'Subordination' versus 'Coordination' in Sentence and Text: A Cross-linguistic Perspective (Studies in Language Companion Series)

‘Subordination’ versus ‘Coordination’ in Sentence and Text

Studies in Language Companion Series (SLCS) This series has been established as a companion series to the periodical Studies in Language.

Editors Werner Abraham University of Vienna

Michael Noonan

University of Wisconsin, Milwaukee

Editorial Board Joan Bybee

Christian Lehmann

Ulrike Claudi

Robert E. Longacre

Bernard Comrie

Brian MacWhinney

University of New Mexico University of Cologne Max Planck Institute, Leipzig University of California, Santa Barbara

William Croft

University of New Mexico

Östen Dahl

University of Stockholm

Gerrit J. Dimmendaal University of Cologne

Ekkehard König

Free University of Berlin

University of Erfurt

University of Texas, Arlington Carnegie-Mellon University

Marianne Mithun

University of California, Santa Barbara

Edith Moravcsik

University of Wisconsin, Milwaukee

Masayoshi Shibatani

Rice University and Kobe University

Russell S. Tomlin

University of Oregon

Volume 98 ‘Subordination’ versus ‘Coordination’ in Sentence and Text A cross-linguistic perspective Edited by Cathrine Fabricius-Hansen and Wiebke Ramm

‘Subordination’ versus ‘Coordination’ in Sentence and Text A cross-linguistic perspective

Edited by

Cathrine Fabricius-Hansen Wiebke Ramm University of Oslo

John Benjamins Publishing Company Amsterdam / Philadelphia

8

TM

The paper used in this publication meets the minimum requirements of American National Standard for Information Sciences – Permanence of Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data Subordination versus coordination in sentence and text : a cross-linguistic perspective / edited by Cathrine Fabricius-Hansen, Wiebke Ramm. p. cm. (Studies in Language Companion Series, issn 0165-7763 ; v. 98) Includes bibliographical references and index. 1. Grammar, Comparative and general--Subordinate constructions. 2. Grammar, Comparative and general--Coordinate constructions. 3. Grammar, Comparative and general--Clauses. 4. Grammar, Comparative and general-Sentences. I. Fabricius-Hansen, Cathrine. II. Ramm, Wiebke. P294.S83    2008 415--dc22

2007052231

isbn 978 90 272 3109 3 (Hb; alk. paper) © 2008 – John Benjamins B.V. No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any other means, without written permission from the publisher. John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa

Table of contents

1. Editor’s introduction: Subordination and coordination from different perspectives    Cathrine Fabricius-Hansen & Wiebke Ramm

1

Part I. General and theoretical issues 2. RST revisited: Disentangling nuclearity    Manfred Stede 3. Subordination and coordination in syntax, semantics, and discourse: Evidence from the study of connectives    Hardarik Blühdorn

33

59

Part II. Cross-linguistic approaches 4. A corpus-based perspective on clause linking patterns in English, French and Dutch    Christelle Cosme

89

5. Sentence splitting – and strategies to preserve discourse structure in German-Norwegian translations    Kåre Solfjeld

115

6. Upgrading of non-restrictive relative clauses in translation: A change in discourse structure?    Wiebke Ramm

135

7. Subordination in narratives and macro-structural planning: A comparative point of view   Mary Carroll, Antje Rossdeutscher, Monique Lambert & Christiane von Stutterheim

161

 ‘Subordination’ verses. ‘Coordination’ in sentence and text

Part III. Monolingual studies 8. German dependent clauses from a constraint-based perspective    Anke Holler

187

9. To the right of the clause: Right dislocation vs. afterthought    Maria Averintseva-Klisch

217

10. Exploring the role of clause subordination in discourse structure: The Case of French avant que    Laurence Delort

241

11. Pseudo-imperatives and other cases of conditional conjunction and conjunctive disjunction    Michael Franke

255

12. From discourse to “odd coordinations”: On Asymmetric Coordination and Subject Gaps in German    Ingo Reich

281

Part IV. Diachronic perspectives 13. Old Indic clauses between subordination and coordination    Rosemarie Lühr

307

14. Rhetorical relations and verb placement in the early Germanic languages: A cross-linguistic study    Svetlana Petrova & Michael Solf

329

Index of subjects

353

Index of names

357

Editors’ introduction Subordination and coordination from different ­perspectives Cathrine Fabricius-Hansen & Wiebke Ramm University of Oslo

1.  Preliminaries The present collection of papers addresses semantic, pragmatic or discourse-oriented aspects of coordination and subordination in a broad sense. Five of the papers approach the topics from a cross-linguistic perspective. The collection is the outcome of a workshop (AG8) on ‘Subordination’ versus ‘Koordination’ in Satz und Text aus sprachverglei­ chender Perspektive / ‘Subordination’ versus ‘Coordination’ in Sentence and Text – from a Cross-linguistic Perspective, which was organised by the editors during the 28th Annual Meeting of the Deutsche Gesellschaft für Sprachwissenschaft in February 2006.1 Coordination and subordination are well-established grammatical terms, but like many other linguistic terms with a long history, they are somewhat fuzzy, both being used in a variety of – mutually related – senses depending on the theoretical context. This holds for other, closely related notions such as parataxis and hypotaxis as well. Thus Lehmann (1988) observes in a paper on the typology of clause linkage: The term subordination is applied, in different schools of linguistics, to different kinds of phenomena. In the broadest use, which may be found in certain trends of ­European structuralism, the size and nature of the subordinate element is of no concern. Here subordination practically means the same as dependency.

.  The research underlying this introduction has been carried out in connection with the project SPRIK (Språk i kontrast / Languages in Contrast) at the University of Oslo, Faculty of Humanities, funded by the Norwegian Research Council (NFR) under project number 158447/530 (2003–2007). Editing was financed by the SPRIK project and by grants from the Department of Literature, Area Studies and European Languages and the Faculty of Humanities. We thank Maria F. Krave and Signe Oksefjell Ebeling for their meticulous work with the manuscripts. And we thank the editors of the Studies in Language Companion Series for giving us and the other workshop participants this opportunity to publish our research presented at the DGfS 06 conference.



Cathrine Fabricius-Hansen & Wiebke Ramm

In the most narrow use, characteristic of classical philology, only finite clauses can be said to be subordinate. Here subordination practically means the same as hypotaxis; and consequently the two latter terms are mostly used interchangeably.  (Lehmann 1988: 221)

In what is probably their most widespread application, ‘subordination’ and ‘coordination’ – along with their adjectival cognates ‘subordinate’, ‘coordinate’, etc. – are syntactic notions denoting relations between parts of a complex syntactic unit. That is, they concern the structure of sentences or clauses and their parts. Subordination is an asymmetric relation – both in linguistics and in everyday life – and by that token intimately correlated with the notion of hierarchical structure: If A is subordinate to B, then B cannot be subordinate to A; but B may, in its turn, be subordinate to a third entity C, and so on. This means that B has a ‘higher’ position in the domain structured by subordination, i.e., it is nearer to the ‘topmost’ element of that structure, which is conventionally identified with an element that is not subordinate to any other element within the domain. In social hierarchies, this kind of asymmetry is typically correlated or associated with (social) ‘importance’, ‘prominence’ and the like. Similarly, the subordinate clause in clausal subordination is often said to contain less prominent or salient information, to have less communicative weight (‘kommunikatives Gewicht’ in German) than the ‘superordinate’ clause (typically the so-called main clause) (see e.g., Aarts 2006; Hartmann 1984; Hetland & Molnár 2001; Peyer 1997; Reis 1993). However, while importance and prominence may be quite transparent concepts with respect to social hierarchies, it is far from clear how the functional notions of prominence, salience or communicative weight can be mapped onto specific structural-syntactic categories as defined by the grammars of different languages. As far as the domain of natural discourse and texts is concerned, it is a common observation in various theoretical approaches that entities of this domain too can be organised hierarchically (‘subordinating’, ‘hypotactically’) or non-hierarchically (‘coordinating’, ‘paratactically’). Relevant examples are Rhetorical Structure Theory (Mann & Thompson 1988) and Segmented Discourse Representation Theory (Asher 1993; Asher & Lascarides 2003). However, relatively few attempts have been made so far to answer the following questions (but see references in Matthiessen & Thompson 1988: 275): i. What kind of relationship holds between the discourse-related and the sentence-re­ lated (syntactic) notions of subordination and coordination or related distinctions? Is the parallelism as close as the terminology may suggest? What impact does the choice between syntactic subordination and coordination have on discourse structure? ii. By what means are hierarchical (‘subordinating’) and non-hierarchical (‘coordinating’) discourse relations signalled across languages? To what degree, and how, do languages differ in this respect? Do languages exploit syntactic subordination (and coordination) differently as a means of organizing information in discourse – ‘information packaging’ – and if so, why and in what ways?

Editors’ introduction

The phrases in sentence and text and a cross-linguistic perspective occurring in the title of the present volume relate to these two sets of questions. An additional important issue concerns methodology: iii. Since the questions raised above are partly of an empirical nature, extensive studies of natural texts – ‘wildlife’ texts, so to speak2 – are needed, including studies of parallel corpora (of original texts and their translations) and comparable texts in different languages: Explorative analyses indicate that text- and corpus-oriented crosslinguistic studies can make an important contribution to the issues specified in (i) and (ii) and thus improve our general understanding of how information packaging on sentence and text level are related (see Doherty 1996, 1999, 2005 and 2006; Granger et al. 2003; Hasselgård et al. 2002; Johansson 2007; Johansson & Oksefjell 1998). These issues were the motivation for our workshop and the articles collected in this volume. We invited contributions based on parallel corpora and/or language comparison, including, but not limited to, topics such as the following: – Syntactically adjoined structures from the perspective of information structuring on discourse level – ‘Subordinating’ versus ‘coordinating’ discourse relations / clause combining, and their realisation – Connectives and punctuation as a means of structuring discourse and signalling discourse relations Apart from punctuation, these topics are all addressed more or less directly in the papers below. The collection is divided into four parts: The two contributions in Part I are concerned with central theoretical questions. Part II contains four corpus-based studies of clause combining and discourse structure, comparing at least two of the languages German, English, French, Dutch, and Norwegian. The five papers in Part III address specific – predominantly semantic – topics relating to German, English or French. Finally, the two papers in Part IV approach the topic of subordination, coordination and rhetorical relations from a diachronic perspective, involving Old Indic and early Germanic languages, respectively. Sections 2–4 of this introduction address some of the general issues raised above more thoroughly, referring to individual papers where relevant. The final section (5) gives a brief presentation of each contribution, outlining how it fits into the overall topic of study. First, however, we need to comment on our distinction between ‘sentence’ and ‘text’. .  We have coined the term ‘wild(life) texts’ inspired by the Norwegian newspaper Aftenposten, which after a very mild winter (2005) introduced the compound vill-snø ‘wild-snow’ referring to natural snow.





Cathrine Fabricius-Hansen & Wiebke Ramm

2.  ‘Sentence’ versus ‘text’ What ‘is’ a sentence? As is well known, this question cannot be answered in a theoryindependent and at the same time precise manner. From a syntactic point of view the sentence is the domain of syntactic theory: a unit that is governed by syntactic principles (constituent structure, dependency, government, binding, movement, etc.). From a discourse perspective a sentence might be a unit representing a single illocutionary act on the part of the speaker/reader and intended to be understood as such by the hearer/reader. Thus, it is a unit the reader/hearer is expected to process ‘in a single step’, assigning to it a communicative purpose and a content to be integrated into her/his representation of the current discourse, i.e., what corresponds to an utterance in spoken language. As mentioned in Section 1, we shall primarily be concerned with written texts. And for the present purpose we shall assume without further discussion that sentence boundaries in the discourse-oriented sense are marked by full stop in written discourse – with question and exclamation marks as non-assertive alternatives. That is, we take the orthographic sentence as defined e.g., by Huddleston & Pullum (2002: 1728) to be the basic processing unit in written texts, and full stop to be the written default counterpart of declarative terminal fall in intonation, signalling to the reader/ hearer that the end of the current communicative unit has been reached and that the common ground can be updated with the content expressed by that unit in the given context (Büring 2007). Comma may signal the end of a syntactic sentence (clause) but cannot by itself mark a sentence boundary in the discourse-oriented sense. For the rest of this introduction we shall use the term ‘sentence’ in that sense unless otherwise indicated, assuming that discourse-sentences by default manifest themselves as orthographic sentences in written language. The terminological decisions made above have some important implications: i. A text consists of a finite number of sentences in succession, with one-sentence texts as a marked category. ii. A sequence of syntactically independent clauses separated by commas is not a sentence sequence but constitutes a single (complex3) sentence if properly demarcated.4 iii. If clause combining (Haiman & Thompson 1988), or  clause linkage (Lehmann 1988), is to be understood as connecting clauses rather than sentences (which in our

.  Sentences having the form of two or more coordinated main clauses are sometimes termed compound rather than complex. .  As Huddleston & Pullum (2002: 1728) put it: “[…] there will often be no syntactically marked distinction between a sentence with the form of a combination of two successive main clauses and a sequence of sentences each of which has the form of a main clause. In writing, one function of punctuation is precisely to indicate whether successive clauses belong together or are to be treated as separate.”

Editors’ introduction

opinion is reasonable) it must be confined to the sentence level. That is, simple juxtaposition of syntactically independent clauses separated by comma, without overt coordination, represents a special case of (paratactic) clause linkage. Corresponding full-stop sentences, on the other hand, are, strictly speaking, not related by any kind of clause combing; they simply succeed each other.5 According to Huddleston & Pullum (2002: 1735f) colon and semicolon, like comma, “normally mark boundaries within a sentence, and hence can be regarded as secondary boundary marks” located between comma and full stop in a hierarchy of “relative strength”. To our knowledge, however, the discourse functions of semicolon and colon have not yet been thoroughly investigated from the perspective of discourse structure or discourse processing, let alone in a cross-linguistic setting (a detailed account of the use and functions of the colon in German, however, can be found in Karhiaho 2003). So, since the topic of punctuation is not addressed in the present collection of papers, we shall refrain from further speculation on this matter.

3.  Coordination and subordination as syntactic notions 3.1  Coordination The literature on the syntax, semantics and pragmatics of coordination – clausal or not – is extensive (see for instance Bátori et al. 1975; Carston 2002; Carston & Blakemore 2005; Haspelmath 2004; Pasch et al. 2003; Redder forthcoming, Schwabe & Zhang 2000 for recent overviews and discussions), so we shall confine ourselves to issues of immediate interest in the present context. As a syntactic relation, coordination (or conjunction, as it is often called6) is traditionally said to hold if the units in question are syntactically ‘equivalent’, ‘have the same status’, ‘play the same role’ in the given syntactic context (see e.g., Lang 1984; Lehmann 1988; Zifonun et al. 1997: 2362ff; Huddleston & Pullum 2002: 1281ff; Haspelmath 2004 with further references; Carston & Blakemore 2005: 353 9; Crysman 2006). Haspelmath (2004), for instance, proposes the definition in (a) below, which he later refines as in (b), leaving out the reference to relative salience (compare Sections 1 above and 4.2 below), on the one hand, and stressing the semantic nature of ‘sameness’ of type, on the other (cf. also Reich forthcoming). This in its turn distinguishes Haspelmath’s second proposal from Lehmann’s definition in (c).

.  The literature on clause combining (e.g., Lehmann 1988; Matthiessen & Thompson 1988) is not very explicit in this respect. .  But see e.g., Cormack & Smith (2005: 396), who distinguish notionally between (logical) conjunction and (syntactic) coordination.





Cathrine Fabricius-Hansen & Wiebke Ramm

a. A construction [A B] is considered coordinate if the two parts A and B have the same status (in some sense that needs to be specified further), whereas it is not coordinate if it is asymmetrical and one of the parts is clearly more salient or important, while the other is in some sense subordinate. (Haspelmath 2004: 3f.) b. The term coordination refers to syntactic constructions in which two or more units of the same type are combined into a larger unit and still have the same semantic relation with other surrounding elements. (Haspelmath 2004: 34) c. Coordination is a relation of sociation [i.e., non-dependency] combining two syntagms of the same type and forming a syntagm which is again of the same type. (Lehmann 1988: 182) In his pioneering work on the semantics of coordination, Lang (1984) explicitly distinguishes between genuine “coordinate conjoining” and “subordinate conjoining”, stressing that the conjuncts have to fulfil both the grammatical condition of “homogeneousness” and semantic conditions concerning the relation between conjunct meanings in order for the construction to qualify as (acceptable) “coordinate conjoining”. The coordinated entities (the conjuncts, coordinates, coordinands, conjoins) may be clauses in a broad sense, i.e., finite or non-finite verb phrases or full clauses, or they may be lower-level categories (phrasal coordination). In this book, we shall predominantly be concerned with the former variety, i.e., clausal coordination (including verb phrase coordination). Coordination, as traditionally understood, is signalled by a coordinating connective (‘conjunction’, e.g., and, or, but), which in the languages studied here introduces non-initial constituents of coordinating constructions;7 that is, coordination represents a syndetic variety of paratactic clause combining (Lehmann 1988). The archetypical coordinating conjunction is and together with its counterparts in other languages – the general terms ‘coordination’ and ‘conjunction’ are, in fact, often used in the restricted default sense of and-coordination/-conjunction. And this is how they will be used in this introduction, as in most of the individual contributions, if nothing is said to the contrary (see Section 3.3). Coordination is a symmetric relation in the formal sense: if A is coordinated to B, then B is also coordinated to A. However, in the literature on coordination the term ‘symmetric’ often seems to be used in a less formal sense, referring to the criterion that the coordinated entities must be of the same type and must have the same relation with surrounding elements. In yet other contexts symmetry is related to the permutativity property of logical conjunction (&) and other operators (intersection and summation) that are taken to represent the meaning of and, depending on the semantic nature of

.  See Haspelmath (2004) for a typological overview. In multiple coordination, comma is a semantically neutral alternative to and and or between all but the last two conjuncts.

Editors’ introduction

the conjuncts (Partee & Rooth 1983). However, the symmetry condition, in one sense or the other, is not always met by constructions that are yet classified as coordinative according to diagnostic coordination tests, as summarized e.g., by Lang (1984), te Velde (2000), Haspelmath (2004: 34f), and Reich (forthcoming).8 In other words, the concept of (syntactic) coordination – like subordination (see 3.2) and many other traditional grammatical notions – has prototype structure: In addition to ‘canonical’ or ‘standard’ coordination, which is symmetric by definition, the literature has registered an impressive variety of asymmetric construction types.9 Opinions vary somewhat as to where exactly one should draw the borderline between coordination (parataxis) and subordination (hypotaxis) in the continuum represented by non-canonical varieties of these categories, i.e., and-constructions exhibiting crucial subordinate properties (“subordinate conjoining” (Lang 1984) or “pseudo-coordination” (e.g., Johannessen 1998)), on the one hand, and coordination-like clause combining formally marked as subordinating (see Section 3.2), on the other hand. At the other end of the continum we find what Zifonun et al. (1997: 2362) call Quasikoordination: “adjacent conjuncts” that do not form a single orthographic sentence or a single “intonation unit”, i.e., cases where a clause containing a coordinating connective follows a major boundary mark in written or spoken language (e.g., full stop and falling boundary tone, respectively). So-called (and-) parenthetical coordination (Blakemore 2005; see also Zifonun et al. 1997: 2363f) represents yet another non-canonical variety of coordination that we cannot take up here. Asymmetry in coordination has been an object of much theoretical discussion, and, as far as clausal coordination is concerned, certain types of asymmetry, in fact, occur so frequently that what is called canonical coordination may seem an abstract ideal rather than a ‘real’ prototype. Thus it has been repeatedly observed that clausal coordination in practice often deviates from standard logical conjunction by being sensitive to permutation of the conjuncts, i.e., that the interpretation may change with the order of the conjuncts, even if overt anaphoric dependences between the second and the first conjunct are adjusted; cf. (1)–(2), taken from Levinson (2000: 121) and Blakemore & Carston (2005: 570), respectively. (1) John turned the key and the engine started. (≠ The engine started and John turned the key.) (2) She [[jumped on the horse] and [rode into the sunset]]. (≠ She rode into the sunset and jumped on the horse.)

.  Coordinate Structure Constraint (CSC), Across the Board Movement (ATB), No Backward Anaphora, the possibility of multiple conjuncts. .  “Unbalanced” coordination according to Johannessen (1998).





Cathrine Fabricius-Hansen & Wiebke Ramm

Here the conjuncts are ‘equal’ at the syntactic level, they are full clauses of the same ‘type’ in (1) and verb phrases in (2); that is, the asymmetry concerns the level of interpretation alone. This phenomenon of conjunction buttressing (Levinson 2000) can be explained as pragmatic enrichment based on the principle of informativeness ­(I-principle) which allows the hearer/reader to choose the strongest interpretation coherent with what is explicitly said: “[…] conjunction is ripe for I-enrichment: when events are conjoined, they tend to be read as temporally successive and, if at all plausible, as causally connected.” (Levinson 2000: 122) Asymmetries of this semantic-pragmatic kind have played a prominent role in Relevance Theory as an argument for the central distinction between what is said and what is implicated (see Blakemore & Carston 2005; Carston 2002: 222ff). In his contribution to the present volume, Solfjeld (Chapter 5) exploits certain aspects of the relevance-theoretic approach to clausal coordination to account for a characteristic translation strategy in Norwegian translations from German. Other asymmetries in clausal coordination overtly concern morpho-syntactic aspects of the initial and non-initial conjuncts: asymmetries with respect to syntactic category (full clause vs. e.g., verb phrase), type of clause (e.g., imperative vs. declarative, verb-end vs. verb- second or verb-first in German), choice of verb form, extraction and binding phenomena, gapping, etc. (see e.g., Johannessen 1998, te Velde 2000, Huddleston & Pullum 2002: 1323ff., Reich to appear). Some examples are seen in (3) – (5): (3) a. Give me your address, and I’ll see what I can do. b. Give me your money, or I’ll shoot. (4)

Gestern [hat er wieder verschlafen und ist zwei Yesterday has he again overslept and is two Stunden zu spät gekommen]. hours too late come. ‘Yesterday he overslept again and arrived two hours late.’

(5) Here is the whisky which I [went to the store and bought].

In (3) an imperative and a declarative clause are coordinated, and in the German example (4) the second conjunct has a subject gap which must be interpreted as co-referential with – or bound by – the overt subject of the first conjunct. The fronted temporal adverbial gestern ‘yesterday’ can, from a semantic perspective, be analysed as ‘common’ to both conjuncts, i.e., as standing outside the coordinate construction.10 (5) is a ‘classical’ example (from J. R. Ross 1967) of asymmetric extraction leaving an object gap in the .  It should be noted that when we front the subject instead of the adverbial, we get what ­appears to be a syntactically canonical VP-coordination: (4)’

Er [hat gestern verschlafen und ist zwei Stunden zu spät gekommen]. he has yesterday overslept and is two hours too late arrived ‘He overslept yesterday and arrived two hours late.’

Editors’ introduction

second conjunct alone and thus violating the so-called Coordinate Structure Constraint (see e.g., the discussion in Johannessen 1998: 214ff; te Velde 2000; Kehler 2002; Crysman 2006 and Reich forthcoming, with relevant references). Asymmetries in clausal coordination are the subject of Franke’s and Reich’s contributions (Chapters 11 and 12). To conclude our discussion of asymmetries in clausal coordination, the following points should be made: i.

On the surface, clauses (or sentences) may be represented as rather simple syntactic structures, depending on syntactic theory. However, from a semantic perspective, they must be conceived as multi-level or multi-layered structures: a full sentence contains a series of nested operators or modifiers (‘functional projections’) – polarity, aspect, tense, modality, illocutionary force – above the predicate expressed by the lexical (main) verb and head of the clause. In the case of clausal coordination, then, it has to be decided at which semantic level the second conjunct attaches to the first – and whether it does in fact constitute a full independent clause from a semantic point of view or rather a semantically reduced clausal variant, i.e., a more or less ‘desententialized’ entity in the sense of Lehmann (1988: 200). For this reason the difference between full clausal and verb phrase coordination should probably be taken more seriously than has hitherto been the case. ii. It is by now generally acknowledged that semantic interpretation is more flexible and that pragmatics plays a more prominent and pervasive role in interpretation than was envisaged some decades ago (Sperber & Wilson 1986, Levinson 2000, Carston 2002, Kehler 2002, Zeevat forthcoming). Against that background it is conceivable that a semantically adequate conjunctive interpretation may be extracted from what seems somewhat ‘odd’ on the surface. At some level the conjuncts must be interpreted as semantic objects of the same type if the meaning of the coordinator is as presumed; otherwise the interpretation would break down.11 iii. Since the first conjunct necessarily precedes and is not structurally subordinated to or embedded in the second conjunct, anaphoric elements occurring in the second conjunct – including zero anaphors – may find their antecedents in the first conjunct. More generally stated: the first conjunct is an accessible context when

But even so, i.e., when placed within the first conjunct, the temporal adverbial gestern ‘yesterday’ is interpreted as setting a temporal frame for both conjuncts. Thus asymmetries involving subject gaps – type (4) ­– might after all be special instances of VP-coordination, despite the counter arguments put forward by Reich (fortcoming). .  “To sum up, extraction out of parts of conjuncts is possible when a link can be established between the conjuncts which can be described in terms of consequence, despiteness, or time (there may be other relationships, too)” (Johannessen 1998: 233). Cf. Kehler (2002, 2004), who argues for a coherence-driven, i.e., pragmatic, approach to asymmetries in coordination. See also e.g., Vuillaume’s (2000) account of asymmetric coordination in German.



 Cathrine Fabricius-Hansen & Wiebke Ramm

the second is processed, but not the other way round (“No Backward Anaphoricity”; cf. Lang 1984; te Velde 2000; Haspelmath 2004 and Reich forthcoming, among others). Consequently, permutativity between conjuncts is to be expected only in the – presumably quite rare cases – where the second conjunct is not linked, explicitly or implicitly, to the first in any other way than by being conjoined with it. iv. As argued e.g., by Lang (1977, 1984), Carston (2002), Blakemore & Carston (2005) and Jasinskaja (2006), coordination itself can be taken to signal that the conjoined clauses are more tightly linked than a corresponding sentence sequence. The coordinating connective has a “procedural” or “operational” meaning triggering the construction of a what Lang (1984: 71) terms a Common Integrators, i.e., “a conceptual entity which encompasses the conjunct-meanings”, on the basis of the conjunct meanings themselves, the linguistic and situational contexts and extralinguistic knowledge systems etc. That is, the coordinator following the first conjunct makes clear that the sentence is not yet finished but that more ‘of the same kind’ as the structure processed thus far will have to be integrated in the interpretation (cf. Redder forthcoming with further references). (i) – (iv) go a long way towards explaining why syntactic-semantic mismatches may occur in natural language coordination and why they are interpreted as they are. But (i) – (iv) cannot, of course, account for language-specific restrictions on syntacticsemantic asymmetry. For the reasons mentioned above, canonical symmetric coordination is probably the exception rather than the norm in language use. (6) and (7) are cases in point.

(6) Tom did his homework and Anna went to the cinema.



(7) I had a paper to finish and my parents needed help.

Here the conjuncts can hardly be understood as connected in any other way than as partial – and jointly exhaustive (Jasinskaja 2006) – answers to a given ‘Question under Discussion’12 like What did the children/ Tom and Anna do last night? for (6) and Why didn’t you join us? for (7). That is, unlike the asymmetric examples discussed above, the coordinate constructions in (6) and (7) cannot be interpreted as self-contained coherent texts. The conjuncts are not linked at the level of eventualities (causally or otherwise); the relation between them is of a purely rhetorical nature. In symmetric coordination, the order of the conjuncts has no bearing on truth conditions. However, this is not to say that the order is arbitrary or irrelevant in general. There may be other reasons for the speaker/writer to prefer one order over the other, e.g., relative importance or salience from a psychological/emotional – or rhetorical – point of view: Only the second, or last, conjunct or the whole coordinate sentence

.  Or Common Integrator in the sense of Lang (1984).

Editors’ introduction

may function as an attachment point for a subsequent sentence; the first conjunct is blocked (see Section 4.2). Like sentence interpretation in general, the interpretation of coordinate constructions interacts with information structure at sentence level (focus-background and topic-comment partition, cf. Krifka (2007)). In spoken language prosody and intonation play a prominent role in this respect (see Lang (2004) for coordination). To our knowledge, however, coordination, in written language at least, has not been investigated specifically from that perspective, except for symmetric coordination exhibiting the characteristic pattern of parallel and contrast seen in (6) (see e.g., Lang 2004, Abraham 2006, Büring 2007). The interplay between information structure / prosody and subordination/coordination in a historical language (Old Indic) is central in the contribution by Lühr (Chapter 13).

3.2  Subordination As pointed out in Section 1, the term subordination – like coordination – may be used in a general sense, as defined e.g., by Lehmann (1988): (a) A grammatical relation R connecting syntagms X and Y is a relation of depen­ dency iff X occupies a grammatical slot of Y or vice versa. In a dependency relation, Y depends on X iff X determines the grammatical category of the complex and thus its external relations. […] Embedding is the dependency of a subordinate syntagm. (Lehmann 1988: 182)

But often – as will be the case here – ‘subordination’ is applied in the restricted sense of clausal subordination, as the asymmetric counterpart of clausal coordination (cf. te Velde 2000): A clause β is subordinate to another clause β if the former is syntactically dependent on the latter (in a sense to be further specified) but not vice versa. In traditional structural terms dependency can be conceived as structural (and functional) embedding: the subordinate clause β is embedded into the superordinate or matrix clause β with a specific syntactic function (complement, adverbial, attribute, etc.) at some level of that clause. Other definitions of clausal subordination rest more directly on the functional properties of the subordinate clause (e.g., Longacre 1985; Matthiessen & Thompson 1988; Cristofaro 2003). Cristofaro (2003) presents a comprehensive typological study of subordination in 80 different languages based on what she terms the Asymmetry Approach. She defines subordination in terms of the cognitive-linguistic notion of ‘profiling’ (Langacker 1991), explicitly equating (non-)profiling with (non-)assertion:13

.  According to Langacker (1991: 436) “[…] a subordinate clause can also be understood more broadly as one that is not a main clause. Is there, then, something that all such clauses have in common that sets them apart from main clauses? I suggest that there is, and that the distinction rests on profiling. Specifically, the main clause is the head of at a particular level of organization,



 Cathrine Fabricius-Hansen & Wiebke Ramm

(b) We are now in a position to propose a functionally based definition of subordination, resting on cross-linguistically applicable and consistent criteria. By subordination will be meant a situation whereby a cognitive asymmetry is ­established between linked SoAs [States of Affairs], such that the profile of one of the two (henceforth, the main SoA) overrides that of the other (henceforth, the dependent SoA). This is equivalent to saying that the dependent SoA is (pragmatically) non-asserted, while the main one is (pragmatically) asserted. (Cristofaro 2003: 33)

In practice Cristofaro’s investigation seems to support approaches that consider subordination to be a multidimensional phenomenon (Lehmann 1988, Verstraete 2007). The definition in (c) below represents a third variant, based primarily on the notion of prominence (cf. Section 1): (c) In a general sense, if an element α is subordinate to an element β, it is less prominent than β and usually α is dependent on β. […] It is a defining characteristic of subordination (also called hypotaxis) that the subordinate element is syntactically at a lower level in the overall structure than the element or string it is subordinate to.  (Aarts 2006: 249)

As is well known, subordinate clauses in the canonical sense of complement, (restrictive) adverbial and (restrictive) relative clauses tend to exhibit specific formal characteristics: constraints on verb form or word order (e.g., verb-final in German), the presence of specific subordination markers – complementizers, subordinating connectives (subordinating conjunctions, subordinators, subjunctions), relativizers – etc.14

(8) She doubted that she would gain access.



(9) If you can’t beat them, join them!

(10) Keiner, der es gesehen hat, wird es je vergessen. nobody rel.pron it seen has will it ever forget ‘Nobody who has seen it will ever forget it’

i.e., the clause that lends it profiling to the composite structure of a multiclausal expression. A subordinate clause is then describable as one whose profile is overridden by that of a main clause.” .  It should be noted that from a structural point of view the entity consisting of subjunction + clause should not, strictly speaking, be considered a clause but a phrase of a different category, headed by the subjunction, i.e., a ‘subjunction phrase’ or the like (cf. Pasch et al. 2003). Similarly for coordination: the second conjunct including the conjunction itself by all relevant phrase structure criteria constitutes a ‘conjunction phrase’ (ConjP) rather than a clause (see in particular Johannessen 1998). Viewed as devices of so-called clause combining, conjunctions and subjunctions are also not considered parts of the clauses they introduce. In this introduction, however, we adhere to the traditional, albeit sloppy use of the term ‘subordinate/coordinate clause’.

Editors’ introduction

Hence the term ‘subordinate’ is sometimes also used with reference to clauses having the relevant formal properties without being syntactically embedded in the strict sense suggested above, i.e., without apparently having the function that is typically associated with those properties. So-called ‘supplementary’ (Huddleston & Pullum 2002: 1356) or ‘sentential’ relative clauses and resultative clauses introduced by so that are cases in point; cf. (11) and (12) from Huddleston & Pullum (2002: 1356, 969). 15 (11) We called in to see Sue’s parents, which made us rather late. (12) Most primary teachers are women so that suitable ‘role’ models, to use the trendy phrase, are more abundant for girls than for boys.

Clauses having subordinate form may even be used as independent sentences with a marked (exclamatory) illocutionary function, and conversely: clauses may be structurally subordinated (embedded) without being formally marked as subordinate (see e.g., Fabricius-Hansen 1992, Reis 1999). Such mismatches between form and function have given rise to much discussion – and terminological confusion – in particular, it seems, with respect to German (see e.g., Fabricius-Hansen 1992 and forthcoming; Holler 2005; Lefèvre 2000; Pasch et al. 2003; Peyer 1997; Reis 1997; Zifonun et al. 1997: 2233ff). They are, however, clear indications of the multidimensionality of the notion of clausal subordination. This is the issue addressed by Holler (Chapter 8) within the framework of Head Driven Phrase Structure Grammar (HPSG). In what follows, we shall use the term ‘subordinate clause’ to refer to clauses that are formally characterized as such, irrespective of whether they are properly embedded or not. From a semantic point of view complement clauses, being arguments of predicates, represent a very different case from adverbial and relative clauses; since none of the papers in this volume is concerned with this category we shall not dwell on it here (see e.g., Noonan 1985 for details). Adverbial and relative subordinate clause types are often subsumed under the notion of (syntactic) adjuncts, together with e.g., non-clausal adverbials and attributive adjectival phrases (Austin et al. 2004; Lang et al. 2003). Semantically, such adjuncts are – grosso modo – said to modify clauses and noun phrases (including determiner phrases), respectively. As for adverbial clauses, [t]his modification can occur at various levels (such as verb phrase, tense phrase, mood phrase) and in various dimensions (such as times and worlds). These variations give rise to a categorization of adverbial clauses (temporal, modal, …) and a subcategorization according to a range of relations within these dimensions, depending on the subjunction. (Sæbø forthcoming)

Modification often amounts to restricting the denotation of the modified constituent. Thus the noun phrase student that passes the exam denotes a proper subset of the set

.  See Zifonun et al. (1997) for a detailed account of such mismatches between function and form.



 Cathrine Fabricius-Hansen & Wiebke Ramm

denoted by the head noun (phase) student, i.e., the property of being both a student and passing the exam – but with priority assigned to the head noun property (Bierwisch 2003). Similarly, the temporal clause in (13) restricts the temporal frame denoted by the future tense of the main clause. (13) I’ll leave when you arrive.

This means that adjunction, including clausal adjunction, may involve the same type of semantic operation as coordination, viz. logical conjunction and related operations (see Section 3.1), but ‘asymmetrically’ restricted to the domain of the modified entity, i.e., semantically ‘downgraded’. In addition to restrictive relative clauses we find non-restrictive (parenthetical, supplementary, appositive) relative clauses as exemplified in (11) above and (14): (14) Dr. Brown, who lives next door, comes from Australia. (Quirk et al. 1985: 290)

However, while there seems to be general consensus as to the syntactic-semantic analysis of the canonical restrictive relative clause this, to the best of our knowledge, does not hold for the non-restrictive varieties. It is not quite clear what their syntactic status is, e.g., whether they should be considered adjuncts to the noun (or determiner) phrase or to the matrix clause itself, nor is it clear how their semantic contribution should be accounted for compositionally. For adverbial clauses (in a broad ragbag sense) we can similarly distinguish between, on the one hand, canonical varieties that are properly integrated into their matrix clause as modifiers at various levels (temporal, conditional, causal, illocutionary, …; cf. the quotation above), on a par with ordinary non-clausal adverbials, and, on the other hand, ‘non-restrictive’ varieties having a more supplementary relation to the subordinate clause, like (12) above and (15): (15) White motorists make up 78 per cent of Maryland highway traffic, while black drivers account for about 17 per cent.

Integration as a dimension structuring the subordination-coordination (parataxis) continuum is discussed in detail in the paper by Holler (Chapter 7); see also Lehmann (1988). However, subordinate clauses that are not properly integrated into their matrix clause, e.g., as a temporal or causal modifier, are ‘less subordinate’ (hypotactic) and ‘more coordinate’ (paratactic) than canonical restrictive varieties. This makes this type of clause linkage particularly interesting in the present context: In order to understand the rationale behind such ‘intermediate’ patterns of clause combining we have to transgress the sentence level and investigate their use, i.e., how they function in natural discourse (see Section 4.2) and across languages. This challenge is taken up by Ramm (Chapter 6). Being non-integrated or somehow detached from the rest of the sentence is a property that certain subordinate clause types share with e.g., (clausal and non-clausal) parenthetical constructions (see Dehé & Kavalova (2007) for a recent collection of articles on parentheticals), appositions and various ‘dislocated’ elements at the (left or right)

Editors’ introduction

sentence periphery. The discourse-structural functions of right dislocation are the topic of Averintseva-Klisch’s contribution (Chapter 9). In restricting our use of the term subordination to clausal subordination we deviate terminologically from Lehmann (1988), who uses hypotaxis and parataxis for the clausal varieties of coordination and subordination in the more general sense. In Matthiessen & Thompson (1988), on the other hand, ‘hypotaxis’ and ‘hypotactic clause (combining)’ refer to a subclass of what is traditionally subsumed under subordination, viz. subordinate clauses that are not embedded (or integrated). Matthiessen & Thompson even suggest that there is no advantage to postulating a grammatical category of ‘subordinate’ clause; rather the grammar of English at least, and perhaps of other grammars as well, suggests that a distinction between what we have been calling ‘hypotaxis’ and embedding is crucial. (Matthiessen & Thompson 1988: 317)

It is, however, not quite clear how embedding is defined in their (systemic-functional) theoretical framework. In practice, the notion seems to subsume complement clauses and clauses functioning as restricting modifiers in noun phrases. This leaves us with hypotaxis covering the heterogeneous group of adverbial clauses in the broadest traditional sense, whether semantically restrictive or not, as well as non-restrictive or ‘appositive’ noun-modifying clauses. It is far from evident that hypotaxis, thus understood, is a natural category. So we prefer equating hypotaxis etc. with subordination etc. in the general sense; parataxis, then, could be a cover term for explicitly marked coordination, as it is understood here (Section 3.1), and mere juxtaposition of mutually independent clauses. But on the whole, we refrain from using these terms. It should be noted, however, that terminology varies somewhat in the present volume as a whole. Thus coordination in the paper by Cosme (Chapter 4) is used in the broad sense of Lehmann (1988), covering syndetic paratactic clause combining (with and etc.) as well as asyndetic juxtaposition of independent clauses (separated by comma) or sentences (separated by full stop) whereas e.g., Solfjeld (Chapter 5) understands ‘coordination’ in the narrower sense outlined in Section 3.1.

4.  ‘Coordination’ and ‘subordination’ at text level Following our survey of current conceptions of syntactic coordination and subordination, this section briefly introduces relevant approaches to the description and representation of discourse structure (4.1) and the discourse functions of syntactic subordination and coordination (4.2).

4.1  Relevant discourse-theoretical distinctions There exists at present a whole series of more or less different conceptions of discourse structure and the relations between discourse units. We shall confine ourselves to models that are applied or discussed in the papers below.



 Cathrine Fabricius-Hansen & Wiebke Ramm

Segmented Discourse Representation Theory (SDRT), as outlined in Asher (1993) and Asher & Lascarides (2003), is an extension of ‘classical’ Discourse Representation Theory (DRT, Kamp & Reyle 1993). While DRT incrementally collapses the content of sentence sequences into one single discourse representation structure (DRS), SDRT aims at modelling discourse coherence and the incremental (dynamic) construction of ‘segmented’ discourse representations (SDRSs) that ‘keep track of ’ the units making up the text in question. This is achieved by representing such units – the discourse constituents – as discourse referents of their own, elementary discourse constituents being identified roughly with (DRSs assigned to) clauses. Discourse relations (Asher 1993, 2004), or rhetorical relations (Asher & Lascarides 2003), are defined for pairs of discourse units immediately succeeding each other. Like many approaches to discourse structure, SDRT acknowledges the fact that discourse units may be organised hierarchically or non-hierarchically by distinguishing between two types of discourse relations, ‘subordinating’ and ‘coordinating’.16 The function of a discourse unit connected to its preceding discourse unit by a subordinat­ ing discourse relation (the prototypical example being Elaboration) is to change the ‘granularity’ of description in the text in the sense of providing more detail to some element present in the preceding discourse unit (Asher & Lascarides 2003: 8, 146; Asher 2004: 172), whereas discourse units connected to the preceding context by a coordinating discourse relation continue the description without changing granularity. Central to the definitions of subordinating vs. coordinating discourse relations is the (formal) notion of a discourse topic (Asher 2004) and of discourse dominance (in subordinating discourse relations the preceding discourse unit dominates the current one). As noted, SDRS construction is understood as an incremental (dynamic) online process. Consequently, it has to be decided for each ‘new’ or ‘incoming’ discourse segment (≈ clause) how it attaches to the SDRS established thus far, i.e., at which point and by which discourse relation(s). In this process the so-called Right Frontier Con­ straint (RFC) is of central importance, in particular with respect to the options for the use of anaphoric expressions (Asher 1993: 270–271; Asher & Lascarides 2003: 10–12): Anaphors can only be resolved if their antecedents can be found on the ‘right frontier’ of the incrementally growing discourse structure which covers the proposition introduced by the prior sentence and any propositions that dominate it (i.e., are in a subordinating discourse relation to it). The concepts of ‘coordinating’ and ‘subordinating’ discourse relations are applied in the papers by Ramm (Chapter 6), Averintseva-Klisch (Chapter 9), Delort (Chapter 10) and Petrova & Solf (Chapter 14), and discussed by Blühdorn (Chapter 3).

.  We shall use the notation ‘coordinating’, ‘subordinating’ when referring to discourse relations rather than syntactic relations.

Editors’ introduction

Rhetorical Structure Theory (RST), as originally presented by Mann & Thompson (1988), aims at modelling textual coherence by assigning discourse relations17 recursively to pairs of neighbouring text spans of increasing length, starting with elementary discourse units of (typically) clause length. The text (passage) is assumed to be coherent if it is possible to assign a discourse relation to all discourse units contained in the text. In RST, the idea that discourse units may be connected hierarchically as well as non-hierarchically is accounted for by the distinction between nucleus-satellite and multinuclear discourse relations. The former are weighted in the sense that the nucleus is conceived as more central to the writer’s purpose than the satellite (i.e., the relation between the discourse units is asymmetric). In multinuclear relations the discourse units are not weighted against each other, i.e., they are of equal importance to the writer’s purpose (they represent a symmetric relation). The notion of nuclearity in RST is the central topic of Stede’s contribution to the present volume (Chapter 2). Seemingly, the fundamental distinction between nucleus-satellite and multinuclear discourse relations is similar to the ‘subordinating’/‘coordinating’ dichotomy in SDRT, and some of the discourse relations do indeed have the same labels in both theories (e.g., Elaboration, Background, Contrast). The underlying criteria for the assignment of a discourse relation, however, are quite different: In SDRT, which aims at an incremental model-theoretic interpretation of discourse, the definitions of SDRT relations are closely related to the construction and maintenance of discourse topics and to the temporal relations between eventualities. In RST, on the other hand, the definitions of discourse relations are based on the writer’s intentions as to which communicative effects s/he wants to reach by uttering a piece of discourse. Yet another concept of discourse organisation can be found in the Quaestio model, as outlined e.g., in Klein & v. Stutterheim (1991, 1992) and Stutterheim (1997), which was originally developed to account for text production rather than discourse analysis. According to this approach, a text answers a specific text question, quaestio, which may be divided into a sequence of sub-questions.18 The global structure or development of the text will be constrained in different ways depending on the nature of the quaestio, e.g., whether it is a narrative or a descriptive task. Sentences or clauses that contribute to answering the text quaestio (by answering a sub-quaestio) are said to belong to the main structure of the text. In addition, a text may contain various kinds of side structures consisting of clauses, sentences or larger text spans that do not directly contribute to the quaestio, but fulfil other functions.

.  The set of discourse relations is not absolutely fixed, but many RST applications operate with a set of approx. 30 relations (see e.g., the relation definitions on the ‘official’ RST website http: //www.sfu.ca/rst/01intro/definitions.html. .  In essence, the notion of quaestio corresponds to the notion of a Question under Discussion (QuD), which was developed in question-answer analysis and which is applied in Jasinskaja’s (2006) approach to discourse relations and the discourse function of and-coordination.



 Cathrine Fabricius-Hansen & Wiebke Ramm

Since it is not quite clear how the side structures are linked to the main structure (if at all), a text may not be formally represented as a connected hierarchical structure (e.g., tree structure) in this model, as is the case in SDRT and RST. But, obviously, side structures are ‘secondary’ to the main structure. The quaestio approach has primarily been applied to narrative texts, i.e., text passages that are constrained by a principle of temporal movement; the paper by Carroll et al. (Chapter 7) provides an example. As observed by Ramm (Chapter 6) the model may not be applied quite as easily to non-narrative texts.

4.2  Syntactic coordination and subordination from a textual perspective Discourse theories differ also with respect to whether, and to what degree, they assume and model a (more or less) direct correlation between coordination and subordination in syntax and on the level of discourse representation. As for syntactic coordination, SDRT is relatively explicit in assuming that syntactic and-coordination is a reliable indicator of coordination on discourse level, i.e., a ‘coordinating’ discourse relation like Narration or Result. Asher & Vieu 2005 (following Txurruka 2000), for example, use and-coordination as a test to distinguish between ‘coordinating’ and ‘subordinating’ discourse relations, although they do point to data suggesting that the inference from and-coordination to discourse coordination may be defeasible in some cases (Asher & Vieu 2005: 598–599; see also Asher 2004: 183–185). RST is less explicit than SDRT in associating syntactic coordination with a specific type of discourse relation, since the theory generally refrains from directly mapping discourse relations to concrete syntactic realisations in a particular language. However, examples given on the RST website (http://www.sfu.ca/rst/), illustrating the multinuclear relation Conjunction, suggest that also RST assumes that the syntactic symmetry of and-coordination mirrors a multinuclear (symmetric, non-weighted) relation on discourse level. Whether such a correlation between syntactic and discourse coordination is a feature that holds universally across languages, however, has been questioned in recent contrastive studies (e.g., Ramm & Fabricius-Hansen 2005; Ramm forthcoming). The studies show that coordination is used somewhat differently in Norwegian versus German and English. In particular, syntactic coordination seems to be compatible with some (‘subordinating’) discourse relations in Norwegian that are blocked in German or English. As concerns possible correlations between syntactic subordination and discourse subordination or asymmetry, RST is slightly more explicit, e.g., in Matthiessen & Thompson (1988: 317), who hold “that hypotactic clause combining is best understood as a grammaticalization of the Nucleus-Satellite relations which characterize the rhetorical organization of certain types of written discourse”. Of course, due to the recursive nature of constructing discourse representations in RST, the concept of

Editors’ introduction

asymmetric nucleus-satellite relations is not restricted to syntactically subordinate structures: a nucleus-satellite relation may also hold between independent sentences or even between paragraphs in a text. But again there seems to be some agreement that syntactic subordination typically implies subordination/asymmetry on discourse level as well. The issue of possible non-correlations between subordination on the syntactic level and on the level of discourse relations is addressed in the papers by Delort (Chapter 10) and Ramm (Chapter 6). Also Stede (Chapter 2) shows that automatically equating syntactic subordinating with discourse asymmetry (in the RST sense) is a simplification that does not do justice to the complexity of natural discourse (cf. Stede 2004). The Relevance-Theoretic work on and-coordination by Blakemore and Carston (Blakemore 1987, 2002; Blakemore & Carston 2005), pointing to the possibility of asymmetric interpretations, has already been mentioned above (Section 3.1; see also Solfjeld, Chapter 5 in this volume). Blakemore and Carston also show that coordination is possible in certain cases while blocked in others and, in particular, that using coordination instead of a sequence of non-coordinated (‘full stop’) sentences sends two types of signal to the reader: (i) The two conjuncts should be processed as a unit, both conjuncts functioning together as premises in the derivation of a joint cognitive effect; (ii) specific inferences are licensed regarding the semanticpragmatic relations holding between them, the first conjunct always functioning as a background to the processing of the second. Thus, in narratives a temporal-causal relation is often inferred (see Section 3.1). A non-narrative use of coordination can be seen in argumentative examples, where the conjuncts make a joint contribution as steps in an argumentation (Blakemore & Carston 2005). Relevance Theory does not distinguish between ‘coordinating’ and ‘subordinating’ discourse relations – and prefers to avoid the notion of discourse relations altogether (Blakemore 2002: Section 5.3) – but most of the narrative, as well as the argumentative examples given in Blakemore & Carston (2005), would probably be classified as ‘coordinating’ relations in the SDRT framework. Klein and v. Stutterheim’s quaestio approach also dispenses with an explicit theoretical concept of discourse relations. But at least in narration, clause coordination with and etc. is generally taken to continue the current quaestio, i.e., mostly the main story line, corresponding to the ‘coordinating’ discourse relations Narration and Continuation in SDRT. Syntactic subordination, on the other hand, is conceived as a means of ‘downgrading’ information (Stutterheim 1997; see also Carroll et al. in this volume), without necessarily assigning the downgraded information to some side structure.19 Associating syntactic subordination with a downgrading or backgrounding function at text level is common practice in more traditional linguistic description (see e.g.,

.  Side structures may be realized in different ways: as subordinate clauses, as parenthetical insertions or other kinds of ‘orphans’, and as independent sentences or text passages.



 Cathrine Fabricius-Hansen & Wiebke Ramm

Aarts 2007; Brandt 1996; Cristofaro 2000; Hartmann 2004; Hetland & Molnár 2001; Peyer 1997). Unfortunately, however, these notions are understood in varying ways, having being explained alternately – and mostly not very explicitly – in terms of e.g., information structure (focus – background), illocutionary function (assertion vs. nonassertion), temporal anchoring (event versus co-temporal state), (lack of) relevance in the given discourse context, attachment options for subsequent discourse units, or in terms of other general terms like salience, prominence or communicative weight. There are exceptions such as Tomlin (1985) and Schecker (2000), who define the notions of foreground(ed) and background(ed), or downgraded, information in experimental text reproduction independently of linguistic form. And they both conclude that (certain types of) subordinate clauses do tend to encode information that is backgrounded or downgraded in the defined sense. On the whole, however, empirical investigations into discourse-functional aspects of subordination and coordination in natural discourse are rare. There is a conspicuous lack of comprehensive studies combining a precise theoretical framework and a sound methodology with extensive empirical data from ‘wildlife’ texts including, but not restricted to, narratives. Matthiessen and Thompson conclude their article on “The structure of discourse and ‘subordination’ ” as follows: [I]f hypotaxis in English is a grammaticalization of rhetorical relations, then it follows that the grammar of clause combining may differ radically from one language to another. Indeed, preliminary discourse-based investigations of such phenomena in unrelated languages strongly suggest that this is the case. If the basic approach to clause combining taken in this paper is correct, then the interesting cross-linguistic issue is how and to what extent the grammar of clause-combining in a given language reflects the rhetorical organization of discourse in that language […]. (Matthiessen & Thomspon 1988: 317)

Since then some empirically oriented contrastive research has been done, indicating that even quite closely related languages (or language communities) may differ considerably in their use of syntactic coordination and subordination (in a broad sense) as a means of information packaging at text level; see the papers by Cosme, Solfjeld and Carroll et al. (Chapters 4–7), with further references. Part of this picture is the difference between a more ‘incremental’ and a more ‘hierarchical’ style of writing, which is mentioned by Cosme (Chapter 4), and taken up more in detail by Ramm (Chapter 6); see also Fabricius-Hansen (2007). Of course, more studies are needed, based on more data, more languages and a more refined assessment of language-specific structural constraints and language-specific means of expressing syntactic subordination and coordination, on the one hand, and (discourse) relations between independent sentences, on the other. But it is our hope that the papers presented here will help reduce the research gaps outlined above, either directly or by encouraging further research.

Editors’ introduction

5.  Outline of contents The overall structure of the following collection of papers is outlined in Section 1. Below we briefly present the individual contributions (in the order in which they appear below): their research topics, theoretical framework and methods of investigation, their relation to the general topics of the present volume, as outlined above, and their relation to each other. Manfred Stede (‘RST revisited: Disentangling nuclearity’, Chapter 2) discusses the central notion of nuclearity in RST (see Section 4.1 above) and other discourse-theoretical approaches, relating it to the broader concept of salience. Reviewing three problematic aspects of RST nuclearity – its relationship to syntactic subordination, its direct association with coherence relations, and its purported pervasiveness – he argues that the complexity of textual coherence and discourse organization is not adequately modelled by hierarchical (tree) structures alone, but demands a multi-level representation. The last part of his paper outlines an annotation framework for multi-level discourse representation. Like Blühdorn (Chapter 3), Stede directly addresses theoretical issues of primary importance in the present context. The RST notion of nuclearity also plays a role in the paper by Ramm (Chapter 6). Hardarik Blühdorn (‘Subordination and coordination in syntax, semantics, and discourse: Evidence from the study of connectives’, Chapter 3) addresses the basic question of whether the syntactic and the discourse notions of subordination and coordination are as closely related as the parallel terminology may suggest (cf. Sections 1 and 4.1 above). Using the syntax and semantics of connectives (in the broad sense of e.g., Pasch et al. 2003) as his point of departure, he argues that the nature of the syntactic and the discourse domains are too different to warrant such parallelism and that syntactic hierarchy should not be considered a general model for the conceptualization of discourse hierarchy, nor vice versa. Blühdorn’s contribution relates to those by Solfjeld (Chapter 5) Ramm (Chapter 6) and Delort (Chapter 10) by pointing to mismatches between syntactic subordination or coordination and ‘subordination’/‘coordination’ at the discourse level. Further, it ties in well with Stede’s paper (Chapter 2) in that both address a primarily theoretical issue. Christelle Cosme (‘A corpus-based perspective on clause linking patterns in English, French and Dutch’, Chapter 4) compares clause-linking patterns and discourse information packaging in English, French and Dutch. Investigating comparable (authentic) as well as translational parallel corpora she concludes that the three languages employ clausal subordination and paratactic clause combining differently: On the scale between a more incremental and a more hierarchical information organization Dutch tends towards the former and French towards the latter, with English situated somewhere in the middle. From a theoretical perspective, Cosme’s study supports the conception of (clausal) coordination and subordination as gradient categories (cf. Section 3 above); these aspects of her paper link it to Holler’s study (Chapter 8). Her contribution is related to the three subsequent papers – Solfjeld, Ramm and Carroll et al. – by its contrastive



 Cathrine Fabricius-Hansen & Wiebke Ramm

perspective and its focus on the use of subordination and paratactic clause combining. It shares two of its object languages (English and French) with Carroll et al. (Chapter 7) but is more closely connected to Solfjeld (Chapter 5) and Ramm (Chapter 6) as far as its research questions, data and methods are concerned. Kåre Solfjeld’s study (‘Sentence splitting – and strategies to preserve discourse structure in German-Norwegian translations’, Chapter 5) is related to the papers by Cosme (Chapter 4) and, in particular, Ramm (Chapter 6): it is contrastive (GermanNorwegian), based on parallel (translational) corpora, and concerned with the relation between syntactic subordination/coordination and discourse interpretation. The particular question Solfjeld addresses is to what extent and by what means the discourse structure of the (German) source text (ST) is preserved in the (Norwegian) target text (TT) when a phrasal adjunct is rendered as an independent sentence or clause. Following a primarily relevance-theoretic line of argumentation, he concludes that simply preserving the linear order of the relevant information units often suffices to ensure an adequate interpretation of the TT, but that clause coordination with og ‘and’ may be used as compensation for missing grammatical options to express ‘backgrounding’ by syntactic subordination (adjunction); and connectives may be added in the second TT clause/sentence if the ST discourse relation is not inferable otherwise. Wiebke Ramm (‘Upgrading of non-restrictive relative clauses in translation – a change in discourse structure?’, Chapter 6) analyses the discourse functions of nonrestrictive relative clauses (see Section 3.1 above) and the discourse-functional effect of ‘upgrading’ (German) non-restrictive relative clauses to independent sentences in (Norwegian) translations. She investigates whether the distinction between appositive/ discontinuative and continuative relative clauses is relevant for how a non-restrictive relative clause is translated into Norwegian (in cases where translation by a corresponding relative clause is not an option). Ramm concludes that upgrading is more problematic with respect to discontinuative/appositive relative clauses, since this gives them too much weight in the discourse structure compared to the original text. The study also reveals certain shortcomings of approaches based on SDRT and the quaestio model. Furthermore, as in Cosme’s paper, the topic of different discourse organisation strategies across languages is taken up: The upgrading of relative clauses makes the translations less hierarchically organised than the original texts – an observation which calls for reflections on the cross-linguistic assessment of coherence within and across sentence boundaries. Mary Caroll, Antje Rossdeutscher, Monique Lambert & Christiane von Stutter­ heim (‘Subordination in narratives and macrostructural planning: A comparative point of view’, Chapter 7) present a comparative study of ‘macrostructural planning’ in English, French and German retellings of a silent animation. Following the quaestio model (cf. Section 4.1 above) they take macrostructural planning to manifest itself in information selection (Which events to narrate?) and strategies of downgrading selected information (e.g., clausal subordination and passive) in accordance with the primary narrative task of advancing the story line. Part of this picture is

Editors’ introduction

the choice of (protagonist versus narrator) perspective, which involves the temporal frame of reference and potential mapping of narrated events onto subordinate rather than main clauses (i.e., downgrading). The first part of the study presents empirical generalizations relating observed significant language-specific differences between the three sets of narratives to well-defined morphosyntactic parameters like subject vs. topic prominence, aspectual marking and word order flexibility. The second part constitutes a first step towards a formal (decision-hierarchical) model of the language-specific principles guiding macrostructural narrative planning. As a cross-linguistic investigation of information packaging at discourse level this paper is linked, in particular, to Ramm (Chapter 6) and Cosme (Chapter 4). The paper by Anke Holler (‘German dependent clauses from a constraint-based perspective’, Chapter 8) focuses on the general issue of non-canonical subordination (see Section 3.1 above), proposing a fine-grained systematization and explication within the constraint-based framework of Head-driven Phrase Structure Grammar (HPSG). She argues that dependent clauses must be distinguished regarding their degree of integration into a putative matrix clause and exemplifies this by means of five instances of non-canonical clause types in German: weil-verb second clauses, continuative wh-relative clauses, verb second relative clauses, free dass-clauses, and dependent verb second clauses. The distinctions between these clause types are modelled in the HPSG framework as constraints and sorts partially ordered in multiple-inheritance hierarchies. By not depending on the position of the finite verb, the analysis goes beyond earlier constraint-based approaches implementing a strict dichotomy between main and subordinate clauses and relying on the V2 property alone. It is linked to other papers concerned with detached constructions (cf. Averintseva-Klisch, Chapter 9) and mismatches between formal and functional aspects of subordination, i.e., De­ lort (Chapter 10) and Solfjeld (Chapter 5). Maria Averintseva-Klisch (‘To the right of the clause: Right dislocation vs. afterthought’, Chapter 9) investigates the discourse functions of two apparently similar constructions at the right sentential edge in German, right dislocation (RD) and afterthought (AT). She shows that RD is syntactically adjoined to its host sentence and functions as a discourse-structuring device marking the topic for the following discourse segment, whereas AT is syntactically independent of its host sentence and serves as a local repair strategy in discourse. Furthermore, she shows that these two functions can be identified in French and Russian data as well. Her paper ties in with other papers in this volume investigating possible mismatches between syntactic and discourse structure, e.g., Ramm (Chapter 6) and Delort (Chapter 10), by pointing out that AT exemplifies a subordinating discourse relation (according to the tests developed by Asher & Vieu 2005), but is detached syntactically as an orphan to its host sentence. Laurence Delort (‘Exploring the role of clause subordination in discourse structure – the case of French avant que’, Chapter 10) investigates discourse relational aspects of a specific subordinating connective, viz. French avant que (‘before’). She ­argues that



 Cathrine Fabricius-Hansen & Wiebke Ramm

avant que, depending on the (sentence-internal and -external) context, may convey at least two different discourse relations between the connected clauses. Adopting relevant SDRT definitions (see Section 4.1), she identifies these two relations with Nar­ ration, which is a ‘coordinating’ relation, and Background, which she classifies as a ‘subordinating’ relation in accordance with Vieu & Prévot (2004).20 The difference manifests itself in different substitution and translation options. Delort’s contribution relates to other SDRT-inspired contributions, viz. Averintseva-Klisch (Chapter 9) and Petrova & Solf (Chapter 14), and to Blühdorn (Chapter 3), Ramm (Chapter 6) and Hol­ ler (Chapter 8), which are also concerned with the mismatch between syntactic and discourse ‘subordination’. Michael Franke (‘Pseudo-imperatives and other cases of conditional conjunction and conjunctive disjunction’, Chapter 11) investigates interpretational asymmetries (‘pragmatic puzzles’) exhibited by so-called pseudo-imperatives (PIs), i.e., and-conjunction and or-disjunction as exemplified in example (3) in Section 3.1. Analyzing PIs on the background of related non-standard conjunction and disjunction, he proposes a pragmatic explanation of PI interpretation that is based on the different discourse segmentation behaviour of the two connectors: Disjunctive PIs are discourse separating, associating the two connected clauses with separate communicative events, while conjunctive PIs may be either discourse integrating or discourse separating, with different consequences for the interpretation of the construction as a whole. In distinguishing between discourse integrating and discourse separating (instances of) conjunctive or disjunctive coordination, Franke implicitly addresses not only the fundamental issue of discourse segmentation (see Section 2 above) but also the question of where, i.e., at which meaning level, coordinate clause combining may take place (see Section 3.1 above). The topic of his contribution – asymmetry in clausal coordination – links it to the paper by Reich (Chapter 12). Ingo Reich (‘From discourse to “odd coordinations” – on asymmetric coordination and subject gaps in German’, Chapter 12) addresses syntactic-semantic asymmetries relating to word order (‘clause type’) and subject gaps in German – so-called Asymmetric Coordination (AC) and SLF-Coordination (SLFC)21 – as illustrated in example (4) in Section 3.1. Combining generative feature-based syntax and formal semantics with discourse-relational considerations, he argues that the conjuncts in both cases are semantically linked by a relation of event subordination, termed Occasion. He defines this relation as a semantic explication of the discourse (‘Contiguity’) relation Occasion as understood by Kehler (2002), introducing it syntactically by way of a specific

.  Asher (1993) and Asher & Lascarides (2003) view Background as a ‘coordinating’ relation; but as argued by Vieu & Prévot (2004) and Behrens & Fabricius-Hansen (forthcoming), it fails to meet decisive (RFC-based) criteria of discourse ‘coordination’. In Asher (2004: 179–180), Background is characterized as ‘subordinating’. .  SLF stands for ‘Subject Lacking in F(ronted) structure’ (Reich forthcoming: 2).

Editors’ introduction

functional projection. Thus, the syntactic-semantic characteristics of the AC and SLFC constructions are explained as a ‘grammaticalization’ of properties that are typically attributed to coordinate structures in discourse. Reich’s contribution is related to other contributions concerned with subordination-coordination mismatches, i.e., Blühdorn (Chapter 3) and Holler (Chapter 8), and to the immediately preceding paper (Franke) on pragmatic asymmetries in coordination. Rosemarie Lühr (‘Old Indic clauses between subordination and coordination’, Chapter 13) addresses the interplay between prosody and subordination/coordination in Old Indic. Her paper aims to explain the distribution of sentences with and without stress on the initial verb and, in particular, the status of sentences that have the prosodic properties of subordinate clauses but are not introduced by a subordination marker. Lühr proposes a solution in terms of information structure, viewing the latter construction type as the Old Indic counterpart of German coordinate structures showing the rise-fall contour that goes along with contrastive coordination. That is, contrary to what has often been suggested in the literature, she takes stress on the first verb in sentences lacking a subordinator to be a main clause phenomenon signaling that a continuation will follow. Lühr’s contribution is the only one in the collection that explicitly takes up prosodic and information-structural aspects of subordinationcoordination mismatches. However, it is linked to the paper by Petrova & Solf (Chapter 14) in applying theories of information structure and discourse relations to historic languages. Svetlana Petrova & Michael Solf (‘Rhetorical relations and verb placement in the early Germanic languages. A cross-linguistic study’, Chapter 14) investigate discourserelated properties of verb placement (V1 versus V2) in Old High German (OHG) and other early Germanic (declarative) main clauses. Adopting the SDRT distinction between ‘subordinating’ and ‘coordinating’ discourse relations, they argue that V1 is a common correlate of discourse ‘coordination’ in all early Germanic languages while ‘subordinating’ discourse relations are signalled by different syntactic means. These differences may be responsible for the development of different word order patterns in the modern languages. Petrova & Solf ’s contribution is closely related to the paper by Lühr (Chapter 13) in being concerned with information or discourse structural aspects of main clause word order in historic Indo-European languages.

References Aarts, Bas. 2006. Subordination. In Encyclopedia of Language & Linguistics, 2nd edn, Keith Brown (ed.), 248–254. Amsterdam: Elsevier. Abraham, Werner. 2006. Topic, focus and default vs. contrastive accent: Typological differences with respect to discourse prominence. In On Information Structure, Meaning and Form. Generalizations across Languages [Linguistik Aktuell / Linguistics Today 100], Kerstin Schwabe & Susanne Winkler (eds), 183–203. Amsterdam: John Benjamins.



 Cathrine Fabricius-Hansen & Wiebke Ramm Asher, Nicholas. 1993. Reference to Abstract Objects. Dordrecht: Kluwer. Asher, Nicholas. 2004. Discourse topic. Theoretical Linguistics 30: 163–202. Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation [Studies in Natural Language Processing]. Cambridge: CUP. Asher, Nicholas & Vieu, Laure. 2005. Subordinating and coordinating discourse relations. Lin­ gua 115: 591–610. Austin, Jennifer R., Engelberg, Stefan & Rauh, Gisa. 2004. Adverbials. The Interplay between Meaning, Context, and Syntactic Structure [Linguistik Aktuell/Linguistics Today 70]. Amsterdam/Philadelphia: John Benjamins. Bátori, István, Pusch, Luise F., Levin, Jurij L., Abraham, Werner, Bublitz, Wolfram & Roncador, Manfred von. 1975. Syntaktische und semantische Studien zur Koordination [Studien zur deutschen Grammatik 2]. Tübingen: Niemeyer. Behrens, Bergljot & Fabricius-Hansen, Cathrine. Forthcoming. The relation Accompanying Circumstance across languages. Conflict between linguistic expression and discourse subordination? In Proceeedings from the Third International Conference on Contrastive Semantics and Pragmatics, Shanghai, September 15–19, 2005. Bierwisch, Manfred. 2003. Heads, complements, adjuncts: Projection and saturation. In Mod­ ifying Adjuncts [Interface Explorations 4], Ewald Lang, Claudia Maienborn & Cathrine Fabricius-Hansen (eds), 113–162. Berlin: Mouton de Gruyter. Blakemore, Diane. 1987. Semantic Constraints on Relevance. Oxford: Blackwell. Blakemore, Diane. 2002. Relevance and Linguistic Meaning. Cambridge: CUP. Blakemore, Diane. 2005. And-parentheticals. Journal of Pragmatics 35: 1165–1181. Blakemore, Diane & Carston, Robyn. 2005. Introduction to coordination: Syntax, semantics and pragmatics. Lingua 115: 353–358. Brandt, Margareta. 1996. Subordination und Parenthese als Mittel der Informationsstrukturierung in Texten. In Ebenen der Textstruktur. Sprachliche und kommunikative Prinzipien, Wolfgang Motsch (ed.), 211–240. Tübingen: Niemeyer. Büring, Daniel. 2007. Semantics, intonation and information structure. In The Oxford Handbook of Linguistic Interfaces, Gillian Raamchand & Charles Reiss (eds), Ch. 15. Oxford: OUP. Carston, Robyn. 2002. Thoughts and Utterances. London: Blackwell. Carston, Robyn & Blakemore, Diane (eds). 2005. Coordination: Syntax, Semantics and Pragmat­ ics [Lingua 115]. Cormack, Annabel & Smith, Neil. 2005. What is coordination? Lingua 115: 395–418. Cristofaro, Sonia. 2003. Subordination. Oxford: OUP. Crysman, Berthold. 2006. Coordination. In Encyclopedia of Language & Linguistics, 2nd edn, Keith Brown (ed.), 183–196. Amsterdam: Elsevier. Dehé, Nicole & Kavalova, Yordanka (eds). 2007. Parentheticals [Linguistik Aktuell/Linguistics Today 106]. Amsterdam: John Benjamins. Doherty, Monika (ed.). 1996. Information Structure: A Key Concept for Translation Theory [Linguistics 34]. Doherty, Monika (ed.). 1999. Sprachspezifische Aspekte der Informationsverteilung [Studia Grammatica 47]. Berlin: Akademie-Verlag. Doherty, Monika (ed.). 2005. Language-specific Discourse Linking [Linguistics 43]. Doherty, Monika. 2006. Structural Propensities: Translating Nominal Word Groups from English into German [Benjamins Translation Library 65]. Amsterdam: John Benjamins. Fabricius-Hansen, Cathrine. 1992. Subordination. In Deutsche Syntax. Ansichten und Aussich­ ten, Ludger Hoffmann (ed.), 458–483. Berlin: de Gruyter.

Editors’ introduction Fabricius-Hansen, Cathrine. 2007. Dreimal (nicht) dasselbe: Sprachliche Perspektivierung im Deutschen, Norwegischen und Englischen. LiLi: Zeitschrift für Literaturwissenschaft und Linguistik 145: 61–86. Fabricius-Hansen, Cathrine. Forthcoming. Subjunktor. In Handbuch der deutschen Wortarten, Ludger Hoffmann (ed.), article C22. Berlin: de Gruyter. Granger, Sylviane, Lerot, Jacques & Petch-Tyson, Stephanie (eds). 2003. Corpus-based Approach­ es to Contrastive Linguistics and Translation Studies [Approaches to Translation Studies 20]. Amsterdam: Rodopi. Haiman, John & Thompson, Sandra D. (eds). 1988. Clause Combining in Grammar and Dis­ course. Amsterdam: John Benjamins. Hartmann, Dietrich. 1984. Reliefgebung: Informationsvordergrund und Informationshintergrund in Texten als Problem von Textlinguistik und Stilistik. Zur Verwendung sprachlicher Mittel zum Ausdruck von Haupt- und Nebeninformation. Wirkendes Wort 4: 305–323. Haspelmath, Martin. 2004. Coordinating constructions: An overview. In Coordinating Construc­ tions [Typological Studies in Language 58], Martin Haspelmath (ed.), 3–40. Amsterdam: John Benjamins. Hasselgård, Hilde, Johansson, Stig, Behrens, Bergljot & Fabricius-Hansen, Cathrine (eds). 2002. Information Structure in a Cross-linguistic Perspective [Language and Computers – Studies in Practical Linguistics 39]. Amsterdam: Rodopi. Hetland, Jorunn & Molnár, Valéria. 2001. Informationsstruktur und Reliefgebung. In Language Typology and Language Universals [HSK 20(2)], Martin Haspelmath & Ekkehard König (eds), 617–633. Berlin: Mouton de Gruyter. Holler, Anke. 2005. Weiterführende Relativsätze. Empirische und theoretische Aspekte [Studia Grammatica 60]. Berlin: Akademie Verlag. Huddleston, Rodney & Pullum, Geoffrey K. 2002. The Cambridge Grammar of the English Lan­ guage. Cambridge: CUP. Jasinskaja, Ekaterina. 2006. Pragmatics and Prosody of Implicit Discourse Relations: The Case of Restatement. PhD dissertation, University of Tübingen. Johannessen, Janne Bondi. 1998. Coordination. Oxford: OUP. Johansson, Stig & Oksefjell, Signe (eds). 1998. Corpora and cross-linguistic research: Theory, method and case studies [Language and Computers – Studies in Practical Linguistics 24]. Amsterdam: Rodopi. Johansson, Stig. 2007. Seeing through Multilingual Corpora. On the Use of Corpora in Contrastive Studies [Studies in Corpus Linguistics 26]. Amsterdam: John Benjamins. Kamp, Hans & Reyle, Uwe. 1993. From Discourse to Logic. Introduction to Modeltheoretic Se­ mantics of Natural Language, Formal Logic and Discourse Representation Theory [Studies in Linguistics and Philosophy 42]. Dordrecht: Kluwer. Karhiaho, Izabela. 2003. Der Doppelpunkt im Deutschen: Kontextbedingungen und Funktionen. PhD dissertation, Göteborg University. Kehler, Andrew. 2002. Coherence, Reference and the Theory of Grammar. Stanford: CSLI. Kehler, Andrew. 2004. Discourse topics, sentence topics and coherence. Theoretical Linguistics 30: 227–240. Klein, Wolfgang & Stutterheim, Christiane von. 1991. Text structure and referential movement. Sprache und Pragmatik 22: 2–32. Klein, Wolfgang & Stutterheim, Christiane von. 1992. Textstruktur und referentielle Bewegung. LiLi 86: 67–92.



 Cathrine Fabricius-Hansen & Wiebke Ramm Krifka, Manfred. 2007. Basic notions of information structure. In Interdisciplinary Studies of Information Structure 6, Caroline Fery & Manfred Krifka (eds). Potsdam University 2007. (Also at URL: http://amor.rz.hu-berlin.de/~h2816i3x/). Lang, Ewald. 1977. Semantik der koordinativen Verknüpfung [Studia Grammatica XIV]. Berlin: Akademie-Verlag. Lang, Ewald. 1984. The Semantics of Coordination [Studies in Language Companion Series 9]. Authorized English translation from Lang (1997). Amsterdam: John Benjamins. Lang, Ewald. 2004. Schnittstellen der Konnektoren-Beschreibung. In Brücken schlagen. Grundla­ gen der Konnektorensemantik, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner (eds), 45–92. Berlin: Mouton de Gruyter. Lang, Ewald, Maienborn, Claudia & Fabricius-Hansen, Cathrine (eds). 2003. Modifying Ad­ juncts. Berlin: Mouton de Gruyter Langacker, Ronald W. 1991. Foundations of Cognitive Linguistics Vol. II. Stanford CA: Stanford University Press. Lefèvre, Michel (ed.). 2000. Subordination in Syntax, Semantik und Textlinguistik. Tübingen: Stauffenburg. Lehmann, Christian. 1988. Towards a typology of clause linkage. In Clause Combining in Gram­ mar and Discourse, John Haiman & Sandra D. Thompson (eds), 181–226. Amsterdam: John Benjamins. Levinson, Stephen C. 2000. Presumptive Meanings. The Theory of Generalized Conversational Implicature. Cambridge MA: The MIT Press. Longacre, Robert E. 1985. Sentences as combinations of clauses. In Language Typology and Syn­ tactic Description II, Timothy Shopen (ed.), 235–286. Cambridge: CUP. Mann, William C. & Thompson, Sandra A. 1987. Rhetorical structure theory: A framework for the analysis of text. IPRA Paper in Pragmatics 1: 79–102. Mann, William C. & Thompson, Sandra A. 1988. Rhetorical Structure Theory: Toward a functional theory of text organization. Text 8(3): 243–281. Matthiessen, Christian & Sandra A. Thompson. 1988. The structure of discourse and ‘subordination’. In Clause Combining in Grammar and Discourse, John Haiman & Sandra A. Thompson (eds), 275–329. Amsterdam: John Benjamins. Noonan, Michael. 1985. Complementation. In Language Typology and Syntactic Description II, Timothy Shopen (ed.), 42–140. Cambridge: CUP. Pasch, Renate, Brauße, Ursula, Breindl, Eva & Waßner, Ulrich Hermann. 2003. Handbuch der deutschen Konnektoren: Linguistische Grundlagen der Beschreibung und syntaktische Merk­ male der deutschen Satzverknüpfer (Konjunktionen, Satzadverbien und Partikeln). Berlin: de Gruyter. Partee, Barbara H. & Rooth, Mats. 1983. Generalized conjunction and type ambiguity. In Mean­ ing, Use and Interpretation of Language, Rainer Bäuerle, Christoph Schwarze & Arnim von Stechow (eds), 361–386. Berlin: de Gruyter. Peyer, Ann. 1997. Satzverknüpfung – syntaktische und textpragmatische Aspekte [RGL 178]. Tübingen: Niemeyer. Quirk, Randolph, Greenbaum, Sidney, Leech, Geoffrey & Svartvik, Jan. 1985. A comprehensive Grammar of the English Language. London: Longman. Ramm, Wiebke. Forthcoming. Discourse-structural salience from a cross-linguistic perspective: Coordination and its contribution to discourse (structure). In Salience. Multidisciplinary Perspectives on its Function in Discourse [Trends in Linguistics. Studies and Monographs (TiLSM)], Michael Grabski et al. (eds). Berlin: Mouton de Gruyter.

Editors’ introduction Ramm, Wiebke & Fabricius-Hansen, Cathrine. 2005. Coordination and discourse-structural salience from a cross-linguistic perspective. In Salience in Discourse: Multidisciplinary Ap­ proaches to Discourse 2005, Manfred Stede, Christian Chiarcos, Michael Grabski & Luuk Lagerwerf (eds), 119–128. Münster: Stichting/Nodus. Redder, Angelika. Forthcoming. Konjunktor. In Handbuch der deutschen Wortarten, Ludger Hoffmann (ed.), article C12. Berlin: de Gruyter. Reich, Ingo. Forthcoming. “Asymmetrische Koordination” im Deutschen. Habilitationsschrift. University of Tübingen 2007. To appear in [Studien zur deutschen Grammatik]. Stauffenburg: Tübingen. Reis, Marga. 1993. Satzfügung und kommunikative Gewichtung: Zur Grammatik und Pragmatik von Neben- vs. Unterordnung am Beispiel ‘implikativer’ und-Konstruktionen im Deutschen. In Wortstellung und Informationsstruktur [Linguistische Arbeiten 306], Marga Reis (ed.), 203–249. Tübingen: Niemeyer. Reis, Marga. 1997. Zum syntaktischen Status unselbständiger Verbzweit-Sätze. In Syntax im Fokus. Festschrift für Heinz Vater, Christa Dürscheid, Karl Heinz Ramers & Monika Schwarz (eds). Tübingen: Niemeyer. Reis, Marga. 1999. On sentence types in German: An enquiry into the relationship between grammar and pragmatics. Interdisciplinary Journal for Germanic Linguistics and Semiotic Analysis 4: 195–236. Schwabe, Kerstin & Zhang, Ning. 2000. Ellipsis in Conjunction [Linguistische Arbeiten 418]. Tübingen: Niemeyer. Schecker, Michael. 2000. Zur kommunikativ-funktionalen Leistung von Nebensätzen. In Subor­ dination in Syntax, Semantik und Textlinguistik, Michel Lefèvre (ed.), 115–122. Tübingen: Stauffenburg. Shopen, Timothy (ed.). 1985. Language Typology and Syntactic Description II. Cambridge: CUP. Sperber, Dan & Wilson, Deirdre. 1986. Relevance: Communication and Cognition. Oxford: Blackwell. (2nd edn, with postface, 1995). Stede, Manfred. 2004. Does discourse processing need discourse topics? Theoretical Linguistics 30: 241–254. Stutterheim, Christiane von. 1997. Einige Prinzipien der Textproduktion: Empirische Untersuchungen zur Produktion mündlicher Texte [Germanistische Linguistik 184]. Tübingen: Niemeyer. Sæbø, Kjell Johan. Forthcoming. Adverbial clauses. In Semantics: An International Handbook of Natural Language Meaning [Handbücher zur Sprach- und Kommunikationswissenschaft/ Handbooks of Linguistics and Communication Science], Claudia Maienborn, Klaus von Heusinger & Paul Portner (eds), Ch. 62. Berlin: Mouton de Gruyter Tomlin, Russell S. 1985. Foreground-background information and the syntax of subordination. Text 5: 85–122. Txurruka, Isabel G. 2000. The semantics of  ‘and’ in discourse. Technical Report ILCLI-00– LIC-9, University of the Basque Country. Velde, John te. 2000. On the structure of coordination. In Ellipsis in Conjunction[Linguistische Arbeiten 418], Kerstin Schwabe & Ning Zhang (eds), 51–78. Tübingen: Niemeyer. Verstraete, Jean-Christophe. 2007. Rethinking the Coordinate-Subordinate Dichotomy. Interper­ sonal Grammar and the Analysis of Adverbial Clauses in English [Topics in English Linguistics 55]. Berlin: Mouton de Gruyter. Vieu, Laure & Prévot, Laurent. 2004. Background in SDRT. In Proceedings of the conference Traitement Automatique du Langage Naturel, 485–494. Fez, Morocco.



 Cathrine Fabricius-Hansen & Wiebke Ramm Vuillaume, Marcel. 2000. Asymmetrische Koordination in subordinierten Strukturen. In Sub­ ordination in Syntax, Semantik und Textlinguistik, Michel Lefèvre (ed.), 45–54. Tübingen: Stauffenburg Verlag. Zeevat, Henk. Forthcoming. Discourse structure in optimality theoretic pragmatics. In Struc­ turing Information in Discourse: The Explicit/Implicit Dimension [Special issue of Linguistics], Cathrine Fabricius-Hansen & Bergljot Behrens (eds). Zifonun, Gisela, Hoffmann, Ludger, Strecker, Bruno et al. 1997. Grammatik der deutschen Sprache Bd. 1–3 [Schriften des Instituts für Deutsche Sprache 7]. Berlin: Walter de Gruyter.

part i

General and theoretical issues

RST revisited Disentangling nuclearity1 Manfred Stede

University of Potsdam The paper discusses the notion of nuclearity as put forward by Rhetorical Structure Theory as a general principle of text organization. On the basis of an inquiry into different kinds of salience phenomena in texts, several problems with the purported role of nuclearity are identified. It is argued that RST trees conflate too much information from different realms of description in a single structure. As an alternative to a more detailed investigation of coherence phenomena, an approach toward multi-level analysis and annotation of text is outlined, which also keeps the various sources of salience apart. Keywords: RST, nuclearity, coherence, multi-level analysis

1.  Introduction Among the various theories of discourse structure, Rhetorical Structure Theory (RST, Mann & Thompson 1988) can probably claim to be the most empirical one: It was developed at the time on the basis of thorough analyses of a variety of texts, and it has since been used for quite different purposes and for different types of text. To characterize it very briefly,2 the idea of RST is to postulate some 20 coherence relations, defined in terms of the effect they are meant to cause in the reader’s beliefs and attitudes, and to claim that a coherent text can be analysed in terms of these relations. This would lead to a tree structure that recursively connects all the “minimal units” and the resulting larger text spans. Following popular practice in later publications, we henceforth call the minimal units EDUs, for “elementary discourse units”. Mann and Thompson acknowledge that coherence relations are “pre-realizational” in the sense .  For their very helpful comments on earlier versions of this paper, I am grateful to Bonnie Webber, John Bateman and two anonymous reviewers. .  A good general introduction to RST, also discussing other work that has been inspired by the theory, is given by Taboada & Mann (2006). See also the website www.sfu.ca/rst.

 Manfred Stede

that they connect mental representations of text spans – but for the purposes of text analysis one regards them as connecting actual portions of surface text. As for the syntactic shape of the EDUs, the authors deliberately do not go into much detail but characterize them as being typically clauses. Mann and Matthiessen (1991: 234) elaborate on this decision: The units of an RST analysis are chosen to fit the purposes of the analysis, and are not theoretically prescribed. Our usual practice in analyzing a text is to regard clauses as the realizations of units, but to merge restrictive relative and complement clauses with their parents and to treat elliptical clauses as if they were non-elliptical.

In the following, we use “text segments” and “text spans” interchangeably as referring either to EDUs or to larger compositions of them. A relation may only join adjacent text spans, and any span can only have one “parent node” in the tree. In the relation set postulated by Mann and Thompson (1988), 21 of the 24 coherence relations join a nucleus and a satellite segment, where the latter is less “important” (to be clarified in the next section). The remaining three relations are multinuclear, i.e., they join segments of equal importance. An important claim is another partitioning of relations into two groups: Instances of presentational relations are supposed to change the reader’s mind in some way, such as motivate her to take a particular action, enable her to perform an action, or encourage her to believe a certain proposition. Subject-matter relations, on the other hand, describe kinds of connections between events happening in the world; here, the effect is merely that the reader “recognizes” the relationship between the two units. As an example, consider the following definition of the (presentational) relation Evidence: – Constraints on nucleus (N): Reader might not believe N to a degree satisfactory to writer – Constraints on satellite (S): Reader believes S or will find it credible – Constraints on the N+S combination: Reader’s comprehending S increases Reader’s belief of N – The effect: Reader’s belief of N is increased – Locus of the effect: N The empirical foundation was the main reason for RST’s considerable popularity among researchers concerned with authentic discourse. And yet, the very fact that the same rhetorical relations can be applied to so many different pairs of text spans implies the danger that definitions become so vague that at many points in a text analysis, several relations can be assumed to hold, which equally well conform to the definitions. Ambiguity is not a priori problematic for text analysis – it is not surprising that portions of text and their relationships can have different readings – but it should be possible for the assumed representations to take notice of such genuine ambiguities

RST revisited

rather than to hide them under the covers of a single tree representation. In earlier work (Grote et al. 1997; Stede 2004b), colleagues and myself argued that both the Concession and the Contrast relation in RST are victims of the tendency toward too unspecific definitions; they conflate a variety of different ways in which two states of affairs can be contrasted with one another. In this paper, I try to make essentially the same point for RST’s notion of nuclearity. Mann and Thompson, in a nutshell, characterized the nucleus of a relation as being “more central to the writer’s purposes”, and my point will be that being central to one’s purposes can be a very different thing from case to case, and that it is worth distinguishing certain notions that occasionally give rise to conflicting predictions on nuclearity. As a consequence, it will be questioned whether it is helpful to assume a single hierarchical structure as an account of text coherence. The paper proceeds as follows. In order to do justice to the original proposals on nuclearity by Mann, Matthiessen and Thompson in the late 1980s, their position will first be summarized in Section 2. Also, some interesting extensions proposed by other authors will be mentioned there. I then turn to a more general discussion of “salience” in text and investigate different ways in which certain segments can be more important/prominent/salient than others (Section 3). This will be my starting point for a critique of RST-nuclearity in Section 4, which will be followed in Section 5 by a proposal for a multi-level approach to discourse structure, which confines different types of information to distinct layers.

2.  Nuclearity in RST 2.1  The original idea: Mann, Matthiessen, and Thompson Towards the end of the most-cited original RST paper, Mann & Thompson (1988) devote a section to clarifying their notion of nuclearity, which they in the previous sections had used more on an intuitive basis when defining the various coherence relations. The trigger for proposing nuclearity as a general principle of text organization was the observation that all but three (Sequence, Contrast, and Joint) of their coherence relations were quite obviously asymmetric. A clear example is Evidence: When A is evidence for B, then B cannot be evidence for A. In carving out this asymmetry for all the relations, Mann and Thompson noticed three commonalities, which for them warranted the general division into nucleus and satellite: (i) Often, one segment depends on the other for the discourse to be comprehensible – for Evidence, the nuclear claim is necessary for the satellite to be understood correctly; without the nucleus, it would be a non-sequitur. Consider the following example, where the relation holds between nucleus (a) and satellite (b), and (a) is more central for the text function than (b):



 Manfred Stede

(1) a. Tomorrow the weather will be nice. b. I listened to the latest forecast on the radio. c. So we can go to the picnic in the afternoon.

(ii) Often, the satellite is prone to substitution – for Evidence, one might replace the satellite with a different piece of evidence, without changing the overall purpose of the text. (iii) Often, the nucleus is more essential to the writer’s purposes than the satellite. The much-cited criterion for making the distinction between nuclei and satellites is Mann and Thompson’s deletion test. The first argument concerns the overall text function: We can determine the “most nuclear” minimal unit of a text by tracing down from the root node to the nucleus at each level all the way to the minimal units. For illustration, see Figure 1,3 which gives the RST analysis of a short text that extends example (1). Nuclei are attached with straight lines to their mother nodes, whereas satellites are linked to their nuclei with curved lines. Moving down from the root node (1–6) to a leaf node (i.e., an EDU) without ever following a satellite branch leads to unit (3), the “most nuclear” one. The deletion test predicts that when removing the most nuclear unit, the overall message of the text typically becomes quite difficult, if not impossible, to infer. This diagnostic is then extended to all minimal units that serve as nuclei in the tree (in Figure 1, these are 1, 3, 4, and 5): When they are removed from the text, the result is an incoherent sequence of utterances. On the other hand, with removal of those minimal units that have been assigned satellite status (2 and 6 in Figure 1), the remainders (even though lacking some cohesion) still convey an idea of what the text is about and often even manage to acceptably communicate the main idea. Volitional-result 1–2

Evidence

Tomorrow the weather will be nice.

I listened to the latest forecast on the radio.

1–6 3–6

Elaboration 4–6

So we can go to the picnic in the afternoon.

Evaluation

4–5 Sequence We can take the bus at 1pm

and come back by 6pm.

That leaves plenty of time for beer and sausages!

Figure 1.  RST analysis of a sample text.

As for the relationship between nuclearity and linguistic structure, Mann and Thompson mention the idea that the nucleus/satellite distinction directly corresponds to the syntactic phenomenon of hypotactic clause combining: “Grammars in many

.  The figure was created with the RSTTool software (http://www.wagsoft.com/RSTTool).

RST revisited

languages draw a distinction between hypotactic and main clauses because of the nucleus-satellite distinction in discourse” (p. 269). We do not elaborate here on the even stronger proposition conveyed in that quotation – that nuclearity might be responsible for grammars having evolved hypotactic clause combining; instead we will later discuss merely the (weaker) correlation that is proposed to hold between grammar and discourse structure. The claim is expanded in more detail by Matthiessen and Thompson (1988), who argue for a direct mapping between subordinate clauses and satellites of coherence relations. Mann and Thompson acknowledge two cases where the nucleus/satellite distinction is rather meaningless: in “enveloping structures” such as conventionalized beginnings and endings of letters, and in parallel structures as discussed by Fries (1981). His examples involve repeated comparisons and contrasts between the same entities. The shape of such structures comparing A and B can be sketched in a simplified way as follows: While A is X1, B is rather Y1. A has a X1, and B has a X2. X3 is highly relevant for A, whereas X4 is more important for B. Regarding the functional interpretation of nuclearity, Mann and Thompson make another distinction between two groups of relations, based on their notion of locus of effect of the relation, which is part of each definition and states the segment that the described effect should arise from:4 –

–

When the locus of effect is only the nucleus (as in Evidence), nuclearity represents the qualitative difference in role between the essential and the inessential, thought and afterthought. The satellite supports the nucleus but does not contribute to it, and the writer intends the reader to notice the distinction introduced by nuclearity. When the locus of effect is both nucleus and satellite (as in Condition or Elaboration), the structural difference between nucleus and satellite represents some distinction in the organization of the subject matter. The distinction is presented as important to the reader, and the significance of the satellite tends to be found in the nucleus – both nucleus and satellite contribute to the result.

The role of locus of effect is discussed in more detail in a less-cited paper by Mann & Matthiessen (1991). Interestingly, here the authors propose a more balanced view of nucleus and satellite in subject-matter relations, for which “it seems quite reasonable that the locus of effect should be the nucleus plus the satellite. The purpose of such relations is to represent the kind of connection prevailing between the nucleus and the satellite, not just to achieve the function of the nucleus” (Mann & Matthiessen 1988: .  This is a part of the definitions that in fact did not always receive enough attention by other authors referring to RST relations and applying them to sample texts. One reason might be that the “locus of effect” field is not present in the definitions that were chosen for the RST website (www.sfu.ca/rst).



 Manfred Stede

244). Mann and Matthiessen proceed to state that the division of relations according to locus of effect and that of presentational versus subject-matter orientation is in fact identical. This minimally diverges from the definitions given in Mann and Thompson (1988), where Concession is in conflict with this view – maybe not surprisingly, because Concession can indeed be seen as a somewhat problematic presentational relation (Grote et al. 1997).

2.2  Nuclei on the run: extensions by other authors The role of nuclearity for discourse structure has been strengthened further by Marcu (2000) who proposed the “strong nuclearity hypothesis”: When a relation is postulated to hold between two spans of text, then it should also hold between the nuclei of these two spans. This move ensures the “upward compatibility” of the idea of nuclearity, from minimal units to arbitrary text spans. At the same time, it generates new constraints for nuclearity decisions at the lower levels: The annotator is encouraged to assign nuclearity status to EDUs in such a way that the combinations of larger units work out correctly. Thus, if in doubt about assigning a relation with a particular nucleus/satellite distribution on a lower level, considering the wider context will often assist in making the decision. Mann and Thompson had argued their case on the basis of small examples with elementary units only; the upward extension was implicitly alluded to but not explicitly proposed. Marcu took this step and supported it, inter alia, with the argument that the application of automatic text summarization would benefit from it: The nuclear portions of a text are assumed to be the more important ones that should become part of an (extractive) summary of the text. In Marcu (1999), he suggests to turn the notion of nuclearity into a scalar one when considering complete texts; the measure for assigning a degree of nuclearity to an EDU considers to what extent it is embedded in larger nuclei and satellites, which results in a partial ordering of all the units according to their “importance”. For example, the aforementioned “most nuclear” EDU(s) will have only nuclear segments along the path from the EDU to the root node of the tree; others also appear in a number of satellites, determining their degree. For our sample tree in Figure 1, this measure would assign the same degree to EDUs 1, 4 and 5, as they all involve one satellite connection when moving up to the root node. Another notable promotion of the role of nuclearity was suggested by Cristea et al. (1998). In their “veins theory”, they argue that antecedents to anaphoric expressions are more likely to be found in nuclear material of the preceding context rather than in satellite material. This appears to echo the distinction between subordinating and coordinating coherence relations in Segmented Discourse Representation Theory (Asher & Lascarides 2003), which to some extent was motivated by constraining anaphoric accessibility – an idea that probably originated with Polanyi’s (1988) “rightfrontier constraint”. While it had originated in this fashion as a linguistically-motivated criterion, SDRT in more recent work seems to regard it as a close relative of nuclearity in RST (cf. Danlos 2008).

RST revisited

An interesting move was made in the annotation guidelines for the RST Treebank (Carlson et al. 2003). The number of coherence relations used in that corpus is 78, including 25 multinuclear ones. This amounts to 32% of all relations, compared to 12.5% in the original Mann/Thompson relation set (where only 3 out of 24 relations are multinuclear). 17 of those 25 multinuclear relations have mononuclear counterparts, i.e., relations with identical definitions except for the assignment of one or two nuclei, respectively. In the RST Treebank, annotators made equal use of both versions: They chose the multinuclear version of a relation 1,905 times (51.4%) and the mononuclear one 1,801 times (48.6%).5 Moreover, according to the guidelines, 8 of the mononuclear relations occur in both variants of nucleus/satellite distribution (as Mann and Thompson had proposed merely for the Cause and Result relations). This indicates that Carlson et al. were acutely aware of the problem that annotators often experience when asked to decide which of two segments should be labelled nuclear. Carlson et al.’s overall solution is to work with a much larger set of coherence relations; to my mind, this hides the fact that coherence arises from distinct sources, which also should be represented separately, and that nuclearity as a purportedly “general” phenomenon is largely orthogonal to the description of the specific coherence relations.

2.3  Other concepts of coherence relations When considering other proposals that explain text coherence in terms of relations, it turns out that not many of them include a notion that corresponds to nuclearity as discussed in RST. None of the works of Hobbs (1979), Sanders et al. (1992), or Kehler (2002) systematically assign different prominence to the segments of relations. On the other hand, Bateman & Rondhuis (1997), who more explicitly draw on RST, propose to regard nuclearity not as a static aspect of the relation definitions, but as a distinct feature that can combine with a set of other features to dynamically create a coherence relation (similar to the decompositional approach of Sanders et al. (1992)). That is, they transfer the idea underlying the Cause/Result relation pair of Mann and Thompson to the entire set of relations, so that any relation can assign nucleus status to either span.6 This amounts to a more radical solution to the above-mentioned problem noted by Carlson et al. (2003), whose response was to partition the overall relation set according to the distribution of nuclearity. Another approach that borrowed nuclearity from RST is “relational discourse analysis” (Moser & Moore 1995), which aims to merge RST with the approach of Grosz &

.  These numbers of annotations in the treebank were provided by Bonnie Webber (personal communication). .  Similarly, a recent proposal of a relation set by Golebiowski (2006) states that the majority of her relations “are basically neutral, and thus able to function both paratactically and hypotactically” (p. 261).



 Manfred Stede

Sidner (1986). These authors use the terms core and contributor for nucleus and satellite, respectively. A recent approach is that of Wolf & Gibson (2005) who draw a distinction between directed and undirected relations, and while they mention RST as related work, they do not seem to regard their directionality as a general principle of text organization. Rather, they treat it as one feature of the semantics of the individual relations. The same holds for the more surface-oriented conjunctive relations proposed by Martin (1992), which will be mentioned again in Section 5. In summary, the idea of nuclearity as a principle of discourse organization based on “centrality for the writer’s purposes” is largely confined to RST and approaches that directly draw on it; a range of other theories of discourse coherence do not postulate an equivalent notion.

3.  Salience in text Having stated the RST position on nuclearity, we now turn to a somewhat broader viewpoint and examine different kinds of relative salience of segments in text – where salience is now meant as a neutral cover term for a variety of phenomena to be investigated. It is beyond doubt that a reader of a text perceives different portions as more or less salient; the crucial question is whether this is due to a single, underlying notion – nuclearity – that should be postulated as an elementary principle of text organization. In this section, I will challenge this idea with observations that resulted from our annotation work with the Potsdam Commentary Corpus (Stede 2004), a collection of German newspaper commentaries, 175 of which have been annotated with RST trees by students trained on the basis of (Mann & Thompson 1988). From time to time, we asked the annotators on what grounds they had made their decision of nucleus assignment between adjacent text spans. After adding some of our own observations we get the following list (a–j) of reasons that annotators can put forward when labelling a text segment as a nucleus. To illustrate the phenomena, in several cases we refer to a sample text, the beginning of the introductory section of a research paper, shown in Figure 2. Sentences are numbered, but note that most sentences contain more than one “minimal unit” of an RST-style analysis. a. One segment turns out to be more supportive of the text’s overall purpose. Constructed example: (2) a. The Labour Party supported the strike. b. The Liberals strongly voted against it. c. Thus we have seen another example of the weakness of the Labour.

Considering (a) and (b) in isolation, none of the two appears to deserve nuclearity status more than the other. But (c) indicates that the text is meant to encourage readers to support the position of the Liberals, thus rendering (b) nuclear. b. One segment supports the intention of the other segment. This is the “classical” situation of RST relations such as Evidence or – in this example – Enablement:

RST revisited

(1) It is widely believed that the best human tutors are more effective than the best computer tutors, (…). (2) A major difference between human and computer tutors is that human tutors use face-to-face spoken natural language dialogue, whereas computer tutors typically use menu-based interaction or typed natural language dialogue. (3) This raises the question of whether making the interaction more natural, such as by changing the modality of the computer tutoring to spoken natural language dialogue, would decrease the advantage of human tutoring over computer tutoring. (4) In fact, as will be detailed below, several potential benefits of spoken tutorial dialogue with respect to increasing learning have already been hypothesized in the literature. (5) One hypothesis is that spoken dialogue may be better at eliciting student behaviors that are believed to accelerate learning, such as student knowledge construction. (6) A second hypothesis is that speech allows tutors to infer a more accurate student model, which similarly is believed to accelerate learning. (7) A third hypothesis is that speech primes a more social interpretation of the tutorial environment, which again is hypothesized to accelerate learning. (8) It is thus important to test whether a move to spoken dialogues is likely to yield increased benefits with respect to learning and other performance measures. (9) Furthermore, if the addition of speech can indeed increase learning gains, it is also important to understand why spoken dialogue accelerates learning. Figure 2.  Excerpt from D. Litman et al.: “Spoken versus typed dialogue tutoring.” Int’l Journal of Artificial Intelligence in Education 16(2), 2006, p. 146. Sentence numbers inserted by MS.



(3) Open the printer cover carefully. The lever is located at the lower right of the backside.

In the sample text in Figure 2, sentence 8 states the main question the paper is set to answer, and 4–7 are meant to support the claim that the question is actually a relevant one. c. Recurrence: Material in one of the two spans is taken up later in the text, and hence appears to be more important for the development of the text: (4) a. Jim was opening a wine bottle when b. his mother rushed in. c. She held a letter in her hands.

Choosing a relation joining (a) and (b) can amount to the decision which proposition is meant to be in the foreground and which in the background; since the mother ­appearing in (b) is also the subject of (c), this local configuration suggests that (b) is more nuclear than (a). d. Repetition: When longer material is repeated, as is the case in both the second clauses of sentences 6 and 7 of the sample text, it does not provide new information but merely reminds the hearer, which renders it less nuclear than a potentially more informative adjacent unit (cf. the deletion test mentioned in Section 2).



 Manfred Stede

e. Digression: When a text segment constitutes a temporary digression from the main topic of the text, it is likely to be only a satellite, which is also supported by the deletion test. f. Meta-discursive elements: A segment can appear to be decidedly non-nuclear when it is a text-internal directive. An example is the clause as will be detailed below in sentence 4 of the sample text. g. Connective: Sometimes, the connective chosen by the author can mark the nucleus/satellite distinction, as with the German zwar (sat) … aber (nuc) (‘true … but’), or nicht nur (sat) … sondern (nuc).” Example: (5) Der Bürgermeister hat zwar keinen Erfolg beim the mayor has true no success with.the Bürokratieabbau, aber er ist ein eifriger bureaucracyreduction but he is an eager Wirtschaftsförderer economysupporter ‘The mayor has no success in cutting red tape, but he is a vivid supporter of the local economy.’

