A. Flammer — W. Kintsch (eds.) Discourse processing.

122 — 136. Hillsdale, NJ: Erlbaum.

662

Kathleen Dahlgren

Polanyi, Livia 1988

"A formal model of the structure of discourse", Journal of Pragmatics 12: 601-638. Reichman, Rachel 1985 Getting computers to talk like you and me. Cambridge, MA: MIT Press. Reichman-Adar, Rachel 1984 "Technical discourse: The present progressive tense, the deictic 'that', and pronominalization", Discourse Processes 7, 337 — 369. Reinhart, Tanya 1982 Principles of Gestalt perception in the temporal organization of narrative texts. [MS.] Rickheit, Gert — Hans Strohner 1985 Inferences in text processing. Amsterdam: North-Holland. Rosch, Eleonor — Carolyn Mervis — William D. Gray — D. M. Johnson — P. Boyes-Braem 1976 "Basic objects in natural categories", Cognitive Psychology 8: 382 — 439. Schänk, R. C. - R. P. Abelson 1977 Scripts, plans, goals and understanding. Hillsdale, NJ: Erlbaum. Schiffren, D. 1987 Discourse markers. Cambridge: Cambridge University Press. Schubert, Len K. — R. G. Goebel — N. J. Cercone 1979 "The structure and organization of a semantic net for comprehension and inference", in: N.V. Findler (ed.) Associative Networks. New York: Academic Press. Seifert, Colleen M. - Scott P. Robertson - John B. Black 1985 "Types of inferences generated during reading", JVLVB 24: 405 — 422. Strawson, P. C. 1953 Individuals. London: Methuen. Tanenhaus, Michael K. — GaryS. Dell — Greg Carlson 1987 "Context effects and lexical processing: A connectionist approach", in: Garfield (ed.) 8 3 - 1 0 8 . Trabasso, Tom — Linda L. Sperry 1985 "Causal relatedness and the importance of story events", Journal of Memory and Language 24.1: 595 — 611. Trabasso, Tom — Tom Secco — Paul van den Broek 1983 "Causal cohesion and story coherence", in: H. Mandl — Ν. S. Stein — T. Trabasso (eds.) Learning and comprehension of text. 83 — 111, Hillsdale, NJ: Erlbaum. Townsend, David J. — C. Carrithers — Thomas G. Bever 1987 "Listening and reading processes in college- and middle school-age readers", in: R. Horowitz — S. J. Samuels (eds.) Comprehending oral and written language, 217 — 242, Hillsdale, NJ: Erlbaum. van Dijk, Teun A. 1987 "Episodic models in discourse processing", in: R. Horowitz — S.J. Samuels (eds.) Comprehending Oral and Written Language, 161 — 196. Hillsdale, NJ: Erlbaum. Van Dijk, Teun — Walter Kintsch 1983 Strategies of discourse comprehension. New York: Academic Press.

Coherence relations

663

Vendler, Zeno 1967 Linguistics in philosophy. Ithaca, New York: Cornell University Press. Wada, Hajime — Nicholas Asher 1986 "BUILDINGS: An implementation of DR Theory and LFG", Proc. COLING 5 4 0 - 5 4 5 . 1988 A computational account of syntactic, semantic and discourse principles of anaphora resolution. To appear. Webber, Bonnie L. 1985 "Discourse model synthesis", in: R. Berwick — M. Brady (eds.) Computational models of discourse. 267 — 330. Cambridge, MA: MIT Press. Wilks, Yorick 1975 "Preference semantics", in: E. Keenan (ed.) Formal semantics of natural language. Cambridge: Cambridge University Press.

Text pragmatics and computational modeling Annely Rothkegel

1. Introduction The question what are the specific kinds of developing theories on language and language use in terms of manipulating data and representations arises. In this paper, I will try to relate this question to several systems of text generation that have been developed or are being developed. During the last ten or fifteen years, text problems, more and more, have been paid attention to by the areas of artificial intelligence and computational linguistics (for surveys see Kempen (ed.) 1987; Zock — Sabah (eds.) 1988; Rothkegel 1989 a, 1989 c). (I only use these two labels for addressing a wide range of approaches of computational modeling. In this context, I do not want to enter a discussion about disciplines and institutional structuring at universities.) With regard to texts, there are three lines of research that can be observed: • One line is characterized by a computer science orientation with knowledge bases in the centre of interest (e.g., Kukich 1983; Swartout 1983; McKeown 1985; Jacobs 1987; Reithinger 1988). • A second line can be found in cognitive research. Conceptual, representational and processual aspects are in the foreground. The idea of schemata is used for memory representations (Schank — Abelson 1977; Schank 1982). Production processes are modeled in Lehnert 1982; Yazdani 1982; Tonfoni 1990). • A third line is a linguistic one (or more precisely a text-linguistic and/ or rhetorical one). In this view, descriptive aspects concerning regularities of language(s) and language use are in the focus of interest: semantic representations and domain knowledge (Danlos 1987; Rösner 1987; Dale 1989; Koch 1988); pragmatic structures (Mann —Thompson 1987; Hovy 1988; Rothkegel 1989 b); research in writing (Yazdani 1987). In recent projects on text understanding or text production, however, these lines begin to come together. Thus, they offer new possibilities of

666

Annely

Rothkegel

text research. Given this background, I will discuss the planned system TEXPERT (TEXt exPERT, Rothkegel 1988) and related systems such as TEXT (McKeown 1985); RESEARCHER (Paris - McKeown 1987); PROTEUS (Grishman 1990); AUTO KOCH (Andric - Hannsson - Koch 1989); PAULINE (Hovy 1988); PENMAN (Mann - Thompson 1987) to make clear what it means to do text research in terms of computational modeling. As a common topic to all systems mentioned above, I will focus on the handling of the informational structure in texts, which means, of object knowledge and its textualization in the view of information processing. Within this framework, the presentation of objects in texts will receive prefered attention. The main idea is that there are particular schemata on several text levels. These schemata provide us with standards of conveying information for specified purposes. In a production model, these schemata are to be reconstructed. I will try to offer one possibility of reconstruction by focusing on three components: the text-oriented domain modeling (chapter 3), the concept of text actions (chapter 4), and some linearization principles of a text grammar (chapter 5). Before the discussion of these topics is entered, there are some more explaining comments on how text pragmatics and computational modeling can be understood (chapter 2).

2.

Discussion of the central notions

2.1. Text pragmatics Pragmatic investigation means, in a global sense, the determination of the relationship between contextual (extralinguistic), grammatical (semantic, syntactic) and lexical factors. (Phonological and phonetic questions will be excluded here.) The general hypothesis is that grammatical and lexical selections in language use are controled by the context, e. g., knowledge and actions of the participants in a particular communicative situation. A lot of studies are concerned with construing models of context and particular situations with respect to the interpretation of single linguistic expressions (e.g., deixis, the meaning of particles, etc.) or of sentences (e. g., illocutionary force of utterances and corresponding grammatical forms of sentences by using parameters of time and space, motivation and purpose, aims and means, etc.). Another view concerns more complex linguistic constructions such as texts. They provide insights into organisational strategies of structuring coherent pieces of knowledge

Computational

modeling

667

in relation to communicative purposes. We will call this domain of pragmatics "text pragmatics" and explicitly contrast it to "sentence pragmatics". Text pragmatics include the modeling of the whole and completed text, its global and local structure. A central question concerns the relationship between knowledge structuring and text structuring. As — in this view — knowledge of the world is always involved, purely grammatical methods should be abandoned, even if the general aim is to develop a text grammar. In this perspective, a new schema of investigation must be developed which integrates principles of domain modeling and principles of textual linearization and lexicalization. Our interest in this paper refers to linearization phenomena and their dependence on the structuring of object knowledge.

2.2. Computational modeling There are three components necessary to meet the needs of linguistic computer systems: • Description of linguistic data: authenticity and expliciteness. Within the framework of text linguistics and/or discourse analysis, there is a lot of ongoing research in finding textual regularities on the basis of empirical investigations. In contrast to "sentence linguistics" which relies on competence of native speakers and on intuition of the researchers about the structure of the linguistic unit concerned, in "text linguistics" such a competence does not exist per se. Texts are artefacts which are results of conventionalised processes of textualization of information. As a consequence, empirical work has to refer to authentic material; artificially construed examples are worthless. The second aspect relates to the presentation of the linguistic description. It has to be shaped in a high degree of explicitness and in terms of formalisms such as algebraic sets, phrase structure grammars, logical expressions or other standards which can be programmed. This aspect may be a help for finding and structuring data on the one hand, while on the other hand it means a very strong restriction with regard to what is described. This has to be kept in mind when we consider the descriptive basis of computer models. • Representation: information processing, schemas, levels of representation and transitions between them.

