Parenthetical Verbs 9783110376142, 9783110376036

Citation preview

Parenthetical Verbs

Linguistische Arbeiten

Edited by Klaus von Heusinger, Gereon Müller, Ingo Plag, Beatrice Primus, Elisabeth Stark and Richard Wiese

Volume 557

Parenthetical Verbs Edited by Stefan Schneider, Julie Glikman and Mathieu Avanzi

DE GRUYTER

Gefördert von der Karl-Franzens-Universität Graz.

ISBN 978-3-11-037603-6 e-ISBN (PDF) 978-3-11-037614-2 e-ISBN (EPUB) 978-3-11-039419-1 ISSN 0344-6727 Library of Congress Cataloging-in-Publication Data A CIP catalog record for this book has been applied for at the Library of Congress. Bibliographic information published by the Deutsche Nationalbibliothek The Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data are available on the Internet at http://dnb.dnb.de. © 2015 Walter de Gruyter GmbH, Berlin/Munich/Boston Printing and binding: CPI books GmbH, Leck ♾ Printed on acid-free paper Printed in Germany www.degruyter.com

Table of Contents Acknowledgements | vii Stefan Schneider, Julie Glikman and Mathieu Avanzi Introduction | 1

Section I: Theoretical Issues Frederick J. Newmeyer Parentheticals and the grammar of complementation | 13 Gunther Kaltenböck, Bernd Heine and Tania Kuteva On theticals: A “rootless” analysis of I think | 39 James Griffiths Speaker and quote reduced parenthetical clauses | 71 Marlies Kluck and Mark de Vries On V2, gaps, and operators in comment and reporting parentheticals | 103

Section II: Diachronic and ontogenetic development Fátima Faya Cerqueiro An approach to parenthetical courtesy markers in requests in Late Modern English | 135 Stefan Schneider and Julie Glikman Origin and development of French parenthetical verbs | 163 Ditte Boeg Thomsen Ontogenetic paths to the parenthetical construction | 189

vi | Table of Contents

Section III: Prosody and its interfaces Nancy Hedberg and Noureddine Elouazizi Epistemic Parenthetical Verb Phrases: C-Command, Semantic Scope and Prosodic Phrasing | 227 225 Sylvie Hanote Are reporting clauses special cases of parentheticals? | 257 259 Güliz Güneş and Çağrı Çöltekin Mapping to prosody: Not all parentheticals are alike | 287 Silvana Abalada and Aida Cardoso The Vocative in European Portuguese: Prosodic Effects of its position in the sentence | 333 Index | 353 355

Acknowledgements The contributions in the present volume, except for the invited article by Kaltenböck, Heine and Kuteva, were presented at the workshop Parenthetical verbs: hypotaxis, parataxis or parenthesis?, held 24–26 May 2012 at the Paris West University Nanterre La Défense. The workshop was supported by the Laboratoire MoDyCo (UMR 7114 and CNRS) and by the Karl-Franzens-Universität Graz. The papers went through a severe three-stage double-blind reviewing process. We thank all the contributors for their patience, their accurate revision work and for respecting the various deadlines during this period. We are also indebted to the numerous reviewers who generously spent their time and energy in reading and commenting on the papers for the workshop and for this volume. We owe particular thanks to the editors of the series Linguistische Arbeiten for their decisive assistance during the selection and revision of papers. Finally, we want to thank Eva Triebl for the preparation of the manuscript and making it ready for printing. Mathieu Avanzi Julie Glikman Stefan Schneider

Stefan Schneider, Julie Glikman and Mathieu Avanzi

Introduction || Stefan Schneider: Institut für Romanistik, Karl-Franzens-Universität. Graz, Austria. Julie Glikman: Faculté des Lettres, LiLPa, Université de Strasbourg. Strasbourg, France. Mathieu Avanzi: Institut des Sciences du Langage et de la Communication, Université de Neuchâtel. Neuchâtel, Switzerland.

1 Presentation of the volume Parenthesis has recently seen a considerable surge in interest, as shown by a number of international workshops and meetings held in the last few years (Bielefeld 2006, Paris 2008, Paris 2010, Paris 2012, Potsdam 2013) and by three edited volumes (Dehé and Kavalova 2007a; Corminboeuf, Heyna and Avanzi 2010; Glikman and Avanzi 2012) resulting from them. Such an abundance of meetings and publications shows, on the one hand, that parenthesis is a current subject and suggests, on the other, that there are some unresolved issues and questions waiting to be investigated. Among the aforementioned volumes, only Glikman and Avanzi (2012) focuses specifically on parenthetical verbs. The present volume, which is the outcome of the international workshop Parenthetical verbs: hypotaxis, parataxis or parenthesis?, held 24–26 May 2012 at the Paris West University Nanterre La Défense, examines parenthetical verbs in several languages. It was the explicit intention of the workshop’s organisers to present the – often contrasting – theoretical positions, to shed new light on parenthetical verbs from different perspectives and to analyse them also in languages other than English and French. This eclectic orientation of the workshop is reflected in the volume at hand. The initial theoretical section hosts two contributions, one by Newmeyer and the other one by Kaltenböck, Heine and Kuteva, presenting diametrically opposed analyses of the syntax of parenthetical verbs. Although the volume is pervaded by (synchronic) syntax, it also gives weight to other analytical levels and approaches: The second section provides insights into the historical and ontogenetic development of parenthetical verbs and the third section deals with prosody and the prosody-syntax or prosody-semantics mappings. The volume covers parenthetical verbs in English as well as in several other languages (Danish, Dutch, French, German, Portuguese and Turkish), some of them rarely taken into consideration when dealing with parentheticals. Another distinguishing

2 | Stefan Schneider, Julie Glikman and Mathieu Avanzi feature of this volume is its strong empirical orientation: With a few exceptions, the contributions are based on data from experiments or from different kinds of corpora (historical vs. contemporary, oral vs. written, child speech vs. adult speech, etc.).

2 Characterisation of parenthetical verbs Although relevant thoughts about the parenthetical use of verbs have been expressed before (e.g. by Schwyzer 1939), it was Urmson (1952) who introduced the term parenthetical verb to characterise a group of English verbs in the first person non-progressive present indicative that occur either in initial position or in intermediate position, interrupting an utterance, or in final position, being appended to an utterance: (1) I suppose (that) your house is very old. (2) Your house is, I suppose, very old. (3) Your house is very old, I suppose.

According to Urmson (1952: 495), the assertion proper is contained in the host clause. By using a parenthetical verb, the speaker indicates how the content of the host fits logically, evidentially and emotionally into the context. Urmson’s (1952) understanding of parenthetical verbs is very broad, including initial governing clauses and final clauses. Instead of attempting to define their exact place in the sentence, he tries to capture their pragmatic function, which is comparable to that of sentence adverbs and which, in his eyes, does not change with position. He therefore frequently uses the expression parenthetical use when referring to this particular function. This function is a property of specific forms of these verbs. Notwithstanding the name parenthetical verbs, Urmson (1952) is of course aware of the fact that not verbs per se are parenthetical, but that specific forms of verbs may be used within a parenthetical clause. A few years after Urmson, Hall (1958) analyses the use of parenthetical I assume in a similar way, as a device for altering, cancelling or altering the speaker commitment, and refers to it as positing word. Benveniste (1966 [1958]), underlining more their function than their syntactic position, calls them verbes d’opération, i.e. ‘verbs of (cognitive) operation’. These three terms already point to a problem arising from these verbs: Should we define and name them according to their discourse function or according to their syntactic property? Urmson (1952) opted for a compromise: He defined parenthetical verbs by their pragmat-

Introduction | 3 Parentheticals and the grammar of complementation

ic function but named them by referring to the syntactic property of parenthesis. In fact, the ambiguity between a discourse-functional concept and a purely syntactic concept still persists. Since then, numerous other terms have been proposed, none of them entirely satisfying (cf. Schneider 2007: 3–7). This terminological heterogeneity is also present in the volume at hand. The authors were free to use the name they preferred. Hence, besides parenthetical verb, which is used despite its limits and mainly because of its notoriety, the reader may find (reduced) parenthetical clause, comment clause, parenthetical (verb) phrase, parenthetical marker or, simply, parenthetical. The common denominator of all but one contribution is the fact that they deal with parentheticals that are made of a single finite verb or whose main part is a finite verb. Therefore, the focus of the volume is on those clauses with finite verbs that may be inserted everywhere in the host without an overt connective linking them to it, whose verbs lack one of the arguments required by their valency, and whose lacking argument can be recovered semantically from the host sentence. Some borderline cases, which do not satisfy these conditions but which are interesting and relevant for discussion, have been included. The contribution by Abalada and Cardoso describes the prosodic properties of Portuguese parenthetical vocatives consisting of proper nouns. The contribution of Güneş and Çöltekin deals with several kinds of Turkish parentheticals, some of which are verbless. To cite another example, Schneider and Glikman’s study on Old and Middle French parenthetical verbs comprises verbs with all arguments present or with connectives linking them to the host.

3 Open questions and issues There are several reasons for the large and heterogeneous set of terms proposed, which suggests the existence of several not completely resolved issues. These are certainly due to the two aforementioned perspectives, syntactic property or pragmatic function, from which these verbs can be described. Similarly, the French term verbe recteur faible (‘weakly governing verb’, cf. BlancheBenveniste 1989) underlines the reduced governing capability of the verbs involved, whereas comment clauses (cf. Biber, Johansson, Leech, Conrad and Finegan 1999) emphasises their pragmatic function. Moreover, there is the question of the syntactic status of these verbs. What kind of syntactic unit do they constitute, that is, are they words (cf. parenthetical verbs), phrases (cf. postposed main phrases in Bolinger 1968) or clauses (cf. reduced parenthetical clauses in Schneider 2007). The same issue also concerns the word class and the

4 | Stefan Schneider, Julie Glikman and Mathieu Avanzi syntactic class: Are these items verbs or some type of adverb or adverbial expression (cf. parenthetic adjuncts in Corum 1975). In addition, linguists do not agree upon the definition of parenthesis. Although there is a minimal common understanding of parenthesis, and most linguists would agree with a description such as “Parentheticals are expressions of varying length, complexity, function and syntactic category, which are interpolated into the current string of the utterance” (Dehé 2009: 307), parenthesis remains a problematic notion (cf. Dehé and Kavalova 2007b: 1–4; Kaltenböck 2007: 25–27; Schneider 2007: 19–35). To mention an example, for several authors, je crois ‘I believe’ in the following example is an instance of parenthesis: (4) C: Ah ben c’est plus libre maintenant, je crois (Beeching 2002, text 16) ‘Oh, well, it’s freer now, I believe’

Not all linguists share this opinion (cf., e.g., Hoffmann 1998). To cite a famous view on this matter, for Bloomfield (1935) the only difference between parenthesis and parataxis is position: With parenthesis one form interrupts the other, with parataxis the two forms occur in succession. From the pragmatic perspective, it seems necessary to review the functions assigned to parenthetical verbs. According to Urmson (1952), their role is to mitigate the propositional content of the utterance to which they are attached. Since the time this statement was made, several authors have shown that the discourse functions of parenthetical verbs are much more multifaceted. Various taxonomies have been proposed, but these are still far from complete (cf. Kaltenböck 2006; Schneider 2007). The issue is important in the light of other items which have similar syntactic properties, as, e.g., speech reporting verbs (cf. Griffiths, this volume; Hanote, this volume; Kluck and de Vries, this volume) and discourse markers (cf. Faya Cerqueiro, this volume). From a syntactic perspective, the nature of the relationship between the parenthetical verb and its host construction needs to be discussed. Are we dealing with syntactic government, with non-government, or with an intermediate form of relationship or dependency (cf. Hedberg and Elouazizi, this volume; Kaltenböck, Heine and Kuteva, this volume; Newmeyer, this volume)? Which empirical tests can help us choose between these options? The issue is particularly tricky in the case of utterance-initial complement-taking predicates, which are the subject of a heated debate. Thompson and Mulac (1991a, 1991b) and even more radically Thompson (2002) defend the idea that the complement clause is central and that the utterance-initial predicate is a mere formulaic stance marker. Newmeyer (2010) supports the view that the finite clausal com-

Introduction | 5 Parentheticals and the grammar of complementation

plement (with or without complementiser) is subordinate to the complementtaking predicate. Boye and Harder (2007) and Schneider (2007: 191–197) underline that the status of the epistemic predicate is highly context-dependent, which is to say that the issue can only be resolved on a case-by-case basis. This is confirmed by Dehé and Wichmann’s (2010) prosodic study on utteranceinitial I think (that) and I believe (that). Data from English and German child language (cf. Diessel and Tomasello 2001; Brandt, Lieven and Tomasello 2010) suggest that some complement-taking clauses, e.g., I think and ich glaube ‘I believe’, are acquired early in development as autonomous chunks (cf. Boeg Thomsen, this volume). Parenthetical verbs are, from a sentential perspective at least, incomplete constructions. One of the valency requirements of the verb is satisfied by the entire host clause, which poses serious problems for syntactic analysis. There is no doubt, however, that the discussion about completeness vs. incompleteness is strongly biased towards a sentential view of grammar. Some recent approaches take a broader perspective (cf. Kaltenböck, Heine and Kuteva 2011; Kaltenböck, Heine and Kuteva, this volume). In such frameworks, there is no need for concepts such as incomplete or reduced parenthesis. The prosodic aspects of parenthetical verbs are still not fully understood. The main questions can be resumed as follows: To which level(s) of unit do parenthetical verbs correspond in the prosodic hierarchy? In other words, do parenthetical verbs always constitute intonation units that are separated and prosodically autonomous from their host or are they integrated into the host in different – and possibly more complex – way? Their prosodic shape and relation with the host certainly depends, among others, on their position in the sentence (initial, medial or final), on the host’s word order, on the type of syntactic subject (pronoun vs. noun phrase), on the parenthetical verbs themselves and on their pragmatic function. There are a number of studies based on English, French and German that try to find answers to these questions (cf., e.g., Dehé 2007, 2014; Döring 2007; Kaltenböck 2008; Dehé and Wichmann 2010; DelaisRoussarie 2010; Avanzi 2012a, 2012b; Hanote, this volume), but it is clear that further analyses, taking into account other type of material and other languages (cf. Abalada and Cardoso, this volume; Güneş and Çöltekin, this volume) are necessary in order to reach reliable conclusions. Only recently, researchers are starting to focus on the diachronic development of these structures, primarily on English and Romance parentheticals (cf. Brinton 2001, 2008; Waltereit 2002, 2006; Company Company 2006; Van Bogaert 2011; Schneider 2011, 2013; Glikman 2009, 2012). The diachronic perspective is necessary to achieve a full understanding of the synchronic phenomenon. The following questions are raised repeatedly: What are the syntactic,

6 | Stefan Schneider, Julie Glikman and Mathieu Avanzi semantic and pragmatic origins of parenthetical verbs? Do they derive from superordinate clauses, subordinate clauses (adverbial clauses) or independent sentences (cf. Glikman and Schneider, this volume)? When and how did they begin to be used as parentheticals (cf. Faya Cerqueiro, this volume)? How did their meanings change during this process? Is the development from full verb to parenthetical verb an instance of grammaticalisation, lexicalisation or pragmaticalisation? Are the hypothesised diachronic paths mirrored by ontogenetic development (cf. Boeg Thomsen, this volume)?

4 Overview of the contributions The volume is divided into three sections. The four contributions following the introduction address fundamental theoretical issues. Although they are mostly based on a single language they are of general relevance. Contributions 6 to 8 look at parentheticals from a developmental perspective and deal with language-specific historical or ontogenetic problems. The final section consists of four contributions dedicated to the prosody, the syntax-prosody or semanticsprosody mappings of particular languages. The two contributions by Newmeyer and by Kaltenböck, Heine and Kuteva address a fundamental question: What is the syntactic nature of parenthetical expressions such as I think? The analyses proposed by the two contributions are diametrically opposed. Newmeyer argues that I think in the following examples is a complement-taking main clause at one point in its derivation: (5) I think we should leave as soon as possible. (6) We should leave, I think, as soon as possible. (7) We should leave as soon as possible, I think.

Kaltenböck, Heine and Kuteva, on the other hand, favour an analysis according to which I think, at least in (6) and (7), is syntactically independent. They classify it as a different kind of syntactic constituent or as no syntactic constituent at all. The following two contributions, by Griffiths and by Kluck and de Vries, focus on two groups of parenthetical verbs (or reduced parenthetical clauses): comment or mitigating clauses and reporting or quote clauses. Griffiths explores to what extent a unified analysis may be applied to both groups of parentheticals, using data from English, Dutch, German, and Turkish. According to Griffiths, such an analysis may account for the vast majority of the disparate properties that these parentheticals display. Likewise, Kluck and de Vries argue,

Introduction | 7 Parentheticals and the grammar of complementation

based on Dutch, English and German data, that various properties exhibited by the two groups of parentheticals, such as word order patterns, possible argument gaps and the interpretation and position of items such as Dutch zo, English and German so and Dutch zoals, English as and German wie, are related and can receive a unified analysis. The diachronic and ontogenetic section starts with Faya Cerqueiro’s contribution on the origin of the parenthetical request markers pray and please in 18th and 19th century English. According to the data from historic corpora, pray originates from the parenthetical use of I pray you, whereas the emergence of please as a parenthetical took place later, when the imperative please to was reanalysed as a pragmatic marker, the particle to was dropped and please spread from pre-verbal positions to any place in the sentence. The contribution by Schneider and Glikman also addresses the origin and development of specific parentheticals, precisely parenthetical verbs derived from croire ‘believe’, cuidier ‘believe’, espérer ‘hope’, penser ‘think’ and sembler ‘seem’ in French texts composed between the 9th and the 16th century. Three possible sources are examined: superordinate clauses, subordinate clauses, and juxtaposed sentences. According to their analysis, juxtaposed sentences seem to offer the most plausible explanation. Boeg Thomsen focuses on children’s parenthetical clauses in Danish. Theoretically, children could categorise parenthetical clauses as complement-taking clauses, adverbial clauses, independent clauses or as autonomous chunks. Based on data from a kindergarten corpus, Boeg Thomsen affirms that children’s parenthetical clauses most likely develop from matrix clauses. These results thus differ markedly from findings on English and German acquisition of parenthetical clauses (cf. Diessel and Tomasello 2001; Brandt, Lieven and Tomasello 2010) where highly frequent complement-taking clauses used in parenthetical position (I think and ich glaube) appear to be acquired as unanalysed chunks. In the section dedicated to prosody and its interface, Hedberg and Elouazizi examine four parenthetical verbs (I believe, I think, I guess, and I suppose) in the Fisher Corpus of American English. The authors argue that there is a transparent relationship between the semantic scope of these verbs and the prosodic phrasing of the constituent to which they are syntactically adjoined. Hanote investigates the intonational phrasing of reporting clauses with the verb say in a corpus drawn from the radio (BBC Radio for British English and National Public Radio for American English). According to her analysis, the clauses mostly are integrated in the rest of the utterance and function as preheads or as tails of intonational phrases. They behave as isolated chunks only when they are initial. Based on laboratory speech, the contribution by Güneş and Çöltekin focuses on two types of Turkish sentence-medial parentheticals: clausal parentheti-

8 | Stefan Schneider, Julie Glikman and Mathieu Avanzi cals and phrasal parentheticals. The results of their experiment show that clausal parentheticals exhibit cues similar to intonation phrase-level cues, whereas phrasal parentheticals exhibit cues similar to phonological phraselevel cues on both edges. Thus, the authors conclude that clausal and phrasal parentheticals do not belong to the same level in prosodic hierarchy: clausal parentheticals are prosodically isolated, while phrasal parentheticals are prosodically integrated. The section is completed by Abalada and Cordoso’s study on Portuguese parenthetical vocatives. According to the data, the vocative’s position plays a crucial part in its prosodic behaviour. Abalada and Cardoso draw a distinction between initial and non-initial vocatives. The former share characteristics with isolated vocatives and the latter form a homogeneous group. Overall, this thematic volume brings together authors with diverse theoretical backgrounds, working on various languages and in different domains of language science. Presenting cutting-edge research on the topic of parentheticals, we hope that this book will pave the way for fruitful future study in this field.

References Avanzi, Mathieu. 2012a. La prosodie des verbes parenthétiques en français parlé. In Julie Glikman & Mathieu Avanzi (eds.). Entre rection et incidence: des constructions verbales atypiques? Ètude sur je crois, je pense et autres parenthétiques (= Linx 61). Nanterre: Université Paris Ouest Nanterre La Défense. 131–144. Avanzi, Mathieu. 2012b. L’interface prosodie/syntaxe en français. Dislocations, incises et asyndètes. Bruxelles: Lang. Beeching, Kate. 2002. Un corpus d’entretiens spontanés. http://www.uwe.ac.uk/hlss/llas/iclru/corpus.pdf. Benveniste, Émile (1966 [1958]). De la subjectivité dans le langage. In Émile Benveniste. Problèmes de linguistique générale. Vol. 1. Paris: Gallimard. [Originally published in Journal de Psychologie 55: 257–265]. 258–266. Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad & Edward Finegan. 1999. Longman grammar of spoken and written English. Harlow: Longman. Blanche-Benveniste, Claire. 1989. Constructions verbales en incise et rection faible des verbes. In Recherches sur le français parlé 9. 53–73. Bloomfield, Leonard. 1935. Language. London: George Allen & Unwin. Bolinger, Dwight. 1968. Postposed main phrases: an English rule for the Romance subjunctive. Canadian Journal of Linguistics 14. 3–30. Boye, Kasper & Peter Harder. 2007. Complement-taking predicates: Usage and linguistic structure. Studies in Language 31. 569–606.

Introduction | 9 Parentheticals and the grammar of complementation

Brandt, Silke, Elena V. M. Lieven & Michael Tomasello. 2010. Development of word order in German complement-clause constructions: Effects of input frequencies, lexical items, and discourse function. Language 86. 583–610. Brinton, Laurel J. 2001. From matrix clause to pragmatic marker. The history of look-forms. Journal of Historical Pragmatics 2. 177–199. Brinton, Laurel J. 2008. The comment clause in English: Syntactic origins and pragmatic development. Cambridge: Cambridge University Press. Company Company, Concepción. 2006. Subjectification of verbs into discourse markers: semantic-pragmatic change only? Belgian Journal of Linguistics 20. 97–121. Corminboeuf, Gilles, Franziska Heyna & Mathieu Avanzi (eds.). 2010. Les parenthèses en français (= Verbum 30: 1). Nancy: Presses Universitaires de Nancy. Corum, Claudia. 1975. A pragmatic analysis of parenthetic adjuncts. In Robin E. Grossman, L. James San & Timothy J. Vance (eds.). Papers from the eleventh regional meeting. Chicago Linguistic Society, April 18–20, 1975. Chicago: Chicago Linguistic Society. 133–141. Dehé, Nicole. 2007. The relation between syntactic and prosodic parenthesis. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam/Philadelphia: Benjamins. 261–284. Dehé, Nicole. 2009. Parentheticals. In Louise Cummings (ed.). The Pragmatics Encyclopedia. London/New York: Routledge. 307–308. Dehé, Nicole. 2014. Parentheticals in spoken English. The syntax-prosody relation. Cambridge: Cambridge University Press. Dehé, Nicole & Yordanka Kavalova (eds.). 2007a. Parentheticals. Amsterdam/Philadelphia: Benjamins. Dehé, Nicole & Yordanka Kavalova. 2007b. Parentheticals: an introduction. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam/Philadelphia: Benjamins. 1–22. Dehé, Nicole & Anne Wichmann. 2010. Sentence-initial I think (that) and I believe (that): Prosodic evidence for use as main clause, comment clause and discourse marker. Studies in Language 34. 36–74. Delais-Roussarie, Elisabeth. 2010. Prosodie incidente et structure prosodique. In Gilles Corminboeuf, Franziska Heyna & Mathieu Avanzi (eds.). Les parenthèses en français (= Verbum 30: 1). Nancy: Presses Universitaires de Nancy. 37–52. Diessel, Holger & Michael Tomasello. 2001. The acquisition of finite complement clauses in English: A corpus-based analysis. Cognitive Linguistics 12. 97–141. Döring, Sandra. 2007. Quieter, faster, lower, and set off by pauses? In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam - Philadelphia: Benjamins. 285–307. Glikman, Julie. 2009. Parataxe et subordination en ancien français. Système syntaxique, variante et variation. Université Paris Ouest/ Universität Potsdam: doctoral dissertation. Glikman, Julie. 2012. Les incises en croire et cuidier en ancien français. In Julie Glikman & Mathieu Avanzi (eds.). Entre rection et incidence: des constructions verbales atypiques? Études sur je crois, je pense et autres parenthétiques (= Linx 61). Nanterre: Université Paris Ouest Nanterre La Défense. 71–85. Glikman, Julie & Mathieu Avanzi (eds.). 2012. Entre rection et incidence: des constructions atypiques? Études sur je crois, je pense et autres parenthétiques (= Linx 61). Nanterre: Université Paris Ouest Nanterre La Défense. Hall, Roland 1958. Assuming: one set of positing words. Philosophical Review 67. 52–75. Hoffmann, Ludger. 1998. Parenthesen. Linguistische Berichte 175. 299–328.

10 | Stefan Schneider, Julie Glikman and Mathieu Avanzi Kaltenböck, Gunther. 2006. Some comments on comment clauses: a semantic classification. In Renata Povolná & Olga Dontcheva-Navratilova (eds.). Discourse and interaction 2. Brno: Masarykova Univerzita. 71–87. Kaltenböck, Gunther. 2007. Spoken parenthetical clauses in English: a taxonomy. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam/Philadelphia: Benjamins. 25–52. Kaltenböck, Gunther. 2008. Prosody and function of English comment clauses. Folia Linguistica 42. 83–134. Kaltenböck, Gunther, Bernd Heine & Tania Kuteva. 2011. On thetical grammar. Studies in Language 35. 852–897. Newmeyer, Frederick. 2010. What conversational English tells us about the nature of grammar: A critique of Thompson’s analysis of object complements. In Kasper Boye & Elisabeth Engberg-Pedersen (eds.). Language usage and language structure. Berlin: Mouton De Gruyter. 3–44. Schneider, Stefan. 2007. Reduced parenthetical clauses as mitigators. A corpus study of spoken French, Italian and Spanish. Amsterdam/Philadelphia: Benjamins. Schneider, Stefan. 2011. Die Entstehung einer neuen Kategorie: Reduzierte parenthetische Teilsätze im Alt-, Mittel- und Neufranzösischen. In Eva Mayerthaler, Claudia E. Pichler & Christian Winkler (eds.). Was grammatische Kategorien miteinander machen. Form und Funktion in romanischen Sprachen von Morphosyntax bis Pragmatik. Festschrift für Ulrich Wandruszka. Tübingen: Narr. 225–244. Schneider, Stefan. 2013. Parenthetische Teilsätze in mittelfranzösischen Texten des 14. und 15. Jahrhunderts. Zeitschrift für romanische Philologie 129. 867–887. Schwyzer, Eduard. 1939. Die Parenthese im engern und weitern Sinne. Abhandlungen der Preußischen Akademie der Wissenschaften. Jahrgang 1939. Philosophisch-historische Klasse. Nr.6. Berlin: Verlag der Akademie der Wissenschaften. Thompson, Sandra A. 2002. Object complements and conversation towards a realistic account. Studies in Language 26. 125–163. Thompson, Sandra A. & Anthony Mulac. 1991a. The discourse conditions for the use of the complementizer that in conversational English. Journal of Pragmatics 15. 237–251. Thompson, Sandra A. & Anthony Mulac. 1991b. A quantitative perspective on the grammaticization of epistemic parentheticals in English. In Elizabeth C. Traugott & Bernd Heine (eds.). Approaches to grammaticalization. Vol. 2: Focus on types of grammatical markers. Amsterdam/Philadelphia: Benjamins. 313–329. Urmson, James O. 1952. Parenthetical verbs. Mind 61. 480–496. Van Bogaert, Julie. 2011. I think and other complement-taking mental predicates: a case of and for constructional grammaticalization. Linguistics 49. 295–332. Waltereit, Richard. 2002. Imperatives, interruption in conversation and the rise of discourse particles: A study of Italian guarda. Linguistics 40. 987–1010. Waltereit, Richard. 2006. The rise of discourse markers in Italian: a specific type of language change. In Kerstin Fischer (ed.). Approaches to discourse particles. Amsterdam: Elsevier. 61–76.

| Section I: Theoretical Issues

Frederick J. Newmeyer

Parentheticals and the grammar of complementation Abstract: This paper argues that sequences of elements that have the superficial structure of main clauses in parenthetical-type sentences really are main clauses. They are not adverbs, orphans, or some other type of independent structure. This result has two interesting consequences. The first is that it reduces the need for a special type of construction that is not integrated into syntax as a whole. The second is that it reinforces the idea of the autonomy of syntax, namely, that syntactic structure has its own algebra, one that is not derivable from phonology, semantics, or language use. Keywords: complementation, displacement, parentheticals, orphan analysis, root clause

|| Frederick J. Newmeyer: Department of Linguistics, University of Washington. Seattle, Washington, USA. Departments of Linguistics, University of British Columbia and Simon Fraser University. Vancouver, BC, Canada.

1 Introduction The principal goal of this paper is a very conservative one, namely to argue that what appear superficially to be main clauses really are main clauses. I focus on what I call ‘apparent main clauses’. In the broadest sense, I mean by that term any subject-verb pair that co-occurs with what looks like it might be an accompanying subordinate clause. Hence the bold-face items in (1) are examples of apparent main clauses: (1) ‘Apparent main clauses’ in boldface: a. I think we should leave as soon as possible. b. We should leave, I think, as soon as possible. c. We should leave as soon as possible, I think.

14 | Frederick J. Newmeyer As is well known, many apparent main clauses have been analyzed as not being in a main clause relationship with what might be considered to be their complement (among the most important analyses are Haegeman 1988/2009; Peterson 1999; Espinal 1991; and Burton-Roberts 1999). Indeed, such is currently the mainstream position. But before proceeding further, I need to point out that terminological differences are a serious impediment to successful communication when it comes to parenthetical-like entities. For example, one might simply describe I think in (1b) or (1c) as a parenthetical, without thereby rejecting the idea that it was syntactically a main clause at one point in its derivation. But for some scholars, labelling something as a ‘parenthetical’ embodies a commitment to an analysis that exempts it from processes or constraints governing nonparentheticals. Let me stress that when I use the term ‘parenthetical’ in this paper, it will be in a purely descriptive sense, namely, one that does not imply a commitment to a particular analysis. In terms of analysis, I contrast two positions: (2) Two positions on apparent main clauses: a. The ‘root analysis’: They are complement-taking clauses at one point in their derivation. b. The non-root (‘rootless’) analysis: They are never complement-taking clauses. In this paper, I argue in favour of the root analysis and against the non-root (or ‘rootless’) analysis. Position (2b) allows for a number of very different ways of looking at apparent main clauses. They are listed in (3a-g): (3) An apparent main clause will be considered ‘rootless’ if it is analyzed as one of the following: a. b. c. d. e. f. g.

syntactically adverbial syntactically subordinate not integrated into the structure fragmentary an ‘orphan’ requiring a special discourse-related node like E base-generated in non-canonical position.

Parentheticals and the grammar of complementation | 15

One could, of course, take the position that sentences like (1a) are structurally ambiguous, manifesting both the root and the rootless structure. While such a position is not unreasonable a priori, I argue below that the published arguments for a rootless structure of such items are very weak. There is a wide range of claims about the syntactic rootlessness of apparent main clauses. A to D present them in order from strongest to weakest: A. All (or almost all) apparent main clauses are rootless. ‘The great majority’ of what have traditionally been analyzed as complement-taking predicates are better analyzed as epistemic/evidential/evaluative (‘e/e/e’) fragments, taking complements that are not grammatically subordinate at all. (Thompson 2002: 136)

B. All apparent main clauses with particular semantic/pragmatic properties are rootless. [Complement-taking predicates are rootless when they] function rather like a certain class of adverbs to orient the hearer aright towards the statements with which they are associated. The ways in which they do this may be roughly indicated as being aids to placing the statements aright against the emotional, social, logical, and evidential background. (Urmson 1952: 491)

C. All apparent main clauses without complementizers are rootless (Thompson & Mulac 1991). (4) a. I think that we should leave. b. I think we should leave. D. All ‘displaced’ parenthetical-like apparent main clauses are rootless. (5) It’s time, I think, to water the lawn. This paper argues that in each of the above cases, the verb in question is, at some point in the analysis, a main-clause verb taking a subordinate clause. In other words, all apparent main clauses are indeed ‘real’ main clauses, syntactically integrated into the overall structure. In most cases I will be able to offer positive evidence for the root analysis. In other cases, I will have to content myself simply with showing that the arguments for the rootless analysis are not very strong. The paper is organized as follows. Section 2 reviews, and attempts to refute, the arguments that apparent main clauses are rootless. Section 3 discusses some broader implications of my results. Section 4 is a brief conclusion.

16 | Frederick J. Newmeyer

2 The arguments for and against a rootless analysis of parentheticals The following four subsections address and attempt to refute positions A, B, C, and D above.

2.1 Refuting the idea that all (or almost all) apparent main clauses are rootless Let us begin by examining the hypothesis that virtually all apparent main clauses are rootless. While this position has rarely if ever been adopted by generative syntacticians, it is widely current among functional and corpus-oriented linguists (Sinclair 1991; Hunston & Francis 2000; Thompson 2002). For example, appealing to semantic and discourse-based evidence, Thompson claimed that only 5% of all seemingly apparent main clauses in her corpus of natural conversation were root clauses. Boye & Harder (2007), however, examined her data and found that she greatly exaggerated the number of predicates with epistemic, evidential, or evaluative readings, which for Thompson was the key piece of support for a rootless analysis. But it would be an interesting exercise, I think, to provide positive linguistic arguments for subordination in the traditional grammatical sense. The first test for subordination in English comes from the use of reciprocal each other as the subject of a tensed verb. Each other never occurs as the subject of a main clause verb, so (6) is impossible: (6)* Each other went to the store together. But each other can occur as the subject of a tensed verb if the clause is subordinate: (7) They were joking that each other were crazy. Hence the that-clause in (7) has to be grammatically subordinate to the main verb joking. Consider now negative polarity items (NPIs). Any and ever are two frequently used NPIs. The (a) sentences of (8-9) are examples with a negative in the same clause, the (b) sentences with the NEG-raising predicate think in a higher clause:

Parentheticals and the grammar of complementation | 17

(8) a. It doesn’t have any deep meaning. b. I don’t think the cat has any allergies. (9) a. It won’t ever go back to being the way it used to be. b. Even I don’t think we’ll ever go back to Iraq. Notice that without the negative element the sentences would be impossible: (10) a. * It has any deep meaning. b. * I think the cat has any allergies. (11) a. * It will ever go back to being the way it used to be. b. * Even I think we’ll ever go back to Iraq. And also notice that NPIs are impossible if they occur in the second of two paratactically linked clauses, even if there is a NEG-raising predicate in the first clause: (12) a. * Do you know what I don’t think?: the cat has any allergies. b. * Here’s what I don’t think: we’ll ever go back to Iraq. In other words, to explain the possibility of the cat has any allergies in (8b) and of we’ll ever go back to Iraq in (9b), one needs to assume that these clauses are subordinate to the predicate think. Finally, consider the mandative subjunctive, that is, uninflected verb forms after predicates like suggest, recommend, insist, demand, and require: (13) a. We should suggest that he speak more clearly. b. I recommend that everyone take a dance class. But in Modern English (whether conversational or literary), we do not find the mandative subjunctive in main clauses: (14) * He be punished for his transgressions. So the only reasonable analysis of the phrase containing the mandative subjunctive is that it is grammatically subordinate. I conclude then, that real subordination exists in English. It is not the case that all or almost all apparent main clauses are rootless.

18 | Frederick J. Newmeyer

2.2 Refuting the idea that all apparent main clauses with particular semantic/pragmatic properties are rootless The position that epistemic, evidential, and evaluative apparent main clauses are rootless is developed by Boye & Harder (2007), which puts forward a series of arguments that such entities are grammaticalized as lacking main clause status. In their view, one important diagnostic for epistemicness is lack of ‘addressability’ or being the main point of the utterance. So for Boye and Harder, structures containing NEG-raised variants are not true main clauses because they supposedly cannot be addressed. Consider (15a) and (15b): (15) a. I think I don’t love her. b. I don’t think I love her. Boye and Harder claim that only if the negation in (15b) is interpreted as not being raised can the content of the higher clause be the main point of the utterance and be addressed by really. But, in fact, both the non-NEG-raised variants and the NEG-raised variants can be addressed by really: (16) A: I think I don’t love her. B: Really? Well make up your mind, do you or don’t you? B’: Really? I’m so sorry to hear that, because you are such a nice couple. (17) A: I don’t think I love her. B: Really? Well make up your mind, do you or don’t you? B’: Really? I’m so sorry to hear that, because you are such a nice couple. If all of the B and B’ responses are felicitous, they have no argument based on NEG-raising for a rootless analysis. The addressability of the statements in the A sentences is independent of whether or not NEG-raising has applied. Boye and Harder’s second argument is based on the distribution of adverbials. They claim that only the semantically epistemic variants of apparent main clauses have adverbial distribution. As one example, they can occur in the same position as epistemic adverbs such as probably: (18) a. The weather is getting better, I think/probably. b. The weather, I think/probably, is getting better. But it is not true that displaced epistemically-interpreted apparent main clauses have the distribution of adverbs. For example, adverbs cannot normally occur

Parentheticals and the grammar of complementation | 19

between a verb and its direct object, while main-clause parentheticals can do so: (19) a. John swallowed, I think, rat poison. b. ? John swallowed probably rat poison. Or contrast the following two sentences: (20) a. We’ll probably win. b. * We’ll I think win. Sentences (20a-b) show that epistemic adverbs and the subject-verb collocations that paraphrase them do not have the same distribution. But a more serious problem with Boye and Harder’s test is the fact that with sufficient pause, adverbs can occur almost anywhere in the sentence. In other words, no test based on adverb position is going to be very convincing. Boye and Harder next claim that only semantically epistemic apparent main clauses allow the addition of a tag-question that relates to the complement clause: (21) I think he fits in very well, doesn’t he? I think that their argument here is that if the apparent main clause is adverbial, then we would expect tags to be formed on what superficially appears to be the embedded clause. The presupposition here is that clauses following epistemic speaker-oriented adverbs allow tags. But do they? Consider sentences (22a-c). (22) a. ? Probably the British team will win, won’t they? b. ? Maybe I’ll be finished by next week, won’t I? c. ? Obviously it won’t rain tomorrow, will it? None of these sound particularly well formed to me, for whatever reason. Yet Boye and Harder predict that they should be grammatical for the same reason that (21) is grammatical — the tag is claimed to refer to the main assertion of the sentence. Given that, the unquestioned well-formedness of sentences like (21) is not particularly good evidence for a rootless analysis of a class of apparent main clauses. Boye and Harder’s next argument has to do with the putative impossibility for epistemic apparent main clauses to allow adverbial modification:

20 | Frederick J. Newmeyer (23) a. * The country is going to the dogs, I never think. b. * The country, I never think, is going to the dogs. But some adverbs modify such predicates without difficulty1: (24) a. The country is going to the dogs, I often/?hardly think. b. The country, I often/hardly think, is going to the dogs. Finally, Boye and Harder point out that epistemically-interpreted apparent main clauses exhibit a more limited range of morphological distinctions and possibilities of syntactic combinations than non-epistemically-interpreted ones. More limited, yes, but not enough to support a radical syntactic analysis. For example, semantically epistemic apparent main clauses can occur in the progressive (25a). And Kate Kearns gives several examples where their subject is in the third person, for example (25b) and (25c): (25) a. I’m thinking that it it’s time to leave, isn’t it? b. The team has improved, he thinks. c. The team, he thinks, has improved.

(Kearns 2007)

Dehé and Wichmann (2010a) give a rather different piece of evidence for an adverbial analysis of semantically epistemic apparent main clauses. Simplifying a little bit, they show that these structures have the prosodic contours of adverbials, and so they argue that they are adverbials grammatically. I do not question their facts, but I do question their conclusions. The basic question is: To what extent do syntactic structure and prosodic structure line up? As long ago as 1968, Chomsky & Halle (1968) in The Sound Pattern of English observed that there is a discrepancy between the output of the syntax and the input to the phonology. For example, the syntax produces a complex left-branching structure like (26a), while the phonological phrasing is as in (26b):

|| 1 I have no explanation to offer for why some, but not all, parentheticals of this type admit adverbial modification. As a descriptive generalization, the more strongly negated the parenthetical clause is, the greater the degree of unacceptability.

Parentheticals and the grammar of complementation | 21

(26) a.

NP NP NP NP NP

POSS N

N ADJ

PLUR AFF

POS

NP NP

N

N POSS

N jɔn

z

sistər z hʌzbənd z

stupid

iti

z

b.                                           NP NP

NP

NP

N

N

N

N POSS ##jɔn # z ##

N POSS sistər z##

N

NP N

PLUR

POSS ADJ AFF

hʌzbənd # z ##+iti # z

Readjustment (Chomsky & Halle 1968; example from Mohanan 1995)

So Chomsky and Halle proposed readjustment rules introducing the juncture symbols, and flattening the highly branching constituent structure. Another good example is (27), where the syntax produces (28a), but the prosodic contours are as in (28b): (27) This is the cat that chased the rat that ate the cheese. (28) a. [This is the cat [that chased the rat [that ate the cheese]]] b. [This is the cat] [that chased the rat] [that ate the cheese]

22 | Frederick J. Newmeyer I do not think that anybody would claim that the surface prosodic bracketing (28b) refutes the idea that at some level structure (27) has a representation as in (28a). In other words, prosody is not a very good guide to syntactic constituency. In examples (26) and (27), one could argue that the demands of pronounceability lead to a prosodic structure that is flatter than the syntactic structure. But semantic and pragmatic factors affect prosody as well. For example, consider (29): (29) You are going to wash the dishes. There is no reason to think that this sentence has more than one syntactic structure. But clearly its intonation pattern, the stress level of individual words, and so on are a function of intended interpretation, whether a simple statement of fact, a threatening command, or even a question. The same argument can be made with respect to sentences like I think we’ll be there tomorrow. If the main point of the utterance is to emphasize the complement, then one would naturally prosodically downgrade the main clause. It does not follow from that fact that I think has ceased to be a main clause.