Similarly, punctuation marks can indicate parenthetical, and hence satellite, information. h. Other lexical marking of salience: Authors can use lexical items to explicitly assign “weight” to utterances, as the phrase it is important to does in sentences 8 and 9 of the sample text. i. Syntactic structure: A main clause often seems more nuclear than an associated subordinate clause, as mentioned in Section 2.

(6) While the speaker struggled through his manuscript, the audience gradually turned to sleep.

Several examples can also be found in the sample text, where subordination correlates with other criteria, e.g., with (f) in sentence 4. j. RST definition: The relation is stated by the RST definition to assign nuclearity in a specific way. For instance, Sentence 9 of the sample text expresses a Condition relation, which according to Mann and Thompson (1988) consists of a condition-satellite and a consequence-nucleus. Similarly, Carlson et al. (2003) had to fix the nucleus assignment for the relation Attribution used in the RST Treebank. It relates a statement to its source, as in The spokesman reported that the company lost two important contracts last year. Or consider sentence 1 of the sample text, where the source is not an individual but “public” opinion. Rather than having annotators decide whether the source’s act of reporting or the content of the report is more important, Carlson et al. define the content portion of such sentences generally to be the nucleus. Notice that these factors, except for (j), are not meant to be hard-and-fast rules for making the nuclearity decision; rather, they are evidential factors. In a suitable context,

RST revisited

any one of them could be cited by an annotator as the decisive factor for labelling one of the adjacent text segments as nuclear. (a)–(j) are obviously not mutually exclusive, but – and that is the point here – they are largely independent of each other: Any one of them alone can in some specific context be taken as the decisive evidence. The observation that the factors are not mutually exclusive suggests that they operate on different levels of description. (a) and (b) are intention-based criteria: When the intention of one segment “dominates” that of the adjacent one, or when the segment intention supports the overall text intention, we have evidence for nuclearity. (c)–(f) have, in different ways, to do with the thematic development of the text. In (c), co-reference and topic-continuity trigger the decision on the level of minimal units, and in (e) topic (dis-)continuity, among minimal or larger units, is responsible. Repeating information in (d) and supplying text-organizing meta-information in (f) can be seen as ways of interrupting the “main” flow of information. Criteria (g)–(i) are the most surface-oriented ones, and they are likely to co-occur with some of the others. Underneath the surface, (h) again is related to intentions: The author, for whatever reason, chooses to emphasize some portion of the text. In contrast to (a) and (b), however, this is not relational but a feature of a single text unit – the reader shall recognize that the author regards the contents of this unit as central for her purposes. This might co-occur with (a) – the unit can also support the intention of a larger segment – but it need not necessarily do so. Finally, (j) is of an entirely different kind: It is an inherent necessity postulated by the underlying theory to assign nuclearity in one particular way once a specific relation has been found applicable. The fact that such a multitude of factors can influence nuclearity assignment is not per se problematic. The question is, however, whether those distinct factors indeed are reflexes of a common underlying notion, a “general principle of text organization”. In the next section I will argue that we have reason to question this.

4.  Problems with nuclearity in RST We now proceed to synthesize the exposition of RST-nuclearity and our inventory of types of salience into a critique and then to an outline of a different approach towards investigating discourse phenomena. To that end, this section critically reviews three central aspects of the RST view of nuclearity: its relationship to syntactic subordination, its direct association with coherence relations, and the purported pervasiveness of nuclearity.

4.1  Syntactic subordination and nuclearity Undoubtedly, the idea to link syntactic subordination to discourse nuclearity has intuitive appeal, and moreover it is not difficult to accumulate evidence in support of such a link: Quite often, a syntactically subordinate clause is in fact “less central to the writer’s purposes” than the matrix clause is. But on the other hand, one can also gather



 Manfred Stede

evidence for the opposite position – the writer’s purposes running against syntactic subordination, which altogether suggests that matters are more complicated, and that the view accentuated especially in Matthiessen & Thompson (1988) generalizes too boldly from a mere tendency to a principle (see Blühdorn, this volume). For one thing, we have pointed out in the previous section that in the absence of any over-arching intentional structure, it is the continuation of discourse – in terms of referential and/ or thematic continuity – which often governs nuclearity assignment. If a discourse referent retains prominence in the subsequent text, the clause containing its original introduction has a good reason to claim nuclearity status. This clause, however, might quite well be a subordinate one. This point is also made by Golebiowski (2006: 262), who uses the sentence While I was feeling really well on Saturday, I was ill on Sunday to illustrate that in a suitable context, either both clauses can be of equal functional prominence (7a) or one can be more important (7b):



(7) a. W  hile I was feeling really well on Saturday, I was ill on Sunday. On Saturday, I had a lot of energy and I (…) managed to have a beautiful meal with my friends and even spent an enjoyable evening with my family. On Sunday, however, I woke up with a splitting headache. In spite of taking some pain relief tablets, my head did not clear up till the evening. I had to cancel all arrangements made for the day and ended up spending most of Sunday in bed. b. While I was feeling really well on Saturday, I was ill on Sunday. I woke up with a splitting headache. In spite of taking some pain relief tablets, my head did not clear up till the evening. I had to cancel all arrangements made for the day and ended up spending most of Sunday in bed.

It seems worthwhile to relate this point to a different issue raised in Mann and Thompson (1988), that of a “standard ordering” of the spans of relations. The authors state that many relations have a clear tendency as to the linear ordering of their nucleus und satellite; for example, both Antithesis and Condition would normally have their satellites occur before their nuclei. Notice that this statement in conjunction with frequent linguistic realization of such relations (“although A, B” and “if A, then B”, respectively) indeed lends support to the purported link between satellite (A) and subordinate clause. But, of course, this “standard” ordering is by no means mandatory – “B, although A” and “B, if A” are perfectly natural. In these cases, however, the subordinate A segments (i.e., the satellites qua relation definition) due to their position at the end of the sentence can play different roles for the continuation of the discourse; for the case of temporal clauses, ordering has been investigated by de Swart (1999) and Schilder and Tenbrink (2001). In summary, for text portions where intentions provide the structural scaffolding, subordination seems to usually mirror the relative weights of such intentions:

(8) While Smith’s approach to the free-will problem was not entirely wrong, the general direction of his philosophy has been completely refuted by now.

RST revisited

But where intentions do not play the critical role,7 linear order takes over as an important force in assigning different degrees of prominence to text segments. Syntactic subordination also plays a role here, but the interactions become complicated, and RST does not offer the means to account for such effects. In addition to the more difficult relationship between subordinate clauses and nuclearity, it is neither the case that, conversely, co-ordinating clauses would naturally lead to multinuclearity. It is well-known that in clauses joined by the contrastive conjunction but the second clause typically carries more “weight” than the first one. An illustrative example was given by Elhadad & McKeown (1990): (9) a. He failed the exam, but he is smart. Let’s hire him. b. ?He is smart, but he failed the exam. Let’s hire him.

4.2  Nuclearity tied to relations For the seven presentational, or intention-based, coherence relations in RST, nuclearity plays the role of “domination” between discourse segment purposes in the theory of Grosz & Sidner (1986). RST generalized this notion to the 16 subject-matter relations as well and defined for each relation which elements would play the roles of nucleus and satellite, respectively. The one exception is the set of causal relations, where RST distinguishes Cause from Result, with the only difference being the mapping from cause/effect to nucleus/satellite. Mann and Thompson thus acknowledge that describing a causal relation between two events does not inherently make either cause or effect more salient than the other. But is such an inherent ascription warranted for the other subject-matter relations? Given the list of factors contributing to nuclearity decisions in Section 3, it is not clear that, for any instance of a relation such as Interpretation or Evaluation, the interpreted or evaluated segment would consistently be more important than the interpreting/evaluating one. Consider again our example text in Figure 1. If the text were continued with a sentence like For me, eating and drinking have always been the major attraction at picnics, the weight distribution for the preceding three segments would change, because now it turns out that the beer-and-sausage sentence is more relevant for the text function than the Sequence segment giving the timing of the trip. Similarly, for the Interpretation relation, it is not difficult to conceive contexts that render the interpretation more prominent than what is interpreted. With other relations, predefining the nuclearity assignment is equally problematic. A case in point (noted also by Bateman and Rondhuis (1997)) is the Purpose

.  This does not mean that intentions can be entirely absent; it is to be understood in terms of the distinction drawn between subject-matter and presentational relations by Mann and Thompson (and also by other authors), as sketched at the very beginning of the paper.



 Manfred Stede

relation, where according to the RST definition the underlying goal of the activity is the satellite and the activity itself is the nucleus. But, given a sentence like (10), why would buying the sports car and impressing the girl-friend not be equally able to play the role of nucleus? (10) Jim bought a red sports car to impress his new girl-friend. Adopting again the argument of continuation possibilities, both elements can easily be picked up by a subsequent sentence – even more so when the linear order of the Purpose sentence is in line with the continuation. Or, independent of a particular continuation, either the sports car or the necessity of impressing the girl-friend can be the main topic of the larger discourse unit, so that criterion (a) from Section 3 would favour it as nucleus. It is evident that the elements of a Purpose relation are clearly distinguishable on semantic grounds, but this should not entail a strict mapping to a nuclearity assignment. (11) If I win the lottery next week, the first thing I’ll buy is a red sports car.

The same point can be made for the Condition relation. In (11), the two clauses clearly express propositions of different epistemic status, and thus the relation is “asymmetric” – but it is not clear in what sense the two clauses should be distinguished on the grounds of a general notion of nuclearity. Example (9) above, at least, demonstrates a difference in salience that is clearly not the same as the one found in such Condition examples. And besides, the “continuation argument” applies to Condition just as well: The following clause can either elaborate the chances of winning the lottery, or the prospective purchase of a sports car (or neither one). Mann and Thompson apparently saw these problems and pointed out that the definitions of relations such as Condition and Purpose would specify as “locus of effect” both the nucleus and the satellite, and therefore the deletion test would not produce satisfactory results for them (which is correct). Resorting to a “locus of effect”, however, brings a new parameter into play and in effect undermines the general nature of the nuclearity idea. It seems more appropriate to question the strict association between (subject-matter) relations and nuclearity assignment altogether. Finally, while Condition and Purpose are clearly marked at the linguistic surface and thus do not create ambiguity for annotators, the situation is different with the aforementioned Interpretation and Evaluation. We pointed out that their fixed nucleus assignment can create conflicts in the annotation process. Notice however, that these relations are much less “visible” at the surface and hence rely more on context-driven interpretation. Unfortunately, when the definition of Interpretation or Evaluation is met for a text span, chances are that other relation definitions are also met, e.g., the notorious Background or the rather vague Circumstance. These two offer a way to turn the nucleus assignment around, and thus annotators can select a “weaker” relation rather than the specific Evaluation/Interpretation, merely in order to achieve a suitable nucleus assignment. This might serve to get the overall discourse structure right (in

RST revisited

terms of degrees of nuclearity of segments; see above), but it is clearly not optimal for describing local coherence.

4.3  Enforced nuclearity One source of complaints from RST annotators is the requirement that nucleus assignment is a “must” for every pair of segments, unless one wants to resort to the multinuclear Joint relation, which is void of any specific content. On the level of minimal units, such cases are encountered, for example, in descriptive texts that characterize various aspects of a topic in an enumerative way. Among those aspects, two may be causally related, but there need not be a consequence of the causality for the writer’s intentions, nor for thematic continuity. The relationship can be marked by a causative adverbial, leaving the surface structure “neutral”:8 (12) Th  e courtyard was very quiet that day. Two teenagers read their books in one ­corner, and a new car was parked beside them. The trees still had all their leaves, and thus the flowers were blooming in a cool shade. A little bench stood in the middle of the lawn. (…)

Teenagers, car, trees, flowers and bench all contribute to the description. The marked causal relation between the third and the fourth clause leaves no doubt as to the “semantic directionality” from cause to effect, but from a textual perspective, it appears to be an artificial move to grant either of the two clauses a nuclear status, for on what grounds would it be more central to the writer’s purposes? Accordingly, an annotator would struggle here to decide between RST’s Cause and Result relations. Recall that the relation set used by Carlson et al. (2003) employs many more multinuclear relations than the original RST set, which indicates that those researchers encountered the same problem (for our specific example, a multinuclear Cause-Consequence relation is provided by Carlson et al.). As indicated earlier, Mann & Thompson (1988) acknowledged that envelope structures (conventionalized beginnings and endings) and certain parallel structures of comparison and contrast are to be exempt from the need to assign nuclei. My sugges­tion here is that this list needs to be extended, which questions the explanatory power of the nuclearity notion. Besides some problematic cases on the level of minimal units as just illustrated, the problem is also prominent higher up in the tree structure, between larger segments, where structural clues indicating a salience distribution are often absent. The severity of this problem differs along the dimension of text type (in the sense of, e.g., Werlich 1975), or what Smith (2003) calls discourse mode. For

.  In German, where a causative adverbial such as deshalb can float quite freely throughout the clause, the impression of perceived “neutrality”, or the lack of a structure-induced weighting, might be somewhat clearer than in English.



 Manfred Stede

argumentative text, the overall structure is often characterized quite well by the RST presentational relations and their nuclearity assignments, which reflect the relative weight and hierarchical structure of the points argued by the author. For narrative, presentational relations are much less relevant, and all the problems with subjectmatter relations, as discussed in the previous section, become especially prominent. A central aspect in understanding narrative is reconstructing the temporal structure of reported events, which cannot be characterized sufficiently with RST relations. And relative “centrality to writer’s purposes” is often not an issue, as illustrated in example (12) above. A similar judgement applies to expository text, whose function is to explain some (concrete or abstract) entity to the reader, by outlining its decomposition into parts, its function, and the like. Here, the reader is typically invited to form a mental image not of a temporal event sequence but of a spatial configuration of objects (even if they are abstract). As with narrative, the linearity of the exposition is very important for ease of processing, while internal structure according to “importance” is only occasionally relevant; nuclearity thus does not explain nearly as much as it does in argumentative text, where the linear order is much less critical.9

5.  Multi-level discourse representation and annotation Researchers investigating the different text types (cf. the citations above) point out that, on the one hand, the types can be distinguished by (relative frequencies of) individual surface-linguistic features, and, on the other, coherence is created in different ways. For example, both Lötscher (1987) and Smith (2003) characterize type-specific strategies of thematic development. And we noted above that the overall function of a text varies according to the type: convey the temporal structure of events; convince the reader by presenting an argument; enable the reader to form a mental image of some, possibly complicated, object or state of affairs, etc., and along with the text function varies the shape of the functional description of the text, which is supposed to capture the phenomena that make the reader experience the text as coherent. To some extent, this can be explained in terms of RST relations: While argumentative text can be characterized by presentational relations, expository text features much fewer instances of those and instead is rich in Elaboration, as the text moves from (sub-) topic to (sub-) topic. In narrative, we might expect an abundance of Sequence relations, but of course the temporal structure can be much more complicated and require different means of

.  We conducted many re-ordering experiments with the texts in the Potsdam Commentary Corpus, indicating that the text function is relatively independent of the linear order of the ­segments, as long as referential chains are not broken. Roughly speaking: In presenting your argument, you have quite a bit of freedom in sequencing and arranging your points.

RST revisited

representation (see, e.g., Mani & Pustejovsky 2004). Thus, different text types favour different means of describing text structure. However, unlike genre, text type is not a category that texts typically belong to in their entirety. While for most texts we can clearly state that it is a recipe, a news report, an instruction manual, etc. (genre), it is relatively rare for a text to be an instance of just one type. Narratives typically contain descriptive, expository or portions of other types; a largely argumentative text may contain a short narration of events that are then being commented on, and so forth. Therefore, a single type-specific description will normally not be sufficient to cover a text. Instead, we need arrangements of descriptive devices that are able to characterize the various dimensions of text coherence simul­taneously. One undisputed facet of this notion is the long-recognized fact that both relation-based and reference-based means play a role in creating coherence. Then, taking the step from co-reference to the more abstract level of thematic development, Knott et al. (2001) voiced their discomfort with RST’s mixing the more “standard” coherence relations with Elaboration (in particular, the object-attribute variant), which does not convey a relationship between propositions, but instead indicates a shifting focus of attention. Accordingly, Knott et al. dismissed that coherence relation and introduced a different representational device, which leads to “entity chains”: Individual portions of the text are captured with standard RST trees, and these portions are linked by a focus shift indicator. My proposal is to take a further step and to seek a text representation that distinguishes various dimensions of coherence from one another. Whereas Knott et al. use a single representation that alternates between RST relations and topic shifts, it seems more appropriate to consider the dimensions in parallel. After all, some referential phenomena are bound to occur also within an RST-portion of an entity chain in the style of Knott et al. (2001) – thematic development and RST-like relations are by no means mutually exclusive. Similarly, temporal relations can easily overlap with nontemporal ones: (13) Tim’s performance at the piano was impeccable. When he had walked onto the stage, people seemed sceptical, but as soon as he finished the sonata, the whole audience burst into applause.

Understanding (13) involves both reconstructing the right order of events (on the basis of verb tense and with the help of world knowledge) and grasping the contrast between the “before” and “after” situations. Motivated by related observations, Wolf and Gibson (2005) responded to the multifaceted nature of local coherence by dismissing RST’s assumption that text structure be represented by a tree; instead, they allow coherence relations to quite freely connect nonadjacent segments, which can also lead to cross-dependencies. Importantly, though, they hold on to coherence relations as the single representational device, merely substituting a graph for the constrained tree structure. It seems to me that incrementally adding more relations, as phenomena are being recognized, is not a promising direction. Rather,



 Manfred Stede

one should explicitly distinguish the various dimensions of discourse structure, so that, for example, the phenomena listed in Section 3 can be attributed to their respective realms. Incidentally, in one RST paper, Mann & Matthiessen (1991) point in this direction when they relate RST relations to the metafunctions of systemic-functional linguistics10 (SFL, Halliday 2004), but they neglect the possibility of allowing different realms of relationships to operate simultaneously, which is the underlying idea of characterizing sentence structure in SFL. Bateman & Rondhuis (1997) took the step to decompose rhetorical relations systematically along the three metafunctions (interpersonal, ideational, textual). A more cautious step in this direction had also been taken by Moore & Pollack (1992) with their suggestion that text needs to be analyzed with subject-matter and presentational relations in parallel, which can possibly lead to two divergent tree structures. However, the thesis that both types of relations always apply simultaneously appears to be too strong, unless one includes in the relation set some very unspecific relations that can be used in cases where just one of the two groups is clearly the appropriate one. In order to make progress with defining an explanatory set of representation levels that accounts for the relevant dimensions of coherence, I propose a data-oriented approach that emphasizes systematic annotation of authentic text, so that ideas can be tested immediately with “real” data and be revised when necessary.

5.1  Annotation framework Research on discourse (in particular: text) structure can be broadly divided into two kinds of approaches: –

–

Work that aims at extending existing theories of sentence syntax and/or semantics to the discourse level. Emphasis is on rigorous formalization and faithfulness to the theories. Usually, such approaches work with constructed examples and are gradually extended to cover more phenomena and, eventually, gain a certain coverage of “real” data (corpora of actual language use). Work that aims at handling “real” data with as much coverage as possible, thus emphasizing an empirical foundation. These accounts suffer from the fact that the theories are much less formalized and definitions are often vague.

Representatives of the first approach are SDRT (Asher & Lascarides 2003), D-LTAG (Webber et al. 2003), and LDM (Polanyi 1988). For the second approach, RST has probably been the most influential; recently, an alternative corpus-oriented approach has been proposed by Wolf & Gibson (2005). In general, recent years have seen a surge of interest in annotating text corpora with discourse-related information. One example is the Penn Discourse TreeBank (Prasad et al. 2004) that builds on a popular Wall Street Journal corpus and provides annotations of connectives and their arguments. For RST,

.  The authors correlate presentational relations with the interpersonal, and subject-matter relations with the ideational metafunction; as for the textual metafunction, they vaguely indicate that the order of text spans in a relation functions textually.

RST revisited

the RST Treebank (Carlson et al. 2003) has been made available, which also covers Wall Street Journal text. These are important steps, because discourse research urgently needs an impetus similar to that which sentence-related annotations (so-called sentence treebanks) have given to syntax research. Such investment of annotation efforts should be carefully reflected, though. Text is a highly multi-faceted means of transmitting information (in the broadest sense), and our discussion of various salience phenomena was meant to illustrate that. Annotations according to RST are a good starting point for exploring coherence phenomena with “real” data, but they clearly have their limitations, as the discussion in the preceding sections demonstrated. Building RST trees is a very complex task that asks annotators to make a range of quite different decisions (segmentation, nuclearity assignment, relation choice) whose individual results are then amalgamated into a single representation. In our experience, annotators are quite often unhappy after finishing a text analysis because of the many ambiguities encountered along the way. In many cases, one could just as well have decided otherwise, and the alternative result would have been equally defendable in terms of the RST definitions. The trouble is that those choice points are all hidden in the tree, not allowing the inspection of the various individual micro-decisions and for considering possible consequences of alternative answers to them. In order to make text annotations maximally useful for research purposes, it therefore seems more helpful to separate the various phenomena from each other and annotate them individually, in the shape of a systematic multi-level annotation (henceforth MLA, see Stede 2007). These levels can then be “mined” for correlations and inform the search for more comprehensive theories of discourse representation and processing. Until recently, this was extremely cumbersome to do (if possible at all), but now with the advances in XML-based software technology, creating and querying MLA corpora is a realistic and fruitful research avenue. One possibility is the architecture developed at Potsdam University, with a generic data exchange format for diverse, task-specific annotation tools and a database that reads and aligns the individual annotations, and allows for manual cross-level analyses, as well as statistical evaluations.11 Importantly, our MLA framework is flexible enough to allow for new levels to be added incrementally to an existing corpus. In the following, I sketch one possible configuration of annotation levels, which is at present being implemented for the Potsdam Commentary Corpus (PCC, Stede 2004), and then I relate this to RST.

5.2  Levels of annotation Sentence syntax. A representation of syntactic structure serves as building blocks for the subsequent levels. In PCC, we use the TIGER schema (Brants et al. 2002), which is designed to be as theory-neutral as possible. It includes a constituent structure as well

.  http://www.ling.uni-potsdam.de/~stede/MLA.html. Technical details on the database can be found in Dipper et al. (2004).



 Manfred Stede

as information on grammatical functions. The dedicated annotation tool that allows for efficient, semi-automatic construction of syntax trees is Annotate.12 Referential structure. Besides discourse relations, co-reference is generally taken to be the most important source of coherence in text. For the time being, our annotations are restricted to links between anaphors (pronouns and definite descriptions) and nominal antecedents; both “bridging” relations and event anaphora are excluded. The units related here – referring expressions – must be licensed by the level of sentential syntax. Technically, the syntax trees are automatically mapped to the input format for the annotation tool (MMAX2,13 specifically designed for co-reference) so that the range of possible “markables” is already presented to the annotator. Thematic structure. As a step of abstraction over the referential chains, we partition the text into zones that annotators identify as dealing with distinct topics. These zones can be embedded, and they can bear subtopic relations to one another, which the annotators assign. This builds on the idea of the “focus stack” used in the attentional structure representation of Grosz and Sidner (1986). In addition, a specific label “Formulaic” is reserved for meta-utterances whose topic is the text itself (e.g., the first clause of sentence 4 of the text in Figure 1) or that contain other, genre-specific, conventionalized formulae – in other words, for utterances that do not have a standard, “content” topic. In future work, the level of thematic structure will be linked to our work on annotating information structure in sentences (Dipper et al. 2007). Conjunctive relations. Along the lines of Martin (1992), this level identifies the “minimal units” of higher-level discourse structure, as well as the connectives linking those units in the text. From the relations proposed by Martin, we chose 10 abstract ones, and hence a set much smaller than that of RST, for labelling the type of connection. Following Martin, this does not include any salience/nuclearity assignment, but relations are in general directed to reflect the (semantically) different roles of the units.14 The range of possible minimal units is not arbitrary but constrained by constituents built on the sentential syntax level.15 For every connective (as defined by the criteria in Pasch et al. (2003: 331), and adding certain prepositions), annotators can mark its semantic scope (similar to the PDTB, Prasad et al. 2004), but they can also leave one or both segments unspecified, if they find it hard to decide on the scope of an

.  http://www.coli.uni‑saarland.de/projects/sfb378/negra‑corpus/annotate.html .  http://www.eml-research.de/english/research/nlp/download/mmax.php ���.  Bateman (2001) proposed a discourse representation combining RST and Conjunctive ­Relations, where the former is in charge of capturing abstract intentions and content, and the latter serves to model local cohesion and thematic development; hence, Bateman’s goal is somewhat similar to ours. .  See Dinesh et al. (2005) for a discussion of possible mismatches between the units of ­syntactic analysis and discourse structure.

RST revisited

adverbial connective such as so. Technically, we have implemented a dedicated annotation tool that allows for marking connectives and scopes semi-automatically (Stede & Heintze 2004). Intentional structure. This most abstract level of representation largely corresponds to the “presentational” relations of RST, with exceptions such as our treatment of Evaluation (see below). Thus it is usually only a partial structure covering not the entire text but only those portions where segments are indeed related via “dominance” to achieve a particular intention. Marking this can involve introducing new segments (larger than minimal units), which have not been found at the conjunctive-relation level because there is no connective present. Technically, we use RSTTool for this step.16

5.3  Example For the short text given in Figure 1, the annotation would proceed as follows. After the syntactic analysis of the individual sentences, co-referential NPs are being identified, which in this case leads only to chain I (sentence 2), we (3), we (4), elided we (5). The event anaphor that (6) and the bridging connections (weather – forecast, afternoon – 1pm, picnic – beer and sausages) are at present not being represented. On the thematic level, the text is divided into segments (1–2), dealing with weather, and (3–6), dealing with picnic. The latter has a subtopic segment (4–5), timing. The assignment of conjunctive relations is a flat, surface-oriented annotation, which here identifies the relations Addition (1, 2), Consequence (1–2, 3), Temporal-successive (4, 5) and Addition (4–5, 6). Finally, the intentional structure corresponds in part to the presentational relations of the RST tree: Evidence (2, 1) and Motivation (1–2, 3). As regards RST-Evaluation, notice that in the example (and in many other cases), there is no clear “dominance” relation between the evaluated and the evaluating segment. In this case, we would annotate 6 as dominating (4–5), in contrast to the RST nuclearity assignment.

5.4  RST trees versus multi-level representation For a short and “friendly” text such as the one in Figure 1, MLA might not seem particularly advantageous when compared to the RST tree, which serves as a compact representation of the text structure and also indicates its function (by showing the “most-nuclear unit”, here 3). But recall, for one thing, that MLA is richer in information (syntax, co-reference), which can be used to analyse correlations between levels. Also, the partial intentional structure of MLA conveys the most “important” segments: the roots of the individual intention-based trees, here 3 and 6. The fact that these two are disconnected seems to do justice to the text, as neither of the two speech acts supports the other.

.  http://www.wagsoft.com/RSTTool



 Manfred Stede

More importantly, the picture changes when we turn to the realistic case of longer texts. In general, when comparing an RST analysis with a representation following MLA, at first sight two kinds of information appear to be missing in the latter: the subject-matter relations, and the existence of a complete tree structure. As for the subject-matter relations, they are in part replaced by the conjunctive relations. These are explicitly intended to be less specific than the RST relations, which have been criticized for their essentially open-ended nature, for example by Grosz and Sidner (1986). After all, the range of possible semantic relationships between events in the world or within people’s minds is not likely to be small. From the discourse perspective, seeking a “complete” inventory of relationships seems neither fruitful nor necessary: In terms of discourse function, reporting a semantic relation between two eventualities is not so different from reporting a single eventuality, so that a fine-grained inventory of semantic relations would not correspond to an equally fine-grained inventory of consequences for text coherence. In short, a dedicated inventory of subject-matter (or semantic) relations seems to belong to the realm of domain knowledge but not to that of text-oriented coherence relations. The loss of a complete tree, which after all was supposed to be the main criterion for coherence postulated by RST, might look like a rather dramatic move. However, the completeness of an RST tree (or that of other discourse representation theories) is often achieved with the help of coherence relations of a somewhat dubious status. RST offers the multinuclear Joint relation for cases where a more specific connection cannot be found; the main reason why Joint is relatively rarely used seems to be the wide applicability of the Elaboration relation, which, as annotators report, fits in very many situations. Elaboration, however, has the disadvantage of masking exactly what is being elaborated on. Typically it is not a relationship between the propositions or eventualities, but between individual entities taking part in those eventualities. As noted by Knott et al. (2001), this amounts to a regular shift of focus of attention, which should be represented as such – in our framework on the levels of referential structure (exactly which entities are co-referent?) and thematic structure (does the overall topic change or stay the same?). The latter has the advantage of explicitly modelling the flow of topics through the discourse, which cannot be read off an RST tree. And recall that we emphasized the need for representing focus shift and coherence relations in parallel (in contrast to the suggestion by Knott et al.), because focus shifts occur throughout the text, and not just in places where no intentional or subject-mater relation is present. In the version of MLA sketched above, the hierarchy information encoded in an RST tree is distributed to three levels: the partial intentional structure (where the role of the hierarchy is largely in analogy to that of an RST tree), the flat segment annotation on the conjunctive-relations level, and the partitioning into content zones, which allow for embedding. The claim here is that this tripartite representation is in fact an advantage, as RST annotators regularly report that the precise assignment of the single hierarchy is very difficult and time-consuming – and the result is typically ambiguous anyway.

RST revisited

When we proceed to merge the three levels (which have been annotated individually) into a single data structure, the result is, in principle, a graph that in various ways violates the constraints of a tree. There can be multiple roots, nodes can have multiple parents, and there may be crossing edges (which is disallowed in RST). To what extent these violations are either annotation artefacts, or indeed reflections of phenomena in the linguistic structure, is subject to empirical investigation. From the corpus-oriented perspective, we should decide on the most appropriate formal shape of text structure not based on a priori considerations of elegance or processing efficiency, but based on a comprehensive picture of the variety of structuring phenomena in discourse, given analyses of authentic data. We argued that the main goal of the multi-level approach is to separate distinct kinds of information from one another, so that they can be individually annotated (resulting in easier coding procedures) and reviewed, and ultimately set into correspondence with one another. Obviously, the amount of work necessary to produce such annotations is larger than with an RST tree, but the gain in useful information should outweigh that additional effort. In other words, a smaller number of texts annotated with a multi-level approach are likely to be more useful for discourse research than a larger number of RST trees, which are prone to the problems indicated throughout this paper.

6.  Conclusion: Salience on different levels of description The possible relationships between distinct levels of discourse representation had been widely discussed in the computational linguistics community in the early 1990s; it had been moved forward especially by research in text generation, where the successive mapping from deep representations of knowledge to the surface text must be made explicit. This discussion ended without a clear result, as the attention shifted – in accordance with the general trend in computational linguistics – to decidedly surface-based methods of discourse processing, then focusing on understanding rather than generation (e.g., Marcu 2000). Today, it is opportune to re-open the discussion of levels and give it a decidedly data-oriented twist: Developments in annotation and retrieval software over the past years have made it possible to systematically collect text annotations on different levels and relate them to one another, either manually, or by employing statistical techniques of pattern recognition. Annotated corpora (in particular the so-called treebanks) have tremendously helped sentence-oriented research over the past decade; on the text level, this development has only started. It will, among other things, enable us to investigate research hypotheses on level interactions that have been put forward, for example, for rhetorical structure and constituent ordering, rhetorical structure and co-reference, or discourse structure and sentential information structure. Finally, we need to consider what happens to the idea of nuclearity in the proposed MLA approach. To some extent, it is being preserved in its original flavour, namely on the level of intentional structure. Here, it should also be possible to identify



 Manfred Stede

the central units of a text (with respect to its overall purpose), as it has been mentioned as an advantage of RST trees. In text portions where intention-based nuclearity is absent, salience can arise from other levels. The referential chains indicate what are the prominent entities under discussion in the text, and in which units they are being mentioned; the thematic structure identifies meta-discursive statements or formulaic portions, as well as temporary topic switches (digressions); and the syntactic structure identifies subordinate clauses, which might generate a mood of non-salience in case none of the other levels points to a different effect. Nuclearity thus becomes an epiphenomenon resulting from different types of sources of salience and non-salience – but it might just as well be absent in certain regions of the text, where segments are equally salient to the author’s purposes.

References Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation. Cambridge: CUP. Bateman, John. 2001. Between the leaves of rhetorical structure: static and dynamic aspects of discourse organization. Verbum 23(1): 31–58. Bateman, John & Rondhuis, Jan-Klaas. 1997. ‘Coherence relations’: Towards a general specification. Discourse Processes 24: 3–49. Brants, Sabine, Dipper, Stefanie, Hansen, Silvia, Lezius, Wolfgang & Smith, George. 2002. The TIGER Treebank. In Proceedings of the Workshop on Treebanks and Linguistic Theories, Sozopol, 24–41. Carlson, Lynn, Marcu, Daniel & Okurowski, Mary Ellen. 2003. Building a discourse-tagged corpus in the framework of rhetorical structure theory. In Current Directions in Discourse and Dialogue, Jan van Kuppevelt & Ronnie Smith (eds), 85–112. Dordrecht: Kluwer. Cristea, Dan, Ide, Nancy & Romary, Laurent. 1998. Veins Theory: A model of global discourse cohesion and coherence. In Proceedings of COLING/ACL’98, Montréal, 281–285. Danlos, Laurence. 2008. Strong generative capacity of RST, SDRT and dependency DAGs for Discourse. To appear in Constraints in Discourse, Anton Benz & Peter Kühnlein (eds). Amsterdam: John Benjamins. Dinesh, Nikhil, Lee, Alan, Miltsakaki, Eleni, Prasad, Rashmi, Joshi, Aravind & Webber, Bonnie. 2005. Attribution and the (Non-) Alignment of Syntactic and Discourse Arguments of Connectives. In Proceedings of the ACL Workshop “Frontiers in Corpus Annotation”, 29–36. Ann Arbor MI. Dipper, Stefanie, Götze, Michael, Stede, Manfred & Wegst, Tillmann. 2004. ANNIS: A Linguistic Database for Exploring Information Structure. In Working Papers of the SFB632, Interdisciplinary Studies on Information Structure (ISIS) 1, Shinichiro Ishihara, Michaela Schmitz & Anne Schwarz (eds), 245–279. Potsdam: Universitätsverlag. Dipper, Stefanie, Götze, Michael & Skopeteas, Stavros (eds). 2007. Information structure in crosslinguistic corpora. Interdisciplinary Studies on Information Structure (ISIS) 9. Potsdam: Universitätsverlag. Elhadad, Michael & McKeown, Kathy. 1990. Generating Connectives. In Proceedings of the 13th International Conference on Computational Linguistics (COLING), 97–101.Helsinki.

RST revisited Fries, Peter. 1981. On the status of theme in English: Arguments from discourse. Forum Linguisticum VI(1): 1–38. Golebiowski, Zofia. 2006. The distribution of discoursal salience in research papers: relational hypotaxis and parataxis. Discourse Studies 8(2): 259–278. Grosz, Barbara & Sidner, Candace. 1986. Attention, Intention and the Structure of Discourse. Computational Linguistics 12(3):175–204. Grote, Brigitte, Lenke, Nils & Stede, Manfred. 1997. Ma(r)king concessions in English and German. Discourse Processes 24(1): 87–118. Halliday, Michael. 2004. An introduction to functional grammar, 3rd edn. London: Arnold. Hobbs, Jerry. 1979. Coherence and co-reference. Cognitive Science 1:67–82. Kehler, Andrew. 2002. Coherence, Reference and the Theory of Grammar. Stanford CA: CSLI. Knott, Alistair, Oberlander, Jon, O’Donnell, Michael & Mellish, Chris. 2001. Beyond elaboration: The interaction of relations and focus in coherent text. In Text Representation: Linguistic and Psycholinguistic Aspects, Ted Sanders, Jost Schilperoord & Wilbert Spooren (eds), 181–196. Amsterdam: John Benjamins. Lötscher, Andreas. 1987. Text und Thema. Studien zur thematischen Konstituenz von Texten. Tübingen: Niemeyer. Mani, Inderjeet & Pustejovsky, James. 2004. Temporal discourse models for narrative structure. In Proceedings of the ACL Workshop on Discourse Annotation, 57–64. Barcelona. Mann, William & Matthiessen, Christian. 1991. Functions of language in two frameworks. Word 42(3): 231–249. Mann, William & Thompson, Sandra. 1988. Rhetorical structure theory: Towards a functional theory of text organization. Text 8 (3): 243–281. Marcu, Daniel. 1999. Discourse trees are good indicators of importance in text. In Advances in Automatic Text Summarization, Inderjeet Mani & Mark Maybury (eds), 123–136. Cambridge MA: The MIT Press. Marcu, Daniel. 2000. The Theory and Practice of Discourse Parsing and Summarization. Cambridge MA: The MIT Press. Martin, James. 1992. English Text – System and Structure. Amsterdam: John Benjamins. Matthiessen, Christian & Thompson, Sandra. 1988. The structure of discourse and “subordination”. In Clause Combining in Discourse and Grammar, John Haiman & Sandra Thompson (eds), 275–330. Amsterdam: John Benjamins. Moore, Johanna & Pollack, Martha. 1992. A problem for RST: The need for multi-level discourse analysis. Computational Linguistics 18(4): 537–544. Moser, Megan & Moore, Johanna. 1995. Using discourse analysis and automatic text generation to study discourse cue usage. In AAAI Spring Symposium on Empirical Methods in Discourse Interpretation and Generation, 92–98. Stanford University. Pasch, Renate, Brauße, Ursula, Breindl, Eva & Waßner, Hermann Ulrich. 2003. Handbuch der deutschen Konnektoren. Berlin: Walter de Gruyter. Polanyi, Livia. 1988. A formal model of the structure of discourse. Journal of Pragmatics 12: 601–638. Prasad, Rashmi, Miltsakaki, Eleni, Joshi, Aravind & Webber, Bonnie. 2004. Annotation and data mining of the Penn Discourse TreeBank. In Proceedings of the ACL Workshop on Discourse Annotation, 88–95. Barcelona. Sanders, Ted, Spooren, Wilbert & Noordman, Leo. 1992. Towards a taxonomy of coherence relations. Discourse Processes 15:1–35.



 Manfred Stede Schilder, Frank & Tenbrink, Thora. 2001. Before and after: sentence-internal and -external discourse relations. In Proceedings of the Workshop From Sentence Processing to Discourse Interpretation: Crossing the Borders, 1–10. Utrecht. Smith, Carlota. 2003. Modes of Discourse. The Local Structure of Texts. Cambridge: CUP. Stede, Manfred. 2004. The Potsdam Commentary Corpus. In Proceedings of the ACL Workshop on Discourse Annotation, 96–102. Barcelona. Stede, Manfred. 2004b. Kontrast im Diskurs. In Brücken schlagen – Grundlagen der Konnektorensemantik, Hardarik Blühdorn, Eva Breindl & Hermann Ulrich Waßner (eds), 255–286. Berlin: deGruyter. Stede, Manfred. 2007. Korpusgestütze Textanalyse. Grunalzüge der Ebenen-orientierte Textlinguistik. Tübingen: Narr. Stede, Manfred & Heintze, Silvan. 2004. Machine-assisted rhetorical structure annotation. In Proceedings of the International Conference on Computational Linguistics (Coling-04), 425– 431.Geneva. de Swart, Henriette. 1999. Position and meaning. Time adverbials in context. In Focus. Linguistic, Cognitive, and Computational Perspectives, Peter Bosch & Rob van der Sandt (eds), 336–361. Cambridge: CUP. Taboada, Maite & Mann, William. 2006. Rhetorical structure theory: Looking back and moving ahead. Discourse Studies 8(3): 423–459. Webber, Bonnie, Joshi, Aravind, Stone, Matthew & Knott, Alistair. 2003. Anaphora and discourse structure. Computational Linguistics 29(4): 545–587. Werlich, Egon. 1975. Typologie der Texte. Heidelberg: Quelle und Meyer. Wolf, Florian & Gibson, Edward. 2005. Representing discourse coherence: A corpus-based analysis. Computational Linguistics 31(2): 249–287.

Subordination and coordination in syntax, semantics and discourse Evidence from the study of connectives1 Hardarik Blühdorn

Institut für Deutsche Sprache, Mannheim This article discusses the question whether the distinction between subordination and coordination is parallel in syntax and discourse. Its main thesis is that subordination and coordination, as they are commonly understood in the linguistic literature, are genuinely syntactic concepts. The distinction between hierarchical and non-hierarchical connection in discourse structure, as far as it is defined clearly in the literature, is of a quite different nature. The syntax and semantics of connectives (as the most prominent morphosyntactic means by which subordination and coordination are encoded) offers little evidence to support the assumption of a structural parallelism between syntax and discourse. As a methodological consequence, sentence and discourse structure should not be mixed up in linguistic analysis. Keywords: Coherence, relation, conjunction, adposition, adverbial connective.

1.  Introduction The distinction between linguistic subordination and coordination, i.e., between hierarchical and non-hierarchical connection in language, has been investigated during the last decades by many authors within various research paradigms. Both types of connection have been examined on sentence level as well as on the level of text and discourse (for syntax-oriented approaches see e.g., Foley & Van Valin 1984: 238ff; Shopen 1985; Wesche 1995; Kortmann 1996; Haumann 1997; Van Valin & LaPolla 1997: 441ff; Johannessen 1998; Cristofaro 2003; Haspelmath (ed.) 2004; for discourse-oriented approaches see e.g., Thompson & Longacre 1985; Mann & Thompson 1988; Polanyi 1988; Günthner 1996; Lefèvre 2000; Asher & Vieu 2005).

.  I am grateful to Bernd Wiese, Anke Holler, Manfred Stede, Ingolfur Blühdorn, Marina Foschi Albert, and Cathrine Fabricius-Hansen, as well as to three anonymous reviewers for their helpful comments on earlier versions of this paper.

 Hardarik Blühdorn

One of the questions addressed in the literature concerns the relation between syntactic connection and discourse connection. Should it be assumed that hierarchical connections of clauses reflect hierarchical connections of discourse units and that non-hierarchical connections of clauses reflect non-hierarchical connections of discourse units (see Hopper & Thompson 1984: 736f; Quirk et al. 1985: 919f; Matthiessen & Thompson 1988; O’Dowd 1992; Schecker 2000; Wegener 2000; Cristofaro 2003: 45ff)? Or should it be assumed that the two domains of connection are structurally independent of each other? The present article contributes to the theoretical discussion of this question from the point of view of the study of connectives, the examples being taken from Modern German. In many languages, connectives (“linkers”; see Quirk et al. 1985: 921) are among the most important means used to establish subordinative and coordinative relations in syntax. Most theories of discourse structure pay particular attention to their role in establishing coherence relations (see e.g., Halliday & Hasan 1976: 226ff; Polanyi 1988: 605; Knott & Dale 1994: 45ff; Knott et al. 2001). In terms of traditional grammar, they belong to the following word classes: subordinating and coordinating conjunctions, adpositions, and adverbs (see Pasch et al. 2003: 38ff; Blühdorn 2007a). There are several other means that serve to encode syntactic subordination in the languages of the world: complementizers, relative pronouns and relative particles, specialized converbs, non-finite verb forms like infinitives, gerunds and participles, as well as inflectional case forms like locatives, instrumentals or ablatives; coordinative relations may also be encoded by asyndetic juxtaposition, plurals or collectives (see O’Dowd 1992; Muller 1996; Kortmann 1996: 5, 73; Zifonun 2001; Cristofaro 2003: 51ff; Heath 2004; Breindl & Waßner 2006; Breindl 2007a, b). These other means will not be looked at in this article, the focus being on prototypical coordination and on adjunct relations (adverbial adjuncts and adverbial subordinate clauses). In particular, I will not be interested in complement relations (subject or object clauses), nor in attributive relations (relative clauses), which are not connective relations in the sense in which the term is used here (see Pasch et al. 2003: 1ff, 38f; on complements and adjuncts see Bierwisch 2003). The research project team Handbuch der deutschen Konnektoren (HdK) at the Institut für Deutsche Sprache in Mannheim has been investigating the syntax and semantics of German connectives for several years (see Pasch et al. 2003; Blühdorn et al. 2004; Pasch 2004). One of the recurring issues discussed in that work has been the syntactic and semantic nature of subordination and coordination (see Pasch et al. 2003: 230ff, 267ff). Based on experience from the HdK project, the thesis to be presented in this paper is: The study of connectives offers little evidence in favour of a general structural parallelism between hierarchical and non-hierarchical connections in syntax and

Subordination and coordination: Evidence from connectives

discourse. Rather, it should be assumed that hierarchical as well as nonhierarchical discourse relations may, in principle, be encoded by both coordination and subordination in syntax. Whatever the relevant factors that control the choice between syntactic subordination and coordination, they should not be identified with the distinction between hierarchical and non-hierarchical connection in discourse.

In support of this thesis, I will present three main arguments: i.

The syntactic distinction between coordination and subordination is neutralized at levels higher than the sentence. On the levels of text and discourse it does not play any relevant role (section 2). ii. In semantics, similarly as in syntax, we can distinguish between hierarchical and non-hierarchical connections. Both types of semantic connection can be encoded by both types of syntactic connection (section 3). iii. Syntax and semantics provide similar models for non-hierarchical connection, but contrasting models for hierarchical connection. There is no reason to believe that the structure of hierarchical relations in discourse should be generally more similar to hierarchical relations in syntax than to hierarchical relations in semantics (section 4). In syntax, hierarchical connection of clauses is traditionally called subordination, and non-hierarchical connection of clauses is called coordination. In line with this tradition, I will use the terms subordination and coordination for the two main types of connection in syntax. In semantics, I will distinguish between symmetrical (non-hierarchical) and asymmetric (hierarchical) connection of conceptual entities (spatial objects, events, propositions, or acts). The terms hierarchical and non-hierarchical themselves will be reserved for the connection of rhetorical units (utterances and speech acts) in discourse. The main question of this article can therefore be reformulated as follows: Is there a parallelism between coordinative vs. subordinative connection of clauses in syntax, symmetrical vs. asymmetric connection of concepts in semantics and non-hierarchical vs. hierarchical connection of rhetorical units in discourse? Or are these three levels of connection independent of each other?

In section 2, this question is approached from a syntactic point of view, in section 3 from a semantic point of view, and in section 4 from the point of view of discourse. Section 5 summarizes the conclusions to be drawn from the arguments presented.

2.  Syntactic connections Recent studies on syntactic coordination and subordination in the languages of the world suggest that there is no sharp dichotomic distinction between these two types of



 Hardarik Blühdorn

connection (see Cristofaro 2003: 15ff). Rather, they should be viewed as prototypical poles on some sort of gradient (see Quirk et al. 1985: 927f; O’Dowd 1992: 68f; Kortmann 1996: 56ff; Johannessen 1998: 237ff). In this section the focus is on the prototypical poles rather than the various intermediate types which occur in German (see Pasch 2000), as well as in many other languages (see Haspelmath 2004: 33ff).

2.1  Coordinative connections Coordinative connections are realized by coordinating conjunctions. The prototype of a coordinating conjunction is und (‘and’): (1) Es gab da überhaupt keine Action-Abteilung, und ich war von Tag zu Tag mieser gelaunt. Und dann erwog ich auch bald meine Abseilung in den ewigen Underground. Und sehn wir uns nicht in dieser Welt, dann sehn wir uns in Bielefeld! (Udo Lindenberg, Rätselhaftes Bielefeld) they didn’t even have an action department / and my mood got worse by the day / and soon I began to think about abseiling / to the eternal underground / and if we won’t meet in this world / then we’ll catch up in Bielefeld

The expressions linked by a coordinating conjunction (coordinator) are typically of the same formal and/or functional category (for exceptions see Johannessen 1998; Osborne 2003: 114ff). In example (1), all conjuncts (or coordinands; see Haspelmath 2004) are V(erb)2-sentences. Opinions differ about what may be the most adequate representation of the syntactic structure of coordination (see e.g., Dik 1972; Wiese 1980; Wesche 1995; Johannessen 1998; Camacho 2003; Osborne 2003, 2006; Eisenberg 2004: 205ff, 377ff). Most but not all approaches assume structures in which both coordinands have equal status in relation to the coordinator or some other category. One of the exceptions is Johannessen (1998: 108ff), who proposes a structure in which one coordinand is the complement of the coordinator, the other being its specifier. Her main interest, however, is in so-called unbalanced, i.e., non-prototypical, coordination. In prototypical coordination, the morphosyntactic format of the coordinands is defined independently of the coordinator. Coordinators neither select coordinands of a specific category nor do they require or attribute specific morphosyntactic features. German has some connectives that behave similarly to coordinators, but nevertheless select relata of a certain morphosyntactic format: denn (‘for’), e.g., can only connect V2-clauses (see Duden 2005: 628), whereas sowie (‘as well as’) can only connect V-final-clauses and constituents less complex than a clause (words or phrases) (see Breindl 2007b). Connectives with a similar non-prototypical behaviour can be found in several languages: for and as well as in English, car and ainsi que in French, gdyż and oraz in Polish, etc.

Subordination and coordination: Evidence from connectives

Prototypical coordinators and similar elements are strongly constrained as to their linear position in relation to the coordinands. In German, they must be positioned in the middle between the coordinands, with a slightly stronger affinity to the right one: (2) a. b.

Ihr kauft ein und wir warten hier an der Ecke. you can go shopping and we’ll wait here at the corner *Und wir warten hier an der Ecke ihr kauft ein. and we’ll wait here at the corner you can go shopping

If two coordinate clauses are separated by a comma or period, then it is invariably put to the left of the coordinator and not to its right: c. d.

Ihr kauft ein. Und wir warten hier an der Ecke. you can go shopping. and we’ll wait here at the corner * Ihr kauft ein und. Wir warten hier an der Ecke. you can go shopping and. we’ll wait here at the corner

In other languages, coordinators may take other linear positions (e.g., Latin ‑que, which is a suffix added to the second coordinand). In general their positions are much more constrained than the positions of any other class of connectives (see Haspelmath 2004: 6ff; also Osborne 2006). We can therefore say that coordinators connect their coordinands basically by linear sequence.

2.2  Subordinative connections One of several means to establish subordinative connections between clauses are subordinating conjunctions (adverbial subordinators; see Kortmann 1996), like während (‘while’): (3) Die Pinguine waren braun-gelb, während die Giraffen schwarz-weiß waren. the penguins were yellow-brown, while the giraffes were black and white

Subordinating conjunctions influence the morphosyntactic format of one of their relata (the subordinate clause). The authors of the Handbuch der deutschen Konnektoren (Pasch et al. 2003: 8ff, 106ff) call this relatum the internal argument. The relation ­between the subordinating conjunction and its internal argument is described as a type of government: in German, subordinating conjunctions select V-final order of their internal argument; in many languages they require certain tense and/or mood forms of the subordinate verb. On the other hand, subordinating conjunctions do not have any influence on the morphosyntactic format of their external argument (the main clause) (see Pasch et al. 2003: 361, 416f). Subordinative connections are therefore structurally asymmetric. While the relata of coordinators are typically of the same morphosyntactic ­category, the relata of subordinating conjunctions typically belong to different



 Hardarik Blühdorn

­ orphosyntactic categories. They can be distinguished in functional terms as main m and subordinate clauses, or formally as clauses with certain morphosyntactic properties, e.g., V2 and V-final. But categorical differences between the relata are by no means obligatory in subordinative connections. Both relata can be of the same category, if, for independent reasons, the external argument is a subordinate clause as well: (4) Maria erzählte, dass [die Pinguine braun-gelb waren, während [die Giraffen schwarz-weiß waren]V-final]V-final. Mary told us that [the penguins were yellow-brown, while [the giraffes were black and white]]

In such cases the morphosyntactic form of the external argument is never determined by the connective. Prototypical coordinators can connect expressions of any morphosyntactic category. Subordinating conjunctions, in contrast, can only connect clauses. This limitation is largely compensated for by adpositions, which are formally and functionally very similar to subordinating conjunctions, but which take noun phrases instead of clauses as at least one of their relata (see Kortmann 1996: 25, 58ff, 66ff): (5) die Vorkommnisse während der Abschlussfeier the occurrences during the leaving party

It is a well-known fact that in German, as in many other languages, clauses can be transformed into noun phrases (see Hopper & Thompson 1984: 737f, 744ff; O’Dowd 1992; Eisenberg 2004: 252ff). The expressions resulting from nominalization must then be connected by adpositions instead of subordinating conjunctions. Adpositions require their internal argument to adopt a specific case form, but they have no influence on the morphosyntactic form of their external argument. Subordinating conjunctions and adpositions have a fixed serial position in relation to their internal, but not in relation to their external argument. In German, they typically take a position at the left margin of their internal argument. They structurally embed their internal argument into the external argument. In the linear structure of the external argument they can be moved rather freely together with their internal argument. In particular, they can be postposed (as in (6/7a)), preposed (as in (6/7b)) and – at least for subordinating conjunctions – even be interposed to the external argument (as in (6c)).

Subordinating conjunction: (6) a. b. c.

Wir warten hier an der Ecke, solange ihr einkauft. we’ll wait here at the corner, while you go shopping Solange ihr einkauft, warten wir hier an der Ecke. while you go shopping, we’ll wait here at the corner Wir warten, solange ihr einkauft, hier an der Ecke. we’ll wait, while you go shopping, here at the corner

Subordination and coordination: Evidence from connectives

Adposition: (7) a. b.

die Aufräumarbeiten nach der Abschlussfeier (waren ermüdend) the cleaning work after the leaving party (was exhausting) nach der Abschlussfeier die Aufräumarbeiten (waren ermüdend) after the leaving party the cleaning work (was exhausting)

Thus, subordinating conjunctions and adpositions (subordinators) do not link their relata by linear sequence, but by government and embedding.

2.3  Adverbial connections Coordinative and subordinative connections of the kinds discussed so far are established by syntactic means such as linear ordering, government and embedding. They clearly fall within the scope of syntax. Semantically equivalent connections can also be encoded by means of adverbial connectives. In the following pairs of examples, the (a)-variants encode the connection by means of a subordinator or coordinator, whereas the (b)-variants encode a semantically equivalent connection by means of an adverbial connective: (8) a. Die Pinguine waren braun-gelb, während die Giraffen schwarz-weiß waren. (subordinating conjunction) the penguins were yellow-brown, while the giraffes were black and white b. Die Pinguine waren braun-gelb. Die Giraffen dagegen waren schwarz-weiß. (adverbial connective) the penguins were yellow-brown. the giraffes, in contrast, were black and white

(9) a. Die Aufräumarbeiten nach der Abschlussfeier waren sehr anstrengend. (adposition) the cleaning work after the leaving party was very exhausting b. Die Abschlussfeier war ein großer Erfolg. die Aufräumarbeiten danach waren sehr anstrengend. (adverbial connective) the leaving party was a great success. The cleaning work afterwards was very exhausting

(10) a. Es gab da überhaupt keine Action-Abteilung, und ich war von Tag zu Tag mieser gelaunt. (coordinating conjunction) they didn’t even have an action department, and my mood got worse by the day b. Es gab da überhaupt keine Action-Abteilung. Zudem war ich von Tag zu Tag mieser gelaunt. (adverbial connective) they didn’t even have an action department. in addition, my mood got worse by the day

Adverbial connectives are syntactic constituent of one of their semantic relata. They are attached to that relatum as adverbial adjuncts. They may influence the tense and/



 Hardarik Blühdorn

or mood of the verb of that relatum, but they do not influence the morphosyntactic format of their other relatum. To their other relatum, they do not bear any syntactic relation (see Pasch et al. 2003: 485). In relation to both connected expressions, adverbial connectives do not have a fixed linear position. Like most adverbials, they can be moved relatively freely within the relatum of which they are a constituent, and though they typically occur in the right (subsequent) relatum they can also occur in the left (antecedent) relatum: (11) Wir warten hier so lange. Ich meine, bis ihr mit dem Einkaufen fertig seid. we’ll wait here for the time being. I mean until you have finished shopping

Adverbial connectives link their relata neither by government and embedding nor by linear sequence. Instead, they connect them semantically or, more precisely, by reference. The semantic representation of an adverbial connective contains a slot for a referent that cannot be identified on grounds of the information provided by the sentence in which the adverbial is a constituent. In order to identify that referent, the interpreter must look for the necessary information in the preceding or following context. Depending on where the required information is placed, we can distinguish between anaphoric (backward oriented) and cataphoric (forward oriented) adverbial connections. Thus, the adverbial connectives dagegen (‘in contrast’), danach (‘afterwards’), and zudem (‘in addition’), in (8b), (9b) and (10b) respectively, connect their relata anaphorically, whereas so lange (‘for the time being’) in (11) connects its relata cataphorically. In many adverbial connectives of German, the referential element is morphologically visible. Such connectives are results of word formation processes in which an adpositional and a pronominal component have been contracted into one word form. In the following examples, the pronominal component is boldfaced, the other one being the adpositional component: da-gegen (‘in contrast’, lit. ‘there-against’), da-nach (‘afterwards’, lit. ‘there-after’), hier-bei (‘on this occasion’, lit. ‘here-at’), hier-für (‘for this’, lit. ‘here-fore’), zu-dem (‘in addition’, lit. ‘to-that’), außer-dem (‘moreover’, lit. ‘outside-that’), in-dessen (‘however’, lit. ‘in-that’), während-dessen (‘in the meantime’, lit. ‘during-that’) etc. Thus, in (8b) the pronominal component da- of the adverbial connective dagegen contained in the second sentence refers anaphorically to the proposition encoded by the first sentence. Similarly in (10b), the pronominal component -dem of the adverb zudem contained in the second sentence refers anaphorically to the proposition encoded by the first sentence. In (9b), the pronominal component da- of the adverb danach refers to an event (the leaving party) described in the preceding sentence. In (11), the pronominal component so refers to an event (finishing shopping) described in the subsequent sentence (on the referential function of da ‘there’ and so ‘so’ see Blühdorn 2003). Some adverbial connectives of German, such as bestenfalls (‘at best’), wenigstens (‘at least’) or anschließend (‘afterwards’) do not contain morphologically explicit pronominal components. We cannot go into the details here, but probably all adverbial connectives can be traced back historically to expressions that involve some sort of

Subordination and coordination: Evidence from connectives

referential element, and all of them are used in exactly the same referential way (see Webber et al. 2003: 548ff). We are therefore justified in assuming that the semantic representation of all adverbial connectives contains a referential slot, even if their morphological form has not preserved a corresponding pronominal element. It is interesting to observe that the pronominal element, where it is visible, quite often maintains case morphology within the contracted form, as in zu-dem (lit.: to-that-DAT), außer-dem (lit.: outside-that-DAT), in-dessen (lit.: in-that-GEN) or während-dessen (lit.: during-that-GEN). The complex morphology of these connectives reveals that their referential linking does not simply substitute syntactic subordination. Rather, the linking force of adverbial connectives builds on a subordinative relation which has become incorporated in their structure. The pronominal component was originally case-governed by the adpositional component: it is, in fact, its internal argument. Consequently, the clause to which the adverbial connective is attached as an adjunct must be its external argument. In the contracted form of the connective, the subordinative force of the adposition has become encapsulated, so that it is no longer able to contribute actively to syntactic structure. But at the same time, the semantic scope of the connective is extended beyond the limits of the sentence by the referential force of the pronominal component. Structurally, the internal argument is incorporated within the connective, but its referent must be found in the context. Thus, in a sense, adverbial connectives are closer in syntax to their external argument (the clause of which they are a constituent) than to their internal argument (the clause to which they establish a link by reference). Table 1 provides an overview of the linking properties of the connectives that have been discussed so far: Table 1.  Overview of the linking properties of connectives

Subordinators

Coordinators

Adverbial connectives

Linking by government + ± and embedding Linking by linear + ± sequence Linking by reference +

Subordinating conjunctions and adpositions (subordinators) link their relata hierarchically, by government and embedding, whereas coordinators link them non-hierarchically, by linear sequence. Adverbial connectives neutralize this difference (see Quirk et al. 1985: 927f). With subordinators, they share government and embedding, but they encapsulate these relations in their morphology, so that they cannot take effect within sentence structure. With coordinators, they share a positional affinity to the right (subsequent) relatum (see ibid.: 921f). But in coordinators this affinity amounts to a strict syntactic rule, while in adverbial connectives it is only a pragmatic preference. The particular linking force of



 Hardarik Blühdorn

adverbial connectives is based neither on government and embedding nor on serial position, but on reference, i.e., on a principle that does not play a crucial role in syntactic coordination or subordination. Reference is a discourse relation, not a syntactic relation. When using referential expressions such as adverbial connectives as cohesive devices, speakers refer to discourse entities, not to other syntactic expressions. This means that adverbial connectives establish connections on discourse level, while subordinators and coordinators establish connections on sentence level. Connections established by adverbial connectives are outside the scope of syntactic rules. The syntactic distinction between coordination and subordination makes sense only within sentence boundaries, i.e., as a means to construct complex and compound sentences (see Quirk et al. 1985: 719). It cannot be transferred to connections established by adverbial connectives. As a consequence, the distinction between hierarchical and non-hierarchical connections in discourse cannot reasonably be supposed to mirror the distinction between coordination and subordination in syntax. A very considerable part of discourse connections is established on a level higher than the sentence, i.e., beyond the scope of syntactic rules. At least for those connections, whether they count as hierarchical or non-hierarchical can only be decided on other than syntactic grounds. An additional observation may be made here about two groups of subordinators with pronominal components. Some German subordinating conjunctions contain such components, e.g., nachdem (‘after’), indem (‘as’) or seitdem (‘since’). Subordinators of this kind derive historically from adverbs. Their pronominal components are residues with a very weak (if any) referential function left in present day use (on nachdem see Blühdorn 2004). A second group are relative adverbs like weshalb (‘wherefore’), wobei (‘whereby’), worauf (‘whereupon’) etc., which contain the pronominal components wes- and wo-. Adverbs of this kind can be used to introduce adverbial relative clauses. They establish a type of connection that combines referential linking with linking by government, though not by embedding (see Pasch et al. 2003: 241ff, 422ff). The existence of this type of linking does not weaken my argument. I do not claim, in fact, that reference has no role to play on sentence level, but that its role is not crucial for syntactic connections. On the other hand, I do claim that syntactic linking mechanisms such as government, embedding and syntactic serialization do not work beyond the boundaries of the sentence, i.e., are neutralized on the level of discourse.

3.  Semantic connections Strictly speaking, the term connective does not refer to a syntactic, but to a semantic category. Connectives are elements of several syntactic classes that share the function of encoding semantically characterized relations between conceptual entities such as events and propositions (see Pasch et al. 2003: 1ff, 38f; Blühdorn 2003; Lohnstein 2004). In semantics, as well as in syntax, we can distinguish between hierarchical and non-hierarchical connections.

Subordination and coordination: Evidence from connectives

According to Lang (1984: 69ff), the semantic relata of coordinative connections must be tied up by a common integrator. This term refers to a superordinate conceptual category, under which both relata can be subsumed, and under which they are in contrast with each other. In example (12), we can construe something like “colours of zoo animals” as the common integrator: (12) Die Pinguine waren braun-gelb und die Giraffen waren schwarz-weiß. the penguins were yellow-brown, and the giraffes were black and white

For the verses by Udo Lindenberg, a possible common integrator seems to be “motives that suggest abseiling to the eternal underground”: (13) Es gab da überhaupt keine Action-Abteilung, und ich war von Tag zu Tag mieser gelaunt. they didn’t even have an action department, and my mood got worse by the day

The examples show that the common integrator need by no means pertain to generally accessible world knowledge. On the contrary, it may be construed ad hoc, according to the demands of each context in discourse. It might seem as if the requirement of a common integrator could be a good criterion to distinguish between hierarchical and non-hierarchical connections in semantics. But on a closer look we find that this requirement is not restricted to semantically symmetrical nor to syntactically coordinative connections. Rather, it is a characteristic of a much more extensive class of connections, which may be symmetrical or asymmetric and which may be encoded by coordination, subordination or any other syntactic mechanism. Also, all kinds of adverbial connections belong to this class, i.e., require a common integrator. According to a well-known model proposed by Lyons (1977: 442ff; 791ff), conceptual entities can be divided into four general categories: spatial (first order) entities, temporal (second order) entities, logical/epistemic (third order) entities and deontic (fourth order) entities (see also Kortmann 1996: 28ff). First order entities are spatial objects, second order entities are states of affairs (including states and events), third order entities are propositions, and fourth order entities are intentional entities, which, for lack of a better term, I will call acts (see Blühdorn 2003: 16ff). The four general categories define four conceptual domains: space, time, logic/episteme (the domain of knowledge) and ethics/deontics (the domain of intentions and acting). It is a general semantic requirement on adverbial connections (subordinative or not), as well as on coordinative connections, that their relata must belong to the same conceptual domain and, consequently, be of the same general category. Thus, in order to connect two relata spatially, both must be spatial objects; in order to connect them temporally, both must be states of affairs; in order to connect them logically, both must be propositions; and in order to connect them deontically, both must be acts (for more details see Blühdorn 2007a, b).



 Hardarik Blühdorn

It can be concluded that the requirement of a common integrator is not an appropriate criterion for distinguishing between hierarchical and non-hierarchical connections in semantics. A better criterion seems to be relational symmetry. Nonhierarchical semantic connections are symmetrical. Their relata have equal semantic functions and equal semantic weight (see Breindl 2007b: 144). One of the syntactic consequences of semantic symmetry is the possibility of inverting the sequence of the relata without a significant change of meaning. The relata in the following example are symmetrically connected. They can be inverted without semantic consequences: (14) a. b.

Die Pinguine waren braun-gelb und die Giraffen waren schwarz-weiß. the penguins were yellow-brown, and the giraffes were black and white Die Giraffen waren schwarz-weiß und die Pinguine waren braun-gelb. the giraffes were black and white, and the penguins were yellow-brown

Hierarchical semantic connections, in contrast, are asymmetric. Changing the syntactic sequence of their relata will significantly change the meaning. Where an inversion of the relata gives rise to such a change of meaning, it can be concluded that the connection is not understood as symmetrical, even if it is syntactically encoded by a coordinator: (15) a. b.

Maria ging in die Bibliothek und sie bekam Hunger. Mary went to the library, and she began to feel hungry Maria bekam Hunger und sie ging in die Bibliothek. Mary began to feel hungry, and she went to the library

In the most plausible reading of these examples, the connected events are ordered in a temporal sequence, which is inverted from (15a) to (15b). It is not uncommon for syntactically coordinative connections to be interpreted in semantically asymmetric ways – an effect that can be explained by very general cognitive and pragmatic principles (see Grice 1981: 185f; Posner 1980: 182ff; Lang 1984: 80ff; Blakemore & Carston 2005; Breindl 2007a). The examples show that syntactic coordination and semantic symmetry must be carefully distinguished. By no means can they be identified with each other. The relata of asymmetric connections cannot be inverted without significant semantic consequences. They have different relational (thematic) roles. One of them is being connected (like a ship that drops its anchor), the other is what it is being connected to (like the sea ground in which the anchor is fixed). In Ronald Langacker’s (1987: 231ff) terminology, the former is called trajector (T) and the latter landmark (L). Three types of asymmetric connections can be distinguished (see Blühdorn 2003: 19f; Blühdorn 2005: 315f): – – –

situating connections conditional connections causal connections

Subordination and coordination: Evidence from connectives

Situating connections are stative. They assign to the trajector a place in a conceptual domain, which is described by a relation to the landmark: (16) Bevor du nach Hause gehst (L), lösch bitte das Licht (T). before you go home (L), please switch off the light (T)

Switching off the light is the trajector that is situated on the time scale in relation to the event of going home (landmark). Switching off the light and going home themselves are evidently non-stative, but their sequential relation on the time scale, encoded by the conjunction before, is stative. Each of the events has its fixed position in time, and the position of the trajector is defined on grounds of the position of the landmark. Conditional connections are dynamic: the landmark event not only situates the trajector event, but it also influences the value to be taken by the trajector event. It is not yet clear if the trajector event will in fact become real (or come true). This depends on the value to be taken by the landmark event: (17) Und sehn wir uns nicht in dieser Welt (L), dann sehn wir uns in Bielefeld (T). and if we won’t meet in this world (L), then we’ll catch up in Bielefeld (T)

The example tells us that the meeting in Bielefeld will take place on the condition that the meeting in this world does not. Causal connections are dynamic as well, but in a causal connection the value of the trajector event is already fixed. The trajector event is real, and the landmark event has influenced the fixing of its value: (18) Ich war von Tag zu Tag mieser gelaunt (T), weil es gab da überhaupt keine Action-Abteilung (L). my mood got worse by the day (T), for they didn’t even have an action department (L)

The inexistence of an action department is the cause which led to the change of the speaker’s mood. The result is presented as factual, i.e., as a state of affairs with a fixed reality value. The distinction between semantically symmetrical and asymmetric connections can be made within the boundaries of the sentence (on intra-sentence level) and also between sentences in discourse (on inter-sentence level). Both kinds of semantic connection can be encoded by coordinators as well as by subordinators and also by adverbial connectives. Some further examples may illustrate this point: (19) a. b. c.

Das ist kein Selbstbedienungsladen, und sonntags ist hier zu. this is no self-service store, and on Sundays we are closed Sonntags ist hier zu, und das ist kein Selbstbedienungsladen. on Sundays we are closed, and this is no self-service store Das ist kein Selbstbedienungsladen. Außerdem ist hier sonntags zu. this is no self-service store. besides, we are closed on Sundays



 Hardarik Blühdorn

d. Sonntags ist hier zu. Außerdem ist das kein Selbstbedienungsladen. on Sundays we are closed. besides, this is no self-service store

The connections in (19a-d) are symmetrical. The sequence of their relata can be inverted without a significant change of meaning. In (19a/b), the connection is encoded by a coordinator (intra-sentence level); in (19c/d), a semantically equivalent connection is encoded by an adverbial connective (inter-sentence level). My argument for considering coordination of two main clauses an intra-sentence connection is the syntactic constraint on the position of the coordinator, discussed in section 2. In addition, we can consider punctuation and intonation. Coordinated main clauses may be separated by a comma instead of a period and may even not be separated by any punctuation mark. In spoken utterances, they can be included into the same intonational phrase. (20) a. b. c. d.

Wenn du keine Lust hast (L), gehe ich allein (T). if you don’t feel like it (L), I’ll go on my own (T) Wenn ich allein gehe (L), hast du keine Lust (T). if I go on my own (L), you won’t feel like it (T) Du hast keine Lust? (L) Dann gehe ich allein. (T) you don’t feel like it? (L) then I’ll go on my own (T) Ich gehe allein? (L) Dann hast du keine Lust. (T) I’ll go on my own? (L) then you won’t feel like it (T)

The examples in (20a-d) illustrate asymmetric connections. Inverting the sequence of their relata leads to a significant change of meaning. The connections in (20a/b) are encoded by a subordinator (intra-sentence level). In (20c/d), semantically equivalent connections are encoded by an adverbial connective (inter-sentence level). On intra-sentence level symmetrical and asymmetric connections can be encoded by coordinators as well as subordinators: (21) a. b. c. d.

Die Pinguine waren braun-gelb und die Giraffen waren schwarz-weiß. the penguins were yellow-brown, and the giraffes were black and white Die Giraffen waren schwarz-weiß und die Pinguine waren braun-gelb. the giraffes were black and white, and the penguins were yellow-brown Die Pinguine waren braun-gelb, während die Giraffen schwarz-weiß waren. the penguins were yellow-brown, while the giraffes were black and white Die Giraffen waren schwarz-weiß, während die Pinguine braun-gelb waren. the giraffes were black and white, while the penguins were yellow-brown

(22) a. Ich war von Tag zu Tag mieser gelaunt (T), weil es gab da überhaupt keine Action-Abteilung (L). my mood got worse by the day (T), for they didn’t even have an action department (L)

Subordination and coordination: Evidence from connectives

b. Es gab da überhaupt keine Action-Abteilung (T), weil ich war von Tag zu Tag mieser gelaunt (L). they didn’t even have an action department (T), for my mood got worse by the day (L) c. Ich war von Tag zu Tag mieser gelaunt (T), weil es da überhaupt keine Action-Abteilung gab (L). my mood got worse by the day (T), because they didn’t even have an action department (L) d. Es gab da überhaupt keine Action-Abteilung (T), weil ich von Tag zu Tag mieser gelaunt war (L). they didn’t even have an action department (T), because my mood got worse by the day (L)

The examples in (21a-d) illustrate symmetrical connections. In (21a/b), the connections are encoded by a coordinator. In (21c/d), the same relata are symmetrically connected by a subordinator. The examples in (22a–d) illustrate asymmetric connections. Inversion of the relata leads to a change of meaning. The connections in (22a/b) are encoded by a (non-prototypical) coordinator. Both relata are syntactically realized as main clauses. In (22c/d), the same relata are asymmetrically connected by a subordinator. One of the relata is realized as a main clause, the other as a subordinate clause. The data discussed in this section are from Modern German, but data from other languages could just as well have been used. From the observations made, it can be concluded that the distinctions between syntactic coordination and subordination and between semantic symmetry and asymmetry are independent of each other. Connections of syntactic units and the connections of the encoded conceptual entities can be either parallel or non-parallel in structure.

4.  Discourse connections Since the 1980s, several models and theories of discourse structure have been proposed, which have tried to give an explicit account of coherence relations in written and spoken text. Most of them distinguish in some way or another between hierarchical and non-hierarchical discourse relations. But they differ considerably in how they motivate this distinction.

4.1  Syntax and semantics as models for understanding discourse structure One obvious hypothesis is that discourse structure might be parallel to syntactic and/ or semantic structure (see Hopper & Thompson 1984: 736f; O’Dowd 1992; Matthiessen & Thompson 1988; Mann & Thompson 1988: 269; Taboada & Mann 2006: 427). The and-variant of this hypothesis has often been assumed for coordinative relations. Coordinated syntactic units are typically of the same formal and functional category, and their



 Hardarik Blühdorn

linear order can be inverted without semantic consequences. Symmetrically connected conceptual units are of the same semantic category and have a common semantic function. Non-hierarchically connected discourse units should belong to the same rhetorical category and have a common discourse function. Many examples discussed in the literature seem to support the view that semantic symmetry and syntactic coordination are natural linguistic means to encode non-hierarchical discourse relations. Yet we have seen in section 3 above that coordinative connections in syntax are typically underspecified for semantic interpretation. Depending on the context of the utterance, they may (or even must) receive an asymmetric reading: (23) Maria ging zu McDonald’s, und sie bekam Hunger. Mary went to McDonald’s, and she began to feel hungry a.→ Außerdem bekam sie Hunger. besides, she began to feel hungry b.→ Dann bekam sie Hunger. then she began to feel hungry c.→ Deshalb bekam sie Hunger. therefore she began to feel hungry

(23a) to (23c) are possible interpretations of the second part of (23). (23a) is a symmetrical reading. The two connected propositions have equal status: two predicates which are true of Mary. (23b) and (23c) are asymmetric readings of (23). (23b) is a situating interpretation: Mary’s going to McDonald’s is the landmark, her beginning to feel hungry is the trajector. (23c) is a causal interpretation: going to McDonald’s is the causal landmark, feeling hungry the causal trajector. Most theories of discourse structure have paid little attention to the multiple interpretability of coordinative syntactic connections. Some authors have assumed structural parallelism between syntax, semantics and discourse also for subordinative adverbial connections (e.g., Matthiessen & Thompson 1988). For most of these connections, however, for general reasons, only the or-variant of the hypothesis is possible, because the syntactic and semantic structures of typical subordinative adverbial connections are inverse to each other. A crucial property of their syntactic structure is embedding. Hierarchically higher relata, e.g., main clauses, embed hierarchically lower relata, e.g., subordinate clauses: (24)

main clause

adverbial subordinate clause

Semantically asymmetric connections, too, can be characterized, in a sense, as embedding relations. The landmark is the embedding part: it forms the conceptual background framework into which the trajector is inserted. Using the terminology of Gestalt psychology, Langacker (1987: 231ff) characterizes the landmark as ground. The trajector,

Subordination and coordination: Evidence from connectives

on the other hand, is the embedded part: it takes a position in relation to the conceptual background. Langacker (ibid.) characterizes it as figure. In terms of embedding, then, the landmark’s position in the semantic hierarchy is higher than the position of the trajector, for the trajector is embedded by the landmark: (25) landmark

trajector

Looking at the mapping relations between hierarchical connections in syntax and semantics, we realize that the semantically embedding landmark is invariably encoded by the syntactically embedded expression (e.g., the subordinate clause) and the semantically embedded trajector by the syntactically embedding expression (e.g., the main clause): (26) Solange ihr einkauft (L), warten wir hier an der Ecke (T). while you go shopping (L), we’ll wait here at the corner (T)

Lohnstein (2004: 143) states, from the point of view of model-theoretic semantics: “Das Wahrheitsintervall des Hauptsatzes [...] wird relativ zum Wahrheitsintervall des Nebensatzes [...] bestimmt, so dass der Nebensatz die Auswertungsdomäne für den [...] Hauptsatz determiniert.” (The truth interval of the main clause is fixed in relation to the truth interval of the subordinate clause, so that the subordinate clause determines the domain of interpretation for the main clause.) Bierwisch (2003) sees the crucial difference between complements and adjuncts in the direction of the attribution of thematic roles. Both complements and adjuncts are syntactically subordinate to their heads, but whilst complements receive their thematic roles from their heads, adjuncts attribute thematic roles to their heads. If we interpret attribution of thematic roles as a manifestation of semantic superordination, we can state that complements are semantically subordinate and adjuncts semantically superordinate to their heads. Applied to adverbial subordinate clauses, this means that they are semantically superordinate to their main clauses. The hierarchies of subordinative syntactic connections and asymmetric semantic connections are thus inverse to each other: the syntactically embedding part is the semantically embedded part and vice versa: (27) main clause

subordinate clause

landmark

trajector

This application of the landmark-trajector distinction on adverbial clauses differs from Langacker’s own proposal (see Langacker 1991: 436). When explaining the subordinate



 Hardarik Blühdorn

status of adverbial clauses, Langacker abandons his analysis in terms of landmark and trajector and stipulates, instead, a general iconicity between syntactic and semantic subordination in terms of profiling (see ibid.: 436f; also Cristofaro 2003: 29ff). This solution appears somewhat ad hoc to me, and Langacker himself does not explain it. As a consequence, questions may also be raised about Cristofaro’s (ibid.) claim that semantic asymmetry provides a more reliable starting point for the analysis of subordination than traditional morphosyntactic asymmetries do.

4.2  Some formal accounts of discourse structure 4.2.1  Rhetorical Structure Theory The hypothesis of a structural parallelism between syntax, semantics and discourse has been of some importance in Rhetorical Structure Theory (RST; see Mann & Thompson 1988; Taboada & Mann 2006; Stede in this volume). RST distinguishes between nuclear information and satellite information within a discourse. Nuclear information is main information, satellite information is secondary information. The difference between them becomes clearest when they are deleted. The deletion of nuclear information will make the discourse less coherent, the remaining parts becoming more difficult to comprehend. The deletion of satellite information will make the discourse less explicit, but the remaining information will still be coherent. Thus, the omission of satellite information may play an important role in summarizing (see Mann & Thompson 1988: 267f). RST distinguishes between two types of relations: nucleus-satellite relations and nucleus-nucleus relations. The former are hierarchical, the latter are non-hierarchical (Mann & Thompson 1988: 246ff, 266). Nucleus-satellite relations are more frequently dealt with in RST-related studies. The examples analysed by Mann & Thompson (ibid.: 252, 261ff) show that they can be encoded by both coordinative and subordinative connections in syntax. Both nuclear (N) and satellite (S) information can be encoded by both main (M) and adverbial subordinate clauses (A): (28) I’ll post more details later (S–M), but this is a good time to reserve the place on your calendar (N–M). (concessive relation; coordinative connection) (29) As your floppy drive writes or reads (S–A), a Syncom diskette is working four ways (N–M). (circumstance relation; subordinative connection 1) (30) A carbon additive drains away static electricity (S–M), before it can attract dust or lint (N–A). (antithesis relation; subordinative connection 2)

Nucleus-nucleus relations are not in the focus of interest in RST. The examples ­given by Mann and Thompson (ibid.: 278f) are not conclusive, but they suggest that ­nucleus-nucleus relations can also be encoded by both coordinative and subordinative connections: (31) Peel oranges (N–M) and slice crosswise (N–M). Arrange in a bowl (N–M) and sprinkle with rum and coconut (N–M). (sequence relation; coordinative connection)

Subordination and coordination: Evidence from connectives

(32) Chill (N–M) until ready to serve (N–A). (sequence relation; subordinative connection)

Satellite-satellite relations (see Asher & Vieu 2005: 592, 594f), which are a second type of symmetrical relations, are not recognized in RST. But the text analysis given in Mann & Thompson (1988: 261ff) shows that they can at least be encoded by syntactic coordination: (33) Strong binders hold the signal-carrying oxides tightly within the coating (S). And the non-woven jacket liner (...) provides thousands of tiny pockets to keep what it collects (S).

One of the main practical problems in RST seems to be how to reliably recognize main information and how to distinguish it from secondary information (see Stede in this volume). As a consequence, the distinction between hierarchical and non-hierarchical discourse connections remains problematic as well. Mann & Thompson (1988: 249f) insist that the definitions of RST relations “do not rely on morphological or syntactic signals. Recognition of the relation always rests on functional and semantic judgements alone. [...] We have found no reliable, unambiguous signals for any of the relations.” Although this affirmation seems to be somewhat idealized (compare the section on explicit signalling of discourse relations in Taboada & Mann 2006: 438ff), RST thus provides at least some evidence for considering discourse structure as independent of syntactic structure. On the other hand, Mann & Thompson (1988: 269) explicitly suggest a functional link between asymmetric discourse relations and hypotaxis in syntax. An unresolved theoretical problem in RST has to do with the concept of a discourse relation. Syntactic subordination and coordination are defined as relations between syntactically categorized units (such as phrases, clauses or sentences) with certain syntactic functions (such as head, complement or adjunct). Symmetrical and asymmetric semantic connections are defined as relations between semantically categorized entities (such as spatial objects, events or propositions) bearing semantic or thematic roles (such as agent, theme, cause, goal etc.). Similarly, hierarchical and non-hierarchical discourse relations should be defined as relations between rhetorically categorized units (such as utterances or speech acts) with certain rhetorical functions (such as topic and focus, or intended communicative effects). But that is not how discourse relations are defined in RST. The basic units of analysis in RST are clauses, sentences or phrases, i.e., syntactic units. The relations between these units are divided into “subject matter” and “presentational” ones (Mann & Thompson 1988: 256f; also Taboada & Mann 2006: 435f). The former are clearly semantic relations such as cause, condition, result etc.; the latter are pragmatic relations between speech acts such as evidence, motivation, justify etc. Only the latter deserve to be called discourse relations in the strict sense of the term. RST is a hybrid model that incorporates syntactic, semantic and rhetorical concepts and categorizations (see Stede in this volume). RST tree diagrams contain information



 Hardarik Blühdorn

of different domains. This makes them suggestive and at the same time non-conclusive in relation to a possible parallelism between sentence and discourse structure. They are suggestive to the extent that their basic units are syntactic instead of rhetorical categories, and they are non-conclusive to the extent that these units are associated with partly semantic and partly rhetorical functions.

4.2.2  The Linguistic Discourse Model A more homogeneous account of discourse structure is offered by Polanyi’s (1988) Linguistic Discourse Model (LDM). In LDM, discourse is segmented into discourse constituent units of different levels of complexity. The units at the elementary level are clauses and so-called discourse operators (assigners, connectives and discourse markers) (Polanyi 1988: 605f). Constituents at the levels of higher complexity are genuine discourse units such as interactions, speech events, stories, plans, question-answer sequences, lists etc. (ibid.: 603). Representations of the elementary level of discourse structure in LDM are relatively close to syntax; representations of higher levels are much closer to conversational analysis. A non-hierarchical (“coordinating”) discourse relation is defined in LDM as a relation between two or more discourse constituents that are all linked by the same relation to a common higher order constituent. The linear sequence in which these constituents are uttered is viewed as motivated by cognitive or communicative principles and therefore as non-random. Thus, the possibility of inverting the sequence of the relata is not a criterion for non-hierarchical discourse connection in LDM. The decisive criterion is the common discourse function in relation to a superordinate constituent (ibid.: 606f). A hierarchical (“subordinating”) discourse relation is defined as a relation in which the information conveyed by a discourse constituent S1 further specifies information conveyed by a preceding constituent S0. More specifically, the same inferences can be drawn from S0 and S1, but some of the information which can be inferred from S1 is more detailed than the information which can be inferred from S0 (Polanyi 1988: 609). Discourse constituents which interrupt preceding constituents or which constitute thematic or interactional digressions are also treated as subordinate, even if no information related to the preceding constituent can be inferred from them (Polanyi 1988: 611, 619). Thus, the concept of discourse subordination in LDM is much broader than the concept of a nucleus-satellite relation in RST. It is striking that in LDM, for purely technical reasons, a subordinate discourse unit can never precede its superordinate unit (Polanyi 1988: 613ff). This constraint seems somewhat unrealistic, when we think of the structural possibilities of natural discourse. Hierarchical discourse relations are also characterized as “embedding” relations in LDM (Polanyi 1988: 613). But the underlying notion of discourse embedding is not related to syntactic embedding as it was discussed in section 2 above, nor to semantic embedding as it was discussed in section 4.1. Although LDM considers clauses as

Subordination and coordination: Evidence from connectives

­elementary units of discourse and distinguishes between coordinating and subordinating connectives as discourse operators, the relation between syntactic structure and discourse structure is not explicitly discussed. In LDM, an incoming discourse unit is processed as subordinate or coordinated in relation to a preceding discourse unit in accordance with contextual information and general world knowledge (Polanyi 1988: 616ff). On the whole, LDM is much more concerned with genuine discourse units and functions than with syntax and semantics. Syntactic coordination and subordination are never explicitly mentioned as criteria for considering a given discourse relation as hierarchical or non-hierarchical. On the other hand, the role of contextual and general world knowledge in the selection of a hierarchical or non-hierarchical discourse relation remains somewhat fuzzy. It seems not impossible that the recognition of syntactic coordination and subordination may contribute indirectly or implicitly to the selection of discourse relations.

4.2.3  Segmented Discourse Representation Theory The approach of Asher & Vieu (2005) is more abstract than both RST and LDM. It does not aim primarily at describing discourse relations in real data, but rather at giving general characterizations of categories of discourse relations such as Narration and Elaboration. Asher & Vieu do not believe in the possibility of defining the distinction between hierarchical and non-hierarchical discourse relations in semantic terms (Asher & Vieu 2005: 598). Instead, they look for syntactic definitions of the two types of relations within the framework of Segmented Discourse Representation Theory (SDRT). They propose formal criteria for testing whether relations of different categories are “subordinating” or “coordinating” (Asher & Vieu: 599ff). The first criterion exploits the so-called right-frontier-constraint on anaphoric linking (see Polanyi 1988: 602, 613ff). If in a sequence of discourse units α, β and γ a discourse relation R connects α and β, and γ can be anaphorically attached to α, then R is subordinating. If γ can only be attached to β, then R is coordinating. The second criterion is based on an SDRT principle called Continuing Discourse Patterns (CDP). If discourse units β and γ are connected by a coordinating discourse relation R2 which requires that its relata “bear the same discourse relation to a dominant constituent” (Asher & Vieu 2005: 595) and if discourse unit α is connected to β by a discourse relation R1 which is different from R2, then it can be inferred that R1 is a subordinating relation and that α and γ are connected by a relation R3 of the same subordinating type as R1. If R2 is not a coordinating relation that requires R1 and R3 to be relations of the same type, then R1 is coordinating. The third criterion is very similar to the first one and can therefore be left out here. The fourth criterion is based on the SDRT assumption that two discourse units α and β cannot be connected simultaneously by two discourse relations one of which is subordinating and the other coordinating. Therefore, if a relation R between α and



 Hardarik Blühdorn

β is proved to be coordinating, α and β cannot be connected simultaneously by a subordinating relation. The second and fourth tests seem to be somewhat circular, since what can be “inferred” from them is little more than a reformulation of their input conditions. The first test is more interesting, but the authors warn that it might not be conclusive in all cases. When applying the tests to some concrete examples, Asher & Vieu recognize that there are more difficulties. An analysis of the discourse relation Result (ibid.: 604ff) suggests that it can be “used” either in a coordinating or in a subordinating manner. The same possibility might exist for other discourse relations as well. Asher & Vieu do not explain what exactly it means for a discourse relation to be “used”. But they draw the conclusion that “the concepts of subordinating and coordinating discourse relations” might be founded neither on semantics nor on syntax. Instead they suggest that these properties might be a matter of information packaging (ibid.: 600, 609).

4.3  Future perspectives An important conclusion to be drawn from the discussion of different models of discourse structure is that discourse should be viewed as a system sui generis. Discourse structure is the arrangement of discourse units (see Polanyi 1988: 603ff), which can be ordered on different levels of constituency as well as on different levels of focality. On higher levels of constituency, the units are structurally more complex, on lower levels they are structurally more simple. On higher levels of focality, units are more central to the communicative goals of the speaker, on lower levels of focality, units have supporting or secondary functions (see Klein & von Stutterheim 1992). Non-hierarchical discourse relations connect units of the same level, either of constituency or of focality (intra-level connection); hierarchical discourse relations connect units of different levels of constituency or of focality (inter-level connection). Discourse should not be viewed from the beginning as something derived from syntax and/or semantics. Hybrid conceptualizations make it more difficult to recognize genuine discourse properties and make discourse studies more susceptible to precipitate conclusions. The interaction between hierarchical and non-hierarchical connections in syntax, semantics and discourse can only be described in appropriate detail on the basis of independent and explicit theories of each of the three domains of language structure. Recent research in the areas of intonational phonology and conversational analysis (see e.g., Chafe 1988; Selting 1995; Peters 2005; Büring 2006; Moroni 2006) has shown ever more clearly that (in non-tone languages) a considerable part of discourse structure is encoded by intonation, even in written discourse (“silent prosody”: Féry 2006). This suggests that the widespread convention of taking the clause as the elementary unit of discourse structure must be seriously questioned. The clause is a syntactic unit. The corresponding discourse unit is the utterance. Clauses have subjects and predicates; utterances

Subordination and coordination: Evidence from connectives

have an information structure which can be described in terms of focus, background and topic (see Jacobs 1988, 2001; Klein & von Stutterheim 1992; Büring 2006; Moroni 2006). Subject, predicate, adjunct etc. are categories that cannot be directly related to the communicative and interactional goals of language users. Focus, background and topic are immediately related to this kind of goals and are therefore more appropriate concepts for analysing discourse structure. As long as we have no independent and comprehensive theory of discourse structure, we should refrain from formulating hypotheses about possible parallelisms between discourse and syntax or other subsystems of language. The examples of text analysis given within different research paradigms (e.g., Mann & Thompson 1988; Klein & von Stutterheim 1992; Asher & Vieu 2005) all indicate that both main and secondary discourse units can, in principle, be encoded by both main and secondary syntactic expressions and that both hierarchical and non-hierarchical discourse connections can be encoded by both coordinative and subordinative syntactic connections. The relation between syntax, semantics and discourse should be viewed as a matter of rhetorical options to be taken by speakers and writers in individual acts of encoding. Languages may fix some of these options in their structure and leave only some of them open to the speakers’ preference, but it seems very improbable that any language should generally identify discourse hierarchy with syntactic subordination and/or with semantic asymmetry. Most discourse connections, at any rate, could not be captured by such a deterministic system: all relations that are encoded referentially, across sentence boundaries, by adverbs and proforms. The connecting function of those elements, as we have seen in section 2, is generally independent of the syntactic distinction between coordination and subordination.

5.  Conclusion The concepts of subordination and coordination in language are ultimately terminological metaphors. The notional content of such metaphors depends very much on the nature of the domain to which they are applied. Syntax and discourse are domains of quite different nature. While discourse structure is about presenting information in order to achieve communicative goals, syntactic structure is about arranging formal expressions in order to facilitate parsing. An appropriate account of discourse structure should preferably be given in a terminology sufficiently different from the one used in syntax, in order to avoid misconceptions. The arguments presented in this paper suggest that syntactic hierarchy should not be considered a general model for the conceptualization of discourse hierarchy, nor vice versa. The data from the study of connectives do not support a view that considers syntactic subordination as a generally specialized means for encoding rhetorical hierarchy. It seems slightly more plausible to expect that syntactic coordination should be a preferable means for encoding non-hierarchical discourse connections. But as we



 Hardarik Blühdorn

have seen, syntactic coordination can (or must) in many cases receive an asymmetric semantic interpretation, and the literature contains numerous examples in which syntactic coordination encodes hierarchical discourse relations. The final conclusion is therefore that we should be careful in assuming too many parallels between syntax and discourse structure. Syntax seems to be designed in such a way that the structural variants it offers can be employed very flexibly to encode the structural variants of discourse. This flexibility in the relationship between syntax and discourse seems to be one of the features of natural languages that render them suitable for infinite use in communication.

References Asher, Nicholas & Vieu, Laure. 2005. Subordinating and coordinating discourse relations. Lingua 115: 591–610. Bierwisch, Manfred. 2003. Heads, complements, adjuncts: Projection and saturation. In Modifying adjuncts, Ewald Lang, Claudia Maienborn & Cathrine Fabricius-Hansen (eds), 113–159. Berlin: Mouton de Gruyter. Blakemore, Diane & Carston, Robyn. 2005. The pragmatics of sentential coordination with and. Lingua 115: 569–589. Blühdorn, Hardarik. 2003. Zur Semantik der Konjunktion als. Paradigmatische und syntagmatische Aspekte. Linguistik online 13: 11–53, URL: http://www.linguistik-online.de/13_01/ bluehdorn.html. Blühdorn, Hardarik. 2004. Die Konjunktionen nachdem und bevor. In Brücken schlagen: Grundlagen der Konnektorensemantik, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner (eds), 185–211. Berlin: Walter de Gruyter. Blühdorn, Hardarik. 2005. Zur Semantik kausaler Satzverbindungen: Integration, Fokussierung, Definitheit und modale Umgebung. Studi Linguistici e Filologici Online. Rivista Telematica del Dipartimento di Linguistica dell‘Università di Pisa (SLiFO) 3.2: 311–338, URL: http:// www.humnet.unipi.it/slifo/2005vol2/Bluh­dorn3.2.pdf. Blühdorn, Hardarik. 2007a. Epistemische Lesarten von Satzkonnektoren – Wie sie zustande kommen und wie man sie erkennt. In Semantik vs. Pragmatik oder Semantik und Pragmatik?, Inge Pohl (ed.). Frankfurt: Peter Lang (in print). Blühdorn, Hardarik. 2007b. Kausale Satzverknüpfungen im Deutschen. Pandaemonium Germanicum 10: 253–282. São Paulo: Humanitas, FFLCH/USP. Blühdorn, Hardarik, Breindl, Eva & Waßner, Ulrich Hermann (eds). 2004. Brücken schlagen: Grundlagen der Konnektorensemantik. Berlin: Walter de Gruyter. Breindl, Eva. 2007a. Die Bedeutung additiver Konnektoren (und, sowohl als auch, sowie). In Semantik vs. Pragmatik oder Semantik und Pragmatik?, Inge Pohl (ed.). Frankfurt: Peter Lang (in print). Breindl, Eva. 2007b. Additive Konjunktoren und Adverbien im Deutschen. In Feldergrammatik in der Diskussion. Funktionaler Grammatikansatz in Sprachbeschreibung und Sprachvermittlung, Joachim Buscha & Renate Freudenberg-Findeisen (eds), 141–164. Frankfurt: Peter Lang. Breindl, Eva & Waßner, Ulrich Hermann. 2006. Syndese vs. Asyndese. Konnektoren und andere Wegweiser für die Interpretation semantischer Relationen in Texten. In Text – Verstehen:

Subordination and coordination: Evidence from connectives Grammatik und darüber hinaus, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner (eds), 46–70. Berlin: Walter de Gruyter. Büring, Daniel. 2006. Intonation und Informationsstruktur. In Text – Verstehen: Grammatik und darüber hinaus, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner (eds), 144–163. Berlin: Walter de Gruyter. Camacho, José. 2003. The Structure of Coordination: Conjunction and Agreement Phenomena in Spanish and Other Languages. Dordrecht: Kluwer. Chafe, Wallace. 1988. Linking intonation units in spoken English. In Clause Combining in Grammar and Discourse, John Haiman & Sandra A. Thompson (eds), 1–27. Amsterdam: John Benjamins. Cristofaro, Sonia. 2003. Subordination. Oxford: OUP. Dik, Simon C. 1972. Coordination: Its Implications for the Theory of General Linguistics, 2nd edn. Amsterdam: North-Holland. Duden. 2005. Die Grammatik [Duden Vol. 4], Dudenredaktion (ed.). 7th edn. Mannheim: Dudenverlag. Eisenberg, Peter. 2004. Grundriss der deutschen Grammatik, Vol. 2: Der Satz. 2nd edn. Stuttgart: Metzler. Féry, Caroline. 2006. Laute und leise Prosodie. In Text – Verstehen: Grammatik und darüber hinaus, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner (eds), 164–183. Berlin: Walter de Gruyter. Foley, William A. & Van Valin Jr., Robert D. 1984. Functional Syntax and Universal Grammar. Cambridge: CUP. Grice, Paul. 1981. Presupposition and conversational implicature. In Radical Pragmatics, Peter Cole (ed.), 183–198. New York NY: Academic Press. Günthner, Susanne. 1996. From subordination to coordination? Verb-second position in German causal and concessive constructions. [Arbeitspapier Nr. 75]. Konstanz: Fachgruppe Sprachwissenschaft der Universität Konstanz. Halliday, Michael A.K. & Hasan, Ruqaiya. 1976. Cohesion in English. London: Longman. Haspelmath, Martin. 2004. Coordinating constructions: An overwiew. In Coordinating Constructions, Martin Haspelmath (ed.), 3–39. Amsterdam: John Benjamins. Haspelmath, Martin. (ed.). 2004. Coordinating Constructions. Amsterdam: John Benjamins. Haumann, Dagmar. 1997. The Syntax of Subordination. Tübingen: Max Niemeyer. Heath, Jeffrey. 2004. Coordination: An adaptionist view. In Coordinating Constructions, Martin Haspelmath (ed.), 67–88. Amsterdam: John Benjamins. Hopper, Paul J. & Thompson, Sandra A. 1984. The discourse basis of lexical categories in universal grammar. Language 60: 703–752. Jacobs, Joachim. 1988. Fokus-Hintergrund-Gliederung und Grammatik. In Intonationsforschungen, Altmann, Hans (ed.), 89–134. Tübingen: Max Niemeyer. Jacobs, Joachim. 2001. The dimensions of topic-comment. Linguistics 39: 641–681. Johannessen, Janne Bondi. 1998. Coordination. Oxford: OUP. Klein, Wolfgang & von Stutterheim, Christiane. 1992. Textstruktur und referentielle Bewegung. Zeitschrift für Literaturwissenschaft und Linguistik 86: 67–92. Knott, Alistair & Dale, Robert. 1994. Using linguistic phenomena to motivate a set of coherence relations. Discourse Processes 18: 35–62. Knott, Alistair, Sanders, Ted & Oberlander, Jon. (eds). 2001. Special issue: Levels of Representation in Discourse Relations. Cognitive Linguistics 12: 197–332. Kortmann, Bernd. 1996. Adverbial Subordination: A Typology and History of Adverbial Subordinators Based on European Languages. Berlin: Mouton de Gruyter.



 Hardarik Blühdorn Lang, Ewald. 1984. The Semantics of Coordination [Studies in Language Companion Series 9]. Amsterdam: John Benjamins. Langacker, Ronald W. 1987. Foundations of Cognitive Grammar. Vol. 1: Theoretical Prerequisites. Stanford CA: Stanford University Press. Langacker, Ronald W. 1991. Foundations of Cognitive Grammar. Vol. 2: Descriptive Application. Stanford CA: Stanford University Press. Lefèvre, Michel (ed.). 2000. Subordination in Syntax, Semantik und Textlinguistik [Eurogermanistik 15]. Tübingen: Stauffenburg. Lohnstein, Horst. 2004. Variable und invariante Strukturmerkmale von Satzkonnektoren. In Brücken schlagen: Grundlagen der Konnektorensemantik, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner (eds), 137–160. Berlin: Walter de Gruyter. Lyons, John. 1977. Semantics. 2 vols. Cambridge: CUP. Mann, William C. & Thompson, Sandra A. 1988. Rhetorical Structure Theory: Toward a functional theory of text organization. Text 8(3): 243–281. Matthiessen, Christian & Thompson, Sandra A. 1988. The structure of discourse and ‚subordination‘. In Clause Combining in Grammar and Discourse, John Haiman & Sandra A. Thompson (eds), 275–329. Amsterdam: John Benjamins. Moroni, Manuela. 2006. Modalpartikeln an der Schnittstelle zwischen Syntax und Prosodie. Doctoral thesis, Università degli studi di Verona, Facoltà di lingue e letterature straniere. Muller, Claude. (ed.). 1996. Dépendance et intégration syntaxique: Subordination, coordination, connexion [Linguistische Arbeiten 351]. Tübingen: Max Niemeyer. O’Dowd, Elizabeth. 1992. The syntactic metaphor of subordination: A typological study. Lingua 86: 46–80. Osborne, Timothy. 2003. The Third Dimension: A Dependency Grammar Theory of Coordination for English and German. PhD dissertation, Pennsylvania State University. Ann Arbor MI: UMI. Osborne, Timothy. 2006. Shared material and grammar: Toward a dependency grammar theory of non-gapping coordination for English and German. Zeitschrift für Sprachwissenschaft 25: 39–93. Pasch, Renate. 2000. Vorschlag für eine Neuordnung der “subordinierenden” Konjunktionen des Deutschen. In Subordination in Syntax, Semantik und Textlinguistik, Michel Lefèvre (ed.), 23–32. Tübingen: Stauffenburg. Pasch, Renate, Brauße, Ursula, Breindl, Eva & Waßner, Ulrich Hermann. 2003. Handbuch der deutschen Konnektoren: Linguistische Grundlagen der Beschreibung und syntaktische Merkmale der deutschen Satzverknüpfer (Konjunktionen, Satzadverbien und Partikeln). Berlin: Walter de Gruyter. Pasch, Renate. 2004. Das “Handbuch der deutschen Konnektoren”. In Brücken schlagen: Grundlagen der Konnektorensemantik, Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner (eds), 11–44. Berlin: Walter de Gruyter. Peters, Jörg. 2005. Intonation. In Duden, Vol. 4: Die Grammatik, Dudenredaktion (ed.), 95–128. Mannheim: Dudenverlag. Polanyi, Livia. 1988. A formal model of the structure of discourse. Journal of Pragmatics 12: 601–638. Posner, Roland. 1980. Semantics and pragmatics of sentence connectives in natural language. In Speech Act Theory and Pragmatics, John R. Searle, Ferenc Kiefer & Manfred Bierwisch (eds), 169–203. Dordrecht: D. Reidel.