668

Annely

Rothkegel

"Representation" is an originally cognitive term and relates to some kind of "psychological reality". This understanding may be maintained in computer systems, but it need not be. In a weaker definition it only relates to formal standards of how to organise information for information processing purposes. Within the information processing perspective, the concept of schemata such as frames have been very productive with regard to knowledge representation. It was applied mainly in expert systems developed in Artificial Intelligence. Frames offer standardized forms of organizing connected concepts and features. They allow the processing of them on the basis of manipulating hierarchies, inheritance and inferencing. The idea of frames which was introduced in studies of the psychology of perception (Minsky 1977) was fruitful in the field of linguistic (computational) investigation too (Metzing 1980). It provides — for instance — some standards of coherent configurations of predicate-argument — structures and has become a basis for domain modeling. With regard to texts, there are — at least — two levels of knowledge. The one level represents all those parts of knowledge which are relevant with respect to the domain that is represented in the text. For this, there may be a particular level of text representation. The other level is this text itself. It may be considered to be a kind of knowledge representation in terms of natural language. Handling text production phenomena, therefore, can be seen in the light of processes of building up several levels of representation of knowledge including the transitions between them. • Construction: tasks, simulation, modularity. "Construction" refers to the overall architecture of a system. There are tasks related to a broader context that determine the selection of what to model and how to model it. For instance, types of knowledge, text structures and text processes are selected and newly composed according to define presuppositions and purposes. Modules and interfaces are to be defined, and the principles of modularity are to be determined. The overall purpose and the theoretical foundations become visible. Hence, it is the constructive basis of a system that reflects kind and manner of what we are doing when we design, develop and use computer systems (Winograd — Flores 1986). Within the framework of coping with the relation between object knowledge and text knowledge, there are two perspectives that can be distinguished in developing text generation systems:

Computational

modeling

669

(i) There exists some knowledge base with respect to particular objects (e.g., "ship" in TEXT) and a particular text structure is looked for in order to linearize information that is extracted from it. In this view the resulted text structure is mainly determined by the structuring of the object in the knowledge base. The approaches in this field are semanticoriented. (ii) There exists some knowledge of text structure, and particular strategies are looked for which can explain the specific kinds of linearization of knowledge about objects (e. g. P E N M A N , TEXPERT). In this view the resulted text structure mainly determines the structuring of the described objects. The approaches in this field are pragmatic-oriented.

3. Domain modeling and text structure "Domain modeling" means kinds of structuring content referring to standardized pieces of knowledge that might be relevant in a particular context. There are two interesting perspectives: what are the structuring principles and what is the relevant content. As usually applied in knowledge bases, the relationships concerned are as follows: (a) class — sub-class (building-house) (b) class — property (house-material = x) (c) part-of (house-roof) (d) class — instance (house — the Empire State building) (e) predicate — arguments (buy (x, house)) What kind of relationship is chosen depends on the object on the one hand, and on the text goal on the other hand. We can distinguish three kinds of approaches: object-driven (TEXT), operation-driven (RESEARCHER, PROTEUS, AUTOKOCH) and communication-driven (PENMAN, PAULINE, TEXPERT). In TEXT (McKeown 1985), ((a) —(d)) play a dominant role. It is a text generation system which is based on a pre-existing database containing knowledge on ships (database of US-army). The system is part of an expert system which has the task to produce answers to questions about ships. The aim is to offer answers in terms of connected sentences instead of single sentences. The text should improve the quality of the conveyed information. According to the structure of the knowledge base, three types of question are possible: what is e (entity)?, what are the properties of e and what is the difference between el and e2? The outcoming text

670

Annely

Rothkegel

enumerates the content available in the knowledge base. The following example makes it clear (McKeown 1985: 180): Question: What is a destroyer? Answer: A destroyer is a surface ship with a DRAFT between 15 and 222. A ship is classified as a destroyer if the characters 1 through 2 of its HUL NO are DD. All destroyers in the ONR database have REMARKS of 0 and FUEL TYPE of BNKR. Instead of classes and properties, in RESEARCHER (Paris —McKeown 1987) the functioning of an object is modeled (e. g. a loudspeaker in a telephone). A net of paths is established which represent the operations. In this case, the operations ( = the paths and not the nodes as in TEXT) provide the input for generating the text (fragment of an example from P a r i s - M c K e o w n 1987: 99): The variation of the current causes the field of the magnet to vary. When the current increases the field increases. Because the field increases the poles of the magnet attract the diaphragm of the loudspeaker. ... The functioning of machines and especially the non-functioning (failures of operations) is focused in the designed system of PROTEUS (Grishman 1990). Similar to A U T O K O C H (Andric - Hansson - Koch 1989; simulation of operations described in recipes), the knowledge net must contain both information about objects (and their parts and properties) and operations. In this context, the kind of operations determines the selection of relevant parameters for structuring the object. "PEALI N G " in AUTOKOCH, for instance, demands the following categories: (type of) operation, object, instrument, container, final quality, arrangement, direction, force, speed, distance, interval (Andric et al. 1989: 53). The question of what the fillings of the semantic net are concerns the empirical perspective (Tannen 1977). In this view, the knowledge sources themselves influence the structure and content of the outcoming texts. It may be that the content of knowledge is already available in a pre-existing data base (e.g., TEXT) or that it must be extracted from other sources, for instance, from pre-existing texts and inferences (general knowledge of the field) that can be drawn from them (PROTEUS, AUTOKOCH). Whereas the systems mentioned above relate to kinds of structuring that are determined by the objects or operations themselves, P E N M A N (Mann — Thompson 1987) focuses on specific communicative purposes which are described by rhetorical structures). But until now, the relationship between the traditional structuring of objects (see (a) —(e)) and the

Computational

modeling

671

rhetorical relations such as "elaboration", "background", etc., is not made explicit (cf. the example of an advertisement in Thompson — Mann 1987: 95 f.). The idea is that exactly those parts and properties of the object are thematized in the text that are relevant for the addressees. In PAULINE (Hovy 1988) the addressees are involved in such a way that the information can be conveyed more or less detailed in dependence on a parameter of "interest". Whereas (i) gives the original figures of a message, in (ii) an inferencing procedure provides a more global information (Hovy 1988: 43): (i) In the primary on 20 February, Carter got 20515 votes. Kennedy got 21850. (ii) In the primary on 20 February, Kennedy narrowed Carter's lead by 1335 votes. The situation becomes even more complex when we do not deal with physical objects but with communicative objects such as books (as texts). An example for this is offered by the designed system TEXPERT (Rothkegel 1988). The text type chosen is the "book announcement". Texts of this type are provided by the publishers in order to inform potential readers about a new publication. We restrict ourselves to announcements of scientific literature. A communicative object is an object which is produced by communication and which is relevant for communicative purposes. It is a part of some specified communication. The book as a text is such a product which is created by communicative acts and which is relevant in other communicative acts, e. g., within the context of professional work. Book announcements are relevant because of the information about a particular book. This information may be important (or qualified as unimportant) for further work of the participants of the professional scene concerned. Another aspect relates to the fact that the information is conveyed by a third participant-role of the communication (besides author and reader of the book): by the writer of the book announcement. Therefore, we have to deal with a process of conveying information which refers to two different knowledge sources: (1) general knowledge about books (available to the writer of the announcement; it includes the (anticipated) expectation of the group of addressees) (2) specific knowledge about a particular book. (1) and (2) can be seen in the relation of class — instance (type —token). (1) provides the general frame (coherent system of slots and values) of

672

Annely

Rothkegel

what is usually relevant in the domain of books. (2) relates to selections of the writer. He/she determines which slots by which values are filled. It is a procedure of abstracting information from a particular book according to the general frame. What is the general frame for books? What method will help us for domain modeling? The hypothesis is that, in the case of communicative objects, text knowledge determines the structuring of object knowledge. In other words: the conventionalized ways of acting in communication and for communicative purposes (by communicative acts, see 4.) determine the selection of categorization which may be relevant for these purposes. Text types then are conventionalised patterns of combined and related categories which are available for language use in (relatively) standardized situations. Authentic texts of book announcements are considered to provide empirical material for abstracting those categories which are relevant for the purpose of I N F O R M I N G . One pattern which is very often used in book announcements reflects the presentation of the content in terms of a report on "what the author (or the book respectively) does". It is a kind of action-description about the production of a book ( = text). For the interpretation of the applied predicate classes and argument classes ("text roles"), a top-down strategy is used which profits from some definitions of "action". The predicates and arguments can be classified according to the different components of actions. We can find all components which are relevant for a general definition of actions such as: participants (agent ( = author of the announced book), recipient ( = anticipated readers), motivation, intention, purpose, means, sub-parts of the action, order of sub-parts, etc.). As a result of the analyses of actual texts, empirical data on language use are available which allow — in connection with the component of the concept of action — the determination of text roles. These text roles, finally, present the domain knowledge of "books". They provide the WHAT of the knowledge base. According to the components of the definition of action there are global typologies which can be specified according to the occurrencies found in the data. The empirical investigation of about a hundred texts (50 in English and 50 in German) had the result that there are recognizable schemata of structuring information about books. Figure 1 shows a fragment of a possible frame concerning the text roles and some sub-frames in terms of attribute-value-pairs. A generalization of classes of predicates relates to types of sequencing the information on books (see chapter 5).

Computational

modeling

673

frame: object "book" BOOK:

WHOLE: PART:

P A R T I (introduction, part I, ...) P A R T II (chapter 2, part II, ...)

SEGMENT:

I N D E X (subjects, authors, ...) BIBLIOGRAPHY

PUBLICATION:

TIME:

FIRST:

LANGUAGE:

RE-PUBL.: LI, L2, ...