2.3 Refuting the idea that all apparent main clauses without complementizers are rootless The most comprehensive arguments for the idea that all apparent main clauses without complementizers are rootless are presented in Thompson & Mulac (1991). The claim is that when that is present we have a real complement structure, but without that, I think is an epistemic adverb. In their view, the apparent main clause without that is a ‘grammaticalized’ version of the clause with that, so, once the pattern has been grammaticalized, the original subject and main verb are free to move around (see the following): 

(30)

[I think [that it’s going to rain]] 

(31) a. [[I think] it’s going to rain]         b. [It’s going to rain [I think]] What licenses the deletion of that in this historical sequence of events? Following earlier work by Noonan (1985) and Underhill (1988), they make two claims.

Parentheticals and the grammar of complementation | 23

First, they claim that that is deleted when the subject of the complement clause is the topic of the discourse and that is retained when the subject of the main clause is the topic. Second, they claim that that is deleted when the writer makes or endorses the assertion of the complement clause and that is retained when the writer does not necessarily endorse the assertion, but attributes it to someone else. Thompson and Mulac’s analysis cannot stand, however. For one thing, a huge number of factors interact to determine omission or retention of that, not just the semantic or pragmatic content. The following is a list I have collated of particular claims about the presence or absence of that. (32) The presence or absence of that is affected by (cf. Bolinger 1972; Quirk, Greenbaum, Leech & Svartvik 1985; Biber, Johansson, Leech, Conrad & Finegan 1999; Hawkins 2001; Dor 2005; Kaltenböck 2006; Kearns 2007; Dehé & Wichmann 2010b): a. b. c. d. e. f. g. h. i. j. k.

the type and frequency of the matrix verb the type of the main clause subject (pronominal vs. full noun phrase) the matrix clause pronouns the length, type, and reference of the embedded subject the position and function of the embedded clause the voice of the main clause (active vs. passive) ambiguity avoidance the linear adjacency or not of the matrix verb and that the speech register the ‘truth claim’ (Dor 2005) to the proposition of the embedded clause the rhythmic pattern of the utterance

Several of these have nothing at all to do with the information content of the utterance. Thompson and Mulac refer to ‘grammaticalization’, which, by definition, is a historical process. But there is no historical support for the grammaticalization analysis. For example, Brinton (1996) showed that in Middle English, the most frequent parentheticals (gesse, leve, undertake, ‘guess’, ‘believe’, ‘undertake’) do not occur most frequently with zero complementizers. And Aijmer (1997) noted that there is no reason to think the that-less particle arose through grammaticalization, since the zero complementizer was the unmarked form throughout Old English and Early Middle English. To put the issue concisely, we have the possibility of retaining the complementizer that when the main clause is clearly not the main piece of information

24 | Frederick J. Newmeyer to be conveyed, and the possibility of omitting the complementizer that when the main clause is informationally prominent. In other words, there is no strong argument based on complementizer omission for thinking that apparent main clauses are anything but real main clauses.

2.4 Refuting the idea that all ‘displaced’ apparent main clauses are rootless The strongest arguments against the root analysis involve cases where the apparent main clause is displaced from initial position. The following sentences illustrate: (33) a. Mary, I think, will win the award. b. Mary will win the award, I think. The majority of scholars who work on parentheticals have rejected the idea that I think in these sentences is ever a complement-taking main clause. My goal now is to resurrect the classical root analysis, in which at some point in the derivation we have a structure like (34): (34) I think [Mary will win the award] I take no position on the precise mechanism by which I think ends up in its surface displaced position. In fact, there are at least two possibilities:2 (35) The non-initial position of the main clause is a result of movement of the complement to the left (Ross 1973) (36) The non-initial position of the main clause is a result of movement of the main clause to the right (Emonds 1973; 1976; 1979) Ross gave a dozen or more arguments for the root analysis and the accompanying movement rule. Many of these arguments no longer apply given today’s assumptions about grammatical processes. Others, frankly, were not terribly convincing even in 1973. But the basic intuition behind this analysis has not || 2 Another early position was that parentheticals are generated in the deep structure as sentence adverbials (Jackendoff 1972).

Parentheticals and the grammar of complementation | 25

changed in 40 years. That is, the postposed apparent main clause has selectional restrictions with the remainder of the sentence. For example, if the element set off with comma intonation starts as a main clause, we have an automatic explanation for the ungrammaticality of (37a-b). In (37a), two semantically similar parentheticals would be excluded from occurring with the same complement. And (37b), given a rootless analysis, would need to consider John to have left as a potential independent clause, thereby vitiating the essence of such an analysis. (37) a. *John, I think, is handsome, I believe. b. *John, I think, to have left. (cf. *I think John to have left) In my view, these facts lead to the conclusion that the analysis embodying a root clause with movement is the default, to be abandoned only under overwhelming evidence to the contrary. Let us examine the arguments against a main clause analysis. Some of them are purely structural. For example, de Vries (2007) gives many examples that seem to show that parentheticals are invisible to c-command. He concludes on that basis that a parenthetical is syntactically independent of its host, as is claimed by the rootless analysis. Consider now a few of de Vries’ examples. First, he points out that there is no wh-movement out of a parenthetical: (38) a. He thinks Mary will win the award. b. Who thinks Mary will win the award? (39) a. Mary, he thinks, will win the award. b. * Who does Mary ___ thinks will win the award? De Vries takes the impossibility of sentences like (39b) as evidence that parentheticals are invisible to syntactic operations. But that would be the case only if the landing site for wh-movement c-commanded the extraction site. If the parenthetical is higher in the tree than the COMP position, then the ungrammaticality of (39b) would follow automatically. Movement of the subject of thinks to SPEC,CP would be predicted to be impossible since the filler would not ccommand the gap. In fact, if the parenthetical were invisible to syntactic opera-

26 | Frederick J. Newmeyer tions, as claimed by de Vries, then movement from object position into the initial SPEC, CP should be possible. As (40) shows, it is not:3 (40) a. John, Mary thinks, ate the entire pie. b. *What does John, Mary thinks, ate? By the same reasoning, de Vries argues for the syntactic independence of parentheticals on the basis of the fact that there is no quantifier binding into a parenthetical. So (41) is impossible: (41) *Every guesti, hei said, had just arrived. But again, since Every guest does not c-command he, we would not expect (41) to be possible. On the other hand, de Vries does seem to predict, incorrectly, the possibility of (42): (42) *What did every guest, he said, eat? If parentheticals were truly invisible to syntactic operations, then what would prevent (42)? Let us look at some more examples of the supposed invisibility of parentheticals. The first is from Asher (2000), and is repeated in Dehé & Wichmann (2010a). Note the pattern of possible responses in (43a-d): (43) a. b. c. d.

A: Mary, I think, left early. B: # Do you? B: Did she? A: I think (that) Mary left early. B: Do you? B: Did she? A: The party, Mary assures us, is over. B: #Does she? A: Mary assures us that the party is over. B: Does she?

|| 3 An anonymous referee has pointed out that German has so called V1-parentheticals, as in (ia), and V2-parentheticals, as in (ib,c): (i)

a. Hans hat, glaubt Maria, den Kuchen gegessen. b. Hans hat, so glaubt Maria, den Kuchen gegessen. c. Hans hat, das glaubt Maria, den Kuchen gegessen.

S/he notes that V1-parentheticals do allow wh-question-formation in the host clause (Was hat, glaubt Maria, Hans gegessen?). Since this type corresponds most closely to the English type exemplified with Mary thinks, s/he wonders what the reason for this difference between English and German might be. That is a question for future research.

Parentheticals and the grammar of complementation | 27

One can question directly the main proposition, but not the parenthetical. That again suggests that parentheticals might be syntactically invisible. But such a conclusion is premature. Questions like Do you? and Does she? are uttered in response to assertions. But I think and Mary assures us are not assertions. What are they? In fact, there are numerous analyses of the pragmatics of parentheticals, but it seems useful to think of them as ‘mitigators’, as they are called in the work of Fraser (1980) and Schneider (2007a, b). The basic point here is that whatever their syntactic analysis, displaced main clauses have a semantics and pragmatics that is different from that when they occur in situ as overt main clauses. In other words, certain facts put forward to defend the syntactic rootlessness of parentheticals are a simple function of their different interpretive properties, with no bearing on their syntactic analysis. Because parentheticals have interpretive properties that differ from that of the same elements that are not displaced, it is not sufficient to think that one has refuted the root analysis simply by showing that there are situations where the main clause cannot be used parenthetically. Doing so would presuppose a matchup between formal properties and interpretive ones that has never been justified (for more development of this point, cf. Newmeyer 1998; 2008; 2009). As a case in point, consider Peterson (1999: 235), which calls attention to the paradigm in (44) (this example is also repeated in Dehé & Wichmann 2010a: 3): (44) a. I hope the rain stops before Sunday. b. * The rain stops, I hope, before Sunday. c. I hope the rain will stop before Sunday. d. The rain will stop, I hope, before Sunday. e. * The rain stops before Sunday. Peterson claims that the impossibility of (44b) argues against a movement analysis. But that would be the case only if one assumed that two sentences with the same underlying structure would necessarily have the same interpretation. We have known since Chomsky (1971), however, that such is not the case. The use of the present tense in English in a surface string to express futurity is possible only when the speaker can control the outcome or at least be reasonably certain of the outcome. That explains why (44e) is impossible. In (44b) the parenthetical, while present structurally, does not affect the interpretation of the main clause that it interrupts. So (44b) is impossible for exactly the same reason as (44e), that is, because one has no control over the stopping of the rain.

28 | Frederick J. Newmeyer Let us turn to ellipsis now. McCawley (1988: 763) and Peterson (1999: 234) have argued against the root analysis on the basis of the apparent fact that ellipsis operations act as if the parenthetical is not there. There are actually two very different kinds of ellipsis: argument ellipsis and VP-ellipsis. Let us examine them in turn. We start with argument ellipsis, as in (45): (45) The Hawks will win, says John, by at least 10 points, and Bill says so too. (the interpretation of so is not ‘the Hawks will win, says John, by at least 10 points’) McCawley and Peterson argue that since ellipsis ignores the parenthetical, the parenthetical ‘isn’t really there’ syntactically. But that argument presupposes that ellipsis picks out a constituent in the first clause. In fact, it is well known that such cannot be the case. So consider (46): (46) Mary turned down the post-doc, but it wouldn’t have happened with a real teaching position. What does it refer to? Certainly not to a constituent in the first clause, since Mary turned down, a subject and verb of a transitive sentence, do not form a constituent. With argument ellipsis, the intended antecedent of the pronoun is coerced from the linguistic and non-linguistic context. That explains why in sentences like (45), no hearer would construct a reading including the parenthetical.4 Verb phrase ellipsis is different. The elided material has to pick out a real VP in the antecedent. Consider the next example: (47) I claimed, Pat said, that the exam was unfair. Notice that Pat said breaks up the VP. Now let us see if ellipsis works: (48) ? I claimed, Pat said, that the exam was unfair and Bill did too.

|| 4 An anonymous referee raises the question: ‘Why is it not possible that so refers to a semantic object which includes the meaning of the parenthetical?’. While I have no answer to that question, the fact that it manifestly does not include that meaning is sufficient to establish my point.

Parentheticals and the grammar of complementation | 29

Sentence (48) seems really strange to me. The explanation seems to be that the former main clause Pat said is really there in the structure and prevents VP ellipsis from working. McCawley (1982: 96; 1998: 751) gave a couple of examples that would appear to show that structures with VP antecedents ignore parenthetical material: (49) John talked, it seems, about literature and Mary did too (did is not interpreted to mean ‘Mary talked, it seems, about literature’). (50) John talked to us, it seems, about literature, but Mary would never do that (do that is not interpreted to mean ‘Mary would never talk to us, it seems, about literature’). I find it interesting that the best examples that McCawley could come up with have an agentless parenthetical, namely it seems. If one uses material like I think, he says, and so on, the sentences are horribly bad. In fact, I judge (49) and (50) to be ungrammatical5. Let us look more closely at the claim that a parenthetical is not a constituent of the host clause. The first, and most obvious, question to ask is how that could be, given that the parenthetical is not only pronounced, but it is often pronounced in the middle of the sentence. In fact, there have been several proposals to handle what is sometimes called ‘linearization’, ranging from semantic association (Peterson 1999), to principles of utterance interpretation (Haegeman 1988/2009), to serialization in the phonetic component (Haider 2005). I do not understand what any of these ideas mean concretely. Nobody disputes that there are phonological and semantic relations between the parenthetical and its host. How can material be integrated pragmatically, semantically, and phonologically, but not syntactically, given the interdependence of the various components of the grammar? I have to agree with Mark de Vries when he writes: As a preliminary, notice that any account will have to be truly ‘integrated’. This means that parataxis must be represented in syntax. Proposals that involve fully ‘radical orphanage’ (Haegeman 1988/2009, Burton-Roberts 1999, Peterson 1999, Fabb 1990) or the attachment of parentheses at some grammatical level beyond LF (Safir 1986) can be rejected

|| 5 Clearly, what is needed is an examination of large corpora of spoken and written English to determine if such sentences ever appear. Defining the appropriate search criteria would be more than a minor challenge, however.

30 | Frederick J. Newmeyer out of hand for a simple reason: parentheses, like any other linguistic material, have both sound and meaning. (de Vries 2007: ms 13)

De Vries proposes a minimalist-based analysis, based on his notion of b-merge, which might be broadly compatible with mine. Those opposed to syntactic integration typically try to show that the facts can be explained in terms of processes that work across sentences, and therefore are not part of sentence grammar. So Ziv (1975) and Kaltenböck (2007) call attention to examples like (51a) and (51b): (51) a. The book is not very interesting, I don’t think b. * The book is very interesting, I don’t think. Ziv says that pragmatic principles are involved, rather than syntactic ones, because the grammaticality or not of (51a) and (51b) reflects that of (52a) and (52b): (52) a. The book is not very interesting. I don’t think so. b. * The book is very interesting. I don’t think so. It seems to me that this argument can be turned against itself. The generalization covering both (51) and (52) it that it is pragmatically odd to deny one’s own assertion. Notice that (53) sounds fine: (53) The book, I don’t think, is very interesting. There is no denying of an assertion here. I do not see how one could explain why (51b) is bad and (53) is good, unless the parenthetical is syntactically integrated. After all, they both contain the subsentences The book is very interesting and I don’t think. In a non-integrated analysis they would be predicted to have the same interpretation, at least without the positing of special assumptions. Now look at the contrast in (54a-b): (54) a. The book is very interesting? I don’t think so. b. * The book is very interesting, I don’t think. (with question intonation on the first clause) The analysis that says that parentheticals are syntactically independent of their host clause cannot explain this contrast. It is precisely because in (54b) the two clauses are integrated syntactically that it is impossible turn the first part of (54b) into a question.

Parentheticals and the grammar of complementation | 31

3 Some implications of the root analysis Consider now some of the implications of the root analysis. First, it seems a priori desirable to be able to eliminate a special category of constructions that have their own particular properties. It is a welcome step to exclude the possibility of a sentence containing pronounced elements that are not integrated into syntactic structure. If we can live without orphans and so on, so much the better! But there is a broader issue that needs discussion. There is a consistent theme underlying most of the arguments for syntactic rootlessness that I have criticized. They all presuppose that there is a one-to-one (or close to a one-toone) relationship between syntactic structure and the structure of some other grammatical component. So if the prosody is X, then the syntactic structure has to reflect X. If the semantics is Y, then the syntactic structure has to reflect Y. If the pragmatics is Z, then the syntactic structure has to reflect Z. I think that that is the wrong way to look at things. One of the most difficult problems in linguistics is the precise relationship between syntactic structure on the one hand and semantic structure and the structure of other grammatical components on the other hand. It would be a serious mistake to presuppose a one-to-one mapping. In fact, there is clearly not in general a one-to-one mapping. Consider a couple of examples. First, look at structures with displaced wh-phrases (who, when, where, why, etc.), that is, constructions like in (55-58): Wh-constructions in English: Questions: (55) Who did you see? Relative clauses: (56) the woman who I saw Free relatives: (57) I’ll buy what(ever) you are selling. Wh (pseudo) clefts: (58) What John lost was his keys.

32 | Frederick J. Newmeyer Here we have a structural generalization that does not map smoothly onto a semantic or discourse-functional one. The structures, in relevant respects, are all the same. In each construction type, the displaced wh-word occupies the same structural position, namely, the left margin of its clause (SPEC of CP, in GB terms): (59)

CP SPEC, CP wh-phrase

C´ C

IP t

Despite their structural parallelism, the wh-phrases in the four constructions play very different semantic and discourse-functional roles. In simple whquestions (55), sentence-initial position serves to focus the request for a piece of new information, where the entire clause is presupposed except for a single element. Relative clauses (56) have nothing to do with focusing. The fronted phrase is a topic, not a focus, and merely repeats the information of the head noun woman. Free relatives (57), by definition, have no head noun. The fronted whphrase seems to fulfill the semantic functions of the missing head noun. Pseudo-clefts (58) are different still. The clause in which the wh-word is fronted represents information that the speaker can assume that the hearer is thinking about. But the function of the wh-word itself is not to elicit new information (as is the case with such phrases in questions), but rather to prepare the hearer for the focused (new) information in sentence-final position. In other words, there is a profound mismatch between a formal generalization on the one hand and meaning and function on the other hand. Sometimes one hears that the fronted wh sets up an operator-variable configuration or acts as a scope marker. In the simplest cases, that might be true, as in the simplest wh-questions like (60): (60) What did Mary eat? = for what x, Mary ate x But when one looks at the full range of wh-constructions, there is no support for the idea that the semantic role of the attracting feature is to encode operator-

Parentheticals and the grammar of complementation | 33

variable relations. So in (61), there is no operator-variable configuration corresponding to trace and antecedent: (61) Hillary Clinton, who ten million Americans voted for in the primaries, was not nominated. Or take questions where the wh-phrase itself is embedded in the phrase that moves: (62) a. Which book did you read? b. To whom did you write an angry letter? Which book and To whom are the phrases that move, though they are not the operators. Does the fronted wh-element act as a scope marker? Not reliably. Look at partial wh-Movement in German and Romani (McDaniel 1989): (63) Wasi glaubt [IP Hans [CP mit wem]i [IP Jakob jetzt ti spricht]]]? what think Hans with who Jacob now speak ‘With whom does Hans believe Jacob is now talking? [[IP o Demìri mislinol [CP [kas]i [IP I Arìfa dikhl’a ti]] (64) Soi what that Demir think who the Arif saw ‘Whom does Demir think Arifa saw?’ In other words, there is profound mismatch between a formal generalization and meaning and function. To give another example, consider the adjective with infinitival complement structure: (65) Noun – Copula – Adjective – Infinitive Whatever the underlying structures of sentences corresponding to (65) may be, there are six logically possible ways on the surface that the subject can relate (in terms of understood grammatical relations) to the Adjective and the Verb. All six occur:

34 | Frederick J. Newmeyer

Relation of NP subject to A to V (66) a. b. c. d. e. f.

Mary is eager to please Mary is easy to please Mary is liable to dance Guns are cheap to buy It is useless to try It is too dark to see

subject subject none object none subject subject object none none subject none

Clearly there is something real about pattern (65) itself, whatever uses it might be put to. In other words, the multiple uses to which any syntactic expression can be put argue for formal patterns being ‘embodied’ independently of their interpretations. In short, the mismatch between some apparent main clauses and their semantics and prosody should not be surprising. That sort of mismatch is the norm for language6. To make a short related point, the mismatch between form and other aspects of grammar leads to interesting hypotheses about language change, in particular grammaticalization. Contrast (67a) and (67b): (67) a. It may be that we’ll finish by noon. b. Maybe we’ll finish by noon. Maybe is, historically speaking, a grammaticalization of (It) may be (for discussion, cf. Fischer 2007). Frequency of the use of ‘it may be’ led speakers to utter that collocation with great rapidity. At first, there was no change in syntactic structure. As we have seen, form and meaning are not always in lockstep. But children — far more than adults — want iconicity among grammatical components, as Dan Slobin pointed out many years ago (cf. Slobin 1973). So at a certain point children reanalyzed the sentential structure as the adverb maybe.

|| 6 In mainstream syntactic theory, the sentences of (66a-f) differ enormously from each other in terms of their underlying structures. Each is different from the other with respect to empty categories, control, empty operators, expletives, and types of movement hypothesized. Be that as it may, tests for surface constituency (e.g. preposing possibilities, surface anaphoric processes, etc.) indicate that they are identical in relevant respects at the interface with PF.

Parentheticals and the grammar of complementation | 35

4 Conclusion To conclude, I have argued that sequences of elements that have the superficial structure of main clauses in parenthetical-type sentences really are main clauses. They are not adverbs, orphans, or some other type of independent structure. This result has two interesting consequences. The first is that it becomes possible to reduce the need for a special type of construction, one that is not integrated into syntax as a whole. The second is that it reinforces the idea of the autonomy of syntax, namely, that syntactic structure has its own algebra, one that is not derivable from phonology, semantics, or language use.

References Aijmer, Karin. 1997. I think — an English modal particle. In Toril Swan & Olaf J. Westvik (eds.). Modality in Germanic languages: Historical and comparative perspectives. Berlin: Mouton de Gruyter. 1–47. Asher, Nicholas. 2000. Truth conditional discourse semantics for parentheticals. Journal of Semantics 17.31–50. Biber, Douglas et al. 1999. Longman grammar of spoken and written English. London: Longman. Bolinger, Dwight. 1972. That’s that. Mouton: The Hague. Boye, Kasper & Peter Harder. 2007. Complement-taking predicates: Usage and linguistic structure. Studies in Language 31.569–606. Brinton, Laurel J. 1996. Pragmatic markers in English: Grammaticalization and discourse functions. Berlin: Mouton de Gruyter. Burton-Roberts, Noel. 1999. Language, linear precedence and parentheticals. In Peter Collins & David Lee (eds.). The clause in English. Amsterdam: John Benjamins. 33–52. Chomsky, Noam. 1971. Deep structure, surface structure, and semantic interpretation. In Danny D. Steinberg & Leon A. Jakobovits (eds.). Semantics: An interdisciplinary reader in philosophy, linguistics, and psychology. Cambridge: Cambridge University Press. 183–216. Chomsky, Noam & Morris Halle. 1968. Sound pattern of English. New York: Harper and Row. de Vries, Mark. 2007. Invisible constituents? In Nicole Dehé & Yordanka. Kavalova (eds.). Parentheses as B-merged adverbial phrases. Parentheticals. 203–34. Amsterdam: John Benjamins. Dehé, Nicole & Anne Wichmann. 2010a. The multifunctionality of epistemic parentheticals in discourse: Prosodic cues to the semantic-pragmatic boundary. Functions of Language 17.1–28. Dehé, Nicole & Anne Wichmann. 2010b. Sentence-initial I think (that) and I believe (that) Prosodic evidence for use as main clause, comment clause and discourse marker. Studies in Language 34.36–74.

36 | Frederick J. Newmeyer Dor, Daniel. 2005. Toward a semantic account of that-deletion in English. Linguistics 43.345– 82. Emonds, Joseph E. 1973. Parenthetical clauses. In Claudia Corum, T. Cedric Smith-Stark & Ann Weiser (eds.). You take the high node and I’ll take the low node: Papers from the Comparative Syntax Festival: The differences between main and subordinate clauses. 333–47. Chicago: Chicago Linguistic Society. Emonds, Joseph E. 1976. A transformational approach to English syntax. New York: Academic Press. Emonds, Joseph E. 1979. Appositive relatives have no properties. Linguistic Inquiry 10.211–43. Espinal, M. Teresa. 1991. The representation of disjunct constituents. Language 67. 726–62. Fischer, Olga. 2007. Morphosyntactic change: Functional and formal perspectives. Oxford: Oxford University Press. Fraser, Bruce. 1980. Conversational mitigation. Journal of Pragmatics 4.341–50. Haegeman, Liliane. 1988/2009. Parenthetical adverbs: The radical orphanage approach. In Benjamin Shaer et al. (eds.). Dislocated elements in discourse: Syntactic, semantic, and discourse perspectives. London: Routledge. 331–47. Haider, Hubert. 2005. Parenthesen – Evidenz aus Bindungsverhältnissen. In Franz J. d’Avis (ed). Deutsche Syntax: Empirie und Theorie. Göteborg: Acta Universitatis Gothoburgensis. 281–93. Hawkins, John A. 2001. Why are categories adjacent? Journal of Linguistics 37. 1–34. Hunston, Susan & Gill Francis. 2000. Pattern grammar: A corpus-driven approach to the lexical grammar of English. Amsterdam: John Benjamins. Jackendoff, Ray. 1972. Semantic interpretation in generative grammar. Cambridge, MA: MIT Press. Kaltenböck, Gunther. 2006. ‘…That is the question’: Complementizer omission in extraposed that-clauses. English Language and Linguistics 10. 371–96. Kaltenböck, Gunther. 2007. Spoken parenthetical clauses in English: A taxonomy. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam: John Benjamins.25–52. Kearns, Katherine S. 2007. Epistemic verbs and zero complementizer. English Language and Linguistics 11. 475–505. McCawley, James D. 1982. Parentheticals and discontinuous constituent structure. Linguistic Inquiry 13. 91–106. McCawley, James D. 1988. The syntactic phenomena of English. Chicago: University of Chicago Press. McCawley, James D. 1998. The syntactic phenomena of English, 2nd edition. Chicago: University of Chicago Press. McDaniel, Dana. 1989. Partial and multiple Wh-movement. Natural Language and Linguistic Theory 7. 565–604. Mohanan, Karuvannur P. 1995. The organization of grammar. In John A. Goldsmith (ed.). The handbook of phonological theory. Oxford: Blackwell. 24–69. Newmeyer, Frederick J. 1998. Language form and language function. Cambridge, MA: MIT Press. Newmeyer, Frederick J. 2008. Semantic evidence and syntactic theory. In Samuel Featherstone & Susanne Winkler (eds.). The fruits of empirical linguistics. Volume 1: Process. Berlin: Mouton de Gruyter. 177–210.

Parentheticals and the grammar of complementation | 37

Newmeyer, Frederick J. 2009. On split-CPs and the ‘perfectness’ of language. In Benjamin Shaer et al. (eds.). Dislocated elements in discourse: Syntactic, semantic, and discourse perspectives. London: Routledge. 114–40. Noonan, Michael. 1985. Complementation. In Timothy Shopen (ed). Language typology and syntactic description. Volume 2: Complex constructions. Cambridge: Cambridge University Press. 42–140 Peterson, Peter. 1999. On the boundaries of syntax: Non-syntagmatic relations. In Peter Collins & David Lee (eds.). The clause in English. Amsterdam: John Benjamins. 229–50 Quirk, Randolph et al. 1985. A comprehensive grammar of the English language. Harlow: Longman. Ross, John R. 1973. Slifting. In Maurice Gross, Morris Halle & Marcel-Paul Schützenberger (eds.). The formal analysis of natural language. The Hague: Mouton. 133–72. Schneider, Stefan. 2007a. Reduced parenthetical clauses as mitigators: A corpus study of spoken French, Italian and Spanish. Amsterdam: John Benjamins. Schneider, Stefan. 2007b. Reduced parenthetical clauses in Romance languages: A pragmatic typology. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam: John Benjamins. 237–58. Sinclair, John M. 1991. Corpus, concordance, and collocation. Oxford: Oxford University Press. Slobin, Dan I. 1973. Cognitive prerequisites for the development of grammar. In Charles F. Ferguson & Dan I. Slobin (eds.). Studies in child language development. New York: Academic Press. Thompson, Sandra A. 2002. ‘Object Complements’ and conversation: Towards a realistic account. Studies in Language 26. 125–64. Thompson, Sandra A. & Anthony Mulac. 1991. A quantitative perspective on the grammaticalization of epistemic parentheticals in English. In Elisabeth C. Traugott & Bernd Heine (eds.). Approaches to grammaticalization, Vol. 2: Focus on types of grammatical markers. Amsterdam: John Benjamins. 313–39. Underhill, Robert. 1988. The discourse consitions for that-deletion. Unpublished MS. San Diego: San Diego State University. Urmson, James O. 1952. Parenthetical verbs. Mind 480–496. Ziv, Yael. 1975. On the relevance of content to the form-function correlation (An examination of extraposed relative clauses). In Robin E. Grossman, L. James San & Timothy J. Vance (eds.). Papers from the Parasession on Functionalism. Chicago: Chicago Linguistic Society. 568–79.

Gunther Kaltenböck, Bernd Heine and Tania Kuteva

On theticals: A “rootless” analysis of I think Abstract: Reviewing a wide range of different linguistic approaches, Newmeyer (2012) argues that the English information unit I think in (1) is an “apparent main clause”, that is, a main clause taking a subordinate clause at some point in its analysis. With this hypothesis, which he calls a “root” analysis, he takes issue with alternative approaches, referred by him as “rootless” (or non-root) analyses. According to the latter approaches, I think is classified as a different kind of syntactic constituent or no syntactic constituent at all. (1) a. We should leave, I think, as soon as possible. b. We should leave as soon as possible, I think. The main purpose of the present paper is to argue in favor of a rootless analysis according to which the “apparent main clauses” in (1) are syntactically independent. Using the framework of Discourse Grammar (Kaltenböck et al. 2011; Heine et al. 2013), it is hypothesized that I think in (1) does not constitute an isolated phenomenon but rather that it belongs to a large pool of theticals. Theticals, which include units such as parentheticals, formulae of social exchange, imperatives, vocatives, and interjections, cannot normally be syntactically embedded, are typically marked prosodically in some way, and their meaning is non-restrictive, that is, it is essentially independent from that of the rest of an utterance. 1 Keywords: cooptation, isomorphism, non-restricted meaning, rootless hypothesis, Sentence Grammar, thetical || Gunther Kaltenböck: Institut für Anglistik, Universität Wien. Vienna, Austria. Bernd Heine: Institut für Afrikanistik, Universität zu Köln. Cologne, Germany. Tania Kuteva: Institut für Anglistik und Amerikanistik, Heinrich-Heine-Universität. Düsseldorf, Germany.

|| 1 We are grateful to three anonymous reviewers who provided us with a number of valuable comments on an earlier version of this paper and to the editors for their assistance in the revision process.

40 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva

1 Introduction In a recent paper, Newmeyer (2012: 2) argues that the English unit I think in (1) “is – at some point in the analysis – a main-clause taking a subordinate clause”. (1) a. We should leave, I think, as soon as possible. b. We should leave as soon as possible, I think. Newmeyer (2012: 8) assumes that in the derivation of the sentences in (2b) and (2c) (cf. (1a) and (1b), respectively) there is a structure like (2a). On the surface, the structure of (2b) is something like (3), and there is a derivation of (3) from (2a) via some kind of movement. (2) a. I think [Mary will win the award]. b. Mary, I think, will win the award. c. Mary will win the award, I think. (3) Newmeyer (2012: 8; simplified) CP CP

Mary

CP

I think

VP

will win the award

Thus, Newmeyer’s (2012) stance is that a comment clause such as I think, referred to by him as an “apparent main clause”, is an integral part of (sentence) syntax and, hence, requires a “root analysis”. In his view, then, all examples of I think in (1) and (2) are complement-taking main clauses at one point in their derivation. He contrasts this view with what he calls “rootless analyses”, where I think is never a complement-taking clause.2 Reviewing a wide range of different approaches, Newmeyer (2012: 1) suggests that an apparent main clause is considered rootless if it is analyzed as one of the following:

|| 2 As noted accurately by an anonymous reviewer, Newmeyer’s distinction suggests that there are no intermediate positions. However, usage-based studies such as Diessel and Tomasello (2001) and Boye and Harder (2007) acknowledge both lexical (“root”) and grammatical (“nonroot”) variants of I think.

On theticals: “rootless” analysis of I think | 41 Parentheticals and theAgrammar of complementation

(4) a. b. c. d. e. f. g.

syntactically adverbial syntactically subordinate not integrated into the structure fragmentary an ‘orphan’ requiring a special discourse-related node like E base-generated in non-canonical position.

A root analysis, as proposed by Newmeyer, however raises questions on the nature of I think in (1) or (2b-c), such as the ones in (5). (5) Questions a. Why is I think typically set off prosodically from the rest of the sentence, frequently signaled in writing by means of commas? b. Why does I think exhibit a word order behavior, i.e. positional mobility, not found in any syntactic category, such as main clauses, subordinate clauses, or adverbials? c. Why does I think have the appearance of an elliptic clause, lacking a complement? d. Why does I think resemble information units such as vocatives in being typically set off prosodically in some way, being syntactically detached from the rest of the utterance, and in being similarly variable in its placement behavior, as can be seen in (6)? 3

(6) a. My friend, we should leave as soon as possible. b. We should leave, my friend, as soon as possible. c. We should leave as soon as possible, my friend. It would seem that answers to these questions will be helpful for understanding the status and the nature of “apparent main clauses” (or comment clauses; Brinton 2008) such as I think in (1), and we will look for answers in Section 4.

|| 3 The term information unit is a cover term for any pairing of form-meaning units that can be separated from the remainder of an utterance by means of semantic, syntactic, and/or prosodic criteria – ideally by all three of them. An information unit can be a word, but it can consist as well of a complex collocation of words (Heine et al. 2013).

42 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva For reasons to be outlined there, we will be restricted to the structures illustrated in (1) (or (2b-c)). The aim of the paper is to critically discuss Newmeyer’s arguments for a root analysis and instead propose an alternative, rootless, analysis. The analysis suggested here is best characterized as the one in (4c). As we argue in Kaltenböck et al. (2011) and Heine et al. (2013), there is, in addition to what we call Sentence Grammar (SG), a second domain of Discourse Grammar, called Thetical Grammar (TG), and the unit I think in (1) or (2b-c) is a thetical, that is, it is not part of Sentence Grammar but rather belongs to Thetical Grammar. As we will see in Section 4.1, theticals consist of a wide range of elements, including the “apparent main clauses” of (1) or (2b-c), referred to variously as discourse markers, comment clauses (Brinton 2008), epistemic parentheticals (Thompson and Mulac 1991), epistemic/evidential/evaluative fragments (Thompson 2002), complement-taking predicates (Boye and Harder 2007), etc., but they also include vocatives, such as the ones in (6) and various other syntactic units. On this view, theticals require a rootless analysis in accordance with (4c). We will attempt to substantiate this assumption in Section 4. This paper essentially consists of two parts. The first part, Sections 2 and 3, looks into the “root analysis” proposed by Newmeyer (2012), with Section 2 critically discussing some of his main arguments and Section 3 being devoted to two more general issues, which are also relevant for an understanding of the framework adopted here. The second part, Section 4, presents an alternative approach to the analysis of I think. Section 4.1 briefly summarizes the framework of Discourse Grammar, described in Kaltenböck et al. (2011) and Heine et al. (2013). In Section 4.2 we relate this framework to the status of I think, and in Section 4.3 we briefly look at the relationship between two domains of Discourse Grammar. The final Section 5 summarizes the findings made and points out problems that we were not able to solve. Throughout the paper we make use of corpus examples from a range of different corpora, notably the British component of the International Corpus of English (ICE-GB, Nelson et al. 2002), the Diachronic Corpus of Present Day Spoken English (DCPSE, http://www.ucl.ac.uk/english-usage/projects/dcpse/), the British National Corpus (BNC, Davies 2004), and the Corpus of Historical American English (COHA, Davis 2010).

On theticals: “rootless” of I think | 43 Parentheticals and theAgrammar of analysis complementation

2 Newmeyer’s main arguments: a critical discussion Newmeyer (2012) presents a number of arguments which are intended to support his root analysis and provide evidence against a rootless analysis. In this section we critically examine some of his main points. In Section 2.1 we first turn to his arguments put forward in favor of a root analysis, but which in our view are not incompatible with a rootless analysis. In Section 2.2 we discuss his arguments against a rootless analysis and present counter arguments to them.

2.1 Newmeyer’s arguments in favor of a root analysis Newmeyer offers various arguments to support his root analysis. However, his discussion suggests that some of the evidence produced by him in favor of a root analysis does not seem to pose a problem for the rootless analysis proposed in Section 4. One piece of this evidence concerns the fact that theticals (or parentheticals) are not allowed in specific structures where corresponding Sentence Grammar counterparts can be used. For example, presenting the examples in (7a-d) (= (55a-d) of Newmeyer 2012: 11),4 he is wondering: “How could the grammar forbid (55d) without [there] being a structural relation between the clauses?” (7) Newmeyer (2012: 11, (55a-d)) a. He hopes Mary will win. b . He hopes for Mary to win. c. Mary will win, he hopes. d. * For Mary to win, he hopes. e. * For Mary to win. There are various ways of analyzing the data in (7) and proposing answers to this question. From our perspective it is the following observation that needs to be taken into account. Sentences are autonomous structures and when they are combined with theticals this does not affect their status. Accordingly, sentences (7a-c) are well-formed, irrespective of whether or not there is a thetical. But (7e)

|| 4 We have added (7e), which is not included in Newmeyer’s paradigm.

44 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva is not an autonomous, well-formed sentence and, accordingly, (7d) cannot be well-formed either. The situation is different in the case of (7c): Since Mary will win would be a well-formed utterance, so is (7c), where a thetical (comment clause) has been added. Thus, the fact that the grammar allows (7c) but forbids (7d) can be predicted on account of the status of theticals, such as he hopes, as syntactically independent information units added to independent host sentences (see Section 4). It would seem that the same observation applies to other examples volunteered by Newmeyer. A second case, based on the data in (2a-b), is presented by Newmeyer (2012: 8–9) to argue on theory-internal grounds that “movement from object position into the initial SPEC, CP should be possible” if the thetical (“parenthetical”) were invisible to syntactic operations. (8) Newmeyer (2012: 9, (40a-b). (40c-e) and bold print, marking theticals, are ours) a. John, Mary thinks, ate the entire pie. b. * What does John, Mary thinks, ate? c. * What does John ate? d. What did John, Mary thinks, eat? e. What did John eat? Both (8a) and (8b) contain the thetical Mary thinks, but (8b) does not appear to be acceptable – for the same reason as in the preceding example: Since the sentence is also unacceptable without this thetical, as (8c) shows, we predict that (8b) must be unacceptable as well: Adding a thetical does not turn an unacceptable sentence into an acceptable one. The relevant question to be asked here – instead – is if (8d) is acceptable. The answer to this question is positive for the same reason: the thetical is added to a host clause which is grammatically correct (cf. 8e). Hence the whole sentence is also acceptable. A third example adduced by Newmeyer (2012: 9), adopted from Peterson (1999: 235; cf. also Dehé and Wichmann 2010b: 3) is the following, based on the data presented in (9). (9) Peterson (1999: 235, (17); bold print is ours) a. I hope the rain stops before Sunday. b. * The rain stops, I hope, before Sunday. c. I hope the rain will stop before Sunday. d. The rain will stop, I hope, before Sunday. e. * The rain stops before Sunday.

On theticals: “rootless” of I think | 45 Parentheticals and theAgrammar of analysis complementation

Both Peterson (1999: 235) and Newmeyer (2012: 9) provide semantic evidence to account for the unacceptability of (9b) and (9e) (which contrast with the acceptability of (9d)) even though they draw contrasting conclusions from this evidence. In the same way as with the preceding example, we are restricted here to a syntactic observation: Since the sentence The rain stops before Sunday is unacceptable as a sentence in (9e) it is also unacceptable when a thetical is added, as in (9b). To conclude, we follow Peterson in arguing that there is “no syntagmatic relationship” between the thetical and its SG host sentence. Hence, we see no justification for a root analysis in any of the three examples discussed in this section: Other than taking an information unit of Sentence Grammar as a host, the thetical does not appear to be syntactically related to its host.

2.2 Newmeyer’s arguments against a rootless analysis In this section we look at those arguments by Newmeyer which are potentially incompatible with our analysis of theticals as syntactically unintegrated elements. Our concern is, first of all, with Newmeyer’s arguments against a rootless analysis of specifically non-initial (in his words ‘displaced’) theticals, that is, comment clauses in the sense of Brinton (2008).

2.2.1 Do selectional restrictions apply? One of Newmeyer’s main arguments in favor of a root analysis is that non-initial theticals such as I think have “selectional restrictions with the remainder of the sentence” (Newmeyer 2012: 8). To illustrate this he gives the examples in (10) (= Newmeyer’s ex. 37), which, in his view, are ungrammatical because they violate selectional restrictions. (10) a. * John, I think, is handsome, I think. b. * John, I think, to have left. (Cf. * I think John to have left) It is not hard, however, to find examples of the structure illustrated in (10a) in actual corpus data, such as the ones in (11). The repetition of I think, in fact, attests to its syntactic independence and semantic non-restrictiveness (see Section 4.1): I think fulfills important discourse organizational functions (Kärkkäinen 2003; Kaltenböck 2010a).

46 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva (11) a. I mean I do know that it does take up time and I do feel that I have to be there for every activity that I set up and I think that’s probably right I think(ICE-GB:s1b-078-214) b. But I think you’ll find it’s it’s a bit tight I think(ICE-GB:s1a-022273) And the ungrammaticality of (10b) can also be explained without recourse to selectional restrictions. As argued in the previous section, (10b) is ungrammatical owing to the ungrammaticality of its host clause (*John to have left), i.e. without thetical.

2.2.2 Not invisible to syntactic operations? Newmeyer also argues against the widely held opinion that theticals, as extraclausal elements, are invisible to syntactic operations (e.g. McCawley 1982, 1998; Peterson 1999; Jackendoff 1977; Emonds 1973; 1976; 1979, Fabb 1990; Espinal 1991; Haegeman 1991; Burton-Roberts 1999; De Vries 2007). To this end he refers to a number of structural tests and operations, which will now be looked at in turn, viz. (i) c-command, (ii) the question test, (iii) the behaviour of negative theticals, and (iv) VP ellipsis. Note that these do not exhaust the list of tests that have been proposed; for a catalog of ten syntactic tests for invisibility, cf. De Vries (2007: 208). (i) Newmeyer (2012: 8) first takes up De Vries’ (2007) claim that theticals are invisible to c-command and argues that this assumption incorrectly predicts that a sentence such as (12) (= Newmeyer’s ex. 42) is possible. (12) * What did every guest, he said, eat? (12) is in fact problematic, but its acceptability can easily be improved by changing the thetical to a pragmatically more plausible one (i.e. more compatible with the question asked by the host clause), as in (13a), and even more so by adding extra weight to the sentence-final predicate, as in (13b) (in accordance with the end-weight principle). (13) a. What did every guest, he wondered / would you say, eat? b. What did every guest, he wondered, have for lunch?