Subordination and coordination: Evidence from connectives Quirk, Randolph, Greenbaum, Sidney, Leech, Geoffrey & Svartvik, Jan. 1985. A Comprehensive Grammar of the English Language. London: Longman. Schecker, Michael. 2000. Zur kommunikativ-funktionalen Leistung von Nebensätzen. In Subordination in Syntax, Semantik und Textlinguistik, Michel Lefèvre (ed.), 115–122. Tübingen: Stauffenburg. Selting, Margret. 1995. Prosodie im Gespräch. Tübingen: Max Niemeyer. Shopen, Timothy (ed.). 1985. Language Typology and Syntactic Description, Vol. II: Complex Constructions. Cambridge: CUP. Taboada, Maite & Mann, William C. 2006. Rhetorical Structure Theory: looking back and moving ahead. Discourse Studies 8(3): 423–459. Thompson, Sandra A. & Longacre, Robert E. 1985. Adverbial clauses. In Language Typology and Syntactic Description, Vol. II: Complex Constructions, Timothy Shopen (ed.), 171–234. Cambridge: CUP. Van Valin Jr., Robert D. & LaPolla, Randy J. 1997. Syntax: Structure, Meaning and Function. Cambridge: CUP. Webber, Bonnie, Stone, Matthew, Joshi, Aravind & Knott, Alistair. 2003. Anaphora and discourse structure. Computational Linguistics 29: 545–587. Wegener, Heide. 2000. Asymmetrische Koordination in subordinierten Strukturen. In Subordination in Syntax, Semantik und Textlinguistik, Michel Lefèvre (ed.), 33–44. Tübingen: Stauffenburg. Wesche, Birgit. 1995. Symmetric Coordination: An Alternative Theory of Phrase Structure. Tübingen: Max Niemeyer. Wiese, Bernd. 1980. Grundprobleme der Koordination. Lingua 51: 17–44. Zifonun, Gisela. 2001. Grammatik des Deutschen im europäischen Vergleich: Der Relativsatz. Mannheim: Institut für Deutsche Sprache.



part ii

Cross-linguistic approaches

A corpus-based perspective on clause linking patterns in English, French and Dutch* Christelle Cosme

Université catholique de Louvain The two traditional clause linking devices, viz. coordination and subordination, are particularly interesting to study from a cross-linguistic perspective. It has often been claimed that principles of information packaging, of which clause linking forms an integral part, are – at least partly – language-specific. Thus, while some languages favour a hierarchical discourse information structure making intensive use of subordinating devices, others more readily employ an incremental discourse information structure, favouring coordinating devices. Close inspection of the contrastive literature reveals that English, French and Dutch exhibit preferred clause linking patterns. The claims found in the contrastive literature are tested empirically on the basis of multilingual corpora. Clause linking, it is ultimately argued, is best treated as a gradient phenomenon. Keywords: clause linking, coordination, subordination, stylistic preferences, multilingual corpora

1.  Introduction Clause linkage strategies have traditionally been divided into two basic types, i.e., coordination and subordination. While subordination organizes discourse hierarchically, coordinating relations are believed to structure discourse in a non-hierarchical, more incremental fashion (see Asher & Vieu’s (2005) treatment within Segmented Discourse Representation Theory and Mann & Thompson’s (1988) Rhetorical Structure Theory). The two clause linking devices are particularly interesting to study from a cross-linguistic perspective. Principles of discourse organization or information

*  This study has been made possible thanks to the support of the Belgian Fund for Scientific Research (Research Fellowship), which I gratefully acknowledge. My thanks also go to Sylviane Granger for her insightful comments on the first draft of the article. Last but not least, I wish to thank the three anonymous reviewers for providing useful feedback.

 Christelle Cosme

packaging, of which clause linking forms an integral part, have indeed been claimed to be – at least partly – language-specific (e.g., Hasselgård et al. 2002; Moder & Martinovic-Zic 2004) (Section 2). The present study takes a cross-linguistic view on sentence-level coordination and subordination in three languages, namely English, French and Dutch. The aim is to test a number of intuition-based claims made in the contrastive literature (Section 3). Such claims will be tested empirically on the basis of multilingual corpus data (Section 4). Another implicit aim of the research presented in this article is to gain new insights into the relationship between coordination and subordination. It will be argued, in line with among others Lehmann (1988) and Cristofaro (2003), that clause linking is best treated as a gradient phenomenon (Section 5). In other words, we hold the view that coordination and subordination should be regarded as the two prototypical poles of a clause linking continuum allowing for a number of in-between constructions.

2.  Information packaging across languages As witnessed by a number of recently published titles such as Discourse across languages and cultures (Moder & Martinovic-Zic 2004) or Information structure in a cross-linguistic perspective (Hasselgård et al. 2002), there is clear evidence that discourse organization differs across languages. In other words, principles of information packaging are – at least partly – language-specific. It appears that the language-specific ways of structuring information are manifest at various layers of discourse organization. To begin with, languages may differ with regard to focus assignment, thematic structure and word order patterns. Comparing English and German, Doherty (1999: 135) argues that German, unlike English, permits the identification of syntactic functions on a strictly morphological basis thanks to its case distinctions. As a result of its nonconfigurational character, German allows for greater variability in word order than English. By comparison, the latter, being a typically configurational language, requires specific structural configurations to distinguish subjects from objects. English subjects are therefore mainly fixed to the position before the verb (i.e., the initial position). As a result, while English sentences opening with an adverbial are marked, beginning a sentence with an adverbial is the unmarked pattern in German. Thus, a word-for-word translation of the German sentence given in example (1), though it is strictly speaking grammatically acceptable, would not be very likely in English. A translation placing the adverbial in sentence-final position appears to be stylistically more appropriate, as it adheres more tightly to the preferences of the target language (i.e., English).

(1) Zehn Jahre nach dem Röhrenversuch konnten Jaffe und Ken die

Beobachtungen bestätigen.

Jaffe and Ken confirmed the observations ten years after the tube experiment. [Doherty 1999: 125]

Clause linking in English, French and Dutch

Language-specific preferences have also been shown to be manifest at the level of informational density. It turns out that what is expressed by a phrase in one language may be expressed by a clause in another language, and vice versa. German, for instance, is generally assumed to have a more ‘nominal’ (as opposed to ‘sentential’) style than English and Norwegian (e.g., Solfjeld 1996). In other words, German is characteristically thought to make more intensive use of phrasal constituents than English and Norwegian, the latter favouring clausal constituents. Accordingly, translating from English into German frequently triggers so-called ‘phrasal reduction’ (Doherty 1998: 239), which can be described as a phenomenon whereby a clause in the source language is turned into a phrase in the target language (see also Doherty 1999 for a similar study). Example (2) may serve as illustration.

(2) Auch werden bei einem verlangsamten Zellwachstum nur geringe Mengen an natürlichem Interferon gefunden. Also when cell growth is slowing down, only small amounts of naturally produced interferon are found.  [Doherty 1999: 116]

Last but not least – and this is of special interest in the present article, languages are also said to differ with regard to their preferred clause linking patterns. Fabricius-Hansen (1998), for instance, observes that some languages (like German) favour a hierarchical (i.e., vertical) discourse information structure making intensive use of hypotactic (i.e., subordinating) devices, while other languages (like English and, to an even larger extent, Norwegian) prefer an incremental (i.e., horizontal or linear) discourse information structure, typically “organiz[ing] the discourse information structure in smaller chunks that can each be processed easily, with the result that a representation of the whole discourse is built up very gradually, step by step” (Fabricius-Hansen 1998: 203). Coordinating or paratactic constructions are typical ways of structuring discourse in an incremental fashion. Translating from a hierarchical language (like German) into a relatively more incremental language (like English or Norwegian) thus, as a rule, involves some degree of so-called ‘information splitting’, i.e., a shift from a high degree of syntactic complexity to a less complex, more paratactical style (Fabricius-Hansen 1996: 521). Example (3) shows how a typical complex sentence in German is turned into two independent sentences divided by a colon in English, and how a German subordinating construction (i.e., in this case, a relative clause) is rendered by means of a more horizontal, incremental construction (i.e., ‘and’-coordination) in English. (3) Wenn die unter jener anderen Landungsbrücke beheimateten Fische mein Kommen bemerkten, kam erschreckend When the fish under the other landing-stage noticed my coming, a startling thing happened: aus dem Dunkel unter dem Steg hervor ein mehrere Meter breites und fast ebenso hohes und viele Male längeres, auf dem besonnten Grund einen tiefschwarzen Schatten werfendes Ungeheuer auf mich zugeschossen,



 Christelle Cosme

from the darkness of the stage emerged a monster several yards high and wide, and many times this length, throwing a deep black shadow on the sunlit sea bottom as it shot towards me, das sich erst beim nahen Herankommen in eine Unzahl freundlicher Purpurmäuler auflöste. and only as it drew very near did it become resolved into a crowd of friendly grunts and snappers. [Fabricius-Hansen 1999: 135]

The above-mentioned language-specific preferences in terms of information packaging have an obvious impact on translation. A good translation has to follow the stylistic norms and preferences of the target language, which sometimes implies a certain amount of information restructuring. As argued by Chuquet & Paillard (1987: 135), following the language-specific strategies of information distribution is of paramount importance in translation, as it contributes substantially – to an even larger extent than lexical and grammatical choices – to the ‘natural-soundingness’ of a translation.

3.  Close-up on English, French and Dutch 3.1  Review of the contrastive literature A review of the contrastive literature dealing with English, French and Dutch brings out major contrasts in the way the three languages typically package information. It should be noted that the contrasts surveyed are located at the level of stylistic preferences. As rightly pointed out by Doherty (2005: 1), the cross-linguistic differences observed are thus “rarely grammaticalized, but determine only preferences in the distribution of information.” It will therefore be argued that, though devices like coordination and subordination are universal, they are employed with varying frequencies across languages. Thus, to convey a given content x, some languages may show a preference for device A (e.g., coordination), while others may favour device B (e.g., subordination).

3.1.1  English vs. French English and French have been claimed to diverge markedly in the way they prefer to package information in discourse. The contrastive literature abounds in claims, such as the following by Chuquet & Paillard (1987: 151), that the explicit linking of clauses tends to be realized through coordination in English and through subordination in French: Lorsque les relations entre les procès sont explicitées dans les deux langues, il est fréquent que ce soit sous forme de coordination en anglais, de subordination en français. On retrouve ici une des manifestations de la plus grande actualisation des procès en anglais […], marquée par l’emploi de formes verbales principales pour désigner des procès repères.

Clause linking in English, French and Dutch

Throughout the contrastive literature the term ‘coordination’ is taken in its very broad sense, as referring not only to explicit coordination by means of a conjunction (e.g., ‘and’), but also to implicit (so-called ‘asyndetic’) coordination by means of a punctuation mark (i.e., a comma, a semi-colon, a colon, or even a full stop). Since it presents the major advantage of allowing for the inclusion of sequences of two independent sentences in the analysis, this broad definition of coordination will also be adopted in the present article. The term ‘coordination’ as used in the present study thus broadly overlaps with the concept of ‘parataxis’ (see e.g., Martin 1992; Halliday & Matthiessen 2004). Two independent sentences strung together in sequence by means of a punctuation mark (i.e., a comma, a colon, a semi-colon, and even a full stop) will therefore be considered as coordination on an equal footing with explicit linking of clauses by means of a coordinating conjunction.1 As illustrated in example (4), the different preferences of English and French are particularly manifest when the relationship holding between the clauses is of a temporal nature. Thus, the English version of example (4), which contains four coordinated predicates in chronological sequence (see bold), is rendered in French by means of a blend of coordinating and subordinating devices.

(4) He stepped aside for me to enter the house, took my overcoat and hat,

guided me to a room on the second floor – Gantvoort’s library – and left me.

Il s’ écarta pour me laisser pénétrer dans la maison et après m’avoir débarrassé de mon pardessus et de mon chapeau, il me fit monter au premier et m’indiqua une pièce (la bibliothèque de Gantvoort) où il me laissa seul. [Chuquet & Paillard 1987: 151]

In other words, English and French differ in the way they relate events in time. English seems to have a strong tendency to view events on a temporal axis (i.e., to present them in a chronological sequence by means of coordination), while French is presented as typically linking events more tightly together, thus making more intensive use of subordinating devices (Chuquet & Paillard 1987: 151).

3.1.2  English vs. Dutch The English-Dutch contrastive literature is much less abundant than that dealing with English and French. There is, however, some evidence that English and Dutch have preferred ‘sentencing styles’ – to use Hannay’s (1997) words – that differ quite strikingly. As suggested by Hannay & Mackenzie (1996: 102), the syntactic devices used in the

.  This broad definition of ‘coordination’ blurs the functional distinction between explicit coordination and (asyndetic) parataxis (see e.g., Ramm & Solfjeld, this volume). The decision to adopt such a broad definition of ‘coordination’ has been primarily motivated by its close adherence to the definition assumed in the English-French and English-Dutch contrastive literature.



 Christelle Cosme

two languages differ in terms of availability, frequency and distribution, and function: “Some devices are not readily available in Dutch; others are more flexible in English than in Dutch; yet others just operate in different ways than their Dutch counterparts.” Generally speaking, the contrastive literature presents English as being syntactically more complex than Dutch. Hannay and Mackenzie, for instance, claim that: English has a tendency in argued text to use more sentences than Dutch which are long (containing say over 30 words) and syntactically complex. It is for instance not at all uncommon in English for sentences to have more than one combined clause; indeed, sometimes a combined clause itself functions as the host clause for another combined clause.  (Hannay & Mackenzie 1996: 118–119)

Sentence length is a first indicator of syntactic complexity. According to Hannay & Mackenzie (1996: 44), “[c]omparison of similar English and Dutch texts has shown that sentences in English, at least in sustained argued prose, tend to be on average somewhat longer than their Dutch counterparts.”2 The greater syntactic complexity of English, it is claimed, also manifests itself in a more intensive use of subordinate clauses. In her investigation of Dutch novels translated into English, Vanderauwera (1985: 78) notes that Dutch favours short and incomplete sentences and typically uses coordinate rather than subordinate clauses. In the English translation, the author observes, textual organization is frequently adjusted to meet the English norms, as is clearly the case in example (5).

(5) Hij gaat naar boven, naar de huiskamer, ziet dat het nu helemaal donker

is, staart nietsziend naar buiten, voelt zich eenzaam, weet nu al in welk restaurant hij straks zal gaan eten. Hij trekt zijn jas aan en gaat naar een bar in de benedenstad om kranten te kopen, iets te drinken en misschien op de pinballmachine te spelen.

Upstairs he goes to the sitting-room; noticing that it is already dark, he gazes outside not seeing a thing. Now he feels lonely, but already he knows at what restaurant he will be eating presently. When he has put on his coat he steps out to make his way to the bar downtown where he will buy newspapers, have a drink and probably end up playing the pin-table.  [Vanderauwera 1985: 101–102]

.  Hannay & Mackenzie (1996: 44) see two underlying reasons for this phenomenon. First, Dutch is more tolerant than English of incomplete sentences. As a result, “what would in Dutch be expressed by an incomplete sentence appears in English as a full sentence” or “the information presented in Dutch as a separate but incomplete sentence is in English integrated into the preceding sentence, making the latter correspondingly longer” (Hannay & Mackenzie 1996: 44). Another reason for the higher sentence length of English texts is the frequent use English makes of participial constructions, for which there is no real equivalent in Dutch.

Clause linking in English, French and Dutch

In relation to such typical textual adjustments, Vanderauwera (1985: 101) points out that “[t]he simple and jerky juxtaposition of syntactically identical and independent clauses suggestive of almost cinematic observation […] is matched here by more syntactically varied and “better” linked narrative portions (by use of such means as subordinate or nonfinite clauses).” It thus appears that while the English text types are characterized by a combining style, the Dutch text types clearly demonstrate a more chopping style (Hannay 1997: 248).

3.2  Research hypothesis The review has brought out major contrasts in the strategies pursued by English, French and Dutch to link clauses. The three languages appear to be located on a continuum ranging from incrementality to hierarchy (see Fabricius-Hansen 1996: 558 for definitions of these two concepts). The two ways of organizing discourse (i.e., incrementality vs. hierarchy) are accurately reflected, at the linguistic level, in the two clause linking devices, i.e., in the distinction between coordination and subordination. An incrementally organized text makes use of a relatively large number of coordinating constructions and, respectively, a relatively low number of subordinating devices; in a hierarchically organized text, by comparison, there is a clear predominance of subordinating structures and, conversely, a relatively low number of coordinating constructions.3 The hypothesis – henceforth referred to as the ‘Dependency Hypothesis’ – thus reads as follows and may be graphically represented as in Figure 1:

INCREMENTALITY Parataxis (coordination)

DUTCH

HIERARCHY Hypotaxis (subordination)

ENGLISH

FRENCH

Figure 1.  Dependency Hypothesis.

.  The distinction between incremental style and hierarchical style coincides with Chafe’s (1982) distinction between fragmentation and integration. A fragmented style, Chafe (1982: 38) argues, is characterized by a more intensive use of coordinating structures, whereas an integrated style typically makes greater use of subordinating constructions (e.g., participial clauses, complement clauses or relative clauses).



 Christelle Cosme

It should be noted that the scale represented in Figure 1 is not to be taken in absolute terms. Rather, incrementality and hierarchy are to be considered as relative concepts. In other words, the hypothesis posits that French, in comparison to English, favours a more hierarchical style and that Dutch, in comparison to English, has a typically more incremental writing style. The different positions of the three languages on the dependency cline represented in Figure 1 have a clear impact on the translation process. It may indeed be postulated that shifts in clause linking will take place. The following shifts may be predicted from the Dependency Hypothesis: a. Majority of shifts from coordination (in the broad sense, including mere juxtaposition by means of punctuation marks) to subordination in translations from Dutch to English, from English to French and from Dutch to French; b. Majority of reverse shifts (i.e., from subordination to coordination) in translations from French to English, from English to Dutch and from French to Dutch. These predictions will be tested against translation corpus data for three of the six possible translation directions, namely (i) from English to French, (ii) from French to English and (iii) from English to Dutch. Before engaging in the corpus analysis proper, a few words about Vinay & Darbelnet’s (1995: 15) distinction between so-called ‘options’ and ‘servitudes’ are in order, as the distinction is highly relevant to the research presented here. It emerges that most of the shifts under study are cases of options rather than servitudes. In other words, the translator could most of the time have maintained a congruent construction in the target language, but chose not do so probably because s/he was aware of the stylistic norms and preferences of the target language. Thus, though translating the English ‘and’-coordination of example (6) with ‘et’-coordination in French would have yielded a grammatically correct sentence in French, a translation by means of a subordinate construction is stylistically more appropriate.

(6) In the course of years several ships had stranded and they now helped to

prop up the bank, […].

Plusieurs navires, qui s’étaient échoués au cours des années, contribuaient maintenant à étayer la rive, […]. [pleci_news]

Though most of the shifts under investigation are optional, a number of them are the result of language servitudes or language-systemic constraints. Thus, to translate the English sentence given in (7), the translator had no choice but to operate a shift. A word-for-word translation would have been grammatically unacceptable here, as adverbial –ing constructions do not have any direct equivalent in Dutch. (7) Ralph spoke to himself, sounding the bass strings of delight. Ralph praatte in zichzelf en bracht de bassnaar van verrukking in trilling. [pleci_news]

Clause linking in English, French and Dutch

4.  Corpus study This section presents the findings of a large-scale corpus study, conducted in an attempt to test and refine the Dependency Hypothesis on the basis of solid empirical data. The reader is referred to Cosme (2004, 2006a) for an overview of preliminary findings concerning the contrast in clause linking between English, French and Dutch.

4.1  Data and methodology The corpus analysis conducted here is two-fold. To start with, a number of preliminary observations are made on the basis of (under ‘corpus’) data (see Section 4.2.1), i.e., original texts in two or more languages that are matched by criteria such as time of composition, register, etc. (see e.g., Johansson & Hasselgård 1999 and Granger 2003 for a definition of comparable corpora). Concretely, the comparable data come from the TRILLED corpus (TRILingual Louvain EDitorials corpus). Currently being compiled by researchers at the Centre for English Corpus Linguistics (University of Louvain, Belgium), the corpus is made up of original newspaper editorials in English, French and Dutch.4 At its present state, it contains ca. 1 million words of English editorials, ca. 750,000 words of French editorials and ca. 500,000 words of Dutch editorials. For the purpose of the present research, we have extracted a random sample of 300 sentences in each language. A careful analysis of these 300 sentences per language, it is believed, should provide useful insights into the frequency of the two clause linking devices (i.e., coordination and subordination) in the three languages. In a subsequent step, use is made of a trilingual (under ‘corpus’) corpus called PLECI (Poitiers-Louvain Échange de Corpus Informatisés) in order to retrieve the translation shifts of interest (see Section 4.2.2). The corpus, which is currently being compiled by the Centre for English Corpus Linguistics (University of Louvain, Belgium) and the University of Poitiers (France), is made up of fiction and newspaper articles. At its present state, it comprises ca. 600,000 words of English texts translated into French, ca. 700,000 words of French texts translated into English and ca. 200,000 words of English texts and their respective Dutch translations. Because of the difficulties connected with the automatic retrieval of the shifts under investigation, it was decided to use only a sample of PLECI, namely ca. 50,000 words from the English ⇒ French component, ca. 50,000 words from the opposite direction (i.e., French ⇒ English) and ca. 25,000 words from the English ⇒ Dutch component. Note that while fiction and newspaper language

.  Comparability in the TRILLED corpus is ensured by selecting editorials from quality newspapers in the three languages. English editorials are taken from The Guardian, The Independent, The Times, The Observer, The Sunday Telegraph, The Daily Telegraph and The Economist; French editorials from Le Monde, Le Figaro and Libération; and Dutch editorials from De Volkskrant, NRC Handelsblad, Trouw, Het Parool, Utrechts Nieuwsblad and Haagsche Courant. Moreover, the editorials were collected during the same period of time.



 Christelle Cosme

are both equally represented in the English-French sub-corpus and in the French-English sub-corpus, the same does not apply to the English-Dutch sub-corpus, which is – at its present state – restricted to fiction.5 In order to reinforce the English ⇒ Dutch part of the study, we have also searched the Oslo Multilingual Corpus (OMC).6 The present study thus combines the respective strengths of comparable and translation data. Johansson and Hasselgård (1999: 146) were among the first to emphasize the benefit of combining comparable and translation corpora, thus taking advantage of the specific merits of both types. The view held in this article is in accordance with Johansson & Hasselgård’s (1999: 146) principle. The basic assumption is that comparable and translation data are not mutually exclusive. Rather the contrary: they can supplement one another. The two types of data will therefore be necessary to gain a deeper understanding of the relationship between the three languages under investigation in our study. While lexis lends itself perfectly to corpus investigations, the same does not always hold for syntactic phenomena, the chief reason being that syntactic structures are usually more difficult to extract from a corpus than simple words (Gilquin 2002: 183). Because of the difficulty connected with the automatic retrieval of macro-linguistic patterns, most contrastive studies dealing with syntax or discourse are still to a large extent dependent on intuition, or resort to corpus examples only to illustrate or support the claims being made. It must, however, be acknowledged that, though it is true that lexis has the lion’s share in present-day contrastive linguistics, much effort has been made in recent years to bridge the gap between lexical contrastive studies, on the one hand, and syntactic or discourse studies, on the other.7 Though it must be admitted that the use of fully automatic methods is generally impossible in corpus-based syntactic or discourse-oriented contrastive research, the

.  The size of the English-Dutch sub-corpus has been deliberately kept to 25,000 words so as to facilitate comparisons with the 25,000-word English-French fiction sub-corpus and the 25,000-word French-English fiction sub-corpus. The decision to include two genres (i.e., fiction and newspaper articles) in the English-French and French-English sub-corpora has been made on the grounds that, at a later stage, we also intend to investigate genre differences (see Cosme 2006a for additional information). .  We hereby thank Stig Johansson and Hilde Hasselgård for allowing us to search the EnglishDutch sub-corpus of the Oslo Multilingual Corpus (http://www.hf.uio.no/ilos/OMC/) compiled within the framework of the Språk i kontrast (Languages in Contrast) project at the University of Oslo (http://www.hf.uio.no/forskningsprosjekter/sprik/english/). 7.  In the Scandinavian countries, for instance, a number of contrastive studies based on the ENPC (English-Norwegian Parallel Corpus) or on the ESPC (English-Swedish Parallel Corpus) have dealt extensively with macro-linguistic phenomena. The SPRIK project (Språk i kontrast / Languages in Contrast; see http://www.hf.uio.no/forskningsprosjekter/sprik/english/), for instance, has given rise to a series of corpus-based syntactic and discourse-oriented contrastive studies (e.g., Fabricius-Hansen 1998).

Clause linking in English, French and Dutch

view held in this article is that it is possible to automate the process to some extent. Computerized methods are thus adopted whenever technically feasible. Thus, in order to ease the retrieval of the structures under study, it was decided to provide both the comparable and the translation corpus data with problem-oriented linguistic annotations using an automated method. Providing greater detail on the annotation scheme and procedure falls beyond the scope of the present article. The reader is referred to Cosme (2003) and Cosme (2006b) for more information on the annotation of comparable and translation data, respectively.

4.2  Corpus findings and hypothesis testing 4.2.1  Preliminary observations A comparison of English, French and Dutch newspaper editorials provides some useful insights into the frequency of the two clause linking strategies, i.e., coordination and subordination, across the three languages. The present section addresses the issue of whether English, French and Dutch employ coordination and subordination differently. The objective is to test the hypothesis that Dutch uses coordination more often than English and French and that French, with its intensive use of subordination, is the most syntactically complex of the three languages. 4.2.1.1  Subordination.  A look at the overall frequency of subordinate clauses – both hypotactic and embedded ones – in the three sub-corpora already reveals something about the different nature of Dutch. While English and French contain a roughly similar number of sub-clauses (464 and 415, respectively), Dutch uses significantly fewer subordinate clauses (255 only). Thus, while the mean number of sub-clauses per sentence (SUB/SEN) in English is essentially identical to that in French (1.55 and 1.38 respectively), it appears to be markedly lower in Dutch (0.85) (see Table 1). An analysis of variance (ANOVA) shows that there is considerable variation in terms of SUB/SEN across the three languages (F(2,897) = 23.29, p ≤ 0.0001). Table 1.  Mean number of subordinate clauses per sentence (SUB/SEN)

ENGLISH

FRENCH

DUTCH

No. of subclauses (SUB) No. of sentences (SEN) Mean no. of subclauses / sentences (SUB/SEN)

464 300 1.55

415 300 1.38

255 300 0.85

Closer inspection of the languages on a two-by-two basis – by means of a t-test procedure – reveals, however, that the difference in SUB/SEN is significant between English and Dutch (t(535) = – 7.19, p = 0.0001) and between French and Dutch



 Christelle Cosme

(t(505) = – 5.14, p = 0.0001) only, hence not between English and French (t(592) = 1.38, p = 0.17).8 The corpus sentences may also be classified according to the number of subordinate clauses they contain. For the purpose of the present research, a three-fold division has been developed. Sentences containing no sub-clause are referred to as ‘zero-sub sentences’; those that contain one single subordinate clause are labelled ‘mono-sub sentences’ and the term ‘multi-sub sentences’ is used for sentences containing two or more subordinate clauses. The exact breakdown of zero-sub vs. mono-sub vs. multisub sentences in all three sub-corpora is summarized in Figure 2. 160

120 100 80

135

133

140

103

102 82

83

114 ENGLISH

84 64

60

FRENCH DUTCH

40 20 0

Zero-sub

Mono-sub

Multi-sub

Figure 2.  Zero-sub vs. mono-sub vs. multi-sub sentences.

It emerges from Figure 2 that Dutch exhibits a larger number of zero-sub and monosub sentences, and conversely, a much lower proportion of multi-sub sentences than English and French. The difference between English and French, on the one hand, and Dutch, on the other, as to their use of zero-sub vs. mono-sub vs. multi-sub sentences proves highly significant (χ² = 40.82, df = 4, p ≤ 0.001). As for the difference between English and French, it is again only slightly (i.e., not significantly) marked as regards their respective use of multi-sub and zero-sub sentences, English using slightly more multi-sub sentences, and respectively slightly fewer zero-sub sentences, than French. This preliminary finding only partially confirms the Dependency Hypothesis. While the corpus study clearly shows that Dutch employs fewer subordinating devices

.  Tests measuring the heterogeneity of variance have been conducted for each language pair. The value of p being always below 0.20 (considered the threshold in such tests), variances have to be considered unequal in all three sub-corpora. The results of the t-tests were thus considered accordingly.

Clause linking in English, French and Dutch 

than English and French (thereby positioning Dutch towards the ‘least hierarchical’, and hence ‘most incremental’, end of the spectrum (see Figure 1)), it makes no strong claims about the English-French contrast. French does not seem to be more hierarchical than English in any respect. On the contrary, as witnessed by its higher mean number of sub-clauses per sentence (see Table 1) and its larger number of multi-sub sentences (see Figure 2), English even seems to be using slightly – though not significantly – more subordinating constructions than French. The role of the translation corpus analysis (Section 4.2.2) will be to refine this preliminary finding.

4.2.1.2  Coordination.  A frequency-based analysis of the coordinating constructions – both explicit/syndetic and implicit/asyndetic ones – in the three sub-corpora reveals that French is markedly less fond of paratactic structures than English and Dutch. Though an analysis of variance (ANOVA) of the mean number of coordinating constructions per sentence (amounting to 0.09 in French, 0.16 in English and 0.14 in Dutch) reveals no significant differences (F(2,897) = 2.32, p ≤ 0.10), the discrepancy between French, on the one hand, and English and Dutch, on the other, is worth pointing to: While the English and Dutch sub-corpora contain, respectively, 50 and 42 instances of coordination, the French sub-corpus scores 28 only. A more delicate analysis of the syntactic patterns displayed by the coordinating constructions indicates, however, that the lower frequency of coordination in French is to a large extent attributable to so-called ‘VP-conjunction’, i.e., constructions in which the conjoined units are not full clauses but predicates sharing a subject or predications sharing a subject and auxiliaries (see example (8)). As appears from Table 2, which displays the raw frequencies of coordination of full clauses (COORFULL; see example (9)) vs. VP-conjunction (VP-CONJ) in the three sub-corpora, French stands out by using a higher percentage of COORFULL, and conversely, a lower proportion of VP-CONJ than English and Dutch.

(8) Palestinian police have arrested three people and ___ are searching for two others as part of their investigation into the bombing that killed three Americans. [trilled] (9) Traditionally, activity increases in the evenings during Ramadan after Muslims break their day-long fast, and shops often stay open until early morning. [trilled]

Table 2.  Coordination of full clauses vs. VP-conjunction

ENGLISH

FRENCH %

#

DUTCH



#

%

COORFULL VP-CONJ Total

26   52.0 19   67.9 24   48.0   9   32.1 50 100.0 28 100.0

#

%

26   61.9 16   38.1 42 100.0

 Christelle Cosme

An analysis of variance (ANOVA) comparing the mean numbers of coordination of full clauses and VP-conjunction across the three sub-corpora shows that, while the differences for COORFULL (0.63 for French and 0.87 for English and Dutch) are not significant (F (2,897) = 0.53, p ≤ 0.59), the differences for VP-CONJ (0.03 for French, 0.07 for English and 0.05 for Dutch) nearly reach significance (F (2,897) = 2.67, p ≤ 0.07). Thus, as clearly emerges from Figure 3, the French ‘under-representation’ of coordinating constructions – as compared to English and Dutch – is especially noticeable as regards VP-conjunction. 60 50 40

24

16

30

10 0

COORFULL

9

20 26

ENGLISH

19 FRENCH

VP-CONJ

26

DUTCH

Figure 3.  Coordination of full clauses vs. VP-conjunction.

As is the case for the analysis of subordination (Section 4.2.1.1), the analysis of coordination thus only partially confirms the Dependency Hypothesis. While French clearly seems to be using fewer coordinating structures than English and Dutch (thereby tending towards the ‘least incremental’, and hence ‘most hierarchical’, end of the spectrum (see Figure 1)), there is no evidence to suggest that Dutch resorts more often to coordination than English, as the English and Dutch sub-corpora behave in a surprisingly similar fashion in that respect. The analysis also enables us to further refine the hypothesis, by postulating that the French ‘underuse’ of coordinating constructions – as compared to English and Dutch – is largely due to VP-conjunction (as opposed to coordination of full clauses).

4.2.2  Translation shifts The aim of the present section is to supplement the comparable corpus analysis by examining another type of data, namely translation corpus data. 4.2.2.1  English ⇔ French.  A cursory glance at the shifts involved in translations from English to French and from French to English seems to corroborate the predictions. Figure 4 shows that in English ⇒ French translations there is a majority of shifts from coordination to subordination, while in French ⇒ English translations there is an overwhelming majority of reverse shifts, i.e., from subordination to coordination.

Clause linking in English, French and Dutch  160 135

140 120 100

90 COORD => SUB

80

65

SUB => COORD

60

60 40 20 0 EN => FR

FR => EN

Figure 4.  Translation shifts in English ⇒ French and French ⇒ English.

In what follows, we will focus exclusively on the two most represented shifts, viz. (a) shifts from coordination to subordination in English ⇒ French and (b) shifts from subordination to coordination in French ⇒ English. A number of recurrent patterns emerge from a careful analysis of the translation corpus data. It appears that the set of French subordinating constructions most frequently corresponding to English coordination (in the broad sense) is fairly limited. Table 3 gives an overview of the most regular patterns of correspondence. Note that the term ‘correspondence’ is meant here to refer not only to French target structures used to translate English coordination, but also to French source structures giving rise to English coordination. Table 3.  French subordinating structures corresponding to English coordination

Relative clauses Participle clauses Present Past ‘Gérondifs’ (‘en’ + present participle) Finite adverbial clauses pour + infinitive Others∗ TOTAL

EN ⇒ FR Target equivalents

FR ⇒ EN Source equivalents

TOTAL

19 20 10 6

49 34 14 15

  68 (30%)   54 (24%)   24   21

4 23 9 19 90

5 25 6 20 135

   9   48 (22%)   15 (7%)   39 (17%) 225

* The category ‘Others’ includes non-finite adverbial clauses (e.g., introduced by avant de (‘before’)), verbless clauses, nominal clauses, etc.

 Christelle Cosme

It emerges from Table 3 that coordination in English most frequently corresponds to a relative clause in French. Relative clauses, it appears, account for nearly a third (30% exactly) of all the shifts mentioned in Table 3. What is particularly striking in these shifts is that in 88.2% of the cases (i.e., 60 cases out of 68), it is the second conjoined clause in English that corresponds to a relative clause in French (see Ballard 1995: 266). Example (10) is thus representative. Cases where it is the first conjoined clause in English that corresponds to a relative clause in French (see example (11)) thus seem to form exceptions. (10) Elles étaient filmées par les services officiels, qui envoyaient ensuite les reportages aux petites stations américaines […]. These were then filmed by the official press services, and the footage was distributed to small US TV stations […]. [pleci_news] (11) Le plus âgé, qui pour la première fois de sa vie n’a pu accompagner son

troupeau dans sa transhumance annuelle, ne parvient pas à comprendre pourquoi je suis venu de si loin.

For the first time in his life the father of the family was unable to accompany his flock on its annual migration this year. He cannot understand why I have journeyed so far. [pleci_news] (12) Il sortit de son portefeuille un instantané jauni qu’il tendit à son visiteur. He drew a yellow snapshot out of his note-case and handed it over. [pleci_news]

Careful analysis of the corpus data also reveals that the functions most generally fulfilled by the relative pronoun in such shifts are those of subject (in 28 cases out of 68, i.e., 41% of the cases) (see example (10)) and – to a lesser extent – direct object (in 11 cases out of 68, i.e 16% of the cases) (see example (12)). Another frequent type of correspondence is that between French participle clauses and English coordination. Participle clauses, it appears, account for about a quarter (24% exactly) of all the shifts mentioned in Table 3. As indicated in Table 3, French participle clauses fall into three types, viz. (a) present participle clauses (see example (13)), (b) so-called gérondifs, formed with ‘en’ plus present participle (see example (14)) and (c) past participle clauses (see example (15)). (13) Ralph stood, one hand against a grey trunk, and screwed up his eyes

against the shimmering water.

Ralph se tenait appuyé contre un tronc gris, plissant les yeux pour regarder la surface miroitante de l’eau. [pleci_news] (14) […] but she only shook her head and muttered about contamination. […] mais elle s’est contentée de secouer la tête en marmonnant quelque chose à propos de la contamination. [pleci_news]

Clause linking in English, French and Dutch 

(15) Etablie par des anciens de la Harvard Business School, elle s’ attacha

avec succès à acclimater les méthodes pratiquées aux Etats-Unis dans un environnement européen.

It was started by Harvard Business School graduates and succeeded in adapting US methods to suit a European environment, with English-language teaching. [pleci_news]

Ballard (1995: 259–260) argues that cases of correspondence between English coordination and French present participle clauses occur predominantly when the semantic relationship holding between the two coordinated clauses is consequential. If the principle applies in certain cases, it is not as generalizable as Ballard’s (1995: 259–260) claim suggests. The English sentence of example (13), for instance, does not involve any cause-and-effect relationship, and yet its French translation is a present participle clause. By contrast, the correspondence between English coordination and French so-called gérondif occurs mostly, Hoarau (1997: 56ff) argues, when the meaning is one of simultaneity. Examination of the corpus data shows that this is nearly always the case. Example (14) is thus representative in that respect, as the shaking event and the muttering event are presumed to take place simultaneously. Hoarau (1997: 56ff) adds that this type of correspondence is more likely to obtain when English coordination joins a verb denoting an activity (e.g., ‘to shake’ in (14)) and a verb of speaking (e.g., ‘to mutter’ in (14)). This claim, however, cannot be substantiated, for lack of sufficient corpus data, example (14) being the only occurrence found in the translation corpus. The correspondence between English coordination and past participle clauses in French (see example (15)) also seems to follow a set pattern. As is the case in example (15) above, the French construction corresponding to English coordination is most of the time (in 62% of the cases exactly) an initial past participle clause premodifying the subject (see Ballard 1995 and Hoarau 1997 for additional examples). English coordination also frequently corresponds to a finite adverbial clause in French. It appears from Table 3 that such correspondences account for 22% of all the shifts mentioned in the table. While relative clauses and participle clauses are not per se more specific (i.e., more explicit) than coordination, the same does not apply to finite adverbial clauses, which generally imply a higher degree of explicitness than coordination. While cases where English ‘and’ is turned into a finite adverbial clause in French are quite common in the corpus data (see example (16)), cases where a finite adverbial clause in French is rendered by means of coordination (in the broad sense, including implicit linking by means of a punctuation mark) in English are equally frequent (see example (17)). (16) Il avait disparu pour se dissoudre dans la lumière jaune, pâle et indécise tandis que la porte se refermait He was gone, dissolving in the pale, uncertain, yellow light, and the door

was closing.

[pleci_news]

 Christelle Cosme

(17) Cette analyse et cette politique devaient conduire à de profondes divergences

avec les Etats-Unis, comme on le vérifia dès la rencontre, le 5 juillet 1958, entre le général de Gaulle et le secrétaire d’Etat américain John Foster Dulles.

His analysis and the policies it inspired resulted in profound disagreement with the US. This became immediately apparent, in July 1958, when De Gaulle met the then US secretary of state, John Foster Dulles. [pleci_news]

Corpus evidence suggests that such shifts mostly occur when the relationship between the clauses is of a temporal nature (see example (16)). Another interesting – though less frequent – type of correspondence is that between English coordination and the construction ‘pour’ + infinitive in French. In French, it is not quite clear whether the construction expresses purpose or mere chronological sequence. There is thus semantic indeterminacy (Chuquet & Paillard 1987: 152). When the subject is endowed with intentionality, as is the case in example (18) for instance, it seems as though the meaning can more readily be interpreted as being one of purpose. Conversely, when the subject cannot be endowed with intentionality (e.g., when it refers to some inanimate being), the meaning is more likely to be one of chronological sequence (see example (19)). This ambiguity of meaning of the construction ‘pour’ + infinitive in a way makes it bear some similarity to coordination. (18) E  n 1991, près de deux millions d’individus ont fait passer de la nourriture, des télévisions, des radios et de l’essence en Pologne pour ramener des vêtements polonais […] et des biens de consommation. In 1991 nearly 2m people ferried food, televisions, radios and petrol to

Poland and brought back Polish clothes […] and consumer goods.

[pleci_news]

(19) Sur la droite, un ruisseau, qu’enjambait un pont de pierre, traversait ce champ pour se perdre dans la mer. To the right of the cottages a stream flowed beneath a low stone bridge, crossed the meadow and lost itself in the sea. [pleci_news]

It should be noted, however, that the above-mentioned typical cases of correspondence are to be interpreted as tendencies rather than strict rules. A number of ‘countertendencies’, i.e., (a) shifts from subordination to coordination in English ⇒ French translations and, vice versa, (b) shifts from coordination to subordination in French ⇒ English translations, may be observed. These shifts, it appears, mainly involve correspondence between French coordination (in the broad sense) and English –ing clauses (see example (20)). (20) -Je ne suis pas une rapporteuse, dit Patricia. Elle me défia du regard. ‘I’m not a tell-tale,’ Patricia said, staring at me defiantly. [pleci_news]

Clause linking in English, French and Dutch 

4.2.2.2  English ⇒ Dutch.  A cursory glance at the shifts involved in translations from English to Dutch seems to confirm the predictions. Figure 5 shows that the majority (69% exactly) of shifts involved in that specific translation direction are shifts from subordination to coordination. COORD => SUB 31%

SUB => COORD

69% Figure 5.  Translation shifts in English ⇒ Dutch.

Close examination of the shifts involved in the English-Dutch sub-corpus of the Oslo Multilingual Corpus (OMC) further confirms this finding, also showing a predominance (67%, i.e., 159 cases out of 239) of shifts from subordination to coordination. A particularly regular pattern stands out from the corpus data, namely the translation of –ing clauses in English by means of coordination in Dutch. This translation strategy – of which examples (21) and (22) are representative illustrations – accounts for about three quarters (73% exactly) of all the shifts from subordination to coordination occurring in the English ⇒ Dutch component of PLECI. (21) T  wee bergketens lopen zo ongeveer van noord naar zuid door de republiek en vormen een aantal dalen en plateaus. Two mountain chains traverse the republic roughly from north to south,

forming between them a number of valleys and plateaux.



[pleci_news]

(22) ‘Dit,’ zei ze, ‘is mijn dochter Sara.’ Ze sprak de naam zo uit dat de eerste lettergreep rijmde op ‘kar’. She said, ‘This is my daughter Sara,’ pronouncing the name so that the first

syllable rhymed with ‘car’.

[pleci_news]

It should be noted that such shifts are mostly cases of servitudes (i.e., constraints), insofar as –ing clauses do not have any direct equivalent in Dutch. As has been shown to be the case for the English-French pair, the shifts observed in English ⇒ Dutch translations are to be interpreted as tendencies. In other words, it is not always the case that translating an English text into Dutch necessarily implies turning all subordinating constructions into coordinating ones. Sometimes, the

 Christelle Cosme

opposite tendency holds. The most typical ‘counter-tendency’ is reflected in example (23), where coordination in English is turned into an infinitival clause of purpose in Dutch. (23) We zouden graag de schildpaden uit hun poel willen halen om ermee te spelen. We would like to get the turtles out of their pool and play with them. [omc]

5.  Clause linking as a gradient phenomenon Besides shedding light on the contrast in clause linking between English, French and Dutch (Section 4), the corpus study also touches upon a more theoretical issue, namely the issue of clause linking as a continuum (rather than as a strict binary system). Important aspects of the relationship between ‘and’-coordination and subordination have emerged. Table 4 summarizes the assumed traditional dichotomy made between coordination and subordination. Table 4.  Clause linking as a binary system 1. Syntactic criteria Dependency Embedding 2. Semantic criteria Backgrounding Explicitness

Coordination

Subordination

– –

+ +

– –

+ +

As pointed out by Cristofaro (2003: 15), the criteria used to distinguish between coordination and subordination are essentially syntactic in nature. Subordination is said to be readily identifiable by a number of properties, such as (a) dependency, i.e., the impossibility for a subordinate clause to occur in isolation, and (b) embedding, i.e., the intrinsic property of a subordinate clause to function as a constituent of the matrix clause. These two features are traditionally thought to differentiate subordination from coordinating constructions, insofar as the latter involve clauses that are both independent and non-embedded. The difference between coordination and subordination can also be accounted for in semantic terms. First, it is generally assumed that subordination is more explicit than ‘and’-coordination, the latter leaving the inference to the reader (Quirk et al. 1985: 1040). Another fundamental difference between coordination and subordination is said to lie in the way in which the states of affairs expressed by the connected clauses are perceived and conceptualized and the status they have in the discourse context. Broadly speaking, “the information in a subordinate clause is

Clause linking in English, French and Dutch 

often placed in the background with respect to the superordinate clause” (Quirk et al. 1985: 919). By contrast, the contents of coordinated clauses are commonly held to have equal informational weight. On closer examination, however, the traditional dichotomy between the two clause linking devices – as expressed in Table 4 – turns out to be not as clear-cut as generally assumed. We agree with Quirk et al.’s (1985: 919) claim that “coordination is related by gradience to subordination.” First, dependency (i.e., the impossibility for a clause to occur in isolation), which is taken to be distinctive of subordination, is not a clearly delineated phenomenon, but is rather a matter of degree (Givón 2001: 327). The degree of dependency of a clause may, for instance, vary according to the type of coordination involved. The clauses conjoined in (24) are syntactically independent of one another, each having its own subject and verb. The second clause of (25), by comparison, exhibits a higher degree of dependency, because it depends on the first clause for the expression of one argument, namely the subject, and lacks an illocutionary force of its own. Likewise, a non-finite subordinate construction like that found in (27) exhibits a higher degree of dependency than a finite subordinate construction like that given in (26), since it lacks tense and an illocutionary force of its own. (24) The winter had come at last, and snow lay thick on the ground. [Quirk et al. 1985: 946] (25) Peter ate the fruit and drank the beer.

[Quirk et al. 1985: 948]

(26) I can’t go out with you because I am studying this evening. [Quirk et al. 1985: 992] (27) Her aunt having left the room, I asked Ann for some personal help. [Quirk et al. 1985: 993] (28) I noticed that he spoke English with an Australian accent. [Quirk et al. 1985: 1049] (29) We have no electricity, because there is a power failure.  [Quirk et al. 1985: 1072]

Embedding is not a clear-cut criterion either. A nominal clause (see example (28)), for instance, is more deeply embedded in the matrix clause than an adverbial clause (see example (29)). Thus, systemic grammar (e.g., Matthiessen & Thompson 1988; see also Givón’s (2001) scale of inter-clausal dependency) makes a sharp distinction between true embedding and hypotaxis. While the nominal clause in (28) belongs to the first category, on the grounds that it functions as an obligatory constituent of the main clause, i.e., as direct object, the adverbial clause in (29) falls into the second, because it is only there to provide additional information on the state of affairs described in the matrix clause and is therefore more loosely attached to the matrix clause (i.e., less embedded). Lastly, there are also cases where the semantic criteria (see Table 4) fail to be distinctive of subordination. Consider example (30), for instance, in which a situation

 Christelle Cosme

that is part of the narrative sequence (i.e., of the foreground) is found in a subordinate adverbial clause (see bold). (30) Only after he stopped smiling and shrieking did he go to Stephanie and hug her. That hug was also interrupted by additional shrieks. Quite a lot of noise from a normally silent chimpanzee! After spending about fifteen minutes with Stephanie, Nim went over to WER, Josh, and Jeannie, and hugged each of them in turn. (Thompson 1987: 443)

There are thus fundamental weaknesses in the traditional binary opposition between coordination and subordination. It emerges that the boundary between the two clause linkage types is blurred and that the various criteria conventionally applied to draw the line between them are not as clear-cut as they might seem at first sight. Rather, coordination and subordination should be considered as gradual phenomena. Cristofaro (2003) summarizes this view – known as the ‘Continuum Approach’ – along the following lines: […] clause linkage types should not be described in terms of the binary opposition between coordination and subordination. Rather, they should be defined in terms of a set of mutually independent and freely combinable features, which form a more or less articulated continuum. Each clause linkage type may be more or less coordinate-like or subordinate-like depending on the parameter taken into account. (Cristofaro 2003: 22–23)

It is therefore much more reasonable to assume a continuum of clause linkage ranging from prototypical coordination to prototypical subordination, with various intermediate degrees in between. A corpus-based contrastive study of the type presented in this article lends support to the hypothesis that the borderline between the two clause linking strategies is less clear-cut than it first seems. Drawing on various functional-typological theories (e.g., Matthiessen & Thompson 1988; Lehmann 1988; Givón 2001), we posit a scale of clause linking, like the one represented in Figure 6. At one end of the spectrum, we find prototypical cases of coordination (such as (1) in Figure 6), i.e., constructions that are [–dependent] and [–embedded]. The opposite pole of the continuum features prototypical cases of subordination (like (5) and (6) in Figure 6) that are clearly [+dependent] and [+embedded]. Note that (6) is more subordinate-like than (5), as non-finite constructions are traditionally assumed to be both more embedded and more dependent than finite ones (see above). In view of the higher degree of subordinateness of non-finite clauses, it is rather surprising to see the relatively large proportion of cases where there is correspondence between a non-finite clause in one language (e.g., a gérondif in French or an –ing clause in English) and a coordinating construction in the other language (Section 4.2.2.1). However, it is less of a surprise if we consider the typical semantic indeterminacy of non-finite adverbial clauses. Like coordinating constructions, non-finite adverbial clauses require more interpretation on

Clause linking in English, French and Dutch  [-dependent] [-embedded]

[+dependent] [+embedded]

PROTOTYPICAL COORDINATION

PROTOTYPICAL SUBORDINATION

coordination of VP-conjunction hypotaxis embedding full clauses (1) (2) finite non-finite finite non-finite (3)

(4)

(5)

(6)

(1) The winter had come at last, and snow lay thick on the ground. (2) Peter ate the fruit and drank the beer. (3) Before I could sit down, she offered me a cup of tea. (4) Driving home after work, I accidentally went through a red light. (5) I noticed that he spoke English with an Australian accent. (6) He likes everyone to relax. (Quirk et al. 1985) Figure 6.  Clause linking as a continuum.

the part of the reader than is the case with finite adverbial clauses, which traditionally involve more explicit linkage. This semantic likeness with coordination, incidentally, earned English –ing adverbial clauses the name ‘ing-coordination’ (Johansson & Lysvåg 1986: 135). VP-conjunction (see example (2) in Figure 6) also falls between the two prototypical poles. Though traditionally accounted for as coordination, it has been shown to exhibit a higher degree of dependency than truly prototypical coordination, i.e., coordination of full clauses (e.g., Cristofaro 2003: 23). It thus comes as no surprise that VP-conjunction is the type of coordinating construction most frequently involved in the shifts investigated. Similarly, hypotaxis (see (3) and (4) in Figure 6) has to be distinguished from embedding on the grounds that such constructions are less embedded and less dependent (hence, less subordinate-like) than truly embedded constructions like (5) and (6) (e.g., Matthiessen & Thompson 1988; Givón 2001). The scale of clause linking posited in Figure 6 turns out to be highly relevant to the investigation of translation shifts. Such shifts indeed prove to occur most commonly with constructions located in the middle of the continuum (from VP-conjunction to finite or non-finite hypotaxis, or vice versa).

6.  Conclusion The primary concern of the article has been to show that languages – even closely-related ones like English, French and Dutch – exhibit preferred clause linking patterns and that these preferences have a clear impact on translation. A corpus-based cross-linguistic

 Christelle Cosme

i­ nvestigation of clause linking proves to confirm the Dependency Hypothesis to a large extent. Though the comparable corpus analysis shows no striking difference between English and French as to their use of subordination, the translation corpus analysis sheds light on a major difference. In line with the predictions, English ⇒ French translations are characterized by a large proportion of shifts from coordination to subordination, whereas reverse shifts are found to be predominant in the other translation direction. As for the contrast between English and Dutch, the translation corpus analysis has enabled us to move one step further from the comparable corpus analysis. English and Dutch do not appear – on the basis of the comparable corpus data – to differ in the frequency with which they resort to coordination. The translation corpus analysis suggests, however, that Dutch has a stronger preference for coordination than English, as English ⇒ Dutch translations typically involve shifts from subordination to coordination. This finding thus confirms the greater incrementality of Dutch, as compared to English. The corpus-based investigation has not only confirmed the predictions, but it has also permitted further refining of the hypothesis. Close examination of the translation shifts reveals a limited set of correspondences. Thus, the shifts involved in English ⇒ Dutch translations have been shown to be predominantly of one type, namely renderings of English –ing clauses by means of coordination in Dutch. Likewise, careful analysis of the shifts involved in English ⇒ French and French ⇒ English translations shows that coordination corresponds to no more than four types of sub-clauses, namely relative clauses, participle clauses, finite adverbial clauses and – to a slightly lesser extent – constructions with ‘pour’ + infinitive. This is in accordance with claims, among others by Beaman (1984), that subordination is not a unitary construct. Therefore, merely stating that one language uses more subordination than another language is somewhat too simplistic. Further investigation ought to include a combined formalfunctional analysis of the different types of sub-clauses that are preferred in one language over another. In addition to providing cross-linguistic descriptions of the preferred clause linking patterns of the three languages, the article has also achieved a more theoretical objective. Clause linking, it has been shown, is best considered as a continuum (rather than as a strict binary system). Some constructions (e.g., VP-conjunction, non-finite clauses) have been said to bear similarities to both coordination and subordination. It has been suggested that this theory, known as the ‘Continuum Approach’, forms an ideal framework for the investigation of translation shifts.

References Asher, Nicholas & Vieu, Laure. 2005. Subordinating and coordinating discourse relations. Lingua 115: 591–610. Ballard, Michel. 1995. La traduction de la conjonction ‘and’ en français. In Relations discursives et traduction, Michel Ballard (ed.), 221–293. Lille: Presses Universitaires de Lille.

Clause linking in English, French and Dutch  Beaman, Karen. 1984. Coordination and subordination revisited. Syntactic complexity in spoken and written narrative discourse. In Coherence in Spoken and Written Discourse, Deborah Tannen (ed.), 45–80. Norwood NJ: Ablex. Chafe, Wallace L. 1982. Integration and involvement in speaking, writing, and oral literature. In Spoken and Written Language: Exploring Orality and Literacy, Deborah Tannen (ed.), 35–53. Norwood NJ: Ablex. Chuquet, Hélène & Paillard, Michel. (eds). 1987. Approche linguistique des problèmes de traduction anglais >< français. Paris: Gap / Ophrys. Cosme, Christelle. 2003. Clause combining in English, French and Dutch. State of the Art and Pilot Study: Towards Annotating Sentences and Clauses. Unpublished MA thesis, University of Louvain. Cosme, Christelle. 2004. Towards a corpus-based cross-linguistic study of clause combining: Methodological framework and preliminary results. Belgian Journal of English Language and Literatures (BELL), New Series 2: 199–224. Cosme, Christelle. 2006a. Clause combining across languages: A corpus-based study of EnglishFrench translation shifts. Languages in Contrast 6(1): 71–108. Cosme, Christelle. 2006b. “Spelling things out” in translations: A corpus-based study of ­explicitation. Paper presented at the SLT-Congress (The Study of Language and Translation), Ghent (Belgium), 12–14 January 2006. Cristofaro, Sonia. 2003. Subordination [Oxford Studies in Typology and Linguistic Theory]. ­Oxford: Oxford University Press. Doherty, Monika. 1998. Clauses or phrases – a principled account of when-clauses in translations between English and German. In Corpora and Cross-linguistic Research: Theory, Method and Case Studies [Language and Computers – Studies in Practical Linguistics 24], Stig Johansson & Signe Oksefjell (eds), 235–254. Amsterdam/Atlanta: Rodopi. Doherty, Monika. 1999. Position and explicitness: Language specific conditions for the use of adverbial clauses in translations between German and English. In Sprachspezifische Aspekte der Informationsverteilung [Studia Grammatica 47], Monika Doherty (ed.), 112–148. Berlin: Akademie Verlag. Doherty, Monika. 2005. Language-specific conditions for discourse linking and appropriateness. Linguistics 43(1): 1–15. Fabricius-Hansen, Cathrine. 1996. Informational density: A problem for translation and translation theory. Information Structure: A Key Concept for Translation Theory, Special issue of Linguistics 34(3): 521–565. Fabricius-Hansen, Cathrine. 1998. Informational density and translation, with special Reference to German-Norwegian-English. In Corpora and Cross-linguistic Research: Theory, Method and Case Studies [Language and Computers – Studies in Practical Linguistics 24], Stig Johansson & Signe Oksefjell (eds), 197–234. Amsterdam/Atlanta: Rodopi. Fabricius-Hansen, Cathrine. 1999. Information packaging and translation: Aspects of translational sentence splitting (German – English/Norwegian). In Sprachspezifische Aspekte der Informationsverteilung [Studia Grammatica 47], Monika Doherty (ed.), 175–214. Berlin: Akademie Verlag. Gilquin, Gaëtanelle. 2002. Automatic retrieval of syntactic structures: The quest for the Holy Grail. International Journal of Corpus Linguistics 7(2): 183–214. Givón, Talmy. (ed.). 2001. Syntax. An Introduction, Vol. II. Amsterdam : John Benjamins. Granger, Sylviane. 2003. The corpus approach: A common way forward for contrastive ­Linguistics and Translation Studies. In Corpus-based Approaches to Contrastive Linguistics

 Christelle Cosme and Translation Studies [Approaches to Translation Studies 20], Sylviane Granger, Jacques Lerot & Stephanie Petch-Tyson (eds), 17–30. Amsterdam/Atlanta: Rodopi. Halliday, Michael A.K. & Matthiessen, Christian M.I.M. (eds). 2004. An Introduction to Functional Grammar, Third Edition. London: Edward Arnold. Hannay, Mike. 1997. Sentencing in Dutch and English. In Studies in English Language and teaching, Jan Aarts, Inge de Mönnink & Herman Wekker (eds), 231–256. Amsterdam: Rodopi. Hannay, Mike & Mackenzie, J. Lachlan. 1996. Effective Writing in English: A resource guide. Groningen: Martinus Nijhoff. Hasselgård, Hilde, Johansson, Stig, Behrens, Bergljot & Fabricius-Hansen, Cathrine (eds). 2002. Information Structure in a Cross-linguistic Perspective [Language and Computers – Studies in Practical Linguistics 39]. Amsterdam/New York: Rodopi. Hoarau, Lucie. 1997. Etude contrastive de la coordination en français et en anglais. Paris: Ophrys. Johansson, Stig & Hasselgård, Hilde. 1999. Corpora and cross-linguistic research in the Nordic countries. In Contrastive linguistics and translation, Sylviane Granger, Ludo Beheydt & JeanPierre Colson (eds), 145–162. Special issue of Le Langage et l’Homme 34(1). Leuven: Peeters. Johansson, Stig & Lysvåg, Per. 1986. Understanding English Grammar, Vol. 1. Oslo: Norwegian University Press (UniversitetsForlaget AS). Lehmann, Christian. 1988. Towards a typology of clause linkage. In Clause combining in grammar and discourse [Typological Studies in Language 18], John Haiman & Sandra A. Thompson (eds), 181–225. Amsterdam/Philadelphia: John Benjamins. Mann, William C. & Thompson, Sandra A. 1988. Rhetorical structure theory: Towards a functional theory of text organization. Text 8(3): 243–281. Martin, James R. 1992. English Text: System and Structure. Amsterdam/Philadelphia: John Benjamins. Matthiessen, Christian & Thompson, Sandra A. 1988. The structure of discourse and subordination. In Clause combining in grammar and discourse, John Haiman & Sandra A. Thompson (eds), 275–329. Amsterdam: John Benjamins. Moder, Carol Lynn & Martinovic-Zic, Aida (eds). 2004. Discourse across languages and cultures [Studies in Language Companion Series 68]. Amsterdam/Philadelphia: John Benjamins. Quirk, Randolph, Greenbaum, Sidney, Leech, Geoffrey & Svartvik, Jan. 1985. A Comprehensive grammar of the English Language. London: Longman. Solfjeld, Kåre. 1996. Sententiality and translation strategies German-Norwegian. Information Structure: A Key Concept for Translation Theory, Special issue of Linguistics 34(3): 567–590. Thompson, Sandra A. 1987. Subordination and narrative event structure. In coherence and grounding in discourse, Russell S. Tomlin (ed.), 435–454. Amsterdam: John Benjamins. Vanderauwera, Ria. 1985. Dutch Novels Translated into English: The transformation of a “minority” literature [Approaches to translation Studies 6]. Amsterdam: Rodopi. Vinay, Jean-Paul & Darbelnet, Jean. 1995. Comparative stylistics of French and English: A Methodology for translation [Benjamins Translation Library 11. Amsterdam/Philadelphia: ‘John Benjamins’ translated and edited by Juan C. Sager and Marie-Jo Hamel.

Sentence splitting – and strategies to preserve discourse structure in German-Norwegian translations Kåre Solfjeld

Østfold University College Adjuncts at NP- and VP-level in German texts are often rendered as separate sentences or clauses in coordinating structures in Norwegian translations. Typically, syntactic subordination in the source version signals that the information should not be interpreted as the main contribution of the sentence. Hence, an important means of guiding the readers to a discourse structure mirroring the original is lost in the target version. This study tries to unveil the strategies used by translators to convey the discourse function reflected in the original when syntactically subordinate source structures are translated by paratactic target structures. In particular, the study focuses on the function of coordination in the target version. Keywords: sentence splitting, translation strategies, discourse structure, subordination in SL vs. coordination in TL

1.  Introduction Empirical studies show that translation from German non-fictional prose into Norwegian often triggers sentence splitting. The syntactic complexity of the German source sentences cannot be felicitously transferred to the Norwegian target version by ­analogous structures, and so adjuncts at NP- and VP-level in the German version often give rise to separate sentences or coordinate clause conjuncts in Norwegian (Fabricius-Hansen 1999, Solfjeld 2000, 2004, Ramm 2005). An illustration of this is (1), where the participle forms of the German original are rendered in a separate sentence in Norwegian: (1) … ging der Amtsrichter gemeinsam mit dem diensthabenden Polizeiarzt in die Wohnung der Frau. Dort trafen sie schon zwei Polizisten, die vom verstörten, ohne Unterlaß schluchzenden Dienstmädchen herbeigerufen worden waren. (dt4, 8–9) ‘… the court judge together with the police doctor on duty went into the woman’s flat. There they already met two police officers who had been called by the upset, [lit.:] without stop crying maid.’

 Kåre Solfjeld

… dro rettsdommeren sammen med vakthavende politilege avsted til fruens leilighet. Der traff de to politifolk som allerede var blitt tilkalt av tjenestepiken. Hun var aldeles forstyrret og gråt uten stans. (nt4, 8) ‘… the court judge together with the police doctor on duty went to the woman’s flat. There they met two police officers who had already been called by the maid. She was extremely upset and sobbed without stop.’

By this process syntactically integrated structures in the source versions have ­syntactically independent informational counterparts in the target version. Syntactic subordination often signals some sort of informational ‘downgrading’ in the sense that the information should not be read as part of the main story line (von Stutterheim 1997: 169, 245). When syntactically subordinated source structures are rendered as separate sentences or conjuncts in the target version, a corresponding informational downgrading may be hard to recover, as syntactic signals to that effect are lacking. A central challenge for the translator seems to be to prevent interpretations where the Norwegian sentence/conjunct corresponding to a syntactically embedded structure in German is falsely understood as continuing the main line of thoughts or events, while it is only meant to give additional or supplementary information of some kind. The main purpose of this study is to look into the strategies Norwegian translators use to prevent the information from being upgraded in a way not intended in the German original. The study should be regarded as a follow-up of the work laid out in Solfjeld (2008), presenting largely the same main claim that the discourse role of the German adjuncts determines target structures. The present study is, however, based on a wider set of data: 13 excerpts of German original texts and corresponding ­authorised Norwegian translations. The material contains a few fictional texts, but most texts are non-fictional, many of them biographies. Each translator is represented only once. The present article is the result of cooperative work within the project ­Languages in ­Contrast (SPRIK), University of Oslo, where (among other topics) the relation between syntactic and discourse functional subordination across languages has been a central topic of research; cf. e.g., Behrens and Fabricius-Hansen (2005), ­Fabricius-Hansen et al. (2005), Ramm and Fabricius-Hansen (2005), Fabricius-Hansen (1996, 2005), Ramm (2005 and in this volume) and Solfjeld (2004).

2.  Main claims and data A main challenge for the translator is to enable the reader to infer a discourse ­interpretation that reflects the original. So it must be avoided that the downgraded ­information of the original is wrongly seen as part of the Hauptstruktur (‘main ­structure’) of the text; that is, it should keep its role as part of the Nebenstruktur (‘side structure’) – as adding supplementary information of some kind – in the terms of the ‘quaestio’-model of Klein and von Stutterheim (1992) and von Stutterheim (1997). Many of the texts on which this

Strategies to preserve discourse structure 

study is based are narratives, where predicates relating a sequence of ­successive events form the main structural line and where the predicate pushing the chain of events one step further makes up the focal part of each sentence (von Stutterheim 1997: 106). Some adjuncts, often state predicates, give information that does not directly contribute to characterising the (focal) event of the sentence. They are not part of the main story line of the narrative. Other adjuncts are more directly integrated into the event predicate pushing the event chain of the narrative one step further. In the material on which this study is based, there is a strong correlation between a position relatively far to the left and background role and a position relatively far to the right and focal role. This is reflected in most of the examples discussed below. The different informational contributions of the adjuncts trigger different translation strategies or put different constraints on the target structures (to be) used. In (2) below the background function of the original adjunct, along with the position of its clause counterpart in the ­Norwegian version, results in a target version where the connectives og (‘and’) and derfor (‘therefore’, ‘for that reason’) are added. This seems to give a felicitous Norwegian version. In (1) above og (‘and’)-coordination cannot be used. (2) Anselmo Rodriguez, der mit dem Mann von Frau Balboa, dem Eigentümer der größten Eisenwarenhandlung der Stadt, seit langem bekannt war, glaubte zuerst an einen privaten Besuch. (dt4, 15) ‘Anselmo Rodriguez, who for long had been acquainted with the husband of Mrs. Balboa, the owner of the biggest hardware store in the town, first believed it to be a private visit.’ Anselmo Rodriguez var en gammel bekjent av fru Balboas mann, innehaveren av byens store jernvareforretning, og han trodde derfor først det dreide seg om et privat besøk. (nt4, 15) ‘Anselmo Rodriguez was an old acquaintance of the husband of Mrs. Balboa, the owner of the biggest hardware store in the town, and he believed it therefore first to be a private visit.’

A fruitful point of departure when discussing the typical translational patterns represented in the material, is found in works within Relevance theory (Blakemore 1987, Carston 2002). They help to explain why different strategies are used under different conditions and shed light on the interplay between the function of the original adjunct and the choice of target structure, e.g., the order of the clauses/sentences in the target version and the use of og (‘and’)-coordination vs. separate sentences. The material comprises 274 cases of sentence splitting in the sense that a German NP- or VP-adjunct gives rise to a separate sentence or a conjunct (as part of a coordinative structure) in the Norwegian target text. Adjunct is understood in a broad sense as pre- and post-modifying structures at NP-level, including relative clauses, prepositional and participle phrases and what is often termed ‘appositions’, i.e., modifying phrases of different sorts separated by comma. The adjuncts at NP-level largely seem to be non-restrictive. Adjuncts at VP-level are primarily adverbials, often spatial or

 Kåre Solfjeld

temporal adverbials, realised as prepositional phrases or clauses. Adjuncts at VP-level also include so-called free predicatives. A basic assumption is that connectives are added in the translation process to ­ensure a discourse interpretation reflecting the original. Among the connectives identified in the material og (‘and’)-coordination stands out as unique in terms of ­frequency. 53 adjuncts in the above sense split into coordination with the conjunction og (‘and’); cf. (2) above. This means that 221 adjuncts split into independent sentences, i.e., sentences separated by full stops or sometimes by comma as in (1) above; cf. the survey in Table 1. Table 1.  Sentence splitting with adjuncts Type of sentence splitting

Number of cases

og (‘and’)-coordination   53 Sentences separated by full stop or comma 221 Total 274

On the whole few other connectives in the traditional sense of conjunctions, ­discourse particles or adverbs creating some link with the preceding text are added in the translation process; cf. the survey in Table 2. Table 2.  Connectives added in Norwegian translations Connective added

Number of cases

derfor (‘hence’, ‘so’) da (‘then’) dermed/-ved (‘therewith’) men (‘but’) nemlig (‘namely’) jo (approx. ‘as you know’) så (‘then’) likevel (‘however’) altså (‘that is’)

7 (3 with og-coordination) 6 (2 with og-coordination) 4 (3 with og-coordination) 4 4 3 1 1 1

In most of the cases listed here no explicit lexical counterpart can be identified for the Norwegian connective in the German original. This supports the assumption that they are added to ensure certain discourse structural effects. However, at the same time these figures show that very often connectives in the traditional sense are not added. 198 of the split sentences result in two separate sentences without any ­connective ­being added at all. The analysis presented in section 3 aims to test whether the material supports the following basic claims: – Adjuncts encoding background information have a tendency to be rendered as clauses/sentences to the left of the sentences/clauses corresponding to the focal parts of the original.

Strategies to preserve discourse structure 

– Og (‘and’)-coordination is frequently used to obtain a downgrading function of the clause encoding the same information as the original adjunct; cf. 3.1. – Adjuncts being integrated into the focal parts of the original sentence have a strong tendency to be rendered as clauses/sentences to the right of the sentences/clauses ­corresponding to (other elements of) the focal parts of the original. They have an elaborating or specifying function and generally og (‘and’)-coordination is not used. Occasionally other connective expressions are added, but the normal case seems to be that no connectives are added at all; cf. 3.2.

3.  Analysis 3.1  P  reserving discourse interpretation when the adjunct plays a background role and is extracted to the left One possible function for the adjuncts is to play some sort of background role giving information that enhances the understanding of the sentence part giving the main informational contribution of the sentence. Often these adjuncts are found in the opening part of the German original. In the Norwegian target version the relative position of the information generally mirrors the original. These are cases of information extraction to the left in the terminology of Fabricius-Hansen (1999). The original adjunct gives rise to a sentence or clause preceding the predicate that corresponds (most closely) to the main predicate of the original. In 81 out of the 95 cases of sentence splittings that are extractions to the left, the source adjunct is found in the so-called prefield of the original; cf. (2) above and also sentence pair (3): (3) Auroras ältester Bruder, der als einziger der Familie noch in El Ferrol ansässig war, schlug am Tag nach Bekanntwerden dieser Aufforderung seine Schwester ins Gesicht. Er kam betrunken aus dem Königlichen Segelklub, wo seine Freunde die Neuigkeit kommentierten und mit ihrem Spott auch ihn nicht verschonten. (dt4, 39) ‘Aurora’s elder brother, who was the only one in the family who still lived in El Ferrol, hit on the day after the announcement of this request his sister in the face. He came drunk from the Royal Sailing Club where his friends had commented on the news and ridiculed him as well.’ Auroras bror var den eneste i familien som ennå bodde i El Ferrol. Dagen etter at oppfordringen var annonsert, slo han søsteren i ansiktet. Han kom full fra den Kongelige Seilforeningen, hvor vennene hadde kommentert nyheten og gjort narr av både ham og henne. (nt4, 36) ‘Aurora’s elder brother was the only one in the family who still lived in El Ferrol. On the day after the request was announced, he hit his sister in the face. He came drunk from the Royal Sailing Club where his friends had commented on the news and ridiculed both him and her.’

 Kåre Solfjeld

Since it can be assumed that the main predicate very often carries the essential information of the sentence, the main challenge for the translators will be to rephrase the target version in a way that keeps this focus intact, in the sense that the first sentence/conjunct must be recognisable as giving background information to the second clause/conjunct, which makes up the focal element(s). In narrative texts the first sentence/clause must be identified as supplying background information for the next step in the chain of events, which is encoded in the second sentence/ clause; cf. (3) above and (4). Semantic constraints seem to play a role in these cases. The fact that the extracted information is conveyed by state predicates ensures the background reading (more about this later). When information is extracted in the adjunct to the left, coordination is a frequent target structure, as illustrated in (2) above and also (4): (4) Micha, wie Charly seit dem heißen Dezember dabei, erklärt die Unterschiede zu den “Instandbesetzern, die meistens viel mehr Häuser haben als wir, mindestens hundert”, … (dt5, 103) ‘Micha, like Charly there since the hot December, explains the differences from the “Instandbesetzern, who normally have many more houses than we, at least a hundred”, …’ I likhet med Charly har Micha vært med siden desember 1980, og han forklarer oss forskjellen mellom oss selv og “Istandbesetterne […], som har mange flere hus enn vi, minst hundre”: … (nt5, 68) ‘Like Charly Micha has been there since December 1980, and he explains us the difference between themselves and “Istandbesetterne, who have many more houses than we, at least a hundred”, …’

Out of the 95 cases of sentence splittings that are extractions to the left, 43 are rendered as coordinate constructions, and in the remaining cases coordination is very often a possible alternative. Relevance theory (Blakemore 1987; Carston 2002) gives interesting clues as to why translators often use coordination when splitting out information to the left. A coordinative structure is processed as a unit, signalling that it is the conjoined proposition that carries optimal relevance, and not each individual conjunct. Lang (1984) presents the same basic ideas, assuming that the conjunction has a procedural meaning that results in the subsumption of the conjuncts under a “common integrator”. Processing a coordinative structure as a unit precludes the possibility that one of the conjuncts may relate to the context separately (Carston 2002: 243), and so no inferences with regard to discourse structure exclusively based on the relation between the first constituent and the preceding text can be made. This may in some cases be important, as a discourse relation established between the proposition of the first constituent individually and the preceding text may affect how the next constituent is interpreted in relation to the discourse interpretation built up so far.

Strategies to preserve discourse structure 

At the same time, processing conjuncts in one step licenses certain pragmatic inferences to the effect that there is a (con)sequential relation between what is conveyed in the propositions of the two conjuncts. Often a temporal sequence can be inferred and/or a weak causal or enablement relation between the first and the second conjunct, at least in a narrative text; cf. the concept of consequentiality in Sandström (1993: 141). This mechanism is obviously exploited by translators. Since the first conjunct of the coordinative structure is interpreted as leading up to the second, the first conjunct is prevented from assuming a position in the discourse hierarchy on an equal footing with the second conjunct. In this way coordination guides readers more effectively to interpretations where the second conjunct is seen as informational focus parallel to the main predicate of the original. Text pair (5) illustrates this observation: (5) a. Aurora hatte wenig Lust, das Elternhaus zu verlassen. Aber es blieb ihr nicht anderes übrig, als ein Zimmer im Hause des Arztes zu beziehen; undenkbar, dass sie sich der Anordnung widersetzte. Außerdem freute sich Ochoas Frau, deren Ehe kinderlos geblieben war, auf weibliche Gesellschaft. Aurora Rodríguez bestand aber darauf, täglich die Bibliothek ihres Vaters aufzusuchen. (dt4, 40) ‘Aurora did not feel much like leaving her parents’ home. But she had no choice but to move into a room in the house of the doctor; impossible that she should resist the arrangement. Furthermore Ochoa’s wife, whose marriage was childless, was happy about female company. Aurora Rodriguez, however, insisted on visiting the library of her father daily.’ Aurora hadde liten lyst til å forlate barndomshjemmet. Men hun hadde ikke annet valg enn å flytte inn i et rom i legens hus, det var utenkelig at hun skulle motsette seg dette. Ochoas ekteskap var barnløst, og hans kone gledet seg over å få kvinnelig selskap. Men Aurora Rodriguez insisterte på i det minste å få besøke farens bibliotek hver dag. (nt4, 37) ‘Aurora did not want to leave her childhood home. But she had no other choice than to move into a room in the house of the doctor; it was impossible that she should resist the arrangement. Ochoa’s marriage was childless, and his wife was happy about female company. Aurora Rodriguez, however, insisted on at least visiting the library of her father daily.’ b. Aurora hadde liten lyst til å forlate barndomshjemmet Men hun hadde ikke annet valg enn å flytte inn i et rom i legens hus, det var utenkelig at hun skulle motsette seg dette. Ochoas ekteskap var barnløst. Hans kone gledet seg over å få kvinnelig selskap. Men Aurora Rodriguez insisterte på i det minste å få besøke farens bibliotek hver dag. ‘Aurora did not want to leave her childhood home. But she had no other choice but to move into a room in the house of the doctor; it was impossible that she should resist the arrangement. Ochoa’s marriage was childless. His wife was happy about female company. Aurora Rodriguez, however, insisted on at least visiting the library of her father daily.’

 Kåre Solfjeld

By choosing a coordinating structure the translator rules out misinterpretations or at least garden-path-effects that might arise from the alternative sentence sequence; cf. my paraphrase in (5b). Hence, coordination can be seen to have discourse structural effects which guide the reader to the right interpretation. The full stop alternative seems to be more open. Since a separate sentence may relate to the context independently, the information that Ochoa’s marriage was childless can – in a first processing step – be read as explaining why it was impossible for Aurora to protest against the arrangement of moving into the doctor’s home. In the next processing step the following, separate sentence can be interpreted as discourse functionally parallel to the preceding one. The information that Ochoa’s wife was happy about female companionship can be read as adding another reason why she could not contest the arrangement. Hence, the full stop alternative can easily create a kind of list reading, and, in turn, blur the role of the information that she was happy about gaining a female companion as the main focus. By choosing a coordinating structure the translator prevents the possibility that the first conjunct may relate to the context separately. There is no processing step which allows Ochoa’s childless marriage to be seen as an independent reason why Aurora could not protest against the arrangement. Coordination ensures that the two conjuncts are processed as a single unit, and as part of the process of linking the conjuncts the readers are licensed to infer a consequential relation between them (Blakemore 1987: 117, Carston 2002: 243). Hence, a coordinating structure guides the readers more effectively to an interpretation where the information that Ochoa’s marriage was childless is read as background for the fact that she was happy about female companionship. In this way the downgraded role of the information given in the syntactically subordinate structure of the original is kept intact. It is the constituent corresponding to the main predicate which adds to the argument that she had no other choice but to move into the doctor’s home. The potential for alternative discourse structural interpretations, not matching the original, obviously has to do with the propositional content of the constituents. Predicates split off to the left in the way illustrated by (2), (3), (4) and (5), are often state predicates supplying some background knowledge which makes the chain of events easier to understand or more plausible. This is a typical pattern, at least in the material on which this study is based: A state/activity predicate – not pushing the chain of events further – functions as background for the predicate which it precedes. Very often sentence splittings with information extraction to the left are rendered as sentence sequences separated by full stops, and the backgrounding function of the state predicate is easily derivable, as in (4) (above). The backgrounding function of the information that Charly has been there since December, is so clear that coordination is only optional; cf. the alternative (4b):

(4) b. I likhet med Charly har Micha vært med siden desember 1980. Han forklarer oss forskjellen mellom oss selv og “Istandbesetterne […], som har mange flere hus enn vi, minst hundre”: …

Strategies to preserve discourse structure 

‘Like Charly Micha has been there since December 1980. He explains us the difference between themselves and “Istandbesetterne, who have many more houses than we, at least a hundred” …’

The recurring translation strategy of extracting the information of adjuncts to the left and rephrasing it as state predicates seems to fit with the observation in Sandström (1993: 168) that state predicates encoding facts that are directly relevant to the interpretation of events are evaluated at a point on the time line (so-called E-state) prior to the incorporation of the event for which it is relevant. It also seems to comply with the observation made in Ramm and Fabricius-Hansen (2005) on the examples of the discourse relation Background in the so-called RST-framework: The so-called satellite constituent giving information with the function of increasing the reader’s ability to comprehend what is conveyed in the nucleus generally precedes this nucleus. The material used in the present study contains many cases where the backgrounding function of a state predicate extracted to the left is self explanatory, whereas the same state predicate in a position to the right of the predicate which corresponds to the main predicate of the original (information extraction to the right, Fabricius-Hansen 1999) obviously blurs the background function. This is illustrated in sentence pair (6) below. In the Norwegian translated version the function of the state predicate is to convey qualities of the woman which clearly lead up to what comes next, whereas this is less clear if the order is reversed. The reversed order would rather seem to induce an interpretation where the two sentences should be seen as a more parallel listing of the qualities of the woman. Consider (6b) below. (6) a. Der Traum zeigt deutlich, in welcher Zwangslage sie sich befindet. Ohne ihre befriedigenden Berufserlebnisse fühlt sich die von Natur aus dynamische, ehrgeizige Frau einsam und isoliert. (dt11, 19) ‘The dream clearly shows what a dilemma she is in. Without her satisfying professional experiences the by nature dynamic, ambitious woman feels lonely and isolated.’ Drømmen viser hvilken tvangssituasjon hun befinner seg i. Av natur er hun en dynamisk, ærgjerrig kvinne. Uten tilfredsstillende yrkesopplevelser føler hun seg ensom og isolert. (nt11, 22) ‘The dream clearly shows what a dilemma she is in. By nature she is a dynamic, ambitious woman. Without her satisfying professional experiences she feels lonely and isolated.’

b. Drømmen viser hvilken tvangssituasjon hun befinner seg i. Uten tilfredsstillende yrkesopplevelser føler hun seg ensom og isolert. Av natur er hun en dynamisk, ærgjerrig kvinne. ‘The dream clearly shows what a dilemma she is in. Without her satisfying professional experiences she feels lonely and isolated. By nature she is a dynamic, ambitious woman.’

 Kåre Solfjeld

3.2  P  reserving discourse interpretation when the adjunct is part of the sentence focus and is extracted to the right Our material suggests that strategies to preserve discourse structures are different when the adjuncts giving rise to sentence splitting are integrated into the focus part of the source sentence, i.e., integrated into the sentence part that gives the main informational contribution of the sentence. In narrative texts, for instance, where the predicate introducing the next step in the event sequence makes up the focused part of the sentence, argument expressions, which refer to individuals and objects participating in the event, will be syntactically dependent on the verb – and so be integral parts of the event predicates. Hence, adjuncts giving information about the individuals or objects participating in the event situations can often also be seen as part of the focus of the sentence (von Stutterheim 1997: 113). Our data show that adjuncts relatively far to the right in the source sentence are often integrated elements of the focused part of the original sentence. In the splitting process these adjuncts typically give rise to separate sentences to the right of the target text predicate which corresponds most closely to the main predicate of the source sentence. Hence, these are examples of what Fabricius-Hansen (1999) has termed information extraction to the right. In the material used for the present study there are altogether 179 cases of information extraction to the right. In 153 of these the source adjunct is found in a position to the right of the main predicate of the original sentences, i.e., in the position to the right of the non-finite verb(s) in a predicate consisting of more verb forms, or to the right of a finite verb when this verb is the only verb form. It should be added that in the latter case more refined criteria might have distinguished between source adjuncts in the so-called middle field and source adjuncts in so-called extraposition. It should also be noted that in complex sentences the adjunct giving rise to sentences or clauses in the Norwegian version may be placed to the left of the predicate of a subclause, as was the case in (1) above. The position of the source adjuncts relative to the main predicate – i.e., in complex sentences exclusively relative to the highest ranking matrix predicate – may give a rough idea. Still, it shows that sentence splitting with source structures in end or near end position of the source sentence (complex) mirrors the sentence splitting in initial position. The information encoded in an adjunct to the right of the main predicate is found in a position to the right of the corresponding predicate in the target version. Hence, the relative position of the information elements are largely kept intact, which strengthens the picture of a linear mapping of the information structure of the monosentential version onto the multisentence or multi-clause version. The target structures are almost exclusively sequences of separate sentences with a full stop between them, but occasionally also separated by comma. When information is extracted from the focal parts of the source sentence the adjuncts and their corresponding target sentences seem to have an elaborating function. The source adjuncts specify or give further information about individuals or objects, and often this information encodes essential features of the event situation. In this way

Strategies to preserve discourse structure 

the adjuncts give important contributions to the focal part(s) of the source sentence. In the target text these adjuncts are rendered as sentences elaborating (in a wide sense) on some individual or object referent. Syntactically, coordination is a rare choice in these cases, but not totally absent: 10 cases of information extraction to the right are rendered as coordinative structures in the Norwegian version. Elaboration seems to block the possibility of attaching the elaborating constituent by coordination, as it gives the answer to a question implicitly posed in the preceding sentence on which it elaborates, and for that reason must relate to the context separately (Carston 2002: 246ff). So, it can be expected that elaborating sentence splittings to the right are normally incompatible with coordinative structures (more about this below). In fact, when information is extracted to the right, not very many other connectives are built into the Norwegian target versions either, although connectives like nemlig (‘in other words’), jo (‘as said’, ‘as you know’), da (‘then’) and altså (‘that is’) occur in a few cases. The strategy used to preserve the discourse structure of the original in these cases is simply to present the elaborating information in a separate sentence directly after the individual or object referent on which it elaborates, ensuring the correct reference relation through anaphoric links. As with the splitting off of background information to the left, the crucial point seems to be to convey efficiently the informationally subordinate role of the elaborating sentences and signal efficiently when the main story line is resumed. In a narrative text it must be clear which sentence provides a new contribution to the chain of successive events. These points are illustrated in text pairs (7) and (8) below. (7) Bei Professor Dements Traumentzugsexperimenten reagierten die Versuchspersonen auf den Traumsentzug mehrerer Nächte mit Reizbarkeit, Unentschlossenheit und Feindseligkeit. Nach traumlosen hundert Stunden wurden aus umgänglichen Menschen plötzlich aggressive Typen, die unter Verfolgungswahn und Erinnerungslücken litten. Sobald man die Testträumer ungestört träumen ließ, waren sie wieder ausgeglichen und normal. (dt11, 8) ‘In professor Dement’s dream denial tests the test persons reacted to several nights of dream denial with hypersensitivity, apathy and hostility. After hundred dreamless hours friendly people became aggressive types, who suffered from paranoia and memory losses. As soon as the test dreamers were allowed to dream undisturbed, they were in balance and normal again.’ Professor Dement hindret sine forsøkspersoner i å drømme. Etter noen netter reagerte de med irritabilitet, ubesluttsomhet og aggresjoner. Etter noen hundre timer ble ellers omgjengelige personer plutselig aggressive. De fikk forfølgelsesforestillinger og hull i hukommelsen. Når forsøkspersonene fikk sove uforstyrret igjen, ble de avbalanserte og normale som før. (nt11, 9) ‘Professor Dement prevented his test persons from dreaming. After some nights they reacted with hypersensitivity, apathy and hostility. After some hundred hours normally friendly persons became aggressive. They suffered from paranoia and memory losses. As soon as the test persons were allowed to dream undisturbed, they were in balance and normal again.’

 Kåre Solfjeld

In (7) the reactions of a group of test persons are rendered as a chain of successive events, where the changing behaviour of the persons – encoded in the predicates including adjuncts characterising the persons at the different stages – make up the focused part of the sentences. So, the relative clause die unter Verfolgungswahn und Erinnerungslücken litten (‘who suffered from paranoia and memory losses’) seems to make up an important part of the informational focus of the sentence in which it is integrated. Together with the adjective aggressive (‘aggressive’) the relative clause ­provides the new relevant information at that stage in the development. In the ­Norwegian version the information of the relative clause is rendered in a separate sentence. The anaphoric link de (‘they’) establishes a connection to the referent personer (‘persons’) in the ­preceding sentence, simply adding to the description of the test persons and with no linguistic expressions pointing to any other discourse role. When in the next sentence the chain of events is pushed one step further, the elaborating function of the Norwegian ­independent ­sentence is clearly conveyed. In line with the idea of the so-called right frontier c­ onstraint (Asher 1993; Asher & Vieu 2005) easily identifiable elaboration relations keep the track open in the sense that the superordinate sentence in the discourse hierarchy remains available as an attachment point. Hence, the chain of successive events is equally ­effectively conveyed in the Norwegian version as in the German one, where each step in the chain of events is mirrored by each sentence in a more iconic way. The temporal adverbials in the surrounding sentences seem to ­support the elaboration reading of the Norwegian independent sentence. In (8) the prepositional phrase in verblaßenden Farben (‘in fading colours’) is part of a ­descriptive passage ­embedded in a narrative text part. Without going into further detail, it can be seen that the splitting off to the right of the information of this prepositional phrase causes no interpretational difficulties. The referents malingen, fargene (‘the paint’, ‘the colours’) establish a clear anaphoric link to gudevogn (‘god cart’). So the sentence simply elaborates further on this cart and contains no linguistic expressions pointing to another discourse function. Hence, the Norwegian version mirrors the German original where the same information is kept together in one sentence. It does not disturb the discourse structure of the passage, where the description part is continued in the next sentence and the event sequence resumed in the last sentence. (8) Rückfahrt […]. – Skriperumpudur. Den Namen dieses Kaffs werde ich nie vergessen. Dort, am Straßenrand, verkam in Wind und Wetter ein hölzerner Götterwagen in verblaßenden Farben, er kann in keiner Prozession mehr mitgeführt werden. – Des unsicheren Lichts wegen knipste ich das einst stolze Fahrzeug mit mit drei Kameras verschiedener Objektive, … (dt13, 192) ‘Travelling back […]. – Skriperumpudur. The name of this hole I shall never forget. There, by the side of the road, fell apart through wind and weather a wooden god cart in fading colours, it cannot be used in any procession any more. – Because of the uncertain light I photographed the once proud vehicle with three cameras with different objectives, …’ Så var det på tide å kjøre tilbake […]. – Skriperumpudur. Denne fattige, lille landsbyens navn kommer jeg aldri til å glemme. Ved veisiden sto det en gudevogn

Strategies to preserve discourse structure 

av tre og ble ett opp av vær og vind. Malingen, fargene var falmet eller avskallet; den kan ikke lenger føres med i noen prosesjon. – På grunn av det dårlige lyset i skumringstimen knipset jeg det en gang så stolte kjøretøyet med tre kamera med ulike objektiver, … (nt13, 192) ‘Then it was time to travel back […]. – Skriperumpudur. The name of this small poor village I shall never forget. There, by the side of the road stood a god cart of wood and was eaten by wind and weather. The paint, the colours were fading or peeling; it can no longer be used in a procession. – Because of the weak evening light I photographed the once proud vehicle with three cameras with different objectives …’

Our data suggest that clear anaphoric links contribute to convey the elaborating functions efficiently, and normally no confusion arises with regard to discourse structure. However, the material contains a few sentence pairs where special connective devices are built into the Norwegian version to convey effectively the elaborating status of the independent Norwegian sentence. This reflects that the information encoded in the source adjunct of the original sentence does not provide contributions to the discourse structures other than elaborating further on (elements in) the sentence in which it is integrated. As with the leftextracted backgrounding cases explicit connective expressions seem to be added in those cases where the propositional content of the extracted information may lead to interpretations deviating from the original or to certain garden-path-effects. Consider text pair (9): (9) a. Die Konferenz endete mit einem Kompromiss. Die ursprüngliche SJVDResolution erhielt zwei Zusätze, in denen die Herausbildung einer neuen internationalen Organisation und die Schaffung einer neuen Jugendinternationale zu Zielen erklärt wurden. Die neue Resolution sollte die Grundlage für das “Büro internationaler Jugendorganisationen” bilden, das spätestens sechs Monate nach der “Liller Konferenz” zu seiner ersten Sitzung zusammentreten sollte. (dt8, 152–153) ‘The conference ended with a compromise. The original SJVD-resolution had two supplements, in which the formation of a new international organisation and the establishment of a new youth international were declared as goals. The new resolution would form the basis of the “Bureau of international youth organisations”, which should get together for its first meeting not later than six months after the “Lille-conference”.’ Konferansen sluttet med et kompromiss. Den opprinnelige SJVD-resolusjonen fikk to tillegg. Målene, het det nå, var å bygge opp en ny internasjonal organisasjon og danne en ny ungdomsinternasjonale. Den nye resolusjonen skulle være grunnlaget for “Det internasjonale byrå for revolusjonære ungdomsorganisasjoner”, som skulle ha sitt første møte senest seks måneder etter “Lille-konferansen”. (nt8, 111) ‘The conference ended with a compromise. The original SJVD-resolution had two supplements. The goals, it was now stated, were to form a new international organisation and to establish a new youth international.

 Kåre Solfjeld

The new resolution would form the basis of the CBureau of international youth organisations”, which should get together for its first meeting not later than six months after the “Lille-conference”.’ (9) b. Konferansen sluttet med et kompromiss. Den opprinnelige SJVDresolusjonen fikk to tillegg. Målene var å bygge opp en ny internasjonal organisasjon og danne en ny ungdomsinternasjonale. Den nye resolusjonen skulle være grunnlaget for “Det internasjonale byrå for revolusjonære ungdomsorganisasjoner”, som skulle ha sitt første møte senest seks måneder etter “Lille-konferansen”. ‘The conference ended with a compromise. The original SJVD-resolution had two supplements. The goals were to form a new international organisation and to establish a new youth international. The new resolution would form the basis of the “Bureau of international youth organisations”, which should get together for its first meeting not later than six months after the “Lilleconference”.’

In text pair (9) the second sentence of the German original version includes a relative clause, which is split off as a separate sentence in the Norwegian version. The ­syntactic structure of the original clearly conveys that the relative clause encodes ­information about the content of the referent Zusätze (‘supplements’), which the relative clause ­modifies. In the Norwegian version the phrase het det nå (‘it was now stated’) is added by the translator. This seems to be needed to ensure an interpretation ­mirroring the ­discourse ­structure of the original. Without het det nå (‘it was now stated’) the ­independent Norwegian sentence corresponding to the relative clause might be ­interpreted as ­giving some kind of reason for the compromise, as målene (‘the goals’) in principle may be anaphorically linked with konferansen (‘the conference’) or possibly kompromiss ­(‘compromise’). By adding het det nå (‘it was now stated’) the readers are guided in the direction of some written document or text. In this way tillegg (‘supplements’), as part of resolusjon (‘resolution’), stands out as a natural anchoring for målene (‘the goals’). The general picture is, however, that no connective expressions are added when information is extracted to the right. Our data suggest that the role as integrated parts of the focused elements is hard to combine with e.g., connectives such as nemlig, jo, altså, which seem to presuppose a more independent backgrounding or commenting function of the information which they modify. Hence, they often seem incompatible with the ‘pure’ function of elaborating on some focused constituent(s), which is what translators, for discourse structural reasons, aim for in most cases of information extraction to the right. Our material includes cases where e.g., nemlig, jo, altså are added in the translation process. For such cases see the discussion in Solfjeld (2008).

4.  Summary and discussion The material confirms that preserving discourse structure is relevant for the choice of target structure when adjuncts at NP- and VP-level in German originals are split

Strategies to preserve discourse structure 

off in the translation process and in this way give rise to independent sentences or conjuncts in the Norwegian target versions. If the extracted information has some kind of ­backgrounding function, it is often moved to the left, preceding the information which provides the contribution to the main line of events or thoughts of the text (part). Often the ordering of the components seems to be enough to ensure an ­interpretation ­equalling the downgraded status of the original adjunct, particularly if the extracted information is a state predicate and can be immediately ­recognised as such. In cases where this is less clear, the backgrounding information is often rendered as the first conjunct of a coordinative structure to prevent discourse functional ­mismatches ­compared with the original. The (con)sequential inferences licensed by the coordination ensure a downgrading effect of the first conjunct and a corresponding upgrading of the second conjunct, which for that reason is ­preserved as focal. Hence, to integrate adjunctive information with a backgrounding function translators ­frequently use the connective og (‘and’). Other connectives are rare, but do occasionally occur – for ­instance derfor (‘so’, ‘hence’) and dermed (‘therewith’). Sometimes they are used in combination with og as illustrated in (2) above. These connectives seem to have much the same downgrading effect on the preceding clause as coordination. When the adjuncts are integrated into the focal part(s) of the original, the strategy is to keep the target sentence corresponding to the original adjunct in a position where clear anaphoric links can be established. Normally, no connectives are added. The position of the extracted information, including the anaphoric links, ensures the intended interpretation, and the discourse function of the incoming sentences seems equally clear in the target version as in the original. Sometimes, however, connective expressions are added, and they seem to be added in those cases where they are needed to prevent the extracted information from being falsely interpreted as a contribution to the main story line. The material shows that translators are generally sensitive to where connectives should be added to prevent unintended discourse functional interpretations. However, in some cases the Norwegian versions are less felicitous or at least more open or confusing with regard to discourse structure than their German counterparts; cf. Ramm (in this volume). The material clearly shows that lexical items of different kinds may be added to obtain a discourse functional effect; cf. (9) above. Interestingly, most cases of sentence splitting seem to work well without the adding of connective expressions for which there are no explicit counterparts in the original. Evidently, the order in which the ­constituents appear (cf. (6)), the use of different sorts of anaphoric expressions and characteristics of the predicates as e.g., states or processes are all elements which ­contribute to convey the discourse structures. These are aspects that have only been briefly touched upon in the present study. The translation patterns in this study ­illustrate, however, that rhetorical structure makes up an important part of the linguistic competence of the translators.

 Kåre Solfjeld

Interestingly, our data suggest a rather linear mapping of the sentence internal i­ nformation structure of the original onto a sequence of sentences or clauses in the target version. Many of our original sentences reflect the traditional claim that sentence openings are the preferred position for background information linking the sentence with the preceding context, whereas components further to the right carry the new, focused information (Dietrich 1994; Nølke 1995). In this way the multisentence or multiclause target versions encompass sequences where some sentences/conjuncts are split off from a front position of the original. These sentences/conjuncts point to the right and give background information for what comes next. Other sentences are split off from focal parts of the original. These point to the left and elaborate on (components of) the preceding sentence. The fact that many such sequences seem to be felicitous without assisting connectives – the translators feel no need to add them – suggests that they are often rather transparent. As the processing runs form left to right, a predicate without any immediate relevance or connection to what has been said, is automatically interpreted as building up to what comes next. In particular state predicates in narratives seem to come with this expectance of looking right. This fits well with the observation that the ties to the left must be clearly stated when elaborating information is split off to the right. Translators do, however, – as seen above – add connective devices to help interpreting such sequences, when misinterpretations might occur. The only frequently occurring connective built into the translation to preserve discourse structure is og (‘and’). In this way, coordination seems to compensate for the more restrictive use of syntactic subordination in Norwegian as compared to ­German. Coordination in Norwegian may also seem to be less restricted as to the ­relation holding between the conjuncts than corresponding und-conjuncts in ­German and and-conjuncts in English (Ramm & Fabricius-Hansen 2005 and ­Fabricius-Hansen et al. 2005). Interestingly, our data include some cases where ­information split off to the right is rendered as a conjunct of a coordinative structure. In (10) the participle form nachsichtig lächelnd (‘smiling indulgently’) is an event which accompanies the event that Tom appears again, and together these event predicates make up the informational focus of the German original. This seems to be typical of the information splittings to the right where the extracted information is part of a coordinative structure. The information extracted to the right is an integrated part of the main informational contribution of the sentence, licensing that it is included in the second conjunct of a coordinative structure, which often carries the focus of a coordinative structure. As the first conjunct of a coordinative structure often seems to convey background information, this translational strategy may mean that some element of a syntactically complex focus in the German original is informationally downgraded to a function of leading up to the second conjunct. This may for instance be the case in (10), where the information that Tom appeared again, can be seen as more of a background information in the Norwegian version than in the German original. (10) Jetzt traf der schwere Ball gegen Johns Kniekehlen. Er fiel um wie eine zu steil gestellte Leiter, erst langsam und dann mit Wucht. Von der Hüfte und vom

Strategies to preserve discourse structure 

Ellenbogen her breitete sich Schmerz aus. Tom stand wieder da, nachsichtig lächelnd. Halblaut sagte er, ohne den Blick von John abzuwenden, etwas zu den anderen, … (dt9, 14) ‘Now the heavy ball hit John’s knee. He fell over like a too steeply set up ladder, first slowly and then with force. From the hip to the elbow the pain spread out. Tom appeared again, smiling indulgently. Half aloud he said, without taking his eyes away from John, something to the others, …’ Nå traff den harde ballen John i knehasene. Han falt om som en stige som stod for steilt, først langsomt og så med tyngde. Smerten bredte seg fra hoften og albuen. Tom stod der igjen og smilte overbærende. Halvhøyt sa han noe til de andre uten å ta blikket fra John, … (nt9, 15) ‘Now the heavy ball hit John’s knee. He fell over like a ladder which was too steeply set up, first slowly and then with force. From the hip to the elbow the pain spread out. Tom appeared again and smiled indulgently. Half aloud he said something to the others, without taking his eyes away from John, …’

Our data confirm that coordination is a frequently used strategy to keep syntactically complex source structures together as units in the target version. To use it felicitously this material at least suggests that the second conjunct must be focal and the first conjunct must play the role of leading up to the second. This may be hard to combine with the challenge of giving an equivalent target version. Linking constituents as ­conjuncts in cases where it is hard to recover a consequence relation may give odd results. In other cases a downgrading which does not mirror the original structure, may be inferred. Furthermore, this leads to the very interesting question, discussed in FabriciusHansen et al. (2005) and Ramm and Fabricius-Hansen (2005), of whether the use of og-coordination in Norwegian really equals corresponding structures in German or English – a question which certainly deserves more cross-linguistic r­ esearch.

Acknowledgments This article is very much the result of joint work within the SPRIK-project, University of Oslo. I would like to thank Cathrine Fabricius-Hansen, Wiebke Ramm and Bergljot Behrens, Oslo, for invaluable cooperation and useful suggestions and comments. I would also like to thank three anonymous reviewers for very constructive and helpful comments.

References Asher, Nicholas. 1993: Reference to Abstract Objects in Discourse [Studies in Linguistics and Philosophy 50]. Dordrecht: Kluwer. Asher, Nicholas & Vieu, Laure 2005. Subordinating and coordinating discourse relations. Lingua 115(4): 591–610.