TEXT TYPE:

MONOGRAPHY

THEME:

HIERARCHY:

DOMAIN:

SCIENCE:

AUTHOR:

LITERATURE: POLICY: QUALIFICATION:

READER:

QUALIFICATION:

M A I N (ΤΙ, T2, ...): S U B (SI, S2, ...):

EXPERT PROMINENT EXPERT NO BACKGROUND

MOTIVATION: INTENTION: EFFECT:

DEFICIT PROBLEM DEMONSTRATION PROOF INFORMATION ENTERTAINMENT

Figure I

As a rule, the actual text realises only a part of the whole pattern. As a consequence, there is a large number of variations of actualizations. These variations form a kind of typology of linearization (scripts, see also chapter 5). The configuration of the predicates are the indicators for a particular type of it. In the following, an example is given for demonstrating the principles. For our purposes, the scripts can be made visible by listing the propositions (predicate —argument —combination) of the actual text. [1] represents an instance of the script-type S E Q U E N C I N G PARTS (BOOK, THEME). ( " ( ) " designates the list of arguments, "[ ]" a repeated argument which is superficially deleted and which refers — as a rule — to the predecessor proposition,"{}" designates the interpretation of text roles. Linkings of propositions which are lexicalized by conjunc-

674

Annely

Rothkegel

tions are considered to be propositions of their own combining two (or more) other propositions). [1] text in full form The volume presents — for the first time in book form in English — the work of two major representatives of the so-called Moscow — Tartu school. The introduction outlines their project for a "poetics of expressiveness" against the background of the structural-semiotic movement of the '60s and '70s. Part I is a systematic exposition of the theory, concentrating on the concepts of theme, expressive device, poetic world, etc. Part II and III apply these concepts to a structuralist portayal of Leo Tolstoy's tales for children ... and of the medieval Latin author Archpoet of Cologne .... The volume is provided with a Bibliography of the poetics of expressiveness and a Glossary of its metalanguage. (Shcheglov, Yu. & A. Zholkovsky, Poetics of Expressiveness: A theory and application. Amsterdam, 1987. ... in: New Book Information, John Benjamins Publishing Company) [la] text as propositional list and interpretations: (1) present

(2) \p.

]

(3) outline

(4) be

(5)

concentrate

(the volume

\

(2)

1

the work of two m. of... school)

{BOOK

|

(2)

1

WHOLE}

([vol. ] {BOOK

| |

[work] 1 for the first time in book form) WHOLE | PUBL: FIRST: for the f . time L: english}

(the

introduction

1

their project for ... against

{BOOK: PART

1

THEME}

(part I

1

syst, exposition of the theory)

{BOOK:PART

1

THEME:MAIN}

(1[p• 1

1

on the concepts of theme, e. device, p. world,

1

THEME: SUB}

{THEME:MAIN (6) apply

(7) be provided

representatives

...)

(5)

(part II and III {BOOK:PART

1 1

these concepts \ THEME:SUB 1 |

(the volume

1

with a Bibliography of... and a Glossary of ...)

{BOOK

1

BOOK:SEGMENT}

etc.)

to a str. portrayal of T. tales ... and TEXT:PART}

of...)

Computational

modeling

675

4. Text actions (TA) Domain modeling and its reflection in the knowledge base represents a general level of knowledge. It is selected and adapted to a specific communicative situation by composing a text. This can be explained on the basis of text knowledge. In TEXT semantic predicates such as DEFINITION, IDENTIFICATION, etc. are introduced in order to relate pieces of the knowledge base and text structure. PENMAN relies on hierarchical relations that are based on a binary organisation (nucleus and satellite) adapted from rhetorics (thesis and argument), and that are used for building up the whole text structure. In the model of TEXPERT, text knowledge relates to processes of building up a text. The text structure is considered to be the result of building up this structure by text actions. There are two sources which are relevant for our theory: speech act theory in the line of Searle (1969) and theory on activity in the line Leont'ev (1971). The interesting point in speech act theory is an action model containing dimensions of agent, intention and purpose and dimensions of the structural connection between illocution (as sustainer of intention, purpose, change), proposition (as the content that is involved in the action) and locution (as the execution by linguistic means). This approach has also been adapted for the representation of text structures (e.g., Gülich 1986; Motsch-Pasch 1986; Viehweger 1991). The psychologically oriented activity theory offers an action model relating to the course and phases of an action (beginning, middle, end) and its achievement as well as to motivation and strategies. A linguistic adaptation was carried out by Rehbein (1977). In our model we will rely on several ideas of these approaches. Both structural and phase-oriented dimensions are integrated into the concept of text actions (see also for legal texts Rothkegel 1985, 1986). It designates particular relations between several knowledge types. In a structural point of view, the concept combines the levels of illocution, proposition, and locution on the one hand (TA in (a) and (a')), and hierarchies of complex text acts (TAc in (b)) as well as linearizations (TAI in (c)) on the other hand. In a process-oriented view of cognitive information processing, text actions represent specified instructions for selecting and composing a communication-oriented informational structure and its linguistic realization by lexical and syntactic means. With regard to the notation, illocutions are considered to be predications on propositions (Häfele 1979). With regard to the implementation they are instructions for selecting the appropriate kind of knowledge pieces and linguistic expressions.

676

Annely

Rothkegel

(a) TA = (I, Ρ, L), where I (illocution) represents the dimension of purpose (goal) in communication, Ρ the content that is represented by a proposition (predicate-argument-structure), and L represents the linguistic means such as lexical expressions and syntactic forms and patterns. For our purpose, we combine I and Ρ in a separate module of interpretation (MI), whereas L forms the module of execution (ME), see (a')· (a') ΤΑ = (MI (I,P)), M E (L)) (b) TAc = (TA(1) : TA(2) ...), where ":" designates several hierarchical relations, e. g. BY-relation (doing χ by doing y) or PART-OF-relation (doing χ consists of doing y, z, ...) (c) TAI = TAi (TAI, TA2, ..., TAn), where "," designates linear sequence (first doing x, then y, then z). In a typology of speech acts (Falkenberg 1988, Dimter 1981) epistemic speech acts play an important role (illocution as propositional attitude). The structure of those speech acts allows a more precise definition of our interpretation module MI because it relates I and Ρ in a specific way. According to Dimter (1981: 71) illocutions can be distinguished on the basis of the relationship between W A N T I N G (in our model: purpose, illocution) and K N O W I N G (in our model: p, propositional content = content of the book frame). Assuming that I N F O R M I N G is a basic text action, it can be paraphrased as follows: INFORMING: writer (w) wants the reader (r) to know that ρ is true or in short form: want (w, know (r, true, (p))) I N F O R M I N G about books can be realized by two different (global) text actions: DESCRIBING and EVALUATING. DESCRIBING means: the labels of the book frame are instantiated by values which correspond to facts concerning the book in question. EVALUATING means: the labels of the book frame are instantiated by values which correspond to norms of the writer (or norms of the social group to which the writer belongs). The different kinds of instantiations are reflected in using different repertories of lexical and syntactic material. EVALUATING can be executed by several means. A frequently used strategy is evaluating by direct ( = lexical) assigning a positive or negative value (EVALUATING by ASSIGNING). Another strategy uses sub-

Computational

modeling

677

patterns of DESCRIBING such as CONTRASTING two entities within a more comprehensive framework (EVALUATING by INTEGRATING) (for an example see [2] in chapter 5).

5. Text grammar As text actions represent the level of the hierarchical text structure, there must be another level concerning the organisation of information in a linear form. There have been investigations, e. g., with respect to schemas combining semantic predicates (in TEXT), to text plans, focus, textual reference and connectedness including the general question of "What's next in the text?" (Fraurud — Hellman 1985). One possibility to establish a general frame for describing these regularities is offered by the concept of script pattern. These patterns can be formalized as a text grammar in terms of rules of combining and sequencing text actions (for a critical view concerning grammar and knowledge schemata such as frames see Wilks 1985). With regard to book announcements, the linearization patterns can be considered to be script-compositions with I N F O R M I N G as a global text action. As sketched in chapter 3, a typology can be established for scripts. Examples are: S E Q U E N C I N G PARTS RELATING RELATING

(BOOK | T H E M E ) (PURPOSE | MEANS) (MOTIVATION | AIM) etc.

The sequences are determined by constraints which can be formulated with respect to selections within the knowledge base. There are two perspectives which can be distinguished: • selections concerning any differentiations between several labels of the frame (e. g. class — sub-class, whole — part, type — token etc.) • selections concerning any differentiations between several purposes of the information (e.g., DESCRIBING-EVALUATING). They have effects on the selections concerning the text lexicon. Whereas the former is oriented towards the structuring of the content, the latter relates to the goal of conveying information. Both are connected by the configuration of illocution, proposition, and locution of the particular text action. [2] gives an example of the (re)construction of a script pattern which is applied to an authentic text.

