Reduced Parenthetical Clauses as Mitigators: A corpus study of spoken French, Italian and Spanish [1 ed.] 9789027292810, 9789027223012

While parentheticals attract constant attention, they very rarely constitute the main subject of monographs. This book p

146 86 2MB

English Pages 252 Year 2007

Report DMCA / Copyright


Polecaj historie

Reduced Parenthetical Clauses as Mitigators: A corpus study of spoken French, Italian and Spanish [1 ed.]
 9789027292810, 9789027223012

Citation preview

Reduced Parenthetical Clauses as Mitigators

Studies in Corpus Linguistics SCL focuses on the use of corpora throughout language study, the development of a quantitative approach to linguistics, the design and use of new tools for processing language texts, and the theoretical implications of a data-rich discipline. General Editor Elena Tognini-Bonelli Consulting Editor Wolfgang Teubert Advisory Board Michael Barlow

Geoffrey Leech

University of Auckland

University of Lancaster

Robert de Beaugrande

Anna Mauranen

Università del Litorale, Capodistria

University of Helsinki

Douglas Biber

Ute Römer

North Arizona University

University of Hannover

Chris Butler

John Sinclair

University of Wales, Swansea

The Tuscan Word Centre

Sylviane Granger

Piet van Sterkenburg

University of Louvain

Institute for Dutch Lexicology, Leiden

M. A. K. Halliday

Jan Svartvik

University of Sydney

University of Lund

Susan Hunston

John Swales

University of Birmingham

University of Michigan

Stig Johansson

H-Z. Yang

Oslo University

Jiao Tong University, Shanghai

Graeme Kennedy Victoria University of Wellington

Volume 27 Reduced Parenthetical Clauses as Mitigators A corpus study of spoken French, Italian and Spanish by Stefan Schneider

Reduced Parenthetical Clauses as Mitigators A corpus study of spoken French, Italian and Spanish

Stefan Schneider University of Graz

John Benjamins Publishing Company Amsterdam/Philadelphia



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.

Cover illustration from original painting Random Order by Lorenzo Pezzatini, Florence, 1996.

Library of Congress Cataloging-in-Publication Data Schneider, Stefan, 1958Reduced parenthetical clauses as mitigators : a corpus study of spoken French, Italian and Spanish / Stefan Schneider. p. cm. (Studies in Corpus Linguistics, issn 1388–0373 ; v. 27) Includes bibliographical references and indexes. 1. Grammar, Comparative and general--Clauses. 2. Grammar, Comparative and general--Parenthetical constructions. 3. Pragmatics. 4. Semantics. I. Title. P297.S36 2007 415--dc22 isbn 978 90 272 2301 2 (Hb; alk. paper)


© 2007 – 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


1 1.1 1.2 1.3 1.4

Preface Abbreviations and notation conventions


Introduction What are reduced parenthetical clauses? Terms, terms ... Some open issues Outline of the book

1 1 3 7 14

2 Parenthesis: a problematic concept 2.1 Origins and modern use 2.2 Principal description criteria 2.2.1 Prosody 2.2.2 Interruption of the host clause 2.2.3 Lack of an overt link 2.2.4 Sententiality 2.2.5 Relationship with the host clause 2.2.6 Communicative function 2.3 Summary

19 19 22 22 25 27 28 29 32 34

3 3.1 3.2 3.3 3.4 3.5 3.6 3.7

37 37 41 44 48 54 57 62

Previous studies Before 1952 Urmson (1952) and Benveniste (1958) RPCs re-discovered Generative analyses RPCs as hedges and as discourse markers Recent contributions Summary



4 Reduced parenthetical clauses in spoken corpora 4.1 An empirical approach 4.2 The analyzed corpora 4.2.1 French corpora 4.2.2 Italian corpora 4.2.3 Spanish corpora 4.3 Selection criteria 4.4 The RPCs selected from the corpora 4.5 The frequencies of RPCs 4.5.1 The three samples 4.5.2 Overall frequencies 4.5.3 Frequencies related to the text type 4.6 Summary 5 5.1 5.2 5.3

Pragmatic analysis: fundamental issues Mitigation Parentheticals versus performatives Summary

6 Pragmatic analysis: functions 6.1 A first basic subdivision 6.2 Mitigating the phrastic 6.3 Indicating the tropic and mitigating the phrastic or neustic 6.3.1 Plain performatives 6.3.2 Hedged performatives 6.4 Directly mitigating the neustic 6.4.1 Believing 6.4.2 Not knowing 6.5 Indirectly mitigating the neustic 6.5.1 Evidentials 6.5.2 Knowing 6.6 Reporting speech 6.7 Summary

65 65 66 67 70 72 73 78 83 83 84 86 89 91 91 100 106 109 109 111 115 115 118 121 121 122 125 125 130 132 134


7 7.1 7.2 7.3 7.4 7.5 7.6

Semantic and pragmatic properties Semantic characterization Influence of person and time RPCs as answers to polar questions Can RPCs be negated? Restricted to declarative sentences? Summary


139 139 143 145 148 151 156

8 Syntactic properties 8.1 Syntactic unit and syntactic function 8.2 Position 8.2.1 Inside the clause 8.2.2 Between clauses 8.2.3 Sentence-final 8.2.4 Sentence-initial 8.2.5 Mobility and interruptive force 8.3 Complementizer deletion or parenthetical? 8.4 Scope 8.4.1 Phrase-limited scope 8.4.2 Clause-limited scope 8.4.3 Sentence-limited scope 8.5 Governing clause versus parenthetical clause 8.6 Summary

159 159 168 169 171 172 173 174 177 184 185 186 190 191 197

9 9.1 9.2 9.3 9.4 9.5

Prosodic properties Prosodic parenthesis versus syntactic parenthesis RPCs in the CRE and CIP RPCs in the CRFP RPCs in the COR-FR, COR-IT, and COR-ES Summary

199 199 204 205 208 209




References Subject index

217 235

PREFACE Parenthesis and the various types of parenthetical expressions have attracted constant but never intense attention. Now and then they are touched on in articles, while rarely constituting the main subject of books. The present monograph is one of the few specifically dedicated to parenthetical verbs or, as I prefer to call them, reduced parenthetical clauses (RPCs) in three Romance languages, namely French, Italian, and Spanish. If asked to declare my linguistic affiliation or to define the approach from which these clauses are examined, I would have some difficulty in giving a short answer. One of the reasons is that in this book I tried to proceed as independently as possible from specific and isolated linguistic traditions and their particular terminologies. It is up to the reader to determine whether this attempt has been successful. Another reason is that in the present monograph RPCs are described from several perspectives and at several analytic levels: pragmatic, semantic, syntactic, and even prosodic ones. Each of these perspectives implies a host of methods and linguistic traditions, none of which offers, in my opinion, a satisfactory framework embracing all levels mentioned. With this preliminary remark, I do not mean to suggest that it is impossible to note the influence of one or more linguistic traditions. In fact, it will be clear to every reader that this study draws extensively on the methodology of contemporary corpus linguistics. It is based on 22 corpora of spoken French, Italian, and Spanish comprising a total amount of 3,975,500 words that document spoken language of the last 30 years. On the other hand, the pragmatic analysis, i. e. the most dominant part of the study, owes much to the framework first sketched out by Urmson (1952) and Hare (1970), and recently further developed by Caffi (1999, 2001). While the semantic and syntactic parts are more eclectic and are not underpinned by a specific linguistic model, the final considerations on the prosodic aspects of parenthetical clauses are partly influenced by theories outlined in Morel & Danon-Boileau (1998) and in Campione (2001a, 2001b). Writing a book is both a solitary and a social exercise. After long and lonely early morning hours in my office at the Institut für Romanistik, I had the luck to meet Gudrun Held, Martin Hummel, Bettina Kluge, Hugo Kubarth, Klaus Lichem, and Muriel Warga; patient colleagues who, during coffee breaks, were willing to discuss aspects of my work or who were generous enough to read drafts of the text, even during their summer vacations. I also had the privilege of presenting two early versions to the colleagues of the Grazer Linguistisches Kolloquium. My work has profited considerably from the suggestions and ideas that came up during these informal gatherings.



This study would have been impossible if I had not had the opportunity to carry out parts of it at the Facultad de Filología of the Universidad de Sevilla and at the Département de Linguistique Française of the Université de Provence. I want to thank the Karl-Franzens-Universität Graz and the Hugo Schuchardt'sche Malvinenstiftung for the financial funding and Catalina Fuentes Rodríguez, Claire Blanche-Benveniste, Sandrine Caddéo, José Deulofeu, Frédéric Sabio, and André Valli for establishing the necessary contacts, their warm reception, the corpora they made available to me and the interest they showed for parenthetical expressions. I also want to thank the members of the Équipe de recherche DELIC (Description linguistique informatisée sur corpus) in Aix-en-Provence, above all Estelle Campione and Jean Véronis, for admitting me to their regular meetings as well as for their valuable suggestions regarding the prosody of parentheticals. Upon completion, the entire manuscript or parts of it have been read by a number of friends, colleagues, and reviewers. Among them I would like to mention Claudia Caffi, Emanuela Cresti, Bruce Fraser, Klaus Hölker, Thomas Krefeld, Guido Mensching, Michele Loporcaro, and Ulrich Wandruszka, who all provided helpful comments. The book has been improved considerably by their care. Since I have not always followed their advice, if faults and infelicities remain, they are certainly my own. I am also grateful to Elena Tognini Bonelli for her support during the publication of the book. As the reader will undoubtedly note, this book has not been written by a native speaker of English. Several persons have given me advice that, hopefully, helped to improve the grammatical consistency and the style of the text. I am especially grateful to Georg Marko and Ursula Lindenberg for their accurate corrections.

ABBREVIATIONS AND NOTATION CONVENTIONS The abbreviations for the corpora are given in section 4.2 (table 4.1). Specific abbreviations used in a table are explained in its caption.

Abbreviations Adv BADIP CHAT Eng. F0 Fr. Ger. IFID It. N NP p RPC S Sp. V VP

Adverb Banca dati dell'italiano parlato Codes for the human analysis of transcripts English Fundamental frequency French German Illocutionary force indicating device Italian Noun Noun phrase Propositional variable Reduced parenthetical clause Sentence Spanish Verb Verb phrase

General notation conventions * ?

[...] qui / non /k∂/

Unacceptable expression Not fully acceptable expression Parts of the example or citation have been omitted Both expressions are acceptable Phonemic transcription



Special notation conventions Corpus or author BEECH BER Bloomfield (1935)

Notation [rire] ++ [,] [.]

Caffi (1999, 2001)

. ? ^ -et: plai*sir ↗, ↘, → -// / [/] [!] &mhm /' //" %ill: %inf: deputati_ # ## , , do

matao " // / # [/] // / / mm.

Campione (2001a, 2001b)






Description Comment Pause of two seconds Pause pitch Final falling pitch of statements Falling intonation Rising intonation Short pause Short pause Pause of medium length Long pause Lengthening Focalization accent Rising, falling, or flat F0 contour Short pause Long pause Utterance limit Limit of intonation unit False start (retracting) Preceding word is stressed Interrupted word End of reported intonation unit End of reported utterance Illocution Information structure Final vowel lengthening Short pause Pause of medium length Unintelligible word(s) Reconstructed word Comment The letter d was not pronounced Direct speech Utterance limit Limit of intonation unit Pause Break caused by a false start Utterance limit Limit of intonation unit Utterance segmentation Sentence-final intonation



Emonds (1976) HCM HOELK Morel & Danon-Boileau (1998) PIXI

Voghera (1992)

... (p=ste) pas: x, xx

- av/le, me/ X, XX, XXX E ... V ... desde … […?] ... °xxx° éno:::rmes " + (5syll) . , ? // iROnico


Pause The expression is pronounced ste Lengthening Unintelligible word(s) Direct speech Short pause Long pause Overlap Interrupted word Both transcriptions are possible Unintelligible syllable(s) Expression Hesitation Lengthening Unintelligible passage Short pause Parenthetical expression Hesitation Interrupted intonation unit Title of book or series Short pause Number of unintelligible syllables Low fall intonation Fall-rise intonation Low rise intonation Limit of intonation unit Accented syllable

INTRODUCTION 1.1 What are reduced parenthetical clauses? In everyday conversations, speakers of French may occasionally insert into their sentences1 expressions such as those boldfaced in the following corpus examples: (1)

(2) (3)

L2 […] on est quelques centaines - à être encore des locuteurs disons naturels de la langue […] (CRFP.GAP-PRI002) [We are a few hundred that are still, let's say, natural speakers of the language] C: Ah ben c'est plus libre maintenant, je crois (BEECH.16) [It's freer now, I believe] FA 13 - […] c'était euh je me rappelle / dans les années après guerre / […] (CREDIF.FA13.351) [That was, I remember, in the years after the war]

These expressions, which I call reduced parenthetical clauses (RPCs), are neither the main clause nor a subordinate clause, but are inserted into or adjoined to the end of the sentence in a way similar to sentence adverbs. Their position is free and there is no overt syntactic link between them and the host sentence or parts of it. They are related to the host only by adjacency and by the fact that their missing argument can be recovered from the host. As other peripheral elements, they are optional, i.e., they can be added as well as dropped freely without endangering the grammatical acceptability of the host. Nevertheless, they are pragmatically connected to it. The use of RPCs is, of course, not limited to French. Speakers of other Romance and non-Romance languages employ them, too. The following two examples show RPCs in Italian and Spanish: (4)

A: senti ti devo lasciare perché ho un collegamento credo da Roma grazie comunque di (CLIP.MC9) [Listen, I must hang up because I have a call, I believe, from Rome, thanks anyway for]

I am of course aware of the fact that acts of speech are performed by utterances and not by sentences or that, as Bar-Hillel (1970) puts it, utterances are sentences paired with a context. For the sake of simplicity, I will use sentence for both 'sentence' and 'utterance' wherever this distinction is not immediately relevant or where the meaning is clear from the context. Occasionally, I will also use utterance as a pre-theoretical term for "any stretch of talk, by one person, before and after which there is silence on part of that person" (Harris 1951: 14) and extend this definition to written language as well (see Lyons 1977: 26). 1

2 (5)


Tendrán ustedes, supongo, periodistas corriendo por ahí ya, ¿no? (COREC.BENT027B) [You have, I suppose, reporters running already over there, don't you?]

French RPCs in most cases require a subject, pronominal or other, Italian and Spanish RPCs mostly omit it. Regarding their status as parentheticals, the boldfaced clauses in (1) - (5) are comparable to the following boldfaced clause: (6)

M: [...] on le notera je l'espère avec une grande ampleur sur le problème du ticket modérateur [...] (LUD.6.118) [It will be remarked, I hope it, on a large scale regarding the problem of the fixed participation]

Yet, from a syntactic point of view, there is a difference: In (1) - (5), the verbs in the parenthetical clauses lack one of the arguments required by their valency and, in this respect, they depend on the host, whereas in (6) this is not the case. For this reason, the clauses in (1) - (5) are reduced parenthetical clauses, while the one in (6) is a complete parenthetical clause. The verbs in the examples (1) - (5) have one argument less than they would have if they were not in a parenthetical position, as, e.g., in the following passage: (7)

FA6 - je crois que ça me restera tout le temps / toute la vie. (CREDIF.FA6.272) [I believe that I'll keep this all the time, all my life]

Whether the conditions of use of the RPCs differ from those of governing clauses above is a question addressed in this book. The syntactic differences are apparent: The boldfaced expressions in (1) - (5) are ungoverned dependents of the sentence or its parts, i.e., they are adjuncts; the expression in (7) governs a (dependent) noun clause, i.e., it is a head. RPCs are paradoxical: Syntactically, their verbs lack one argument but they do not actually govern the host sentence; semantically, the lacking argument can be recovered by referring to the host. The pragmatic function most commonly associated with RPCs concerns the speaker's claim to truth. In other words, the speaker, while making a statement, uses an RPC to reduce his or her claim to truth and hence the burden of responsibility that comes with it. Reducing one's claim to truth is a way of mitigating a statement (cf. Caffi 1999, 2001). As all speech acts, statements and the resulting utterances are complex structures composed of several parts contributing to their meanings. RPCs can thus be classified according to the utterance component they focus on. E.g., disons in (1) is concerned with the appropriateness or precision of the referring term in its scope, whereas je crois in (2) dir-



ectly operates on the speaker's commitment to the truth of the proposition. Je me rappelle in (3) indicates the evidential circumstances of the speaker's statement and only indirectly reduces his or her commitment. RPCs may fulfill other discourse functions besides the mitigation of statements. For example, they may serve as phatics, i.e., as expressions that invite the addressee to cooperate and thus maintain the effectiveness of communication, as vous savez in the following French example: (8)

B: […] Saint Lunaire lui-même est anglais, vous savez? […] (BEECH.23) [Saint-Lunaire itself is English, you know?]

Furthermore, RPCs containing utterance verbs are commonly used to report utterances made by other speakers, as in the following Italian sequence: (9)

A: e si realizza dice Calvino attraverso dei riti di? F: iniziazione (CLIP.RC9) [A: And it is achieved says Calvino through rites of? B: Initiation]

1.2 Terms, terms ... Quite a few terms have been proposed for these clauses, none of them completely satisfying. This is due in part to the fact that parenthesis is an elastic concept in linguistics and in part to the fact that linguists did not exactly know how to classify these expressions syntactically. The terminological variety is testimony to the manifold approaches and perspectives scholars have adopted regarding RPCs. Some were intended only for RPCs, while others refer to a set of related phenomena of which these clauses are a part. The labels are mostly composed of adjectives attached to terms such as sentence, clause, phrase, predicate, verb, and in the following pages they will be presented in this sequence. In the first work entirely dedicated to parenthesis, Schwyzer (1939: 31) defines it a phenomenon in which a sentence (grammatischer Hauptsatz) interrupts another sentence. A few pages later (1939: 40), he discusses the incidental use of Classical Greek dokô, Latin puto, and Swiss German glaub and mein and calls these expressions parenthetische Verbalsätze. Generative linguists differ in their treatment of English parentheticals like I think, I believe, and others. Espinal (1991) proposes a differentiated syntactic classification of parentheticals, ranging from sentence to phrases, in which I guess or the President announced are defined as disjunct sentences. Grevisse (1993: 273, 1579) calls the sentences (Fr. phrases) that are united by coordination or by inserting a parenthetical into a host sous-phrases. A sous-phrase incidente is a complete parenthetical clause, a sous-phrase incise a reduced one. According to Grevisse, sousphrases, unlike subordinate clauses, are not essentially different from simple



sentences. Influenced by Hare's (1970) theory, which postulates as an essential component of utterances the neustic or sign of subscription, I proposed to call parentheticals such as It. credo, penso, etc. frasi parentetiche sottoscrittive (see Schneider 1997a). Since parentheticals such as credo, penso, etc. express only a reduced, mitigated or modalized subscription, I argued later (see Schneider 1999: 43) that frase modalizzante was more appropriate than frase sottoscrittiva. In a few introductory lines to a chapter dedicated to parenthetic clauses, Jespersen (1937: 72) presents his symbolization of sentences containing parenthetical I think, it seems, etc.; however, there are no other remarks that hint at a possible definition of his notion of parenthetical clause, which therefore remains cloudy. In the linguistic literature of the last three decades the term, as well as its cognate parenthetical clause, is fairly widespread and covers many types of incidental constructions, though typically authors cite as most representative examples medial or final I think or I believe (see Emonds 1976; Levinson 1983: 266). Similarly, Italian linguists influenced by generative grammar usually employ frase parentetica (see Lonzi 1981; Borgato & Salvi 2001).2 Quirk, Greenbaum, Leech, & Svartvik (1985: 1112ff.) call parenthetical I believe and a series of related expressions comment clauses and write: Comment clauses are parenthetical disjuncts.3 They may occur initially, finally, or medially, and thus generally have a separate tone unit. (Quirk, Greenbaum, Leech, & Svartvik 1985: 1112)

More specifically, I believe is subsumed under the category of content disjuncts, i.e., those clauses expressing the speaker's comment on the content of the host sentence. Biber, Johansson, Leech, et al. (1999: 197, 864ff.) also refer to parenthetical I think, you know, it seems, and similar expressions as comment clauses and categorize most of them as stance adverbials. Cornulier (1978) calls parenthetical clauses with inverted subjects incises (inverties).4 Parenthetical clauses with preposed subjects like in (2) and in (3) are incises progressives. In Cornulier's (1978: 87) eyes, there is a radical difference between the incises mentioned so far and the parenthetical clause in (6), which he, following an established French tradition, labels incidente. Most importantly, though, for my considerations regarding Urmson's (1952) terminology, he states that the formal notion of parenthetical verb does not make much sense. For Bolinger (1968), incidental or final I believe, I think, and I guess are postposed main phrases. Likewise, a couple of years later, Ross (1973: 133), In contrast to the usual Italian linguistic terminology, here frase means 'clause' instead of 'sentence' (see Lonzi 1981: 397; Emonds 1976: 55). 3 In Quirk, Greenbaum, Leech, & Svartvik (1972: 778) comment clauses "may be classed as disjuncts or conjuncts". 4 Cornulier (1978) never clearly states that incises and incidentes are clauses, but he probably respects the traditional French use that classifies them as propositions (= clauses) (see Mounin 1974: 171f.). 2



working within the framework of generative grammar, defines the final parenthetical in Max is a Martian, I feel a phrase, without referring to any particular kind of phrase. Recently, Kärkkäinen (2003: 38-45), in a corpus study of English epistemic stance markers, has taken up this characterization by defining parenthetical I think, I suppose, and other similar verb-based expressions epistemic phrases. Hooper (1975) proposes the term assertive predicates for these expressions, because they characterize and sometimes govern clauses that are assertions and should be distinguished from factive predicates, which introduce clauses whose truth is presupposed (see Kiparsky & Kiparsky 1970). The term assertive can lead to misunderstandings: Especially when used as parentheticals, I believe, etc. do not assert at all, the assertion being expressed by the sentence into which they are inserted. Furthermore, it is of course problematic to classify I believe, etc. as predicates, as this would imply that the pronominal subject (e.g., Engl. I, Fr. je) is part of the predication. Since in Spanish incidental creo, pienso, etc. often occur without pronominal subjects, Spanish linguists are not reluctant to call them predicados parentéticos or verbos parentéticos (see for example Kovacci 1994 [1979]; Alcaraz Varó & Martínez Linares 1997: 593; Maldonado González 1999: 3571ff.). This also happens in Italy, where verbi or predicati parentetici is the most frequent term. Benveniste (1966a [1958]) calls verbs like Fr. croire, supposer, présumer, and conclure, which in first person present tense express the speaker's attitude towards the sentence and may occur in parenthetical position, verbes d'opération. Blanche-Benveniste (1989: 60f.), examining a corpus of spoken French, found Fr. je crois bien being used as a parenthetical and as an expression governing a que-clause. Because of the phonologically weak (or even missing) que between je crois bien and the following clause, she defines it as a verbe recteur faible. From a completely different perspective, the philosopher Hall analyzes the clause I assume. When used as governing clause or as parenthetical clause it is a member of [...] a set of special linguistic devices for adding to, cancelling or altering the commitments which statements made about them would have involved, indicating the sorts of grounds statements they introduce may have [...]. (Hall 1958: 54)

Due to this function of I assume in statements, Hall calls it a positing word. Among English speaking linguists, the best-known and most widely used term certainly is parenthetical verbs,5 proposed by Urmson (1952). The philosopher starts with three examples: (10)

I suppose that your house is very old.

RPCs are mentioned in Matthews' (1997) The Concise Oxford Dictionary of Linguistics under this entry. 5



(11) (12)

Your house is, I suppose, very old. Your house is very old, I suppose.

These are followed by a short definition: A verb which, in the first person present, can be used, as in the example above, followed by 'that' and an indicative clause, or else can be inserted at the middle or end of the indicative sentence, is a parenthetical verb. (Urmson 1952: 481)

As follows from this quotation, Urmson's understanding of parentheses and of RPCs is an unusually broad one: Final clauses and even initial governing clauses are included. Instead of attempting to define their exact place in the sentence, he tries to capture their pragmatic function, which, in his eyes, does not change with their positions and is comparable to that of sentence adverbs. He therefore frequently uses the expression parenthetical use when referring to this specific function. For him, this function is a property of single forms of these verbs, mainly of the first person singular of the non-progressive present indicative. Although he does not insist on the aspect, Urmson is of course aware of the fact that, strictly speaking, to believe or to suppose are not per se parenthetical verbs, but that single forms of these verbs may be used within a parenthetical clause. Apart from this, there are two further problems with his terminology: He defines RPCs through their discourse function but names them by referring to the formal property of parenthesis and, moreover, his understanding of this formal property is also very special. In fact, since Urmson (1952), it has not been clear anymore if parenthesis is a discourse-functional concept or just a notion referring to word order and syntax. The difficulty in determining the syntactic unit and the syntactic function of these expressions combined with the occasional reluctance to deal with this issue at all has led many authors to use a hypernym that avoids the problem. Terms such as parenthetical or incidental construction, parenthetical or incidental expression, parenthetical or incidental structure, elemento parentético (Fuentes Rodríguez 1990a: 105), or simply inciso (Cresti 2000a) are very common. Corum (1975: 133, 136) calls I believe, Harvey says or I think placed at the end or in the middle of the clause simply parentheticals, which together with sentence adverbs, adverbial phrases, some non-restrictive relative clauses and rhetorical tag questions are said to constitute the group of pragmatically defined parenthetic adjuncts. Vigara Tauste (1992: 397ff.) also takes a pragmatic standpoint and calls them incisos de opinión. Similarly, in García Cid's (1995: 45f.) framework, where communicative function plays a much more prominent role than syntax, incidental expressions such as Sp. me acuerdo are simply Kommentare (über einen Sprechakt). Although one should not attach excessive importance to the different labels scholars have found convenient for RPCs, the terminological uncertainty mirrors the general attitude towards them. Each one of the names we have seen so



far captures one or more of the following aspects of RPCs: their pragmatic function, the fact that they contain verbs, the syntactic features of these clauses, their position in the host sentence, and the syntactic unit they represent. None of them seems quite to the point, though; either the term is not correct or, if it is correct, it lacks precision. Due to this uncertainty, I briefly want to illustrate my understanding of reduced parenthetical clause; a thorough discussion will follow in chapter 4. The point of departure of my research is Urmson (1952); that means that I focus, roughly, on those clauses with finite verbs that may be inserted everywhere in the host, that are not overtly linked to the host, whose verbs lack one of the arguments required by their valency, and whose lacking argument can be recovered semantically from the host sentence. Section 4.3 specifies these and other criteria in more detail. I am aware of the fact that these criteria are closely tailored to my needs, however, this allows to restrict the field of research and to select the RPCs from the corpus texts. The list of proposed criteria is open to discussion; occasionally, I will take into account borderline cases that do not satisfy all of the proposed conditions. Although the point of departure is Urmson (1952), I cannot adopt his terminology. Parenthetical verbs has a long tradition and is a well-known label, however, I cannot call the expressions this book is about verbs. As will be illustrated in section 8.1, from the perspective of the syntactic unit, incidental Fr. je crois, It. credo, Sp. creo, etc. are not verbs, i.e., simple words, but clauses, albeit reduced ones. Moreover, as anticipated already by Urmson (1952) and as underlined by Cornulier (1978), there exists no lexically definable class of parenthetical verbs. In this study I adhere to the following convention: Parenthesis and parenthetical, very much like incidental, are used as syntactic terms. Therefore I do not use parenthetical in Urmson's sense but as a general term not specifying a particular discourse function. Whenever I want to refer to Urmson's sense, i.e., whenever I speak of those RPCs that express a mitigation, I will adopt the term mitigating RPCs. 1.3 Some open issues RPCs give rise to several questions. The most important of these will briefly be outlined in this section, some others will turn up throughout the book. Not only the term of the phenomenon at the center of my study is controversial but also its definition. This is due mainly to two facts. Firstly, the concept of parenthesis is itself controversial. According to the narrow definition, a parenthesis is a syntactically complete and autonomous sentence interrupting the syntactic coherence of another sentence. Broad definitions include a great variety of constructions (e.g., interjections, vocatives and other forms of address, elements dislocated to the left or right, appositions, certain elliptical expressions, adjectival and participial supplements to noun phrases, relative clauses, and communicat-



ive formulas). The understanding of the concept of parenthesis depends crucially on a series of criteria which have been adopted by the authors and on the weight assigned to each of them. I have identified at least six criteria and properties: prosody, interruption of the host clause, lack of an overt link between parenthetical and host clause, sententiality, relationship between parenthetical and host clause, and communicative function. The second fact I want to mention is the persisting ambiguity between the semasiologic and the onomasiologic approach to RPCs. This ambiguity begins with Urmson (1952). On the one hand, he gives a morphological and syntactic definition (first person singular of the non-progressive present indicative, either governing or parenthetical) and calls the verbs parenthetical, on the other hand, he himself appears to be interested primarily in their communicative role and writes that parenthetical is merely a convenient term (1952: 481). It is necessary to separate the semasiologic from the onomasiologic perspective and choose one of the two as the analytical starting point. The first perspective can only be taken with a clear definition and delimitation of the clauses involved. According to Urmson (1952), the function of some RPCs is to weaken the claim to truth implied by the host sentence. He neither clearly defines these clauses nor takes into consideration other possible functions RPCs may assume. The majority of authors accepted Urmson's ideas about statements and thus also adopted his interpretation of the role of RPCs in the latter. This role was believed to be the most important, if not the only one. It seems, though, that some revision is needed, mainly because in the spoken corpora, there are RPCs which in previous studies have not been considered (e.g., those based on utterance verbs). On the one hand, there are RPCs that satisfy the formal criteria but do not function as mitigators, on the other, there are RPCs with verbs that are not in the first or third person singular of the present indicative but do act as mitigators. After having explained my understanding of statements, how the speaker's claim to truth is weakened and what the consequences of this communicative strategy are, I want to examine and classify all RPCs according to their communicative roles. The high frequency of RPCs with utterance verbs leads to a question often discussed: Are there performative parentheticals? In Urmson's (1952) view performatives and parentheticals have much in common; but he points to an important difference: Performatives express speaker commitment rather than illocution when used as parentheticals. According to Urmson (1952: 494), we cannot treat the following sentences as a guarantee and a bet, respectively: (13) (14)

He'll come to a bad end, I guarantee. He'll forget to come, I bet.

Urmson adds that we have here a case of the borrowing of another sort of verb for parenthetical uses. From these remarks, we may get the impression that in



Urmson's view a verb that occurs in parenthetical position acts exclusively as mitigator, in other words, there are no performative parentheticals. Venier (1991: 66f.) expresses the same opinion when she says that in the following Italian sentence dico expresses speaker commitment rather than illocutionary force: (15)

È proprio così, dico. [It's just like this, I say]

Lyons (1977: 738; see also Holdcroft 1978: 64 and Récanati 1984: 347), by contrast, states that a parenthetical may very well act as a performative and cites as a counter example the following sentence: (16)

I'll be there at two o'clock, I promise you.

According to Fava (2001: 45ff.) the same is possible in Italian. From a syntactic point of view, RPCs pose several difficulties. The first concerns the syntactic unit they represent. Are they full-fledged sentences? If they are not, as most authors seem to believe, what else do they represent? A glance at the terms mentioned in section 1.2 reveals that any unit from simple word to sentence has been deemed appropriate by at least some author. We will see the most relevant options: sentence, as proposed by Schwyzer (1939) and Espinal (1991); Grevisse's (1993: 273, 1579) sous-phrase; phrase, as suggested by Bolinger (1968) and Ross (1973); predicate (see Hooper 1975; Kovacci 1994 [1979]), which in most cases can be equated with a verb phrase; verb (see Urmson 1952; Benveniste 1966a [1958]); and, finally, clause, as in Jespersen (1937: 72), Quirk, Greenbaum, Leech, & Svartvik (1972: 778ff., 1985: 1112ff.), Emonds (1976), and Levinson (1983: 266). A second syntactic problem posed by RPCs concerns their function. Whereas it is quite clear that they are adjuncts depending on a part of or on the entire sentence, the question remains of exactly how they become adjuncts or, perhaps more correctly, how their adjunctival status is signaled to the hearer. This is important because clauses like Fr. je crois, It. credo, or Sp. creo are usually exactly the opposite of a dependent, i.e., they are heads. When they are in parenthetical position, there is no word, morpheme or other overt functional element that indicates or causes the change of syntactic function. The third problem, more than the preceding two, also has pragmatic significance. When occurring in medial or final position there is no doubt about the function of RPCs, but what happens when they are initial, as governing verbs introducing a declarative noun clause? Are they part of the statement itself or do they mitigate the statement? Urmson sees no pragmatic difference between the two structures unless context changes significantly (see Urmson 1952: 481). Hooper (1975: 101f.), on the other hand, proposes the idea that the type of



statement might condition the interpretation of a main clause like I believe that. Likewise, according to Quirk, Greenbaum, Leech, & Svartvik (1985: 1113), verbs like believe and think, when governing a that-clause, have two meanings, a "more definitive" and a hedging one, while they have only a hedging meaning when they occur in parenthesis. The question has a syntactic and a pragmatic side and accordingly allows, depending on the perspective adopted, different answers. There is no doubt that sentences with governing I believe or its Romance equivalents and sentences with the same verb in parenthetical position differ syntactically, but does this difference always correspond to a pragmatic one in English and Romance? This study focuses on pragmatics and syntax but there is at least one prosodic issue that I would like to discuss. It is frequently assumed that the prosody of RPCs does not follow a particular pattern setting them apart from other instances of parenthesis. However, there are facts indicating that some RPCs have specific intonation properties. Reinhart (1983: 179; see also Ziv 1985: 181f.) distinguishes between two types of reporting final parentheticals: those whose subjects coincide with that of the host clause and those in which this is not the case. The two types display markedly different intonation patterns. The first RPC is pronounced with a low, steady pitch and is preceded by a pause. The second one is pronounced without a preceding pause, like a sentence adverb such as of course. Another observation was made by Voghera (1992: 146), who notes that in the following Italian sentence spero carries no accent (see also a remark by Borillo 1978a: 17): (17)

//sei iROnico spero// [You're being ironic, I hope]

She concludes that spero must be attached to the preceding intonation unit, which implies that the sentence above consists of only one intonation unit. Morel & Danon-Boileau (1998) distinguish systematically between parenthesis, which they call incise, and the use of expressions like je crois, j'espère, il me semble, disons, tu vois, moi je trouve, and je sais pas. The authors divide oral speech into paragraphes oraux on the basis of suprasegmental features. This corresponds essentially to a segmentation into discourse units. The typical paragraphe oral has the following structure: préambule+rhème+(postrhème). The préambule is characterized by a slowly rising F0 which continues without interruption throughout the rhème (see Morel & Danon-Boileau 1998: 36). The facultative postrhème is also characterized by the absence of a pause following the rhème a low F0, the absence of modulation of F0, and the number of syllables ranging between seven and twelve (see Morel & Danon-Boileau 1998: 29, 62). The préambule and the postrhème may both contain RPCs like those previously mentioned. Morel & Danon-Boileau's (1998) work constitutes the



clearest statement of the fact that some RPCs do not interrupt the utterance's intonational contour but are just attachments to other intonation units. The last open issue I want to mention here is less a problem of theory than one related to the empirical basis of the different accounts. Although there is a considerable amount of literature on the subject, there are only three monographs dedicated specifically to Romance RPCs: Cornulier's (1973) unpublished doctoral thesis about French parentheticals, Venier's (1991) philosophically oriented study of mitigating adverbs and parenthetical clauses in English and Italian and Andersen's (1997) unpublished doctoral thesis about parenthetical clauses and subordination in spoken French. Cornulier (1973) and Venier (1991) are not corpus-based. The list of Italian parentheticals contained in Venier (1991: 67f.), though useful from a theoretical point of view, is based on translations from English. Andersen (1997) is based on a fair number of texts. Their total amount of words is, however, largely inferior to that of the French corpora I analyzed. Moreover, the author opts explicitly for a qualitative analysis and focuses on a few selected parentheticals. No list of French RPCs, let alone indications of frequency and distribution, can be found in the study. French is the Romance language whose RPCs have attracted most attention. Among Romance speaking linguists, French researchers were moreover the first to show interest in the phenomenon. Research started with the contributions by Cornulier (1973, 1978) and Borillo. Borillo's work on modal sentence adverbs and other linguistic means to mitigate a statement is especially interesting because it introduced a systematic line of research analyzing - within a coherent pragmatic framework - a series of interrelated aspects, such as modal adverbs and RPCs, negation, and question (see Borillo 1976, 1978a, 1978b, 1979, 1982). In the latter work cited, Borillo affirms that there are more than 200 French verbs that may be postposed (1982: 34). She then tests those of them that may be used as answers to polar questions (see section 7.3 for the description of her test). Their differing behavior in these questions yields a classification that I summarize in table 1.1. These studies established a theoretical basis for further inquiries into RPCs, taking into account their dominant areas of use (as main clauses or as parenthetical clauses) and data about their frequencies of occurrence in spoken vs. written texts. Even the preparation of an exhaustive list of these clauses based on corpora of contemporary French has started. Delomier & Morel (1986) analyzed prosodic and syntactic characteristics of parentheticals in a corpus recorded by themselves; Blanche-Benveniste (1989) examined the relation between RPC and dependent clause in the corpus of the Groupe aixois de recherches en syntaxe (GARS). Andersen (1997), a corpus-based study of parentheticals and subordination in spoken French, discusses and classifies a number of parentheticals that are summarized in table 1.2. Despite its empirical orientation, Andersen's study takes only a selected number of parentheticals into account and does not include statistical information regarding frequencies or textual distribution.


REDUCED PARENTHETICAL CLAUSES Groupe 1 croire avoir l'impression penser trouver sembler on dirait

Groupe 2 supposer espérer imaginer présumer craindre il paraît être probable

Groupe 3 savoir noter observer entendre dire remarquer faire attention sentir voir se rendre compte se souvenir se rappeler être au courant avoir (aucune) idée avoir (aucun) souvenir être évident être sûr être certain Table 1.1: Classification of French parenthetical verbs (cf. Borillo 1982: 52f.) Propositions parenthétiques prototypiques il me semble je t'assure j'ai l'impression je crois je pense je trouve

Propositions parenthétiques non prototypiques 1ère personne 2ème personne je me souviens à l'indicatif à l'impératif je me rappelle tu sais regarde vous savez regardez tu vois écoute vous voyez écoutez voyez remarque tu comprends remarquez comprenez attends attendez dis dites Table 1.2: French parenthetical clauses (cf. Andersen 1997)

Italian RPCs received much less attention, although the Grande grammatica italiana di consultazione edited by Renzi, Salvi, & Cardinaletti (2001) contains a chapter on complete and reduced parenthetical clauses. In Schneider (1999), I examined the use of the subjunctive in the Italian CLIP corpus. In governing position, verbs such as credere, parere, pensare, and sembrare turned out to occur predominantly in the forms (io) credo, (mi) pare, (io) penso, and (mi) sembra. E.g., 90 % of the occurrences of credere as a governing verb are in the first person singular present indicative. Having recognized that these are precisely the forms which also tend to be used in incidental position, I looked through all constructions in the CLIP governing noun clauses and checked which ones may



be used as parentheticals and as affirmative answers to polar questions (a test isolating the mitigating forms from other parenthetical ones). This resulted in a list of mitigating verb forms (see table 1.3). In the table, the verbs marked by ° do not actually appear in parenthetical position in the CLIP. Parentheticals were not the main subject of Schneider (1999), I therefore dedicated only a few pages to them; nonetheless, besides Andersen (1997), the list in table 1.3 is the only one based on a corpus of a spoken Romance language. °mi auguro, io mi auguro credo, io credo si dice, dicono, direi, io direi immagino ho l'impressione pare, mi pare penso, io penso presumo °ritengo mi sa mi sembra, sembra, sembrerebbe spero, °io spero suppongo °trovo si vede Table 1.3: Italian mitigating verb forms (see Schneider 1999: 69)

Spanish RPCs have been studied even less. The recent monumental Gramática descriptiva de la lengua española edited by Bosque & Demonte (1999) does not dedicate a separate section to parenthesis, let alone to RPCs, although it contains occasional remarks on parentheticals, e.g., on parenthetical relative clauses and relative clauses interrupted by parentheticals (see Brucart 1999: 408ff., 475), on RPCs in direct speech (see Maldonado González 1999: 3571ff.), and on performatives in parenthetical position (see Garrido Medina 1999: 3894f). To find a slightly more detailed account, one has to go back to Hooper's (1975) article On assertive predicates, where five English predicate classes (weak and strong assertive, non-assertive, true factive, and semi-factive) governing clauses are distinguished on semantic and pragmatic grounds. The predicates belonging to the first, second, and fifth classes can be postposed, i.e., used as parentheticals. Hooper claims that the meanings of the predicates govern the syntactic processes that the predicates and their depending clauses may undergo. In the final section of her article, the classification is presented and discussed with respect to Spanish. Table 1.4 lists Hooper's three classes of predicates that can be used in parenthetical position.