 Kåre Solfjeld Behrens, Bergljot & Fabricius-Hansen, Cathrine 2005. The relation accompanying circumstance across languages. Conflict between expression and discourse subordination? [SPRIKreport 32]. URL: http://www.hf.uio.no/forskningsprosjekter/sprik/docs/pdf/bb/Sprik-Report32bb-cfh.pdf. Oslo: University of Oslo. Blakemore, Diane 1987. Semantic Constraints on Relevance. Oxford: Basil Blackwell. Carston, Robyn 2002. Thoughts and Utterances. The Pragmatics of Explicit Communication. ­Oxford: Blackwell. Dietrich, Rainer 1994. Wettbewerb – aber wie? Skizze einer Theorie der freien Wortstellung. In Was determiniert Wortstellungsvariation? Studien zu einem Interaktionsfeld von Grammatik, Pragmatik und Sprachtypologie, Brigitta Haftka (ed.), 33–47. Opladen: Westdeutscher Verlag. Fabricius-Hansen, Cathrine 1996. Informational density: A problem for translation and translation theory. Linguistics 34: 521–565. Fabricius-Hansen, Cathrine 1999. Information packaging and translation. Aspects of translational sentence splitting (German – English/Norwegian). In Sprachspezifische Aspekte der Informationsverteilung, Monika Doherty (ed.), 175–213. Berlin: Akademie Verlag. Fabricius-Hansen, Cathrine 2005. Elusive connectives. A case study on the explicitness dimension of discourse coherence. Linguistics 43(1): 17–48. Klein, Wolfgang & von Stutterheim, Christiane 1992. Textstruktur und referentielle Bewegung. Zeitschrift für Literaturwissenschaft und Linguistik 86: 67–92. Lang, Ewald. 1984. The Semantics of Coordination [Studies in Language Companion Series 9]. Amsterdam: John Benjamins. Nølke, Henning 1995. Utterance Focus. Elements of a modular theory. In Information Structure [Copenhagener Studies in Language 18], 75–114. Michael Herslund (ed.) København/ Frederiksberg: Samfundslitteratur. Ramm, Wiebke 2005. Satzrelativsätze im Deutschen und ihre Übersetzung ins Norwegische. Zum Zusammenhang zwischen Satzteilung und Diskursfunktion. In Estudios Filológicos Alemanes. 9, Universidad de Sevilla: Revista del Grupo de Investigación de Filología Alemana. 259–277. Also available as SPRIKreport Nr. 28. URL: http://www.hf.uio.no/ forskningsprosjekter/sprik/docs/pdf/wr/WiebkeRammReport28.pdf. Oslo: University of Oslo. Ramm, Wiebke & Fabricius-Hansen, Cathrine 2005. Coordination and discourse-structural salience from a cross-linguistic perspective. In Salience in Discourse: Multidisciplinary ­Approaches to Discourse 2005, Manfred Stede, Christian Chiarcos, Michael Grabski & Luuk Lagerwerf (eds). Münster: Stichting/Nodus, 119–128. Also available as SPRIKreport Nr. 30. URL: http://www.hf.uio.no/forskningsprosjekter/sprik/docs/pdf/wr/WR-CFH-Report30. pdf. Oslo: University of Oslo. Sandström, Gørel 1993. When-clauses and the Temporal Interpretation of Narrative Discourse. Umeå: Department of General linguistics, University of Umeå. Solfjeld, Kåre 2000. Sententialität, Nominalität und Übersetzung. Eine empirische Untersuchung deutscher Sachprosatexte und ihrer norwegischen Übersetzungen [Osloer Beiträge zur Germanistik 26]. Frankfurt: Peter Lang. Solfjeld, Kåre 2004. Informationsspaltung in Sachprosaübersetzungen Deutsch-Norwegisch. In Kommunikasjon, Eva L. Björk & Sverre Vesterhus (eds), 111–130. Halden, Høgskolen i Østfold. Solfjeld, Kåre. 2008. Sentence splitting and discourse structure in translations (German-Norwegian). Languages in Contrast 8.1, 21–46.

Strategies to preserve discourse structure  Stutterheim, Christiane von 1997. Einige Prinzipien des Textaufbaus. Empirische Untersuchungen zur Produktion mündlicher Texte. Darmstadt: Niemeyer.

Source literature Cremer, Peter 1982. Ali Cremer. U 333. Berlin – Frankfurt/M – Wien: Ullstein, 225–240 (dt1). Norw. translation (bokmål) 1989. U 333. Vendepunktet. Translated by Tore Bjørn StornæsNilsen, Oslo: Faktum, 107–128 (nt1). Däniken, Erich von 1985. Habe ich mich geirrt? München: Bertelsmann, 166–194 (dt2). Norw. translation (bokmål) 1986. Har jeg tatt feil? Translated by Ådne Goplen, Oslo: Mortensen, 168–195 (nt2). Franz, Uli 1987. Deng Xiaoping. Chinas Erneuerer. Stuttgart: Deutsche Verlags-Anstalt (dt3), 125–144. Norw. translation (bokmål) 1988. Deng Xiaoping. Kinas sterke mann. Translated by Iver Tore Svenning. Oslo: Schibsted, 114–132 (nt3). Hackl, Erich 1987. Auroras Anlass. Zürich: Diogenes, 7–42 (dt4). Norw. translation (bokmål) 1988. Auroras motiv. Translated by Lasse Tømte. Oslo: Cappelen, 7–39 (nt4). Haller, Michael (ed.) 1981. Aussteigen oder Rebellieren. Authors of selected parts: Hans Halter and Walter Tauber, Hamburg: Spiegel, 100–116 (dt5). Norw. translation (bokmål) 1982. Isolasjon eller opprør. Translated by Truls Wyller. Oslo: Gyldendal, 66–80 (nt5). Hermlin, Stephan 1979. Abendlicht. Berlin: Klaus Wagenbach, 58–87 (dt6). Norw. translation (bokmål) 1980. Kveldslys. Translated by Carl Fredrik Engelstad. Oslo: Gyldendal, 53–78 (nt6). Lorentz, Konrad 1963. Das sogenannte Böse. Zur Naturgeschichte der Aggression. Wien: BorothaSchoeler,19–33 (dt7). Norw. translation (bokmål) 1968. Den såkalte ondskap. Om aggresjon hos mennesker og dyr. Translated by Brynjulf Valum. Oslo: Cappelen, 21–35 (nt7). Lorenz, Einhart 1989. Willy Brandt in Norwegen. Kiel: Neuer Malik, 132–158 (dt8). Norw. translation (bokmål) 1989. Willy Brandt i Norge. Translated by Anne-Lise Risø. Oslo: Tiden, 97–115 (nt8). Nadolny, Sten 1983. Die Entdeckung der Langsamkeit. München – Zürich: Piper, 9–17 (dt9). Norw. translation (bokmål) 1985. Mannen som ilte langsomt. Translated by Ellen Harboe Meisingset. Oslo: Cappelen, 11–17 (nt9). Schreiber, Hermann 1978. Singles Allein leben. Besser als zu zweit? München: Bertelsmann, 115– 133 (dt10). Norw. translation (bokmål) 1980. Leve alene. Bedre enn å være to? Translated by Leif Toklum. Oslo: Gyldendal, 89–103 (nt10). Senger, Gerti 1985. Frauenträume Männerträume und ihre Bedeutung Niedernhausen/Ts: Falken-Verlag, 8–29 (dt11). Norw. translation (bokmål): Kvinnedrømmer Mannsdrømmer. Translated by Tormod Ropeid. Oslo: Teknologisk Forlag, 9–32 (nt11). Prekop, Jirina 1988. Der kleine Tyrann. München: Kösel-Verlag, 49–67 (dt12). Norw. translation (bokmål) 1989. Den lille tyrann. Translated by Peter Normann Waage and Eivind Tjønneland. Oslo: Ex libris, 55–76 (nt12). Wallraff, Günter 1985. Ganz unten. Köln: Kiepenheuer & Witsch, 84–101 (dt13). Norw. translation (bokmål) 1986. Aller nederst. Translated by Ola Johnsrud. Gjøvik: Pax, 73–87 (nt13).

Upgrading of non-restrictive relative clauses in translation A change in discourse structure?1 Wiebke Ramm

University of Oslo The paper investigates the upgrading of non-restrictive German relative clauses to independent main clauses in Norwegian translations. The question is raised whether the discourse function of the relative clause has any impact on the interpretation of the upgraded translation. Previous research has revealed that German non-restrictive relative clauses can serve different discourse functions: a discontinuative/appositive or a continuative discourse function. These correlate with different types of discourse relations. A corpus study supports the hypothesis that upgrading is more problematic with respect to discontinuative/appositive relative clauses. However, the study also reveals certain shortcomings of the applied approaches. The upgrading of relative clauses makes the translations less hierarchically organised than the original texts. This allows for some reflections on the cross-linguistic assessment of coherence within and across sentence boundaries. Keywords: Translation of relative clauses, subordination in syntax vs. discourse, discourse organisation strategies, coherence.

1.  Introduction In his contribution, Solfjeld (in this volume) shows how relatively complex structural changes in a translation may nevertheless lead to similar interpretations of a source

.  The research presented here was developed as part of my PhD project funded by the Faculty of Humanities, University of Oslo. I am also grateful to the project ‘SPRIK (Språk i kontrast/ Language(s) in contrast)’ funded by the Norwegian Research Council (project number NFR 158447/530), for supporting my participation at the DGfS-06 conference in Bielefeld. I would like to thank Cathrine Fabricius-Hansen for her helpful comments on earlier versions of this paper and Barbara Clare Dalton, Bergljot Behrens and Signe Oksefjell Ebeling for correcting my English. Last but not least, I owe gratitude to the three reviewers, for their useful comments and suggestions for improving the paper.

 Wiebke Ramm

language (SL) and a target language (TL) text. In this paper, we will take the opposite case as a starting point and show that seemingly small adjustments in the sentence structure may nonetheless have significant consequences for the discourse interpretation of the TL text for example as regards anaphora resolution. Examples of German non-restrictive relative clauses upgraded to independent main clauses in Norwegian translations will be taken from a German-Norwegian parallel corpus of popular science articles. We will investigate how the interpretation of the translation differs from the SL text and whether the discourse function of the relative clause (RC) in the original text has an impact on the interpretation of the (upgraded) TL version. Section 2 explains why non-restrictive RCs in German represent a problem when translated into Norwegian and summarises two approaches addressing the discourse function of (German) non-restrictive RCs. These approaches provide the framework for the description of the discourse structures of German text fragments containing a non-restrictive RC and for the comparison with their sentence-structurally upgraded Norwegian translations. Section 3 presents a hypothesis about the consequences of upgrading German RCs to independent sentences in translation and the predictions that follow from it. We then present and evaluate the results of a corpus study testing the hypothesis and its predictions. Two different discourse organisation strategies are considered in relation to the RC examples and their translations in Section 4, and some possible implications of the different strategies for the coherence of a text are discussed. Finally, Section 5 summarises the findings, concluding with an outlook on how the analysis of relatively small structural adjustments in translation – such as those discussed in this paper – can contribute to a better understanding of a crosslinguistically valid notion of coherence.

2.  Non-restrictive relative clauses in German 2.1  Non-restrictive relative clauses as a problem for translation Due to typological differences between the two languages, it is not unusual for nonrestrictive German RCs to be translated as independent main clauses in Norwegian.2 German relative markers with nominal antecedents are pronouns with referential properties, and the case, number and gender agreement requirement helps to disambiguate potential antecedents in the matrix clause. By contrast, the most frequent Norwegian

.  In a corpus study investigating sentence boundary adjustments in translations of popular science texts between Norwegian and German, non-restrictive RCs turned out to be the most frequent cause of sentence splitting (i.e., the translation of one SL clause or clause complex as two or more independent sentences in the TL) for the translation direction German-Norwegian (Ramm, in preparation).

Upgrading of non-restrictive relative clauses in translation 

relative marker, som, is non-inflecting and usually considered as a subjunction (Faarlund et al. 1997: 866). This lack of potential for disambiguation, sometimes in interaction with other typologically motivated changes of word order in the matrix clause, often leaves the upgrading of an RC to an independent main clause and the reintroduction of the referent as the only option for the Norwegian translation, as in example (1a) vs. (1b). (1) a. Von 1905 bis zum Ersten Weltkrieg existierten kaum außenpolitische Berührungspunkte zwischen dem Deutschen Reich und Norwegen, dessen Regierung nach der Unabhängigkeit eine permanente außenpolitische Neutralität anstrebte. […] [BB] Literal (linear) English gloss: ‘From 1905 until World War I existed hardly any foreign-political points of contact between the German Reich and Norway, whose government after the independence a permanent foreignpolitical neutrality strived for.’ b. Fra 1905 til utbruddet av første verdenskrig fantes det få utenrikspolitiske berøringspunkter mellom Det tyske riket og Norge. Etter uavhengigheten ønsket den norske regjeringen utenrikspolitisk nøytralitet, […].[BBTN] ‘From 1905 until the outbreak of World War I existed few foreign-political points of contact between the German Reich and Norway. After the independence wished the Norwegian government foreign-political neutrality, […].’

RCs with the whole matrix clause as antecedent, so-called Satzrelativsätze (‘clause-­related RCs’), are another type of RC frequently rendered as an independent sentence in Norwegian translations. Norwegian equivalents of German relative adverbials − such as wobei (lit. ‘where-by’), weshalb (‘wherefore’, ‘for what reason’), wodurch (lit. ‘where-through’) and womit (lit. ‘where-with’) − are missing in Norwegian, or at least they are highly marked or old-fashioned. Consequently, the only option available to the Norwegian translator is often to render the RC in the form of an independent sentence and to express the discourse relation encoded by the relative adverbial in German by a corresponding connective (example (2a) vs. (2b)). Alternatively, the discourse relation can be left implicit (example (3a) vs. (3b)). The relative marker was (‘what/which’) is the only relative marker which has a quasi-equivalent in Norwegian, noe (som) (‘something (which/that)’), which makes upgrading dispensable in most of these cases. (2) a. Das militärstrategische Primat wird auch daran erkennbar, dass die Militärbehörden durch die Rüstungsprogramme über erhebliche Kompetenzen in wirtschaftlichen Fragen verfügten, weshalb deutsche Firmen auch mit diesen kooperieren mussten, obwohl die Wirtschaftspolitik formell durch das Reichskommissariat gestaltet wurde. [BB] ‘The military-strategic primacy becomes also [daran (pron. adv., lit. ‘there-on’)] visible, that the military authorities through the arms programs considerable competences in economic questions had, [weshalb] German companies with these cooperate must, although the economic policy formally through the Reichskommissariat formed/organised was.’

 Wiebke Ramm

b. Et forhold som reflekterer det militær-strategiske primatet, var at rustningsprogrammene hadde gitt de militære myndighetene en betydelig kompetanse i økonomiske spørsmål. Derfor måtte de tyske firmaene samarbeide med dem, selv om den økonomiske politikken formelt sett ble utformet av rikskommissariatet. [BBTN] ‘A fact that reflects the military-strategic primacy, was that the arms programs had given the military authorities considerable competence in economic questions. Therefore must the German companies cooperate with them, although the economic policy formally seen was formed by the Reichskommissariat.’ (3) a. Neben diesen engen wirtschaftlichen Kontakten bestanden rege kulturelle, technische und wissenschaftliche Beziehungen, wobei für das norwegische Bildungsbürgertum die deutsche Literatur und die deutsche Sprache den Bezugspunkt darstellte. [BB] ‘Besides these close economic contacts existed active cultural, technical and scientific relations, [wobei] for the Norwegian educated classes the German literature and the German language the point of reference represented.’ b. Ved siden av disse tette handelsforbindelsene fantes det også viktige kulturelle, tekniske og vitenskapelige kontakter. Tysk språk og litteratur stod sterkt i de utdannede borgerlige kretser. [BBTN] ‘Besides these close trade connections existed also important cultural, technical and scientific contacts. German language and literature had a strong position in the educated middle-class circles.’

The question is whether such modifications of clause linkage make a difference for the discourse interpretation of the original vs. translated versions of the text, and if they do, in what way?

2.2  The discourse function of non-restrictive relative clauses German noun-related and clause-related non-restrictive RCs such as those presented in the previous section are often described as weiterführende Relativsätze (‘continuative RCs’), a subtype of non-restrictive RCs which can only appear in sentence-final position, and, in studies on German RCs, their similarity to independent main clauses is often emphasised (Brandt 1990: 46–52; Peyer 1997: 141–142; Laux 2002: 199–204). In contrast to an adverbial or complement clause, a continuative RC is not syntactically embedded − it does not realise a syntactic function within the matrix clause − and neither is it pragmatically integrated into the matrix clause (signalled by intonation, for instance), having an information structure (focus-background structure) of its own (Fabricius-Hansen 1992: 479; Holler 2005: 131; Holler in this volume). The only obvious difference between a weiterführende Relativsatz and a main clause is its syntactic dependency on the matrix clause, signalled by the clause-final position of the finite verb in German. We are interested in specifying its discourse function, and in

Upgrading of non-restrictive relative clauses in translation 

establishing the means by which this function is retained – or not retained – when it is upgraded to an independent sentence in the Norwegian translation. Laux (2002) and Holler (2005) investigate non-restrictive RCs in German – Laux focussing on their temporal interpretation in narrative texts and Holler providing a formal discourse-semantic account and computational-linguistic modelling in HPSG – and develop similar criteria for the identification and description of weiterführende Relativsätze. Both authors investigate the temporal structure of the matrix clause and the RC in order to discover in what way the RC represents a continuation of the events described in the matrix clause. Both use the distinction between Hauptstruktur (main structure) and Nebenstruktur (side structure) as delineated by Klein and v. Stutterheim’s (1987, 1991) ‘quaestio’ approach in order to determine how the RC contributes to answering the ‘quaestio’, the text question guiding the development of the respective text fragment; and, finally, they both use the model of Segmented Discourse Representation Theory, SDRT, (Asher 1993; Asher & Lascarides 2003) to describe the discourse relations holding between the matrix clause and the RC. (4) a. Emil traf einen Bauern, der übrigens einen weiten Mantel trug. [Ihn fragte er nach dem Weg.] ‘Emil met a farmer, who by the way a wide coat wore. [Him asked he for the way.]’ b. Emil traf einen Bauern, den er dann nach dem Weg fragte. [So konnte er ohne Probleme den Gasthof finden.] ‘Emil met a farmer, whom he then asked for the way. [So was he able without problems the inn to find.]

Example (4a) instantiates what Laux (2002) calls a discontinuative (diskontinuativ) and what Holler (2005) calls an appositive (appositiv) RC.3 The RC is not seen as part of the main structure of the text, as it merely contains additional or background information and does not push forward the development of the (narrative) text. ­Applying Holler’s (2005) approach, an asymmetric, i.e., subordinating, discourse relation (Elaboration) can be identified between the matrix clause and the RC. The distinction ­between subordinating and coordinating discourse relations in the SDRT model (Asher & Lascarides 2003; Asher & Vieu 2005) reflects an assumption common to many approaches to discourse structure − that discourse has a hierarchical structure. A central criterion for distinguishing between the two types of discourse relation is the possible attachment points for new information in a discourse structure: if two discourse units are linked by a subordinating discourse relation (the prototypical example being Elaboration), then both units provide possible attachment points to new information; whereas with a coordinating discourse relation between two segments (as in the prototypical case of Narration), the last but one unit is not available for the attachment of incoming new

.  Example (4a) and (4b) are adapted from Holler (2005: 211).

 Wiebke Ramm

information, i.e., it is not at the right end of the incrementally constructed discourse representation. This is roughly what is called the right-frontier constraint (RFC) in SDRT (Asher 1993: 270–271; Asher & Lascarides 2003: 10–12). Other important factors entering into play here are the temporal relations holding between the eventualities described in the discourse units and how discourse topics are constructed and maintained. Laux’s (2002) main criterion to distinguish between continuative (kontinuativ) and discontinuative RCs is whether a Consequentiality relation in the sense of Sandström (1993) – which is defined similarly to Narration in SDRT – can be identified between the clauses. In contrast to Holler’s approach, not only is the discourse relation holding between the matrix clause and the RC taken into account here, but so is the relation holding between the RC and the following sentence. According to Laux’s definition (2002: 189–190), a RC is continuative only when a Consequentiality relation holds between the matrix clause and the RC, and between the RC and the following sentence (given in square brackets in the examples). Example (4a) is not continuative on this premise, as there is neither a Consequentiality relation between the matrix clause and the RC, nor between the RC and the following context. The RC in (4b) would be classified as weiterführend in Holler’s and as kontinuativ in Laux’s terminology. The RC has the same communicative weight as an independent main clause and is part of the main structure of the text, since it makes an individual contribution to answering the ‘quaestio’ of the text fragment (here something like ‘What happened?’). According to Holler’s definition, a symmetric i.e., coordinating, discourse relation (Narration) holds between the matrix clause and the RC, while Laux’s method leads to the assignment of a Consequentiality relation between the matrix clause and the RC, and between the RC and the following sentence. Example (4a) and (4b) illustrate a difference in the mapping between syntactic structure and discourse organisation which is interesting from the perspective of translation. While in (4a) syntactic subordination correlates with a subordinating (or asymmetric) discourse relation, syntactic subordination in (4b) does not correspond to subordination or asymmetry on the discourse level. Thus we have a mismatch between syntactic structure and discourse organisation in this example (for similar observations regarding the possible non-correspondence of subordination in syntax and discourse see Delort in this volume). Although approaches operating with a concept of ‘subordination’ or ‘asymmetry’ of discourse relations such as SDRT or Rhetorical Structure Theory (RST) (Mann & Thompson 1988) often emphasise that there is no direct mapping between a particular type of discourse relation and its lexico-grammatical realisation, there seems to be a common understanding that syntactically subordinate clauses are typically subordinated or less salient in discourse as well. Matthiessen and Thompson (1988), for example, view hypotaxis as the grammaticalisation of a property of the hierarchical structure of the discourse itself, and also Brandt (1990: 128) concludes that – at least in the default case – the subordinate clause form of a continuative RC is used to indicate that an information unit is less important than its matrix clause. A similar view of the downgrading function that syntactic subordination has in

Upgrading of non-restrictive relative clauses in translation 

discourse can be found in Carroll et al. (in this volume). The potential contribution of subordinate vs. main clauses is also different as regards information-structure. Unless explicitly marked to the contrary,4 a main clause always contains some new (in the sense of ‘asserted’) information, whereas this is not necessarily the case for a subordinate clause. In this way, syntactic upgrading may be accompanied by a change in the information status of a discourse referent from being presupposed (given) to asserted (new). However, while potentially relevant for the translations of the RC examples discussed in this paper, this aspect of discourse interpretation will not be taken up in the following discussion. As mentioned above, what distinguishes a continuative RC from a ‘normal’ subordinate (or relative) clause is precisely its information-structural independence and similarity to a main clause. We will now turn to the question of whether the different functions a non-restrictive RC may have in discourse also play a role in how they are translated into Norwegian, with particular regard to cases where a corresponding RC in the Norwegian translation is not an option.

3.  Translational upgrading of relative clauses – a corpus study 3.1  Possible consequences of upgrading of relative clauses Transforming a RC into an independent sentence is a translational operation that requires relatively little syntactic restructuring in relation to the SL text. In many cases it involves only smaller adjustments of the anaphoric links (such as replacing the relative pronoun by a personal pronoun or a full NP), and the verbal predicate (for example, the upgrading may require that a RC in the passive voice – typically starting with durch (‘through’) + relative pronoun – is transformed into active voice in the translation). Thus, if it holds that, in general, a translator tries to conserve as much of the ordering and structuring of the SL text as possible (cf. Solfjeld in this volume), then it should be expected that the syntactic upgrading of a RC will be one of the preferred translation choices. Yet even seemingly minor modifications in the syntactic dependency structure of a sentence may have clear ramifications for its interpretation, depending on the contribution the modified element makes to discourse development. One hypothesis following from the discussion above is the following: Hypothesis (H): The closer the RC is to an independent main clause and the more it contributes to answering the quaestio, the less translational upgrading will change the interpretation. This suggests that continuative RCs are less

.  For example, Fabricius-Hansen (2000: 73–74) discusses an example of a non-restrictive German RC upgraded in the Norwegian translation, where givenness is marked by adding altså (‘so’, ‘thus’) in order to block an interpretation as new information.

 Wiebke Ramm

problematic for discourse interpretation when syntactically upgraded in translation than discontinuative/appositive RCs are.5

Example (5a) and (5b) – intra-lingually upgraded variants of (4a) and (4b) – illustrate the intuition behind this hypothesis. (5) a. Emil traf einen Bauern. Der/dieser trug übrigens einen weiten Mantel. [Ihn fragte er nach dem Weg.] ‘Emil met a farmer. He/[dieser (dem.)] wore by the way a wide coat. [Him asked he for the way.]’ b. Emil traf einen Bauern. Den/diesen fragte er dann nach dem Weg. [So konnte er ohne Probleme den Gasthof finden.] ‘Emil met a farmer. Him/[diesen (dem.)] asked he then for the way. [So was he able without problems the inn to find.]

In example (5a), where the discontinuative/appositive RC of (4a) has been upgraded, the new independent sentence makes it more difficult to reconstruct the anaphoric link between einen Bauern (‘a farmer’) in the first sentence and the masculine pronoun ihn (‘him/it’) in the sentence following the upgraded RC. In fact, ihn (‘him/it’) in (5a) is ambiguous, either referring to einen weiten Mantel (‘a wide coat’) – an interpretation supported by adjacency – or (the intended interpretation) to einen Bauern (‘a farmer’) – an interpretation supported by parallelism of semantic roles. There is no such ambiguity in the RC variant in (4a), due to the expectation of topic continuation from the matrix clause in the first sentence of (4a) to the second sentence, i.e., due to the syntactically downgraded nature of the RC. Example (5b), the upgraded version of the continuative example (4b), however, does not cause any problems for reference resolution, and the two versions seem to have very similar interpretations. Hypothesis H allows for some interesting predictions regarding the practice of how German RCs typically are translated into Norwegian: Prediction (P1): Translators avoid upgrading discontinuative/appositive RCs in Norwegian translation and use other translation strategies instead. Prediction (P2): If discontinuative/appositive RCs nevertheless are upgraded in translation (e.g., because no alternative translation options are available), this leads to a less coherent text or to a different interpretation.

.  In the remainder of this paper we will use the terms ‘continuative RC’ and ‘discontinuative/ appositive RC’ as they are defined in Laux (2002) and Holler (2005). In theory, there could be examples which would be classified as continuative following Holler (2005) but not according to Laux (2002) – namely, when a Consequentiality relation holds between the matrix clause and the RC, but not to the following sentence (a constellation Laux (2002: 193) calls ‘blind end’ (Sackgasse) – in practice, however, the correlation of the classification of the RC examples is relatively good, at least in the examples discussed in this paper.

Upgrading of non-restrictive relative clauses in translation 

To test these predictions, data from two parallel corpora of German original texts translated into Norwegian were investigated: a corpus of three (popular-)scientific articles on the subject of historical relations between Norway and Germany (consisting of approximately 15000 words / 574 sentences), and the German-Norwegian part of the Oslo Multilingual Corpus (OMC) containing fictional and non-fictional texts (13 fictional and 17 non-fictional texts, about 40000 words in total at the time this study was undertaken).6 As concerns the testability of the two predictions, it should be pointed out that P1 and P2 have a different status: Whereas P1 should be testable on the basis of a corpus study, namely by checking whether the relevant RCs are upgraded or not, statements regarding P2 have to be more tentative, since a purely linguistic analysis has its limits as regards judgements about the interpretation and cognitive processing of a piece of discourse. The corpus data are discussed in the following two sections: In Section 3.2, corpus examples of clause-related RCs upgraded in the Norwegian translation are analysed and Section 3.3 addresses the translation of German RCs with a nominal antecedent. Section 3.4. evaluates the results of the corpus study with respect to the hypothesis and predictions formulated in this section. A general observation as regards the translation of German RCs is that restrictive RCs appear to be never upgraded to independent sentences and almost always correspond to RCs in the Norwegian versions of the texts. This does not come as a surprise, since restrictive RCs denote properties (restricting the possible denotations of the NP) and do not have so-called ‘root-clause’ properties. This implies that they cannot be transformed into independent sentences without bringing about a semantic change (Holler 2005: 30, Holler in this volume). But also many non-restrictive RCs are translated by RCs in Norwegian, as long as grammatical contrasts between the two languages do not block this option.

3.2  Translational upgrading of clause-related relative clauses As regards clause-related RCs, some evidence confirming prediction P1 was found in the OMC data which consist of a mixture of fictional (narrative) and non-fictional texts,7 i.e., here the translators apparently avoid the upgrading of discontinuative/ appositive RCs: 42 out of 61 clause-related RCs found in the German SL texts were discontinuative/appositive, and only 6 of these examples (14,3 %) were upgraded to an independent sentence in the Norwegian translation. Paratactic clause coordination

.  See http://www.hf.uio.no/forskningsprosjekter/sprik/english/corpus/index.html for a description of the Oslo Multilingual corpus (OMC). .  The data for clause-related RCs were retrieved from the OMC database by lexical search for the individual relative adverbs. The relative marker was (‘what/which’) was excluded from the study, because it was not possible to automatically filter out the instances of was not corresponding to a relative marker.

 Wiebke Ramm

with og (‘and’), as in example (6b) below, is a pattern used relatively frequently as an alternative translation strategy.8 (6) a. Meine Eltern untereinander sprachen deutsch, wovon ich nichts verstehen durfte. Zu uns Kindern und zu allen Verwandten und Freunden sprachen sie spanisch. [EC] ‘My parents with each other spoke German, [wovon (lit. ‘where-of ’)] I nothing to understand was allowed to. To us children and to all relatives and friends they spoke Spanish.’ b. Mine foreldre snakket tysk med hverandre, og det var ikke meningen at jeg skulle forstå noe av det. Til oss barn og til alle slektninger og venner snakket de spansk. [ECTN] ‘My parents spoke German with each other, and it was not intended that I should understand anything of this. To us children and to all relatives and friends they spoke Spanish.’

In the purely non-fictional corpus of scientific articles, however, evidence confirming P1 was more tentative: 18 of 32 clause-related RCs found in the German SL texts were discontinuative/appositive, but in several cases categorisation was problematic, cf. the discussion below. 8 of these examples (44 %) were upgraded as an independent sentence in the Norwegian translations. One factor complicating the interpretation of these results is that it seems that the different relative adverbs (wobei, weshalb, wodurch, etc.) behave differently. Their individual meanings, often in combination with the occurrence of other connectives (adverbs and particles), determine a continuative vs. discontinuative interpretation of the RC in question. The relative adverb wobei (lit. ‘where-by’), for example, typically introduces a clause-related RC describing some accompanying circumstance, which in many cases leads to a classification as discontinuative/appositive as in example (3a), repeated as (7a) below – 9 out of 11 wobei-examples in the corpus are discontinuative/ appositive. In her analysis of the German connective dabei (lit. ‘there-by’), which is similar to wobei in that it consists of an anaphoric component combined with the preposition bei (‘with, at’), Fabricius-Hansen (2005: 40) points out that “a main function of dabei is to prevent the narrative from ‘moving forward’”. Therefore dabei is often added in German translations of English ing-progressive forms, a grammatical resource which is not found in German. It is reasonable to assume that wobei has a similar effect to dabei of blocking a continuative reading, not least because wobei additionally triggers syntactic subordination (as do English ing-adjuncts), which is a further indicator of a downgraded, discontinuative interpretation.

.  On the relative frequency and special use of clause coordination with og (‘and’) in Norwegian texts in general, see Ramm and Fabricius-Hansen (2005), Fabricius-Hansen et al. (2005) and Ramm (forthcoming).

Upgrading of non-restrictive relative clauses in translation 

(7) a. Neben diesen engen wirtschaftlichen Kontakten bestanden rege kulturelle, technische und wissenschaftliche Beziehungen,[I] wobei für das norwegische Bildungsbürgertum die deutsche Literatur und die deutsche Sprache den Bezugspunkt darstellte.[II] [Zahlreiche skandinavische Studenten studierten in Deutschland, und zwischen den protestantischen Kirchen existierten enge Bindungen.][III] [BB] ‘Besides these close economic contacts existed active cultural, technical and scientific relations,[I] [wobei] for the Norwegian educated classes the German literature and the German language the point of reference represented.[II] [Numerous Scandinavian students studied in Germany, and between the protestant churches existed tight connections.][III]’ b. Ved siden av disse tette handelsforbindelsene fantes det også viktige kulturelle, tekniske og vitenskapelige kontakter.[I]Tysk språk og litteratur stod sterkt i de utdannede borgerlige kretser.[II] [Utallige skandinaviske studenter studerte i Tyskland, og det fantes tette bånd mellom de protestantiske kirkene.][III] [BBTN] ‘Besides these close trade connections existed also important cultural, technical and scientific contacts.[I] German language and literature had a strong position in the educated middle-class circles.[II] [Innumerable ­Scandinavian students studied in Germany, and there were tight connections between the protestant churches.][III]’

Example (7) seems to give evidence for the validity of prediction P2 as the Norwegian translation (7b) appears to be more ‘chopped up’ than the German (7a). One possible explanation for this is that the relation of temporal and spatial overlap of the eventualities described in the matrix clause and the RC, signalled by wobei in (7a), is not signalled by a corresponding connective in (7b) and thus has to be inferred from the context only. Explicit marking of co-temporality and co-spatiality between the matrix clause and the RC in (7a), together with the marking of syntactic dependency of the RC, favours an interpretation of the matrix clause as being more salient than the RC, which in turn assists the interpretation of the following sentence (7a[III]) as continuing the matrix clause. Such an interpretation is much harder to get in (7b), since the attachment of (7b[III]) in the discourse structure is not as obvious, i.e., it is less obvious that the first sentence (7b[I]) is still on the ‘right frontier’ when processing (7b[III]), than it is in (7a). While the meaning of wobei itself clearly guides the interpretation of the RC as ­appositive/discontinuative, it seems that adverbial connectives such as allerdings (‘however’), freilich (‘admittedly’), letztlich (‘ultimately’) or bereits (‘already’), in combination with the relative marker and the semantics of the verbal predicate, often trigger a continuative reading of the respective RC. In the scientific prose corpus, 9 out of 14 clause-related RCs introduced by the semantically underspecified relative marker was (‘which/what’) contain an additional adverbial connective, and 6 of these examples are continuative, as in (8a) below. This accords with Laux’s (2002: 260) observation that

 Wiebke Ramm

clear signals (e.g., lexical) are usually required to indicate the continuative discourse function of a RC. Interestingly, in many cases the discourse relation expressed by an adverbial connective in the German version is realised as a corresponding conjunction in the (upgraded) Norwegian translation, as in (8), where allerdings (‘however’) is translated by men (‘but’) in the Norwegian version of the text. (8) a. Vor allem nach der britisch-französischen Kapitulation auf der Konferenz von München im Herbst 1938 mehrten sich die Anzeichen, dass das Deutsche Reich eine wirtschaftliche und politische Interessensphäre in Skandinavien anstrebte, was allerdings wegen der starken britischen ökonomischen Präsenz in der Region nur wenig Erfolg versprechend war. [Dies zeigt, wie weit die deutsche Außenwirtschaftspolitik sechs Jahre nach der so genannten Machtergreifung noch davon entfernt war, eine Großraumwirtschaft zu etablieren.] [BB] ‘First and foremost after the British-French capitulation at the conference of Munich in autumn 1938 increased the indications, that the German Reich an economic and political sphere of interest in Scandinavia strived for, which however because of the strong British economic presence in the region only little promising was. [This shows, how far away the German foreign-economic politics six years after the so-called seizure of power was, a Großraumwirtschaft to establish.]’ b. Framfor alt etter den britisk-franske kapitulasjonen under Münchenkonferansen høsten 1938 ble det stadig tydeligere at Det tyske riket siktet mot å opprette en økonomisk og politisk interessesfære i Skandinavia. Men på grunn av den sterke britiske tilstedeværelsen i området hadde en slik plan få utsikter til å kunne realiseres. [Dette viser hvor langt den tyske utenrikspolitikken fortsatt befant seg fra å kunne etablere en økonomisk kontinentalblokk, selv seks år etter nazistenes maktovertakelse.] [BBTN] ‘First and foremost after the British-French capitulation autumn 1938 became it more and more clear that the German Reich aimed to establish an economic and political sphere of interest in Scandinavia. But because of the strong British presence in the region had such a plan few chances to be realised. [This shows how far away the German foreign policy still was from to be able to establish an economic continental block, even six years after the Nazi’s seizure of power.]’

A second factor that may blur the results of the corpus study is that, in many cases, it is difficult to determine whether the RC is continuative or discontinuative – in regard to both Laux’s and Holler’s approaches. A case on point is example (2a), from a text discussing the role of Norway in German military strategy between 1914 and 1945. The example is repeated as (9a) below.

(9) a. Das militärstrategische Primat wird auch daran erkennbar, dass die Militärbehörden durch die Rüstungsprogramme über erhebliche Kompetenzen in wirtschaftlichen Fragen verfügten, weshalb deutsche Firmen auch mit

Upgrading of non-restrictive relative clauses in translation 

diesen kooperieren mussten, obwohl die Wirtschaftspolitik formell durch das Reichskommissariat gestaltet wurde. [Dadurch wurde nicht die zukünftige Friedenswirtschaft eines ‘Großraumes’ vorprogrammiert, sondern Norwegen in die aktuellen Bedürfnisse der deutschen Rüstungsproduktion eingebaut.] [BB] ‘The military-strategic primacy becomes also [daran (pron. adv., lit. ‘thereon’)] visible, that the military authorities through the arms programs considerable competences in economic questions had, [weshalb] German companies with them/these cooperate must, although the economic policy formally through the Reichskommissariat formed/organised was. [Dadurch (‘with this’) was not the future peace economy pre-programmed, but Norway in(to) the actual needs of the German arms production built in.]’ b. Et forhold som reflekterer det militær-strategiske primatet, var at rustningsprogrammene hadde gitt de militære myndighetene en betydelig kompetanse i økonomiske spørsmål. Derfor måtte de tyske firmaene samarbeide med dem, selv om den økonomiske politikken formelt sett ble utformet av rikskommissariatet. [Følgelig ble ikke Norge programmert for en plass innenfor en fremtidig kontinental økonomisk blokk, men snarere utviklet i samsvar med de aktuelle behovene til den tyske rustningsproduksjonen.] [BBTN] ‘A fact that reflects the military-strategic primacy, was that the arms programs had given the military authorities considerable competence in economic questions. Therefore must the German companies cooperate with them, although the economic policy formally seen was formed by the Reichskommissariat. [Consequently was not Norway programmed for a place within a future continental economic block, but rather developed in correspondence with the actual needs of the German arms production.]’

It is unclear whether the RC (together with the following concessive clause) continues the events described in the matrix clause – which would correspond to a coordinating discourse relation – or whether it comments on them – which would rather support a subordinating discourse relation. Furthermore, this ambiguity is not resolved in the following context, since the pronominal adverb dadurch (‘with this’) in the sentence following the weshalb-clause is also ambiguous, i.e., it is open whether it refers back to the whole clause complex (the intended interpretation?9) or only to the clause-related RC (which would classify the RC as continuative according to Laux’s definition). The følgelig (‘consequently’) in the Norwegian version (9b), however, has its most likely antecedent in the sentence before, which is the upgraded counterpart of the German weshalb-sentence. As a result, the connection between the economic

.  Due to space limits it is not possible to include more of the context before the example ­sentence. This context, however, would have made it clearer that taking dadurch as referring to the whole preceding sentence, not only the RC, leads to a more plausible interpretation here.

 Wiebke Ramm

influence of the German military administration mentioned in the German matrix clause, and the consequences this had on the role Norway played (economically, after the First World War – which is what this text paragraph is about), are much harder or almost impossible to infer in the Norwegian translation. What seems to make syntactic upgrading of the RC additionally problematic in this example is the fact that the RC in the German version is syntactically subordinate to a matrix clause which is itself a subordinate clause. Translating the RC by an independent sentence thus ‘lifts’ the clause up two syntactic levels, possibly giving it too much prominence in discourse structure when compared with the SL version.

3.3  Translational upgrading of relative clauses with a nominal antecedent The examples of clause-related RCs illustrate that translational upgrading of a syntactically dependent clause to an independent sentence can change the discourse interpretation and lead to problems, e.g., for the reconstruction of the referential structure of the text passage. However, it is not clear to what extent this is due to the discontinuative nature of the RC upgraded in the translation. This turned out to be even more difficult to decide in the case of non-restrictive relative clauses with a nominal antecedent in the matrix clause, since German RCs – particularly those in non-fictional texts of the type seen in the scientific prose corpus – may be very complex. For instance, examples with a RC depending on a matrix clause which is in itself a dependent clause (as in (9a) above) seem to be even more frequent than is the case with clause-related RCs. A noun-related non-restrictive RC may express very different types of things, depending on the nature of the entities the RC is anaphorically linked to and the contribution of the RC itself. This makes the application of either Holler’s or Laux’s criteria for distinguishing between discontinuative/appositive and continuative RCs almost unfeasible in many instances. Whether the situation time in the matrix clause is completed, and whether there is a shift of topic time between the matrix clause and the RC – two of Holler’s diagnostic features for continuative RCs with a nominal antecedent (Holler 2005: 162) – is often difficult to resolve for complex examples in ‘wildlife’ texts, not least as in some sentences it can even be difficult to identify the topic itself. Example (10a) is a complex instance of a RC with a nominal antecedent, where the RC contains additional information on a nominal discourse participant (Hitler); but this description is not continued in the following sentence (although it could have been), rather the claim in the matrix clause is taken up − namely that Rosenberg’s ‘Außenpolitische Amt’ did not have any influence on the ‘Auswärtige Amt’ and Hitler. (10) a. Das am 1. April 1933 unter Alfred Rosenberg gegründete Außenpolitische Amt der NSDAP blieb in der Folge fast ohne Einfluss gegenüber dem Auswärtigen Amt und Hitler,[I] der seit Ende 1933 punktuell in die außenpolitischen Entscheidungsprozesse einzugreifen begann.[II] [Rosenberg scheiterte wie auch Darré, der selbst ernannte völkische Parteiexperte für

Upgrading of non-restrictive relative clauses in translation 

den Agrarsektor, damit, seiner Dienststelle in der sich herausbildenden polykratischen Struktur des nationalsozialistischen Staates ein Profil und eine Machtbasis zu verschaffen.[III]] [BB] ‘The on the 1. April 1933 under Alfred Rosenberg founded foreign-political office of the NSDAP remained in the following time almost without influence on the Foreign Office and Hitler,[I] who since the end of 1933 punctually into the foreign-political decision processes to intervene started.[II] [Rosenberg failed as also Darré, the self-appointed national (völkisch) expert for the agrarian sector, [damit (pron. adv.)] for his department in the growing polycratic structure of the National Socialist State a profile and a basis of power to provide.[III]]’ b. 1. april 1933 ble NSDAPs utenrikspolitiske kontor grunnlagt under ledelse av Alfred Rosenberg, men den fikk svært liten innflytelse over utenriksministeriet og Hitler.[I] Sistnevnte begynte fra slutten av 1933 av og til å gripe inn i de utenrikspolitiske avgjørelsene. [II] [På samme måte som Darré, den selverklærte parti-ideologen for jordbruksspørsmål, mislyktes Rosenberg i å profilere sitt kontor innenfor den flersidige, kompliserte maktstrukturen som var iferd med å ta form innenfor den nasjonalsosialistiske staten.[III]] [BBTN] ‘1. April 1933 was the foreign-political office of the NSDAP founded under leadership of Alfred Rosenberg, but it gained very little influence over the Foreign Office and Hitler.[I] The latter started from the end of 1933 occasionally to intervene into the foreign-political decisions.[II] [In the same way as Darré, the self-appointed party ideologist for agrarian questions, failed Rosenberg to distinguish his office within the multi-sided complicated power structure which was about to take form within the National Socialist State.[III]]’

In terms of discourse relations as defined in SDRT (Asher and Lascarides 2003), the RC would be an Elaboration of the matrix clause, which is the prototypical case of a subordinating discourse relation. This means that the matrix clause as well as the RC would be at the ‘right frontier’ of the discourse representation and thus could serve as attachment points for information in the following discourse. In the German version, syntactic subordination, here correlating with a subordinating discourse relation, helps the reader to infer the appropriate attachment point for the following sentence (i.e., the matrix clause), thus facilitating the processing of the complex information presented in this text fragment. In the Norwegian translation in (10b), however, syntactic subordination as a discourse structuring signal is lost, and the discourse structure after (10b[II]) seems to be more open as to where following sentences could attach, when contrasted to the German version. This leads to interpretation problems (garden path readings) regarding (10b[III]), i.e., one first tries to interpret the sentence as a continuation of (10b[II]) until one has to revise it upon encountering the subject ‘Rosenberg’. Placing ‘Rosenberg’ in sentence-initial position as in the German version would possibly facilitate the processing of this referent, but would not necessarily assist recognition of the

 Wiebke Ramm

hierarchical discourse structure. Thus, upgrading the RC does not block the right frontier after (10b[I]) semantically − the same subordinating discourse relation can in the end be inferred between (10b[I]) and (10b[II]). Nevertheless it causes difficulties for the identification of discourse referents in the subsequent context, making the Norwegian version appear less coherent than the German one, at least as far as reference structure is concerned. So this example can be seen as evidence of the validity of P2, that translational upgrading of discontinuative RCs in some way changes the interpretation of the texts.

3.4  Evaluation As illustrated by the examples discussed so far, the upgrading of a RC almost always changes the referential structure in the translation in some way. Often discourse referents have to be reintroduced or explicitated as a non-pronominal noun phrase as in (10b), or the upgrading may force the disambiguation of referents which are (deliberately) ambiguous in the source text, possibly leading to further interpretative problems as in (9b). Furthermore, the discourse-structuring signal to process the matrix clause and the RC together as one unit – also signalled by intonation if the text is read aloud – is lost. This may cause ‘more trouble at the right frontier’, i.e., more garden path readings of the context following the upgraded RC as in (10b), or it may lead to a change of the discourse relations between the clauses involved, as occurs in (7). Although it seems intuitively plausible that the continuative vs. discontinuative/ appositive discourse function of a non-restrictive RC should matter for its translation into Norwegian, the predictions following from this hypothesis have proved difficult to attest. This is probably due to the fact that the criteria for distinguishing between the two discourse functions were often inoperable for the complex examples in this type of text. One problem seems to be that approaches to discourse functions and discourse relations often concentrate on texts of the narrative type, where the event structure and its temporal development are decisive factors for the structuring and progression of discourse. At least this holds for Klein and v. Stutterheim’s (1987, 1991) quaestio approach and for Laux’s (2002) study of non-restrictive RCs. In fact, most of the classical examples illustrating the concept of coordinating and subordinating discourse relations in SDRT in Asher and Lascarides (2003), are narrative. Many text passages in the actual non-fictional corpus are, however, of non-narrative text types, where other criteria, such as the description of (abstract) objects and their properties and relations or the unfolding of argumentation, may guide the structuring of the discourse, as in (11a) below. (11) a. Unabhängig davon begannen im Herbst 1916 deutsche U-Boote in großem Stil damit, den Dampferverkehr nach Russland um das Nordkap herum anzugreifen,[I] wobei innerhalb von nur vier Monaten 143 norwegische Schiffe versenkt wurden, ohne dass die Entente militärischen Schutz bieten konnte.[II] [Diese brutale Form der Kriegsführung richtete sich nicht gegen

Upgrading of non-restrictive relative clauses in translation 

Norwegen, obwohl […], sondern stellte einen deutschen Test dafür dar, ob ein verschärfter U-Boot-Krieg gegen Handelsrouten der Entente Erfolge erzielen konnte.[III]] [BB] ‘Independently from this started in autumn 1916 German submarines in large scale [damit (‘with this’, pron. adv.)], the steamboat traffic to Russia around the Northcap to attack,[I] [wobei] within four months 143 Norwegian ships were sunk, without that the entente military protection to offer able was.[II] [This brutal form of warfare was aimed not against Norway, although […], but represented a German test for whether an intensified submarine war against trade routes of the entente success achieve could.[III]]’ b. Uavhengig av dette innledet høsten 1916 tyske ubåter omfattende angrep mot skipsfarten på Russland rundt Nordkapp.[I] I løpet av bare fire måneder ble 143 norske skip senket uten at ententen kunne tilby noen form for militær beskyttelse.[II] [Denne brutale krigføringen var ikke rettet mot Norge, selv om […], men var et tysk forsøk på å finne ut om en skjerpet ubåtkrig mot ententens handelsruter kunne lykkes.[III]] [BBTN] ‘Independently from this initiated autumn 1916 German submarines large-scale attacks against the shipping to Russia around the Northcap.[I] In the course of only four months were 143 Norwegian ships sunk, without that the entente could offer any form of military protection.[II] [This brutal warfare was not aimed against Norway, although […], but was a German attempt to find out whether an intensified submarine war against the entente’s trade routes could succeed.[III]]’

The wobei-clause expands (‘elaborates’) on the German attack of the steamboat traffic to Russia mentioned in the matrix clause, thus a subordinating discourse relation should be assigned to the two clauses, following Holler’s (2005) approach. Further, no Consequentiality relation can be identified according to Laux’s (2002) definition, neither does the following sentence allow a Consequentiality interpretation. Nevertheless it clearly relates to the clause-related RC (which cannot be omitted without the text becoming incoherent). The demonstrative subject NP diese brutale Form der Kriegsführung (‘this brutal form of warfare’) in (11a[III]) summarises the content of the preceding clause complex and creates an attachment point for the rest of this sentence which interprets the events described in the preceding clauses, (11a[I]) and (11a[II]), in this way pushing forward the argumentation of the text. In the Norwegian translation (11b), the wobeiclause is upgraded as an independent main clause without any explicit signalling of the discourse relation. However, the Elaboration relation can be inferred from the context without problem, and the interpretation of the Norwegian text diverges only slightly from the German version in that the sentence following the upgraded RC, (11b[III]), is interpreted as being related to (11b[II]) only, whereas the German original is more ambiguous as to whether this sentence relates to the whole clause complex, ((11a[I]) and (11a[II])), or to the RC in (11a[II]) only. What this illustrates is that the RC in this example seems in some way to be continuative, notwithstanding that Laux’s and ­Holler’s

 Wiebke Ramm

criteria of definition are left unsatisfied, which may be related to the criteria being tailored towards the typical discourse progression in narrative texts. If this observation is correct, namely that this RC would be continuative on a definition of continuativity adapted to a non-narrative text type, this example would also support the validity of the hypothesis presented in Section 3.1, viz. that translational upgrading of continuative RCs does not significantly affect discourse interpretation.

4.  Discourse organisation strategies and coherence The syntactic upgrading of German RCs in the Norwegian translations (for which in some cases there is no alternative) raises some interesting questions regarding the language-specificity of discourse organisation strategies. The loss of syntactic subordination renders the Norwegian translation less hierarchic (hypotactic) and more paratactic than the German original in that there is a shift towards more syntactic autonomy regarding the clause-linking pattern (Lehmann 1988: 189; see also Cosme in this volume on the characteristic shifts of clause-linking patterns in translations between Dutch, French and English). Splitting the SL clause complex into a sequence of independent sentences at the same time increases the incrementality of the text, a parameter of information packaging that accounts for the amount of new information introduced per sentence (Fabricius-Hansen 1999). According to Fabricius-Hansen, Norwegian translations of German scientific prose appear to follow the principles of incremental discourse organisation (PIDO) stated as follows: (1) minimise the number of new discourse referents introduced per sentence, (2) minimise the information stated per new or reactivated discourse referent in the sentences used to declare or reactivate them in the discourse, and (3) minimise “the use of presuppositions triggering expressions whose presuppositions are not justified by the preceding context” (Fabricius-Hansen 1999: 184). The discourse organisation of the German original versions, however, is instead characterised by what Doherty (2006: 60) terms the strategy of incremental parsimony (SIP). This implies that new information should be attached to an appropriate point of attachment in the sentence under construction, if possible. Thus, whereas for a text in Norwegian it is important not to ‘squeeze’ too much information into a single sentence, the guiding principle of German texts is to avoid starting a new sentence if a piece of information can be attached to the sentence under construction. According to Doherty (who investigates translations from English into German), the two discourse organisation strategies PIDO and SIP compete with each other and are regulated by Sperber and Wilson’s (1986) Principle of Relevance, aiming at an optimal balance between processing effort and cognitive effect (Doherty 2006: 60). The trade-off between the two strategies may differ from language to language, the availability of appropriate attachment points being one possible source of divergence – as pertaining to the non-restrictive RCs discussed in this paper.

Upgrading of non-restrictive relative clauses in translation 

The question is whether these discourse organisation strategies tell us anything about the coherence and/or comprehensibility of the respective texts. Is a typical ­hierarchically organised German text consisting of complex clauses, where a lot of complex information is expressed within the single sentence in a syntactically structured way, more coherent than a less hierarchically organised Norwegian version of the text? And does this mean anything for the comprehensibility of a text? The examples discussed so far could give the impression that the less hierarchical organisation of the Norwegian versions of the texts automatically makes them less coherent than the German original texts, as illustrated by the problems with keeping track of discourse referents and due to the greater effort required to integrate separate sentences into the incrementally constructed discourse representation. However, this is not always the case. The relative clause in (12a) below explains (‘elaborates’) an ‘action-reaction scheme’ (referring to the German strategy in World War II) mentioned in the matrix clause, which is a central topic of the whole paragraph. In this case the following sentence takes up the topic of derartige historische Interpretationen (‘such historical interpretations’), i.e., what is said in the relative clause and in the following contrastive während-adverbial clause. (12) a. Hubatsch betrachtet die Besetzung des Landes von 1940 ebenfalls innerhalb eines Aktions-Reaktions-Schemas,[I] bei dem das Deutsche Reich aus strategischen Gründen den Briten zuvorkommen musste, um die Rohstoffzufuhr aus Schweden sicherzustellen, während offensive seestrategische Gesichtspunkte für die Kriegsführung gegen England keine Rolle gespielt hätten. [II] [Erst in der Fischer-Kontroverse während der späten 60er Jahre wurden derartige historische Interpretationen in der Bundesrepublik außer Kraft gesetzt.][III] [BB] ‘Hubatsch looked at the occupation of the country of 1940 as well within an action-reaction scheme,[I] [bei dem (‘where’)] the German Reich (nom.) due to strategic reasons the British (dat.) would have to forestall, in order to secure the commodity supplies from Sweden, while offensive sea-strategic considerations played no role for the warfare against England. [II] [Not until the Fischer controversy in the late sixties were such historic interpretations in the Federal Republic suspended.][III]’ b. Hubatsch tolket også okkupasjonen av 1940 som en reaksjon.[I] I denne tolkningen måtte Tyskland av strategiske grunner komme britene i forkjøpet for å sikre råvareleveransene fra Sverige, mens offensive sjøstrategiske synspunkter ikke hadde hatt noen betydning for krigføringen mot England.[II] [Det var først under Fischer-debatten mot slutten av 1960-årene at slike historiske tolkninger ble satt ut av kraft i Forbundsrepublikken.] [III] [BBTN] ‘Hubatsch interpreted as well the occupation of 1940 as a reaction. [I] In this interpretation had Germany from strategic reasons to forestall the British in order to secure the commodity supplies from Sweden, while offensive sea-strategic considerations played no role for the warfare against

 Wiebke Ramm

England. [II] [It was only under the Fischer debate in the late sixties that such historic interpretations were suspended in the Federal Republic.] [III]’

In this example, the upgrading of the RC in the Norwegian translation (12b) does not lead to difficulties in interpretation. On the contrary, it seems that the nominal style of the original German version (which is not untypical of this genre) – the RC being syntactically only a NP modifier of Aktions-Reaktions-Schema – is quite ponderous: Aktions-Reaktions-Schema is not taken up as a referent in the following sequence of subordinate clauses (due to the nature of the RC construction chosen), until derartige historische Interpretationen refers back to it in the following sentence (in non-initial position). In the Norwegian translation, betrachtete […] innerhalb eines AktionsReaktions-Schemas (‘looked at […] within an action-reaction scheme’) of the German matrix clause is rendered as tolket […] som en reaksjon (‘interpreted […] as a reaction), a translation which is licensed by the previous context. The RC is upgraded from an informationally ‘heavy’ and complex NP modifier to an independent sentence by anaphorically linking (12b[II]) to (12b[I]) with the sentence-initial adverbial i denne tolkningen (‘in this interpretation’). Thus, the Norwegian version more explicitly keeps track of the discourse referent ‘interpretation as a reaction’ than the German original does and at same time is a smart translation solution for marking the indirect speech indicated by the conjunctive mood of hätte (‘would have’) in the German version. The example illustrates that the SIP strategy is not always applied appropriately or may be overused, especially in texts with high information density such as scientific articles.10 The tendency to use a nominal style attributed to German informationoriented texts may lead to incongruence with or inadequate perspectivisations of the states of affairs to be expressed, as in example (12a) where the bei dem-RC (structurally a NP modifier) is a linguistic realisation that is rather difficult to recover in proportion to the relative salience of its content. As to the question of whether a hierarchical vs. non-hierarchical organisation of discourse gives an indication of how coherent a text is, different factors seem to interact and compete with each other and are constrained by the choices available in the grammatical systems of the individual languages. One factor is the degree of ease in establishing and maintaining (accommodating) referential relations in a text. It can be argued (following Holler 2005: 211) that, with respect to this dimension of coherence, non-restrictive RCs such as those in the German SL texts are superior to their ­Norwegian translations by a corresponding sequence of independent sentences, ­because the syntactic link between the matrix clause and the RC restricts the search space for the antecedent of the anaphor (the relative marker) introducing the RC. The analysis of examples (9) and (10) may indicate that the search space for antecedents

.  One possible explanation for this is related to the space constraints authors often have to meet.

Upgrading of non-restrictive relative clauses in translation 

also plays a role for the resolution of anaphors in the sentence following a RC construction (as in the German original versions), as opposed to anaphors in the sentence following a sentence sequence (as in the Norwegian translations). Further evidence that syntactic subordination as a RC has an impact on the accessibility of possible antecedents for anaphora in the following context comes from research based on Centering Theory (Grosz et al. 1995). Miltsakaki (2005) analyses English and Greek non-restrictive RCs and investigates whether a pronoun taking up the topic (which typically means subject position in English) in the sentence following a clause complex containing a non-restrictive RC has its more likely antecedent in the matrix clause or the RC. The study has revealed that processing a clause complex containing a non-restrictive RC as a single unit together with the matrix clause in most cases leads to a more coherent interpretation11 of the discourse segment than processing the matrix clause and the RC as a separate discourse unit each. In the present study, the referential relations between entities in the matrix clause, the RC and the following sentence are, in many cases, more complex (i.e., not restricted to the topic only), therefore possibly not all of Miltsakaki’s results are comparable. ­Nevertheless, the present study also shows that syntactic subordination can operate to signal a downplay in the salience of a discourse unit, but this is not always the case, as the continuative RC examples demonstrate. In any case, subordination instructs the reader to process the subordinate clause as one unit together with the matrix clause. If a subordinate clause is (or must be) upgraded to a main clause in translation, then the possibility of structurally signalling its relative prominence may be lost. If in addition this upgraded clause is separated from the translation of the matrix clause by a full stop (and not by a semicolon or a comma, as in (6b) above), the signal to process the two clauses as one unit is also lost which may lead to a more ‘chopped up’ text, as in example (7b). Assessing the coherence of a text is not only a matter of referent accessibility but just as much a question of the recoverability of other types of content relations (conceptual relations and discourse relations). Complex hierarchical information packaging and high information density within a sentence do not imply that the information can be easily unpacked by the reader, as illustrated by example (12a). Conceptual and discourse relations are sometimes easier to retrieve when explicitly (lexically) signalled, but this usually makes a sentence longer and, at some point, the limits of what

.  In Centering Theory, coherence is measured in terms of the type of centering transition computed between the topic update units. Centering transitions describe the type of topic continuation vs. topic shift in a discourse, where topic continuation is counted as the most coherent transition (Miltsakaki 2005: 4). According to Miltsakaki (2005: 5), the question of whether each tensed (main or subordinate) clause should count as a topic update unit (an utterance), or the complex clause as a whole should be processed as a unit, was left open to empirical investigation in the original version of Centering Theory – which is what Miltsakaki does in her study.

 Wiebke Ramm

can be expressed within a single sentence are reached. Moving pieces of information into a new sentence, however, may cause new problems for keeping track of referents, as seen above.

5.  Summary and conclusions This paper has looked at the relation between subordination in syntax vs. discourse from a translation perspective, taking upgrading of non-restrictive German RCs to independent main clauses Norwegian – a frequent cause of sentence splitting in translations from German into Norwegian – as an example. Firstly, it has been shown that contrasts between the grammatical systems of the two languages in regard to the realisation of RCs are an important factor motivating this type of adjustment affecting the (syntactic) hierarchical structuring of the SL vs. TL text and frequently also its segmentation into sentences (discourse units separated by full stops). We have taken a closer look at the discourse functions of German non-restrictive RCs. On the basis of two approaches, viz. Laux (2002) and Holler (2005), we made some predictions regarding the potential effects of translational upgrading of the different types of non-restrictive RCs upon the interpretation of the TL text. The two accounts consider the temporal interpretation and the discourse relations holding between the matrix clause and the RC, and Laux further takes into account the sentence following the clause complex containing the RC. The main hypothesis was that continuative RCs (weiterführende Relativsätze) would be less problematic when upgraded as independent main clauses than discontinuative/appositive RCs because they are more similar to main clauses than discontinuative/appositive RCs are. Continuative RCs are non-restrictive RCs where, according to Holler’s (2005) approach, a symmetric/coordinating discourse relation (e.g., Narration) can be identified between the matrix clause and the RC (applying the SDRT model of Asher & Lascarides 2003). Alternatively, following Laux’ (2002) approach, a Consequentiality relation (as defined in Sandström 1993) holds between the matrix clause and the RC as well as between the sentence containing the RC and the following context. However, the hypothesis − although intuitively plausible − proved difficult to test on the ‘wildlife’ data of two German-Norwegian parallel corpora, one reason being the difficulties in determining the continuative vs. discontinuative/appositive nature of the RC in several of the (non-narrative) text examples. As a possible explanation for these difficulties, it was suggested that the criteria delineating the (dis)continuativity of a RC are based on the typical discourse progression found in narrative texts, and that the application of such criteria is not entirely appropriate for non-narrative text passages, such as those in many of the corpus examples. With modified (i.e., text-type adjusted) criteria for the definition of discourse continuity, the validity of the hypothesis seems to be better supported by the corpus data.

Upgrading of non-restrictive relative clauses in translation 

Translational upgrading of RCs changes the clause linking pattern from hypotaxis (clause subordination) to parataxis (clause coordination), cf. Lehmann (1988), and in many cases – when an upgraded clause is separated from the preceding sentence by full stop – also the incrementality of the text, i.e., it reduces the amount of new information introduced or accommodated per sentence. Language-systematic contrasts concerning the realisation of RCs can thus contribute to the preference of Norwegian information-oriented texts (or at least of texts translated from German) to be organised according to the PIDO principles of incremental discourse organisation (Fabricius-Hansen 1999), whereas it is rather SIP, the strategy of incremental parsimony (Doherty 2006), which is followed in the German texts. The two discourse organisation strategies potentially impact upon the establishment and maintenance of discourse coherence in texts. Nominal referents for example may be linked anaphorically within or across sentence boundaries, a difference which influences their recoverability. Inherent hierarchical or dependency relations between concepts may be expressed by means of the sentence-internal syntactic structure (by adjunction or syntactic dependency), they may be signalled lexically (e.g., by connectives), or they may be contextually inferable between clauses or sentences. Finally, the communicative weight (discourse prominence) that an utterance is intended to have, may be indicated by sentence-internal structure (e.g., by syntactic subordination) or by lexical means (across sentence boundaries). Viewing coherence as a multi-dimensional phenomenon (cf. Stede in this volume for a similar view), and taking into account the language-specific options for the realisation of the different dimensions, provides evidence for Doherty’s (2006: 60) claim that the trade-off between the two discourse organisation strategies is language-specific. This implies that coherence is a concept that is best measured against language-specific options and preferences for discourse structuring within and across sentence boundaries. What this means for text comprehensibility and text comprehension, however, is much more opaque. The methodology used in this paper is not suitable to assess whether languagespecific preferences for hierarchical vs. non-hierarchical discourse organisation have any implications for how texts are actually processed and understood. Whether (from the perspective of native speakers) a hierarchically structured German text could be processed and understood with equal ease as a less hierarchically structured Norwegian text is only possible to test within the framework of a psycholinguistic methodology.

References Asher, Nicholas. 1993. Reference to Abstract Objects in Discourse [Studies in Linguistics and Philosophy 50]. Dordrecht: Kluwer. Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation [Studies in Natural Language Processing]. Cambridge: CUP.

 Wiebke Ramm Asher, Nicholas & Vieu, Laure. 2005. Subordinating and coordinating discourse relations. Lingua 115: 591–610. Brandt, Margareta. 1990. Weiterführende Nebensätze: Zu ihrer Syntax, Semantik und Pragmatik [Lunder germanistische Forschungen 57]. Stockholm: Almqvist & Wiksell. Doherty, Monika. 2006. Structural Propensities: Translating Nominal Word Groups from English into German [Benjamins Translation Library 65]. Amsterdam: John Benjamins. Faarlund, Jan Terje, Lie, Svein & Vannebo, Kjell Ivar. 1997. Norsk referansegrammatikk. Oslo: Universitetsforlaget. Fabricius-Hansen, Cathrine. 1992. Subordination. In Deutsche Syntax: Ansichten und Aussichten, Ludger Hoffmann (ed.), 458–483. Berlin: de Gruyter. Fabricius-Hansen, Cathrine. 1999. Information packaging and translation: Aspects of translational sentence splitting (German – English/Norwegian). In Sprachspezifische Aspekte der Informationsverteilung, Monika Doherty (ed.), 175–213. Berlin: Akademie. Fabricius-Hansen, Cathrine. 2000. Übersetzen mit Stil - ein unmögliches Ziel? In Übertragung, Annäherung, Angleichung: Sieben Beiträge zu Theorie und Praxix des Übersetzens [Osloer Beiträge zur Germanistik 25], Cathrine Fabricius-Hansen & Johannes Østbø (eds), 65–95. Frankfurt: Peter Lang. Fabricius-Hansen, Cathrine. 2005. Elusive connectives. A case study on the explicitness dimension of discourse coherence. Linguistics 43(1): 17–48. Fabricius-Hansen, Cathrine, Ramm, Wiebke, Solfjeld, Kåre & Behrens, Bergljot. 2005. Coordination, discourse relations, and information packaging: Cross-linguistic differences. In Proceedings of the Symposium on the Exploration and Modelling of Meaning (SEM-05), Biarritz, France, November 14–15 2005, Michel Aurnague, Myriam Bras, Anne Le Draoulec & Laure Vieu (eds), 85–93. Grosz, Barbara, Joshi, Aravind & Weinstein, Scott. 1995. Centering: A framework for modelling local coherence in discourse. Computational Linguistics 21(2): 203–225. Holler, Anke. 2005. Weiterführende Relativsätze: Empirische und theoretische Aspekte [Studia Grammatica 60]. Berlin: Akademie. Klein, Wolfgang & von Stutterheim, Christiane. 1987. Quaestio und referentielle Bewegung in Erzählungen. Linguistische Berichte 109: 163–183. Klein, Wolfgang & von Stutterheim, Christiane. 1991. Text structure and referential movement. Sprache und Pragmatik 22: 1–32. Laux, Britt Dalen. 2002. Die nicht-restriktive Relativsatzkonstruktion im Deutschen: Diskursfunktion und temporale Interpretation. PhD dissertation, NTNU Trondheim. Lehmann, Christian. 1988. Towards a typology of clause linkage. In Clause Combining in Grammar and Discourse, John Haiman & Sandra A. Thompson (eds), 181–225. Amsterdam: John Benjamins. Mann, William C. & Thompson, Sandra A. 1988. Rhetorical structure theory: Toward a functional theory of text organization. Text 8: 243–281. Matthiessen, Christian & Thompson, Sandra A. 1988. The structure of discourse and ‘subordination’. In Clause Combining in Grammar and Discourse, John Haiman & Sandra A. Thompson (eds), 275–329. Amsterdam: John Benjamins. Miltsakaki, Eleni. 2005. A centering analysis of relative clauses in English and Greek. In Proceedings of the 28th Penn Linguistics Colloquium, University of Pennsylvania, Philadelphia. Also available as U. Penn Working Papers in Linguistics 11.1: 1–15. URL: http://www.cis.upenn. edu/~elenimi/CenteringAnalysisOfRelativeClauses.pdf.

Upgrading of non-restrictive relative clauses in translation  Peyer, Ann. 1997. Satzverknüpfung – syntaktische und textpragmatische Aspekte [Reihe Germanistische Linguistik 178]. Tübingen: Niemeyer. Ramm, Wiebke. Forthcoming. Discourse-structural salience from a cross-linguistic perspective: Coordination and its contribution to discourse (structure). In Salience. Multidisciplinary Perspectives on its Function in Discourse [Trends in Linguistics. Studies and Monographs (TiLSM)], Michael Grabski et al. (eds). Berlin: Mouton de Gruyter. Ramm, Wiebke. In preparation. Satzgrenzenveränderungen in der Übersetzung: Zum Zusammenhang von Informationsportionierung und Diskursstruktur am Beispiel der Sprachen Norwegisch und Deutsch. PhD dissertation, University of Oslo. Ramm, Wiebke & Fabricius-Hansen, Cathrine. 2005. Coordination and discourse-structural salience from a cross-linguistic perspective. In Salience in Discourse: Multidisciplinary Approaches to Discourse 2005, Manfred Stede, Christian Chiarcos, Michael Grabski & Luuk Lagerwerf (eds). Münster: Stichting/Nodus, 119–128. Also available as SPRIKreport Nr. 30. URL:http:// www.hf.uio.no/forskningsprosjekter/sprik/docs/pdf/wr/WR-CFH-Report30.pdf. Sandström, Görel. 1993. When-clauses and the Temporal Interpretation of Narrative Discourse. PhD dissertation, University of Umeå. Sperber, Dan & Wilson, Deirdre. 1986. Relevance: Communication and Cognition. Oxford: Blackwell.

Corpus texts referred to in the examples [BB]: Barth, Boris 1999. Norwegen und der Norden in der deutschen Strategie 1914 bis 1945. In Deutschland – Norwegen. Die lange Geschichte, Jarle Simensen (ed.), 165–186. Oslo: Tano Aschehoug. [BBTN] Translation: Hobson, Rolf. Norge og Norden i tysk strategi, 1914–1945. In Tyskland – Norge. Den lange historien, Jarle Simensen (ed.), 145–163. Oslo: Tano Aschehoug. [EC]: Canetti, Elias 1977. Die gerettete Zunge. München: Carl Hanser. [ECTN] Translation: Qvale, Per 1982. Den reddede tungen. Oslo: Gyldendal.

Subordination in narratives and macro-structural planning A comparative point of view Mary Carroll, Antje Rossdeutscher, Monique Lambert and Christiane von Stutterheim

University of Heidelberg/ University of Stuttgart/ University of Paris VIII/ University of Heidelberg Based on a series of cross-linguistic studies covering Germanic (Dutch, English, German) as well as Romance languages (French, Italian, Spanish), the present paper addresses questions concerning the nature of the decisions made at the level of macro-structural planning when producing a narrative. Focusing on English, German, and French, it presents evidence for a hierarchy of factors and associated constraints that are both grammatically based and perspective driven. The comparison shows how some of the constraints at issue lead to crosslinguistic differences in the way subordination as a downgrading operation is implemented in the narratives. Keywords: Macro-structural planning principles, decision hierarchy, downgrading operations, crosslinguistic study.

1.  Introduction The analyses presented in this paper are based on retellings of a silent animation film (Quest) where speakers were asked to tell ‘what happened’ (20 per group). The type of information organisation required in carrying out a task of this kind can be described at different levels of analysis. There are questions involving information selection (deciding what to say), thematic continuity (e.g., topic assignment), referential framing, which relates to predicate-argument structures and how they are anchored with respect to times, worlds, and spaces. In order for a sequence of propositions to be coherent, these referential properties have to be related in consistent terms across utterances. The question is: are the issues solved for each sentence anew or are there macro-structural principles that guide the speaker at each relevant stage in the narrative? For example, is the decision to map an agent of an action as subject of a main or subordinate clause made individually at each point in

 Mary Carroll et al.

the narrative, or are there principles that decide the issue on a default basis for the narrative sequence as a whole? As mentioned above, the present paper is based on cross-linguistic findings that show how the types of decisions that speakers preferably make correlate with grammaticised means in the respective languages. The present objective is to show how factors guiding decision making in information structure are hierarchically organised at the level of macro-structural planning. Given the assumption that sequencing principles form the underlying conceptual structure in a narrative task (‘tell what happened’), the narratives are grouped in the following according to the underlying temporal frames of reference and associated perspectives. Although the present focus is placed on English, French, and German, results are included in some cases for Dutch, Spanish, and Italian in order to illustrate the overall typological trend (cf. von Stutterheim, Carroll & Klein 2003; Carroll & Lambert 2003, 2006; Rossdeutscher & von Stutterheim 2006). The following section presents a summary of the relevant findings, while section 3 gives an overview of the associated quantitative analyses on which these are based; section 4 shows how patterns in information structure can be traced to a set of grammaticallydriven principles that guide decisions at the level of macro-structural planning in narrative tasks and is followed by the first step towards a formal model of possible hierarchical orders in section 5.

2.  The narrative task Speakers were asked to retell the content of the silent film Quest which lasts approx. 9 minutes. It portrays the adventures of a clay figure, the only animate-like protagonist in the film, who tries to find his way in a hostile world ‘peopled’ by natural forces (high winds, flying rocks, huge sheets of paper) – which in some way or another get in his way in his quest for salvation. For the analysis of information structure the constellation is simple. There is one main character with a well defined status (animate-like) with respect to the other major players (inanimate), so that mapping patterns in information structure can be compared accordingly. Although the speakers were asked to tell ‘what happened’, the present tense was selected spontaneously across all the languages studied (20 speakers per group).

2.1  Temporal frame with a protagonist-based perspective Information structure for the narrative sequence may centre on a temporal frame of reference that is organised on the basis of temporal shift. This constitutes the preferred option in the German data and is marked by the following factors: i. The main character is accorded a higher status compared to other agents. It is eligible for mention as the subject of a main clause. The other agents (inanimate entities

Subordination in narratives and macro-structural planning 

such as gusts of wind, falling rocks, etc.) are downgraded in that selection for mention is restricted. Furthermore, when inanimate entities compete with the protagonist for mention as subject of the clause (inanimate entity is agent, for example), they are typically mapped into a passive construction as the logical or underlying subject. This ensures maintenance of the status of the main character at the level of thematic continuity (see line 016 in example 1 below). In this sense the main character/protagonist is assigned ‘topic’ status. ii. In reference maintenance there is additional evidence for informational status accorded to the protagonist since the means used involve zero anaphora (cf. lines 16, 17, and 19, 20 in example 1 below). iii. In the narrative sequence, the main structure of the text, events are presented as constituting a change in state. This is typically brought about by an animate agent who acts to bring about the change in state. The temporal relation is anaphoric in that events are construed as having reached a point of completion (pc). Linkage is based on the relation what happens after pc, whereby the post time of the preceding event is taken as the topic time of the subsequent event. This is typically expressed by the temporal shifter dann, ‘then’ (cf. in detail von Stutterheim, Carroll, Klein 2003; Rossdeutscher & von Stutterheim 2006). For German speakers, a protagonist-based perspective and temporal shift form the relevant criterion in organising narrative progression. Before looking at the implications with regard to decisions at the level of macro-structural planning, we will take a closer look at examples of this pattern in information structure. Use of the passive to ensure thematic continuity is illustrated in example 1 below (line 16). It occurs when another agent comes into competition with the protagonist for mention as subject of a main clause, as mentioned. The rate of occurrence for the passive in this kind of context is high at 72.7 per cent (Murcia Serra 2001). (Ex.1) de21 [12] das Männchen fällt runter in eine Papierwüste mit lauter einzelnen quadratischen Blättern the little-man falls down into a paper-desert with many single square sheets [13] und in diesem Raum weht ein starker Wind and in this place blows a heavy wind [14] und wirbelt die Papierblätter durch die Gegend and whirls the paper-sheets through the area [15] und das Männchen muss dauernd ausweichen and the man must constantly avoid [16] und wird dann schließlich von einem Blatt getroffen and gets then finally by one sheet hit

 Mary Carroll et al.

[17] [18] [19] [20]

und fällt zu Boden and falls to ground dann steht es wieder auf then gets it again up und findet eine nasse Stelle im Papierboden and finds a wet place in the paper-floor fängt an nach dem Wasser zu graben begins after the water to dig

Dutch retellings show similar principles with regard both to perspective taking and the temporal frame (same task). In other words, similar preferences in information structure were observed in the two languages which share the ‘verb second’ constraint for word order. In these languages the finite verb constitutes the second main constituent in main clauses. Significantly, the verb can be preceded by only one main constituent, and this need not be the subject. The ‘slots’ created by the verb second constraint are referred to as the ‘Vorfeld’ and ‘Mittelfeld’. Both are relevant in topic assignment (Frey 2000, 2005). We assume that there is an interdependency between the constituents that may typically occur in the ‘Vorfeld’ in V2 languages in the given task. For narratives (answering the quaestio ‘what happened’) these are linguistic means that encode temporal relations (such as dann, (‘then’), as well as the syntactic subject. Evidence for an interdependency lies in the fact that in information selection in both Dutch and German (deciding what to say), entities and events are filtered with respect to their ability to accommodate temporal relations such as shift. Events that do not accommodate this relation are less likely to be selected for mention (cf. in detail Carroll & Lambert 2003). In the narrative task the protagonist is accorded ‘topic’ status on a global basis for the task as a whole, since an intentional agent is more likely to be involved in events that reach a goal or point of completion, thus accommodating temporal shift (typically expressed by dann (‘then’)). The next example is taken from the English data and illustrates a temporal frame based on anaphoric shift, and a ‘protagonist-based’ perspective with comparable use of the passive in downgrading agents involving natural forces (cf. line 11). In contrast to German, however, this is not the preferred pattern in information structure in retellings in English, as will be illustrated below. (Ex.2) ee03 [9] ok the man arrives in a paper world [10] and eh everywhere is covered with paper [11] and he gets hit by the flying piece of paper [12] and then he walks / he hustles around [13] and he finds a damp piece of paper [14] and he pushes the paper [15] and then he falls through the paper

Although information organisation is similar at the temporal level, there are marked differences compared to the German texts. In the English retellings maintenance of the protagonist as subject across adjacent clauses does not constitute a criterion for zero anaphora (VP-coordination). Pronouns are used and zero anaphora is restricted to events that are very closely related in causal terms, regardless of the type of agent.

Subordination in narratives and macro-structural planning 

This was also found in the larger set of retellings in English (see below) as well as in French retellings (cf. Carroll & Lambert 2006 on causality and the contexts in which zero anaphora is licensed in English and French). In German, by contrast, zero anaphora predominates and is closely bound to the event sequence and the temporal means used to establish it (this will be taken up again below). While the protagonistbased perspective is representative of a large group of coherent German narratives, the English example belongs to a small set (3/20 speakers).

2.2  Temporal frame deictic with narrator-perspective For the majority of English speakers (17/20) the relevant indicators in information structure can be summarised as follows: iv. All agents, both animate and inanimate, are selected for mention and accorded the same status in that they can be mapped as the subject of a main clause; use of downgrading procedures for this category is thus very low, compared to German (see below). v. In the temporal frame events are linked to a topic time which is externally anchored and deictic (‘now you see’; or ‘then you see’). All types of situations can be hooked up to this temporal anchor, both bounded and unbounded, and ongoing events (expressed by the progressive ‘–ing’) may form an integral part of the sequence; there are strict constraints in switching from potentially bounded events (simple tense) to ongoing events within an event sequence; significantly, the temporal relation expressed by ‘then’, where this occurs, relates to the left boundary of the preceding event, and not to the right boundary; it entails precedence but not necessarily completion of the prior event; this principle is crucial since it facilitates the integration of ongoing events into the story line (cf. Carroll & Lambert 2006). vi. Causal relations (‘so’, ‘because’) predominate in advancing the story line. vii. The means used in reference maintenance to the main character are pronouns. Zero anaphora occurs when the core relation in advancing the story line (causal) holds between the events in question (as in French retellings). In this sense use of zero anaphora in information structure may be viewed as linked to the type of relation at issue as much as to the entity to which reference is maintained. In information organisation in German topic assignment and anaphoric shift coincide, while in English and French causal relations and zero anaphora coincide. Starting with narratives with the deictic anchor ‘then you see’, line 44 gives an example of natural forces acting as the agent of an action. In contrast to the protagonistbased perspective, it is mapped into a main clause. (Ex.3) ee13 [34] then you see the sand man [35] falling from the sky onto this surface [36] and ­apparently falling through another world from the world of sand [37] and he gets up

 Mary Carroll et al.

[38] and he walks around [39] trying to figure out where he is [40] and it’s windy [41] and those papers are blowing [42] there is a little sort of tornado going by him [43] as he is walking [44] a piece of paper flies in his face [45] and he falls down [46] and he gets up [47] and he takes the paper away from his face [48] and he / in front of him sees water drip onto the surface onto the part of the surface [49] which is wet [50] and he goes over to it [52] and he kneels / gets down on his hands and his knees in front of it [53] and he feels the paper [54] he feels that it’s wet [55] so he reaches his hands up to wait for the next drop of water [58] so where its wet on the paper surface [59] he just starts to dig [60] like he was digging in the sand [61] and he only digs for a little while [62] and because the paper is wet [63] and he ends up falling through the paper.

Occurrence of natural forces as subject of a main clause is the predominant pattern (61.1%) – when these forces act as agents and compete with the protagonist for mention as subject. In the remaining cases a passive construction or subordination is used. The following example illustrates use of the deictic anchor ‘now you see’. As in the first example for English above, zero anaphora is confined to events that are closely linked, as in 48/49. (Ex.4) ee02 [34] okay eh now you see [35] that the man has fallen onto another like really bleak landscape [36] and it looks [37] as if it as all made out of paper [38] and there are like sheets of paper [39] lying on the floor [40] and he falls down onto this ground [41] and eh gradually he stands up again [42] and he can hear a wind [43] blowing a little bit [44] and these pieces of paper keep flying past him [45] and he can still hear the sound of eh water [46] falling [47] and dropping onto the ground [48] so he gets up [49] and goes towards the sound [50] and as he does this [51] you see like [52] it’s a paper [53] flying past him [54] and they’re quite big [55] they’re like the size of him [56] so one knocks him over.

In contrast to protagonist-based narratives where temporal shift is defined over events with right boundaries, temporal sequencing, in this strict sense, is often underspecified. The relation between one event and the next is supported by causal relations (so, as in lines 48 and 56). In contrast to the use of dann (‘then’) in German, the temporal shifter then relates to the precedence of the prior event (its left boundary), and does not entail that this event has reached a point of completion before the next one begins, as mentioned above. This factor, along with the presence of the deictic anchor ‘you see’, facilitates the integration of unbounded ongoing events at any point in the story line. Transitions from a bounded to an ongoing event are mediated by reference to a state or by the narrator (‘you see’): a pile of stones pushes him up out of the ground and you see him looking around; unmediated sequences where a bounded event is immediately followed by an ongoing event are not licensed; compare – x pushes him up, he is looking around to the example above (cf. in detail Carroll & Lambert 2006). The narrator as deictic anchor ensures adequate integration of ongoing events into the sequence, as in lines 49, 50, 51, and 52 in the example above,

Subordination in narratives and macro-structural planning 

where a transition of this kind is mediated by ‘you see’, combined with the marking of simultaneity in (50).

2.3  Protagonist- and narrator-based perspective On the whole, speakers of French follow a consistent frame of reference which can be described as incorporating both dimensions to form a specific cluster. English and French share the feature ‘subject prominent’ in information structure: i. All agents, both animate and inanimate are selected for mention and mapped as subject of a clause, as in English. ii. However, events involving natural forces are downgraded and related entities are mapped as subject of a subordinate clause when advancing the story line; this means that non-intentional agents are downgraded, in contrast to English. iii. The change in status accorded to different event types (switch from main clause to a subordinate clause) leads to disruptions in the temporal chain of bounded events given in main clauses, since anaphoric linkage, which relates to the right boundary of an event in an adjacent clause, is interrupted. iv. Predicates relating to intentions, attitudes and perception are accorded a predominant status in that they allow for the introduction of a causal chain, thus bridging events that differ in informational status. v. Events are also linked externally as components of an interpretative frame; although interpretation occurs in all languages, a ‘narrator perspective’ is present to a significantly high degree in French retellings. Ex. 5 illustrates a typical sequence where many events are unbounded, showing how this temporal property is often underspecified. Lines 63 to 68 express situations of perception or reflection by the protagonist. Furthermore the text illustrates the relative predominance of causal relations over temporal relations (donc, (‘so’) i.e., causal) as stated above. (Ex. 5) ef08 [61] il s’aperçoit he realises [62] que c’est mouillé partout that it is wet all around [63] donc là aussi il est un petit peu perplexe, on peut dire so again he is a little bit astonished, one might say [64] d’abord il essaie de récupérer l’eau par terre first he tries to collect the water from the ground [65] il s’aperçoit he realises [66] que c’est pas possible that this is not possible

 Mary Carroll et al.

[67] [68] [69] [70] [71] [72] [73] [74]

donc il essaie plutôt de l’attraper vers le haut so he rather tries to catch it from above comme s’il est en attente d’une offrande ou quelque chose comme ça as if he were waiting for an offering or something like that et il essaie à nouveau de retrouver cette goutte d’eau and he tries again to find this drop of water et il s’engouffre à nouveau and he is engulfed again oui, il creuse en fait yes, he digs basically pour essayer de trouver in order to find out où va cette eau where this water goes to et il s’enfonce à nouveau dans un gouffre and he sinks into the abyss again

The following example again illustrates this overall pattern, showing how events with non-intentional agents are mapped into subordinate clauses, although the events in question form part of the story line (e.g., 62, 63, 64). (Ex.6) ef02 [57] et c’est à ce moment là qu’il trempe ses mains and in this moment he immerses his hands [58] puisqu’il s’est dit que because he said to himself that [59] comme il ne pleut il ne pleuvait pas since it is not raining/was not raining [60] l’humidité ne pouvait venir que de ce sol the wetness could not come from somewhere else than the ground [61] et donc il trempe ses mains dans cette feuille humide and so he immerses his hands into the wet paper [62] qui se froisse which crumbles [63] et se déchire and tears [64] et l’emporte en profondeur and takes him into the depths

The analysis also shows that, although both dimensions are present, speakers may differ with respect to the relative weight attributed to the two worlds (narrator, protagonist). Some speakers refer to the narrator on a marginal level in introductions (6 out of 18), taking the protagonist as the global topic, while 12 out of 18 speakers rely on an interpretative frame, with inclusion of the narrator as witness on a consistent basis. In addition to the narrator as an interpreting subject and prominent source for implicit or

Subordination in narratives and macro-structural planning 

explicit causal relations in the latter group, the main character (the clay figure) can take on this role as well, given its salient characteristics as a human-like being. This gives a ‘two-layered’ form of macro-structural organisation with both the narrator as well as the protagonist as interpreting subject, both with the same potential in establishing coherence between events. In the following we will give a brief quantitative overview over the relevant features in the French data with respect to expression of attitude or stance (il est perplexe; ‘he is perplexed’; il s’aperçoit, ‘he realises’) and the type of relation used in linking states of affairs (causal, temporal): In French, verbs expressing attitude or stance occur more frequently, as compared to German (French 40.2%; German 18.1%). Causal relations (donc (‘so’), for example) predominate in French (12.6%), in contrast to temporal shift (3.8%) as expressed by et puis (‘then’). In German, by contrast, temporal shift is the most frequent means expressing linkage (28.2%), compared to causal relations (3.8%), as mentioned above.

3.  D  owngrading operations in French, German, and English: Quantitative overview This section provides a quantitative overview of the results presented above. We have illustrated how the use of downgrading, in the form of subordinate clauses and passive constructions, is driven by the temporal frame and thematic continuity. In French and English it was shown how all agents are eligible for mention as subject of a clause. They are downgraded in French by means of subordination (subject of a subordinate clause), even though the events in question serve to advance the story line. Downgrading does not occur on a systematic basis in English, reflecting differences that can be linked to the grammaticised means available to express temporal relations. The following figures on subordination in the narratives, covering both perspectives, reflect the scale on which this form of downgrading is implemented in the three languages. As indicated above, overall occurrence is highest in French: English 742/3886 utterances (19.1%); German 366/2740 utterances (13.3%); French 850/2297utterances (37.0%). The preferences shown do not constitute an isolated phenomenon but fall in line with other decisions made in information structure. In the following summary we will indicate how they form part of a set of principles, starting with information selection. 3.1 

Downgrading at the level of information selection

In German, where the story line is advanced by means of the temporal relation based on anaphoric shift (relating to events with a right boundary), speakers do not mention natural forces to the same extent, as illustrated above. The following overview illustrates differences in information selection across the different languages. In order

 Mary Carroll et al.

to illustrate the overall preferences, it takes into consideration another member of the V2 group (Dutch), as well as other members of languages that are ‘subject prominent’ in information structure (Italian and Spanish). The following figures cover mentions in all clause types. References to inanimate entities (natural forces) as agent or experiencer in an event (e.g., sheets of paper are swirling around; rocks push up out of the ground; a sheet of paper knocks him over), amount to 34.5% in English, 35.8% in French, 37.1% in Italian, 30.9% in Spanish. But there is a significant drop in both German 24.5% and Dutch 23.6%. The difference between speakers of German and Dutch and the other languages is statistically significant: t-test English – German p = 0.006 highly significant; Spanish – German p = 0.04 significant; while English – Spanish p = 0.25 is not significant; English – French p = 0.21 is also not significant. These figures cover mentions as subject of both main and subordinate clauses and are based on average values, in percent, for 20 speakers per group (cf. in detail Carroll & Lambert 2003). So speakers of German and Dutch not only downgrade by use of the passive, they also ‘defocus’ at a higher level, so to speak, in that in many cases some entities and events are not selected for mention at all. The criterion for selection is relevance for the protagonist, as will be shown in the examples below. By contrast, in English and French speakers mention minor events, even when they are not of direct relevance for the protagonist – events that pose a threat but do not actually affect the protagonist. The first example illustrates the relevance principle and speakers of all the languages studied mention the following situation in which a natural force is the agent of the action. English: It’s very windy with a lot of paper and a very big sheet hits him in the face and knocks him down: Mention of this event (given in percent) amounts to 90.0 in English and 60.0 in German. Mention in German is markedly lower however, if there are no immediate consequences for the protagonist, as when the rock in the next event misses him, for example: English: The first thing we see is a rock dropping from the sky directly towards his head. Mention in English is high at 80.0, compared to 40.0 in German. Another example is given with the event where a saw slightly damages the clay man’s foot . English: As he was lying there this saw came right by and slightly damaged part of his toe. The rate in this case is 70.0 in English compared to 30.0 in German. Relevance with respect to the protagonist is not at issue in information selection in English (or French and Spanish), in contrast to German.

3.2  Mapping inanimate entities as subject of a main clause or not The next set of figures gives the extent to which events involving natural forces are mapped into a main clause within the narrative sequence in the different languages. Again, figures for Italian, Spanish, as well as Dutch are included in order to underline overall trends. Mention of natural forces as subject of a main clause are as follows: English 26.0, Spanish

Subordination in narratives and macro-structural planning 

24.1, while the values in French and German are lower with French 8.0, German 14.0. The figures for German and French reflect downgrading by means of the passive, subordination, or omission (cf. Carroll & Lambert 2003). (The figures are percentages and references to the remaining option, the protagonist or the narrator amount to: English 74.0, French 92.0, German 86.0.

3.3  Reference introduction Focusing and downgrading patterns can be observed with the means used in the introduction of natural forces. In English where natural forces are eligible for mention as subject of a main clause throughout the narrative, this status correlates with the way in which they are typically introduced. The referents are introduced as the main content of a clause. In other words, there is a clause which serves to introduce the entity at issue (presentationals). This contrasts with the second option in which referents are embedded in clauses involving actions of the protagonist. Taking, for example, the first mention of new entities in the case of ‘sheets of paper’, the preferred means in English are as follows: (Ex. 7) There are sheets of paper swirling around (main content of clause)

This contrasts with German (as well as French) where first mentions are more frequently embedded: (Ex. 8) Das Männchen fällt nun auf eine neue Ebene (…), die mit Blättern (…) also keine Baumblätter sondern Schreibblätter große Blätter überdeckt ist. The little-man falls now onto a new level (…), which with leaves (…), not treeleaves but sheets of paper big sheets covered is. (Ex. 9) Dieses Wesen scheint in die nächste Ebene gefallen zu sein, wo da der

Fußboden aus lauter einzelnen Papierstücken besteht.

This creature seems into the next level to have fallen, where the floor out of lots of single sheets of paper consists.

As illustrated in the examples, the entities at issue are often introduced in German narratives as a property of the place where the protagonist lands. Overall figures for these options are as follows: In English, introductions as the main content of a clause, given in percentages, is 83.1 while the embedded form amounts to 16.9; (Spanish: main content 70.9, embedded 29.1). In contrast to English, these options in reference introduction occur with comparable frequency in French and German: In French, introductions as main content of the clause amount to 54.6, while the embedded form is 45.4; in German, introduction as the main content of the clause accounts for 58.6 of the cases and the embedded form for 41.4. (cf. Carroll & Lambert 2003). Up to now we have isolated some of the main downgrading operations, showing how they are implemented in consistent terms in the different languages.

 Mary Carroll et al.

3.4  N  arrator-perspective and downgrading at the level of the narrative as a whole With this perspective, speakers explicitly refer to the narrator’s world by means of then you see (example 3 above, line 34), or now you see. The events in the story world referred to in lines 35 and 36 in ex. 3 above are mapped into dependent clauses. The subsequent story world descriptions, as a whole, are subordinate in the sense of modal subordination, where all events in the narrative are presented as witnessed (cf. Roberts 1989). Subordination can take the form of a complement clause introduced by that. This can be compared with other possible patterns of subordination at this level, taking, for example, the following retelling of a German speaker, one of the very few who produces a series of narrative descriptions with syntactic subordinate clauses, i.e., relative clauses. The pattern is one of the few in German in which the perspective is narrator based. (Ex.10) (de013) [1] am Anfang sieht man so ne Art Pappmachemännchen im Sand at-the beginning see you a kind of paper-man in the sand [2] liegen vor ‘ner Flasche lying in-front-of a bottle [3] das dann aufwacht who then wakes up [4] sich anfängt zu bewegen starts to move [5] die Flasche in die Hand nimmt the bottle in the hand takes

This is found in English as well, though relative clauses are not used to the same extent. All speakers starting out in this way end up sooner or later with modal subordination. In theory one could have narratives where descriptions of the narrator’s world are downgraded or subordinated with respect to the story line, as in und er hebt den Arm, den man vorher nicht gesehen hat (‘he raises an arm which one did not see before’). There is no evidence, however, in the present database of a pattern of this kind, at least not on a systematic basis. In sum, one of the main types of subordination in the English narratives arises from the status accorded to the narrator. In French subordination of this kind is also found for the narrative as a whole, in addition to its use in downgrading certain types of events when advancing the story line. 4. 

 vidence for macro-structural principles underlying E information s­ tructure

The cross-linguistic comparison between English, French, and German allows isolation of a number of core parameters that are relevant for macro-structural planning.

Subordination in narratives and macro-structural planning 

Starting with the role of temporal-aspectual means, English and French are similar typologically in that they are subject prominent in information structure (and both have fixed word order), they differ, however, with respect to the grammatical means available to encode aspectual relations (present tense). Similarities in information structure can be observed in that all agents, whether animate or inanimate, are selected for mention. The languages differ, however, with respect to the way in which these events are integrated into the story line, taking both degree and type of event integration. This finding points to the relevance of the progressive in English and the associated deictic anchor (‘now you see’; ‘then you see’) in accommodating events of different types. Both languages mention events that are not fine tuned with respect to temporal shift. Both languages rely on causal relations in advancing the story line. However, French speakers downgrade events in which the protagonist is not mapped as subject, and make prominent use of the narrator in an interpretative role in establishing coherence. In this sense there is what has been termed a ‘narrator perspective’. Speakers of English can rely on a deictic ‘now you see; then you see’ in integrating agents of all kinds. We can thus identify core parameters that are well defined given the nature of the structural means available – parameters that can be investigated at the level of macro-structural planning for information structure as a whole. The presence of an aspectual marker to code the distinction ‘ongoing’ allows speakers to establish a temporal frame which delivers a default setting with respect to decisions concerning event integration. However, the comparison with French shows that a temporal frame of this kind may, but need not be the sole determining factor in information selection. This question was tested with a comparison across different languages that are ‘subject prominent’ in information structure: in the cross-linguistic analyses English, French, Italian, Spanish all show a similar pattern or default case in information selection – there are no restrictions. This shows that in English and French information selection (e.g., events) is not dictated by the nature of the temporal frame. However, event integration clearly is. Prominence is accorded to the protagonist in the narrative sequence in French (not downgraded), in contrast to English, a principle in information structure that is found in all the languages in the present study that do not make use of aspectual distinctions such as the progressive in the temporal frame. If events that advance the story line (inanimate agents) are systematically downgraded, as in French, an external point of reference is implemented (narrator as witness). We assume that this point of reference serves to bridge differences (events with the protagonist as subject in main clauses vs. natural forces as subject in subordinate clauses) and allows the speaker to establish coherence via other relations (causal, intentional, etc). The fact that the speakers of English choose between two frames is useful in the analysis of determining factors since the small number of speakers who implement the relation temporal shift in the temporal frame evidence patterns in information integration (downgrading) that differ markedly from the main group (deictic frame). This provides evidence for the assumption that the temporal frame of reference ­constitutes the most basic component at the level of macro-structural planning in a

 Mary Carroll et al.

narrative task (tell ‘what happened and why’). If speakers of English decide to recount the narrative using the simple tense form (with little or no use of the progressive), events which are not fine tuned for temporal shift are filtered out and not selected for mention, as in German and Dutch (in as far as one can rely on results for 3 narratives). As mentioned, this finding points to the relevance of the temporal frame and associated sequencing principle in delivering a default for decisions regarding the status assigned to entities in topic assignment, reference maintenance, and downgrading (passive, subordination). All of these factors correlate with the temporal frame used in sequencing events: (i) causal relations and a deictic anchor allowing incorporation of ongoing events (for the main group in English) versus (ii) temporal shift (relating to events with a right boundary) in frames that do not include the progressive but rely on the simple tense form. Significantly, there is no external anchor or reference point (i.e., no witness as reference point or deictic anchor) in this latter option in English so that sequencing principles in the temporal frame proceed on the basis of internal means only. Information structure in German, and Dutch, is fine tuned so as to accommodate the relation temporal shift on an internal basis so to speak. In German and Dutch the syntactic subject, and the information it encodes, shares its potential status (encode topic information) with other categories which map into the ‘Vorfeld’ or pre-verbal ‘slot’. These are temporal relations, e.g., the times introduced by temporal adverbials such as dann which are associated with change in state events and encode temporal shift. We assume that information selection and topic assignment is defined in these terms, and with this the decision as to what type of entity can typically be mapped as subject of a main clause: there is a focus on candidates that can deliver on temporal shift. Entities that do not qualify are either filtered out in information selection, or when selected, are downgraded on a systematic basis (passive), as revealed in the analyses for Dutch and German. The cross-linguistic comparison shows that this pattern in macroplanning is not excluded in English, for example. However, the extent to which it is found in both V2-languages provides evidence for the role of grammaticised means in driving this option, as given with the ‘V2’ constraint for word order. In keeping with these overall patterns, zero anaphora in reference maintenance correlates with events involving temporal shift in Dutch and German, while zero anaphora is restricted to events that are closely related in causal terms in English and French.

5.  Towards a model of macro-structural planning principles Based on the empirical findings for macro-structural principles in the preceding sections the present section presents the first step towards a formal model of principles that guide macro-structural planning. However, this initial step is confined to the ­subset presented above, i.e., language-specific principles that are driven by grammatical means and operate at the macro-structural level when advancing the story line.

Subordination in narratives and macro-structural planning 

Principles relating to the narrator’s world and how the two dimensions in story telling are combined will be addressed in a future paper.

5.1  Comparability of the speaker’s knowledge base We assume that speakers build up a knowledge base relating to the events seen in the film, and that knowledge at this level is comparable across the speakers. Although this assumption is clearly a simplification, it allows a common point of departure in the cross-linguistic comparison. All speakers saw the same film and, other things being equal, we can ask what factors are then imposed by language structure at the level of macro-structural planning in re-narrating the content of the film. In order to give this idea more substance let’s assume that all speakers, irrespective of their languages, have access to the same set of events in their knowledge base. For the sake of argument let us assume that this set contains every event ever selected by any speaker. In particular it contains ‘minor events’ without immediate relevance for the protagonist. Events that remain unmentioned in the individual narratives are modelled as successfully filtered by global principles when organising information for expression. The comparison of production principles will entail a comparison of principles that govern this kind of filtering. In the present framework these principles will be stated in very general terms.

5.2  Producing as deciding Assuming that the speakers have access to the same set of events we further assume that global planning principles apply event by event. Each event of the set will at some point form the current event in the sequence of events under planning. For each event the speaker must decide whether it should be selected for mention and how to encode it. Assume for a starting point that the speaker has to decide no more than three things: (i) whether to select the event or not (ii) which verbal predicate to select in encoding the event (iii) which participant to select for subject. This is written as the following set of three decisions:

(1) {<select (e), not select(e)>, <predicate(e)>, <subject(e)>}

Binary decisions are represented by ordered pairs in (1), constructive decisions are one-tuples. The set (1) is an unordered set of decisions. However, models of macrostructural planning will be ordered sets of decisions. This section deals in particular with possible hierarchical orders and explores the extent to which the relevant set of decisions are ordered in different ways in different languages. Here is a first example: The option of selecting the current event for mention can determine the selection of the subject of the related clause in a straightforward manner: The speaker can select as subject the participant that is highest on the scale for ‘agentivity’. As the findings show, the majority of English speakers proceed in this way. Another order is given when the decision to select and encode an event depends on whether a certain entity can be mapped as subject. In other words, if the current event does not allow the speaker

 Mary Carroll et al.

to map information into a clause with the ‘predetermined’ entity as subject it will be omitted. Again this provides a criterion that can operate globally since it will apply for every current event under description. We have seen that German speakers follow this principle. These interdependencies between selection for mention and subject selection in English (E) and German (D) can be presented as follows: (1’) E <select(e), not select(e)> G <subject(e)>   <subject(e)>   <select(e), not select(e)>

The question arises as to why we have converse ranking in German and English. The answer cannot be delivered on the basis of (1) since there are more obligatory binary options that have to be decided for the description than the three in (1). We will go into the other obligatory options in the next subsections thereby following three main assumptions: i. If a language demands an obligatory binary decision on the linguistic form of a sentence predication these decisions will be made at the outset for the whole unit of macroplanning. ii. Languages differ with respect to the set of obligatory options on the decision agenda, given constraints rooted in the grammatical system. iii. This leads to different decision hierarchies for the individual languages.

5.3  Structurally-determined options in German In German the pre-verbal position can be filled by different constituent types. As the analyses show, the decision includes whether or not to fill it with the syntactic subject, since this may, but need not, be placed in this position. The subject may also be placed sentence-internally, thereby mapping another constituent into the pre-verbal position.