678

Annely

Rothkegel

[2] text in full form Instead of composing a continuous survey of the development of linguistic techniques of analysis, this book focuses on particular 'landmark' texts in the history of linguistic thought. This strategy allows the authors to consider each theory in depth, rather than simply placing them in a historical narrative. Each chapter contains a short extract from a landmark text, followed by a commentary in which its ideas are clarified for the modern reader and in which the text is situated in its appropriate social and intellectual context. [2a] text as propositional list including interpretations (1) —(3)

EVALUATING by C O N T R A S T I N G (p2, p3)

(1)

instead (2)

((2) I (3)) I N T E G R A T I N G (p2)

(2)

compose

(3)

DESCRIBING (p3)

focus

(this book

(4)

DESCRIBING (p4)

I

a continuous survey of the development of linguistic techniques of analysis,)

on a particular 'landmark' text in the history of I. thought)

(5)-(7)

(this strategy | the authors) EVALUATING by CONTRASTING (p5, p7)

(5)

I N T E G R A T I N G (5)

consider rather

([authors] \ ((5) I (6)) ASSIGNING (p7)

(7)

(7) simply place ([authors] (8) - (12) (8)

I

(3)

(4) allow

(5) (6)

([b- ]

contain

each theory

| in depth)

[theory]

I in a historical narrative)

DESCRIBING (SEQUENCING (SEGMENTS)) (each chapter

a short extract from a 1. text)

Computational

(9) be followed (10) and (11) clarify

([extract] | by a commentary ((11) | (12)) (its (1. text) ideas for ...)

(12) be situated

(the text

|

in its a.

modeling

679

(10))

context)

The status of the information in this text is mainly evaluative, which means, the process of conveying information is mainly related to the standards of knowledge of the text producing system (writer of the announcement). The illocutionary structure (in an isolated form of sequences as one part of a text grammar) is as follows: (1) —(3)

(4) (5)-(7)

(8)-(12)

E V A L U A T I N G (P) BY C O N T R A S T I N G (P) BY I N T E G R A T I N G (P2) AND BY D E S C R I B I N G (P3) D E S C R I B I N G (P4) EVALUATING (P) BY C O N T R A S T I N G (P6) BY I N T E G R A T I N G (P5) AND BY A S S I G N I N G / N E G (P7) D E S C R I B I N G ( P R E S E N T I N G (SEGMENTS))

The analysis shows that it is the combined illocutionary and propositional structure which determines the sequences of linearization. For a (formal) description of patterns of linearization it would be necessary to consider those compositions correlating content and goal and to establish corresponding repertories of linguistic means in a text lexicon.

6. Concluding remarks The relationship between object knowledge and text knowledge in the view of text production can be reflected in different ways. Nevertheless, there are two questions that are common for several approaches: what are the principles of selection and composition and how can they be modeled adequately. The constructive method of computational modeling (see 2.2.) brings together descriptive, representational and modular aspects which allow theoretical questions to be set in the context of particular tasks. The action-oriented perspective allows the relation of system-

680

Annely

Rothkegel

atic aspects of pragmatics and aspects of what we are doing when we use language. In our model, we want to try some integration of the several levels of object knowledge and text knowledge within the paradigm of computeroriented information processing. It is presumed that the text itself is a kind of knowledge representation. It can be considered to be the result of (linguistic) actions and executions of actions (operations). This result can be reached by building up an intermediate level of representation: a semantic-pragmatic representation covering all information that is necessary for realizing the global text structure. It also provides the basis for local (lexical and syntactic) realizations. The process-orientation is reflected in an action-based theory of text actions that allows the formulation and formalization of the interrelationship between illocutionary and propositional aspects and their implications on lexical and syntactic means on the one hand, and on principles of linearization on the other hand. Hence, it offers promising possibilities for modeling text planning by virtue of relating a knowledge base, a text grammar and text patterns. Furthermore, the planning process itself can be re-constructed in terms of constructing specified transitions between these components. With regard to prospective developments, the building up of representations can serve as the basis for modeling human writing processes and the development of a theory of text-writing.

References Andric, Barbro — Kerstin Hansson — Wolfgang Koch 1989 Prozeßsteuerung und Verbsemantik in einem computer simulier ten Küchenmodell: Teilen, Mischen und Erwärmen. S & P , Sprache und Pragmatik, Arbeitsberichte Nr. 14, University of Lund. Antos, Gerd — Peter Krings (eds.) 1989 Textproduktion. Ein interdisziplinärer de Beaugrande, Robert 1984 Text Production. Toward a science of Boscolo, Pietro (ed.) 1989 Writing: Trends in European research. Dale, Robert 1989 "Generating recipes: An overview of

Überblick. Tübingen: Niemeyer, composition.

Norwood: Ablex.

Padova: Upsel Editore. EPICURE", Paper presented at the

First D A N D I Workshop, Lugano, 2 6 - 2 8 October 1989. Danlos, Laurence 1987 The linguistic basis of text generation. Cambridge: Cambridge University Press.

Computational

modeling

681

Dimtcr, Matthias 1981 Textklassenkonzepte heutiger Alltagsspraehe. Tübingen: Niemeyer. Falkcnberg, Gabriel 1981 "Epistemische O p e r a l o r e n und illokutionäre K r a f t " , in: G ö t z Hindelang - Werner Zillig (eds.) 1 9 3 - 2 0 3 . F r a u r u d . Kerstin — Christian Hellman 1985 What's next in the text. A method for investigating discourse processing. Papers from the Institute of Linguistics, University of Stockholm, 54. Freedle, Roy O. (ed.) 1977 New directions in discourse processing. N o r w o o d : Ablex. Grishman, Ralph 1990 Domain Modeling for Language, in: Ulrich Schmitz —Rüdiger Schütz —Andreas Kunz (eds.), Linguistic Approaches to Artificial Intelligence, 41 —57. F r a n k f u r t : Peter Lang. Gülich, Elisabeth 1986 "Reformulierungshandlungen als Mittel der Textkonstitution. Untersuchungen zu französischen Texten aus mündlicher K o m m u n i k a t i o n " , in: Wolfgang Mötsch (ed.), 1 9 9 - 2 6 1 . Häfele, Josef 1979

Der A u f b a u der Sprachkompetenz. Untersuchungen zur G r a m m a t i k des sprachlichen Handelns. Tübingen: Niemeyer. Hindelang, G ö t z — Zillig, Werner (eds.) 1981 Sprache: Verstehen und Handeln. Tübingen: Niemeyer. Hovy, Eduard H. 1988 Generating natural language under pragmatic constraints. Hillsdale: Lawrence Erlbaum Ass. Jacobs, Paul 1987 " K I N G : A knowledge-intensive natural language generator", in: Gerard Kempen (ed.), 2 1 9 - 2 3 0 . Kempen, Gerard (ed.) 1987 Natural language generation. New results in artificial intelligence, psychology and linguistics. Dordrecht: M a r t i n u s Nijhoff Publishers. Koch, Wolfgang 1988 Weltwissen, Sprachsystem und Textstruktur in einem integrierten Modell der automatischen Sprachverarbeitung. Zum Beispiel Kochrezepte. S & P, Sprache und Pragmatik, Arbeitsberichte Nr. 4, University of Lund. Kukich, Karen 1983 Knowledge-based report generation: a knowledge-engineering approach to natural language report generation. [PhD thesis, University of Pittsburgh.] Lehnert, Wendy G. 1982 "A narrative summarization strategy", in: Wendy Lehnert — Μ. H. Ringle (eds.), 3 7 5 - 4 1 2 . Lehnert. Wendy G. — M a r t i n Ringle (eds.) 1982 Strategies for Natural Language Processing. Hillsdale (NJ): Erlbaum Ass. Leont'ev, Aleksej A. 1971 Sprache — Sprechen — Sprechtätigkeit. Stuttgart: Klett-Cotta (Moskau 1969).

682

Annely

Rothkegel

Mann, William — Sandra A. Thompson 1987 "Rhetorical structure theory: description and construction of text structures", in: Gerard Kempen (ed.), 85 — 95. McKeown, Kathleen R. 1985 Text generation. Using discourse strategies andfocus constraints to generate natural language text. Cambridge: Cambridge University Press. Mellish, Christian 1988 "Natural language generation from plants", in: Michael Zock — Gerard Sabah (eds.), 1 3 1 - 1 4 5 . Metzing, Dieter (ed.) 1980 Frame Conceptions and Text Understanding. Berlin: de Gruyter. Minsky, Marvin 1977 "Frame System Theory", in: P. N. Johnson — P. C. Wason (eds.), Thinking: Readings in cognitive science, 355 — 377. Cambridge: Cambridge University Press. Mötsch, Wolfgang — Renate Pasch 1986 "Illokutive Handlungen", in: Wolfgang Mötsch (ed.), 1 1 - 7 9 . Mötsch, Wolfgang (ed.) 1986 Satz, Text, sprachliche Handlung. Berlin: Akademie Verlag. Paris, Cecile L. — Kathleen R. MacKeown 1987 "Discourse strategies for describing complex physical objects", in: Gerard Kempen (ed.), 9 7 - 1 1 5 . Rehbein, Jochen 1977 Komplexes Handeln. Elemente zur Handlungstheorie der Sprache. Stuttgart: Metzler. Reithinger, Norbert 1988 POPEL: Α parallel and incremental natural language generation system. 4th IWG, Santa Catalina Islands, 7, 1988. Rothkegel, Annely 1985 Text Acts in Machine Translation. L. A. U. T. Paper No. 133, University of Trier. 1986 TEXAN. Eine semantisch-pragmatische Textanalyse. Arbeitsbericht des Projekts A3, TEMA, Nr. 1, Sonderforschungsbereich 100, Elektronische Sprachforschung, University of Saarbrücken. 1988 Knowledge Representation and Text Processing. L. A. U. D. Paper No. 219, University of Duisburg. 1989 a "Textualisierung von Wissen. Einige Forschungsfragen zum Umgang mit Wissen im Rahmen computerorientierter Textproduktion", LDV-Forum - Bd. 6, Nr. 1: 3 - 1 3 . 1989b "Textual knowledge and writing", in: Pietro Boscolo (ed.), 172 — 181. 1989c "Maschinelle Textproduktion", in: Gerd Antos — Peter Krings (eds.), 483-501. Rösner, Diemar 1987 "The automated news agency: SEMTEX — a text generator for German", in: Gerard Kempen (ed.), 133 — 148. Schänk, Roger C. 1982 "Reminding and Memory Organization: an introduction to MOPs", in: Wendy Lehnert - Μ. H. Ringle (eds.), 4 5 5 - 4 9 3 .