REDUCED PARENTHETICAL CLAUSES Weak assertive Strong assertive creer decir pensar contar estar seguro decidir ser verdad afirmar ser cierto explicar Semi-factive darse cuenta enterarse saber ver comprender Table 1.4: Classification of Spanish parenthetical predicates (see Hooper 1975: 121f.)

Hooper's list of Spanish parenthetical verbs is small, tentative, and not based on a corpus of spoken or written language since she was essentially looking for facts confirming her semantic-based classification of English predicates that govern clauses. It is, however, the first approach to Spanish RPCs, and in some aspects my analysis is a continuation and expansion of Hooper's pioneering work. 1.4 Outline of the book In sections 1.1 and 1.2, I defined and delimited the subject of the analysis, briefly mentioned the discourse functions RPCs assume in utterances, and illustrated the numerous terms which have been proposed to refer to RPCs. In section 1.3, I presented some open issues raised by RPCs. Part of them pertain to general linguistics, part of them to Romance linguistics. Most of them will also be touched upon in the research history: the relationship between performatives and parentheticals, the linguistic unit they represent, the question of how their syntactic function or class is brought about, the semantic and pragmatic conditions under which these clauses in governing position can have a mitigating reading, the question concerning the specific prosodic properties distinguishing them from other instances of parenthesis, and the lack of a fairly exhaustive corpus-based list of French, Italian, and Spanish RPCs. Five chapters are dedicated to the analysis of reduced parenthetical clauses from the perspective of a specific linguistic level, namely the pragmatic level, the semantic level, the syntactic level, and the prosodic level, while the other chapters are introductory or they combine more than one analytical perspective. Since I consider the pragmatic approach to RPCs the most fruitful and insightful, more space is dedicated to this perspective. Chapter 2 investigates the status of the linguistic concept of parenthesis and provides the background necessary to place RPCs within a wider context. It reviews the various definitions available for the notion of parenthesis and the descriptive criteria adduced to support these. After a thorough examination of the most important criteria or properties involved, I conclude that the concept, as it has been used so far, is hardly useful, since it entails a never-ending cycle of misunderstandings. Yet, the exploration of the criteria firstly shows that



Urmson's (1952) formal criteria are insufficient and secondly provides a number of elements for the selection of RPCs from the corpus texts. Chapter 3 contains an account of previous research on the subject. It starts with thoughts and contributions formulated before Urmson (1952), continues with an introduction to his seminal article and afterwards illustrates the most important linguistic studies dedicated specifically to RPCs which started to be produced around 1970. Apart from the discourse function of parenthetical clauses, which always constituted the main issue, these latter works addressed a host of additional questions: RPCs and the Romance subjunctive, their behavior with respect to negative raising, the use of these clauses as answers to polar questions, the properties shared by modal adverbs and mitigating RPCs, the relationship between parenthesis on the one hand and parataxis and hypotaxis on the other, the level of linguistic representation at which one should deal with RPCs, and the difference in truth conditions between these clauses in governing position and in incidental position. Chapter 4 starts with an illustration of the empirical approach and the data on which the research is based. The 22 Romance corpora used in the study are briefly presented. In the following section, I explain and discuss the selection criteria, often by taking up the criteria already investigated in chapter 2. I then proceed to illustrate how the RPCs were selected from the corpora and which RPCs I eventually found for each language. Finally, I examine the frequency of these previously-defined RPCs by taking into account the discourse situations and domains of use in which the texts were produced as well as the degree of monologicity of discourse. In chapter 5, I outline the pragmatic framework according to which the RPCs selected from the corpora are analyzed. Two heuristically useful distinctions are at the center of the chapter: Hare's (1970) phrastic, tropic, and neustic, as the fundamental components contributing to the meaning of an utterance, and Caffi's (1999, 2001) classification of mitigators in bushes, hedges, and shields. The chapter also addresses the question of how and whether at all performative clauses can be distinguished from RPCs. Initially, I propose to distinguish between two types of performatives: those that are obligatory constituents of the speech act, which I call saying-is-doing clauses, and those that are illocutionary force indicating devices but not necessary elements of the speech act performed, which I call saying-and-doing clauses. I then summarize the most important differences between performatives and parentheticals. Finally, I discuss Urmson's (1952) and Venier's (1991) point of view, who affirm that performatives in parenthetical position lose their performative function. In the following chapter, I distinguish, based on the deictic orientation and the sentence type (declarative, imperative, interrogative), RPCs alleviating speaker responsibility from those removing and those dividing speaker responsibility. I also mention briefly a group of addressee-centered imperative and interrogative clauses that do not reduce speaker responsibility but invite the address-



ee to cooperate and control the effectiveness of communication. Hare's (1970) and Caffi's (1999, 2001) distinctions and classes are then applied to Romance RPCs. Based on the logical structure of utterances and the scope of RPCs, I identify five main classes: 1. RPCs mitigating the phrastic (propositional content); 2. RPCs indicating the tropic (illocution) and mitigating the phrastic or the neustic (speaker commitment); 3. RPCs directly mitigating the neustic; 4. RPCs indirectly mitigating the neustic; 5. RPCs reporting discourse. Several RPCs have more than one component in their scope. That is, they may alternate between two scopes (typically phrastic or neustic) or operate on two components at the same time (e.g., tropic and phrastic, tropic and neustic). The analysis shows that, although many RPCs occur in statements, a number of others also occur in questions or directives. Moreover, reducing speaker responsibility is not their exclusive discourse function. Besides the phatic function, one has to take into account several other functions, e.g., highlighting expansions, explanations, repetitions or self-corrections, focusing, and indicating the illocution. The illustration is complemented by a synoptic table of the functions of French, Italian, and Spanish RPCs. The next chapter discusses several issues at the interface between syntax, semantics, and pragmatics. Regarding their semantic characterization, the RPCs found in the corpora contain verbs belonging to the following semantic macrotypes: 1. verbs referring to the act of speaking, predominantly derived from decir; 2. doxastic verbs (belief verbs); 3. verbs referring to inferences and other mental operations; 4. verbs expressing sensory or mental sensation or perception; 5. epistemic verbs, i.e., verbs referring to the mental state of knowledge, derived from saber. Other issues dealt with are the influence of person, time, and mood, the use of RPCs as answers to polar questions, and the question of whether RPCs are restricted to declarative sentences. Chapter 8 starts with a brief discussion of the notions syntactic unit and syntactic function and reviews the possible syntactic units RPCs could be assigned to (sentence, clause, phrase, word). The following section analyzes the positions RPCs may occupy within their host. In section 8.3, I defend the view that, in sentence-initial position and in the absence of a complementizer, RPCs do not govern the following clause. Section 8.4 is dedicated to questions of scope. I maintain that the scope of an RPC is largely predetermined by its pragmatic function and the structure and meaning of the host. Finally, I discuss the pragmatic differences between governing clauses and parenthetical clauses. Rather than to ask which linguistic phenomena signal a mitigating function, I



deem it more appropriate to ask which phenomena indicate that a main clause assumes an assertive function. The description of the formal aspects of RPCs is completed in chapter 9, where prosodic properties, namely intonation, pauses, increased rate of speech, length, etc., are discussed. The main question is the phonological status of these clauses, i.e., whether they constitute separate prosodic segments or are just attachments to other segments. The findings of the study are summarized in the concluding chapter.

PARENTHESIS: A PROBLEMATIC CONCEPT 2.1 Origins and modern use The Classical Greek word parénthesis, a compound of para 'next to, by the side of', en 'in, into' and a nominalization of the verb tithénai 'put, place', was originally used in rhetoric to refer to the insertion of a thought into a continuing discourse. In 9.3.23 of Institutio Oratoria, the most influential description of classical rhetoric written during antiquity (95 B. C.), Quintilian mentions two Greek terms, parénthesis and parémptôsis, and two Latin ones, interpositio and interclusio, and writes that they are used when "[...] continuationi sermonis medius aliqui sensus intervenit", i.e., some intermediate meaning or thought (medius aliqui sensus) intervenes in a continuing discourse. A casual idea may be inserted before, during or after the main thought; to be precise, only the second one of these possibilities is denoted by these terms. In the past, scribes and printers adopted various graphic means to mark these insertions, one of them being a pair of brackets. Hence, during the development of letterpress printing, the word parenthesis began to be used also for 'pair of (round) brackets', a meaning now found in English as well as in the Romance languages. While the Greek parénthesis implies an intentional and conscious insertion, parémptôsis, composed of para 'next to, by the side of', en 'in, into' and a nominalization of the verb píptein 'fall', underlines the casual, incidental and unplanned way by which an optional thought may come to mind while another thought is evolving. This second term later fell into disuse, probably because it was not consistent with the concept of the rhetorical figure. In classical rhetoric, a figure is a form of expression in which the normal use of language is intentionally manipulated, extended or altered in order to enhance the effect a speech may have on the audience. It is important to note that parenthesis or paremptosis was considered a figura sententiae, that is, a figure of thought or a semantic figure theoretically independent from any particular linguistic form (see Lausberg 1960: 375). This also implies the idea that the inserted thought should be expressible by a complete sentence.6 The corresponding figura elocutionis, i.e., the formulation in the speech, is the hyperbaton. The boundaries between figures of thought and figures of speech were not as clear in classical rhetoric as one might think, therefore it cannot be said that the hyperbaton was always the exact linguistic equivalent of the parenthesis. In several rhetoric treatises another distinction seemed to be more important: Parenthesis was used for the insertion of a complete sentence into another one, "Die interpositio 'Parenthese' ist die konstruktionsfremde Zwischenschaltung eines Satzes (und damit eines Gedankens) in einen Satz." (Lausberg 1960: 427) 6



while hyperbaton was used when the normal word order of a sentence was disrupted by the insertion of sentence parts.7 In modern linguistics, a number of terms are in use for the phenomenon in question. Particularly the German language abounds with such expressions. In grammars, introductions to linguistics, dictionaries of linguistics, and works on the subject in general, in addition to Parenthese one may find Einfügung (see Homberger 2000: 373), Einschaltung (see Behaghel 1928: 537ff.; Conrad 1988: 61), Einschub (see Winkler 1969; Erben 1972: 308f.; Lewandowski 1985: 752; Abraham 1988: 566; Zifonun, Hoffmann, Strecker, et al. 1997: 2214), Schaltsatz (see Helbig & Buscha 1989: 648f.; Bußmann 1990: 665; Homberger 2000: 373). The general notion Zwischensatz (see Conrad 1988: 281; Homberger 2000: 651) sometimes explicitly includes parenthetical clauses (see Wackernagel 1897: 21-27). The meanings of these terms are not exactly the same; moreover the meaning of every single term may vary from author to author. García Cid (1995), whose working paper about parenthesis in spoken Spanish is written in German, describes two different types of parenthesis, one called Parenthese, the other Einschub. For Hoffmann (1998: 300), a Schaltsatz is an RPC. Moreover, whereas Schaltsatz and Zwischensatz imply that the parenthetical must be a sentence or a clause, other terms do not imply such a restriction. If they mention it at all,8 English texts on the subject limit themselves to parenthesis (see Matthews 1997: 265f.), parenthetic clause (see Jespersen 1937), parenthetical clause (see Emonds 1976), parenthetical structure (see O'Grady, Dobrovolsky, & Katamba 1997: 669) or simply parenthetical (see O'Grady, Dobrovolsky, & Katamba 1997: 670; Biber, Johansson, Leech, et al. 1999: 137 f.; Burton-Roberts 2006). Incidental clause / expression / construction, etc. although possible as linguistic terms,9 do not occur in English dictionaries of and introductions to linguistics. In French, incise and the related terms (élément / proposition / sous-phrase) incident(e) and incidence10 are more frequent than parenthèse and parenthétiques (see Mounin 1974: 171f.; Robert 1985; Delomier & Morel 1986; Grevisse 1993; Dubois, Giacomo, Guespin, et al. 1994: 242, 344; Nølke 1996). 11 Occasionally, one may also find proposition intercalée and intercalation (see Mounin 1974: 171). Often parenthèse merely denotes the typographic device, i.e., the brackets, whereas parenthétique, incise, or incidente refer to the inserSee Lausberg 1960: 428: "[...] das eigentliche Hyperbaton [...] ist ein Satzteil-Hyperbaton, die Parenthese ist ein Gedanken-Hyperbaton (Ganzsatz-Hyperbaton)". We find the same distinction in Quilis (1999: 443-445). 8 Crystal (1997), for instance, contains no entry for the subject. 9 At least, that is what I conclude after a search in the World Wide Web, where some linguistic documents containing the sequence incidental clause may be found. 10 In Guillaume's theory of tense and aspect, the terms incidence and incidentielle have nothing to do with parenthesis; cf., e.g.: "Deux phases sont à envisager dans le développement de la perspective ouverte par la conjonction. La première se rapporte à la venue du procès dans le temps: c'est la phase incidentielle; la seconde, aux conséquences du procès survenu, ou censé survenu dans le temps: c'est la phase conséquentielle." (Guillaume 1929: 42) 11 In Mounin's Dictionnaire de linguistique (1974), there is no entry for parenthèse. 7



ted expression itself (see Dubois, Giacomo, Guespin, et al. 1994: 242, 344; Wilmet 1998: 575f.; Riegel, Pellat, & Rioul 2002: 95f, 460f.). In addition, French texts commonly differentiate between medial or final quotatives, i.e., devices reporting something said by another person, which are called incises, and other types of incidental constructions, often called incidentes (see Mounin 1974: 171f.; Grevisse 1993: 568ff.; Robert 1985; Wilmet 1998: 574; Riegel, Pellat, & Rioul 2002: 460f.). The first ones are reduced clauses, the second ones are often complete clauses. Italian and Spanish linguistic terminology does not share this differentiation. Italian linguists speak of inciso and (proposizione / frase) incidentale / parentetica (see Serianni 1991: 625f.; Cardona 1988: 166, 229; Beccaria 1996: 385; Dardano & Trifone 1997: 427f.; Borgato & Salvi 2001: 165ff.). Parentesi is rarely employed; in Italian dictionaries of linguistics it is only used to refer to 'pair of (round) brackets' (see Beccaria 1996: 544; Cardona 1988: 229). Quite similarly, in the Spanish linguistic literature one may find inciso (see Lázaro Carreter 1968: 233; Bosque & Demonte 1999), inciso oracional (see Alarcos Llorach 1994: 317), oración incidental (see Lázaro Carreter 1968: 233; Real Academia Española 1991: 150), simply incidental (see Fernández Fernández 1993;12 Alcaraz Varó & Martínez Linares 1997: 297), oración parentética (see Lázaro Carreter 1968: 314), and enunciado parentético (see Quilis 1999: 445; Fuentes Rodríguez 1999); paréntesis13 (see Lázaro Carreter 1968: 314; Navarro Tomás 1974: 82ff.; Alcina Franch & Blecua 1991: 469) may also occur in reference to these constructions. From a communicative point of view, Vigara Tauste (1992: 397ff., 423ff.) distinguishes between incisos de opinión, with which the speaker declares that the proposition expresses his or her personal opinion and that he or she is preparing himself or herself for a possible refutation, and paréntesis asociativo, which is a momentary, syntactically unconnected insertion into an information block. As in classical times, in modern linguistics parenthesis continues to be a rather flexible concept (for a recent overview, see Kaltenböck 2005: 25ff.). There is a general tendency to see it not as a discourse technique, but as a concrete piece of language. The title of Schwyzer (1939), Die Parenthese im engern und im weitern Sinne, shows this flexibility, implying that there are a narrow and a broad definition of parenthesis. Schwyzer, at one point, defines it narrowly, saying that it is a grammatical main clause that interrupts the grammatical continuity or coherence of another sentence.14 The variety of the phenomena examined in his treatise suggests, however, a broad definition.15 According to the narrowest characterization one may find, a parenthesis is, very much as in The definition of incidental in Fernández Fernández (1993), though, is rather specialized and limited to incidental adjectives, nouns, and adverbs, as well as to incidental adverbial clauses with an overt syntactic link. 13 In Spanish, the word paréntesis is masculine, not feminine as in French or Italian. 14 See Schwyzer (1939: 31): "Vom formal grammatischen Standpunkt ist eine Parenthese ein grammatischer Hauptsatz, der den grammatischen Zusammenhang eines andern Satzes unterbricht". 12



Schwyzer (1939: 31), a syntactically complete and autonomous sentence that interrupts the syntactic coherence of the sentence into which it is inserted (see also Faulseit & Kühn 1969: 155; Abraham 1988: 566).16 Very broad definitions, those based only on prosody or communicative function, without determining either its syntactic class or its position in the host, include constructions that do not match our intuitive understanding of parenthesis: interjections, vocatives and other forms of address, left or right dislocated elements, appositions, certain elliptical expressions, adjectival and participial elements appended to noun phrases, relative clauses, and communicative formulas; at one point, Schwyzer (1939: 45f.) even sees a relationship with cases of anacoluthon, i.e., with sentences that switch from one construction to another. Most definitions fall in between these two extremes. There is nevertheless a core group of constructions which are subsumed under parenthesis by most authors, while there is no general agreement on the classification of a number of other constructions. Most definitions do not specify whether they apply to spoken or written language. There are, of course, some contributions explicitly based on spoken corpora; those based on written documents are dedicated to the use of parenthesis in literary texts in especially Sanskrit, Classical Greek, and Latin (e.g., Wackernagel 1897; Schwyzer 1939). 2.2 Principal description criteria In the following overview, I will present the most important criteria on which descriptions and definitions are founded. Exploring and discussing them is essential for the definition of my selection criteria in chapter 4. At least six criteria or properties are involved, both in spoken and written language: prosody, interruption of the host clause, lack of an overt link between parenthetical and host clause, sententiality, the relationship between parenthetical and host clause, and communicative function (cf. also Lampert 1992: 35ff.). They cover almost all levels of linguistic analysis from phonology to pragmatics and are those most often mentioned when parenthesis is defined. 2.2.1 Prosody Prosodic features, primarily fundamental frequency (F0), intensity, rate of speech, and pauses, are considered important by almost all authors.17 Variations in F0 recurring over stretches of speech are associated with patterns. These distinctive patterns, which are sometimes associated with particular meanings, are called intonation. Many authors assume that parenthetical constructions display See Schwyzer (1939: 8): "Die Parenthese im landläufigen Sinne ist nur eine Seite einer viel umfassenderen Spracherscheinung, zur der auch manches gehört, was unter andern Namen geht". 16 Faulseit & Kühn (1969: 155) use Aussage instead of Satz. A literal interpretation of Aussage would restrict the meaning of parenthesis even more, i.e., to the insertion of a declarative sentence. 17 There are some notable exceptions to this: prosody plays no role in Hooper (1975) and Venier (1991); Urmson (1952) does not even mention the matter. 15



such a distinctive pattern. In fact, terms such as intonation parenthétique (see Nølke 1996: 320; Campione 2001a: 377; Riegel, Pellat, & Rioul 2002: 460), parenthèse intonative (see Rossi 1999), incise prosodique (see Simon 2004: 188-196, 225-249), intonazione parentetica (see Borgato & Salvi 2001), entonación del paréntesis (see Navarro Tomás 1974), or Parenthese-Intonation (see Helbig & Buscha 1989: 676) occur frequently. The level of pitch or of its acoustic correlate, the fundamental frequency, is considered to be the most important property of the parenthetical intonation. Navarro Tomás (1974: 83) describes the entonación del paréntesis as being six or seven semitones lower than the intonation of the rest of the sentence. According to Bolinger (1989: 186), English parentheticals are spoken with a lower pitch than the matrix sentence and with a rising terminal; final parentheticals display a low level pitch. Cruttenden (1997: 123, 140, 173) adds to these properties a reduced variation in pitch range and a reduced intensity. Quilis (1999: 445) states that the Spanish medial reporting parenthetical me contaba Juan is pronounced with a low pitch that rises towards the end. Canepàri (1999: 224f.) confirms this, although he assumes that the F0 of parentheticals is not completely flat (= absence of modulation) but displays - on a reduced scale - the intonational patterns possible in Italian. The few parenthetical constructions specifically analyzed and described in Campione (2001a: 375ff.) display an "intonation plate en plage basse", i.e., a low-level flat intonation. Pauses, although possible before and after a parenthesis, are often not treated as being an essential prosodic property. Not all researchers, however, are convinced that parentheticals display a low-level flat intonation. Nemni (1979: 108), in an experimental study on the intonation of French parentheticals, found out that F0 varies considerably throughout the parenthetical sequence and might even reach levels higher than those of the host (cf also Blanche-Benveniste 2000: 122; Cresti 2000a: 143). The aforementioned distinctive patterns of variation in F0 recur over stretches of speech that I call, following Hirst & Di Cristo (1998), intonation units.18 In phonology, the segmentation of utterances in intonation units is a much debated issue. There is no general agreement among phonologists on the phonetic correlates of the boundaries of these units (e.g., pauses, initial shortening, final lengthening, pitch of unaccented syllables) or the factors (e.g., length, syntactic unit, semantic coherence) defining intonational grouping. The majority of researchers nevertheless agrees that at least medial parenthetical expressions Due to the many perspectives from which this linguistic unit can be considered, the terminology is fairly diverse in this respect. In English publications, one may also find intonation group, intonational phrase, phonological phrase, tone group, tone unit or breath group (see Cruttenden 1997: 29; Hirst & Di Cristo 1998: 18f:). French terminology is similarly prolific; besides unité intonative (see Campione 2001a), there are at least groupe rythmique, mot phon(ét)ique, groupe prosodique, groupe de souffle (see Meisenburg & Selig 1998: 121), and groupe intonatif (see Simon 2004: 65ff.). Cresti (1987: 39ff.; 2000a) calls it unità tonale, Voghera (1992: 90ff.) gruppo tonale,, Nespor (1993: 205ff.) sintagma intonativo, Canepàri (1999: 191ff.) intonía, Quilis (1999: 419f.) grupo de entonación. 18



constitute separate intonation units (see Selkirk 1984: 382; Nespor 1993: 206; Cruttenden 1997: 71). Only a few specific accounts of the prosody of parenthesis are based on corpora of spoken language. Some of these adopt prosody as a primary criterion of definition. According to Winkler (1969: 282ff.), in spoken German one recognizes an incidental construction by the fact that the speaker all of a sudden changes the intonational contour of his or her utterance by making a marked drop in pitch and then, after the interruption, continues by resuming the previous pitch level. Usually, he or she accompanies this particular contour by an increased rate of speech, a short pause before and a longer pause afterwards. Interestingly, after the parenthesis the pitch returns to exactly the same level and direction (falling or rising) as before. For Delomier & Morel (1986: 142f.), who examine the prosody of French parentheticals, the general characteristics are the so-called décrochement, i.e., the detaching of the intonational contour, a usually lower pitch that rises slightly towards the end, an increased rate of speech, and pauses before and especially after the parenthetical. This is confirmed by Morel & Danon-Boileau (1998: 59ff.). According to their study of spoken French, incidental sequences feature the lowering of the fundamental frequency (F0) to level H2 (for the description of the levels of F0 see section 9.1), the absence of modulation of F0, frequently the acceleration of the rate of speech and the maintenance of the level of intensity, and the rising of F0 towards the end. Interestingly, they do not mention pauses as a constitutive feature. Morel & Danon-Boileau's (1998) study is exceptional insofar as they are the first to develop a coherent framework in which the incise is treated separately from expressions like je crois, j'espère, il me semble, disons, tu vois, moi je trouve, and je sais pas (see section 9.1). Cresti (2000a: 143) states that parentheticals usually have a lower F0 than the host, but it may also be the case that the F0 is higher. The detaching from the host is further underlined by a lower or higher rate of speech. Firenzuoli & Tucci (2003: 188) and Tucci (2004: 10) mention the lowering of F0, an increased rate of speech, and a flat intonation profile. Both authors add that the mean length of syllables in parentheticals (0.142 ms) is significantly shorter than elsewhere. According to Simon (2004: 188, 225ff.), the décrochage mélodique may consist in a lowering or a raising of the F0. It may be accompanied by pauses, an increased rate of speech, and a decrease in intensity. Winkler (1969) and Delomier & Morel (1986) emphasize that parentheticals may not manifest all of these properties in the same way (cf. also Bolinger 1989: 186); some may be lacking, e.g., pauses, increased rate of speech, some may be realized differently, by, e.g., F0 as a rise instead of a drop. An important conclusion can be drawn from the examples of Einschub, incise,19 and inciso referred to in Winkler (1969), Delomier & Morel (1986), MoIn Delomier & Morel (1986) and Morel & Danon-Boileau (1998) the term incise refers to incidental constructions in general, not only to incidental quotatives. 19



rel & Danon-Boileau (1998), Cresti (2000a), Firenzuoli & Tucci (2003), Tucci (2004), and Simon (2004): Practically any syntactic construction may be marked in speech by the aforementioned prosodic properties (see also Dessaintes 1960: 72-97, 102-112; Delattre 1966: 5, 8, 12f.; Alisova 1972: 181ff.; Nølke 1996: 317f.; Cruttenden 1997: 71; Canepàri 1999: 224f.; Borgato & Salvi 2001: 165). This is the case, for example, for constructions that are either clause-final or cannot otherwise be said to interrupt the syntactic coherence of the host. Winkler (1969: 289) even quotes a parenthetical that is a free sentence. Several of the aforementioned sources cite parentheticals that contain an overt syntactic link to the host clause. Furthermore, some of the mentioned parentheticals do not contain a predicator, i.e., they are simple noun phrases or prepositional phrases (see especially the table in Winkler 1969: 290; see also Delomier & Morel 1986: 145f.). Regarding the incise, Morel & Danon-Boileau (1998: 60) and Simon (2004: 189) explicitly confirm this conclusion. In Wunderli (1987) and Rossi (1999), the parenthetical intonation is a characteristic of all detached constructions (e.g., topics dislocated to the left or right, appositions, modal adverbs, RPCs, and others). 2.2.2 Interruption of the host clause Interruption of the host clause or sentence20 or position within the host is the second property typically adduced to describe parenthesis. Occasionally, it is considered the defining factor. As will be explained in section 3.1, for Bloomfield (1935: 186) and for Behaghel (1928: 493, 537ff.), parenthesis is a variety of parataxis in which one form interrupts the other. In both cases, two forms are united by the use of only one sentence pitch. The only difference between parataxis and parenthesis thus is position. Parentheticals are clause-medial: (1)

Won't you please come?

In Bloomfield's opinion, one of the two members of a clause-medial appositional group is a parenthetical, as the poor boy in the following example:21 (2)

John the poor boy ran away.

As becomes clear from these examples, in Bloomfield's conception parenthesis is not dependent on the syntactic class (sentence, clause, phrase, word) of the interrupting expression. According to Matthews (1997: 265), the host does not necessarily have to be either of the two. In his words, a parenthesis is "a syntactic unit which interrupts a larger unit". Previously, Espinal (1991: 728) had expressed the same opinion. 21 1998, in a personal communication, W. Mayerthaler criticized Urmson's (1952) term parenthetical verb as syntactically useless and stated that the construction in (ii) is appositive not parenthetical: (i) Your house is, I suppose, very old. (ii) Your house is very old, I suppose. 20



Although I treat the two aspects together, I am of course aware that medial position is not the same as interruption, if this means interruption of a close syntactic relationship or of a word order pattern. Please in (1) is medial, but does not actually interrupt a close syntactic relationship or a word order pattern. Medial position is recognizable immediately, whereas interruption is a less straightforward issue. As Havers (1931) suggested, associative thinking, a thinking mode in which imaginations and impressions are ordered loosely without being limited by sequence rules, is responsible for parenthesis. Zifonun, Hoffmann, Strecker, et al. (1997: 2364) state that parenthesis is above all a device that operates with linear order and makes it possible to structure information in a way not provided for by the word order rules of a language (cf. also Lampert 1992; Hoffmann 1998; Burton-Roberts 2006). While illustrating German word order, they discuss several clauses which in their framework have to be parenthetical only because of their exceptional position (Zifonun, Hoffmann, Strecker, et al. 1997: 1580f.). Espinal (1991: 730ff.) is actually expressing the same thought when, in mentioning German word order, she writes that the syntactic independence of the parenthetical and its lack of relations to the host is demonstrated by the fact that it does not exert influence on the latter's word order. In German declarative sentences, finite verbs occur in second position, so there must be another constituent (e.g., the impersonal pronoun es or a temporal adverb) in initial position: (3) (4)

Es wurde gestern getanzt. [Somebody danced yesterday] Gestern wurde getanzt. [Yesterday somebody danced]

Whereas a second preverbal constituent is ungrammatical, a presentential parenthetical occurring before the first preverbal constituent is acceptable: (5) (6)

*Gestern es wurde getanzt. [Yesterday, somebody danced] Ehrlich, es wurde getanzt. [Honestly, somebody danced]

Mayerthaler, Fliedl, & Winkler (1998: 262) and Ackema & Neeleman (2004: 97) cite similar German and Dutch examples that underline the extra-sentential status of parentheticals. Apart from Espinal (1991) and Zifonun, Hoffmann, Strecker, et al. (1997), I know of no other relevant work on the matter that has gone beyond general statements and examined in practice the possibility of defining parenthesis through word order.22 Interestingly enough, Quintilian, when defining parenthesis in 9.3.23 of Institutio Oratoria, explicitly refers to the "continuationi sermonis", in which the parenthesis intervenes, although classical Latin had rather free word order. 22



García Cid (1995: 13f.) defines Parenthese explicitly as discourse unit that interrupts syntactic coherence. There is no clue, though, as to how this coherence is to be understood. Moreover, the author on purely graphic grounds excludes certain expressions that could be classified as interruptions (e.g., vocatives and other forms of address, interjections, communicative formulas). 23 Einschub, on the other hand, is a digression, excursion, or aside that interrupts semantic or thematic coherence, introducing a unit belonging to one discourse type into a portion of another discourse type (1995: 19ff.). The classification of discourse types is taken mainly from Grimes (1975). The thematic digression may last for one or more predications. In the texts analyzed, all Einschübe express information that is non-sequential with respect to the time line or, as García Cid puts it, secondary predication. Parenthese and Einschub, as defined by García Cid, do not necessarily coincide. Probably as a result of Urmson's (1952) influence, many contributions of the last 50 years promote the view that clause-medial position is not a necessary condition, which contradicts the original meaning of parenthesis. However, this opinion is also found in works preceding Urmson (1952) (e.g., Spitzer 1922; Schwyzer 1939: 34) or coming from a different tradition, as Grevisse (1993: 569). 2.2.3 Lack of an overt link The third criterion I want to mention is the lack of an overt link between parenthetical and host. In Bloomfield's (1935: 171, 186) understanding, parataxis and parenthesis are syntactic relations between successive or interrupting units marked only by intonation, that is, without an overt linking device. Bayer (1973: 74) excludes any sequence overtly linked to the host clause from his study of parenthesis in spoken German, even if its communicative role is comparable to that of an asyndetic sequence. Sommerfeldt (1984: 243), Helbig & Buscha (1989: 648f.), and Borgato & Salvi (2001: 165f.) state likewise that parentheticals lack an explicit syntactic relationship between them and the host. For other authors, though, an incidental construction does not necessarily have to be asyndetic.24 I have already quoted examples of Winkler (1969), Delomier & Morel (1986), and Morel & Danon-Boileau (1998), where parentheticals contain an overt syntactic link. Hofmann (1926: 115) states that explanatory parentheses introduced by conjunctions such as nam, enim, and etenim were frequent in spoken Latin. Although Schwyzer (1939: 31) believes that originally parenthesis was asyndetic, he explicitly mentions the possibility that it may be linked to the host clause by a particle or a conjunction, a construction which was frequent in Classical Greek and Latin. The most striking example of a definSee García Cid (1995: 14): "Im Gegensatz dazu gilt in der vorliegenden Arbeit eine engere Definition von Parenthese als Grundlage. Diese umfasst hauptsächlich nur jene Diskursteile, die in der geschriebenen Sprache als Einschaltung mitten im Satz, umgeben von Klammern oder Gedankenstrichen, erscheinen würden". 24 Similarly, Lehmann (1988: 210ff.) holds that the presence or absence of a connective device between two clauses has nothing to do with parataxis vs. hypotaxis. 23



ition where asyndesis does not play a role is Urmson's (1952). In his eyes, even sentence-initial main clauses with a complementizer can be parenthetical clauses.. Serianni (1991: 625) distinguishes between incidentali primarie with a conjunction and incidentali secondarie with a subordinating or coordinating conjunction. In a corpus-based account of the function of parce que in contemporary spoken French, Debaisieux (1994) dedicates a whole chapter to incidentals introduced by this conjunction and cites numerous examples. Similarly, Dardano & Trifone (1997: 428) emphasize that parenthetical clauses may occasionally contain a subordinating conjunction. Zifonun, Hoffmann, Strecker, et al. (1997: 1581), whose main criterion is position, repeatedly quote incidental clauses with overt syntactic links. 2.2.4 Sententiality Sentential or clausal character is the fourth factor occasionally named in the description of parenthesis. According to many authors, parentheticals are complete or reduced clauses. Terms like parenthetic(al) clause, proposition incidente, frase incidentale / parentetica, inciso oracional and oración incidental all presuppose this. For those German linguists who share this standpoint, Parenthese and Schaltsatz are synonyms (see, e.g., Homberger 2000: 373), for those who do not, Parenthese has a much broader meaning. The sentential character never, of course, constitutes the sole criterion, but implicitly it plays a role in many accounts. The narrow definitions of parenthesis I referred to above see it as a necessary condition. Even authors who do not explicitly mention it often cite sentences or clauses as prototypical examples of parenthesis (see Schreiter 1988; Matthews 1997: 265f.; Borgato & Salvi 2001). The opposing standpoint is equally diffused. In phonological approaches, any syntactic class may possess the properties that characterize parentheticals (see Winkler 1969; Delomier & Morel 1986; Morel & Danon-Boileau 1998; Cruttenden 1997: 71; Borgato & Salvi 2001: 165). In Bloomfield (1935: 186), any type of expression (sentence, clause, phrase, word) may be parenthetical. In Grevisse (1993: 568ff.), the nature of the éléments incidents is extremely variable and comprises most constructions from word to clause. Similarly, Espinal (1991: 727ff.) writes that parentheticals, considered in isolation, are not characterized by a particular structure (see also Kaltenböck 2005: 27); they may be sentences, noun phrases, adjectival phrases, prepositional phrases or adverbial phrases. The fact that defines them is that they are disjuncts with respect to some other syntactic structure, that is, their nonconfigurationality with regard to a host structure. Espinal adds that the parenthetical contains less grammatical information than the host. Unfortunately she does not pursue the idea that parenthetical expressions, though not definable by themselves, are grammatically, semantically, and pragmatically reduced with respect to the host sentence. That would mean that they can never be full-fledged sentences.



It is not clear, though, in which respect a parenthesis is pragmatically reduced. According to Bassarak (1987) parentheses have autonomous illocutionary forces, which are coordinate, subsidiary, or autonomous with respect to the illocution of the host. Schreiter (1988: 126) also states that the host and the parenthetical clause may have different illocutions. García Cid (1995) assumes that parentheses fulfill autonomous communicative functions. Similarly, Debaisieux (1994) says that a parenthesis may have a modality that is distinct from that of the host. There is no doubt that, at least with respect to modality, the parenthesis and the host may differ: (7) (8)

Inf. B.- Me quitó tres años, oye, de la manera más idiota. […] (HCM.18.331) [I gave up three years, listen, in the most stupid way] Y cazuelas, cazuelas... sí. Y había cazuelas, ¿sabes? antes de estas de barro [...] (COREC.CCON021B) [And saucepans saucepans yes and there were saucepans, you know, before these of clay]

In the Spanish examples above, the declarative sentence hosts an imperative and an interrogative parenthetical, respectively. 2.2.5 Relationship with the host clause Related to the question of an overt link is the one regarding a possible syntactic and / or semantic interaction between the parenthetical and the host. Though never adduced as a defining criterion, the type of relation has been discussed repeatedly. According to Espinal (1991: 735), a number of arguments show that parentheticals are syntactically independent constituents. They do not participate in syntactic processes that take place within the domain of the sentence, and they are not subcategorized by verbs. Likewise, at the level of semantic representation (Logical Form in generative terms), the meaning of the parenthetical is not integrated into the meaning of the host. Only at the moment of utterance processing, that is, at the pragmatic level, is an interaction between the parenthetical and the host taking place. To a certain degree, all linguists share this opinion, that is, all assume that parentheticals, though independent from the host at a lower level of representation, maintain some kind of relationship with the host at a higher level (see, e.g., Schreiter 1988: 125). The disagreements regard the point at which such a relationship becomes active. Schreiter (1988) and Espinal (1991) set it rather high, i.e., not before the pragmatic level, other linguists (e.g., Ross 1973; Emonds 1976; Zifonun, Hoffmann, Strecker, et al. 1997) posit overt or covert relations at lower levels. As Blakemore (2006: 1670ff.) explains, this has even led to a distinction between parentheticals that are regarded as performance phenomena



('disfluencies') and parentheticals that can be accommodated within grammar. Any solution to the problem also depends on the approach regarding the 'division of labour' between grammar and pragmatics (see Blakemore 2006: 1672). If pragmatics penetrates into grammar, parentheticals can be accounted for earlier and at lower levels; if grammar is autonomous from pragmatics, the interpretation of parentheticals will have to wait until the level of discourse interpretation is reached. Unfortunately, the question has rarely been posed in such clear terms as in Espinal (1991) and Blakemore (2006). Usually the question that attracted most attention was the relationship, hypotactic or paratactic,25 between the host and the parenthetical. While there is no doubt that hypotaxis presupposes a relationship on the syntactic level, authors disagree on the subject of parataxis. Some consider it a syntactic relationship (see Bloomfield 1935; Mayerthaler, Fliedl, & Winkler 1998: 262), some do not (see Matthews 1981; Lepore & Loewer 1989). Authors of the second group may see a difference between coordination and parataxis (see Matthews 1981: 220; López García 1999: 3513), insofar as coordination presupposes a syntactic relationship, while parataxis does not. It is also possible that some authors are unaware of the potential ambiguity of parataxis and, by interpreting parenthesis as parataxis, simply want to stress that parenthetical expressions have to be interpreted as being outside the domain of the host clause. Especially in older works on the subject, parentheticals are seen as paratactically connected to the host: Diez (1872: 340ff.) and Meyer-Lübke (1899: 578) consider incidental Italian credo and prego as independent sentences; Wackernagel (1897: 21-27) also treats sentences asyndetically inserted into other sentences as cases of parataxis; Hofmann (1926: 114ff.) and Behaghel (1928: 493, 537ff.) share this view; for Bloomfield (1935: 186), parenthesis is just a variety of parataxis; In some initial generative accounts (see, e.g., Emonds 1976), sentences with parentheticals also form paratactic pairs. The nature of the syntactic relationship is frequently made dependent on the position of the parenthetical. For many authors (e.g., Jespersen 1914: 32; Nilsson-Ehle 1947; Wanner 1981; Hand 1991, 1993; Mayerthaler, Fliedl, & Winkler 1998: 262, 298), initial parentheticals lead to hypotactic structures (i.e., in reality, they are not parentheticals but governing clauses), while medial or final parentheticals lead to paratactic structures. This standpoint has been questioned by the so-called paratactic analysis of sentence-initial verbs of indirect quotation (see Davidson 1968-1969) and propositional attitude verbs (see Lepore & Loewer 1989), which was originally designed to apply to sentences with an explicit complementizer (e.g., Engl. that), but which can also be applied to sentences with complementizer deletion (see Thompson & Mulac 1991). According to Schreiter (1988: 124f.) this is, however, a futile question: "Parenthetische Strukturen [...] lassen sich nur unbefriedigend als komplexe Sätze erklären, weil sie sich weder völlig in den syntaktischen Rahmen einer subordinativen Beziehung noch in den einer koordinativen Beziehung fügen." 25



A number of authors suggest that there may be an implicit hypotactic relationship between the parenthetical and the host. Most of them consider the parenthetical as a subordinate. Damourette & Pichon (1933: 459f.), e.g., state that the incise in the following example is a subordinate: (9)

Celui-là, dit Lalie en montrant Bernard, n'est plus notre frère. [That one over there, says Lalie indicating Bernard, isn't our brother anymore]

According to the two linguists, its function as a subordinate is sometimes evidenced by the insertion of que in spoken French (cf. also Grevisse 1993: 574): (10)

Celui-là n'est plus notre frère que dit Lalie en montrant Bernard. [That one over there isn't our brother anymore, that says Lalie indicating Bernard]

Schwyzer (1939) does not express a clear opinion on the matter, but at one point (p. 44) he says that some parenthetical expressions stand in a subordinative relation to the host sentence. For Helbig & Buscha (1989: 648f.), a Schaltsatz formally seems to be an asyndetically coordinate sentence, whereas in reality it is a clause subordinate to the host clause; therefore it may be transformed into a syndetically subordinate clause: (11)


Die Prüfungstermine sie waren vorverlegt worden beunruhigten die Studenten.26 [The exam dates, they had been brought forward to an earlier date, worried the students] Die Prüfungstermine, die vorverlegt worden waren, beunruhigten die Studenten. [The exam dates, which had been brought forward to an earlier date, worried the students]

In their short remarks, Helbig & Buscha (1989: 648f.) do not pursue the question further, but they are certainly aware that there are parenthetical clauses that cannot be transformed into subordinate clauses. Borgato & Salvi (2001: 165f.) affirm that there is no explicit syntactic relationship between the parenthetical clause and the host. Semantically, though, the parenthetical clauses are subordinate to the host, but not governed by it. They function as modalizing adjuncts, as circumstantial adjuncts (= adjuncts expressing time, cause, etc.) or as appositive relative clause. Regarding parentheticals as in (11) and (12), which are essentially non-restrictive relative clauses, Hofmann (1926: 118) affirms that they were common in spoken Latin and historically represent the initial state out of which regular relative clauses, i.e., those with relative pronouns, developed. See also Lehmann (1988: 194). 26



The opposite view, however, can also be found. As will be explained in section 3.4, Ross (1973) assumes that the parenthetical is a superordinate clause in the deep structure and that its surface structure is derived by a transformation that deletes the complementizer and fronts the complement clause. According to Riegel, Pellat, & Rioul (2002: 470), the incises and incidentes in the following French examples semantically govern their respective hosts: (13) (14)

Quand, me demanda-t-il, reviendras-tu? [When, he asked me, will you return?] L'été, je le crains, sera chaud. [The summer, I'm afraid of it, will be hot]

Finally, according to Sommerfeldt (1984: 243), a Schaltsatz, though syntactically coordinate, may be semantically either subordinate or superordinate to the host clause. 2.2.6 Communicative function The last criterion I want to discuss is the communicative role a parenthetical sequence may have. Most authors at some point refer to the communicative function of parenthesis and try to describe it; few of them actually base their definition of parenthesis on it. One of these few seems to be Schwyzer (1939: 31, 33, 44): In cases of doubt, he suggests resorting to the meaning of the parenthetical clause within the context. Parentheses contain additional information which is not absolutely necessary. This view requires, on the one hand, the exclusion of those clause-medial elements that express necessary content, e.g., utterance verbs, from the concept of parenthesis and, on the other hand, the inclusion of medial syndetic adverbial clauses with redundant information (cf. Hoffmann 1998: 304). We owe the first thorough examination of the communicative function of at least a part of parenthesis to Urmson (1952). As he was interested mainly in statements, he confined his analysis to this speech act and examined the role of parentheticals therein. Their main function is to limit the speaker's commitment regarding the truth of the asserted proposition. What is relevant for us there is Urmson's view that the interruption of the host clause is not a constitutive property of parenthesis. He explicitly mentions the following two possibilities for using a parenthetical (1952: 481): (15) (16)

Your house is I suppose very old. Your house is very old I suppose.