5.4  Structurally-determined options in English There is no option and thus no obligation in deciding where to place the subject in the English narratives. Word order constraints at the syntactic level determine its pre­verbal position – constraints which also hold for French. In contrast to both German and French, English is a special case in that the question of temporal aspect, i.e., use of the simple tense or the progressive, arises with every event and associated sentence predication. There is no obligation to mark temporal perspective by grammatical means either in German or French, at least not in the present tense (the preferred tense in the present task in all three languages). The (unordered) set of decisions on the agenda differs in German (G), French (F), and English (E): (2) G {<select(e), not select(e)>, <pred(e)>, <subj(e)>,   < [subj |V| … ] , […|V| subj]>}

Subordination in narratives and macro-structural planning 



(3) F {<select(e), not select(e)>, <pred(e)>, <subj(e)>} (4) E {<select(e), not select(e)>, <pred(e)>, <subj(e)>, <prog, not prog>}

Following (3) the set of decisions for French with respect to our model is smaller than in German and English. The consequences of this finding will be treated following our presentation of the constraints on planning principles in German and English.

5.5  Implementation of these principles in context In German the syntactic subject, and the information it encodes, shares its potential status (encode topic information) with other categories which map into pre-verbal position, the ‘Vorfeld’ or pre-verbal ‘slot’. These are temporal relations introduced by anaphoric temporal adverbials which are associated with change in state events and encode temporal shift (cf. Rossdeutscher & Stutterheim, 2006, on dann). Thus (3) can be given more substance for the set of decisions on the agenda in practice. There is the option of either placing the time relation or discourse referent (the latter as subject) sentence-initially or sentence-internally, respectively. Thus (3) can be rewritten as (5). (5) G {<select(e), not select(e)>, <pred(e)>, <subj(e)>,   <[subj(e) |V| t-rel], [t-rel |V| subj(e)]>}

‘t-rel’ represents relational descriptions of temporal discourse referents of the anaphoric type. The temporal discourse referent has an antecedent in the context of the narrative sequence. In English decisions with an obligatory status relate to the temporal domain. The main impact of the global decision on verbal aspect relates to the temporal frame. The common feature of the narratives of the speakers which include use of both the progressive tense as well as the simple tense in the narrative sequence is (i) implementation of a temporal deictic anchor and (ii) a narrator-oriented perspective. In Rossdeutscher & Carroll (2006), this form of temporal structure is defined as follows: the times which the event descriptions relate to (also referred to as ‘topic times’ in Klein 1994) are subintervals of evaluation times of the propositional attitude which the narrator has about the events he has witnessed. Those speakers who do not choose this anchor and implement the simple tense only in the narrative sequence follow a principle of anaphoric shift in sequencing events, i.e., it is non-deictic. Thus in practice the aspectual option involves (i) implicit or explicit reference to some witness time, t-witness, around which the states in the story world obtain and within which events are located, or (ii) implicit or explicit reference to some relational time t-rel, in which the states obtain and within which the events are located. Leaving states aside, we write ‘e in t‘ when the event is located in the period t. The global decision to adopt a deictic anchor is written as ‘e in t-witness‘ and the global decision for anaphoric shift as ‘e in t-rel’. The grammatical option <prog, not prog> in (4) is substituted by the more ­general decision in perspective taking that correlates significantly with the decision regarding aspect. Thus (4) can be rewritten more substantially as (6):

 Mary Carroll et al.

(6) E {<select(e), not select(e)>, <pred(e)>, <subj(e)>,    <e in t-witness, e in t-rel>}

5.6  Decision hierarchies Taking the set of decisions which the speaker has to solve, we will now show how ranking is established within these sets. In German, eligibility for mention in pre-verbal position constitutes a core aspect with respect to information structure: Constituents on the left periphery play a role in linking the new predication to the preceding context (cf. Lambrecht 1994; Frey 2000, 2005). If the decision is made to ‘fill’ this position by a stable or recurring discourse referent, one form of coherence is fulfilled at the level of macroplanning. The conditions that allow use of zero anaphora are complex (cf. Klein 1993; Reinhart 1980) but one obvious factor, of course, is that the utterances at issue must share discourse referents (or else referential linkage must be marked in more explicit terms). The default condition for use of zero anaphora in German instantiates a global decision of this kind that at the same time outranks event selection. Event selection is outranked by subject selection also, if the subject (protagonist) is placed sentence internally. In order to display the ranking we will introduce an irreflexive and transitive ordering ‘>’ that obtains between subsets of the decision sets. The ordering is displayed in addition in different lines. At the level of macroplanning for the German narratives, the decision as to which individual referent forms the subject belongs to the subset of decisions that outrank the decision as to whether or not an event is selected for mention. Nothing will be determined at present about the relative ranking between subject selection and subject position: the decision regarding selection has the same ranking as the decision on where to place the referent. We also assume that predicate choice and event selection are of the same ranking. The choice of predicate cannot outrank the decision regarding sentence-internal or pre-verbal position but is definitely outranked by subject selection. This is the case since German speakers choose passive descriptions even if a natural force constitutes the discourse referent with the highest status on the agentivity scale for the event at issue. The ranking so far can be displayed as (7). (7) {{<subj(e)>, <[subj(e) |V| t-rel], [t-rel |V| subj(e)]>} > {<select(e), not select(e)>, <pred(e)>}}

The ranking in (7) predicts our findings for German where events that do not support the selection of the protagonist as subject tend to be filtered out. In English we are concerned with two different rankings: For the few speakers who do not select a deictic temporal anchor for the temporal frame, subject selection and non-deictic anchoring outrank event selection and predicate selection. Again the relative ranking between the former two must be left open at this point. Assuming here that the choice of the event-dependent predicate is independent of the option between deictic vs. anaphoric anchoring and corresponding tense usage, we include

Subordination in narratives and macro-structural planning 

the decision in the subset of outranked decisions, giving (8). (The decision to implement anaphoric shift (and thus not to use the other sequencing principle) is presented as ‘<--- , e in t-rel>’; a decision to implement a deictic temporal anchor is presented as ‘<e in t-witness, --->’ (see (9) below). (8) E {{<subj(e)>, <---, e in t-rel>} >    {<pred(e)>, <select(e), not select(e)>}}

The parallels between the English and the German narratives (based on temporal shift) are mirrored in (7) and (8) in that subject selection outranks event selection. For the larger group of English speakers we get a reverse ranking with regard to event selection and subject selection, see (9). (Compare also (1’)) (9) E {{<select(e), not select(e)>, <e in t-witness, --->} >    {<pred(e)>, <subj(e)>}}

The principle of setting eligibility for mention as subject in terms of agentivity (highest candidate), for the given event, is on a par with the decision involving a deictic temporal anchor. In (9) there are no constraints on the selection of events, which mirrors our findings that events are eligible for mention irrespective of the role the protagonist may play. Event selection is not constrained by subject selection. The question is: which set of criteria is relevant in deciding whether to select an event or not? This will be one of the issues dealt with in the next subsection.

5.7  Strengthening the decision hierarchies Up to now we have been dealing with arguments for the hypothesis that the differences in information structure in German and English are the consequence of grammatically-driven hierarchies within a set of decisions. We will now turn to the question as to whether we can strengthen the hierarchies and discuss which principles may be involved. For example, could some ranking on the basis of ‘protagonist-is-subject’ and ‘temporal frame-is-deictic’ outrank the decision as to whether an event should be selected or not? That is to say: Is a hierarchy of the following form to be expected? (m stands for the protagonist, the clay man) (10) E {{<subj(e) = m>, <e in t-witness, --->} >    {<pred(e)>, <select(e), not select(e)>}}

To give a brief answer: None of the English narratives can be reconstructed on the basis of global planning principles of this kind. Why is this so? We believe that a combination of this kind is excluded in global planning by general principles. Let’s assume for a moment that (10) could form a possible decision hierarchy (even if this is contradicted by the facts). We could then also assume that either the global decision ‘protagonist m is subject’ outranks the global decision for a deictic perspective or else the latter decision outranks the former.

 Mary Carroll et al.

i. If the speaker opts in the first place for the construction of a narrative sequence that is protagonist-centred, events can be linked by anaphoric means such as then. In this case the temporal relation between two events – selected on the basis of the fact that the protagonist can be chosen as subject, is always specified by anaphoric connectors. Any period succeeding the last mentioned event qualifies as a then-interval. Temporal anaphoric anchors provide temporal succession whether or not the described events are inherently, e.g., causally, connected. And the events chosen on protagonistcentred grounds are not causally connected in general. ii. If, on the other hand, the speaker decides to establish a deictic perspective in the first place, he is committed to establishing the story line with predications in which the topic times constitute an extended time span (e.g., ‘now’) of the witness time of the narrator. Sets of events can be described as witnessed during a time span of this kind if they are continuous events, for example. In Rossdeutscher and von Stutterheim (2006) we define two events as continuous if the consequent event overlaps in part with the state or result state of the antecedent. There are in principle three ways of accomplishing a narrative task involving a sequence of events that qualify as continuous: The speaker could produce a state by state description, each one corresponding to one witness time (‘now’) only. However, witness time can involve an extended ‘now’ and include temporal updates of the topic times of events. The events at issue can be conceptualised as continuous. Furthermore, the speaker can decide not to omit any of the events in his knowledge base and select each one. Continuity can be accomplished in this way if the events in the knowledge base are continuous. But this strategy would not go far. Given the constraints in (10) it cannot work at all: the events do not generally qualify for a description with the protagonist as subject. There is a third solution: A sequence of events can be conceptualised as continuous if conceptualised as causally connected or as part of a common plan on the part of the protagonist. But this third option of predicating a causal chain of events cannot be easily resolved by relying on protagonist-centred descriptions either. In establishing a causal chain the speaker may want to mention events that interfere with the current states and plans of the protagonist. Events of this kind need not have the protagonist as a participant and may have an antagonist as a likely subject. We sum up the results for macroplanning principles following deconstruction of the non-existent (10): If <subj(e) = m> would outrank <e in t-witness, ---> the task would be hard to solve (if solvable at all) and the speaker would be better off with the anaphoric option <----, e in t-rel> in (8), excluded in (10). In this case the speaker is free to choose the times for which he predicates properties and changes relating to the protagonist. If <e in t-witness, ---> outranks <subj(e)> the extended ‘now’ of the witness time can be filled by describing a continuous sequence of events. Restricting the selection further so as to reduce the set to events that allow for ‘protagonist-is-subject’ would be too rigid.

Subordination in narratives and macro-structural planning 

The majority of English speakers opt for a deictic perspective which outranks selection; selection is restricted by filling the witness time with causally connected events. The subject is selected with regard to each particular, taking the candidate with the highest ranking on the agentivity scale as subject. (We display this as ‘subj(e) = x’.) This option is modelled as the appropriate strengthening of (9) yielding (11): (11) E {{<e in t-witness, ---->} >    {<select(e), not select(e)>} >    {<pred(e)>, <subj(e) = x>}}

The decision hierarchy for the minority of English speakers displayed in (8) can be also strengthened in accordance with the general principles treated in this subsection. As already discussed, the highest ranking for ‘protagonist-is-subject’ leads to anaphoric anchoring. There is another possible ranking – the one where the option for anaphoric anchoring outranks any other decision in the decision set, subject selection in particular. We claim that the option for anaphoric linkage leads to restrictions in subject selection. Note first that it allows restriction of the set of events to be chosen. Any time succeeding the last topic time may be chosen as the predication time of the event. But any time is not a permissible feature in this context. If all options are open the speaker has a high burden at every point in the chain with regard to selection of the ‘next’ event (whether to omit it or not). Restricting the set by restriction via subject selection is a possible option. If the speaker chooses this solution the protagonist is the only individual discourse referent at issue. Which of the two rankings is actually appropriate cannot be decided. The latter procedure would allow for the theoretically interesting situation that production principles in English are the consequences of a binary decision on the way to link the times to which the event descriptions relate. We end this subsection by displaying this solution in (12). (11) and (12) are the two decision hierarchies in our model for macroplanning principles in English so far. (12) E {{<---- , e in t-rel>} >    {<subj(e) = m>} >    {<select(e), not select(e)>, <pred(e)>}}

5.8  German Are there any consequences for German, given the claims made for English above? It should be clear that the restriction of event selection via restrictions for the subject of the clause would be a consequence of anaphoric temporal anchoring. And the option <subj(e) = m> entails anaphoric anchoring as its consequence. Whatever the speaker decides with respect to the question as to where to place the subject is in accordance with a possible strengthening of (7), repeated here as (13). (13) G {{<subj(e) = m>, <[subj(e) = m |V| t-rel], [t-rel |V| sub(e) = m]>} >    {<select(e), not select(e)>, <pred(m)>}}

 Mary Carroll et al.

The constraints given with ‘verb second’ involves more than mere constraints on subject selection, however. As mentioned above, the dual role in information structure in German of both the syntactic subject and constituents that encode temporal relations (temporal shift with dann, ‘then’, channeled by the verb second constraint), may focus attention in information selection on a specific subset of entities in events that indicate a point of completion and thus set the preconditions for temporal shift.

5.9  French French differs from German and English in that there are no further elements in the decision set besides the basic ones to which our comparison was confined. There are no requirements which have to be considered in solving the narration task concerning aspect or subject position. Subject position is clear and aspectual distinctions are not marked, as the analyses reveal. There is no evidence of any grammatically-driven constraints on event selection (other than the absence of restrictions found for all ­subject-prominent languages in the cross-linguistic studies). Event selection outranks any other decision and the speaker must solve this decision problem in an appropriate way, given the narrative task. The question is solved via subject selection: the protagonist qualifies as subject in main clauses when advancing the story line. (Other entities do not, even though they also advance the story line.) In contrast to German, sentence structure in French (SVO) does not give rise to the case where temporal adverbs enter the stage of global sentence planning. In French the story line, as the analysis shows, is established via relations such as ‘means/goal’, ‘reactions’ and ‘plans’, often ascribed by the narrator. The intensional dimension of the story is in focus and the preferred properties predicated, in relation to the protagonist, are attitudinal properties. But as we have already discussed in connection with the English narratives, a causal chain will not provide a basis for omitting reference to antagonists, the natural forces which interfere with the current beliefs, intentions of the protagonist, especially where they appear as causal elements. In both SVO languages, the events in question are eligible for mention and are not filtered out in contrast to the V2 languages (German and Dutch). So word order constraints in French and English do not drive a restriction at this level, in contrast to the V2 languages. In French omission would mean filtering out events that can contribute to progression at a causal level. They are downgraded at a temporal level (subordinated) but not in causal terms. In English a deictic temporal anchor allows integration of any event that fills the extended ‘now you see’ or ‘then you see’ of the witness time. There are no grounds at this level in English to filter out a certain set of events. So in both languages eligibility for mention may be guided by the fact that they qualify as candidates in forming a causal chain. The discussion of planning principles in French ends with the decision hierarchy resulting from our considerations. In the notation, <subj = m> states that subject selection is not event dependent. The restriction to intensional properties is presented in <pred(attitude(m)>.

Subordination in narratives and macro-structural planning 

(14) F {{<subj = m>} >    {<select(e), not select(e)>, <pred(attitude(m)>}}

The ranking predicts the restriction given for event sequences.

6.  Conclusions The present analysis provides evidence for downgrading procedures and their determinants as going beyond the functional description of subordination as a downgrading device that operates across well established divisions such as foreground and background (Hopper 1979, 1995; Tomlin 1985; Cristofaro 2005). It shows how determining factors can be linked to grammaticised means that are language specific – a verb second constraint at the level of word order, the presence or absence of aspectual distinctions such as the progressive. It also illustrates other forms of downgrading that are not random, or subject to questions such as ‘style’, but form part of a set of general principles that hold at different levels in information structure. Analyses of the related factors and associated constraints, that pair with relevant typological contrasts, point to a hierarchically ordered set of principles at the level of macro-structural planning that are grammatically driven.

References Carroll, Mary & Lambert, Monique. 2003. Information structure in narratives and the role of grammaticised knowledge: A study of adult French and German learners of English. In Information Structure and the Dynamics of Language Acquisition, Dimroth, Christine & Marianne Starren (eds), 267–287. Amsterdam: John Benjamins. Carroll, Mary & Lambert, Monique. 2006. Reorganizing principles of information structure in advanced L2s: French and German learners of English. In Educating for Advanced Foreign Language Capacities, Heidi Byrnes, Heather Weger-Guntharp & Katherine Sprang (eds), 54–73. Washington DC: Georgetown University Press. Cristofaro, Sonia. 2005. Subordination. Oxford: OUP. Frey, Werner. 2000. Über die syntaktische Position des Satztopiks im Deutschen. In Issues on Topics. ZAS Papers in Linguistics 20: 137–172. Frey, Werner. 2005. Pragmatic properties of certain German and English left peripheral constructions. Linguistics 30: 5–55. Hopper, Paul. J. 1979. Aspect and foregrounding in discourse. Syntax and Semantics 12: Discourse and Syntax. Talmy Givon (ed.), 213–241. New York: Academic Press. Hopper, Paul J. 1995. The category ‘Event’ in natural discourse and logic. In Discourse Grammar and Typology, Werner Abraham, Talmy Givon & Sandra Thompson (eds), 139–153. Amsterdam: John Benjamins. Klein, Wolfgang. 1993. Ellipse. In Syntax. Ein internationales Handbuch zeitgenössischer Forschung, Joachim Jacobs (ed.), 763–798. Berlin: de Gruyter.

 Mary Carroll et al. Klein, Wolfgang. 1994. Time in Language. London: Routledge. Lambrecht, Knud. 1994. Information Structure and Sentence Form. Cambridge: CUP. Murcia Serra, Jorge. 2001. Grammatische Relationen im Deutschen und Spanischen. Frankfurt: Peter Lang. Reinhart, Tanya. 1980. Conditions for text coherence. Poetics Today 1: 161–180. Roberts, Craige. 1989. Modal subordination and pronominal anaphora in discourse. Linguistics and Philosophy 12: 683–721. Rossdeutscher, Antje & Carroll, Mary. 2006. How to cope with narration tasks in German and English. University of Heidelberg. URL: http://www.ims.uni-stuttgart.de/~antje/narration-task. pdf. Rossdeutscher, Antje & von Stutterheim, Christiane. 2006. Semantische und pragmatische Prinzipien der Positionierung von ‘dann’. Linguistische Berichte 206: 29–60. Stutterheim, Christiane von, Carroll, Mary & Klein, Wolfgang. 2003. Two ways of construing complex temporal structures. In Deictic Conceptualization of Space, Time, and Person. Friedrich Lenz (ed.), 97–133. Berlin: de Gruyter. Tomlin, Russell S. 1985. Foreground-background information and the syntax of subordination. Text 5(1–2): 85–122.

part iii

Monolingual studies

German dependent clauses from a constraint-based perspective Anke Holler

University of GÖttingen In this article the main focus is on German non-canonical clause linkage. Inspecting five instances of non-canonically linked clauses (i.e., weil-verb second clauses, continuative wh-relative clauses, verb second relative clauses, free dass-clauses, and dependent verb second clauses) it is argued that subordination should be treated as a multidimensional phenomenon. By means of the aforementioned clauses, the article empirically investigates several factors underlying the compositeness of subordination. To account for the presented clause linkage facts a non-derivational analysis in the framework of Head-driven Phrase Structure Grammar is developed employing constraints and sorts partially ordered in multiple-inheritance hierarchies as conceptual devices to model several degrees of clause linkage. Keywords: non-canonical clause linkage, subordination, HPSG

1.  Introduction Traditional, structural as well as generative approaches to grammar usually act on the assumption that clauses that can stand alone as complete sentences differ grammatically from the ones that are dependent on a prior clause. This becomes manifest in the ­distinction of main and subordinate clauses. Related to this opposition is a stipulated contrast between two syntactic linkage relations: hypotaxis and parataxis. On the other hand, several studies (cf. e.g., Haiman & Thompson 1984; van Valin 1984; Lehmann 1988; König & van der Auwera 1988; Fabricius-Hansen 1992; Reis 1997; Peyer 1997) have shown that a dichotomic view on clause types and syntactic linkage in complex sentences oversimplifies the linguistic facts. In this article I will revisit this problem. I will firstly review relevant aspects of the debate by discussing pertinent empirical data and influential theoretical approaches. On the basis of clause linkage data from ­German I will take up the position that the opposition of two syntactic linkage relations should be abandoned entirely, and that clauses should be grammatically distinguished in terms of the degree they are integrated into a potential host clause ­depending on their ­syntactic form, their interpretation, and their functional usage. Further, I will ­elaborate on the proposal that typical non-integrated clauses are best analysed as ­being orphan

 Anke Holler

c­ onstituents that are linked to the prior clause only discourse-structurally. I will ­provide a constraint-based analysis in the framework of Head-driven Phrase Structure Grammar (Pollard & Sag 1994) arguing that a ­sign-based linguistic theory ­employing sorted feature structures and multiple-inheritance sort hierarchies as ­conceptual ­devices seems to be well suited to account for the presented clause linkage facts. The article is organised as follows. In the next section some more general issues concerning the nature of clause linkage and the distinction between main and subordinate clauses are addressed. Section 3 recalls data of German dependent clauses that challenge any approach implementing a twofold differentiation between main and subordinate clause types. Section 4 sketches a constraint-based analysis that organizes clause types according to the way they are grammatically linked to the prior clause. Section 5 concludes the article.

2.  Clause linkage and subordination This section discusses two essential issues concerning the nature of clause linkage and subordination, viz. (i) the question whether there exist more than one relation linking a putative subordinate clause to its matrix clause, and (ii) by which grammatical means a subordinate clause can be identified at all.

2.1  Is there more than one subordinating linking relation? Complex sentences are usually analysed as being composed of clauses joined in a ­hypotactic or paratactic relation. Roughly speaking, a clause is viewed as hypotactically linked to a prior clause if it is subordinated to this clause; it is viewed as paratactically linked to a prior clause if it is coordinated with it.1 At the syntactic level subordination can be expressed by specific connectives such as subordinating conjunctions like dass (‘that’), (relative or interrogative) pronouns and relative adverbs. A coordinate ­relationship is realised by coordinating conjunctions like und (‘and’) and oder (‘or’) or by ­punctuation marks. Whereas it seems to be general consensus that clauses ­being in a ­coordinate relationship are sequentially added together, it is much more under ­debate what ­subordination exactly means. One way to look at subordination is to define ­subordinate clauses as clauses that are embedded as a constituent of a matrix sentence. ­According to this view a subordinate clause functions as a part of another clause, for instance ­being an argument or an attribute. Another approach is to say that subordination is based on ­dependence, plus assuming that subordinate clauses are characterised by internal formal properties such as a certain position of the finite verb (e.g., a clause final position in German) or the presence of a subordinating conjunction. At first glance defining

.  The notions of parataxis and coordination are not identical, cf. Pasch et al. (2003).

German dependent clauses 

subordination involving either embeddedness or dependence seems to be equivalent. In fact it is not. Being connected with a prior clause does not automatically result in having a syntactic function in that clause. For instance, this can be seen in examples like (1). The dependent clause, although looking like a relative clause, is not syntactically embedded in the prior clause, cf. Holler (2005). (1) Emma gewann die Schachpartie, was Oskar ärgerte. Emma won the chess.match which Oskar annoyed ‘Emma won the chess match, which annoyed Oskar.’

In addition, typologists studying non-Indo-European languages have persuasively ­argued that clause chaining is evidence that dependence and embeddedness are ­distinct parameters in defining syntactic relations in clause linkage, see e.g., van Valin (1984), Haiman & Thompson (1984), Longacre (1996). Thus, seeking for language universals also rejects identifying embeddedness and dependence. On the basis of these results it is striking that clauses must generally be evaluated according to at least two separate criteria: (i) whether they are linked to another clause (dependence), and (ii) whether they fulfil a syntactic function in another clause (embeddedness).2

2.2  How to identify a subordinate clause? Even if embeddedness and dependence are recognized as isolable and independent parameters of the relationship that two adjacent clauses can have, it is not yet clarified which set of properties characterises subordinate clauses. That this is in fact a complex issue shows work done in the eighties and nineties on various Germanic languages, e.g., by Lehmann (1988), König & van der Auwera (1988), Fabricius-Hansen (1992), Santorini (1992) and Peyer (1997). It has been demonstrated in the course of these studies that a clear-cut separation of the main clauses from the subordinate ones seems to be impossible. Despite this comprehensive work, it is still often taken for granted that German clause types could be clearly distinguished by surface-related syntactic means such as the placement of the finite verb. This point of view has been particularly revitalised in construction-oriented and/or constraint-based approaches to Germanic languages, which usually act on the assumption that the finite verb is generally fronted in main clauses, while it has to occur clause-finally in subordinate clauses. For instance, this ­position has been adopted in Uszkoreit (1987), Kathol (1995), Netter (1998), or ­Ginzburg & Sag (2000), mostly consulting examples similar to (2) as evidence. (2) a.

Oskar ist vom Stuhl gefallen. Oskar has from.the chair fallen ‘Oskar has fallen from a chair.’

.  This view complies well with the approaches proposed for instance by Lehmann (1988) and Pasch et al. (2003) in order to account for phenomena of German clause combining.

 Anke Holler

b. Emma bezweifelt, dass Oskar vom Stuhl gefallen ist. Emma doubts that Oskar from.the chair fallen is ‘Emma doubts that Oskar is fallen from a chair.’

These data form the basis of stipulating that the position of the finite verb is ‘hardwired’ to the representation of a certain clause type. In other words, the mentioned constraint-based approaches follow the idea that a fronted finite verb generally marks main (root) clauses, whereas its final position generally signals a subordinate clause. This perspective involves the additional assumption that the set of main clauses comprises all clauses and only those clauses that can be uttered independently. See Kathol (1995) for an explicit proposal elaborating on this idea. Splitting clause types into main and subordinate depending on the position of the finite verb and interconnecting this with the (in-)dependence of the respective clauses yields an approach that classifies dependent verb second clauses such as (3a) as main clauses, but independent verb final clauses such as (3b) as subordinate clauses, predicting – contrary to the facts – that the verb second clause is uttered independently but not the verb final one. (3) a. Emma glaubt, Studierende schlafen lange. Emma believes students sleep long ‘Emma believes that students sleep long.’ b. Ob er noch kommt? Whether he till omes ‘I wonder whether he will still come?’

Although approaches acting on a dedicated relation between the placement of the finite verb and the classification of a clause as main or subordinate are obviously flawed, most linguists probably share the intuition that the position of the finite verb may somehow contribute to the main-subordinate distinction. This is not contradictory if we assume that a clausal typology is not based on a single phenomenon, but is rather associated with a number of different aspects of grammar. From this point of view it is conceivable that a phenomenon being normally considered as typical for main clauses (such as verb second) may occur in dependent clauses (cf. example (3a)),3 or may not be realised at all in an independent clause (cf. example (3b)). If we look this way on a clausal typology, it comes as no surprise that the competent user of language only seems to be able to identify a clause as main or subordinate by consulting a bunch of grammatical properties. This means that the terms ‘main clause’ and ‘subordinate clause’ must be taken as labels for a set of several independent linguistic properties. In canonical cases, each of these properties can be identified; in non-canonical cases,

.  Germanic verb second is a so-called main clause phenomenon. It has been studied e.g., by Vikner (1995), Santorini (1992), Wechsler (1991), Reis (1997), Meinunger (2004), Truckenbrodt (2006) and many others. Emonds (1970), Hooper & Thompson (1973) as well as Green (1976) are concerned with related phenomena in English, i.e., constructions involving root phenomena. For reasons of limited space, the reader is referred to the original work.

German dependent clauses 

however, only a subset of the formal properties that are associated with the one or the other clause type is realised. This is the reason why clauses may differ in degree they are viewed as main or subordinate, as I will further discuss below. Taking subordination as a multidimensional phenomenon in the aforementioned way raises the question of the types of factors underlying the compositeness of subordination. Following Haiman & Thompson (1984), Lehmann (1988) as well as Fabricius-Hansen (1992) and Reis (1997), I propose that these properties pertain to syntactic form, interpretation, and functional usage. On the basis of empirical data of German complex clause constructions I will discuss the essential factors in the next section. The data by themselves are not new, but the consequences drawn from them may contribute to the theoretical discussion on the analysis of clausal subordination.

3.  The empirical challenge In this section pertinent complex clause constructions, primarily drawn from German, are investigated with respect to a set of selected grammatical and pragmatic criteria that are widely accepted as being relevant for the distinction of main and subordinate clauses, cf. among others Haiman & Thompson (1984), Fabricius-Hansen (1992), Reis (1997). The following criteria are employed: (i) intonation and prosodic features, (ii) internal form of the clause, (iii) external syntax and topological placement, (iv) interpretational aspects as well as (v) information structure and association with illocutionary force.

3.1  Weil-verb second clauses The so-called weil-verb second clauses4 (henceforth weil-V2 clauses) as exemplified by (4a)5 are an alternative type of standard causal adverbial clauses such as (4b), which has been substantiated by work of Wegener (1993), Keller (1995), Pasch (1997), ­Uhmann (1998) among others. At the syntactic surface, weil-V2 clauses are characterized by an introducing conjunction weil (‘because’) and by a finite verb in second position.6 (4) a.

Peter kommt zu spät, weil er hat keinen Parkplatz gefunden. Peter comes too late because he has no parking.space found ‘Peter is late because he could not find a parking space.’

.  Weil-verb second clauses are mainly attested for colloquial German, but can be observed in ­written German as well, cf. Wegener (1993) and Uhmann (1998). Both authors have convincingly ­argued that verb second in weil-clauses is not an irregularity of performance but a systematic part of speakers’ language competence. .  The example is taken from Uhmann (1998). .  Uhmann (1998) citing Altmann (1997) points out that there are cases where the finite verb occupies an initial position, e.g., Ich brauche jetzt unbedingt neue Kleider, weil – möchtest du ­eigentlich noch so mit mir ausgehen? (‘I absolutely need new clothes, because – do you really want go out with me that way?’)

 Anke Holler

b.

Peter kommt zu spät, weil er keinen Parkplatz Peter comes too late because he no parking.space gefunden hat. found has ‘Peter is late because he could not find a parking space.’

Example (5) indicates that weil-V2 clauses are topologically restricted to a final position, which means that they neither stand in front of the clause they are combined with (cf. (5a)) nor within it (cf. (5b)). This is clearly contrary to standard weil-adverbial clauses as (6) shows.7 (5) a. b. (6) a. b.

*Weil er hat keinen Parkplatz gefunden, kommt Peter zu spät. because he has no parking.space found comes Peter too late *Peter kommt, weil er hat keinen Parkplatz gefunden, zu spät. Peter comes because he has no parking.space found too late Weil er keinen Parkplatz gefunden hat, kommt Peter because he no parking.space found has comes Peter zu spät. too late ‘Peter is late since he could not find a parking space’ Peter kommt, weil er keinen Parkplatz gefunden hat, Peter comes because he no parking.space found has zu spät. too late ‘Peter is late since he could not find a parking space’

In contrast to canonical adverbial clauses weil-V2 clauses do not have a syntactic function in the clause they are combined with. Evidence for this comes from the fact that it is impossible to refer to a weil-V2 clause using a correlative or to attach them by a supplement, as (7) demonstrates.8 (7) a.

Peter kommt (*deswegen) zu spät, weil er hat keinen Parkplatz Peter comes therefore too late because he has no parking.space gefunden. found ‘Peter is late because he could not find a parking space.’

.  The described topological restrictions may be violated if the weil-V2 clause has parenthetical function. This does not blur the argument concerning the mentioned topological restrictions, but rather substantiates the view that weil-V2 clauses show a root-like behaviour: ?Peter ist – weil er hat keinen Parkplatz gefunden – schon wieder zu spät. (‘Peter is, because he could not find a parking space, late again.’) .  There is further evidence that weil-V2 clauses do not function as a component part (Satzglied) of the clause they are linked to. Contrary to causal weil clauses, weil-V2 clauses cannot be transferred into an adverbial phrase as the following example illustrates: (i) can only paraphrase (ii), but not (iii).

German dependent clauses 

b.

*Peter kommt Peter comes Parkplatz parking.space

zu spät, und.zwar weil er hat keinen too late namely because he has no gefunden. found.

In addition, weil-V2 clauses behave exceptionally with respect to their internal syntax. Uhmann (1998:111) cites corpus data showing that weil-V2 clauses allow left dislocation and topicalisation, which are typical root phenomena. Apart from the mentioned syntactic peculiarities, weil-V2 clauses deviate semantically and pragmatically from their adverbial counterparts. It is a well-known fact that weil clauses in general may be interpreted either causally or epistemically (cf. Sweetser 1990; Keller 1995; Blühdorn forthcoming). The preferred reading of weil-V2 clauses is however an epistemic one. According to this interpretation, the speaker gives reasons for the assumption expressed in the host clause. There are in fact cases that can only get the epistemic interpretation. In example (8) for instance, the speaker concludes that it hailed from the fact that her car has dints. Obviously, the damaged car cannot be a reason for hailing, but the damage is a plausible reason for the speaker’s assumption that it hailed. For further discussion of the epistemic reading of weil-V2 clauses I refer the interested reader to the literature, e.g., Wegener (1993), Keller (1995), and Blühdorn (forthcoming). (8) Es hat gehagelt, weil mein Auto hat Dellen. it has hailed because my car has dints ‘My car is damaged because it has hailed.’

Another semantic characteristic of weil-V2 clauses is that they may behave differently from causal weil clauses with respect to scope of negation and quantifier binding. As Wegener (1993) has shown, the content of a weil-V2 clause is not negated if the host clause contains a negative particle: Whereas the content of the weil clause in (9a) is not in the scope of the matrix negation, (9b) can be interpreted such that it is denied that Peter went home because of a head ache.9 (9) a. b.

Peter ist nicht nach.Hause gefahren, weil er hatte Kopfweh. Peter is not home driven because he had a.head.ache ‘Peter did not drive home because he had a head ache.’ Peter ist nicht nach.Hause gefahren, weil er hatte, Peter is not home driven because he had

(i) Wegen seines Versprechens kommt Max. (‘Max comes because of his promise to do it.’) (ii) Max kommt, weil er es versprochen hat. (‘Max comes because he has promised it.’) (iii) Max kommt. Weil – er hat es versprochen. (‘Max comes. Because he has promised it.’) .  Note that negation generally does not have scope over epistemically interpreted weil clauses. The verb final weil clause in example (9b) is actually ambigious between a causal and an epistemic reading.

 Anke Holler



Kopfweh (sondern …). a.head.ache but ‘Peter did not drive home because he had a head ache, (but …).’

Example (10) also taken from Wegener (1993) demonstrates that a quantifier in the host clause does not scope over a weil-V2 clause: (10a) justifies why the speaker ­believes that some guests will come, while (10b) means in a preferred reading that some guests will come because of the sunny weather.10 (10) a. b.

Einige Gäste werden kommen, weil heute scheint die Sonne. some guests will come because today shines the sun ‘Some guests will come, because today the sun is shining.’ Einige Gäste werden kommen, weil heute die Sonne scheint. some guests will come because today the sun shines ‘Some guests will come, because the sun is shining today.’

With respect to the pragmatic properties of weil-V2 clauses, several authors emphasise that weil-V2 clauses have an illocutionary force that is independent of that of the host clause, see e.g., Wegener (1993), Keller (1995), Uhmann (1998). In other words, a weilV2 clause is not part of the speech act expressed by the prior clause. This is ­illustrated in example (11). The weil-V2 clause in (11a) expresses a statement and is clearly not part of the question. This contrasts with canonical causal weil clauses as shown in (11b). (11) a. b.

Kommt Peter? Weil er hat es versprochen. comes Peter because he has it promised ‘Is Peter coming? Because he promised to.’ Kommt Peter, weil er es versprochen hat? comes Peter because he it promised has ‘Is Peter coming because he promised to?’

Data like (11) furthermore provide evidence that the intonational unit of a weil-V2 clause is separated from that of the host clause. This is also observed in Uhmann (1998:121). For instance, she points out that the sequence in (12a) is only acceptable if separated in two intonation units marked by a rising boundary tone at the final boundary of the first intonation phrase. In contrast, (12b) indicates that the same sequence is unacceptable if phrased as one single intonation unit. This is, however, not the case for causal weil clauses as (12c) illustrates. (12) a.

Ist Peter zu spät gekommen? Weil er hat keinen is Peter too late came because he has no

.  Example (10b) is in fact ambiguous. There is a second reading of the verb final weil clause which is similar to the interpretation of example (10a).

German dependent clauses 

b. c.

Parkplatz gefunden. found parking.space ‘Is Peter coming too late? Because he has not found a parking space.’ * Ist Peter zu spät gekommen, weil er hat keinen is Peter too late came because he has no Parkplatz gefunden. parking.space found Ist Peter zu spät gekommen, weil er keinen Parkplatz is Peter too late came because he no parking.space gefunden hat? found has ‘Did Peter come too late because he has not found a parking space?’

As a matter of course, the aforementioned intonational differences of causal weil ­clauses and weil-V2 clauses are closely related to their interpretation. For instance, the question operator scopes over the weil clause only in (12c), but not in (12a). ­Consequently, the weil-V2 clause in (12a) states a reason for asking the question, which is a speaker ­attitude. In (12c), however, the reason for Peter’s being late is asked about. For a ­profound ­discussion of these data see Uhmann (1998: 120 et seqq.).

3.2  Continuative wh-relative clauses There is a subclass of non-restrictive relative clauses that behave all about the same as weil-V2 clauses. I call this class continuative wh-relative clauses or short wh-relatives because at the syntactic surface these clauses can be recognised by a possibly complex left-peripheral wh-expression as exemplified in (1), repeated here as (13).11 (13) Emma gewann die Schachpartie, was Oskar ärgerte. Emma won the chess.match which Oskar annoyed ‘Emma won the chess match, which annoyed Oskar.’

At first glance, wh-relatives look like canonical subordinate clauses since the finite verb is placed in final position. On closer inspection, however, wh-relatives show a number of properties that are usually not considered as being characteristic of subordinate clauses. First, wh-relatives are prosodically detached from their host clause as they establish a separate intonation unit. They have to be preceded by a pause, and the prior clause ends with falling intonation. (14) Emma kaufte einen TEURen Schrank.(\), was ÄRGerlich ist. Emma bought an expensive cupboard which annoying is ‘Emma bought an expensive cupboard, which is annoying.’

.  For a comprehensive empirical and theoretical analysis of the class of continuative relative clauses see Holler (2005).

 Anke Holler

Secondly, the described prosodic behaviour forms the basis for the pragmatic autonomy of the wh-relatives. Their focus domain is independent of that of the prior clause, cf. (15). (15) Was ist passiert? (‘What happened?’) #[Emma kaufte einen teuren Schrank, was ÄRGerlich ist.]F Emma bought an expensive cupboard which annoying is ‘Emma bought an expensive cupboard, which is annoying.’

The construction in (15) cannot be uttered as an ‘all-focus’ answer to the question What did happen?, which is usually taken as an argument that the wh-relative clause is not integrated into the information structure of the host. That the whole construction ­cannot be uttered as an ‘all-focus’ sentence is attributed to the fact that every clause of the construction needs a focus accent, and thus projects a separate focus-background-structure.12 From examples like (16) comes evidence that wh-relatives have an autonomous illocutionary force. On the one hand, they are not part of the content that is asserted or asked for, cf. (16a). On the other hand, they may be used to accomplish a speech act, cf. (16b). (16) a. * Hat Emma einen Schrank gekauft, was Oskar erstaunte? has Emma a cupboard bought which Oskar astonished b. Hat Emma einen neuen Schrank gekauft? Was Oskar has Emma a new cupboard bought which Oskar nämlich sehr erstaunen würde, weil sie doch so namely very astonish would because she PART so geizig ist. mean is ‘Did Emma bought a new cupboard? This would astonish Oskar because she is very mean.’

Example (17) demonstrates that epistemic expressions, performative indicators, ­modal particles can appear in wh-relatives. This is further evidence for the illocutionary independence of wh-relatives. (17) Die Firma handelt mit Waffen, weshalb ich {sicher/ the company deals with weapons that’s why I {certainly/ hiermit/wohl} kündige. hereby/well} hand.in.my.notice ‘The company deals with weapons, and that’s why I {certainly/ hereby/well} hand in my notice.’

Thirdly, wh-relatives do not have a syntactic function in their host clause because they neither receive a theta-role from the matrix predicate nor may function as an adverbial adjunct. They rather express a commentary or provide additional back-

.  For a similar argumentation see Brandt (1990).

German dependent clauses 

ground ­information. Thus, wh-relatives can be fairly freely added to a clause. Example (18) also illustrates that wh-relatives are not a component part of the prior clause since they cannot be linked to it using a supplement expression like und zwar. (18) *Oskar verlor die Schachpartie, und.zwar was ihn ärgerte. Oskar lost the chess.match namely which him annoyed

Fourthly, wh-relatives show a scope behaviour that is characteristic of non-integrated clauses as they disallow variable binding from outside, and a negative particle placed in the host clause does not scope over the wh-relative. (19) a. Oskari hat die Schachpartie nicht gewonnen, was ihni ärgerte. Oskar has the chess.match not won which him annoyed ‘Oskar did not win the chess match, which annoyed him.’ b. * Kein Spieleri gewann die Schachpartie, was ihni ärgerte. no player won the chess.match which him annoyed

Fifthly, wh-relatives are topologically restricted occuring only at the very end of a complex sentence, as demonstrated in (20) and (21).13 (20) a. b.

Oskar verlor die Schachpartie, was ihn ärgerte. Oskar lost the chess.match which him annoyed ‘Oskar lost the chess match, which annoyed him.’ * Was ihn ärgerte, verlor Oskar die Schachpartie. which him annoyed lost Oskar the schess.match

(21) a. * Oskar verlor die Schachpartie, was ihn ärgerte, die er Oskar lost the chess.match which him annoyed that he unbedingt gewinnen wollte. absolutely won wanted.to b. Oskar verlor die Schachpartie, die er unbedingt gewinnen Oskar lost the chess.match that he absolutely won wollte, was ihn ärgerte. wanted.to which him annoyed ‘Oskar lost the chess match that he absolutely wanted to win, which annoyed him.’

On the basis of the presented empirical properties of wh-relatives we can conclude that continuative wh-relative clauses seem to be in accordance with weil-V2 clauses with respect to all principal criteria investigated.

3.3  Verb second relative clauses Gärtner (2001) investigates a further peculiar subclass of relative clauses in ­German, which he dubs integrated-verb second relative clauses (henceforth V2 ­relatives).­

.  Note that there are cases where continuative wh-relatives appear inside the host clause. However, this is possible only if they are pronounced parenthetically.

 Anke Holler

Using criteria along the lines pursued in the present article, he shows that V2 relatives such as (22) are clauses that “form a hybrid between hypotaxis and parataxis” (Gärtner 2001:29). (22) Das Blatt hat eine Seite, die ist ganz schwarz. the sheet has one side that is completely black ‘The sheet has one side that is completely black.’

Gärtner argues that V2 relatives show to some extent grammatical properties that are normally taken as evidence for embeddedness and are hence attributed to subordinate clauses: (i) V2 relatives have attributive function and get a restrictive interpretation. This can be seen in example (23) indicating that the content of the relative clause is picked up by the indefinite pronoun eins (‘one’) if this pronoun substitutes the nominal phrase. (ii) V2 relatives can be prosodically integrated into the prior clause, cf. (24), and (iii) V2 relatives are able to constitute a single informational unit with the host clause, which is according to Brandt (1990) definable as a single partition into focus and background. This demonstrates the ‘all-focus’ sentence in (25). (23)

Anna hat ein Buch, das liest sie nicht, und Emma hat Anna has a book that reads she not and Emma has auch eins. also one ‘Anna has a book that she does not read and Emma also has one.’

(24) Anna hat ein Buch.(/), das liest sie nicht. Anna has a book that reads she not ‘Anna has a book that she does not read.’ (25)

Was ist passiert? (‘What happened?’) [Anna hat ein Buch, das liest sie nicht]F Anna has a book that reads she not ‘Anna has a book that she does not read.’

In terms of the aforementioned facts V2 relatives should be treated on a par with ­restrictive relative clauses, but things are more intricate as Gärtner (2001) argues. He provides evidence that V2 relatives have properties that diverge from canonical ­(restrictive) relative clauses: (i) The finite verb occurs in second position; (ii) V2 ­relatives must remain clause final as (26) suggests since V2 relatives cannot precede ordinary verb final restrictive relative clauses; (iii) a quantifier cannot bind a variable in a V2 relative, cf. (27); (iv) V2 relatives cannot be in scope of a negation operator realised in the ­matrix clause, cf. (28); and (v) V2 relatives seem to be illocutionary independent, because (29a) is ungrammatical indicating that the relative clause cannot be subsumed under the question operator. Additionally, epistemic expressions and modal particles may occur in V2 relatives, cf. (29b). Gärtner (2001) concludes from these data that V2 relatives must have an “assertional nature”.

German dependent clauses 

(26) a. b.

Anna hat Bücher, die sie sehr schätzt, die liest Anna has books that she very appreciates that not sie nicht. reads she ‘Anna has books that she very appreciates that she does not read.’ *Anna hat Bücher, die liest sie nicht, die sie sehr schätzt. Anna has books that reads she not that she very appreciates

(27) a. *Keine Studentini hat ein Buch, das liest siei nicht. no student has a book that reads she not (28) Anna findet ein Buch nicht, das hat Grass geschrieben. Anna finds a book not that has Grass written ‘Anna does not find a book that Grass has written.’ (29) a. b.

*Hat Anna ein Buch, das liest sie nicht? has Anna a book that reads she not Anna hat ein Buch, das liest sie {sicher/wohl} nicht. Anna has a book that reads she {certainly/well} not ‘Anna has a book that she {certainly/well} does not read.’

The presented facts, thus, suggest that V2 relatives seem to oscillate grammatically between main and subordinate clauses.

3.4  Free dass-clauses A further class of clauses having properties residing between main and subordinate clauses are the so-called free dass-clauses, which have been thoroughly described by Reis (1997).14 Example (30) illustrates this clausal class. (30) Max ist größenwahnsinnig, dass er jetzt noch.einen Porsche kauft. Max is megalomaniac that he now another Porsche buys ‘Max is megalomaniac to buy another Porsche now.’

Free dass-clauses demonstrate that the presence of a complementizer is another parameter traditionally taken to indicate subordination that cannot be generally valid. Apart from the clause final position of the finite verb, free dass-clauses behave only with respect to their prosody and information structure like subordinate clauses. According to Reis (1997), it is possible to prosodically integrate a free dass-clause into its host clause, which results in one single informational unit. In other words, free dassclauses do not necessarily form a separate phonological phrase and focus domain as is

.  I follow the terminology of Reis (1997). The term ‘free dass- clause’ is sometimes also used for independent sentences like Dass Du mir nicht wegläufst. (‘Don’t run away.’), which I do not discuss in this paper.

 Anke Holler

illustrated in example (31) showing that free dass-clauses can in principle be uttered as ‘all-focus’ sentences.15 (31)

Was ist passiert? (‘What happened?’) [Max ist GRÖßenwahnsinnig, dass er jetzt noch.einen Porsche kauft.]F Max is megalomaniac that he now another Porsche buys ‘Max is megalomaniac to buy another Porsche now.’

On the other hand, free dass clauses show properties that are not characteristic of subordinate clauses. Contrary to dass-complement clauses, free dass-clauses are not subcategorized by the host clause’s predicate. Hence, free dass-clauses do not allow correlatives or supplements as (32) exemplifies. (32) a. b.

Maria ist (*es) blöd, dass sie kommt. Maria is (it) kind.of.stupid that she comes ‘Maria is kind of stupid to come.’ *Maria ist blöd, und.zwar dass sie kommt. Maria is kind.of.stupid namely that she comes

Free dass-clauses are also more restricted than dass-complement clauses with respect to their topological behaviour. As noticed by Reis (1997), they are obligatorily extraposed, cf. (33). (33) a. b. c. d.

Du musst verrückt sein, dass du kommst. you must crazy be that you come ‘You must be crazy that you come.’ *Dass du kommst, musst du verrückt sein. that you come must you crazy be Was ist denn gerade los, dass er so schreit? what is part just.now the.matter that he like.that screams ‘What is wrong just now, that he screams like that?’ *Was ist denn, dass er so schreit, gerade los? what is part that he like.that screams just.now the.matter

The fixed order is also evidence that free dass-clauses do not behave like subordinate adverbial adjunct clauses, either, which for instance can occupy the clause initial ­position, as (34) indicates. (34) Weil er so schreit, beschweren sich die Nachbarn. because he likes.that screams complain refl the neighbours ‘Because he screams like that the neighbours complain.’

In addition, free dass-clauses occur before rather than after adverbial clauses introduced by weil (‘because’) as Reis (1997) shows by means of examples like (35).

.  Note that it is generally also possible to decompose the construction into two informational units: Max ist GRÖßenwahnsinnig, dass er jetzt noch einen PORsche kauft.

German dependent clauses 

(35) a. b.

Du bist blöd, dass du Ernas Nerzmantel bezahlst, weil you are kind.of.stupid that you Erna’s mink.coat pays.for because sie nämlich selbst genügend Geld auf dem Konto hat. she namely herself enough money on the account has ‘You are kind of stupid to pay for Erna‘s mink coat because she has enough money on her account.’ *Du bist blöd, weil sie nämlich selbst genügend you are kind.of.stupid because she namely herself enough Geld auf dem Konto hat dass du Ernas money on the account has that you Erna’s Nerzmantel bezahlst. mink.coat pays.for

Moreover, Reis (1997) argues that free dass-clauses cannot appear as sentence ­fragments in question-answer pairs, cf. (36a), which is again in contrast to canonical subordinate adverbial clauses introduced by e.g., weil, cf. (36b). (36) a. b.

Wieso/Warum ist Fritz blöd? (‚Why is Fritz stupid?‘) *Dass er Ernas Nerzmantel bezahlt. that he Erna‘s mink.coat pays.for Wieso/Warum ist Fritz blöd? (‘Why is Fritz stupid?‘) Weil er Ernas Nerzmantel bezahlt. because he Erna‘s mink.coat pays.for ‘Because he’s paying for Erna’s mink coat.’

The semantico-pragmatic properties of free dass-clauses are not completely understood, yet. But it can be observed that free dass-clauses are not generally interpreted in the scope of negation or negative predicates contained in the host clause, cf. (37), and that a quantifier in the host clause may not bind a variable occurring in the free dass-clause, cf. (38).16, 17 (37) *Max ist nicht größenwahnsinnig, dass er jetzt noch.einen Porsche kauft. Max is not megalomaniac that he now another Porsche buys (38) *Fast jeder Politikeri ist größenwahnsinnig, dass eri jetzt almost every politician is megalomaniac that he now noch.einen Porsche kauft. another Porsche buys

.  These empirical facts contrast to Reis’s original results. She claims by means of examples like Jederi war blöd, dass eri darauf eingegangen ist. (‘Everybody was stupid to accept it.’) that free dassclauses can contain a pronoun that is bound by a quantifier in the matrix clause. Having consulted several speakers of German, I do not think that this example is acceptable in the intended reading. .  Examples (37) and (38) improve a lot if the particle so (‘so’) is added in front of the host clause’s adjective, but, in my opinion, this modification results in a new construction type.

 Anke Holler

Reis (1997) further noticed that free dass-clauses give reasons for a presumption or an ­assessment expressed in the prior clause based on the fact that they denote, and that free dass-clauses are to some extent illocutionary independent of the preceding clause. For ­instance, in example (39a) the content of the free dass-clause cannot be part of the question expressed in the host clause. Uttering (39a) the speaker wonders whether Max is megalomaniac and additionally gives a reason for her question/wonder, viz. that Max bought another Porsche. Examples (39b) and (39c) illustrate contrary to this that a matrix clause’s question operator may scope over a complement dass-clause or a subordinate adverbial clause. (39) a. b. c.

Ist Max größenwahnsinnig, dass er jetzt noch.einen Porsche kauft? is Max megalomaniac that he now another Porsche buys ‘Is Max is megalomaniac to buy another Porsche now?’ Sagt Max, dass er jetzt noch.einen Porsche kauft? says Max that he now another Porsche buys ‘Does Max say that he is buying another Porsche now?’ Ist Max größenwahnsinnig, weil er jetzt noch.einen Porsche kauft? is Max megalomaniac because he now another Porsche buys ‘Is Max megalomaniac because he is buying another Porsche now?’

Moreover, it is generally possible to add expressions typically indicating illocutionary force (such as epistemic expressions, modal particles or performative indicators) to a free dass-clause as (40) shows. (40)

Max ist größenwahnsinnig, dass er {sicher/wohl/hiermit} Max is megalomaniac that he {certainly/well/hereby} noch.einen Porsche kauft. another Porsche buys ‘Max is megalomaniac to buy {certainly/well/hereby} another Porsche now?’

In a sense, free dass-clauses remind of V2 relative clauses in terms of their illocutionary force potential. To account for V2 relatives Gärtner (2002) assigns them an “assertional proto-force”, thereby claiming that V2 relatives share this kind of illocutionary force with similar “embedded root” constructions. Although Gärtner (2002), who has not studied free dass-clauses, establishes the hypothesis that assertional proto-force is triggered in embedded V2 declaratives only, I suppose that free dass-clauses are another clausal class having assertional proto-force since they show a certain independent ­assertionality being “parasitic” on the force of their matrix clause. If this assumption is correct, proto-force construal is not reducible to V2 phenomena anymore. I have to leave this interesting issue for further research.

3.5  Dependent verb second clauses German dependent verb second clauses (henceforth dependent V2 clauses) such as (41) have attracted attention in a number of studies.

German dependent clauses 

(41) Oskar denkt, er gewinnt das Spiel. Oskar thinks he wins the match ‘Oskar think that he will win the match.’

This clausal class has been recently discussed for instance by Vikner (1995), Reis (1997), Auer (1998), Steinbach (2007), Meinunger (2004), and Truckenbrodt (2006). The mentioned work reports unanimously that dependent V2 clauses similarly show properties of subordinate and main clauses. In the following, I briefly summarize the fundamental results. First of all, dependent V2 can be observed only in the complements of a certain set of predicates.18 Since the present article does not aim at classifying predicate groups that allow or do not allow for dependent V2 clauses, I simplify the matter a little bit by assuming that dependent clauses with V2 order are restricted to the class of bridge verbs.19 There is an ongoing debate as to whether the dependent V2 clause indeed syntactically functions as an object clause satisfying an argument of the matrix predicate. The main evidence speaking for its complement status is obvious: The matrix predicate remains otherwise unsaturated.20 In addition, a rising tone at the end of the matrix predicate signals that it awaits completion, as (42) exemplifies. (42) Sie hatte geglaubt.(/), er SCHLIEfe. she had believe he slept[SUBJ] ‘She had believed that he would sleep.’

.  Some nouns also allow for dependent V2 clauses, such as die Befürchtung, Max verspäte sich (‘The apprehension that Max is[SUBJ] late’). Lexically, these nouns are mostly related to verbs showing verb second as well. I have nothing to say on this issue in the present article. .  In their classical paper on the applicability of root transformations Hooper & Thompson (1973) already distinguished verb classes with respect to their ability to embed root phenomena. Classifications for German verbs combining with dependent V2 clauses that have been developed more recently are recapitulated in Meinunger (2004). .  Extraction data like (i) are usually adduced as further evidence for the argument status of V2 dependent clauses. Reis (1997), however, invalidates this argument by claiming that (i) is not a case of extraction, but results from parenthetical insertion of the matrix clause glaubt er (‘believes he’) into the dependent V2 clause. Even if one does not follow this argumentation, data like (ii) indicate that extraction out of dependent V2 clauses is not allowed unconditionally. I thank one of the anonymous reviewers for pointing me to the latter example. (i) Wo glaubt er wohnt man billig? where believe he lives one inexpensively ‘Where do you believe one can live inexpensively?’ (ii) *Wen glaubt er, Maria wird einladen? whom believe he Maria will invite

 Anke Holler

Reis (1997), however, seriously questions that dependent V2 clauses realise an argument of the matrix predicate in the usual way. She concludes this from the fact that the grammatical behaviour of dependent V2 clauses clearly deviates from the behaviour of verb final dass-clauses, whose complement status is proven beyond doubt. One of the various arguments Reis (1997) discusses is a topological one: Dependent V2 clauses are restricted to a final position within the matrix clause.21 Consequently, they cannot appear in subject or topic position. This is demonstrated by example (43), which takes advantage of binding facts. Unlike a verb final dass-clause appearing in initial position, a topicalised dependent V2 clause may not contain a pronoun bound by a quantifier in the matrix clause. (43) a. b. c.

Jederi möchte gern glauben, eri sei unheimlich beliebt. everyone want.to gladly believe he is[SUBJ] extremely popular ‘Everyone would like to believe that he is extremely popular.’ *Eri sei unheimlich beliebt, möchte jederi gern glauben. he is[SUBJ] extremely popular want.to everyone gladly believe ‘Everyone would like to believe that he is extremely popular.’ dass eri unheimlich beliebt ist, möchte jederi gern glauben that he extremely popular is want.to everyone gladly believe ‘That he is extremely popular, everyone would like to believe.’

Besides the aforementioned topological difference, dependent V2 clauses and verb ­final dass-clauses behave differently with respect to correlatives and und zwar-supplements, as the contrast in (44) illustrates. (44) a. b.

Hans hat (*es) geglaubt, Peter geht dahin zu Hans has it believed Peter goes there on ‘Hans believed Peter goes on foot there.’ Hans hat (es) geglaubt, dass Peter dahin zu Hans has it believed that Peter there on ‘Hans believed that Peter goes on foot there.’

Fuß. foot Fuß geht. foot goes

Whether dependent V2 clauses may be in scope of a negated matrix predicate is still an open empirical issue. Acceptable examples like (45) suggest that dependent V2 clauses may appear as complements of negated verbs. (45) Sie behauptet ja nicht, er sei Kommunist, [sondern ...]. she claims part not he is[SUBJ] communist but ‘She does not claim that he is a communist, but …’

.  An exception to this generalisation are V2 clauses that are dependent on nouns, e.g., Die Befürchtung, Max verspäte sich, war völlig unberechtigt. (‘The apprehension that Max is[SUBJ] late was completely unsubstantiated.’).

German dependent clauses 

On the other hand, data like (46) substantiate the widely accepted view that V2 is blocked under a negated matrix predicate. (46) *Er glaubt nicht, Maria möchte das Theorem beweisen. he believes not Maria wants.to the theorem prove ‘He doesn't believe Maria wants to prove the theorem.’

Conducting some corpus research, Auer (1998) observed that only a small group of verbs ­ egated. Mei(primarily verba dicendi) may license V2 in a dependent clause if they are n nunger (2004), mainly referring to corpus work done by Butulussi (1991), also ­presents a list of apparent counter examples to the conventional observation that V2 is ruled out under negation. Discussing the problem at length he in a way affirms Auer’s original result and additionally states that (i) only subjunctive mood can save a dependent V2 clause under negative elements, and that (ii) all of the listed negated matrix clauses “must not be interpreted as an assertion, but must be interpreted as a command to give up a certain belief.” (Meinunger 2004: 213). In the light of these very restrictive conditions, advocating the view that negation blocks V2 still seems to be justified.22 Although the facts discussed so far concerning order, supplements and negation point to a root-like behaviour of dependent V2 clauses, it should be noted that dependent clauses with V2 order also show typical properties of subordinate clauses. As I have mentioned earlier, their phonological shape as well as the thematic relationship between the matrix predicate and the dependent V2 clause are characteristic for a subordinate status. Binding facts provide even more positive evidence: Since binding is ­usually impossible across a sentence boundary, examples like (47) indicate that the dependent V2 clause must be structurally analysed as being deeper than the ­matrix clause; otherwise the pronoun could not be c-commanded by the quantifier and, hence, not be bound. (47) Fast jederi glaubt, eri sei der Beste. almost everyone believes he is[SUBJ] the best ‘Almost everyone believes that he is the best one.’

In addition to admitting variable binding from outside, dependent V2 clauses may appear as complements to predicates that are questioned, which is also in accordance with ordinary verb final dass-complement clauses, cf. (48). (48) a. b.

Behauptet sie, er sei Kommunist? claims she he is[SUBJ] communist ‘Does she claim that he is a communist?’ Behauptet sie, dass er Kommunist sei? claims she that he communist is[SUBJ] ‘Does she claim that he is a communist?’

.  The matter is even more complicated since dependent V2 clauses cannot generally be ­combined with negative predicates like bezweifeln (‘doubt’) either, cf. Auer (1998), Steinbach (2007). For reasons of limited space I will not go into this issue here.

 Anke Holler

The observation that the interrogative operator may take scope over a dependent V2 clause meets Reis’s claim that dependent V2 clauses may be information-structurally integrated into their matrix clause, meaning that the matrix clause and the dependent V2 clause are part of one and the same focus domain. In the face of the ­aforementioned phonological facts, this information-structural behaviour comes as no surprise. Much more sensation creates the interaction of verb second in German with the assignment of illocutionary force. Details apart, many researches have argued that the movement of the finite verb to the second position has its reason in the assignment of assertional illocutionary force, cf. e.g., Wechsler (1991), Lohnstein (2000), Gärtner (2002), Meinunger (2004), Truckenbrodt (2006). This theoretical assumption is based on the well-known empirical observation that dependent V2 clauses may be used to assert a certain proposition, cf. (49). (49) Emma glaubt, Max wählt rechts. Emma believes Max votes.for right ‘Emma believes that Max votes for right-wing candidates.’

It is widely accepted that two assertions can be made uttering (49): (i) Emma believes something, and (ii) Max votes for right-wing candidates. Thus, any account of dependent V2 clauses must explain this assertional force (potential). Summing up, the presented facts reveal that dependent V2 clauses show mixed properties in terms of the main-subordinate-clause distinction. Before turning to a formal analysis accounting for the grammatical behaviour of the discussed five clause types, let me give a short overview of the outlined facts.

3.6  Summary of the empirical results The clause types discussed in section 3.1 to 3.5 are various instances of dependent clauses that relate to their prior clause in a non-canonical way. Table 1 gives an overall picture of these facts in comparison to canonically linked clauses, i.e., standard complement clauses such as verb final dass complement clauses or adjunct clauses such as ordinary adverbial clauses or restrictive relative clauses.23 Table 1 makes perfectly clear that the behaviour of non-canonical clauses cannot be reduced to the position of the finite verb. This result rejects all existing constraint-based

.  It is not intended to exhaustively represent all existing German clause types. Table 1 rather summarises the subset of data discussed in the present article according to the criteria that have been applied. Apart from the listed properties the table hence does not make any specific assumption concerning the grammatical behaviour of canonical subordinate clauses. In particular, it says nothing about the relation between complement clauses on the one hand and adjunct clauses on the other hand.

German dependent clauses 

Table 1. Comparison of empirical properties of canonical and selected non-canonical dependent clauses in German Non-canonical clause types Canonical clause types Dep. V2 cl. Phonologically integrable Information structurally integrable Syntactic function Variable binding from outside Within scope of negation Topologically free Dependent illocutionary force

+ + + + + + +

+ + + + – – –

Free dass-cl. V2 rel. cl. + + – – – – –

Wh-rel. cl. Weil-V2 cl. – – – – – – –

approaches relying on the V2 property as a means for distinguishing main (independent) and subordinate (dependent) clauses. For instance, influential ­approaches such as Uszkoreit (1987), Kathol (1995), Netter (1998) and Ginzburg & Sag (2000) are explicitly based on this flawed assumption. In addition, it is striking that the relation between a dependent clause and its host clause may be differently tight, which confirms earlier generalisations saying that clause integration in German is not a clear-cut phenomenon but underlies a certain gradation. See, for instance, König & van der Auwera (1988), Lehmann (1988), ­Fabricius-Hansen (1992), and Reis (1997).

4.  A constraint-based approach to non-canonical clause linkage To account for the presented facts several theoretical proposals have been developed. One consists in exploiting prototype theory by making a distinction between prototypical and marginal cases of subordination (e.g., Fabricius-Hansen 1992). Another approach is to organise clause types in an elaborate taxonomy. Reis (1997), for instance, defines a set of grammatical and pragmatic criteria upon which she distinguishes three major classes of clauses: integrated clauses, i.e., all clauses being licensed by the matrix predicate and having hence a syntactic function in the matrix clause, and unintegrated clauses, which do not fulfil this criterion. Reis (1997) further subdivides the latter class into a subclass of ‘relatively’ unintegrated clauses consisting of free dass-clauses and dependent V2 clauses, and a second subclass of ‘absolutely’ unintegrated clauses comprising, for instance, continuous wh-relative clauses. A third solution, proposed inter alia by Lehmann (1988) and Peyer (1997), is to evaluate clause types with regard to a number of independent scalar parameters, and thus to analyse clause linkage as a

 Anke Holler

semanto-syntactic continuum. While being sympathetic with the idea to regard subordination as a composite term of various linguistic properties, I believe that none of the sketched approaches offers a satisfactory solution for the described non-canonical clause linkage phenomena in German. Therefore I will next develop a sort-driven constraint-based analysis couched in the framework of Head-driven Phrase Structure Grammar (Pollard & Sag 1994) that accounts for the data presented in this article.

4.1  Why is HPSG eligible for modelling degrees of clause linkage? The architecture of Head-driven Phrase Structure Grammar (henceforth HPSG) qualifies very well to capture a graded clause linkage. Firstly, all linguistic objects (i.e., words, subsentential phrases and sentences) are described as internally structured complex signs consisting of phonological, syntactic, semantic, discourse, and phrasestructural information, thereby positing multiple levels of structures. The linguistic information is represented using hierarchically ordered attribute-value pairs. A sign is assumed to have at least the attributes Phon(ology), Syn(tax)-Sem(antics), and (in the case of phrases) D(augh)t(e)rs. The Synsem value distinguishes Loc(al) and Nonloc(al) ­features. Whereas Nonlocal information is central in the analysis of unbounded ­dependencies, Loc information has in turn three attributes of its own called Cat(egory), Cont(ent), and Con(te)x(t). The Cat value includes categorical information of a word and its ­required grammatical arguments. The Cont value constitutes the word’s semantic contribution, and the Conx value contains context-dependent semantico-pragmatic information. This feature architecture on a par with a non-derivational perspective on grammar allows a simultaneous access to information stored at different structural levels.24 ­Secondly, a grammar cast in the framework of HPSG is formulated as a declarative system of constraints, which restrict the set of well-formed linguistic structures. Thirdly, the ­linguistic structures employed in HPSG are sorted, which means that they are labelled with a sort symbol. The finite set of sort symbols is partially ordered resulting in a multipleinheritance-hierarchy. Combining the latter two conceptual devices enables to model ­effects of constraint ranking, and, thus, to formulate restrictions that can be used to represent degrees of clause linkage. In addition, the sign-based setup of  HPSG allows extending the application of general grammatical constraints to more and more ­‘peripheral’ clause

.  Although there are essential differences in the nature of structural levels posited in HPSG on the one hand and in approaches influenced by Chomsky’s Government and Binding Theory and its descendants on the other hand, we can regard Phon and Dtrs as rough analogs of the GB levels phonetic form and S-structure; Cat plays a role roughly analogous to that of GB’s D-structure, and Cont is most closely analogous to the GB level of logical form. However, the Conx value, which contains linguistic information that bears on certain context-dependent aspects of semantic interpretation, does not map directly to a GB level.

German dependent clauses  phrase

independent

dependent

integrated fully-integr

non-integrated

weakly-integr weakly-non-integr fully-non-integr

Figure 1.  Partition of phrase in terms of the dimension Dependence. linkage phenomena without being forced to posit a bifurcation ­between ‘core’ and ‘peripheral’ constructions. In the following I will sketch an analysis based on the hypothesis that dependent clauses differ in the way they are combined with the preceding clause.

4.2  The proposed analysis Following Green (1996), I firstly assume that objects of sort phrase may be partitioned in terms of a Dependence dimension into independent and dependent objects, as depicted in figure 1. While the sort independent comprises all independently uttered sentences,25 the sort dependent describes all objects that are somehow attached to their linguistic ­surrounding applying to all (canonical and non-canonical) clausal types ­presented in table 1. ­Building on the empirical facts presented above, the sort dependent is ­further partitioned by two subsorts, which are called integrated and non-integrated. The sorts fully-integr(ated) and weakly-integr(ated) are in turn subsorts of the sort integrated, ­whereas the sorts fullynon-integr(ated) and weakly-non-integr(ated) are subsorts of the sort ­non-integrated. By formulating restrictions on these sorts we can easily model the empirical result that dependent clausal objects differ in the degree they are integrated into a prior clause.26 It is thereby assumed that the sort fully-integrated describes ­canonical dependent clauses and the sort weakly-integrated dependent V2 clauses. The sort weakly-non-integrated represents

.  The class of objects of this sort also includes independent verb final clauses such as Ob er noch kommt? (‘I wonder whether he will still come?’), which cannot be investigated in this article. See for instance Oppenrieder (1991) for discussing this phenomenon. .  Unfortunately, it cannot be discussed in this article to which extent this distinction can be used for constituents other than clauses. At least, there is evidence from German and English that nominal left-peripheral elements also need to be classified regarding their degree of (non-) integrateness into a clause, cf. Shaer & Frey (2004).

 Anke Holler

free dass-clauses as well as V2 relative clauses. Last but not least, the sort fully-non-integrated accounts for continuative wh-relative clauses and weil-V2 clauses.

4.2.1  Clauses of sort integrated To begin with, clauses of sort integrated share the property of being related to the prior clause by a selection or modification relation as they are either subcategorized by the matrix predicate (complement clauses, dependent V2 clauses) or function as an adjunct to it (adverbial clauses). In other words, these clauses are ­embedded into their preceding clause. Assuming that canonical subordinate clauses of sort f­ ully-integrated are analysed in a standard way (see e.g., Pollard & Sag 1994), I shall say nothing in particular about them. I will rather concentrate on the analysis of non-canonical types of dependent clauses starting with clausal phrases of sort weakly-integrated, i.e., dependent V2 clauses. Apart from the V2 property27, they differ in one major aspect from canonical complement clauses: the assertional (proto-) force assigned to them. This is subject to the constraint shown in figure 2 which says that all clauses of sort weakly-integrated must contain an assert relation in their B(ac)kgr(ound) set.28 weakly-integrated Æ SYNSEM | L OC

C AT | H EAD v2 CONX | BKGR {assert_rel}

Figure 2.  Constraint restricting clauses of sort weakly-integrated (i.e., dependent V2 clauses).

On the other hand, constraint inheritance guarantees that dependent V2 clauses behave like ordinary complement clauses in receiving a theta-role from the matrix predicate since the sort weakly-integrated is defined as subsort of sort integrated. That dependent V2 clauses are outside of the scope of negation is modelled as a lexical property of negative particles requiring that they do not take scope over clausal phrases of sort weakly-integrated. The peculiar topological behaviour of dependent V2 clauses is formalised by linear precedence rules. I will come back to both issues later.

4.2.2  Clauses of sort non-integrated

Since the empirical results show that clauses of sort non-integrated do not fulfil a syntactic function in the prior clause and are not in its binding domain, they are neither a complement

.  For the sake of simplicity I say nothing specific about the way the V2 property is modelled in HPSG. For a sort-driven proposal to represent verb final and verb second clauses see, for instance, Holler (2005). The analysis presented here is, however, compatible with any other ­account to verb placement in German. .  The Background feature is defined for objects of sort context. This sort represents the value of the Context feature modelling discourse and pragmatic information. The value of Background is a set of parametrized states of affairs. See Green (2000) for a proposal for ­structuring the Background set to appropriately treat illocutionary force.

German dependent clauses 

nor an adjunct of the matrix predicate. Thus, I assume at first that clauses of sort weaklynon-integrated modify the complete CP representing the prior clause. Correspondingly, this is realised by restricting the Mod(ified) value of objects of this sort to CPs, as the constraint in figure 3 shows. This constraint additionally states that assertional (proto-) force must be assigned to clauses of sort weakly-non-integrated, which is again achieved by manipulating the Background set.

weakly-non-integrated Æ SYNSEM | L OC

C AT

H EAD

verb M OD CP

S UBCAT 〈〉 CONX | BKGR { assert_rel }

Figure 3.  Constraint restricting clauses of sort weakly-non-integrated (i.e., V2 relative clauses, free dass-clauses).

To account for clauses of sort fully-non-integrated, I adapt an approach to ­English peripheral adverbials by Haegeman (1991). Haegeman argues that peripheral ­adverbials can best be treated as being outside the syntactic representation of the ­sentences that they modify. Elaborating on a proposal for non-restrictive relative ­clauses made by ­Safir (1986), she pursues the hypothesis that peripheral adverbials are syntactically ­unattached “orphan constituents”. Under this view orphan constituents are neither ­attached at D- or S-structure nor at logical form. Nevertheless, syntax may ­constrain the interpretation of orphan constituents. For instance, Haegeman (1991) suggests that ­peripheral ­adverbials have to be coindexed with a sentential modifiee. She further claims that the interpretation of orphan constituents must be dealt with on the basis of a theory of ­utterance ­interpretation. I propose to analyse clauses of sort ­fully-non-integrated as orphan ­constituents in the terms of Haegeman (1991). The ­sign-based architecture of HPSG described in section 4.1 integrating ­grammatical, pragmatic and phrase-structural ­information as well as HPSG’s representational ­formalism employing sorted feature structure ­descriptions are best suited for implementing such an orphan approach. It ­allows to built constituents that are not ­syntactically but rather (discourse-)interpretationally linked. Such constituents can be conceived of suprasentential units combining sentential constituents that, although syntactically unattached, are ­interdependent on the level of utterance ­interpretation. Under this view, the highest unit of linguistic description is not the sentence (i.e., a CP or an object of sort clause), but a minimal ­discourse unit paratactically linking sentential phrases. Units of this kind have been stipulated before, for instance by Banfield (1973), Fabb (1990) or Gärtner (2001). Due to the HPSG setup the hypothesis outlined here, however, differs from earlier ­approaches in not requiring that there is a syntactic phrase marker immediately ­dominating both sentential ­constituents. It is an asset of the HPSG formalism that it allows within one object of sort sign, i.e., a linguistic object, establishing a discourse-structural relation without being forced to ­simultaneously ­establish a syntactic ­relation. The just described fundamental idea can be formally realised in three steps. In the first step, phrases are distinguished in the dimension Headedness whether they have

 Anke Holler phrase

hd-phrase hd-nexus-phrase

non-hd-phrase

hd-fill-phrase host-orphan-phrase […]

Figure 4.  Part of the partial sort hierarchy in the dimension Headedness.

a head or not (following Ginzburg & Sag 2000), and for phrases of sort non-h(eade)dphrase a subsort called host-orphan-phrase is stipulated.29 See figure 4. In the second step, the constraint depicted in figure 5 is formulated. It defines ­fully-non-integrated clauses30 as part of a phrase of sort host-orphan-phrase and ­restricts this phrase in an appropriate way. The constraint under discussion specifies that a fully-nonintegrated clause figures as an orphan daughter. It additionally states that orphan constituents do not syntactically modify another phrase because the orphan daughter’s Mod feature is specified by the value none, and the host daughter’s Subcat list is required to be saturated (empty). Thus, there is no clause to which the orphaned clause may attach ­syntactically. However, there is a discourse-structural relation that is established between the host and the orphan constituent. This is expressed by structure sharing the Content value of the host daughter with an element of the Background set of the orphan daughter. The semantics of the host hence acts as the contextual background of the orphan constituent.31

.  I leave it open which further subsorts of non-headed-phrase should be defined. .  With Uhmann (1998) I assume that two lexical entries for weil (‘because’) must be ­stipulated. In Holler (2005) I have argued that continuative wh-relative clauses are projected from a peculiar non-restrictive relativizer. Note that these two elements trigger the value f­ ully-non-integrated for weil-V2 clauses and continuative wh-relative clauses, respectively. .  One reviewer raised the question of how the proposed discourse-based analysis may differentiate between example pairs like (i) Peter kommt zu spät, weil er hat keinen Parkplatz gefunden. (‘Peter is late because he could not find a parking space.’) vs. Peter kommt zu spät. Er hat nämlich keinen Parkplatz gefunden. (‘Peter is late. He namely could not find a parking space.’), and (ii) Emma gewann die Schachpartie, was Oskar ärgerte. (‘Emma won the chess match, which annoyed Oskar.’) vs. Emma gewann die Schachpartie. Das ärgerte Oskar. (‘Emma won the chess match. This annoyed Oskar.’) because each second clause of the four example constructions in (i) and in (ii) relates discourse-functionally to the prior clause in a similar way. This, however, is no objection against the proposed analysis for one important reason: in both example pairs, the sort dependent is assigned only to the latter clause of the complex clause construction, i.e., the first construction, but not to the latter clause of the sequence of clauses, i.e., the second construction. Hence, only the first construction in (i) and (ii) may figure as an object of sort host-orphan phrase, and is thus described as a sign. The sequence of clauses, however, must be described by

German dependent clauses 

In the third step, a constraint on objects of sort fully-non-integrated is assumed positing that all clauses of this sort contain an assert relation in their Background set, and thus may have illocutionary force, cf. figure 6.

HOST-DTR | SS | LOC

CAT

HEAD verb SUBCAT 〈〉

CONT 1

host-orphan-phrase Æ

fully-non-integrated ORPH-DTR

SS | LOC

CAT | HEAD| MOD none CONX | BKGR { 1 , assert_rel}

Figure 5.  Constraint defining clauses of sort fully-non-integrated (i.e., weil-V2 clauses, continuative wh-relative clauses) as orphan constituents.

fully-non-integrated Æ

SYNSEM | LOC

CAT | HEAD

verb {assert_rel}

CONX | BKGR

Figure 6.  Constraint restricting clauses of sort fully-non-integrated (i.e., weil-V2 clauses, ­continuative wh-relative clauses).

If we now reinspect the facts concerning the scope of negation, we can generalise that a negative particle may only scope over clauses of sort fully-integrated, as can ­easily be specified in the lexicon. Last but not least let me comment on a further advantage of the sort-driven ­approach as pursued here. Employing linear precedence rules, it allows accounting for the observed topological facts in a plausible way. See for instance the following set of rules: i. Objects of sort weakly-non-integrated must precede objects of fully-non-integrated. ii. Objects of sort fully-integrated must precede objects of weakly-integrated. iii. Objects of sort integrated must precede objects of non-integrated.

two separate sign objects. Formally, this different treatment is achieved by the lexical specification of the introducing elements (weil (‘because’) in weil-V2 clauses and a wh-relativizer in continuative wh-relatives) in each dependent clause of the complex clause construction in (i) and (ii). Since these elements are not present in the sequence of clauses, in these cases each clause is separately assigned the sort independent. Furthermore, that a dependent clause may be linked to a preceding clause by one and the same discourse function as underlying two subsequent independent clauses is not exceptional and hence says nothing about the dependence status. Compare for instance Maria weint. Max hat ihr Spielzeug weggenommen. (‘Maria screams. Max has taken away her toys.’) with Maria weint, weil Max ihr Spielzeug weggenommen hat. (‘Maria screams because Max has taken away her toys.’)

 Anke Holler

In a nutshell, the proposed analysis has three essential ingredients: (i) Clauses are differentiated in terms of their dependence on a putative matrix clause. (ii) Dependent clauses that link non-canonically to their prior clause are partitioned into three sets, each labelled by a specific sort. (iii) By constraining these sorts the empirical properties of the discussed clause types are modelled non-derivationally, thereby realising effects of graded clause linkage.

5.  Conclusion Taking German as the main example, I have argued that dependent clauses must be ­distinguished regarding their degree of integration into a putative matrix clause. This has been empirically shown by means of five instances of non-canonical clause types: weil-verb second clauses, continuative wh-relative clauses, verb second relative clauses, free dass-clauses, and dependent verb second clauses. Accounting for the presented data, a constraint-based analysis has been developed that formulates restrictions on ­partially ordered sorts, which are assigned to certain clause types. In accordance with the facts, the analysis does not depend on the position of the finite verb, and hence ­rejects earlier constraint-based approaches that implement a strict dichotomic ­distinction ­between main and subordinate clauses relying only on the V2 property. In particular, the ­presented analysis allows distinguishing the (in)dependence relation from other grammatical factors contributing to the multidimensional phenomenon of subordination.

Acknowledgements The paper is a considerably revised version of Holler (2005). I am grateful to the audience of the HPSG 2005 conference in Lisbon and the International CoGETI workshop on constraint-based grammar 2005 in Bremen for inspiring discussions. I also thank Cathrine Fabricius-Hansen and Wiebke Ramm for giving me the opportunity to present the core ideas of this paper at the workshop “‘ Subordination’ vs. ‘Coordination’ in Sentence and Text” at the 2006 DGfS meeting in Bielefeld and the audience there for helpful comments. I would like to thank furthermore Hardarik Blühdorn, Danièle ­Godard and three anonymous reviewers for critical remarks and stimulating ­suggestions on an earlier version of this paper. All errors and shortcomings are mine.

References Altmann, Hans. 1997. Verbstellungsprobleme bei subordinierten Sätzen in der deutschen Sprache. In Syntax im Fokus. Festschrift für Heinz Vater, Christa Dürscheid, Karl Heinz Ramers & Monika Schwarz (eds), 69–84. Tübingen: Niemeyer. Auer, Peter. 1998. Zwischen Parataxe und Hypotaxe: ‘abhängige Hauptsätze’ im gesprochenen und geschriebenen Deutsch. Interaction and Linguistic Structures 2.

German dependent clauses  Banfield, Ann. 1973. Narrative Style and the grammar of direct and indirect speech. Foundations of Language 10: 1–39. Blühdorn, Hardarik. Forthcoming. Epistemische Lesarten von Satzkonnektoren – Wie sie ­zustande kommen und wie man sie erkennt. To appear in Semantik vs. Pragmatik oder Semantik und Pragmatik?, I. Pohl (ed.). Tübinjen: Niemeyer. Brandt, Margareta. 1990. Weiterführende Nebensätze. Stockholm: Almquist & Wiksell International. Butulussi, Eleni. 1991. Studien zur Valenz kognitiver Verben im Deutschen und Neugriechischen. Tübingen: Niemeyer. Emonds, Joseph E. 1970. Root and Structure Preserving Transformations. PhD dissertation, MIT. Fabb, Nigel. 1990. The difference between English restrictive and non-restrictive relative clauses. Journal of Linguistics 26: 57–78. Fabricius-Hansen, Cathrine. 1992. Subordination. In Deutsche Syntax. Ansichten und Aussichten, Jahrbuch des IDS 1991, L. Hoffmann (ed.), 458–483. Berlin: Walther de Gruyter. Gärtner, Hans-Martin. 2001. Are there V2 –.relative clauses in German? Journal of Comparative Germanic Linguistics 3(2): 97–141. Gärtner, Hans-Martin. 2002. On the force of V2 declaratives. Theoretical Linguistics 28: 33–42. Ginzburg, Jonathan & Sag, Ivan A. 2000. Interrogative Investigations: The Form, Meaning, and Use of English Interrogatives. Stanford CA: CSLI. Green, Georgia M. 1976. Main clause phenomena in subordinate clauses. Language 52: 382–397. Green,  Georgia M. 1996. Distinguishing main and subordinate clause: The root of the problem. Unpubl. Ms., University of Illinois. (URL: http://www.linguistics.uiuc.edu/g-green/ marseille.ps) Green, Georgia M. 2000. The nature of pragmatic information. In Grammatical Interfaces in HPSG, R. Cann, C. Grover & P. Miller (eds), 113–138. Stanford CA: CSLI. Haegeman, Liliane. 1991. Parenthetical adverbials: The radical orphanage approach. In Aspects of Modern English Linguistics, S. Chiba, A. Ogawa, Y. Fuiwara, N. Yamada, O. Koma & T. Yagi (eds), 232–254. Tokyo: Kaitakusha. Haiman, John & Thompson, Sandra A. 1984. ‘Subordination’ in Universal Grammar. In Proceedings of the Tenth Annual Meeting of the Berkeley Linguistics Society, C. Brugman et al. (eds), 520–523. Berkley CA: Berkeley Linguistics Society. Holler, Anke. 2005. Weiterführende Relativsätze. Empirische und theoretische Aspekte. Berlin: Akademie Verlag. Hooper, Joan & Thompson, Sandra A. 1973. On the applicability of root transformations. ­Linguistic Inquiry 4: 465–497. Kathol, Andreas. 1995. Linearization-Based German Syntax. PhD dissertation, Ohio State ­University. Keller, Rudi. 1995. The epistemic weil. In Subjectivity and Subjectivisation. Linguistic Perspectives, D. Stein & S. Wright (eds), 16–30. Cambridge: CUP. König, Ekkehard & van der Auwera, Johan. 1988. Clause integration in German and Dutch conditionals, consessive conditionals, and concessives. In Clause Combining in Grammar and Discourse, J. Haiman & S.A. Thompson (eds), 101–133. Amsterdam: John Benjamins. Lehmann, Christian. 1988. Towards a typology of clause linkage. In Clause Combining in ­Grammar and Discourse, J. Haiman & S.A. Thompson (eds), 181–225. Amsterdam: John Benjamins. Lohnstein, Horst. 2000. Satzmodus – kompositionell. Zur Parametrisierung der Modusphrase. Berlin: Akademieverlag. Longacre, Robert. 1996. The Grammar of Discourse. New York NY: Plenum.

 Anke Holler Meinunger, André. 2004. Verb position, verbal mood and the anchoring (potential) of sentences. In Syntax and Semantics of the Left Periphery, H. Lohnstein & S. Trissler (eds), 313–341. Berlin: Mouton de Gruyter. Netter, Klaus. 1998. Functional Categories in an HPSG for German. PhD dissertation, German Research Center for Artificial Intelligence (DFKI) & University of Saarland. Oppenrieder, Wolfgang. 1991. Von Subjekten, Sätzen und Subjektsätzen. Untersuchungen zur Syntax des Deutschen. Tübingen: Niemeyer. Pasch, Renate. 1997. Weil mit Hauptsatz – Kuckucksei im denn-Nest. Deutsche Sprache 25(3): 252–271. Pasch, Renate, Brauße, Ursula, Breindl, Eva & Waßner, Ulrich H. 2003. Handbuch der deutschen Konnektoren. Linguistische Grundlagen der Beschreibung und syntaktische Merkmale der deutschen Verknüpfer (Konjunktionen, Satzadverbien und Partikeln). Berlin: de Gruyter. Peyer, Ann. 1997. Satzverknüpfung: syntaktische und textpragmatische Aspekte. Tübingen: ­Niemeyer. Pollard, Carl & Sag, Ivan. 1994. Head-Driven Phrase Structure Grammar. Stanford: CA CSLI. Reis, Marga. 1997. Zum syntaktischen Status unselbständiger Verbzweit-Sätze. In Syntax im Fokus. Festschrift für Heinz Vater, C. Dürscheid, K.H. Ramers & M. Schwarz (eds), ­121–144. Tübingen: Niemeyer. Safir, Ken. 1986. Relative clauses in a theory of binding and levels. Linguistic Inquiry 17(4): 663–689. Santorini, Beatrice. 1992. Variation and change in Yiddish subordinate clause word order. ­Natural Language and Linguistic Theory 10: 595–640. Shaer, Ben & Frey, Werner. 2004. Integrated and non-integrated leftperipheral elements in ­German and English. ZAS Papers in Linguistics 35(2), 465–502. Steinbach, Markus. 2007. Integrated parentheticals and assertional complements. In parentheticals, N. Dehé & Y. Kavalova (eds), 53–87. Amsterdam: Benjamins. Sweetser, Eve. 1990. From Etymology to Pragmatics. Metaphorical and Cultural Aspects of Semantic Structure. Cambridge: CUP. Truckenbrodt, Hubert. 2006. On the semantic motivation of syntactic verb movement to C in German. Theoretical Linguistics 32–3: 257–306. Uhmann, Susanne. 1998. Verbstellungsvariation in weil-Sätzen. Zeitschrift für Sprachwissenschaft 17(1): 92–139. Uszkoreit, Hans. 1987. Word Order and Constituent Structure in German. Chicago IL: Chicago University Press. Van Valin, Robert. 1984. A typology of syntactic relations in clause linkage. In Proceedings of the Tenth Annual Meeting of the Berkeley Linguistics Society, C. Brugman et al. (eds), 542–558. Berkley CA: Berkeley Linguistics Society. Vikner, Sten. 1995. Verb-Movement and Expletive Subject in the Germanic ­Languages. New york: Oxford University Press. Wechsler, Stephen. 1991. Verb second and illocutionary Force. In Views on Phrase Structure, K. Leffel & D. Bouchard (eds), 177–191. Dordrecht: Kluwer. Wegener, Heide. 1993. Weil – das hat schon seinen Grund. Zur Verbstellung in Kausalsätzen mit weil im gegenwärtigen Deutsch. Deutsche Sprache 21: 289–305.

To the right of the clause Right dislocation vs. afterthought* Maria Averintseva-Klisch University of Tübingen

In my paper I will introduce two different, although apparently similar constructions at the right sentential edge in German: right dislocation (RD) and afterthought (AT). I show that RD is a discourse-structuring device: it marks the topic for the following discourse segment. AT, on the contrary, is a local repair strategy. I assume that a syntactically independent repair like AT and a syntactically adjoined discourse-structuring device like RD are cross-linguistically two options of the right sentential edge, and test this assumption for French and Russian. Keywords: Right dislocation, discourse topic, afterthought, discourse relation

1.  Introduction ‘German right dislocation’ is since Altmann (1981) a term for a construction consisting of an NP1 at the right edge of the clause (i.e., after the formal completion of the clause) and a coreferent intraclausal pro-form, as in (1): (1) Hast Du ihn schon gesehen, (ich meine) den Karl? have you himi already seen, (I mean) the Karli (Schindler 1995: 44)

*  This paper emerged as part of my doctoral research conducted first at the DFG-funded graduate school “Economy and Complexity in Language” at the Humboldt University Berlin, and then at the University of Tübingen. I would like to thank my supervisor Claudia Maienborn (Tübingen), Manfred Consten (Jena), Mareile Knees (Jena) und Fabienne Salfner (Berlin) for numerous valuable comments on my work, the organizers and participants of the AG “ ‘Subordination’ versus ‘coordination’ in sentence and text – from a cross-linguistic perspective” for stimulating discussions, and Cathrine Fabricius-Hansen as well as three anonymous reviewers for their helpful remarks on the first draft of this paper. .  Altmann (1981) also mentions PP- and CP-right dislocations; these are not considered in this paper. See, however, Averintseva-Klisch & Salfner (2007) for PP-RD.

 Maria Averintseva-Klisch

Traditional analyses of the German right dislocation assume that this is a construction of spoken German that serves to resolve a potentially unclear pronominal reference (Altmann 1981; Auer 1991; Schindler 1995; Selting 1994; Uhmann 1993; 1997). The speaker of (1) notices that the use of the pronoun ihn (‘him’) is unclear in the context (it might be that there are either several referents that are equally suitable as an antecedent, or no referent is activated highly enough for the reference of the pronoun to be resolved effortlessly by the addressee). This observation causes the speaker to provide as an “afterthought” information that should make it easier to identify the intended referent. However, there are abundant cases where a right dislocation cannot serve the reference clarification, as the reference is pretty clear, cf. (2): (2) a.

(“Der The “Da There

Taifun!” rief Lukas dem Kapitän zu. typhoon.mask called Lukas the captain to ist er!”) Ja, da war er, der Taifun. is he Yes there was hei the typhooni (Michael Ende, Jim Knopf und die Wilde 13: 190)

b.

(Den Tag, den vergess‘ ich nicht,) (That day, d-pron forget I not) der war viel zu schön, der Tag. d-proni was much too wonderful the dayi.

(Altmann 1981:129)

Here the right dislocation marks the referent of the NP (the typhoon in (2a) and the day in (2b)) as being especially important for the discourse. To be more exact, the referent is marked as the discourse topic in the sense of aboutness (see section 2.2.). I argue that (1) and (2) are instances of two functionally and formally different constructions that have been subsumed under the label of right dislocation. I name these constructions ‘right dislocation proper’ (RD), cf. (2), and ‘afterthought’ (AT), cf. (1), following the distinction made for other languages, e.g., in Ziv (1994) and Ward & Birner (1996) for English and Fretheim (1995) for Norwegian. In this paper, I will first dwell upon the functional and formal differences between RD and AT in German. In section 2, I will briefly introduce prosodic and syntactic features of RD and AT (section 2.1.), and then turn to their respective discourse functions (section 2.2.). I will propose that AT is a syntactically independent ‘orphan’ (Haegeman 1991), that gets introduced into the discourse via a discourse relation Afterthought formulated for this case (section 2.3.). As for RD, it is syntactically integrated into its host sentence, and serves to mark the discourse topic for the following discourse segment. In section 3, I will address the general issue of the use of the right clausal edge.2 I will

.  I use the expression “right clausal edge” here and later to refer to constructions coming after the syntactic completion of the clause irrespectively of the fact whether they really pertain syn-

To the right of the clause 

argue with data from German, Russian and French that RD (i.e., a syntactically integrated global discourse-structuring device) and AT (i.e., a syntactically non-integrated local repair) are two cross-linguistically possible options of the use of the right clausal edge. Finally, in section 4, I will sum up and draw some conclusions.

2.  RD vs. AT: Form and function In this section, I introduce the prosodic and syntactic differences between RD and AT. Many of the facts presented here have been already pointed out in the previous research, e.g., in Altmann (1981), but have mostly been done up with as exceptions from a certain pattern of a so-called “right dislo­cation”. However, distinguishing between RD and AT allows to dispence with most exceptions, and to describe distinct patterns for RD and AT instead. The formal differences between RD and AT suggest that RD is prosodically and syntactically part of its host sentence, while AT is not. After describing these differences I will specify the discourse functions of RD and AT which have been mentioned above.

2.1.  RD vs. AT: Prosodic and syntactic differences RD is prosodically integrated into its host sentence (3a), i.e., it continues the tone movement of the host sentence and thus does not build a prosodic unit (an intonational phrase in terms of Selkirk (1978)) of its own, whereas AT builds an intonational phrase (optionally divided from the clause by a pause) with a tone movement and a clause-like accent of its own (cf. Uhmann 1997), (3b)3: (3) a. b.

[Ich MAG siei nicht, die Brigittei]. [Ich MAG siei nicht], | [die BriGITtei]. Ilike her not, the Brigitte. ‘I don’t like her, Brigitte.’ (|: pause; [ ]: intonational phrase; CAPITALS: primary accent)

RD AT

Crucially, prosody is only one of several criteria for distinguishing between RD and AT. Prosodic differences go along with syntactic differences, which are briefly listed

tactically to the clause (being base-generated or moved right IP-adjuncts) or whether they are syntactically autonomous. That is why I do not use the term “right periphery” here, as this term implies that a right-peripheral construction belongs syntactically to its host sentence. “Right clausal edge” in this sense covers right peripheral items as well as autonomous afterthought additions, which are syntactically not part of the host sentence, as I argue below. .  See Fretheim (1995) for a similar analysis for Norwegian: he shows that in Norwegian, as in German, prosodically integrated structures are RDs, and prosodically non-integrated ones ATs.

 Maria Averintseva-Klisch

below; they all suggest that RD belongs in a much more straightforward way to its host sentence than AT. I. Strict morphological agreement (in case, gender and number) between the clause-internal pro-form and the NP is necessary for RD, while at least gender agreement is only optional for AT, cf. (4) vs. (5): (4)

(Ach ja, mein Nachbar! Er hat gestern wieder einen Wutanfall bekommen, nur weil die Kinder im Garten gespielt haben. ) ‘(Oh yes, my neighbour! He went completely berserk yesterday, only because the children were playing in the garden.’) Ich kann ihn nur bedauern, den Mann /        RD I can himakk.mask only regret the manakk.mask / *die Giftspritzei / *der Manni. *the spitfireakk.fem / *the mannom.mask

(5) a. b.

Esi ist ein bisschen gleichförmig, | deine Melodiei.4 it.neutr is a bit homogeneous your melody.fem AT (Attested oral data) Und dann passierte das Unglücki, (ich meine) And then happened [the misfortune]neutr (I mean) AT dieser schreckliche Autounfalli / ich meine [this terrible traffic.accident]mask.nom / I mean diesen schrecklichen Autounfalli.5 [this terrible traffic.accident] mask.akk

II. A subordinate clause between the clause-internal pro-form and the NP is not possible for RD and possible for AT, cf. (6): (6) Ach, diese Münchner Stadtväter! a. Es könnte ihneni ja wirklich aufgefallen sein, it could themi yes really standing.out be

RD

.  It is not quite clear, whether es and the AT-NP deine Melodie really corefer, or whether the neutral weak personal pronoun es here is a so-called “anticipatory es”, which refers in a very abstract way (Smith 2002: 95). In the latter case coreference in the strict sense of the word does not take place. For RD in any case coreference between the RD-NP and the pro-form is required. Thus, irrespective of the fact whether grammatical incongruence corresponds to noncoreference here or not, (5a) is only possible as AT, and not as RD. .  If an AT is introduced with the addition ich meine (‘I mean’), the case of the AT-NP can vary between nominative and accusative. Altmann (1981) argues against the assumption that the verb meinen (‘to mean’) constituting the part of the addition is really able to govern the accusative case, so that the case variation cannot be explained with the case marking through the verb.

To the right of the clause 

b.

daß dazu das Geld fehlt, *den Münchner Stadtväterni. that for.it the money lacks *the of.Munich city.fathersi Es könnte ihneni ja wirklich aufgefallen sein, daß dazu it could themi yes really standing.out be that for.it das Geld fehlt,| ich meine (damit) die M. Stadtv.i. AT the money lacks I mean (with.it) the of.M. city.fathersi (Altmann 1981: 115)

‘Oh, these city fathers of Munich! Theyi really could have noticed, that the money for this purpose is lacking, (I mean) these city fathers of Munich / *these city fathers of Munichi!’

Furthermore, optional additions (ich meine (‘I mean’), also (‘that is’), tatsächlich (‘really’) etc.) between the clause-internal pro-form and the NP are possible for AT and not for RD, cf. (7) vs. (8):6 (7) (“Der Taifun!” rief Lukas dem Kapitän zu. „Da ist er!“) ‘“The typhoon!” Lukas called to the captain. “Here it comes!”’ Ja, da war er, (*ich meine /*also/*tatsächlich) der Taifun. the typhooni yes, there was iti (*I mean /*that.is/*really) (8) (Lisa und Melanie haben sich gestritten.)



RD

‘Lisa and Melanie quarrelled.’

Dann ist sie weggelaufen, | then is shei run.away

(ich (I

meine / also) Lisa. mean / that.is) Lisai

AT

III. The NP is not tied to the right-peripheral position in the case of AT, but can have a fairly free position in its host sentence, while RD is only possible at the right periphery, cf. (9) vs. (10): (9) a. b. c.

Ich habe ihn gestern nur mit Mühe wiedererkannt, | ich I have him yesterday only with effort recognized, I meine den Peter. AT mean the Peter Ich habe ihn, | ich meine den Peter, | gestern nur mit Mühe wiedererkannt. Ich habe ihn gestern | ich meine den Peter, | nur mit Mühe wiedererkannt. ‘I hardly recognized him yesterday, I mean Peter.’

.  As the examples show, the (im)possibility of additions with RD and AT is not due to the meaning and function of the addition, as one might be tempted to believe in the case of ich meine / also (‘I mean’ / ‘that is’), which are additions explicitly assisting the reference-clarifying function of AT. Also such additions like natürlich (‘of course’), tatsächlich (‘really’) etc., which are insensitive to the functional difference between RD and AT, are bad with RD and perfectly acceptable with AT. Thus, this difference seems to be a syntactic one.

 Maria Averintseva-Klisch

(10) Den Tag, den vergess’ ich nicht, the dayi d-proni forget I not a. deri war viel zu schön, der Tagi. d-proni was much too beautiful the dayi (Altmann 1981: 129) b. *deri, der Tagi, war viel zu schön. RD *d-proni the dayi was much too beautiful ‘That day, I will never forget it, it was so beautiful, the day.’

To summarize: there is ample evidence that RD belongs prosodically and syntactically in a much more straightforward way to its host sentence than AT. Prosodically, RD is a part of its host sentence’s tone contour. Con­sidered syntactically, morphological agreement of the RD-NP with the clause-internal pro-form suggests that NP is part of the clause, as morpho­lo­gical agreement is a sentence-bound phenomenon.7 Moreover, RD occu­pies a fixed position in the host sentence (at its right periphery), and does not allow insertions (neither subordinate clause insertion nor optional additions of any kind) between the host sentence and the RD-NP. The former constraint, i.e., not allowing subordinate clause insertion, is since Ross (1967) known in the literature as “Right Roof Constraint”: Ross assumes rightward movement to be possible, but never across a sentence boundary. For reasons of length and thematic contiguity of this paper I am not going to discuss the issue whether RD is a product of syntactic movement or base-generated at the right periphery.8 I use the term ‘right dislocation’ without implying any syntactic analysis, but rather as it is the term generally accepted in the literature for this construction, cf. e.g., Ziv (1994), Fretheim (1995), Ward & Birner (1996), Lambrecht (2001). However, RD in German seems to be island-sensitive, cf. (11), which might be considered an argument for the movement analysis: (11) Über Meyer1 werden öfters Gerüchte verbreitet. Mal soll er1 Beziehungen zur Regierung haben, mal werden seine1 Erfolge mit Kontakten zur Mafia erklärt. ‘There are lots of rumours about Meyer. One hears that he is bound to have connections to the government, or even that his success is due to his contacts to the mafia.’

.  Consten (2004: 91) shows that intersentential anaphoric resumption is generally possible without gender agreement, whereas it is impossible for intrasentential anaphora. .  I am not aware of any detailed syntactic analysis of German RD, but see Ross (1986/1967), Kayne (1994), and Culicover / Jackendoff (2005), to name but a few, for different possibilities of syntactic analysis of RD in general. Ross assumes that RD is a result of a rightward movement, whereas for Kayne RD emerges through the movement of the remaining clause to the left of the RD-NP. Villalba (2000) modifies this analysis for RD in Catalan slightly, still assuming that RD involves the leftward movement of its host clause. In contrast, Culicover / Jackendoff (2005) advocate the base generation of the RD to the right of its host sentence.

To the right of the clause 

a. *Ich habe schon oft [NP das Gerücht, dass ihm1 die Mafia I have already often [NP the rumour that him1 the mafia geholfen hat] gehört, dem Meyer1. helped has] heard the Meyer1 b. Ich habe schon oft [NP das Gerücht ta] gehört, [dass ihm1 I have already often [NP the rumour ta] heard [that him die Mafia geholfen hat]a, dem Meyer1. the mafia helped has]a the Meyer ‘I have already often heard the rumour, that the mafia has assisted him, this Meyer.’