Computational modeling

683

Schänk, Roger C. — Robert P. Abelson 1977 Scripts, plans, goals and understanding. Hillsdale: Erlbaum. Searle, John P. 1969 Speech Acts. An Essay in the Philosophy of Language. New York: Cambridge University Press. Swartout, W. 1983 "XPLAIN: a system for creating and explaining expert consulting systems", Artificial Intelligence, 21: 285 — 325. Tannen, Deborah 1977 "What's in a frame? Surface evidence for underlying expectations", in: Roy O. Freedle (ed.), 1 3 7 - 1 8 1 . Thompson, Sandra A. — William C. Mann 1987 Rhetorical structure theory: a framework for the analysis of texts. Papers in Pragmatics, Vol. 1, no. 1, 79 — 105. Tonfoni, Graziella 1990 Text Representation Systems. Helsinki: Eurographica. Viehweger, Dieter 1991 Illokutionsstrukturen und Subsidiaritätsrelationen, in: Sprache und Pragmatik, Arbeitsberichte 24, 62 — 76. Wilks, Yorik 1985 "The structures and knowledge structures." Quaderni di semantica 6,2: 335-344. Winograd, Terry — Fernando Flores 1986 Understanding computers and cognition. A new foundation for design. Reading etc. Reading, Ma.: Addison-Wesley Publ. Comp. Yazdani, Masoud 1982 "How to write a story", Proceedings of the First European Conference on Artificial Intelligence. Orsay, France. 1987 "Reviewing as a component of the text generation process", in: Gerard Kempen (ed.), 1 8 3 - 1 9 0 . Zammuner, Vanda L. 1988 "Discourse Planning and production: an outline of the process and some variables", in: Michael Zock — Gerard Sabah (eds.), Vol. II: 144—160. Zock, Michael — Gerard Sabah (eds.) 1988 Advances in natural language generation. An interdisciplinary perspective. Vol. I, II. London: Pinter Publishers.

Integrating knowledge sources for the generation of referential expressions Russell Block

1. Introduction In this paper on generating referential expressions, I would like to discuss some of the research conducted in the W I S B E R project at the University of H a m b u r g . W I S B E R is a G e r m a n acronym for Wissensbasierter natürlich-sprachlicher Dialog 'Knowledge-based natural-language dialog'. W I S B E R was a human-machine communication project in the field of Artificial Intelligence. The goal of the project was to produce a prototype advisory system in the domain of financial investment which would be capable of counselling the user through written natural language dialog conducted in German. As a linguist a m o n g computer scientists, it was my j o b to reduce linguistic theories a b o u t reference to working computational models, given the resources we had at hand. In this paper, I will provide a very brief overview of some of the problems we faced and some of the solutions we found. At the end I will briefly address the question of the relevance of the computer and Artificial Intelligence for linguistic research.

2. Artificial intelligence and linguistic data processing Let me begin by devoting a few words to Artificial Intelligence. Thanks to extensive publicity, particularly in connection with the Japanese Fifth Generation computer project (Feigenbaum — McCorduck 1984), most people who are interested in computer applications have heard of it, but many may still not be quite sure what it is. There are many definitions of Artificial Intelligence (cf. Schank 1987), but my favorite definition is the one popularly attributed to Terry Winograd, one of the pioneers in the field: Artificial Intelligence is concerned with getting machines to do things that humans do better. To understand this, consider for a moment some of the things a machine can do better. N u m b e r crunching, carrying out enormously complex

686

Russell

Block

calculations, is an obvious example. More interesting from the linguist's point of view is a problem like listing all the words in the English language that end in the suffix -ary. With a personal computer and a good word processing program, you can call the spelling checker, type in the pattern *ary, and within seconds generate the required list. Doing the job by hand, leafing through the dictionary, would take hours or even days and even then, you could not be sure that you had captured all the relevant data. This is clearly one of the things that machines do better. It falls under the rubric of Linguistic Data Processing. Resolving and generating referential expressions in texts is one of the things that humans do far better than machines and thus belongs to the area of Artificial Intelligence. Why is this so? The answer is simple, even if the solution is not. The information required for processing referential expressions is not necessarily contained in the expressions themselves, but must be derived from contextual information and world knowledge, often by means of fairly complex and sometimes fallible reasoning processes. What we attempted to do in WISBER was to create a microcosm — a tiny world concerned solely with questions of financial investment — in which we could model and process all of the information necessary for conducting a human-machine dialog in natural language, in German to be exact. Although some of my colleagues would certainly disagree, I see this as the logical extension of the paradigm introduced some 30 years ago by Noam Chomsky for the study of syntax. Just as Chomsky attempted to develop a set of formal rules which would characterize all and only the syntactically acceptable sentences of natural language, we are attempting to develop formalisms for the communicative process as a whole — a performance model. If you consider the difficulties that Chomsky, his supporters, and detractors have had in developing a formalism for syntax alone, you can perhaps appreciate the difficulties involved in developing a communication model.

3. Approaches to the problem How do we approach the problem of reference in a communication model? Essentially there are two possibilities: top down and bottom up. The top down approach takes formal logic as its point of departure and attempts to expand and enrich it in order to encompass the kinds of phenomena that actually occur in natural language. Since formal logic

Referential

Expressions

687

was designed precisely to avoid the vagueness and ambiguities of natural language, this generally leads to insuperable difficulties (Block 1988). In Hamburg, we took the bottom up approach, first defining a relatively broad spectrum of natural language phenomena we wished to cover and then attempting to develop formal methods to deal with these phenomena. This is, of course, no substitute for a theory of reference or of anything else, but it was our hope that the outlines of a theory would emerge from a set of procedures that was actually capable of accounting for the data.

4. The feature system In this spirit I initially defined a set of 19 semantic and syntactic features necessary for generating natural-sounding referential expressions in German and English. This list laid no particular claims to completeness or originality since our purpose was to observe the interactions between knowledge sources and the data flow in the system. The feature system is explained and illustrated at length in Block (1987). We will go into some of the more interesting details below.

5. The knowledge sources The knowledge sources available to the reference generation component included the following: 1. A Lexicon listing the words known to the system along with grammatical information about those words. 2. A Terminological or T-Box containing structured conceptual knowledge about the domain. 3. An Assertional or Α-Box containing assertions about objects in the dialog. 4. The Functional structure containing the output of the sentence parse on analysis and the verbalization on generation. 5. The Internal Representation Structure, which is loosely based on predicate logic, and accompanies each utterance on its way through the system. Before proceeding, we can take a brief look at each of these components.

688

Russell Block

5.1. The lexicon For the sake of simplicity, we maintained the principle: "one word — one concept" in constructing the lexicon. That is, the word Pfandbrief 'bond' referred exclusively to the concept BOND (concepts are indicated by capitalization) or one of its concrete instances like BOND 123 (instances are identified by an arbitrary code number). No synonyms (like couch and sofa for the same kind (or piece) of furniture) were allowed. Referential paraphrases like dieses Wertpapier 'this security' for BOND 123 on subsequent mention were bound by the same principle, the word Wertpapier again corresponding to the concept SECURITY or one of its concrete instances.

5.2. The T-Box It is important to observe that intensional knowledge about concepts in the domain is stored separately from assertions about actual objects. Intensional knowledge is captured in a hierarchical semantic network of concepts qualified by roles and value restrictions on roles (cf. Bergmann — Gerlach 1987). For instance, the concept BOND is a subconcept of SECURITY. BOND has roles like HAS-ISSUER, HAS-YIELD, HAS-TERM, etc. Some of these roles are inherited from the superconcept SECURITY (for example, all securities have issuers and yields), others are specific to BOND (not all securities have terms). In addition the roles have value restrictions which define the range of possible values for role fillers. For instance, WISBER believes that a BOND may have a yield between 3 and 8 per cent. If the yield is less than 3 or more than 8 per cent, it's not a bond. This feature is useful for correcting user misapprehensions. If the user inquires about a bond with a 12 per cent interest rate, WISBER could, in principle, reply that there are no such bonds. On the other hand, if the user requests a bond with an interest rate of 7 per cent and the system does not happen to have a bond with this particular interest rate in its knowledge base, it can reply that no such bond is available.