In Urmson's view, the communicative function of I suppose remains the same. This pragmatic approach exerted a strong influence on the considerable body of linguistic research on the subject published from 1970 onwards. In Corum (1975), for example, the position in the sentence does not play a major role.



Clause-final, but also clause-initial expressions, even without a verb, are called parenthetical because of their communicative function (see Corum 1975: 134):27 (17)

Obviously, Patty Hearst is in Guatemala.

I have to add that primarily in the decade from 1970 to 1980, the debate between generative linguists about the deep structure of sentences with parentheticals and the type of transformation necessary to derive the surface structure as well as the two proposed analyses (complement-fronting hypothesis and proform deletion hypothesis) were based on sentences with final parentheticals (see, e.g., Ross 1973; Emonds 1976). Not all modern publications on the discourse role of parenthesis, of course, follow Urmson (1952). Bayer (1973) examines a corpus of spoken German and extracts all sequences which are contained in a clause, but which can neither be described as a part of it (i.e., they are not syndetic) nor be identified as slips of the tongue or false starts.28 Typically for a pragmatically oriented approach, the position in the clause is treated as irrelevant, though sequences which constitute complete clauses are excluded if they are initial or final. Like Schwyzer (1939: 45f.) and Betten (1976), Bayer sees a relationship with anacoluthon and draws attention to the fact that, especially in corpora containing spontaneous conversation, it is difficult to extract parentheses, since host clauses often are incomplete or even not recognizable as such. As I will illustrate in section 3.3, he then classifies the extracted sequences according to a revised version of Malinowski's (1923) and Jakobson's (1960) model of communication that distinguishes between kommentierenden Parenthesen (expressing speaker comment on the content, code, form, and structure of the message) and the kontaktbezogenen Parenthesen (establishing, maintaining, controlling and closing the physical and psychological contact between interlocutors). The second group significantly expands the potential realm of parenthesis, since it comprises vocatives and other forms of address, interjections and communicative formulas.29 Nevertheless, as Betten (1976: 219) rightly points out, Bayer's definition of parenthesis is primarily syntactic and communicative only on a second level, notwithstanding his initial declarations (see Bayer 1973: 66ff.). Lampert (1992: 106, 132) says that in written texts a parenthetical construction can be recognized by four properties, of which textual defocalization is the See Corum (1975: 133): "Parenthetic adjuncts (PAs) are used to strengthen or weaken the force of an assertion. [...] PAs include sentential adverbs, adverbial phrases, parentheticals, some non-restrictive relative clauses, and rhetorical tag questions". 28 See Bayer (1973: 71): "Untersucht werden [...] alle Sequenzen im Bereich eines Satzes, die nicht als dessen Teil beschrieben werden können und auch nicht als 'Versprecher' oder sprachliche Fehlplanungen identifizierbar sind". 29 Betten (1976: 219) expresses terminological doubts and furthermore sees no point in comprising under the heading of parenthesis a group of expressions which have a long research tradition and have been described and examined independently. 27



most salient one. That is, a parenthesis is primarily seen as a textual strategy that defocalizes information. Lampert (1992: 134ff.) goes on to describe three substrategies and their corresponding parenthetical constructions. Associative constructions embody the parenthetical principle most clearly and allow the speaker to place a secondary informational focus wherever he or she likes (cf. also Apothéloz & Zay 1999: 23). Reformulating constructions specify, exemplify, generalize, or paraphrase previous information, and regulative ones express subjective evaluations of statements. On pragmatic and semantic grounds, Borgato & Salvi (2001) distinguish between two types of frasi parentetiche, both of which may have in their scope the entire host sentence or parts of it (a phrase or word). The first type modalizes the host sentence or its parts with respect to speaker commitment, evaluation or the mode in which the speech act is carried out. Syntactically, these parenthetical clauses have the same function as sentence adverbials. The second type, which Borgato & Salvi consider similar to a subordinate clause, has a temporal, causal, consecutive, or other relation, not explicated by a conjunction, to the host sentence or its parts. Syntactically, these parentheticals correspond to circumstantial adjuncts or to appositive relative clauses. Borgato & Salvi (2001: 170) explicitly mention the end of the sentence as one of three positions a parenthetical clause may occupy. In concluding this subsection, I want to return briefly to the study of spoken French by Morel & Danon-Boileau (1998). Although they base their description of incise on intonational features, they also make some useful comments on its discourse function. Contrary to what is normally said, they think that parentheses are not a useful means either for the insertion of additional details or for the recovery of constituents which have not been placed in the right position (see Morel & Danon-Boileau 1998: 60ff.). There are other devices more appropriate for this (see Morel & Danon-Boileau 1998: 75-93). Parenthesis is a rather specialized device that enables speakers to comment on their discourse or to express their standpoint and allows them to put forward some information or an argument without directly submitting it to the hearer's judgment (cf. also Simon 2004: 232; Potts 2005: 7). 2.3 Summary Even in classical rhetoric, parenthesis was a concept which could already be described from more than one viewpoint: intentionality versus accidentality, thought versus actual linguistic formulation, insertion of a complete sentence versus insertion of sentence parts. In modern linguistics, other viewpoints and criteria have been added. I have illustrated and explored the six most important criteria or properties involved: prosody, interruption of the host clause, lack of an overt link between the parenthetical and the host clause, sententiality, relationship between the parenthetical and the host clause, and communicative function.



Due to the great variety of these criteria, the concept continues to be very flexible, heavily dependent on the initial choice of the respective author, and therefore not particularly useful. The sequences selected according to one criterion do not correspond to those selected according to another one. In general, syntactic definitions are narrower. The most striking discrepancies are found between syntactic and prosodic descriptions and between discourse-related descriptions and syntactic descriptions. According to the prosodic criterion, practically any syntactic construction may be parenthetical. Pragmatic definitions also select sequences that are not covered by most syntactic definitions, e.g., sentence-initial syndetic main clauses, sentence-initial adjuncts, syndetic adverbial clauses. The concept of parenthesis is certainly vague, yet the exploration of the six criteria shows that those RPCs that Urmson (1952) had in mind are not prototypical manifestations of parenthesis. They are not complete clauses, they are ungoverned dependents of the host and they are not always prosodically isolated from the host.

PREVIOUS STUDIES 3.1 Before 1952 Strictly speaking, the research literature on RPCs does not date farther back than 1952, when Urmson recognized that these clauses constitute a coherent group and defined their most salient features. It was his article that established the notion and provided the first term for it. Nonetheless, the matter has been touched upon before that date, mostly within the broader context of parenthesis in general; important remarks on these clauses can also be found in philosophical and linguistic contributions about the logic of reasoning, statements, mood and modality, hypotaxis vs. parataxis, and sentence adverbs. Ideas and considerations on RPCs were a by-product of studies aimed at related and usually more general phenomena. Urmson (1952) was the first and has up to this day remained one of the few contributions exclusively concerned with these clauses, although it also represents a major advance in the understanding of how statements work. In the following outline, I shall first present some early works which, without treating it as a coherent phenomenon, touched on relevant aspects of the subject, then give a short introduction to Urmson's article and afterwards describe the major problems with which later linguistic research had to deal with. RPCs are a phenomenon at the interface between linguistics and philosophy, and it is no coincidence that the French Port-Royal scholars Arnauld and Lancelot in La logique ou l'art de penser (1662) already recognized an essential aspect of these clauses (see Récanati 1984: 330). After having described non-restrictive relative clauses as propositions incidentes, Arnauld & Lancelot (1662: 153) cite the following sentence: (1)

Je soutiens que la terre est ronde. [I affirm that the earth is round]

Expressions like je soutiens are in their opinion also [...] propositions incidentes qui ne regardent que la forme de la proposition [...]. (Arnauld & Lancelot 1662: 153)

They add that je soutiens […] n'est qu'une proposition incidente, qui doit faire partie de quelque chose dans la proposition principale; & cependant il est visible qu'elle ne fait partie ny du sujet ny de l'attribut: car cela n'y change rien du tout, & ils seroient conceus entierement de la mesme sorte si ie disois simplement, la terre est ronde. (Arnauld & Lancelot 1662: 154)



In these remarks, the Port-Royal scholars anticipate a fundamental aspect underlined by Regula (1925: 177f.) and then by Urmson (1952) almost three centuries later: In the sentence above, the asserted proposition is La terre est ronde and not Je soutiens que p. Similarly innovative is their definition of je soutiens as incidental clause.30 Comparative philologists of the late 19th century also had an interest in the phenomenon of parenthetical clauses and left us some fruitful observations about it. This should not surprise us too much. As Schwyzer (1939) showed some years later, the Sanskrit, Classical Greek, Latin, etc. literary texts studied by comparative philologists contain numerous examples of parenthesis. The diachronic perspective adopted by these scholars almost automatically led to insights that synchronic oriented linguists of the following century gained with much greater difficulty. Interesting are Wackernagel's (1897: 21-27) remarks on parataxis, which in many aspects also touch upon parenthesis and parenthetical clauses. He states that paratactic clauses which are semantically subordinate or secondary with respect to the clause they are connected to may develop into discourse particles. Three types of initial states and subsequent processes can be distinguished according to Wackernagel: 1. A word which originally was an autonomous sentence develops into an introductory particle of the following sentence or clause, e.g., Latin Quare? 'Why?' > Fr. car. 2. A governing main clause first turns into what Wackernagel calls a Zwischensatz, i.e., a clause inserted into another clause, and subsequently is downgraded to a discourse particle. He explains that quite a few Latin, Classical Greek and Sanskrit clauses with utterance verbs, opinion verbs and verbs expressing wishes tended to be used as main clauses as well as medial or final parentheticals. In medial or final position, they lost their 'verbal power' and essentially became particles indicating the type of sentence; e.g., in Latin incidental credo already meant something similar to 'perhaps, maybe' and was typical of declarative sentences. As examples of this process, Wackernagel also cites cases of discourse particles and adverbs that developed out of main clauses; e.g., Latin vel 'or' was originally the imperative of velle and meant 'Choose!' or 'Take what you want!' 3. As in the second process, a governing main clause turns into a parenthetical one which is inserted into the former depending clause together with the subordinating conjunction; e.g., Latin Fors sit an ... 'May it happen that ...' becomes the adverb forsitan or forsan 'maybe, probably'. According to Chomsky (1966: 75ff.), who re-discovered the Port-Royal works and interpreted them as proto-transformational texts, Arnauld and Lancelot were heavily influenced by the philosophy of Descartes. Lakoff (1969b) has shown, however, that the decisive inspiration for Arnauld and Lancelot's theory of language did not come from Descartes, but rather from the book Minerva, seu de causis linguae Latinae, written by the Spanish grammarian Sanctius and published in 1585 (cf. also Huchon 2002: 185). 30



One aspect of Wackernagel's remarks needs to be underlined. He states at a certain point that RPCs do not themselves express communication, opinion or wishes, but indicate, as we would say today, the type of speech act that is being performed. Both ideas (RPCs do not assert, express wishes, etc. and are illocutionary force indicating devices) are hinted at, but are not fully developed in his remarks. In the following decades, the illocutionary force indicating function has never been considered as pertaining to RPCs. On the other hand, contrary to modern contributions, Wackernagel neither restricts the use of RPCs to declarative sentences nor does he mention their function as mitigators. If Wackernagel had further pursued his important insight about sentence content versus sentence indication, he might have reserved a chapter separate from that of parataxis for some of his Zwischensätze. But like a number of linguists after him (e.g., Hofmann 1926: 114ff.; Behaghel 1928: 493, 537ff.; Bloomfield 1935: 186), he seems to consider parenthesis as being merely a variety of parataxis; actually, he does not even use the term parenthesis. In his understanding, two coordinate sentences can be successive or one can be inserted into the other. The decades between the two World Wars produced some of the most significant works in linguistics. The understanding of parenthesis and parenthetical clauses profited from this activity. In his Handbuch der erklärenden Syntax (1931) Havers, a forerunner of the communicative-pragmatic approach, tries to explain the concrete phenomena of language in their cultural, social and psychological contexts. In some very useful notes he discusses the pressures and influences incidental expressions are exposed to (suppression of pauses, pressure of increased rate of speech, stereotypical character of many expressions, types of ellipsis). More importantly, in a chapter on the psychological factors influencing language structure, he illustrates the roles of what he calls successive thinking, i.e., a mode opposed to simultaneous thinking, and associative thinking, i.e., a mode in which ideas and impressions are organized loosely, and states that parataxis and parenthesis are their most obvious manifestations. It should be remembered that, five years earlier, Hofmann had published a pioneering work about spoken Latin where he describes, among other phenomena, the influence subjective states of mind exert on sentence construction. In the part dedicated to paratactic devices, he also mentions parenthesis and states that it hampers the natural progress of thinking and linguistic expression (1926: 114). The relationship between parenthesis and parataxis also predominates Bloomfield's thoughts on the subject. He proposes the first formal definition of parenthesis. In his theory, where intonation plays a major role, parenthesis, parataxis, coordination, and apposition are closely related constructions; the importance of any of these is understood only in relation to the others. In English and many other languages, a specific pitch marks the end of sentences. Parataxis is a construction in which […] two forms united by no other construction are united by the use of only one sentence-pitch. Thus, if we say It's ten o'clock [.] I have to go home [.] with the final fall-


REDUCED PARENTHETICAL CLAUSES ing pitch of a statement on o'clock, we have spoken two sentences, but if we omit this final-pitch (substituting for it a pause-pitch), the two forms are united, by the construction of parataxis, into a single sentence: It's ten o'clock [,] I have to go home [.]. (Bloomfield 1935: 171)

Note that, for Bloomfield, coordination is not limited to sentences and usually implies the use of a coordinator (1935: 195). Parenthesis thus […] is a variety of parataxis in which one form interrupts the other; in English the parenthetic form is ordinarily preceded and followed by a pause-pitch: I saw the boy [,] I mean Smith's boy [,] running across the street [.] In a form like Won't you please come? the please is a close parenthesis, without pause-pitch. (Bloomfield 1935: 186)

The only difference between the two constructions is position, according to Bloomfield's definition. Closely related to parataxis and parenthesis is apposition, in which […] paratactically joined forms are grammatically, but not in meaning, equivalent, e.g. John [,] the poor boy. When the appositional group appears in included position, one of its members is equivalent to parenthesis: John [,] the poor boy [,] ran away [.]. (Bloomfield 1935: 186)

Schwyzer's short treatise on Die Parenthese im engern und im weitern Sinne (1939) is rooted in the comparative philological tradition and concentrates on the use of incidental constructions in written literary texts (cf. also Lampert 1992: 30ff.). Although present in Vedic, Sanskrit, Avestan and Old Persian texts, the major development of parenthesis took place in the classical Greek and Latin literature, where it represented a common stylistic technique. In later texts Schwyzer mentions Old Icelandic, Anglo-Saxon, Old and Middle High German, Celtic and Slavic ones - its importance seems to have diminished. Schwyzer (1939: 31) also gives a definition of of his understanding of parenthesis: It consists of a sentence (grammatischer Hauptsatz) that interrupts the grammatical coherence of another sentence, though a few pages later (p. 44) he corrects himself insofar as he states that subordinate clauses may also occur in parenthesis. In most cases, the two sentences are linked together asyndetically, i.e., there is no overt sign indicating a relation between them. Parenthetical clauses are usually inserted where there would be a punctuation mark in written language. In spoken language, more important than pauses, which are often suppressed, is the lower (or, occasionally, higher) pitch that contrasts with the pitch of the host sentence. Spoken parentheses are also characterized by an increased rate of speech. The coherence of the interrupted sentence may be re-established by repeating parts of it or by inserting connective discourse markers after the interruption. When all the characteristics mentioned so far are lacking, Schwyzer suggests as a last resort the adoption of meaning as a criterion: Parentheses contain additional information which, strictly speaking, is not necessary for the meaning of



the sentence. He explicitly excludes those medial clauses from his notion of parenthesis that express necessary content, e.g., reporting clauses with utterance verbs (1939: 33). According to Schwyzer (1939: 38f.), parentheses are either formally sentences or they at least resemble sentences in their functions. In the second case they are diachronically or synchronically related to sentences and may easily be expanded into full sentences. Schwyzer mentions as diachronic examples names in the vocative case, vocative particles, and interjections that originally were sentence-like forms of address or exclamations. The Latin introductory particle ceterum 'by the way' in ceterum censeo ... first meant 'the rest is: I am of the opinion that ...' (Schwyzer 1939: 41; Wackernagel 1897: 21f.). Only a few paragraphs are dedicated to RPCs. As I said before, Schwyzer (1939: 40) classifies parenthetical Classical Greek dokô, Latin puto, and Swiss German glaub and mein as parenthetical verbal sentences. He also draws attention to the weakening and reduction processes that short sentences of this type undergo when they are used parenthetically. They change their form, e.g., Classical Greek dokô moi > dokô, Standard German glaube ich > glaub ich > Swiss German glaub, but, most importantly, their meaning and function are reduced to that of adverbs. As trivial as it might seem, the fact that mitigating RPCs are essentially sentence adverbs was first and most clearly stated by linguists belonging to the comparative philological tradition, like Wackernagel and Schwyzer, and is thus an insight mainly owing to the historical approach. Regula (1925: 177f.), a Romance philologist, went even farther: In his view, a sentence-initial governing clause like Fr. je crois in Je crois que ça vous met dans l'embarras, is also downgraded to a modal sentence adverb in certain contexts. 3.2 Urmson (1952) and Benveniste (1958) As mentioned, the 'discovery' of RPCs can be credited to Urmson's seminal paper Parenthetical verbs (1952), which also represents the first comprehensive analysis exclusively dedicated to them. Contemporary linguistic research on the subject usually takes his work as starting point. Being a philosopher, he leaves aside most of the linguistic aspects, namely prosodic and syntactic properties, and examines the workings of the speech act statement and the role of verbs used in it. Urmson is aware that uttering a statement presupposes that the speaker is convinced of its truth, unless of course there are specific reasons to the contrary. He formulates this fundamental insight in the following way: Unless we are acting or story-telling, or preface our remarks with some such phrase as 'I know I'm being silly, but ...' or, 'I admit it is unreasonable, but ...' it is, I think, a presupposition of communication that people will not make statements, thereby implying their truth, unless they have some ground, however tenuous, for those statements. (Urmson 1952: 487)



It should be noted that the presupposition of communication and the implied claim to truth postulated by Urmson anticipate Grice's (1975, 1978) cooperative principle and quality maxim (cf. Bublitz 2001: 171). By using an RPC, the speaker does not modify the meaning of the statement, but tells the hearer how its content fits logically, evidentially and emotionally into the context. Urmson puts it as follows: [...] when these verbs are used in the first person of the present tense, as is very clear when they occur grammatically in parenthesis, the assertion proper is contained in the indicative clause with which they are associated, which is implied to be both true and reasonable. They themselves have not, in such a use, any descriptive sense, but rather function as signals guiding the hearer to a proper appreciation of the statement in its context, social, logical, or evidential. (Urmson 1952: 495)

We do not state with an RPC, but with the sentence with which it is associated; with an RPC "we show rather than state", as Urmson writes (1952: 484). The examination of the meaning of the verbs of RPCs within the social and logical context leads Urmson (1952: 485f.) to divide them into three groups, associating each of these to a different group of functionally corresponding (but not synonymous) adverbs: Group 1: "give emotional orientation when used parenthetically" verbs: adverbs: regret happily rejoice luckily unfortunately Group 2: "signal how the statement is to be taken as fitting logically into the discussion" verbs: adverbs: assume admittedly concede consequently confess presumably deduce infer maintain presume presuppose Group 3: "indicate the evidential situation in which the statement is made" verbs: adverbs: believe certainly estimate possibly feel probably guess know suppose suspect Table 3.1: Parenthetical verbs and corresponding adverbs (cf. Urmson 1952: 485f.)

The basic function of some RPCs consists in the expression of varying degrees of speaker commitment to the truth, i.e., in the downgrading of speaker commitment:



[...] the whole point of some parenthetical verbs is to modify or to weaken the claim to truth which would be implied by a simple assertion p. (Urmson 1952: 484)

Although Urmson does not formulate it in these terms, we might say that if simple or categorical assertions, i.e., statements without a mitigating RPC or adverb, by default presuppose the maximum of speaker commitment, a mitigating RPC by its mere presence diminishes that commitment. Urmson does not say which RPCs weaken the claim to truth but probably he was thinking of those composed of the verbs of the second and third group. I have to add that the first group of Urmson's verbs and adverbs are of a different type: Rather than expressing the degree of confidence in the truth of the sentence, they express an emotional evaluation of what is being stated. Whereas evaluative adverbs are common in the languages I am examining, RPCs with an additional evaluative meaning are rare in Romance; e.g., there is no Romance parenthetical equivalent to regret or German bedauern. The few Romance verbs with an, albeit not purely, evaluative meaning which can be used as parentheticals are equivalents of find and hope. As Urmson repeatedly underlines, it is convenient to speak of a parenthetical use of these clauses and to differentiate it from their descriptive use. I will address the question in section 8.5, where I will quote an example in which, according to Urmson, I suppose is used descriptively. In the light of the previous contributions by comparative philologists, Urmson's ideas, although not entirely new, mean a decisive breakthrough. His philosophic standpoint enables Urmson to arrive at a much more precise understanding of the pragmatic aspects of mitigating RPCs. The framework he outlines allows us to see these clauses as a coherent phenomenon restricted to statements (although he does not explain exactly why) and to understand what kind of functions they share with modal sentence adverbs. It provides a criterion for assigning them a special position within the large group of incidental constructions in general, as well as for differentiating in principle between (at least a part of) parenthesis on the one hand and parataxis on the other. In a short but insightful article with the title De la subjectivité dans le langage (1966a [1958]), Benveniste develops, independently of both Urmson and Austin,31 ideas about the subjective element in language, that is, about the speaker's influence on the utterance, which leads him to the recognition of two classes of verbs that in the first person present share an important property: Instead of describing actions or states of affairs, they act as indicateurs de subjectivité, i.e., as devices with which the speaker makes an utterance and at the same time expresses his or her attitude to what is being said. Benveniste's words immediately bring to mind Urmson's: Although Urmson's article had appeared earlier and Austin had given lectures on his view of ordinary language usage at the Universities of Oxford and Harvard from the end of World War II on, it is unlikely that Benveniste was familiar with the ordinary language philosophy typical of the Oxford school before the famous Colloque de Royaumont in 1958 (see Benveniste 1966b [1963]). 31


REDUCED PARENTHETICAL CLAUSES Puis-je considérer ce je crois comme une description de moi-même au même titre que je sens? Est-ce que je me décris croyant quand je dis je crois (que ...)? Sûrement non. L'opération de pensée n'est nullement l'objet de l'énoncé; je crois (que ...) équivaut à une assertion mitigée. En disant je crois (que ...), je convertis en une énonciation subjective le fait asserté impersonnellement, à savoir le temps va changer, qui est la véritable proposition. (Benveniste 1966a [1958]: 264)

Verbs like croire, supposer, présumer, and conclure are named verbes d'opération, and although the Benveniste does not explicitly mention the role of parenthesis, he indicates his awareness of it by using the example of Vous êtes, je suppose, Monsieur X. The second class consists of the verbes de parole, e.g., jurer, promettre, garantir, certifier. In a few lines Benveniste captures the essentials of what came to be known as performative utterance: L'énonciation je jure est l'acte même qui m'engage, non la description de l'acte que j'accomplis. En disant je promets, je garantis, je promets et je garantis effectivement […] L'énonciation s'identifie avec l'acte même. (Benveniste 1966a [1958]: 265)

3.3 RPCs re-discovered No immediate linguistic research resulted from Urmson's and Benveniste's pioneering considerations. As the example of Bolinger (1968) shows, linguists even had to re-discover and re-name the phenomenon in question. Only in the late 1960 did contributions begin to appear that touched on the same subject. Quite a few of their authors had an interest in Romance languages. The majority of them examined syntactic aspects. Bolinger (1968; see also Wüest 1980: 231) points to a parallel between English RPCs and the Romance subjunctive (with the exception of Italian; see Schneider 1999). It can be summarized in a simple rule, sometimes called the Bolinger test (e.g., in Gsell & Wandruszka 1986: 49): If in an English complex sentence it is possible to drop that and postpone the main clause, then the verb in the corresponding Romance noun clause will be in the indicative: (2) (3) (4)

I believe they're ready. They're ready, I believe. Creo que están listos.

If postposition is not possible in English, the Romance noun clause will be in the subjunctive mood: (5) (6) (7)

I don't believe they're ready. *They're ready, I don't believe. No creo que estén listos.

Nowhere in his paper does Bolinger speak about parenthesis, parenthetical verbs, or RPCs, though it is obvious that by postposed main phrases he essen-



tially refers to the same phenomenon. Considering the research history, it is interesting to note that Bolinger neither makes reference to nor seems to be familiar with Urmson's article. Nevertheless, his conclusions about the adjunctival character and the pragmatic function of postposed main phrases are almost identical to those formulated by Urmson. Equally important is the fact that in his article he makes, almost incidentally, a series of original remarks about the syntax of RPCs that anticipate many of the points discussed in the following years. Apart from the impossibility of governing Romance subjunctive noun clauses, there is another feature of which Bolinger (1968: 17ff.) first took notice. With verbs that have a mitigating as well as an attitudinal reading (e.g., understand), postposition of the main clause gives its verb a purely mitigating meaning, whereas nominalization of the noun clause gives the governing verb a purely attitudinal meaning: (8) (9) (10)

I understand that you like him (= 'this is my information' [mitigating] or 'I can accept it' [attitudinal]). You like him, I understand (mitigating). I understand your liking him (attitudinal).

This amounts to saying that subordinate subject or object clauses whose governing verb could be postposed or used parenthetically cannot be nominalized. Whereas Bolinger (1968), without seeming to know Urmson (1952), arrives at conclusions very similar to the latter's, Bayer (1973) neither refers to him nor is his understanding of the meaning and function of RPCs comparable to Urmson's. Bayer (1973) examines parenthesis in general and has no special interest in reduced clauses; nonetheless his remarks on them, in addition to the fact that he examines a corpus of spoken German, make his study worth mentioning. Starting from the assumption that parentheses are on a meta-communicative level with respect to their hosts, he classifies those extracted from his corpus according to a modified version of Malinowski's (1923) and Jakobson's (1960) models of communication and linguistic functions. The majority of them are assigned to the referential function (representing states of the world), the phatic function (maintaining contact between people), and the meta-linguistic function (elucidating the used language itself). Regarding the referential function, Bayer furthermore notes that parentheses always constitute a comment on the representation contained in the host sentence. He adds that meta-linguistic parentheses, besides commenting on the used code, may also refer to the form and structure of the message. Therefore, Bayer suggests distinguishing kommentierende Parenthesen, which express speaker comment on the content, code, form and structure of the message, from kontaktbezogene Parenthesen, which are used to establish, maintain, control and close the physical and psychological contact between interlocutors.



Oddly, mitigators such as glaub(e) ich, ich glaub(e) and mein(e) ich, ich mein(e) are said to belong to the second group (Bayer 1973: 110ff.). Since Bayer does not give reasons for this decision it is difficult to evaluate it. One explanation might be that he assigns a lot of importance to frequency and to the degree to which a parenthetical is stereotyped and formulaic. If an expression occurs frequently as a fixed formula, its use is likely to be triggered both by situational and informational factors. A frequent expression that neither modifies nor adds information is felt to have a contact-related function. This seems to be the case with glaub(e) ich, etc. Putting aside its functional interpretation, Bayer's finding about the stereotyped character of RPCs supports Havers' (1931: 22, 27) statement and is in line with what I found in a study (see Schneider 1999: 69) when compiling a list of Italian mitigating verb forms (see section 1.3): The number of RPCs as well as the number of verb forms actually employed is extremely limited. Bayer's study moreover shows the strength and advantages of a corpus-based approach. According to Bolinger (1968: 25), there are some RPCs which are "completely adverbialized, regardless to position" and others whose status depends on the context and the interpretation of the sentence. Hooper (1975) takes up this suggestion and by means of several syntactic tests arrives at a grouping of English RPCs, which she then applies to Spanish (see table 1.4 in section 1.3; cf. also Haverkate 1994: 125ff.). RPCs, assertive predicates in her terms, allow their complement clauses to be partly or completely preposed - a shift of the perspective in comparison to Bolinger's postposed main phrases. The five syntactic tests (scope of tag questions, negative raising, pronominalization with either so or it, fronting of subject noun clauses and scope of factive sentence adverbs) are all applied to complex sentences and are designed to identify the semantically dominant proposition, i.e., the assertion. Negative raising, for example, indicates that there must be two classes of RPCs (Hooper 1975: 97f., 105ff.). Those that are completely transparent with respect to negation (weak assertive predicates) allow the negative to be raised into the RPC, while keeping the host clause as its scope: (11) (12)

I think these living conditions are not suitable. I don't think these living conditions are suitable.

Owing to the reduced semantic content of I think and its transparency to negation, the two sentences are synonymous. That means that the negative, regardless of its position, always negates the host clause. Parentheticals belonging to the second class (strong assertive predicates) are less transparent. The following two sentences are not necessarily synonymous: (13) (14)

He said the door wasn't closed properly. He didn't say the door was closed properly.



According to Hooper, this is due to the fact that a sentence with a strong assertive contains two assertions: […] a sentence with a strong assertive predicate contains two assertions. [...] negating the main clause may negate any part of either assertion. (Hooper 1975: 106)

As is clear from the examples above, Hooper sees no difference between a sentence-initial parenthetical and a sentence-initial main clause followed by that and a complement clause. To show the different scopes of the negative with strong assertive predicates, Hooper (1975: 98) uses two examples with a complementizer: (15) (16)

The boss didn't say that we have to hire a woman, he only implied it. The boss didn't say that he wanted to hire a woman, he said he had to hire a woman.

In the first example, the negative in the main clause negates say, in the second one, it negates wanted. Bolinger (1968: 11f.) had already noticed that English RPCs like I imagine, I reckon, or I suppose and a few others may be used as synonyms of yes: (17)

Can you get up all right? I imagine / reckon / suppose.

Borillo (1982) examines postposition of main clauses in French and counts more than 200 verbs and corresponding clauses that can undergo this transformation. By testing their behavior as answers to polar questions (the test will be explained in detail in section 7.3), she arrives at a classification which reflects the one proposed by Hooper. Depending on whether the negative is possible or not, the answer may be of three different types (see table 1.1.with the three groups in section 1.3). Clauses with verbs like croire and penser may be used with or without negative, depending on the answer, affirmative or negative, one wishes to give. The clauses with the verbs of the second group, e.g., supposer and espérer, may be used only in the unnegated version. The clauses with the verbs of the third group, e.g., savoir and être au courant, may be used as answers to a polar question only if negated. Borillo's first two groups correspond to Hooper's weak and strong assertive predicates. With a few exceptions, both authors classify the verbs of the third group as semi-factives (see Karttunen 1971). Borillo's question test represents a useful supplement to Hooper's analysis. It is easy to execute and, regarding the first two groups of verbs, it yields a very neat distinction between those verbs that are transparent to negation and those that are not.



3.4 Generative analyses Especially between 1970 and 1980, incidental constructions, primarily with the verbs I have discussed so far, also enjoyed the attention of generative grammar. The main issue under discussion was the source of parentheticals, i.e., the deep structure of sentences with parentheticals, and the type of transformation necessary to derive the surface structure. Within a few years, at least three analyses were proposed. Jackendoff (1972: 95ff.) starts with the observation that some relationship must be established between the following two sentences, but that any transformation that derives (18) from (19) will run into problems: (18) (19)

John is, I think, a fink. I think John is a fink.32

A transformation fronting a part of the complement clause would involve the non-constituent John is, a transformation lowering the main clause to the complement would involve the non-constituent I think. Another difficulty Jackendoff points out is that in a transformation there is no convenient way to state that the parenthetical cannot be negated. Furthermore, he observes that in certain cases, a transformation deriving (18) from (19) does not preserve meaning: (20) (21)

Helen thinks that John is a fink, but he isn't. John is a fink, Helen thinks, (*but he isn't).

In Jackendoff's opinion, these difficulties can be solved if the relation between (18) and (19), rather than being syntactic, is semantic and if the parenthetical clause is generated as a sentence adverbial. This would account for the fact that a parenthetical like I think and sentence adverbs like truthfully, honestly and sincerely share several syntactic and semantic properties. The next two analyses I want to present are called complement-fronting hypothesis and pro-form deletion hypothesis. Both are based on clauses with final parentheticals.33 Ross (1973) does not take up Jackendoff's stimulating semantic interpretation34 and returns to a purely syntactic analysis. He assumes that the parenthetical is a main clause in the deep structure, as in (22), and that its surface structure (23) is derived by a transformation, called Slifting (= sentence lifting),35 that deletes the complementizer and fronts the complement clause: Jackendoff's (1972: 96) phrase marker of the sentence contains the complementizer that. Emonds (1976: 43-51) suggests deriving internal parentheticals from those with final parentheticals by a transformation that moves a final phrasal constituent of the non-parenthetical clause to the right across the sentence final parenthetical. 34 See also Reinhart's (1983: 184ff.) evaluation of Jackendoff's proposal. 35 This is one of the tamer and more transparent terms proposed by the generative semanticists. Later, names for rules became completely arbitrary, e.g., Irving, Ludwig, or Richard (see Harris 1993: 202f.). 32 33




S1 V












NP a


S0 S2 VP




S1 NP a






In his view, various arguments support the postulation of such a rule. One of these is that the analysis accounts for the fact that only verbs that take thatclauses appear in parentheticals.36 As another confirmation of the correctness of his hypothesis, Ross (1973: 152ff.) mentions pronominalization with so. An object clause of verbs like think or believe, if identical to the object clause of a preceding verb of the same class, may be pronominalized with so: (24)

Max thinks that apricot paste has no calories, and his doctor thinks so too.

Although so cannot pronominalize an unembedded clause, it can pronominalize a clause followed by an RPC: (25) (26)

*Rufus is flatulent, and his wife thinks so too. Rufus is flatulent, I think, and his wife thinks so too.

Cornulier (1978) and Banfield (1982: 45f.) show that the rule does not always apply to quotative RPCs, i.e., there are quotatives which are exclusively parenthetical and cannot govern a noun clause (see also the final part of this section). 36



In Ross' opinion, this provides striking evidence for the hypothesis that the main clause Rufus is flatulent of (26) was at one time an embedded object clause of I think. The pro-form deletion hypothesis, proposed among others by Emonds (1976), represents an alternative analysis of the source of clause-final parentheticals. It claims that the surface structure (27) is close to the deep structure (28), which consists of two successive independent clauses concatenated without coordinating conjunction. The second clause contains a pro-form referring back to the first: (27)



John came later than Sue (28)

I think E





John came later than Sue;

I think soi (anyway)

In his view, the same deep structure underlies paratactically connected clauses, which in writing are separated by semicolons: (29)

John came later than Sue; I think so anyway.

By a transformation that deletes the pro-form from the second S and attaches this S to the right of the first S, the surface structure is derived. By postulating a category E (= expression), Emonds takes up Banfield's (1973) and Partee's (1973) proposal about the deep structure of direct quotations and introduces a level above the sentence which is only optionally expanded to a sentence. Thus



By God and Yes, yes in the following examples are generated under E, but outside of any S: (30) (31)

By God, I know that he is guilty. Yes, yes, that would appeal to John.