Complex NPs being movement islands in German might explain the ill-formedness of (11a) as opposed to (11b). In this paper, I solely assume that RD is syntactically part of its host sentence, i.e., it belongs to the ‘Nachfeld’ (“afterfield”) in terms of Zifonun, Hoffmann & Stecker (1997). For the purposes of this paper I assume that RD is a right IP-adjunct.9 As far as the structural position of the clause-internal pro-form is concerned, RD does not impose any additional placement constraints on it (to the structural position of unstressed personal pronouns in general see Abraham 2007: 185 ff.)10 AT, on the contrary, can vary its position in its host sentence. Furthermore, AT does not strictly require morphological agreement between the NP and the clauseinternal pronoun, and it allows various insertions between the host sentence and ATNP. Besides, AT is not island-sensitive, cf. (12): (12) Alles spricht jetzt nur noch über die Freilassung von Meyer1 und Müller2, und es gibt die wildesten Gerüchte. ‘Everybody is talking about the discharge of Meyer1 and Müller2, and the strangest rumours go around.’ Ich habe schon oft [NP das Gerücht, dass ihm1 die Mafia I have already often [NP the rumour that him1 the mafia geholfen hat] gehört, | ich meine dem Meyer1. helped has] heard I mean the Meyer1

All in all, AT appears to be syntactically fairly independent of its host sentence. That is why I propose to analyse AT as an ‘orphan’ in terms of Haegeman (1991), Shaer (2003) and Shaer & Frey (2004). ‘Orphans’ are linguistic units that are syntactically totally

.  As for rightward adjunction there are no syntactic tests allowing to discern the exact functional layer that serves as the adjunction site, I keep the notation “IP” to refer to the functional domain above VP. This does not mean that I theoretically disagree with the split IP assumption, but solely that there is virtually no possibility to account for the adjunction site of the RD-NP in a more precise way. .  In this respect German RD is expected to differ from clitic RD like in Catalan, Italian or Greek, where clitics are subject to structural constraints different from those on full pronouns.

 Maria Averintseva-Klisch

independent from their host sentence and get integrated into it only at the level of the discourse, as other utterances do. One illustration: according to Shaer (2003), certain adverbials like with his X-ray vision in (13) are ‘orphans’ in this sense: (13) With his X-ray vision, John located the files.

(Shaer 2003: 458)

Shaer shows that the PP with his X-ray vision in (13) is syntactically independent. For instance, it does not show any effects that a syntactic movement would produce (e.g., it violates island constraint), so that a movement-based analysis of such adverbials fails. Furthermore, this PP can occur parenthetically at different positions in the clause, as in (14), which is expected if it does not have any syntactically determined position in its host sentence: (14) a. With his X-ray vision, John located the files. b. John, with his X-ray vision, located the files. c. John located the files, with his X-ray vision.

etc.

Besides, an ‘orphan’ has to build a prosodic unit of its own. As I have argued above, the same characteristics apply to AT: it does not show any features typical for elements syntactically dependent on its host sentence. Thus, AT allows morphological non-agreement, and does not obey the island constraint. Prosodically, it also has to build a unit of its own. In other words, AT appears to be a syntactic ‘orphan’. In terms of Zifonun, Hoffmann & Stecker (1997) AT is a construction of the ‘rechtes Außenfeld’ (“right outer field”) that embraces syntactically independent additions to the clause. The assumption that AT is syntactically non-integrated can directly account for its appearance after an explicit marking of the sentence boundary, the latter being either marked graphically with an appropriate punctuation mark, e.g., a full stop, in the case of written language, cf. (15a), or, in spoken language, by an intervention of another speaker (so-called ‘alien-initiated repairs’ in terms of Uhmann (1993)), like in (15b): (15) a.

(Der Koch war schon an Bord, der Matrose ebenfalls.) ‘The cook was already on board, the sailor too.’ Er aß die Fliegen. Der Koch, nicht der Matrose. He ate the flies. The cook not the sailor (Martel Yann, Schiffbruch mit Tiger: 364)

b.

A: A: A: A:

Ich weiß nicht, wann sie kommt. B: wer? I know not when shei comes. B: who? na die Anna. interj the Annai.

(Attested oral data)

After showing how the formal differences between RD and AT correspond to functional differences, I will turn to the issue of how the discourse integration of AT takes place.

To the right of the clause 

2.2  RD vs. AT: Functional differences My proposal is that RD is used to mark the discourse topic. I understand discourse topic informally as the discourse referent that is stably activated in the discourse representation during the reception of a particular discourse segment,11 or, in other words, the referent about which the current discourse segment is, cf. (16), where the old shoemaker is the discourse topic:12 (16) A broad ray of light fell into the garret, and showed the workman with an unfinished shoe upon his lap, pausing in his labour. [...] He had put up a hand between his eyes and the light, and the very bones of it seemed transparent. So he sat, with a steadfastly vacant gaze, pausing in his work. [...] (Charles Dickens, The Tale of Two Cities, Ch. 6)

Each particular language has preferred options of referring to discourse topics. Thus, personal pronouns are generally considered to be the cross-linguistically preferred anaphorical means of reference to discourse topics (see e.g., Bosch et al. 2003). Also in (16) the discourse topic referent is constantly referred to with the personal pronoun he. Besides, the so-called Left Dislocation in English, the construction consisting of an NP to the left of a clause and a coreferent pro-form inside it as in (17), has been argued to set the current discourse topic (e.g., Frey 2004):13 .  I understand discourse segment intuitively as a relatively small span of a discourse (minimally one utterance) that is characterized through a fairly tight thematic contiguity. In written language a discourse segment mostly corresponds to a paragraph. .  The problem of the status of discourse topic has been extensively and controversely discussed in literature (see Büring (2003); Asher (2004a) and (2004b); Kehler (2004); Oberlander (2004); Stede (2004) and Zeevat (2004), to name just a few). However, in spite of theoretical controversies and unclarities concerning the exact definition of the discourse topic, the existence of some kind of entity that is most salient at a given stage of the discourse and that is relevant for establishing coherence seems to be uncontroversial. This is the intuition shared by the papers in the recent issue of Theoretical Linguistics dedicated to discourse topics, although the authors use different terms for the same intuition of “the thing” that “chunks of text are about” (Asher 2004b: 255). My understanding of the discourse topic as being valid on the local level of a discourse segment corresponds more or less to the concept of ‘local topic within discourse segments’ in Kehler (2004). .  Generally a comparison of the left and right dislocation seems promising, even if it is out of reasons of space impossible in this paper. In short, both dislocation constructions are topicrelated devices, although there are differences. In general, LD seems cross-linguistically to be functionally more tightly connected with its host sentence than RD. Due to the linear order of the discourse, LD is bound to set the topic for the discourse segment including its host sentence, whereas RD is more important for the segment following its host sentence. In German, a special variant of left dislocation, the so-called ‘Hanging topic’ like in (a), has been argued to be a discourse-topic-marking device, cf. Frey (2004):

 Maria Averintseva-Klisch

(17) Annai, I like heri a lot. Shei [...]

Similarly, RD in German serves to mark the discourse topic, but is has the peculiarity that the topic is marked not only to be the current one, but also as the topic for the discourse segment following the RD. In (18), Madame Dutitre is set as “what is being talked about” for the segment following the host sentence of the RD: (18) Und als der König seine Frau verloren hatte, bedauerte ihn die Dutitre: “Ach ja, für Ihnen is et ooch nich so leicht [...].” ‘And when the king lost his wife, Dutitre pitied him: “Dear me, I should say, for you things aren’t that easy either [...]” ’ Siei war ein Original, die Madame Dutitrei. shei was an original the Madame Dutitrei Sie verstand nie, warum man über ihre Aussprüche lachte. Sie war eben echt und lebte, wie alle wirklich originalen Menschen, aus dem Unbewussten. Kein falscher Ton kam deshalb bei ihr auf. ‘Shei never understood why everybody always laughed at heri remarks. Shei was genuine and lived unconsciously, as all unique people do. Shei never came across as being artificial.’ (Siegfried Fischer-Fabian, Berlin-Evergreen: 125)

Thus, RD in German is a global discourse strategy in the sense that it helps to structure a whole discourse segment by marking its topic. More specifically, RD can either promote a discourse-old referent to the discourse topic, or to signal maintenance of the old discourse topic (especially after a change of the narration perspective).14 Importantly, it is a forward-looking strategy, as it structures the segment following the host sentence. RD being situated at the right periphery of the sentence predestines it to influence the subsequent segment. This means that in the default case the referent of the RD-NP is the discourse topic for the following segment, as in (19a). Interestingly, RD is also possible at the very end of a discourse like in (19b). This only seems to contradict my claim: the crucial point is that no other discourse referent is available as topic as long as the discourse continues, cf. (19c), which is a pragmatically unsuitable continuation in the context of



(a)

Peter, I like him a lot.

See also Zybatow & Junghanns (1997) for Russian and Lambrecht (1981) for French counterparts to Hanging topic, as well as Averintseva-Klisch (2006) for similarities and differences between Hanging topic and RD. .  Thus in (18) the discourse topic is maintained over the change of the perspective introduced by the quotation; as this maintenance cannot be taken for granted, it is explicitly signalled through the RD.

To the right of the clause 

the RD (marked with “#”). This means that even in (19b) the referent of the RD-NP remains the only discourse topic that is hypothetically possible: (19) a. Siei war ein Original, die Madame Dutitrei. shei was an original the Madame Dutitrei Sie verstand nie, warum man über ihre Aussprüche lachte. ‘Shei never understood why everybody laughed at heri remarks.’ b. Ja, siei war ein Original, die Madame Dutitrei. yes shei was an original the Madame Dutitrei ‘Yes, she was unique, that Madame Dutitre.’ (context: e.g., as the last sentence of the chapter about Madame Dutitre.) c. Monsieur und Madame Dutitre waren damals ziemlich berühmt, er für seine Pferde, sie für ihre Sprüche. Sie war ein Original, die Madame Dutitre. #Er hatte die besten Pferde Berlins in seinen Ställen. ‘Monsieur1 and Madame2 Dutitre were rather famous at that time, she2 in particular for her2 bon-mots, and he1 for his1 race horses. She2 was unique, that Madame D.2. #He1 had the best race horses of Berlin in his stables.’

As for AT, it is a repair strategy used to resolve a potentially unclear reference in the host sentence, as in (20): (20) (Sie [Die Mutter] hat den Wohnzimmerschrank aber auch nicht leiden können,[...], aber mein Vater hat sich auf keine billigen Sachen mehr eingelassen,) ‘Mother hated the wardrobemask, [...], but my father didn’t want to have any more cheap things around.’ er ist ihr auch zu dunkel gewesen, der Wohnzimmerschrank [...] hei is for.her also too dark been the wardrobei (Birgit Vanderbeke, Das Muschelessen)

Contrary to RD, AT is a local strategy, as it is used to repair its host sentence and does not have any impact on the global discourse structure. In this sense it is backwardlooking, as it refers back to its host sentence and does not influence the following discourse segment. In the next section, I will introduce my proposal for the integration of syntactically independent AT into its host sentence.

2.3  Syntax-discourse asymmetry with AT Being detached syntactically, an orphan gets attached to its host sentence at the level of the discourse (cf. Haegeman 1991). This attachment occurs for orphans in a regular way used for discourse units (cf. Shaer 2003). I assume with Asher & Lascarides (2003) that discourse units get attached to each other via discourse relations. In AverintsevaKlisch (forthc.), I have argued that a special discourse relation Afterthought has to be formulated for the attachment of ATs. What is relevant for the present issue is that

 Maria Averintseva-Klisch

Afterthought is argued to be a subordinating discourse relation in terms of Asher & Vieu (2005). Asher and Vieu put forward four tests to distinguish between subordinating and coordinating discourse relations, cf. (21): (21) Given are two constituents, α and β, a relation R (α, β), and a possible extension with a constituent γ; the nature of R is to be tested: 1. Attachment Test: If it is possible to attach γ to α, then R is subordinating; if attachment is possible only to β, then R is coordinating. 2. Continuation Test: if γ continues β in its relation to α, then R is subordinating; if it is impossible, then R is coordinating. 3. Anaphora Test: if for any γ attached to β no pronominal element in γ can be bound by referents in α, then R is coordinating; if some can, then R is subordinating. 4. ‘Prototype’ Test: if R can co-occur with Narration (as prototypical coordinating relation), then R is coordinating; if R can co-occur with Elaboration (prototypical subordinating relation), then R is subordinating. (Summed up from Asher & Vieu 2005)

Test 4 is not applicable to Afterthought, as the semantics of Afterthought is not compatible with Elaboration. In short: firstly, Afterthought differs crucially from Elaboration in its impact on the truth conditions of the whole sentence: AT first makes the establishing of the truth conditions for an utterance possible; due to the lack of referential clarity , it is not possible before the adding of the AT takes place. Secondly, Elaboration and Afterthought are different kinds of discourse relations. Asher & Lascarides (2003) distinguish between ‘content-level’ discourse relations and ‘cognitive-level’ discourse relations. For the former, it is only the content of the utterances building up a discourse that matters; this is the case with Elaboration. For the latter not only the content of the utterances, but also the intentions of the speaker and the addressee are important for defining their semantics. This seems for me to be the case with Afterthought (see also Averintseva-Klisch (forthc.)). Because of these differences between Elaboration and Afterthought Test 4 cannot be applied to Afterthought. The applicable tests all yield the same results: Afterthought is subordinating, cf. (22)-(24): (22) Attachment Test: a. Dann ist sie weggelaufen, (α) α b. (ich meine) die Serena. (β) ↓ Explanation c. Das macht sie immer wenn sie wütend ist. (γ) γ ‘Then she ran away (α), (I mean) Serena (β). That‘s what she always does when she is angry (γ).’

According to the Attachment-Test, the relation R (α, β) is bound to be subordinating, because γ gets attached to α (and not to β) via discourse relation Explanation: the last constituent explains the first one.

To the right of the clause 

(23) Continuation Test a. Dann ist sie weggelaufen, (α) α   ↓ b. (ich meine) die Serena. (β)   β Afterthought   ↓ c. Also die Kleine mit blonden Zöpfchen. (γ) γ ‘Then she ran away (α), (I mean) Serena (β). That is the little one with blonde pigtails (γ).’

Here, it is possible to proceed with a constituent that continues β in its Afterthought relation to α; thus, R again fulfils the conditions for subordination. (24) Anaphora Test , (α) attachment site a. Dann ist sie weggelaufen b. (ich meine) die Serena. (β) c. Das war nicht besonders schlau. (γ)... das ... c’. Das macht sie immer wenn sie wütend ist. (γ) ... das ... ‘Then she ran away (α)1, (I mean) Serena (β). That1 was not very clever (γ). / That1’s what she always does when she is angry (γ).’

It is possible to resume the event token (24c) as well as the event type (cf. Asher 1993) of the constituent α (24c’) with the pronoun das (‘that’) in γ. To sum up: the tests 1–3 all show that Afterthought is a subordinating discourse relation. According to the analysis presented here, corrections like (25) are a subtype of afterthoughts: (25) A: John failed his exams. B: No, he didn’t, he got 60%. A: I meant John Smith. (Asher & Lascarides 2003:305)

Asher and Lascarides (2003) do not consider in detail cases like (25), but they seem tacitly to handle them in a similar way as the instances of the discourse relation Correction, cf. (26): (26) A: John distributed the copies. B: No, it was Sue who distributed the copies. (Asher & Lascarides 2003: 470)

However, for Correction the constituents involved are per definition required to be inconsistent with each other (Asher & Lascarides 2003: 469). This is not the case in (25), where the constituent With “John” I mean John Smith is in no way inconsistent with the constituent John failed his exams. Besides, with respect to their function corrections like (25) are exactly like ‘alien-initiated repairs’ (Uhmann 1993), where the hearer explicitly signals his inability to resolve the pro-form, like in (27) (cf. also example (15b)):

 Maria Averintseva-Klisch

(27) A: I don’t like her at all. B: Whom? A: Anna I mean.

That is why I propose to subsume corrections like (25) under afterthoughts. Reflecting the character of AT as described above, the discourse relation Afterthought can be informally stated as in (28): (28) Afterthought is a subordinating discourse relation, which holds whenever the speaker of the host sentence and the AT supplies the AT with the intention of clearing the reference of a discourse referent x that has been introduced in the host sentence by establishing a relation x=z, where z is a discourse referent introduced in the AT, and the reference of z in the discourse representation is assumed to be unambiguous.

A point that I would like to make here is the following: syntax and discourse attachment do not mirror each other in the case of AT: syntactically, AT is an orphan, i.e., not attached at all. At the level of the discourse structure, however, it gets attached via a subordinating discourse relation. This supports the widely assumed hypothesis that generally the symmetry of syntactic relations and discourse relations between two discourse units is not necessarily required (see e.g., Blühdorn in this volume and Holler in this volume), although in some cases syntactic and discourse relations might go hand in hand. As for RD, I assume that it does not constitute a separate discourse unit, but is a part of the unit containing the host sentence. In Averintseva-Klisch (2006) I show how exactly RD contributes to the semantics of its host sentence. Summing up this section one might say that the right clausal edge in German holds two options: it is either used for syntactically integrated RD, or for syntactically non-integrated AT repairing some intraclausal reference.15 The former is a global discourse-structuring device, in particular concerning the immediately following discourse segment. The latter, on the contrary, is a local (i.e., related to the host sentence) repair strategy. In the next section I will argue that these are cross-linguistically two options the right clausal edge might have.

3.  Right periphery in the discourse: an outlook The issue of interest now is whether the usage of the right clausal edge for backward-looking local repairs and for forward-looking discourse topic marking is a peculiarity of German or whether these two options (local repair and global discourse structuring) are used .  Strictly speaking, RD and AT hold two different positions: while I assume that RD being part of the sentence is located at its right periphery, AT comes after the sentence boundary; being syntactically fully independent, it cannot be analysed as a right-peripheral construction. Zifonun, Hoffmann & Stecker (1997) introduced the term rechtes Außenfeld (‘right outer field’) for non-integrated phrases as the right edge of the clause (as compared to Nachfeld (‘afterfield’) for syntactically integrated ones).

To the right of the clause 

cross-linguistically. The following is to be understood as an outlook for further research. My hypothesis is that local repair and discourse topic marking are cross-linguistically available due to the general properties of the right clausal edge. This position is on the one hand the last possibility to add locally to the clause or to comment on some information within it. On the other hand, by virtue of its placement it relates the clause to the following discourse. As the first step to test this hypothesis I will in this paper very briefly compare German to French and Russian with respect to the use of RD and AT. I will begin with AT.

3.1  Afterthought in Russian and French AT is expected to be cross-linguistically generally available (primarily as an option of the spoken language). Not being syntactically attached at all, ATs do not depend on the syntactic characteristics of a language, and in fact nothing should prevent the possibility of adding repairs after the actual end of the clause. As expected, ATs are possible in Russian, cf. (29), as well as in French, cf. (30): (29)

Ego žena revnujet ego k Tereze Lido. ‘His wife is jealous of Teresa Lido.’ Ona puskajetsja za nim v pogonju, | ego žena. Shei starts for him in pursuit his wifei. ‘Shei pursues him, his wifei’

(Attested oral data)

Here, the speaker retells the contents of a movie, and the sentence preceding the clause with the pronominal reference ona (‘she’) introduces two female referents, “his wife” and “Teresa Lido”. The speaker uses a personal pronoun to refer to the wife, but then believes it might be unclear to the hearer whom she actually means, and she adds an explicit reference resolution device. Similarly in (30), where the reference of the pronoun il (‘he’) is unclear in the context, it is resolved explicitly with an afterthought NP: (30) (context: Jean and Jacques are eating.) Il a déjà mangé la soupe, | Jean. he has already eaten the soup Jean

As in German, the AT-NPs in Russian are those constructions involving NPs to the right of the clause that have been investigated most often.16 Now, I will turn to the less clear and more interesting issue of RD proper in Russian and French.

3.2  Right dislocation in Russian and French RDs in Russian have to my knowledge barely been considered in the literature. The only analysis I am aware of distinguishing between a repair and another kind of .  Lapteva (1976) assumes that adding a NP having a coreferent pro-form inside the clause after a syntactically complete clause has a function of “explanation”. Cf. also Zemskaja (1973) and Švedova et al. (1982) who are talking about the function of “specification of the pronoun” for the added NP.

 Maria Averintseva-Klisch

right-peripheral NP is given in Lapteva (1976). She assumes that besides AT there is another kind of added NPs with a coreferent pro-form, which seems “not to have any function at all” (Lapteva 1976: 267). In (31), however, one could hardly assume that the right-peripheral NP does not have any function; it is clearly used to enhance the intended meaning of the passage in that the wall is marked as the discourse topic: (31) Meier Wolf vsju žizn’ kopil den’gi, čtoby uvidet’ Stenu Plača. ‘MW saved money his whole life long to see the Wailing Wall’ On videl ejë teper’, ėty stenu. he saw heri now this wall.femi Ona naxoditsja za linijej železnoj dorogi [...]. Ėto prostaja požarnaja stena [...] ‘Now he has seen iti, this walli. Iti is situated behind the railways [...]. Iti is a simple fire protection wall [...]’ (Aleksandr Galič, Matrosskaja Tišina)

Characteristically, the reference of the pronoun in RD is often quite clear in the context, as in (31) and (32). That is why I argue that both are cases of RD proper. (32) A: A Petrova ušla čto-li? A: and Petrovai is.gone interr.pron B: Net, ona po-mojemu ne byla Petrova B: no shei I.believe not was Petrovai ‘A: Is Petrova already gone? B: No, I believe she has not been here at all, Petrova.’ (Lapteva 1976: 267)

In (32) the referent of Petrova is clearly the discourse topic, so that (31) and (32) are similar and in both the right-peripheral NP is an RD in the sense of a right-peripheral discourse topic marking construction.17 RD is also attested for French (Lambrecht 1981, 1987), e.g., (33):18 (33) a. b.

Il hei Il hei

a mangé la soupe, Jean. has eaten the soup Jeani est beau, ce tableau! is beautiful, this picture.maski

(Lambrecht 1981: 80)

Formally, RD is prosodically and syntactically integrated in Russian as well as in French, (see Lapteva (1976) resp. Lambrecht (1981)). For example, morphological agreement is required for Russian RD (Zemskaja 1973), cf. (34) as well as for French RD (35):

.  In (32) the discourse topic is similarly to (18) marked as maintained over the change of a perspective (here the change of the speaker). .  The French data presented in section 3 are taken from Lambrecht (1981), (1987) and Ashby (1988); the glosses are mine.

To the right of the clause 

(34)

Mne eje podarili nedavno, ėtu knigu / To-me her.acc presented.3.ps.plur recently this.acc book.acc / *ėta kniga. *this.nom book.nom ‘I recently got it as a present, this book.’ (Zemskaja 1973: 166)

(35)

Il-faut y aller quand il-fait chaud à la plage / *la one.should therei go when it.is warm to the beachi / *the plage. beach.nomi (Lambrecht 1981: 78)

In (34), the default nominative case is not available for the RD-NP if the intraclausal pro-form is in the accusative. In (35) RD-phrase has to preserve the morphological marking which an argument phrase would have in its canonical position (Lambrecht 1981: 79).19 This suggests that RD in Russian and French is syntactically integrated in the same way as RD in German. As for the function of RD, I argue for Russian as well as for French that in these languages RD serves in a way similar to German RD to mark the discourse topic. Observations made in the previous research on French RD are compatible with my analysis of RD as a discourse topic marking device. Lambrecht argues that RD is used to refer to the “previously established topic” (Lambrecht 1987: 237). Ashby (1988) claims that at least sometimes the referent of the RD “continue[s] to be talked about in succeeding sentences” (Ashby 1988: 216),20 which would correspond to its being the discourse topic for the following segment, cf. (36): (36)

Et puis ils étaient méchants, les Allemands. A la fin, quand and then theyi were angry the Germansi at the end when ils ont vu que la situation était perdue [...] theyi have seen that the situation was lost [...] (Ashby 1988: 214)

Here the speaker tells about the end of World War II in France and introduces the new discourse topic, the Germans, with the help of the RD. Generally, Ashby (1988) differentiates between 6 pragmatic functions of French RD: turn closing, filler,

.  In (35) the dislocated constituent is a PP and not an NP. However, Averintseva-Klisch & Salfner (2007) argue that PPs like in (35) that are prepositional objects behave in a RD exactly like NPs as far as their function is concerned. Here it is only due to the subcategorization frame of the verb aller that requires a PP, that a PP instead of an NP is dislocated. .  It is slightly surprising, that Ashby does not explicitly state referring to the discourse topic as a function of RD, although he observes that 93% of RDs in his corpus refer to discourse topics (Ashby 1988: 216, Table 4).

 Maria Averintseva-Klisch

clarification, topic shift, contrast and epithet. However, these, with possible exception of clarification, appear to me to be secondary functions accompanying the discourse topic marking. Thus, turn closing, which Ashby claims to be the most important function of RD, seems to me to be a special case of the discourse topic marking: the discourse topic gets “passed on” to the next speaker, so to speak: the first speaker wants to get sure that his communication partner continues to talk about the same entity as he did. So, in (37) “his son” is the discourse topic for both speakers: (37)

A: A: B: B:

Il he Ah Ah

est en dernière is in last mais, oui, c’est but yes it.is

année, son year his en dernière in last

fils. son. année. year. (Ashby 1988: 222)

As for the cases analysed as clarification in Ashby, these cases I expect to be ATs. This is however, an issue I cannot decide upon, as Ashby intro­duces only one example of clarification, and it does not have enough context for me to be able to decide on this issue. In other words, an informal cross-linguistic comparison supports the claim that RD marks the discourse topic. An important additional evidence for this claim coming from French are the pronominal RDs like in (38a), cf. Ashby (1988); these also occur in Norwegian (Fretheim 1995, 2001), cf. (38b):21 (38) a. b.

Madame X, elle est née ici, elle. Mrs. Xi shei is born here shei Scott heter Glenn til etternavn, han. Scotti is.named Glenn as surname hei

(Ashby 1988: 204)

(Fretheim 2001: 62)

Pronominal RDs seem to me to be the clearest case of a discourse topic marking function of the RD. Firstly, adding a pronominal reference definitely cannot be a reference clarification. Secondly, personal pronouns are traditionally assumed to be the preferred means for referring to discourse topics (e.g., Zifonun, Hoffmann & Stecker (1997), Bosch et al. (2003), Consten & Schwarz-Friesel (forthc.)), so it is to expect that also right dislocated personal pronouns refer to discourse topics. This cross-linguistic affinity of the discourse topics to the right periphery might seem unexpected, as especially in the conversation analysis the right clausal edge is usually seen as the position used for turn taking or for delaying of turn taking (cf. e.g., Auer (1991)). In accordance with this analysis information coming at the right periphery is not important, so that its potential loss through a turn-taking is not problematic. However,

.  Note that my analysis with respect to the function of RD differs from the actual proposals in Ashby (1988) and Fretheim (2001).

To the right of the clause 

it seems to me that the repetition of information that is already given and is thus seemingly “unnecessary” must be something more than just a strategy to retain turn-taking. Rather with this repetition the importance of the corresponding referent for the discourse is made clear, i.e., the referent is explicitly set as the current discourse topic.

3.3  Discussion: RD vs. AT cross-linguistically A comparison of German, Russian and French with respect to their use of RD and AT shows that these languages have AT and RD as the options of the use of the right clausal edge. These two uses of the right clausal edge are due to its twofold character. On the one hand, for reasons of linearity the right periphery constitutes a point of intersection with the following discourse. As such it is predestined to host NPs referring to the discourse topics that pertain in the following discourse segment. On the other hand, the position immediately after a clause is the last possibility to add something locally to this clause; that is why afterthought NPs are possible. Moreover, in all three languages observed RD is syntactically integrated. A similar observation was made in Lambrecht (2001: 1068), who stated that RD is cross-linguistically “more tightly connected with the predicateargument structure of the clause” than left-dislocated elements. In the languages under discussion AT is syntactically non-integrated. That is to be expected, as a speaker adds AT as a repair device when he has already completed the clause and only after that notices that the clause might be unclear. The question that remains is: Why is RD cross-­linguistically prosodically and syntactically integrated? Why a forward-looking discourse topic marking prefers a realization through a syntactically integrated NP at the right periphery is a challenging question for further research. It would be especially interesting to compare in a systematic way RD in languages featuring clitic RD (like e.g., Catalan or Greek) with RD in languages having only full pronouns like German. It seems that clitic RD differs from non-clitic RD in that multiple dislocation is non-restrictedly allowed (Vallduvì 1992: 85 for Catalan). As for Left dislocation, Grohmann (2003) proposes different syntactic analyses for non-clitic and clitic left dislocation. For Romance languages it is often assumed that RD is either derived from left dislocation (e.g., Frascarelli & Hinterhölzl (2007) for Italian) or that both constructions are symmetrical (Beninca & Poletto (2003) for Italian and Vallduvì (1992) for Catalan), which would imply also different analyses for clitic and non-clitic RD. If the syntax of the clitic and non-clitic RD also differs, then one would expect functional differences, too. Still, it seems that also clitic RD is a topic-marking construction, cf. Villalba (2000: 20), Lambrecht (2001: 1072), so that the generalization made above, viz. that the right periphery is a position designated for hosting expressions referring to discourse topics, still holds.

4.  Summing up and conclusions In my paper, I first introduced two prima facie similar, but formally and functionally different constructions of German, right dislocation and afterthought. I argued that

 Maria Averintseva-Klisch

whilst AT is a local reference repair strategy, RD is an important discourse-structuring device used to mark the discourse topic for the segment following RD. At the level of syntax, RD is part of its host sentence, presumably a right IP-adjunct. AT, on the contrary, is syntactically independent from its host sentence, and gets integrated into it only at the level of the discourse via a special discourse relation. Thus, AT might be seen as an illustration of the general asymmetry of syntactic and discourse relations: AT is added via a subordinating discourse relation, although there is no subordination on the level of syntax, AT being syntactically not attached at all. I assume that local repair and global discourse-structuring devices are crosslinguistically two options of the use of the right clausal edge, and put forward the question whether AT is always syntactically detached, whereas RD is always syntactically integrated. To follow up this point, I compared German to Russian and French with respect to the use and characteristics of RD and AT. I found that all three languages feature AT as an ‘orphan’ and RD as a syntactically integrated construction used to mark the discourse topic for the following segment. The former is due to the character of AT as a local repair. As for RD, this discourse-topic marking device at the right clausal periphery seems to be available cross-linguistically in a similar way. Discourse topic, contrary to sentence topic, is not a sentence-bound notion and thus also not a syntactic, but a pragmatic category. Accordingly, there cannot be a prototypical position for discourse topic inside the clause. However, it seems that right periphery, being a syntactically unnecessary and thus entirely pragmatically ruled position, is cross-linguistically preferred for the placement of NPs referring to discourse topics. This leads to the hypothesis that there is a cross-linguistic interdependence of the discourse function and the syntax. I assume that RD is cross-linguistically part of its host sentence, situated at the right periphery. Respectively, AT and other kinds of repair constructions are crosslinguistically ‘orphans’, i.e., syntactically independent units generated after a sentence is completed. Testing of this assumption for more languages with different syntactic character would be a challenging task for further research.

References Abraham, Werner. 2007. Topic, focus and default vs. contrastive accent. In On Information Structure, Meaning and Form: Generalizations across Languages, Kerstin Schwabe & Susanne Winkler (eds), 183–203. Amsterdam: John Benjamins. Altmann, Hans. 1981. Formen der “Herausstellung” im Deutschen: Rechtsversetzung, Linksversetzung, Freies Thema und verwandte Konstruktionen [Linguistische Arbeiten 106]. Tübingen: Niemeyer. Ashby, William J. 1988. The syntax, pragmatics and sociolinguistics of left- and right-dislocations in French. Lingua 75: 203–229. Asher, Nicholas. 1993. Reference to Abstract Objects in Discourse. Dordrecht: Kluwer.

To the right of the clause  Asher, Nicholas. 2004a. Discourse topic. Theoretical Linguistics 30: 163–201. Asher, Nicholas. 2004b. Troubles with topics: Comments on Kehler, Oberlander, Stede and ­Zeevat. Theoretical Linguistics 30: 255–262. Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation. Cambridge: CUP. Asher, Nicholas & Vieu, Laure. 2005. Subordinating and coordinating discourse relations. ­Lingua 115: 591–610. Auer, Peter. 1991. Vom Ende deutscher Sätze. Zeitschrift für Germanistische Linguistik 19: 139-157. Averintseva-Klisch, Maria. 2006. ‘Separate performative’ account of German right dislocation. In Proceedings of the Sinn und Bedeutung 10 [ZAS Papers in Linguistics 44]. Christian Ebert & Cornelia Endriss (eds), 15–28. URL: http://www.zas.gwz-berlin.de/papers/zaspil/ articles/zp44/Averintseva.pdf. Averintseva-Klisch, Maria. Forthcoming. German right dislocation and afterthought in discourse. In Constraints in Discourse, Anton Benz & Peter Kühnlein (eds). 213–235. Amsterdam: John Benjamins. Averintseva-Klisch, Maria & Salfner, Fabienne. 2007. German PP-dislocations to the left and to the right: against a symmetric approach. In Proceedings of Sinn und Bedeutung 11, Estela Puig-Waldmüller (ed.), 46–62. Barcelona: Universitat Pompeu Fabra. URL: http://mutis. upf.es/glif/pub/sub11. Beninca, Paola & Poletto, Cecilia. 2003. Topic, focus and V2: Defining the CP sublayers. In The Structure of CP and IP, Luigi Rizzi (ed), 52–75. Oxford: OUP. Bosch, Peter, et al. 2003. Demonstrative pronouns and personal pronouns. German der vs. er, Proceedings of the EACL2003, Workshop on The Computational Treatment of Anaphora. Budapest. Büring, Daniel. 2003. On D-trees, beans, and B-accents. Linguistics and Philosophy 26(5): 511–545. Consten, Manfred. 2004. Anaphorisch oder deiktisch? Zu einem integrativen Modell domänengebundener Referenz. Tübingen: Niemeyer. Consten, Manfred & Schwarz-Friesel, Monika. 2007. Anapher. In Wortarten des Deutschen, Ludger Hoffmann (ed.). 265–292. Berlin: de Gruyter. Culicover, Peter & Jackendoff, Ray. 2005. Simpler Syntax. Oxford: OUP. Frascarelli, Mara & Hinterhölzl, Roland. 2007. Types of topic in German and Italian. In On ­Information Structure, Meaning and Form: Generalizations across Languages, Kerstin Schwabe & Susanne Winkler (eds), 87–116. Amsterdam: John Benjamins. Fretheim, Thorstein. 1995. Why Norwegian right-dislocated phrases are not afterthoughts. Nordic Journal of Linguistics 18: 31–54. Fretheim, Thorstein. 2001. The interaction of right-dislocated pronominals and intonational phrasing in Norwegian. In Nordic Prosody, Wim A. van Dommelen & Thorstein Fretheim (eds), 61–75. Frankfurt: Peter Lang. Frey, Werner. 2004. Notes on the syntax and pragmatics of the German left dislocation. In The Syntax and Semantics of the Left Periphery, Horst Lohnstein & Susanne Trissler (eds), ­203–233. Berlin: de Gruyter. Grohmann, Kleanthes K. 2003. Prolific Domains. On the Anti-Locality of Movement Dependencies. Amsterdam: John Benjamins. Haegeman, Liliane. 1991. Parenthetical adverbials: The radical orphanage approach. In Aspects of modern English linguistics: Papers presented to Masatomo Ukaji on his 60th birthday, S. Chiba et al. (eds), 232–254. Tokyo: Kaitakusha.

 Maria Averintseva-Klisch Kayne, Richard. 1994. The Antisymmetry of Syntax. Cambridge MA: The MIT Press. Kehler, Andrew. 2004. Discourse topics, sentence topics, and coherence. Theoretical Linguistics 30: 227–240. Lambrecht, Knud. 1981. Topic, Antitopic and Verb Agreement in Non-Standard French [Pragmatics and Beyond II-6]. Amsterdam: John Benjamins. Lambrecht, Knud. 1987. On the status of SVO sentences in French discourse. In Coherence and Grounding in Discourse, Russel S. Tomlin (ed.), 217–262. Amsterdam: John Benjamins. Lambrecht, Knud. 2001. Dislocation. In Language Typology and Language Universals /  Sprachtypologie und sprachliche Universalien. An International Handbook, Vol. 2, Martin Haspelmath et al. (eds), 1050–1078. Berlin: de Gruyter. Lapteva, Ol’ga A. 1976. Russkij razgovornyj sintaksis. Moskva: Nauka. Oberlander, Jon. 2004. On the reduction of discourse topic. Theoretical Linguistics 30: 213–225. Ross, John R. 1986. Infinite Syntax! Norwood: Ablex. [= Ross, John R. 1967. Constraints on Variables in Syntax. PhD dissertation, MIT]. Schindler, Wolfgang. 1995. Zur Topologie von Herausstellungen und Zusätzen am rechten ­Satzrand. Sprache & Sprachen 16: 44–56. Selkirk, Elisabeth. 1978. On prosodic structure and its relation to syntactic structure. In Nordic prosody II, Thorstein Fretheim (ed.), 111–140. Trondheim 1981: TAPIR. Selting, Margret. 1994. Konstruktionen am Satzrand als interaktive Ressource in natürlichen Gesprächen. In Was determiniert Wortstellungsvariation?, Brigitta Haftka (ed.), 299-318. Opladen: Westdeutscher Verlag. Shaer, Benjamin. 2003. An ‘orphan’ analysis of long and short adjunct movement in English. In WCCFL 22 Proceedings, G. Garding, G. & M. Tsujimira (eds), 450–463. Sommerville MA: Cascadilla Press. Shaer, Benjamin & Frey, Werner. 2004. ‘Integrated’ and ‘non-integrated’ left-peripheral elements in German and English. ZAS Papers in Linguistics 35(2): 465–502. Smith, Michael. 2002. The polysemy of German es, iconicity, and the notion of conceptual distance. Cognitive Linguistics 13(1): 67–112. Stede, Manfred. 2004. Does discourse processing need discourse topics? Theoretical Linguistics 30: 241–253. Švedova, Natalija et al. (eds). 1982. Russkaja Grammatika. Moskva: Nauka. Uhmann, Susanne. 1993. Das Mittelfeld im Gespräch. In Wortstellung und Informationsstruktur, Marga Reis (ed.), 313–354. Tübingen: Niemeyer. Uhmann, Susanne. 1997. Grammatische Regeln und konversationelle Strategien. Fallstudien aus Syntax und Phonologie. Tübingen: Niemeyer. Vallduvi, Enric. 1992. The Informational Component. New York NY: Garland. Villalba, Xavier. 1999. Symmetry and antisymmetry in syntax. Syntaxis 2: 1–25. Ward, Gregory & Birner, Betty J. 1996. On the discourse function of rightward movement in English. In Conceptual Structure, Discourse and Language, Adele Goldberg (ed.), 463–479. Stanford CA: CSLI. Zeevat, Henk. 2004. Asher on discourse topic. Theoretical Linguistics 30: 203–211. Zemskaja, Elena A. (ed.). 1973. Russkaja razgovornaja reč. Moskva: Nauka. Zifonun, Gisela, Hoffmann, Ludger & Stecker, Bruno. (eds). 1997. Grammatik der deutschen Sprache. Vol. 1 & 2. Berlin: de Gruyter.

To the right of the clause  Ziv, Yael. 1994. Left and right dislocations. Discourse functions and anaphora. Journal of Pragmatics 22: 1–17. Zybatow, Gerhild & Junghanns, Uwe. 1998. Topiks im Russischen [Sprache und Pragmatik 47]. Lund: University of Lund.

Exploring the role of clause subordination in discourse structure The Case of French avant que Laurence Delort Université Paris 7

The goal of this paper is to explore the role of clause subordination in discourse structure. Through the study of the French subordinating conjunction avant que (‘before’) and its interaction with discourse context, I will attempt to show that clause subordination can affect temporal structure and also discourse structure, by conveying either subordination or coordination between discourse units. Keywords: clause subordination, subordinating and coordinating discourse relations, discourse structure, avant que

1.  Introduction This paper deals with the interaction between the French subordinating conjunction avant que (‘before’) and discourse context, and with the effect of this interaction on both temporal structure and discourse structure. In a discourse C1 avant que C2,1 the subordinate clause has an adverbial function: it temporally anchors the eventuality conveyed in the main clause, (Kamp & Reyle 1993). This function allows the subordinate clause to be replaced by another temporal adverbial, as illustrated in (1) and (2). (1) (2)

Paul a trouvé la solution avant que Marie la lui donne.2 ‘Paul found the solution before Marie gave it to him.’ Paul a trouvé la solution {avant la tombée de la nuit / avant 20h00}. ‘Paul found the solution {before nightfall / before 8.00 pm}.’

.  Or C1 avant de V2, if the subject is identical in both clauses. Therefore, this paper is c­ oncerned with one-sentence discourses rather than discourses made up of sentence sequences. Notation: Ci is a clause, and Vi is an infinitive verb, which denotes an eventuality (also called situation – state or event, (Asher 1993)) noted ei. .  The fact that the subordinate clause has a factual or a counterfactual interpretation is not my point here. Whatever the interpretation of the subordinate clause, this discourse has a circumstantial interpretation.

 Laurence Delort

Interestingly, this adverbial function is lost in some discourse contexts. Indeed, in ­discourses (3a) and (3b), the subordinate clause cannot be replaced by a temporal adverbial, cf. (4a) and (4b) respectively. (3) a. b. (4) a. b.

Paul a d’abord cherché la solution avant que Marie la lui donne. ‘Paul first sought the solution before Marie gave it to him.’ Paul a longtemps cherché la solution avant que Marie la lui donne. ‘Paul sought the solution for a long time before Marie gave it to him.’ #Paul a d’abord cherché la solution avant la tombée de la nuit. ‘Paul first sought the solution before nightfall.’ #Paul a longtemps cherché la solution avant la tombée de la nuit. ‘Paul sought the solution for a long time before nightfall.’

From this observation, several questions arise: how can a temporal subordinate clause lose its adverbial function? What does the contrast between discourses such as (1) and (3) hide? Does the loss of adverbial function have an effect on temporal structure and discourse structure? I will try to provide some answers in this paper. The paper is organized as follows. First, I propose an analysis of the interaction between the conjunction avant que and discourse context which gives rise to nonadverbial uses of the subordinate clause, as exemplified by discourses (3a–b). Then, I present some effects of this interaction on temporal structure (in particular, avant que conveys a strongly constrained temporal relation, rather than a simple one), and on discourse structure (in particular, avant que conveys coordination, rather than subordination, between discourse units).

2. 

Interaction between avant que and discourse context

Discourse (1) exhibits a ‘circumstance’ interpretation, that is, the subordinate clause ­introduced by avant que has a temporal adverbial function. As observed in the ­discourses in (3), the same subordinate clause can lose this adverbial function. Avant que does not introduce an eventuality providing a temporal anchor for the eventuality conveyed in the main clause. If a ‘circumstance’ interpretation is impossible or not plausible, what does a discourse C1 avant que C2 express? My hypothesis is that ­interaction between the semantics of avant que and discourse context gives rise to two interpretations: – –

‘continuation’: e1 and e2 are in a narrative sequence, e2 being the continuation of e1; ‘pre-condition’: e1 is the (necessary) condition of the realization of e2, i.e., realization of e2 depends on e1.

In these two interpretations, as well as in a ‘circumstance’ interpretation, the temporal relation conveyed by avant que is the same: a temporal precedence between the

Clause subordination in discourse structure 

­eventualities, i.e., e1 < e2. The central idea behind this work is that it is discourse context, and sometimes also extra-linguistic knowledge, that allows avant que to appear in ­non-circumstantial discourses (and trigger coordination in discourse, see section 3). To argue for this idea, the next two subsections investigate what kind of discourse context supports ‘continuation’ (section 2.1) and ‘pre-condition’ (section 2.2).

2.1  Continuation ‘Continuation’ can be inferred thanks to linguistic cues capturing either a topic ­elaboration or a preparatory event.

2.1.1  Topic elaboration There are some linguistic cues, such as verbs and adverbs, which unambiguously ­express a continuation between two eventualities. These cues can be found in ­combination with avant que: for instance, the adverb d’abord (‘first’) in (5), or the verb commencer par (‘to start off by’) in (6). (5) Permettez-moi d’abord de présenter quelques réflexions générales avant d’entrer dans les détails. (A. de Tocqueville – Correspondance avec H. Reeve)3 ‘Let me first present some general comments before going into the details.’

(6) Ils commencèrent par se dévisager, les uns les autres, avant de se parler. (P. Loti – Mon frère Yves) ‘They started off by staring at one another, before speaking.’

When made explicit by such lexical items, ‘continuation’ is constrained by a common theme, called topic, shared by the eventualities, (cf. Danlos, 2005). That is, avant que conveys ‘continuation’ only if the eventualities linked share a topic (supported by ­identical subjects). For instance, in (7), there is no thematic link between e1 ‘to try to bring me out’ and e2 ‘to blow up’. In fact, e1 is continued by e3, introduced by ensuite (‘then’). Avant que in (7) does not convey ‘continuation’, but ‘circumstance’. (7) Eh bien, elle a d’abord essayé de m’entraîner hors du club avant que cela n’éclate. Ensuite, elle a fait preuve d’un sang-froid […] que bien peu de femmes possèdent en pareille occasion. (M. Droit – Le retour) ‘She first tried to bring me out of the club before it blew up. Then, she kept her head, such as few women do in such a situation.’

The cues in (5) and (6) are cataphoric, because they call for a succeeding eventuality. Sometimes, avant que is used in combination with anaphoric cues, calling for a preceding

.  Most of examples are taken from the French literature database Frantext (http://www. frantext.fr/). If there is no reference to author, the discourses are constructed. All examples in French are followed by their translation in English.

 Laurence Delort

eventuality, as illustrated in (8) and (9), with the verbs ajouter la suite (‘to add’) and the verb conclure (‘to conclude’), respectively. (8) Il répondit en reprenant une phrase de Virginie : “Faut pas se plaindre...” avant d’ajouter la suite : “…y’a plus malheureux que nous !”. (R. Sabatier – David et Olivier) ‘He answered by borrowing a sentence from Virginie: “ There is no need to complain…” before adding: “There are people more unfortunate than us!”.’

(9) Pour ma part je me demandai surtout si je pouvais arrêter ma carrière – avant de conclure que non. (M. Houellebecq – La possibilité d’une île) ‘I was wondering if I could stop my career – before concluding that I couldn’t.’

All these lexical cues for ‘continuation’ reflect the fact that e1 cannot occur after e2. The eventualities occur in a certain temporal order, with respect to a topic, on which they elaborate. So, these discourses C1 avant que C2 have nothing to do with a circumstantial interpretation. Lexical cues are necessary with avant que when a topic elaboration with ‘continuation’ is used. If there are cues, e1 can be either instantaneous or durative, cf. (10a) and (10b) respectively, and there is no ambiguity. (10) a. b.

Paul a d’abord donné la solution à Marie avant de la donner à Lisa. ‘Paul first gave the solution to Marie before giving it to Lisa.’ Paul a d’abord cherché la solution avant que Marie la lui donne. ‘Paul first sought the solution before Marie gave it to him.’

If there are no cues, when e1 is instantaneous, there is an ambiguity between ­‘continuation’ and ‘circumstance’, cf. (11a). When e1 is durative, there is no ambiguity, cf. (11b). (11) a. b.

Paul a donné la solution à Marie avant de la donner à Lisa. ‘Paul gave the solution to Marie before giving it to Lisa.’ Paul a cherché la solution avant que Marie la lui donne. ‘Paul sought the solution before Marie gave it to him.’

Discourse (11b) reflects another discourse configuration where avant que conveys ‘continuation’.

2.1.2  Preparatory Event Some discourses display no thematic link between eventualities, i.e., there is no topic elaboration. Nevertheless, they express ‘continuation’ with avant que in combination with lexical semantics of verbs. In particular, there is ‘continuation’ when e1 is a ­durative situation, as in (12) and (13). (12) Il erra longtemps avant de s’asseoir sur un banc de pierre. (R. Sabatier – Les noisettes sauvages) ‘He wandered a long time before sitting down on a stone bench.’

Clause subordination in discourse structure 

(13) Quand je suis entré, elle a attendu quelques secondes avant de lever les yeux vers moi. Je l’avais encore jamais trouvée aussi belle. (P. Djian – 37.2 Le matin) ‘When I entered, she waited a few seconds before looking at me. I had never found her so beautiful.’

One can observe that none of the subordinate clauses can be replaced by a temporal adverbial, cf. (14) and (15) respectively. (14) #Il erra longtemps avant 20h00. ‘He wandered a long time before 8.00 pm.’ (15) #Elle a attendu quelques secondes avant 20h00. ‘She waited a few seconds before 8.00 pm.’

The eventuality expressed by the main clause can be interpreted as a preparatory phase for e2: there is no thematic link between e1 and e2, but e1 leads to e2 (not necessarily naturally). This preparatory phase is even clearer when one observes discourses such as (16) where e1 is an interval of time between two events e0 ‘the noise echoes through the whole house’ and e2 ‘he walks down’. (16) La porte claque très fort et le bruit résonne dans toute la maison. Quelques minutes s’écoulent avant qu’il redescende. (B. Clavel – La maison des autres) ‘The door bangs very loudly and the noise echoes through the whole house. A few minutes go by before he walks down.’

The situation e1 can be iterative, as illustrated by (17). (17) La porte battit derrière lui, oscilla plusieurs fois avant de s’immobiliser. (M. Genevoix – Eva Charleboix) ‘The door banged behind him, hovered several times before coming to a standstill.’

In this case, the subordinate clause can be replaced by a temporal adverbial in (18), but it does not imply that (17) and (18) share the same meaning. In (17), e1 leads to e2: ‘to hover several times’ leads to ‘to come to a standstill’. While in (18), e1 does not lead to e2, e1 is only temporally anchored by the adverbial phrase. (18) La porte oscilla plusieurs fois avant 20h00. ‘The door hovered several times before 8.00 pm.’

In discourses (12), (13), (16) and (17), lexical semantics of verbs (conveying that e1 is a durative/iterative situation leading to e2) explicitly represents the semantic relation ‘continuation’ between the eventualities. The next interpretation, ‘pre-condition’, arises from this interpretation: e1 is a preparatory (durative or iterative) event that is realized with the purpose of ­realizing e2.

 Laurence Delort

2.2  Pre-condition ‘Pre-condition’ means that e1 is the necessary condition of the realization of e2. It can be either explicit, through linguistic clues, or inferred, through the semantics of eventualities (in particular e1). Again, avant que interacts with discourse context to trigger a non-circumstantial interpretation.

2.2.1  Explicitness of ‘pre-condition’ ‘Pre-condition’ can be expressed via lexical cues such as modal verbs, in addition to expression of durative or iterative situations. These cues can be found either in C1, as e.g. devoir (‘must’, ‘to have to’) in (19) and (20), or in C2, as e.g. pouvoir (‘to be able to’) in (21) and (22). (19) Arlequin dut tambouriner longtemps avant que la porte s’ouvrît. (M. Tournier – Le medianoche amoureux) ‘Arlequin had to drum a long time before the door opened.’ (20) L’air siffla entre les lèvres de Bensoussan qui dut s’y reprendre à deux fois avant d’évacuer un nom : - Toni. (A. Page – Tchao Pantin) ‘Bensoussan had to try twice before saying a name: - Toni.’ (21) J’ai attendu encore une ou deux secondes avant de pouvoir débloquer mes mâchoires. (P. Djian – 37.2 Le matin) ‘I waited one or two seconds longer before I could unlock my jaw.’ (22) J’en étais tellement déconcertée et froissée qu’il se passa une minute avant que je puisse réagir. (P. Labro – Des bateaux dans la nuit) ‘I was so stumped and piqued that an entire minute went by before I could react.’

In all these discourses, e1 must be realized in order to make e2 happen. Modal verbs trigger this interpretation unambiguously. For instance, if a modal verb is inserted in the ‘continuation’ discourse (16), cf. (23a), e1 is interpreted as a necessary ­condition for his walking down, cf. (23b). See also the parallel between (13) and (21), and ­between (16) and (22). (23) a. b.

Quelques minutes s’écoulent avant qu’il redescende. ‘A few minutes go by before he walks down.’ Quelques minutes s’écoulent avant qu’il puisse redescendre. ‘A few minutes go by before he can walk down.’

Some contexts do not display lexically specified modalities for conveying ‘pre-condition’. The semantics of eventualities and extra-linguistic knowledge represent clues for the inference of ‘pre-condition’.

2.2.2  Inference of ‘pre-condition’ Sometimes, modalities are not expressed but e1 is such that it is easily understood as the necessary condition of the realization of an event, which is e2. Corpora show

Clause subordination in discourse structure 

numerous ‘continuation’ discourses (where e1 is a preparatory event) parallel to ‘precondition’ discourses. For instance, discourses (12) and (24) (see (19) with modality) both display a durative situation. However, in (12), e1 leads (with no intent) to e2, while in (24), e1 is realized in order to realize e2. The same observation can be made on discourses (17) and (25) (see (20) with modality). Both display an iterative situation but in (17), the repeated hovering is not realized in order to come to a standstill (it just leads to it), while in (25), he tried twice in order to speak. Finally, discourses (13) and (26) (see (21) with modality) show the same distinction between ‘continuation’, on the one hand, when there is no intention, and ‘pre-condition’, on the other hand, when there is intention. (24) Lucie tambourina longtemps à la porte de sa chambre avant qu’il se décidât à ouvrir. (J. Rouaud – Les champs d’honneur) ‘Lucie drummed a long time at the door before he decided to open.’ (25) Il s’y reprit à deux fois avant de parler, mais enfin il dit : - Pourquoi? (L. Guilloux – Le pain des rêves) ‘He tried twice before speaking, but at last he said: - Why?’ (26) J’ai attendu que la bonne femme se décide à aller chercher son argent avant de considérer le boulot comme terminé. (P. Djian – 37.2 Le matin) ‘I waited for the woman to take her money before considering the job done.’

Interaction between avant que and lexical semantics cues licenses ‘continuation’ or ‘pre-condition’. Differences with ‘circumstance’ pertain to semantics, as we have seen, but also to temporal structure and discourse structure, as we will see in the next section.

3.  Effects on temporal structure and discourse structure Avant que appears in several discourse contexts and I put forward that this subordinating conjunction licences different interpretations: ‘circumstance’, ‘continuation’ or ­‘pre-condition’. These three interpretations are summed up respectively by the discourses (27) (=(1)), (28) (=(3)), and (29), which are examples constructed from discourses taken from the Frantext database (see note 3), and illustrate the previous analysis in section 2. (27) Paul a trouvé la solution avant que Marie la lui donne. ‘Paul found the solution before Marie gave it to him.’ (28) a. b.

Paul a d’abord cherché la solution avant que Marie la lui donne. ‘Paul first sought the solution before Marie gave it to him. Paul a longtemps cherché la solution avant que Marie la lui donne. ‘Paul sought the solution for a long time before Marie gave it to him.’

 Laurence Delort

(29) a. b.

Paul a dû longuement insisté avant que Marie lui donne la solution. ‘Paul had to insist strongly before Marie gave the solution to him.’ Paul a longuement insisté avant que Marie lui donne la solution. ‘Paul insisted strongly before Marie gave the solution to him.’

‘Circumstance’ and ‘continuation’ echo two discourse relations from Segmented ­Discourse Representation Theory (SDRT) as found in Asher and Lascarides (2003), viz. Background and Narration, respectively. Since I did not find a detailed description of a possible corresponding discourse relation for ‘pre-condition’ in SDRT, or in Mann and Thompson’s (1988) Rhetorical Structure Theory (RST), I put this interpretation aside in this section. Nevertheless, one interesting aspect arises from Background (for ‘circumstance’) and Narration (for ‘continuation’): they do not involve the same effects on temporal structure and discourse structure. Does avant que convey Background in (27) and Narration in (28)? If this proves to be the case, it would mean that the subordinating conjunction avant que has an important role at the discourse level. After a general definition of Background and Narration in SDRT (section 3.1), I try to show that avant que can be a marker of these discourse relations in contexts similar to those in (27) and (28) (section 3.2).

3.1  Definition of Background and Narration in SDRT 3.1.1  Background in SDRT Background is defined as follows in Asher and Lascarides (2003: 460): “this relation holds whenever one constituent provides information about the surrounding state of affairs in which the eventuality mentioned in the other constituent occurred”. That is, the eventuality described in the first clause is the main event, and the eventuality described in the second clause is the (e.g., temporal, spatial) circumstance of this main event. Since temporal progression is broken, it is a subordinating discourse relation (Vieu & Prévot, 2004). Discourse (30), taken from Asher and Lascarides (2003), illustrates Background between the two constituents π1 and π2 (noted as Background(π1, π2)), representing the semantics of e1 and e2, respectively. (30) Max opened the door. The room was pitch dark.

The semantic effect of Background is a temporal overlap between the eventualities, as axiom (A1) expresses, namely that e2 and e1 overlap. (A1) Background(π1, π2) → overlap(e2, e1)

3.1.2  Narration in SDRT Narration is defined as follows in Asher and Lascarides (2003: 462): “this relation holds if the constituents express eventualities that occur in the sequence in which

Clause subordination in discourse structure 

they were described”. That is, the eventuality described in the first clause occurs, then, the ­eventuality described in the second clause occurs. Since temporal progression is ­continuous, it is a coordinating discourse relation. Discourse (31), taken from Asher and Lascarides (2003), illustrates Narration(π1, π2). (31) Max came into the room. He sat down.

This discourse relation implies semantic effects on discourse interpretation. The first effect is the temporal relation between the denoted events, cf. axiom (A2) proposed in Bras et al. (2001). This axiom means that, when Narration holds between π1 and π2, “post(e1) persists up to the beginning of e2, and pre(e2) starts when (or before) e1 ends”, i.e. there is a strong contiguity between the events expressed in the units linked by Narration. That is, no relevant event can occur between the events. Hence, this temporal relation has to be distinguished from the simple temporal relation eα < eβ. (A2) Narration(π1, π2) → e1⊃⊂(post(e1)∩pre(e2))⊃⊂e2

The second effect of Narration is both semantic and structural: it reflects the need for a common topic between the events, and the insertion in the discourse structure of a constituent corresponding to this topic, cf. axiom (A3). If Narration holds, then there should exist a unit (the topic, noted π*) summarizing the units linked by Narration, and structurally dominating the complex unit (π’) made by Narration(π1, π2) via the discourse relation Topic. (A3) Narration(π1, π2) → ∃π*(π*=π1∩π2)∧Topic(π*,Narration(π1, π2))

This topic constraint is a means for ensuring coherence in a narrative text. For ­instance, the representation in SDRT of discourse (31) is given in Figure 1. It illustrates that, in SDRT, a coordinating relation like Narration is drawn with a horizontal arrow, while a subordinating relation like Topic (or Background) is drawn with a vertical arrow, cf. Asher and Lascarides (2003: 146–147). π* Topic π¢ π1

Narration

π2

Figure 1.  SDRT structure for discourse (31).

3.2  Is avant que a marker of Background and Narration? 3.2.1  Avant que and Background Given the definition of Background in SDRT, one can easily see that avant que t­ riggers this discourse relation in (27): π1 describes the main event (conveyed by the main

 Laurence Delort

clause), and π2 describes the temporal circumstance (conveyed by the subordinate clause introduced by avant que). It is also the case in discourse (32), where the ­subordinating nature of Background is indubitable because of the attachment of the subsequent constituents (π3 and π4) on π1 via Narration, with respect to the right frontier constraint, (cf. Polanyi 1988). (32) Il remit soigneusement sa casquette avant de sortir, salua de nouveau et ouvrit maladroitement la porte. (R. Sabatier – David et Olivier) ‘He carefully put on his cap before leaving, said good-bye once again and opened the door awkwardly.’

Moreover, the temporal effect of Background is compatible with the semantics of ­discourses conveying ‘circumstance’. For instance, in (27) and (32), e1 does not really occur before e2, but e1 occurs during an event occurring before e2. In more formal terms: there is not simply e1 < e2, but overlap(e, e1) and e < e2, i.e. e overlaps e1, and e occurs before e2 (same temporal overlapping operator as in axiom (A1)). Nevertheless, a detailed investigation remains to be done. Because of the subordinating nature of Background, the fact that avant que is a marker of Background exemplifies the hypothesis of the mapping between clause subordination and discourse subordination made by Matthiessen and Thompson (1988). This issue is discussed in the next subsection.

3.2.2  Avant que and Narration There is some linguistic evidence showing that the temporal precedence relation in discourse (27), on the one hand, and discourse (28), on the other hand, is not the same. First, the temporal relation can be modified in (27) but not in (28), cf. (33) and (34) respectively. (33) Paul a trouvé la solution trois heures avant que Marie la lui donne. ‘Paul found the solution three hours before Marie gave it to him.’ (34) a. #Paul a d’abord cherché la solution trois heures avant que Marie la lui donne. ‘Paul first sought the solution three hours before Marie gave it to him.’ b. #Paul a longtemps cherché la solution trois heures avant que Marie la lui donne. ‘Paul sought the solution for a long time three hours before Marie gave it to him.’

This adverbial modification with trois heures (as well as with juste, quelques secondes, longtemps (‘just’, ‘a few seconds’, ‘a long time’, etc.), bearing on the temporal relation, works with ‘circumstance’, but not with ‘continuation’. This reflects the fact that not only is it impossible to modify the temporal relation, but it is also impossible to ­extend the temporal distance between the eventualities. That is, with ‘circumstance’, the ­distance between the eventualities can be quantified, while with ‘continuation’, this distance is constrained such that e2 occurs immediately after e1, i.e. there is no distance between e1 and e2.

Clause subordination in discourse structure 

Inserting a third eventuality between e1 and e2 can test this constraint, cf. Bras et al. (2001): in (35), e3 can occur between e1 and e2, whereas in (36), it cannot. (35) Paul a trouvé la solution avant que Marie la lui donne. Entre-temps, il l’avait donnée à Lisa. ‘Paul found the solutions before Marie gave it to him. Meanwhile, he had given it to Lisa.’ (36) a. Paul a d’abord cherché la solution avant que Marie la lui donne. #Entre-temps, il l’avait trouvée. ‘Paul first sought the solution before Marie gave it to him.’ #Meanwhile, he had found it.’ b. Paul a longtemps cherché la solution avant que Marie la lui donne. #Entre-temps, il l’avait trouvée. ‘Paul sought the solution for a long time before Marie gave it to him.’ #Meanwhile, he had found it.’

I did not come across discourses such as (36) in a corpus. However discourses like (37) have been attested in corpora, where the third eventuality always corresponds to an eventuality occurring during the time interval of e1. So, this is not a contradicting observation. (37) J’ai passé quinze ans à donner des cours et à écrire des livres avant de comprendre que je n’étais pas fait pour cela, mais je ne peux vraiment me retirer, car, entre-temps, j’ai acquis quelque renommée et les collègues me retiennent. (J. Kristeva – Les samouraïs) ‘I spent fifteen years lecturing and writing books before understanding that I am not made for this, but I cannot withdraw because, meanwhile, I had gained a reputation and my colleagues retain me.’

It seems clear that the temporal relation between the end of e1 and the beginning of e2 is constrained. With ‘continuation’, the temporal relation is like the one proposed by Bras et al. (2001) for describing the temporal effect of Narration, see axiom (A2). So, with respect to temporal structure, in discourse (28), and in other discourses pertaining to ‘continuation’, Narration is used. The topic constraint, expressed in axiom (A3), is satisfied in topic elaboration, but not in a preparatory event. These two subtypes of ‘continuation’ share the same ­temporal relation but not the topic constraint. Nevertheless, they imply the same structural ­effect: coordination between discourse units. Finally, by conveying Narration, avant que is a challenge to the hypothesis of a mapping between clause combining and discourse ­structure. To put it in a nutshell, avant que is a subordinating conjunction that can trigger (at least) two discourse relations, involving different temporal and structural effects. On the one hand, avant que can trigger Background, which implies a temporal overlap relation between the main event and its temporal location, and subordination in discourse (i.e. narrative digression). On the other hand, avant que can trigger Narration,

 Laurence Delort

which implies a temporal precedence relation between two main events, and coordination in discourse (i.e. narrative progression).

4.  Conclusion and perspectives The goal of this paper was twofold. First, one aim was to shed light on the fact that the subordinating conjunction avant que can convey several interpretations when it interacts with discourse context. Second, it aimed at showing that avant que has an important role in discourse structure, by triggering either subordination or coordination between discourse constituents. It follows that there is no direct mapping between clause combining and discourse structure: a subordinating conjunction can convey a coordinating discourse relation. If avant que is a cue-phrase of Narration, a comparison between avant que and puis would be interesting, as the following observations show. When there is a topic elaboration, avant que can be translated by and then, as observed in the bilingual ­database TransSearch,4 cf. (38). (38) a. En deux heures à peine, l’espace aérien nord-américain a été plongé dans le chaos le plus complet avant d’être complètement fermé. b. In the space of a few hours, North American air space was thrown into complete chaos and then shut down completely.

When ‘continuation’ occurs, and e1 is durative, one can find puis in place of avant que, cf. (39). (39) Il a attendu un moment puis m’a touché l’épaule et s’est relevé. - Je sors par les cuisines, il a fait. (P. Djian – 37.2 Le matin) ‘He waited for a while and then touched my shoulder and got up.’

But puis cannot always be replaced by avant que without changing the original meaning of the discourse. The ‘continuation’ interpretation of (40a) is totally lost with avant que: (40b) rather conveys ‘circumstance’ despite the thematic link (topic) between eventualities. (40) a. b.

Il frappa à la porte. Puis il entra. ‘He knocked at the door. Then he entered.’ Il frappa à la porte avant d’entrer. ‘He knocked at the door before entering.’

This confirms that avant que needs a particular discourse context to convey a discourse relation such as Narration. But further investigation is needed.

  TransSearch is a database of translations between English and French accessible from http:// www.tsrali.com/.

Clause subordination in discourse structure 

Finally, this paper gives an account of several interpretations illustrated by unambiguous discourses. But corpora display many ambiguous discourses for which it is hard to find a clear interpretation, especially to decide if a discourse represents ‘circumstance’ (or Background), or ‘continuation’ (or Narration). ‘Pre-condition’ ­discourses are not ambiguous because there are always (extra-) linguistic clues for inferring such an interpretation. However, from discourses without lexical clues (such as d’abord, ­commencer par) or no possible semantic inference, ambiguity arises. Future research will concentrate on other clues for inferring the correct interpretation of avant que, such as taking into account a wider discourse context or discourse attachment on the content of the subordinate clause, as discourses (41) and (42) illustrate. In (41), e2 is part of a set of events, all elaborating the topic expressed in the first sentence. In (42), the last sentence is to be attached to the subordinate clause and not to the main clause (as in discourse (32), for instance). (41) Heureusement, Mme Bernard arracha les autres à la torpeur en jouant la meneuse de jeu. Elle entraîna les jeunes dans une partie de croquet, puis organisa une partie de mikado avant de faire des tours de cartes et de montrer des jeux avec une simple ficelle. (R. Sabatier – Les filles chantantes) ‘Fortunately, Mrs Bernard acted as a leader. She dragged the youths into a match of croquet, then organised a game of pick-up-sticks before doing card tricks and showing games with a simple piece of string.’ (42) Olivier trouva là une heureuse occasion de promenade. Il jeta un regard dans la glace avant de sortir en sifflotant. Il faisait beau. (R. Sabatier – David et Olivier) ‘Olivier found an opportunity for a stroll. He peeked in the mirror before going out whistling. The sun was shining.’

Acknowledgments I would like to thank the reviewers for their many valuable comments, and André ­Bittar for checking my English.

References Asher, Nicholas. 1993. Reference to Abstract Objects in Discourse. Dordrecht: Kluwer. Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation. Cambridge: CUP. Bras, Myriam  Le Draoulec, Anne & Vieu, Laure. 2001. Temporal information and discourse relations in narratives: The role of French connectives puis and un peu plus tard. In Proceedings of the 39th Annual Meeting of ACL, 49–56, Toulouse, France. Danlos, Laurence. 2005. Partition of an entity with aspectuo-temporal operators. In Proceedings of the Third International Workshop on Generative Approaches to the Lexicon (GL’2005), Geneva, Switzerland. Kamp, Hans & Reyle, Uwe. 1993. From Discourse to Logic. Introduction to Model-theoretic ­Semantics of Natural Language, Formal Logic and Discourse Representation Theory. D ­ ordrecht: Kluwer.

 Laurence Delort Mann, William & Thompson, Sandra. 1988. Rhetorical structure theory: Toward a functional theory of text organization. Text 8(3): 243–281. Matthiessen, Christian & Thompson, Sandra. 1988. The structure of discourse and ‘subordination’. In Clause Combining in Grammar and Discourse [Typological Studies in Language vol. 18], John Haiman & Sandra Thompson (eds), 275–329. Amsterdam: John Benjamins. Polanyi, Livia. 1988. A formal model of the structure of discourse. Journal of Pragmatics 12: 601–638. Vieu, Laure  & Prévot, Laurent. 2004. Background in SDRT. In Proceedings of the conference Traitement Automatique du Langage Naturel, 485–494. Fez, Morocco.

Pseudo-imperatives and other cases of conditional conjunction and conjunctive disjunction* Michael Franke

Universiteit van Amsterdam Pseudo-imperatives, a special kind of sentential conjunctions and disjunctions, display a surprisingly divergent preference bias. This paper aims to explain this pragmatic preference puzzle based on the different discourse segmentation behavior of conjunction and and disjunction or. To lend credence to the suggested explanation, related non-standard uses of conjunction and disjunction will be discussed in detail. Keywords: pseudo-imperatives, discourse relations, non-standard uses of conjunction/disjunction

1.  Pseudo-imperatives and a pragmatic puzzle 1.1 

The problem of pragmatic asymmetry

There is an interesting pragmatic asymmetry in the following pattern (cf. van der ­Auwera, 1986):1 (1)

a. b. c. d.

Close the window and I will kiss you. Close the window and I will kill you. Close the window or I will kill you. ?Close the window or I will kiss you.

(A and P+) (A and P–) (A or P–) (A or P+)

*  A lot of people have helped me considerably in writing this text including the anonymous reviewers. I’m thankful to all of them. Anke Lüdeling kindly provided me with access to the corpora that I have used and I am particularly grateful for that, as well as for the stimulation I have received from discussions with Robert van Rooij, Martin Stokhof, Paul Dekker and Magdalena Schwager. Thanks to Tikitu de Jager for proofreading. Needless to say, all errors are my own. .  I assume here for the sake of the example that (it is common knowledge between interlocutors that) the hearer wants to be kissed by the sender, but not killed. I will write P+ (P–) for declarative clauses that denote (un‑)desirable propositions in this sense.

 Michael Franke

I will call the sentences in (1) pseudo-imperatives,2 or PIs for short, which I take to be a label for a particular surface form: an imperative clause3 A is followed by either and or or which is followed by a declarative clause P. If we assume that in a given context the hearer has a clear preference whether the content expressed by the declarative clause P of a pseudo-imperative should become true or not, each sentence (1a–1c) is a natural argument for or against the act named by A: (1a) and (1c) are arguments for, (1b) is an argument against. However, sentence (1d) is not a natural thing to say at all: for fixed hearer-desirability of P it is simply infelicitous. In particular, the infelicitous (1d) is not an argument against the act named in A, although this might be expected when looking at the conditional sentences in (2) which correspond loosely to each sentence in (1).

(2)

a. b. c. d.

If you close the window, I will kiss you. If you close the window, I will kill you. If you don’t close the window, I will kill you. If you don’t close the window, I will kiss you.

(If A, then P+) (If A, then P–) (If not-A, then P–) (If not-A, then P+)

For the conditional sentences in (2) the picture is entirely symmetric: positive (negative) consequences of act A are an argument for (against) it, while positive (negative) consequences of not-A are an argument against (for) A. This symmetry in (2) is broken in (1) by the infelicity of (1d). This paper takes this Problem of Pragmatic Asymmetry (PoPA) observed in pseudoimperatives as its starting point. The puzzle can be approached from two sides. Firstly, (i) we should ask why there are instantiations of variable X so that (3a) becomes an argument against A, while there are no instantiations of Y that turn (3b) into an argument against A. (3) a. Do A and X will be the case/happen. b. Do A or Y will be the case/happen.

Secondly, (ii) we should ask why instantiations of X in (3a) may be desirable or undesirable to the hearer, while instantiations of Y in (3b) may not be hearer-desirable.4

.  This does not quite match Clark’s (1993) terminology where the term was chosen for its particular connotation. In this paper I do not want to hint at a possible difference between genuine imperatives and mere pseudo-cases. Here “pseudo-imperative” is just a name for a particular linguistic form. .  Let me justify this categorization: It is indeed not entirely self-evident that the first connect in sentences (1) is really an imperative clause. It might as well be an infinitive or a bare VP (see Bolinger, 1979, for some early discussion). Although this paper deals with English, I take the fact that other languages, such as German or modern Greek (cf. Han, 1998), have parallel constructions where the first connect morpho-syntactically is undeniably an imperative clause to be reason enough to reject the idea that in English these forms should be anything else than imperative clauses (contra Russell, 2007). .  Strictly speaking, these two questions really address two different problems, or at least two different challenges from the same problem set. Nevertheless, there is an obvious connection.

Pseudo-imperatives and other cases of conjunction/disjunction 

1.2  Some possible solution strategies There are several ways in which the PoPA might be solved. Let me sketch some possibilities in order to briefly discuss previous accounts and to situate the current contribution more clearly in the greater scheme of things.5

1.2.1  Pragmatics only Perhaps the most appealing account for the PoPA would be entirely pragmatic in terms of principles of how to say what and what not to say when, especially in order to influence someone else’s behavior or judgment.6 Here is a rough sketch of such a purely pragmatic account. Suppose that we start with the minimal assumption that the pseudo-imperatives in (1) are semantically equivalent to the conditional sentences in (2) one by one. Then, obviously, whatever pragmatic account for the infelicity of (1d) we would like to give, we would have to take into consideration that, after all, (2d) is acceptable. So we need to acknowledge at least some difference between the disjunction (1d) and the conditional (2d). In Franke (2005a) I suggested that the difference might be found in topicality or aboutness. With some intuitive plausibility we could say that conditionals (2c) and (2d) mention, refer to or talk about not-A. Disjunctions (1c) and (1d), on the other hand, mention or talk about A instead while still giving conditional information about what happens when A is not performed. Based on this intuition we might endorse the following Mention-Principle: Mention-Principle: Do not bring to attention an action that you do not want to be performed (because mentioning choices just makes them salient and more probable to be chosen), unless you immediately discredit what you mention (by stating negative consequences of it, for instance).

It is not crucial to find the Mention-Principle convincing as long as it helps illustrate what a purely pragmatic approach could look like. For it is palpable that only (1d)

For instance, suppose we have a satisfactory independent explanation for the preference bias in (3b) as an answer to question (ii). Then an answer to question (i) may be given on top of that. We could say that in order to be an argument against A, (3b) needs a desirable Y, because this is how conditional information influences choice of action. But (3b) cannot get a positive Y by assumption. Hence it cannot be an argument against A. But clearly not every such argument that connects (i) and (ii) needs to be accepted. In this sense, the two questions should be kept apart and each account of the PoPA should make clear whether the chicken or the egg part of the puzzle is solved first and how we get from one to the other. .  The present approach restricts itself to semantic and pragmatic considerations. More syntactic considerations are left for another occasion. .  Both van der Auwera (1986) and Clark (1993) propose a solution that fits into this category. For discussion and criticism of these and other accounts see Franke (2005b) and Schwager (2006).

 Michael Franke

v­ iolates this principle while all other sentences in (1) and (2) agree with it. So here we have a rough sketch of a pure pragmatic solution to (aspect (ii) of) the PoPA. One of the driving ideas of this paper is that a pure pragmatic explanation of the PoPA as outlined here, however plausible on its own, would ignore a considerable amount of (at least prima facie) relevant empirical data. It is therefore the central concern of this paper to give further, mostly authentic and in this context hitherto overlooked data in order to put PIs into a broader linguistic perspective. Doing so will support a particular alternative approach, namely the Connector Hypothesis which I will introduce in section 1.3 and whose plausibility I will discuss critically thereafter.

1.2.2  Forceless imperative clauses To appreciate the central hypothesis of this paper, it is advisory to contrast it with a superficially similar, but different solution strategy. The solution strategy I have in mind here is based on the following idea. Suppose that we can make plausible that imperative clauses are associated with directive force in most, but not all linguistic contexts. In particular, consider the following Force Hypothesis:7 Force Hypothesis: If an imperative clause is followed by conjunction and and a declarative it will not be associated with directive force, but if followed by disjunction or and a declarative clause it will.

Indeed, the Force Hypothesis gives us an answer to question (i) of the PoPA: PIs with conjunction, call them conjunctive pseudo-imperatives (ConjPIs), are not associated with directive force, but disjunctive pseudo-imperatives (DisjPIs) are. Hence, a ConjPI (3a) can be an argument for and against the content of its imperative clause A, but a DisjPI (3b) cannot be an argument against A, because here the imperative clause is always associated with the directive force that A should be performed. Obviously, (any serious careful formulation of) the Force Hypothesis needs independent support. What is needed is independent morpho-syntactic evidence that imperative clauses lose their directive force exactly in those contexts where they can be used to argue against the performance of the act they describe. Perhaps such independent evidence can be found. Again, I do not want to argue that there is no plausible continuation of this idea. At present, I want to suggest a similar, but different hypothesis for which independent evidence can be given easily in terms of parallel examples that do not involve imperatives at all. And once more, I argue that this approach has an

.  For any serious proposal along these lines more severe qualifications are necessary and, of course, have been spelled out by followers of this strategy such as Han (1998) and Russell (2007). To name just one example of such a refinement, we obviously have to require the absence of speech-act triggers such as please (see Bolinger, 1979). I omit the details and refer the reader again to the discussion in Franke (2005b) and Schwager (2006).

Pseudo-imperatives and other cases of conjunction/disjunction 

advantage also over the Force Hypothesis in that it places our explanandum in a wider context of relevant examples.

1.3  The Connector Hypothesis Let me start by suggesting and evaluating critically an account of the PoPA based on what I will call the Connector Hypothesis (CH). The idea behind the CH is that it is not pragmatic principles of debate or the pragma-semantic properties of imperatives that are responsible for the pragmatic asymmetry in PIs, but rather properties of the sentential connectors and and or. I want to suggest that connectors and and or in PIs have different discourse segmentation properties. To spell out the CH, I will therefore first briefly introduce this difference in discourse segmentation behavior in section 1.3.1. Subsequently, in section 1.3.2, I will investigate authentic PIs in order to assess, again on an intuitive basis only, their discourse segmentation behavior. This finally leads to a concrete formulation of the CH in section 1.3.3.

1.3.1  Discourse segmentation According to Mann and Thompson (1987), discourse interpretation is a three-step procedure consisting of: Segmentation: What are the units of discourse? Relation: Which units of discourse relate to each other and what is the relation between them? Coordination: Which hierarchical structure holds between related units of discourse; which ones, if any, are subordinate?

Especially in the context of sentential connections such as PIs whose connects are of different clause types the segmentation step is not trivial and deserves extra attention. For a simple motivating example, consider the sentences in (4).

(4) a. Drive past the gas station and then, after five minutes, you will see the roundabout. b. Drive past the gas station until you see the roundabout.

Intuitively, at speech-act level we have two discourse units in (4a), but only one in (4b): (4a) gives advice about driving in a certain direction followed by the statement that, if that direction is followed, the roundabout will be reached. In clear contrast, (4b) is just a directive to drive in a certain direction for a specific amount of time. In other words, the clause “you will see the roundabout” in (4b) is not realized as a speech-act in its own right, because of the (semantically subordinating) connector until, while in (4a) both connects of the (presumably semantically coordinating) connector and (then) give rise to two separate speech-acts. This simple observation suggests the following intuitive distinction: (an occurrence of) a sentential connection “X ⊗ Y”, where X and Y are sentences or clauses and

 Michael Franke

⊗ is a sentential connector such as until or and, is discourse separating if the connects X and Y are treated as separate discourse units and are thus associated with separate communicative events, or speech acts (in a particular discourse context); otherwise, if (an occurrence of) “X ⊗ Y” is treated as one communicative event and both connects together are associated with one speech act (in a particular discourse context), (this occurrence of) “X ⊗ Y“ is discourse integrating.8

1.3.2  Pseudo-imperatives in the wild The distinction between discourse separating and discourse integrating sentential connections is based merely on introspection. At present, I am not interested in theoretical refinement. Rather, I suggest to apply naive intuition about discourse segmentation behavior of connections to a selection of examples of PIs collected from various sources. To start with, here are three cases of ConjPIs:

(5) a. The creature sighed, ‘Come closer and I will tell you a great secret.’ I moved closer, until I could see my own breath condense on its skeletal shoulder. (http://elfwood.lysator.liu.se/libr/s/h/shalene/ random.html, 17.8.2006) b. Please send me the issues you are experiencing, and I will update this page, along with information when they are resolved. (http://www.bloglet.com, 17.8.2006) c. “And now rearrange yourself,” I said, “and in the meanwhile I will go and wash the baby.”    (http://www.worldwideschool.org/library/books/lit/ shortstories/ThroughRussia/Chap1.html, 17.8.2006)

The examples in (5) are all authentic examples corresponding more or less to the artificial (1a): for instance, under the likely assumption that the addressee wants to be told a great secret, the whole sentence(5a) urges him or her to come closer. The other examples are similar in this respect. But now ask yourself: are the examples in (5) discourse separating or discourse integrating connections? For instance in (5a), does the creature give a directive followed by a (conditional) statement or does the creature only assert a (conditional) statement? Maybe the following three readings are most plausible: (6) a. inform(“if you come closer, I will tell you a secret”) b. direct(“send issues”) & inform(“if you send, I will update”) c. direct(“rearrange yourself ”) & inform(“I will wash the baby”)

.  The ideas of discourse integrating connections and non-veridical discourse relations in Segmented Discourse Representation Theory (Asher and Lascarides, 2003) are related but not identical: the former is a property of a linguistic form (token), the latter is a property of a semantic object (token).

Pseudo-imperatives and other cases of conjunction/disjunction 

Example (5a) could well just be an assertion of a conditional, (5b) gives a clear directive followed by a conditional statement, and (5c) gives a directive followed by a nonconditional statement. There is plenty of room for disagreement about the judgments in (6a). But this is not crucial for the argument that I would like to make. There are other examples where things are less controversial. Here is one: (7) Say one more word and I’ll scream!  (From the song ‘Show me’ from the musical My Fair Lady by Frederik Loewe and Alan J. Lerner)

To my mind, example (7), which corresponds to (1b), can only be treated as a discourse integrating connection: a statement of a conditional as in (8a). (8) a. inform(“if you say one more word, I’ll scream”) b. direct(“say one more word”) & inform(“(if you do) I will scream”) c. direct(“don’t say one more word”) & inform(“(if you do) I will scream”)

Clearly, (7) does not get reading (8b), and even (8c) seems implausible, because it is unclear how an imperative clause can give rise to a directive with the exact opposite content of what is mentioned.9 To remove all doubt that there are examples of discourse integrating conditionallike ConjPIs consider finally the following example:

(9) ‘You see what kind of people they are,’ he said: his eyes moved around restlessly, he did not seem to be speaking to anyone in particular. ‘You think there is nothing and all the time the ground beneath your feet is ­rotten with tunnels. Look around a place like this and you would swear there wasn’t a living soul in miles. Then turn your back and they come crawling out of the ground. […]’ (J.M. Coetzee, Life and Times of ­Michael K, pp. 121–2)10

The last ConjPI in (9) certainly does not get reading (10b) or (10c), but simply (10a). (10) a. inform(“if you turn your back, they come crawling out”) b. direct(“turn you back”) & inform(“(if you do) they come crawling out”) c. direct(“don’t turn you back”) & inform(“(if you do) they come crawling out”)

.  In fact, both van der Auwera (1986) and Clark (1993) maintain that under certain circumstances an imperative clause can be interpreted like an ironic or sarcastic remark to mean the opposite of what has been said literally. Still I find it implausible to argue for analysis (8c) on these grounds, because in order to be perceived as ironic or sarcastic it is necessary that the literal content is assessed as if meant literally after all. But this is not intuitive for cases like (7) where it is not the case that the imperative clause is first taken literally and then reinterpreted along pragmatic principles. .  Page numbers refer to the Vintage 2004 paperback edition.

 Michael Franke

Example (9) is special, not only because it clearly does not involve an imperative speech act, but also because it involves a different modality than the examples that we have considered so far: other examples referred to the concrete immediate future, but (9) states a generic relationship between events. We could speak of generic ConjPIs here. The crucial observation is that generic ConjPIs are clear cases of discourse integrating ConjPIs with a conditional-like interpretation. The reason why this is crucial is because it contrasts with DisjPIs. First of all, look at some genuine examples: (11) a. The relevant Minister for Finance and the Budget says, ‘Put the brakes on or we will lose our European Union aid!’ (From corpus: Europarl (en) (EU-EN), 2049840) b. Don’t bother to resist, or I’ll beat you (From the song ‘The Beautiful People’ by Marylin Manson) c. Bush Tells UN, Make War or I Will (http://www.thenation.com/blogs/capitalgames?pid = 100, 17.8.2006)

In all examples in (11) the speaker urges the addressee to bring about some state of affairs or perform some action and further enhances this urge by a threat. In example (11a), for instance, the minister demands that the breaks be put on and threatens that if the brakes are not put on something bad will happen: (12) direct(“put brakes on”) & inform(“if brakes are not put on, we lose aid”)

In other words, sentence (11a) is a discourse separating connection. This is similar for the other examples in (11) and, it seems, for most, if not all DisjPIs. There is but one class of DisjPIs for which it is not clear whether they are discourse separating or discourse integrating. These are sentences like (13) which we could call generic DisjPIs because, like generic ConjPIs, they too refer to a generic conditional relation between events. (13) a. I’m telling you, working for a successful start-up is no fun. The atmosphere is so tensed. Agree to everything he says, or your boss fires you immediately. b. Speak at least six languages or you are not a cosmopolitan.

I will come back to generic DisjPIs in section 4 where I discuss whether these cases are a threat to the CH in its strong formulation given next.

1.3.3  The Connector Hypothesis: formulation & application The PoPA can be accounted for if we assume that conjunction and and disjunction or, as they occur in PIs, have a different discourse segmentation behavior: Connector Hypothesis (CH): While ConjPIs can be discourse integrating connections, in which case they get a pure conditional-like reading such as (6a), or discourse separating connections, in which case they get a speech-act ­conjunction

Pseudo-imperatives and other cases of conjunction/disjunction 

reading such as (6c), DisjPIs can be discourse separating connections only and they always get a speech-act conjunction reading as in (12).

Based on the CH the following pragmatic account of the PoPA can be given. If a DisjPI like (3b) is a discourse separating connection like (12), it will always contain a directive to perform act A. Hence, the whole disjunction cannot possibly be an argument against A. This answers part (i) of the PoPA, the question why a DisjPI cannot be an argument against the content of its imperative connect. As for part (ii) of the PoPA, the presence of a directive to perform A also explains why a (hearer-)desirable declarative connect Y is pragmatically infelicitous in (3b). If Y is desirable then the conditional statement “If not-A, then Y” urges the addressee not to bring about or perform A, but at the same time the addressee is told to do so by the directive. This is clearly incongruous, whence the infelicity. Like the Force Hypothesis, the CH also requires independent evidence in its support. In particular, there are three issues to be addressed in order to lend credence to the CH: firstly, the claim that there are discourse integrating conditional-like readings of English and needs support; this will be given in section 2. Secondly, we need evidence for the stipulated discourse separating readings of English or. To this end, section 3 gives corroborating examples of discourse separating or which, to the very best of my knowledge, have not been addressed in the relevant literature in this form. Finally, we have to justify the claim that DisjPIs can only be discourse separating. This is the Achilles tendon of the CH and I will address this issue critically when I turn to a closer investigation of generic DisjPIs in section 4.

2.  Conditional conjunction 2.1  The problem of conditional conjunction It is not surprising that and can realize a speech-act conjunction as in (6b) and (6c). Similarly it is not surprising that, in (6b), the declarative connect P of a ConjPI is interpreted in the context of A, as it may be considered a modal subordination phenomenon (Roberts, 1989) and has parallel examples in contexts different from ConjPIs such as in (14). (14) “I’m leaving one for all the others,” said Rabbit, “and telling them what it means, and they’ll search too. I’m in a hurry, good-bye.” And he had run off.  (A.A. Milne, The House at Pooh Corner, ‘In which Eeyore finds the Wolery and Owl moves into it’)

In (14) Rabbit’s prognosis that the others will search too is interpreted in the context of, i.e., conditional upon, the others receiving and understanding “Owl’s Mysterious Missage” (the subject of conversation at that particular point in the story).

 Michael Franke

What is much more surprising is the behavior of discourse integrating conjunction with its conditional-like readings, as in (6a). However, it has been observed before that conditional conjunctions, as we might want to call them, occur in a variety of contexts (cf. Culicover and Jackendoff, 1997; Schwager, 2006). Apart from ConjPIs, there are three further contexts in which and is naturally interpreted to express a conditional relationship between connects. First of all, we find conjunctions of declaratives (15), either with the simple present tense in both connects as in (15a), where, in the present case, we find a generic interpretation, or with a simple present tense in the first connect and a will-future in the second (15b), where, in the present case, we find a conditional interpretation relating to the immediate future course of events. (15) a. I am not sure he is wholly of this world. One tries to imagine him running a staging post for insurgents and one’s mind boggles. (J.M. Coetzee, Life and Times of Michael K, p. 130) b. ‘And if I climb the fence? What will you do if I climb the fence?’ ‘You climb the fence and I’ll shoot you, I swear to God I won’t think twice, so don’t try.’ […] ‘You climb the fence and I’ll shoot you dead, mister. No hard feelings. I’m just telling you.’ (J.M. Coetzee, Life and Times of Michael K, p. 85)

Second, conjunctions of NPs with declaratives, often with a negative polarity item expressing minimality (16), such as one more or any, also give rise to conditional readings (Culicover, 1970, 1972). (16) Bob: I’m real disappointed in you, Modesto; pullin’ a gun on an old saddle pal like that. Chico: One more word and I will kill you! Bob: One more word, huh? Let me see if I can think of one. How about g-r-e-a-s-e-r? Greaser? (Passage from the film One-Eyed Jacks (1961) by Marlon Brando)

The third and last case of conjunctions with a conditional reading are conjunctions of sufficiency-modal constructions, such as only have to or it’s sufficient to, and a declarative (17). (17) You only have to install the MSDTC once in Cluster Administrator and MSDTC will be configured on all nodes in the cluster. You do not have to install MSDTC manually on each node. (http://support.microsoft.com/kb/301600/, 21.8.2006)

In conclusion, there are indeed a number of examples for conditional conjunctions beyond PIs. This then supports part of the CH.

2.2  Semantics of conditional conjunction Of course, the observation that there are discourse integrating conditional readings of and also raises an interesting question: how can one of our prototypical coordinators

Pseudo-imperatives and other cases of conjunction/disjunction 

give rise to a conditional-like, i.e., a subordinate-clause-like, meaning? In particular, what do all the contexts in which and can get conditional readings have in common that together with a single reasonable semantic analysis of and the conditional readings can be explained? In order to address this problem, Schwager (2006), who assumes and applies the thesis that imperatives denote necessity modals, proposes that the first connects contribute a necessity modal, not only for ConjPIs but also for other examples of conditional conjunctions. Necessity modals in turn denote, in simplified terms, a necessity operator Nec(P,Q) that takes two propositions as arguments, a restrictor P and a domain Q, with the following obvious semantics: Nec(P,Q) = λ w . (w ∈ P → w ∈ Q). Schwager now proposes that due to the (topical) intonational properties of conditional conjunctions (deaccenting of first connect and lack of a low boundary tone before and), the content of the first connect is not mapped onto the domain of the necessity operator, but onto its restrictor while the content of the second connect is mapped onto its domain, thus yielding a conditional reading. However, Schwager herself notices some obvious shortcomings of this idea. First of all, it is unclear why (18a) does not get the same conditional reading as (18b).11 (18) a. You have to drink one more beer and I’ll leave. b. Drink one more beer and I’ll leave.

It is moreover not entirely obvious why sufficiency modals such as (17) should be analyzed as contributing a necessity modal. Similarly, Schwager’s own example (73b) of chapter 12.4, here (19), clearly contains a possibility modal. (19) You can even call him at MIDnight and he won’t be angry.

In the light of this, the stipulation of a covert necessity operator in the first connect of, for example, generic simple present tense clauses, as in (15a), is drawn into doubt as well. So how could we improve on the analysis? Let me sketch a very rough alternative idea. Culicover and Jackendoff (1997) proposed to account for the conditional reading of and in terms of a generic operator Gen(P)(Q). That a generic operator can be tweaked to model conditional readings is clear: very roughly Gen(P)(Q) means the same as the above necessity operator, namely that under all normal circumstances, if the restrictor P is met, Q is true.12 That the content of the first connect of conditional conjunctions is then mapped onto the restrictor of the generic operator can again be justified with reference to the

.  Notice that (18a) may get a different conditional reading that we are not interested in: “If you have to drink one more beer, then I’ll leave.”. .  Interestingly, in an entirely unrelated account, Krifka (1995), for instance, uses a generic operator with a semantics that is basically equivalent to the semantic analysis of conditional sentences as proposed, among others, by Kratzer (1991).

 Michael Franke

above-mentioned topical intonation properties. So suppose we had a satisfactory ­account for generic conditional conjunctions. Maybe it would not be too unlikely to assume that a metaphysical condition, i.e., a result relation of (immediate) possible future events, is just a special case of a generic condition, which relates possible events in a relevant time interval that is not restricted to the immediate future.13 A solution along these lines would help to explain why (18a) does not get the same conditional reading as (18b). Since we no longer have to stipulate a necessity operator hidden in the first connect, we are free to assume that the overt necessity operator in (18a) has to be part of the propositional content that enters the restrictor, if at all.14 Unfortunately however, the sketched idea does not help to account for the conditional readings of (17) and (19) either, and I will have to leave the issue as an interesting open problem for future analysis.

3.  Conjunctive disjunction There are certain standard cases of natural language disjunction that seem to have informed the notion of logical, truth-conditional disjunction. (20) is a case in point. (20) I don’t know exactly what John did. He either went to the cinema, or he stayed at home all night.

In (20) the connects of the disjunction are epistemic alternatives to one another. The speaker commits himself to the truth of neither. This is different for a class of examples that I would like to discuss in this section, where the speaker commits himself to both connects. The speaker in (21), for instance, intuitively says two things: that coherence is required and that lack thereof has undesirable consequences. (21) Therefore, we must start to be coherent in our intentions or we will be indulging in fine theoretical discourses which have absolutely no substance outside this Chamber. (Europarl (en) (EU-EN), 2698860)

The discourse separating or as in (21) gets a speech-act conjunction reading. It is in this respect that I will speak of conjunctive disjunctions.15

.  If this assimilation of metaphysic conditions to generic conditions could be verified, there are hardly any cases left uncovered, because epistemic conditions (“If the gardener has not killed the baroness, then the butler has.”) or speech-act related conditions (“If you’re hungry, there are biscuits on the shelf.”) cannot be expressed by conditional conjunctions, if only very marginal. .  We can thus account for the fact that the only available conditional reading is the one mentioned earlier: “If you have to drink one more beer, then I’ll leave.” .  An anonymous reviewer correctly remarks that my terminology may be confusing. So let me clarify: conditional conjunctions are conjunctions with a conditional reading at content level, whereas conjunctive disjunctions are conjunctions at speech-act level.

Pseudo-imperatives and other cases of conjunction/disjunction 

To the best of my knowledge, the fact that disjunction can get such conjunctive readings has received little, if any, attention in the literature, although this phenomenon is far from infrequent, as I will show with a multiplicity of wild examples. I would like to substantiate this claim in the following by a cartography of conjunctive disjunctions. The main aim, still, is to corroborate the CH. But, in addition to this, I think that the examples discussed in this section might be of independent interest for semantic analyses of disjunction. That is why I will go into this in some more detail.

3.1  Hypothetical conjunctive disjunctions The most frequently occurring examples of conjunctive disjunction are, like (21), ­connections with expressions of necessity in the first connect. Interestingly, necessity can be expressed not only by an overt modal as in (21), but also rather indirectly as in (22): (22) a. The key is to keep it simple or it will not work, it will not remove the obstacles. (Europarl (en) (EU-EN), 272055) b. With regard to control – as I said at the end of the speech – we propose that the report we present every Spring to Parliament should include an assessment of the sustainability element of the decisions taken. This is crucial, or we will end up with conflicting decisions. (Europarl (en) (EU-EN), 5718224)

All of the examples in (21) and (22) are discourse separating connections: speech-act conjunctions, in analogy to our previous analysis of DisjPIs in (12). A speech-act conjunction analysis of conjunctive disjunctions is further corroborated by the examples in (23) where the first connect is an explicit performative. (23) a. I would therefore ask you to clarify this point, or I will not be a bearer of good news for the Italian farmers. (Europarl (en) (EU-EN), 2637413) b. Mr Spencer is asking for the floor again. I would ask him not to make personal allusions or we will never finish. (Europarl (en) (EU-EN), 18178233)

The same holds of example (24) where the speaker commits himself to some future action in the first connect and gives a reason for his preference for this course of action in the second connect. (24) Mr President, I shall not now echo the compliments paid to Mr Lamy or he will start to blush. (Europarl (en) (EU-EN), 28355082)

Similar to this are cases where the speaker makes a promise in the first connect and then gives a reason why he made it:

 Michael Franke

(25) a. No cookies, no pop-ups and no evil tracking devices, I promise. Or you can have a bite of my dinner. (http://absolutely-fuzzy.com/blog/?m=200406, 16.7.2006) b. I will have these things judged by this weekend! Po will make me do it!!!!! I promise or you can kick my butt!!!!!! (http://www.writersco.com/the%20Day%20Before%20The% 20 World%20Ends, 16.7.2006)

As a final example, consider (26) where the first connect expresses a hope and the second connect delivers a reason for having that hope. (26) Senator Petten and I met quite frequently, and we would discuss the business that was to be done that week. We would then say, ‘I hope to God the leaders do not find out or we will never get it done.’ Therefore, do not tell them what you are planning for that week, and you will get things accomplished. (Hansard Senate e (HANSARD_SENATE_E), 1733312)

All the examples, (11) and (21)–(26), have a particular discourse format in common: they are all conjunctive disjunctions “X or Y” whose first connect X presents a topic χ, which I will refer to henceforth as the χ-component (of the disjunction “X or Y”), as necessary, preferred or desirable. The second connect is interpreted in the context of not-χ. χ is a mere (immediate-)future possibility which is why we could speak of hypothetical conjunctive disjunctions here in contrast with a factual variety that will be discussed in the next section. Crucially, the χ-component of “X or Y” need not be identical to the semantic content of X. It may not even be contained in or referred to explicitly in X, as example (27) makes clear. (27) That is enough points of order, or we will never get on to the items on the agenda. (Europarl (en) (EU-EN), 4859513)

The χ-component of the conjunctive disjunction in (27) is the content of an indirect speech act associated with the first connect, namely that one should stop collecting more points of order. So, if we assume that α is the speech act associated with X in context, we can represent the speech-act conjunctive reading of “X or Y” as in (28). (28) α(χ) & inform(“if not χ, then Y”)

(28) gives the general discourse format of a hypothetical conjunctive disjunction. It now seems plausible that DisjPIs with their discourse format in (12) are a special case of hypothetical conjunctive disjunctions. For hypothetical conjunctive disjunctions “X or Y” whose first connect X presents a not-yet-realized state or action χ as necessary, preferred or desirable, we expect that the content of the second connect Y has a negative connotation. Indeed, the preference bias noted for DisjPIs generalizes to all hypothetical conjunctive disjunctions, for a (hearer‑)desirable second connect seems impossible:

Pseudo-imperatives and other cases of conjunction/disjunction 

(29) ? It is necessary/preferred/desirable that this happens, or you will get my ­Ferrari.

This raises the question whether we also find the reversed pattern: are there hypothetical conjunctive disjunctions “X or Y” whose χ-component is presented as undesirable in X, but whose second connect Y has a positive connotation? Although conceivable in principle, this pattern does not seem to occur. Constructed examples sound decidedly odd too: (30) ? It is undesirable that this happens, or you will get my Ferrari.

That means that the previously attested bias puzzle in connection with DisjPIs is part of a larger whole. Hypothetical conjunctive disjunctions all present their χ-component as desirable and have a negatively connoted second connect which is interpreted as an epistemic alternative to χ, i.e., have a double preference bias.16

3.2  Factual conjunctive disjunctions This double preference holds only for hypothetical conjunctive disjunctions. There is another class of conjunctive disjunctions where this preference bias is not present. These could be called factual conjunctive disjunctions, because, as we will see shortly, their χ-components refer to true, established or presupposed states of affairs. There are furthermore two kinds of factual conjunctive disjunctions: one is evaluative, the other evidential. In evaluative factual conjunctive disjunctions some true state of affairs χ is presented as fortunate or unfortunate in the first connect, while the second connect states an epistemic alternative to χ which is either negative or positive. As χ is presented as a true state of affairs, the second connect usually contains a counterfactual would-modal. The examples in (31) present some true state of affairs χ as fortunate and give a negatively connoted counterfactual epistemic alternative. (31) a. “Lucky we know the Forest so well, or we might get lost,” said Rabbit half an hour later, and gave the careless laugh which you give when you know the Forest so well that you can’t get lost. (A.A. Milne, The House at Pooh Corner, ‘Tigger is unbounced’) b. . . . . . . . . the wilderness . . . . . . . . remember the jumping over the drains and the endless footpath? :lol: …..that was fun :) yup. Luckily we did that, or we could have been walking farther away up to the Kallang Basin and Nicoll Highway! :eek: (http://skyscrapercity.com/archive/index.php/t-106394.html, 16.7.2006)

.  Gómez-Txurruka (2002) also stipulated this double preference bias for readings of or that she called “conditional alternative”.

 Michael Franke

In contrast, the examples in (32) present some true state of affairs χ as unfortunate and give a positively connoted counterfactual epistemic alternative. (32) a. It is truly unfortunate that you are so far from us, or we would be weekly guests at the gallery. (http://www.cordair.com/accolades.htm, 12.7.2006) b. Christine, it’s unfortunate that their version of the Creative Commons license does not allow derivative versions to be created or we could fix this problem (and also correct their numerous misspellings and other typos). (http://www.foresight.org/nanodot/?p=2295, 21.8.2006)

In all examples in (31) and (32) the second connect with its particular negative or positive connotation seems to give a reason why the state of affairs χ is fortunate or unfortunate, respectively. A preference bias, as for hypothetical conjunctive disjunctions, does not exist. Finally, there are evidential factual conjunctive disjunctions such as (33) which do not involve preferences at all, but which are entirely epistemic in character. (33) a. It is not a very exciting topic to be discussing but, as we all know, it is very important or we would not be here. (Europarl (en) (EU-EN), 11272804) b. Finally, I would also appeal to our colleagues in the national parliaments, since they apparently have little interest in the matter either, or they would be pressing their governments and raising a debate in the national parliaments on why the Member States are so perverse when it comes to taking the necessary decisions. (Europarl (en) (EU-EN), 16403391)

Whereas in the evaluative cases, (31) and (32), the second connect gives a reason for the evaluative judgment expressed in the first, in the evidential cases (33) the second connect gives a reason for the epistemic judgment that the χ-component is believed true, or at least highly probable. The discourse format of factual conjunctive disjunctions in (34), be they evaluative or evidential, is basically the same as that of the hypothetical variant in (28). (34)

α(χ) & inform(“if had not χ, then would Y ”)

The only difference is that, due to the factuality of χ and evidenced by the frequent wouldmodal in the second connect, the conditional relation involved is c­ ounterfactual.