On theticals: “rootless” analysis of I think | 47 Parentheticals and theAgrammar of complementation

(ii) On the other hand, Newmeyer (2012: 9) criticizes the use of tests such as the question test (e.g. Huddleston & Pullum, 2002: 896; Asher, 2000: 33; Dehé and Wichmann 2010a), which is intended to show that questioning produces different results for complement-taking predicates in initial and medial position, as illustrated in (14) (= Newmeyer’s ex. 43). (14) a. b. c. d.

A: Mary, I think, left early. A: I think (that) Mary left early. A: The party, Mary assures us, is over. A: Mary assures us that the party is over.

B: #Do you? B: Did she? B: Do you? B: Did she? B: #Does she? B: Does she?

Newmeyer (2012: 9) dismisses this test on the grounds that “questions like Do you?, Does she? are uttered in response to assertions. I think and Mary assures us are not assertions, they are ‘mitigators’...”. Newmeyer’s mitigator argument is, however, weakened by the fact that an epistemic/mitigator function has also been identified for I think in initial position (e.g. Thompson 2002; Kärkkäinen 2003; Schneider 2007). Nonetheless, we agree with Newmeyer that results obtained from structural tests such as these are generally not very conclusive. Compare, for instance, theticals in final position, which – unlike their medial counterparts in (14) – allow questioning, as in (15).5 The question test could thus be used to argue for equal syntactic status of initial and final complementtaking predicates (CTPs). (15) a. A: Mary left early, I think. b. A: The party is over, Mary assures us.

B: Do you? B: Does she?

This fact can be explained in terms of the difference in degree of theticality, whereby medial placement of theticals appears to be indicative of the highest degree of theticality: since interrupting an utterance in the middle is the most unnatural one (in terms of cognitive processing), we would expect it to be done mainly by well-established, i.e. prototypical theticals. On the other hand, both initial and final position of a thetical represent more natural points of insertion (i.e. easier to process) into the flow of information encoded by the host sentence.

|| 5 As noted by a reviewer, such questioning may be possible on paper but is not necessarily natural, even more so with a negated main clause: Mary didn’t leave, I think. - #Do you?. Tests such as these therefore have to be taken with some caution, as will be discussed in Section 3.1.

48 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva The situation is further complicated if we compare initial comment clauses with and without that-complementizer. The question test has been used to show that questioning of the comment clause is possible with an explicit complementizer, as in (16a), but is problematic without a complementizer, as in (16b) (cf. Huddleston & Pullum, 2002: 896; Asher, 2000: 33). (16) a. A: I believe that John is in Paris. B: Really. Do you? b. A: I believe John is in Paris. B: ?Really. Do you? Clearly, the results of these tests have to be taken with care. Much seems to depend on the type of predicate used and its relative degree of grammaticalization (and more narrowly, desemanticization). Compare for instance believe, which is well-known for its semantic ambivalence (cf. Hooper 1975: 100–101; Quirk et al., 1985: 1113; Huddleston & Pullum, 2002: 896), with more fully grammaticalized predicates such as I think. Moreover, acceptability of the test sentences depends to a large extent on context and prosodic delivery. (iii) Newmeyer (2012: 11) also refers to the behavior of negative theticals, as in the examples in (17) (= Newmeyer’s ex. 51). Ziv (1985: 187) originally used these sentences to argue that the difference in grammaticality can be attributed to independently motivated pragmatic principles (rather than intra-sentential syntactic constraints), as the same restriction applies across sentence boundaries with non-parenthetical negative verbs, viz. The book is not very interesting. I don’t think so vs. *The book is very interesting. I don’t think so (cf. also Kaltenböck 2007: 37). (17) a. The book is not very interesting, I don’t think. b. * The book is very interesting, I don’t think. Newmeyer (2012: 11) rightly criticizes the pragmatic plausibility of denying one’s own assertion in (17b) (and the corresponding non-parenthetical I don’t think so), pointing out that the sentence in (18) is fine “since there is no denying of an assertion”. (18) The book, I don’t think, is very interesting. The sentence in (18) is, however, not directly comparable to the ones in (17), precisely for pragmatic reasons: Although syntactically negative, I don’t think in (17a) is pragmatically positive, reinforcing the (negative) meaning of the host

On theticals: “rootless” of I think | 49 Parentheticals and theAgrammar of analysis complementation

clause (just as I think would), while in (18) it is pragmatically negative, negating the (positive) meaning of the host clause. Newmeyer further argues that an analysis of theticals as syntactically independent of their host clause cannot explain the difference between (19a) and (19b) (= Newmeyer’s ex. 54). (19) a. The book is very interesting? I don’t think so. b. * The book is very interesting, I don’t think. (With question intonation on the first clause) While for Newmeyer the unacceptability of (19b) is the result of the two clauses being syntactically integrated, it can equally well be attributed to pragmatic reasons, as it seems odd to ask a question and answer it within the same sentence. To conclude, what counts is not so much the syntactic form but the pragmatic function of the thetical unit. This has also been pointed out by other authors. As Knowles (1980: 388) shows, for example, syntactically negative theticals may be acceptable in combination with a positive host, provided the semantic ‘sum’ of the thetical is positive, e.g. The book is very interesting, I don’t deny. (cf. also Kaltenböck 2007: 37–38). (iv) Another one of Newmeyer’s (2012: 10) arguments concerns VP ellipsis, which, as he points out, has to pick out a real VP in the antecedent (unlike argument ellipsis). Contrary to McCawley (1982: 96; 1998: 751) and Peterson (1999: 234), Newmeyer argues that parentheticals are not really invisible to VP ellipsis, and he questions the acceptability of examples such as in (20) and (21) (= Newmeyer’s ex. 49, 50). For Newmeyer (ibid.) they are at best marginal and “[if] one uses material like I think, he says, and so on, the sentences are horribly bad”. (20) John talked, it seems, about literature and Mary did too. (did is not interpreted to mean ‘Mary talked it seems about literature’) (21) John talked to us, it seems, about literature, but Mary would never do that. (do that is not interpreted to mean ‘Mary would never talk to us, it seems, about literature’) The problem with acceptability judgments is that they may vary considerably for different speakers (cf. section 3.1) and there are many for whom a sentence such as (22) is fine:

50 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva 50 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva

(22) John talked, I think, about literature and Mary did too. (22) John talked, I think, about literature and Mary did too. Clearly, the position of I think between a verb and PP modifier is amply attested Clearly, position of I think between a verb andand PP modifier attested in corpusthe data, as shown by the examples in (23), it is hardistoamply see why ellipin asshould shownnot by the examples inthe (23), and it isinhard siscorpus in thesedata, cases be possible (cf. addition 23b) to see why ellipsis in these cases should not be possible (cf. the addition in 23b) (23) a. Something brief should be said, I think, about Haddam, where I’ve lived (23) a. Something briefyears should said,live I think, about Haddam, where I’ve lived these fourteen andbe could forever. these fourteen years and could live forever. (COHA:1986:FIC:Sportswriter) b. (COHA:1986:FIC:Sportswriter) you asked, I think, about the, about the Bridgit side of it […and John did b. you I think, about the, about the Bridgit side of it […and John did too] asked, (BNC:K71:S_speech_unscripted) too] (BNC:K71:S_speech_unscripted) Similarly, Newmeyer (2012: 10) argues that a sentence such as (24a) (= NewmeySimilarly, 10)said argues thatup a sentence suchit as (24a) (= Newmeyer’s ex. 48)Newmeyer is strange (2012: as “Pat breaks the VP”. But can be improved by er’s ex.a 48) is strange as “Pat said upwhich the VP”. But the it can be improved using different parenthetical, as breaks in (24b), makes sentence easier by to using a different parenthetical, as in (24b), which makes the sentence easier to process. process. (24) a. ? I claimed, Pat said, that the exam was unfair and Bill did too. (24) a. I claimed, PatI said, b. ? She claimed, think,that thatthe theexam examwas wasunfair unfairand andBill Billdid didtoo. too. b. She claimed, I think, that the exam was unfair and Bill did too. Compare also the following corpus examples where adding an ellipsis does not Compare the following corpus examples where adding an ellipsis does not seem to bealso a problem: seem to be a problem: (25) a. Tibetans believe, he said, that human beings can become perfect on this (25) a. earth Tibetans believe, said, that beings can become perfect on this […and otherhe Buddhists dohuman too] (COHA:1992:NEWS:SanFran) earthgovernment […and otherhas Buddhists (COHA:1992:NEWS:SanFran) b. The decided,do hetoo] said, that social spending can only in b. The government has decided, he said, that social spending can only in crease […and parliament did too] (COHA:1992:NEWS:WashPost) crease […and parliament did too] (COHA:1992:NEWS:WashPost) The same is true for examples where the parenthetical intervenes between a The true examples where parenthetical between a mainsame verb is and its for NP object, as in (26), the which nonethelessintervenes allow for VP ellipsis main verb and its NP object, as in (26), which nonetheless allow for VP ellipsis (cf. also Peterson 1999: 234). (cf. also Peterson 1999: 234). (26) a. A: John swallowed, I think, rat poison [= Newmeyer’s ex. (19a)] (26) a. B: A: John swallowed, I think, poison [=to Newmeyer’s ex. (19a)]I think rat No, he didn’t (did is not rat interpreted mean ‘swallowed B: No, he didn’t (did is not interpreted to mean ‘swallowed I think rat poison’) poison’) b. A: And we will see I think the introduction of a new tax next year b. (adapted A: And we willICE-GB:s2a-003-088) see I think the introduction of a new tax next year from (adapted from ICE-GB:s2a-003-088)

On theticals: “rootless” analysis of I think | 51 Parentheticals and theAgrammar of complementation

B: No, we won’t (will is not interpreted to mean ‘see I think the introduction of a new tax’) The above examples show that Newmeyer’s dismissal of the VP ellipsis test does not seem to be entirely justified and, more generally, that acceptability judgments have to be taken with care. We will look at the question of acceptability test in more detail in the following section.

3 General issues After examining specific arguments used by Newmeyer, let us now turn to some more general issues that arise in the discussion of the two analyses. The first of these concerns the type of evidence used in support of a particular theoretical position (Section 3.1), the second concerns the question of possible isomorphism between different components of language structure (Section 3.2).

3.1 On evidence As may have become apparent in the preceding sections there are a number of points where we favor a theoretical stance that differs from that of Newmeyer (2012). In the present section we wish to draw attention to one of these points which concerns the kind of evidence to be employed for decisions on linguistic taxonomy. In the subsequent Section 3.2 we will then deal with the question of whether there are systematic structural correlations between different components of grammar. Newmeyer is restricted to syntactic analysis, even if he occasionally draws also on semantic and pragmatic evidence to substantiate his root hypothesis.6 The procedure he adopts is the one most widely employed in mainstream linguistics, namely drawing on the distinction between grammatically acceptable vs. non-acceptable sentences, to define discontinuities of linguistic categorization. That this distinction can at times be problematic has been demonstrated in a large body of works. Suffice it here to draw attention to the following: As more recent corpus-based studies suggest, the boundary between what are accepta|| 6 Concerning the latter, cf. e.g. his discussion (Section 2.2.2) of mitigators in the interpretation of grammatical acceptability.

52 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva ble and non-acceptable utterances is highly fuzzy, and to some extent controversial (e.g. Schütze 1996: 98–169, Keller 1998: 4–8, and references therein). It would seem that this observation has a bearing on how to evaluate the evidence adduced by Newmeyer in favor of the root analysis. This can be shown with a few examples taken from Newmeyer’s (2012: 3–5) argumentation to refute the “rootless” view of Boye and Harder (2007) according to which all evaluative/epistemic complement-taking predicates (CTPs) are rootless.7 One of the arguments adduced by Boye and Harder is based on the positional distribution of adverbials, which they claim is the same as the one for epistemic CTP clauses. While we share Newmeyer’s problems with this claim (see our discussion of frankly in 3.4 below), we have reservations accepting his claim, illustrated in (27), that comment clauses (“apparent main clauses”) like I think can occur between a verb and its direct object whereas adverbs cannot. Conversely, he argues that adverbs but not comment clauses can occur in the environment of (28). (27) a. John swallowed, I think, rat poison. (= Newmeyer’s ex. 19). b. ? John swallowed probably rat poison. (28) a. We’ll probably win. (= Newmeyer’s ex. 20) b. * We’ll I think win. A quick corpus search suggests, however, that Newmeyer’s generalization must be taken with care, as text examples such as (29) (as against (27b)) and (30) (as against (28b)) show. (29) a. I would have said probably Radio Brittany or Radio Rouane. (BNC:G5J:S_interview) b. Both of her hands waved out now, indicating, maybe, defeat. (COHA:1971:FIC:DreamMuseum) (30) a. And that role will, I think, soothe Lydia’s own excitable nature. (COHA:1998:FIC:FantasySciFi) || 7 The classification of Boye and Harder’s (2007) analysis as “rootless” is Newmeyer’s. As pointed out by a reviewer, this does not really do justice to the position taken by Boye and Harder, who acknowledge three different types of complement-taking predicates: (i) lexicalpotentially primary, (ii) lexical+invariably secondary, (iii) grammatical+conventionally secondary. The lexical type can be considered a “root” variant, while the grammatical can be considered “non-root”.

On theticals: “rootless” analysis of I think | 53 Parentheticals and theAgrammar of complementation

b. What I have to say about the case of the Muslims will, I think, have some relevance for all moral individuals in our times. (COHA:1998:NF:RethinkingWorld) Another argument by Boye and Harder (2007: 579–80) is that only the grammatical variants of complement-taking predicates, such as (31a), allow the addition of a tag-question relating to the complement clause, whereas the more lexical variants, such as (31b), do not allow it. (31) a. I think he fits in very well, doesn’t he? (Newmeyer’s ex. 21) b. * I regret I killed her, didn’t I? (= Boye and Harder 2007, ex. 14) Newmeyer does not contest this particular difference but draws a parallel to adverbs, arguing that “clauses with epistemic speaker-oriented adverbs do not allow tags”. His argument thus seems to run as follows: Given the distributional and semantic parallel between adverbs and grammatical CTPs (cf. Boye and Harder’s argument above), sentences such as in (32), which Newmeyer marks as questionable, should be acceptable. (32) a. ? Probably the British team will win, won’t they? (= Newmeyer’s ex. 22) b. ? Maybe I’ll finish by next week, won’t I? c. ? Obviously it won’t rain tomorrow, will it? Contrary to his assumption, however, such combinations of sentence adverb and tag question are not at all problematic, as attested by numerous corpus examples such as the following: (33) a. Obviously David I’m not going to give him any of the personal stuff am I.(ICE-GB:s1a-092-323) b. Well clearly what we have here is a a very subjective statement isn’t it. (DCPSE:DL-D02-063) c. But surely from the point of view of the farmer it’s it’s all to do with the hard ecu and and the hard facts of of driving tractors across large fields isn’t it. (ICE-GB:s1b-037-021) To conclude, while (non-)grammaticality judgements no doubt provide an important source of analysis, they cannot necessarily be taken to be strong evidence for or against a given analysis when there is contradicting evidence from actually spoken or written discourse data. And this applies especially to judge-

54 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva ments on theticals (or parentheticals) which are not rooted in the syntactic structure of a sentence but rather in the situation of discourse, as we will argue in Section 4.

3.2 On isomorphism An important point addressed by Newmeyer (2012: 11–13) concerns isomorphism between different components of language structure. He observes that most arguments that have been adduced for syntactic rootlessness “presuppose a close to one-to-one relationship between syntactic structure and the structure of some other grammatical component”. On the basis of a range of examples he demonstrates that it would be a mistake to assume that there is such a one-toone relationship. We agree, but the examples he adduces to substantiate this assumption concern essentially only Sentence Grammar material. The situation is different when theticals are involved (see Section 4). We may illustrate this with the following example. In (34a), the item frankly is an adverb of Sentence Grammar modifying the meaning of the predicate, that is, it is integrated in the sentence syntactically, semantically, and prosodically. In (34b), by contrast, it is a thetical, called a stance adverbial by Biber et al. (1999: 133), a sentence adverb by Brinton and Traugott (2005: 139), or a disjunct by Quirk et al. (1985: 648, 613). (34) a. She spoke frankly about herself now and then. b. Frankly, Kris didn’t want to know. (Biber et al. 1999: 132) When used as a thetical, as in (34b), it would seem in fact that there is “isomorphism” involving syntax, phonetics (prosody), and semantics. First, frankly is syntactically detached. It is not licensed by the syntax of the sentence and is positionally more mobile than that in (34a), or any other adverb. For example, as an adverb in (34a) frankly does not allow the alternative placements illustrated in (35), while frankly of (34b) readily does, cf. (36); as Quirk et al. (1985: 627) observe, disjuncts such as frankly “can appear at almost any place in clause structure.”8

|| 8 One has to be aware, however, that even SG adverbs generally allow for some positional flexibility, as a corpus search may demonstrate. It also has to be pointed out that theticals – despite their positional flexibility – usually have certain preferred ‘niches’ for insertion. These

On theticals: “rootless” analysis of I think | 55 Parentheticals and theAgrammar of complementation

(35) a. * Frankly she spoke about herself now and then. b. * She frankly spoke about herself now and then. (Biber et al. 1999: 132) (36) a. Kris, frankly, didn’t want to know. b. Kris didn’t want to know, frankly. (Biber et al. 1999: 133) c. I don’t frankly know. (Quirk et al. 1985: 628) Second, as a thetical (or disjunct), frankly is typically set off prosodically, it is likely to be separated by a pause and an independent intonation contour, being marked off by commas in writing. And third, its meaning is non-restrictive: Rather than determining the meaning of the predicate, disjuncts such as frankly in (34b) “have a scope that extends over the sentence as a whole” (Quirk et al. 1985: 613). Finally, frankly in (34a), but not in (34b), can be elicited by a question form of the type ‘How did S V?’ (where S = subject and V = verb; Biber et al. 1999: 133).9 To conclude, it would seem that in the case of theticals such as frankly in (34b) there is a one-to-one mapping between syntax, semantics, and prosody – in accordance with our definition in (39) below. That this is not an isolated case of “isomorphism” can be illustrated with another kind of conceptual thetical serving an appositional function, taken from the Comprehensive Grammar of English by Quirk et al. (1985: 1306–7). In (37a), the information unit a very good teacher is an object complement, that is,

|| are typically ‘weak spots’ in the syntax of the host construction, such as between subject NP and VP. 9 One of the reviewers notes that frankly in a sentence such as I object to his hearty and, frankly, crude behaviour differs from the other examples in that it has scope only over the adjective crude, not the sentence as a whole (cf. Boye’s 2012: 250–257 propositional modifiers occurring without a fully-fledged proposition in scope; also Kaltenböck’s 2008 phrasal scope comment clauses). While this is certainly the case, frankly here is still no SG modifier of the adjective crude as it has semantically non-restrictive meaning. The difference becomes clearer by comparing the following two uses of frankly (Quirk et al. 1985: 647), where (i) is a SG modifier (adjunct) and (ii) a thetical (disjunct): (i) His frankly admitted mistakes were pardoned. (ii) His frankly extraordinary attitude dismayed his friends. Unlike (i), the use of frankly in (ii) is syntactically and prosodically independent and semantically non-restrictive and therefore qualifies as thetical in accordance with our criteria in (39) below, despite its restricted scope over the following adjective. In terms of discourse function, frankly in (ii) expresses the attitude of the speaker, whereas in (i) it is ascribed to the subject of the sentence.

56 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva it is licensed by the syntax of the sentence. In (37b), by contrast, it is syntactically independent from the rest of the utterance: Being neither an argument nor an adjunct, it is technically known as a non-restrictive appositive. But it is also prosodically independent from the rest of the sentence. Thus, the information unit a very good teacher is part of the prosody of the sentence in (37a) while in (37b) it is separated from the preceding clause by a tone unit boundary in spoken English and by a comma in written English (Quirk et al. 1985: 1307). And third, there is also a difference in meaning. Whereas the meaning of a very good teacher in (37a) is determined by its syntactic function as a complement of the sentence, in (37b) it is fairly independent from the sentence meaning, that is, it is non-restrictive: It can be understood e.g. as extra information, an elaboration on one participant of the utterance. Thus once more, there is “isomorphism” between syntactic, prosodic, and semantic structure. (37) a. They considered Miss Hartley a very good teacher. b. They considered Miss Hartley, a very good teacher. And much the same observation can be made when other thetical categories are involved, such as the vocative unit my friend that we discussed in (6): The unit is syntactically and prosodically set off from the rest of the sentence and its meaning does not relate to the structure of the sentence but rather to the situation of discourse – more precisely to speaker-hearer interaction (see (40) below). To conclude, we concur with Newmeyer (2012) that syntactic, semantic, and prosodic boundaries need not coincide in clausal, phrasal, or other units. But in the case of theticals there is a high degree of a close to one-to-one relationship between different components of grammar that we take to be diagnostic of the structural independence of theticals.

4 An alternative approach A paradigm strategy of linguistic description is to compare the structure to be analyzed with other structures assumed to be related in some way or other. The way this strategy is applied determines to quite some extent the results obtained. Newmeyer (2012) relates the analysis of “apparent” main clauses such as I think first and above to “actual” main clauses, that is, to a structure where I think forms the matrix clause of a main clause-complement clause construction.

On theticals: “rootless” of I think | 57 Parentheticals and theAgrammar of analysis complementation

It therefore may not be surprising that, like many others before him, he comes to the conclusion that the two are syntactically related. This is not the procedure adopted here. Rather than looking at I think from a syntactic perspective, we are interested in exploring its role and functions as a recurrent piece (or fragment) of discourse, be that spoken or written discourse. And rather than comparing thetical I think, as in (1b-c), with its homophonous counterpart serving as a matrix clause we argue that there is a more plausible alternative. According to this alternative perspective, thetical I think has – despite its indubitable etymological relationship to matrix-clause I think – functionally more in common with information units such as vocatives, as we observed in (5d). In the present section we describe this alternative perspective in more detail. It will be argued that “apparent main clauses” such as I think in (1) should be analyzed as belonging to a separate domain of grammar, called Thetical Grammar. To this end, we deal with this perspective in Section 4.1 by The-tical drawing on the framework of Discourse Grammar. Section 4.2 relates this framework to such “apparent main clauses”, and Section 4.3 is concerned with the issue of interaction between different domains of Discourse Grammar.

4.1 Discourse Grammar As proposed by Kaltenböck et al. (2011) and Heine et al. (2013), Discourse Grammar is composed of all the linguistic resources that are available for constructing spoken or written (or signed) texts; an outline of its architecture is provided in Figure 1.10 Discourse Grammar is based on the assumption that there are two domains of discourse organization that need to be distinguished, referred to respectively as Sentence Grammar (SG) and Thetical Grammar (TG). SG is well documented; it has been the main or the only subject of theories of mainstream linguistics. It is organized in terms of parts of speech or constituent types such as sentences, clauses, phrases, words, and morphemes, plus the syntactic and morphological machinery to relate constituents to one another. || 10 The figure raises a number of questions, in particular the following: What justification is there for assigning SG and TG to the same domain, namely to Discourse Grammar? Are theticals different enough from SG units to be excluded from the domain of SG? Are theticals similar enough to one another to justify their analysis as a domain of their own? Is the inventory of categories distinguished exhaustive? What is the nature of the boundaries of categories: Are they discrete or gradient? How to identify theticals in isolation? The reader is referred to Heine et al. (2013, Section 4), where these questions are discussed.

58 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva 58 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva

Discourse Grammar

Sentence Grammar

Conceptual theticals

Thetical Grammar

Vocatives

…

Imperatives Formulae of

Interjections

…

social exchange Fig. 1: A sketch of Discourse Grammar. Fig. 1: A sketch of Discourse Grammar.

TG consists of a catalog of thetical formulae and constructions as well as the TG consists of a catalog of thetical formulae and constructions as well as the ability to design new theticals and to deploy them for structuring discourse. The ability to design new theticals and to deploy them for structuring discourse. The main categories of theticals distinguished so far are illustrated in (38) (like main categories of theticals distinguished so far are illustrated in (38) (like 11 elsewhere, theticalsare areprinted printedininbold). bold). 11 elsewhere, theticals (38) ofThetical TheticalGrammar Grammar(Kaltenböck (Kaltenböck 2011; Heine (38) Categories Categories of etet al.al. 2011; Heine andand Kaltenböck 2012) Kaltenböck 2012) a. He was aa man friends. thetical a. manwho, who,unaccountably, unaccountably,had hadfew few friends. Conceptual Conceptual thetical b. Good morning! Formula of social b. morning! Formula of social exchange exchange 12 c. Today’s Today’s topic, c. topic,ladies ladiesand andgentlemen, gentlemen,isisastrophysics. astrophysics. Vocative Vocative12 d. Hold Hold on, Imperative d. on, are are we welate? late? Imperative e. Damn, we’ve missed the bus. Interjection e. Damn, we’ve missed the bus. Interjection

|| || 11 These categories are distinguished on the one hand on the basis of their respective dis11 These categories are distinguished on the one hand on the basis of their respective discourse function and on the other hand with reference to their status as theticals, according to course function and on the other hand with reference to their status as theticals, according to which they are set off syntactically, prosodically, and semantically (see (39)). Like our definiwhich they are set off syntactically, prosodically, and semantically (see (39)). Like our definition of theticals, categories are prototypical rather than based on necessary and sufficient tion of theticals, categories are prototypical rather than based on necessary and sufficient criteria. criteria. 12 There are a number of languages, such as Latin, that express vocatives by means of case 12 There are a number of languages, such as Latin, that express means of case affixes. Nevertheless, in their external syntax, vocatives are as a rule vocatives not part ofbythe sentence affixes. in their external syntax, vocatives are as a rule not part of the sentence structureNevertheless, in such languages. structure in such languages.

On theticals: “rootless” analysis of I think | 59 Parentheticals and theAgrammar of complementation

Theticals differ from SG units in a principled way, their defining properties are listed in (39).13 Note that this definition is not based on necessary and sufficient conditions, that is, not all of the properties are necessarily present in a given case (cf. Kaltenböck et al. 2011: 853–5). (39) Properties of theticals (Kaltenböck et al. 2011: 853) a. They are syntactically independent.14 b. They are typically set off prosodically from the rest of an utterance.15 c. Their meaning is non-restrictive. d. They tend to be positionally mobile. e. Their internal structure is built on principles of SG but can be elliptic.16 To conclude, information units of the two domains can be separated on the basis of semantic, prosodic, and semantic properties. Nevertheless, they are not entirely independent. First, as noted also by Newmeyer (2012: 11), there is not always a one-to-one relationship between syntactic structure and some other component.17 Accordingly, the match stipulated in (39) between different components of grammar may not be complete in exceptional cases. And second, there are various forms of interaction between them (see 4.3 below). As the term “non-restrictive” in (39c) indicates (Huddleston and Pullum 2002: 1352), non-restrictive meaning is not determined by the structure of a sentence or its constituents but rather by reasoning processes grounded in the || 13 The term thetical must not be confused with that of “thetic” statement (Kuroda 1972; Sasse 1987; 2006; Lambrecht 1994; cf. Kaltenböck et al. 2011, Fn. 6). Following Wehr (2000), we assume that the use of the term, at least as proposed e.g. by Sasse (1987; 2006), is redundant since “thetic statements” appear to be more appropriately analyzed in terms of information structure: They represent a construction type that lacks a sentence topic. 14 While syntactic independence is fairly obvious with stand-alone theticals (such as formulae of social exchange, vocatives, and interjections), various tests have been proposed in the literature to show the syntactic independence of conceptual theticals (i.e. parentheticals such as I think). For overviews cf., for instance, Peterson 1999, De Vries 2007, Kaltenböck 2007. 15 Typically theticals represent independent tone units. It is possible however for frequently used theticals to be re-integrated into the prosodic contour of the host construction over time. An example of such a prosodically integrated element is He was furious he was, as noted by an anonymous reviewer. 16 “Ellipsis” is a theory-dependent term that can be, and has been defined in a number of different ways. We are restricted here to comparing theticals with corresponding expressions of SG, where the former frequently have the appearance of reduced forms of the latter. 17 We are hesitant to accept the following somewhat radical position of Newmeyer (2012: 11) on the relationship between grammatical components: “It would be a serious mistake to presuppose a one-to-one mapping. In fact, there is clearly not in general a one-to-one mapping.”

60 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva situation of discourse.18 To be sure, any act of linguistic communication requires a situation of discourse, but in the structure of SG its impact is minimal, being restricted to a few grammaticalized phenomena like spatial, temporal and personal deixis, deontic modality, etc. Such restrictions do not appear to exist when TG is involved. The situation of discourse consists of a network of interlocking components, namely the ones listed in (40). (40) Components of the situation of discourse (Kaltenböck et al. 2011: 861) − − − − − −

Text organization Source of information Attitudes of the speaker Speaker-hearer interaction Discourse setting World knowledge

In accordance with this distinction, SG units differ from theticals in their semantic-pragmatic scope potential: Whereas the former have scope over some constituent of the utterance, theticals may have wider scope, typically extending over the entire situation of discourse (Kaltenböck et al. 2011: 861; Heine et al. 2013, Section 2.2).19 Accordingly, the meaning of theticals has been described with reference to notions such as “metatextual”, “metacommunicative”, “metadiscursive”, “metapragmatic”, or “meaning beyond the sentence” (relating to the component of text organization), “subjectivity” (attitudes of the speaker), or “intersubjectivi|| 18 An anonymous reviewer comments that (39c) is problematic in view of sentences such as Legally, they’re married, which “restrict the reference conditions of the sentence”. It is useful, however, to contrast this use of legally (classified by Quirk et al. 1985: 568 as viewpoint subjunct) with its use as adjunct in They’re legally married. In the latter, legally has clear restrictive meaning, restricting the meaning of marry to its legal form (not including e.g. marriage on stage, etc.). In the former use, on the other hand, legally modifies the sentence as a whole, qualifying the speaker’s position with regard to the proposition expressed. Here the speaker signals that s/he is speaking from a legal perspective (cf. also the possible paraphrases Legally speaking, from a legal point of view), thus expressing some speaker attitude (cf. (40)) similar to style disjuncts such as generally, personally, frankly. These uses are non-restrictive in Huddleston and Pullum’s (2002: 1352) sense, as used in their characterization of supplements (which largely correspond to our notion of theticals): “By virtue of not being integrated into the syntactic structure, supplements are necessarily semantically non-restrictive”. 19 Scope is, however, not a defining criterion of theticals (cf. 39 above). As noted earlier (footnote 9), theticals may also occasionally have narrow scope, e.g. over an adjective.

On theticals: “rootless” of I think | 61 Parentheticals and theAgrammar of analysis complementation

ty” (speaker-hearer interaction) (e.g., Traugott and Dasher 2002; Brinton 2008; cf. also Mithun 2008). The relationship between the two domains of Discourse Grammar is complex; it is shaped most of all by a mechanism called cooptation, whereby a chunk of SG, such as clause, a phrase, a word, or any other unit is deployed for use as a thetical (Kaltenböck et al. 2011: 874–5). Cooptation, in other words, is a spontaneous mechanism which ‘lifts’ an element out of SG and uses it as a thetical. As a result, its meaning is no longer constrained by the rules of SG but is shaped by the situation of discourse (see above).

4.2 Back to I think What is important for our discussion is the hypothesis underlying the structure of TG according to which a thetical comment clause such as I think in (1) is part of a larger pattern of conceptual theticals. What distinguishes this comment clause from many other lexically related theticals is that it tends to be used as a formulaic thetical, that is, as a non-compositional, morphosyntactically unanalyzable information unit, and this fact accounts for some features of I think in (1) (Kaltenböck et al. 2011: 871). But instead of the formulaic thetical in (1) there can be any kind of instantaneous (i.e. spontaneously designed) thetical, like the units printed in bold in (41a).20 Instead of a declarative form, the thetical can take e.g. an imperative (41b) or an interrogative form (41c). Finally, the thetical is also not restricted to a (reduced) clause; rather, it may have a complex internal structure, as in (41d). (41) a. b. c. d.

We should leave, as I would think, as soon as possible. We should leave, don’t forget, as soon as possible. We should leave, don’t you think, as soon as possible. We should leave, and I think that there is no alternative, as soon as possible.

Note further that the thetical is not necessarily restricted to the positions illustrated in (1), namely after the verb, and sentence-finally. Rather, it may also appear in other positions, such as after the subject (42a) or the auxiliary (42b).21

|| 20 Concerning the distinction between instantaneous, constructional, and formulaic theticals, cf. Kaltenböck et al. (2011: 870–2). 21 We are ignoring here differences in meaning associated with this positional variability.

62 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva 62 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva And much the same applies to theticals other than I think, as the examples in (43) suggest. And much the same applies to theticals other than I think, as the examples in (43) suggest. (42) a. b. (42) a. b. (43) a.

We, I think, should leave as soon as possible. We should, I think, leave as soon as possible. We, I think, should leave as soon as possible. We should, I think, leave leave as soon possible. We, don’t forget, should as as soon as possible.

b. We should, don’t forget, leave as soon as possible. (43) a. We, don’t forget, should leave as soon as possible. b. We should, leave as soonsuch as possible. To conclude, I thinkdon’t in (1),forget, like other theticals as the ones in (41), conforms to our definition in (39): Unlike SG categories, such as verbs or adverbs, they are To conclude, I think in (1), like other theticals such as the ones in (41), conforms syntactically independent, not being licensed by the syntax of the sentence, to our definition in (39): Unlike SG categories, such as verbs or adverbs, they are they tend to be prosodically independent, signaled in writing by commas, and syntactically independent, not being licensed by the syntax of the sentence, to be positionally highly variable. And (39e) also applies: I think is built on printhey tend to be prosodically independent, signaled in writing by commas, and ciples of SG but has the appearance of an elliptic clause, lacking a complement. to be positionally highly variable. And (39e) also applies: I think is built on principles of SG but has the appearance of an elliptic clause, lacking a complement.

4.3 Interplay between the two domains 4.3 Interplay between the two domains

Obviously, when two different phenomena serve the same overall function, are made of the same material, and share the same space of operation they will not Obviously, when two different phenomena serve the same overall function, are remain entirely isolated from one another. Very likely, there will be interaction made of the same material, and share the same space of operation they will not between the two; the presence of coarticulation in phonetics is a paradigm exremain entirely isolated from one another. Very likely, there will be interaction ample of massive interaction. between the two; the presence of coarticulation in phonetics is a paradigm exTG and SG are in principle syntactically independent of one another. Neverample of massive interaction. theless, they do not operate in a vacuum; rather, both serve human communicaTG and SG are in principle syntactically independent of one another. Nevertion, draw on essentially the same linguistic material, and share the same linear theless, they do not operate in a vacuum; rather, both serve human communicachannel. It therefore may come as no surprise that there is also interaction betion, draw on essentially the same linguistic material, and share the same linear tween SG and TG when they combine to produce linearized texts (cf. Heine et al. channel. It therefore may come as no surprise that there is also interaction be2013: Section 3). tween SG and TG when they combine to produce linearized texts (cf. Heine et al. The exact nature and extent of interaction is a subject of future research; 2013: Section 3). suffice it here to illustrate its arguably most salient form, namely crossThe exact nature and extent of interaction is a subject of future research; referencing. Usually it is the SG host or some constituent of it (typically a noun suffice it here to illustrate its arguably most salient form, namely crossphrase) that is cross-referenced in the thetical by some pronominal form (cf. e.g. referencing. Usually it is the SG host or some constituent of it (typically a noun Ackema and Neeleman 2004 for Dutch). This form can be a personal pronoun, phrase) that is cross-referenced in the thetical by some pronominal form (cf. e.g. like it in (44a), a relative clause marker in the case of non-restrictive relative Ackema and Neeleman 2004 for Dutch). This form can be a personal pronoun, clauses, cf. who in (44b), or some other pronominal such as the manner prolike it in (44a), a relative clause marker in the case of non-restrictive relative clauses, cf. who in (44b), or some other pronominal such as the manner pro-

On theticals: “rootless” of I think | 63 Parentheticals and theAgrammar of analysis complementation

noun so in (44c).22 But the pronominal may be a characteristic of the host while the thetical presents the full form, as is commonly the case in detached phrases such as left-dislocations (44d) or right-dislocations (44e). Finally, the two units can also be linked in that e.g. the thetical repeats some components of its SG host, as in (44f). (44) a. John is, would you believe it?, a fink. b. John, who is living on a small income, is still a bachelor. (Espinal 1991: 726) c. John is, at least I think so, a fink. d. The book I lent you – have you finished it yet? (Quirk et al. 1985: 1416) e. Hei is here, my unclei. f. I agree with every word you’ve said – every single word. (Quirk et al. 1985: 1416) Interaction affects all levels of language structure. While theticals are as a rule “invisible” to syntactic or semantic interpretation, this is not always so. Take the following example, where presence vs. absence of the thetical loosely speaking (called a stance adverb by Powell 1992: 78) affects the truth of the utterance. (45) a. A whale is a fish. (False) b. A whale is, loosely speaking, a fish. (True) (Adapted from Lakoff 1972 and Powell 1992: 78) Conceivably, a number of the features that have been adduced as evidence for a root hypothesis can more profitably be interpreted as being due to interaction between the two domains. But this is a subject that requires much further research. On the other hand, what is also obvious is that the framework proposed here is not, and cannot be, a “genuinely ‘rootless’ analysis” since the two domains share the same space of discourse and are related to one another in a number of ways, most of all by means of the mechanism of cooptation (Section 4.1).23

|| 22 In all these uses the thetical is still seen as syntactically unintegrated, with the link to the host construction being a non-syntagmatic one which operates only on the semantic-pragmatic level (cf., e.g. Burton-Roberts 1999, Peterson 1999; Kaltenböck 2007 for overviews). 23 We are grateful to an anonymous reviewer for having drawn our attention to this issue.

64 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva

5 Conclusions The goal of the paper has been to argue against a “root” analysis and in favor of a “rootless” approach along the lines outlined in Section 4. To this end, three main arguments were put forward. First, it was argued that not much would be gained by reducing thetical I think, as illustrated in (1), to its syntactic manifestations. Rather, what is required is a better understanding of its status as a fragment of discourse structure. To deal with this structure we proposed the framework of Discourse Grammar in Section 4. Due to space limitations we were not able to discuss its functions as a discourse marker in this paper. That these functions differ dramatically from those of I think as a matrix-clause has been demonstrated abundantly in the course of the last decade (cf., e.g., Holmes 1990; Aijmer 1997; Simon-Vandenbergen 2000; Mindt 2003; Kärkkäinen 2003; 2010; Kaltenböck 2010a, 2013; Rodríguez Louro & Harris 2013).24 Second, it was argued that not much would be gained by reducing thetical and matrix-clause I think to the same basic, or underlying, syntactic structure (Section 3). There are simply too many structural differences separating the two, and Discourse Grammar provides a tool to account for these differences. And third, we drew attention to the fact that the two domains of Discourse Grammar do not operate in a vacuum. SG units and TG units occupy the same linear space and regularly co-occur and interact with one another (Section 4.3). It therefore comes as no surprise that the two cannot always be neatly separated: Due to constant interaction, theticals may adopt syntactic, prosodic, and/or semantic features of their SG host sentences. As we observed in Section 4.1, our definition of theticals in (39) is not based on necessary and sufficient conditions, that is, not all of the features are necessarily present in a given case. Accordingly, if there are in fact syntactic features that thetical and matrix-clause I think share then this may well be due to such kinds of interactions; more research is needed on this issue. But ignoring central issues of theoretical orientation, Newmeyer’s analysis is not entirely incompatible with our approach to Discourse Grammar. His notion of “derivation” can well be likened to our mechanism of cooptation (Section 4.1), e.g. when he argues that the information unit think in the examples of || 24 As pointed out by an anonymous reviewer, this also leaves open the question of a possible link between the different functions, i.e. I think used as a discourse marker and as a matrix clause. A discussion of how these uses may be linked can be found in Boye and Harder (2007), who advocate three different types on a scale of gradience, and Kaltenböck (2010b, 2011), who argues for a taxonomic link in terms of a Construction Grammar approach.

On theticals: “rootless” analysis of I think | 65 Parentheticals and theAgrammar of complementation

(1) “is – at some point in the analysis – a main-clause taking a subordinate clause”: There is a transfer from one kind of structure (which in our framework is an SG structure) to another kind of structure (belonging to TG)25 and second, the coopted unit may have the appearance of an “elliptic” piece compared to the corresponding structure of SG (for a discussion of how cooptation relates to omission, see Section 2.5 in Kaltenböck et al. 2011: 867–70). And even when coopted as an elliptic piece, the unit can inherit valency features, although such features may no longer relate to surrounding text pieces but rather to the situation of discourse in general. Moreover, both derivation and the development of comment clauses may involve several stages (or “points”; Newmeyer 2012: 8): Once coopted as a thetical, I think developed further, eventually giving rise to an invariable formulaic thetical in some of its uses. When drawing attention to such parallels, however, we do not wish to belittle the fundamental differences that exist between the two perspectives: The “root” hypothesis presented in Newmeyer (2012) differs drastically from the “rootless” hypothesis argued for in this paper. The problem that we have with Newmeyer’s analysis is in particular that it does not appear to provide plausible answers to the questions that we raised in (5) of the introduction. We contend that an explanation of the grammatical and ontological status of I think in examples such as (1) needs to account for why I think differs from constituents of Sentence Grammar in not being licensed by the syntax of the sentence, not being linked to the host sentence by any morphological or syntactic means, in showing a prosodic behavior that tends to differ from that of the rest of the utterance, and in lacking a complement that otherwise would be expected to be present. But perhaps more importantly, such an account would also need to account for why I think in (1), while differing from phrases or clauses of Sentence Grammar, shows the same kind of behavior as the vocative unit in (6) or many others of the theticals mentioned in Section 4.1 and described in more detail in Kaltenböck et al. (2011) and Heine et al. (2013). Our analysis of theticals is in line with the notion of supplementation of Huddleston and Pullum (2002: 1350), which is characterized by “lack of integration into the syntactic structure”. Supplementation, which they contrast with || 25 An anonymous reviewer raises the question of how this transfer from SG to TG fits in with language acquisition data presented by Diessel and Tomassello (2001) and Brandt et al. (2010) according to which children first use I think as an unanalysed formulaic chunk and only later as a matrix clause. In our terminology I think is a “formulaic thetical” (see 4.2), i.e. a noncompositional and essentially invariable unit. As high-frequency items they are firmly conventionalized (entrenched) and can therefore be expected to be acquired by children at a very early stage.