The category E represents the attempt by transformational grammar to account for sequences that are not sentences in a strictly syntactic sense, but which nevertheless are autonomous utterances and behave like sentences from a pragmatic or functional point of view. In Emonds' view (1976: 53), intonation is generated under E; this allows him to assign a "semicolon intonation" (sequence of two E's) to (28) and a "comma intonation" (one E) to (27). In commenting on (27), Lonzi (1981: 394) defines sentences with parentheticals as follows: They have a single primary stress, a single addressee and contain a single assertion. The debate about the complement-fronting hypothesis and pro-form deletion hypothesis shows both the shortcomings and strengths of generative syntax in this period. Unlike some of the other generative and non-generative approaches illustrated so far, the RPC is presented as a simple coordinate sentence in the surface structure. Furthermore, there is no indication whatsoever of a possible adverbial function of the parenthetical with respect to the other sentence. Ross (1973: 133) even explicitly excludes interpretations such as the following: (32)

S1 Adv S2 Max is a Martian I feel

On the other hand, ignoring the technical aspects peculiar to transformational syntax of this period, one gets the impression of a quite valuable debate about the relationship between parenthesis on the one hand and parataxis and hypotaxis on the other and, most importantly, about the level of linguistic representation at which RPCs should be dealt with. The level of linguistic representation is still a matter of debate in recent works of generative grammar. The proposals we have seen so far (see Jackendoff 1972; Ross 1973; Emonds 1976) treat RPCs as somehow syntactically attached to the host sentence. That is, the parenthetical is dominated by a node of the representation in which it occurs. More recent generative ap-



proaches postulate the absence of any syntactic link to the host sentence. In these analyses, the parenthetical is not part of the host's structure (cf. Ackema & Neeleman 2004: 96f.). According to Espinal (1991), the problem posed by RPCs cannot be solved in sentence grammar in the way in which it has been developed in generative grammar: In my view, a solution to the problem of how to represent disjunct constituents requires an innovation in phrase-structure theory, in that the standard assumption that every syntactic structure has a single root node should be abandoned. If my analysis is correct, the grammar is going to generate structures that have two or more independent root nodes. (Espinal 1991: 741)

According to Espinal's (1991: 747) hypothesis, a sentence like the following receives three independent phrase-markers, one for The manager has gone to another company, one for I think (with an empty category in the complement position), and one for that's life: (33)

The manager has gone, I think - that's life! - to another company.

The tree diagrams share no syntactic node, but interact at three levels: phonetic form (PF), deep structure (D-structure), and utterance interpretation. At the PFlevel, the final linearization between the parenthetical and the host sentence takes place; at D-structure level, the structural relations between the terminal nodes are defined; in the final process of utterance interpretation, the linguistic meaning of the host is projected into the empty argument position of the RPC. Espinal (1991: 751-757) then formulates a series of association principles and conditions: Syntactic conditions allow the insertion of parentheticals in certain positions only, conceptual conditions control the mapping of syntactic constituents onto conceptual constituents, and pragmatic processes enable the hearer to fill the empty argument positions on the basis of the most accessible contextual information. Espinal's proposal is similar to Haegeman's (1991) radical orphanage approach, which postulates that parentheticals are unattached at deep structure (D-structure), syntactic structure (S-structure), and logical form (LF) and that their interpretation has to take place within a theory of the utterance.37 Both proposals have been taken up by Shaer (2003: 241ff.), who advocates a similar approach for fronted, parenthetical, and afterthought adverbs. As Ackema & Neeleman (2004: 96ff.) explain, neither the view that parentheticals are integrated in the host structure nor the opposite one that they are generated separately are satisfactory because both are unable to explain that [...] a parenthetical cannot affect the syntax of the host clause, but grammatical requirements imposed by material in the parenthetical can be satisfied by elements in the host clause. (Ackema & Neeleman 2004: 99) 37

For a discussion of Espinal's and Haegeman's approaches see Burton-Roberts (1999, 2006).



In other words, both views are unable to cope with the semantic one-way-dependency which is characteristic of RPCs. A supposed relation between RPCs and corresponding main clauses has always been at the center of generative and other analyses. Cornulier (1978) questions this relation and objects to the transformational derivation of parentheticals from governing clauses, as hypothesized, e.g., by Ross (1973). He takes into account only French incises inverties, that is, RPCs with inverted subjects. Most of them are reporting RPCs. This, however, does not impair the fundamental validity of his critique. Cornulier notes that there are RPCs that are not capable of governing a noun clause: (34) (35)

On se moquait d'elle, l'interrompit-elle. [She was derided, she interrupted him] *Elle l'interrompit qu'on se moquait d'elle. [She interrupted him that she was derided]

This objection, of course, is neither limited to French (see Banfield 1982: 45f.) nor to the verb interrompre. A number of other reporting RPCs are acceptable only in parenthesis (e.g., fit-elle, s'esclaffa-t-elle, renchérit-elle). Cornulier (1978: 58) also points to Latin inquit and quoth in Shakespearean English, which appear only in parenthesis. I wish to add that there are also non-reporting parentheticals that do not govern noun clauses. E.g., in the Italian corpora I found It. diciamo così, which is monovalent (see Hölker 2003: 134) and occurs either as an independent sentence or as a parenthetical: (36)

A: […] non si può trarre affatto la convinzione che vi sia stata diciamo così una sconfitta storica eh definitiva di quest'ideale […] (CLIP.RD13) [One cannot absolutely deduce the conviction that there has been, let's say it like this, a definitive historical defeat of this ideal]

Cornulier (1978: 58f.) also notes that there are host structures that cannot be complements: (37) (38)

Encore une faute d'orthographe, me semble-t-il. [Another spelling error, it seems to me] *Il me semble qu'encore une faute d'orthographe. [It seems to me that another spelling error]

He adds an important observation to which I will return on several occasions. Whereas a governing clause always has the entire dependent clause in its scope, an RPC may have either parts or the entire host sentence in its scope. Influenced by the French linguistic tradition, Cornulier distinguishes RPCs with postposed subjects (incises inverties) from those with preposed subjects



(incises progressives). Neither of the two can occur sentence-initially. Also as a result of this influence, Cornulier (1978: 87) is probably the first to isolate RPCs from complete parenthetical clauses, which he calls incidentes. Another valuable contribution by Cornulier (1978) to the study of RPCs consists in his critique of the notion of parenthetical verb. Generative and other linguists seemed to presuppose the existence of a lexically definable class of parenthetical verbs. Cornulier attacks this conception directly: Le propos du présent exposé est d'abord de montrer que l'existence de la classe des verbes parenthétiques est douteuse, et corrélativement que la notion formelle de verbe parenthétique n'a guère de sens. (Cornulier 1978: 55)

He contends that parentheticality is dependent on a number of non-lexical conditions and phenomena that make it practically impossible to draw up an exhaustive list of parenthetical verbs. Cornulier mentions verbal aspect, negation, the presence of additional modifiers, the type of subject, and the semantics and sentence type of the host. He also points to the possibility of making a verb parenthetical that normally is not: (39) (40)

*Il pleuvra demain, sait le jardinier. [It will rain tomorrow, knows the gardener] Il pleuvra demain, me fait savoir le jardinier. [It will rain tomorrow, tells me the gardener]

Furthermore, he states that theoretically almost any verb, given the appropriate context, may be acceptable in parenthesis, even a verb like clignoter 'flash': (41) (42)

Le Pape est mort, clignote l'affiche lumineuse. [The pope died, flashes the neon sign] When, was the question, have you smoked dope?

Finally, the second example above shows, according to Cornulier, that there are RPCs whose parentheticality depends more on the noun phrase they contain than on the verb. 3.5 RPCs as hedges and as discourse markers During the same period in which generative grammar began to deal with RPCs, Lakoff (1972) proposed a concept that later proved to have considerable success in semantics and pragmatics: the hedge. Lakoff's starting point is fuzzy logic and the problem of fuzziness and vagueness. In that context, he discusses words whose job it is to make referring terms and predicates fuzzier. These words, e.g., sort of, kind of, rather, technically, often and several others, Lakoff (1972: 196) calls hedges. Initially, they were seen as devices merely acting on the precision of the propositional content; in the following years it became clear



that by reducing the precision of reference, the speaker at the same time reduces his or her claim to truth. In other words, the concept of the hedge was extended from the relation between referring term and referent to the relation between speaker and utterance (see Caffi 2001: 47). The first step in this extension was to involve the indication of the illocution, i.e., the performative clause. Taking up an idea already hinted at in Lakoff (1972: 213), Fraser (1975) develops the concept of hedged performative, i.e., performatives like I can promise you that p, I must advise you that p, and so forth. Later, Brown & Levinson (1978, 1987) introduce a further extension. In their ground-breaking work on politeness and face-threatening acts, hedges act not only on the propositional content or on the explicit performative clauses but also on illocutionary force. They are mainly expressions indicating that the speaker is not fully adhering to one of Grice's (1975, 1978) maxims. In fact, Brown & Levinson (1978: 169-176, 1987: 164-171) illustrate hedges on quantity, on quality, on relevance, and on manner (see also Yule 1996: 38f.). What is relevant for us is that Brown & Levinson (1978: 154-167, 1987: 146-162) mention as hedging devices, among others, the English RPCs I guess, I suppose, I'm afraid, and it seems to me. In Brown & Levinson's framework, RPCs are strategies of positive politeness as well as of negative politeness. In the first case, they are used as [...] redress directed to the addressee's positive face, his perennial desire that his wants (or the actions/acquisitions/values resulting from them) should be thought of as desirable [...]. (Brown & Levinson 1987: 101)

They are elements within general positive politeness strategies such as the use of in-group markers, avoiding disagreement, complementing, showing an interest in the hearer's needs, involving the hearer in the activity, joking, and others. In the second case, RPCs are means by which the speaker tries to minimize his or her imposition on the hearer or intrusion into the hearer's space. Negative politeness strategies address the hearer's need to have his or her freedom unhindered and his or her attention unimpeded. In his study of understatement and hedging in English, Hübler (1983) starts from Hare's (1970) and Lyons' (1977: 749ff.) differentiation between phrastic (propositional content), tropic (illocutionary force), and neustic (sign of subscription). He insists particularly on the difference between modalized and unmodalized statements. As an example, Hübler (1983: 11) cites the following pair of sentences: (43) (44)

It's snowing in the mountains. It's snowing in the mountains, I suppose.

Whereas they have a common phrastic and tropic, they differ with respect to the neustic. The first sentence represents a straightforward statement, the second a weakened one which becomes an assumption. Hübler redefines the neustic in



terms of the speaker-hearer relationship, i.e., as the component of an utterance that expresses the attitude of the speaker to the hearer regarding the proposition. In the case of statements, this attitude consists in an expectation of ratification by the hearer. This feature of statements is called modal by Hübler (1983: 97). Since modality is not a part of the proposition, but operates on the proposition, it cannot be negated. Statements where the speaker is sure of what he or she says and does not expect refutation by the hearer contain categorical (= unmodalized) assertions. Hübler (1983: 97) characterizes them as quasi-factive. Nevertheless, in spite of the speaker's expectation, a categorical assertion is potentially refutable or negatable, which is why Hübler says that its modality is implicit or unmarked. Statements containing modalized assertions, by contrast, are characterized by the presence of an explicit modal expression which produces a neustic indetermination. Hübler discusses three classes of modal expressions: RPCs, modal adverbs, and modal verbs. With respect to the first class, he adheres to Urmson's (1952) and Hooper's (1975) theoretical framework, in particular to the distinction between strong and weak assertive predicates: Strong assertive verbs convey more than just the attitude of the speaker to the propositional content and the validity claim arising from it, whereas weak assertive verbs convey only this. (Hübler 1983: 117)

Weak assertive predicates are therefore purely modal, i.e., non-propositional, whereas the additional semantic component of most strong assertive predicates makes them form a separate proposition. Only be certain, be sure, be clear, be obvious, and be evident are singled out for not showing this semantic enrichment and not conveying anything but speaker commitment. Hübler's (1983: 148) analysis of the relationship between categorical assertions and assertions modalized by be certain, be sure, and so forth is really original. Although an assertion with a modal expression of certainty is more emphatic than an unmodalized assertion, in his view there is no increase in probability above the degree present in categorical assertions. By using parentheticals like be certain or be sure, the speaker mainly prevents the hearer from negating the propositional content. Insofar as modal expressions of certainty prevent the hearer from refuting the asserted proposition they partly assume an imperative function. There thus are several aspects that make Hübler's innovative work relevant for us. They relate to the functioning of statements in general rather than being constrained to RPCs. One of the job of hedges is to make referring terms and predicates fuzzier. In other words, they generate vagueness. Vagueness in language is often seen as a defect. Recent approaches to vagueness are taking a different standpoint. Channell (1994: 3) maintains that vagueness is neither good nor bad, Jucker, Smith, & Lüdge (2003: 1737, 1766) even argue that in everyday conversation vague expressions are a pervasive phenomenon and that interlocutors rarely have



problems in understanding them. Vague expressions are not just imperfect substitutes of precise ones. On the contrary, they may be more effective than precise ones, insofar as they guide the hearer in determining how much processing effort he or she should devote to a referent. The publication of Weydt's (1969) study of German discourse markers and their French equivalents created an important research area. Whereas Weydt takes into account German ja, doch, denn, aber, nur, etc. but no verb-based expressions, later studies treat a few selected RPCs as discourse markers. Östman (1981) and Holmes (1986) are discourse marker studies specifically dedicated to Engl. you know. Schiffrin (1987) includes Engl. you know and I mean within the group of discourse markers. In his work on French correction and end markers, Hölker (1988) analyzes specifically tu sais, vous savez, and je veux dire. Stenström's (1994: 59) list of English "interactional signals and discourse markers" includes common RPCs such as I mean, I see, I think, you know, and you see. Bazzanella (2001) discusses, among others, the Italian discourse markers diciamo, sai, sa, and sapete.38 For Andersen (1996: 314) all RPCs are discourse markers. Martín Zorraquino & Portolés Lázaro (1999) mention forms such as Sp. digo, mira, mire, oiga, oye, vale, venga, and vamos. Aijmer (2002: 48) defines Engl. I think as a phatic discourse marker with an evidential meaning. Kärkkäinen (2003: 175ff.) argues quite convincingly that, according to the criteria outlined in Schiffrin (1987: 328), I think must be a full-fledged discourse marker. Evidently, a lot depends on the definition adopted. Yet there is no commonly accepted definition (see Hansen 1998: 4). Initially, great importance was placed on the notion of particle: Discourse markers were seen as words that do not take inflections. Later definitions assumed a discourse-functional approach and this was the point at which RPCs began to be considered. Recently though, there has been a tendency to divide the functions of discourse markers into two major domains, i.e., the relationship between clauses on the one hand and the relationship between interlocutors on the other. According to Fraser's (1999, 2004) narrow definition, discourse markers are exclusively elements that signal relationships, e.g., contrast, elaboration, implication, temporality, between successive discourse units (cf. also Hansen 1998: 73ff.). In this framework, RPCs are excluded from the class of discourse markers. 3.6 Recent contributions Blanche-Benveniste (1989) focuses on the governing capabilities of those French verbs that govern noun clauses introduced by que. Unlike previous syntactic studies, this contribution stems from a long tradition of work dedicated to spoken language and corpus linguistics, which constituted the primary interest of the Groupe aixois de recherches en syntaxe (GARS). Blanche-Benveniste Likewise, in an article on discourse markers in Old Italian, Bazzanella (2003: 257ff.) mentions "suppositional parentheticals", but unfortunately provides no example. 38



(1989: 62, 72) notes an inverse proportionality between the pronominalization of the noun clause and the postposition of the verb. When the clause can be pronominalized, the governing verb cannot be placed within or after the clause, and vice versa: (45) (46) (47) (48) (49) (50) (51) (52)

Je vous ai prouvé que c'était dans le journal. [I proved to you that it was in the newspaper] Je vous l'ai prouvé. [I proved it to you] Je vous l'ai prouvé, que c'était dans le journal. [I proved it to you that it was in the newspaper] *C'était dans le journal, je vous ai prouvé. [It was in the newspaper, I proved to you] Il paraît qu'il est malade. [It seems that he is ill] ? Il le paraît. [It it seems] *Il le paraît, qu'il est malade. [It it seems that he is ill] Il est malade, il paraît. [He is ill, it seems]

In the framework of the approche pronominale (see Blanche-Benveniste, Deulofeu, Stéfanini, & Van den Eynde 1984), the possibility of pronominalizing is the sole indicator of valency, that is, only those elements that can be pronominalized are said to be required by the verb's valency (i.e., are complements). A verb like je vous ai prouvé in the example above is called recteur fort by Blanche-Benveniste (1989: 61ff.), because it governs the following noun clause, while a verb like il paraît is defined as a recteur faible. According to BlancheBenveniste (1989: 63), even when preposed, the recteurs faibles do not actually govern the following clause: Je dirait que, dans les emplois de recteurs faibles, le verbe ne régit pas vraiment la séquence qui suit. Il garde des apparences de verbe recteur, mais il est dépourvu d'une des caractéristiques essentielles des verbes recteurs forts.

She adds that some verbs, notably je crois, can have both functions, while others, e.g., il paraît, on dirait, and je crois bien, are only recteurs faibles. I will return to the difference between je crois and je crois bien in section 8.5. Venier (1991) is one of the three existing monographs specifically dedicated to RPCs in Romance (and the only one on Italian RPCs). Being written from a philosophical perspective, without taking into account corpus data or spoken language, the study is important mainly from a theoretical point of view. Many of the issues I have touched upon so far are discussed in this book. It consti-



tutes a thorough introduction to the subject to which I will refer repeatedly. Venier examines the properties shared by modal sentence adverbs and mitigating RPCs and compares the two types of expressions in Italian. Apart from free position in the sentence, characteristic prosody, and use as answers to polar questions, there is another common property underlined by Venier: In Italian, as presumably in Spanish and French, neither modal sentence adverbs nor mitigating RPCs may be negated: (53) (54)

*Non probabilmente il treno arriverà in orario. [Not probably the train will arrive on time] *Il treno arriverà in orario, non credo. [The train will arrive on time, I don't believe]

There are counterexamples regarding the negation of English modal sentence adverbs (Venier 1991: 93) and of English RPCs (see Bolinger 1968: 23; Baker 1970; Ross 1973: 136f., 155ff.; Hooper 1975: 107; Quirk, Greenbaum, Leech, & Svartvik 1985: 1112f.; Ziv 1985: 186f.), but in Venier's opinion they do not invalidate the general principle that, pragmatically, it is possible to downgrade the claim to truth, but impossible to negate it. Another property discussed by Venier (1991: 37ff.) is the question of whether in Italian and English other speech acts than statements may be mitigated. She concludes that in spite of some apparent exceptions mitigation is limited to statements. The book contains also a list of Italian parentheticals (see Venier 1991: 67f.), which, though based on translations from English, is useful from a theoretical point of view. Another monograph on Romance RPCs is Andersen's (1997) study of parentheticals and subordination in spoken French, whose main starting point is Blanche-Benveniste (1989). Despite its corpus orientation, it takes into account only a limited number of parentheticals (see table 1.2 in section 1.3). Moreover, the study does not include statistical information on frequency or textual distribution. Andersen refines and completes the distinction, already present in French linguistics, between incidente and incise and introduces a third type of parenthetical, namely the proposition parenthétique. Andersen's definitions of incise and proposition parenthétique are similar; most importantly, both expressions are reduced clauses. The main differences are the position within the host, the position of the subject, limitations regarding person, tense, and mode of the verb, meaning and discourse function. As defined by Andersen (1997: 127-146), the propositions parenthétiques may occupy three different positions (initial, medial, and final) in the host, the incises only the medial and final positions. The propositions parenthétiques display the sequence subject-verb; furthermore, their verbs are, with few exceptions, in the first person singular of the present indicative, belong to the semantic group of opinion verbs and psychological verbs, and have a modal discourse function. Other important properties of the



propositions parenthétiques mentioned by Andersen are the impossibility to negate them (apart from the case of je ne doute pas) or to question them, the absence of adverbials, the lack of one argument and the semantic bleaching of their verbs. Pragmatically, they weaken an assertion by reducing speaker commitment. Andersen also adopts the pronominalization method of the approche pronominale (see Blanche-Benveniste, Deulofeu, Stéfanini, & Van den Eynde 1984) in order to determine the constituents governed by the verb. Regarding sentences like (50) above, she assumes an even more radical standpoint than Blanche-Benveniste (1989: 61ff.). In Anderson's (1997: 140) eyes, the rection faible is actually a non-rection. In addition to these propositions parenthétiques prototypiques, Andersen (1997: 179-200) describes another group of parenthetical clauses, which she calls propositions parenthétiques non prototypiques and whose verbs are in the second person of the indicative or imperative. She also remarks that a third group, containing verbs in the third person that report direct speech, could be added. The pragmatic function of the clauses of the second group concerns the structure of the utterance and the maintenance of the contact with the interlocutor (phatic function). In Schneider (1999), I examine the noun clauses in the CLIP, in particular the mood of their verbs and their degree of sententiality or nominality. The sententiality of a noun clause depends on the main clause by which it is governed. Main clauses that may be used as parentheticals, that cannot govern a preposed noun clause, that cannot govern noun phrases, and whose verbs rarely occur as the most rhematic element in simple sentences are defined as being sentential (see Schneider 1999: 188f.). Sentential clauses can be substituted by sì or no; moreover they permit the extraction of phrases and their dislocation to the left of the main clause. Nominal clauses are governed by main clauses that do not share these features. Most importantly, they can easily be nominalized or expanded by il fatto. I thus establish a direct relationship between the parentheticality of the main clause and the sentential character of the noun clause. The next step consists in the analysis of all noun clauses in the CLIP (about 3,200). A considerable number of these (about 500) stem from a small set of stereotyped main clauses, e.g., (io) credo, (mi) pare, (io) penso, and (mi) sembra (see table 1.3 in section 1.3). These are precisely those that can be used as parentheticals and that are followed by clauses with the highest degree of sententiality. On the other hand, the clauses with the highest degree of nominality are introduced by mi dispiace, io mi meraviglio, sono contento, and some other expressions (cf. the figure in Schneider 1999: 189). There are two functions typically associated with the Italian (and Romance) subjunctive: indication of modality, indication of a high degree of subordination and dependence. The two functions are necessarily in opposition: The indication of modality means the indication of illocution or speaker commitment, the sub-



ordination and dependence implies the neutralization of illocution or speaker commitment. According to my findings, the subjunctive appears in clauses with sentential character in the majority of cases. The subjunctive is primarily a marker of modality and only secondarily a marker of subordination. In Italian, unlike in French and Spanish (cf. Gsell & Wandruszka 1986: 47ff., 69; Hummel 2001: 137ff., 200ff.), 55 % of the clauses governed by (io) credo, (mi) pare, (io) penso, and (mi) sembra are in the subjunctive. Evidentiality, i.e., everything concerned with the source, type and reliability of the speaker's information, is defined very broadly by Ifantidou (2001) and includes all types of mitigation. This is the reason why her book on evidentials and relevance comprises an extensive chapter about parentheticals. She starts with a critical review of the speech act account of parentheticals and then sketches how these expressions can be analyzed within the framework of relevance theory. Urmson (1952) treats RPCs as not contributing to the proposition expressed by the utterance in which they occur, i.e., as non-truth-conditional. Besides, at least in Urmson's view, there is no difference between parentheticals and their main clause counterparts. Ifantidou (2001: 125ff.) criticizes this assumption and states that only genuine parentheticals are non-truth-conditional. Her main argument is the fact that only the main clause I think, not the parenthetical I think, fall within the scope of operators like 'if' or 'either … or' (see section 8.5). It is the merit of relevance theory to have underlined the distinction between lexical or conceptual meaning and procedures providing instructions as to how concepts are to be interpreted, i.e., the distinction between conceptual and procedural meaning. Blakemore (1991) suggests - rightly - that parenthetical and performative clauses guide the hearer in processing a proposition. Ifantidou (2001: 129), on the other hand, holds that parentheticals encode conceptual meaning and asks if parenthetical I think can be challenged as untruthful: (55)

Peter: John is waiting at the airport, I think. Mary: That's not true; you don't think anything of the sort.

In her eyes, I think, though not contributing to the truth conditions of the sentence it is attached to, can in fact be challenged as untruthful, i.e., can be true or false in its own right; therefore, it encodes a conceptual representation. Ifantidou's example is not taken from a corpus but invented, a fact which weakens her argument.39 Moreover, as will be shown with comparable exchanges (see section 7.3), it is possible that the speaker Mary, by negating you think, actually negates John is waiting at the airport. Blanche-Benveniste (1989: 65f.) cites a corpus-based example which only supports partly Ifantidou's argument: (i) - l'atmosphère s'est dégradée, paraît-il. - non, c'est pas paraît-il; c'est qu'il s'est dégradé. Blanche-Benveniste adds that the speaker wants to refute the parenthetical, but since he cannot apply a negative to it, he says c'est pas paraît-il. 39



It has been pointed out that it is possible to create complex RPCs (cf. section 7.1). The conceptual rather than procedural nature of a parenthetical like think is also shown, according to Ifantidou (2001: 129f.), by its potential compositionality: (56)

John is, I increasingly tend to think, a fool.

If this complex parenthetical encodes conceptual meaning, it can easily be analyzed as semantically composed. It is not clear how it could be analyzed in procedural terms. Ifantidou advocates a double-speech-act approach to RPCs. In her opinion, it is already implicit in Urmson (1952) that they give rise to complex speech acts, in which one part comments on the other. With genuine performatives, by contrast, only a single speech act is performed. According to Ifantidou (2001: 137) a sentence like the following makes two assertions, which are given in (58) and (59): (57) (58) (59)

That painting is, I think, the best in the museum. That painting is the best in the museum. The speaker thinks that painting is the best in the museum.

The function of the assertion in (59) is to guide the interpretation of (58), in other words, (59) encodes the higher-level explicatures of (58). It is debatable whether Ifantidou's (2001) hypothesis can be maintained in this strong version. In any case, her interpretation of Urmson (1952) is erroneous; in fact, the passage she cites affirms exactly the contrary: They are not part of the statement made, nor additional statements, but function with regard to a statement made rather as 'READ WITH CARE' functions in relation to a subjoined notice […] They help the understanding of what is said rather than being a part of what is said. (Urmson 1952: 495f.)

By "nor additional statements", Urmson is clearly saying that RPCs do not constitute additional higher-level assertions. There is no doubt that the host sentence and the RPC are both predications, nevertheless there is a pragmatic difference between them. The issue raised by Ifantidou (2001), though, requires consideration. If RPCs do not constitute full-fledged assertions, what else do they constitute (cf. also Jayez & Rossari 2004: 219f.)? The question is related to the debate, conducted mostly within generative grammar (see section 3.4), about the level of linguistic representation at which RPCs should be dealt with. 3.7 Summary I started by reviewing thoughts and contributions formulated before Urmson (1952), by the French Port-Royal scholars Arnauld and Lancelot in La logique



ou l'art de penser (1662), by comparative philologists like Wackernagel (1897) and Schwyzer (1939), and by general linguists like Bloomfield (1935). These contributions touched upon RPCs without integrating them into a coherent theoretical framework. This was exactly the merit of Urmson's (1952) seminal article, in which for the first time RPCs and modal adverbs are analyzed within a sound theory of statements. I also mentioned that a couple of years later Benveniste (1966a [1958]) developed, independently, ideas about these clauses and the verbs they contain which support Urmson's views. Linguists took notice of RPCs only shortly before 1970, when RPCs started to attract much interest and several important papers - but very few monographs - were dedicated specifically to them. Apart from the discourse function of these clauses, which has been the major controversial issue, these works addressed a host of additional problems: RPCs and the verbs they contain or, more specifically, the relation of these verbs with the Romance subjunctive and the impossibility of nominalizing clauses governed by these verbs; the behavior of RPCs with respect to negative raising; the use of RPCs as answers to polar questions; the properties shared by modal sentence adverbs and mitigating RPCs; the syntactic analysis of sentences containing incidental expressions; the relationship between parenthesis on the one hand and parataxis and hypotaxis on the other; the level of linguistic representation at which one should deal with RPCs and the difference in truth-conditionality between these clauses in governing position and in incidental position. The overview provided by this chapter shows that RPCs have been studied from several perspectives. While the analyses based on a pragmatic, supra-sentence level approach have led to the most fruitful insights, there is no doubt that a conclusive understanding of RPCs is to be reached only by an integrated approach that considers the interaction of more than one level.

REDUCED PARENTHETICAL CLAUSES IN SPOKEN CORPORA 4.1 An empirical approach A proper description of Romance RPCs can only be achieved on the empirical basis of authentic texts. While obtaining samples of written language has never presented any serious difficulty and nowadays is made easier by the electronic format in which many texts are available, using spoken samples has frequently implied a considerable technical and organizational effort. Fortunately, the analysis of spoken language is gaining in importance in all fields of linguistics, and in the past years there have been a number of projects aimed at collecting corpora of languages other than English. Today the major Romance languages are all represented with publicly available corpora of spoken discourse. Surveys of Romance spoken language corpora are found in Koch & Österreicher (1990: 30-49) and Pusch (2002). Over-reacting to the research tradition dominating in American linguistics before 1950, Chomsky (1957: 15) condemned corpus-based analysis, holding that any corpus is "finite and somewhat accidental". While this critique was theoretically justified, in reality it turned out that intuitive acceptability judgments provided an even less trustworthy empirical basis. In the past, the introspective method was partly justified by technical limitations and the research object of these studies, i.e., a small set of grammatically acceptable sentences, nowadays the shift towards the analysis of all sorts of verbal utterances requires a more empirical and accountable way by which linguistic data are obtained and presented. The need for authors to quote data from an accessible corpus of written or spoken language as evidence supporting a particular theory is, in my eyes, especially pressing in a field like Romance linguistics, where many researchers are not native speakers of the languages they are examining. Basically, the approach of this study is empirical, that is, 'bottom-up'. The selection of RPCs from the corpora is based on formal, i.e., semasiologic criteria, and important hypotheses and conclusions are derived by inductive reasoning, i.e., inferred as probable from the data of the corpora. Yet, as in any study of this kind, at least an equal amount of inferences are drawn deductively from premises that represent a priori knowledge or assumptions. These constitute a framework of categories through which the data are interpreted. In my case, these are, apart from general linguistic premises, the assumptions on mitigation illustrated in section 5.1 and largely owing to Hare's (1970) and Caffi's (1999, 2001) distinctions of mitigating devices. These constitute the pragmatic framework. Notwithstanding the theoretical difficulty posed by the hermeneutic circle, in this study I therefore alternate semasiologic with onomasiologic criteria in order to arrive, step by step, at a thorough understanding of RPCs.



Another one of my a priori assumptions is that RPCs are basically similar in French, Italian, and Spanish, but also in English and German. This assumption allows me to apply to one language inferences drawn from another language. Obviously, whenever necessary and useful I seek confirmation by referring to more than one language and point to differences by comparing them. 4.2 The analyzed corpora This study is based on 22 corpora of spoken French, Italian, and Spanish, comprising a total amount of 3,975,500 words. With the exception of the CRFP, they are all publicly available, either in printed or in electronic form. They were collected between 1968 and 2002. Among the 22 corpora, there are at least two large reference corpora for each language. The considerable differences between the total numbers of words for each language reflect only partly the actual corpus scene. There is no doubt that, from a purely quantitative standpoint, Spanish is best represented through corpora. To a certain extent, this is also true as far as diastratic variation is concerned. With respect to situational and diatopic variation, however, the French and Italian corpora offer a far broader variety. In French and Italian corpus linguistics, the phonological aspects of transcription have always played a more prominent role, which explains the generally superior quality of the French and Italian transcriptions. Apart from a few invented examples and those quoted from other authors or from literary works, the examples cited in this book stem from the corpora listed in table 4.1. They are followed by a bracketed expression indicating the corpus, the text, and, if a printed version is available, the page number, e.g., LUD.3.63 = Texte des gesprochenen Französisch published in Ludwig (1988), third text, page 63. In two printed corpora, BER and CAF, the final number indicates the line, in the CREDIF it indicates the utterance number. In order to enhance the readability of the examples, some symbols and notational particularities have been omitted when not necessary for interpretation.



Corpus French corpora Eschmann Hölker Ludwig Transcriptions d'entretiens Beeching Choix de textes de français parlé Corpus de référence du français parlé French part of C-ORAL-ROM Italian corpora Brandi Cresti Pixi corpora Corpus del lessico di frequenza dell'italiano parlato Corpus di italiano parlato Caffi Bertoli Sand Italian part of C-ORAL-ROM Spanish corpora El habla de la ciudad de Madrid Encuestas del habla urbana de Sevilla. Nivel culto Encuestas del habla urbana de Sevilla. Nivel popular Encuestas del habla urbana de Sevilla. Nivel medio Corpus oral de referencia del español contemporáneo Spanish part of C-ORAL-ROM






1984 1988 1988 1989 2002 2002 not published

9 31 11 25 95 36 132

24,000 35,000 22,000 213,000 155,000 66,700 440,000


2005 Total French

206 545

300,000 1,255,700


1987 1987 1990 1993

4 1 229 469

5,800 2,700 22,000 490,000


2000 2001 2002 2005 Total Italian

49 3 3 204 962

58,300 13,500 19,500 300,000 911,800


1981 1983

24 24

141,000 67,000













169 763 2,270

1,808,000 3,975,500


2005 Total Spanish Total French, Italian, and Spanish Table 4.1: The analyzed corpora

4.2.1 French corpora Eschmann (1984) (ESCH) is a small corpus with approximately 24,000 words. Its 9 texts were recorded between 1968 and 1972; the speakers (students from university and secondary schools, teachers, housewives, bus drivers, agricultural engineers) come from Paris, Lyon, and from the départements of Haute-Loire and Tarn. Most of them knew that they were being recorded. The conversations are unguided and are concerned with the student revolt which took place in May 1968, holiday photographs, cattle-raising, school, and radio communication between bus drivers and a control center. The notation is orthographic, with punctuation; elided sounds, liaison, unintelligible words, overlaps, and pauses are marked. Hölker's (1988) study of French discourse markers is based primarily on the transcriptions of 31 doctor-patient-conversations. These texts, which constitute



a homogeneous corpus of approximately 35,000 words (HOELK), are added as an appendix to the book. In order to protect the privacy of the patients, no background information regarding time, place, and persons of the recordings is provided. From at least two conversations (4.7 and 7.3), it can be deduced that they were made in 1982. In fact, Hölker (personal communication) stated that the 20 hours of recording took place in 1981-82 in several urban areas all over France, though mainly in its central and southern regions. The corpus texts are characterized by a low degree of communicative distance. The notation is orthographic, with punctuation and special symbols for long and short pauses and unintelligible parts. Between 1979 and 1984, Ludwig (1988) collected the Texte des gesprochenen Französisch (LUD). Its 11 texts total approximately 22,000 words and were recorded in three discourse situations (family conversation, political discussion on the radio, and university course) with gradually rising communicative distance. During the family conversations, one participant was not aware of being recorded, in all other recordings the participants knew they were being taped. The recordings of the family conversations took place in Grenoble. The orthographic transcription is in the musical score format, which permits a better representation of overlaps and simultaneous speech, and marks numerous speech phenomena such as vowel lengthening, pauses, interruptions, loudness, and emphasis. It is completed by additional extra-linguistic comments and background information on each text. A research group of the Centre de recherche et d'étude pour la diffusion du français, directed by Michel Martins-Baltar, in 1984 collected the 213,000word Transcriptions d'entretiens (CREDIF) (see Martins-Baltar, Cintrat, James, Mochet, & O'Neil 1989). The corpus, which is very homogeneous, contains 25 guided conversations that touch upon themes such as school, family, profession, death, mass media, food, domestic animals, drugs, and so forth. The recordings took place either at the interviewee's home or work place. The interviewer is always the same person, the interviewees constitute a representative sample with respect to age, sex, profession, region, and number of inhabitants of the town or city. They knew that they were being taped; each conversation, though, contains a final part in which the interviewee is convinced that the tape recorder has been shut. The orthographic transcription contains punctuation marks, extra-linguistic comments, and symbols for vowel lengthening, segmentation, final intonation, pauses, emphatic accent, word truncation, and unintelligible passages. Each text is preceded by a summary which contains basic information about the informant, the conversation, and the situation. Between 1980 and 1990, Kate Beeching recorded 95 interviews with speakers from Paris, Brittany, and the départements Lot, Aude, and Hérault. 45 interviews were conducted with men and 50 with women. In terms of age and level of education, the speakers constitute a representative sample (see Beeching 2002: 69). The corpus, which contains 155,000 words, is freely accessible (see



). In the interviews, the interviewer, almost always Kate Beeching herself,40 usually only suggests topics by asking an initial question but then allows speakers to develop the topics as they like. This procedure guarantees considerable spontaneity. The broad orthographic transcription, with punctuation, includes some indication of pauses but does not systematically mark prosodic features. The 66,700-word Choix de textes de français parlé (CHOIX) (see BlancheBenveniste, Rouget, & Sabio 2002) contains 36 texts of varying length collected by the Groupe aixois de recherches en syntaxe (GARS) of the Université de Provence. Some were recorded between 1991 and 2000, others before 1991. 41 The speakers (children, adults, farmers, professionals, employees, politicians, teachers), discourse types (interviews, conversations, explanations, and narrative texts; private or radio recordings), and themes (films, work, life, economy, hunting, car accidents, etc.) vary; unfortunately the interview style predominates. The speakers come from different social groups and different parts of France, although some départements of Southeastern France (e.g., Bouches-duRhône, Var) are overrepresented. The orthographic transcription, without punctuation, is quite narrow and includes many particularities of pronunciation. In addition, there are a number of special symbols for long or short pauses, interruptions, unintelligible or missing syllables, and overlaps.42 The 440,000-word Corpus de référence du français parlé (CRFP) contains 132 texts deriving from 36.5 hours of transcribed recordings (see Campione 2001a: 91ff.; Bilger 2002: 49f.; Équipe DELIC 2003). Its collection and transcription were carried out in 1999-2000 by a group of researchers of the Université de Provence in Aix-en-Provence coordinated first by Claire BlancheBenveniste and then by Jean Véronis. The recordings took place in 37 cities distributed all over France as well as in five zones of the Paris area. Only two of these cities have less than 10,000 inhabitants. Although geographically representative, the corpus essentially covers only one type of dialogue structure, namely the interview, which can be assigned to three main categories: private interviews regarding personal or professional experiences of the interviewee, interviews in the workplace, and public speech situations (public interviews, radio or television programs, conferences, assemblies, round tables, etc.). Two thirds of the texts belong to the first category, the rest is distributed equally among the other two. Although most of the texts are formally interviews, they feature an almost equal distribution of speaking time between interviewee and interviewer (see Équipe DELIC 2003).43 The interviewees participating in the private interFor this reason, the RPCs contained in the questions of the interviewer of BEECH have been excluded from the analysis. 41 Many texts have a date of transcription (between 1989 and 2000), but no date of recording. 42 See also Blanche-Benveniste (2000: 28-34) for the transcription conventions of the CHOIX and the CRFP. 43 In order to quantify the distribution of speech, the research group has defined a fragmentation index (paramètre de fragmentation de la parole), which varies between 0 (monologue) and 1 (speech distributed equally between speakers). 40



views come from age-groups between 18 and over 65 and have a variety of educational backgrounds. 55 % of them are men, 45 % are women. The orthographic transcription (see Campione 2001b: 9ff.) is basically the same as in the CHOIX. The CRFP is the only corpus used in this research that has not been published. The recent C-ORAL-ROM collection (see Cresti & Moneglia 2005) consists of 772 texts deriving from 121.5 hours of speech and comprises four corpora (French, Italian, Portuguese, and Spanish). I used three of them (COR-FR, COR-IT, and COR-ES) in the research. Since they are comparable in structure and size (roughly 300,000 words for each language), they will presented only in this subsection. The collection is the main outcome of the C-ORAL-ROM project coordinated and managed by the LABLITA laboratory of the University of Florence. Being a multimedia corpus, the DVD on which it is published contains the audio files aligned with the transcripts as well as software for the acoustic and linguistic analysis of the corpus. Most of the texts were recorded between 2000 and 2002, although some of them date back to as early as 1980. They vary according to several semiotic and social parameters (dialogue structure, social context, channel, domain of use, register, and speaker characteristics), but not diatopically, that is, the French texts derive mainly from Southern France, the Italian ones mainly from Tuscany, and the Spanish ones mainly from Madrid and Castile (see Moneglia 2005: 8ff.). The orthographic transcription basically follows the conventions of the CHAT-format (see MacWhinney 1995). Intonation units as well as other prosodic features are marked. The whole corpus is partof-speech-tagged and lemmatized. 4.2.2 Italian corpora Brandi's (1987) article on the verbal interaction between teacher and pupils, in a volume published by the Accademia della Crusca, has an appendix with the 5,800 word transcripts of four school lessons. The recordings took place in 1984 in different classes of two Italian Scuole medie (12-14-year-old pupils) in Prato (Tuscany). They were transcribed orthographically. Punctuation is reduced to question and exclamation marks; short and long pauses, and overlaps are marked. Most importantly, the stretches of transcribed speech are segmented into intonation units. In the same volume, Cresti (1987) illustrates her analysis of information structure in spoken Italian, which is mainly based on the attached, 2,700-word transcript of a spontaneous breakfast conversation between four adults and two children. The conversation was recorded in 1984 in Florence. The orthographic transcription follows the same conventions as Brandi (1987). The Pixi corpora (PIXI) published in Gavioli & Mansfield (1990)44 contain the transcripts of 229 bookshop encounters recorded in Italy (as well as 150 enAn electronic edition of its texts may be obtained at the Oxford Text Archive (see ) 44



counters recorded in England). 180 encounters took place in a large bookshop in Bologna, the others in three different bookshops situated in three different cities of Northern and Central Italy. In Gavioli & Mansfield (1990), there is only an approximative indication of the time in which the recordings were made; probably between 1984 and 1988. The transcripts of the Italian encounters amount to approximately 22,000 words. The shop assistants knew they were being recorded (they carried the tape recorder in their jacket pocket), while the customers were unaware of it. The bookshops being of the self-service type, the exchanges usually occurred when the customer needed help. Most of them concern a specific book or a category of books which the costumer is not able to locate. The orthographic transcription of the exchanges contains punctuation, comments on paralinguistic features, speech rhythm, loudness, the marking of unintelligible words, pauses, overlaps, syllable lengthening and stress. The 490,000-word Corpus del lessico di frequenza dell'italiano parlato (CLIP) was collected in 1990-1992 by a group of linguists coordinated by Tullio De Mauro at the University of Rome La Sapienza. The aims of the initiative were twofold: firstly, to guarantee the necessary basis for a frequency dictionary of spoken Italian that was prepared in collaboration with the research division of IBM Italia and, secondly, to compile a standard reference corpus for spoken Italian. Both were published in one volume shortly afterwards (see De Mauro, Mancini, Vedovelli, & Voghera 1993). The corpus is sufficiently varied to provide a good representation of contemporary spoken Italian: Its 469 texts were collected in the four major cities of Italy (Milan, Florence, Rome, Naples) and represent five basic discourse situations (see De Mauro, De Palo, De Renzo, et al. 1992: 86ff.; De Mauro, Mancini, Vedovelli, & Voghera 1993: 35ff.). The CLIP was originally published on two discs (attached to the frequency dictionary) which contained the broad orthographic transcripts of the recordings. The transcription is orthographic without punctuation; pauses, overlaps, unintelligible words, final vowel lengthening, phatic and hesitation phenomena are noted. A version of the CLIP, enriched with POS-tags and lemmata, can be queried online at the website of the Banca dati dell'italiano parlato (BADIP) (see ; Schneider 2002). The 58,300 word Corpus di italiano parlato (CIP) was collected in the period 1973-1998 and published in printed form and on a CD-Rom in Cresti (2000b). It contains 42 short and 7 long texts covering a broad variety of discourse situations (monologue and dialogue with free or regulated turn-taking, family conversation, discussion in the workplace, interview, debate on television, lesson at school or at university, etc.), and themes (school, university, music, politics, film, etc.). The speakers are aged 2-80 years, have educational backgrounds ranging from primary school to university level and all come from Tuscany. The orthographic transcription follows the CHAT-format (see MacWhinney 1995) and has been enriched by the notation of intonation units as well as of other prosodic features.