5.3. The A-Box While the T-Box contains knowledge about general concepts, the information that a specific bond mentioned in the dialog (e.g., BOND 123)

Referential

Expressions

689

has the issuer Dresdner Bank, has the yield 5 per cent and a term of 4 years is listed separately as a series of assertions in the Α-Box (Poesio 1988). Within the Α-Box a distinction is drawn between system knowledge, e. g., information about specific securities the system knows about and can recommend to the user, and mutual knowledge, i.e., things mentioned in the course of the dialog (e. g., the interest rate of BOND 123) and thus known to both system and user. The basic division between concepts and objects is familiar from Platonic philosophy. The T-Box is Plato's world of ideas; the Α-Box contains the concrete reflections of those ideas in the real world.

5.4. The functional structure Roughly speaking, the functional structure contains grammatical information about grammatical relations like subject, object, etc., at the sentence level. This information is potentially useful for generating or resolving sequential references like the following: (1)

a. John punched Harry, then John kicked Harry, b. John punched Harry, then he kicked him.

Assuming noncontrastive stressing, the underlying structure (la) can be pronominalized as in (lb) since, other things being equal, the subject of the follow-up sentence will be assumed to refer to the subject of the initial sentence and the object of the follow-up sentence to the object of the initial sentence. For some discussion and caveats see Block (1989).

5.5. The internal representation structure The internal representation structure (Bergmann et al. 1987) is expressed in a special formalism called IRS (for interne Repräsentationssprache 'Internal Representation Language'). IRS is essentially a "machine-readable" formalism for encoding grammatical, semantic and pragmatic information contained in utterances. The internal representation structure of a given utterance is continually transformed during processing. On generation, for instance, the deepest internal representation structure encodes the system's intentions — what the system wants to say (Sprenger — Gerlach 1988). After processing by various components, the expanded structure serves as the input to the verbalization component, which casts the utterance in words.

690

Russell

Block

6. The reference generation component The reference generation component makes use of the knowledge sources discussed above and includes in addition two other knowledge sources and a processor: 1. S E M M E M O R Y — containing listings of objects and propositions that are potentially referable along with their semantic features like ANIMACY or C A R D I N A L I T Y 2. R E F O M E M O R Y — containing listings of the names of objects that have been referred to by the user or the system along with their grammatical features like G E N D E R and N U M B E R . These are listed in order of occurrence in the dialog. Once again we create a useful dichotomy — this time between objects and their names. The reason for this should be clear. The object BOND 123 remains BOND 123 throughout the dialog. Any additional information about BOND 123 that is brought to light must be listed under this unique heading. On the other hand, BOND 123 may have many different names during the course of the dialog: der Pfandbrief 'the bond' (with masculine pronominal reference er), dieses Wertpapier 'this security' (with neuter pronominal reference es), or even, somewhat metaphorically: der Grund all meiner Sorgen 'the source of all my troubles'. The SEMMEMORY, containing semantic information about objects mentioned in the dialog, is located in a special compartment of the ABox. Grammatically relevant information like the fact that BOND 123 is semantically definite (known to the system and the user) is listed here, while contingent information such as the fact that BOND 123 is issued by the Dresdner Bank is listed elsewhere. This division reflects the fact that natural languages have special means for referring to objects that are mutually known to the speaker and hearer (e. g., the definite article among others in German and English), but no grammatically relevant way of distinguishing between bond-isuing institutions like the Dresdner Bank and the Commerzbank. We will discuss the use of contingent knowledge for generating unique references further below. The REFOMEMORY, dealing as it does with the names of objects, is located outside the Α-Box. Listings are arranged sequentially and cross referenced to the actual objects in the SEMMEMORY. The usefulness of this arrangement is explained below. The heart of the reference generation system is the processor SEMLISTERS (SEM from semantics and LISTERS from Interlisp-D, the

Referential

Expressions

691

computer language in which our part of WISBER is implemented). SEMLISTERS contains the procedures for culling the information necessary for generating referring expressions from the available knowledge sources.

7. Data flow In this section I will present a few simple examples to illustrate the kinds of problems the computer has to deal with and perhaps inspire the reader with new wonder for the capacities of the human mind and its ability to deal with problems of reference generation and resolution.

7.1. Grammatical reference The simplest case arises when it is possible to generate references purely on the basis of grammatical information available in the R E F O M E MORY: (2)

Die Frau kaufte einen Pfandbrief. Sie hielt ihn für preiswert. 'The woman bought a bond. She considered it a good buy'.

In this context the pronoun sie can only refer to die Frau (grammatical feminine singular) and the pronoun ihn to the masculine singular Pfandbrief. Grammatical information alone is sufficient here to generate (or resolve) pronominal references. In this case, then, we can pronominalize the underlying full NPs in the second sentence of (2).

7.2. Semantic complications It is easy, however, to find cases in which grammatical information is not sufficient for pronominalization decisions. The simple formula: singular = cardinality I does not always hold: (3)

Drei Frauen kauften einen Pfandbrief. 'Three women bought a bond.'

Here, three women may have pooled their money to buy a single bond, in which case the follow up statement (4) is appropriate:

692

(4)

Russell

Block

Er war teuer. 'It was expensive.'

Or else three individual women bought one bond each, in which case the follow up is as in (5): (5)

Sie waren teuer. 'They were expensive.'

Evidently, in this case pronominal reference depends on the semantic feature cardinality rather than the grammatical singular/plural dichotomy. On generation, the computer has to verbalize the semantic feature cardinality, listed in the S E M M E M O R Y rather than the grammatical feature number listed in the REFOMEMORY. This is a prima facie example of how semantic features can influence grammatical form. Of course, not all semantic features influence the form of referring expressions. Neither German nor English has, for example, a special pronominal form for "edible" objects. In some other languages, however, there is a far more extensive system of semantic-grammatical correlations which would have to be taken into account.

7.3. A more complex case On generation at least, the case just considered is rather trivial. The computer knows what it wants to say and must be simply instructed to look for cardinality rather than number in making its pronominalization decision. Sometimes the decision process is not quite so straightforward. Consider the feature [ + openset], which is included in the basic feature list. Most grammar books tell us that the interrogative pronoun who is used for humans and the interrogative what is used for nonhumans. In fact, things are not quite so simple. Who/what are used for open sets, the choice depending on the value of the feature [ + human], also included in the feature set, and which is used for closed sets irrespective of the value for the feature [ ± human]. Consider the following: (6)

a. Who broke the window? (open set) b. Which of the boys the window? (closed set) c. *Who of the boys broke the window?

The problem for the system is determining whether an open or closed set is involved. Here the genitive attribute of the boys is a dead giveaway for

Referential

Expressions

693

the closed set and SEMLISTERS would have no difficulty in generating (6b) rather than (6c). The next example is somewhat more problematic: (7)

a. What is your name? (open set) b. Which is your name? (closed set)

The first sentence is the normal way of asking someone to identify him/ herself. The second, on the other hand, is appropriate for asking someone to point to his/her name in a list (closed set). Here there is no specific mention of the reference set to guide the choice between what and which. SEMLISTERS has to play the odds and opt for the more common open set unless a genitive attribute is present to indicate closed set. This is a good example of the kind of problem that humans are rarely aware of, but which causes the computer no end of headaches.

7.4. Using the T-Box SEMLISTERS utilizes intensional (T-Box) knowledge when concep-role relations are exploited in order to generate definite descriptions. To make this clear consider the following example: (8)

A: Uncle Otto bought himself a bond. B: Where did he get the money?

If the system is represented by B, it must clearly be able to determine that the noun phrase the money is definite (roughly speaking know to the speaker and hearer). In this case the noun phrase is not definite because it has been mentioned before in the dialog (which could easily be determined by checking the REFOMEMORY), but rather because money belongs to a buying act. If a buying act is mentioned in the dialog, then the money involved counts as "known" even if it has not been explicitly verbalized. Such cases can be handled with ease by consulting the T-Box. The role MEANS-OBJECT belongs to the concept BUY. A possible role filler for MEANS-OBJECT is MONEY. In general the system must see to it that instances of role fillers are marked as definite if an instance of the concept involved has been mentioned. Here an instance of BUY has been mentioned and the role filler M O N E Y counts as known and is marked as such with the definite article in English and German. (It should be mentioned in passing that not every instance of the definite article indicates that a concept is "known" (cf. Block 1988)).

694

Russell

Block

The successful application of this technique depends heavily on the modeling of concepts and roles in the T-Box and the search procedure. It is not just the roles immediately under an instantiated concept that trigger definite marking as the following example shows: (9)

A: Uncle Otto stole a bond. B: Where did he get the courage?

"Having courage" does not belong to "stealing" in quite the same way as "having money" belongs to "buying". Money is there to buy things, but any number of actions in addition to stealing demand courage. The concept "courage" is considerably more general in application and belongs higher up in the T-Box conceptual hierarchy. In principle the inheritance mechanism built into the T-Box search procedure could establish the necessary connection between "stealing" and "courage". However, important questions of modeling and limiting search procedures are involved, which we were not able to address during the term of the project.