66 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva dependency constructions and coordination, exhibits significant overlaps with our domain of TG. In the present paper we had to ignore a number of problems that would need to be considered when dealing with the distinction between root and rootless analyses. One problem concerns the nature and structure of Discourse Grammar. We could hardly do more other than provide a brief sketch of it in Section 4.1 (cf. Heine et al. 2013 for details). Accordingly, we were unable to discuss other factors that are also relevant for understanding the differences separating the information unit I think as a matrix clause from its uses as a thetical and, especially, as a formulaic thetical, commonly described as a discourse marker (cf. Aijmer 1997; Kärkkäinen 2003; Mindt 2003; Simon-Vandenbergen 2000; Kaltenböck 2010a, 2013). Another kind of issue relates to the present state of research, that is, to problems that have not been appropriately investigated so far. In particular, we did not deal with the question of where the boundary between the two domains of Discourse Grammar is to be traced. This includes, for instance, the question of diachronic development, such as the possible reintegration of a thetical unit into the domain of SG over time (e.g. Boye and Harder 2009 on the grammaticalization of parentheticals into SG items).26 Most of all, however, we ignored the question of what the status of sentence-initial I think in (2a) is: Is it a matrix clause without complementizer, and hence a part of SG, or is it a comment clause, i.e. a manifestation of TG? Or, as aptly pointed out by an anonymous reviewer, is it perhaps a matrix clause in some of its uses but a thetical in other uses? As observed in Kaltenböck et al. (2011: 882), this question cannot be answered satisfactorily due to lack of appropriate evidence: Both analyses are possible (cf. also Kaltenböck 2009; 2011; Dehé and Wichmann 2010a 2010b). Hence we must leave such questions for future research.

Corpus Data Corpus of Historical American English (CHA), British National Corpus (BNC), International Corpus of English – British component (ICE-GB), Diachronic Corpus of Present-Day Spoken English (DCPSE). || 26 An anonymous reviewer notes that “the paper ignores diachrony”. In fact the paper does so for good reasons. Considering the complex nature of the diachronic development of theticals (cf. Heine 2013), this is an issue that needs much more research and a separate treatment.

On theticals: “rootless” analysis of I think | 67 Parentheticals and theAgrammar of complementation

References Ackema, Peter & Ad Neeleman. 2004. Beyond Morphology: Interface Conditions on Word Formation. Oxford: Oxford University Press. Aijmer, Karin. 1997. I think – an English modal particle. In Toril Swan, Toril & Olaf J. Westvik (eds.). Modality in Germanic languages. Historical and comparative perspectives. Berlin: Mouton de Gruyter. 1–47. Asher, Nicholas. 2000. Truth conditional discourse semantics for parentheticals. Journal of Semantics 17. 31–50. Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad & Edward Finegan. 1999. Longman Grammar of Spoken and Written English. London: Longman. Boye, Kasper. 2012. Epistemic meaning. Berlin: Mouton de Gruyter. Boye, Kasper & Peter Harder. 2007. Complement-taking predicates: Usage and linguistic structure. Studies in Language 31, 3. 569–606. Boye, Kasper & Peter Harder. 2009. Evidentiality: Linguistic categories and grammaticalization. Evidentiality in language and cognition. Special issue of: Functions of Language 16, 1. 9–43. Brandt, Silke, Elena Lieven & Michael Tomasello. 2012. Development of word order in German complement-clause constructions: Effects of input frequencies, lexical items, and discourse function. Language 86. 583–610. Brinton, Laurel J. 2008. The Comment Clause in English: Syntactic Origins and Pragmatic Development. Cambridge: Cambridge University Press. Brinton, Laurel J. & Elisabeth C. Traugott. 2005. Lexicalization and Language Change. Cambridge: Cambridge University Press. Burton-Roberts, Noel. 1999. Language, linear precedence and parentheticals. In Peter Collins & David Lee. 1999. The Clause in English. Amsterdam, Philadelphia: Benjamins. 33–52. Davies, Mark. 2004-. BYU-BNC. (Based on the British National Corpus from Oxford University Press). http://corpus.byu.edu/bnc/. Davies, Mark. 2010. The Corpus of Historical American English: 400 million words, 1810-2009. http://corpus.byu.edu/coha. de Vries, Mark. 2007. Invisible constituents? Parentheses as B-merged adverbial phrases. In Nicole Dehé & Yordanka Kavalova 2007a. Parentheticals. Amsterdam, Philadelphia: Benjamins. 203–34. Dehé, Nicole & Anne Wichmann. 2010a. Sentence-initial I think (that) and I believe (that): Prosodic evidence for use as main clause, comment clause and discourse marker. Studies in Language 34, 1. 36–74. Dehé, Nicole & Anne Wichmann. 2010b. The multifunctionality of epistemic parentheticals in discourse: Prosodic cues to the semantic-pragmatic boundary. Functions of Language 17, 1. 1–28. Diessel, Holger & Michael Tomasello. 2001. The acquisition of finite complement clauses in English: A corpus-based analysis. Cognitive Linguistics 12, 2. 97–141. Emonds, Joseph. 1973. Parenthetical clauses. In Claudia Corum et al. (eds.). You take the high node and I’ll take the low node. Chicago: Linguistic Society. 333–47. Emonds, Joseph. 1976. A Transformational Approach to English Syntax. New York: Academic Press.

68 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva Emonds, Joseph. 1979. Appositive relatives have no properties. Linguistic Inquiry 10, 2. 211– 43. Espinal, Teresa M. 1991. The representation of disjunct constituents. Language 67. 726–62. Fabb, Nigel 1990. The difference between English restrictive and non-restrictive relative clauses. Journal of Linguistics 26. 57–78. Haegeman, Liliane. 1991. Parenthetical adverbials: the radical orphanage approach. In Shuji Chiba et al. (eds.). Aspects of Modern English Linguistics: Papers Presented to Masatomo Ukaji on his 60th Birthday. Tokyo: Kaitakusha. 232–54. Heine, Bernd. 2013. On discourse markers: Grammaticalization, pragmaticalization, or something else? Linguistics 51, 6. 1205–1247. Heine, Bernd & Gunther Kaltenböck. 2012. On the placement of conceptual theticals. Typescript. Heine, Bernd, Gunther Kaltenböck, Tania Kuteva & Haiping Long. 2013. An outline of Discourse Grammar. In Shannon Bischoff & Carmen Jeny (eds.). Reflections on Functionalism in Linguistics. Berlin: Mouton de Gruyter. 155–206. Holmes, Janet. 1990. Hedges and boosters in women’s and men’s speech. Language and Communication 10 (3). 185–205. Hooper, Joan B. 1975. On assertive predicates. In John P. Kimball (ed.) Syntax and Semantics, Vol. 4, 91-124. New York: Academic Press. Huddleston, Rodney & Geoffrey K. Pullum. 2002. The Cambridge Grammar of the English Language. Cambridge: Cambridge University Press. Jackendoff, Ray S. 1977. X’ Syntax: A Study of Phrase Structure. Cambridge MA: MIT Press. Kaltenböck, Gunther. 2007. Spoken parenthetical clauses in English: a taxonomy. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam: Benjamins. 25–52. Kaltenböck, Gunther. 2008. Prosody and function of English comment clauses. Folia Linguistica 42, 1. 83–134. Kaltenböck, Gunther. 2009. Initial I think: main or comment clause? Discourse and Interaction 2, 1. 49–70. Kaltenböck, Gunther. 2010a. Pragmatic functions of parenthetial I think. In Gunther Kaltenböck, Wiltrud Mihatsch & Stefan Schneider (eds.). 2010. New Approaches to Hedging. Bingley: Emerald Publishers. 243–70. Kaltenböck, Gunther. 2010b. Comment clauses as constructions. Vienna English Working Papers 19, 1&2. 21–46. https://typo3.univie.ac.at/fileadmin/user_upload/dep_anglist/weitere_Uploads/Views/ Views19_1and2_2010.pdf Kaltenböck, Gunther. 2011. Explaining diverging evidence. The case of clause-initial I think. In Doris Schönefeld (ed.). Converging evidence: methodological and theoretical issues for linguistic research. Amsterdam: Benjamins. 81–112. Kaltenböck, Gunther. 2013. Development of comment clauses. In Bas Aarts, Joanne Close, Geoffrey Leech & Sean Wallis (eds.). The English verb phrase: investigating recent language change with corpora. Cambridge: Cambridge University Press. 286–317. Kaltenböck, Gunther, Bernd Heine & Tania Kuteva. 2011. On thetical grammar. Studies in Language 35, 4. 848–893. Kärkkäinen, Elise. 2003. Epistemic Stance in English Conversation. A Description of its Interactional Functions, with a Focus on I think. Amsterdam: John Benjamins.

On theticals: “rootless” analysis of I think | 69 Parentheticals and theAgrammar of complementation

Kärkkäinen, Elise. 2010. Position and scope of epistemic phrases in planned and unplanned American English. In Gunther Kaltenböck, Wiltrud Mihatsch & Stefan Schneider (eds.). New approaches to hedging. Bingley: Emerald. 207–241. Keller, Frank. 1998. Grammaticality judgments and linguistic methodology. Research Paper EUCCS-RP-1998-3. Knowles, John. 1980. The tag as a parenthetical. Studies in Language 4, 3. 379–409. Kuroda, Shigeyuki 1972. The categorical and the thetic judgment. Evidence from Japanese syntax. Foundations of Language 9. 153–85. Lakoff, George. 1972. Hedges: A study in meaning criteria and the logic of fuzzy concepts. Chicago Linguistic Society 8. 183–228. Lambrecht, Knud. 1994. Information Structure and Language Form. Cambridge: Cambridge University Press. McCawley, James D. 1982. Parentheticals and discontinuous constituent structure. Linguistic Inquiry 13. 91–106. McCawley, James D. 1998. The Syntactic Phenomena of English. 2nd edn. Chicago: The University of Chicago Press. Mindt, Ilka. 2003. Is I think a discourse marker? In Ewald Mengel et al. (eds.). Proceedings Anglistentag 2002 Bayreuth. Trier: WVT. 473–483. Mithun, Marianne. 2008. The extension of dependency beyond the sentence. Language 84, 1. 69–119. Nelson, Gerald, Sean Wallis & Bas Aarts. 2002. Exploring Natural Language. Working with the British Component of the International Corpus of English. Amsterdam/Philadelphia: Benjamins. Newmeyer, Frederick J. 2012. Parentheticals, ‘fragments’, and the grammar of complementation. Les verbes parenthtétiques: hypotaxe, parataxe or parenthèse? (Conference). Université Paris Ouest Nanterre. 24–26 May2012. Peterson, Peter. 1999. On the boundaries of syntax: non-syntagmatic relations. In Peter Collins & David Lee (eds.). The Clause in English: In Honour of Rodney Huddleston. Amsterdam/ Philadelphia: John Benjamins. 229–250. Powell, Mava Jo. 1992. The systematic development of correlated interpersonal and metalinguistic uses of stance adverbs. Cognitive Linguistics 3. 75–110. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech & Jan Svartvik. 1985. A Comprehensive Grammar of the English Language. London/New York: Longman. Rodríguez Louro, Celeste & Thomas C. Harris (2013). Evolution with an attitude: The grammaticalisation of epistemic/evidential verbs in Australian English. English Language and Linguistics 17,3. 415–443. Sasse, Hans-Jürgen. 1987. The thetic/categorical distinction revisited. Linguistics 25. 511–580. Sasse, Hans-Jürgen. 2006. Theticity. In Giuliano Bernini & Marcia L. Schwartz (eds). Pragmatic Organization of Discourse in the Languages of Europe. Berlin/New York: Mouton de Gruyter. 255–308. Schneider, Stefan. 2007. Reduced Parenthetical Clauses as Mitigators. A Corpus Study of Spoken French, Italian and Spanish. Amsterdam: Benjamins. Schütze, Carson T. 1996. The Empirical Base of Linguistics: Grammaticality Judgments and Linguistic Methodology. Chicago/London: University of Chicago Press. Simon-Vandenbergen, Anne-Marie. 2000. The functions of I think in political discourse. International Journal of Applied Linguistics 10 (1). 41–63.

70 | Gunther Kaltenböck, Bernd Heine and Tania Kuteva Thompson, Sandra A. 2002. ‘Object complements’ and conversation: Towards a realistic account. Studies in Language 26. 125–64. Thompson, Sandra A. & Anthony Mulac. 1991. A quantitative perspective on the grammaticization of epistemic parentheticals in English. In Elisabeth C. Traugott & Bernd Heine (eds.) 1991. Approaches to grammaticalization Vol 2. Amsterdam: John Benjamins. 313–339. Traugott, Elizabeth C. & Richard B. Dasher. 2002. Regularity in semantic change. Cambridge Studies in Linguistics 96. Cambridge: Cambridge University Press. Wehr, Barbara. 2000. Zur Beschreibung der Syntax des français parlé (mit einem Exkurs zu “thetisch” und “kategorisch”). In Barbara Wehr & Helga Thomaßen (eds.). Diskursanalyse: Untersuchungen zum gesprochenen Französisch. Frankfurt/Berlin: Peter Lang. 239– 89. Ziv, Yael. 1985. Parentheticals and Functional Grammar. In A. Machtelt Bolkestein et al. (eds.). Syntax and Pragmatics in Functional Grammar. Dordrecht: Foris. 181–199.

James Griffiths

Speaker and quote reduced parenthetical clauses Abstract: In this paper, I present a syntactico-pragmatic delimitation and subsequent analysis of reduced parenthetical clause constructions that is based upon Schneider’s (2007a) superordinate and binary division between reduced parenthetical clauses that modify quotes, and those that modify utterances of any other illocutionary type. I argue that the parentheticals in question can receive a unified analysis, and that, by splitting them asunder according to the type of host they modify, the vast majority of the seemingly disparate properties they display can be readily accounted for. Keywords: parenthesis, comment clauses, report clauses, quotation

|| James Griffiths: Center for Language and Cognition, University of Groningen. Groningen, The Netherlands.

Introduction Studies such as Reis (1997, 2000, 2002) and Schneider (2007a) have revealed the large diversity in syntactic and pragmatic properties displayed by the parentheticals exemplified by the italicised strings in (1) and (2). (1) a. Claude is, I think, the best candidate for the job. b. Who has, do you reckon, won the race? c. I declare this store open, I am happy to say. (2) a. “I have,” Kevin exclaims, “won the lottery!” b. “Do you know why,” asked Ysobell, “they are making these redundancies?”

Reis (1995), Schneider (2007a,b) and Griffiths (2014) illustrate that such clauses display dissimilar properties depending upon (i) their degree of prosodic integration into their host, (ii) their pragmatic function, and (iii) their position of interpolation within their host. That such variety exists has led scholars in re-

72 | James Griffiths cent years to concentrate on well-delimited subclasses of these parentheticals (cf., for example, Steinbach 2007 and Fortmann 2007). This paper is an exercise in generalisation. Here, I explore to what extent one can apply an encompassing analysis to parentheticals of the type exemplified by (1) and (2), using data from English, Dutch, German, and Turkish and analytical tools from generative syntax and Discourse Representation Theory (DRT, Kamp 1981). I adopt the following approach to this task: (i) outline a plausible analysis that provides broad but incomplete empirical coverage, and (ii) explore to what extent this account’s plausibility diminishes when one aims for full empirical coverage. By undertaking (i) and (ii), I aim to mark out the path that I believe one should tread towards a generalised analysis that accounts for the vast majority of the disparate properties that these parentheticals display. My investigation is guided by two dicta from Schneider (2007a). The first is Schneider’s delimitation of the object of inquiry. He delimits the parentheticals in (1) and (2) as reduced parenthetical clauses (RPCs), and defines them as follows (ibid.:7): (3) ‘[RPCs are] 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.’

The second is his superordinate and binary distinction between RPCs that modify (that is, ‘have an interpretative effect upon’) host clauses that are quotes (2), and those that modify host clauses that are of any other type of speech act (1). I call the former quote RPCs, and latter speaker RPCs. The paper proceeds as follows. In §1, I advance a typology that delimits RPCs according to type of host they modify. In §2, I outline my analysis of how RPCs and their hosts relate to one another. In §3, I outline my analysis of the internal syntax of RPCs. In §4, I discuss utterance-initial RPCs, and explore the extent to which the analysis outlined in §2-3 can account for their distribution. §5 concludes.

1 Reduced parenthetical clauses Throughout this paper, I maintain that RPCs can be delimited across two dimensions: (i) report versus attitude type, and (ii) quote versus speaker use. Let us examine these two divisions in turn.

Speaker andand quote clauses | 73 Parentheticals thereduced grammarparenthetical of complementation

1.1 Report versus attitude type Report RPCs describe the actions of the speaker or another agent. These actions can occur either concurrently or asynchronously with speaker time (the time at which the speaker utters the host). In English, a report RPCs’ verb permits any tense, and its subject permits any person. RPCs in which passivation is displayed are also report RPCs. (4) a. Bobi’ll make chief cameraman by July, hei reckons. b. Clint mustn’t, I thought yesterday, blame himself. c. Dirk must, I’ve been told, re-mortgage his house.

Attitude RPCs express an attitude of the speaker. Adoption of this attitude runs concurrent with speaker time. Attitude RPCs display active voice and a first person subject. An attitude RPC verb is typically marked in the simple present. However, certain combinations of modality and tense are permitted provided they have no semantic import. (5) a. It’ll be shot in analogue, I hope. b. All of Fassbinder’s films, I declare, are utter rubbish. c. Eastwood will retire at ninety, I’d have thought.

(where the RPC means ‘I think’)

1.2 Quote versus speaker use In certain situations, the speaker may take no responsibility for the content of the utterance that she voices. In these environments, the speaker is a spokesperson that relays utterances that are said or thought by others (or herself, at an epistemic distance). When the speaker functions in this manner, she employs quotation. The utterance that is relayed is a quote. If the speaker wishes to provide extraneous information about the quote’s creator or the manner in which the quote was created, she utilises a report RPC. Quotes that the speaker voices are either originally created before, after, or concurrent with the speaker’s broadcast of the quote. Considering the variation in creator and time, (6) provides an illustrative list of quote/report RPC combinations (where t = speaker time, the time at which the speaker broadcasts the quote).

74 | James Griffiths

(6) a. “Orson,” I {think / say}, “must be fired.”

[speakeri] [creatori, at t]

b. “Orson,” Mank {thinks / says}, “must be fired.”

[speakeri] [creatork, at t]

c. “Orson,” I {thought / said}, “must be fired.”

[speakeri] [creatori, before t]

d. “Orson,” Mank {thought / said}, “must be fired.” [speakeri] [creatork, before t] e. “Orson,” I will {think / say}, “must be fired.”

[speakeri] [creatori, after t]

f. “Orson,” Mank will {think / say}, “must be fired.” [speakeri] [creatork, after t]

Note that, in quotation of the type illustrated in (6), deictic expressions displayed in the quote are interpreted relative to the quote’s creator. This is direct quotation. When the quote’s deictic centre shifts to the speaker, free indirect quotation is obtained (Rooryck 2001, Sharvit 2008). Modulo this shift in deictic centre, both types of quotation are identical with respect to their syntax and semantics (Maier to appear). Quotes are demonstrations (Clark & Gerrig 1990) of previously-performed illocutionary, locutionary, or non-linguistic acts. Quote RPCs may modify demonstrations of any type.1 (7) a. “My husband is an awful cook,” Mary admits.

Host = quoted illocutionary act

b. “Who is,” Adam asked, “coming to my party?”

Host = quoted illocutionary act

c. “Doo-wop skoobie-doobie,” he scatted.

Host = quoted locutionary act

d. “Sacrebleu!” he exclaimed.

Host = quoted locutionary act

e. , the rhythm will go.

Host = quoted non-linguistic act

Direct and free indirect quotation comprise the quote use of RPCs. RPCs used to modify quotes are quote RPCs. RPCs are employed for their speaker use whenever they are not employed for their quote use. Therefore, whenever a RPC modifies a proposition that is utilised by the speaker to commit an assertive, erotetic, declarative, commissive or optative illocutionary act (inter alia) – i.e. any act that is not a quotation –, that RPC is a speaker RPC. (8) provides some exemplary cases. (8) a. Lucyi should, {I reckon / Pete tells me / shei’s been told}, sell the house. b. Who has, {do you wonder / does Fred believe}, solved the problem?

Assertive Erotetic

c. I hereby name this ship the Paralus, I am happy to say.

Declarative

d. I do swear that I will bear allegiance to the Queen, I declare.

Commissive

|| 1 I assume here that mixed quotations are not demonstrations, however, based on the fact that quote RPCs cannot modify mixed quotation (i). (i) * Bush has an “ecelectic”, he says, reading list.

Speaker andand quote clauses | 75 Parentheticals thereduced grammarparenthetical of complementation

e. Live long, I {wish / hope}, and prosper!

Optative

With respect to pragmatic function, speaker RPCs fall into three main classes: mitigative, speech act and evaluative. These classes correspond to Cinque’s (1999) distinction between evidential/epistemic, speech act, and evaluative adverbs (adverbs such as probably, frankly, and fortunately, respectively). Mitigative RPCs alleviate the speaker’s responsibility for the truth of proposition denoted by the host. In English, Dutch, and German, report and attitude RPCs can be employed to serve this mitigative function.2 If an attitude RPC is employed, the speaker shifts the burden of responsibility to her own attitude. If a report RPC is employed, the speaker shifts the burden of responsibility to a third party. (9) a. John will, I believe, be late. b. John will, Mary heard, be late.

attitude RPC report RPC

In Turkish, attitude RPCs alone may serve this mitigative function. This is evidenced by (10), where the host must be interpreted as an assertion and not a quote when an RPC occupies the host’s postverbal area. In such cases, report RPCs are prohibited. (10) a. Ali bir hırsız-dır, {san-ıyor-um/ * san-ıyor-Ø} Ali a

thief-COP

ki.

Turkish

believe-PROG-1S/ believe-PROG-3S ki

‘ Alii is a thief, {I / *hek} believe(s).’ b. Ali bir hırsız-dır, {san-ıyor-um/ Ali a

thief-COP

* san-dı-m}

believe-PROG-1S/ believe-PST-1S

‘ Ali is a thief, I {believe / * believed}.’

ki. ki (Griffiths & Güneş 2013)

Speech act RPCs ‘type’ the host that they modify for a particular illocutionary force (11a). Evaluative RPCs express the speaker’s emotional stance towards a host (11b). To perform either function, only attitude RPCs can be utilised. (11) a. I will, I {promise / swear / declare}, always love you. b. My article should, I {hope / pray}, be accepted.

|| 2 In English, Dutch, and German, only first and second person report RPCs can be employed in speaker RPC constructions.

76 | James Griffiths The ‘demarcation lines’ between mitigative, speech act and evaluative RPCs are not sharp. Some RPCs, such as those in (12), inherently function as evaluative and speech act RPCs simultaneously. Others change function according to the context. In (13a) for example, when the speaker vows, a ‘speech act’ reading is obtained. However in (13b), when the newspapers vow, the RPC is understood as shifting the burden of responsibility for the truth of the host from the speaker to a third party. Hence, in (13b), the newspapers vow is interpreted as mitigative. (12) a. John will, I regret to say, be late. b. John will, I’m happy to concede, be late. c. Who will, I’m obliged to ask, be late? (13) a. He will, I vow, be acquitted by Friday b. He will, the newspapers vow, be acquitted by Friday.

Thus, the terms mitigative, evaluative, and speech act most likely denote positions on a cline of evidential meanings (Rooryck 2001), and are not discrete classes (Schneider 2007b:243). Which type of speaker RPC may modify which type of host depends upon whether pragmatic congruity is obtained when hosts and RPCs of certain types are combined. Table 1 below provides an overview of permissible combinations in English (see appendix 1 for corresponding examples).

Host type3 Assertive Erotetic Declarative Commissive Optative

Mitigative RPCs     

Speech Act RPCs     

Evaluative RPCs   *  

Table 1: Possible host/speaker RPC combinations *

Possible only with evaluative/speech act combos: I declare this store open, I am {happy/sad} to {say/pronounce}.

|| 3 This refers the illocutionary force that the host would express if the RPC were absent.

Speaker andand quote clauses | 77 Parentheticals thereduced grammarparenthetical of complementation

1.3 Combining report versus attitude type and quote versus speaker use Table 2 summarises how an RPC’s type (attitude or report) and use to which it is put (speaker or quote) may interact in English, Dutch, and German. Quote

Speaker

Speaker

Speaker

attitude RPCs



(mitigative) 

(speech act) 

(evaluative) 

report RPCs



4





Table 2: Possible ‘type’/’use’ RCP combinations

Table 2 highlights two sources of ambiguity in English, Dutch, and German. The first arises from the fact that for every attitude RPC there exists a homophonous report RPC counterpart. The second arises from the fact that report RPCs can be used in both quote and speaker – specifically, mitigative – RPC constructions. Combined, these two sources of ambiguity engender structures like (14), whose hosts can be interpreted as either quotes or assertions in neutral contexts. (14) a. Pete should, I think, quit his job. b. Susie has, Pete says, finally finished her thesis.

attitude or report RPC report RPC

Fortunately, there are a number of means by which examples like (14) can be disambiguated. A subset of these means are language-specific (I concentrate solely on English here). The first concerns disambiguation of the RPC’s verb. In (14a), think is ambiguous between an attitude that is synonymous with believe (the ‘attitude RPC’ reading) and an action that is synonymous with contemplate (the ‘report RPC’ reading). This ambiguity renders the host ambiguous between a quote and an assertion, as only second and third person report RPCs can be used in speaker RPC constructions in English. Thus, the addition of disambiguators such as to himself, which disambiguates think as an action, disambiguates the host as a quote. (15) Pete should, I think to myself, quit his job.

|| 4 As per the examples in (10), this cell is checked in with a  rather than an  in Turkish.

78 | James Griffiths In English, provided certain restrictions are obeyed (cf. Collins & Branigan 1997), subject-verb inversion may occur in quote RPCs but not speaker RPCs. As such, the presence of subject-verb inversion in an example such as (14b) disambiguates it as a quote RPC construction.5 (16) Susie has, says Pete, finally finished her thesis.

In English, provided (i) the host contains sentential negation, (ii) the RPC contains a common verbum sentiendi, and (iii) the RPC’s subject is a first person pronoun, a semantically vacuous instantiation of not is optionally licensed in speaker RPCs, but not quote RPCs. Thus, if an instance of not in an RPC such as the one in (17) is interpreted as semantically vacuous, the host is disambiguated as an assertion. (17) John won’t, I shouldn’t think, be late.

Provided the context is sufficiently rich, point of view may help disambiguate speaker from quote RPCs. Point of view describes through whose eyes the host content should be understood. Most often, speaker RPC constructions represent the speaker’s point of view (as speech acts tend to convey the opinions of their creator, which in this case is the speaker), while quote RPC constructions represent the RPC’s subject’s point of view (as quotes tend to be faithful replicas). In context A, the host in (18) is interpreted as an assertion in which the speaker’s point of view is represented. This interpretation arises because Julie is unlikely to hold the opinion that her best friend is an idiot. In context B, the host in (18) is interpreted as a free indirect quote in which the RPC’s subject’s point of view is represented. This interpretation arises because Julie is likely to use idiotic, while an impartial speaker is unlikely to (unless employed expressively (Potts 2005) or restrictively, idiotic’s use by the speaker violates the Maxim of Quantity in this context). (18) Mary’s idiotic brother, Julie tells me, has eloped with a schoolgirl. Global context: Mary only has one brother. Context A:

The speaker dislikes Mary’s brother. Mary’s brother and Julie are best

Context B:

The speaker is impartial to all parties. Julie dislikes Mary’s brother.

friends.

|| 5 As Collins & Branigan (1997) note, subject-verb inversion is impossible with pronouns in English. Consequently, subject-verb inversion cannot be employed as a disambiguator in the case of (14a).

Speaker andand quote clauses | 79 Parentheticals thereduced grammarparenthetical of complementation

Adult speakers tend to avoid illeism, and quotes tend to be faithful replicas. These two tendencies provide another means of disambiguation that involves pronominal elements. If (19a) were a quote RPC, John would have committed illeism when he originally uttered the assertion depicted by the quote. As this is unlikely, (19a) is most naturally interpreted as a speaker RPC construction. Conversely, if (19b) were a speaker RPC construction, the speaker would violate constraints on specificity that demand that co-referent entities follow an Rexpression > pronoun linear order across utterances.6 As this is unlikely, (19b) is most naturally interpreted as a quote RPC, where the desire to avoid illeism overrides constraints on discourse specificity (where (19b) is interpreted as free indirect quotation). (19) a. Johni will, hei said, be late. b. Hei will, Johni said, be late.

Ambiguous constructions like (14) can also be delimited by their prosody. While the majority of speaker RPCs may be optionally integrated into the prosodic domain of a host constituent (Dehé 2007), there exist a subset of speaker RPCs in English that must be integrated. These are constituent-modifying RPCs (Griffiths 2014). (20) John and I think BillF are coming to the party.

(where F = focus marking)

Quote RPCs are obligatorily isolated from the prosodic domain of the host in English (Reinhart 1983). Consequently, if I think in (20) is pronounced as prosodically isolated, as in (21), only a quote interpretation is obtainable (if at all). (21) ?John and, I think (to myself), BillF is coming to the party.

Finally, context may distinguish between the ambiguous constructions in (14). In neutral contexts, the hosts in (14a-b) are interpreted as assertions. This is because, as a default, speakers assert propositions rather than quote them. However, if (14a-b) are embedded within a chunk of narrative, a quote interpretation is more readily obtained.

|| 6 That I appeal to constraints of discourse specificity to explain the distribution of pronominal elements in (19b) implies that RPCs and their hosts constitute separate utterances related only across the discourse. This is precisely the view for which I argue in §2.

80 | James Griffiths

(22) From a novel yet to be written by this paper’s author: I suddenly awake from my period of narcissistic self-analysis, and consider the plight of my friends. I begin to run through my opinions about the people I love. Firstly, Pete should, I think, quit his job. As I think this, the man himself bursts through the door. He looks excited. Susie has, Pete says, finally finished her thesis. I tell that I am relieved at this news, and suggest we celebrate.

1.4 An aside: point of view and Reinhart (1983) Following Schneider (2007a), I adopted in §1.2 the idea that dissimilar host types (quotes versus non-quotes) is the marker by which RPCs should be split asunder. As §1.3 illustrated, certain epiphenomena follow from this distinction, such as (i) attitude versus action RPC verb, (ii) speaker versus subject oriented point of view, and (iii) pronominal distribution. That RPCs should be delimited according to the type of host they modify does not enjoy unanimous support. Reinhart (1983) adopts point of view – an epiphenomenon according to the current approach – as the marker by which RPCs are split asunder. Unlike the current approach however, Reinhart’s distinction is not jointly exclusive, and is hence a false dichotomy. Joint exclusivity is not achieved because there exist RPC constructions that are neither speaker nor subject oriented, as (23) and (24) illustrate. First, consider the exchange in (23). As (23A´-B´) make clear, (23B) employs a portion of (23A) as an echo, in order to question the presupposition implied by (23A). In (23B), the opinion that the thesis is stupid is held by neither the speaker nor the subject of the RPC (who, in this case, co-refer). (23) A: Is your stupid thesis due in today? B: No – my STUPID thesis is, or so I think, due tomorrow. A´: No need get defensive! It IS stupid! Why would anyone write a book on soil? B´: Evidently you are not an Ecologist!

Second, consider (24). The context for this exchange is the following: B’s supervisor Henry is a well-spoken British professor unlikely to employ American English slang. (24B) makes a joke of this fact, utilising A’s opinion of B’s paper as kick-ass (a slang term) in the quote B attributes to Henry. In (24B) kick-ass is neither the opinion of the speaker nor the RPC’s subject; it is A’s opinion.

Speaker andand quote clauses | 81 Parentheticals thereduced grammarparenthetical of complementation

(24) A: What did Henry say when he read your kick-ass paper? B: “Your kick-ass paper,” said Henry when I saw him today, “is super-awesome.” A´: Very funny… what did he ACTUALLY say?

The existence of ‘unoriented’ RPCs like (23B) and (24B) undermines Reinhart’s attribution of mutually exclusive properties to speaker and subject oriented RPCs. (23B) displays ‘or so’, while (24B) displays subject-verb inversion: properties her approach attributes solely to speaker and subject oriented RPCs respectively. The current approach makes a weaker claim than Reinhart’s, asserting merely that speaker RPC constructions tend to be speaker oriented, while quote RPC constructions tend to be subject oriented. If complicated contexts like those in (23) and (24) override these tendencies, then point of view cannot be employed as a reliable disambiguator of speaker and quote RPCs. However, other cues remain. For instance, that the RPC in (23B) is an attitude RPC suggests that its host is an assertion, while that the RPC in (24B) displays subject-verb inversion suggests that its host is a quote. This type of reasoning is unavailable on Reinhart’s approach, where the existence of ‘unoriented’ RPCs necessitates a tripartite division of RPCs. Thus, to utilise point of view as the marker of superordinate delimitation is to unnecessarily conflate Schneider’s (2007a) dichotomy between quote and non-quote RPC constructions. On the current approach, this conflation is avoided. Thus, the current approach should supplant Reinhart’s.

1.5 Summary: descriptions versus demonstrations RPCs form natural classes according to (i) their attitude versus action specification, and (ii) the type of host they modify. In the remainder of the paper, I aim to demonstrate that one can appeal to (ii) in order to account for the vast majority of dissimilarities observed across RPC constructions. Ultimately, the dissimilarities listed below in §2-3 arise from that fact that quotes constitute distinct communicative acts to all other types of speech acts. Clark & Gerrig (1990) maintain that communicative acts are conveyed by three methods: indication, description, and demonstration. Suppose Lucy wishes John to know how actor x walks. She can point to footage of actor x’s gait (indication), predicate relevant properties of actor x in an assertion (description), or alter her gait to exemplify actor x’s (demonstration). As mentioned in §1.2, quotes are demonstrations. They are vocal exemplifications that depict linguistic objects (LOs). All other speech acts under discus-

82 | James Griffiths sion here are descriptions. To describe, one utilises the content (i.e. the propositionality) of a linguistic object like John is tall. To demonstrate, the content of John is tall is irrelevant. Demonstrations utilise John is tall solely as an entity in the world that consists of a string of sounds whose likeness can be depicted. In this respect, descriptions and demonstrations must differ with respect to their mereology. Assertions and questions, for example, can be composed of assertions and questions (Krifka 2014). Demonstrations, on other hand, are indivisible (i.e. demonstrations are not composed of ‘sub-demonstrations’).

2 The external syntax of RPC constructions In this section, I advance my analysis of how RPCs and their hosts are related to one another. I argue that RPCs are simultaneously clausal adjuncts and independent speech acts, and demonstrate that certain differences in distribution displayed by speaker and quote RPCs can be explained by appealing to the constraints that determine how speech acts are organised at the level of the discourse.

2.1 The proposal I propose that RPCs are clausal adjuncts that do not influence the semantic composition of their host. On such an approach, concatenation of an RPC and a maximal projection γ returns γ unaltered.7 (25) a. John1 didn’t [VP [VP t1 help], [RPC Bill says]].

|| 7 To ensure that the concatenation of the RPC and γ returns γ unaltered, one may either posit a compositional rule that achieves this (as in Potts’ 2005:66 isolated CI application), or posit that the RPC is dominated by a semantically vacuous functional projection that pair-Merges with γ (as in De Vries’ 2007, 2008, 2012 par-Merge approach).

Speaker andand quote clauses | 83 Parentheticals thereduced grammarparenthetical of complementation

b.

TP

(ignoring tensions and tense)

¬help (John) John

NegP ¬help(x1) not

VP help(x1) VP

RPC

help(x1)

says(p, Bill)

t1 help

Bill says p

Because RPCs are not involved in the semantic composition of their host, (25) predicts that RPCs are unaffected by c-commanding operators, and are unsuitable targets for Internal Merge (Chomsky 2004). This prediction is borne out. RPCs escape the scope of sentential negation (26), models (27), and cannot host reflexives (38). Moreover, RPCs are locked islands (to use Postal’s 1998 terminology) (29). (26) John won’t be late, I reckon. Interpretation: a. reckon(p, I)  ¬[will-be-late(John)]

b. * ¬[reckon(p, I)  will-be-late(John)]

(27) John might, I fear, be late. Interpretation: a. fear(p, I)  [be-late(John)]

b. * [fear(p, I)  be-late(John)]

(28) John will, {he/ *himself} thinks, be late. (29) * Who1 will John, t1 thinks, be late?

(25) also engenders a description of how RPCs ‘modify’ their hosts. If RPCs and their hosts are semantically unrelated (as (25) suggests), then they constitute independent speech acts (Potts 2005:68). Thus, (30aA) is the 2-tuple of speech

84 | James Griffiths acts in (30b). The order of the speech acts in (30b) is dictated by derivational timing. Interpretation is bottom-up compositional and therefore dominated XPs (i.e. RPCs) suitable for Transfer (Chomsky 2004) will be Transferred before undominated XPs (i.e. root clauses). If XP α is Transferred before XP β, α precedes β in the discourse structure. (30) a. A: Marxi, Engelsk told the press, wrote most of their manifesto. B: No, hei/#k didn’t. b. [α Engels told the press β.], [β Marx wrote most of their manifesto.]

(30b) is a monologue, albeit a short one. In monologues, utterances are interpreted relative to the context created by the utterances that precede them. In (31) for example, δ’s felicity is contingent upon the truth of γ; γ’s felicity is contingent upon the truth of β; and so on. Moreover, as the monologue continues, the alleged veracity of assertions made ‘early on’ becomes harder to question unless such assertions are targeted specifically (as illustrated by speaker B’s and B´’s responses in (31)). (31) A:

[αThe football season began yesterday.], [β Man United played Chelsea.], [γ The

score was 1-0.], [δ Man United’s new striker scored the game’s only goal.] B: # That’s not true! (referring to β)

B´: While it’s true that Man United won 1-0 and that their new striker scored, it’s NOT true that Man United played Chelsea. They played Arsenal!

In (31), α, β and γ are context restrictors. If quick enough to interrupt A, B may reject A’s proposal to update the context at any juncture: after α is uttered, after β is uttered or after γ is uttered. If no rejection is forthcoming at any of these points, the common ground is updated. But there is no time for B to voice her dissent in (30aA). By virtue of the fact that α and β in (30aA) are voiced synchronously, B can never question the veracity of α before β is uttered (as illustrated by (30aB)). In this respect, the truth of α is imposed on the common ground. In sum: RPCs ‘modify’ their hosts by altering the context in which their host is interpreted.8 || 8 This formulation of RPCs as context restrictors requires abandonment of Potts’ (2005:32) claim that one requires a four-valued system of truth evaluation to interpret utterances that contain parentheticals. According to Potts, parentheticals can be false whilst their host is true. He says that one can still ‘recover’ from (i) that Lance Armstrong has won the 2003 Tour de France, even though the appositional material – i.e. Lance is an Arkansan – is false (he is a Texan). Thus one requires that a system in which the truth values 0 (for the apposition) and 1

Speaker andand quote clauses | 85 Parentheticals thereduced grammarparenthetical of complementation

(25) treats the RPC verb as selecting for a propositional object variable p (in the case of speaker RPCs) or the entity-denoting variable x (in the case of quote RPCs). If RPCs and their hosts are separate utterances, p/x is an anaphor. As an anaphor, p/x’s resolution is subject to the same accessibility restrictions that govern pronouns. According to DRT, the referents of a discourse representation structure (DRS) K1 are accessible from K2 only if K1 = K2 or K1 subordinates K2.9 To provide an example, she cannot corefer with woman in (32) because the referent s cannot find a suitable antecedent within K1. The referent w in K2 is inaccessible to s because the DRS that contains s subordinates K2. (32) John kissed every woman. # She’s happily married. j, s

K1

John(j) happily-married(s) w

K2

woman(w)



kissed(j, w)

s=?

While the p/x of RPCs refers to either propositions or quotes – i.e. DRSs themselves – the constraints on accessibility remain the same. In (33) for example, only K2 provides an accessible antecedent for p. K3, on the other hand, is subordinate to the DRS containing p, and thus an inaccessible antecedent for anaphor resolution.10 (33) [John regrets that [Sam, Mary reckons pi/*k, will be late]k]i

|| (for the host) provided by (i) can still be defined. I disagree, and argue that the host is ‘recoverable’ in (i) only if the apposition’s content is ignored for the truth evaluation of the host. (i) Lance Armstrong, an Arkansan, has won the 2003 Tour de France!

(Potts 2005:32)

9 For formal definition of subordination in DRT, cf. Asher (1993). Here, the idea that K2 is subordinate to K1 if K2 appears inside K1 will suffice. 10 The DRS is hugely simplified, and employs non-standard orthography for exposition. Cf. Asher (1993) for a detailed exposition of precisely how anaphors that denote abstract objects and their antecedents interact.

86 | James Griffiths

m, p Mary(m)

K1

j

K2

John(j)

reckons(m, p) p = K2

regrets(j,

s

K3

)

Sam(s) will-be-late(s)

Anaphora, whether they denote a proposition or an entity, resolve to their most local suitable accessible antecedent. This fact explains certain scope dissimilarities that pertain between speaker and quote RPCs. As mentioned in §1.5, speech acts that are not quotes such as assertions can themselves be composed of assertions. Examples are conjoined assertions (Krifka 2014) (34) and constructions containing asserted adverbial clauses (Reis 2000, Haegeman 2006) (35). (34) a. [α [β Harry kissed Chloe] and [γ Sue, I hope, kissed Nathan]]. Interpretation: b. Harry kissed Chloe and I hope that Sue kissed Nathan. c. * I hope that Harry kissed Chloe and Sue kissed Nathan. (35) a. [α [β Zoë has eaten all the cream cakes], while [γ Tom, I fear, hasn’t eaten any]]. Interpretation: b. Zoë has eaten all the cream cakes, while I fear that Tom hasn’t eaten any. c. * I fear that Zoë has eaten all the cream cakes, while Tom hasn’t eaten any.