Caffi (2001) has an appendix that includes the transcripts of one doctor-patient-encounter and of two psychotherapy sessions that took place in the period 1993-1996 in Northern Italy. The three texts together contain approximately 13,500 words. No information regarding time, place, and identity of participants is provided. The participants knew they were being recorded. The orthographic transcription with punctuation contains symbols for basic intonation, vowel lengthening, loudness, pause length, emphatic stress, speech rate, interruptions, overlaps and unintelligible words. The transcripts also include comments with information about situational factors. Bertoli Sand's (2002) study of negation raising is based on three written and spoken corpora, among them three recordings of Italian television talk-shows. The transcripts (BER), containing together about 19,500 words, are included on the CD-ROM attached to the volume which contains Bertoli Sand's contribution. The talk-shows took place in 1995 and each lasted about 60 minutes. The main participants are the show-master and an Italian politician (on two occasions Massimo D'Alema, the head of the Partito Democratico di Sinistra, on one occasion Augusto Fantozzi, the treasury minister); in two talk-shows, one or more other persons occasionally participate in the conversation. The orthographic transcription without punctuation follows the musical score format and marks basic intonation, vowel lengthening, loudness, stress, pause length, interruptions, and unintelligible words. The texts also contain comments with minimal background information. For the COR-IT, refer to subsection 4.2.1. 4.2.3 Spanish corpora The El habla de la ciudad de Madrid (HCM) (see Esgueva & Cantarero 1981) contains approximately 141,000 words and was collected in the 1970s. It consists of 24 transcribed dialogues between speakers from Madrid. The largest section of the compilation are 16 interviews conducted by a linguist with speakers of either sex aged between 15 and 86 years and mostly with a high level of education, either secondary or university level. This section is followed by four texts spoken by university students who talked with each other knowing that they were being recorded and four further conversations between students that were recorded secretly. The subjects of the interviews and conversations are free and concern the speakers' everyday lives. The texts are characterized by repetitions, false starts and syntactic breaks, which indicate a moderate communicative spontaneity. The transcription is orthographic and includes punctuation. Silent pauses due to hesitation, filled pauses, interruptions, and unintelligible passages are noted. The Encuestas del habla urbana de Sevilla. Nivel culto (HUSNC) (see Pineda 1983), the Encuestas del habla urbana de Sevilla. Nivel popular (HUSNP) (see Ropero 1987), and the Encuestas del habla urbana de Sevilla. Nivel medio (HUSNM) (see Ollero Toribio & Pineda Pérez 1992) are the most im-



portant products of a research project initiated in the early 1970s and dedicated to the study of the sociolinguistic situation of Sevilla. The first corpus contains about 67,000, the other two each about 100,000 words. The first corpus was recorded in the period 1972-1973, the second in the period 1984-1986 and the third one presumably around 1983.45 The structure is identical for the three corpora: Their texts derive from interviews with 24 male and female speakers coming from selected areas of Sevilla (center, Los Remedios, Triana, El Cerro del Aguila, etc.) and belong to three distinct age groups (younger than 30, between 30 and 45, older than 45). Each corpus represents a distinct social class defined by level of education, namely speakers with a university background, speakers with secondary education, speakers with primary education. The speakers knew that they were being taped. The subjects of the interviews mostly concern the interviewees' biographies and everyday lives. The transcription is orthographic with punctuation; pauses are noted. The 1,100,000 word Corpus oral de referencia del español contemporáneo (COREC), an electronic corpus which can be downloaded from a website (), was collected in Spain in the period 1991-92 by a research group headed by Francisco Marcos Marín. It consists of 498 texts collected in 18 social contexts (e.g., administrative, scientific, conversational, educational, etc.). The context can be deduced from the second, third, and fourth letter identifying every text (e.g., acon, ccon, econ, pcon = conversational; acie, ccie, ecie, pcie = scientific). Numerous texts derive from recordings of television and radio programs broad-casted by stations located in Madrid. In many of these, speakers from other Spanish cities intervene. The transcription is orthographic with punctuation; interruptions, hesitation phenomena, phatic communication, onomatopoeic and other sounds, overlaps, and other discourse phenomena are registered. For the COR-ES, refer to subsection 4.2.1. 4.3 Selection criteria The discussion of the criteria in chapter 2 shows that there is no agreement or common understanding concerning what parenthesis is. The applied criteria and the sequences extracted according to these vary considerably. We have also seen that quite a few of the points mentioned, notably prosody, interruption of the host clause, relationship between parenthetical and host clause, and communicative function have not been sufficiently explored. The vagueness of the notion of parenthesis is responsible for the broad range in the contemporary understanding of what RPCs are. Hence, I cannot base my research on a ready-made definition. The criteria provide me, however, with enough elements to define own selection criteria. The following points summarize the properties that according to Urmson characterize a small set of English verbs: Ollero Toribio & Pineda Pérez (1992) do not indicate when the interviews were recorded, but from information given by an interviewee on p. 151 it can deduced that at least that particular interview was recorded in 1983. 45

74 1.


They are in the first person singular of the non-progressive present indicative form. 2. They are either in the initial position and govern an indicative clause introduced by that or are in medial or final parenthetical position. 3. They have emotional, inferential, doxastic or epistemic meanings. 4. Like some modal sentence adverbs (e.g., luckily, consequently, presumably, certainly, probably), they weaken the speaker's claim to truth in statements. Urmson's definition was both semasiologic, i.e., morphological and syntactic, and communicative. Its syntactic side was unconventional, though, and lacked precision - Urmson considered parenthetical merely a convenient term (1952: 481). In the following decades, authors tended to separate the semasiologic from the communicative aspect. Inevitably, this broadened the understanding of what RPCs are. The ambiguities of the notion of parenthesis contributed to this extension. Some authors took up only the communicative function and then ignored form in their classifications of parentheticals (e.g., Corum 1975), others left aside communicative function and adopted only Urmson's semasiologic criteria (e.g., Ross 1973; Emonds 1976; Fava 2001: 45ff.). Like the second group of authors, I start from a semasiologic perspective in this book and in the process of the analysis I will attempt to find out what the communicative roles of the selected expressions are. That means that, initially at least, I do not want to restrict the set of verbs to those with emotional, inferential, doxastic or epistemic meanings. Verbs in parenthetical position may also have to do with perception or speaking. Neither do I adopt the fourth of Urmson's criteria; I am looking for formal criteria that do not make a priori restrictions regarding pragmatic functions. As I said in section 1.1, I treat parenthesis and parenthetical (like incidental) as merely syntactic terms. Therefore I use reduced parenthetical clause as a general term independent of discourse function. Which are the elements we can retain from the discussion in chapter 2? The prosodic criterion alone would select a set of expressions which is far too heterogeneous (cf. Andersen 1997: 127), apart from the fact that the properties illustrated in subsection 2.2.1 are, as already suggested, not typical of the expressions I have in mind. The communicative function, on the other hand, must be excluded as a possible criterion for the reasons mentioned above. The remaining points, though, contain elements which can be used for the selection criteria. I want to preserve the spirit of Urmson's characterization, while introducing some modifications and a few additional specifications. Most importantly, it is not convenient to restrict the group of selected expressions to those verbs in the first person singular of the present indicative. In Romance (but also in English and German), some verbs in parenthetical position are not in the first person singular (e.g., Fr. on va dire), some are not in the present tense (e.g., It. ho detto), others are not in the indicative (e.g., Sp. digamos). Too many interesting and



relevant verbs would be left out. Therefore, the only morphological restriction I introduce is that the parenthetical clauses have to be based on a finite verb. The second important criterion I use is the lack of an overt link. I take into account only those clauses that are not overtly linked to the host clause. There must be neither a link by which the parenthetical clause governs the host clause or a part of it nor a link by which the host dominates the parenthetical (see clauses like It. come penso or Sp. según creo). This criterion also excludes cases - frequent especially in spoken Spanish - where the main clause verb together with the complementizer occurs inside the dependent clause: (1)

I: […] Las gaviotas parece que son como los murciélagos, que ven venir el proyectil, el peligro, antes […] (HUSNP.18.409) [The seagulls seem to be like the bats that see the projectile, the danger before]

Lack of overt linking is not explicitly mentioned by Urmson (1952); the initial clauses in his examples are linked, those in medial or final position are not. A fundamental feature of RPCs, which has been underlined by all authors, is positional mobility, i.e., the clauses can be employed in several different places of the host. In Romance, though, the initial position is far less common than in English or German. Therefore, I cannot require that a clause be acceptable in all three major positions (initial, medial, and final). Hence, the third criterion is that the clause must be acceptable in medial as well as in final position. For the selection procedure, this means that the clause must occur at least twice, once in medial and once in final position, in order to be listed in one of the three tables in the next section. As I said in subsection 2.2.2, the medial position is related to but not the same as interruption. In my view, the parenthetical status also implies that an expression has the ability to interrupt a close syntactic relationship. For the purpose of this study, I deem all relations up to the level of the clause nucleus to be close enough, that is, it is sufficient that a clause occurs between the subject and the predication, between the verb and its complement, or inside any type of phrase in order to be selected. The relation between the clause nucleus and an adjunct is not deemed to be close enough. In the case of subordinate clauses, the syntactic closeness is defined by referring to the clause immediately surrounding the parenthetical. Accordingly, e.g., the relation between the predication and an adjunct of an object clause is not sufficiently close. In order to prove its interruptive potential and to be selected as RPC, a clause has to interrupt a nucleus at least once. The fifth criterion is that the host has to be acceptable without the incidental clause, i.e., the structure of the host has to be totally independent from the parenthetical. This excludes cases such as the following Italian utterance:

76 (2)


C: […] credo si chiami via Luigia San Felice quella parte lì (CLIP.NA12) [I think the part over there is called Via Luigia San Felice]

In Italian, si chiami via Luigia San Felice quella parte lì is never acceptable as an autonomous utterance because it cannot be interpreted without the preceding credo. Note that this criterion only requires that the host is able to fulfill its function in discourse also without the RPC, not that the host is an independent and syntactically complete clause. As I said before, in spoken language, the host may also be a phrase or a single word. The fifth criterion also excludes those formulas that permit the speaker to continue the host clause and to add new interpretations, as Fr. c'est-à-dire in the following utterance: (3)

L1 […] l'important pour moi - euh - quand une personne est en fin de vie c'est de savoir lui dire au revoir c'est-à-dire d'accepter le fait qu'elle n- qu'elle nous quitte […] (CRFP.AMI-PRI002) [The important thing for me, when a person is dying, is to be able to say goodbye to her, that is, to accept the fact that she is leaving us]

C'est-à-dire links two formulations of a constituent (word, phrase, clause) and is an element necessary for the interpretative resumption of a preceding piece of discourse (see Murat & Cartier-Bresson 1987; regarding Sp. es decir see Martín Zorraquino & Portolés Lázaro 1999: 4124). Due to its interpretative function, it cannot appear in final position. The sixth criterion is inherent in Urmson's (1952) and other researchers' accounts: The verb of the parenthetical clause lacks one argument. That means that, although the clause's verb may have governed or ungoverned dependents or a subject, one of its arguments (usually either the subject or the object) must be missing. Due to this criterion, complete parenthetical clauses such as, e.g., Fr. je l'espère are not considered in this research: (4)

M: [...] on le notera je l'espère avec une grande ampleur sur le problème du ticket modérateur [...] (LUD.6.118) [It will be remarked, I hope it, on a large scale regarding the problem of the fixed participation]

For the same reason I excluded It. diciamo così, which, contrary to simple diciamo, neither lacks an argument nor governs noun clauses in the corpora: (5)

A: […] non si può trarre affatto la convinzione che vi sia stata diciamo così una sconfitta storica eh definitiva di quest'ideale […] (CLIP.RD13) [One cannot absolutely deduce the conviction that there has been, let's say it like this, a definitive historical defeat of this ideal]



Note that this restriction is not necessarily implied by the common understanding of parenthesis and therefore sets RPCs apart from other instances of parenthesis. According to narrow definitions, the structure of a parenthetical has to be syntactically complete, that is, identical to that of independent sentences of the respective language. My restriction is explainable, though, by the attempt to select only those clauses that must necessarily be attached to sentences and cannot themselves be used as autonomous, free-standing sentences. The sixth criterion also implies that it is the verb itself, not some other element in the clause, whose valency is not satisfied. Clauses whose verbs neither carry the main semantic load nor actually lack an argument, like those containing a copula plus an adjective (e.g., Sp. estoy seguro or es cierto) or an auxiliary verb plus a nominal object (e.g., Fr. j'ai l'impression or It. ho l'impressione), are therefore excluded from the analysis. The seventh criterion is related to the preceding one: The verb's missing argument should be represented semantically by the host itself, i.e., the missing information can be recovered from the host. Usually, when verbs governing noun clauses are in parenthetical position and one of their arguments is missing, one can assume that this is represented semantically by the host. There is at least one exception, i.e., Sp. perdona. The verb governs noun clauses. However, in the parenthetical occurrences I have found, the missing argument cannot be recovered from the host. The transformation of the host into a governed clause would be syntactically possible but semantically inappropriate: (6)


Doscientas ochenta y ocho quilogramos. [Two hundred and eighty-eight kilograms] ¿Ah! Ha pesado, perdona, doscientas ochenta y ocho quilos […] (COREC.BCON007B) [He weighted, excuse me, two hundred and eighty-eight kilograms] Perdona que haya pesado doscientas ochenta y ocho quilos. [Excuse me that he weighted two hundred and eighty-eight kilograms]

Since there is an overlap and H1 intervenes correcting H2's statement, perdona in (6) must govern an implicit clause meaning something like 'excuse me if I have to correct your statement'. I am now able to summarize the criteria according to which the clauses were selected from the corpora: 1. The parenthetical is based on a finite verb. 2. Between the parenthetical and the host clause there is no overt syntactic link. 3. The parenthetical occurs at least once in medial and at least once in final position. 4. In medial position, the parenthetical at least once interrupts a close syntactic relationship (phrase or clause nucleus).




The host structure (sentence, clause, phrase) is structurally self-sufficient, i.e., it does not depend on the parenthetical. 6. The parenthetical's verb lacks one of the arguments required by its valency. 7. The missing argument can be semantically recovered from the host. Since each example had to fulfill all of these conditions, the order in which they are presented above has no significance. 4.4 The RPCs selected from the corpora Even in electronic corpora there is no easy automatic method by which an RPC can be identified without knowing its exact form, not even in the few POStagged spoken corpora available for French, Italian, and Spanish. I therefore adopted a method that combined 'manual' with electronic research. For each language, I read a considerable number of texts selected from the 22 corpora (totaling 1,350,000 words) and registered in a database all parentheticals. For many parentheticals, there were at least two examples that satisfied the criteria illustrated in the preceding section, for others I had to search all the electronic corpora to check if there were additional examples. I also ran queries on the electronic corpora if I found an RPC in one language and suspected that a comparable one might exist in another. For the queries I used two concordancers, WordSmith Tools 3.00 and Contextes 1.0.0, as well as the query page of the BADIP website. There is a problem that all researchers encounter while analyzing stretches of spontaneous speech. In many instances the free syntactic structure of spontaneous texts makes it difficult if not impossible to figure out the limits of a sentence (cf. Hansen 1998: 113ff.). This applies especially to initial and final parentheticals. In other cases there is no doubt about the parenthetical status of the clause, but it is uncertain where it is attached to. I also want to mention that in spoken corpora many utterances are not syntactically complete sentences; nevertheless, they may host an RPC. Most typically, the host utterance is only a noun phrase. Occasionally, RPCs also appear in dependent clauses that lack a clear connection to a main clause. 1,939 RPCs (tokens) in the corpora were found to satisfy the criteria. These were assigned to 23 French, 25 Italian, and 33 Spanish forms (types). Tables 4.2-4.4 list these forms and the infinitives of the corresponding verbs. In order to assess the frequency of each form, I compiled three comparable samples of systematically analyzed texts totaling approximately 250,000 words for each language. The structure of the samples is explained in more detail in subsection 4.5.1. The percentages in the tables 4.2-4.4 are based on the three samples; a percentage of 0 in the tables means that a form occurs only in texts that are not contained in the sample. Except in the imperative, French verbs require a pronominal or nominal subject, Italian and Spanish verbs do not. Those Italian and Spanish RPCs with an additional (facultative) noun or pronoun are listed with



NP or PRO. The noun phrase or the pronoun may precede or follow the verb, e.g., It. Il Corriere della Sera dice or dice Calvino and Sp. yo pienso or pienso yo. In all three tables, Fr. dire and its Italian and Spanish equivalents are the verbs with the greatest varieties of forms and with several high frequency forms. The frequency of It. diciamo is particularly impressive, especially in comparison to that of the other Italian RPCs, and suggests that in Italian mitigation is heavily centered on this form. Whereas its Spanish equivalent digamos is also quite frequent, Fr. disons plays a minor role. Apart from this verb, I found only two other utterance verbs, i.e., Sp. insistir and repetir. Another group of RPCs centers on belief verbs, that is, on doxastic verbs such as Fr. croire and penser. Fr. savoir is concerned with the presence of knowledge, i.e., it is an epistemic verb. However, two of its forms (Fr. je sais pas and je sais plus) express absence of knowledge and therefore indirectly act as expressions of belief. RPCs such as Fr. j'imagine, mettons, je me rappelle, and je suppose are based on verbs that are best characterized as denoting mental operations. Another group of parentheticals are based on verbs expressing sensory (Fr. voir) or mental (Fr. paraître and sembler) perception. Comparing Borillo's (1982) classification of infinitives (see table 1.1 in section 1.3) with the RPCs in table 4.2, one notes, besides a basic correspondence, some differences. Borillo takes only those predicates into account that may be used as answers to polar questions (see section 7.3 for the description of her test), therefore RPCs such as, e.g., Fr. disons or mettons, are not included in her table. On the other hand, I consider only clauses whose verbs carry the main semantic load, not sequences containing a copula plus adjective or an auxiliary verb plus a nominal object; hence Fr. il est probable, j'ai l'impression, etc. are not contained in my list. These reasons also account for the main differences between the Italian table 1.3 in section 1.3 and table 4.3 in the present section. Andersen (1997) (see table 1.2 in section 1.3) includes among the propositions parenthétiques non prototypiques a series of forms in the second person most of which I excluded because it turned out that they do not interrupt phrases or clause nuclei.


REDUCED PARENTHETICAL CLAUSES Infinitive Form Percentage croire je crois 10.2 dire disons 6.3 dire dites-vous 0.3 dire il faut dire 0.3 dire je dirais 4.5 dire je veux dire 11.4 dire on dirait 0.3 dire on va dire 0.6 espérer j'espère 0.6 imaginer j'imagine 1.5 mettre mettons 1.5 paraître paraît-il 0.9 penser je pense 3.9 savoir je sais pas 7.2 savoir je sais plus 1.2 savoir tu sais 9.3 savoir vous savez 11.4 sembler me semble-t-il 0.0 se rappeler je me rappelle 0.3 supposer je suppose 1.2 trouver je trouve 1.5 voir tu vois 19.8 voir voyez-vous 5.7 Table 4.2: French reduced parenthetical clauses

Infinitive Form Percentage credere credo 6.4 dire devo dire 2.1 dire dice 6.2 dire dice NP 1.7 dire diciamo 41.0 dire dico 7.5 dire direi 3.4 dire ho detto 1.1 dire voglio dire 5.8 dire volevo dire 0.6 figurarsi figurati 1.1 immaginare immagino 0.2 parere mi pare 1.1 pensare penso 1.9 pensare penso PRO 0.9 porre poni 0.9 sapere mi sa 0.9 sapere non so 5.2 sapere sa 1.9 sapere sai 3.6 sembrare mi sembra 2.4 sembrare sembra 0.2 sperare spero 1.9 vedere vede 0.4 vedere vedi 1.5 Table 4.3: Italian reduced parenthetical clauses



Hooper's (1975) (see table 1.4 in section 1.3) strong assertives are primarily speech activity verbs. Apart from decir, I have not found parenthetical examples of contar, decidir, afirmar, and explicar in the corpora. Hooper's semi-factives comprise saber plus verbs of sensory or mental perception. In Romance it is necessary to distinguish between the perfective meaning of saber in the past tense (e.g., supe) and past perfect tense (e.g., he sabido) and its imperfective meaning in the present tense (e.g., sé) and imperfect tense (e.g., sabía). Perfective saber means 'come to know, learn, hear' and can be related to the verbs of mental perception; it is a semi-factive verb. Neither in Spanish nor in French or Italian did I find RPCs based on the perfective verb form. Imperfective saber ('know') is a cognitive factive verb (see subsection 6.4.2) and a frequent parenthetical in the corpora. I have not found parenthetical examples of darse cuenta. Enterarse and comprender occur as parentheticals, though never in an interrupting position. Infinitive Form Percentage acordarse me acuerdo 0.0 creer creo 2.9 creer creo PRO 5.7 decir decía NP 0.0 decir dice 6.1 decir dicen 0.0 decir digamos 22.0 decir digo 17.1 decir digo PRO 2.0 decir diría 0.8 decir diría PRO 0.4 decir diríamos 1.6 decir ha dicho 0.4 decir ha dicho NP 0.0 decir me han dicho 0.4 decir podríamos decir 0.4 decir quiero decir 3.3 decir ya digo 2.4 decir ya te digo 2.9 fijarse fíjate 0.4 imaginar imagino 0.0 imaginar imagínate 0.0 imaginarse me imagino 1.6 insistir insisto 0.8 parecer me parece 3.7 parecer me parece a mí 0.8 parecer parece 0.4 pensar pienso PRO 0.4 repetir repito 0.4 saber no sé 4.5 saber sabes 17.1 suponer supongo 1.2 ver ya ves 0.0 Table 4.4: Spanish reduced parenthetical clauses

The tables above contain a subset of those clauses that occur in parenthetical position. In the main, two conditions are responsible for the exclusion of clauses



from the tables above: the condition that the clause has to occur in final position and that the clause has to occur in an medial position interrupting a close syntactic relation. One case for each was sufficient to satisfy these conditions. Yet, for each language there are several excluded clauses that satisfy only one of the two. Table 4.5 lists the most frequent of these forms, their respective infinitives, and the number of examples found. Those marked with yes interrupt at least once a close syntactic relation but do not occur in final position, those without yes satisfy only the second condition. Infinitive


dire voir dire voir dire sembler savoir dire

je dis voyez je dis PRO vous voyez j'allais dire il me semble je sais pas moi j'ai dit

sentire ripetere dire mettere credere capire dire pensare

senti ripeto dicevo mettiamo io credo capisci m'ha detto pensi

Interrupting? French yes

yes yes Italian yes yes yes yes yes yes Spanish

Occurrences 26 10 8 8 8 7 7 7 85 19 16 12 9 8 4 3

oír oye 15 saber ya sabes 7 parecer parece ser 4 saber ya se sabe 4 comprender comprendes 3 ver verás 3 pensar pienso 3 decir dijéramos yes 2 Table 4.5: Parenthetical clauses satisfying all criteria but one

The reasons why these and other forms do not achieve the status of RPCs are either the limitations of a still too small empirical basis or the fact that certain forms apparently have a restricted positional variability. With larger corpora, I might have found, e.g., an interrupting Fr. j'ai dit, a final It. mettiamo, or an interrupting Sp. pienso. These forms achieve the status of RPC in at least one other Romance language. On the other hand, despite a large number of attested cases and the fact that Sp. oye appears sometimes in final position, It. senti probably does not possess the positional mobility characteristic of parentheticals. It never appears at the end of an utterance. It is used mainly at the beginning, in the few medial cases it is placed between an initial adverb and the clause nucleus; in one occurrence only does it interrupt the nucleus.



4.5 The frequencies of RPCs In what follows, I examine the frequency of RPCs, first in general and then by taking into account the discourse situations and domains of use in which the texts were produced. I also consider the degree of fragmentation of the texts, that is, if they represent mainly monologic or mainly dialogic discourse. Discourse situation, domain of use, and degree of fragmentation are external criteria, which group texts according to properties other than lexical or grammatical features (cf. Günthner 1995: 203ff.; Lee 2002: 248ff.). 4.5.1 The three samples In order to permit frequency comparisons, I assembled three samples of systematically analyzed texts totaling approximately 250,000 words for each language. They are similar in terms of discourse situation, domain of use, and degree of fragmentation; other parameters, e.g., age, profession, educational background, and gender, have not been taken into account. Future research, based on a corpus containing detailed metatextual data, such as, e.g., the C-ORAL-ROM, should analyze the impact of these factors. Each text was assigned to one out of four discourse situations, whose definitions are similar to those adopted in the Italian CLIP (see De Mauro, De Palo, De Renzo, et al. 1992: 86ff.; De Mauro, Mancini, Vedovelli, & Voghera 1993: 35ff.): A B C D

Discourse situations bidirectional discourse face to face with free turn-taking bidirectional discourse not face to face with free turn-taking bidirectional discourse face to face with regulated turn-taking

Examples conversations at home, at work, at school or university telephone conversations legislative and other types of assemblies, oral exams at school or at university, interrogations in the courtroom, interviews school or university lectures, presentations, unidirectional discourse sermons, political speeches, unidirectional discourse on radio or television Table 4.6: The four discourse situations

Discourses of type A and B are dialogic, while those of type D are monologic. The discourses of type C are quite heterogeneous: In numerous texts, one participant dominates in such a way as to transform the dialogue into a quasimonologue. In order to account for these cases and inspired by the paramètre de fragmentation de la parole in the CRFP (see Équipe DELIC 2003), I assessed the degree of fragmentation (high or low) of the texts of type C. As domains of use, I adopted those defined for the British National Corpus (see ):


REDUCED PARENTHETICAL CLAUSES Examples lectures, news broadcasts, classroom discussion, tutorials sales demonstrations, trades union BU business meetings, consultations, interviews sermons, political speeches, council IP institutional and public meetings, parliamentary proceedings sports commentaries, after-dinner LE leisure speeches, club meetings, radio phoneins Table 4.7: The four domains of use EI

Domains of use educational and informative

The three parameters, discourse situation, degree of fragmentation, and domain of use, produce a grid with 20 text types. Table 4.8 shows the grid and the amount of text (in number of words) for each language. EI BU IP LE Total


6355 35096 8538 28224 78213


5968 38082 6253 28520 78823


9032 26199 8378 32745 76354


French C­high­frag C­low­frag D 0 369 32338 7549 27334 0 0 17157 1580 7597 12683 20351 15146 57543 54269 Italian 0 27253 11004 5349 10342 9670 0 16671 16900 11647 15083 2670 16996 69349 40244 Spanish 0 12392 9292 4190 13014 22595 0 17424 9605 12712 20630 4160 16902 63460 45652 Table 4.8: The three samples

24063 3370 17430 0 44863


63125 73349 44705 68855 250034

24247 2756 17671 0 44674

68472 66199 57495 57920 250086

30184 2125 15530 0 47839

60900 68123 50937 70247 250207

The three samples are structured in such a way so as to produce comparable quantities of text for the totals of each macro-type. Due to the limitations and heterogeneity of the corpora, it was not always possible to compile comparable quantities for every single text type (e.g., French+C-high-frag+EI vs. Italian+Chigh-frag+EI). 4.5.2 Overall frequencies To start with a general statement, in the systematically analyzed texts I found a higher frequency of RPCs for Italian than for French and Spanish. In figure 4.1, the relative frequency of RPCs in the Italian texts is 0.186 % (approximately one RPC every 500 words), 0.133 % in the French ones and 0.097 % in the Spanish ones. The ratio does not change significantly if we take into account only mitigating RPCs, i.e., parentheticals exclusively reducing speaker responsibility (not phatic or reporting RPCs; cf. table 6.1 in section 6.1).



0.200 0.175

All RPCs Mitigating  RPCs with 'let's  say' Mitigating  RPCs without  'let's say'

0.150 0.125 0.100 0.075 0.050 0.025 0.000 French



Figure 4.1: RPC percentages for each language

There is, however, one important thing to note: As becomes apparent from table 4.9 on the next page, the frequency of Italian RPCs is partly the result of the persistent use of diciamo (41 % of Italian RPCs), a word that dominates Italian conversation. In fact, if we exclude Fr. disons, It. diciamo, and Sp. digamos from the group of mitigating RPCs, the difference between French and Italian becomes insignificant. Table 4.9 combines tables 4.2-4.4, but presents the RPCs in order of decreasing frequency. The percentages are based on the three samples. Each value indicates, separately for each language, the percentage of a form out of the total of occurrences of the listed forms. A percentage of 0 means that a form occurs only in texts that are not contained in the sample. According to the quantitative study by Biber, Johansson, Leech, et al. (1999: 667ff.), based on a large corpus of spoken English, the frequency of RPCs in English conversation must be at the level of Italian or even higher. The authors state that the single most common verb controlling that-clauses is think (mainly in the form I think and with the complementizer being regularly omitted). This verb alone has a frequency of about one occurrence per 500 words. Firenzuoli & Tucci (2003: 187) and Tucci (2004: 5ff.) measure the frequency of parentheticals of all sorts in a subset of the Italian CIP corpus. They do not exactly indicate the texts chosen nor do they give the total number of words of the subset. I estimate it to be approximately 54,000. They count 248 avverbi modalizzanti; that means according to my estimate about one modal adverb in 218 words. Their classification of parentheticals is different from mine; in their framework, there are no RPCs. The class that corresponds closest to RPCs appears 154 times, i.e., approximately one parenthetical in 351 words. These data also suggest that the frequency of Italian RPCs corresponds to my figures.



French tu vois je veux dire vous savez je crois tu sais je sais pas disons voyez­vous je dirais je pense j'imagine mettons je trouve je sais plus je suppose paraît­il on va dire j'espère dites­vous il faut dire on dirait je me rappelle me semble­t­il

Italian 19.8 diciamo 11.4 dico 11.4 credo 10.2 dice 9.3 voglio dire 7.2 non so 6.3 sai 5.7 direi 4.5 mi sembra 3.9 devo dire 1.5 penso 1.5 sa 1.5 spero 1.2 dice NP 1.2 vedi 0.9 ho detto 0.6 figurati 0.6 mi pare 0.3 penso PRO 0.3 poni 0.3 mi sa 0.3 volevo dire 0.0 vede immagino sembra

Spanish 41.0 digamos 7.5 digo 6.4 sabes 6.2 dice 5.8 creo PRO 5.2 no sé 3.6 me parece 3.4 quiero decir 2.4 creo 2.1 ya te digo 1.9 ya digo 1.9 digo PRO 1.9 diríamos 1.7 me imagino 1.5 supongo 1.1 diría 1.1 insisto 1.1 me parece a mí 0.9 diría PRO 0.9 ha dicho 0.9 me han dicho 0.6 podríamos decir 0.4 fíjate 0.2 parece 0.2 pienso PRO repito me acuerdo decía NP dicen ha dicho NP imagino imagínate ya ves Table 4.9: Frequencies of French, Italian, and Spanish RPCs

22.0 17.1 17.1 6.1 5.7 4.5 3.7 3.3 2.9 2.9 2.4 2.0 1.6 1.6 1.2 0.8 0.8 0.8 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.4 0.0 0.0 0.0 0.0 0.0 0.0 0.0

4.5.3 Frequencies related to the text type As I said before, each text of the three samples was assigned to one of four discourse situations (see table 4.6). The texts of type C were further subdivided into C-high-frag and C-low-frag according to their degree of fragmentation. The texts were also assigned to one of four domains of use (see table 4.7). This produced a grid with 20 text types for each language (cf. table 4.8). Only those text types with at least 5,000 words of text were taken into account. Thus, in French there are data for 13 text types, in Italian for 15, and in Spanish for 14. Figure 4.2 shows the frequencies (in percents of the total numbers of words of the text type) of French mitigating RPCs (not phatic or reporting RPCs) related text types:





0.225 0.200 0.175 0.150 0.125 0.100 0.075 0.050 0.025 0.000 c­hi­ fr+ip



c­hi­ fr+le


c­hi­ fr+bu




c­lo­ fr+le


c­lo­ fr+ei


Figure 4.2: Percentages of French mitigating RPCs related to text types

The French data show a clear difference between highly fragmented dialogic texts (C-high-frag and A), where mitigators occur more frequently, and less fragmented or monologic texts (C-low-frag and D) with a lower percentage of mitigators. Interestingly, telephone conversations (B) show comparably low percentages. There is also a tendency, less clear though, for texts of the institutional-public domain to show higher percentages. Figure 4.3 shows the frequencies of Italian mitigating RPCs related to text types: 0.275


0.250 0.225 0.200 0.175 0.150 0.125 0.100 0.075 0.050 0.025 0.000 c­hi­ c­hi­ c­lo­ a+ip fr+ip fr+le fr+ei

a+ei c­lo­ c­hi­ a+le c­hi­ a+bu d+ip fr+ip fr+bu fr+ei

d+ei b+bu b+le c­lo­ fr+bu

Figure 4.3: Percentages of Italian mitigating RPCs related to text types



The tendencies mentioned above are also present in the Italian data. Texts of types C-high-frag and A have higher percentages than texts of type D; C-lowfrag texts, however, achieve high percentages. Again, telephone conversations (B) show low percentages and texts of the institutional-public domain show relatively high percentages. Regarding the role of social distance in speech behavior, two general hypotheses have been proposed. According to Brown and Levinson's (1978, 1987) fundamental work, indirectness and degree of politeness increase proportionally with social distance. E.g., in a conversation between status unequals or between strangers they maintain that we should expect a higher frequency of mitigating expressions and strategies than in a conversation between intimates. Wolfson (1988, 1989) proposes the 'bulge' model to account for the role of social distance in speech behavior. According to her findings, people are direct and brusque with intimates and with strangers, but indirect and polite with people in the middle of the spectrum of social distance, such as colleagues and friends. The explanation for this rather unexpected tendency lies in the 'relative certainty' of the relationship between the speakers. Between intimates, as well as total strangers, there is a 'fixed state' in which the speakers know what to expect from one another. Relationships among status-equal friends and acquaintances are dynamic, open to negotiation, and thus relatively unstable. In these relationships, people take special care to signal solidarity and to avoid confrontation. E.g., compliments are more frequent between speakers who are neither intimates nor total strangers. If we apply them to mitigating RPCs, the two hypotheses make different predictions. Following Brown & Levinson's (1978, 1987) model, we should expect frequencies to increase with social distance; being low between intimates, higher between coworkers and between acquaintances, and highest between strangers. According to the 'bulge' model, on the other hand, the frequencies should be low between intimates, high between coworkers and between acquaintances, and low between strangers. Although my classification of texts is still very coarse-grained, it is likely that social distance increases from texts of type A to texts of type D. Under this assumption, the French and Italian data are more in line with Wolfson's (1988, 1989) model. But there must be a second important factor, namely the fragmentation of discourse. The French and Italian data suggest that the frequency of mitigating RPCs rises when the discourse is fragmented and dialogic. In other words, a speaker is more likely to use an RPC in an exchange with one or more actively participating interlocutors. Andersen (1997: 18) confirms this view, BlancheBenveniste (2000: 22) though holds that at least incidentes are more frequent in monologues than in dialogues. Figure 4.4 shows the frequencies of Spanish mitigating RPCs related to the text types:





0.120 0.100 0.080 0.060 0.040 0.020 0.000 c­lo­ fr+ei




c­hi­ fr+bu


c­lo­ fr+ip


c­hi­ fr+ei



c­hi­ fr+ip

c­hi­ c­lo­ fr+le fr+bu

Figure 4.4: Percentages of Spanish mitigating RPCs related to text types

Unfortunately, the tendencies present in the French and Italian data are neither confirmed nor disproved by the Spanish data, that is, in the Spanish texts it is difficult to recognize a clear tendency, regarding the discourse situations or the domains of use. Texts of type C-high-frag have middle or even low percentages; texts of type A have high or low percentages; texts of type D have middle percentages; telephone conversations (B) reach relatively high percentages; Clow-frag texts have very high or very low percentages. The distribution of the domains of use is characterized by the same indeterminacy. 4.6 Summary The chapter starts by illustrating the empirical research approach. The selection of RPCs from the corpora is based on formal, i.e., semasiologic properties. Then the 22 Romance corpora used in the study are presented. They comprise a total amount of 3,975,500 words. In the following section, I explain and discuss my understanding of reduced parenthetical clause, often with reference to the criteria already investigated in chapter 2. Seven selection criteria are formulated and used to extract the RPCs from the corpora. 1,939 RPCs (tokens) in the corpora were found to satisfy the criteria. These were assigned to 23 French, 25 Italian, and 33 Spanish forms (types). A series of other forms satisfied all but one of the seven conditions: Either none of their tokens occurred in final position or none of their tokens interrupted a clause nucleus. Fr. dire and its Italian and Spanish equivalents are the verbs with the greatest varieties of forms derive and with several high frequency forms. Finally, I examine the frequencies of these RPCs, in general and by taking into account the discourse situations and domains of use in which the texts were produced as well as the degree of fragmentation of discourse. In general, the frequency of RPCs is high in Italian, lower in French and lowest in Spanish. However, this is due mainly to the persistent use of diciamo in the Italian texts. If we exclude Fr. disons, It. diciamo,



and Sp. digamos from the group of mitigating RPCs, there is no difference between French and Italian. Whereas in the French and Italian data mitigating RPCs achieve high percentages in dialogic texts of type C (face to face, regulated turn-taking) produced in the institutional-public domain and low percentages in monologic texts produced in all kinds of domain, no clear tendency can be observed in the Spanish data.