7.5. Using the A-Box The Α-Box, which contains assertions about objects in the dialog world, is utilized in two ways for reference generation. First of all, the SEMMEMORY, which stores grammatically relevant semantic information, is part of the Α-Box. If, for instance, pronominalization is imminent and the verbalization component needs to know whether the N P to be pronominalized is animate, inanimate or propositional, it consults the appropriate listing in the SEMMEMORY. The importance of this can be seen in the following examples: (10)

(11)

(12)

Erinnern Sie sich an ihren letzten Finanzberater? (animate) Ja, ich erinnere mich an ihn/*daran. 'Do you remember your last financial advisor. Yes, I remember him'. Erinnern Sie sich an Ihre letzte Geldanlage? (inanimate) Ja, ich erinnere mich an sie/daran. 'Do you remember your last financial investment. Yes, I remember it'. 1929 habe ich mein ganzes Vermögen verloren, (proposition) Ja, ich erinnere mich *an es /dar an. 'In 1929 I lost my entire fortune. Yes, I remember'.

Referential

Expressions

695

In German animates require preposition + pronoun, inanimates may have preposition + pronoun, but are more likely to utilize the pronominal adverb construction and propositions require the pronominal adverb. In English the distributional facts are somewhat different and the rules for utilizing semantic features would have to be altered accordingly.

7.6. Using contingent knowledge So-called contingent knowledge in the Α-Box can also play an important role in generating unique descriptions of objects. Suppose that there are two objects of the same type in the dialog, e. g., two different bonds. In many instances a referential expression like the bond or this bond is sufficient to identify which object is meant. If, for instance, the system enumerates the characteristics of BOND 1 and then proceeds to a discussion of BOND 2, the user will relate such references to the last bond mentioned, BOND 2. If, however, the dialog proceeds to some other objects say, TREASURY-CERTIFICATE 1 and then returns to one of the two bonds, simple reference by means of the bond will be ambiguous and confusing. Some other identidying characteristic is needed to distinguish the two bonds already mentioned from one another. In such a case the system can use contingent knowledge about the two objects stored in the Α-Box to generate an unambiguous description. If, for instance, the two bonds have different issuers and this fact is already known to the user, the system can generate the bond issued by the Dresdner Bank ( = BOND1) as opposed to the bond issued by the Commerzbank ( = BOND 2). The process is complex. First the system has to check the R E F O M E M O R Y to see whether the simple reference would suffice. If this is not the case, it has to check contingent knowledge in the A-Box to find a distinguishing characteristic which is already part of mutual knowledge. The first check is necessary because we want the system to generate natural sounding references. That is, we would not want the system to refer to the bond issued by the Dresdner Bank every time this bond is mentioned simply because some other bond has been mentioned at some time during the dialog although no confusion would result in the given context.

696

Russell Block

7.7. T-Box —A-Box interaction As a final example of the complex interaction of knowledge sources needed for making pronominalization decisions, let us consider the following case: (13)

Das Mädchen aß ein Butterbrot, dann lief es weg. 'The girl ate a sandwich, then 'it' ran away'.

As everyone who has read Mark Twain's essay on the terrible German language knows, das Mädchen 'the girl' is neuter in German, just like das Butterbrot 'the sandwich'. A simple gender-number check would rule against pronominalization as possibly ambiguous, although native speakers do not find it ambiguous at all and would find a repetition of the full N P the girl instead of the pronoun unnatural. One way of dealing with this problem is to make use of the intensional knowledge in the T-Box. A T-Box check makes it clear that a girl, but not a sandwich, can run away, hence pronominalization can go through. This, of course, is very reminiscent of the concept of selectional restrictions in Chomsky (1965: 149 — 160). Artificial Intelligence has not reinvented the wheel, rather it is attempting to make it turn within a very complex system of interlocking gears. More interesting than the case above is one like the following: (14)

Das Mädchen tötete ein Kaninchen, dann lief es weg. 'The girl killed a rabbit, then 'it' ran away'.

The same problem arises here since das Kaninchen 'the rabbit' is, like the girl and the sandwich, neuter. This time the T-Box would lead us to believe that pronominalization would lead to ambiguity since inherently both a girl and a rabbit can run away. There is, however, no real ambiguity here since common sense tells us that a dead rabbit can't run away. For the computer this means that an assertion in the Α-Box, namely that the rabbit, having been killed, is dead) would override intensional knowledge. This introduces a further complication. The utilization of role-filler restrictions (or, if you like, selectional restrictions) in the T-Box must be subject to a check to determine whether inherent semantic characteristics have been changed by assertions made in the dialog. This is major complication for our knowledge representation system although this check is, of course, only necessary where changeable features are involved. Normally it would be superfluous to inquire whether some assertion in the dialog had altered the fact that gold is heavier than lead.

Referential

Expressions

697

8. Results and perspectives At this point, we must ask just what Artificial Intelligence has to offer to linguistics. It is certainly not the case that computer scientists have in two decades solved the problems that linguists have been wrestling with for the last two thousand years. And it would be unfair to expect them to have done so. The real advantage of the computer is perhaps that it can, in fact, do certain things much better than humans — it can keep track of all the details and interactions in a complex performance model — something no human linguist could ever hope to do. In the past, linguists have been limited to three sources of information for constructing and evaluating their theories. We can attempt to evaluate a large corpus of data, which may or may not contain the crucial evidence we need and, in any case, informs us at best about the realm of possibilities — never about what is impossible. We can question native speakers about what is possible and what is not, assuming we know what to ask. And we can rely on our own intuitions about acceptability, assuming we have the schizophrenic ability to split ourselves into experimental object and objective observer. The computer can offer us a fourth possibility. It can allow us to test our theories with merciless rigor and we can learn from its mistakes. If the computer does not react as expected, then we know there is something wrong with the theory we've used to program it. This is not an opportunity we can afford to ignore. Nevertheless, it is clear that we will have to teach the computer a great deal about linguistics before the computer begins to pay back its debt in kind. My personal estimation is that it will be a while before we can go to the bank and have a heart-to-heart talk with the computer about investing our savings. But, when that day arrives, we can be sure that the investment will have paid off a thousandfold in increased knowledge about human language and communication.

References Bergmann, Henning — Michael Fliegner — Heinz Marburger — Masimo Poesio 1987 IRS - The Interna! Representation Language. WISBER Bericht Nr. 14, Universität Hamburg. Bergmann, Henning — Michael Gerlach 1987 QUIRK: Implementierung einer Τ-Box zur Repräsentation begrifflichen Wissens. (2te erweiterte Auflage). WISBER Memo 11, Universität Hamburg.

698

Russell Block

Block, Russell 1987 1988 1989

Papers on Reference and Knowledge Representation.WISBER Bericht Nr. 20, Universität Hamburg. "Indefinite Noun Phrases and the Limits of Logic", in: Wolfgang Hoeppner (ed.), 1 0 4 - 1 1 1 . Generating Referential Expressions. WISBER Bericht Nr. 46, Universität Hamburg.

Chomsky, Noam 1957 Syntactic Structures. The Hague: Mouton. 1965 Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press. Feigenbaum, Edward — Pamela McCorduck 1984 The Fifth Generation, (revised, edition) London: Pan Books. Hoeppner, Wolfgang (ed.) 1988 Künstliche Intelligenz, G WA 1-88, Proceedings. Berlin: Springer-Verlag. Poesio, Massimo 1988 The QUARK Reference Manual. WISBER Memo Nr. 22, Universität Hamburg. Schank, Roger 1987 "What is AI, Anyway?", AI Magazine Winter 1987. Sprenger, Michael — Michael Gerlach 1988 Expectations and Prepositional Attitudes — Pragmatic Issues in WISBER. WISBER Bericht Nr. 28, Universität Hamburg.

Subject Index

Abstract sentences 136 Accented letters 19, 27 Action categories 609 — 610, 614—617, 619,622-624, 626, 628 Adjacency-condition 133 Alphabet 22. See also Writing system Roman 16 Semitic 19 Anaphors 9 1 - 9 2 , 1 1 4 - 1 1 6 , 1 2 2 - 1 2 3 , 3 7 6 - 3 7 9 , 6 3 5 - 6 3 8 , 649 Arabic 21, 27 Articulation 153 — 155 Artificial intelligence 6 8 5 - 6 8 6 , 697 Cataphores 1 1 5 - 1 1 6 , 1 2 2 - 1 2 3 , 3 6 4 - 3 6 6 , 370, 3 7 8 - 3 7 9 Chinese 1 5 - 2 8 , 51 Coherence relations 631—658 discourse coherence 530, 540 — 541, 545 Communicational situation 129, 140—142, 145 — 150. See also Spatial situational model Computational processing 21, 22, 26 — 27, 631, 634, 6 3 6 - 6 3 7 , 644, 652, 6 6 6 - 6 6 9 , 672, 675, 6 7 9 - 6 8 0 , 686, 689 Concretization 136, 140, 1 4 6 - 1 4 7 , 339 Connectives 3 7 5 - 3 7 9 , 385 Contextualization devices 49 Conversational devices 53 Converzational maxims 260, 261, 267, 2 7 0 - 2 7 2 , 578 Deixis 2 0 4 - 2 0 7 , 2 1 1 - 2 1 2 , 215, 2 1 9 - 2 2 7 , 239, 240, 245, 246, 3 7 8 - 3 7 9 , 391 Dependency hypothesis 31 — 32, 35, 60, 62. See also Independency hypothesis Diactitics 19, 27 Dictionaries 22, 26—27 Diglossia 23 Digraphs 19, 27 Discourse segmentation 635, 649 Dislocation 561 - 569