If propositional anaphora resolve to their most local suitable antecedent, one expects that the RPCs in (34) and (35) can be interpreted as modifying γ, but not α or β. This expectation is borne out, as (34b-c) and (35b-c) illustrate. As demonstrations, quotes cannot be composed of quotes (see §1.5). This predicts that quote RPCs are always interpreted as modifying the entirety of their host. This prediction is borne out, as (36) and (37) show.11 || 11 Many native speakers of English judge (36c) and (37c) to be entirely acceptable, contrary to my prediction. (36c) and (37c)’s acceptability arise due to the confounding effect of mixed quotation. To illustrate that a quote RPC does indeed modify the entirety of its host – regardless

Speaker andand quote clauses | 87 Parentheticals thereduced grammarparenthetical of complementation

(36) a. [α [β Harry kissed Chloe] and [γ Sue, says Pete, kissed Nathan]]. Interpretation: b. Pete says “Harry kissed Chloe and Sue kissed Nathan.” c. * Harry kissed Chloe and “Sue,” says Pete, “kissed Nathan.” (37) a. [α [β Zoë has eaten all the cream cakes], while [γ Tom, says Frank, hasn’t eaten any]]. Interpretation: b. Frank says “Zoë has eaten all the cream cakes, while Tom hasn’t eaten any.” c. * Zoë has eaten all the cream cakes, while “Tom,” says Frank, “hasn’t eaten any.”

From (25) arise predictions about what types of verbs RPCs may exhibit. For instance, (25) predicts that factive verbs (Hooper & Thompson’s 1973 class C & D verbs) like regret and deny are prohibited as speaker RPC verbs (38a). To see why, consider (39) below. (38) a. # John will, I {deny/regret}, be late. b. (39) a.

[α I {deny/regret} β.], [β John will be late.]

[α John kissed his cousin.], [β He regrets that he kissed his cousin terribly.]

b. # [α John regrets that he kissed his cousin.], [β He kissed his cousin.]

(39a-b) are monologues, like (31A). In (39a), β is felicitously asserted because the verb regret selects for presupposed clauses and the content of regret’s complement – i.e. α – is rendered presupposed by virtue of β being uttered in context in which α’s truth is guaranteed. In (39b) however, β is infelicitously asserted because its content is understood as presupposed in α. By asserting β after α in (39b), the speaker attempts to offer for truth evaluation a proposition whose truth is already taken for granted. (38a) creates the 2-tuple in (38b). In terms of discourse structure, (38b) mirrors (39b): the content of propositional variable in (38b) must be presupposed, and yet the variable’s content (i.e. β) follows α as an assertion. This results in an incoherent discourse. || of what the orthographic quotation marks suggest – consider (i) and (ii). Here it is clear that the quote RPC is permitted only if the entirety of the host is quoted. (i) “That Ii am well-liked at school,” says Maryi, “comforts mei.” (ii) * “That Ii am well-liked at school,” says Maryi, comforts heri.

88 | James Griffiths Note that the same reasoning applies to semifactive RPCs verbs (Hooper & Thompson’s class E verbs). Semifactive verbs are permitted in speaker RPCs only if their complement is not presupposed (compare (40) to (41)). (40) a. I understand that John will be late. (interpretation = John will be late; at least according to my understanding of the situation) b. John will, I understand, be late. (41) a. I understand that two plus two equals four. (interpretation = I have come to sufficiently grasp the undisputed fact that 2+2 =4) b. # Two plus two, I understand, equals four.

Because quotes are demonstrations (and not assertions), the restrictions observed in (38) to (41) do not apply to quote RPCs. Consequently, factive, semifactive and nonfactive verbs are all permitted in quote RPCs (42). If the action denoted by a verb can be interpreted as an action that creates something demonstrable, that verb is permitted in quote RPCs. Thus, intransitive bodilymovement verbs are permitted, yet transitive verbs like forget are not (43). (42) a.

“Ii am not alone in this house,” Johni instinctively knew.

b.

“Ii wish Ii could have saved her,” Johni regrets.

c.

“Ii am not the killer,” Johni denies.

d. (43) a. b.

There are 11 planets, Max realizes.

(Ross 1973:138)

“Ii have,” grins Terryi, “found a great new job”. The pope dies, flashes the neon sign.

(Schneider 2007a:54)

c. * “I am,” forgot Mary, “due in court today.”

The choice of quote RPC verb is not entirely unrestricted, however. It appears that quote RPCs containing verbs like ask and wonder may modify questions alone (44a-b), while quote RPCs containing verbs like declare and assert may not modify questions (44c). One might suggest that (44a-c) show that the variable for which the quote RPC verb selects is sensitive to the [±Q]-status of the host. This suggestion is untenable. (44d) illustrates that verbs like said or whispered – which select only for [–Q] complements in subordination environments (45) – may select for [+Q] hosts. Thus, one cannot appeal to syntactic clause-

Speaker andand quote clauses | 89 Parentheticals thereduced grammarparenthetical of complementation

typing to explain this restriction. Future investigation is required to discover precisely why this restriction holds. (44) a.

“Who will come?” {wonder/asks} John.

b. * “Mary will come,” {wonders/asks} John. c. * “Who will come?” {declares/asserts} John. d.

“Who will come?” {said/whispered} John.

(45) * John {said/whispered} [[+Q] whether Mary will come].

(25) also predicts that quote RPCs cannot interpolate into demonstrations with no internal linguistic structure, as in such hosts no points for adjunction are available. This prediction is borne out, as (46a) illustrates. If used as LOs, such demonstrations’ opacity demands that adjunction is available only at their edge (46b). (46) a. * , goes the rhythm, . b.

[LO [LO ] [RPC goes the rhythm]].

2.2 Summary In §2.1, I outlined my proposal for the external syntax of RPC constructions. This proposal confers a number of benefits. From a conceptual perspective, it maintains Chomsky’s (1995 et seq.) views that (i) the post-syntactic components of the grammar are interpretative alone (i.e. no reorder or insertion operations occur there), and that (ii) a tight correspondence pertains between the syntax and linear order.12 Also, my proposal describes how RPCs ‘modify’ their host in a manner that does not appeal to the apparent exceptional informational status of an RPC as a conventional implicature (Potts 2005) or as information that updates the common ground in a manner distinct to how assertions do (AnderBois et al. 2011). RPCs, like many assertions, are difficult to question solely because of their place in a structured discourse. From an empirical perspective, my proposal correctly accounts for a number of properties (many hitherto undocumented) displayed by speaker and

|| 12 Note that analyses of how RPCs and their host are externally related such as Espinal (1991) and Haegeman (2009) violate condition (i) in the main text, as they each require that the postsyntactic components of the grammar may order or reorder linear strings of lexical items.

90 | James Griffiths quote RPC constructions by utilising general constraints on anaphor resolution and discourse structure. For ease of reference, these properties are recapitulated in table 3. speaker RPCs

quote RPCs

c-command

no c-command relations with host

no c-command relations with host

modification

unable to modify non-root clauses

able to modify all demonstrations (discourse congruity permitting)

Scope

local scope

widest scope

RPC verb

only nonfactive

(almost) no restriction

Prosody

integrated or isolated

isolated

Table 3: Properties of speaker and quote RPCs

3 The internal syntax of RPCs In this section, I advance my analysis of the internal syntax of speaker and quote RPCs. First, let us consider the relevant data. In both speaker and quote RPCs in Dutch and German, the subject and the finite verb must invert. (47) a. Hans hat, {(so) glaube ich (*so) / * ich glaube}, das Auto repariert. Hans has

so believe I

so

I

believe the car

German

repaired

‘Hans has, I believe, repaired the car.’ “Hans hat das Auto repariert,” {(so) sagt Rudi (*so) / *Rudi sagt}. Hans has the car

repaired

so says Rudi so

Rudi says

‘“Hans has repaired the car,” says Rudi.’ b. Joop heeft, {(zo) denkt Jan (*zo) / * Jan denkt}, zijn brood gegeten. Joop has

so thinks Jan

so

Dutch

Jan thinks his bread eaten

“Joop heeft zijn brood gegeten,” {(zo) denkt Jan (*zo) / * Jan denkt}. Joop has

his bread eaten

so thinks Jan

so

Jan thinks

‘Joop’s eaten his bread, Jan thinks.’

In mitigative and quote RPCs, so can be optionally realised, but in the first position alone (47 – 48).

Speaker andand quote clauses | 91 Parentheticals thereduced grammarparenthetical of complementation

(48) a.

John will be late, (so) Pete says (*so).

b. “John will be late,” (so) says Pete (*so).

In both speaker and quote RPCs, unacceptability ensues if the RPC verb is contained within a strong syntactic island.13 (49) a. John will, Mary reckons that Pete said, be late. b. *John will, Mary heard [ISLAND the rumour that Pete says], be late. (50) a. “Ii will,” Mary was told that Petei whispered, “be late.” b. *”Ii will,” Mary was told [ISLAND the claim that Petei whispered], “be late.”

In both speaker and quote RPCs, parasitic gaps are licensable. (51) a. John will be late, Mary instinctively knows _ without articulating _. b. “Ii will be late,” Johni instinctively knew _ without articulating _. c. She instinctively knew it without articulating *(it).

Based on the observations listed in (47) to (51), my analysis of the internal syntax of speaker and quote RPCs is provided in (52). Here, Ø represents the phonologically null counterpart of so, which A´-moves to a topic position within the RPC’s C-domain (cf. Suñer 2000:543).

|| 13 One may argue that the cause of unacceptability in (49b) and (50b) is simply an increase in syntactic complexity (i.e. subordination), rather than an island violation. Examples like (i), where clausal complements are subordinated under bridge verbs, may be utilised to support this view. (i) ? John will, Mary heard that Pete said that many think likely, be late. While syntactic complexity no doubt degrades the acceptability of an RPC, the examples in (49b) and (50b) demonstrate that when only one level of subordination is observed, acceptability judgements differ sharply according to whether an island-boundary is displayed in the RPC or not. If one considers (ii) – (iv), where no A’-movement occurs, one observes no degradation in acceptability between (ii) and (iii) (even though there is an island boundary in (iii)), while one observes degradation in acceptability in (iv), where complexity is increased. These observations, taken together, suggest quite convincingly that island-violations occur in (49b) and (50b). (ii) John will – and Mary heard [ISLAND the rumour that Pete denies it] – be late. (iii) John will – and Mary reckons that Pete denies it – be late. (iv) ? John will – and Mary heard that Pete says that many think it likely – be late.

92 | James Griffiths

(52) a. [{Ø/so}1 [Mary [[says epro] t1]]]

English, Turkish14

b. [{Ø/so}1 [says2 [Mary [[t2 epro] t1]]]] Dutch, German, English quote RPCs (optionally)

Assuming that subject-verb inversion in Dutch and German is triggered by A´movement to the C-domain (Zwart 1997), (52) explains the obligatory subjectverb inversion in (47). That so is optionally displayed in mitigative and quote RPCs but must appear in the first position rather than after the verb (as in (53)) is accounted for by (52), where so or its phonologically null instantiation Ø is treated as a VP-adjunct of say.15 (53) Pete thinks that Nixon was a crook. (*so) Mary thinks *(so), too.

According to (52), the A´-movement of {so/Ø} the C-domain causes the unacceptability observed in (49b) and (50b). Similarly, parasitic gaps are licensed in (51a-b) because of A´-movement. The proposal that A´-movement to the C-domain occurs in speaker and quote RPCs is well-supported by the evidence in (47) to (51). This proposal is further strengthened by the fact that in Irish RPCs COMP must be realised as aL – the realisation that encodes the fact that an A´-moved element (or its trace) occupies the C-domain.16 (54) Bhí an traen mall, a

mheas Klaus.

Irish

Was the train late, COMP thought Klaus ‘The train was late, Klaus thought.’

In (52), {so/Ø} is an adjunct of VP, not the direct object of the RPC verb (contra Suñer 2000, Corver & Thiersch 2001, Van Maastricht 2011). A number of observations support this conclusion. Firstly, so does not receive an objective case in a language that always assigns them, such as Turkish. This suggests that so is not an argument of the RPC verb. || 14 In Turkish, a wh-in-situ language, A′-movement to the C-domain is covert. Thus, öyle ‘so’ is phonologically realised in its base position. 15 It should be noted that so cannot be realised in evaluative or speaker RPCs (i), and yet these RPCs still exhibit the properties listed in (47) to (51). From this observation one must conclude that the phonological realisation of Ø as so is dependent upon some unknown but extraneous factors that are sensitive to the pragmatic function of the RPC’s verb. (i) John will, (*so) I {hope/assert}, be late. 16 Thanks to Jim McCloskey (p.c.) for the Irish data.

Speaker andand quote clauses | 93 Parentheticals thereduced grammarparenthetical of complementation

(55) Ali Ayşe-yi Ali Ayşe-ACC

[Hasan öyle-{*yi/*ye} san-ıyor Hasan so-ACC/DAT

ki] nazikçe öp-tü.

believe-PROG ki gently

Turkish

kiss-PST

‘Ali, so Hasan believes, kissed Ayşe gently.’

Secondly, unlike it, so cannot be the subject of passive RPCs. Moreover, it and so may co-exist in passive RPCs. (56) a. John, {it/ *so} has been said (by Pete), will be late. b. John, (so) it has been said (by Pete), will be late.

Thirdly, so is optionally displayed in quote RPCs that contain bodily-movement verbs such as grin (57a). In utterance-initial RPCs, grin cannot select for a demonstrative object (unlike say) (57b-c). This suggests that in RPCs grin is intransitive. If true, so cannot be the realisation of grin’s direct object, because grin does not select for a direct object. (57) a. “I will,” so grins John, “start my new job tomorrow.” b. Paul says this: “I am glad to be home.” c. Paul grins (*this): “I am glad to be home.”

Fourthly, unintegrated German speaker RPCs (unlike their integrated counterparts) license the correlative pronoun es (which moves to the left periphery of Mittelfeld).17 If so and es were both direct objects, one would expect them to display a complementary distribution. This expectation is not borne out. In such RPCs, so is licensable. If present, so occurs in the first position and triggers subject-verb inversion. (58) Theo kam – so sagt es Paul – mit seinem Hund. Theo came so says it Paul

with his

‘Theo brought – so says Paul – his dog.’

German

dog (Fortmann 2007:99)

In summary: both speaker and quote RPCs are root clauses in which A´movement of a host-denoting element occurs to a topic position within the RPC’s C-domain.18 This A´-movement is evidenced by (i) subject-verb inversion || 17 Cf. Reis (1995, 2000) for the distinction between integrated and unintegrated RPC constructions. 18 Note that this generalised description of the internal syntax of RPCs also applies to [+Q]RPCs like do you reckon in (i). Two potential dissimilarities do pertain between the RPC in (i)

94 | James Griffiths in Dutch and German, (ii) island effects, (iii) parasitic gaps, and (iv) the presence of the aL complementiser in Irish.

4 Extending the analysis in §2-3: utteranceinitial clauses In §2-3, I outlined a generalised analysis of RPCs that treats them as clausal adjuncts that display internal A´-movement of a host-denoting anaphor to a topic position within the RPC’s C-domain. While this approach accounts for many of the properties RPCs display, it cannot be understood as an exhaustive analysis. This is because my analysis so far makes no mention of utteranceinitial clauses such those in (59). In this section, I explore the extent to which utterance-initial clauses can be accommodated into the analysis from §2-3, and whether this particular level of inclusion is predicted by it. (59) a. [INITIAL_CLAUSE I think] [CONSEQUENT_CLAUSE John will be late] b. [INITIAL_CLAUSE Johni thinks] [CONSEQUENT_CLAUSE “Ii will be late”]

The first point to make with regards to the constructions in (59) concerns the diagnostics for parataxis (i.e. ‘RPC-hood’). One must distinguish between diagnostics for parataxis and diagnostics for parataxis in a particular linear position. It was shown in §3 that utterance-medial/final RPCs (i) optionally display so, (ii) obligatorily display subject-verb inversion in Dutch and German, (ii) display island effects, and (iv) license parasitic gaps. Utterance-initial clauses like those in (59) display none of these properties. They cannot display so (in either the first position or VP-internally) (60); subject-verb inversion is prohibited in Dutch and German and in English constructions containing quoted consequent clauses (60); no island effects are observed (61); and parasitic gaps cannot be licensed (62).

|| and its [-Q]-counterparts, however. The first is that wh-movement is focus rather than topic movement (Breul 2004), and the second is that the moved element Ø might be base-generated in the direct object position. (i) Who has, do you reckon, won the prize?

Speaker andand quote clauses | 95 Parentheticals thereduced grammarparenthetical of complementation

(60) a. {(*So) ich glaube (*so) / * glaube ich} es wird regnen. so I

believe so

believe I

German

it will rain

‘I believe it will rain.’ b. {(*Zo) Jan zegt (*zo) / * zegt Jan} “het boek ligt op tafel”.19 so Jan says so

Dutch

says Jan the book lies on table

‘Jan says “the book is on the table”. c. * (So) said John (so): “I will be late.”20 (61) a. Mary heard [ISLAND the rumour that Pete said] John will be late. b. Mary was told [ISLAND the claim that Petei whispered]: “Ii will be late.” (62) a. Mary instinctively knows __ without articulating (*it) that John will be late. b. Johni instinctively knew __ without articulating (*it): “Ii will be late.”

If so is an adjunct in utterance-medial/final RPCs that A´-moves to a topic position within the RPC’s C-domain (see §3), then so’s absence in utterance-initial clauses like those in (60) – and consequently the absence of A´-movement, island effects and parasitic gap licensing – can be easily explained: so cannot be licensed in utterance-initial clauses because (i) so (or its null counterpart Ø) is not a cataphor, and (ii) even if so were a cataphorically licensed, the consequent clause is not a topic, and thus A´-movement of so to a topic-position within the utterance-initial clause would be unnecessary. Resultantly, the dissimilarities between utterance-initial clauses and medial/final RPCs listed in (60) to (62) arise because of the dissimilarity in linear position: utterance-initial clauses precede their ‘host’, while utterance|| 19 In Dutch, zo is permitted VP-internally on its ‘thus’ interpretation (De Vries 2006:222), as illustrated in (i) below. (i) Joop verdedigde zich Joop defended

aldus /zo/ {met deze woorden}: “ Ik heb het niet met opzet gedaan.”

himself thus / so/ with these words

I have it

not

with intent done

‘Joop defended himself with the words, “I did not do it on purpose.”‘

20 The use of subject-verb inversion in utterance clauses like (60c) is not unacceptable in English, but rather archaic. An illustration of this comes from Hesse’s Siddhartha, translated into English in 1951, where subject-verb inversion is observed with the archaic-sounding quoth (i). (i) ‘Quoth Siddhartha: “What should I possibly have to tell you, oh venerable one? Perhaps that you’re searching far too much? That in all that searching, you don’t find the time for finding?”‘

96 | James Griffiths medial/final RPCs do not. Thus, the presence of so, obligatorily subject-verb inversion in Dutch and German, and so on are diagnostics of parataxis in a particular position. The absence of these properties in utterance-initial clauses like (60) bears no relevance to the question of whether the clauses in (60) subordinate or are paratactically related to the clause that they precede (contra Griffiths 2014). If the observations from §3 provide unsuitable diagnostics of parataxis (as has just been shown), we must look to the predictions from §2. The analysis from §2 predicts that, if utterance-initial clauses are RPCs, c-command dependencies cannot be established across the initial/consequent clause boundary. This diagnostic applies straightforwardly to the Turkish data, and illustrates that utterance-initial clauses are indeed RPCs in Turkish. Regular subordination constructions in Turkish display quantifier-binding (63a), wide wh-scope (64a) and exceptional case marking (65a): all dependencies that require c-command (cf. Şener 2010 for arguments that Turkish ECM requires c-command). That ‘enclitic-ki’ constructions do not license any of these dependencies (63b – 65b) illustrates that the enclitic-ki clause does not c-command its consequent clause and is thus an RPC.21 (63) a. Herkesi

söyle-di.22

[prok/i geç gel-eceğ-in]-i

Everyone pro

Turkish

late come-FUT.NOM-POSS-ACC say-PST

‘Everyonei said that (hek/i) will be late.’ b. [Herkesi dedi ki] {o/pro}k/*i geç gel-ecek everyonei said ki he/pro

late come-FUT

‘Everyone said (hek/*i) will be late.’ (64) a. Ahmet-in

[kim-i

öp-tüğ-ü-n]-ü

san-ıyor-sun?

Ahmet-GEN who-ACC kiss-NOM-POSS-ACC believe-PROG-2S ‘Whom1 do you believe Ahmet kissed t1?’ b. *[San-ıyor-sun

ki] Ahmet kim-i

öp-tü?

believe-PROG-2S ki Ahmet who-ACC kiss-PST ‘Whom1 do you believe ki Ahmet kissed t1?’

(intended)

|| 21 For evidence that (A) enclitic-ki clauses adjoin to their consequent clauses, rather than are coordinated with them (as Kesici 2013 maintains), and (B) the tests in (63) to (65) apply equally to constructions containing quoted consequent clauses, cf. Griffiths & Güneş (2014). 22 Only pro or the reflexive kendi can be bound by universal quantifiers in Turkish.

Speaker andand quote clauses | 97 Parentheticals thereduced grammarparenthetical of complementation

(65) a. Merve

[ben-Ø/i

plaj-a

git-ti-m] san-ıyor.

Merve-NOM I-NOM/ACC plaj-DAT go-PST-1S believe-PROG ‘Merve believes that I went to the beach.’ b. *[Merve san-ıyor

ki] ben-Ø/*i

plaj-a

git-ti-m.

Merve believe-PROG ki I-NOM/ACC plaj-DAT go-PST-1S ‘Merve believes I went to the beach.’

Of course, the application of these and similar c-command tests to English, Dutch, and German and analyses that arise from the results obtained therefrom pervade the generative literature. While there is general accord that utteranceinitial clauses like (59b) are quote RPCs (cf. Banfield 1982 and De Vries 2006, among others) – as evidenced in part by the fact that quotes are locked islands (66) and arise ‘complete’ (67) – the conclusion that utterance-initial clauses like (59a) are speaker RPCs enjoys little support (but cf. Schneider 2007a:177–184). (66) *Who1 did Johni say “Ii have met t1”? (67) a. John read the recipe to Mary: “flour, sugar, butter and eggs, all in equal measure. Stir.” b. Arniei said (this): “Ii’ll be back.” c. Terry grinned (*this): “I’ve finally finished my thesis!”

As is well-known, the class of verbs that purportedly c-select for embedded root clauses in German equate roughly to those permitted in utterance-medial/final RPCs (i.e. nonfactives and certain semifactives) (cf., for example, Reis 1997 and Scheffler 2009). This observation and others lead Reis (1997) to the conclusion that so-called ‘embedded root clause’ constructions in German (68a) are actually paratactically related to an utterance-initial speaker RPC (68b). (68) a. [CP1 Ich glaube, [CP2 Maria hat Fieber]].

German

b. [[CP Ich glaube,] [CP Maria hat Fieber]]. I

believe

Maria has fever

‘I think Mary has a fever.’

To what extent the same analysis can be extended to English and Dutch is uncertain. For instance, standard Dutch does not permit embedded root clauses whatsoever. If parenthetical (i.e. evidential) function and paratactic syntax are in any sense related, one expects the absence of semantic bleaching of bang

98 | James Griffiths ‘afraid’ in a construction like (69). The expectation is not borne out, however – bang can indeed be understood as bleached. Such an observation might make one wonder whether V2/V-final word order is a reliable indicator of hypotaxis/parataxis, or whether some extraneous factor (perhaps the presence or absence of the complementiser) is the true trigger for V-finality. (69) Ik ben bang dat je

kat dood is.

Dutch

I am afraid that your cat dead is ‘I’m afraid that your cat is dead.’

Such issues and speculations thereon will continue to arise if one adopts the analysis in §2, as, according to this approach, the existence of utterance-initial RPCs is expected. Nothing rules them out. While Turkish displays both utterance-initial speaker and quote RPCs, only quote RPCs are evidenced in English, Dutch, and German (from a consensus viewpoint, at least). Thus, the current approach to RPCs is incomplete in one of two ways. Either (i) additional mechanisms must be sought to adequately explain why utterance-initial speaker RPCs are unevidenced in English, Dutch, and German, or (ii) additional evidence must be sought to show that a subclass of what look like subordination constructions in English, Dutch, and German are actually root clauses with utterance-initial speaker RPCs attached to them. Whether (i) or (ii) is correct must remain an issue for future investigation. To summarise: the analysis outlined in §2-3 predicts the existence of utterance-initial RPCs. In this section, I demonstrated how the analysis of the internal syntax of utterance-medial/final RPCs from §3 impedes one’s discovery of utterance-initial RPCs, as the analysis in §3 renders many of the internal dissimilarities displayed by utterance-initial clauses and utterance-medial/final RPCs irrelevant to the evaluation of the ‘RPC-hood’ of utterance-initial clauses. Without having these tools available, diagnosing parataxis in utterances like those in (59) is challenging. Thus, future investigation must decide whether the current analysis is correct or incorrect to predict the presence of RPCs in all linear positions.

5 Conclusion I have endeavoured in this paper to generalise over RPC constructions from English, Dutch, German and Turkish. I argued that RPCs are clausal adjuncts whose adjunction has no semantic import on the truth conditions of their host.

Speaker andand quote clauses | 99 Parentheticals thereduced grammarparenthetical of complementation

RPCs ‘modify’ their host solely by restricting the context in which their host is interpreted to that context in which the proposition denoted by the RPC is true. That speaker RPCs are utterance-bound while quote RPCs modify the entirety of their host was explained by appealing to general mechanisms on anaphor resolution across the discourse and by appealing to the fact that speech acts like assertions can be composed of assertions, while quotes – as demonstrations – cannot be composed of quotes. I also argued that RPCs are independent clauses which either (i) display A´-movement of a host-denoting anaphor to a topic position (if the RPC interpolates into or follows its host), or (ii) displays no A´movement or host-denoting anaphor (if the RPC linearly precedes its host). The upshot of my analysis of the internal syntax of RPCs is that their internal properties have no bearing upon their external relation to their host – i.e. one cannot appeal to the idiosyncrasies of their internal syntax to support a paratactic analysis of RPCs. The major consequence of this is that the current approach requires that both utterance-initial speaker and quote RPCs should be observed in each of the languages under investigation. While this requirement can be readily demonstrated to be fulfilled in Turkish, it is difficult to judge whether it is met in English, Dutch, or German. Further research is required to discover whether this requirement is met in these languages, or whether the current requires amendment (or abandonment) if it is shown that this requirement can never be met.

100 | James Griffiths

References AnderBois, Scott, Adrian Brasoveanu & Robert Henderson. 2013. At-issue Proposals and Appositive Impositions in Discourse. Journal of Semantics. Asher, Nicholas. 1993. Reference to abstract objects in discourse. Studies in Linguistics and Philosophy. Vol. 50. Dordrecht: Kluwer. Banfield, Anne. 1982. Unspeakable sentences: Narration and representation in the language of fiction. London: Routledge & Kegan Paul. Breul, Carsten. 2004. Focus Structure in Generative Grammar: An integrated syntactic, semantic and intonational approach. Amsterdam: John Benjamins. Chomsky, Noam. 1995. The Minimalist Program. Cambridge: MIT Press. Chomsky, Noam. 2004. Beyond explanatory adequacy. In Adriana Belletti (ed.). Structures and beyond. The cartography of syntactic structures. Vol. 3. Oxford: Oxford University Press. 104–131. Clark, Herbert & Richard Gerrig. 1990. Quotes as demonstrations. Language 66. 764–805. Collins, Christopher & Phillip Branigan. 1997. Quotative inversion. Natural Language and Linguistic Theory 15. 1–41. Corver, Norbert & Craig Thiersch. 2001. Remarks on parentheticals. In Mark van Oostendorp & Elena Anagnostopoulou (eds.). Progress in grammar: Articles at the 20th anniversary of the comparison of grammatical models group in Tilburg. Utrecht: Roquade. Dehé, Nichole. 2007. The relation between syntactic and prosodic parenthesis. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam: John Benjamins. 261–284. Espinal, M. Teresa. 1991. The representation of disjunct constituents. Language 67. 726–762. Fortmann, Christian. 2007. The complement of reduced parentheticals. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam: John Benjamins. 89–119. Griffiths, James. 2014. Parenthetical verb constructions, fragment answers, and constituent modification. Natural Language and Linguistic Theory. [link.springer.com/article/10.1007/s11049-014-9256-6, accessed 10/14] Griffiths, James & Guliz Güneş. 2014. Ki issues in Turkish: Parenthetical coordination and adjunction. In Marlies E. Kluck, Dennis Ott & Mark de Vries (eds.). Parenthesis and Ellipsis: Cross-Linguistic and Theoretical Perspectives. Berlin: Mouton de Gruyter. 109–145. Haegeman, Liliane. 2006. Conditionals, factives and the left periphery. Lingua 116 (10). 1651– 1669. Haegeman, Liliane. 1991/2009. Parenthetical Adverbs: The radical orphanage approach. In Benjamin Shaer, Philippa Cook, Werner Frey & Claudia Maienborn (eds.). Dislocated Elements in Discourse. New York: Routledge. 331–347. Hooper, Joan & Sandra A. Thompson. 1973. On the Applicability of Root Transformations. Linguistic Inquiry 4. 465–497. Kamp, Hans. 1981. A theory of truth and semantic representation. In Jeroen Groenendijk, Theo Janssen & Martin Stokhof (eds.). Formal Methods in the Study of Language. Amsterdam: Mathematisch Centrum. 277–322. Kesici, Esra. 2013. Ki-Clauses in Turkish: A Paratactic Analysis. In Hyun Kyoung Jung & Jessamyn Schertz (eds.). Proceedings of the Arizona Linguistics Circle 6. Tucson: University of Arizona.

Speaker andand quote clauses | 101 Parentheticals thereduced grammarparenthetical of complementation

Krifka, Manfred. 2014. Embedding speech acts. In Tom Roeper & Peggy Speas (eds.). Recursion: Complexity in Cognition. Studies in Theoretical Pyscholinguistics 43. Dordrecht: Springer. 59–87. Maastricht, Lieke van. 2011. Reporting and Comment clauses: A cross-linguistic study. MA thesis: University of Groningen. Maier, Emar. To appear. Quotation and Unquotation in Free Indirect Discourse. Mind & Language. Postal, Paul. 1998. Three Investigations of Extraction. Cambridge: MIT Press. Potts, Christopher. 2005. The Logic of Conventional Implicatures. Oxford: Oxford University Press. Reinhart, Tanya. 1983. Point of View in Language – The Use of Parentheticals. In Gisa Rauh (ed.). Essays on Deixis. Tübingen: Müller & Bass. 169–194. Reis, Marga. 1995. Wer glaubst du hat recht? On so-called extractions from verb-second clauses and verb-first parenthetical constructions in German. Sprache & Pragmatik 36. 27–83. Reis, Marga. 1997. Zum syntaktischen Status unselbständiger Verbzweit-Sätze. In Christa Dürscheid, Karl Heinz Ramers & Monika Schwarz (eds.). Syntax im Fokus: Festschrift für Heinz Vater. Tübingen: Niemeyer. 112–144. Reis, Marga. 2000. Wh-movement and integrated parenthetical constructions. In Jan-Wouter Zwart & Werner Abraham (eds.). Studies in Comparative Germanic Syntax, Proceedings from the 15th workshop on Comparative Germanic Syntax. Amsterdam: John Benjamins. 3– 40. Reis, Marga. 2002. On the Parenthetical Features of German Was…W-Constructions and How to Account for Them. In Uli Lutz, Gereon Müller & Arnim von Stechow (eds.). Wh-Scope Marking. Amsterdam: John Benjamins. 359–407. Rooryck, Johan. 2001. State of the article: Evidentiality Part I & II. GLOT International (4&5). 125–133 & 161–168. Ross, John Robert. 1973. Slifting. In Maurice Gross, Morris Halle & Marcel Schützenberger (eds.). The Formal Analysis of Natural Languages. Amsterdam: Rodopi. 147–159. Scheffler, Tatjana. 2009. Evidentiality and German Attitude Verbs. In Laurel MacKenzie (ed.). Proceedings of the 32nd Annual Penn Linguistics Colloquium Volume 15(1). University of Pennsylvania Working Papers in Linguistics. Schneider, Stefan. 2007a. Reduced parenthetical clauses as mitigators. A corpus study of spoken French, Italian and Spanish. Amsterdam: John Benjamins. Schneider, Stefan. 2007b. Reduced parenthetical clauses in Romance languages. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam: John Benjamins. 237–257. Sharvit, Yael. 2008. The puzzle of free indirect discourse. Linguistics and Philosophy 31. 353– 395. Steinbach, Marcus. 2007. Integrated parentheticals and assertional complements. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam: John Benjamins. 53–88. Suñer, Margarita. 2000. The Syntax of Direct Quotes with Special Reference to Spanish and English. Natural Language & Linguistic Theory 18. 525–578 Şener, Serkan. 2010. Cross Clausal Licensing of Accusative Case on Subjects of CPs in Turkish. In Suzi Lima, Kevin Mullin & Brain Smith (eds.). Proceedings of the 39th Annual Meeting of the North East Linguistics Society. GLSA: Amherst. Vicente, Luis. 2013. In search of a missing clause. Paper presented at the Parenthesis and Ellipsis workshop of the 34th Annual Meeting of the German Society of Linguistics. Potsdam, March 2013.

102 | James Griffiths Vries, Mark de. 2006. Reported Direct Speech in Dutch. Linguistics in the Netherlands 23. 212– 223. Zwart, C. Jan-Wouter. 1997. Morphosyntax of Verb Movement. A Minimalist Approach to the Syntax of Dutch. Dordrecht: Kluwer.

Appendix (A1) a. John, I think, will be late.

[Assertive host, mitigative RPC]

b. John, I swear, will be late.

[Assertive host, speech act RPC]

c. John, I hope, will be late.

[Assertive host, evaluative RPC]

(A2) a. Will John be late, do you think?

[Erotetic host,mitigative RPC]

b. # Will John be late, do you assert?

[Erotetic host,speech act RPC]

c. # Will John be late, do you hope?

[Erotetic host,evaluative RPC]

(A3) a. # I hereby dub this store open, I reckon.

[Declarative host, mitigative RPC]

b. I hereby dub this store open, I declare.

[Declarative host, speech act RPC]

c. # I hereby dub this store open, I fear.

[Declarative host, evaluative RPC]

(A4) a. # I do swear that I will uphold the law, I believe. [Commissive host, mitigative RPC] b. I do swear that I will uphold the law, I declare.

[Commissive host, speech act RPC]

c. # I do swear that I will uphold the law, I prey.

[Commissive host, evaluative RPC]

(A5) a. # Live long, I’m told, and prosper!

[Optative host,mitigative RPC]

b. # Live long, I assert, and prosper!

[Optative host,speech act RPC]

c. Live long, I wish, and prosper!

[Optative host,evaluative RPC]

Marlies Kluck and Mark de Vries

On V2, gaps, and operators in comment and reporting parentheticals Abstract: Parenthetical comment and reporting clauses exhibit various interesting properties, which are shown to be related. Three different word order patterns are attested in Dutch and German: apparent V1, V2 and V-final; the internal argument of the parenthetical verb is usually missing; there is an optional so/zo element; the construction is island-sensitive. These patterns are explained by means of an operator that is A´-moved inside the parenthetical, and which can optionally be lexicalized in the first position. The V-final pattern arises when a complementizer is present, which is the case in propositional asparentheticals. Keywords: comment clause, reporting clause, as-parenthetical, operator, anaphoric so, word order, verb second (V2)

|| Marlies Kluck: Department of Linguistics, University of Groningen. Groningen, The Netherlands. Mark de Vries: Department of Linguistics, University of Groningen. Groningen, The Netherlands.

1 Introduction and brief overview We investigate the internal syntax of parenthetical comment and reporting clauses, and potential generalizations over various subtypes.* Thereby, we discuss word order patterns, the nature of possible argument gaps, and the interpretation and position of elements such as zo/so and zoals/wie/as primarily in Dutch, while pointing out striking parallels with German and English, as well as some differences between the three languages. A classic puzzle concerns the position of the finite verb in intercalated or utterance-final comment clauses. Three variants are attested: apparent V1, V2, || * Thanks to the organizers and the audience of Parenthétiques 2012 (Paris Nanterre), and in particular to Stefan Schneider. We are grateful to the anonymous reviewers for useful comments and questions. This research was carried out as part of the project Incomplete Parenthesis, financially supported by the European Research Council.

104 | Marlies Kluck and Mark de Vries and V-final (modulo extraposed elements). This is illustrated in Dutch in (1a-c), where the relevant verb is underlined. Similar word order patterns are used in German. (1) a. Bob is, vermoed ik, een echte charmeur. Bob is suspect

I

a

true charmer

‘Bob is, I suspect, a true charmer.’ b. Bob is, zo vermoed ik, een echte charmeur. Bob is so suspect I

a

true charmer

‘Bob is, so I suspect, a true charmer.’ c. Bob is, zoals ik al Bob is as

I

vermoedde, een echte charmeur.

already suspected

a

true charmer

‘Bob is, as I already suspected, a true charmer.’

Such comment clauses, and reporting clauses likewise, have a parenthetical status. The zero hypothesis, then, is that they are main clauses, which trigger verb second in Dutch, unless there is an internal complementizer that gives rise to a subordinate clause pattern with a final verb. Indeed, examples (1a/b) clearly show an interrupting main clause with an optional first element zo, which we analyze as a kind of operator. Section 2 works out these claims. Section 3 highlights the problem of the base position of the zo-operator, and discusses its interpretation and status in more detail. Section 4 concerns the parenthetical type in (1c), which is somewhat different. Here, the word order is that of a subordinate clause. This is due to the fact that the complementizer position is occupied by the comparative als. We argue that this element may fuse with the moved operator zo. Furthermore, we make a comparison of the Dutch construction type with English as-parentheticals and German wie-parentheticals. Section 5 is the conclusion.

2 Verb second and the operator zo/so/ø in comment and reporting clauses 2.1 Introductory remarks and delimitations Comment and reporting clauses can occur in various positions, as is illustrated in (2a-c). In (2a) the assertion can be that Bob is a real charmer (depending on the context); in (2b/c) this is necessarily the case. However, there are more im-

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 105

portant qualitative differences between the initial position on the one hand, and the medial or non-derived final position on the other hand. The last two involve parenthetical insertion of the comment/report clause within the host sentence, whereas (2a) involves regular subordination of the communicated proposition under the commenting/reporting verb.1 It is only the parenthetical construction type that we are interested in, for reasons that will become clear shortly. (2) a. i)

I suspect (that) Bob is a real charmer.

[initial position]

ii) Anna said: “Bob is a real charmer”. iii) Anna said (that) Bob is a real charmer. b. i)

Bob, I suspect, is a real charmer.

[medial position]

ii) “Bob”, said Anna, “is a real charmer”. c. i)

Bob is a real charmer, I suspect.

[final position]

ii) “Bob is a real charmer”, said Anna.

As a preliminary precaution, note that a final position of the comment or report clause can sometimes be derived from the initial position simply by topicalizing the object clause (of course, it requires some context to make this felicitous). This is not what we are after; see further below, and see also Griffiths (this volume) for relevant discussion of such potential confusion. Notice, incidentally, that contrastive topics or preposed foci can easily be detected by a so-called ‘Baccent’ on the preposed object, that is, an L+H*L-H% intonational contour, as was pointed out already in Jackendoff (1972). In Dutch, there are clear (morpho)syntactic differences between main and subordinate clauses, which makes it easier to distinguish the relevant cases than in English. In (3a), there is an obligatory complementizer dat ‘that’, and the finite verb is ‘is’ is in the final position; this is typical for subordinate clauses. In (3b), the comment clause happens to be linearly final. Still, it is clearly the matrix itself, considering that the first (associated) clause is formally a subordinate clause in every respect, similarly to the situation in (3a). Therefore, (3b)

|| 1 In various examples, we transparently use the verb say as a (directly) reporting verb, but evidently it can also function as a parenthetical verb in a comment clause, depending on the context. In (i), for instance, said Anna can be taken to mean ‘according to Anna’. The exact phrasing of the host is, however, the responsibility of the current speaker, not Anna. (i) Bob is a real charmer, said Anna. In other words, a parenthetical verb construction is necessarily an epistemic or evidential comment if the host is not a direct quote. We will return to this issue.

106 | Marlies Kluck and Mark de Vries must involve clausal topicalization, and we can set it aside as a derivational variant of (3a) that is irrelevant to the present discussion. (3) a. Ik vermoed dat Bob een echte charmeur is. I

suspect that Bob a

true charmer

[initial position  subordinative]

is

b. Dat Bob een echte charmeur is vermoed ik. [derived final pos  subordinative] that Bob a

true charmer is suspect I

By contrast, the comment clauses in (4a/b) are parenthetically construed in a sentence-medial or final position, and hence do not directly participate in the syntax of the host clause. Here, the host is a main clause, as is evidenced by V2 and the absence of a complementizer.2 (4) a. Bob is, vermoed ik, een echte charmeur. Bob is suspect I

a

b. Bob is een echte charmeur, vermoed ik. Bob is a

[parenthetical]

true charmer

true charmer

[non-derived final pos parenthetical]

suspect I

Let us briefly highlight four differences between the subordinative and the parenthetical verb construction, illustrated in (3) versus (4):3

|| 2 Depending on the intonation (recall the remarks below (2c)), the word order pattern in (4b) might alternatively be derived by topicalizing an embedded V2 clause without a complementizer. Although this cannot be excluded for certain examples in German (e.g., vermuten ‘suspect’ allows for an embedded V2 complement clause), it is highly unlikely in Dutch, where apparent cases of embedded V2 are much less frequent, and generally perceived as a performance-related restart of the sentence rather than a grammatical construction type. In addition, we should note that Ross’s (1973) famous ‘slifting’ (sentence lifting) analysis is in fact untenable for parenthetical comment and reporting clauses. First, this would require stipulative operations that move a regular subordinate clause from its purported base position as the complement of the ‘parenthetical’ verb, formally turn it into a main clause and change the original matrix into a parenthetical – hence transforming (3a) into (4b), for instance. Second, intermediate positions for parentheticals, as in (4a), can simply not be derived in this way. For more discussion, see Reis (1995, 2002), Van Maastricht (2011), and Griffiths (this volume), among others. 3 In this contribution, we do not consider the possibility of ‘initial parentheticals’, but see Blanche-Benveniste (1989), among others, for a different take on the (then apparent) subordinative construction. For the reasons indicated in the main text, it is clear, however, that the two construction types have different properties, and our analysis only targets the parenthetical one.