PRAGMATIC ANALYSIS: FUNDAMENTAL ISSUES 5.1 Mitigation Hare (1970) was the first to emphasize a distinction regarding the logical structure of utterances that is extremely valuable for the understanding of statements. One of the few linguists to recognize its importance early on was Lyons (1977: 749ff.). According to Hare, who refines a binary distinction already introduced into philosophy by Frege (1891) and Russell & Whitehead (1910) and into linguistics by Bally (1932) (cf. dictum versus modus), there are three parts that contribute to the meaning of an utterance. The phrastic conveys the propositional content, that is, the part of an utterance common to corresponding assertive, interrogative, and directive utterances. This term is certainly preferable for focused questions which do not express complete propositions (see Levinson 1983: 242).46 The tropic classifies a sentence according to the speech act it is characteristically used to perform, that is, it indicates the illocutionary force of an utterance. Hare also calls it mood sign; in fact in many languages it is grammaticalized in the category of mood. The neustic is the sign of subscription to the speech act that is being performed and expresses the speaker's commitment to the factuality, desirability, etc. of the propositional content (cf. also Heger 1976: 277). Hare's distinction between tropic and neustic is, in its main outlines, the same as the distinction between illocutionary point and illocutionary force proposed by Searle (1976: 3ff.): The point or purpose of a type of illocution I shall call its illocutionary point. Illocutionary point is part of but not the same as illocutionary force. Thus, e.g., the illocutionary point of requests is the same as that of commands: both are attempts to get hearers to do something. But the illocutionary forces are clearly different.

However, in Searle's taxonomy of illocutionary acts differences in the force with which an illocution is presented do not play a significant role (see Bublitz 2001: 108ff.). In most works on speech acts, the terms illocutionary point and illocutionary force are used interchangeably (see Holmes 1984: 345). This means that in standard speech act theory it is not clear if the illocutionary point or the illocutionary force may vary in the strength of the speaker's commitment. In other words, it is not clear if within the same speech act type there are variations of force. The concept of illocutionary force indicating device (IFID) is affected by the same uncertainty. Do different IFIDs indicate different speech acts or just different strengths of the same speech act? For a recent critical exam of the notions of proposition and propositional content, see Vignolo (2001). 46



Hare's (1970) differentiation between tropic and neustic resolves this uncertainty and defines the two extremes of what is probably a continuum: on the one hand, the plain and simple indication of the type of speech act, on the other hand the variation of its strength in terms of the involvement and commitment of the speaker and hearer. As we will see, in reality, mitigators such as RPCs always contain elements of both: They indicate a certain tropic and mitigate the associated neustic. They do this, however, to different extents, that is, they represent different points on the aforementioned continuum. For this reason, I find Hare's distinction heuristically useful and maintain it throughout my analysis. Moreover, this differentiation corresponds very closely to the one between modalité d'énonciation (declarative, interrogative, etc.) and modalité d'énoncé (evaluation by the speaker) one can find in some French grammars (e.g., Riegel, Pellat, & Rioul 2002: 580f.). A second heuristically useful distinction has been proposed by Caffi (1999, 2001) for the function of mitigating devices in statements and in directives. I have to mention in advance that in Caffi's framework the pragmatic notion mitigation assumes a broad sense. Whereas for Fraser (1980), 47 who introduced the notion into modern pragmatics, it only means the reduction of the unwelcome effects that a speech act may have on the addressee, for Caffi it covers all macro- and micro-strategies by which the speaker tries to avoid the risks arising from linguistic interaction. One of these risks may be, for example, that the speaker is held responsible for what he or she is stating. According to the utterance component they focus on, mitigating devices can be grouped into three types. Extending Lakoff's (1972) felicitous botanic metaphor and pursuing an idea already present in Prince, Frader, & Bosk (1982), Caffi (1999, 2001) calls them bushes, hedges, and shields. Bushes focus on the propositional content: They make less precise or minimize referring terms or predicates and thus may affect the truth value of the phrastic, if it is asserted (cf. Prince, Frader, & Bosk 1982: 86). In the following example taken from Caffi (1999: 892, 2001: 309), the Italian diminutive suffix -ino attached to un problema mitigates the doctor's diagnosis: (1)

D: ma quello è un problemino. - non è mica un problema grosso. [But that's a small problem - that's not a big problem]

The inventory of Italian devices that function as bushes comprises diminutive suffixes (cf. Dressler & Merlini Barbaresi 1994)48 as well as expressions like un po', una specie di, un certo, and a few others; English bushes are, for example, On p. 348f., however, Fraser (1980) discusses the use of RPCs (as one of several ways by which the speaker can indicate his or her intent to mitigate). 48 Butt & Benjamin (2004: 551ff.) point to the pragmatic function of the Spanish diminutive suffixes. For two Spanish (Chilean) examples, see Puga Larraín (1997: 96, 100): (i) Tratemos de bajar por atrasito, atrás 'tá la bajá, por favor, no ve que por aquí pasan parte. (ii) Está un poquito gordito. 47



a certain, a sort of (cf. Aijmer 2002: 175-209), and a kind of. Bushes operate, in Hare's (1970) terms, on the phrastic. Hedges focus on the illocutionary force indication and / or on the speaker's commitment. They do not affect the truth value of the phrastic, even if it is asserted (cf. Prince, Frader, & Bosk 1982: 89). Caffi (1999: 893, 2001: 279ff., 311) cites as examples the Italian hedged performative (see Lakoff 1972: 213; Fraser 1975) direi che, the modal adverb probabilmente, and the parenthetical mi sembra: (2) (3)

M: […] ma direi che va abbastanza bene eh?. [But I'd say it's quite all right, isn't it?] D: probabilmente è - dove c'è l'attaccapanni - probabilmente è una conseguenza di un problema intestinale che è cominciato con l'influenza eh. [Probably it is - where the clothes-stand is - it is probably the consequence of an intestinal problem that began with the flu]

Whereas the conditional in direi che acts on the indication of the illocutionary force, probabilmente acts on the speaker's commitment. In commenting on (3), Caffi writes: Here the scope of the mitigation is (that aspect of the illocution which is) the speaker's epistemic commitment to the propositional content. Probabilmente weakens the speaker's degree of certainty about the proposition […]. (Caffi 1999: 893)

The role of the modal adverb is analogous to that of the parenthetical mi sembra (see Caffi 2001: 280). The illocutionary-centered hedges operate, in Hare's (1970) terms, on the tropic and / or the neustic. In shields, the mitigating operation takes place in various manners; either by shifting one of the deictic components of the utterance (ego, hic, nunc) or by suspending its literal interpretation or by backgrounding one of its topics. The deictic displacement may be due to a lexicalized expression and to operations affecting syntax, as in passive transformations or in the shift from first person singular pronouns to other person pronouns.49 An actantial shield is based on the ego and is typically represented by utterances ascribed to a source other than the actual speaker or with the author being deleted, as in impersonal or passive constructions (see Caffi 1999: 896, 2001: 315). The following example illustrates what Caffi calls a non-ego strategy of objectivization: (4)

D: c'è un'iperplasia estrogenica - c'è scritto qui. [There's an estrogenic hyperplasia - it's written here]

A narrativization shield involves the nunc and describes present events or state 49

For Spanish examples, see Haverkate (1994: 129ff.).



of affairs as if they had happened in the past. Displacements involving another possible world (the 'as if') are fictionalization shields, those created by hypothetical sentences are eventualization shields. While the deictic shields substitute one of the aspects of the deictic triad with its negation, the following two shields operate on expressions that remain in the utterance together with the mitigating device. In quotational shields the speaker opens up a meta-level, by using expressions like Italian fra virgolette, per così dire or diciamo così or English so to speak or let's say, and distances him- or herself from what he or she is saying (see Caffi 1999: 900, 2001: 321). The suspension of the literal meaning implies that the speaker's commitment is suspended too. With a topical shield, the speaker backgrounds an embarrassing, painful or delicate topic; either by making a strategic digression (see expressions such as Italian tra l'altro or per caso or English by the way or incidentally) or by giving a strategic example (see Caffi 1999: 901f., 2001: 322f.). Caffi does not classify shields according to Hare's (1970) distinction. It is quite clear, though, that some actantial and quotational shields indirectly involve the neustic (cf. Prince, Frader, & Bosk 1982: 89). By adding, e.g., c'è scritto qui, the speaker makes it clear that he or she is not the (only) source of the statement and that he or she cannot be held fully responsible for it. Thus the speaker reduces implicitly his or her commitment. Obviously, mitigation can be interpreted within the framework of Grice's (1975, 1978) cooperative principle, especially with respect to the quality maxim. In this view, mitigators such as RPCs are expressions indicating that the speaker is not fully adhering to the maxim. In other words, they indicate an implicature.50 Mitigation may also be seen from the perspective of felicity conditions (see Holmes 1984: 347f.; Caffi 2001: 291ff., 448ff.). By stating, the speaker commits him- or herself to the truth of the proposition and tries to make the hearer believe the proposition. This is an essential condition of statements (see Searle 1969, 1976) and also one of its constitutive rules (see Rawls 1955; Searle 1969), because the condition creates or constitutes the speech act. Mitigation often acts on this condition, either by adding fuzziness to the proposition or by deintensifying the speaker's attempt to make the hearer believe something. Regarding the second point, Caffi (2001) sees a parallel with directives, where mitigation often acts on the intensity of the speaker's attempt to get the hearer to do something. A general felicity condition of speech acts is that the procedure must be executed correctly and completely (see Austin 1976 [1962]: 15). Bushes, which render the propositional content less precise, indicate that this condition is not completely satisfied (see Caffi 2001: 307, 450). In concluding the illustration of Hare (1970) and Caffi (1999, 2001), I wish to add three remarks. Firstly, besides mitigation there also is the possibility of reinforcement. However, as I will outline below, speaker commitment cannot be Jayez & Rossari (2004) hypothesize that parentheticals convey conventional implicatures, i.e. lexically triggered constraints on interpretation. 50



reinforced above the degree present in a categorical assertion. Therefore, RPCs cannot but mitigate speaker commitment.51 Secondly, Hare (1970: 20) concentrates on statements, but says in a short remark that the neustic is present in other speech acts as well: For this sin I will now try to atone by using the term "neustic" more narrowly for the sign of subscription to an assertion or other speech act […]

As we will see in the examples (7) and (8), the neustic component does not exclusively apply to statements in Hübler's (1983) framework. In Caffi (1999, 2001), mitigation also pertains to illocutionary macro-types, namely to constative-verdictives as well as to directive-exercitives, and reduces speakers' obligations in the first type of speech act and hearers' obligations in the second type. This makes it possible to unify mitigation related to epistemic modality and mitigation related to deontic modality. Similarly, in my framework, mitigation is a notion relevant for both macro-types, although I focus on statements. The obligations generated by the two types of speech acts have different aims: Constative-verdictives require the existence of a state of affairs, whereas directive-exercitives oblige to a future verbal or non-verbal action (cf. Langner 1994: 45). Thirdly, now that we have isolated the meaning components of statements, we know that there are different scopes for mitigation. In fact, wherever it is helpful and possible, I specify the function of the RPC, e.g., say it acts preferably on the phrastic or on the tropic and so on. However, in many instances it is difficult, if not impossible, to determine the function of an expression; as we will see, RPCs may, furthermore, have more than one function. Particularly in statements, the effect of mitigation acting on the phrastic and mitigation acting on the neustic is comparable; the distinction may have little conceptual importance (see Brown & Levinson 1978: 152, 1987: 147). It also has been pointed out (see Sbisà 1990; Caffi 2001: 62ff.) that in terms of the linguistic devices employed, the relationship between the mitigation of the neustic and the indication of the tropic is very close. For example, an adverb such as please downgrades the force of the request and at the same time characterizes the act as a request (see Blum-Kulka 1985; House 1989). RPCs with doxastic verbs mitigate the neustic and by being limited to assertive tropics also imply that the utterance is a statement.52 In the case of such illocution-bound mitigators (see Caffi 1999: 886, 2001: 297), specialized for a certain illocution, it is convenient to treat Holmes (1984: 359) is among the few linguists to state that "I think and I believe are parenthetical verbs which, with different patterns and in different contexts, may boost or attenuate the force of the utterance they modify". 52 In the literature about RPCs based on epistemic verbs, illocutionary force indication is rarely  mentioned as function of these clauses. As I pointed out in section 3.1, a rather early exception  is Wackernagel (1897), who remarks on 23f. of his contribution about parataxis in Classical  Greek and Latin: "Eine Bitte oder eine Aussage kann man in einem (sic) Nebensatz geben, der  von einem Verben (sic) des Bittens oder des Sagens oder Meinens abhängt. Giebt (sic) man sie  aber in einem (sic) Hauptsatz und schiebt das Verbum des Bittens u.s.w. ein, so erhält dieses  oft sehr schwachen Ton und nähert sich seinem Wesen nach einer die Satzart anzeigenden Par­ tikel". 51



neustic mitigation not as superficially adjoined to an independently indicated tropic, but as closely connected to or even identical with it (see Sbisà 2001). Mitigation can be interpreted as a strategy or process (nomen actionis) or as a result or effect (nomen acti) (see Caffi 1999: 884). I will now illustrate the effects mitigation may have on linguistic interaction. Hübler (1983) takes up Hare's distinction and carefully analyzes what he calls assertorical sentences (statements about the actual state of affairs) and practical sentences (fiats, i.e., statements about a desirable state of affairs). E.g., the following assertorical sentences (see Hübler 1983: 11) have the same phrastic and the same tropic, but differ with respect to the neustic: (5) (6)

It's snowing in the mountains. It's snowing in the mountains, I suppose.

Similarly, the following practical sentences differ only with respect to the neustic insofar as the order in (7) becomes a request in (8): (7) (8)

Open the window! I ask you to open the window.

Hübler redefines and deepens the neustic component in terms of the speakerhearer relationship, i.e., as the component of an utterance that expresses the attitude of the speaker to the hearer regarding the proposition. In the case of a statement, this attitude amounts to expecting agreement, in the case of an order or a request to expecting compliance. Hübler notes that any statement requires acceptance and ratification by the hearer and thus reveals its negatability: The hypothetical nature of the sentence is based on the fact that the speaker has a relatively free choice between various sentences. Of course, it follows that the choice of one particular sentence must entail a decision against all other possible sentences. However, such a decision does not mean that the others no longer exist, but rather that they are retained as alternatives [...]. The negatability of any (uttered) sentence is based on the existence of such alternative sentences. This negatability actually becomes manifest in the hearer's right to refute a sentence. (Hübler 1983: 12)

Only presuppositions are almost impossible to negate, everything else of the propositional content is negatable and must be ratified by the hearer. The difference between categorical (= unmodalized) assertions and assertions modalized by think, believe, etc. can then be defined in terms of the degree of anticipation on the part of the speaker regarding the negatability of his or her sentence: Modal assertory sentences, where the probability is explicit, anticipate negatability, whereas categorical assertions with implicit probability simply do not rule out the possibility of negation [...]. The latter are neutral and the former biased towards negation [...]. (Hübler 1983: 151)



Hübler (1983: 97) characterizes categorical assertions also as quasi-factive. Nevertheless, in spite of the speaker's expectation, a categorical assertion is potentially refutable or negatable, which is why Hübler says that its modality is implicit or unmarked. Modalized statements, in contrast, are characterized by the presence of an explicit modal expression. I wish to add to Hübler's reasoning that, once a sentence is negated by the facts and therefore refuted by the hearer, the question of the speaker's responsibility automatically arises. Hence, anticipating or expecting the refutation of one's own sentence is tantamount to anticipating the risk of being held responsible for what one is stating. Mitigators allow speakers to minimize this risk by making an utterance as acceptable as possible to the hearer without at the same time giving up their standpoint. They are fine-tuning devices that produce a compromise between what the speaker wants to say and what the hearer is willing to accept. Furthermore, by making possible a refutation and by tuning in to the standpoint of the interlocutor, the speaker assigns a more important role to the interlocutor and concedes space for interventions to him or her. The next step in Hübler's (1983: 98f., 141ff.) reasoning is the observation that the functions of statements and questions sometimes overlap (see also Bolinger 1957: 61; Lyons 1977: 800). E.g., in statements with the modal adverb possibly, the questive element predominates to such a degree that the speaker is able to retreat from his position by adding but I don't know: (9) (10)

Sue possibly lied, but I don't know. *Sue lied, but I don't know.

On the other hand, there are questions whose function consists in expressing an opinion, rather than in obtaining knowledge. E.g., in rhetorical questions the state of affairs formally questioned is actually taken for granted. In Hübler's eyes, statements combine assertive and questive elements. Depending on which elements predominate, the sentence has an assertory or a questory function. In Sue lied the speaker has an affirmative attitude with respect to the truth of the proposition, in Sue possibly lied a neutral attitude. Hübler's standpoint on the role of expressions of certainty (e.g., be certain, be sure, etc.) is highly innovative. Although an assertion with a modal expression of certainty is more emphatic than a categorical assertion, in his view there is no increase in probability above the degree present in categorical assertions, because what is probable can never be factive: Thus the emphasis turns out to be a meta-communicative device aimed at forestalling any possible argument about the validity of a propositional content. By using such adverbs and verbs of certainty, the speaker is calling upon the addressee not to question the propositional content but simply to accept it as being true. At the same time, the claims made by the modal degree of certainty in an assertion uttered by the speaker prevents the hearer from considering any of the alternative propositions which have been considered and rejected by the speaker. Preventing the hearer from negating the propos-


REDUCED PARENTHETICAL CLAUSES itional content is meta-communicative in that the primary communicative function of asserting something is accompanied by instructions as how to process the assertion. (Hübler 1983: 148)

By preventing the hearer from refuting the asserted proposition, modal expressions of certainty partly assume an imperative function. The question of the truth of an asserted proposition, i.e., the question of its matching a state of affairs, has to be kept distinct from the speaker's evaluation of the relationship between a proposition and a state of affairs and the guarantee he or she implicitly or explicitly offers for the existence of a correspondence between them. A proposition is either true or false, i.e., it matches the world or not, whereas the possibilities the speaker has for expressing evaluation and commitment are more nuanced. Regarding the truth-value, sentences with and those without an RPC like Engl. I suppose do not differ; the former are not more true or false than the latter nor vice versa. What changes is merely the evaluation of their truth by the speaker and the responsibility he or she is willing to accept (cf. Prince, Frader, & Bosk 1982: 89). Most philosophers (see Urmson 1952: 483; Searle 1969: 64ff.; Hare 1970: 22; Grice 1975, 1978) assume that it is a constitutive property of statements that the speaker claims to say the truth and that this is implied every time a statement is made, as long as there are no supplementary indications to the contrary. Hence, it comes as a necessary consequence that a categorical or unmodalized assertion, i.e., a statement with no comment whatsoever regarding its truth, possesses the maximum of commitment to the truth the speaker is able to offer. In fact, Lyons (1977: 809) states that […] there is no epistemically stronger statement than a categorical assertion […]

Lyons (1977: 594f.) proposes a reasoning about how this constitutive property together with Grice's (1975) cooperative principle might explain how someone who says I think that p communicates that he or she does not know that p: Taken together, the maxims of quantity and quality can be invoked […] to account for the fact that, if someone says I think it's raining or It may be raining, he can be held to have implied that he does not know for certain that it is raining […]. According to the maxim of quantity, we should be as informative as we need to be. The proposition "It may be raining" is less informative than the proposition "It is raining", since it is compatible both with "It is raining" and with "It is not raining". […] It follows that, by saying either I think it's raining or It may be raining, the speaker can normally be held to imply (i.e. to implicate) that he does not have the evidence that would enable him to make the more informative assertion It is raining.

Grice (1975: 49) sees the above-mentioned situation as a clash of maxims: The speaker is unable to fulfill the maxim of quantity (Be as informative as required) without violating the maxim of quality (Have adequate evidence for what you say). Such a clash typically gives rise to a conversational implicature, which can be of the type sketched by Lyons.



The initial assumption about the constitutive property of statements implies furthermore that speaker commitment can be reduced or maintained at the same level, but never increased by additional linguistic material. This, too, can be explained by a conversational implicature, according to Lyons (1977: 595): […] not only It may be raining and I think it's raining, but also It must be raining and I know it's raining, involve a weakening of the speaker's commitment to the truth of the proposition "It is raining" […]. […] if the speaker's evidence is unpeachable […], he will not feel obliged to make explicit the fact that it is so. By being more informative, in this respect, than he need be, he draws the addressee's attention to the possibility that the evidence for p is not as strong as it might be.

That is to say, neither an additional I think nor an additional certainly or surely raise the speaker's commitment above the degree of the corresponding categorical assertion. On the face of it this seems paradoxical, yet by interpreting sentences with certainly or surely as violations of the maxim of quantity, as outlined in Lyons (1977: 595), it seems quite reasonable. In fact, Lyons (1977: 809) repeats his standpoint and underlines that […] the very fact of introducing must, necessarily, certainly, etc., into the utterance has the effect of making our commitment to the factuality of the proposition explicitly dependent upon our perhaps limited knowledge.

Not everybody accepts this standpoint. Quirk, Greenbaum, Leech, & Svartvik (1985: 485) believe that certainly enhances the truth value or force of a sentence. In Palmer's (1986: 28f.) opinion […] certainly is often used to emphasize the commitment over and above what would be indicated by a simple declarative in English.

In his eyes the declarative is epistemically unmarked or neutral; it is the expression of a proposition with no direct indication of its epistemic status. It is clear that Quirk, Greenbaum, Leech, & Svartvik (1985) and Palmer (1986) are conditioned by the semantics of certainly, etc. while Lyons (1977) has pragmatic functions in mind. Hübler's (1983: 148) hypothesis about the role of expressions of certainty could elegantly resolve this question, which is raised in many contributions on epistemic modality and statements. He also assumes a pragmatic perspective and shows a possible solution by introducing the effect on the hearer as a third element besides truth value and speaker commitment. Following his reasoning, the speaker, by using an additional expression of certainty, does not raise his or her commitment above the level of the corresponding categorical assertion, but calls upon the hearer to accept the propositional content as being true, i.e., to accept the speaker's evaluation. Summing up, whereas an RPC like I think is an evaluation by the speaker, certainly and surely are instructions how to process the evaluation. It is difficult to prove Hübler's hypothesis. However, there is one phenomenon that can be interpreted as evidence against Quirk, Greenbaum, Leech, &




Svartvik (1985) and Palmer (1986). In spoken Spanish, seguramente, and also ciertamente, are frequently used in sentences with necessarily reduced speaker commitment: (11)

Y ... seguramente mañana se va a consumar […] (COREC.PENT001E) [And ... surely tomorrow it will be consumed]

In the utterance above, the speaker is evidently making a supposition about a future state of affairs; seguramente cannot change its presumptive or doxastic status. On the contrary, it seems that the utterance gains in 'authority' if the modal adverb is left out. The following Italian example even contains an explicit expression (penso) that qualifies the statement containing sicuramente as an assumption:53 (12)

B: […] a me me l'ha detto la signora penso che sicuramente lo saprà meglio di noi (CLIP.NB55) [It was the lady who told me. I think she surely knows it better than us]

5.2 Parentheticals versus performatives The presence of so many RPCs based on forms of decir and other utterance verbs in my list raises the question of whether and how they are to be distinguished from performative clauses. Before I address this issue, I have to make clear my understanding of three separate though interlinked notions: performative clause, performative utterance, and illocutionary clause. According to Searle, sentences are made up of a proposition and a series of illocutionary force indicating devices (IFIDs) that show [...] how the proposition is to be taken, or put it another way, what illocutionary force the utterance is to have; that is, what illocutionary act the speaker is performing in the utterance of the sentence. (Searle 1969: 30)

IFIDs include phenomena such as word order, intonation, verb mood, and sentence adverbs. Usually performative clauses are cited as the obvious device for indicating the illocutionary force (cf. Yule 1996: 49; Bublitz 2001: 130f.). 54 By beginning the sentence with, e.g., I state, I apologize, I warn or I promise, the speaker indicates unambiguously the kind of act he or she is performing. Needless to say, verbs like state, apologize, warn act and promise act as IFIDs only in the first person singular of the present indicative, i.e., only in that form are they performatives. It is therefore more correct to speak of the performative Lyons (1977: 808) discusses a similar English sentence: (i) He may certainly have forgotten. 54 Long before Austin (1976 [1962]), Spitzer (1922: 33), in referring to Italian dire as introductory formula, had already recognized that speaking is acting: "[...] der Mann aus dem Volke liebt es, sein Sagen zu betonen. [...] Das Sagen ist eine Tat, nicht nur im höfischen Kreise". 53



use of a verb.55 In everyday speech, utterances introduced by performative clauses are rare; the illocutionary force is clear from the context and it is unnecessary to use a clause to indicate it. Liedtke (1997: 208f.; see also Bublitz 2001: 130f.) shows that, in everyday speech, the use of a performative is often due to special circumstances. Performative clauses may even be misleading, as Bublitz (2001: 134) proves with an example from the British National Corpus: (13)

If you don't put that knife away, I promise you shall die a criminal's death under the law.

Notwithstanding the presence of I promise, the context suggests that the utterance above is not a promise but a warning. It is thus important to note that performative clauses like those cited above are not necessary elements of the speech act (see Urmson 1977), that is, they are in my terminology saying-anddoing clauses but not saying-is-doing clauses. Certain conventional actions and ceremonies are carried out by pronouncing a formula which in most cases has a rather fixed wording. It has the grammatical form of a statement but it is neither true nor false. It usually contains a performative clause,56 which in this case does not merely act as a facultative IFID, but as an obligatory constituent of the speech act, that is, as a saying-is-doing clause. The action does not take place without the pronouncement of the formula and the performative clause. Austin calls these formulas performative utterances and characterizes them as follows: A. they do not 'describe' or 'report' or constate anything at all, are not 'true or false'; and B. the uttering of the sentence is, or is a part of, the doing of an action, which again would not normally be described as, or as 'just', saying something. (Austin 1976 [1962]: 5)

One of Austin's examples of performative utterances is: (14)

I name this ship the Queen Elizabeth.

An Italian example for a performative utterance would be the following, pronounced at university by the chairman of an examining commission (see Fava 2001: 27): (15)

In nome della legge, per l'autorità conferitami dal Magnifico Rettore, La proclamo Dottore in Lettere. [In the name of the law, by the authority conferred on me by the rector, I declare you Master of Arts]

The notion of performative verb is basically as mistaken as the notion of parenthetical verb. That is the reason why I prefer the terms performative clause and illocutionary clause. 56 A performative utterance without a performative clause is French Rien ne va plus!, by which in French casinos the croupier closes the betting (see Urmson 1977: 266). 55



Needless to say, these sentences achieve their goals because there are specific conventions that link the words to institutional procedures; they are therefore essential ingredients of conventional actions and can only be assessed as felicitous or infelicitous. Moreover, the action or ceremony can only take place and have an effect in the appropriate circumstances, i.e., if certain felicity conditions are met. E.g., in the last sentence, the person who confers the title must have the authority to do so. Illocutionary clauses or illocutionary verbs (see Searle 1969), also called speech activity verbs (see Ballmer & Brennenstuhl 1981) or speech act verbs (see Wierzbicka 1987), describe linguistic activities, that is, they are terms for different types of speech acts. They do not coincide with the class of performative clauses, although many of them may be used as performatives (see Searle 1976: 6). For instance, whereas state, apologize and warn can be used performatively, other illocutionary verbs, e.g., boast, insult and threaten, cannot: (16) (17) (18) (19) (20) (21)

I hereby state that the airplane has arrived on time. I hereby apologize for what I have done. I hereby warn you of possible failure. *I hereby boast that I am the best. *I hereby insult you for what you have done.57 *I hereby threaten you that I will beat you up.

In addition to boast, insult and threaten, a few other verbs referring to linguistic activities, e.g., to insincere speech activities (lie, show off), complex speech activities (deliberate, chatter, joke), special aspects of speech activities (whisper, shout), perlocutionary speech activities (upset, tease, alarm, seduce), interactive speech activities (quarrel, discuss), do not allow a performative use (Ballmer & Brennenstuhl 1981: 16; see also Bublitz 2001: 71, 103f.). A parallelism between parenthetical and performative clauses was already noted by Urmson, who stated that [...] parenthetical and performatory verbs have much in common against ordinary descriptive verbs. (Urmson 1952: 490)

As I have said before in the short research overview (see section 3.2), Benveniste (1966a [1958]), independently both of Urmson and Austin, pointed to the same parallelism and stressed the non-descriptive role as indicateurs de subjectivité of both classes of clauses, i.e., as devices whereby the speaker makes an utterance and at the same time expresses his or her attitude to what he or she is saying. The common property that struck Urmson and Benveniste was that both classes of verbs are used primarily in the first person singular of the present indicative form and do not describe psychological states or actions. Yet BenvenFor similar examples in Spanish and French, see Fuentes Rodríguez (1987: 155) and Kerbrat-Orecchioni (2001: 25). 57



iste (1966a [1958]: 265) sees a fundamental difference: It is only with performative clauses (verbes de parole in Benveniste's terms) that saying is doing.58 The similarities between the two groups led to some confusion (e.g., Lakoff 1969a; Palmer 1986: 167ff.), but also stimulated a debate on the distinction between the two classes of clauses and their main functions, mitigation of the neustic on the one hand, indication of the illocutionary force on the other (see Venier 1991: 137ff.; Ifantidou 2001: 120ff.). Vendler (1970: 82; see also Venier 1991: 118f., 140ff.) also notes the similarity between the two verb classes in the first person singular of the present indicative, but observes that the temporal aspect provides a distinctive criterion: It is quite clear that whereas the simple present tense, in the case of a performative, singles out the moment at which the illocutionary act occurs, in the case of a propositional attitude verb, the same tense does not indicate a unique moment, but an indefinite time span which includes the moment of the utterance. […] performatives are achievement verbs, but propositional verbs are state verbs […].

The difference between the achievement character of performative verbs (saying verbs) and the state character of parenthetical verbs (thinking verbs) is evident when the verbs occur together with a time adverb like still: (22) (23)

I still believe. I still promise.

According to Vendler (1970: 83), the first sentence needs no explanation, the second does because it might mean that the speaker has not withdrawn his or her promise or that he or she is still willing to promise, but certainly not that his or her act of promising has not ended yet. Cattel (1973: 620ff.; see also Hooper 1975: 102ff.) points to other obvious differences between Urmson's parenthetical clauses and performative clauses. Firstly, think and suppose do not behave like common performatives insofar as they may not be preceded by hereby: (24) (25)

*I hereby think the Yankees will win. *I hereby suppose John will come back soon.

Secondly, think and suppose permit that a tag question be formed for their complements: (26) (27) (28)


I think this car needs a tune-up, doesn't it? I suppose the Yankees will lose again this year, won't they? I propose that we build a new clubhouse, *don't I? / *do I? / *don't we? / * do we?

See Wittgenstein (1958: 144): "Saying that I believe is not believing".



Performatives like propose, demand, advocate, suggest, etc. do not allow tag questions to be formed from their complement sentences. Most, if not all, researchers assume that performative clauses may occur in parenthetical position (e.g., Urmson 1952; Fraser 1975: 188; Holdcroft 1978: 64; Récanati 1984: 347). Lyons (1977: 738) points to the obvious similarity between parentheticals and performatives and proposes to modify Urmson's definition of parenthetical verbs in order to encompass parenthetically used performatives. In his opinion, the semantic and grammatical relations between the following pairs of sentences are exactly the same: (29) (30) (31) (32)

She's in the dining-room, I think. I think (that) she's in the dining-room. I'll be there at two o'clock, I promise you. I promise (that) I'll be there at two o'clock.

According to Fava (2001: 45ff.), four of Austin's five classes of performative clauses may be used as final parentheticals (exercitives, commissives, behabitives, and expositives; see Austin 1976 [1962]: 148-164). Fava cites the following Italian examples: (33) (34) (35) (36)

Adesso dimmelo, te lo ordino. [Now tell me, I order it] Vengo senz'altro, te lo prometto. [I come certainly, I promise you] Non posso farlo, mi dispiace. [I can't do it, I regret] Non sono stato a Roma, lo nego. [I haven't been in Rome, I deny it]

Kerbrat-Orecchioni (2001: 37) does not mention RPCs explicitly, but includes postposed clauses like the following in the class of French IFIDs: (37)

Aidez-moi je vous en supplie. [Help me, I implore you]

As we know, the range of contemporary conceptions of what is a parenthetical position and what are RPCs is very broad. To start with, according to the selection criteria illustrated in section 4.3, only one of the performative clauses above (I promise you) would count as a reduced clause, since none of the arguments of the other verbs is missing, i.e., they are complete clauses. Moreover, in their present form, only the utterances with Engl. I promise you, and It. te lo prometto and mi dispiace could be transformed into a structure with a subordinate noun clause. I also have doubts regarding the positional variabilities of some of



them, e.g., I am unsure if, in actual speech, It. te lo ordino, mi dispiace, and lo nego and Fr. je vous en supplie are acceptable in medial position, let alone in a position interrupting the clause nucleus. Furthermore, from a prosodic point of view, the performatives cited above cannot be equated with parentheticals like Engl. I think, It. credo or Fr. je crois. The performatives constitute intonation units of their own, separated from the host clause by a pause and with a separate tonic syllable.59 One question which has been debated repeatedly is the function of performatives when they occur in parenthetical position. Is the speaker doing by saying with them? Urmson (1952: 494) does not think so. In his view, performative clauses with verbs such as guarantee, bet, and warrant express speaker commitment rather than illocution when in parenthetical position. We cannot treat the following sentences as a guarantee or a bet, respectively: (38) (39)

He'll come to a bad end, I guarantee. He'll forget to come, I bet.

Urmson adds that […] we have here another case of the borrowing of another sort of verb for parenthetical uses. (Urmson 1952: 494)

That is, a verb which occurs in parenthetical position acts as mitigator, in other words, a parenthetical does not express an illocution. Blakemore (1991: 202ff.) discusses examples with parenthetical I admit, I bet, and I promise and arrives at the same conclusion. Venier (1991: 139ff.) also shares Urmson's standpoint and states that in comparable Italian utterances, sentence-final and sentence-initial performatives are used as neustic indicators rather than to perform a speech act: (40) (41)

Scommetto che arriverà di nuovo in ritardo. [I bet that he'll be late again] Farà una brutta fine, te lo garantisco. [He'll come to a bad end, I guarantee]

The proof lies, according to Venier, in the variability of the word order and in the temporal aspect (see Vendler 1970: 82f.) of the verbs scommetto and te lo garantisco. The state character of both may be shown by adding a durative adverb or verb: (42)

Scommetto tuttora che arriverà di nuovo in ritardo. [I still bet that he'll be late again]

Andersen (1996: 309) confirms this view regarding the intonation of sentences with parenthetical Fr. je regrette. 59

106 (43) (44) (45)


Continuo a scommettere che arriverà di nuovo in ritardo. [I continue to bet that he'll be late again] Farà una brutta fine, te lo garantisco tuttora. [He'll come to a bad end, I still guarantee] Farà una brutta fine, continuo a garantirtelo. [He'll come to a bad end, I continue to guarantee]

In her opinion, the acceptability of the sentences above shows that performative verbs may be exploited as signs of subscription, i.e., may act on the neustic: Si è dunque qui di fronte allo sfruttamento di verbi di dire come verbi di pensiero, direi ad una scommessa e ad una garanzia metaforiche. Dal momento che si scommette o si fa da garanti su quanto si sa essere vero, dire che si scommette o si garantisce implica che l'oggetto di questa scommessa o questa garanzia siano da intendersi come veri. Dire che si scommette o si garantisce diventa in questo modo indicare il proprio grado di certezza in quanto si afferma. (Venier 1991: 143)

In other words, in these utterances the function of scommetto and te lo garantisco is equivalent to that of RPCs with doxastic verbs like Italian credo or penso. In my opinion, Urmson's and Venier's arguments are convincing, at least regarding saying-is-doing clauses. In none of the examples above are the boldfaced expressions performatives. In fact, they can easily be substituted for an alternative device reducing responsibility to an equivalent degree (e.g., It. sicuramente), without substantially altering the meaning of the utterance. What about saying-and-doing clauses? Garrido Medina (1999: 3894) cites two Spanish examples with performatives in parenthetical position: (46) (47)

Ven aquí, te lo ruego. [Come here, I ask you] Juan está en casa, digo yo. [Juan is at home, I say]

and remarks that the corresponding speech acts are not performed but rather described or interpreted by the parentheticals. The speech acts could also be performed without te lo ruego and digo yo. To this remark I would like to add that a saying-and-doing clause can always be omitted, without compromising the execution of the speech act. Its parenthetical position does not change this characteristic. By describing their respective speech acts, the two clauses also indicate these acts, in other words, they maintain their IFID function. I return to the question in subsection 6.3.1. 5.3 Summary This chapter outlines the framework and the basic scheme according to which the RPCs selected from the corpora will be analyzed in the next chapter. Two heuristic distinctions are at its center: Hare's (1970) phrastic, tropic, and



neustic, as the fundamental components contributing to the meaning of an utterance, and Caffi's (1999, 2001) classification of mitigators in bushes, hedges, and shields. Caffi's classes are based on Hare's distinction and cannot be fully appreciated without it. The distinction between phrastic, tropic, and neustic is also necessary to understand Hübler's (1983) analysis and re-definition of the neustic. He defines its role within statements as the degree of anticipation on part of the speaker regarding the negatability of his or her sentence. I add to this definition that anticipating the refutation of one's own sentence is tantamount to anticipating the risk of being held responsible for what one is stating. Furthermore, by making possible a refutation and by tuning in to the standpoint of the interlocutor, the speaker assigns a more important role to the interlocutor and concedes space for interventions to him or her. The second section focuses on a possible distinction between performative clauses and parenthetical clauses. Initially, I state that there are two types of performatives: those that are not necessary elements of the speech act performed (I call them saying-and-doing clauses) and those that are obligatory constituents of the speech act (I call them saying-is-doing clauses). Then I summarize the most important differences between performatives and parentheticals: 1. The first contain achievement verbs, while the second contain state verbs. 2. Parenthetical clauses cannot be introduced by hereby. 3. Only parentheticals permit the formation of a tag question from their complements. These properties were identified by Vendler (1970) and Cattel (1973) regarding English but may be also found in Romance. Finally, I pose the question if there are performative parentheticals and discuss Urmson's (1952) and Venier's (1991) standpoint, who think that, when saying-is-doing performatives are in parenthetical position, they lose their performative function. I find their arguments convincing. Regarding saying-and-doing clauses, I believe that in parenthetical position they maintain their optional IFID function.