3.3  Remarks on the semantics of disjunction The above exposition raises the question whether conjunctive disjunctions and the discourse analysis that I suggested in (28) have any relation to free-choice readings of disjunction (Ross 1941; von Wright 1968) and recent conjunctive semantics for

Pseudo-imperatives and other cases of conjunction/disjunction 

disjunction (Zimmermann 2000; Geurts 2005) that have been proposed to account for free-choice readings.17 Let me briefly comment. The free-choice reading of disjunction (35a), which is present to a lesser degree also in (35b), is the conjunctive reading according to which the speaker permits that the hearer may take an apple and that he may take a pear (but not both). (35) a. You may take an apple or a pear. b. You may take an apple or you may take a pear.

In very vague outline, this free-choice effect can indeed be conceived of as a special case of a conjunctive disjunction. The analysis of (35) according to the scheme in (28) would then be a reading in which the modal may gets a performative reading (Lewis, 1979) to the effect that a permissive speech act is given: (36) permit(“take an apple”) & inform(“if you don’t take an apple, you may take a pear”)

We might then assume that the modal in the asserted conditional gets a performative reading too, so that the assertion in (36) becomes a conditional permission and the whole sentence comes out as a speech-act conjunction where the speaker allows the hearer to take an apple and a pear.18 Suffice it to say that it is at least not entirely implausible that the free-choice conjunctive reading of disjunction could be conceived of as a discourse separating disjunction. However, it is widely held that the free-choice effect does not arise in all contexts, or that it can be cancelled (Kamp 1978) as in (37) in which case we get an epistemic reading with the modal used descriptively: the speaker does not know what is permitted (by some other source). (37) a. You may take an apple or a pear. But I don’t know which. b. You may take an apple or you may take a pear. But I don’t know which.

Now it is interesting to see that for most, if not all, examples of conjunctive disjunction that I have given above the performative reading with its discourse format (28) is most salient, but that an epistemic reading is also available. For instance, example (21), repeated here, could also get an epistemic reading in the vein of (20).

.  Thanks to various commentators and an anonymous reviewer for raising this issue. .  A more complete investigation has to wait for another occasion. But let me nevertheless make two short comments: (i) Depending on the interpretation of the conditional permission we may account for the felt exclusivity that the hearer may not take both apple and pear at the same time. (ii) Notice however that this sketchy idea depends on taking the long form (35b) as basic for the free-choice effect and the short form (35a) as elliptical. This is at odds with the observation that the free-choice reading is not the preferred reading of the long form (35b).

 Michael Franke

(21) Therefore, we must start to be coherent in our intentions or we will be indulging in fine theoretical discourses which have absolutely no substance outside this Chamber. (Europarl (en) (EU-EN), 2698860)

In this case, the speaker of (21) would say that the true state of affairs is one out of two possible candidates: either coherence is required or the addressees will be indulging in a particular form of discourse. It seems, then, that conjunctive disjunctions have performative readings, not only of certain modals, but also of explicit performatives (23), commitments (24), permissions (25) etc. All of these may in principle also have an epistemic reading, no matter how implausible the context would be that would give us such a reading. So the relation between free-choice disjunction, whose leading examples are permissions and obligations, and conjunctive disjunctions seems to be that the former are a special case of the latter. Do we need a specialized semantics of disjunction as a conjunction of modalized propositions in the vein of Zimmermann (2000) or Geurts (2005)? Not necessarily, it seems.19 For the difference between the performative and the epistemic reading need not necessarily lie in the semantics of the disjunction. The difference may simply be the difference between a performative or a descriptive use of some linguistic form.20 In case of a performative use of the disjunction “X or Y” we associate with X some speech-act α(χ) with the content χ. It is this χ that is then filtered out as the epistemic content of the speech-act α(χ) and taken up by the disjunction. In contrast to this, in case of a descriptive reading of the disjunction “X or Y” the whole propositional content of X is the epistemic unit that the disjunction is sensitive to. It is then perhaps not too far-fetched to assume that in both performative and descriptive readings of a disjunction “X or Y” the semantics of the disjunction are the exact same, only that different propositions are considered the epistemic content of the first connect, depending on its use. So for a performative reading of X in “X orY” we get (38a), while for a descriptive reading of X in “X or Y ” we may get (38b). (38) a. α(χ) & inform(“χ or Y ”) b. inform(“X or Y ”)

This behavior of or is plausible, if we think of it as a connector of epistemic possibilities: disjunction seems sensitive to the epistemic minimal unit (cf. Pasch et al. 2003) of the first connect.21

.  In fact, it is hard to conceive how such an approach could deal with (27) where the χ-component is not even mentioned in the first connect. .  Whether an expression is interpreted as being used performatively or descriptively may very well be a matter of whether the speaker is taken to be competent or not (cf. Zimmermann, 2000). .  We are now in a position to explain the relation between or and otherwise: or may only refer to the epistemic minimal unit of the first connect, whereas otherwise has much wider

Pseudo-imperatives and other cases of conjunction/disjunction 

3.4  Embedded conjunctive disjunctions Let us finally note a set of examples that slightly complicate the picture sketched so far. An analysis of conjunctive disjunctions as discourse separating connections, i.e., as speech-act conjunctions, seems very plausible for all examples discussed. If correct, examples of embedded conjunctive disjunctions will require special scrutiny. (39) a. They are there, threatening campesinos that they must plant coca or they will die. (Europarl (en) (EU-EN), 7760017) b. Aerie forces you to tell her you love her or the romance is over. (http://www.gamebanshee.com/baldursgateii/npcs/ aerie.php, 26.6.2006) c. That person showed me a letter from a banking institution asking for additional guarantees or the institution would demand full payment of its loan within 48 hours. (Hansard House e (HANSARD_HOUSE_E), 8151944)

To account for embedded occurrences of conjunctive disjunctions the notion of discourse separation simply has to encompass reported speech.

3.5  Intermediate summary In conclusion, conjunctive disjunctions are a highly interesting topic with a lot of challenges for semantics which are beyond the scope of this paper. At present our main concern is still the PoPA. This section tried to corroborate the part of the CH that claimed that there are discourse-separating disjunctions, by providing a wide range of parallel cases of conjunctive disjunctions.

4.  The Generic Challenge 4.1  The problem of generic DisjPIs So far we have made plausible an account of the PoPA based on the CH by pointing out that conditional conjunctions and conjunctive disjunctions are phenomena that go well beyond PIs. We should therefore accept the claim that some ConjPIs are conditional conjunctions and that some DisjPIs are conjunctive disjunctions. What is left to argue is that all DisjPIs are conditional disjunctions.

anaphoric possibilities. This is shown nicely by the following examples from Webber et al. (2003): i. a. If the light is red, stop, otherwise you might get hurt. b. If the light is red, stop, or you might get hurt. ii. a. If the light is red, stop, otherwise you may proceed. b. ? If the light is red, stop, or you may proceed.

 Michael Franke

I have already indicated that this might not be clear for generic DisjPIs (13), repeated here for convenience: (13) a. Working for a successful start-up is no fun. The atmosphere is so tensed. Agree to everything he says, or your boss fires you immediately. b. Speak at least six languages or you are not a cosmopolitan.

The problem with generic DisjPIs is that it is not entirely clear what discourse format generic DisjPIs have, in particular whether they are discourse separating connections or not. Intuitively, we would probably like to conceive of them as one communicative event, but then not as a statement with mere conditional content, but rather as a single goal-oriented directive that presents something as necessary or required for a particular purpose. But if generic DisjPIs are to be interpreted as one communicative event only, albeit some sort of informative directive, this contradicts the CH in its present formulation which claimed that all DisjPIs are discourse separating connections. In order to decide the question whether generic DisjPIs are discourse separating or not, it does not suffice to rely on intuition alone. Therefore I would like to consider evidence for and against semantic subordination in conditional conjunctions and conjunctive disjunctions that was proposed by Culicover and Jackendoff (1997). I will extend their discussion with special emphasis on generic DisjPIs, but conclude that the evidence is not decisive.

4.2  Pseudo-coordination vs. asymmetric coordination Since we are in doubt whether all occurrences of DisjPIs are discourse separating, we should not speak of conjunctive disjunctions when referring to all of them. So, let us call the use of or in all DisjPIs and other clear cases of conjunctive disjunction more generally ‘explanation-or’. Culicover and Jackendoff (1997) argue that conditional conjunctions are pseudocoordinate, i.e., coordinate in syntax, but subordinate in conceptual structure, while explanation-or is, though asymmetric, coordinate on both levels of analysis. Two arguments are given for why conditional conjunction is a subordination in semantics, but explanation-or is not. These two arguments rest on observations concerning NPIlicensing (see section 4.2.1) and the possibility of cataphoric binding by quantifiers across connects (see section 4.2.2), respectively.22

.  Culicover and Jackendoff assume that if a connection behaves like a conditional with respect to NPI-licensing and cataphoric binding, then this is an argument for it being a case of subordination semantically. It is a further step to take conditional-likeness as evidence for discourse integration, but this is what I will do. If this further step is incorrect, so much the better for the Connector Hypothesis.

Pseudo-imperatives and other cases of conjunction/disjunction 

4.2.1  NPI-licensing The first argument for semantically subordinate and, but coordinate or in PIs is that only the former allows negative polarity items (NPIs) in the first connect. (40) is Culicover and Jackendoff ’s (53c). (40)

Say anything and/*or I’ll call the police.

As conditionals quite clearly allow NPIs in their if-clauses, Culicover and Jackendoff argue that this parallel is evidence for subordination in case of conditional conjunction and for coordination in case of explanation-or. However, the impossibility of NPIs in the first connect of explanation-or is not an argument against subordination. This is so, because the parallel in NPI-licensing behavior between conditionals and PIs is actually more complicated than suggested by Culicover and Jackendoff ’s argument (cf. Lawler, 1975). NPIs are licensed in ConjPIs only if the imperative connect gets a negative or, to some minor extent, a generic interpretation. This then is entirely parallel to the NPI-licensing in the if-clauses of the corresponding conditionals, as evidenced in (41) and (42). (41) (42)

a. b. a. b.

If you say anything to anyone about this, I’ll kill you. Say anything to anyone about this and I’ll kill you. ? If you say anything to anyone about this, I will give you my Ferrari. ? Say anything to anyone about this and I will give you my Ferrari.

The crucial point here is that not all conditionals license NPIs in their if-clause. This was observed by Lakoff (1970) who gave the contrast pair in (43), which shows how NPI-licensing has a pragmatic dimension and is susceptible to the kind of argument that is being made with the assertion of a conditional. (43) a. If you eat any loxo, I’ll {batter you / ??give you whatever you like}. b. If you eat some loxo, I’ll {?batter you / give you whatever you like}.

But then, if there are conditionals which do not license NPIs in their if-clauses due to some rhetorical effect of NPIs, then the impossibility of NPIs in the first connect of explanation-or-connections should not be an argument against subordination, unless one were willing to claim that also certain conditionals are not subordinate in semantics. Rather, NPI-licensing should be dismissed as a criterion for semantic subordination for these reasons, or should at least not be counted as an argument against a subordination in the semantics of explanation-or-connections.

4.2.2  Cataphoric binding Culicover and Jackendoff (1997, pp. 203 – 4) present interesting binding data to support the idea that conditional conjunction is subordinating in conceptual structure, while explanation-or is not. They observe that conditional conjunction allows for cataphoric binding by a quantifier phrase with every across the connects, just as conditionals do (44a), but that apparently explanation-or does not (44b) (=(53b) in Culicover and Jackendoff 1997).

 Michael Franke

(44) a. If you give himi enough bribes, every senatori will vote for the president’s proposal. b. Give himi enough bribes and/*or every senatori will vote for the president’s proposal.

Possibilities of cataphoric binding by quantifiers across connects might prima facie seem like a genuine test for subordinateness, but there is room for doubt that this is so. Although the or-variant of sentence (44b) seems odd, especially after having read the and-variant, this might be due to quite different reasons. Very likely we are inclined to assume by default that senators need to be bribed to vote for the president’s proposal, not against. This scenario is made salient by the and-variant of (44b), but renders the or-variant of (44b) unintelligible. In contrast to that, there are other examples of explanation-or with cataphoric binding across connects where little doubt about their acceptability arises. Drawing on juvenile gender stereotypes, for instance, (45a) was judged acceptable by 7 out of 9 informants, marked by 2, while no one found it entirely unacceptable on a scale including ‘acceptable’, ‘odd’, ‘marked’ and ‘unacceptable’. This was, nota bene, the exact same distribution as that obtained for the conditional sentence in (45b). (45) a. Tell heri every now and then that you love heri, or every girli will leave you sooner or later. b. If you don’t tell heri every now and then that you love heri, every girli will leave you sooner or later.

From this we should conclude, contra Culicover and Jackendoff, that at least under certain circumstances cataphoric binding by quantifiers across disjuncts is possible. As a matter of unfortunate fact, the situation is, as usual, far more complicated. Cataphoric binding possibilities seem to depend crucially on the kind of modality that is at stake. Unlike the examples in (45) where a generic relationship is expressed, examples (46) relate to the concrete future course of events, i.e., they express a metaphysical conditional relationship. The striking observation is that although 7 out of 9 informants found the conditional (46b) acceptable or odd, 7 out of 9 informants found the DisjPI (46a) marked or unacceptable. (46) a. Invite heri, or every girli from our rugby team will feel neglected. b. If you don’t invite heri, every girli from our rugby team will feel neglected.

Moreover, in the group of 7 informants who found (46b) acceptable or odd, 5 found it odd. This means that there is an acceptability contrast not only between DisjPIs and conditionals in case of metaphysical modality, but also between generic and metaphysical variants of either sentence. One possible conclusion to be drawn from all this is that we have to be careful with Culicover and Jackendoff ’s second test for subordination, as judgments seem to depend on a variety of factors, such as default expectations and the kind of conditional relationship that is expressed. Yet if we were to hold on to the test, which had some prima facie

Pseudo-imperatives and other cases of conjunction/disjunction 

plausibility on its side, another possible conclusion is that there are some conditionallike, subordinate instances of DisjPIs, namely those that express a generic idea.

4.3  The Connector Hypothesis revisited I will leave the matter whether generic DisjPIs refute the CH undecided. But suppose that generic DisjPIs are not discourse separating connections. If we nevertheless wanted to account for the PoPA, we could resort to a weaker formulation of the CH which includes the possibility of discourse integrating DisjPIs. We could then try to defend that such discourse integrating DisjPIs are not merely unbiased conditional statements, but rather a conventionalized way of expressing purpose-related necessity. This would still explain the preference bias and accord with intuition. But unfortunately, this weaker hypothesis could no longer be made plausible by parallel uses of or alone. In its defense we would have to find support for the claim that if a disjunction communicates a conditional relationship it communicates purpose-related necessity. In conclusion, if we accept that there are discourse integrating, conditional-like uses of disjunction, like generic DisjPIs, the PoPA is replaced by a different problem, namely the problem of conditional disjunctions: why do generic DisjPIs express purpose-related necessity only and not also unbiased conditional relationships? Seen in this light, the CH is a possible explanation with independent evidence for the bulk of cases that fall under the PoPA but also gives rise to an interesting, more refined, followup problem.

5.  Conclusion This paper addressed PIs as sentential connections and presented their diverse discourse formats in detail. The Connector Hypothesis was advanced as the basis of a possible solution to the Problem of Pragmatic Asymmetry. In support of the CH, PIs were shown to be embedded in a broader context of non-standard conjunctions and disjunctions. Conditional conjunctions and conjunctive disjunctions presented themselves as overarching hitherto neglected linguistic troublemakers with their own respective semantic and pragmatic challenges. This paper contributed an in-depth classification of conjunctive disjunctions and showed that the PoPA too is part of a larger whole, as all hypothetical DisjPIs display a double preference bias. Doubts about the generality of the CH finally arose from generic DisjPIs. These examples appeared, contrary to the CH, discourse integrating connections with just one informative communicative event. I concluded that if generic DisjPIs were indeed discourse integrating, this would diminish the explanatory scope of the CH, but it would also leave us with a new, smaller and more concrete explanandum which we might hope is easier to account for than the one we started out with.

 Michael Franke

References Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation. Cambridge: CUP. van der Auwera, Johan. 1986. Conditionals and speech acts. In On Conditionals, Elizabeth Closs Traugott, Alice ter Meulen, Judith Schnitzer Reilly & Charles A. Ferguson (eds), 197–214. Cambridge: CUP. Bolinger, Dwight. 1979. Is the imperative an infinitive? In Meaning and Form, Dwight Bolinger (ed.), 152–182. London: Longman. Clark, Billy. 1993. Relevance and ‘pseudo-imperatives’. Linguistics and Philosophy 16: 79–121. Culicover, Peter W. 1970. One more can of beer. Linguistic Inquiry 1: 366–369. Culicover, Peter W. 1972. OM-sentences. Foundations of Language 8: 199–236. Culicover, Peter W. & Jackendoff, Ray. 1997. Semantic subordination despite syntactic coordination. Linguistic Inquiry 28(2): 195–217. Franke, Michael. 2005a. How and how not to employ discourse relations to account for pseudoimperatives. In Proceedings of the 15th Amsterdam Colloquium, Paul Dekker & Michael Franke (eds), 83–88. Amsterdam: ILLC. Franke, Michael. 2005b. Pseudo-imperatives. MA thesis, Universiteit van Amsterdam. Geurts, Bart. 2005. Entertaining alternatives: Disjunctions as modals. Natural Language Semantics 13: 383–410. Gómez-Txurruka, Isabel. 2002. The semantics of natural language disjunction or. Ms, ILCLI Donostia-San Sebastián. Han, Chung-Hye. 1998. The Structure and Interpretation of Imperatives. PhD dissertation, ­University of Pennsylvania. Kamp, Hans. 1978. Semantics versus pragmatics. In Formal Semantics and Pragmatics for ­Natural Languages, Franz Guenthner & Siegfried Josef Schmidt (eds), 255–287. Dordrecht: Reidel. Kratzer, Angelika. 1991. Modality. In Semantik: Ein internationales Handbuch der zeitgenossischen Forschung, Arnim von Stechow & Dieter Wunderlich (eds), 639–650. Berlin: Walter de Gruyter. Krifka, Manfred. 1995. Focus and the interpretation of generic sentences. In The Generic Book, Gregory Norman Carlson & Francis Jeffry Pelletier, (eds), 238–264. Chicago IL: Chicago University Press. Lakoff, Robin. 1970. Some reasons why there can’t be any some-any rule. Language 45: 608–615. Lawler, John M. 1975. Elliptical conditionals and/or hyperbolic imperatives: some remarks on the inherent inadequacy of derivations. In Papers from the 11th Regional Meeting. Chicago Linguistic Society, 371–382. Chigaco Linguistic Society. Lewis, David. 1979. A problem about permission. In Essays in the Honor of Jaakko Hintikka, Esa Saarinen, Risto Hilpinen, Ilkka Niinluoto & Merrill Provence Hintikka (eds), 163–175. Dordrecht: Reidel. Mann, William C. & Thompson, Sandra A. 1987. Rhetorical structure theory: A framework for the analysis of text. IPRA Paper in Pragmatics, 1: 79–102. Pasch, Renate, Brauße, Ursula, Breindl, Eva & Waner, Ulrich Herman. 2003. Handbuch der deutschen Konnektoren. Berlin: de Gruyter. Roberts, Craige. 1989. Modal subordination and pronominal anaphora in discourse. Linguistics and Philosophy 12: 683–721. Ross, Alf. 1941. Imperatives and logic. Theoria 25(7): 53–71.

Pseudo-imperatives and other cases of conjunction/disjunction  Russell, Benjamin. 2007. Imperatives in conditional conjunction. Natural Language Semantics 15: 131–166. Schwager, Magdalena. 2006. Interpreting Imperatives. PhD dissertation, Johann-Wolfgang-Goethe Universtität zu Frankfurt am Main. von Wright, Georg Henrik. 1968. An Essay on Deontic Logic and the Theory of Action. Amsterdam: North-Holland. Webber, Bonnie, Stone, Matthew, Joshi, Aravind & Knott, Alistair. 2003. Anaphora and discourse structure. Computational Linguistics 29(4): 545–587. Zimmermann, Thomas Ede. 2000. Free choice disjunction and epistemic possibility. Natural Language Semantics 8: 255–290.

From discourse to “odd coordinations” On Asymmetric Coordination and Subject Gaps in German Ingo Reich*

University of Tübingen Subject gaps in coordinate structures (i.e., SLF-Coordination, or simply SLFC) and Asymmetric Coordination (AC) in the sense of Höhle (1983, 1990) exhibit specific semantic (a ‘fusing’ interpretation) and syntactic properties (e.g., verb fronting in non-initial conjuncts). This paper ties together these characteristics by considering them an immediate consequence of the fact that in AC and SLFC properties typically attributed to ‘coordinate’ structures in discourse show up in ‘grammaticalized’ form. As far as the analysis of SLFC is concerned, it is furthermore argued that the dropping of the subject (and only the subject) is in a sense parasitic on the special fusing semantics of the construction. That dropping the subject systematically results in an overt V1-structure is traced back to the way covert operators are licensed and identified in specifier positions in German (paralleling Topic Drop). Keywords: asymmetric coordination, subject gaps, ellipsis, event semantics

1.  Introduction German (like some other Germanic languages, for example, Dutch) exhibits an ­interesting, and at the same time recalcitrant construction, which we can describe somewhat sloppily as Subject Gaps in Coordinate Structures (SLFC).1 As it turns out, SLFC is closely related to another kind of ‘non-standard’ coordination observable in German (see Höhle 1990): so-called Asymmetric Coordination (AC).

*  I’d like to thank Veronika Ehrich, Cathrine Fabricius-Hansen, Angelika Kratzer, Claudia Maienborn, Jürgen Pafel, Wiebke Ramm, Marga Reis, Arnim von Stechow, Wolfgang Sternefeld, Hubert Truckenbrodt, Angelika Wöllstein and three anonymous referees for helpful comments and related discussion. Thanks to Kirsten Brock for checking my English. .  The term “SLF-coordination” is due to Höhle (1990) and is shorthand for “Subject Lacking in F-structure,” where “F-structure” in turn stands for “Fronted structure,” a technical term that refers to sentences which show overt verb fronting; see Höhle (1986) for details.

 Ingo Reich

While previous approaches to SLFC and AC have focused almost exclusively on their syntactic side (Reis 1993, and Frank 2002, being notable exceptions), this paper takes as its starting point the constructions’ semantic and pragmatic peculiarities, and seeks to pull these together with their more formal ­characteristics. I will argue that properties typically attributed to coordination in discourse show up in AC and SLFC in ‘grammaticalized’ form, and thus determine both the constructions’ ­specific semantic and syntactic properties (among other things, verb fronting in non-initial ­conjuncts). Before laying out the details of my proposal, however, let me first introduce these characteristics of SLFC and AC, characteristics that in particular delineate this construction from what we may call ‘standard’ (symmetric) coordination.

1.1 

Subject gaps in coordinated structures (SLFC)

With this aim in view, consider the particularly interesting example (1), taken from a German online newspaper. Following Höhle (1983) SLFC can be characterized by three distinct properties: (i) Within non-initial conjuncts the finite verb is fronted; see tut (‘pretends’) in (1). (ii) Even though property (i) tells us that SLFC is most likely some case of sentential coordination (verb fronting requires a left periphery), the noninitial conjunct lacks a subject.2 (1)

da stellt sich jemand vor die Mikrofone und there stands refl somebody in front of the microphone and tut so, [als könne er etwas erklären] pretends part, [as if could he something explain] ‘Somebody stands there in front of the microphone, and pretends to be able to explain something’ (zeit.de, 26 January 2006)

But it is property (iii) that distinguishes SLFC from phrasal C’-coordinations like the one in (3), and which establishes SLFC as a separate phenomenon: Whereas in the case of (3) the subject jemand (‘somebody’) – which fills the ‘shared’ prefield (“Vorfeld”) position left adjacent to the position of the fronted finite predicate – c-commands (and, thus, binds) the base positions t1 of the fronted subject in both conjuncts, this is not the case with SLFC; see (2).

(2) da [[stellt sich jemand vor die Mikrofone] und [tut SL so, …]]



(3) jemand1 [[stellt t1 sich da vor die Mikrofone] und [tut t1 so, …]]

.  In a recent paper, Fortmann (2005) argues that it is not the subject per se, but in fact the highest argument of the verb (which is not necessarily the subject in German) that is dropped in SLFC; see, however, Reich (2007a) for critical discussion and a rejection of this position.

Odd coordinations 

In (2) the subject jemand of the first conjunct is located after [!] the fronted finite ­predicate stellt (‘stands’), i.e., within the middle field (“Mittelfeld”) of the first ­sentence.3 If we now try to analyze (2) along the lines of (3), we need to assume that (2) – like (3) – is a symmetrically coordinated V1-structure sharing a common prefield filled with da (‘there’). This time, however, the subject of the first sentence does not ­c-command the base position SL of the ‘dropped’ subject in the second conjunct, which means that we cannot construe the subject gap SL in (2) as a trace that is bound by the subject of the first sentence. In other words, given phrasal C’-coordination, the subject gap in (2), in contrast to the one in (3), cannot be taken to be the result of ‘simultaneously’ ­moving the subject jemand out of both conjuncts – a movement process which is known as “across the board (ATB) movement” (see Williams 1978). There is one alternative, though, that we have to take into consideration here. If we suppose that examples like (3) are not the result of C’-coordination followed by ATB-movement, but that they are the result of CP-coordination followed by deletion of the subject, then we might still put forward a uniform analysis in terms of ellipsis (see, e.g., Wilder 1997): Both structures could be taken to display instances of so-called “forward deletion,” an ellipsis process deleting strings at the left periphery of noninitial conjuncts under identity. In the case of (2) and (3) this amounts to saying that (i) the subjects in non-initial conjuncts are to be located in the prefield, and (ii) that they are deleted in this position under identity with the subject in the first conjunct; see (4) and (5).

(4) [[da stellt sich jemand vor die Mikrofone] und [jemand tut …]]



(5) [[jemand stellt sich da vor die Mikrofone] und [jemand tut …]]

There is good evidence, however, that the lack of the subject in (1) cannot be due to “forward deletion” or some similar kind of deletion process. In fact, it is all but clear whether such a deletion process exists at all (see, e.g., Höhle 1991; Hartmann 2002). If it does, it certainly does not behave in crucial respects the way clear cases of ellipsis in coordinate structures do. To see this, contrast the interpretation of elided indefinites in Gapping – see (6) below – with the interpretation of elided indefinites in structures like (4) and (5) above. Whereas in the case of Gapping the interpretation of the indefinite etwas (‘something’) remains constant irrespective of whether etwas is realized overtly or covertly – in both cases, etwas in the second conjunct refers (in its preferred reading) to a different object than etwas in the first conjunct – this is not true of (4) and (5) above: In (4) and (5), it is claimed that there is one person who both stands in front of the microphone and who pretends to be able to explain something, i.e., the elided indefinite jemand is .  This is, of course, one of the reasons why in subject-initial languages like, for example, English SLFC seems not to exist: the only syntactic structures in English that comply with property (iii) are inverted structures as used in English interrogatives.

 Ingo Reich

systematically interpreted as being ‘coreferent’ with its antecedent. Their overt counterparts in (7) and (8), however, are far more liberal: they may be, and in fact are taken as attributing the properties of standing in front of the microphone and of pretending to be able to explain something to two different persons. (6) Peter schenkte Anna etwas und Anna schenkte Peter etwas Peter gave Anna s’thing and Anna gave Peter s’thing ‘Peter gave Anna something, and Anna Peter’

(7) da stellt sich jemand vor die Mikrofone und jemand tut so, …



(8) jemand stellt sich da vor die Mikrofone und jemand tut so, …

Besides, “forward deletion” is subject to another important formal restriction not mentioned yet: It is not only the target of deletion that has to be positioned at the left periphery of its conjunct, but also its antecedent (see, e.g., van Oirsouw 1993). This, however, immediately contradicts defining property (iii) of SLFC; and somewhat loosening that restriction would obviously lead to considerable overgeneration. Now, since the lack of the subject in non-initial conjuncts in SLFC seems to be due neither to ellipsis nor to ATB-movement, this property of SLFC remains – for the time being – quite puzzling.

1.2  Asymmetric Coordination (AC) As Höhle (1990) observes, SLFC seems not to be restricted in its use to root clause phenomena like (1) above, but also occurs in dependent contexts as, for example, in the antecedent of a conditional; see (9). (9)

Wenn du nach Hause kommst und siehst den Gerichtsvollzieher vor if you back home come and see the bailiff at der Tür, dann […] the door, then […] ‘If you come home, and you see the bailiff at the door, then […]’

Since SLFC requires fronting of the finite verb in non-initial conjuncts, we encounter, however, an unexpected word order asymmetry: While the finite verb is fronted in the second conjunct (overt V1-position), it is – because of the complementizer wenn (‘if ’) – in final position (VE) in the first conjunct. This kind of word order asymmetry is in fact more general and is also found in coordinations that do not lack a subject in the second conjunct; see (10). (10)

Wenn du nach Hause kommst und der Gerichtsvollzieher steht if you back home come and the bailiff stands vor der Tür, dann […] at the door, then […] ‘If you come home, and the bailiff is standing at the door, then […]’

Odd coordinations 

Coordinations that show this kind of word order asymmetry have been dubbed ­“Asymmetric Coordination (AC)” – in a narrower sense – in Höhle (1990). As it turns out, SLFC and AC share many crucial properties (see Reich 2007a, for a detailed discussion), the two most prominent probably being (i) verb fronting, see (11a), and (ii) a fusing interpretation, see (11b). (11) a. b.

Verb fronting In non-initial conjuncts the finite verb is fronted. Fusing semantics SLFC and AC are subject to a ‘one-event’ interpretation.

That AC as well as SLFC do in fact require fronting of the finite predicate in non-initial conjuncts is easily illustrated by the following pair of examples: (12) a. *Kommst du nach Hause und der GV vor der Tür steht, […] b. *Da stellt sich jemand vor die Mikrofone und so tut, […]

The fact that AC and SLFC are subject to some kind of one-event interpretation is somewhat harder to illustrate. Höhle (1983: 22) gives the contrast in (13). (13) a. b.

Stehen da wieder welche rum und verteilen Flugblätter? stand there again folks around and hand out flyers? ‘Are, again, folks standing around and handing out flyers?’ Stehen da wieder welche rum und verteilen sie FB? stand there again folks around and hand out they flyers? ‘Are, again, folks standing around and are they handing out flyers?’

Whereas the SLFC in (13a) is conceived of as a single question focusing on the distribution of flyers, the symmetric coordination in (13b) is understood as two basically independent questions asking about two different things. It is tempting here to blame the overt subject in (13b), for it seems that VPcoordination also shows some kind of one-event interpretation; see the pair of examples in (14), modeled after an example taken from Reis (1993). (14) a. b.

Jetzt geht er in Buchhandlungen und liest Neuerscheinungen. now goes he in bookstores and reads new publications ‘Now, he goes in bookstores and reads new publications’ Er geht jetzt in Buchhandlungen und liest Neuerscheinungen. he goes now in bookstores and reads new publications ‘Now, he goes in bookstores and reads new publications’

There is one difference though, and this difference is crucial: In the case of SLFC the one-event interpretation is obligatory, whereas it is only optional in the case of VP-coordination. This is – for reasons related to sentence mood – best illustrated with SLFC in a conditional clause. Consider (15).

 Ingo Reich

(15) a. b.

*wenn er Neuerscheinungen if he new publications wenn er Neuerscheinungen if he new publications

liest reads liest reads

und and und and

geht in Buchhandlungen goes in bookstores in Buchhandlungen geht in bookstores goes

In (15), the conjuncts are switched, which has the effect of blocking a one-event ­interpretation (since it is a natural course of events to go in a bookstore and read a new publication, but not vice versa). This change in interpretation does not affect symmetric VP-coordination in (15b), but it does affect asymmetric SLFC in (15a), which is clearly marginal, if not ungrammatical. This shows several things: (i) The one-event interpretation associated with SLFC and AC cannot simply be reduced to whatever mechanism is responsible for one-event ­interpretations in the case of symmetric VP-coordination. (This seems obvious with AC.) (ii) Since the one-event interpretation in AC and SLFC is obligatory, it is most likely a semantic (and not a pragmatic) property. (iii) Since the other characteristic property of AC and SLFC – fronting of the finite verb in non-initial conjuncts – is obligatory, too, it seems quite obvious that both properties are in fact two sides of the same coin. If this is correct, we have to answer the question of how to characterize the fusing semantics of AC and SLFC, and how this semantics is linked to verb fronting (in AC and SLFC). This is the topic of the following section.

2.  From discourse to “odd coordinations” But where to start when trying to characterize the fusing semantics of AC and SLFC? The fact that AC and SLFC frequently occur with conjunctional adverbs like dann (‘then’), dabei (‘thereby’), etc., which explicitly refer to coherence relations connecting utterances in discourse, suggests that a somewhat closer look at possible coherence relations may be promising here.

2.1  Discourse structure and the “Occasion” relation A more recent classification of coherence relations is found in Kehler (2002).4 Kehler (2002), building on work done by David Hume and Jerry Hobbs, subdivides coherence relations into three different classes; see table 1. Table 1.  Classification of coherence relations in Kehler (2002) Resemblance

Cause-Effect

Contiguity

Parallel Contrast …

Result Explanation …

Occasion

.  But see Kehler (2002) and this volume for (a discussion of) alternative classifications.

Odd coordinations 

Whereas resemblance relations and cause-effect relations are construed as relations between propositions – structured and unstructured, respectively – contiguity ­relations are to be conceived of according to Kehler (2002) as relating events. The class of contiguity relations is in fact a singleton class consisting only of the coherence ­relation occasion, which is informally characterized as follows: “Occasion can be seen as a mechanism for communicating a complex situation in a multi-utterance discourse by using states of affairs as points of connections between partial descriptions of that situation” (Kehler 2002: 22). This seems to be exactly what we are looking for. So let’s have a closer look at the formal definition. Kehler (2002: 22) gives the following characterization of the occasion relation (see also Hobbs 1990): Occasion (i): Infer a change of state for a system of entities from S1, inferring the final state for this system from S2. Occasion (ii): Infer a change of state for a system of entities from S2, inferring the initial state for this system from S1.

Kehler (2002: 23), however, considers this to be only a first approximation to the ­intuition underlying occasion, since “temporal progression in the absence of a common scenario connecting the events is insufficient in and of itself.” In this respect, I completely agree with Kehler (2002). What we do not agree upon is his classification of coherence relations. The fact that (i) the class of contiguity ­relations is a singleton, and (ii) the occasion relation relates events rather than ­propositions ­suggests that occasion is in fact not a coherence relation in the strict sense, but a discourse phenomenon operating on a different level, interacting with genuine coherence relations. If this is correct, we end up with a much simpler and homogeneous classification of coherence relations; see table 2. Table 2.  Coherence relations as relations between (un)structured propositions Structured Coherence

Unstructured Coherence

Parallel Contrast …

Result Explanation …

This, of course, does not discharge us of specifying exactly what we mean when talking about “occasion”.5 To get a better idea of what “occasion” is all about, consider the following sequence of utterances: (16) a. At 6:30 a.m., the lights are turned on in house No. 3. b. Then, at 6:32 a.m., the lights are turned on in house No. 10. c. Then, at 7:01 a.m., the sun comes up.

.  In the following, the typographical difference between occasion and “occasion” is meant to reflect the aforementioned different theoretical perspectives on the occasion relation.

 Ingo Reich

There are two ways to interpret the sequence of utterances in (16): (i) simply as a sequence of events e1, e2, e3, connected by temporal progression, or (ii) as a complex event e – say the ‘awakening’ of (for example) Downing Street in London – consisting of the subevents e1, e2, e3 as perceived of by some person observing the street, and what happens there. What we are interested in is, of course, scenario (ii). Another way to describe what’s going on in (16) under scenario (ii) is as follows: The first sentence (16a) sets the scene by describing part of a possibly complex event e; the second sentence (16b), speaking about another event e2, in a sense ‘extends’ the description of e by subordinating e2 to e. This process of ‘event subordination’ is then reiterated with (16c), resulting in a complex event e consisting of (at least) the events e1, e2, and e3.6 We thus can define “occasion” as some kind of event subordination as follows: (17) Occasion (event subordination) Occasion is to be construed as event subordination, i.e., for any two events e1, e2, Occasion(e1, e2) holds iff it is the case that e2 ≤ e1.

Schematically, the situation in (16) can be depicted as follows: (18)

e S1 (conj) S2 (conj) S3 … p1 co-rel p2 co-rel p3 … e ≥ e2 ≥ e3 …

(discourse syntax) (coherence relations) (event subordination)

The basic idea is, thus, clear: Sentence S1 introduces an event variable e. If the discourse stops here, e denotes the minimal event described by the sentence S1, i.e., the event e1. If, however, an utterance S2 is added, and S2 is meant to continue the description of a (complex) event introduced by S1, then the minimal event e2 referred to by S2 is taken to be a subevent of e – which is guaranteed by the condition e2 ≤ e on e2 and e. This is, I think, exactly what “occasion” does: While “occasion” ensures that S1 and S2 build a complex structure in the first place, coherence relations like contrast, result, etc. specify the way the propositions denoted by S1 and S2 relate to each other. As an alternative to event subordination, one could think of (recursively) merging two events e1 and e2, i.e., we could define a merging operation ‘⊕’ which gives us for e­ very

.  The term ‘event subordination’ is reminiscent of the term ‘modal subordination’, which is introduced in Roberts (1989). In fact, event subordination and modal subordination do share some important formal characteristics. However, as the following contrast between the modal subordination example in diesem Haus lebt kein Mann oder es ist gut versteckt (‘There is no man living in this house or is well hiddden’) and the SLFC construction *in diesem Haus lebt kein SL ist gut versteckt shows, modal suboradination and event subordination are in principle two independent phenomena; see Reich (2007a) for further discussion.

Odd coordinations 

e1 and e2 a complex event e = e1 ⊕ e2.7 The problem with such an approach, ­however, is that it is all but clear how to define ‘⊕’: What exactly does it mean to merge two events? And given two arbitrary events e1 and e2, is there always a complex event e which can be considered to be the result of merging e1 and e2? Event subordination avoids such problems in a transparent way: First of all, the subevent relation ‘≤’ is (one of) the most basic relation(s) in any reasonable event mereology; see, e.g., Schein (1993). Second, the subevent relation ‘≤’ can be defined by spatial-temporal inclusion, without losing any of the events’ internal structure: This is simply because the event variable e always directly refers to and is conceptualized as an immediately accessible event e, ­without having to build e from other events e1 and e2.

2.2  Discourse structure and verb order in German So much for the specification of the Occasion relation in discourse. But how does this relate to AC and SLFC? To see this, consider examples (19) and (20). It is common knowledge that the prototypical verb order of an independently used clause in German is V2 or V1, see (19a–b), whereas the prototypical verb order of dependently used clauses is VE, see (20). (19) a. b.

Lucky Luke sprang Lucky Luke jumped Sprang Lucky Luke jumped Lucky Luke

auf onto auf onto

sein Pferd his horse sein Pferd? his horse?

(20) Rantanplan wußte, dass L.L. gleich auf sein Pferd springt. Rantanplan knew, that L.L. part onto his horse jumps ‘Rantanplan knew that L.L. was about to jump onto his horse’

This, of course, is not to say that it is absolutely impossible to use VE-clauses in ­German independently – as (21) shows this is quite possible: (21) Dass Lucky Luke ständig auf sein Pferd springt! that Lucky Luke always onto his horse jumps!

There are important restrictions on the use of such clauses though: First of all, they never carry the force of an assertion, irrespective of whether they are to be classified as declarative clauses or not. Second, they are so to speak ‘islands in discourse’, i.e., they do not directly relate to previous utterances in the usual sense. This suggests that there is an intimate connection between establishing “occasion” (and “occasion”-related ­coherence relations) on the one hand and word order on the other: Fronting the finite predicate

.  A merging operation along these lines has been proposed, for example, in Eckhardt (1998).

 Ingo Reich

seems to be a necessary – though not necessarily sufficient (see, for example, the case of embedded V2-declaratives in German) – condition on establishing “occasion”. From this point of view, the fact (i) that both AC and SLFC always show fronting of the finite predicate in non-initial conjuncts, and (ii) that this behavior systematically correlates with the presence of some kind of fusing semantics does not look like a coincidence. The question, then, is in what way these processes – one syntactic, the other ‘semantic’ – relate to each other. Though I’m not (yet) in a position to give a general answer to this question, I do have a concrete proposal concerning the special case of SLFC and AC.

2.3  The functional projection “OccP” The crucial idea of the analysis to follow is that in AC and SLFC the two conjuncts are in fact systematically linked by the “occasion” relation, which we know well by now from complex structures in discourse. Whereas in discourse, however, the linking of two utterances with the help of “occasion” is optional, and thus most likely pragmatically mediated, this is not the case with AC and SLFC. As we saw above, the fusing semantics of AC and SLFC – now construed as event subordination triggered by ­“occasion” – is an essential characteristic of these constructions. Therefore, establishing “occasion” in AC and SLFC seems to be a semantic rather than a pragmatic process.8 But if it is semantic, there needs to be some syntactic expression or feature that introduces the semantics of “occasion” into the structure of AC and SLFC. Let’s call this feature for lack of a better term [occ], which is obviously short for “occasion”. In the following, I will assume that this very feature ­projects – along with the feature [f] introduced below – a functional projection OccP (more precisely, OccP/FP). In this sense, “properties typically attributed to coordinate structures in discourse show up in AC and SLFC in ‘grammaticalized’ form.” Syntax. Before laying out the details of “OccP,” however, something needs to be said about the specifics of verb movement and sentence structure in German. In the following, I will assume – following recent work in syntax – that verb movement in German is triggered by a (strong) syntactic feature [f] that needs to be checked by a finite verb in overt syntax. It is important to see here that this feature is neither directly connected to force nor to sentence mood, for there are also – as mentioned above – dependent uses of V1- and V2-sentences in German. I, therefore, take it that [f] is uninterpretable.9 I furthermore assume that [f] is the head of a functional projection FP, which distinguishes V1- and V2-sentences from VE-sentences (which are headed by the complementizer projection CP only). A simple sentence like Jan verpetzt Maria (“Jan tells on Mary”) is thus analyzed along the lines of (22).

.  There is no need, however, to assume that coherence relations are established semantically. .  But see Truckenbrodt (2006) for a possible interpretation of V2 in German.

Odd coordinations 

(22) a. Jan verpetzt Maria. (“Jan tells on Maria”) b. FP DPi

Jan

F’

F

verpetzt

vP

ti

v’

v

VP

DP

V

Maria

tv

The syntactic structure in (22b) furthermore illustrates that I essentially follow Larson (1988) and others in locating subjects within a functional projection vP, i.e., external to VP. I cannot enter into a discussion on the VP-internal vs. VP-external hypothesis here for obvious reasons; in what follows, however, it will become apparent why this seems to be the most adequate way of treating subjects – at least from the perspective of AC and SLFC. After this brief excursion on sentence structure in German, let’s come back to the analysis of AC and SLFC. As I have already mentioned several times, one of the crucial characteristics of AC and SLFC is the V1/V2-property of non-initial conjuncts. But what exactly is it that triggers fronting of the finite verb in these cases? As we argued above, this property seems to be intimately connected to the process of establishing “occasion” between two sentences. In other words, it is the feature [occ], denoting the “occasion” relation, which is to be made responsible for this behavior. In terms of a feature-based syntax, this is – given our assumptions about verb movement – ­essentially equivalent to saying that the feature [occ] selects for the feature [f]; that is, we have (23). (23) Verb fronting in AC and SLFC [occ] selects for [f].

If the feature [occ] projects a functional projection, and if it furthermore selects for the feature [f], there are in principle two syntactic configurations where we can expect [occ] to occur: Either it projects a functional projection of its own, thereby

 Ingo Reich

taking FP as its syntactic complement, see (24a), or [occ] attaches to the head F, and they project a functional projection OccP/FP, see (24b). (24) a. [OccP Spec [Occ’ Occ [FP Spec [F’ F [ … b. [OccP/FP Spec [Occ’/F’ Occ/F [ …

As far as I can see, there is no compelling independent evidence for the more elaborate structure in (24a). I, therefore, assume that non-initial conjuncts in AC are assigned the syntactic structure in (24b). If we now reconsider example (10) from section 1.2 above, repeated here as (25) for convenience, (25)

Wenn du nach Hause kommst und der Gerichtsvollzieher steht if you back home come and the bailiff stands vor der Tür, dann […] at the door, then […] ‘If you come home, and the bailiff is standing at the door, then […]’

we end up with a syntactic analysis along the lines of (26). (26) CP

CP

OccP/FP

wenn…kommst und

OccP/FP

Occ’/F’

DPi

der GV

Occ/F

vP

steht

ti vor der Tür tv

Two comments seem to be in order here. First, whatever the correct analysis of ‘standard symmetric coordination’ is (see, e.g., Progovac 1998; Reich 2007a), the use of und (‘and’) in AC and SLFC seems to be somewhat different. This is apparent from the fact that ‘standard (symmetric) coordination’ allows for ATB-movement, while AC and SLFC do not (Höhle 1983). I, therefore, follow Büring and Hartmann (1998) in analyzing coordination as adjunction in the case of SLFC (and AC). As we will see in a moment, this assumption is in fact an immediate consequence of the semantics of [occ].

Odd coordinations 

The other comment concerns the adjunction site of OccP/FP. In the case of ­ ependent uses of AC and SLFC as in (25/26), OccP/FP seems to (have a strong d ­tendency to) adjoin sentence-externally, i.e., to adjoin to the whole CP. Even though this is somewhat unexpected from a semantic point of view, there is good evidence from the binding of pronouns (quantifiers in the first conjunct do not easily bind ­pronouns in the second conjunct) and scope of negation (the scope of negation is restricted to the conjunct it is part of) that this is in fact correct; see Reich (2007a,b). Here, I just want to touch on this matter by illustrating the behavior of negation in this respect. Consider (27). (27)

#Wenn du nicht die Kronjuwelen klaust und (du) wirst dabei if you not the crown jewels steal and (you) get thereby erwischt, dann hast du Glück gehabt caught, then have you luck had

From a pragmatic point of view, the only reading that makes sense here is the one with wide scope of negation: If it is not the case [that you steal the crown jewels and you get caught red-handed], then you are lucky. But as a matter of fact, this reading is not available in (27). The only reading available is the self-contradictory one with scope of negation restricted to the first conjunct: If [it is not the case that you steal the crown jewels] and you get caught red-handed, then you are lucky. This is consistent with the assumption that the second conjunct attaches to CP. Because of space limitations, I cannot go into details here, but the reader is referred to Reich (2007a,b) for further discussion, including possible implications for the semantics of conditionals. Semantics. Having introduced the main syntactic properties of [occ], let’s have a closer look at its semantics. We already argued in some detail that [occ] denotes the “occasion” relation, i.e., that it implements event subordination. If we want to spell out this idea in a formal setup, we need to assume – in accordance with the literature on lexical semantics and ‘Aktionsarten’ in the tradition of Donald Davidson – that each verb introduces an event variable e, i.e., that sentences are taken to denote sets of (possible) events. Given these assumptions, we can specify the semantics of [occ] as given in (28). (28) Semantics of [occ] ||[occ]|| = λQλ¡λPλe∃e2[ ¡(Q(e2))(P(e)) & e2 ≤ e ]

According to (28), the feature [occ] implements two conditions: (i) The second c­ onjunct Q is conjoined with the first conjunct P by some appropriate coordinating conjunction ¡; this relates to the condition ¡(Q(e2))(P(e)), which can be rewritten as “P(e) & Q(e2)”, if ¡ is instantiated by and. (ii) The feature [occ] furthermore states that the event e2 corresponding to the second conjunct Q is a subevent of the event e introduced by the first conjunct P. While condition (i) prevents the coordinating conjunction ¡ from projecting its own functional projection (otherwise we run into a type mismatch), and thus is to be made responsible for the adjunction structure of AC and SLFC, condition (ii) implements the by now familiar event subordination “e2 ≤ e”.

 Ingo Reich

If we take (essentially following Kratzer 1991) the complementizer wenn (‘if ’) to be semantically empty, it is straightforward to see that the proposed semantics derives the interpretation given in (29) for the AC in (26), where CBH stands for the complex predicate “come back home,” and SAD is shorthand for the other complex predicate “stands at the door.” (29) λe∃e2[ [CBH(e, you) & (SAD(e2, bailiff )] & e2 ≤ e ]

According to (29), the AC in (26) denotes that set of (complex) events e which are ­directly characterized as ‘you are coming back home’ events – because of the condition CBH(e, you) – and indirectly as being at the same time also ‘the bailiff is standing at the door’ events – because of the condition SAD(e2, bailiff ) mediated by event subordination e2 ≤ e. This seems to be a good and reasonable approximation to the fusing semantics of AC and SLFC. The ‘Vorfeld’ of non-initial conjuncts in AC. Before switching over to SLFC let me point out an interesting prediction of this approach that concerns overt realizations of the prefield of non-initial conjuncts in AC. Following Frey (2004), I assume that the German prefield can be filled in three different ways: (i) by base generation (of some specific elements, e.g., the prefield-es), (ii) by formal movement (FM), and (iii) by real A-bar-movement. Whereas real A-bar-movement is semantically ­motivated by some interpretable feature in the left periphery ([topic], [contrast], [wh], etc.) and may, in principle, start from any position within the middle field, FM is not triggered semantically (it is just ‘formal’), and it is restricted to short movement of the leftmost expression in the middle field (which is quite often, but not always the subject). Now, what about the prefield of non-initial conjuncts in AC? As we saw above, the prefield is, in this case, not simply the specifier position of some FP, but the specifier position of an OccP/FP; and since the feature [occ] is interpretable, we ­expect movement to [Spec,OccP/FP] to be restricted as follows: While base generation and FM to [Spec,OccP/FP] should be fine (see, e.g., (30)), semantically motivated movement should be out if (and only if) the semantics of the triggering feature is incompatible with the semantics of [occ]. This definitely excludes wh-movement to this position; but it also prevents real A-bar-movement of a (lower ranked) object, which – according to Frey (2004) – always involves some kind of contrastive interpretation; see (31). (30) Wenn du nach Hause kommst und dir ist schlecht if you back home come and you.dat feel sick (31)

*Wenn if spricht speaks

du you der the

nach Hause kommst und mit deiner Frau back home come and with your wife Gerichtsvollzieher, dann … bailiff, then …

Odd coordinations 

There is far more to say about this intricate topic; alas, I have to refer the reader once more to Reich (2007a) for a more detailed discussion.

2.4  Subject gaps in coordinate structures As I argued in section 1, it seems safe to assume that AC and SLFC are – apart from some minor differences – essentially one and the same phenomenon. Or to put it another way: SLFC is basically AC with a subject gap. If this is correct, we have to answer at least two questions: (i) Why is it that in AC subjects (and only subjects) can be dropped, resulting in an SLFC? (ii) What is the nature of the subject gap – is it just an instance of some well-known kind of ellipsis or is it due to something special? As we saw in the introduction, there seems to be no (straightforward) way to ­derive the gap from ATB-movement or forward deletion. Moreover, the fact that subject gaps (as the name suggests) are restricted to subjects – a restriction which is, to say the least, not very widespread in ellipsis phenomena – suggests that the second option, i.e., that the subject gap is something special, is true. The fact that subject gaps in coordinate structures are limited to AC further suggests that they are closely tied to some idiosyncratic property shared by these constructions. The hypothesis I want to pursue in the following is that this very property is the ‘fusing semantics’ of AC and SLFC, i.e., it is event subordination, implemented with the help of [occ], that licenses dropping the subject. The crucial idea of the approach to follow is thus that subject gaps are possible in AC (and only in AC), since event subordination enables us to ­semantically reconstruct the subject of the second conjunct from the subject of the first conjunct. This will follow from the ‘subject condition’. Syntax and semantics of vP. Before I am in a position to precisely state the ‘subject condition’, however, I have to introduce some more assumptions about the syntax and semantics of sentence structure in German. In the previous section, I made the assumption that subjects – understood as nominative DPs that constitute the highest thematic argument – are to be located VP-externally in a functional projection called vP. Since there are verbs in German that do not allow for subjects in this strict sense (see, e.g., mir[dat] ist komisch zumute ‘I’m in a funny mood’), this means that vP does not project independently, but is triggered by a (non-interpretable) feature [v] on the relevant verbs. In this section, I propose – borrowing ideas from Kratzer (1996) – that the externality of subjects is not restricted to syntax, but extends to semantics. Let me illustrate this with the following example. In (32a) the subject der Gerichtsvollzieher is severed from the verb’s semantic representation, and the VP vor der Tür steht (‘stands at the door’) already denotes an (almost) saturated semantic object of the type ‘set of events,’ namely the set of events that are ‘stands at the door’-events. The head [v] itself is ­semantically vacuous, see (32b), but it carries (by assumption) a feature [subj] which is interpretable, and which

 Ingo Reich

states that the specifier of vP (i.e., x) is a subject (to e), see (32c);10 this statement is then conjoined with the denotation of its complement, see (32d).11 (32) a.

vP

DP

der GV m

v’

v [subj] ||[subj]||

VP

vor der Tür steht λe.stands-at-the-door(e)

b. ||v|| = λQ.Q c. ||[subj]|| = λQλxλe[subj](e)(x) & Q(e)] d. ||(32a)|| = λe[subj(e)(m) & stands-at-the-door(e)]

The vP as a whole thus denotes the set of events e such that e is a ‘stands at the door’event, and the bailiff is the subject in e (i.e., he realizes the highest thematic role in the ‘stands at the door’-event e). This seems correct. Two notes: First, as a consequence of severing the external argument from its verb, both VP and vP are now of the same semantic type, i.e., they both denote sets of events. Second, weather verbs like regnen (‘rain’) do not select for a thematic argument, but for an expletive subject es (‘it’). In such cases [v] does not carry the feature [subj], i.e., the vP contributes nothing to semantic interpretation; [v] just assigns nominative case to the expletive es. The Subject Condition. Let’s consider now an AC with an overt subject in its second conjunct which is coreferential with the subject in the first conjunct; see (33). Its semantic representation is given in (34). (33)

Wenn du nach Hause kommst und du siehst den if you back home come and you see the Gerichtsvollzieher vor der Tür, dann […] bailiff at the door, then […]

.  The property of being a subject relative to a given event e needs to be independently ­defined, of course. Roughly speaking, x is a subject to an event e – i.e., ||subj(e)(x)|| = 1 – iff x refers to the highest thematic role participating in the event e; see Reich (2007a) for details. .  In a step by step fashion the interpretation of vP proceeds as follows: First, the denotation of VP (λe.stands-at-the-door(e)) functions as an argument to the interpretation λQλxλe[subj(e) (x) & Q(e)] of the feature [subj]; this results in λxλe[subj(e)(x) & stands-at-the-door(e)] as the denotation for the node v’. The denotation of v’ then takes the subject m in [Spec,vP] as its argument, and we end up with the denotation λe[subj(e)(m) & stands-at-the-door(e)].

Odd coordinations 

(34) λe∃e2[subj(e)(you) & come-back-home(e) & subj(e2)(you) & see-at-the-door(e2)(bailiff ) & e2 ≤ e ]

The semantic representation in (34), of course, contains two characterizations of the subject you, namely ‘subj(e)(you)’ and ‘subj(e2)(you)’. Since we know that we can drop you in the second conjunct, what we’d like to be able to say is that the second characterization is in some sense redundant – for if it were, then the subject in the second conjunct could be semantically reconstructed on the basis of the subject in the first conjunct, and the gap would be licensed. However, since the event variables e and e2 do not necessarily refer to the same event, ‘subj(e)(you)’ does not entail ‘subj(e2)(you)’ nor vice versa. But wait: Since one of the crucial characteristics of AC is that AC systematically triggers event subordination ‘e2 ≤ e’, it could very well be that this property of AC is the right piece of information that bridges the gap and licenses the entailment; that is, the question to ask is the ­following: Is it possible to conclude from the fact that you is the subject of e and the fact that e2 is a subevent of e that you is also the subject of e2? Again, the answer is “no”. If the answer were “yes”, then it would be impossible to have an AC with an overt, non-coreferent subject in the second conjunct. But, as (25) shows, this is possible. So it seems that we’re stuck. But in fact we are not. What I just showed is that the conditions ‘subj(e)(you)’ and ‘e2 ≤ e’ do not strictly [!] entail – in the sense of a ­material conditional – the condition ‘subj(e2)(you)’. It is well known, however, that entailments in natural language typically are not strict, but non-monotonic (see, e.g., Asher & Lascarides 2003). To see this, consider (35). (35) a. If John comes, the party will be a success. b. If John and Mary come, the party will be a success.

If entailments in natural language adhered to the material conditional, it should be possible to conclude from the truth of (35a) the truth of (35b), since it follows from ‘p1 → q’ that ‘p1 & p2 → q’ for any proposition p2. But in fact we cannot strictly conclude from (35a) that (35b) is true as well – it may very well be that John behaves completely different in the presence of Mary. So let’s assume that the entailment we are after is in fact a non-monotonic entailment (symbolic: ‘»’), i.e., that the ‘subject condition’ in (36) holds. (36) Subject condition For any two events e, e2 and any individual x the following holds: subj(e)(x) & e2 ≤ e » subj(e2)(x)

What (36) tells us is that – in the absence of any other information – we can conclude from the fact that x is the subject of e, and that e2 is a subevent of e, that x is also the subject of e2. If we assume the validity of (36), then a dropped subject can be semantically reconstructed from its antecedent – and this seems to be (almost) all that is needed for licensing a subject gap in the second conjunct of an AC. Note, by the way, that (36) does not license subject gaps in initial conjuncts, for event subordination is an asymmetric relation.

 Ingo Reich

That the assumption of (36) is in fact consistent is easily shown, for there are only three cases to consider. Either the subject of the second conjunct is realized, or it isn’t. If it is, then it is either coreferent, or it is not. Suppose we have a non-coreferent overt subject, then the conditions ‘subj(e2)(y)’ and ‘x ≠ y’ are added to the antecedent of (36); in this scenario the conclusion is blocked, since the added conditions in the antecedent contradict the conclusion. Suppose, alternatively, that we have an overt, coreferent subject, then the condition ‘subj(e2)(x)’ is added, and the entailment is trivialized. The last scenario is the one with a covert coreferent subject, i.e., the ­scenario stated in (36). Subject gaps and properties of vP. That there are in fact only three scenarios to consider follows from the syntax of subject gaps: What I do have to assume – and what I will, henceforth, assume – is that to drop the subject in an AC means to drop the feature [subj]; see (37). (37) Subject gaps (Part I) In an SLFC the head [v] of the second conjunct lacks [subj].

As we saw above, dropping the feature [subj] is equivalent to saying that no condition of the form ‘subj(e2)(x)’ is introduced in the relevant conjunct, i.e., from the point of view of syntax no thematic subject is realized. In this respect, the situation in SLFC seems to be – at least to some extent – comparable to the one with weather verbs like regnen (‘rain’). There is one important difference, though: Whereas in the case of weather verbs there simply is no highest thematic role in the event structure of e (the ‘raining event’) that could be realized in syntax, the event structure corresponding to the relevant SLFC conjunct does of course include a highest thematic role, which is in need of syntactic realization; nevertheless, the syntactic realization of this argument can be suppressed in SLFC, for we are – thanks to the ‘subject condition’ – able to link the not realized thematic role of the second conjunct to the subject of the first one. In this sense, a coreferent subject in the second conjunct of an AC is redundant, and its dropping is ‘parasitic’ on the semantics of AC. What we have not yet talked about is the nature of the gap itself. As it turns out, the nature of the gap is already completely determined by the assumptions we made above. It is clear from what has been said that the subject gap is to be located in the specifier position [Spec,vP] of vP. This position, however, cannot be completely empty, since the task of the head [v] is to assign nominative to some element in [Spec,vP], and there is no good reason to assume that this is somehow suppressed in SLFC. Since the head [v] is not accompanied by the feature [subj], the element in [Spec,vP] needs to be some kind of expletive (if it were referential, it would have to be conjoined with the verbs’ interpretation, but there is no way to conjoin an expression of type e with an expression of type t without the mediating semantics of the feature [subj]). The only expression fulfilling all these criteria we know of seems to be the expletive ­subject es. So, is the subject gap a covert variant of es? Most certainly not, since es carries in addition to the feature nominative the features 3rd person singular; the finite verb in the second conjunct of an SLFC, however, is not restricted to 3rd person singular, but seems to agree in person and number with the subject in

Odd coordinations 

the first conjunct.12 It follows thus that the subject gap is unspecified with respect to person and number. To sum up, what we know about the nature of the gap is the following: It is covert, it is an expletive, and it is unspecified with respect to person and number, but it is specified with respect to case – it carries the case nominative. The only ­expression I know of that has exactly these properties is the feature [nom] itself. So let’s assume the following restriction: (38) Subject gaps (Part II) In an SLFC [Spec,vP] of the second conjunct is filled by [nom].

In fact, (38) is exactly what we expect in a feature-driven syntax, where each feature [x] has to be matched with its counterpart to avoid a crashing derivation: The only task of the head [v] is to match its feature [nom] with a feature [nom] in [Spec,vP] (i.e., it ‘assigns nominative’ to [Spec,vP]); since there are no other processes going on within vP in the case of SLFC, there is no need (and it is even impossible) to realize a different feature than [nom] in [Spec,vP]. Subject gaps and the V1-property of non-initial conjuncts. There is one last q ­ uestion we need to answer: Why is it that dropping the subject in an AC systematically ­results in a V1-structure (rather than a ‘gapped’ V2-structure)? Or to put it differently: Why does a subject gap block movement to and base generation in the prefield of a V ­ 2-clause? First note that the subject gap [nom] is a covert expression which has to be – ­according to, e.g., Lobeck (1995) – (i) licensed, and (ii) identified. There are two configurations in which covert expressions are licensed: either in a complement relation, or in a spec-head configuration. Since the subject gap is not in a complement, but a specifier position, licensing needs to take place in a spec-head configuration. The phenomenon of ‘Topic Drop’ in German tells us that licensing covert expressions in such a configuration is in fact possible in German. Essentially following Fries (1988), the syntax of Topic Drop can be sketched as follows: Dropping the topic is (i) licensed by the matching feature [topic] in the head [f], and it is identified by the finite verb in [f]; see (39). FP

(39) DP1 du you

F’ F[TOPIC]

vP

hast are

t1 ja recht t1  right

.  This is, of course, a tough nut to crack in approaches like the one proposed in this paper. There are, as far as I can see, at least two promising ways to tackle the problem: Either there is some syntactic process like ‘long distance agreement’ available (see, e.g., Sternefeld 2006b, chap. V.3) or person and number agreement is semantically mediated (see, e.g., Reich 2007a).

 Ingo Reich

Now consider Wunderlich’s (1988) example (40) of a typical root clause SLFC. (40) In den Wald ging der Jäger und schoss einen Hasen in the forest went the hunter and shot a rabbit

What we are primarily interested in is, of course, the syntactic structure of the ­second conjunct. If the licensing of the subject gap in examples like (40) is in fact essentially structurally parallel to the licensing of Topic Drop, then there needs to be some ­interpretable feature corresponding to [topic] that is located in the head [v] of vP; the obvious candidate here is of course the feature [occ], the core ­characteristic of AC, which denotes event subordination, and thus feeds the licensing ‘subject ­condition’ in (36). The fact that the subject gap is identified by the fronted finite verb then follows immediately from the syntax of the feature [occ]: Since [occ] selects for the feature [f], and [occ] is to be located in the head [v] of vP (the subject gap in [Spec,vP] is licensed in a spec-head configuration), it follows that [f] is to be located in the head [v] of vP, too. In other words: Whereas an AC with an overt subject is a projection of OccP/FP (see the previous section), an SLFC is an even more ‘compressed’ projection OccP/FP/vP; see the syntactic structure given in (41), and its semantic interpretation in (42). (41)

[FP In den Wald [F’ gingi [vP der Jäger [v’ ti,[SUBJ] VP

VP

ti

OccP/FP/vP

und

OccP/FP/vP

[nom]

Occ’/F’/v’

Occ/F/v

VP

schossj

einen Hasen tj

(42) λe∃e2[subj(e)(the-hunter) & [went-in-the-forest(e) & shot(e2)(a-rabbit)] & e2 ≤ e ]

Odd coordinations 

In (41) the subject gap [nom] is (i) licensed by the interpretable feature [occ], and (ii) identified by the overt fronted finite verb schoss (‘shot’). Since the prefield of the fronted structure in (41) is identical with the specifier position of OccP/FP/vP, and this position is already (and necessarily) filled with the case feature [nom] (required by the head [v] of vP), movement to this position as well as base generation of some other element in this position is blocked. On where to attach the second conjunct of an AC. If we compare the analysis of the root SLFC in (41) with the analysis of the dependent AC in (26), we observe another, somewhat surprising difference: Whereas in (26) the second conjunct is adjoined sentence-externally, i.e., to CP, it is adjoined sentence-internally in (41). This corresponds to the empirical observation that in cases like (26) negation cannot (easily) have scope over the second conjunct of an AC, but in cases like (43), taken from Höhle (1983), it can. (43) Hoffentlich sieht er uns nicht und zeigt uns an! hopefully sees he us not and reports us part! ‘Hopefully he doesn’t see us and report us.’

A closer look at the facts reveals that the empirical generalization seems to be as ­follows: If the finite verb in the initial conjunct is fronted (V1/V2), then there is a strong tendency to attach the second conjunct sentence-internally, i.e., to VP; if it is not (VE in the first conjunct), then there is a strong tendency to attach the second conjunct sentence-externally, i.e., to CP. This is, however, not a strict rule, for there are obvious – but nevertheless systematic – exceptions to this generalization (see Reich 2007a, for further discussion).

3.  Summary Starting from semantic and pragmatic considerations, I have proposed an analysis of Asymmetric Coordination (AC) and SLF-Coordination (SLFC) that ties together their semantic and syntactic characteristics by considering them an immediate ­consequence of the fact that in AC and SLFC properties typically attributed to coordinate structures in discourse show up in ‘grammaticalized’ form. Concretely, I have suggested that in AC and SLFC both conjuncts are linked by the Occasion relation, which I have ­furthermore argued to express ‘event subordination’. In a feature-based syntax this has been implemented with the help of a syntactic feature [occ] that selects for the feature [f], which is, in turn, responsible for the fronting of the finite verb in non-initial ­conjuncts. Finally, the dropping of the subject in the case of SLFC has been argued to be in a sense ‘parasitic’ on the special ‘fusing’ semantics of SLFC: Given event subordination, a coreferent subject in the second conjunct of an AC is redundant in the sense that it can be reconstructed with the help of a non-monotonic inference – the ‘subject condition’ – from the semantics of the first conjunct. On the syntactic side, the gap has been construed as the case feature [nom], which – being a covert expression – has to be

 Ingo Reich

licensed (paralleling Topic Drop in German) by the “occasion” feature [occ] in a spec/ head-configuration and identified by the fronted finite verb. The fact that the prefield – i.e., Spec of OccP/FP/vP – is filled with the feature [nom] derives the V1-property of non-initial conjuncts in SLFC.

References Asher, Nicholas & Lascaride, Alex. 2003. Logics of conversation. Cambridge: CUP. Büring, Daniel & Hartmann, Katharina. 1998. Asymmetrische Koordination. Linguistische ­Berichte 174: 172–201. Eckhardt, Regine. 1998. Adverbs, events, and other things. Tübingen: Niemeyer. Fortmann, Christian. 2005. Die Lücken im Bild von der Subjektlücken-Konstruktion. ­Linguistische Berichte 204: 441–476. Frank, Annette. 2002. A (discourse) functional analysis of asymmetric coordination. In Proceedings of the LFG02 Conference, M. Butt & T. King (eds), 174–196. Frey, Werner. 2004. The grammar-pragmatics interface and the German prefield. Sprache & Pragmatik 52: 1–39. Fries, Norbert. 1988. Über das Nulltopik im Deutschen. Sprache & Pragmatik 3: 19–49. Hartmann, Katharina. 2002. Right Node Raising and Gapping: Interface conditions on prosodic deletion. Philadelphia, Amsterdam: John Benjamins. Hobbs, Jerry R. 1990. Literature and Cognition. Stanford: CSLI. Höhle, Tilman N. 1983. Subjektlücken in Koordinationen. Ms., Tübingen. Höhle, Tilman N. 1986. Der Begriff  ‘Mittelfeld’. Anmerkungen über die Theorie der ­topologischen Felder. In Textlinguistik contra Stilistik. Akten des VII. Internationalen Germanisten-Kongresses Göttingen 1985, Bd. 3, W.E. Weiss, H.E. Wiegand & M. Reis (eds), 329–340. Tübingen: Niemeyer. Höhle, Tilman N. 1990. Assumptions about asymmetric coordination in German. In Grammar in Progress: Glow Essays for Henk van Riemsdijk, J. Mascaró & M. Nespor (eds), 221–235. Dordrecht: Foris. Höhle, Tilman N. 1991. On reconstruction and coordination. In Representation and Derivation in the Theory of Grammar, H. Haider & K. Netter (eds), 139–197. Dordrecht: Kluwer. Kehler, Andrew. 2002. Coherence, reference, and the theory of grammar. Stanford: CSLI. Kratzer, Angelika. 1991. Conditionals. In Semantics: An International Handbook of Contemporary Research, A. von Stechow & D. Wunderlich (eds), 651–656. Berlin: De Gruyter. Kratzer, Angelika. 1996. Severing the external argument from its verb. In Phrase Structure and the Lexicon, J. Rooryck & L. Zaring (eds), 109–137. Dordrecht: Kluwer Acadamic publisher. Larson, Richard. 1988. On the double object construction. Linguistic Inquiry 19: 335–391. Lobeck, Anne. 1995. Ellipsis. Functional Heads: Licensing, and Identification. NewYork, Oxford: OUP. Oirsouw, Robert van. 1993. Coordination. In Syntax: An International Handbook of Contemporary Research, J. Jacobs, A. von Stechow, W. Sternefeld & T. Vennemann (eds), 748–763. Berlin, NewYork: De Gruyter. Progovac, Ljiljana. 1998. Structure for coordination I/II. Glot International 3: 3–6/9.

Odd coordinations  Reich, Ingo. 2007a. Asymmetrische Koordination im Deutschen. Habilitationsschrift, Universität Tübingen. Reich, Ingo. 2007b. What asymmetric coordination in German tells us about the syntax and semantics of conditionals. In Proceedings of Sinn & Bedeutung 11, Barcelona. Universitat Pompeu Fabra. Reis, Marga. 1993. Satzfügung und kommunikative Gewichtung: Zur Grammatik und Pragmatik von Neben- vs. Unterordnung am Beispiel ‘implikativer’ und-Konstruktionen im Deutschen. In Wortstellung und Informationsstruktur, M. Reis (ed.), 203–249. Tübingen: Niemeyer. Roberts, Craige. 1989. Modal subordination and pronominal anaphora in discourse. Linguistics and Philosophy 12: 683–721. Schein, Barry. 1993. Plurals and events. Cambridge, Mass. The MIT Press. Sternefeld, Wolfgang. 2006a. Syntax. Eine morphologisch motivierte generative Beschreibung des Deutschen. Band 1. Tübingen: Stauffenburg. Sternefeld, Wolfgang. 2006b. Syntax. Eine morphologisch motivierte generative Beschreibung des Deutschen. Band 2. Tübingen: Stauffenburg. Truckenbrodt, Hubert. 2006. On the semantic motivation of syntactic verb movement to C. Theoretical Linguistics 32: 257–306. Wilder, Chris. 1997. Some properties of ellipsis in coordination. In Studies on Universal Grammar and Typological Variation, A. Alexiadou & T.A. Hall (eds), 59–107. Philadelphia, Amsterdam: Benjamins. Williams, Edwin. 1978. Across the board rule application. Linguistic Inquiry 9: 31–43. Wunderlich, Dieter. 1988. Some problems of coordination in German. In Natural Language Parsing and Linguistic Theories, U. Reyle & C. Rohrer (eds), 289–316. Dordrecht: Reidel.

part iv

Diachronic perspectives

Old Indic clauses between subordination and coordination1 Rosemarie Lühr University of Jena

In Old Indic main clauses, the verb is usually unaccented. If it has a prefix, the stress lies on the prefix. However, in subordinate clauses that are introduced by a complementizer or a relative pronoun, the verb does bear an accent, whereas any prefix remains unstressed. Apart from that, we find clauses in Old Indic that have the same prosodic properties as subordinate clauses, but are not introduced by a complementizer or a relative pronoun. This paper deals with two questions: Are sentences with an accented verb but no complementizer main sentences or subordinated sentences and under what conditions do sentences with or without stress on the first verb appear. We will see that information structure plays an important role in explaining all these phenomena. Keywords: verb accentuation, Old Indic, information structure, subordination, coordination

1.  The Problem This paper is about a certain verb accentuation in Old Indic that at first glance seems to indicate subordination rather than coordination. A possible solution is influenced by the information-structural and prosodic description of a certain German coordinated clause type. If the following analysis is correct, the description of the German phenomenon is relevant for other languages as well and can be applied not only to modern Indo-European languages but also to historic ones. In Old Indic main clauses, the verb as a rule does not bear the accent. If it has a prefix, the prefix bears the accent. However, in subordinate clauses that are introduced by a complementizer or a relative pronoun, the verb does bear accent, whereas any prefix remains unstressed. (1) main clause subordinate clause verb préfix verb prefix

vérb vérb

.  I thank Augustin Speyer and Jonathan Gress-Wright for the translation into English.

 Rosemarie Lühr

(2) RV V,31,6 a. prá te puØrvāп» i káraп» āni pfx of thee-gen.sg early-acc.pl.n deeds-acc.pl.n Vocam prá nuØtanā I will proclaim-1.sg.inj.aor.act pfx newest-superl.acc.pl.n maghavan rich in giving-voc.sg b. yaØ cakártha / which-rel.acc.pl.n hast done-2.sg.ind.pfv.act c. śáktīvo yád vibhárā powerful-voc.sg that-sub dividest-2.sg.sbjv.pres.act2 ródasī ubhé worlds-acc.pl.n both-acc.pl.n ‘I will proclaim your former deeds, the most recent ones that you, full of gifts, have done, that you, powerful one, divided both halves of the world.’

In Old Indic there are also clauses which are stressed like subordinate clauses, but which do not have an overt complementizer or relative pronoun to introduce them. (3) clause verb préfix verb (4) RV V,45,3 a. ví3 párvato jíh¿Fta apart-pfx mountain-nom.sg.m opens-3.sg.inj.pres.med saØdhata4 dyáur reaches goal-3.sg.inj.pres.med sky-nom.sg.m/f b. āvívāsanto dasayanta trying to win-prt.pres.act.nom.pl.m they wear down themselves 3.pl.inj.pres.med bhuØma earth-acc.sg.n ‘The mountain opens up, the sky reaches the goal. To win the earth, they (the An. giras) wear down themselves (with spells).’

Clauses such as (4) are viewed controversially in Indo-European philology: Dunkel (1985: 51ff.) thinks they are paratactic and interprets the accent on the verb as a sign of emphasis. Hettrich (1988: 166f.), on the other hand, assumes that “Verbalbetonung […] vom vedischen Sprachteilhaber primär als Signal der Unterordnung empfunden [wurde], sobald der implizit gegebene semantische Bezug des betreffenden .  The subjunctive refers to past events here (Hoffmann 1967: 244). .  The prefix is stressed because of emphasis in initial position. .  On the stress of saØdhata see below.

Between subordination and coordination 

Satzes zu seinem Nachbarsatz dies zuließ”5. But these main clauses that are stressed like subordinate clauses are “illokutiv selbständig”6. Therefore he dubs such clauses “Ergänzungssätze”7 (Hettrich 1988: 157). Oldenberg (1906) thought on similar lines: Inasmuch as it is about “Nebeneinander- oder Gegenüberstellung”8 he regards these clauses – which are stressed like subordinate clauses – as subordinate.9 In that he, for part of the clauses, follows Delbrück (1888: 42), who wrote in his Altindische Syntax: “The first clause is the basis for the second one”. Some examples can in fact be interpreted in this way. The relation postea ergo propterea, which is depicted in the order of the clauses (Abraham 1976: 14; Lühr 1989), could underly (5): (5) RV VI 4,7 tvaØm. hí mandrátamam thee-acc.sg namely most pleasing-superl.acc.sg.m arkaśokaír vavr® máhe songs of flames-instr.pl.m we have chosen-1.pl.ind.pf.med máhi nah· śrós·y agne / big-acc.sg.n of us listen-2.sg.imp.aor Agni-voc.sg.m ‘For you, the most pleasing one, we have chosen with songs of flames, [because we wish], hear our big [prayer], oh Agni!’

In that case it would be a final relation: The consequence would be in accordance with the wishes of the referent and the proposition of the hí-clause is something that is – in the speaker’s opinion – the condition for the consequence in the background sentence (“As we worshipped Agni with songs of flames, our wish that he hears our prayer is legitimate”) and that is therefore realized by the referent (Pasch 1987: 138ff.). But if one interprets these two hypotheses in an information-structural way and first scrutinizes the subordination hypothesis, then the difference in verb stress could be seen in connection with the in-focus-form and out-of-focus-form found in many languages (Drubig & Schaffar 2001: 1096ff.; cf. also Gussenhoven 1992). Alternatively one could see the structural strategy of different verb stress at text level as a signal of ‘embossing’, in that the syntactic difference between main and subordinate clause corresponds to the communicative difference between foreground and background

.  ‘The speaker of Vedic saw verb stress primarily as a sign of subordination, if the implicitly given semantic reference of this sentence to its neighboring sentence allowed for that’ (trans. RL). .  ‘independent with respect to their illocution’ (trans. RL). .  ‘supplement clauses’ (trans. RL). .  ‘juxtaposition or opposition’ (trans. RL). .  Klein (1992) has discussed Oldenberg’s material and uses intonation as explanation for the different verb stresses, as I do in this paper. But he does not refer to information structure.

 Rosemarie Lühr

information (Hartmann 1984; Klein & von Stutterheim 1992: 70ff.). As with all other subordinate clauses, the speaker would “durch die Wahl der Nebensatzform zu erkennen [geben], dass es sich nicht um assertierte Information handelt, sondern um eine Voraussetzung zur Einschätzung der eigentlichen Aussage, die ihrerseits in Form eines Hauptsatzes präsentiert wird” (Hetland & Molnár 2001: 630).10 Hartmann (1984) goes one step further: By using a subordinate clause a speaker can signal that the content of the subordinate clause will not be part of the subsequent discourse. In the other case, i.e., the hypothesis of verb emphasis, contexts must be given in which the verb is actually emphasized. In German we find such cases. The Verum Focus, for example: Here the feature [FOC] is assigned to the finite verb. (6) A: Jura will never agree to rework his article. B: Jura WILL rework the article.

In such cases the emphasis lies on the fact that is associated with the verb. The verb must be previously mentioned (Szucsich 2002: 221). As Anita Steube (2001: 236) showed with sentences like in (7), Verum-focus is one of several correction sentences with contrastive focus: (7)

a. b. c. d. e.

Peter hat es [geSAGT]CF corrects lexical meaning Peter [HAT]CF es gesagt corrects time Peter [HAT]CF es gesagt - Verum-focus corrects sentence force Peter [SAGte]CF es corrects lexical meaning Peter [sagTE]CF es corrects time etc.

I had better say right now that neither of the two hypotheses, neither the subordination hypothesis nor the hypothesis of verb emphasis, can be applied to the following material. This can be seen if a main clause and a clause of the problematic type, i.e., with stress typical for a subordinate clause, are compared. Note however that also in real main clauses the verb can be stressed under certain circumstances. It is stressed, for example, if it is the first word of the clause (cf. above to sádhata in (4)).

2. 

Analysis

2.1  Obligatory Stress True, among the Old Indic sentences that are stressed like a subordinate clause there are no sentences such as (7) where the verb is in contrastive focus (cf. also Hettrich 1988: 165f.). Contrast, however, is a conditioning factor for verb stress in any case.

.  ‘[the speaker would] signal by choosing the subordinate mode that it is not asserted information but a presupposition of the actual statement that is presented in the main clause’ (trans. RL).

Between subordination and coordination 

This is demonstrated by certain bipartite structures. They consist of a preposed clause with accent on the verb and a main clause that contains a verb, regularly unstressed. Regularly there is verb stress in the first conjunct of bipartite structures introduced by vā – vā ‘either …. or …’; cf. (8): (8) VII 104,9 yé pākasam ˘˙ sám viháranta · · which-nom.pl.m simple speech-acc.pl.n dishevel-3.pl.ind.pres.med évair yé vā bhadrám · as usual-instr.pl.m which-nom.pl.m or good-acc.sg.n dūsáyanti svadhaØbhih· / · make bad-3.pl.ind.pres.act by own impetus-instr.pl.f áhaye vā taØn pradádātu snake-dat.sg.m either the-acc.pl.m shall surrender-3.sg.imp.pres sóma       aØ vā dadhātu Soma-nom.sg.mpfx or shall bring-3.sg.imp.pres nírr® ter upásthe destruction-gen.sg.f ‘Those who conscientiously twist simple speech or wantonly make bad the good, those Soma either shall surrender to the snake or bring to the womb of destruction.’

In addition, there are the examples with ca – ca ‘X as well as Y’; cf. (9): (9) RV I 120,9 rāyé ca no mimītám · wealth-dat.sg.m and us-dat.pl help-2.du.imp.pres.act vaØjavatyai/ is· é ca no bringing reward-dat.sg food-dat.sg.f and us-dat.pl mimītam dhenumátyai · help-2.du.imp.pres.act rich of milk-dat.sg ‘Entitle us to wealth, bringing reward, and entitle us to food and cows, full of milk!’ (Oldenberg 1906: 716) (but cf. 2.3. to the identical verbs).

Whereas the first vā or ca occurs here after an element of the first conjunct, simple vā ‘or’ and ca ‘and’ appear after a word of the second conjunct; cf. (10) and (11). (10)

RV I 6,9 átah· parijmann aØ gahi thence wandering-voc.sg.m here-pfx come-2.sg.imp.pres.act divó vā rocanaØd ádhi sky-gen.sg.m or light-abl.sg.n from ‘Come here from there, oh wandering one, or from the sky’s empire of light!’

(11) RV I 10,5 ukthám song-nom.sg.n

índrāya Indra-dat.sg.m

śám ˘˙ syam · must be recited-nom.sg.n

 Rosemarie Lühr

várdhanam purunis· s· ídhe / nourishment-nom.sg.n giving many gifts-dat.sg śakró yáthā sutés· u n· o powerful-nom.sg.m so that Soma-juices-loc.pl.m of us-gen.pl rārán· at sakhyés· u ca may have fun-3.sg.subj.pres.act societies-loc.pl.n and ‘A song be recited for Indra for his nourishment; him, to whom we are obliged, so that the powerful one be joyful in our company, with our soma-juice.’

Thus, in bipartite structures with double coordinator, the position of vā and ca in the first conjunct signals to the hearer that a second conjunct is to come. A third kind of sentences with fixed verb stress on the first clause are those containing anyá- – anyá- ‘the one – the other’. Here the first verb is always stressed, when both this clause and the following clause form a Pāda, i.e., a verse (12) (Oldenberg 1906: 724): (12) RV I 123,7 ápānyád11 éty abhy away-pfx-one-nom.sg.n goes-3.sg.ind.pres.act to it-pfx ànyád eti12 vís· urūpe other-nom.sg.n comes-3.sg.ind.pres.act in different shapes-nom.dual.n áhan¿F sám carete / · day and night-nom.dual.n together-pfx go-3.dual.ind.pres.med ‘The one (half of the day) goes, the other comes: both dissimilar halves of the day meet.’

The hearer knows that it is about day and night both from the context (‘the burning flames have risen. The shining Us· as, the dawns, bring to light the treasures, worth to be desired, that have been hidden in darkness’) and from his world knowledge. Types (8) through (9) and (12), i.e., the types with fixed verb stress, have in common that the hearer expects a second conjunct: ‘either X … or Y’, ‘both X … and Y’, ‘the one … the other’. The hypothesis presented here is that the verb stress in the first conjunct is caused by the expectation of a second conjunct. Searching for comparable structures in German, we have to refer to Ewald Lang’s (2004: 58) analysis of bipartite coordinated clausal structures for (13):

→

→

(13) Was machen denn deine Eltern? l*h l*h h % [[[Mein VAter ]T [ist ernsthaft KRANK

]FIP]

.  Cf. (19) as to stressing the prefix. .  Cf. Tichy’s (2000: 43) assumption that verb stress in main clauses is caused by an ‘antithetic accent’.

Between subordination and coordination 

→

→

l*h h*l l % [0 [meine MUtter ]T [geht ARbeiten ]FIP]U] (Lang & Umbach 2002: 155) ‘What are your parents doing? – My dad is seriously ill, my mom goes to work.’ (Transl. RL)

The two essential properties of this construction are:

→

→

→

a. Within the utterance domain [...U] the subjects mein Vater, meine Mutter are evoked by the previously mentioned deine Eltern and thus belong to the background. Prosodically they are contrastive topics, marked by “ ” and “L*H H”. b. Both conjuncts form prosodically symmetrical Intonational Phrases (IPs) with a difference at the end of both conjuncts: The first conjunct has , the second . Also “der steigende Akzent bildet ikonisch13 Offenheit bzw. Unabgeschlossenheit ab und evoziert dadurch eine Alternativmenge mit einer nicht leer laufenden offenen Proposition …”14 (Molnár & Rosengren 1996: 82; Jacobs 1997: 124). We now have to check whether we find these properties also in the Old Indic example (12)! Property (a) is present in any case: ‘The one day-half goes, the other comes’ contains two contrastive topics whose referents are known from the context and because of world-knowledge. About property (b) my approach is: The accent on the clause-final éti ‘he goes’ in the first conjunct is comparable to the rising contour L*H H in the German example.15 In fact, the Old Indic accent marked by the Udātta must be a rising contour, as the name (‘high’, ‘exalted’) suggests. The Udātta in the contrastive topics anyád – anyád ‘the one – the other’, on the other hand, was probably accompanied “von einer Stimmverstärkung”16 (Wackernagel 1896: 284; Klein 1992: 86). A contrast can be expressed also by antonyms, as exemplified in (14). The structure is: …

vérb / … préfix verb

→

(14) first conjunct

second conjunct

.  The term ‘indexalic’ is to be preferred here (O.Panagl, p.c.). .  ‘the rising accent denotes openness iconically or that it is not yet finished and by that evokes an alternative set with an open proposition that is going to be filled’ (trans. RL). .  Klein (1992: 38ff.) also traces back the verb accent in such bipartite structures to “incompleteness”. He refers to it as “intonational accent”, or as “anti-final contour” (1992: 89). In addition he assumes – justifiably – “two fundamentally different phonetic bases for verbal accentuation in the Rigveda: salience/emphasis and heightened intonation” (1992: 91). .  ‘by a rise in volume and/or pitch’ (trans. RL).

 Rosemarie Lühr

(15)

RV I,152,3 r® tám píparty ánr·tam · truth-acc.sg.n fosters-3.sg.ind.pres.act untruth-acc.sg.n ní tār¿F down-PFX presses-3.sg.ind.pres.act ‘Truth he fosters, untruth he suppresses.’

Since (15) as a whole is the answer to the focus question ‘what does the god do?’, the antonyms ‘truth’ and ‘untruth’ are contrastive foci. As in other languages too, ‘contrast’ is no intrinsic property of either focus or topic constituents (Molnar 2002; Vallduvi & Vilkuna 1998). Thus the consequence of this interpretation of the Old Indic verbal stress in the first conjunct is: The Udātta on the verb serves to structure the text.17 By choosing a rising verb accent on the verb of the first intonational phrase it is hinted that this is an initial conjunct of a coordinated bipartite structure and that a final member is to follow.18 … … prefix

vérb vérb

/… verb second conjunct / préfix verb

(17) first conjunct

…

vérb

/ … préfix verb second conjunct

→

→ →

(16) first conjunct

Let us look at two other comparisons of Vedic and German. Firstly, Delbrück (1871) utters the following to the state of affairs in Proto-Indoeuropean (but compare his contradicting statements quoted in 2.1): Haupt- und Nebengedanken und folglich Haupt- und Nebensätze existirten schon in der Periode des einfachen Satzes vor der Entstehung des Relativums und der Conjunctionen, nur dass sie kein sprachliches Zeichen hatten, ausser dem freilich sehr mächtigen und mannigfaltigster Nüançen fähigen Satzbetonung19.  (Delbrück 1871: 98)

.  Klein (1992: 88), too, interprets verb stress in Old Indic subordinate clauses as a signal that “the accented clause is incomplete”. If, however, the emphasis on the subordinate clause verb is a signal that the hearer should expect another clause, in this case a main clause (as ‘two-verb’ coordinated structures suggest, Lühr 2007), this verb emphasis would have to have started in preposed subordinate clauses and from there would have to have been transferred to postposed subordinate clauses (H.Hettrich, Th.Krisch, p.c.; Klein 1992: 91ff.). .  Cf. German, when guiding person into a parking space: Weiter!, Weiter!, Weiter!, Stopp! ‘On! On! On! Stop!’ (Peters 2005: 111). .  ‘main thoughts and subordinate thoughts, consequently main clauses and subordinate clauses, were in existence already in the period of the simple sentence, before complementizers and relative pronouns came into being. They only had no linguistic sign assigned to them, save for the powerful and finely grained sentence stress’ (transl. RL).

Between subordination and coordination 

Furthermore, Delbrück assumed that the special stress of a verb in a subordinate German clause is comparable to the Vedic accentuation and that it has something to do with importance (Kümmel, p.c.): Der Nebensatz [enthält] sehr oft gerade das …, was im Zusammenhange der Rede das Allerwichtigste ist, und im Sanskrit, wo z.B. die Relativsätze zum allergrössten Theile ‘nothwendige’ sind, ist das besonders häufig der Fall. Dadurch allein schon ist eine stärkere Betonung des Nebensatzverbums gerechtfertigt. Nun kommt noch die Gewohnheit des Sanskrit hinzu, die Nebensätze voranzustellen, wodurch unläugbar in dem Hörenden eine Spannung auf den Hauptsatz hervorgerufen (Delbück 1871: 96ff) wird.20

At least the last assumption – that the hearer expects a following phrase, if a verb is stressed in Vedic – is compatible with our solution, while the importance criterion of the subordinate clause does not convince. Secondly Tichy (2000: 43) assumes that clausal structures in Vedic and ProtoIndo-European had a similar intonation as a certain Modern German sentence type. She compares: (18) Hätt’ ich’s gewusst (/), hätt’ ich’s gesagt (\). ‘If I had known it, I would have said it.’

(Transl. RL)

Under certain circumstances also the main clause verb could have been stressed, “so bei der Gegenüberstellung mit einem inhaltlich kontrastierenden Verbum (‘antithetischer Akzent’)”21: (19) Der eine kommt (/), der andre geht (\) ‘One comes, the other leaves.’

(Transl. RL)

and “ferner in Verbindung mit affirmativen (= bekräftigenden, wie ved. íd) oder kausalen (= begründenden, wie ved. hí) Partikeln;22 vgl. nhd. Ich hab’s ja gewusst (/).23

.  ‘The subordinate clause often contains exactly that which is most relevant in the context of the text, and this goes especially frequently for Sanskrit, in which e.g., nearly all relative clauses are ‘necessary’. This alone justifies a stronger emphasis on the subordinate clause verb. In addition there is the tendency of Sanskrit to prepose the subordinate clauses. This, beyond any doubt, evokes suspense for the main clause in the hearer’s mind.’ (Transl. RL). .  ‘if put in opposition to another verb with contrasting content (antithetic accent)’ (transl. RL). .  ‘furthermore in connection with affirmative (as Vedic íd) or causal (as Vedic hí) particles’ (transl. RL). .  ‘Well, I knew it’ (transl. RL).

 Rosemarie Lühr

Diese Verhältnisse [seien] aus dem Vedischen erschlossen, wo die angegebenen Regeln gelten (allerdings gibt es dort keine konjunktionslosen Nebensätze mehr)”.24 However, we do not gain much from a comparison with Modern German clauses that contain a complex predicate (Hätt’ ich’s gewusst (/), hätt’ ich’s gesagt), as ProtoIndo-European had certainly no such predicates. The assumption that main clause verb stress started from cases with contrastive focus on this part of speech (Der eine kommt (/), der andere geht (\)) is not very likely either, for there are no examples in ­Vedic where the verb is in contrastive focus as mentioned above. Expectation of another proposition seems to be a much more plausible starting point. After having dealt with the conditions under which sentences having an accented verb but no complementizer appear we can give an answer to the question if those sentences are main clauses or subordinated ones: The rising contour in the first intonational phrase does not signal subordination but coordination.

2.2  Optional stress In the examples (8) through (9), (12) and (15) the rising contour on the verb (together with lexical means such as the bipartite vā … vā, ca … ca, anyá- … anyá-, or antonyms) causes an expectation for a second conjunct. We now have to scrutinize cases in which the bipartiteness in both conjuncts containing a verb is expressed purely by accent and not by lexical means. The relevant examples are of the following kind: The verb of the first conjunct is stressed and in final position, as in (12). The verb of the second conjunct is stressed as well. It is in clause-initial position and consequently shows the already mentioned main clause stress. …

vérb

→

(20) first conjunct

/ vérb

…

second conjunct

Oldenberg (1906: 728; cf. Klein 1992: 33ff.) describes such structures as follows: In front of the first verb and behind the second verb usually there are elements that correspond to each other, such as two subjects or two objects. One also finds heterogenous material on both sides, e.g., first place nominative, second place instrumental. The middle of the Pāda, “wo die beiden [betonten] Verben zusammenstoßen”25 he terms culmination point (“Gipfelpunkt”). But I assume also here that the Udātta in conjunct-final position on the verb of the first conjunct is a signal for bipartiteness, cf. (21): (21) RV I 135,8 yám aŚvatthám upa­tís» t» hanta which-acc.sg.m AŚvatta-tree-acc.sg.m reach-3.pl.ind.pres.med

.  ‘This state of affairs has been concluded from the Vedic situation where the afore mentioned rules apply; there are, however, no more subordinate clauses without covert complementizer’ (transl. RL). .  ‘where both (stressed) verbs clash’ (transl. RL).

Between subordination and coordination 

jāyávo 'smé té winners-nom.pl.m at us-loc.pl those-nom.pl.m santu jāyávah» / sākám» shall be-3.pl.imp.pres.act winners-nom.pl.m simultaneously gaØvah» súvate cows-nom.pl.m/f give birth-3.pl.ind.pres.med pácyate yávo ripens-3.sg.ind.pres.med grain-nom.sg.n ‘Those who stepped underneath the AŚvattha‑tree [tree of life] as winners, those shall be winners with us. At this time cows give birth and the grain ripens.’

The hearer knows that the soma sacrifice causes wealth and happiness for the human beings. Thus ‘cows’ and ‘grain’ are again contrastive topics, and both conjuncts are the answer to the focus question ‘what wealth is given to the human beings?’ Example (22) also refers to the situation of a sacrifice: (22) RV I 135,7 yátra graØvā vádati tátra where millstone-nom.sg.m sounds-3.sg.ind.pres.act thither gachatam» gr® hám índraś 26 ca come-2.dual.imp.pres.act house-acc.sg.m Indra-nom.sg.m and gachatam / ví sūnr®' tā come-2.dual.imp.pres.act pfx glory-nom.sg.f dádr® śe rØF yate will be seen-3.sg.ind.pres.med flows-3.sg.ind.pres.med ghr® tám lard-nom.sg.n ‘Where the stone sounds, thither come, you and Indra, come into the house! The glory is to be seen, lard is flowing.’ (Kümmel 2000: 233)

The focus question here could be: ‘what happens at the soma sacrifice, if rightly executed?’ In ví sūnr®' tā dádr® śe rØF yate ghr® tám not only is the verb dádr® śe in the first conjunct stressed, but also the initial prefix ví. This is in accordance with Old Indic language usage: Prefixes standing at the top of the clause are optionally stressed. As for optionality, cf. (23): (23) RV I 40,8 upa ks» atrám pr® ñcītá pfx government-acc.sg.n might multiply-3.sg.opt.pres.med

.  A peculiar use of ca in the RV is to add one noun (nearly always in the nominative) to another, which has to be supplied.

 Rosemarie Lühr

hánti raØjabhir hits-3.sg.ind.pres.act kings-instr.pl.m ‘He may extend his government, he hits (the enemy) with the kings.’

Thus (23) has the following intonational structure:

→

(24) First conjunct prefix … vérb / vérb … second conjunct

Now we are in a position to analyse example (4), which also has the accent on the clause-initial prefix. The example is repeated below. (4) RV V 45,3 a. asmaØ ukthaØya párvatasya this-dat.sg.n saying-dat.sg.n mountain-gen.sg.m gárbho mahØF nām· janús· e pūrvyaØya / body-nom.sg.m exalted-gen.pl.f birth-dat.sg.n first-dat.sg.n b.i ví párvato jíhīta apart-pfx mountain-nom.sg.m opens-3.sg.inj.pres.med b.ii saØdhata dyáur reaches goal-3.sg.inj.pres.med sky-nom.sg.m/f ‘The body of the mountain opens up to this spell for the birth of the exalted (dawns). The mountain opens up, the sky reaches the goal.’27

(4b.i) here refers to the preceding context (4a). Note that the missing verb in (4a) must be supplemented from (4b.i): ‘The body of the mountain opens up to this spell for the birth of the exalted (dawns). The mountain opens up’. Thus (4b) is an answer to the focus question: ‘What happens if the body of the mountain [does X] to this spell for the birth of the dawns?’ Answer: ‘The mountain opens up, the sky reaches the goal’. Whereas the examples up to now have been coordinations of two conjuncts, there are also examples with four conjuncts (25): The preceding context is: ‘The lion’s thunders are rolling from afar, when Parjanya makes rain-bringing clouds.’ (25)

RV V 83,4 prá vaØtā vaØnti  / forward-pfx winds-nom.pl.m blow-3.pl.ind.pres.act patáyanti vidyútah· úd fall-2.pl.ind.pres.act flashes-nom.pl.f upward-pfx ós·adhīr jíhate  / pínvat plants-nom.pl.f rise-3.pl.ind.pres.med streams-3.sg.ind.pres.med

.  Hettrich (1988: 160) here assumes a modal-adverbial relation and illustrates this by the following translation: “ebenso wie der Berg sich auftut, [gelangt] zum Ziel der Himmel” (Hoffmann 1967: 174: ‘Der Berg tut sich auf und gelangt zum Himmel’).

Between subordination and coordination 

svàh· / sky-nom.sg.m/f ‘The winds start blowing, flashes fall, the plants rise, the sun overflows.’

Everybody knows what is going to happen if Parjanya excites a thunderstorm. The poet formulates the answer in two parallel Pādas with two conjuncts each and a rising tone at the end of the first conjunct. The structure is:

→ →

(26) first conjunct préfix … vérb third conjunct prefix … vérb

/ vérb … second conjunct / vérb … fourth conjunct

If the structure is tripartite, only the first and the second conjunct have the intonational structure with an Udātta in the first conjunct, as in the answer to the focus question ‘What does the Old Indic god Savitar effect?’: (27)

RV I 35,9 ápaØmīvām baØdhate away-pfx-pressure-acc.sg.f dispel-3.sg.ind.pres.med véti suØryam abhí drive-3.sg.ind.pres.act sun-.acc.sg.m towards-pfx kr® s·n· éna rájasā dyaØm black-instr.sg.n mist-instr.sg.n sky-acc.sg.m/f r® n· oti reaches-3.sg.ind.pres.act ‘He dispels sickness, drives the sun; he reaches the sky with his black mist.’

→

(28) a. first conjunct prefix … vérb / vérb … second conjunct b. third conjunct préfix … verb

Structure (28) with a stressed verb before a stressed verb in the middle of the Pāda is not compulsory, as (29) shows: In the first conjunct the verb carati is unstressed before the stressed verb of the second conjunct titvis· é. The passage is about Indra’s battle in the Vala-myth: (29) RV I 52,6 pár- īm· ghr® n· aØ carati  / around-pfx him heat-nom.sg.f goes-3.sg.ind.pres.act titvis.é Śávo is incited-3.sg.ind.pf.med power-nom.sg.n ‘The glowing heat goes around him, his power is incited.’ (Kümmel 2000: 224) (30) first conjunct

préfix

…

verb

/ vérb … second conjunct

By using the unstressed verb in the first conjunct the poet obviously signals that he regards both conjuncts as confined informational units. The subjects of the two conjuncts differ without being in contrast to one other. Thus, if no contrast is expressed

 Rosemarie Lühr

lexically, in the first conjunct both structures are possible, the one with verb stress and the one without. This is true if both conjuncts meet in the middle of the Pāda.

→

(31) a. first conjunct prefix … vérb / vérb … second conjunct b. first conjunct prefix … verb / vérb … second conjunct

But type (31a) is more common (Oldenberg 1906: 726ff.). Furthermore, verb stress in the first conjunct is optional if both conjuncts are extended over several Pādas. Whereas in (32) the verb in the first conjunct is stressed, in (33) it is not: (32) RV I 164,20 dvaØ supan» aØ sayújā two-nom.dual.m birds-nom.dual.m connected-nom.dual.m sákhāyā samān» nám. vr® ks·ám companions-nom.dual.m together-adv tree-acc.sg.m pári s·asvajāte / around-pfx keep hugged-3.dual.ind.pf.med táyor anyáh. of them-gen.dual.m the one-nom.sg.m píppalam. svādv átty berry-acc.sg.n sweet-acc.sg.n eats-3.sg.ind.pres.act ánaŚnann anyó abhí not eating-prt.pres.act.nom.sg.m the other-nom.sg.m towards-pfx cākaŚĪti looks on-3.sg.ind.pres.act ‘Two birds, close companions, are hugging the same tree. One of them eats the sweet berry, the other looks on, not eating.’ (Kümmel 2000: 591) (33) RV I 93,6 aØnyám» divó mātaríŚvā PFX-the one-acc.sg.m sky-gen.sg.m/f Mātarisvan-nom.sg.m jabhāra ámathnād has brought-3.sg.ind.pf.act tore-3.sg.ind.ipv.act anyám pári Śyenó ádreh· the other-acc.sg.m hither-pfx eagle-nom.sg.m rock-gen.sg.m ‘MātariŚvan brought one from the sky, the eagle tore the other away from the rock.’ (Kümmel 2000: 338ff.)

The lack of lexical marking of contrast and the distribution of conjuncts on two Pādas thus cause optional verb stress in the first conjunct. It is unknown whether this is a metrical or a linguistic phenomenon, or both.28

.  We know from other areas of Vedic syntax that the poets artificially equated pada and sentence, so we get pada-initial topicalization (Hale 1987: 8ff.).