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 107

−

−

−

−

The verb vermoeden ‘suspect’ is obligatorily transitive. In (3), it takes a clausal object, and the sentence is overtly complete. In (4), however, the internal argument of the parenthetical verb seems to be lacking. Example (3a) displays the standard word order pattern in simple main clauses: S–V2–O; in (3b) there is subject-verb inversion due to topicalization. Surprisingly, the parenthetical comment clauses in (4a/b) also show obligatory subject-verb inversion, resulting in V1. The subordinative construction in (3) leads to prosodic integration of the embedded material with the matrix verb; here, the object clause contains the sentence accent. By contrast, parentheticals as in (4) can always be prosodically isolated, and, crucially, they never attract the sentence accent, independently of the question whether they contain a pitch accent of their own.4 The subordinative construction has a canonical word order as in (3a), with the selecting verb adjacent to the object clause. Their relative positions can only be altered due to well-defined movements of either the verb (say, V2 in more complex main clauses) or the object clause (e.g., topicalization as in (3b)). By contrast, in parenthetical verb constructions there is no designated syntactic position for comment clauses: like regular parentheticals, they have no predefined structural position. Where exactly they surface is primarily determined by pragmatic and prosodic considerations. Next to the final position in (4b), various intermediate positions are possible, as in (4a), for instance.5

Thus, it is clear that parenthetical comment clauses must be distinguished from subordinative constructions, and also that they display properties that require further explanation. Analogously, we can safely assume the same for reporting clauses. Comment clauses can sometimes be interpreted at the constituent level rather than at the sentence level, depending on the right intonational grouping. This is illustrated in (5) and (6) for English and Dutch, respectively:

|| 4 See Dehé (2009) and Güneş & Çöltekin (this volume) for current studies on the prosody of parentheses. 5 We will not elaborate on this here. Some frequency of use effects concerning various potential positions in Dutch and German are reported in Schelfhout, Coppen & Oostdijk (2004) and Stoltenburg (2003), respectively.

108 | Marlies Kluck and Mark de Vries

(5) a. John gave the book to [I think Mary], yesterday. b. John gave the book to [Mary, I think,] yesterday. (6) a. Jan heeft het boek gisteren aan [ik dacht Jan has

the book yesterday to

Marie] gegeven.

I thought Marie given

b. Jan heeft het boek gisteren aan [Marie, dacht Jan has

the book yesterday to

ik,] gegeven.

Marie thought I

given

Griffiths (to appear) argues that such intrusive constructions are ‘fragment amalgams’, in line with Kluck’s (2011, 2013) analysis of Horn amalgamation, which involves clausal ellipsis (cf. I think it was Mary who John gave the book to yesterday). In addition, he distinguishes between the subordinative and the parenthetical kind of fragment amalgam, parallel to the difference between (3) and (4) above. Accordingly, only (5b) and (6b) involve parenthetical attachment of the comment clause within its local syntactic environment – irrespectively of how complex that might be. What is relevant for us here, is that the internal structure of a parenthetical comment clause as dacht ik ‘I thought’ in (6b), for instance, is equivalent to that in (4). Finally, it is worth mentioning that the ‘gap’ in reporting and comment parentheticals does not necessarily involve the direct object; in exceptional cases, it can also be the subject, as is illustrated with Dutch examples in (7): (7) a. “Bob is”, (zo)werd gezegd, “een echte charmeur.” Bob is

so was said

a

true charmer

‘“Bob is”, so [it] was said, “a true charmer”‘ b. Bob is, (zo) wordt algemeen aangenomen, een echte charmeur. Bob is so is

generally assumed

a

true charmer

‘Bob is, so [it] is generally believed, a true charmer.’ c. Bob had, (zo) was wel

duidelijk, een blunder begaan.

Bob had so was AFFIRMATIVE clear

a

blooper done

‘Bob had, [this] was clear, made a blooper.’ d. Bob had, zoals wel Bob had as

AFFIRMATIVE

duidelijk was, een blunder begaan. clear

was a

blooper done

‘Bob had, as was clear, made a blooper.’

The first two examples involve a passive construction, and the last two a copular construction with a propositional subject. In each case, the gap still counts as the internal argument of the predicate. For obvious reasons, a parenthetical

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 109

gap cannot correspond to an external argument, since propositions or speech acts cannot be semantic Agents. The construction types mentioned in (5) through (7) do not affect the analysis to be developed below, and we will set them aside in order to prevent unnecessary complications of exposition. In the remainder of this contribution, we will focus on the internal syntax of parenthetical reporting and comment clauses. We will simply take for granted that parentheticals are non-restrictive additions to the sentence, but evidently the external syntax of parentheticals more generally is a topic of interest as well; see the introductions in Dehé & Kavalova (2007) and Kluck, Ott & De Vries (2014) for a general overview, and see De Vries (2012a), Griffiths & De Vries (2013), and Kluck (to appear) for a more particular take on the matter.6

2.2 Word order and the optional presence of zo/so With these preliminaries and delimitations in mind, let us now return to the primary examples, repeated in (8) for convenience: (8) a. Bob is, (zo) vermoed ik, een echte charmeur. Bob is so suspect I

a

true charmer

b. “Bob is”, (zo) zei Anna, “een echte charmeur”. Bob is

so said Anna

a

true charmer

A number of questions concerning the internal structure of the parenthetical comment clause immediately arises: − − −

How can we explain the word order? Where is the internal argument? What is the status of the element zo?

|| 6 We envision structural incorporation of parentheses as a kind of non-restrictive adjuncts with respect to the host by means of a distinct syntactic operation (Parenthetical Merge) that enables linear integration at the PF-interface, but blocks c-command relationships, thus ensuring semantic ‘orphanage’ in effect.

110 | Marlies Kluck and Mark de Vries In the remainder of this section and the next one, we will address these issues in more detail, starting with word order. In the parentheticals in (8), the finite verb and the subject are inverted. In Dutch, this is the only possible order:7 (9) a. * Bob is, ik vermoed, een echte charmeur. b. * Bob is, zo ik vermoed, een echte charmeur. (10) a. * “Bob is”, Anna zei, “een echte charmeur.” b. * “Bob is”, zo Anna zei, “een echte charmeur.”

By extending the comment or reporting clause, it can easily be shown that it is a main clause. As already indicated in the introduction, this is also what one would expect by default for a parenthetical. In (11), adverbials follow the inverted subject, suggesting that the verb is in the regular position for finite verbs in main clauses. If so, we expect it to shift to the final verb position when we insert an additional auxiliary. This is indeed the case, as is shown in (12). (11) a. Bob is, (zo) vermoed ik al

lange tijd, een echte charmeur.

Bob is so suspect I already long time a

true charmer

‘Bob is, I’ve suspected for a long time now, a true charmer.’ b. “Ik wil I

met je

dansen”, (zo) zei

want with you dance

Bob gisteren.

so said Bob yesterday

‘“I want to dance with you”, Bob said yesterday.’ (12) a. Bob is, (zo) heb ik al

lange tijd vermoed, een echte charmeur.

Bob is so have I already long time suspected a b. “Ik will met je dansen”, I want with you dance

true charmer

(zo) heeft Bob gisteren gezegd. so has Bob yesterday said

The word order pattern with both left and right verbal ‘sentence brackets’ (Dutch ‘polen’, German ‘Satzklammern’) is typical for main clauses. In a subordinate clause all verbs surface in the final position:8

|| 7 In English, subject-verb inversion in such parentheticals is dependent on the predicate, and it is never obligatory. It can be considered a remnant of V2 in an earlier stage of the language. For an illustration, see (16) below in the main text. 8 Recall that in V2 languages, finite verbs are – at least descriptively – in complementary distribution with complementizers. See Den Besten (1977) and Zwart (1994) for elaborate discussion.

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 111

(13) ... dat Bob dat gisteren gezegd heeft. that Bob that yesterday said

has

In comment and reporting clauses, the element zo always surfaces in the first position, immediately followed by the finite verb. This implies that an adverbial cannot precede zo or intervene between the fixed elements; see (14), for instance: (14) Bob is, (*eigenlijk) zo(*eigenlijk) vermoed ik (eigenlijk), een echte charmeur. Bob is actually so actually suspect I

actually

a

true charmer

Evidently, these parentheticals show regular V2 effects associated with main clauses. However, if zo is absent, the verb is linearly first inside the parenthetical. As we have seen, the presence of zo is optional in Dutch. This is similar in German:9 (15) a. Hans, (so) glaubt

Martin, möchte

das Theorem beweisen.

Hans so believes Martin would.like the theorem prove ‘Hans, Martin thinks, would like to prove the theorem.’ b. “Hans ist”, (so) sagte Martin, “ein richtiger Charmeur”. Hans is

so said Martin

a

true

charmer

‘“Hans is”, (so) said Martin, “a true charmer”.’

|| 9 We are aware of one potential counterexample, where the purported comment clause figures inside a question (cf. Steinbach 2007). The sentence is acceptable only without so: (i) Welches theorem,(*so) glaubt Hans, möchte Martin beweisen? which theorem so believes Hans would.like Martin prove

[German]

The reason might be that an overt anaphoric relationship with an open set of referents is inherently problematic. However, it is not immediately clear to us why an alternative analysis in terms of extraction from an embedded V2 clause would be excluded here. If so, there is no comment clause to begin with, hence no so; and it is also clear why the equivalent in Dutch is eschewed altogether (as Dutch virtually lacks embedded V2; cf. footnote 2). Furthermore, we would predict that a different site of attachment is excluded, which is correct: (ii) *Welches Theorem möchte, glaubt Hans, Martin beweisen? This contrasts with the situation in declarative contexts. For instance, (15a) can be rephrased as (iii): (iii) Hans möchte, (so) glaubt Martin,das Theorem beweisen. Hans would.like so believes Martin the theorem prove

112 | Marlies Kluck and Mark de Vries In English, the use of so is more restricted due to the fact that it can only surface with those predicates that allow for subject-verb inversion, which is relatively rare because of the general loss of V2 in this language. Nevertheless, in the relevant cases, so is optionally present as well (albeit slightly archaic): (16) a. “John is”, (so) said Mary, “a real charmer”. b. “John is”, (so) Mary said, “a real charmer”.

Thus, the pattern is quite consistent. In the absence of zo/so, the verb is arguably still in the same structural position. Consider the Dutch data. First, there is S–V inversion, and second, nothing else can be preposed/topicalized. Compare (17) to (14), for instance: (17) Bob is, (*eigenlijk) vermoed ik (eigenlijk), een echte charmeur. Bob is actually suspect I

actually

a

true charmer

By contrast, in regular declarative main clauses, adverbs can be put in the first position without any difficulty: (18) Eigenlijk vermoedde ik dat al. actually suspected I that already ‘Actually, I already suspected that.’

An obvious way to account for all these data is to assume that the relevant zoless parentheticals are only apparently V1, and have the same syntactic structure as V2 parentheticals. This implies that there is a silent counterpart of zo/so. Thus, we postulate a clause-initial operator (OP) inside the parenthetical, in line with ideas by Collins & Branigan (1997) and various others concerning a ‘quotative operator’ in reporting clauses. This immediately explains the obligatory inversion of the subject and the finite verb in V2 languages, and we can maintain that the parenthetical is a bona fide V2 main clause: (19) Bob is, OP/zo vermoed ik, een echte charmeur.

To be perfectly clear, we cannot assume an abstract operator in addition to zo because there is only one available first position. Apparent V1 in parentheticals can now be compared to various other V1 main clauses for which silent material has been proposed: yes/no questions, imperatives, topic drop, and so on. See the examples in (20), for instance:

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 113

(20) a. Heb jij

dit boek gelezen?

have you this book read? ‘Have you read this book?’ b. Lees dit boek! Read this book ‘Read this book!’ c. Heb ik al

gelezen.

have I already read ‘I’ve read [it] already.’

The question then is where the operator in (19) – whether overt zo or silent OP – originates and what it means. It appears that we can relate this to another issue, namely the absence of the internal argument. As we already noticed above, many verbs of communication and cognition are obligatorily transitive and normally take a clausal or (pro)nominal complement; see (21):10 (21) a. Ik vermoed [DP dit/iets] I

suspect

this/something

b. Ik zei [DP dit/iets] I

/

said this/something

/

[CP dat Bob Bea kent] that Bob Bea knows [CP dat Bob Bea kent] that Bob Bea knows

/

*ø . *ø

/ ““ /

*ø .

*ø

At least at first sight, this strongly suggests that OP/zo in a comment or reporting clause is somehow associated with the internal argument of the parenthetical verb. Semantically, the pronominal OP/zo is anaphoric to the host clause. It may thus be considered a pronominal placeholder for the proposition or speech act expressed by the host (depending on whether it concerns direct quotation or not); compare I think so, he said so, or I don’t believe that. A reasonable first hypothesis to test is therefore that OP/zo syntactically instantiates the internal argument. As such, it can be generated in the regular object position, and subsequently be A´-moved to the first position (SpecCP, i.e., the specifier position of the complementizer phrase), which results in an argument variable. This idea is depicted in (22): (22) a. Bob is, OPi/zoi vermoed ik ti , een echte charmeur. b. “Bob is”, OPi/zoi zei Bea ti , “een echte charmeur”.

|| 10 Here, DP stands for ‘determiner phrase’, and CP for ‘complementizer phrase’ (i.e., a full clause).

114 | Marlies Kluck and Mark de Vries In section 3, we will suggest an alternative variant, in which OP/zo is generated as a low adverb (related to a potential manner interpretation of zo/so). What is relevant for now is that both analyses require topicalization of OP/zo, which should be detectable by standard movement diagnostics.11 Chomsky (1977) presents island sensitivity and unboundedness as hallmark characteristics of A´movement. Expectedly, this is what we find (see also suggestions in Potts 2002 and De Vries 2006). Example (24), for instance, would involve movement out of a complex noun phrase, which is unacceptable; compare a textbook example as in (23). (23) * Whati did you talk to someone who suspected ti ? (24)* Bob is, OP/zoi sprak ik [iemand

die

vermoedde ti ], een echte charmeur.

Bob is OP/so spoke I someone who suspected

a

true charmer

Similarly, movement across a complex NP boundary is barred in reporting clauses: (25) * “Bob is”, OPi/zoi ken

ik [iemand

Bob is Op/so know I

die

gezegd ti had], “een echte charmeur”.

someone who said

had

a

true charmer

Crucially, it is not the case that long distance movement is excluded per se. Unbounded movement of an object across bridge verbs is acceptable (26); and this is also the case in parenthetical constructions; see (27) and (28), which contrast clearly with (24) and (25): (26) Whati did you say/hear/claim [that John bought ti today]? (27) Bob is, OPi/zoi hoorde ik [dat Piet vermoedde ti], een echte charmeur. Bob is OP/so heard I that Piet suspected

a

true charmer

(28) “Bob is”, OPi/zoi beweerde Piet [dat Anna gezegd ti had], “een echte charmeur”. Bob is

OP/so claimed Piet that Anna said

had

a

true charmer

|| 11 An issue that we will leave aside here, is the general question what triggers displacement in A´-movement constructions. A common theoretical assumption is that operators (whether overt or covert) are assigned an abstract feature (for instance, [+wh], [+rel], [+top]) that needs to be checked in the complementizer domain. The explanatory power of this is limited, however, and it is being called into question in current Minimalist thinking.

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 115

Thus, there is evidence for A´-movement of OP/zo,12 which corroborates the proposal in (22).13 Notice also that these facts are highly problematic for alternative approaches that make reference to an implicit, syntactically suppressed argument variable that is non-canonically licensed in semantics (Reis 1995, Steinbach 2007), which would lead to ‘genuine V1’; see also Fortmann (2007) for critical discussion.14 A further advantage of the proposed view is that we can link it to the phenomenon of topic drop, a possibility also discussed in the literature just mentioned. For pronominal arguments in Dutch, it is the case that they can be dropped on two conditions: the first is an obvious requirement of contextual (or deictic) recoverability, the second is that the (usually demonstrative) pronoun must be moved to the first position before deletion, and – relatedly – can function as the sentence topic. This is shown in (29). The in situ pronoun in (29a) cannot be dropped, but if it is fronted (29b), it can be: (29) about a particular book: a. Ik heb *(dat) gelezen. I

have that read

b. (Dati) heb ik ti gelezen.

|| 12 In German, the corresponding judgments for unbounded movement appear to be slightly less clear. According to our informants, a complex comment clause is acceptable (i) – depending on the right context –, but a complex reporting clause is quite marginal (ii): (i)

Peter ist, so Peter is so (ii) ?? “Peter ist”, Peter is

habe ich gehört dass have I heard that so behauptete Hans so claimed Hans

Anna behauptet hat, ein echter Charmeur. Anna claimed has a true charmer dass Anna gesagt hätte, “ein echter Charmeur” that Anna said had a true charmer

Despite this, there is still a contrast with complex parentheticals containing island boundaries (comparable to the Dutch examples in (24) and (25) in the main text), which are completely unacceptable, as expected. 13 Further evidence can be obtained by studying particular properties of individual languages. For instance, Irish complementizers have different possible morphological realizations (cf. McCloskey 1979 and subsequent work), which distinguish between regular and A´movement contexts (the ‘agreeing’ type, where SpecCP is filled). James Griffiths (p.c.) informed us that in parenthetical constructions, the second type is used, which is in accord with our expectations. For reasons of space, we cannot go into detail, here. 14 In Fortmann’s (2007) own proposal, there is an empty pro-form in the parenthetical object position, which stays in situ. He argues rightly against base-generation of the eventual main clause as the parenthetical internal argument and subsequent deletion or movement (see also footnote 2 for discussion). However, this cannot be taken as evidence that there is no A´movement at all. As we showed in the main text, it is the pro-form/operator that is moved.

116 | Marlies Kluck and Mark de Vries Note that fronting can be detected even if the object is silent because of subjectverb inversion in combination with V1 (i.e., it is not another constituent that is topicalized). Topic drop can be considered as a special kind of ellipsis, brought about by deletion at the phonological interface (PF). Unlike core syntactic processes, it is optional and context-dependent – that is, the possibility of topic drop is conditional (hence depending on certain discourse conditions, it can but need not be done). So far, we established a number of things concerning zo. Firstly, the realization of zo is optional. At least in Dutch, all apparent V1 parentheticals can be optionally turned into overt V2 parentheticals by spelling out OP as zo, and vice versa. Secondly, we noticed that zo is anaphoric, hence pronominal (demonstrative, even), and its antecedent is clearly recoverable, as it involves the immediate syntactic context. Thirdly, we argued that zo is fronted within the parenthetical. To this we can add that a comment clause is rather obviously about what it comments upon, and a reporting clause is about what it reports. Therefore, zo can be considered the sentence topic. All conditions for topic drop are fulfilled, and hence the morpho-phonological realization of zo will be optional. It should be emphasized that whether topic drop of zo applies or not, is arbitrary for our assumptions: if zo is dropped at PF, there will still be an empty operator with the same syntactic and semantic function.15 Note that Steinbach (2007) presents a few arguments against a topic drop analysis, which we think fail on closer inspection. First, he assumes that a topi|| 15 Interestingly, there appears to be a small grey area between comment clauses and certain and-parentheticals (or quasi-coordinative parentheses), namely those in which a clauseanaphoric demonstrative is topicalized, where the predicate coincides with a comment or reporting verb, and where the coordinator can be left out (resulting in parenthetical juxtaposition). A minimal quadruple is provided in the examples (i) through (iv): (i) De directeur had –en dat geloofden wij allen– gefraudeerd. the manager had and that believed we all committed.fraud ‘The manager had – and we all believed that – committed fraud.’ (ii) De directeur had – dat geloofden wij allen – gefraudeerd. (iii) De directeur had, zo geloofden wij allen, gefraudeerd. (iv) De directeur had, geloofden wij allen, gefraudeerd. It is hard to tell whether the regular comment clause in (iv) is derived from topic drop in (ii) or (iii), and it is likely that both options are possible. Nevertheless, it is clear that the similarity between these sentences is accidental. Generally, comment clauses cannot be transformed into and-parentheticals or vice versa at all; and usually an overt zo/so in a comment clause cannot be replaced by dat/das. Therefore, the claim in the main text is justified that topic drop in parenthetical verb constructions normally involves the operator zo/so and not a preposed d-pronoun.

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 117

cal d-pronoun would compete with the optional so for the first position, which is impossible. However, we do not postulate an additional demonstrative: so itself is the sentence topic. The fact that so/zo cannot replace das/dat/that in all contexts (Steinbach 2007:71/2, fn.15) is irrelevant. Due to its operator status in the pertinent examples, we do not expect so to surface in situ (Martin glaubt das/*so [German] ‘M. believes that’). Furthermore, the example Das/*so glaube ich nicht [German] ‘I don’t think so’, if anything, can rather be interpreted as an argument in favor of the ‘so-drop’ analysis because the corresponding negative parenthetical is unacceptable, too.16 Secondly, and this is a more fundamental issue, Steinbach argues at length that the antecedent of the dropped pronoun is (or can be) an assertion and does not have a discourse-topical status. If we are not mistaken, this is a misunderstanding of the topic drop phenomenon: it is the dropped pronoun itself that must be the sentence topic, the antecedent does not need to be topical. If it is not, we are facing a shift of topic, which is perfectly possible, at least with sentential antecedents, and probably in various other cases as well: (30) A: Mieke is leuk.

B: Dat vind ik ook.

Mieke is nice

that think I too

‘Mieke is nice’

‘I agree.’

(31) A: Mieke vond het boek interessant.

B: Dat is het ook.

Mieke found the book interesting

that is is PRT

‘Mieke found the book interesting.’

‘It certainly is.’

See also De Korte (2008) for examples and discussion. || 16 One may wonder why topic drop in negative sentences is allowed in certain nonparenthetical contexts, but not in parentheticals, as illustrated in (i) and (ii). However, as (iii) shows, the parenthetical is not acceptable with an overt pronoun, either; so the question is misguided. (i) A: Jan is ziek. B: Dat geloof ik niet. Jan is ill that believe I not (ii) # Jan is ziek, zo/dat geloof ik niet. Jan is ill so/that believe I not (iii) # Jan is ziek, zo/dat geloof ik niet. We think the reason why (ii) and (iii) are out is simply that there is only one speaker involved, contrary to the situation in (i). Therefore, the negative parenthetical, whether it involves topic drop or not, would lead to self-contradiction, which is pragmatically odd. Hence, this issue, however interesting in itself, is irrelevant for the argumentation in the main text.

118 | Marlies Kluck and Mark de Vries To summarize, we showed that all regular comment and reporting parentheticals are V2 main clauses (but see section 4 concerning as-parentheticals). There is a clause-initial operator that can be either null due to topic drop or it can be spelled out as so/zo. This explains subject-verb inversion in Dutch and other V2 languages, as well as the appparent V1 pattern in case an empty operator is used. Adverbs and other elements can be part of the parenthetical, but they cannot be preposed. We argued that the operator is A´-moved, and also suggested that it might originate as the apparently missing internal argument of the parenthetical verb, which is consistent with the anaphoric function of OP/zo. The next section discusses certain complications for this view, and introduces a potential solution.

3 More on the interpretation and base position of zo/so In the examples used so far, the parenthetical verb is transitive, and there appears to be an empty internal argument with which zo/OP is associated. It may therefore come as a surprise that we find reporting clauses without an available argument position, either because the verb is intransitive, or because the argument slot is already filled with another noun phrase. These possibilities are illustrated in (32), based on earlier observations in Schelfhout (2006) and De Vries (2006). (32) a. “Bob is”, (zo) ontplofte Mieke, “een gemene charmeur!” Bob is

so exploded Mieke

a

nasty

charmer

‘“Bob is a nasty charmer,” Mieke exploded.’ b. “Dit is waardeloos”, (zo) becommentarieer de ze This is worthless

so commented.on

mijn artikel.

she my

article

‘“This is worthless”, she commented on my article.’

The verb ontploffen ‘explode’ in (32a), is unaccusative, hence intransitive. Consequently OP/zo cannot be analyzed as the direct object. But is this firm conclusion true? We acknowledge that the argument structure of lexical predicates is not always fixed, and can sometimes be used creatively. In this case, one might hypothesize that ontploffen ‘explode’ is syntactically (not just pragmatically) reinterpreted as ‘say angrily’, which would make it transitive. However, if that

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 119

were possible, we would expect (33a-c) to become acceptable, too, which is definitely not the case: (33) a. * Mieke ontplofte een vloek. Mieke exploded a

curse

‘[*]Mieke exploded a curse.’ b. * Mieke ontplofte dat Bob gemeen deed. Mieke exploded that Bob mean

did

‘[*]Mieke exploded that Bob was doing mean.’ c. * Mieke ontplofte “Ga weg!” Mieke exploded

go away

‘[*]Mieke exploded “Go away!”

The problem thus stands as it is. Furthermore, a flexible view on transitivity would not solve the issue in (32b), where there is an object slot, but it is taken by the noun phrase mijn artikel ‘my article’. We would like to stress that the relevant data are not a quirk of Dutch; comparable examples were tested in French and Spanish by Van Maastricht (2011), and judged acceptable by most speakers. Crucially, the meaning of zo shifts to a manner interpretation in (32). Example (32a), for instance, indicates roughly that the way Mieke exploded is by saying “Bob is a nasty charmer”. Notably, a manner interpretation of zo is independently available; see (34): (34) Ik heb het zo gedaan. I

have it

so done

‘I did it in that way.’

The pronominal manner adverb zo is either deictic or anaphoric. Notice that it can also be topicalized/fronted: (35) Zoi heb ik het ti gedaan. so have I it

done

‘In that way, I did it.’

For these reasons, it makes sense to analyze OP/zo as an anaphoric manner adverbial rather than a direct object in (32). Importantly, there is evidence for A´-movement even in these cases. First of all, it is evident that there is subject-verb inversion. However, we can also show that the construction is island-sensitive, for instance. This is illustrated in (36):

120 | Marlies Kluck and Mark de Vries

(36) * “Bob is”, zoi sprak ik [iemand Bob is so spoke I

die ti ontplofte], “een

someone who

exploded

a

gemene charmeur”. nasty

charmer

‘[*]”Bob is”, so did I speak to someone who exploded, “a nasty charmer!”

The structure for (32) is therefore as sketched in (37). As always, the subject is moved from its argumental base position within the verb phrase to the structural subject position, and the finite verb is moved to the second position. Furthermore, OP/zo is generated as a manner adverbial, which could be adjoined to the verb phrase (or have its own projection in the sentential spine – that is irrelevant for our purposes), and consequently fronted within the parenthetical, similarly to the situation in (35). (37) “Bob is”, OP/zoi ontploftev Miekes [VP ti [VP tv ts]], “een gemene charmeur”.

Thus, there is a clear solution for the cases in (32), which seemed problematic at first sight. We now have to consider at least the possibility that zo is always a manner adverb, even in regular parentheticals of the type in (8), where an object position is available. One might speculate that OP/zo is able to license an empty argument in those cases, as suggested earlier for reporting clauses in De Vries (2006). In (38a/b), [e] indicates the empty internal argument of vermoeden ‘suspect’ and zeggen ‘say’, respectively; t is the original position of the fronted adverb OP/zo. (38) a. Bob is, OP/zoi vermoed ik ti [e]i , een echte charmeur. Bob is op/so suspect

I

a

true charmer

‘Bob is, I suspect, a true charmer.’ b. “Bob is”, OP/zoi zei Anna ti [e]i , “een echte charmeur”. Bob is op/zo said Anna

a

true charmer

‘Bob is, Anna said, a true charmer.’

Interestingly, the configuration is reminiscent of parasitic gap constructions: the moved operator c-commands both its trace (the adverb position) and the empty argument slot.17 A common example of a parasitic gap is (39): || 17 There is also a difference: in parasitic gap constructions such as (39), there is no ccommand between the two variables (the trace and the gap). In (38), however, it seems that t ccommands [e], depending on the details of the analysis. We will not elaborate on this issue, here.

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 121

(39) [Welk boek]i heb

je

opgeborgen ti zonder [e]i te lezen?

which book have you filed

without

to read

‘Which book did you file without reading?’

Thus, we might hypothesize that the operator is also directly related to both variables in (38). We leave further discussion of the parallel with parasitic gap constructions open here. There is additional evidence that OP/zo can be used as an adverb. Fortmann (2007:99ff) shows that in prosodically unintegrated parentheticals in German, so can be combined with an object pronoun es, with certain restrictions; see (40). It seems that such examples combine two types of anaphoric relations, namely reference to a proposition as well as the way it is expressed (the speech act). (40) Theo kam – so sagt es Paul – mit seinem Hund. Theo came

so said it Paul

with his

[German]

dog

‘Theo brought – so says Paul – his dog.’

It is somewhat difficult to find similar types of examples in Dutch, but they do exist, especially in explicit contexts of so-called mixed quotation (where, again, mention and use overlap). This is illustrated in (41), where het ‘it’ must be prosodically weak. On a side-note, the German word order with the pronoun preceding the subject is impossible in Dutch. (41) Anna was helemaal

‘flabbergasted’, zo zei Bea

Anna was completely flabbergasted so said Bea

het althans. it

at.least

‘Anna was completely “flabbergasted”, at least, that’s how Bea put it.’

In this sentence, too, zo can only be understood in the manner reading. This is precisely what is expected in our analysis.18 In (40) and (41) a verb of communication (say) is used as a comment, a possibility we already commemorated in footnote 1. In the German example in (42), || 18 Although the present work is limited to comment and reporting parentheticals in Germanic languages, we would like to point out that Turkish data may offer additional evidence in favor of the present hypothesis. As Güliz Güneş (p.c.) pointed out, the word oyle ‘so’ can be used in Turkish reporting parenthetical clauses, but it is not inflected for accusative case, unlike other objects; this is indicative of an adverbial status. Clearly, however, this suggestion requires careful examination in the context of other related phenomena in Turkish.

122 | Marlies Kluck and Mark de Vries which we owe to an anonymous reviewer, so has a pertinent manner interpretation, too. Here, the object position is occupied by a full noun phrase, comparable to the situation for reporting clauses in (32b). Although we have trouble assigning a non-quotative meaning to the equivalent in Dutch, we accept a parenthetical as in (43) as a comment. (42) Der Minister, so erzählte man die Geschichte, habe plagiiert. the minister so told

one the story

has

plagiarized

‘The minister, so people told the story, has plagiarized.’ (43) Bob heeft, zo deed het verhaal de ronde, een grote vis gevangen. Bob has

so did the story

the tour

a

big

fish caught

‘Bob caught a big fish, so the story went.’

It should be noted, however, that a forced manner reading of zo due to the lack of an available object position is not available in comment clauses with a regular epistemic verb. Since vermoeden ‘suspect’ in (44) is a transitive verb, an overt object like een complot ‘a conspiracy’ might be expected to occur next to adverbial zo. However, such is not the case. (44) *Bob ging, zo vermoedde Bea een complot,

weleens

vissen.

Bob went so suspected Bea a conspiracy sometimes fish.INF ‘[*]Bob went, so did Bea suspect a conspiracy, fishing once in a while.’

It seems to us that this restriction is due to the interpretation of the predicate. Verbs of communication can easily be modified with a manner adverb. Therefore, such predicates can be part of a wh-question with manner-how; see (45a). By contrast, predicates with an inherent epistemic import cannot be modified in such a way; see (45b). For ease of representation, we illustrate in English, but the observation extends to Dutch straightforwardly. (45) a. A: How did Bea report the conspiracy / that there was a conspiracy? B: Reluctantly. b. A: How did Bea suspect the conspiracy / that there was a conspiracy? B: # Reluctantly.

Thus, comment clauses are slightly more restricted in use than reporting clauses.

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 123

To wrap up, in this section we have elaborated on the interpretation and position of zo, taking our assumption that zo is an operator in section 2 as a point of departure. We have demonstrated a shift to manner-interpretation in case there is no internal argument position available in the parenthetical clause – a possibility in reporting clauses, but not in comment clauses. In the relevant cases, OP/zo may originate as an anaphoric manner adverb ‘in this way’. When it is used as such, OP/zo is demonstrably A´-moved, as in regular parentheticals: it causes inversion, is sensitive to islands, and so on. Finally, we considered the possibility to generalize the solution for the problematic cases to all comment and reporting clauses. This would imply that OP/zo is always generated in an adverb position, and that it is able to license an empty object when needed. Given the suggested parallel with parasitic gaps, and other considerations, this seems a reasonable way to go, but at this stage of our research it is hard to tell what would be better from a theoretical point of view: a generalized but somewhat complex analysis, or a simpler analysis for regular cases, and a somewhat different analysis for the anomalous cases only.

4 As/wie/zoals-parentheticals Let us now turn to as/wie/zoals-parentheticals, as announced in the introduction. Before anything else, it should be pointed out that these are a special variety of comment clauses; even examples in which the parenthetical verb is a verb of saying cannot have a quotative reading:19

|| 19 Consequently, examples with intransitive parenthetical verbs or transitive constructions with an occupied object slot are normally excluded because they only occur in true quotative contexts, as we discussed; see (i) and (ii). However, there are some possibilities with mixed quotation if a verb of communication is used; see (iii): (i) * Bob is, zoals Mieke ontplofte, een gemene charmeur. Bob is as Mieke exploded a nasty charmer ‘[*] Bob is, as Mieke exploded, a nasty charmer’ (ii) * Bob is, zoals Mieke een complot vermoeddde, een gemene charmeur. Bob is as Mieke a conspiracy suspected a nasty charmer ‘[*] Bob is, as Mieke suspected a conspiracy, a nasty charmer.’ (iii) Bob is, zoals Mieke het zou zeggen, ‘een gemene charmeur’. Bob is as Mieke it would say a nasty charmer ‘Bob is a “nasty charmer”, as Mieke would say it.’

124 | Marlies Kluck and Mark de Vries

(46) Bob is, as Bea said, a true charmer.

Interestingly, as-parentheticals are consistently verb-final in German and Dutch, witness the following examples: (47) a. Bob is, zoals ik al Bob is as

I

vermoedde, een echte charmeur.

already suspected

a

true

b. Bob ist, wie ich vermutete, ein echter Charmeur. Bob is as I

suspected a

true

[Dutch]

charmer [German]

charmer

‘Bob is, as I already suspected, a true charmer.’

It has been widely acknowledged in the literature that the gaps in English asparentheticals may correspond to CPs as well as any predicate-denoting phrase (Potts 2002, LaCara 2013). Dutch and German lack this last type of asparentheticals. This follows straightforwardly if this type is analyzed in terms of verb phrase (VP)-ellipsis (LaCara 2013), since Dutch and German lack VPellipsis altogether. The discussion here is therefore restricted to the CP-gap type, in relation to the comment clauses discussed above. The primary issue we are concerned with is the aberrant V-final word order in these cases. Taking the analysis of zo as an operator further, we will argue that als (‘as’) is a complementizer. In addition, we will briefly discuss some cross-linguistic data and speculate about the implications of our proposal for the meaning of asparentheticals. Parentheses – by definition – are not hypotactically construed (i.e., subordinated with respect to the host) in the usual way. The zero hypothesis is therefore that they have ‘root status’ (in a pretheoretical sense), and by extension that clausal parentheticals classify as main clauses, unless there is an internal clue for a subordinated pattern.20 The observed V2 (or V1) order in regular Dutch and German comment and reporting clauses – as well as and-parentheticals and the like (with or without an actual discourse linker) – comes as no surprise, then. However, the V-final order in (47) requires an explanation. We think that the Dutch complex form zo-als (lit. ‘so’ + ‘as/if/like’) provides an important clue. The basic idea is completely straightforward. Building on work by Hoekstra

|| 20 See Espinal (1991), among others, for extensive argumentation and further references concerning the ‘root’ status of parentheses (think of scopal independence, speaker orientation, and so on). De Vries (2012b) explicitly discusses the issue in the light of another interesting case, namely appositive relative clauses (which are V-final in Dutch) versus ‘quasi-relatives’ or ‘V2-relatives’.

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 125

(1993) and Zwart (2000), we claim that -als is a complementizer. Since clauses introduced by a complementizer are necessarily V-final (Den Besten 1977), the word order in zoals-parentheticals (47a) and the regular Dutch dat-clause (48b) can be explained in a similar fashion: (48) a. Bob kuste Bea.

[V2]

Bob kissed Bea ‘Bob kissed Bea.’ b. … dat Bob Bea kuste. C

[V-final]

Bob Bea kissed

‘… that Bob kissed Bea.’

We take the ability for the morpheme als to introduce a finite, often complementizer-less clause as a clear argument that it is to be seen as the complementizer itself. Consequently, the clausal complement is then the inflectional or tense phrase (IP/TP), or a lower layer of CP if there are several shells (e.g., in als-of ‘as if’ comparative clauses).21 Notice that this runs counter to assumptions in Emonds (1985), Potts (2002) and LaCara (2013), who take the English counterpart as to be a preposition with a CP complement, but without convincing evidence.22 We do however concur with the idea that as-morphemes (in English as well as in Dutch and German) may denote comparative meaning as one of several interpretation types. Undoubtedly, as-parentheticals are comment clauses with a comparative meaning aspect. The question is then how this can be reconciled with the idea that zo is an operator, which was central to the previous sections. Our hypothesis is as follows:

|| 21 Just to be clear, we do not claim that prepositional or other nonverbal elements cannot select for clauses per se. However, if they do, the selected clause is a CP with an overt or at least active COMP domain in all cases we are aware of. 22 Potts (2002:639) acknowledges that it is particularly difficult to establish prepositionhood of as-morphemes; the arguments suggested are merely indirect. Specifically, Potts draws a parallel between the use of as and the use of than in comparatives. Interestingly, some of the data concerns inversion in English than- and as-clauses, reminiscent of inversion patterns traditionally associated with main clauses (Emonds 1976, Hooper & Thompson 1973). The fact that than and as pattern alike may reveal something about the type of clause they are associated with; it is not clear why this should imply that they are prepositions.

126 | Marlies Kluck and Mark de Vries (49) Hypothesis Zoals in Dutch as-parentheticals is a morpho-phonological combination of the operator zo and the complementizer als. As was mentioned above, the interpretation of zoals-parentheticals involves a comparison of the contents of the host sentence with the understood embedded proposition – in (47a) what the subject of the parenthetical suspected. In our analysis, the latter is expressed by the anaphoric operator zo. Maintaining the general idea that comment clauses involve operator movement, the structure is then as in (50): (50) …[CP zoi [C´ als [IP ik al so

as

vermoedde ti ]].

I already suspected

‘… as I suspected.’

The V-final pattern follows automatically because C, the position of the finite verb in V2 clauses, is occupied by an overt complementizer. The contrast in (51) follows directly from our assumptions as well: (51) a.

..., zo-als ik gisteren so-as I

al

vermoedde, ...

yesterday already suspected

‘… as I already suspected yesterday, …’ b. *..., gisteren

zo-als ik al

yesterday so -as

I

vermoedde, ...

already suspected

That is, preposed adverbials (or other phrases) give rise to ungrammaticality. This is expected for various reasons. If zo occupies SpecCP, there is no room for a topicalized phrase; apart from that, positioning of non-operator material directly before a lexical complementizer – als, in this case – is generally disallowed. We turn to a potential concern regarding the ‘doubly filled COMP’ filter below. It is worth mentioning that zo-als can sometimes be preceded by a modifier targeting zo, as is illustrated in (52): (52) ..., [precies zo]-als ik al exactly so- as I

vermoedde, ...

already suspected

This nicely confirms the inherent phrasal status of OP/zo. In (52) it is the phrase precies zo ‘exactly so’ that is fronted.

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 127

If it is indeed the case that zo in zoals-parentheticals is essentially the same operator as zo in comment clauses, we predict the same effects regarding unboundedness and island-sensitivity as were observed in (23)-(25) above. The data in (53) and (54) show that this is indeed the case: (53) ..., zoi -als i k gehoord had so -as

I

heard

had

[dat Bob ti vermoedde], ... that Piet

suspected

‘… as I heard that Bob suspected,…’ (54) *..., zoi-als ik iemand so-as

I

gesproken had [die ti vermoedde], ...

someone spoken

had who suspected

Thus, movement of zo may cross the boundary of a finite clause (53) but is sensitive to islands – here a relative clause (54). The idea pursued here offers an interesting perspective for cross-linguistic comparison. As it turns out, there is some language variation. First, notice that the combination so wie (‘so as’) exists in German as well (55). However, contrary to Dutch, the use of so is optional in the German as-parenthetical (56): (55) Bob ist,(so) wie ich vermutete, ein echter Charmeur. Bob is so as

I

suspected

a

true

[German]

charmer

‘Bob is, as I already suspected, a true charmer.’ (56) Bob is, *(zo-) als ik vermoedde, een echte charmeur. Bob is

so

as I

expected

a

true charmer

‘Bob is, as I expected, a true charmer.’

In present-day English the combination of so as no longer exists in the relevant sense:23 (57) Bob is, (*so) as I suspected, a true charmer.

|| 23 The combination so as does exist in another sense, namely purpose. For instance, in (i) the meaning of so as is comparable to in order to: (i) Wash your hands carefully so as to decrease the risk of getting sick.

128 | Marlies Kluck and Mark de Vries Nevertheless, it seems reasonable to generalize the approach suggested for Dutch to English, especially since English as, like Dutch als, can be used independently as a (temporal) complementizer: (58) I heard a dog barking as I rang Bob’s doorbell.

Why the operator must remain silent in English is an issue that may be related to the Doubly Filled COMP Filter. This is a superficial constraint that states that only one position in the complementizer domain can be lexicalized, that is, either the C head or the SpecCP position. Relative clauses, for instance, can be introduced by a relative pronoun (e.g., who) or a complementizer (that), but not both (*who that). Similarly, the combination of so plus as in an as-parenthetical can be excluded. Since the complementizer contains an essential meaning component in this case, it cannot be deleted, but the pronoun can. The Doubly Filled COMP Filter is not universal, and it can be argued that it is not active in Dutch, considering topicalization plus V2 in main clauses, or the optional presence of complementizers in embedded questions (in many regional variants of Dutch); see (59), for instance: (59) Piet vroeg wie (of) het gedaan had. Piet asked who if

it

done

had

‘Piet asked who did it.’