PRAGMATIC ANALYSIS: FUNCTIONS 6.1 A first basic subdivision If we look at the RPCs in section 4.4 and take into account their deictic orientation, the mood of the verb and the sentence type (declarative, imperative, interrogative), we are able to make a first rough subdivision (cf. also Brown & Levinson 1978, 1987; Holmes 1984; Koch & Österreicher 1990: 51-72; Andersen 1996: 314, 1997: 147), which is summarized in the following table: Person 1st Sg 1st Sg 1st Sg 1st Sg 3rd Sg 3rd Sg 3rd Sg 3rd Pl 1st Pl 1st Pl 1st Pl 2nd Sg 2nd Polite 2nd Sg 2nd Sg 2nd Sg 2nd Polite 3rd Sg 3rd Sg 3rd Pl 1st Sg

Tense Mood Mod. verb? Sentence Type of downgrading Example Pres Ind No Decl Alleviate responsibility Fr. je crois Pres Ind Yes Decl Remove responsibility Fr. je veux dire Pres Cond No Decl Remove responsibility Fr. je dirais Past Ind Yes Decl Remove responsibility It. volevo dire Pres Ind No Decl Remove responsibility Fr. paraît-il Pres Ind Yes Decl Remove responsibility Fr. il faut dire Pres Cond No Decl Remove responsibility Fr. on dirait Pres Ind No Decl Remove responsibility/Report Sp. dicen Pres Imp No Imp Remove+Share responsibility Fr. disons Pres Cond No Decl Remove+Share responsibility Sp. diríamos Pres Cond Yes Decl Remove+Share responsibility Sp. podríamos decir Pres Ind No Decl Share responsibility Fr. tu vois Pres Ind No Decl Share responsibility Fr. vous savez Pres Imp No Imp Share responsibility It. poni Pres Ind No Interr Phatic (No downgrading) Fr. tu vois? Pres Imp No Imp Phatic (No downgrading) It. figurati! Pres Ind No Interr Phatic (No downgrading) Fr. voyez­vous? Pres Ind No Decl Report (No downgrading) It. dice NP Past Ind No Decl Report (No downgrading) Sp. ha dicho NP Past Ind No Decl Report (No downgrading) Sp. me han dicho Past Ind No Decl Report (No downgrading) It. ho detto Table 6.1: Basic subdivision of reduced parenthetical clauses

As can be seen from the table, some RPCs downgrade speaker commitment, i.e., they are mitigating RPCs, while others function as phatic or as reporting devices. There are three ways in which the downgrading of speaker commitment can be accomplished: by alleviating responsibility, by removing responsibility, and by sharing responsibility. The RPCs with verbs in the first person singular of the present indicative focus on the speaker. With respect to sentence types, they are all declaratives. With them the speaker directly reduces his or her burden of responsibility, that is, they are responsibility alleviating devices. With the exception of the forms in the second person, most of the RPCs whose deictic centers lie outside the unmarked ego, hic, nunc center have the purpose of shifting the burden of responsibility away from the speaker, that is, they are responsibility removing devices. This is also true of the verbs in the first person singular of the conditional form (e.g., Fr. je dirais) because they



make the truth of the utterance (and the responsibility of the speaker) dependent on implicit external circumstances (If q, I'd say p). Some clauses are based on verbs that are in the first person plural of the imperative (e.g., Fr. disons). They remove responsibility, since the statement is presented as if it had been requested or called for, and at the same time they explicitly include the addressee and thus divide the responsibility between speaker and addressee, i.e., they are also responsibility dividing devices. The deictic centers of reporting RPCs also lie outside the unmarked ego, hic, nunc center (e.g., It. ho detto). As I will point out in section 6.6, reported statements do not actually reflect the judgment of the speaker and hence are outside the realm of speaker commitment. That is why I prefer to treat reporting parentheticals as a separate group not connected to mitigation. The RPCs with verbs in the second person are directed towards the addressee. They may be either declaratives (e.g., Fr. dites-vous, vous savez), imperatives (e.g., It. figurati60, Sp. fíjate) or interrogatives (e.g., Fr. voyez-vous, Fr. vous savez61). The declaratives distribute responsibility to both speaker and addressee. The imperatives and interrogatives mostly have a phatic function, i.e., they are signals by which the speaker invites the addressee to cooperate and controls and maintains the effectiveness of communication (see Koch & Österreicher 1990: 57f.). They do not modify the speaker's responsibility. The subdivision is, admittedly, approximative; however, the differentiation between these three ways of reducing speaker responsibility is useful and will represent an important reference point for the discussion of the pragmatic functions of RPCs in this chapter. The criteria in table 6.1 also permit me to isolate the RPCs with purely phatic function from those operating on the responsibility of the speaker, which constitute the main subject of this book. I wish to draw attention to the fact that in my understanding, the phatic function is limited to invitations to cooperate and to the maintenance of the effectiveness of communication. However, there are researchers who prefer a broader definition of this function. E.g., Koch & Österreicher (1990: 57f.) and Bazzanella (2001: 236ff., 253ff.) do not distinguish between declarative and interrogative parenthetical Fr. tu sais, It. sai, and Sp. sabes and classify both as phatic signals (German Kontaktsignale or It. Fatismi) that establish and maintain the communicative contact. Bazzanella (2001: 236) defines the phatic function as follows: In the case of It. figurati, the invitation to imagine a state of affairs may imply the conclusion that this state of affairs cannot possibly exist (see Devoto & Oli 2000: 814; Zingarelli 2000: 703). 61 In those corpora with punctuation marks, imperative RPCs are transcribed with or without exclamation marks, depending on the type of the host sentence: (i) Pareces humorista ¡qué bien lo haces esto, oye! (COREC.CLUD025B) (ii) Se lo digo a Paula, oye. (COREC.ACON006C) Interrogative RPCs are transcribed mostly with question marks. In tables 4.2-4.4, I did not punctuate the verbs. This means, of course, that some of them, e.g., Fr. vous savez, actually consist of two forms, a declarative and an interrogative. 60



Alcuni segnali discorsivi sottolineano l'aspetto fàtico, cioè di "coesione sociale", intesa come strumento per creare, consolidare o evidenziare l'appartenenza di un individuo ad un gruppo.

And regarding the function of It. capisci, sai, come sai, lo sapete, etc. she writes (2001: 237): Fanno parte di questo gruppo anche i segnali discorsivi che sottolineano la "conoscenza condivisa", cioè l'insieme di conoscenze comuni al parlante in corso e agli interlocutori […].

Aijmer (2002: 48) has an even broader understanding of the phatic function and defines Engl. I think as phatic connective with an evidential meaning. I prefer to treat declarative sai, and Fr. vous savez, etc., which underline shared knowledge, not as phatic but as responsibility dividing device, although I am aware of the fact that one side-effect of the use of declarative sai is to draw the addressee's attention to the utterance that follows (see section 6.5). RPCs whose primary and exclusive function is phatic are not at the center of this book. I will discuss, however, several RPCs that occasionally, in addition or alternatively to reducing speaker responsibility, may serve as phatic devices (e.g., expressing hesitation, attracting attention). In the following sections, I will define the pragmatic function of each class of RPCs and also the component of the utterance to which it refers. It is obvious that in the case of the phatic parentheticals the distinction between phrastic, tropic, and neustic is not relevant. 6.2 Mitigating the phrastic The main RPC operating on the propositional content by affecting the appropriateness or precision of the referring term or the predicate in its scope is the imperative Fr. disons (cf. Hölker 1988: 120f.; Authier-Revuz 1995: 183ff., 650ff.; Morel & Danon-Boileau 1998): (1)

L1 [...] ça faisait partie de ces - disons données culturelles [...] (CHOIX.19.111) [That was part of these, let's say, cultural data]

It is neither an epistemic verb nor is it in the first or third person singular of the present indicative. For this reason, traditional works on RPCs did not take into account Engl. let's say or comparable Romance expressions. The function of Fr. disons corresponds closely to those Bazzanella (1994: 162, 2001: 250f.) and Caffi (2001: 280, 293, 449) found for the Italian equivalent diciamo. Hölker's (2003) detailed analysis dedicated specifically to this verb form in the Italian CLIP corpus confirms their findings. In addition to the uses I mention here, he cites examples where diciamo introduces paraphrases or additional information about a referent. Its Spanish counterpart digamos has a similar function (see Fuentes Rodríguez 1990a; Guillén Sutil 2001: 110; Schneider



2004: 41f.).62 Caffi (2001: 448f.) classifies It. diciamo as a bush operating on predicates, seemingly excluding the possibility that it also operates on referring terms. However, according to my findings, It. diciamo and its French and Spanish equivalents operate very often on referring terms. With respect to form, Fr. disons, It. diciamo, and Sp. digamos are imperatives in the first person plural. That is, they should have the illocutionary force of proposals, which could be accepted or rejected by the interlocutor. However, their form does not exactly correspond to their function, because in reality the speaker does not propose to say something together with the interlocutor, and the interlocutor cannot accept or reject the speaker's proposal. Yet, by presenting an expression as a proposal, the speaker is able to distance him- or herself from it and to reduce the communicative responsibility that arises from his or her utterance. With Fr. disons and its Italian and Spanish equivalents, the speaker presents a predicate or a referring term as if it were a proposal that is not wholly satisfying neither to him- or herself nor to the interlocutor and that could therefore be improved. They are primarily bushes, but indirectly also hedges (see Hölker 2003: 150), because the speaker, by using a referring expression that is explicitly inappropriate, reduces his or her commitment. They remove responsibility, since the statement is presented as if it were a proposal. Furthermore, they include the addressee and expand the deictic center of the utterance (actantial shield), i.e., they share responsibility with the addressee.63 Morel & DanonBoileau (1998) observe that Fr. disons, notwithstanding its first person plural form, underlines that it is the speaker who proposes an imperfect formulation. The two authors add: On peut gloser ainsi sa valeur 'je propose une formulation de compromis pour faire simple'. Le renoncement à une formulation plus appropriée est lié, nous semble-t-il, à l'anticipation d'une incompréhension: 'je pourrais mieux dire, mais en disant mieux je crains de vous perdre, donc je dis moins bien de façon à ce que nous continuions à garder une consensualité'. (Morel & Danon-Boileau 1998: 104)

Occasionally, the referent itself, rather than the referring term in the scope of Fr. disons, It. diciamo, and Sp. digamos, may be deemed vague: (2)

Inf. - […] eso ha sido durante muchos años … V … el … el modo que el … se ha planteado el matrimonio en España, en Italia, en Francia, en todos los países … desde … hace unos treinta años, digamos (HCM.4.69) [This has been during many years the way how people conceived of marriage in Spain, in Italy, in France, in all countries since about thirty years ago, let's say]

See Pusch (2003: 254f.) on the use of the Occitan equivalent disem. See Bazzanella (2002: 244-249) for an analysis of the different meanings of 'we' in relation to its context. 62 63



In the utterance above (see also Hölker 2003: 138f. for Italian examples), not the relation between the referent and its referring term is deemed inappropriate but the referent itself is vague; this is marked by unos. The referring expression hence corresponds to 'some thirty years'. The vagueness of the referent is also marked by digamos. It might also happen, although rarely, that Fr. disons, It. diciamo and Sp. digamos by themselves act as markers of vagueness. In example (49), in the preceding chapter, Fr. disons has nearly the same meaning as Fr. environ oder Engl. about. In fact, Fuentes Rodríguez (1990a: 105) clearly states that Sp. digamos may mean 'approximately': Su posición puede variar. Intercalado, sirve para precisar un elemento, o una idea. A veces equivale a 'aproximadamente' […] A veces puede ir al final, como un apoyo también, con valor de 'aproximadamente esto es lo que quiero decir, pero no es lo justo'.

Fr. disons may also serve to indicate that the expression in its scope is intended as a hypothesis or an example. The same is true for Sp. digamos (see Hölker 2003: 138 for an Italian example of this use): (3)

Inf. - […] una persona, digamos más, más o menos culta ¿eh?, entonces … V … que le guste el arte de la lectura, entonces va a, por ejemplo, una traducción de "Archipiélago Gulag" y ve que es muy mala ¿no? […] (HCM.17.296) [A, let's say, more or less educated person, well, who likes reading, well, he sees, for example, a translation of the "Gulag Archipelago" and sees that it is very bad]

So far we have seen Fr. disons and its Italian and Spanish equivalents as mere signals of the inappropriateness of an expression. But by saying that an expression is imperfect, the speaker opens up the possibility of re-formulating, expanding or correcting it. In the following Spanish utterance, the referring adjective in the scope of digamos is re-formulated immediately afterwards: (4)

Inf. - […] cuando se trata de obras clásicas, pero adaptadas digamos, remozadas. […] (HCM.6.102) [When we have to do with classic but adapted, let's say, refreshed pieces]

In the example above, the scope of Sp. digamos includes the inappropriate expression, not the reformulation. However, besides being markers of imprecise or not completely appropriate referring expressions, Fr. disons and its Italian and Spanish equivalents may be markers of self-corrections. In the following French example, the reformulation is in the scope of disons:

114 (5)


B: […] on essaie d'élaborer ou ... disons d'adapter les enseignements de la Bible à leur niveau […] (BEECH.15) [We try to develop or, let's say, to adapt the lesson of the Bible to their level]

The utterance shows that disons, besides proposing imperfect terms, may also propose reformulations or self-corrections. It. diciamo (see Koch & Österreicher 1990: 62; Bazzanella 1994: 162; Hölker 2003: 141ff.) and Sp. digamos may be employed similarly. It is important to note, though, that the reformulation remains a proposal and involves a reduced commitment by the speaker. Fr. disons and its Italian and Spanish equivalents are the phrastic parentheticals that are most frequent and that cover the broadest spectrum of functions. Other parentheticals operating on the phrastic are either less frequent or more specialized. E.g., Fr. on va dire is limited to the marking of terms not completely appropriate: (6)

L1 […] une société c'est on va dire sur le plan sociologique c'est une organisation […] (CRFP.PCR-PRO001) [A society, that is, we'll say, on the sociological level, that is an organization]

As Fr. disons, the RPC on va dire removes responsibility in two ways: The statement is presented as if not (only) attributable to the speaker64 and as if only to be said in the near future. Based on Caffi (1999: 869, 2001: 315), we could say that it is a double deictic shield, namely an actantial shield involving the ego and an anticipation shield involving the nunc. Fr. on dirait, as an RPC operating on the phrastic or a reporting expression, also removes responsibility twice; by acting on the ego and presenting the statement as if it were dependent on implicit external circumstances. Two Spanish parentheticals, diríamos and podríamos decir, may also mark terms not completely appropriate. They, too, are responsibility removing devices that make the truth of the utterance dependent on implicit external circumstances. Furthermore, they include the addressee and thus divide responsibility, i.e., they are actantial shields. As I have said above, Fr. disons may also indicate that the expression in its scope is intended as an example or a hypothesis. In Italian, to mark an expression as an example, one may also use the second person singular imperative of porre: (7)


B: cioè c'è poni lo scaffale di inglesi o scaffale di francesi lo scaffale di italiani_ letterature vocabolari_ così (CLIP.MB4) [That is, there's, suppose, the book-shelf for English or the book-shelf for French, the book-shelf for Italian, literature, dictionaries, so]

In spoken French, on has often the same meaning as nous and hence includes the speaker.



Poni is exceptional insofar as it is a second person imperative that is not employed as a phatic device. In French, examples or hypothetical terms may be introduced by parenthetical mettons (see Dostie 2004: 160f., 178):65 (8)

L1 […] dans un bateau qui faisait peut-être euh quinze seize mètres mettons dix-sept mètres madame de long […] (CHOIX.32.166) [In a boat that was maybe fifteen sixteen, suppose, seventeen meters, madam, long]

In French conditional clauses introduced by si, one may find an additional mettons underlining the hypothetical character of the clause. In that case, since mettons marks a whole proposition as a possible hypothesis, it becomes a device directly operating on speaker commitment, i.e., on the neustic. 6.3 Indicating the tropic and mitigating the phrastic or neustic 6.3.1 Plain performatives In section 5.2, I expressed doubts regarding the parenthetical status of the performatives in (31) - (37). In the corpora I examined, there are, however, some performatives occurring in parenthetical position. All of them meet the conditions outlined in section 4.3. Moreover, according to tests with native speakers,  their intonation is like that of other RPCs. Hence, at least from a formal point  of view, they are RPCs. Sp. insisto and repito appear in only a few utterances. The first one indicates that something is being very firmly said a second time, the second, simply that an utterance is being repeated: (9) (10)

[…] va a continuar, insisto, en los próximos minutos […] (COREC.ADEB003A) [It will continue, I insist, in the next minutes] […] sobre eso, repito, en unos pocos minutos vamos a tener ... pues esta fotografía […] (COREC.APUB007B) [About this, I repeat, in a few minutes, we'll have this photo]

Contrary to the clauses cited by Urmson (1952), Engl. I guarantee and I bet, and Venier (1991), It. scommetto and te lo garantisco, the parentheticals in the passages above are saying-and-doing performatives, that is, they are IFIDs indicating, in combination with other linguistic means, the speech act that is being performed. Note, however, that statements as well as other speech acts can be performed firmly and can thus be marked by Sp. insisto, like, e.g., the questions in the following passage:

Besides diciamo, Bazzanella (1994: 163) cites the Italian "indicatori di esemplificazione" facciamo, mettiamo and prendiamo. 65

116 (11)


[...] pero insisto, ¿cómo serían? ¿tasadas en el tiempo? […] (COREC.AJUR017A) [But, I insist, how are they going to be? Taxed according to time?]

Sp. insisto and probably also repito are IFIDs that are not specific to a macrotype of speech act (statement, question, directive); rather, they mark subtypes. As I wrote in section 5.2, according to Urmson (1952) and Venier (1991), performatives in governing position are IFIDs; they lose this function and become mitigators in parenthetical position. Does the parenthetical position automatically turn Sp. insisto and repito into pure neustic mitigators, as hypothesized by Urmson (1952) and Venier (1991) in the case of the English and Italian performatives above? This is highly improbable. In (11), insisto cannot be substituted for a modal sentence adverb that reduces responsibility like, e.g., Sp. seguramente. In (9) and (10), such a substitution, though grammatically possible, would substantially alter the meanings of the utterances. An important meaning component, i.e., either firmness and repetition of the utterance or focus on a part of the utterance, would be lost. I thus have to conclude that in parenthetical position insisto and repito maintain their IFID function. The most frequent performative RPCs are It. dico and Sp. digo, digo PRO, ya digo and ya te digo. Although they are performatives, they express very little or nothing about the type of speech act. Even more than Sp. insisto and repito, they are focusing devices directing the addressee's attention to a part of the utterance. The type of speech act is made clear by other IFIDs. In fact, in a short remark, Fuentes Rodríguez confirms that Sp. digo in parenthetical position may express 'lo que está sujeto a mi punto de vista' (1990a: 113). Are parenthetical It. dico and Sp. digo, digo PRO, ya digo and ya te digo responsibility reducing devices? Based on Urmson's (1952: 494) reasoning, Venier (1991: 66f) writes that in the following Italian (invented) sentence dico expresses speaker commitment not illocutionary force: (12)

È proprio così, dico. [It's just like this, I say]

In fact, this possibility cannot be totally excluded (cf. also Spitzer 1922: 35f.). In the following utterance, dico could be substituted for a modal sentence adverb or an RPC like credo: (13)


A: [...] perché a questo punto bisognerebbe riuscire a partire secondo me_ dico o comunque [...] (CLIP.FB5) [Because, at this point, we should manage to leave, in my opinion, I say, or anyway] A questo punto bisognerebbe riuscire a partire secondo me credo. [At this point, we should manage to leave, in my opinion, I think]



Likewise, pointing to the use of parenthetical digo PRO (mostly final digo yo) in spoken Spanish, Fuentes Rodríguez (1990a: 112, cf. also Fuentes 1998: 154; Garrido Medina 1999: 3894) says: Aquí digo se usa generalmente como apéndice al final de un enunciado, en la fórmula digo yo que pretende atenuar lo anteriormente dicho y reducirlo al ámbito de lo opinable por el hablante. Equivale a 'eso es lo que pienso', 'en mi opinión'.

In fact, in the Spanish corpora I examined, there are such utterances with parenthetical digo PRO: (15)

[…] cirujano de ovejas o cirujano de ... de mujeres. Porque si ha matao a una mujer será cirujano de mujeres, digo yo. (COREC.PENT007D) [Surgeon for sheep or surgeon for women. Because if he has killed a woman he must be a surgeon for women, I say]

In the example, digo yo may be substituted for a responsibility reducing RPC (e.g., creo, pienso) without substantially changing the utterance's meaning. The other parentheticals (It. dico and Sp. digo, ya digo and ya te digo), however, are mostly used as focusing or highlighting expressions acting on the phrastic. For example, Sp. digo represents a typical device for highlighting selfcorrections (cf. Martín Zorraquino & Portolés Lázaro 1999: 4128): (16)

Inf. B. - […] Como Marga, que se ha metido en Literatura italiana, bueno, en Filología italiana, digo (HCM.20.392) [Like Marga, who began Italian literature, well, Italian philology, I say]

With Sp. digo the speaker may also signal an increase in the precision of a term and thus focus on an aspect of it. Nilsson-Ehle (1947: 20) mentions this use also in relation to It. dico.66 Occasionally, Sp. digo and ya digo do not indicate a correction or an expansion but the mere repetition of a part of the proposition: (17)

Me va a costar, dame, ya te digo, dame unos días para ... para llegar a alguna conclusión. (COREC.BENT026F) [It will cost me, give me, I say, give me some days in order in order to arrive to a conclusion]

It. dico and Sp. digo, ya digo and ya te digo may even be used to mark a term An Italian example with the repetition and explanation of a preceding term may even be found in Dante, La Divina Commedia, Inferno, IV, 66: (i) Non lasciavam l'andar perch'ei dicessi, ma passavam la selva tuttavia, la selva, dico, di spiriti spessi. 66



that is neither a correction nor an explanation or repetition of a preceding term, but which occurs for the first time within the proposition. In the examples above, the speaker employs the RPC as a focusing device that directs the addressee's attention to a part of the utterance. As we have seen in section 6.2, Fr. disons and its Italian and Spanish equivalents partly have the same purpose, yet only It. dico and Sp. digo, ya digo and ya te digo may occasionally function as a pure focusing expression, i.e., without reformulation, self-correction, or explanation. In that case, it is questionable whether It. dico and Sp. digo, ya digo and ya te digo are still responsibility reducing expressions. Focusing parts of an utterance is not necessarily related to speaker commitment, that is, it does not alter the speaker's burden of responsibility. My doubt is confirmed by the fact that these parentheticals also appear in sentences other than statements. In (17), Sp. ya te digo occurs in a directive and marks a part of or the whole utterance. As shown by the following Italian example, parenthetical dico is possible also in a question: (18)

B: c'arrivano tanti come lei perché non ci deve arriva' anche lei dico così (CLIP.FB16) [Many like her make it. Why shouldn't she make it, I say]

As a sentence-initial performative, It. dico never introduces questions, in other words, it is not an IFID for questions. The examples suggest that It. dico and Sp. digo, ya digo and ya te digo, when in parenthetical position, tend to lose their illocutionary force indicating potential completely. Occasionally, these forms, besides being focusing devices, are clearly hesitation forms by which the speaker holds the floor, i.e., keeps his or her turn and gains time to plan the utterance (see Koch & Österreicher 1990: 60f.).67 6.3.2 Hedged performatives Hedged performatives (see Lakoff 1972: 213; Fraser 1975) are more common than plain performatives as parentheticals. The hedge may either consist of a modal or auxiliary verb or be morphologically marked or both. All hedged performatives I found are based on Fr. dire and its Italian and Spanish counterparts. Frequently, the performative is hedged with 'want to', as in Sp. quiero decir (see Koch & Österreicher 1990: 63) or It. voglio dire (see Bazzanella 1994: 162): (19)


- ¿Cree que ha variado mucho el tipo de enseñanza actual de la antigua?. Vamos, de la antigua, de la anterior quiero decir. (HUSNC.C3H1.252) [Do you think the contemporary type of teaching is very different from the antique one? Come on, from the antique one, from the previous one, I mean]

a footnote, Spitzer (1922: 37) mentions that the frequent use of dico in Italian, and especially of digo in the Venetian dialect, has been seen by foreigners as characteristic of the Italian way of speaking and adds that in Hungarian this word may even be used to refer to Italians.



In the example, the hedged performative is used to mark a replacement repair and follows the expression in its scope. But hedged performatives with 'want to' may also mark further explanations (cf. Morel & Danon-Boileau 1998: 105). In the following example, Fr. je veux dire is positioned after the explaining term in its scope (sur le plan oculaire): (20)

Ä: [...] c'est le seul problème que vous avez, sur le plan oculaire, je veux dire (HOELK.I.6.267) [That's the only problem you have, regarding your eyes, I mean]

This hedged performative signals a further explanation or a specification of a term; it also implies, though, that the explanation or specification is not completely equivalent to the initial term (see Hölker 1988: 114). It intervenes on the phrastic level of the utterance. The 'want to' may be further hedged by moving it into the past, as in the Italian volevo dire. According to Fraser (1975: 204), performatives with Engl. want to provide the hearer with the option of rejecting the implied force. While explanatory Fr. je veux dire in initial position may actually provide the option to reject the assertive force of the utterance, in postposed position it seems to have lost its illocutionary force indicating power. Otherwise we could not explain why it is quite common that the postposed parenthetical corrects or explains terms contained in questions, as in (19) and (20). This phenomenon brings to mind the discussion about It. dico and Sp. digo and ya te digo in the preceding subsection. Regarding its function, initial Fr. je veux dire in part resembles the English expression I mean. Schiffrin (1987: 295-311), who cites only examples with initial I mean, treats it as discourse marker that signals the speaker's upcoming modification of the meaning of his or her prior talk. These modifications include both expansions of ideas and explanations of intentions. In addition to this, she writes that I mean introduces replacement repairs. I also found a performative hedged by 'must', namely It. devo dire: (21)

A: […] spessissimo a Napoli devo dire si incoraggiano quelli che dovrebbero essere non incoraggiati ma resi al silenzio […] (CLIP.NE14) [Very often, in Naples, I must say, they encourage those that should not be encouraged but silenced]

By using 'must', the speaker says that he is obliged to perform a particular speech act and, at the same time, he implies that he wants to be partly relieved of the burden of its consequences (Fraser 1975: 196). In the case of a statement, this means that he wishes to reduce his claim to truth. Normally, It. devo dire occurs in statements expressing personal opinions or beliefs and mitigates the neustic.



The French RPC il faut dire only partly fits into this subsection. From a formal point of view, it is not a performative; in the following example, though, it is employed just as It. devo dire: (22)

A: […] y avait pas beaucoup de monde au cours, il faut dire, […] (ESCH.1.18) [There weren't many people in the course, I must say]

If still treated as a performative, it is double-hedged: by the 'must' and by the deictic displacement of the speaker. A performative may also be hedged morphologically, e.g., by putting it into the conditional form. Fr. je dirais, It. direi, Sp. diría and diría PRO (yo diría and diría yo) are typical RPCs of this type. In the following Italian example68, the performative directly reduces the commitment of the speaker: (23)

AA […] Si può buttar via direi (PIXI.BOF.130) [You can throw it away, I'd say]

Since conditional direi makes the truth of the statement (and the responsibility of the speaker) dependent on an implied condition, I treat it as a responsibility alleviating and removing device. In the Italian example above, the RPC acts on the neustic. In medial position, the RPC tends to assume a function limited to the phrastic. That is, Fr. je dirais, It. direi, Sp. diría and diría PRO may also be employed to highlight expressions not deemed completely satisfying by the speaker: (24)

L1 […] - la plupart du temps ils n'arrivent pas à faire le lien entre l'enseignement euh - je dirais théorique […] (CRFP.LYO-PRO001) [Most of the time, they can't make the connection between the, I'd say, theoretical teaching]

The two different pragmatic functions (and scopes) of this hedged performative can also be clearly seen in the following Spanish passage: (25)


Bueno, yo diría que se han estado quejando durante mucho tiempo de que ... en nuestros hospitales, realmente los medios de que se disponían eran unos medios, yo diría todavía muy precarios […] (COREC.CENT014A) [Well, I'd say that they have been complaining for a long time that the resources actually at the disposal of our hospitals are, I'd say, still very precarious]

For a Spanish example, see Vigara Tauste 1992: 398.



The governing yo diría que reduces the commitment of the speaker regarding the truth of the following object clause, while the parenthetical, noun phrasemedial yo diría marks todavía muy precarios as an expression not considered completely appropriate by the speaker. 6.4 Directly mitigating the neustic 6.4.1 Believing RPCs with belief verbs directly operate on the neustic, that is, on the speaker's commitment to the truth of the proposition, and directly alleviate the speaker's burden of responsibility. Since they are limited to assertive tropics, they imply that the utterance is a statement. The corpus analysis shows that the clauses used for this purpose constitute a relatively closed set of stereotyped forms derived from a few verbs. Fr. croire and its Italian and Spanish equivalents are the basic verbs, while forms deriving from Fr. penser, It. pensare, and especially Sp. pensar are less common. Sp. creer comes in three varieties, namely creo, yo creo and creo yo. The position of the personal pronoun is mainly dependent on the position of the RPC: Those with postposed pronouns are preferred in clause-final position, those with preverbal pronouns in clause-initial position; the same tendency can be noted regarding Sp. pienso PRO and It. penso PRO (for examples see section 8.2). Two other important belief verbs that have parenthetically used forms are Fr. supposer and Sp. suponer as well as Fr. imaginer, It. immaginare and Sp. imaginar. For It. suppongo I did not find a sufficient number of cases that satisfied all the selection criteria defined in section 4.3. In the Italian and French corpora, however, there are sufficient numbers of parenthetical spero and j'espère: (26)

D: [...] # la cosa_ su cui voglio intervenire su questo spero primo intervento è questo concetto di democrazia # [...] (CLIP.RC3) [The issue I'd like to address in this, I hope, first talk is this concept of democracy]

These are the few forms in my list of Romance RPCs with an evaluative as well as a doxastic meaning (see also section 7.1). In Romance, as in German and English, the counterparts of Engl. find may acquire a doxastic meaning and govern noun clauses (see Ducrot 1980: 57-92). Only in French, however, did I find the doxastic je trouve employed as a parenthetical: (27)

L1 [...] intellectuellement c'est quand même - plus enrichissant je trouve - d'échanger avec d'autres [...] (CRFP.PNO-PRI004) [Intellectually, it is however more rewarding, I find, to exchange with others]



In contrast to Fr. j'espère and It. spero, Fr. je trouve is not an evaluative itself but an RPC specialized for evaluative statements. 6.4.2 Not knowing Besides using a doxastic parenthetical, the speaker may employ another device in order to directly mitigate the neustic component and alleviate his or her burden of responsibility, namely saying that he or she does not know. There are two aspects making the Romance equivalents of know relevant for parenthetical clauses and mitigation: The absence of knowledge of the speaker is the subject of the present subsection, the knowledge attributed to the addressee will be discussed in subsection 6.5.2. Know is a cognitive factive verb (see Kiparsky & Kiparsky 1970; Norrick 1978); so is It. sapere (see Arbasi 1986; Schneider 1999: 153-158), and this is certainly also true for its Spanish and French equivalents. Factive predicates presuppose the truth of their dependent clause. That means that, if the predicate is negated, the truth of the dependent clause remains intact. Unlike in emotive and evaluative factives, the presupposition triggered by cognitives is merely a speaker presupposition; that is, it is not shared by the subject of the main clause. X knows that p asserts (not presupposes) that X has the belief p and presupposes that the speaker has the same belief. For that reason, in sentences where subject and speaker coincide, i.e., in the first person of the present indicative, the cognitive factive verb cannot be negated (see Kiparsky & Kiparsky 1970: 148; Norrick 1978: 18). It would be a contradiction to negate that the speaker / subject has a certain belief and presuppose that the subject has this belief. Borillo's (1982) polar question test, which I will explain in detail in section 7.3, groups evidentials like Fr. noter, observer, voir and se souvenir together with epistemics like savoir and être au courant. Both evidentials (= semi-factives) and epistemics (= cognitive factives) are acceptable answers only if negated. Vet (1994: 60) adds that savoir is not capable of governing an affirmative or negative particle:69 (28)

Est-ce que Pierre est partie? *Je sais que qui / non. [Has Pierre left? I know that yes / no]

Borillo's test plays down, though, the profound differences between evidentials and epistemics. If she had tested the governing behavior of negated savoir she would have noted that, contrary to je n'ai pas noté (see subsection 6.5.1), the only possible option after je ne sais pas is an interrogative clause: Vet (1994: 60) also notes that savoir may be in the scope of croire, but not vice versa: (i) Je crois savoir que Pierre est divorcé. (ii) *Je sais croire que Pierre est divorcé. 69




Est-ce que les prix ont baissé? a) *Je ne sais pas qu'ils aient baissé. b) Je ne sais pas s'ils ont baissé. [Have the prices fallen? a) I don't know that they have fallen. b) I don't know if they have fallen]

Contrary to comparable evidentials, je ne sais pas is not ambiguous with respect to factivity, that is, it only allows a non-factive interpretation; je ne sais pas is not a semi-factive (see Karttunen 1971). This changes when the verb is in the third person, i.e., when speaker and subject of savoir do not coincide: (30) (31)

Jeanne ne sais pas que les prix aient baissé. [Jeanne doesn't know that the prices have fallen] Jeanne ne sais pas si les prix ont baissé. [Jeanne doesn't know if the prices have fallen]

In the first example, the speaker knows that prices have fallen (factive interpretation), in the second one, he or she does not know (non-factive interpretation). In most accounts (see Urmson 1952, 1988: 23; Hooper 1975; Borillo 1982; Quirk, Greenbaum, Leech, & Svartvik 1985: 1114; Venier 1991; Ifantidou 2001), Engl. know or its Romance equivalents savoir, sapere, and saber are parenthetical verbs. A few authors, however, e.g., Bronzi (1977: 436), Cornulier (1978: 56, 62f.), Lleó (1979: 170), do not share this view. Vet (1994: 59ff.) compares the following two sentences: (32) (33)

Pierre n'est pas là, je crois. [Pierre isn't there, I believe] Pierre n'est pas là, je sais. [Pierre isn't there, I know]

and states that, in the second one, je sais is pronounced with the intonation of an independent sentence. Hence, we don't have "une phrase suivie d'une assertion, mais deux assertions". It has to be said that, unlike Vet (1994), most authors take into consideration the infinitive, without further pursuing the matter by analyzing specific forms (cf. Schneider 1997b, 1999: 67f.). In French and other Romance languages, it is necessary to distinguish between the perfective meaning of savoir ('come to know, learn') in the past tense (e.g., je sus) and past perfect tense (e.g., j'ai su), and its imperfective meaning ('know') in the present tense (e.g., je sais) and imperfect tense (e.g., je savais). According to Cornulier (1978: 62f.), the parentheticality of Fr. savoir depends on the aspect. In his eyes, only the perfective forms are acceptable in parenthesis. However, in the corpora, there is



only one case where a perfective form, namely It. ho saputo, can be interpreted as a parenthetical: (34)

B: [...] per esempio ho saputo quando dici ad una persona vai vai ad ammazzarti fa morire [...] (CLIP.MA15) [For example, I learned, when you say to a person go go to kill you, this causes her to die]

It. ho saputo could be classified as an evidential parenthetical, because it indicates that an information has been obtained (without saying exactly how). On the other hand, in the analyzed corpora, the imperfective forms Fr. je sais pas70, je sais plus, It. non so and Sp. no sé are frequent parenthetical expressions that fulfill the conditions defined in section 4.3.: (35) (36)

A: […] Vous devez avoir, je sais pas, un petit problème local […] (HOELK.4.2.214) [You must have, I don't know, a small local problem] I: […] la Feria era, no sé, mucho más bonita en todos los aspectos […] (HUSNP.348) [The fair was, I don't know, much nicer in all aspects]

Their unnegated counterparts, though, are completely absent in Italian71 and Spanish,72 and very rarely used in French. In none of the few French cases does je sais interrupt the clause nucleus. Negation sheds light on an important difference between Fr. savoir, It. sapere, and Sp. saber and the other parenthetical verbs: It is the only verb which appears with a negative in parenthetical position. What is the meaning and function of Fr. je sais pas, je sais plus and its Romance equivalents? In (35) and (36), the clauses with the negative directly act on the commitment of the speaker and ease his or her responsibility (cf. Fuentes Rodríguez 1990b: 163). Rarely, though, do they indicate complete ignorance or unawareness (see Morel & Danon-Boileau 1998: 105). Rather, the RPCs may in these cases be substituted for other RPCs directly alleviating speaker responsibility, e.g., je crois or creo. In several other cases, the action of this RPC focuses on a specific referring expression. As with the RPCs mitigating the phrastic, it may affect the precision of Je ne sais pas is practically absent in the French corpora. Interestingly, the only occurrences I encountered are contained in the interviewer's (Kate Beeching herself) questions of BEECH. 71 In the Italian corpora, I found one utterance with the complete parenthetical lo so: (i) H: devo fare tre file lo so (CLIP.RA06) However, as is always the case with final complete parentheticals, in the example above lo so might as well be an independent utterance. I found no parenthetical example of Spanish lo sé. 72 Already Lerch (1919: 65f.) and Tobler (1921: 122f.) pointed to the parenthetical use of French que je sache, which means 'as far as I know'. This expression is absent in my French and Italian corpora, in the Spanish ones, though, the equivalent que yo sepa occurs frequently: (i) [...] Bueno, ¿qué pasa con la actuación de Israel en esa zona? Es decir, de eso no se ha escrito nada que yo sepa todavía. [...] (COREC.CDEB033A) 70



a term. In the following French example, the meaning of je sais plus is 'approximately': (37)

L1 - [...] j'ai bossé avec lui pendant je sais plus trois semaines un mois [...] (CRFP.LIL-PRO001) [I worked with him for, I don't know anymore, three weeks, a month]

That is, this RPC may be employed, in Caffi's (1999, 2001) terms, either as a hedge or as a bush. In the Italian corpora, the bush function of non so predominates. The RPC often indicates that the expression in its scope is intended as a hypothesis or an example, as does Sp. digamos in example (3) cited in section 6.2: (38)

A: […] supponete non so il presidente della camera dei deputati_ eh assegna il progetto di legge a una commissione […] (CLIP.FD2) [Suppose, I don't know, the president of the chamber of deputies assigns the bill to a commission]

In the example above, instead of non so, one could also use per esempio or the RPC diciamo. I also found Sp. no sé being sometimes used in this way. In addition to alleviating responsibility and highlighting examples, Koch & Österreicher (1990: 61) propose hesitation as a function of Sp. no sé and its French and Italian counterparts (cf. also Fuentes Rodríguez 1990b: 163). 6.5 Indirectly mitigating the neustic 6.5.1 Evidentials In the spoken corpora, several parenthetical forms can be found by means of which the speaker refers to the circumstances that may confirm or disprove his or her statement, e.g., by saying that the addressee can see a state of affairs or by describing the source of information. These forms may derive from verbs expressing sensory perception or the obtaining of knowledge, which can be interpreted as mental perception, or from utterance verbs. Forms deriving from verbs expressing the semblance or appearance of a state of affairs also belong to this group. Those in the first person singular of the present indicative ease the speaker's responsibility. The other forms can be subsumed under Caffi's (1999: 896, 2001: 315) class of actantial shields, insofar as they are based on the deictic ego and either divide responsibility between speaker and addressee or remove it entirely from the speaker. All RPCs in this subsection are evidential expressions, because they specify the evidence the asserted proposition relies on. Since they do not directly refer to the speaker's belief or absence of knowledge but to the evidential circumstances of his or her statement, I say that they indirectly mitigate the neustic (cf. Prince, Frader, & Bosk 1982: 89).