Epistemology 639 — 640 Ethiopic 18 Etymology 58 Fable structure 3 3 9 - 3 4 6 , 3 4 9 - 3 5 2 actantial roles of agents, 341, 347 — 352 motifemes 341, 343 spatio-temporal structures 341 Fictional discourse 257 — 394 Formal semantics 637 — 638, 645, 648 — 649, 653 French 27, 225 Generative grammar 335 — 353, 596—600, 609, 633, 6 8 5 - 6 9 7 Generative semantics 339 — 341, 343, 345 Genre theory 336 Genres 50, 633, 649, 6 5 1 - 6 5 2 . See also Written fiction and Story grammar commentary 651 —652 conversation 3 1 - 3 2 , 34, 5 1 - 5 4 , 5 8 - 5 9 , 6 1 - 6 2 , 232, 2 3 6 - 2 4 1 , 251, 642 drama 335 fairy tales 3 8 - 4 8 , 5 0 - 5 1 legal texts 15 novel 3 4 3 - 3 5 3 , 593 oral directions 54 — 57, 63 personal letters 35, 37, 49 — 50 scientific texts 15, 38 sonnet, 361—371. See also Poetic discourse short story 3 4 6 - 3 5 3 , 365 written correspondence 32 written directions 35, 38, 44, 4 9 - 5 1 , 57 German 22, 58, 116, 119, 124, 222, 590 Grammar 31 - 3 2 , 687, 689, 691 Graphemic structures 17 — 18, 26 Implicature 2 6 0 - 2 6 2 , 265, 2 6 7 - 2 6 9 , 271, 5 7 7 - 5 7 8 , 580, 583 Independency hypothesis 31 —32, 35, 60, 62. See also Dependency hypothesis Inserted clause 373 — 394 Interference 144, 146, 1 5 3 - 1 5 5

700

Subject Index

Interjections 51 Interpropositional relations 598 Inversion 575 — 584 Japanese 18, 20—22, 26 Language change 20, 590 Language processing 129, 136, 142, 149 Latin 25, 198 Lexical borrowing 20 Lexical semantics semantic classifier 16 — 18 Lexicon 687-688 Literacy 15, 19, 2 1 - 2 3 , 27 Literary criticism 336 — 337, 342 Literary theory 335 — 337 Medieval texts 163-168 Memory 130-134, 151-155, 236-239 short term memory 617, 619, 627 — 628 Meta-narration 341, 345, 347 Metadiscursive function 380 — 384, 390 Modality 639 Modifiers 610-611, 614-615, 628 Mohawk 179-200 Morphology 16-18, 22, 62, 649 Multisentential structures 589, 601 Naive semantics 644—647, 654 Organon model 69 Paradigm changes 202 Paradigm 581 Paratones 49. See also prosodic paragraphs Parsing 609-629 Pauses in oral discourse/speech 34 — 35, 38, 51, 5 5 - 5 6 , 57 in reading 35, 38, 41 - 4 4 , 4 9 - 5 0 , 57 in writing 3 5 - 3 7 , 41, 47, 4 9 - 5 0 , 5 7 - 6 1 Persian 549-570 Phonology 16-17, 19, 2 3 - 2 6 , 32, 3 4 - 3 5 , 38, 4 0 - 4 1 , 4 4 - 4 7 , 4 9 - 5 1 , 53-54, 5 8 - 6 0 , 147, 374, 377, 384, 549-552 Pflifting 596, 600-601 Poetic discourse 361 —372 Pragmatics 261

Presupposition 363, 364, 366, 370, 383 Principle of least linguistic effort 577 Pronominalization 593 Proper names 19—20 Proposition 130-156, 675, 677 Prosodic paragraphs 4 7 - 4 9 , 51, 53, 5 7 - 5 8 . See also Paratones Punctuation 377, 601 Quantifiers 201-202 Reaction time 130, 131, 140-142 Reading aloud processes 34, 40—41, 4 4 - 4 5 , 4 7 - 5 1 , 57, 5 9 - 6 0 Reading process 83, 223 incremental reading 105, 107, 109-111, 119 processing constraints 83, 92 processing time 110, 111 reading time 134 Russian 27 Rheme 381 Scope phenomena 589 — 603 Self-monitoring ability 16, 21—22 Semantics 32, 337, 691. See also Truth conditions semantic components 632 semantic properties 634, 647 semantic system 145 — 150 Sense relations 576 — 579 Sentence-level grammar 609, 616 Slifting 596, 602 Spanish 18, 27 Spatial situational model 132 — 135 Spatio-temporal structures 341, 349 Speech perception 34 (re)production 3 4 - 3 5 , 3 7 - 3 8 situation 204-206, 208, 210 spontaneous 33 — 34, 50 Speech act felicitous discourse 631 felicity conditions 259, 260, 579 illocution 259-260, 461, 462, 464, 675-677, 679 indirect speech act 267 locution 362-366, 461, 675, 677

Subject Index macro speech act 261—263, 269, 271 micro speech act 261—262, 269, 272 neo-Gricean pragmatic principles 576, 583 perlocution 462, 464, 465 planned/unplanned perlocutionary effects 463, 466 speech act theory 639 Standardization 24 — 25 Stimulus 130, 141, 145 Story grammar 609 — 617. See also Genres Stratificational grammar 336, 339 Subjunctive 590 Suprasentential units 589-590, 596, 602 Surface constituent 591 Surface information 131 — 150 Syntactic disambiguation 637, 645 Syntactic markers 634 Syntactic niche 593, 596, 600 Syntax 374-377 Text acceptability 632 actions 666, 675 grammar 677 knowledge 668, 672, 679, 680 linearization 666-669, 673, 675, 677, 679 processing 15-16, 19, 103, 105, 114, 221-222 reception 61 understanding 86, 105, 130-156, 316, 320-326, 493-526, 636-637, 641, 645, 652

701

Text types. See also Genres Textual surface 339—340 Thematic idea 339, 341 Theme 381 Topicalization 556 — 569 Truth-conditions 575, 577, 636-637, 642, 644, 646 - 648, 650 Turn-taking 33, 51, 53, 58 turn-taking rules 54 Verbal system 145 — 155 Vernacular varieties 25 Visual Imagery Questionaire 140 Visual information coding 18 Visual system 145 — 155 Writing processes 34, 49, 57, 5 9 - 6 0 . See also Reading aloud processes Writing system 15-28, 180 alphabetic 1 8 - 1 9 21, 27. See also Alphabet building blocks of 16 — 17 cenemic 16-17, 20, 2 4 - 2 5 economy of 17 — 19 morpho-syllabic 16 — 17 phonetic 23 pleremic 16-17,20, 2 4 - 2 5 . See also Chinese systematic properties of 15, 23, 25 transfer problems 19 Written fiction, 241-253

mi Dieter Stein The Semantics m of Syntactic Change m Aspects of the Evolution of do in English

m m

m m m m

m m m m m

1990.15.5 x 23 cm. XIV, 444 pages. Cloth. ISBN 3-11-011283-3 (Trends in Linguistics. Studies and Monographs 47) This research monograph presents a foundation for the examination of the mechanism of language change. As an example, a detailed analysis of the change in the semantics and syntax of English do is presented, showing that the Modern English syntactic pattern is the result of a complex interaction of language internal and external factors. The study is based on an analysis of a wide range of texts. A series of interrelated changes in English was located in a period of beginning mass literacy and a written standard, with the demotion of dialectal uses to stigmatized forms. The microanalysis of the internal evolution of the change shows a permanent dialectic between language internal forces and such external forces as style, prestige and other societal forces. The clustering of certain types of constructions in certain literary styles shows that the types of data obtained for a study depend critically on the choice of texts to be analyzed. The results of the study suggest a complex and heterogeneous methodology for the explanation of syntactic change, results applicable to other questions and languages.

m

m m

mouton de gruyter

Berlin · New York

m Andreas H. Jucker m Social Stylistics m imi m m m m m m m

Syntactic Variation in British Newspapers 1992. 23x15.5 cm. xxii, 297 pages. Cloth. ISBN 311012969 8 (Topics in English Linguistics 6) This research monograph presents a foundation for the study of syntactic variability by combining research tools from sociolinguistics and traditional stylistics.

While traditional stylistics has focussed on the density of specific features within a given stretch of text, it has lacked a generally accepted methodological framework and has neglected lines of inquiry for which sociolinguistics has developed methodologies, such as the differentiation of the axes of social and regional variation. These approaches are examined and synthesized to produce a comprehensive methodology that can be used for an integrated analysis of texts. As an example, the structure, the complexity and the variability of noun-phrases are analysed in a large corpus consisting of samples of eleven British daily newspapers aimed at all segments of the population. The results of this study suggest that the noun-phrase is a very sensitive style marker. Its structure and complexity correlate directly with the socioeconomic profile of the audiences addressed by the different papers.

m imi m m m m m

mouton de gruyter Berlin · New York