Thus, deletion of OP is not necessarily expected. A further question is why it cannot be deleted. We suggest that this has to do with morphological incorporation with the complementizer. In the end, this is an arbitrary lexical choice. Finally, the question arises how German wie-clauses fit into the picture. It is tempting to postulate an analysis parallel to Dutch, which would require that wie is a complementizer in this construction (hence not a question word in SpecCP; pace Fortmann 2007). The fact that so can precede wie could be taken as an argument for that. However, the wh-morphology of wie seems problematic from this perspective. An interesting alternative view may be that wie is analyzed as a relative pronoun, and so is the antecedent of an adverbial relative clause. This in turn raises questions about the status of the parenthetical as a whole (which is then a complex adverbial phrase containing a relative clause) and the semantics of the construction. We will leave the matter undecided here for reasons of space and time. Summing up, in this section we have taken the basics of our proposal for comment clauses further, and extended our analysis to capture the Dutch vari-

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 129

ant of as-parentheticals, i.e., parentheticals introduced by zoals. We have proposed that als/as in as-parentheticals is a complementizer. Specifically, Dutch zoals is a combination of the complementizer als (in C) and the operator zo, which has A´-moved to SpecCP. The presence of an overt complementizer directly explains why these parentheticals exhibit the V-final pattern rather than the V2 pattern in other Dutch comment clauses. Comparing to English and German, we attested some cross-linguistic variation, but none of this is inconsistent with the claims we have made. In order to gain more insight in this respect, further comparative and diachronic studies are needed.

5 Conclusion We investigated the internal syntax of parenthetical comment and reporting clauses in Germanic, with some emphasis on Dutch data, and with a specific excursion to as-parentheticals. Our central questions concerned the variable position of the finite verb, the various types of predicates that are involved, and the status of the element zo/so. We argued in favor of a generalized analysis that involves a clause-initial operator inside the parenthetical. The operator can be either null due to topic drop or spelled out as so/zo. This straightforwardly explains the obligatory subject-verb inversion in Dutch and other V2 languages, as well as the apparent V1 pattern if zo/so is dropped. The idea is corroborated by the fact that adverbs and other elements can be part of the parenthetical, but they cannot be preposed: OP/zo occupies the first sentence position (SpecCP). More specifically, the operator is A´-moved into this position, and base-generated within the verb phrase either as the complement of the parenthetical verb or as a low adverb adjoined to the verb phrase. The analysis is further supported by the fact that movement of OP/zo is unbounded and island-sensitive. In regular cases, OP/zo is anaphoric to the host clause, the embedded verb is obligatorily transitive, and there is an object gap. Nevertheless, there are also cases where no internal argument position is available for the anaphoric pro-form, especially in reporting parentheticals. These can involve intransitive parenthetical verbs, such as the unaccusative ontploffen ‘explode’, or transitive constructions with an otherwise satiated internal argument. Crucially, zo shifts to a manner interpretation in those cases, which requires reanalysis of OP/zo as an anaphoric manner adverbial. We also speculated how this might be generalized to all parentheticals under discussion.

130 | Marlies Kluck and Mark de Vries Concerning as/wie/zoals-parentheticals, which are a particular kind of comment clauses, we showed that these are consistently verb-final in German and Dutch. This word order is different from regular parenthetical comment and reporting clauses. We take the Dutch complex form zo-als (lit. ‘so’ + ‘as/if/like’) to provide an important clue in explaining this fact. We argued that -als is a complementizer here. The presence of a complementizer in the C position automatically induces the V-final pattern. Furthermore, the interpretation of these parentheticals involves a comparison of the contents of the host sentence with the understood embedded proposition. In our analysis, the latter is expressed by the anaphoric operator zo, which is A´-moved as in the other parenthetical constructions, and then morphologically combined with the complementizer. We showed that the movement of zo in Dutch zoals-parentheticals is similarly unbounded and sensitive to islands. Generalizing the approach, we suggested that English as is a complementizer, too (hence not a preposition), and we speculated about the status of German wie. In this context, we observed some interesting variation between English, German and Dutch: zo is obligatory in Dutch zoals-parentheticals, so is optional in German (so)-wie-parentheticals, but so has become impossible in present-day English as-parentheticals. In conclusion, the word order puzzle we started out with can be solved unproblematically. We advanced a generalized syntactic analysis of various types of parenthetical comment and reporting clauses, making use of mechanisms that are independently available in the grammar, and thus reducing the need to stipulate construction-specific rules and constraints.

References Besten, Hans den. 1977. On the interaction of Root Transformations and Lexical Deletive Rules. In Werner Abraham (ed.). On the Formal Syntax of the Westgermania. Amsterdam: John Benjamins. 47–131 Blanche-Benveniste, Claire. 1989. Constructions verbales “en incises” et rection faible des verbes. Recherches sur le français parlé 9. 53–73. Chomsky, Noam. 1977. On Wh-Movement. In Peter Culicover et al. (eds.). Formal Syntax. New York: Academic Press. 71–132. Collins, Chris & Phil Branigan. 1997. Quotative Inversion. Natural Language & Linguistic Theory 15. 1–41. Dehé, Nicole. 2009. Clausal parentheticals, intonational phrasing, and prosodic theory. Journal of Linguistics 45. 569–615. Dehé, Nicole & Yordanka Kavalova (eds.). 2007. Parentheticals. Amsterdam: John Benjamins. Emonds, Joseph. 1976. A transformational approach to English syntax: root, structure preserving, and local transformations. New York: Academic Press.

On V2, gaps, andOntogenetic operators inpaths comment and reporting parentheticals to the parenthetical construction | 131

Emonds, Joseph. 1985. A Unified Theory of Syntactic Categories. Dordrecht: Foris. Espinal, M. Teresa. 1991. The Representation of Disjunct Constituents. Language 67. 726–762. Fortmann, Christian. 2007. The complement of reduced parentheticals. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam: John Benjamins. 89–119. Griffiths, James. (this volume). Speaker-oriented & quotative reduced parenthetical clauses. In Stefan Schneider, Julie Glikman & Mathieu Avanzi (eds.). Parenthetical Verbs. Berlin: De Gruyter. Griffiths, James. (to appear). Parenthetical verb constructions, fragment answers, and constituent modification. Natural Language & Linguistic Theory. Published online 12 August 2014. Griffiths, James & Mark de Vries. 2013. The syntactic integration of appositives: evidence from fragments and ellipsis. Linguistic Inquiry 44. 332–344. Güneş, Güliz & Çağrı Çöltekin. (this volume). Prosody of Parentheticals in Turkish. In Stefan Schneider, Julie Glikman & Mathieu Avanzi. (eds.). Parenthetical Verbs. Berlin: De Gruyter. Hoekstra, Eric. 1993. Dialectal Variation inside CP as Parametric Variation. In Werner Abraham & Josef Bayer (eds.). Dialektsyntax [Linguistische Berichte, Sonderheft 5]. 161–179. Hooper, Joan & Sandra Thompson. 1973. On the application of root transformations. Linguistic Inquiry 4. 465–497. Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. Cambridge, MA: MIT Press. Kluck, Marlies. 2011. Sentence amalgamation. University of Groningen: PhD thesis. Kluck, Marlies. 2013. A sluicing account of amalgams. University of Groningen: Manuscript. Kluck, Marlies. (to appear). On representing anchored parentheses in syntax. In Josef Bayer, Andreas Trotzke & Antje Lahne (eds.). Sentential Complexity across Interfaces and Frameworks/Interface Explorations. Berlin: Mouton. Kluck, Marlies, Dennis Ott & Mark de Vries (eds.). (2014). Parenthesis and Ellipsis: CrossLinguistic and Theoretical Perspectives. Berlin: Mouton de Gruyter. Korte, Siebe de. 2008. Dutch topic drop as a PF phenomenon. University of Amsterdam: MA thesis. [online]. LaCara, Nicholas. 2013. Deletion in As-Parentheticals. University of Massachusetts: Manuscript. Maastricht, Lieke van. 2011. Reporting and comment clauses: a cross-linguistic study. University of Groningen: MA thesis. McCloskey, James. 1979. Transformational syntax and model theoretic semantics: A case study in Modern Irish. Dordrecht: Reidel. Potts, Christopher. 2002. The Syntax and Semantics of As-Parentheticals. Natural Language & Linguistic Theory 20. 623–689. Reis, Marga. 1995. Wer glaubst du hat recht? On so-called extractions from verb-second clauses and verb-first parenthetical constructions in German. Sprache und Pragmatik 63. 27– 83. Reis, Marga. 2002. Wh-movement and Integrated Parenthetical Constructions. In Jan-Wouter Zwart & Werner Abraham (eds.). Studies in Comparative Germanic Syntax. Amsterdam: John Benjamins. 3–40. Ross, John. 1973. Slifting. In Maurice Gross et al. (eds.). The Formal Analysis of Natural Languages. The Hague: Mouton. 133–169. Schelfhout, Carla. 2006. Intercalations in Dutch. Radboud University Nijmegen: PhD thesis.

132 | Marlies Kluck and Mark de Vries Schelfhout, Carla, Peter-Arno Coppen & Nelleke Oostdijk. 2004. Finite comment clauses in Dutch: a corpus-based approach. Journal of Germanic Linguistics 16. 331–349. Steinbach, Markus. 2007. Integrated parentheticals and assertional complements. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam: John Benjamins. 53–88. Stoltenburg, Benjamin. 2003. Parenthesen im gesprochenen Deutsch. InLiSt – Interaction and Linguistic Structures 34. Vries, Mark de. 2006. Reported Direct Speech in Dutch. Linguistics in the Netherlands 23. 212– 223. Vries, Mark de. 2012a. Unconventional Mergers. In Myriam Uribe-Etxebarria & Vidal Valmala (eds.). Ways of Structure Building. Oxford: OUP. 143–166. Vries, Mark de. 2012b. Parenthetical main clauses – or not? On appositives and quasirelatives. In Lobke Aelbrecht, Liliane Haegeman & Rachel Nye (eds.). Main Clause Phenomena: New Horizons. Amsterdam: John Benjamins. 177–201. Zwart, Jan-Wouter. 1994. Dutch is Head Initial. The Linguistic Review 11. 377–406. Zwart, Jan-Wouter.2000. A Head Raising Analysis of Relative Clauses in Dutch. In Artemis Alexiadou, et al. (eds.). The Syntax of Relative Clauses. Amsterdam: John Benjamins. 349– 385.

| Section II: Diachronic and ontogenetic development

Fátima Faya Cerqueiro1

An approach to parenthetical courtesy markers in requests in Late Modern English Abstract: The Late Modern English period witnessed the replacement of the preferred courtesy marker in requests. At the beginning of the period, forms including the verb pray were commonly used parenthetically in initial position, and led to the pragmatic marker pray through a process of grammaticalisation. However, the use of pray declined during this period, while the new form please became more common with the same function. While a conditional origin of the parenthetical if you please has been suggested, other studies have identified an imperative construction as the main source of the Present-day English please, which would imply that its parenthetical usage would have appeared as a later development. This study analyses the main request markers in Late Modern English, the variability of strategies available and their development in this period. Their processes of change are analysed in terms of grammaticalisation, using data from three genres: novels, drama and letters. Keywords: please, pray, pragmatic markers, requests, grammaticalisation

|| Fátima Faya Cerqueiro: Center for Language and Cognition, University of Groningen. Groningen, The Netherlands.

1 Introduction At the beginning of the Late Modern English period2, pray was the preferred courtesy marker in requests. This form of French origin (cf. OED s.v. pray v.) had undergone a process of grammaticalisation that led to its reduction from con|| 1 The author acknowledges generous funding from the following institutions: European Regional Development Fund, Spanish Ministry for Economy and Competitiveness (grant FFI2011-26693-C02-01), and Autonomous Government of Galicia (Directorate General for Scientific and Technological Promotion, grants CN2011/011 and CN2012/012). Thanks are also due to the anonymous reviewers for their careful reading and their constructive comments. 2 This period generally includes the eighteenth and nineteenth centuries in a broad sense (cf. Dossena & Jones 2003; Beal 2004).

136 | Fátima Faya Cerqueiro structions such as I pray or I pray you, which were very commonly used parenthetically in initial position. The interesting development of this form has been tackled by scholars such as Akimoto (2000), who shows that it became more frequent in the seventeenth century, later to decline in frequency in the nineteenth century. Pray came to replace the pragmatic function of a native construction, the parenthetical ic bidde (cf. Traugott 2000). During the Late Modern English period, pray started to fall in disuse while the new form please was gaining ground with the same function. Please was also a form introduced into English from French (cf. OED s.v. please v.), and, in fact, the Present-day French main request marker, the conditional construction s’il vous plaît, still retains a cognate of please. Many authors have identified a similar conditional origin in English, namely the parenthetical if you please (cf. Traugott and Dasher 2002 or Brinton 2006). Nevertheless, other studies have identified an imperative construction, followed initially by a to-infinitive (as in be pleased to and please to), as the main source of the Present-day English please, which would imply that its parenthetical usage would have appeared as a later development (cf. Tieken-Boon van Ostade and Faya Cerqueiro 2007). Pray and please show different features with regard to the experiencer, which has led some scholars to consider that the change of marker may be related to a shift in politeness, since pray forms are more focused on the speaker’s feelings, whereas please forms pay more attention to the addressee’s willingness (cf. Busse 1999, Traugott and Dasher 2002). The case of please parentheticals is particularly remarkable, since they show a shift in their pattern from an experiencer object — found only in highly formulaic expressions, as in and please you — to an experiencer subject, as in if you please. The term ‘parenthetical’ is generally applied to expressions such as pray, please and longer forms like I pray or if you please (cf. OED, s.v. pray v., Huddleston and Pullum 2002: 930, Brinton and Traugott 2005: 137).3 Parentheticals show similar features to comment clauses. In fact, comment clauses are included as a type of parenthetical (cf. Dehé and Kavalova 2007: 2–3, Kaltenböck 2007, 2008). Kaltenböck (2007: 33–45) proposes three formal criteria for the identification of parentheticals, namely (i) syntactic form, excluding nonclausal parentheticals (and therefore restricting the group to those members following the principle of clausal constituency), (ii) lack of syntactic attachment, and (iii) positional flexibility (which only applies to core members of the

|| 3 The label used by Bloomfield in this sense to refer to clause-internal please seems remarkable: “In a form like Won’t you please come? the please is a close parenthesis, without pausepitch” (1933/1984: 186).

An approach to parenthetical markers inconstruction requests ... | 137 Ontogenetic paths tocourtesy the parenthetical

group). If we apply Kaltenböck’s distinctive features, we could say that the courtesy marker please in Present-day English would fail to satisfy the first criterion, while it meets the last two criteria (even the last one, applied only to central members), whereas longer constructions of the parenthetical verb pray found in Late Modern English like I pray (you) would satisfy all the criteria. A related term used in the literature is that of ‘thetical,’ as proposed by Kaltenböck et al. (2011). These authors identify a slight conceptual difference between thetical and parenthetical, using the former as a hypernym (2011: 856). Theticals “consist of a word, a phrase, a clause, or even a chunk that does not form any syntactic constituent. But they may as well have a more complex structure involving clause combining” (2011: 857). These elements show several defining features: with regard to external structure, they are independent from the utterance as regards syntax and prosody, and they usually show mobility; concerning internal structure, theticals have “non-restrictive” meaning, and “their internal structure is built on principles of S[entence] G[rammar] but can be elliptic” (2011: 857). Kaltenböck et al. (2011: 875–876) distinguish a group of ‘formulaic theticals,’ which generally are very frequently used: “short chunks, morphosyntactically unanalyzable, [they] tend to be positionally flexible and to express functions that are mostly procedural, and they relate to the situation of discourse rather than to sentence syntax” (2011: 875). Among formulaic theticals, which would also include many discourse markers, there is a subgroup of formulae of social exchange, such as good morning, hello and please. There were other formulaic structures with parenthetical verbs used in requests already before the Late Modern English period, such as I beg and I beseech (cf. Sönmez 2005), but first pray and later please were the most frequent courtesy markers in this speech act, probably due to the degree of conventionalisation and grammaticalisation reached by both of them. Whereas several other request markers (I pray, I beg, I beseech) were used as parenthetical verbs, this was not the case with please, which became a parenthetical only when it was grammaticalised. Thus, the pragmatic function of polite request markers has shown a need for renewal in the history of English, although it has generally been carried out by parenthetical constructions: Old English ic bidde was replaced by (I) pray (you/thee) in Early Modern English, and please took over from Late Modern English onwards. This study analyses (i) the main request markers in Late Modern English, (ii) the variability of strategies available and their development in this period, and (iii) how the most relevant features generally associated with grammaticalisation match their processes of change. For this purpose several corpora have been taken into account, focusing on three genres: novels, drama and letters, which are described in Section 2. Quantitative and qualitative anal-

138 | Fátima Faya Cerqueiro ysis of examples from these corpora reveal interesting findings as regards the changes affecting the evolution of these request markers in the Late Modern English period. Thus, Section 3 provides a description of the main courtesy markers in requests in this period. An account in terms of grammaticalisation is presented in Section 4.

2 Selection of text-types A preliminary study in the multi-genre corpus ARCHER in the period 1850–1959 (649,170 words) of the parenthetical courtesy markers please, pray and if you please revealed that these items were found mainly in three text-types: fiction, letters and drama (cf. Faya Cerqueiro 2011). Courtesy markers in requests can only occur when some sort of interaction is present, and in these three texttypes interaction acquires special relevance: drama is essentially dialogic, novels generally contain a considerable amount of dialogue, and the epistolary genre assumes a necessary communication between writer and reader. In order to undertake a more in-depth study of these forms and to look for possible sources of please, several corpora were used in order to include fiction, and more specifically novels, drama and letters. As regards novels, data from Chadwyck-Healey Eighteenth-Century Fiction (1700–1780) and NineteenthCentury Fiction (1882–1903) databases was used. A selection of three decades from these collections was made in order to enable a diachronic analysis: 1710– 1720 (646,032 words); 1800–1810 (1,368,202 words); and 1890–1900 (1,728,129 words). Plays were selected from the drama section in A Corpus of Irish English,4 which goes from the sixteenth to the twentieth century (c. 500,000 words). This corpus includes prominent Irish and non-Irish playwrights, such as William Shakespeare, Oscar Wilde and Thomas Sheridan, prestigious figures in English literature as regards both linguistic and literary criteria (Hickey 1994, 2003). Two main epistolary collections were used, namely the Corpus of Late Eighteenth-Century Prose (1761–1790)5 (ca. 300,000 words) and A Corpus of Late Modern English Prose (1860–1919) (100,000 words). A selection of eighteenth-

|| 4 I had access to some of these corpora (A Corpus of Irish English and A Corpus of Late Modern English Prose) with the kind permission of the VARIENG Unit at the University of Helsinki. 5 I am grateful to David Denison for providing me access to this corpus.

An approach to parenthetical markers inconstruction requests ... | 139 Ontogenetic paths tocourtesy the parenthetical

century letter-writing manuals from ECCO database was also considered in order to provide more information on this genre.6

3 Main request markers in Late Modern English 3.1 Requests markers under analysis The analysis in the selection of corpora described in Section 2 included all the different forms of please, taking into consideration its uses as a verb and as a pragmatic marker. There is no agreement among scholars as to the origin of the courtesy marker please. The OED proposes three possible different origins (s.v. please adv. and int.): –

“As a request for the attention or indulgence of the hearer, prob. originally short for please you (your honour, etc.);”

–

“But subsequently understood as short for if you please;”

–

“As a request for action, in immediate proximity to a verb in the imperative, probably shortened from the imperative or optative please followed by the to-infinitive.”

The second choice, namely the reduction of if you please, has usually been favoured in the literature (cf. Traugott and Dasher 2002: 255–257; Brinton 2006: 326). After an analysis of all the different patterns including verbal please, special attention was devoted to the set of structures which looked appealing as the ultimate source of the pragmatic marker please, namely conditional parentheticals, such as if you please and if it please you, among others, and both imperative forms of the verb please followed by a to-infinitive and imperative be pleased to patterns. In fact, in many cases instances of please and be pleased followed by a to-infinitive introduce polite requests. As opposed to what other authors suggest, our data identifies different reasons to exclude a development of the pragmatic marker please from conditional forms. In the data analysed, structures with an experiencer object such as if it

|| 6 I am grateful to the Leiden University Centre for Linguistics for providing me access to this collection.

140 | Fátima Faya Cerqueiro please you had a low frequency already at the beginning of the Late Modern English period. Other experiencer object expressions such as (and) please your honour included a group of honorific noun phrases as the experiencer, and their use decreased from the beginning of the eighteenth century onwards. Parenthetical if you please is also quite fixed as regards the experiencer, which is mainly restricted to the second-person pronoun you. It maintains a similar frequency throughout the period in several genres, as in novels (cf. Tieken-Boon van Ostade and Faya 2007: 432), which does not vary in spite of the emergence of the new pragmatic marker. If you please shows a great mobility in the sentence, being common in middle position in the nineteenth century, as opposed to please, which appears mainly sentence-initially in the same period. In addition, the politeness function of if you please implies a higher degree of formality than Present-day English please and is often close to if you don’t mind. Conversely, there are important changes in the frequency of be pleased to and please to imperatives, which experience a decrease at the same time as please emerges and begins to become a frequent pragmatic marker (cf. TiekenBoon van Ostade and Faya 2007: 435). Other reasons to consider these patterns as a plausible source for the request marker please are shown in Section 3.3. As for pray, instances of the verb were excluded from the study since this form was already grammaticalised in the Late Modern English period as a request marker (cf. Busse 1999, 2002). In fact, instances of the verb were not very frequent in the corpora, and the Corpus of Late Eighteenth-Century Prose revealed no examples of verbal pray. However, the analysis considers all the different forms of the pragmatic marker available in this period, such as (I) pray (you/thee) and (I) prithee, since there was an interesting variability of forms, reflecting several layers in its grammaticalisation (cf. Section 4). The cooccurrence of pray and please as requests markers in Late Modern English, described in the following section, deserves special attention since both forms were used as polite mitigators in requests.

3.2 Pray and please During the Late Modern English period, two main courtesy markers, namely pray and please, were used with the same pragmatic function in requests, as can be observed in (1) and (2), which date back to the late nineteenth century: (1) ‘Pray forgive me!’ he murmured humbly, leaning forwards towards the girl with eyes which deprecated her displeasure. (Novels, 1891, George Gissing, New Grub Street)

An approach to parenthetical markers inconstruction requests ... | 141 Ontogenetic paths tocourtesy the parenthetical

(2) I want to go home very much; so please send the Williamses all away & have the house cleansed. (Letters, 1872, Lord Amberley to wife Kate) The presence of these pragmatic markers mitigates the illocutionary force of the directive speech act, in order to be interpreted as a request. The ability of both pray and please to be used in the same context (i.e. followed by a bare verbal form in a polite request) is found in an early example from an eighteenthcentury grammar, which offers the following comment referring to verbal moods: It may not be indicative and imperative, but subjoined with an uncertainty, a doubt, wish, request—as, if the Sun set—it is necessary that it set—perhaps it may set—I wish it may or may it set—please or pray give—This is speaking in the Subjunctive, Potential, Optative. (Bayly 1772: 34) [italics original; bold mine]

Nevertheless, instances including please followed by a base form are still scarce in the eighteenth century. Some of them are found in The Corpus of Late Eighteenth-Century Prose, which features early instances that only show pre-verbal please, as in (3) and (4) below: (3) You’ll please be so good to Let me know p~ return post when the Lease will be ready (Letters, 1771, Jeffery Hart)7 (4) (Please see over) (Letters, 1788, James Hammond) Evidence of this pragmatic synonymy is clearly due to the fact that several authors or characters use both markers in similar contexts. It seems, nevertheless, that a distinction was already underway by the late nineteenth century or at the beginning of the twentieth century, when speakers started to regard pray as an old-fashioned form. The following figures trace the frequencies of the courtesy markers pray and please in Late Modern English in the three genres that have been considered. Figure 1 offers the evolution in novels, in a selection of decades from the Eighteenth-Century Fiction and the Nineteenth-Century Fiction collections. Figure 2 includes these courtesy markers in letters, in the periods covered by the Corpus of Late-Eighteenth Century Prose and A Corpus of Late Modern English Prose, and Figure 3 represents their frequencies in drama:

|| 7 For the conventions used in the Corpus of Late Eighteenth-Century Prose, see van Bergen & Denison (2007).

142 | Fátima Faya Cerqueiro

Fig. 1: Please and pray in novels

Fig. 2: Please and pray in letters

Ontogenetic paths to the parenthetical construction | 143 An approach to parenthetical markers inconstruction requests ... | 143 Ontogenetic paths tocourtesy the parenthetical

Fig. 3: Please and pray in drama Fig. 3: Please and pray in drama

As shown in the graphs, pray was very frequent in the eighteenth century and probably until thegraphs, mid-nineteenth when in please started to century take over. As shown in the pray was century, very frequent the eighteenth and Both pray and please were used interchangeably as courtesy markers in over. reprobably until the mid-nineteenth century, when please started to take quests probably second of the eighteenth century until wellinadBoth pray and from pleasethewere usedhalf interchangeably as courtesy markers revanced into the nineteenth century. quests probably from the second half of the eighteenth century until well advanced into the nineteenth century.

3.3 Imperative patterns with verbal please 3.3 Imperative patterns with verbal please

Other structures of verbal please are used in the Late Modern English period. Thus, was competing onlyare against thethe pragmatic marker please, but Otherpray structures of verbal not please used in Late Modern English period. also against all the range of expressions with politeness functions in requests Thus, pray was competing not only against the pragmatic marker please, but using verball please. Thus,of weexpressions find instances the patterns be pleased to and also the against the range withofpoliteness functions in requests please to, which were often used in imperative constructions, as in (5) and using the verb please. Thus, we find instances of the patterns be pleased to (6) and below: please to, which were often used in imperative constructions, as in (5) and (6) below: (5) & when You have done Yours in Cheshire, be pleased to order the man to come over to me, & he shall do ours to Buxton &c. (Letters, 1773, John (5) & when You have done Yours in Cheshire, be pleased to order the man Dickenson) to come over to me, & he shall do ours to Buxton &c. (Letters, 1773, John Dickenson) (6) His gray hair streamed on the wind, as he held out the tattered remains of a hat, and said, “Please to help me Lady.---I am very poor.” (Novels, (6) His gray hair streamed on the wind, as he held out the tattered remains 1810, Mary Brunton, Self-Control) of a hat, and said, “Please to help me Lady.---I am very poor.” (Novels, 1810, Mary Brunton, Self-Control)

144 144| |Fátima FátimaFaya FayaCerqueiro Cerqueiro The Theinstances instancesinin(5) (5)and and(6) (6)reveal revealthat thatboth bothpatterns patternscould couldbebeused usedininsimilar similar contexts. contexts.These Theseforms formsofofplease pleaseconveyed conveyedthe thesame samepragmatic pragmaticfunction, function,even even though thoughbebepleased pleasedtotowas wasmore morefrequent frequentduring duringthe theeighteenth eighteenthcentury. century.Figure Figure4 4 below belowrepresents representsthe thetimelines timelinesofofthe themain mainforms formsavailable availableininLate LateModern ModernEngEnglish, lish,based basedon onour ourcorpus corpusanalysis. analysis.Black Blackareas areasindicate indicatethe theperiods periodswhen whenthose those forms formswere wereininnormal normaluse, use,whereas whereasgrey greyand andthinner thinnerareas areassuggest suggestthe theemeremergence genceofofthe theforms formsand andtheir theirdecay decayperiods: periods:

Fig. Fig.4:4:Main Mainrequest requestmarkers markersininLate LateModern ModernEnglish English

Although Althoughwe wemay maysay saythat thatthese theseforms formswere werepragmatic pragmaticsynonyms, synonyms,there theremight might have havebeen beenstylistic stylisticdifferences differencesamong amongthem, them,which whichcaused causedsome someofofthese theseexexpressions pressionstotobebeperceived perceivedasasexcessively excessivelypolite politeatata agiven givenpoint pointinintime. time.ProgresProgres8 8 sively, sively,bebepleased pleasedtotoexpressions expressionsshow showa ahigher higherdegree degreeofofnegative negativepoliteness politeness than thanthe thenewer newerform formplease pleaseto, to,which whichisiseventually eventuallyidentified identifiedasaspoliter politerthan thanthe the proper propercourtesy courtesymarker markerplease. please.AtAtthe thesame sametime, time,pray prayfollows followsa asimilar similarprogresprogression, sion,since sinceititisisassociated associatedwith witholder oldergenerations. generations.Finally, Finally,bebepleased pleasedto, to,please please totoand andpray prayforms formswear weardown, down,asascan canbebeobserved observedininexamples examplesininwhich whichthey theyare are placed placednext nexttotoeach eachother, other,totoreinforce reinforcethe theintended intendedpolite politerequests; requests;we wemay mayfind find pray praytogether togetherwith withimperatives imperativesbebepleased pleasedtotoororplease pleasetotoalready alreadyininthe theeighteighteenth eenthcentury. century. Let Letususobserve observenow nowthe thefollowing followingexamples examplesfrom fromthe theconsecutive consecutivepages pagesofof the thesame sameletter-writing letter-writingmanual. manual.InIn(7) (7)please pleaseisisfollowed followedbybythe thebase baseform formofofa a verb, verb,whereas whereas(8) (8)includes includesananinstance instanceofofthe theverb verbplease pleasefollowed followedbybya atotoinfinitive. infinitive.Both Bothinstances instancesare arefound foundininthe thesame sameletter-writing letter-writingmanual, manual,and andthe the

|| || 8 8Negative Negativepoliteness politenessinvolves involvesthe theassumption assumptionofofa acertain certaindegree degreeofofimposition impositionononthe theadaddressee. dressee.There Thereisisusually usuallya adetachment detachmentfrom fromthe theaddressee, addressee,which whichisismade madeexplicit explicitbybyshowing showing more moredeference deferenceand andmore morerespect. respect.InInrequests requestsit itisisoften oftenconveyed conveyedthrough throughconventional conventionalin-indirectness, directness,pessimism pessimismand andreduction reductionofofthe theimposition. imposition.

An approach to parenthetical markers inconstruction requests ... | 145 Ontogenetic paths tocourtesy the parenthetical

language used in the first letter, including please, is criticised, while the language used in the second one, including please to, is praised: (7) If have not insured, please omit the same till hear farther; (Letterwriting manuals, 1756, The complete letter-writer: or, new and polite English secretary, p.6) (8) Please to send, per first Ship, 150 Chests best Seville, and 200 Pipes best Lisbon, white. (Letter-writing manuals, 1756, The complete letterwriter: or, new and polite English secretary, p.7) The manual criticises different aspects of the letter in which (7) is included, whereas the letter containing (8) is regarded as a much better example of a business letter. Regarding the author of the first one, the manual suggests that “[t]here is nothing in all this Letter though appearing to have the Face of a considerable Dealer.” The intended message is also questioned in the following terms: “By such a way of Writing, no Orders can be binding to him that gives them, or to him they are given to.” By contrast, the second letter is praised, including a revealing comment on the use of please to, as opposed to please in the previous letter, as an expression granting the request and leaving choices to the addressee: “Here is the Order to send a Cargo, with a please to send; So the Factor may let it alone, if he does not please” [italics original]. The intended benefit of the speech act is more likely to be obtained in the second letter, since “Orders ought to be plain and explicit; and he ought to have assured him, that, on his drawing on him, his Bills should be honoured, that is, accepted, and paid” (1756: 6–7). These comments suggest that pre-verbal please was probably felt to be a phenomenon among the lower classes in the eighteenth century, but that it was indeed in use. Both letters contain business language, which generally includes direct strategies. Thus, business letters tend to be shorter than those for other purposes (i.e. family letters), they make use of very immediate language, and are generally written among equals. The requestive strategy with please to, which is highly productive in eighteenth-century letters, was favoured as a more refined courtesy marker from the eighteenth century until probably the end of the nineteenth or the beginning of the twentieth century. In fact, by the turn of the century, please to was felt to be “more formal” than please, as can be inferred from the following comment in a journal: Please To or Please? The imperative ‘‘please’’ may or may not be followed by ‘‘to’’ before an infinitive. Milton’s ‘‘Heavenly stranger, please to taste these bounties’’ is of course more formal, less colloquial, than our everyday ‘‘Please taste this.’’ – Ladies Home Journal, October 1901 (Gold 2006: 108)

146 | Fátima Faya Cerqueiro

3.4 Emergence of parenthetical please Although, as shown above, the OED mentions imperative please followed by a to-infinitive as one of the possible origins of the courtesy marker please, this option has not been considered as a plausible source in the literature. The OED provides a possible explanation of the relationship between this construction with to-infinitive and the first instances of the courtesy marker please followed by a verb form: Examples with bare infinitive complement are now usually analysed as please adv. followed by an imperative. This change probably dates from the development of the adverb, which may stand at the beginning of a clause modifying a main verb in the imperative. (OED, s.v. please v. 6d)

Please would have emerged in sentence-initial positions, in the same contexts where be pleased to and please to patterns occurred. Gradually, be pleased to structures became less common and were replaced by please to patterns. Finally, the to-particle was dropped and please eventually came to precede the bare form of the verb. Taking into consideration the evolutions of the different structures, the following path has been proposed (cf. Tieken-Boon van Ostade and Faya Cerqueiro 2007): be pleased to > please to > please Like other parenthetical pragmatic markers which have emerged in clauseinitial position (cf. Thompson and Mulac 1991, Aijmer 1997 on I think), please has become more mobile within the sentence, and by the end of the nineteenth century it could also be found in clause-final positions, as in (9) below: (9) ‘Ask him to come up, Mrs. Thompson, please.’ (Novels, 1891, George Gissing, New Grub Street) In addition to the different imperative constructions of the verb please, the courtesy marker was exposed to other influences. One of them was the courtesy marker pray itself, which was also typically used preceding a base form of the verb, as shown above. The fact that the different patterns of please coexisted with the already grammaticalised courtesy marker pray could have played a role in the emergence of the new marker. In addition, pre-verbal please occurs very often, not only in imperative constructions, but also in sentences with modal will (cf. example (3) above). The presence of will in requests with please would add a certain nuance to the speech act, since the speaker is taking into consid-

An approach to parenthetical markers inconstruction requests ... | 147 Ontogenetic paths tocourtesy the parenthetical

eration the addressee’s willingness, and would be used as a mitigator in the utterance. This would be related to the shift observed by Busse (2002) in Shakespeare’s time: [A]t least in colloquial speech a shift in polite requests has taken place from requests that assert the sincerity of the speaker (I pray you) to those that question the willingness of the listener to perform the request (please). (Busse 2002: 212)

4 Grammaticalisation of pray and please The developments of both pray and please show typical features of grammaticalisation. Different scholars have suggested that their development constitutes a case of grammaticalisation. The older courtesy marker in requests analysed here, pray, is studied by Akimoto from this perspective. Akimoto (2000) focuses on the evolution of pray, prithee and their different forms from the fifteenth to the nineteenth centuries, and identifies Hopper’s (1991) five principles of grammaticalisation in the development of pray (2000: 78–80). Busse (1999, 2002) examines the occurrence of pray and prithee and their variants in Shakespeare as well as the pronouns used with each marker, considering both the presence/absence of the subject (I) and the inclusion of T or V object pronouns (thee/you). He shows that these markers were already grammaticalised by Shakespeare’s time, and that pray constructions had already acquired adverb-like qualities, since they no longer behaved as main clauses: the verbal qualities of pray are lost and the expression may occupy different positions within the clause. He proposes the following path for the development of both markers: †I pray you/thee > †pray you/thee > prithee/pray! (1999: 496). Thus, according to Busse, both pray and prithee have undergone recategorialisation, semantic bleaching and pragmatic strengthening, while prithee also shows phonetic reduction. He also suggests that differences in politeness and discourse functions may have conditioned the evolution of both pray and prithee, and that the shift from positive to negative politeness suggested by Kopytko (1993, 1995) for Elizabethan English may have played a role in the replacement of pray by please. Culpeper and Archer (2008: 74–76) nevertheless consider that pray was not yet fully grammaticalised in Early Modern English, since it typically occurred as a pre-support move, usually in initial position. In this respect, pray and prithee differ from please, which may occupy different slots within the sentence. This fixedness as regards word order is further supported by Corbet’s comment on the syntactic properties of pray in an eighteenth-century grammar, “Syntax

148 | Fátima Faya Cerqueiro teaches us to range or place our Words in a proper Order, as Pray, Sir, dine with me To-day; not Dine with me To-day, pray, Sir” (Corbet 1784: viii). Traugott and Dasher (2002) take a similar stand on the development of the pragmatic marker pray, although grammaticalisation is not explicitly mentioned. They consider that pray developed from a main clause performative expression (I pray you) into a parenthetical expression and finally into a pragmatic marker with social deictic function (2002: 252–255). Similarly, Brinton (2006, 2007a, 2008, 2010) appeals to grammaticalisation to account for the development of pray. She observes that the matrix clause I pray (you) could be followed by different complement clauses at the earlier stages. Progressively, the set of complement clauses was reduced, the object omitted and eventually the subject as well. Thus, after an intermediate stage between matrix clause and parenthetical, pray became a pragmatic marker. As mentioned above, Akimoto (2000: 78) identifies Hopper’s (1991) five principles of grammaticalisation in the development of pray. Nevertheless, it should be noted that the same principles are mentioned in the literature as defining features of pragmaticalisation (Aijmer 1997, 2002), and, in fact, Claridge & Arnovick (2010) base their description of pray in terms of pragmaticalisation on Akimoto’s claims. As regards layering, Akimoto (2000) mentions forms such as I pray you, I pray thee or I pray, which were used contemporaneously. This variation is reflected in the following examples extracted from the same play in A Corpus of Irish English, the parenthetical verb structures include I pray and I pray thee as in (10) and (11), while in (12) pray is alone and in (13) the courtesy marker prithee is used: (10) Hip. Well, sir, I pray lets see you Master Scholler. (Drama, 1605/1630, Thomas Dekker, The Honest Whore Part II) (11) Bryan. Come I pray dee, wut come sweet face? Goe. (Drama, 1605/1630, Thomas Dekker, The Honest Whore Part II) (12) Cand. What’s that he saies, pray’ Gentlemen? (Drama, 1605/1630, Thomas Dekker, The Honest Whore Part II) (13) Bryan. I faat, I doubt my pate shall be knocked: but so crees sa me, for your shakes, I will runne to any Linnen Draper in hell, come preddy. (Drama, 1605/1630, Thomas Dekker, The Honest Whore Part II) Akimoto (2000) also mentions the principle of divergence, which is observable in the development of the courtesy marker pray, since alongside the courtesy marker, the verb pray is also found, as in (14), where the religious sense of the verb is clear:

An approach to parenthetical markers inconstruction requests ... | 149 Ontogenetic paths tocourtesy the parenthetical

(14) DORINDA Morrow, my dear sister, are you for church this morning? MRS. SULLEN Anywhere to pray, for heaven alone can help me. (Drama, 1707, George Farquhar, The Beaux’ Stratagem.) Concerning specialisation, Akimoto lists other forms such as I entreat you, which could have developed along a similar path¸ but notes that only I pray you specialised as a courtesy marker in requests. Late Modern English used several verbs with meanings related to ‘request’ in the speech act of requests, most of which were used performatively. Examples include I beg, I beseech and I entreat (cf. Akimoto 2000; Sönmez 2005). The core lexical meaning of the verb pray, ‘to ask earnestly,’ is shared with most of these requestive verbs. Among all the forms available, only pray was grammaticalised as a courtesy marker, thus undergoing Hopper’s (1991) principle of specialisation. Nevertheless, in eighteenth-century letters some examples of the performative beg can be found without an explicit subject, followed by a to-infinitive, a pattern similar to the one found with please, as in (15). Note that in (16), beg is found in final position, which clearly indicates that it is a parenthetical: (15) Beg now mr Grimsw, to send my Goose to London when convenient to him. (Letters, 1783, John Amson) (16) Please to send me a ansr. to the Inclosd by first as We shall not go before Wensday fail not I Beg (Letters, ?1790, Thomas Gaskell) Note that example (16) also includes the imperative of please followed by a toinfinitive, showing again the formulaic character of business epistolary language. Instances with beg are very frequent in the Corpus of Late EighteenthCentury Prose, but this verb was used in an ample range of expressions, such as beg pardon, beg excuse, beg leave (OED, s.v. beg v. 3a), and did not develop further as a parenthetical courtesy marker. According to Akimoto, persistence is also observed in the courtesy marker pray, which “still retains part of the original meaning of the verb in the sense of supplication” (2000: 78). The requestive value of the verb is even more obvious in the full construction I pray you, as in (17): (17) I have now all responsibility for Jamie on my head, and I have to do what my conscience tells me I should do; only, I pray you, do not take offence at what I have said.” (Novels, 1892, Sabine Baring-Gould, In The Roar Of The Sea) Furthermore, the courtesy marker pray undergoes de-categorialisation. Akimoto observes that over the time pray loses its verbal features and behaves more like

150 | Fátima Faya Cerqueiro an interjection. This is also visible in prithee, which underwent further decategorialisation from I pray thee to become an “interjectional marker” (2000: 78). The development of the courtesy marker pray also shows some of the parameters and processes proposed by Lehmann (1995), which were checked against our findings. In the case of the courtesy marker pray, we observe a similar semantic loss in its development from the verb to pray, while phonological attrition is clearly present in the development from I pray thee to prithee, which shows univerbation. The second of Lehmann’s parameters is paradigmaticity and its related process is paradigmaticisation. Although this process of integration into a paradigm is usually disregarded in the literature as a typical feature in the development of pragmatic markers (cf. Brinton 1996; Wischer 2000), as long as we think of pragmatic or courtesy markers as members of a word-class, such as Biber et al.’s (1999) ‘inserts,’ pray would show this quality. The process affecting the parameter of paradigmatic variability is obligatorification, another feature questioned in the literature as a defining criterion of grammaticalisation (Wischer 2000). This status was probably not reached by pray due to the paradigmatic variability in the whole Late Modern English period. As regards the syntagmatic parameters, the first deals with structural scope, and its related process is condensation. This parameter is not typically associated with the development of pragmatic markers (cf. Brinton 1996; Traugott 1997; Tabor and Traugott 1998), which show an increase in scope, since once grammaticalised these forms can affect more complex constituents. No reduction in scope is shown by the courtesy marker pray in the Late Modern English period. The parameter of bondedness, and its consequent process of coalescence is also a problematic feature of grammaticalisation. It has often been questioned in the literature and, generally, not identified in the development of pragmatic markers (Brinton 1996; Traugott 1997; Wisher 2000). Though absent from the grammaticalisation path of pray, coalescence is observed in the development of prithee from I pray thee, as noticed by Claridge and Arnovick (2010). Finally, the last parameter proposed by Lehmann (1995) is syntagmatic variability. The variability of an item is progressively reduced, until a form reaches fixation. The pragmatic marker pray in some corpora is attested in initial position only, as in the Corpus of Late Eighteenth-Century Prose (cf. Faya Cerqueiro 2007), showing a tendency to occupy a fixed slot in the sentence. Several authors have also paid attention to the development of please as an example of grammaticalisation. Following Kopytko (1993) and on the basis of the data offered by the OED, Busse considers that please has also undergone grammaticalisation. He proposes the following steps in its development: †please it you (= may it please you) > †please you > please! (1999: 496). Traugott and

An approach to parenthetical markers inconstruction requests ... | 151 Ontogenetic paths tocourtesy the parenthetical

Dasher (2002: 255–258) explain that please could have originated in an expression such as if you please, in which the experiencer occupies a subject position, both this experiencer and the conditional conjunction were left out and the expression developed into the social deictic. Brinton (2006, 2007a, 2008, 2010) also sees the development of please in terms of grammaticalisation. Brinton follows Allen’s (1995) explanations on the development of impersonal and personal constructions with please (if it please you and if you please) in the seventeenth century. She identifies an adverbial clause with a nominative Experiencer, namely if you please, as the possible source of please, which replaced pray at the beginning of the twentieth century. Hopper’s (1991) principles, which are generally identified at the beginning of the process of grammaticalisation, are found in the development of please. Taking into consideration all the data gathered and the diachronic development of please, it seems that please is a prototypical example of this process of change. In terms of layering, the development of the courtesy marker please illustrates layering in the two conceptions of the term. On the one hand, we identify the emergence of layers within the functional domain (cf. Hopper 1991: 23) of courtesy markers in requests. In fact, pray and please coexist from the late eighteenth century until the beginning of the twentieth century, a period when both markers, the old and the new form, were available with exactly the same function, as in (18) and (19). In addition to the courtesy marker pray, the whole range of forms including the verb please, and especially be pleased to and please to, were used in similar contexts with the same pragmatic function as the courtesy marker please, as in (20) and (21), taken from family and business letters. (18) Pray let me know in a Line, whether you are better or worse, whether I am honest or a Knave, and whether I shall live or die. (Letter-writing manuals, 1765?, The British letter-writer, p. 112) (19) Please let me know whether I may see you. (Novels, 1893, George Gissing, The Odd Women) (20) If you have any particular cause for insisting on it sooner, be pleased to let me know, and I will endeavour to borrow the money; (Letterwriting manuals, 1782?, Every man his own letter-writer, p. 19) (21) Monday the 16th Inst:, is the day appointed for another meeting at Prescot, to sign the articles &c; which I hope will be convenient for you

152 | Fátima Faya Cerqueiro to attend. if it is not, please to let me know; (Letters, 1778, Thomas Penwick) On the other hand, layering can be understood as the coexistence of new and old meanings and uses of a form (cf. Hopper and Traugott 2003: 49). In this sense, the verb please and the courtesy marker please have been available in English since the emergence of the latter. This coexistence is also related to the principle of divergence: whereas the form please has developed into a pragmatic marker, the verb please is still kept as an autonomous lexical item in different syntactic structures. Concerning specialisation, different forms of pray and please coexisted for a few centuries, but gradually the choice of forms narrowed down, until please became the only courtesy marker in requests. The inclusion of please is almost compulsory in some contexts from a pragmatic perspective, though not morpho-syntactically as in obligatorification (Lehmann’s 1995 process). In fact, the presence of please makes a question easily identifiable as a request (BlumKulka 1985), a property which is no longer available for other forms, unless they convey an exaggerated and marked tone of negative politeness or even irony. The principle of persistence is also shown by this form, since the original lexical meaning of the verb please ‘to have the will or desire’ (OED, s.v. please v. 4b) may have restricted the functions of please as a courtesy marker to the field of requests. In a previous step, the adoption of the passive meaning ‘be pleased’ by the verb please could have contributed to the semantics and pragmatics of please. This value is clearer in pre-verbal instances, such as (22), which is an obvious appeal to the hearer. The head act of the request could be paraphrased as will you have the will to thank her: (22) I have a long letter from Beatrice – will you please thank her for it if you’re seeing her... (Letters, 1917, Gertrude Bell to father) The development of the verb please towards the homonymous pragmatic marker also illustrates another of Hopper’s parameters, namely de-categorialisation. In this case it entails the change from a major class (verb) to a minor class (adverb/pragmatic marker). The courtesy marker please lacks verbal properties, it loses all the morphological and syntactic properties of verbs, that is, it no longer shows inflectional properties, it cannot take subjects, objects or complements, and the form becomes fossilised. As regards semantics, the form please loses its descriptive meaning (i.e. ‘to be agreeable to’) and develops procedural meaning when it is understood as part of a polite request.