Examples of divided responsibility between speaker (who affirms) and addressee (who sees) can be seen in the following examples: (39) (40)

L1 [...] ils sont plus ou moins tu vois débiles [...] (CRFP.PSO-PRO001) [They are more or less, you see, weak] I: […] para mí los tíos, ya ves, son todos amigos míos […] (HUSNP.9.212) [For me the guys, you see, are all my friends]

Depending on their intonation, Fr. tu vois and It. vedi may either be declaratives, as in the French example above, or interrogatives and, therefore, phatic devices. Sp. ya ves is only possible as declarative. Even as declaratives, however, the three RPCs are borderline cases between phatics and evidentials (cf. also Spitzer 1922: 90). In the examples given above, they are probably used as rhetoric devices. The respective addressees probably do not actually see but, rather, should see (see also subsection 6.5.2, regarding 'you know'). Responsibility may be either divided between speaker and addressee or removed from both. A Spanish example of the second strategy is the following: (41)

Inf. A. - [...] Es que es un veneno, dicen, la montaña; […] (HCM.19.352) [It's a poison, they say, the mountain]

In the example above, dicen73 describes, though in an unspecific way, the source and thus the reliability of a statement. It does not report a statement made by a particular person (it is even possible that such a statement has never actually been made), that is, in this particular case it is not a device for reported speech. The primary source of the evidence on which a statement is based is direct observation. Other possible sources are what somebody else had told the speaker, that is, hearsay, and the speaker's inferences (see Chafe & Nichols 1986; Willett 1988; Frawley 1992; Ifantidou 2001: 5ff.; Squartini 2001). Although not normally treated as such, expressions that indicate (one's own) memory as a source of information of a statement should also be classified as evidentials (see Ifantidou 2001: 6f.). Fr. je me rappelle and Sp. me acuerdo74 are evidential parentheticals of this type that one may find, albeit rarely, in the spoken corpora analyzed in this study: (42)

FA 13 - […] c'était euh je me rappelle / dans les années après guerre / […] (CREDIF.FA13.351) [That was, I remember, in the years after the war]

See Guillén Sutil (2001: 103): "Más que significar 'enunciar' o 'expresar el pensamiento con palabras' parece ser que el verbo toma aquí otro valor como es el de 'opinar'". 74 The few examples of parenthetical recuerdo I have found (not enough occurrences to be included, failing to satisfy the criteria) show that this form is employed as a performative. 73




I: [...] hubo uno, me acuerdo, que tenía necesidad, por cuestiones familiares, de ir a su pueblo, [...] (HUSNP.19.421) [There was one, I remember, who had to go, due to family problems, to his village]

Since these expressions are in the first person singular of the present indicative, they alleviate (indirectly, though, as all evidentials do) the speaker's responsibility regarding the truth of the sentence. An important and large group of evidential RPCs derives from verbs expressing the semblance or appearance of a state of affairs. These RPCs either alleviate the speaker's responsibility (e.g., Fr. me semble-t-il) or remove it (e.g., Fr. paraît-il): (44)


L4 […] un éclairage une vision qui évidemment est fort éloigné(e) me semble-t-il - de certains éclairages de certaines visions [...] (CRFP.AIXPUB001) [An illumination a vision that, evidently, is far away, it seems to me, from certain illuminations certain visions] L2 [...] c'était dans la région d'Aigues-Mortes où il y a eu - paraît-il les premières vignes [...] (CRFP.MON-PRI001) [It was in the region of Aigues-Mortes where there have been, it seems, the first vineyards]

Besides parece and me parece (cf. Haverkate 1994: 126), the Spanish texts contain me parece a mí, which underlines the speaker's involvement and responsibility even more. Among the parentheticals referring to the semblance or appearance of a state of affairs, there is an interesting Italian form that stands out from the others. In spoken language, one may find mi sa (or me sa in texts from Rome) governing noun clauses: (46)

A: [...] mi sa che l'ha chiamato proprio lei (CLIP.RA9) [It seems to me that exactly you called him]

The meaning of the expression mi sa, which can roughly be translated as 'it seems to me, I have the impression, I have the feeling', may be explained with the fact that the Italian verb sapere has maintained the two meanings of 'know' and 'taste' it had originally.75 The expression with the meaning given above is used regularly as a parenthetical expression reducing the responsibility of the speaker:

Latin sapēre (or *sapĕre) meant 'taste' and also 'be sage, wise' (see Cortelazzo & Zolli 1999: 1438). 75

128 (47)


O: era il primo mi sa che aveva avuto_ questa_ A: questa intuizione (CLIP.FC6) [O: It was the first one, it seems to me, who had that A: that intuition]

Obviously, mi sa has nothing to do with 'know', i.e., it is not an epistemic RPC. The nature and limits of evidentiality are the subject of debates, a point of major concern being the distinction between evidentiality and epistemic modality. Quite obviously, there are overlaps between the speaker's evaluation of the truth of his or her statement and the speaker's description of the evidence on which the truth of the statement is based. Expressions for sensations (e.g., Fr. me semble-t-il) and inferential expressions (e.g., Fr. je suppose) are typical borderline cases. Albeit indirectly, evidential expressions express speaker commitment, since the quality of evidence indicates what kind of responsibility the speaker is able to assume for his or her statement. Moreover, there are many languages in which the two types of information (speaker commitment and evidence) are coded by the same means, while pure evidential systems are much rarer (Palmer 1986: 66). This is the reason why evidential, epistemic, and doxastic parentheticals are sometimes put under the same umbrella (cf. Ifantidou 2001; Kärkkäinen 2003: 19). Urmson (1952) does not mention parentheticals such as you see, they say, or I remember; but, when dividing the parenthetical verbs into three groups, he writes that verbs such as know, believe, guess, suppose, suspect, estimate, and (in a metaphorical use) feel and adverbs such as certainly, possibly, etc. are […] used to indicate the evidential situation in which the statement is made (though not to describe that situation) and hence to signal what degree of reliability is claimed for, and should be accorded to, the statement to which they are conjoined. (Urmson 1952: 485)

However, already Prince, Frader, & Bosk (1982: 89ff.) distinguish between hedges that involve doubt of the speaker and those that simply attribute a belief to someone other than the speaker, the speaker's commitment being only indirectly inferable. In subsections 6.4.2 and 6.5.2, I will illustrate why I treat the epistemic parentheticals based on Fr. savoir, It. sapere, and Sp. saber separately from doxastic parentheticals. Now I focus on the differences between evidentials and the doxastics illustrated in subsection 6.4.1. Firstly, most evidential parentheticals are not in the first person singular of the present indicative, i.e., there is a morphological criterion that helps to distinguish them from doxastics. Secondly, evidentials only indirectly act on speaker commitment, that is, on the neustic. It is furthermore legitimate to ask whether some of the evidential parentheticals I cited (e.g., Fr. je me rappelle or Sp. me acuerdo) increase rather than reduce speaker commitment. I touched upon this issue in section 5.1, when discussing



the role of expressions of certainty. As pointed out by Hübler (1983: 148), expressions of certainty and, in my view, also evidential parentheticals cannot increase speaker commitment above the level of the corresponding categorical assertion. What they do, however, is to prevent the hearer from considering alternative propositions and to negate the asserted proposition. Another reason for introducing a separate class of evidential parentheticals is that, with the exception of Fr. me semble-t-il and its Italian and Spanish equivalents, these do not pass Borillo's (1982) polar question test (see section 7.3). Finally, the governing capabilities of evidential parentheticals are quite different from those of doxastic parentheticals and point to a semantic property. The difference, however, only becomes manifest when they are negated. Borillo (1982: 50f.; see also Hooper 1975: 117) notes that the complement clause after je n'ai pas noté and other evidentials may either be a declarative with que or an interrogative with si: (48)

Est-ce que les prix ont baissé? a) Je n'ai pas noté qu'ils aient baissé. b) Je n'ai pas noté s'ils ont baissé. [Have the prices fallen? a) I haven't noticed that they have fallen. b) I haven't noticed if they have fallen]

In relation to a similar example, Venier (1991: 72f.) says that a potential ambiguity of evidentials surfaces here. In the first answer, the truth of the dependent clause is presupposed; in the second, it is not. This fact is a consequence of their semi-factivity, a property first observed by Karttunen (1971) in relation to verbs like discover, realize, find out, see, notice, etc., later becoming a much-debated issue (see Hooper 1975; Terrell 1976; Klein 1977; Lleó 1979: 168-175; Arbasi 1986: 172-215; Venier 1991; Schneider 1999: 158ff.). Semi-factivity means that, in certain contexts, these verbs allow both a factive (see Kiparsky & Kiparsky 1970) and a non-factive interpretation. Evidential parentheticals, though, always rely on the non-factive interpretation of the verb.76 In the corpora, the factive version of an evidential is rarely used to govern a clause. That means that an affirmative evidential like, e.g., Sp. me acuerdo governs declarative clauses, while its negative no me acuerdo usually governs polar or focused interrogative clauses. When negated, evidentials lose their indirect mitigating function entirely. They cease to be evidentials because there is no statement for which supporting evidence is needed. Differences between evidentials and doxastics also appear when one reacts with a parenthetical to another speaker's statement: According to Lleó (1979: 170) certain semi-factives are not even acceptable as parentheticals: (i) *El Honorable, noto, va a cultivar su jardín. (ii) *El Honorable, me he enterado, va a cultivar su jardín. 76

130 (49)


Hay mucha diferencia entre el Árabe de primero y el de segundo. a) Me acuerdo. b) *Creo. [There is much difference between the Arab course in the first and in the second year. a) I remember. b) I believe]

Only evidentials but not doxastics may be used to confirm a statement by saying that one has obtained evidence of its truth. On the other hand, only with a doxastic may the speaker refute the interlocutor's statement: (50)

Hay mucha diferencia entre el Árabe de primero y el de segundo. a) *No me acuerdo. b) No creo. [There is much difference between the Arab course in the first and in the second year. a) I don't remember. b) I don't believe]

6.5.2 Knowing In subsection 6.4.2, I illustrated how the speaker may reduce commitment and responsibility by saying that he or she does not know. Now I discuss another aspect which makes the Romance equivalents of know relevant for mitigation: The speaker may attribute to the addressee a certain knowledge. By saying that the addressee also knows, the speaker divides responsibility for what he or she is stating. Instead of referring to his or her belief or absence of knowledge, the speaker refers to external circumstances that might corroborate the statement. As in the case of evidential parentheticals, this is a way to indirectly mitigate the neustic. The interlocutor may be addressed with the familiar second person singular or with a non-familiar form, as in the following two examples with Sp. sabes and It. sa: (51)


Huy, a mí me tocó una mesa ... una mesa de ... de cajón grande. Grandita como ... […] Y luego, sabes, me tocó ... (COREC.CCON021B) [I had to take a table, a table with with a big drawer, quite big as .... and then, you know, I had to take] A: embe' ci vuole sa Filidoro_ (CLIP.RB12) [Well, we need, you know, Filidoro]



In the Spanish example, H1 repeats a preceding utterance almost literally, hence H1 can count on the shared knowledge and use the declarative sabes to refer to it. We should not forget, though, that all the forms mentioned in this subsection may also be interrogative parentheticals with phatic function, inviting the addressee to cooperate and controlling the effectiveness of communication (for Spanish ¿sabes? cf., e.g., Briz Gómez 1998: 224ff.). In that case, they do not modify the speaker's responsibility. Brown & Levinson (1978: 125, 1987: 120) observe that Engl. you know (or ya know) may not claim that the hearer's knowledge is as detailed as the speaker's, but it suggests that the hearer knows the kind of situation in general. This is probably the case in the two Spanish and Italian passages above. Davoine (1981) distinguishes three uses of Fr. tu sais and vous savez: the "tu sais cognitif", the "tu sais d'identification", and the "tu sais de justification". The first use corresponds closely to the knowledge and responsibility dividing exemplified in the two passages above. In the identifying use, a term which is not sufficiently clear to identify its referent is followed by a complementary and explanatory term. It is shown by the following example with Fr. vous savez (see also Hölker 1988: 74f.): (53)

P: [...] ça faisait une petite plaque A: Qui P: chais pas, cinq millimètres sur trois, quoi, vous savez A: Qui P: avec un peu de sang dedans, quoi […] (HOELK.5.5.238) [P: There was a small plaque A: Yes P: I don't know, five by three millimeters, you know A: With some blood inside]

In the justifying use, the explanation introduced by tu sais or vous savez is new, that is, the explanation does not refer to a preceding term (for Spanish cf. also Briz Gómez 1998: 226): (54)

EF - […] j'ai vu des maisons qui avaient que cinq étages alors comme x m'a dit qu'il y avait six étages je me suis dit non c'est parce que le central était derrière tu sais (CREDIF.HE20.286) [I saw houses that had only five floors. Since X had told me that it had six floors, I said to myself "My God, I went to the wrong number". No, that's because the central was behind, you know]

Hölker (1988: 74) remarks that both the identifying tu sais and the justifying tu sais mark explanations. In my opinion, it is more important that in the examples



above, the addressee actually does not know or knows only partly.77 There is a fundamental difference, then, between a 'you know' where the addressee actually knows the contents of the host sentence and a 'you know' where the addressee in reality does not know. In the second case, the speaker may imply, furthermore, that the addressee should know. In fact, according to Morel & Danon-Boileau (1998: 96f.), Fr. tu sais frequently implies that the speaker should know and that actually it should not be necessary to draw the addressee's attention to that particular point. In other words, the RPCs discussed in this subsection may serve as rhetorical devices aimed at forestalling possible refutations by the interlocutor (cf. Hübler 1983: 148; Bazzanella 2001: 253f.): That the addressee often does not know is indirectly confirmed also by Bazzanella (2001: 254), who notes that declarative It. sai as a parenthetical near the beginning of a turn is often used to draw the addressee's attention to the following utterance: (55)

B: sai io chiedo e poi (CLIP.NB63) [You know, I ask and then]

Here, too, the addressee does not know. Undoubtedly, this use of sai is a borderline case between the reduction of speaker responsibility and a call for attention, i.e., a phatic (cf. also Spitzer 1922: 80f.). 6.6 Reporting speech In the analyzed corpora, there are a few RPCs that serve to mark a passage as previously said by another or by the same speaker. All of them are based on Fr. and It. dire, and Sp. decir: (56)


L1 [...] la droite dites-vous n'a pas pu apporter une réponse satisfaisante aux aux données [...] (CRFP.POI-PUB001) [The parties of the right, you say, couldn't provide a satisfactory answer to the data] [...] "Después de ducha ..." ¡Oooh! "... me pongo túnica mora", ha dicho su mujer. [...] (COREC.ALUD007A) ["After the shower", oh!, "I put on the Moorish tunic", said his wife]

In the examples above, an utterance is reported - partly or completely - as it was said before. Only in (57), however, can it be clearly seen that the deictic orientation of the reported utterance remains unchanged, that is, (57) has two deictic centers, one pertaining to the reporting RPC, the other one to the reported utterance. As outlined in Schneider (1997b, 1999: 163), there are many instances in which It. sapere che does not necessarily mean that the addressee knows, although it is a cognitive factive. In these cases the clause is rhematic not thematic. 77



Whereas (57) illustrates the use of free direct speech, in the following example, the boldfaced RPC reports a sequence of free indirect speech: (58)

C: gliel'hai ridata indietro quella scheda riassuntiva che c'aveva lei? A: sì perché dice c'aveva solo quella copia [...] (CLIP.FA4) [C: Did you give her back that summary card she had? A: Yes, because she says she had only that copy]

In free indirect speech, some or all deictic elements of the reported sentence are oriented towards the deictic center of the parenthetical (see Bally 1912a, 1912b; Hummel 1999; Maldonado González 1999: 3551f.; 3572; Mortara Garavelli 2001: 464ff.). Usually, free direct speech and free indirect speech both refer to speech reported without an introductory verb and a complementizer (Weinrich 1982: 801ff.; Hummel 1999: 1636; Andersen 2000: 144). Free indirect speech may be transformed into indirect speech, without changing the orientation of its deictic elements, by adding a complementizer. Reporting RPCs in the present tense often refer to utterances actually said in the past (historic present). In the following Italian example, dice reports such an utterance and refers to the same time and person as the introductory ha detto che: (59)

B: # ha detto che ha preso un sei nel nel compito dice questa bambina dice_ potrebbe fare di più dice ma è una bambina molto timida (CLIP.FA14) [She said that she got a six on the test, she says, this girl, she says, could do more, she says, but she's a very shy girl]

Similar examples can also be found with It. dico and Sp. digo, which are usually plain performatives. Palmer (1986: 67, 71) considers reporting expressions (sentence-initial governing verbs and RPCs) as being part of the evidential system. However, in my view, reporting parentheticals, especially those reporting direct speech, should be set apart from evidentiality. Palmer (1986), in fact, does not mention direct speech or cite examples of it when writing about reporting expressions. Reporting parentheticals are an extreme form of suspending speaker commitment. Since the speaker in reported speech explicitly marks the statement as not being his or her own, he or she neither has to nor is able to give a guarantee for what he or she is saying. Although the speaker is saying a sentence, he or she is not actually stating, asking, requesting or ordering. The illocution (statement, question, directive) expressed by the sentence is not the illocution of the actual but that of the reported speech act. Reported statements do not reflect the judgment of the speaker and hence should be considered as cases of objective modality (see Lyons 1977: 797ff.). They are actually outside the realm of mitigation.



There is a crucial difference between reporting another person's utterance without being committed to it, on the one hand, and making a statement and downgrading commitment by describing the evidence one has for it, on the other. That is why in this book I prefer to treat reporting parentheticals as a group separate from evidentials. I admit, however, that a separation of reported speech from evidentiality and from mitigation in general can be problematic. E.g., since in free indirect speech a statement is reported from the deictic perspective of the speaker, objectivity becomes formally lost and the distinction between merely reporting the statement of another person and expressing one's own statement is blurred. In other cases, there is ambiguity between marking an utterance as being reported and merely citing as evidence for a statement the fact that others have said it before. E.g., in the following sentence invented by Reinhart (1983: 179), Ed will be late can either be a reported or stated: (60)

Ed will be late he said.

According to Reinhart's terminology, if Ed will be late is stated, it is speakeroriented; in that case, the parenthetical he said is functionally equivalent to a modal sentence adverb. In section 9.1, I will illustrate the phonological aspects of Reinhart's reasoning. Due to their objective character, reporting RPCs are not specific to a certain sentence type, that is, they report statements as well as questions (and presumably also directives): (61) (62)

C2w E allora - cerco io qui, dice? (PIXI.BOF.215) [And now, I search here, you say?] ¿Cuántos espectadores ha dicho? (COREC.AENT033A) [How many spectators, you said?]

6.7 Summary Taking in consideration the deictic orientation and the sentence type (declarative, imperative, interrogative), I distinguished clauses alleviating speaker responsibility from those removing and those dividing speaker responsibility. All these clauses reduce communicative responsibility and they are at the center of this book. Initially I also took into account RPCs with phatic function, i.e., I isolated a group of addressee-centered imperative and interrogative clauses which do not reduce speaker responsibility but invite the addressee to cooperate and control the effectiveness of communication. The rest of the chapter focuses on mitigating RPCs. Taking into account the perspective of the logical structures of utterances and the functions of RPCs, I identified and described four main classes of RPCs that reduce communicative responsibility:


1. 2.


clauses mitigating the phrastic (propositional content); clauses indicating the tropic (illocution) and mitigating the phrastic or the neustic (speaker commitment); 3. clauses directly mitigating the neustic; 4. clauses indirectly mitigating the neustic. Several clauses mitigate more than one utterance component. That is, they may alternate between two components (typically phrastic or neustic) or operate on two at the same time (e.g., tropic and phrastic, tropic and neustic). Furthermore, I identified a class of clauses reporting discourse. Since, ideally, these clauses take communicative responsibility completely away from the speaker and speaker commitment is not at stake, they constitute an extreme form of responsibility reduction or lie even outside it. The most prominent representatives of the first class are Fr. disons, It. diciamo, and Sp. digamos. The second class consists of plain performatives, notably It. dico and Sp. digo, and hedged performatives, e.g., Fr. je veux dire, It. voglio dire, Sp. quiero decir, Fr. je dirais, It. direi, Sp. diría and yo diría. Besides indicating the tropic, these parentheticals always do something else: They either mitigate another component of the utterance or have some other additional function (hesitation, self-correction, etc.). The third class consists of clauses expressing belief, i.e., they are the classical RPCs in Urmson's sense (e.g., Fr. je crois and je pense), and of forms of the verb Fr. savoir, It. sapere, and Sp. saber that express absence of knowledge on behalf of the speaker. The fourth class comprises evidential clauses, i.e., clauses that describe the source, type or reliability of a statement, and forms of Fr. savoir, It. sapere, and Sp. saber that attribute knowledge to the addressee or to everybody. The evidential parentheticals derive from verbs expressing sensory perception (e.g., Fr. voir) or the obtaining (or activation) of knowledge (e.g., Fr. je me rappelle), from an utterance verb (e.g., Sp. decir), or from verbs expressing the semblance or appearance of a state of affairs (e.g., Fr. sembler). The analysis indicates that reducing speaker responsibility is not the sole discourse function of RPCs. Besides the phatic function, the indication of the illocutionary force, and reporting someone else's utterance, one has to take into account several other functions which are loosely or not at all related to responsibility reduction. These, e.g., highlighting self-corrections, explanations or repetitions, concern one of the general felicity conditions of speech acts, namely that the procedure must be executed correctly and completely (see Austin 1976 [1962]: 15; Caffi 2001: 450). Tables 6.2-6.4 are parallel to tables 4.2-4.4 and assign to each RPC one or more of the functions illustrated in this chapter. The following abbreviations are used in tables 6.2-6.4: pha = phatic, phr = phrastic, p_per = plain performative, h_per = hedged performative, dox = doxastic, not_k = not knowing, evi = evidential, k = knowing, rep = reporting.



Infinitive Form pha phr p_per h_per dox not_k evi k rep croire je crois + dire disons + dire dites­vous + dire il faut dire + + dire je dirais + + + dire je veux dire + + dire on dirait + + dire on va dire + espérer j'espère + imaginer j'imagine + mettre mettons + paraître paraît­il + penser je pense + savoir je sais pas + + savoir je sais plus + + savoir tu sais + + savoir vous savez + + se rappeler je me rappelle + sembler me semble­t­il + supposer je suppose + trouver je trouve + voir tu vois + + voir voyez­vous + Table 6.2: Pragmatic functions of French reduced parenthetical clauses

Infinitive Form pha phr p_per h_per dox not_k evi k credere credo + dire devo dire + + dire dice dire dice NP dire diciamo + dire dico + + dire direi + + + dire ho detto dire voglio dire + + dire volevo dire + + figurarsi figurati + immaginare immagino + parere mi pare + pensare penso + pensare penso PRO + porre poni + sapere mi sa + sapere non so + + sapere sa + + sapere sai + + sembrare mi sembra + sembrare sembra + sperare spero + vedere vede + + vedere vedi + + Table 6.3: Pragmatic functions of Italian reduced parenthetical clauses

rep + + + +



Infinitive Form pha phr p_per h_per dox not_k evi k acordarse me acuerdo + creer creo + creer creo PRO + decir decía NP decir dice decir dicen + decir digamos + decir digo + + decir digo PRO + + decir diría + + decir diría PRO + + + decir diríamos + decir ha dicho decir ha dicho NP decir me han dicho decir podríamos decir + decir quiero decir + + decir ya digo + + decir ya te digo + + fijarse fíjate + imaginar imagino + imaginarse imagínate + imaginarse me imagino + insistir insisto + parecer me parece + parecer me parece a mí + parecer parece + pensar pienso PRO + repetir repito + saber no sé + + saber sabes + + suponer supongo + ver ya ves + Table 6.4: Pragmatic functions of Spanish reduced parenthetical clauses


+ + + +

+ + +

SEMANTIC AND PRAGMATIC PROPERTIES 7.1 Semantic characterization The issues I want to discuss in this section are the lexical meanings of the verbs in the different classes of RPCs as well as the question as to what extent these meanings play a role in determining which kind of verb might be used for which pragmatic function. Parentheticals mitigating the phrastic are predominantly based on a single utterance verb, namely on Fr. and It. dire, and Sp. decir. The few remaining clauses are based on verbs which do not constitute a homogeneous group. Fr. je sais pas and je sais plus, as well as their Italian and Spanish equivalents, have a doxastic meaning, although they derive from an epistemic verb, while forms such as Fr. mettons and It. poni denote mental operations. The performative parentheticals, i.e., those acting as saying-and-doing performatives and mitigating the phrastic or the neustic, are based on utterance verbs. This might be speaking in general or some particular aspect of it, i.e., saying something very firmly or repeating something. Hedged performative RPCs are based exclusively on Fr. and It. dire, and Sp. decir. Doxastic verbs, i.e., verbs referring to the mental state of belief, are particularly appropriate for the purpose of directly mitigating the neustic and alleviating speaker responsibility. French, Italian, and Spanish each have RPCs based on those verbs (e.g., je crois and je pense). Fr. j'espère and It. spero are based on a doxastic verb with an additional evaluative component. I will return to it below. Verbs denoting mental operations are also employed as devices acting on the neustic (e.g., Fr. je suppose). Fr. j'imagine does not denote an inference but still a mental operation. The operation consists in representing or depicting a state of affairs in one's mind. Originally at least, Fr. je trouve meant 'I find or discover something mentally', that is, it indicated an act of mental perception and was an evidential. Today, however, this meaning has been lost and the verb now is a doxastic specialized for evaluative statements. Fr. savoir, It. sapere, and Sp. saber are epistemic verbs. An epistemic preceded by a negative may be employed like a doxastic. Thus Fr. je sais pas may be used to directly mitigate speaker commitment. Parentheticals that refer to the evidential circumstances of a statement and, hence, indirectly weaken speaker commitment derive from three semantic types: from a verb expressing some kind of perception or sensation (e.g., Fr. tu vois, je me rappelle, paraît-il), which may be sensory, mental, or unspecific, from an utterance verb (e.g., Sp. dicen) or from an epistemic that attributes knowledge to the addressee or to everybody (e.g., Fr. tu sais). The parentheticals of the first



type mostly describe how information is obtained, those of the second type indicate where it comes from, and those of the third type indicate who else knows it. Finally, reporting parentheticals, which release the speaker from his or her responsibility, are all based on Fr. and It. dire, and Sp. decir. Summing up, the RPCs in the corpora contain verbs of the following semantic types: 1. utterance verbs, predominantly Fr. and It. dire, and Sp. decir; 2. doxastic verbs (belief verbs); 3. verbs referring to inferences and other mental operations; 4. verbs expressing sensation or perception (sensory or mental); 5. epistemic verbs, namely Fr. savoir, It. sapere, and Sp. saber. Does semantics play a role in determining which type of verb might be used for which function and for which utterance component? In other words, is there a correspondence between the lexical meaning of the base verb and the function of the RPC? My answer is only a qualified yes. Two functions are fulfilled by homogeneous sets of RPCs: The indication of the tropic (performative parentheticals) and the literal reporting of an utterance (reporting parentheticals) are both accomplished by RPCs with utterance verbs. The other functions are fulfilled by sets which from a semantic point of view are less homogeneous. Parentheticals directly mitigating speaker commitment mainly derive from doxastic verbs or from the negative form of an epistemic; there are also some parentheticals deriving from inferential verbs. With the partial exception of Fr. je trouve, there are no parentheticals directly mitigating speaker commitment that center on perception or sensation verbs. The indirect mitigation of the neustic is done with perception or sensation verbs, with one utterance verb (Sp. decir) and with epistemic Fr. savoir, It. sapere, and Sp. saber. The parentheticals mitigating the phrastic also constitute a heterogeneous group. They are based on utterance verbs, on inferentials, as well as on (negated) epistemics. With the exception of direct weakening of speaker commitment, an utterance verb is possible in all functions I have identified. Due to the negation, epistemic verbs are also quite versatile. Other semantic types are more specialized. Restrictions of use due to lexical meaning of the base verb are not absolute and may be overruled by combining verbs belonging to different semantic types. Urmson states that it is possible to [...] manufacture parenthetical uses of verbs which are not normally parenthetical, even in the first person present, by the addition of the infinitive to say. Thus the verb 'I am sorry that' is normally a formula of apology or self-reproach; but we can convert it into a parenthetical verb by the addition of to say: 'He is, I am sorry to say, unwell'. (Urmson 1952: 493f.)

Other expressions of this kind are I am glad to say, I regret to hear, I am sorry to conclude (see Urmson 1952: 494). According to Bolinger (1968: 20) the



same construction is possible in Spanish, e.g., siento decir. Borillo (1982: 34) points to the fact that in French not only simple verbs like croire may be postposed but also formulas containing croire, e.g., j'ai tout lieu de croire, tout me porte à croire, j'ai tendance à croire (see also Cornulier 1978: 65f.). In connection with discourse signals or discourse markers, the relation between lexical meaning and actual meaning has been examined repeatedly. In an article on Romance imperatives as discourse signals, Lamiroy & Swiggers (1991) point to the functional displacement of a word class and to the diachronic loss of word meaning. E.g., the French imperative Tiens! in most cases neither is used to convey an order, nor shows grammatical agreement with the addressee (cf. Vous ici? Tiens! Tiens! Tiens!), nor establishes a constant semantic relation with tenir. In the case of RPCs, the question of functional displacement and loss of word meaning plays a minor role, although the verb meaning undoubtedly is bleached once the clause is placed in parenthetical position. There are two RPCs directly mitigating the neustic that need special attention: Fr. j'espère and It. spero. Unlike the other doxastic verbs I have mentioned so far, they have an additional evaluative meaning, i.e., they positively or negatively evaluate a supposed state of affairs. As I have pointed out in Schneider (1999: 111f.), It. spero (and supposedly Fr. j'espère) has a volitive component but should not be confused with purely volitive verbs, e.g., It. voglio. Spero, but not voglio, may govern a noun clause referring to the past or be used as parenthetical: (1) (2)

(3) (4)

A: […] spero che non abbiano fatto errori […] (CLIP.MA28) [I hope that they haven't made mistakes] D: [...] # la cosa_ su cui voglio intervenire su questo spero primo intervento è questo concetto di democrazia # [...] (CLIP.RC3) [The issue I'd like to address in this, I hope, first talk is this concept of democracy] *Voglio che non abbiano fatto errori. [I want that they haven't made mistakes] *La cosa su cui voglio intervenire su questo, voglio, primo intervento è questo concetto di democrazia. [The issue I'd like to address in this, I want, first talk is this concept of democracy]

That sentences with parenthetical spero are (mitigated) statements and not requests is further corroborated by the following utterance: (5)

A: […] forse uno dei pochi libri eh a cui possiamo fare riferimento collettivo spero credo eh o [...] (CLIP.ND4) [Maybe one of the few books to which we all can refer, I hope, I believe, or]



These facts suggest that sentences with spero are statements, that is, their words have to match the world (cf. direction of fit in Searle 1976), while those with voglio are directives, whose illocutionary point is to get the world to match the words. Hübler (1983: 137) classifies modal expressions, i.e., parenthetical constructions, modal adverbs, and modal verbs, according to their degree of probability. Two groups of parenthetical constructions are taken into consideration: 1. the verbs think, believe, suppose, guess, seem, and appear (weak assertive predicates, see Hooper 1975); 2. the constructions with copula plus adjective be certain, be sure, be clear, be obvious, and be evident (strong assertive predicates, see Hooper 1975). The spectrum of probability ranges from 'uncertain' (extreme doubt), via 'probable' (moderate doubt) and 'certain' (no doubt), to 'certain' (quasi-presuppositional). The RPCs with verbs of the first group are assigned to 'probable'. They have the same degree as, e.g., the adverb probably, are more certain than, e.g., possibly, and less certain than, e.g., certainly or evidently. The five constructions with copula and adjective are assigned to 'certain' (no doubt). What is relevant for us is that, in Hübler's scheme, RPCs have a very limited range. They neither occupy the possibility pole nor the certainty pole. Hübler affirms that the reason lies in negative raising. As I explained in section 3.3, Hooper (1975: 97f., 105ff.) tests the behavior of parentheticals with respect to the negative not: Certain parentheticals, i.e., the weak assertive predicates listed above, allow the negative to move freely within the host without changing the host's meaning. According to Hübler (1983: 138), this feature pertains only to expressions of the 'probable' degree: (6) (7) (8)

It's probable that not p = It's not probable that p It's possible that not p ≠ It's not possible that p = It's certain that not p It's certain that not p ≠ It's not certain that p = It's possible that p

As we shall see in section 7.3, the behavior of the negative can also be tested when the RPC is used as an answer to a polar question. Only a few clauses in my lists allow the negative to move freely and are, if we adopt Hübler's standpoint, expressions of probability. The question remains, though, as to where in Hübler's scheme we have to position the other clauses. There is strong evidence that the Romance parentheticals do not fall into the 'uncertain' (extreme doubt) category. In section 5.1, I illustrated Hübler's (1983: 98f., 141ff.) ideas about the assertory and questory functions of statements. Questory statements are those in which the speaker has a neutral attitude towards the truth of the proposition. They contain expressions of extreme doubt, e.g., Engl. possibly, perhaps, or maybe, which in English may be further enhanced by a following but I don't know. Hübler's test is easy to apply to sen-



tences with Romance RPCs. Sentences with parentheticals mitigating the phrastic, with plain performatives, with hedged performatives, with parentheticals directly mitigating the neustic and with evidentials cannot be continued with Fr. mais je (ne) sais pas, It. però non (lo) so, or Sp. pero no (lo) sé. A second property of statements containing expressions of extreme doubt is that they allow the formation of an equivalent indirect speech version introduced by he wondered whether (see Hübler 1983: 143f.): (9) (10) (11)

Sue lied. Sue possibly / perhaps / maybe lied. He wondered whether Sue lied.

The indirect speech form with he wondered whether is appropriate only for questory statements, that is, only for (10). Likewise, Romance indirect speech versions introduced by Fr. il s'est demandé si, It. si è chiesto se, and Sp. se preguntó si are possible only if the sentence contains an expression of uncertainty in addition to the parenthetical. In the following French example, (13) but not (12) may be transformed into an indirect speech version: (12) (13) (14)

Vous avez, je sais pas, un petit problème local. [You have, I don't know, a small local problem] Vous devez avoir, je sais pas, un petit problème local. [You must have, I don't know, a small local problem] Le médecin s'est demandé si le patient avait un petit problème local. [The doctor wondered if the patient had a small local problem]

According to Hübler (1983: 137ff.), the modal verb may is one such expression of extreme doubt. It allows the continuation with but I don't know and the indirect speech counterpart with he wondered whether. 7.2 Influence of person and time According to Urmson (1952), parenthetical verbs are exclusively in the first person singular of the non-progressive present indicative form. I found several Romance verbs in parenthetical position that are not in the first person or not in the present tense or not in the indicative. Yet, if we take into account only those parentheticals that directly mitigate the neustic, Urmson's view is basically correct. What about RPCs with doxastic verbs in forms other than the first or third person singular of the present indicative? Urmson writes regarding comparable cases in English: [...] while 'He believed that it was lost', if said by me, does not imply a modified truth claim to reasonableness by me for the statement that it is lost, it does allege exactly that of the man to whom I refer. [...] The point of these verbs remains as a kind of orientat-



ing signal, but when not in the first person present they report the statement-cum-signal rather than making it. (Urmson 1952: 492)

That means that, by saying He believed that it was lost, the speaker reports a proposition that was asserted and mitigated (= statement-cum-signal) by another person. There is one aspect that proves Urmson's point immediately. Clauses governed by, e.g., Sp. cree may contain a modal sentence adverb, whereas those governed by creo are only marginally acceptable with it (see Schneider 1999: 70f. for Italian examples): (15) (16)

[...] a lo mejor cree que es que le tiene que tratar de otra manera [...] (COREC.CCON029A) [Maybe he thinks he has to treat him differently] ? A lo mejor creo que es que le tiene que tratar de otra manera. [Maybe I think he has to treat him differently]

(16) is acceptable only if intended as self-description, i.e., as a statement of the speaker's own belief. The reason is clear: In (15), the main clause is asserted and can be mitigated by a lo mejor. In (16), the main clause does not assert but mitigate Es que le tiene que tratar de otra manera. The mitigation of a mitigation would be pragmatically unacceptable (see also Ziv 1985: 191ff.). Note that this does not exclude the possibility of two mitigators acting on the same assertion: (17)

HR24 y a une certaine pudeur peut-être je sais pas / (CREDIF.HR24.549) [There is maybe a certain sense of decency, I don't know]

Fr. je crois que, peut-être, and je sais pas express reduced commitment on behalf of the speaker. Two mitigations of the same assertion and with the same function are pragmatically acceptable. Can doxastic verbs, in forms other than the first person singular of the present indicative, occur in parenthetical position? In Urmson's (1952: 488) opinion, believe in the third person may indeed occur in this position: (18)

X is, Jones believes, at home.

According to Venier (1991: 120), in Italian even the progressive form (= stare + gerund) is possible in parenthetical position:78 (19)

Verrò a trovarti, sto pensando. [I'll come to see you, I am thinking]

Urmson (1952: 491), though, writes: "[…] one cannot say 'I was believing', 'he is believing', 'I was knowing', 'he was knowing', 'I was suspecting' or 'he was suspecting'." 78



I do not regard these sentences as unacceptable. As a matter of fact, however, parenthetical occurrences of doxastic verbs in forms other than the first person singular of the present indicative79 are practically non-existent in my corpora of spoken Romance. This is partly due to a general tendency in spoken discourse to use these verbs, even if in governing position, only in the first person singular of the present indicative. For example, 84 % of the 496 occurrences of croire in the CREDIF are je crois, 82 % of the 473 occurrences of credere in the CLIP are (io) credo, and 91 % of the 1306 occurrences of creer in the COREC are (yo) creo. 7.3 RPCs as answers to polar questions It has been noted repeatedly (see Bolinger 1968: 11f.; Borillo 1976, 1982; Weinrich 1982: 694, 703; Conte 1987: 32; Venier 1991; Gil 1995: 97f.; Schwarze 1995: 265; Schneider 1999: 48, 59ff.) that certain RPCs and modal sentence adverbs may be used as affirmative or negative answers to polar questions. Borillo (1982) tested the behavior of French RPCs as answers to polar questions. Depending on whether the negative is possible or not, the answer may be one of three different types. Croire, avoir l'impression, penser, trouver, sembler, and on dirait may be used with or without the negative, depending on the answer, affirmative or negative, one wishes to give: (20)

Est-ce qu'il est rentré? a) Je crois. b) Je ne crois pas. [Did he return? a) I believe. b) I don't believe]

Note that with Je ne crois pas one negates the implicit dependent clause qu'il est rentré and not the clause Je crois. What is being negated is not the fact that I believe, but the fact that he has returned. This becomes evident when one takes into account the equivalence between Je ne crois pas and Non. In fact, in the answers above Je crois and Je ne crois pas are functionally (not semantically) equivalent to Qui and Non respectively: (21)

Est-ce qu'il est rentré? a) Qui. b) Non. [Did he return? a) Yes. b) No]

Spanish parecer, French sembler and paraître, and Italian parere and sembrare are verbs referring to sensations. 79



Je crois is transparent to negation; the negative, although adjacent to je crois, does not operate on it. According to Borillo (1982), the verbs of the second group, e.g., supposer, espérer, imaginer, présumer, craindre, and il paraît may be used as answers only in the unnegated version: (22)

Est-ce qu'il est rentré? a) Je suppose b) *Je ne suppose pas. [Did he return? a) I suppose. b) I don't suppose]

These verbs are not transparent to negation because the latter affects their truth value. With Je ne suppose pas I negate that I suppose, not that he has returned. Although grammatical, the negated answer is contextually inappropriate and not equivalent to Non. If the speaker wants to use an equivalent to Non, he or she must say Je suppose que non80. The following Spanish example shows this point clearly, since it contains a transparent and a non-transparent verb side by side: (23)

[...] ¿cree uste que Pepe Itarburi finalizará alguna vez alguno de sus chistes? No creo, supongo que no. (COREC.BLUD032A) [ Do you think that Pepe Itarburi will finish some day one of his jokes? I don't believe, I suppose not]

The verbs of the third group, e.g., savoir, noter, remarquer, sentir, voir, se rendre compte, etc., may be used as answers to a polar question only if negated: (24)

Est-ce qu'il est rentré? a) *Je sais. b) Je ne sais pas. [Did he return? a) I know. b) I don't know]

While I share Borillo's (1982) view regarding supposer - in the corpora, there are no cases of je (ne) suppose pas -, a correction is necessary for espérer. As documented by several cases in my corpora, this French verb is transparent to negation: (i) L2 c'était pas quand même un caractère j'allais dire normand pour revenir au au profil