Between subordination and coordination 

2.3  Obligatory non-stress Turning to cases in which the verb in the first conjunct is compulsorily unstressed, we see that there are only two patterns: Firstly, the verb in the first conjunct is unstressed if it is identical to the verb of the second conjunct. (34) RV I 103,5 sá gaØ avindat he-nom.sg.m cows-acc.pl.m/f found-3.sg.ind.ipv.act só avindad áŚvān he-nom.sg.m-part found-3.sg.ind.ipv.act horses-acc.pl.m sá ós» adhĪh· só apáh· he-nom.sg.m plants-acc.pl.f he-nom.sg.m-part water-acc.pl.f sá vánāni he-nom.sg.m trees-acc.pl.n ‘He (Indra) won the cows, he won the horses, the plants, the rivers and seas, the trees.’ f. also (35): C (35) RV X 94,1 praíté vadantu forward-pfx-these-nom.pl.m shall speak-3.pl.imp.pres.act prá vayám. vadāma forward-pfx we-nom.pl want to speak-1.pl.subj.pres.act ‘They shall start to speak, we want to start to speak.’ (36) RV I 134,5 tvám víśvasmād bhúvanāt pāsi · thou-nom.sg all-abl.sg.n world-abl.sg.n protect-2.sg.ind.pres.act dhárman· ā -suryàt pāsi kind-instr.sg.n Asura-power-abl.sg.m protect-2.sg.ind.pres.act dhárman· ā kind-instr.sg.n ‘You protect from all the world, as it befits your kind, you protect from the Asura-power, as it befits your kind.’

With regard to deaccenting, the examples above are comparable to German sentences showing broad focus and deletion of an identical element or gapping (Hartmann 2000: 171), as for instance in (37): (37) a.

weil [IP HánsF [VP [ eine ÈRDbeere]f isstf]f]f und because John          a STRAWberry eats and [PéterF [VP eine KÌRsche]f isstf]f]f]f. Peter a CHERry ‘because John eats a strawberry, and Peter a cherry’

Those sentences, in which the second verb is deleted, are mostly to be found in spoken New High German. They may be reputed to be a marked option in comparison with

 Rosemarie Lühr

(37)(b) with deletion of the first verb. This is usual in written language. In the literature also the terms “Einsparung” (Duden 2005: 912) or “coordinate ellipsis” (Schwabe 2000: 248ff.) are used: (37) b.

weil [IP HánsF [VP [ because John und [ PéterF [VP eine and Peter a

eine ÈRDbeere]F isstF]F]F a STRAWberry KÌRsche]f isstF]F]F]F. CHERry eats

Nevertheless, sentences like (37a) do occur. Cf. also examples (38)–(39) with distinct morphological features (Lang & Umbach 2002: 161): (38) ER trank BIER und WIR tranken WEIN. he drank beer and we drank wine (39) MAX wurde KRANK und WIR wurden GESUND. Max became ill and we became healthy ‘Max went ill and we recovered.’ Here, the two identical verbs are focus-marked by focus-projection rule (40ii): (40) Focus Projection i. F-marking of the head of a phrase licenses the F-marking of the phrase. ii. F-marking of an internal argument of a head licenses the F-marking of the head (Selkirk 1995: 555; Hartmann 2000: 126).

(40ii) means, that the whole utterance can only be focused (maximal or wide focus) if the constituent carrying the nuclear accent (focus exponent) is in its base position and in the sister position of the verbal head, as for instance in (41): (41)

[CP Maria hat behauptet, [CP dass [vp die Tante [V’ die Nichten Maria has claimed    that     the aunt the nieces [vo begrüßt hat]]]] welcomed has ‘Maria claimed that the aunt welcomed the nieces.’ Maria hat behauptet, dass [die Tante [die NICHten begrüßt hat] ‘Maria claimed that the aunt welcomed the NIEces.’

Moreover, (41) shows that the main stress of a phrase is assigned to its most deeply embedded constituent. If there is no other focus structural information given, the focus projects (Stolterfoht & Bader 2004: 261). The main factor in our examples, however, is maximal contrast of the determiner phrases in (42): (42) Maximal Contrast Principle In a Gapping construction the number of contrasting remnant-correspondent pairs is maximized.

Between subordination and coordination 

This principle ensures that there is no remnant without an adequate correspondent. A verb will always remain unaccented, as long as it is not a focus (Hartmann 2000: 165, 170). Thus, the Maximal Contrast Principle concerns phonological deaccenting in contrastive structures. It applies if two constituents are identical. In that case, gapping either of the first or the second constituent takes place. The result is increasing stress on the remaining contrasting pairs, here the determiner phrases. In Old Indic such contrastive accents are obviously created intonationally by the complete lack of accent on identical verbs. While both verbs are preserved here, in German deaccenting results in deletion of one verb on the level of Phonological Form in consequence of source and target identity (Hartmann 2000). On the other hand, we expect a ‘fortition of voice’ on contrasting corresponding pairs, following the Maximal Contrast Principle as in the German examples. An exception from our deaccenting rule are cases involving ca … ca ‘both … and’ and identical verbs, as in example (9), repeated here: (9) RV I 120,9 rāyé ca no mimītám · wealth-dat.sg.m and us-dat.pl help-2.du.imp.pres.act vaØjavatyai / is· é ca no bringing reward-dat.sg food-dat.sg.f and us-dat.pl mimītam dhenumátyai · help-2.du.imp.pres.act rich of milk-dat.sg ‘Entitle us to wealth, bringing reward, and entitle us to food and cows, full of milk!’

Cf. also (43): (43) RV I 123,12 áśvāvat¿Fr gómat¿Fr bringing horses-nom.pl.m bringing cows-nom.pl.m viśvávārā yátamānā wished by all-nom.pl.m contesting-part.pres.med.nom.pl.f raśmibhih· suØryasya / párā ca yánti rays-instr.pl.m Sūrya-gen.sg.m away-pfx and go-3.pl.ind.pres.act púnar aØ ca yanti bhadraØ naØma again-adv hither and go-3.pl.ind.pres.act lucky names-acc.pl.n váhamānā uśaØsah· bringing-prt.pres.med.nom.pl.f dawns-nom.pl.f ‘Bringing horses and cows, wished by all, contesting with Sūrya’s rays the dawns go and come, with names bringing luck.’

Remember that the Vedic Udātta on the verb of the first conjunct is not considered as a means of marking prominence but as a signal to expect another conjunct. But since ca already signals a following conjunct, stress on the identical verb of the first conjunct

 Rosemarie Lühr

seems to be redundant. Thus, there might exist a rule which overrides deaccenting of the first of two identical verbs in the surroundings of ca … ca.29 In Optimality Theory such a rule would be comparable to the constraint MAX(LEX),30 which is not dominated by other constraints.31 On the contrary, expected deaccenting appears in the first conjunct with ná … ná ‘neither… nor’, nearly without exception.32 This is our second pattern for obligatory non-stress; cf. the sentences (44) and (45), which contain antonyms: (44) RV X 129,1 naaØsad ās¿Fn nó neither-nonexistence-nom.sg.n was-3.sg.ind.ipf.act nor-part sád ās¿Ft tadaØn¿Fm · naØs¿Fd being-nom.sg.n was-3.sg.ind.ipf.act then-adv not-was-3.sg.ind.ipf.act rájo nó vyòmā paró yát / air space-nom.sg.n not-part sky-nom.sg.n above-adv ‘There was neither non-existence nor existence; neither was air space nor the sky above it.’ (45)

RV I 113,3 ná methete ná not are in mutual fight-3.dual.ind.pres.med not tasthatuh· stand still-3.dual.ind.pf.act ‘They do not start a fight, they do not stand still’

(Oldenberg 1906: 717)

In both examples sentence negation occurs, in (44) in combination with constituent negation. It has been unclear until now why the first verb is unstressed. We propose an explanation following Trautwein’s (2005: 216f.) analysis of the semantics of sentence negation. Here, his assumption that negative sentences “tend to denote almost nothing” is decisive. Looking at a single negated sentence like (46) Eli did non run/laugh/wait.

.  The assumption that accentuation of the verb following the first ca is due to fact that there is no determiner phrase to put the stress on holds for example (43) but not for example (9), cf. rāyé at the beginning of the sentence. .  Cf. Wunderlich (2003): Max (lexF): Every lexically assigned (positively-valued) feature in the input has a correspondent in the output. .  Admittedly, as we are dealing with a dead language claims about obligatoriness are uncertain. We cannot extend the corpus by tests (as we do when dealing with living languages, combining intuition and corpus search). But every language has rules that are fixed in the lexicon. .  Counterexamples are to be found in RV I 62,12 und III 53, 14 with “Doppelgegensatz” (Oldenberg 1906: 717 footnote 2).

Between subordination and coordination 

we find that almost everything could be the case in the situations denoted by (46). Although Eli did not run, it is possible nevertheless that she was moving somehow (driving, cycling etc.). Only the presupposed configuration which would also precede the situation denoted by the corresponding positive sentence remains “as the possible truthmaker of the negative propositional description”. But many verbs, including several state and activity verbs presuppose no specific scenario at all. Anyway, the most important semantic feature is that sentence negation neutralizes the complete aspectual information. Therefore, no continuation of any event or state is expected if you do not have double coordinators like English neither … nor or German weder … noch containing different lexemes but only a general expression meaning ‘not’ and no specific word order information is available. Exactly this is the case in Old Indic (for word order cf. the Old Indic examples with double vā … vā or ca … ca). The verb of the first conjunct may be completely deaccented since nothing lets one expect a continuation of the first conjunct following the regularly accented ná ‘not’ at the top of the sentence. However, besides the negator the negated determiner phrases are stressed. Thus, negated sentences are instances of contrastive structures, too. They are similar to the above mentioned structures, which contain identical unstressed verbs but contrastive stress on determiner phrases.

3.  Summary All Old Indic examples that have been put forward are cases of coordination and not of subordination, although the verb in the first conjunct tends to be stressed as in subordinate clauses. Verb stress in the first conjunct, however, functions as a signal that a second conjunct is to follow. It is a rising tone, as is the usual realisation of the Old Indic Udātta. We find an Udātta of that kind obligatorily in bipartite phrases of the types ‘either X or Y’, ‘both X and Y’, ‘the one… the other…’ and with antonyms. The Udātta is optional if such lexical means are missing and two verbs clash in the middle of the verse, for instance. Generally, the Udātta is not present if the two verbs of the conjuncts are identical, as here the contrastive accents on the determiner phrases seem to be the main factor, similar to German gapping constructions. But sentences with ca … ca ‘both X and X’ and identical verbs show redundancy, since the first verb is stressed. Not only the semantics of ca but also the verbal accent signal expectation of a second conjunct. To explain this phenomenon there was no other way left but a lexical rule. On the other hand, in sentences with ná – ná ‘neither – nor’ both verbs are unstressed. Here, the determiner phrases are maximally contrasted and bear contrastive accents as in the gapping constructions. Probably, the specific semantics of sentence negation is of importance. This paper is an attempt to make statements about information structure and accentuation even for very old languages.33 .  Cf. Klein 1992: 95.

 Rosemarie Lühr

References Abraham, Werner. 1976. Die Rolle von Trugschlüssen in der Diachronie von Satzkonnektoren. In Opuscula slavica et linguistica. Festschrift für A. Issatschenko [Schriftenreihe Sprachwissenschaft. Universität für Bildungswissenschaften], Heinz Dieter Pohl & Nikolai Salnikow (eds), 11–72. Klagenfurt: Heyn. Delbrück, Berthold. 1871. Der Gebrauch des Conjunctivs und Optativs im Sanskrit und Griechischen [Syntaktische Forschungen 1]. Halle: Waisenhaus. Delbrück, Berthold. 1888. Altindische Syntax [Syntaktische Forschungen 5]. Halle: Waisenhaus. Drubig, Hans Bernhard & Schaffar, Wolfram. 2001. Focus constructions. In Language Typology and Language Universals [Handbücher zur Sprach- und Kommunikationswissenschaft (HSK) 20(2)], Martin Haspelmath & Ekkehard König (eds), 1079–1104. Berlin: de Gruyter. Duden. 2005. Duden. Die Grammatik [Der Duden in zwölf Bänden 4]. Mannheim: Dudenverlag. Dunkel, George. 1985. IE hortatory *ey, *eyte: Ved. éta … stávāma, Hitt. ehu=wa it, Hom. εi δ’ ˘ " >ε. Münchener Studien zur Sprachwissenschaft 46: 47–79. α Gussenhoven, Carlos. 1992. Sentence accents and argument structure. In Thematic Structure. Its Role in Grammar, Iggy Roca (ed.), 79–106. Berlin: Foris. Hale, Mark. 1987. Syntax in the Comparative Syntax of the Oldest Indo-Iranian Languages. PhD dissertation, MIT. Hartmann, Dietrich. 1984. Reliefgebung: Informationsvordergrund und Informationshintergrund in Texten als Problem von Textlinguistik und Stilistik. Zur Verwendung sprachlicher Mittel zum Ausdruck von Haupt- und Nebeninformation. Wirkendes Wort 4: 305–323. Hartmann, Katharina. 2000. Right Node Raising and Gapping. Interface Conditions on Prosodic Deletion. Amsterdam: John Benjamins. Haspelmath, Martin & König, Ekkehard (eds). 2001. Language Typology and Language Universals [Handbücher zur Sprach- und Kommunikationswissenschaft (HSK) 20.2]. Berlin: de Gruyter. Hetland, Jorunn & Molnár, Valéria. 2001. Informationsstruktur und Reliefgebung. In Language Typology and Language Universals [Handbücher zur Sprach- und Kommunikationswissenschaft (HSK) 20.2], Martin Haspelmath & Ekkehard König (eds), 617–633. Berlin: de Gruyter. Hettrich, Heinrich. 1988. Untersuchungen zur Hypotaxe im Vedischen [Untersuchungen zur indogermanischen Sprach- und Kulturwissenschaft 4]. Berlin: de Gruyter. Hoffmann, Karl. 1967. Der Injunktiv im Veda. Eine synchrone Funktionsuntersuchung. Heidelberg: Winter. Jacobs, Joachim. 1997. I-Topikalisierung. Linguistische Berichte 168: 91–133. Klein, Jared. 1992. On verbal accentuation in the Rigveda [American Oriental Society: Essay nr. 11]. New Haven CT: American Oriental Society. Klein, Wolfgang & von Stutterheim, Christiane. 1992. Textstruktur und referentielle Bewegung. Zeitschrift für Literaturwissenschaft und Linguistik 86: 67–92. Kümmel, Martin Joachim. 2000. Das Perfekt im Indoiranischen. Eine Untersuchung der Form und Funktion einer ererbten Kategorie des Verbums und ihrer Weiterentwicklung in den altindogermanischen Sprachen. Wiesbaden: Reichert. Lang, Ewald. 2004. Schnittstellen bei der Konnektoren-Beschreibung. In Brücken schlagen. Grundlagen der Konnektorensemantik [Linguistik – Impulse & Tendenzen 5], Hardarik Blühdorn, Eva Breindl & Ulrich Hermann Waßner (eds), 45–92. Berlin: Walter de Gruyter.

Between subordination and coordination  Lang, Ewald & Umbach, Carla. 2002. Kontrast in der Grammatik: Spezifische Realisierungen und übergreifender Konnex. In Sprachtheoretische Grundlagen der Kognitionswissenschaft: Sprachliches und nichtsprachliches Wissen [Linguistische Arbeitsberichte 79], Anita Steube (ed.), 145–186. Leipzig: University of Leipzig. Lühr, Rosemarie. 1989. Ist die Basiskategorie TEMPORAL als Ausgangspunkt für konjunktionale Bedeutungen ein Universale? Zur Polyfunktionalität von Konjunktionen in den indogermanischen Sprachen. Historische Sprachforschung 102: 153–173. Lühr, Rosemarie. 2007. Verbakzent und Informationsstruktur. In Struktur und Semantik der Verbalphrase, Rosemarie Lühr & Susanne Zeilfelder (eds). Molnár, Valéria. 2002. Contrast from a contrastive perspective. In Information Structure in a crosslinguistic perspective, Cathrine Fabricius-Hansen et al. (eds), 147–162. Amsterdam: Rodopi. Molnár, Valeria & Rosengren, Inger. 1996. Zu Jacobs’ Explikation der I-Topikalisierung. Sprache & Pragmatik: Arbeitsberichte 41: 49–88. Oldenberg, Hermann. 1906. Vedische Untersuchungen. 16. Die Verbalenklisis im Rigveda. Zeitschrift der Morgenländischen Gesellschaft 60: 707–760. Pasch, Renate. 1987. Illokutionsstrukturtypen und Typen der Verknüpfung von Inhalten illokutiver Handlungen. In Satz, Text, sprachliche Handlung [Studia Grammatica 25], Wolfgang Motsch (ed.), 119–161. Berlin: Akademie-Verlag. Peters, Jörg. 2005. Intonation. In Duden. Die Grammatik, 7 Aufl. [Duden Band 4], 95–128. Mann­­ heim: Dudenverlag. Schwabe, Kerstin. 2000. Coordinative ellipsis and information structure. In Ellipsis in Conjunction [Linguistische Arbeiten 418], Kerstin Schwabe & Ning Zhang (eds), 247–269. Tübingen: Niemeyer. Selkirk, Elisabeth. 1995. Sentence prosody: Intonation, stress, and phrasing. In The Handbook of Phonological Theory, John A. Goldsmith (ed.), 550–570. Cambridge: Blackwell. Steube, Anita. 2001. Correction by contrastive focus. Theoretical Linguistics 27: 215–249. Stolterfoht, Britta & Bader, Markus. 2004. Focus structure and the processing of word order variation. In Information Structure. Theoretical and Empirical Aspects [Language, Context and Cognition 1], Anita Steube (ed.), 259–275. Berlin: de Gruyter. Szucsich, Luka. 2002. Informationsstruktur in komplexen russischen Sätzen. In Sprachtheoretische Grundlagen der Kognitionswissenschaft: Sprachliches und nichtsprachliches Wissen [Linguistische Arbeitsberichte 79], Anita Steube (ed.), 213–233. Leipzig: University of Leipzig. Tichy, Eva. 2000. Indogermanistisches Grundwissen für Studierende sprachwissenschaftlicher Disziplinen. Bremen: Hempen. Trautwein, Martin. 2005. The Time Window of Language. The Interaction between Linguistic and Non-Linguistic Knowledge in the Temporal Interpretation of German and English Texts [Language, Context and Cognition 2]. Berlin: de Gruyter. Vallduvi, Eric & Vilkuna, Marina. 1998. On rheme and contrast. In The Limits of Syntax [Syntax and Semantics 29], Peter W. Culicover & Louise McNally (eds), 79–108. San Diego CA: Academic Press. Wackernagel, Jacob. 1896. Altindische Grammatik, I: Lautlehre. Göttingen: Vandenhoeck & ­Ruprecht. Wunderlich, Dieter. 2003. Optimal case patterns: German and Icelandic compared. In New Perspectives on Case Theory, Ellen Brandner & Heike Zinsmeister (eds), 331–367. Stanford CA: CSLI.

Rhetorical relations and verb placement in the early Germanic languages A cross-linguistic study Svetlana Petrova and Michael Solf Humboldt-Universität zu Berlin

The paper investigates pragmatic principles determining clause structure in the early Germanic languages. Previous observations on discourse-related properties of V1 vs. V2 in Old High German are reconsidered on a larger scale and compared with evidence from other early Germanic languages. It is claimed that the position of the inflected verb is a device of marking coordination vs. subordination as the two basic types of discourse relations according to the model of SDRT (Asher & Lascarides 2003). Keywords: Early Germanic, verb placement, information structure, discourse relations

1.  Introduction Throughout the mass of literature exploring the principles of word order in the early Germanic languages, numerous remarks suggest that the placement of the inflected verb is sensitive to a complex set of factors pertaining to information packaging and discourse organization (Ries 1880; Behaghel 1932; Fourquet 1974; Lenerz 1984). ­Following this, Hinterhölzl & Petrova (2005) take a first attempt at describing word order variation in early Germanic in a dynamic model of discourse relations as outlined in the Segmented Discourse Relation Theory (SDRT) by Asher & Lascarides (2003). On the basis of data from Old High German (henceforth OHG), it is claimed that the position of the finite verb is a device of differentiating coordination vs. ­subordination as the two major types of rhetorical relations outlined in the framework of SDRT. This function mainly manifests itself in the opposition of verb-initial (V1) vs. verb-second (V2) as the two most common root patterns in OHG. V2 structures with ­referential material placed before the verb typically occur in contexts giving supportive, e.g., ­explanatory or descriptive information and therefore constitute discourse parts that are subordinated in text structure. By contrast, V1 and functionally equivalent ­patterns in which only a frame-setting element (a particle or an adverbial) precedes the verb are generally found in sentences carrying the main story-line of the narration and are therefore viewed to attribute to the coordinating type of linking.

 Svetlana Petrova and Michael Solf

Following these observations it may be concluded that syntactic patterns other than the widely assumed basic verb-final order started to emerge in early Germanic for ­reasons of discourse organization and rhetorical explicitness. In earlier studies, it has been already observed that verb fronting in early Germanic is related to phenomena like illocution and sentence mood. Kiparsky (1996) and Van Kemenade (1997) ­postulate residual V2 in the case of syntactic operators like interrogative phrases, ­negation elements or sentence adverbials (e.g., þa/þonne in Old English) situated in the left periphery of the clause and regularly attracting the verb to the second ­position in the sentence. ­Additionally, Eythórsson (1996: 111) assumes a phonologically empty operator yielding verb-initial structures in imperatives, while Lenerz (1984) – building upon Fourquet’s (1974) idea that verb fronting highlights the contents of the whole sentence – accounts for special stylistic usages of V1 in declaratives. In an overview on verb placement in Old English, Pintzuk (1996: 380) assigns to all these cases the overall label of some common discourse-related functions. Following this, we aim at discussing further type of data, predominantly declaratives, to provide more empirical support for the claim that verb placement serves special discourse needs in the early Germanic languages. In this respect, we are indebted to previous observations by Hopper (1979) on Old English and Leiss (2000: 84–109) on Old Norse who relate verb placement to grounding in discourse and show how this interacts with the aspectual reading of the verbs involved. Both Hopper (1979) and Leiss (2000) state that V2-clauses provide background information in Old English and Old Norse respectively. Hopper relates foregrounding to peripheral verb placement, which means both Vend and V1. He also observes a strong tendency of perfective verbs to appear in foregrounding (i.e., V1) units. Leiss draws the attention at the complementary distribution of the “historic present” and V1 in different types of sentences serving the aim of foregrounding the message. Her main claim is that V1 is a device of perfectivization in Old Norse. However, this view cannot be applied to all cases of V1 in older Germanic. As will be shown later, there are examples in which verbs in initial position retain their durative semantics, cf. (1a), (17b) and (21a) below. Therefore, we aim at deriving a common functional definition of V1 with both perfective and statal/durative predicates in early Germanic which in our view is achievable if we look at word order from the perspective of text structure and discourse analysis. The paper is organized as follows: starting from an analysis of the distribution of V1 and V2 in OHG, we compare our findings with the situation in the remaining early Germanic languages attested from nearly the same period of time and then propose a formal model of discourse-semantics which is suitable to account for the distinctions observed.

2.  The initial hypothesis: verb placement in the Old High German Tatian In order to investigate the role of information structure in the syntax of OHG, ­Hinterhölzl et al. (2005) pursue an approach that especially concentrates on the

Rhetorical relations and verb placement in early Germanic 

r­ elationship between the given/new status of discourse referents and their placement with respect to the inflected verb in the sentence. For several methodological reasons outlined in Hinterhölzl et al. (2005: 4–6), the empirical basis of the analysis is ­restricted to examples from the Tatian translation (9th century) in which the vernacular text departs from the word order of the underlying Latin original. The analysis provides significant points in favour of the interdependence between verb placement and information structure in OHG which can be best demonstrated on sentences representing the thetic vs. categorical type of judgements (Sasse 1995). By definition, categorical sentences have a bipartite structure divided into a predication base, or topic of the sentence and a comment on this topic. By contrast, thetic utterances are “monominal predications” (Sasse 1995: 4) in which no particular constituent is taken as the predication base of the utterance; rather, the entire sentence, including all participants, is asserted as a unitary whole. Therefore, thetic utterances are viewed as all-focus sentences in which no topic-comment division applies (see also Drubig 1992). Typical representatives of the thetic kind of judgements are presentational sentences introducing a new discourse referent to the context (Lambrecht 1994, pp. 127– 131, 137–146 and 177–181). In OHG, such sentences basically display two syntactic patterns. In the first one, the finite verb1 occupies the position at the beginning of the entire sentence yielding V1, see (1a). This use of V1 has been already reflected in previous studies, e.g., Lenerz (1984: 151–153) or Ramers (2005: 81) who claim that V1 in OHG is typical for utterances containing new information only. Apart from V1, a second pattern occurs in which a particle or an adverbial is allowed before the verb, see tho ‘then’, thar ‘there’ in (1b): (1) a. b.

uuas thar ouh sum uuitua was there too [a] certain widow ‘There was a widow there, too’ vidua autem quædam erat tho uuas man In hierusalem then was [a] man in Jerusalem ‘There was a man in Jerusalem’ homo erat In hierusalem

(T 201, 2)

(T 37, 23)

In both patterns, the newly added referent is placed after the finite verb, see sum uuitua ‘a widow’ in (1a) and man ‘a man’ in (1b). As for the structure of the left periphery of these sentences, there is evidence to assume that pure V1 is the preferred pattern in these contexts since particles or adverbials added against the original are also found in

.  The finite verb in both OHG and Latin, as well as in all instances from other early Germanic languages considered later in this paper, is underlined for clarity. A slash stands for end of verse or text line according to the graphical representation of the instances in the manuscripts ­respectively.

 Svetlana Petrova and Michael Solf

postverbal position thus preserving V1, see thar in (1a). From this we may conclude that V1 is strongly associated with the functional type of presentational, or thetic/allfocus sentences in the early period of German. By contrast, categorical sentences are systematically realized as V2 structures against various word order patterns occurring in the Latin original. See the second conjunct in (2) predicating on the discourse entity guot hirti ‘a good shepherd’ introduced in the previous sentence: (2)

(ih bin guot hirti. ‘I am a good shepherd’) guot hirti/ tuot sina sela furi siniu scaph good shepherd does his soul for his sheep ‘The good shepherd gives his soul for his sheep.’ bonus pastor/ animam suam dat pro ouibus suis

[T 225, 16]) (T 225, 16–17)

A general property of V2-instances like those in (2) distinguishing them from ­structures like (1b) consists in the type of phrase that occupies the pre-verbal domain. In V2structures of the categorical kind, the inflected verb singles out a discourse ­referent that is meant to provide the starting point, or the aboutness topic of the ­utterance (Reinhart 1981). This does not hold for structures like (1b) where the sentence-initial elements provide the temporal or local orientation of the situation, thus sharing ­properties of frame-setting elements. This analysis strongly suggests that the position of the finite verb in OHG is sensitive to the information structure as reflected by the status of the discourse referents in the sentence. This allows for an initial generalization concerning the placement of the finite verb in early German. In all cases considered above, the verb occupies one and the same position, namely the one at the beginning of the new-information focus ­domain, by additionally setting apart the aboutness topic from the rest of the ­utterance. This generalization is represented in (3): (3) thetic a. b. Frame categorical TOP[DRgiv]

FOC[Vfin

… DRnew …] [Vfin … DRnew …] FOC [Vfin ...] FOC

Given this conclusion, however, we are in need of an explanation for V1 in sentences containing discourse-given material as in (4): (4)

bigonda ther phariseus innan imo/ ahtonti queden began this Pharisee inside him thinking [to] speak ‘This Pharisee began to speak thinking by himself ’ Phariseus autem coepit intra se/ reputans dicere

(T 126, 4–5)

Here, the definite expression ther phariseus ‘this pharisee’ refers to a discourse-given entity which is a suitable topic candidate. Nevertheless, it occupies a position shown as typical for new referents in all-focus sentences. A further examination of the Tatian as well as a look at other OHG texts reveals that V1 with discourse-given material is

Rhetorical relations and verb placement in early Germanic 

a wide-spread phenomenon at this stage of German. Consequently, it is unjustified to restrict V1 to sentences with new information only. Rather, we have to look for the conditions leading to V1 and the subsequent postverbal realization of discourse-given subjects in root declaratives in OHG. On closer inspection, cases of V1 in OHG show a clear positional and functional distribution. First, it is well-known that they typically occur in text-initial sentences or at the beginning of new episodes (Schrodt 2004: 199). A similar fact is reported for some colloquial registers of modern German as well as for the beginning of some orally transmitted genres like jokes etc. (Lenerz 1984: 153; Önnerfors 1997: 53). For the Bavarian dialect, Simon (1998: 145) describes a similar function of V1 sentences used to open a new dialogue or to continue a dialogue after a break.2 In Tatian, which is a bilingually attested translation of a gospel harmony, episode onsets, for instance the shift to another place of reference in the source text of the New Testament, are signalled by concordance notes in the left-hand margin of the Latin column or between the Latin and the OHG text. Additionally, it is known that Carolingian manuscripts of both Latin and vernacular texts use to mark the beginnings of new text units by means of punctuation and graphical representation in order to distinguish coherent parts in written discourse (Bästlein 1991: 59 and 1991: 214–242). As for the manuscript of the Tatian, Simmler (1998: 306–307) remarks that the strategy of dividing episodes and sub-episodes through initial capital letters predominantly applies for the Latin section of the text and only rarely occurs in the OHG part. At the same time, we observe that the graphical distinction of new episodes in the Latin original correlates with the regular pre-posing of the finite verb in the OHG translation, see (5a–b): (5) a. b.

uuard thô gitân In then tagon [it] became then done in those days ‘It happened in those days’ Factum est autem In diebus illis Intfiengun sie tho the heilantes lichamon took they then theGen SaviourGen bodyAcc ‘Then they took the body of Jesus’ Acceperunt autem corpus ihesu

(T 35, 7)

(T 321, 29)

This suggests that the syntactic means of verb fronting systematically applies for marking episode boundaries in OHG as a functional equivalent of the graphical highlighting of the episode onsets in the Latin original. The strong preference for V1 at the

.  But note that Simon points at further restrictions on V1 in Bavarian. Unlike the situation in the standard language, V1 in Bavarian appears to be most common with modals, less common with auxiliaries and highly restricted with main verbs (Simon 1998: 148). Moreover, V1 is preferred in sentences highlighting the reference to an addressee. As such, they represent no declarative statement but fulfill different illocutionary functions in the domain of request (Simon 1998: 149).

 Svetlana Petrova and Michael Solf

beginnings of new episodes not only accounts for the post-verbal position of full DPs as in (4) but also for the positioning of pronominal subjects inserted against the Latin original, see sie ‘they’ in (5b). Furthermore, it has been observed that V1 regularly occurs with certain verb ­lexemes in OHG (for an overview see Schrodt 2004: 199) which according to our data classification constitute some stable classes of predicates. The most common group of predicates favouring V1 – apart from those in presentational sentences – are motion verbs (6a), verbs of saying (6b) as well as punctual, especially inchoative verbs (6c) signalling the initiation of a new state of affairs (very often this is a new physical or cognitive state of the referent): (6) a. b. c.

quam thara gotes engil came there GodGen angel ‘There appeared God’s angel’ & ecce angelus domini antlingota thô sîn muoter Inti quad responded then his mother and said ‘Then his mother responded and said’ & respondens mater eius & dixit uuard tho giheilit ther kneht in thero ziti became then healed the boy in these times ‘Then the boy was healed at this very moment’ & sanatus est puer in illa hora

(T 35, 32)

(T 30, 24)

(T 84, 7)

How can the initial position of the finite verb in these examples containing discoursegiven material be re-unified with the fact that the same structure occurs in presentational ­sentences with brand-new referents as well? One suggestion that might be uttered in this respect is that the predicates favouring V1 are intransitive unaccusatives in which the ­surface subject is an underlying object actually3. However, our data provides examples of V1 with transitive verbs as well, see (7a–b); consider that the insertion of the subject pronoun as well as of the adverbial tho in OHG does not affect the initial position of the inflected verb: (7) a. Quad her tho zi then giladoten/ ratissa (T 180, 9) said he then to the guests parableAcc ‘Then he told a parable to the guests’ Dicebat autem & ad Inuitatos/ parabolam b. furstuont siu thó in ira lihhamen/ (T 95, 14) understood she then in her body (thaz siu heil uuas fon theru suhti ‘that she was healed from her suffering’  [T 95. 15])

.  See Sasse (1995: 6) for a brief discussion on this matter.

Rhetorical relations and verb placement in early Germanic 



‘Then she realized with her body’ & sensit corpore/ quod sanata ess& a plaga

We suggest that a plausible explanation of the postverbal position of both given and new subjects in V1 sentences discussed so far can be gained if next to the ­informational status of referents, discourse relations among sentences are considered as well. From the point of view of discourse organisation, examples (4)–(7) do not act as categorical sentences providing a comment on a given referent but rather as event-reporting sentences answering implicit questions like “What happened then?/How does the story go on?” etc. This makes clear that the discourse referents contained in the instances under scrutiny are not mentioned as the starting point or the aboutness topic of the utterances but as being involved in the new state of affairs reported here. From this perspective, sentences (4)–(7) have to be viewed as all-focus sentences just like the presentational ones given in (1). Since the post-verbal position is associated with newinformation focus (see (3) above), fronting of the finite verb is used as a special strategy to highlight the entire proposition and to disable a topic-comment separation, which otherwise would have applied, especially as far as discourse-given material is contained in the sentence. Another point in favour of this account on V1 comes from the lexical meaning of the predicates involved. Motion verbs, inchoatives predicates as well as verbs of ­saying affect the main characteristics providing the deictic orientation of the narrative situation according to ‘place’, ‘time’ and ‘participant/perspective’. In discourseanalytic approaches like Brinton (1996) these characteristics are considered relevant for distinguishing episode boundaries in running discourse. Changes with respect to one of these characteristics mark the beginning of a new episode in the structure of the text. This observation fits to the fact that the shift of place and/or participant is a typical instance in which episode onset is marked by an initial capital letter in Carolingian manuscripts (see Bästlein 1991: 168 and 1991: 192). Following these considerations, we can conclude that predicates like those favouring V1 in OHG operate as inherent indicators of episode boundaries. Presentational sentences appear to be a subset of this group of sentences establishing a new situation via change of personnel.

3.  Comparison to other early Germanic languages The hypothesis about the role of verb placement as a discourse-structuring device in OHG would gain strong support if it turned to hold in other early Germanic languages as well. In the following analysis, the contexts and factors favoring V1 vs. V2 in OHG shall be reconsidered on the basis of data from other early Germanic languages. For this reason, texts from Old English (henceforth OE), Old Saxon (henceforth OS) and Old Norse (henceforth ON) shall be considered. Remarks on age,

 Svetlana Petrova and Michael Solf

size and genre of the evaluated texts will be given at the beginning of each section, respectively.4

3.1  Old English For OE, we chose the text of the Beowulf poem, the oldest epic narrative of all early Germanic literature.5 The text, comprising some 3.000 alliterative lines, is composed in the 8th century in the Anglian dialect but written down in the later half of the 10th century in Late West Saxon – the standard OE dialect at that time – though a number of original Anglian forms remain (see Lehnert 1960, vol. I, p. 43). Starting with a review of typical V1-occurrences in Beowulf, it is important to note that these are by no way rare or uncommon for this text of the early Germanic period. First of all, Beowulf tends to use V1 regularly in text-initial position as well as at the ­beginning of a new text section. New chapters (called ‘fits’) are easily detected in Beowulf since they are marked by Roman numbers in the manuscript. At the beginning of such chapters, V1 goes with all types of main-verb predicates. So in (8a) the sentence at the beginning of a fit XII describes a statal (durative) condition, whereas in (8b) the predicate at the beginning of fit XIX clearly describes a punctual event: (8) a. b.

Nolde eorla hleo […] / cwealm-cuman not wanted noblemenGen protector murderous visitorAkk cwicne forlætan (Beo 791f.) alive let go ‘The protector of the warriors did not wish to let the monster go alive’ Sigon þa| to slæpe (Beo 1252) sankPl then to sleep ‘They sank then into sleep’

Next to episode onsets, V1 in Beowulf appear to favour the same groups of verbs which were also described for OHG above. Among these, motion verbs constitute the

.  Gothic, which apart from the runic inscriptions provides us with the earliest written records of the Germanic group, leaves only little ground for any reliable conclusions about authentic word order. The basic text of the corpus, the translation of the New Testament from Greek made by Wulfila in the 4th century and attested in fragments of copies from the 5th and 6th century (see Braune & Heidermanns 2004, § E5, p. 6), shows an overwhelming identity with the word order of the parallel Greek text. If this text is indeed the source for the Gothic translation, future work could address especially sentences deviating from the syntactic structure of the original, an approach similar to the one pursued for the Tatian above. According to Fourquet (1938: 234–281), such sentences in Gothic really exist. A full sample of these would provide a basis for a subsequent analysis of verb placement in Gothic, too. .  On the possibility to acquire suitable material for syntactic analysis from Bewoulf, see ­Pintzuk (1996: 386).

Rhetorical relations and verb placement in early Germanic 

­ verwhelming part. In such instances, both context-given and context-new referents o are involved, see (9a–b). Note that in (9b) Wealhtheow, the wife of the Danish king Hrothgar, is mentioned for the first time in the narrative: (9) a. b.

Com þa| to lande lid-manna helm came then to land sailorsGen protector ‘Then the protector of the sailors approached the shore’ eode wealh-þeow forð cwen hroð-gares went Wealhtheow forth wife HrothgarGen ‘Then came Wealhtheow, Hrothgar’s wife’

(Beo 1623)

(Beo 613)

V1 is also common with verbs of saying appearing both at the beginning of a fit (10a) as well as within one (10b): (10) a. b.

Heht ða þæt heaðo-weorc to hagan biodan (Beo 2893) ordered then that battle-toilAcc to castleDat announce ‘He ordered to tell the people in the castle about this difficult fight’ Spræc/ ða ides scyldinga (Beo 1168) spoke then [the] queen [of the] DanesGenPl ‘Then the queen of the Danes spoke’

With verbs of saying, the V2 pattern as in (11) is also very common: (11) Bio-wulf maðelode bearn ecg-ðioes (Beo 1999) Beowulf spoke child EcgtheowGen ‘Beowulf spoke, the son of Ecgtheow’.

However, a typical property of this type of expression is that a discourse-given ­referent, mainly a changing interlocutor in a dialogue, is placed before the verb and an apposition follows it. Apparently, the verb lexeme in this type of structure never changes which indicates a kind of idiomatic expression standing beyond the analysis of word order variation in this functional domain. Furthermore, V1 also correlates with different main verbs sharing the property of perfective, punctual semantics which – used within a fit – denote the beginning of a new state of affairs. This pattern is especially frequent when a new or extraordinary important event is announced, e.g., a turning point in the course of the narration. See (12) which relates that Beowulf suddenly detects the weapon with which he is going to win the battle against Grendel’s mother: (12) GE-seah ða on|searwum sige-eadig bil saw then in battle triumphantAcc bladeAcc ‘In the middle of the battle he saw a triumphant blade’

(Beo 1558)

Also in striking parallelism to the picture drawn for OHG above, V1 sentences may ­contain particles or adverbials like OE þa ‘then’ placed after the verb. V1 as an indicator of a change within the narrative setting is thus preserved. Additionally, instances of

 Svetlana Petrova and Michael Solf

sentence-initial þa followed immediately by the verb co-occur as functional equivalents to V1-sentences in all contexts described so far,6 see (13): (13) a. b.

Đa com of more […] gre-/ ndel gongan Then came from moorDat […] Grendel goPastPart ‘Then from the moorland […] Grendel came’ ða| wæs swigra secg sunu/ eclafes then was more silent man son EcglafGen ‘Then this man, the son of Ecglaf, became more silent’

(Beo 710)

(Beo 980)

To sum up, the context features and predicate groups distinguished as triggers of V1 in OHG appears to be a wide-spread pattern in the Beowulf as well. In the previous ­literature, the placement of a finite verb before all arguments has already been ­associated with the purpose of highlighting “a new or surprising subject” (Stockwell 1984: 576). Due to the property of V1 sentences to carry further the discourse, we may now extend the function of this pattern to that of focussing not only the subject but the entire proposition. Turning to sentences of the categorical kind and comparing the results from OE with the preferred V2 in OHG, we encounter a basic difference between the two languages. V2 with a left-peripheral topic constituent dominating in this pragmatic domain in OHG is indeed found in part of the evidence from OE. This is the case in (14) which clearly allows for an interpretation as an identificational sentence answering the preceding question ‘Who are you?’: (14) we| synt gum-cynnes/ geata leode we are kinGen GeatsGen people ‘We are by kin of the clan of Geats’

(Beo 260)

The same kind of topic marking also occurs in parenthetic constructions providing additional information on an entity just mentioned: (15) wulfgar maþelode þæt wæs wendla leod Wulfgar spoke this was WendelsGen chieftain ‘Wulfgar spoke – this was the Wendles’ chieftain’

(Beo 348)

Due to the pragmatic status of the referents and on the basis of discourse interpretation, it can be concluded that the finite verb in structures of the type in (14) and (15) is set to distinguish the aboutness topic from the new information supplied by the rest of the sentence. Nevertheless, patterns other than V2 also appear in categorical sentences of OE. Consider the following small discourse: after his return to his home land, Beowulf

.  For a detailed investigation on the discourse functions of OE þa see Enkvist & Wårwick (1987) who describe parallel functions of sentence-initial þa followed by the verb as those claimed here for V1 in OE.

Rhetorical relations and verb placement in early Germanic 

relates his adventures with the Danes and is eager to present the gifts that he has ­obtained from them as an award for his successful fight against Grendel. So he asks to bring these gifts and as they lay in front of his counterparts, he utters the sentence quoted in (16): (16) me| ðis hilde-sceorp/ hroðgâr sealde me these ornaments of war Hrothgar gave ‘These ornaments of war were given to me by Hrothgar’

(Beo 2155)

The context of this text passage bears strong indications for the interpretation of the sentence as a categorical one. It is also clear that the definite expression ðis hildesceorp ‘these ornaments of war’ best qualifies to be the aboutness-topic of the utterance because due to the preceding context it is the expected starting point of the next sentence providing more information on these weapons. However, this constituent is not separated from the rest of the utterance by means of verb placement as demonstrated for the parallel cases in (2) from OHG and (14)–(15) from OE. Rather, the aboutness topic of the utterance shares the same syntactic domain as referents belonging to different information-structural categories, for example the familiarity topic me ‘me/to me’ as well as the focus of the sentence, the donor of the weapons Hrothgar. The examples discussed in this section provide important points concerning the interaction between verb placement and discourse structure in OE. On the one hand, the cases of V1 confirm the findings for OHG. On the other hand – in contrast to the situation in OHG – V2 turns out not to serve as a topic-marking device in categorical statements.

3.2  Old Saxon After having pointed at a crucial difference between the syntactic realization of categorical sentences in OHG and OE, we turn to the investigation of evidence from the OS period. The data is based on the most representative text of the OS corpus – the Heliand – a 9th/10th-century poetic gospel harmony comprising 5.983 alliterative lines (see Rauch 1992: 1). It has been pointed out in the previous literature that sentences in which the ­finite verb precedes all arguments in main clauses are extremely frequent in OS syntax. Rauch (1992) estimates sentences initiated by a particle followed immediately by the finite verb to be the most common pattern in OS and therefore accounts them to be “[t]he unmarked word order of the OS independent declarative sentence” (Rauch 1992: 24), followed in number by pure V1 sentences. As early as in the revealing work of Ries (1880), the kind of logical relations between sentences in context have explicitly been accounted for as factors triggering this kind of fronting of the finite verb in OS. Looking at the correlation between the finite verb form and the pragmatic features of discourse referents, we encounter a slightly different situation in OS in comparison to OHG. On the one hand, in presentational contexts, the type of clause-initial ­particles followed by the finite verb seems to be more frequent than pure V1, see (17a)

 Svetlana Petrova and Michael Solf

vs. (17b). Thus, these two patterns in presentational sentences establish a quantitative relation opposite to that in OHG: (17) a. b.

Than uuas thar ên gigamalod mann then was there an old-aged man ‘Then it was an old-aged man there’ Lag thar ên felis bio( an lay there a stone upon ‘A stone lay there upon [the entry of the tomb]’

(Hel 72)

(Hel 4075)

On the other hand, categorical sentences – for example those directly following ­presentational ones – exhibit the structure established for OHG already, i.e., they use to fill a single position before the finite verb form with the topic of the utterance, most usually in form of an anaphoric pronoun referring backwards to the entity just introduced to context, see (18): (18) a. b.

that uuas fruod gomo that was wise man ‘that was a wise man’ That uuas sô sâlig man that was so blessed man ‘This was such a blessed man’

(Hel 73)

(Hel 76)

This situation leads to some conclusions about the interaction between information structure and syntax in OS. First, it points at a higher stage of generalization of the V2-rule in OS as this structure appears in different contextual types of main sentences despite of the type of constituent or the pragmatic status of the referents involved. Second, it shows that new information in all cases follows the finite verb and therefore confirms the view of a right-peripheral focus domain in early Germanic as stated so far. This is also demonstrated by other instances bearing a brand-new referent, for example in the object position of a transitive verb, see (19): (19) Thar fundun sea ênna gôdan man there foundPl they aAcc goodAcc manAcc ‘There they found a good man’

(Hel 463)

However, OS displays some more peculiarities. There are cases showing patterns other then V1 or V2, see (20): (20) Ic is engil bium I his angel am ‘I am his angel’

(Hel 119)

The sentence provides more information on a discourse-given referent, thus it ­classifies as a categorical one, with the pronoun OS ic as the aboutness topic of the utterance. Nevertheless, the sentence shows a mixed topic-focus-domain in front of the finite

Rhetorical relations and verb placement in early Germanic 

verb, a peculiarity which reminds us of comparable instances in OE like (16) repeated here for convenience: (16) me|ðis hilde-sceorp/ hroðgâr sealde ‘These ornaments of war were given to me by Hrothgar’

(Beo 2155)

Like OE, OS does not show a tendency to distinguish the aboutness-topic from the rest of the utterance as was observed for OHG, thus confirming OS to linger on a ­continuum between the other West Germanic dialects. On the other hand, although V1 turns out to be less frequent in presentational ­contexts, it is interesting to look for further utilizations of this pattern aside from these classical cases of all-focus sentences. As a matter of fact, such instances really occur in OS exactly in the conditions under which they systematically appear in OHG and OE discussed above. Also quite similar to the situation in these languages, V2 with a preceding frame adverbial acts as an optional variant in the pragmatic domains of V1-sentences. First, we shall turn to sentences at the beginning of a new text section. There is no problem to isolate such instances in Heliand as the text is divided in chapters termed ‘fits’ (‘per vitteas’) in the Latin preface of the poem. V1 occurs with all sorts of predicates describing both states and actions, see (21a) vs. (21b): (21) a. b.

Stôd imu thô fora themu uuîhe  (Hel 3758) stood PronRefl then in front of the temple Crist uualdandeo almighty Christ ‘The almighty Christ stood in front of the temple’ Uurðun thô thea liudi umbi thea êra Cristes, / umbi becamePl then the people about the doctrines ChristGen about thiu uuord an geuuinne (Hel 3926f.) the words in quarrel ‘Then these people began quarreling about the message of these words’

Second, V1 regularly occurs in sentences containing the predicate groups distinguished as triggers of verb-initial placement in OHG above. Like in OE, the most common ­examples are provided by sentences containing motion verbs. In general, these ­occupy the initial position in the sentence despite of the pragmatic status of the referent involved or the position of the utterance in global text structure. As (22a) vs. (22b–c) show, both new and given referents are found in post-verbal position, and only (22b) is one at the beginning of a new fit, the rest of the examples signal a change of the situation within one and the same episode: (22) a. b.

Thô quam thar ôc ên uuîf gangan (Hel 503) then came there also a woman goPastPart ‘Then a woman came there, too’ Giuuitun im thô eft an Galilealand Ioseph endi wentPl PronRefl then back to Galilee Joseph and

 Svetlana Petrova and Michael Solf

c.

Maria Maria ‘Then Joseph and Mary went back to the land of Galilee’ Fôrun thea bodon o(ar all wentPl the messengers over all ‘The messengers went all over the country’

(Hel 780)

(Hel 350)

The same may be seen to apply to verbs of saying. V1 occurs both at the beginning of a fit as well as within one, see (23a–b). Sentence-initial particles or adverbials followed by the verb provide a functional equivalent of this pattern, see (23c): (23) a. b. c.

Hiet man that alla thea elilendiun man iro ordered PronIndef that all the foreign people their ôðil sôhtin (Hel 345) home country serche3PlSubjII ‘It was ordered that all these foreign people went to their native country’ Hêt imu helpen thô / uualdandeo Krist (Hel 4101) ordered him helpInf then almighty Christ ‘The almighty Christ ordered to help him’ Thô sprak thar ên gifrôdot man (Hel 208) then spoke there a wise man ‘Then a wise man spoke there’

Finally, the same structural variation holds for predicates pointing at the beginning of a new state of affairs, e.g., a physical or cognitive state of affairs of a given referent, see (24a–c): (24) a. b. c.

uuarð ald gumo / sprâca bilôsit became old man speechGen bereft ‘The old man became bereft of speech’ Uuarð Mariun thô môd an sorgun became MariaDat then heart in worries ‘Then Mary became anxious’ Thô uuarð hugi Iosepes, / is môd giuuorrid then became mind JosephGen his heart worried ‘Then, Joseph became worried’

(Hel 172f.)

(Hel 803)

(Hel 295)

Apart from these instances in which – quite similar to the other investigated old ­Germanic languages – OS makes use of verb syntax to signal progress in narration, OS itself offers a number of special cases in this direction. One of these involves the use of the V1 in explanatory parts of the narration denoting durative conditions on a discourse-given referent. In the cases analysed so far, we witnessed regular V2 in OHG and parts of the OE examples, but no V1. However, the OS data give reason to believe that V1 in explanatory parts of the narration does not occur randomly but ­applies exactly in sentences announcing an outstandingly important event or property,

Rhetorical relations and verb placement in early Germanic 

i.e., a state of affairs which is crucial for the further development of the narrative (see Ries 1880: 19 for a similar interpretation of such instances). Applying our previous account on cases like these, we may conclude that fronting of the finite verb to the beginning of the sentences is a syntactic means used to highlight the importance of the entire proposition in relation to the surrounding units of discourse. V1 and V2 form an ­interesting opposition, as will be shown by the following examples taken from the story about the nativity of John the Baptist (Luke 1): (25) a. Than uuas thar ên gigamalod mann, / then was there an old-aged man b. that uuas fruod gomo […] that was wise man c. That uuas sô sâlig man / […] that was so blessed man d. was im thoh an sorgun hugi, / that sie was him however in worries mind that they er( iuuard êgan ni môstun heir own Neg could (Hel 72–86) ‘Then there was an old-aged man, this was a wise man […] This was such a blessed man […] But they had great sorrow, for they had no child’

The categorical sentences in (25b–c) provide additional information about the discourse referent introduced in (25a), the information-structural domains of Topic and Focus are clearly distinguished by means of verb placement. In (25d), however, a characterization of the referent is given which is more important than the already provided information about his age and his wisdom. Zachariah’s and Elizabeth’s lack of a child is crucial for the further development of the story; it is a condition which is going to change and to motivate a chain of subsequent events constituting the further course of the narrative. So in this case, a special utilization of verb position is shown for OS. Whenever an important feature of a referent, or a crucial event of the story is narrated, the language switches to V1 in order to highlight the whole proposition and to set it up against other, not so important parts of the discourse. As a further peculiarity of OS, V1 appears not only at episode edges but within episodes of temporally successive events called ‘period’ according to the ancient grammatical tradition (see Kusmenko 1996: 147). Consider the following example: (26) a. b. c.

Thô uuarð thar an thene gastseli / megincraft mikil then became there in the guesthall crowd big manno gesamnod […] menGen gathered Quâmun managa / Iudeon an thene gastseli; came many JewsGen in this guesthall uuarð im thar gladmôd hugi, / blîði became they there gladhearted mind happy

 Svetlana Petrova and Michael Solf

an iro breostun: […] in their hearts d. Drôg man uuîn an flet / skîri mid scâlun […] carry PronIndef wine in room pure with bowls e. Uuas thes an lustun landes hirdi, / huat was thisGen in joy landGen shepherd what he themu uuerode mêst te uunniun gifremidi. / he theDat crowdDat most for joy did f. Hêt he thô gangen forð gêla thiornun (Hel 2733–2745) ordered he then goInf forth gay maidenAcc ‘There was a mighty crowd of men gathered together in the guest hall [...] Many people came into that guest hall; there they became glad-hearted [...]. Wine was brought to the room [...]. The herdsman of the land bethought him with joy [...]. He hight to go forth the gay maiden’

The example gives a chain of chronologically ordered events all exposing no dependency relation among each other but being equally situated on the layer of main action. Again, V1 signals that each sentence in a narrative chain like the one in (26) reports a new event and thus forms a situation on itself in which no topic-comment distinction applies.

3.3  Old Norse ON, although exhibiting a rich amount of authentic text material, starts its written records only very late, in the 12th century (Ranke & Hofmann 1988: 13–18), i.e., at a time at which in other Germanic dialects we already speak of Middle High German or Middle English respectively. Moreover, most of the prose is attested to us in copies of a later time – something we should keep in mind comparing the ON evidence against the West Germanic. Although the runic inscriptions show us a mixed picture of word order, see (27) as an example of V-last in a main sentence from Proto-Norse, ON has a very strong tendency towards V2 in all kinds of sentences, see (28). (27) ek hlewagastiR holtijaR horna tawido I Hlewagast HoltGen hornAcc made ‘I, Hlewagast, from/son of Holt, made the horn’ (28)

gud mínn þu ert híalpari mínn God my you are helper my ‘My lord, you are my helper’ Deus meus adiutor meus

(Golden horn of Gallehus)

(VienPs 17,3)

Whereas a sentence like (27) was possible centuries before the ON literal tradition started, ON abandoned this pattern in favour of a more rigid syntax, as (28) demonstrates: The Vienna Psalter shows how even an interlinear version tends to build V2-sentences, here by inserting the subject pronoun and the finite copula verb. In other words, ON shows less evidence that the position of the verb differs according to the pragmatic properties of

Rhetorical relations and verb placement in early Germanic 

sentence constituents. New discourse referents for instance are introduced in sentences showing regularly V2, see (29), which stands as typical for the Edda prose: (29) Garðarr hét maðr Garðarr was called man ‘Garðarr was the name of a man’

(Ldn 3)

Findings like these do not differ from instances of anaphoric reference: (30) Garðarr siglði umhverfis landit Garðarr sailed around land-the ‘Garðarr sailed around the land’

(Ldn 3)

Unlike in the West Germanic dialects, we may suppose, the V2-order seems to be much more fixed. As for the functional domains of V1 outlined for OHG, ON uses this verb pattern too, but shows no regular connection between the content of a verb and its position. Thus we find examples like (31–33) with a verb of motion, and (34–36) with a verb of saying: (31) oc fór hann útan þat sumar and went he out that summer ‘and he went out that summer’

(Íslb X, 10)

(32) þá fór hann á braut then went he away ‘then he went away’

(Íslb VII, 3)

(33) En et næsta sumar epter fóro þeir austan but one next summer after wentPl they eastAcc ‘and the next summer they went eastwards’

(Íslb VII, 6)

Even though there are instances of V1 (31) and equivalent V2-order after a frame ­adverbial like in OHG, OE and OS (32), the occurrence of V2 sentences with the same group of verbs (33) clearly speaks against a functional positioning of the finite verb ­according to a West Germanic pattern. The same is true for, e. g., verbs of saying: (34) oc mællte því mange í gegn and spoke thatDat many against ‘and there were many speaking against that’

(Íslb X, 9)

(35) En þá hóf hann t lo sína upp [...] oc sagþe (Íslb VII, 13) but then raised he voiceAcc his up and said ‘and then he raised his voice [...] and said’ (36) En hann sagþe cuningom Ólafe but he said kingDat OlafDat ‘and he said to king Olaf ’

(Íslb 7)

 Svetlana Petrova and Michael Solf

Here, V1 (34), sentence-initial frames followed by the verb (35) and V2 with referential constituents in preverbal position (36) may serve the same discourse function. As can be seen from the connective elements, the sentences in (31) and (34) – and this is quite often the case – are noninitial conjuncts. We often find the typical group marker oc (see Kotcheva 2000: 154) in particular to connect shorter or longer passages of V1-sentences to chains of conjuncts of the kind already encountered in the OS Heliand, see (37). (37) Þá var fj rðrinn fullr af veiðiskap, ok gáðu þeir then was fjord-the full of catch and carePl they eigi fyrir veiðum at fá heyjanna, ok dó alt kvikfé not for acres to make hay-the and died all cattle þeira um vetrinn. (Ldn 5) their in winter-the ‘There was the fjord full of fish, and because of the fishing they didn‘t care to make hay, and all their cattle died in the winter’

This phenomenon, known as ‘Narrative Inversion’ (see Sigurðson 1994) often ­applies within a passage to link closely related and chronologically successive events to consistent units of running discourse called ‘period’ after the ancient Latin syntactic tradition, see Heusler (1977, § 508, p. 173), Kusmenko (1996) and Donhauser et al. (2006). It is a reminiscence of the common Germanic discourse-sensitive properties of verb placement in ON although ON allows less freedom of word order than the remaining early Germanic languages were shown to do.

4.  Theoretical implications The foregoing analysis revealed that the placement of inflected verb in the early ­Germanic languages depends on properties of discourse organization and text structure. The aim in this part is to account for a theoretical model reflecting these properties of early Germanic syntax. There is a great variety of approaches developed to capture matters of discourse organization and text coherence. Two recent and well accepted models, the Rhetorical Structure Theory RST (Mann & Thompson 1988) and the Segmented Discourse ­Relation Theory SDRT (Asher & Lascarides 2003) share the basic assumption that ­discourse ­coherence is achieved only if each utterance makes an illocutionary contribution to ­another utterance in the context. This is achieved when discourse units establish different kinds of rhetorical relations among each other thus creating a dynamic, hierarchical structure in discourse. According to the models presented above, the rhetorical relations linking together the contents of single discourse units can be basically of the following two kinds: a. two units can display no dependency relation among each other but share the same level of discourse hierarchy thus creating a multi-nuclear relation in the terms of RST or a relation of coordination in the terms of SDRT

Rhetorical relations and verb placement in early Germanic 

b. two units can build a dependency relation creating a hierarchical structure in ­discourse, i.e., a nucleus-satellite relation due to RST or a relation of subordination due to SDRT. In order to show how verb placement participated in achieving discourse hierarchy in texts of the early Germanic tradition, we chose the model of SDRT and provided some additional characteristics. Although the inventory of individual discourse ­relations is still under discussion, there is overwhelming agreement on the basic features distinguishing coordination vs. subordination as the two basic types of linking. Both are associated with prototypical rhetorical relations displaying some complementary features (Asher & Vieu 2005). Coordination, which is prototypical for units situated on an equal level of discourse hierarchy, is typically represented in the relation of narration. Narration is established when e.g., two discourse units (α, β) display a temporal relation of succession and β continues the narrative sequence in discourse. By contrast, subordination is typically represented in elaboration, i.e., when a unit β provides more detail on another unit α situated on a higher level of discourse hierarchy. In this case, the two events (α, β) temporally overlap. Further, the rhetorical relation of continuation applies when two or more subsequent units β and γ are equally situated on a lower level of dependency with respect to a higher unit α so that both β and γ represent elaboration on α. Looking at the distinctive features of coordination vs. subordination in SDRT, we discover a number of parallels between them and the discourse properties of the word order patterns discussed in the foregoing data analysis. Turning to the distribution of the V2-pattern in OHG first, we recall the fact that it appears in sentences providing explanatory or supportive information like descriptions, characterizations, motivation etc. on a referent or action previously mentioned in the discourse. The events described in such sentences temporally overlap with the state of affairs of the governing situation. Thus, V2 in OHG indicates relations of subordination, i.e., elaboration and continuation in discourse. By contrast, V1 in OHG never occurs in elaborative passages. In its first canonical domain – that of presentational or text-opening sentences, V1 establishes the basis for subsequent elaboration, whereas in its second domain – episode onsets within the text – it signals that a previous sequence of elaboration or continuation is suspended and discourse returns to the level of main action. The same function is also observed in patterns in which the verb is preceded only by a frame adverbial. Our analysis on the relation between this pattern and pure V1 in OHG gives reason to believe that the former pattern is an innovation in the system of OHG which has major effects on the consolidation of the V2-phenomenon in root clauses of modern German while V1 was marginalized in certain domain of colloquial style. Comparing this picture with the one in the remaining early Germanic languages, we have to observe that elaborative parts display patterns in which the preverbal domain is not restricted to the aboutness topic of the utterance as was shown for OHG. However, in each of the languages examined, V1 was detected in different functional domains providing progress in narration. Like in OHG, the pattern displaying a

 Svetlana Petrova and Michael Solf

sentence-initial adverbial or particle also occurs in this function as an equivalent of V1. However, V1 sentences may also contain frame adverbials or a particle in postverbal position. This indicates that V1 is a genuine Germanic pattern and a common formal correlate of coordination in discourse in all early Germanic languages. In this respect, the results from ON are especially suggestive. Although ON ­displays a much later stage of language development towards a generalized V2, we nevertheless encounter domains in which V1 serves to carry further the discourse by providing chronologically sequenced units failing to establish a hierarchical order among them. In these cases known as ‘Narrative Inversion’ V1 still displays typical properties of ­coordination in discourse by establishing the level of main action in text structure.

5.  Conclusions The present study investigates some discourse-related properties of verb placement in the early Germanic languages. It reconsiders a number of already known ideas about the syntax of some individual Germanic languages by providing a larger empirical basis and a more fine-grained classification of the discourse-related properties of ­particular word order patterns as well as the conditions for their use. Above all, the pragmatic value of the word order patterns was seen from the comparative perspective in order to show that we have to deal with no isolated phenomena but with properties of a common Germanic heritage. The most important conclusion from the empirical part of the investigation is that apart from OHG, other early Germanic languages also provide evidence for the claim that verb placement plays a role in text-structuring and discourse organization. In West Germanic languages other than OHG, instances of V1 as a main device of discourse segmentation show up more or less regularly in the functional domains outlined for OHG, though with a slightly different distribution and frequency than in OHG. A crucial domain of difference in the early Germanic situation was discovered in the field of supportive text parts providing more information on a preceding referent or situation. Here, V2 used in OHG but broadly generalized in OS seems to co-occur with verb-final structures in OE. These differences in the early Germanic situation could be made responsible for the development of different word order patterns in the modern systems of these languages. In order to explain the role of verb placement in early Germanic, we invoke the distinction between coordinating and subordinating discourse relations as developed in Asher & Lascarides (2003) and claim that at a certain stage in the history of the Germanic languages, the position of the verb was a means for distinguishing the type of rhetorical relation the sentence implies with respect to the previous context. We outlined the main characteristics of coordination and subordination in the framework of SDRT and pointed to the parallels which special word order types display with ­regard to these types of discourse linking. In the overall comparison between the

Rhetorical relations and verb placement in early Germanic 

languages investigated, we discovered that the V1 pattern is a common formal correlate of coordination in discourse while relations of the subordinating kind are realized by different syntactic means in the languages of the early Germanic tradition.

Acknowledgement The present paper is an extended and revised version of a talk held at the 28th annual meeting of the German Linguistic Society (DGfS) on February 24th, 2006 in Bielefeld/Germany, within the program of Workshop 08 “‘Subordination’ vs. ‘coordination’ in sentence and text from a cross-linguistic perspective”. We thank all participants of the workshop for questions and discussions as well as the anonymous reviewers for comments on an earlier version of this paper.

Primary texts [Beo] Beowulf. Reproduced in facsimile from the unique manuscript British Museum Ms. Cotton Vitellius A. XV. With a Transliteration and notes by Julius Zupitza. Second edition containing a new reproduction of the manuscript with an introductory note by Norman Davis. Published for the Early Text Society. London: Oxford University Press. 1959. [Hel] Heliand und Genesis. Hg. v. Otto Behaghel. 10. überarb. Aufl. v. Burkhard Taeger. Tübingen: Niemeyer. 1996. [Íslb] Íslendingabók. Hg. Wolfgang Golther. 1923. Ares Isländerbuch. 2., neu bearb. Aufl. Halle: Niemeyer. [Ldn] Landnámabók Ìslands. Ug. Finnur Jónsson. København: Thiele. 1925. [T] Die lateinisch-althochdeutsche Tatianbilingue Stiftsbibliothek St. Gallen Cod. 56. Hg. von Achim Masser, Göttingen: Vandenhoek & Ruprecht. 1994. (Studien zum Althochdeutschen, Bd. 25). [VienPs] Vienna Psalter Der Wiener Psalter. Hg. Heiko Uecker. Cod. Vind. 2713. Kopenhagen: Reitzel. 1980.

References Asher, Nicholas & Lascarides, Alex. 2003. Logics of Conversation. Cambridge: CUP. Asher, Nicholas & Vieu, Laure. 2005. Subordinating and coordinating discourse relations. Lingua 115(4): 591–610. Bästlein, Ulf Christian. 1991. Gliederungsinitialen in frühmittelalterlichen Epenhandschriften. Studie zur Problematik ihres Auftretens, ihrer Entwicklung und Funktion in lateinischen und volkssprachlichen Texten der Karolinger- und Ottonenzeit [Europäische Hochschulschriften. Reihe I. Deutsche Sprache und Literatur, 1167]. Frankfurt: Peter Lang. Behaghel, Otto. 1932. Deutsche Syntax. Eine geschichtliche Darstellung. Bd. IV. Wortstellung. ­Periodenbau. Heidelberg: Carl Winter.

 Svetlana Petrova and Michael Solf Braune, Wilhelm & Heidermanns, Frank. 2004. Gotische Grammatik. Mit Lesestücken und Wörterverzeichnis. 20 edn, [Sammlung kurzer Grammatiken germanischer Dialekte. A. Hauptreihe Nr. 1]. Tübingen: Max Niemeyer. Brinton, Laurel J. 1996. Pragmatic Markers in English. Grammaticalization and Discourse ­Function. Berlin: Mouton de Gruyter. Donhauser, Karin, Solf, Michael & Zeige, Lars Erik. 2006. Informationsstruktur und Diskursrelationen im Vergleich Althochdeutsch – Altisländisch. In Grenzgänger. Festschrift zum 65. Geburtstag von Jurij Kusmenko [Berliner Beiträge zur Skandinavistik 9], Antje Hornscheidt et al. (eds), 73–90. Berlin: Nordeuropa-Institut. Drubig, Bernhard. 1992. Zur Frage der grammatischen Repräsentation thetischer und kategorischer Sätze. In Informationsstruktur und Grammatik [Linguistische Berichte. Sonderheft 4], Joachim Jacobs (ed.), 142–195. Opladen: Westdeutscher Verlag. Enkvist, Nils Erik & Wårvik, Brita. 1987. Old English þa, Temporal Chains, and Narrative ­Structure. In Papers from the 7th International Conference on Historical Linguistics, Anna Giacalone-Ramat, Onofrio Carruba and Giuliano Bernini (eds), 221–237. Amsterdam: John Benjamins. Eythórsson, Thórhallur. 1996. Functional categories, cliticization, and verb movement in the early Germanic languages. In Studies in Comparative Germanic Syntax, Höskuldur ­Thráinsson, Samuel David Epstein & Steve Peter (eds), 109–139. Dordrecht: Kluwer. Fourquet, Jean. 1938. L’ordre des éléments de la phrase en germanique ancien. Étude de syntaxe de position. Paris: Les Belles Lettres. Fourquet, Jean. 1974. Genetische Betrachtungen über den deutschen Satzbau. In Studien zur deutschen Literatur und Sprache des Mittelalters. Festschrift für Hugo Moser zum 65. Geburtstag, Werner Besch, Günther Jungbluth, Gerhard Meissburger & Eberhard Nellmann (eds), 314–323. Berlin: Erich Schmidt. Heusler, Andreas. 1977. Altisländisches Elementarbuch. 7 edn. Heidelberg: Carl Winter. Hinterhölzl, Roland & Petrova, Svetlana. 2005. Rhetorical relations and verb placement in the early Germanic languages: Evidence from the Old High German Tatian translation (9th century). In Salience in Discourse. Multidisciplinary Approaches to Discourse, Manfred Stede et al. (eds), 71–79. Münster: Stichting / Nodus. Hinterhölzl, Roland, Petrova, Svetlana & Solf, Michael. 2005. Diskurspragmatische Faktoren für Topikalität und Verbstellung in der ahd. Tatian-Übersetzung (9. Jh.). Interdisciplinary Studies on Information Structure 3: 143–182. Hopper, Paul. 1979. Aspect and foregrounding in discourse. In Syntax and Semantics, Talmy Givón (ed.), 213–241. New York NY: Academic Press. Kemenade, Ans van. 1997. V2 and embedded topicalization in Old and Middle English. In ­Parameters of morpho-syntactic change, Ans van Kemenade & Nigel Vincent (eds), 326–352. Cambridge: CUP. Kiparsky, Paul. 1996. The shift to head-initial VP in Germanic. In Studies in Comparative ­Germanic Syntax, Höskuldur Thráinsson, Samuel David Epstein & Steve Peter (eds), ­140–179. Dordrecht: Kluwer. Kotcheva, Kristina. 2000. Om konjunktionerna og och en i isländskan. In Kors och tvärs i nordistiken. Föredrag från det 4:e studentkollokviet i nordisk språkvetenskap [Kleine Schriften des Nordeuropa-Instituts 20], Jurij Kusmenko & Sven Lange (eds), 24–35. Berlin: Nordeuropa-Institut der Humboldt-Universität. Kusmenko, Jurij. 1996. Die Periode – Syntaktische Haupteinheit der altisländischen Prosa. In Berkovs bók. To Honour of Prof. V. P. Berkov, Jurij Kusmenko & Lilja Popowa (eds), 145–160. Moscow: Impeto.

Rhetorical relations and verb placement in early Germanic  Lambrecht, Knud. 1994. Information Structure and Sentence Form. Topic, Focus and the Mental Representations of Discourse Referents. Cambridge: CUP. Lehnert, Martin. 1960. Poetry and Prose of the Anglo-Saxons, Vol. I: Texts. 2nd, revised edn, Halle (Saale): Max Niemeyer. Leiss, Elisabeth. 2000. Artikel und Aspekt. Die grammatischen Muster von Definitheit [Studia Linguistica Germanica 55]. Berlin: Walter de Gruyter. Lenerz, Jürgen. 1984. Syntaktischer Wandel und Grammatiktheorie. Eine Untersuchung an Beispielen aus der Sprachgeschichte des Deutschen [Linguistische Arbeiten 141]. Tübingen: Max Niemeyer. Önnerfors, Olaf. 1997. Verb-erst-Deklarativsätze. Grammatik und Pragmatik. Stockholm: Almquist & Wiskell International. Mann, William C. & Thompson, Sandra A. 1988. Rhetorical structure theory: Toward a functional theory of text organization. Text. An Interdisciplinary Journal for the Study of Discourse 8: 243–281. Pintzuk, Susan. 1996. Clitization in Old English. In Approaching Second. Second Position Clitics and Related Phenomena, Aaron L. Halpern & Arnold M. Zwicky (eds), 375–409. California CA: CSLI. Ramers, Karl Heinz. 2005. Verbstellung im Althochdeutschen. Zeitschrift für germanistische Linguistik 33: 78–91. Ranke, Friedrich & Hofmann, Dietrich. 1988. Altnordisches Elementarbuch. 5 edn. Berlin: de Gruyter. Rauch, Irmengard. 1992. The Old Saxon Language. New York NY: Peter Lang. Reinhart, Tanya. 1981. Pragmatics and linguistics: An analysis of sentence topics. Philosophica 27: 53–94. Ries, John. 1880. Die Stellung von Subject und Prädicatsverbum im Hêliand. Straßburg: Trübner. Sasse, Hans-Jürgen. 1995. “Theticity” and VS order: A case study. Sprachtypologie und Universalienforschung 48(1–2): 3–31. Schrodt, Richard. 2004. Althochdeutsche Grammatik II. Syntax [Sammlung kurzer Grammatiken germanischer Dialekte. A. Hauptreihe, Nr. 5/2]. Tübingen: Max Niemeyer. Sigurðson, Halldór Ármann. 1994. Um frásagnarumröðun og grundvallarorðaröð í forníslensku. Reykjavík: Málvísindastofnun Háskóla ĺslands. Simmler, Franz. 1998. Makrostrukturen in der lateinisch-althochdeutschen Tatianbilingue. In Deutsche Grammatik. Thema in Variationen. Festschrift für Hans-Werner Eroms zum 60. Geburtstag, Karin Donhauser & Ludwig M. Eichinger (eds), 299–335. Heidelberg: Carl Winter. Simon, Horst J. 1998. “KinnanS Eahna fei heid gfrein.” – Über einen Typ von Verb-ErstAussagesätzen im Bairischen. In Deutsche Grammatik. Thema in Variationen. Festschrift für Hans-Werner Eroms zum 60. Geburtstag, Karin Donhauser & Ludwig M. Eichinger (eds), 137–153. Heidelberg: Carl Winter. Stockwell, Robert P. 1984. On the history of the verb-second rule in English. In Historical ­Syntax, Jacek Fisiak (ed.), 575–592. Berlin: Mouton de Gruyter.

Index of subjects

A Across the Board Movement (ATB)  7, 283–4, 292, 295 adjunct  13, 22, 60, 65–7, 75, 77, 81, 115–7, 126–9, 144, 210–1, 223, 226 adjunction  14, 22, 157, 223, 292–3 adposition  59, 60, 64–67 afterthought (AT)  23, 117–22 anaphoric shift  164–5, 169, 177, 179 and  3, 6–10, 15–9, 37, 45, 53, 62–5, 69–77, 101, 108, 110, 117–9, 129–30, 144, 188, 275 ATB see Across the Board Movement attachment point  126, 139, 149, 151–2 B Background  17, 24, 46, 123, 248–53 Background  210–3 backgrounding  19, 22, 108, 122, 123, 127–9 binding  4 cataphoric  274–6 of pronouns  293 quantifier  193 variable  197, 204, 207 C Centering Theory  155 clause  115–129, 163, 167–81 adverbial  104–10, 153, 202 complement  138, 172 coordinate  12, 63, 114 dependent vs. independent  148, 187–214 main  4, 9, 12, 35, 42, 72–5, 162–7, 170–4, 182 superordinate  2, 109

see relative clause, see subordinate clause, see verb second clause clause combining (linkage, linking)  1–7, 11–24, 89–111, 138, 187–9, 207–8, 214 asyndetic vs. syndetic  6, 15, 60 canonical (prototypical) vs. non-canonical  187–209 hypotactic  15 paratactic  5, 7, 14–5, 21–2 clause linkage shift (in translation)  96–97, 102–13, 152 see sentence splitting coherence  16–17, 22, 33, 39–40, 136, 152–7, 169, 173, 178, 225, 249, 266, 272 Common Integrator  10, 69–70, 120 communicative weight  2, 20, 140, 157 conjoining coordinate vs. subordinate  6–7 see conjunction, see coordination conjunct  10, 62, 115–22, 129–131, 293 conjunction  8, 12, 59, 63–4, 67, 71, 118, 120, 26–4 conditional  24, 263–6 coordinating  60, 62, 65, 93, 293 logical  5–7, 14 subordinating  63–5, 68, 241, 245–8, 251–2, 258 verb phrase (VP) see coordination connection  58

asymmetric(al) vs. symmetric(al)  61 discourse  73–76 hierarchical vs. non-hierarchical  58–69, 70, 81 semantic  68–73 syntactic  61–8 see relation connective  59–60, 62–8, 71–72, 78–81, 117–9, 127–30, 145–146 Consequentiality  121, 140–2, 151, 156 constraint ranking  208 continuum  7, 341 clause-linking  14, 90, 108–12 semanto-syntactic  208 Continuum Approach  110, 112 conversational analysis  78, 80 coordinate construction  6–14, 120 ellipsis  322 relationship  188–9 structure  290, 295, 301 Coordinate Structure Constraint (CSC)  7, 9 coordination  24, 59, 60–2, 68–74, 77–82, 115–22, 129–31, 281–6 and-  6–7, 17–9, 24, 130 asymmetric (AC)  9, 24–5, 274, 281–301 asyndetic (implicit) vs. syndetic (explicit)  93, 101 canonical (standard, prototypical) vs. non-canonical  7–10 clause (clausal)  7–11, 15, 21, 24, 101–2, 112 multiple  6 og- (Norwegian)  117–9, 144 parenthetical  7

  Index of subjects phrasal  6 pseudo-coordination  7, 274 symmetric  282, 285, 292 unbalanced  7, 62 verb phrase (VP)  6–9, 101–11, 164, 285–6 see conjunction, see SLFC coordinator  62–3, 65, 70–3 corpus  96–102,143,156, 246, 251, 253 parallel corpora  3, 21, 98, 136, 143 156 cross-linguistic  89–92, 112–3, 131, 161–2, 172–5, 182 CSC see Coordinate Structure Constraint D decision hierarchy  161, 176, 179–81 deictic frame  165 Dependence  187–214 dependency  1–11, 95–7, 100–112, 138, 141, 145, 157 dependency cline  96 disjunction conjunctive  24, 255, 266–70 free-choice reading of  271 or-  24 discourse constituent  78, 252 discourse organization  22, 89, 90, 135–6, 140, 152–3, 157 discourse relation  61, 68, 73–4, 77–82, 120, 123, 135, 217–8, 227– 30, 236, 248, 249–51 asymmetric  77, 140, 329 continuative vs. discontiuative see relative clause coordinating vs. subordinating  16, 23, 79, 139–50, 158, 228, 236, 241, 248–9, 252, 348 non-veridical  260 multi-level  21 multinuclear  17, 18, 34, 39, 45, 47, 54 nucleus-satellite  76, 78 symmetric  140, 156 see Background, see Consequentiality,

see Elaboration, see Occasion discourse representation  16 multi-level  21, 48–55 Discourse Representation Theory (DRT)  16 discourse segmentation  24, 51, 156, 255, 259–62, 348 discourse structure  59–60, 73–82, 115, 120–30, 136, 227, 230, 241–2, 247–52 discourse topic  16, 217–18, 225–7, 230–6 discourse unit  60, 74, 80 disjunction  24, 255, 257, 262, 263, 266–74, 277 free-choice reading of  270–1 downgrading  116, 119, 129, 131, 161, 164–5, 169–74, 183 Dutch  161, 162, 164, 170, 174, 182 E Elaboration  16–7, 36–37, 48–9, 54, 79, 125, 139, 149, 151, 228  embedding  11, 15, 54, 65– 8, 74–9, 109–12 English  130–1, 161–82 epistemic minimal unit  272 event  116–7, 120–30, 163–7, 170–83 landmark vs. trajectory  71 one-event interpretation  285–6 event integration  173 event selection  178–82 event subordination  24, 288–90, 293–7, 300–1 see Occasion explicitness  106, 109, 246, 330 external argument  63, 64, 67, 296 F focality  80 focus  120–6, 130, 162, 174, 182 of attention  49, 54 forward deletion  283–4, 295 free dass-clause  23, 187, 199– 202, 207, 210–11, 214

French  3, 21–3, 63, 89–113, 146, 152, 161–77, 182, 217, 219, 226, 232–35, 241, 243, 252 fusing semantics  281–95, 301 G gapping  8, 283, 321–5 genre  49, 52, 98, 154, 336 German  115–9, 126–31, 161–82 government  4, 63–8 gradient (category, phenomenon)  21, 62, 89–90, 108 H Hauptstruktur  116 see main structure Head-driven Phrase Structure Grammar  13, 23, 139, 187–9, 208–11, 214 hierarchy  75, 81, 121, 126, 161, 178–9 HPSG see Head-driven Phrase Structure Grammar hypotaxis  1–2, 7–15, 20, 77, 95, 111, 140, 157, 187, 198 I incrementality  95–6, 112, 152, 157 information density  154–5 information extraction (in translation) to the left  119, 122 to the right  124, 125, 128 information packaging  80, 89–92, 152, 155 information structure  11, 52, 124, 130, 138, 141, 162–73, 178–83, 307, 325 discourse  81, 89–91 integration  165–6, 173, 182 internal argument  63–64, 67 intonation  72, 80 J juxtaposition  5, 15 60, 95–6, 309 see clause combining L landmark  70–1, 74–6 Latin  63, 331–4, 341, 346

Index of subjects  linear sequence  63–7, 78 Linguistic Discourse Model  78 M macroplanning  174–181 main clause phenomenon  25 main clause word order  25 main structure  116, 139, 140, 163 movement island  223 multi-level annotation (MLA)  50–56 N narrative  117, 120–1, 124–6, 130, 161–82 Nebenstruktur  116 see side structure non-integrated constituent  14, 219, 224, 230, 235 see orphan non-monotonic entailment  297 NPI (Negative Polarity Item)licensing  274–5 nuclearity  17, 21, 33–56 nucleus  36–9, 44, 47, 123 O Occasion  24, 66, 286–9, 301–2 OE see Old English OHG see Old High German Old English  330, 335–6 old (older, early) Germanic (languages)  3, 25, 329–49 Old High German  25, 329–48 Old Indic  3, 11, 25, 307, 308, 310, 313, 314, 317, 319, 323, 325 Old Norse  330, 335 Old Saxon  321, 335, 339–48 orphan  19, 23, 187–8, 211–3, 218, 223–4, 227, 230, 236 orthographic sentence  4, 7 OS see Old Saxon otherwise  272–3 P parataxis  1, 7, 14–5, 93, 157, 187–8, 198 see coordination, see juxtaposition passive  163–4, 166, 169–71, 174, 178

Penn Discourse Tree Bank  50 PI see pseudo-imperative PIDO (Principles of Incremental Discourse Organization)  152, 157 Potsdam Commentary Corpus  40, 51 procedural (operational) meaning  10, 120 prominence  2, 12, 20, 23, 39, 44–5, 148, 155, 157, 173, 323 prosody  11, 25, 80, 199, 219 pseudo-imperative (PI)  24, 255–8, 260 punctuation  3–5, 42, 72, 93, 188, 224, 333 colon  5, 91 comma  4–6, 15, 63, 72, 117–8, 124 full stop (period)  4–7, 15, 19, 63, 72, 93, 118, 122–4, 155, 157, 224 semicolon  5, 93, 155 see juxtaposition, see orthographic sentence Q quaestio  116, 164 approach  18–9, 139, 150 model  17, 22, 116 Question under Discussion (QuD)  10, 17 R RC see relative clause referential linking  67–8 referential structure  148, 150 relation adjunct  60 asymmetric(al)  2, 297 causal  19, 45, 165–9, 173–4 coherence  34–5, 38–9, 45, 49, 54, 60, 73 conjunctive  40, 52–4 enablement  121 rhetorical  3, 16, 20, 25, 34, 80, 329, 346–8 semantic  6, 54, 77, 245 symmetric(al)  6, 16–17, 77 temporal  6, 16–17, 77 see connection, see discourse relation

relative clause (RC)  14, 60, 68, 106, 113, 117, 126, 128 appositive  14–5, 22, 135, 142, 156 clause-related (‘Satzrelativsatz’)  143–8 continuative vs. discontinuative  22, 141–2, 148, 152, 155–6 non-restrictive  22, 126–9, 135–8, 143, 148, 150–6 restrictive  12, 14, 198, 206 sentential  13 verb second  187, 197, 202, 214 wh-  23, 187, 195, 197–207, 210–4 ‘weiterführender Relativsatz’  138–9, 156 see weshalb, see wobei Relevance Theory  8, 19, 117, 120 RFC see Right Frontier Constraint Rhetorical Structure Theory (RST)  17–21, 33–56, 76–8, 89, 140, 248, 346–47 right dislocation (RD)  217–22, 235 Right Frontier Constraint  16, 24, 38, 79, 126, 140, 250 RST see Rhetorical Structure Theory RST Treebank  39, 42, 51 S salience  2, 5, 10, 20–1, 33, 35, 40–7, 51–6, 154–5, 313 satellite  36–8, 44, 76–7, 123 scope of negation  193, 198, 201, 207, 210, 213, 293 Segmented Discourse Representation Theory (SDRT)  2, 15–9, 22–5, 38, 50, 54, 79, 89, 139–40, 150, 156, 248–9, 329, 346–8 sentence  4–5 categorical vs. thetic  331–332 all-focus  196–200, 331–332 complex  187–88

  Index of subjects compound  4, 68 identificational  338 presentational  331–332, 335 sentence boundary  4, 68, 81, 136, 205, 222–30 sentence length  94 sentence splitting  115–8, 124, 129 servitude  96, 108 side structure  17–18, 116 SIP see Strategy of Informational Parsimony SLFC (Subject-Lackingin-Fronted-structure Coordination)  24–5, 281–302 state predicate  117, 120–3, 129–30 Strategy of Incremental Parsimony (SIP)  152, 154, 157 Strong Nuclearity Hypothesis  38 stylistic preferences  89, 92 subject gap  8, 24, 281–3, 295, 298–301 subject selection  176–82 subjunction (subordinator)  12, 13, 63–73, 137 subordinate clause  2, 9–15, 25, 42, 63–4, 73, 75, 108, 109, 140–1, 161, 167, 169, 220–2, 241–5, 250, 253 canonical vs. non-canonical  12–23, 187–209

detached  14, 23, 195, 236 embedded  15 integrated vs. non-integrated  14, 197–209 see free dass-clause, see relative clause, see subordination subordination  59–62, 68–9, 73, 76–81, 115–6, 161–83 clause (clausal)  157, 241, 251 modal  172 semantic  76 syntactic  60–1, 67, 81, 115, 130, 140–1, 144, 149, 152, 155–7 see event, see subordinate clause subordinator see subjunction syntactic complexity  91, 94, 115 syntactic structure  62, 67, 74, 77– 81, 98, 128 T temporal sequence  70, 121 temporal shift  148, 164, 166–182 text comprehensibility  153, 157 text type  47–9, 95, 150, 152 thematic development  43, 48–9, 52 thematic role  75–7, 296, 298 thematic structure  54, 56, 90 topic  116, 161–8, 174, 177, 180–1 aboutness  332–347

topic-comment division (separation)  331, 335 topic-focus-domain  340 topic shift  49, 54,155, 162, 234 topic time  148,  163, 165, 181 trajector  70–6 translational patterns  117 translational upgrading  141, 148, 150, 152, 156 U und  61, 130, 292 und zwar  204 upgrading (translational)  22, 129, 135–157 utterance  4, 80, 155, 161, 169, 178, 339 see sentence V verb accentuation  307 verb placement (position)  25, 210, 329–1, 335–6, 339, 343, 347–8 verb-second (V2) clause  191–5, 202–5 dependent  23, 187, 202–6, 209–10, 214 relative  197 weil-  23, 187, 191, 214 W weshalb (German)  68, 137, 144 wobei (German)  137, 144 word order asymmetry  284–5 word order constraint  176–83

Index of names

A Aarts, Bas  2, 12, 19, 114 Abraham, Werner  11, 223, 236, 309 Altmann, Hans  191, 217–22 Asher, Nicholas  2, 16–24, 38, 50, 59, 77–81, 89, 126, 139–40, 149–50, 156, 225–9, 247–9, 260, 297, 345–8 Auer, Peter  203, 205, 218, 234 Austin, Jennifer R.  13 van der Auwera, Johan  187, 189, 207, 255, 257, 261 B Bader, Markus  322 Ballard, Michel  104–6 Banfield, Ann  211 Bästlein, Ulf Christian  333, 335 Bateman, John  37, 39, 45, 50, 52 Bátori, István  5 Beaman, Karen  112 Behaghel, Otto  329 Behrens, Bergljot  24, 116, 135 Bierwisch, Manfred  14, 60, 75 Blakemore, Diane  5, 7, 8, 10, 19, 70, 117, 120, 122 Blühdorn, Hardarik  16, 21, 25, 44, 60, 68–70, 193, 214, 230 Bolinger, Dwight  256, 258 Bosch, Peter  225, 234 Brandt, Margareta  20, 138, 140, 196, 198 Brants, Sabine  51 Breindl, Eva  60, 62, 70 Brinton, Laurel J.  335 Büring, Daniel  4, 11, 80–81, 225, 292 Butulussi, Eleni  205

C Carlson, Gregory Norman  39, 42, 47, 51 Carston, Robyn  5, 7–10, 19, 70, 117, 120, 122, 125 Chafe, Wallace L.  80, 95 Chuquet, Hélène  92, 93, 106 Consten, Manfred  222 Cosme, Christelle  15, 20–3, 97–9, 152 Cristea, Dan  38 Cristofaro, Sonia  11–2, 20, 59–62, 76, 90, 108–12, 183 Crysman, Berthold  5, 9 Culicover, Peter W.  222, 264–5, 274–6 D Danlos, Laurence  38, 243 de Swart, Henriette  44 Dik, Simon  62 Dipper, Stephanie  51, 52 Doherty, Monika  3, 90, 91, 92, 152, 157 Donhauser, Karin  346 Drubig, Bernhard  309, 331 Dunkel, George  91, 308 E Eckhardt, Regine  289 Eisenberg, Peter  62, 64 Elhadad, Michael  45 F Fabb, Nigel  211 Fabricius-Hansen, Cathrine  13, 18, 20, 59, 91–2, 95, 98, 115–6, 119, 123–4, 130–1, 138, 141, 144, 152, 157, 187, 189, 191, 207, 214 Fourquet, Jean  329, 336 Frank, Annette  282 Franke, Michael  24, 25, 257–8

Frey, Werner  164, 178, 209, 223, 225, 294 Fries, Norbert  37, 299 G Geurts, Bart  271, 272 Gibson, Edward  40, 49–50 Gilquin, Gaëtanelle  98 Ginzburg, Jonathan  189, 207, 212 Givón, Talmy  109–11 Golebiowski, Zofia  44 Gómez-Txurruka, Isabel  269 Granger, Sylviane  3, 89, 97 Greenbaum, Sidney  28 Grosz, Barbara  39, 45, 52, 54, 155 Grote, Brigitte  35, 38 Gussenhoven, Carlos  309 H Haegeman, Liliane  211, 218, 223, 227 Haiman, John  4, 187, 189, 191 Halliday, Michael A.K.  50, 60, 93 Hannay, Mike  93–5 Hartmann, Dietrich  2, 20, 310 Hartmann, Katharina  283, 292, 321–3 Haspelmath, Martin  5–7, 10, 59, 62–3 Hasselgård, Hilde  3, 90, 97–98 Heintze, Silvan  53 Hetland, Jorunn  2, 20, 310 Hettrich, Heinrich  308–10 Heusler, Andreas  346 Hinterhölzl, Roland  235, 328, 330–1 Hoarau, Lucie  105 Hobbs, Jerry  39, 286, 287 Hoffmann, Karl  308, 318 Hofmann, Ludger  223–4, 230, 234

  Index of names Höhle, Tilman N.  281–285, 292, 301 Holler, Anke  13–4, 23, 24–5, 59, 138, 139–43, 146, 148, 154, 156, 189, 195, 210, 212, 214, 230 Hopper, Paul  60, 64, 73, 183 Huddleston, Rodney  4–5, 8, 13 J Jackendoff, Ray  264–5, 274–6 Jacobs, Joachim  81, 313 Jasinskaja, Ekaterina  10 Johannessen, Janne Bondi  7–9, 12, 59, 62 Johansson, Stig  3, 97–8, 111 K Kamp, Hans  16, 241, 271 Karhiaho, Izabela  5 Kathol, Andreas  189–90, 207 Kavalova, Yordanka  14, 216 Kehler, Andrew  9, 24, 39, 57, 225, 286–287 Keller, Rudi  191, 193, 194 Kiparsky, Paul  330 Klein, Jared  309, 313–4, 316, 325 Klein, Wolfgang  17, 19, 80–1, 116, 139, 150, 162–3, 177–8 Knott, Alistair  49, 54, 60 König, Ekkehard  189, 207 Kotcheva, Kristina  346 Kratzer, Angelika  265, 278, 281, 294, 295 Krifka, Manfred  11, 265 Kusmenko, Jurij  343, 346 L Lakoff, Robin  275 Lambrecht, Knud  178, 222, 232–5, 331 Lang, Ewald  5–7, 10–3, 69–70, 120, 158, 313, 322 Langacker, Ronald W.  11, 74–6 Larson, Richard  291 Lascarides, Alex  2, 16, 38, 50, 139–40, 149–50, 227–9, 237, 248–9, 260, 297, 329, 346, 348 Lawler, John M.  275 Lefèvre, Michel  13, 58

Lehmann, Christian  1, 2–6, 9–15, 90, 110, 152, 157, 187, 189, 191, 207 Lehnert, Martin  336 Leiss, Elisabeth  330 Lenerz, Jürgen  329–31, 333 Levison, Stephen C.  7–9 Lewis, David  271 Lobeck, Anne  299 Lohnstein, H.  68, 75, 206 Longacre, Robert  11, 59, 189 Lötscher, Andreas  48 Lühr, Rosemarie  11, 25, 309, 314 Lysvåg, Per  111 M Mackenzie, J. Lachlan  93–4 Maienborn, Claudia  217, 289 Mann, William C.  2, 17, 33–50, 59, 73, 76–7, 81, 89, 140, 248, 259, 346 Marcu, Daniel  38, 55 Martin, James R.  40, 52, 93 Matthiessen, Christian M.  2,  5, 11, 15, 18, 20,  34–8, 44, 50, 60, 73–4, 93, 109–12, 140, 250 McKeown, Kathy  45 Meinunger, André  190, 203–6 Miltsakaki, Eleni  155 Molnár, Valéria  2, 20, 310, 313 Moore, Johanna  39, 50 Moser, Megan  39 N Netter, Klaus  189, 207 Noonan, Michael  13 O Oberlander, Jon  225 Oirsouw, Robert van  284 Oksefjell, Signe  3 Oldenberg, Hermann  309–12, 316, 320, 324 Önnerfors, Olaf  333 Oppenrieder, Wolfgang  209 P Paillard, Michel  92–3, 106 Partee, Barbara H.  7 Pasch, Renate  4, 12–3, 21, 52, 60–69, 188, 191, 272, 309

Peters, Jörg  80, 314 Petrova, Svetlana  16, 25,  232, 329 Peyer, Ann  2, 13, 20, 138, 187, 189, 207 Pintzuk, Susan  330 Polanyi, Livia  50, 59, 60, 78–80, 250 Pollack, Martha  50 Pollard, Carl  188,  208, 210 Prasad, Rashmi  50, 52 Prévot, Laurent  24, 248 Progovac, Ljiljana  292 Pullum, Geoffrey K.  4–5, 8–9, 13 Pustejovsky, James  49 Q Quirk, Randolph  14, 60, 62, 67–8, 85, 108, 109, 111 R Ramers, Karl Heinz  331 Ramm, Wiebke  14, 16, 18–24, 93, 115–6, 123, 129–31, 136, 144, 214 Ranke, Friedrich  344 Rauch, Irmengard  339 Redder, Angelika  5, 10 Reich, Ingo  5, 7–10, 24, 25, 281–2, 285, 288, 292–3, 295–6, 299, 301 Reinhart, Tanya  178, 332 Reis, Marga  2, 13, 187, 190–1, 199–207, 282, 285 Reyle, Uwe  16, 241, 253 Ries, John  329, 339, 343 Roberts, Craige  172, 263, 288 Rondhuis, Jan-Klaas  39, 45, 50 Rooth, Mats  7 Rosengren, Inger  313 Ross, John R.  8, 222, 270 Russell, Benjamin  256, 258 S Sæbø, Kjell Johan  13 Safir, Ken  211 Sag, Ivan A.  188–9, 207–8, 210, 212 Sanders, Ted  39 Santorini, Beatrice  189–90 Sasse, Hans-Jürgen  331, 334 Schaffar, Wolfram  309

Index of names  Schecker, Michael  20, 60 Schein, Barry  289 Schilder, Frank  44 Schrodt, Richard  333–4 Schwabe, Kerstin  5, 322 Schwager, Magdalena  257–8, 264–5 Schwarz-Friesel, Monika  234 Selkirk, Elisabeth  219, 322 Shaer, Ben  209, 223–4, 227 Shopen, Timothy  59 Sidner, Candace  40, 45, 52, 54 Simmler, Franz  333 Simon, Horst J.  333 Smith, Carlota  47 Smith, Michael  220 Solf, Michael  16, 24, 25 Solfjeld, Kåre  8, 15, 19–23, 91, 93, 115–6, 128, 135, 141 Sperber, Dan  9, 152 Stede, Manfred  19, 21, 35, 40, 51, 53, 76–8, 157, 225 Steinbach, Markus  203, 205 Sternefeld, Wolfgang  299 Steube, Anita  310 Stockwell, Robert P.  338 Stolterfoht, Britta  322 Stutterheim, Christiane von  17, 19, 22, 80–1, 116–7, 124, 162–3, 180, 310

Sweetser, Eve  193 Szucsich, Luka  310 T Taboada, Maite  33, 73, 76–7 Tenbrink, Thora  44 Thompson, Sandra A.  2, 4, 11, 15, 17–8, 20, 33–47, 59–60, 64, 73–77, 81, 109, 110–2, 140, 187, 189–91, 203, 250, 259, 346 Tichy, Eva  315 Tomlin, Russell S.  20 Truckenbrodt, Hubert  190, 203, 206, 290 Txurruka, Isabel G.  18 U Uhmann, Susanne  191–5, 212, 218–9, 224, 229 Umbach, Carla  313, 322 Uszkoreit, Hans  189, 207 V Vallduvi, Eric  314 Van Valin, Robert  59, 187, 189 Vanderauwera, Ria  94–95 Velde, John te  7–11 Verstraete, Jean-Christophe  12

Vieu, Laure  18, 23–4, 59, 77, 79–81, 89, 126, 139, 228, 248, 347 Vikner, Sten  190, 203 Vilkuna, Marina  314 Vinay, Jean-Paul  96 W Wackernagel, Jacob  313 Webber, Bonnie  39, 50, 67, 273 Wechsler, Stephen  190, 206 Wegener, Heide  60, 191, 193–4 Werlich, Egon  47 Wilder, Chris  283 Williams, Edwin  283 Wilson, Deirdre  9 Wolf, Florian  40, 49–50 Wright, Georg Henrik von  270 Wunderlich, Dieter  324 Z Zeevat, Henk  9, 225 Zhang, Ning  5 Zifonun, Gisela  5, 7, 13, 60, 223–4, 230, 234 Zimmermann, Thomas Ede  271–2

Studies in Language Companion Series A complete list of titles in this series can be found on the publishers’ website, www.benjamins.com 103 Josephson, Folke and Ingmar Söhrman (eds.): Form and Functions. Apsect, tense, mood, diathesis, and valency. Expected August 2008 102 Goddard, Cliff (ed.): Cross-Linguistic Semantics. 2008. xvi, 356 pp. 101 Stolz, Thomas, Sonja Kettler, Cornelia Stroh and Aina Urdze: Split Possession. An areallinguistic study of the alienability correlation and related phenomena in the languages of Europe. x, 546 pp. Expected May 2008 100 Ameka, Felix K. and M.E. Kropp Dakubu (eds.): Aspect and Modality in Kwa Languages. 2008. ix, 335 pp. 99 Høeg Müller, Henrik and Alex Klinge (eds.): Essays on Nominal Determination. From morphology to discourse management. xviii, 366 pp. + index. Expected May 2008 98 Fabricius-Hansen, Cathrine and Wiebke Ramm (eds.): 'Subordination' versus 'Coordination' in Sentence and Text. A cross-linguistic perspective. 2008. vi, 357 pp. 97 Dollinger, Stefan: New-Dialect Formation in Canada. Evidence from the English modal auxiliaries. 2008. xxii, 355 pp. 96 Romeo, Nicoletta: Aspect in Burmese. Meaning and function. 2008. xv, 289 pp. 95 O’Connor, Loretta: Motion, Transfer and Transformation. The grammar of change in Lowland Chontal. 2007. xiv, 251 pp. 94 Miestamo, Matti, Kaius Sinnemäki and Fred Karlsson (eds.): Language Complexity. Typology, contact, change. 2008. xiv, 356 pp. 93 Schalley, Andrea C. and Drew Khlentzos (eds.): Mental States. Volume 2: Language and cognitive structure. 2007. x, 362 pp. 92 Schalley, Andrea C. and Drew Khlentzos (eds.): Mental States. Volume 1: Evolution, function, nature. 2007. xii, 304 pp. 91 Filipović, Luna: Talking about Motion. A crosslinguistic investigation of lexicalization patterns. 2007. x, 182 pp. 90 Muysken, Pieter (ed.): From Linguistic Areas to Areal Linguistics. 2008. vii, 293 pp. 89 Stark, Elisabeth, Elisabeth Leiss and Werner Abraham (eds.): Nominal Determination. Typology, context constraints, and historical emergence. 2007. viii, 370 pp. 88 Ramat, Paolo and Elisa Roma (eds.): Europe and the Mediterranean as Linguistic Areas. Convergencies from a historical and typological perspective. 2007. xxvi, 364 pp. 87 Verhoeven, Elisabeth: Experiential Constructions in Yucatec Maya. A typologically based analysis of a functional domain in a Mayan language. 2007. xiv, 380 pp. 86 Schwarz-Friesel, Monika, Manfred Consten and Mareile Knees (eds.): Anaphors in Text. Cognitive, formal and applied approaches to anaphoric reference. 2007. xvi, 282 pp. 85 Butler, Christopher S., Raquel Hidalgo Downing and Julia Lavid (eds.): Functional Perspectives on Grammar and Discourse. In honour of Angela Downing. 2007. xxx, 481 pp. 84 Wanner, Leo (ed.): Selected Lexical and Grammatical Issues in the Meaning–Text Theory. In honour of Igor Mel'čuk. 2007. xviii, 380 pp. 83 Hannay, Mike and Gerard J. Steen (eds.): Structural-Functional Studies in English Grammar. In honour of Lachlan Mackenzie. 2007. vi, 393 pp. 82 Ziegeler, Debra: Interfaces with English Aspect. Diachronic and empirical studies. 2006. xvi, 325 pp. 81 Peeters, Bert (ed.): Semantic Primes and Universal Grammar. Empirical evidence from the Romance languages. 2006. xvi, 374 pp. 80 Birner, Betty J. and Gregory Ward (eds.): Drawing the Boundaries of Meaning. Neo-Gricean studies in pragmatics and semantics in honor of Laurence R. Horn. 2006. xii, 350 pp. 79 Laffut, An: Three-Participant Constructions in English. A functional-cognitive approach to caused relations. 2006. ix, 268 pp. 78 Yamamoto, Mutsumi: Agency and Impersonality. Their Linguistic and Cultural Manifestations. 2006. x, 152 pp. 77 Kulikov, Leonid, Andrej Malchukov and Peter de Swart (eds.): Case, Valency and Transitivity. 2006. xx, 503 pp.