An approach to parenthetical markers inconstruction requests ... | 153 Ontogenetic paths tocourtesy the parenthetical

Some of the parameters and processes proposed by Lehmann (1995) are also present in the development of please. Among paradigmatic parameters, the first process is attrition, which relates to the loss of integrity, both phonological and semantic. The semantic change of please develops from ‘please’ to ‘be pleased’ (cf. OED, s.v. please v.), and this shift in meaning would be accompanied by phonological loss. Thus, in the development from be pleased to > please to, proposed here as the major source of the courtesy marker, the verb to be is dropped. Moreover, the final plosive of the participle is also lost, thus making it possible for the verb to be reinterpreted as either a bare form or as an imperative. In the stage please to > please, the particle to is lost, and this loss enables the reanalysis of the verb as a courtesy marker. The whole process reduces the semantic weight of the initial construction, also reducing its descriptive meaning. Therefore, if the developmental path proposed in this study is correct, the grammaticalisation of the courtesy marker please illustrates attrition.9 As regards paradigmaticity and the process of paradigmaticisation, this process of integration into a paradigm is also found in please. As mentioned above, please may be necessary in certain contexts to understand a question as a request or to mitigate the threat of an order to the negative face. The choice of forms available during the Late Modern English period which offer the speaker paradigmatic variability was progressively reduced until please became the only unmarked courtesy marker in polite requests. Since its omission could affect the pragmatic value of an utterance, we may refer to a certain degree of obligatorification in its development, which would not be reached until the twentieth century, when please becomes almost the only choice for the speaker (cf. examples (18)–(21) above). As for syntagmatic parameters, structural scope, and its related process condensation, in pre-verbal position, the courtesy marker please would affect exactly the same constituents as the verb in the please to construction did (Would you please tell me?/Would you please to tell me), but once it moves to different positions, especially sentence-initially or sentence-finally, it may affect requests with even more complex head acts. Regarding bondedness, and its process, far from showing any tendency to coalescence, please is a prominent prosodic element in requests (cf. Wichmann 2004, 2005) and it may be used as a stand-alone word in certain contexts, such as the acceptance of an offer. Concerning Lehmann’s (1995) syntagmatic variability, in the case of please, fixation

|| 9 Within this process, the property of phonological attrition is questioned in the literature as necessary for a form to be regarded as a case of grammaticalisation (cf. Brinton 1996: 273).

154 | Fátima Faya Cerqueiro as regards word order would only be a tendency in the initial stages of grammaticalisation, when please usually occupies a slot as a clause-initial element, or more accurately a pre-verbal position. However, as is the case with other pragmatic markers, please shows increased mobility within the sentence, even though its presence affects the meaning of the whole head act regardless of its position. Thus, the development of please (and pray) fails to show some alleged properties of grammaticalisation, which are not found in the developmental paths of other pragmatic markers either; Lehmann’s syntagmatic parameters, namely condensation, fixation and coalescence are not identified in the development of the courtesy marker please. In addition to these typical features in the grammaticalisation of pragmatic markers, Brinton (2005: 291–293, 2007b: 62) mentions conventionalisation of conversational implicatures. In this respect, Blum-Kulka (1985) shows that the presence of please causes a question to be understood as a request (cf. also Sadock 1974, Geukens 1978, Leech 1983). Following this approach, Aijmer (1996: 144) mentions indirect requests with please as instances of conventionalisation, and, in fact, dictionaries include its function as a request mitigator. Grammaticalisation requires the existence of bridging and critical contexts (cf. Diewald 2002, Heine 2002, Traugott 2010). It is very likely that (23) represents a bridging or critical context, in which the form could be interpreted as a verb or as a pragmatic marker. The earliest instances of please found in the eighteenth century correspond to similar contexts. Thus, pre-verbal positions would favour a later reinterpretation of please as a courtesy marker. (23) So you’ll please pay me the two hundred pounds which he owed to Mr John Dykes. (Novels, 1810, Mary Brunton, Self-Control) Once the form is reanalysed, please can move from former ambiguous strings, as in (23), to new, non-ambiguous contexts, as in (24): (24) SUSIE For the future, during the period you are patients here, I am to be addressed as “Nurse Monican”, and not as “Susie”. Remember that, the pair of you, please. (Drama, 1928, Sean O’Casey, The Silver Tassie) The development proposed here matches Brinton’s (2007a, 2008, 2010) description of some clausal pragmatic markers, namely those pragmatic markers which originate in a matrix clause, and more specifically, one showing a ‘secondperson imperative’ (imperative matrix clause > indeterminate structure > parenthetical discourse marker), since it is likely that the most prominent sources are the imperative constructions be pleased to and please to. Imperatives are the

An approach to parenthetical markers inconstruction requests ... | 155 Ontogenetic paths tocourtesy the parenthetical

origin of other pragmatic markers in English, such as in look, see and listen, all of them sensory verbs with attention-getter functions, which retain the original imperative nuance. Another example of a pragmatic marker originating in an imperative is mind/mind you (Brinton 2007a: 57) as in We was still hard up, mind you (1987, OED, s.v. mind v. 12b). Again, it seems to also keep an attention-getter effect on the hearer. There are several examples in the literature of pragmatic markers deriving from imperative verb constructions in other languages, such as Italian guarda (Waltereit 2002) or Spanish oye/oiga (Pons Bordería 1998) and dale (Company Company 2006). Dostie (2004) identifies the same pattern in French discourse markers écoute, regarde, voyons, tiens or disons, all of them originated in imperatives, and understood as instances of pragmaticalisation by this author. Subjectification, a process commonly associated to grammaticalisation, is observed in the developments of pray and please as courtesy markers in requests. These pragmatic items help the speaker to mark the illocutionary force of the directive speech act, and to determine its intended illocutionary strength. However, the use of please and pray as pragmatic markers is closely related to the hearer, since these forms can only take place in the social exchange. Please as a request marker occurs when the speaker shows deference and respect towards the hearer, taking into account his/her willingness to perform the requested act. Thus, the courtesy marker illustrates both subjectification and intersubjectification, processes which are clearly present in both requestive and non-requestive uses of please. In the latter the speaker expresses his/her attitudes and feelings, and what he/she says affects the addressee’s feelings and opinions as well. In this way, when the illocutionary force is not related to politeness, please is an attitude marker. Ironic uses10 may reflect a further step of grammaticalisation (and subjectification), since they are only found once the form is conventionalised as a request marker. This only happens to please in the corpora analysed in the twentieth century, whereas there are no instances of pray with such a pragmatic value, probably because it fell into disuse before reaching such a step in the grammaticalisation process, or maybe its semantic connotations did not facilitate this function. Both pray and please have been referred to in the literature as examples of different processes. Some authors have accounted for pray and please in terms of grammaticalisation (cf. Akimoto 2000), while others have opted for pragmat-

|| 10 Note the irony in this instance extracted from the OED, which belongs to Rushdie’s Ground beneath her Feet (1999): What, they should let the kids decide what they put on the air? Please (OED s.v. please adv. and int.).

156 | Fátima Faya Cerqueiro icalisation (cf. Claridge and Arnovick 2010). The difference between pragmaticalisation and the grammaticalisation of pragmatic markers has to do mainly with how broadly the concept of grammar is interpreted by the scholar, but the characteristics identified in both processes remain roughly the same. Therefore, in a more restricted view of grammar, the label pragmaticalisation could be used to define the developments of please or pray as courtesy markers, although such a distinction does not seem necessary to account for the peculiarities of these pragmatic items. Although the theoretical grounds of pragmaticalisation as understood by Aijmer (1997) and Claridge and Arnovick (2010), among others, match the development of pragmatic features, the need for a new label is still debatable.

5 Concluding remarks Taking into consideration the data gathered from different corpora and databases, it seems very likely that the courtesy marker please emerged in a conventionalised imperative construction of the please to type. When the to-particle was dropped, please could be interpreted either as a courtesy marker or as an imperative verb. Early instances of the courtesy marker please are found in pre-verbal position, and most commonly in initial position, where imperatives usually occur. As opposed to pray, which originates in a parenthetical verb pattern (I pray you), the emergence of please as a parenthetical would have taken place later. Only then, the imperative was reanalysed as a pragmatic marker, and spread from pre-verbal positions to any place in the sentence. Thus, the development of please would match Brinton’s (2010) cline of parentheticals originating in imperative clauses: imperative matrix clause > indeterminate structure > parenthetical discourse marker. The grammaticalisation processes of both pray and please show similar features to those found in the development of other pragmatic markers. Akimoto (2000) identifies Hopper’s (1991) principles in the grammaticalisation of pray, and they are also present in the grammaticalisation of please. Lehmann’s (1995) syntagmatic parameters are not observed in the development of these courtesy markers, although some of his paradigmatic parameters can be noticed, namely attrition and paradigmaticisation. Other typical characteristics of grammaticalisation observable in the development of please are the conventionalisation of conversational implicatures as well as subjectification and intersubjectification. Further steps in the grammaticalisation of please are shown by the acquisition

An approach to parenthetical markers inconstruction requests ... | 157 Ontogenetic paths tocourtesy the parenthetical

of new pragmatic uses, which developed only when the politeness requestive function was conventionalised.

References Aijmer, Karin. 1996. Conversational Routines in English: Convention and Creativity. London: Pearson Education (Longman). Aijmer, Karin. 1997. I think – an English modal particle. In Toril Swan & Olaf Jansen Westvik (eds.). Modality in Germanic Languages: Historical and Comparative Perspectives. Berlin: Mouton de Gruyter. 1–47. Aijmer, Karin. 2002. English Discourse Particles. Amsterdam/Philadelphia: John Benjamins. Akimoto, Minoji. 2000. “The grammaticalization of the verb ‘pray’.” In Olga Fischer, Anette Rosenbach & Dieter Stein (eds.). Pathways of Change. Grammaticalization in English. Amsterdam/Philadelphia: John Benjamins. 67–84. Allen, Cynthia L. 1995. On doing as you please. In Andreas H. Jucker (ed.). Historical Pragmatics. Amsterdam/Philadelphia: John Benjamins. 275–308. Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad & Edward Finegan. 1999. Longman Grammar of Spoken and Written English. London: Longman. Bloomfield, Leonard. 1984/1933. Language. Chicago: The University of Chicago Press. Blum-Kulka, Shoshana. 1985. Modifiers as indicating devices: The case of requests. Theoretical Linguistics 12 (2/3). 213–229. Bayly, Anselm. 1772. A Plain and Complete Grammar of the English Language. London. http://galenet.galegroup.com/servlet/ECCO Beal, Joan. 2004. English in Modern Times: 1700-1945. London: Hodder Arnold Publication. Biber, Douglas, Stig Johansson, Geoffrey Leech, Susan Conrad & Edward Finegan. 1999. Longman Grammar of Spoken and Written English. London: Longman. Brinton, Laurel. 1996. Pragmatic Markers in English: Grammaticalization and Discourse Functions. Berlin/New York: Mouton de Gruyter. Brinton, Laurel. 2005. Processes underlying the development of pragmatic markers: The case of (I) say. In Janne Skaffari, Matti Peikola, Ruth Carroll, Risto Hiltunen & Brita Wårvik (eds.). Opening Windows on Texts and Discourses of the Past. Amsterdam and Philadelphia: John Benjamins. 279–299. Brinton, Laurel. 2006. “Pathways in the development of pragmatic markers in English.” In Ans van Kemenade & Bettelou Los (eds.). The Handbook of the History of English. London: Blackwell. 307–334. Brinton, Laurel. 2007a. What’s more: The development of pragmatic markers in the modern period. In Javier Pérez-Guerra et al. (eds.). ‘Of Varying Language and Opposing Creed’: New Insights into Late Modern English. Bern, etc.: Peter Lang. 47–75. Brinton, Laurel. 2007b. The development of I mean: Implications for the study of historical pragmatics. In Susan Fitzmaurice & Irma Taavitsainen (eds.). Methods in Historical Pragmatics. Berlin/New York: Mouton de Gruyter. 37–79. Brinton, Laurel. 2008. The Comment Clause in English: Syntactic Origins and Pragmatic Development. Cambridge: CUP.

158 | Fátima Faya Cerqueiro Brinton, Laurel. 2010. Discourse markers. In Andreas H. Jucker & Irma Taavitsainen (eds.). Historical Pragmatics. Berlin/New York: Mouton de Gruyter. 285–314. Brinton, Laurel & Elizabeth Closs Traugott. 2005. Lexicalization and Language Change. Cambridge: CUP. Busse, Ulrich. 1999. ‘Prithee now, say you will, and go about it.’ Prithee vs. pray you as discourse markers in the Shakespeare corpus. In Fritz-Wilhelm Neumann & Sabine Schilling (eds.). Anglistentag 1998, Erfurt: Proceedings. Trier: Wissenschaftlicher Verlag. 485–500. Busse, Ulrich. 2002. Linguistic Variation in the Shakespeare Corpus. Morpho-syntactic Variability of Second Person Pronouns. Amsterdam/Philadelphia: John Benjamins. Claridge, Claudia and Leslie Arnovick. 2010. Pragmaticalisation and Discursisation. In Andreas H. Jucker & Irma Taavitsainen (eds.). Historical Pragmatics. Berlin/New York: Mouton de Gruyter. 165–192. Company Company, Concepción. 2006. Zero in syntax, ten in pragmatics: Subjectification as syntactic cancellation. In Angeliki Athanasiadou, Costas Canakis & Bert Cornillie (eds.). Subjectification: Various Paths to Subjectivity. Berlin/New York: Mouton de Gruyter. 375– 398. Corbet, John, of Shrewsbury. 1784. Concise System of English Grammar. Shrewsbury. Eighteenth Century Collections Online. Gale Group. . Culpeper, Jonathan & Dawn Archer. 2008. Requests and directness in Early Modern English trial proceedings and play-texts, 1640–1760. In Andreas H. Jucker & Irma Taavitsainen (eds.). Speech Acts in the History of English. Amsterdam/Philadelphia: John Benjamins. 45–84. Dehé, Nicole & Yordanka Kavalova. 2007. Parentheticals: An introduction. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam/ Philadelphia: John Benjamins. 1– 22. Diewald, Gabriele. 2002. A model for relevant types of contexts in grammaticalization. In Ilse Wischer & Gabriele Diewald (eds.). New Reflections on Grammaticalization. Amsterdam/ Philadelphia: John Benjamins. 103–120. Dossena, Marina &Charles Jones. 2003. Introduction. In Marina Dossena & Charles Jones (eds.). Insights into Late Modern English. Bern, etc.: Peter Lang. 7–17. Dostie, Gaétane. 2004. Pragmaticalisation et Marqueurs Discursifs: Analyse Sémantique et Traitement Lexicographique. Brussels: De Boeck and Duculot. Faya Cerqueiro, Fátima. 2007. The courtesy markers pray and please in the late 18th century: Evidence from the Corpus of Late Eighteenth-Century Prose. In María Losada Friend, Pilar Ron Vaz, Sonia Hernández Santano & Jorge Casanova (eds.). Proceedings of the 30th International AEDEAN Conference. Huelva: Servicio de Publicaciones de la Universidad de Huelva. [CD-Rom] Faya Cerqueiro, Fátima. 2011. Request markers in drama: Data from the Corpus of Irish English. In Carrió Pastor, María Luisa & Miguel Ángel Candel Mora (eds.). Las tecnologías de la información y las comunicaciones: Presente y futuro en el análisis de córpora. Actas del III Congreso Internacional de Lingüística de Corpus. Valencia: Universitat Politècnica de València. 535–542. Geukens, Steven K.J. 1978. The distinction between direct and indirect speech acts: Towards a surface approach. Journal of Pragmatics 2. 261–276. Gold, David L. 2006. A little-noticed English construction: Subjectless imperative please to + infinitive. Neophilologus 90. 107–117.

An approach to parenthetical markers inconstruction requests ... | 159 Ontogenetic paths tocourtesy the parenthetical

Hickey, Raymond. 1994. A Corpus of Irish English. In Merja Kytö, Matti Rissanen & Susan Wright (eds.). Corpora across the centuries. Amsterdam/ Atlanta: Rodopi. 23–32. Hickey, Raymond. 2003. Corpus Presenter: Software for Language Analysis with a Manual and A Corpus of Irish English as Sample Data. Amsterdam/ Philadelphia: John Benjamins. Heine, Bernd. 2002. On the role of context in grammaticalization. In Ilse Wischer & Gabriele Diewald (eds.). New Reflections on Grammaticalization. Amsterdam/Philadelphia: John Benjamins. 83–101. Hopper, Paul. 1991. On some principles of grammaticization. In Elisabeth C. Traugott & Bernd Heine (eds.). Approaches to grammaticalization. Vol. 1. Amsterdam: John Benjamins. 17– 35. Hopper, Paul & Elizabeth Closs Traugott. 2003. Grammaticalization. 2nd edn. Cambridge: CUP. Huddleston, Rodney &Geoffrey K. Pullum. 2002. The Cambridge Grammar of the English Language. Cambridge: CUP. Kaltenböck, Gunther. 2007. Spoken parenthetical clauses in English: A taxonomy. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam and Philadelphia: John Benjamins. 25–52. Kaltenböck, Gunther. 2008. Prosody and function of English comment clauses. Folia Linguistica 42 (1). 83–134. Kaltenböck, Gunther, Bernd Heine & Tania Kuteva. 2011. “On thetical grammar.” Studies in Language 35 (4). 852–897. Kopytko, Roman. 1993. Polite Discourse in Shakespeare’s English. Poznan: Adam Mickiewicz University Press. Kopytko, Roman. 1995. Linguistic Politeness Strategies in Shakespeare’s Plays. In Andreas H. Jucker (ed.). Historical Pragmatics. Amsterdam/Philadelphia: John Benjamins: 515–540. Leech, Geoffrey. 1983. Principles of Pragmatics. London: Longman. Lehmann, Christian. 1995. Thoughts on Grammaticalization. Munich/Newcastle: Lincom Europa. OED = Oxford English Dictionary online. . Pons Bordería. 1998. Los apelativos oye y mira o los límites de la conexión. In Martín Zorraquino, Mª Antonia and Estrella Montolío (eds.). Marcadores discursivos: Teoría y práctica. Madrid: Arco. 213–228. Sadock, Jerrold M. 1974. Towards a Linguistic Theory of Speech Acts. New York: Academic Press. Sönmez, Margaret J-M. 2005. A study of request markers in English family letters from 1623 to 1660. European Journal of English Studies 9 (1). 9–19. Tabor, Whitney & Elizabeth Closs Traugott. 1998. Structural scope expansion and grammaticalization. In Anna Giacalone Ramat and Hopper, Paul J. (eds.). The Limits of Grammaticalization. Amsterdam/Philadelphia: John Benjamins. 229–272. Tieken-Boon van Ostade, Ingrid & Fátima Faya Cerqueiro. 2007. Saying please in Late Modern English. In Javier Pérez-Guerra et al. (eds.). ‘Of varying language and opposing creed’: New insights into Late Modern English. Bern, etc.: Peter Lang. 421–444. Thompson, Sandra A. & Anthony Mulac. 1991. A quantitative perspective on the grammaticization of epistemic parentheticals in English. In Elisabeth C. Traugott & Bernd Heine. (eds.). Approaches to grammaticalization. Vol. 2. Amsterdam/Philadelphia: John Benjamins. 313–329.

160 | Fátima Faya Cerqueiro Traugott, Elizabeth Closs. 1997. The role of the development of discourse markersin a theory of grammaticalization. ICHL XII, Manchester 1995, Version of 11/97. [online] . (accessed October 2013) Traugott, Elizabeth Closs. 2000. Promise and Pray-Parentheticals. Eleventh International Conference on English Historical Linguistics (ICEHL XI), Santiago de Compostela, September 2000. Traugott, Elizabeth Closs. 2010. Grammaticalization. In Andreas H. Jucker & Irma Taavitsainen (eds.). Historical Pragmatics. Berlin/ New York: Mouton de Gruyter. 97–126. Traugott, Elizabeth Closs & Richard B. Dasher. 2002. Regularity in Semantic Change. Cambridge: CUP. van Bergen, Linda & David Denison. 2007. A Corpus of Late Eighteenth-Century Prose. In Joan C. Beal, Karen P. Corrigan & Hermann Moisl (eds.). Creating and Digitizing Language Corpora. Vol. 2: Diachronic corpora. Basingtroke: Palgrave Macmillan. 228–246. Waltereit, Richard. 2002. Imperatives, interruption in conversation, and the rise of discourse markers: A study of Italian guarda. Linguistics 40 (5). 987–1010. Wichmann, Anne. 2004. The intonation of Please-requests: A corpus-based study. Journal of Pragmatics 36. 1521–1549. Wichmann, Anne. 2005. Please: From courtesy to appeal: The role of intonation in the expression of attitudinal meaning. English Language and Linguistics 9 (2). 229–253. Wischer, Ilse. 2000. Grammaticalization versus lexicalization: ‘Methinks’ there is some confusion. In Olga Fischer, Anette Rosenbach & Dieter Stein (eds.). Pathways of Change: Grammaticalization in English. Amsterdam/ Philadelphia: John Benjamins. 355–370.

An approach to parenthetical markers inconstruction requests ... | 161 Ontogenetic paths tocourtesy the parenthetical

Corpora 

ARCHER = A Representative Corpus of Historical English Registers 3.1. 1990– 1993/2002/2007/2010. Compiled under the supervision of Douglas Biber and Edward Finegan at Northern Arizona University, University of Southern California, University of Freiburg, University of Heidelberg, University of Helsinki, Uppsala University, University of Michigan, University of Manchester, Lancaster University, University of Bamberg, University of Zurich, University of Trier, University of Salford, and University of Santiago de Compostela. Denison, David (ed.). 1994. A Corpus of Late Modern English Prose. University of Manchester: Department of English and American Studies, formerly Department of English Language and Literature. (Text of The Letters of Ernest Dowson: Associated University Presses, NJ, USA). Denison, David (dir.) and Linda van Bergen (ed.). 2003. Corpus of Late Eighteenth-Century Prose. University of Manchester Faculty of Arts and John Rylands University Library of Manchester. Eighteenth-Century Fiction. 1996. Electronic Book Technologies Inc./Chadwyck-Healey. Cambridge. ECCO = Eighteenth Century Collections Online. Gale Group. . Hickey, Raymond. 2003. Corpus Presenter: Software for Language Analysis with a Manual and A Corpus of Irish English as Sample Data. Amsterdam and Philadelphia: John Benjamins. Nineteenth-Century Fiction. 1999–2000. Electronic Book Technologies Inc./Chadwyck-Healey. Cambridge.

Stefan Schneider and Julie Glikman

Origin and development of French parenthetical verbs Abstract: Regarding the emergence of parenthetical verbs, three possible sources have been proposed in the literature: superordinate clauses, subordinate clauses, and juxtaposed sentences. This contribution analyses their validity with respect to parenthetical verbs derived from croire ‘believe’, cuidier ‘believe’, espérer ‘hope’, penser ‘think’ and sembler ‘seem’ in Old, Middle and Renaissance French texts composed between the 9th and the 16th century. According to the results of the analysis, the least probable source for modern parentheticals are subordinate clauses, followed by superordinate clauses as a likelier origin, while juxtaposed sentences seem to offer the most plausible explanation. Keywords: Epistemic parenthetical verb, superordinate clause, subordinate clause, juxtaposed sentence, French

|| Stefan Schneider: Institut für Romanistik, Karl-Franzens-Universität. Graz, Austria. Julie Glikman: Faculté des Lettres, LiLPa, Université de Strasbourg. Strasbourg, France.

1 Research question and outline This contribution* examines epistemic parenthetical verbs1 in Old, Middle and Renaissance French texts, that is, in texts that were composed between the 9th and the 16th century. Regarding the emergence and development of parenthetical verbs, three possible sources or paths have been proposed in the literature: superordinate clauses, subordinate clauses, and juxtaposed sentences (cf. Schneider 2011). The hypotheses have been advanced to explain the origin and || * We are grateful to two anonymous reviewers, to Mathieu Avanzi and to the editors of the series Linguistische Arbeiten, who provided us with useful comments and suggestions on earlier versions of the paper. 1 With this choice, we do not want to suggest that only epistemic verbs have a parenthetical use in French neither that all parenthetical verbs have an epistemic value. We merely choose to concentrate on verbs that are well known to have a parenthetical use in modern French (cf. Blanche-Benveniste 1989; Blanche-Benveniste and Willems 2007; Schneider 2007).

164 | Stefan Schneider and Julie Glikman development of English parentheticals (cf. Thompson and Mulac 1991a, 1991b; Brinton 1996, 2008; Fischer 2007). The present paper intends to review and discuss their validity with respect to parenthetical verbs in Old, Middle and Renaissance French. In Old, Middle and Renaissance French, as in Modern French and other languages, parenthetical verbs are items that occur with some frequency even in written language. However, their precise definition and delimitation poses some difficulty. Even the denomination parenthetical verb is far from uncontroversial in contemporary linguistics (cf. Schneider 2007: 3–7). The issue cannot be solved satisfactorily within the scope of the present study. For the time being, we basically follow the characterisation of reduced parenthetical clauses adopted in Schneider (2007). This paper thus deals with short detached or unintegrated clauses containing a finite verb that are capable of interrupting fundamental constituents of the host. The focus is on those clauses that have no connective and that lack one of the arguments required by their verbs (the valency requirements of the parenthetical verb being satisfied by the host). However, in order to be able to fully trace the development of these items, we occasionally take into account elements with a connective or with all arguments present. Although present in Vedic, Sanskrit, Avestan and Old Persian texts, the major development of parenthesis took place in the classical Greek and Latin literary and non-literary language, where it represented a common stylistic technique (cf. Hofmann 1926: 114–119; Schwyzer 1939: 14–19). Parenthesis was a figure of thought frequently employed in Greek and Latin and has been thoroughly described in the works of classical rhetoric (cf. Lausberg 1960: 427f.). Schwyzer (1939: 40) mentions the use of parenthetical verbs and cites as examples Classical Greek dokô ‘I believe’ and Latin puto ‘I reckon’. Furthermore, he draws attention to the weakening and reduction processes that short sentences of this type might undergo when used parenthetically (e.g., Classical Greek dokô moi → dokô). Of course, puto was not the only Latin verb to appear as parenthetical. Hofmann (1926: 106f.) also mentions credo ‘I believe’, opinor ‘I opine’, and censeo ‘I deem’. Some occurrences of the verb credo are, for example, found in Cicero’s letters to members of his family: (1) (Latin, Cicero, Epistulae ad Familiares, book 3, letter 7, Perseus Digital Library, Purser 1901–1902) [...] non poteram, credo,

ante

hiemem [...]

not I.could I.believe before winter ‘I couldn’t, I believe, before winter’

Origin and development of French parenthetical verbs | 165 Ontogenetic paths to the parenthetical construction

Hence, we can safely say that this communicative strategy and the associated constructional template existed in informal written Latin. This leaves us with two options. The first one, which seems far more plausible, is that this template had always been available in informal written language and in spoken language, from Latin to Gallo-Romance, Proto-French, Old French and so forth, and had found its way into written texts later on. If this is the case, the template is not an innovation coming from spoken and written Old French. The second option, which seems less plausible, would be that the Latin constructional template had been completely lost during the long period of transition and had to be reinvented in Old French. Even if the constructional template were available, it had to be filled with the new verbs and syntactic constructions that evolved during the various historical periods. In addition, due to the profound structural changes of French throughout the period under examination, the syntactic features of the construction underwent a continuous process of adaptation. For these reasons, the period comprises several crucial points in the development of French parenthetical verbs. In a nutshell, we concentrate on the emergence of certain parenthetical verbs, their structural features and their evolution throughout the period and do not specifically look at the origin and rise of a communicative strategy and a constructional template. The paper is structured as follows: In the next section, we describe the three possible sources of parenthetical verbs. In section 3, we present the analysed texts and verbs, i.e. croire, cuidier (both meaning ‘believe’), espérer ‘hope’, penser ‘think’ and sembler ‘seem’. The following section contains an overview of the parenthetical constructions2 in which these verbs can be found. In section 5, the three aforementioned sources of parenthetical verbs are then evaluated in the light of French data. The results and their implications are discussed in the final section.

|| 2 We use construction as a general term for a set of elements and the relations between these elements and not in the specific sense it has in Construction Grammar.

166 | Stefan Schneider and Julie Glikman

2 Three possible sources of parenthetical verbs: superordinate clauses, subordinate clauses and juxtaposed sentences Thompson and Mulac (1991a, 1991b) propose superordinate clauses as source of parenthetical verbs in English. The parenthetical use of I think and other epistemic verbs originates from contemporary complex sentences through an intermediate stage of asyndetic complementation. In spoken American English, I think occurs as superordinate clause with or without that and as medial or final parenthetical (cf. Thompson and Mulac 1991b: 313): (2) I think that we’re definitely moving towards being more technological. (3) I think 0 exercise is really beneficial, to anybody. (4) It’s just your point of view you know what you like to do in your own spare time I think.

The frequent omission of the complementizer that leads to a reanalysis of the superordinate clause that causes syntactic ambiguity. In fact, (3) can be analysed either as [VERB [XYZ]CLAUSE]SENTENCE or as [VERB XYZ]SENTENCE. Eventually, the second interpretation predominates and VERB may appear in any position within the sentence. The whole process can be summarised as follows: (5) [VERB [COMP XYZ]CLAUSE]SENTENCE  [VERB [XYZ]CLAUSE]SENTENCE  [VERB XYZ]SENTENCE  [XYZ VERB]SENTENCE

Hence, unlike the generative sentence lifting (slifting) analysis proposed by Ross (1973), the process hypothesised by Thompson and Mulac (1991a, 1991b) implies the lowering of a clause. A key role is taken by the intermediate stage [VERB [XYZ]CLAUSE]SENTENCE. Following Evans and Wilkins (2000: 550), we can call it the bridging context. Since [VERB [XYZ]CLAUSE]SENTENCE opens the possibility of reanalysis, the construction is syntactically ambiguous (cf. Waltereit 1999: 21). The diachronic scenario described above is supported by data from other languages. These data are of two types: the first type supports both the scenario above and the scenario which assumes paratactic sentences as source for paren-

Origin and development of French parenthetical verbs | 167 Ontogenetic paths to the parenthetical construction

theticals, the second type supports unambiguously the scenario described above. Thompson and Mulac (1991b: 318) mention the Afrikaans epistemic adverb glo ‘allegedly, presumably’, which has evolved from the Afrikaans verb glo ‘believe’. Boye and Harder (2007: 591) observe that, contrary to parenthetical I think, glo in its adverbial use has even lost the pronominal subject ek ‘I’. In addition, Thompson and Mulac (1991b: 327) refer to an epistemic adverb [atiŋ] ‘maybe’ in Tok Pisin, Bislama and Salomons pidgin that has been derived from English I think (cf.also Keesing 1988: 101). It is clear, however, that the source of these epistemic adverbs could also have been an autonomous paratactic or juxtaposed sentence rather than a superordinate clause. In the second type of data, there can be no doubt that the epistemic adverb derives from a superordinate clause because it contains a trace of the former complementizer. The evidential adverb dizque ‘supposedly’, which can be found in American Spanish, derived from dice que ‘he/she/one says that’ (cf. Company Company 2006: 107– 110). Note that the adverb may even occur sentence-finally, although the former complementizer is clearly recognisable. Quite similarly, in several Sardinian varieties, we find the evidential adverb nachi, nanchi, naca or nanca. It stems from the univerbation (cf. Brinton and Traugott 2005: 48–50) of the third person singular or third person plural of the verb narrere ‘tell, say’ (from Latin narrare ‘tell, say’) followed by the complementizers chi (from Latin quid) or ca (from Latin quia) (cf. Pitau 1972: 90, 181; Blasco Ferrer 1984: 258). Chruschina (2010) discusses several Sicilian modal adverbs (e.g., dicica ‘it is said’, parica ‘it seems, apparently’, penzica ‘probably’) that are formed with a verb and the complementizer ca. Other examples of this type are Romanian parcă ‘it seems, apparently’ (from se pare că ‘it seems that’) and cică (from se zice că ‘it is said that’) (cf. Iliescu and Popovici 2013: 326; Zafiu 2013: 582) and Latin forsitan or forsan ‘maybe’ (from fors sit an ‘may it happen that’) (cf. Bodelot 2009: 92–95). A second source of parenthetical verbs could be subordinate clauses. Brinton (1996, 2008) states that Thompson and Mulac’s (1991a, 1991b) hypothesis is not confirmed by Old English and Middle English data and proposes a different developmental path. According to Brinton (1996: 239–253, 2008: 44f.; cf. also Fischer 2007: 300–305), the starting point is represented by Old English clauses with þæs. This particle, corresponding roughly to contemporary so or thus, introduces adverbial clauses. Brinton (2008: 45) refers, e.g., to the following example: (6) (Old English, Beowulf, v. 270-272, Krapp and Dobbie 1931-1952) / [...] ne sceal þær dyrne sum nor shall there secret anything

/ wesan, þæs be

ic wene [...] /

so.of.this I think

‘nor shall anything there be secret, so I think’

168 | Stefan Schneider and Julie Glikman Adverbial clauses following this pattern can also be found in Middle English, although with a different introductory particle (as and occasionally so). Since the particles þæs, as and so do not have the same origin, the continuity from Old to Middle English concerns a constructional template, not a particular expression. Brinton (2008: 45) observes that the verbs occurring in these adverbial clauses are also found in constructions without introductory particle, occasionally in the very same Middle English text: (7) (Middle English, Chaucer, The Canterbury Tales, B. NP 3453-3454, Benson 1987) Thee

were need of hennes, as I wene

You.Dat was need of hens

as I think

‘you have need of hens, as I think’ (8) (Middle English, Chaucer, The Canterbury Tales, E. Cl. 1173-1174, Benson 1987) I wol with lusty herte ,fresh and grene, / Seyn you a song to glade I will with lusty heart fresh and green

you, I wene

say you a song to gladden you I think

‘I will with lusty heart, fresh and green / say to you a song to gladden you, I think’

This fact leads Brinton (1996, 2008) to propose adverbial clauses introduced by þæs (Old English) and as or so (Middle English) as the origin of parenthetical verbs. We interpret and summarise the diachronic development hypothesised by Brinton (1996, 2008) in the following way: (9) [XYZ [CONN VERB]CLAUSE]SENTENCE  [XYZ [VERB]CLAUSE]SENTENCE  [XYZ VERB]SENTENCE

Brinton (1996, 2008) does not mention the issue of bridging contexts, but according to our interpretation such a context could only be represented by (8), where I think can be analysed either as clause or as single adverb. Brinton (1996, 2008) does not refer data from other languages supporting her hypothesis and, in fact, historical data and research on epistemic adverbials following the model of as I think are scarce. The only additional information we can provide is that such adverbials are also attested in Old Italian:

Origin and development of French parenthetical verbs | 169 Ontogenetic paths to the parenthetical construction

(10) (Old Italian, Dante, Vita nuova, XII, 11, Bárberi Squarotti, Cecchin, Jacomuzzi and Stassi 1983) / però que quella che ti

dee

audire,

but that the.one who you.ACC should hear / sì com’io credo, so as. I

è ver

di me adirata: /

believe is really of me angry

‘but that the one who should hear you is, as I believe, angry with me’

Referring to a similar example, Bazzanella (2003: 253) notes that sì come (io) credo ‘so as I believe’ is used in Old Italian instead of modern credo ‘I believe’. The third possible source of parenthetical verbs is represented by sentences juxtaposed to other sentences. According to Fischer (2007: 310f.), the parenthetical use of certain verbs has a long tradition in English, probably dating back to Old English. I think and similar items were used as formulaic tokens, therefore undergoing lexicalization rather than grammaticalization. Originally, they were not part of a complex sentence, neither as superordinate constituents, as proposed by Thompson and Mulac (1991b), nor as subordinate constituents, as proposed by Brinton (1996). According to Fischer (2007: 311), epistemic expressions like I think were used either as independent sentences (in oral language) or as superordinate clauses (in written language): They probably occurred both in independent clauses and with complement clauses from the very beginning, the former being most frequent in spoken, the latter in written discourse.

The two uses existed side by side, without any direct historical relationship, that is, the independent use did not derive from the superordinate use. Fischer (2007) proposes a possible developmental path, but does not go into details. We can hypothesise the following scenario: In the initial stage, the sentence with the epistemic verb precedes another sentence. The epistemic sentence may contain a cataphoric pronoun referring to the sentence to follow, similar to this I think or so I think. At the next stage, the epistemic sentence is postposed, due to a reorganisation of the information structure of the whole sequence. The sentence with the epistemic verb is short and invariable, which contributes to its transformation into a formulaic expression, so that at the end of the process it loses its sentential status. Once the pronoun no longer has its referential function, the position of the formulaic expression is relatively free; even the sentence-medial position is possible. Eventually, the pronoun is omitted. The development can be summed up in the following way:

170 | Stefan Schneider and Julie Glikman

(11) [PRO VERB]SENTENCE [XYZ]SENTENCE ↓ [XYZ]SENTENCE [PRO VERB]SENTENCE ↓ [XYZ PRO VERB]SENTENCE ↓ [XYZ VERB]SENTENCE

If this scenario is correct, the bridging context is represented by [XYZ]SENTENCE [PRO VERB]SENTENCE, which can also be interpreted as [XYZ PRO VERB]SENTENCE. It is not clear, though, whether parenthetical verbs similar to modern I think existed in Old English. Brinton (1996: 239–242) is convinced that the few examples of sentence-initial epistemic verbs without complementizer do not provide sufficient evidence for the conclusion that there were medial and final epistemic parentheticals. Fischer (2007: 311), on the other hand, assumes that it is likely that medial and final epistemic parentheticals already existed in Old English. Both authors base their positions on data from Gorrell (1895), which can be interpreted in either sense. The data from other languages cited by Thompson and Mulac (1991b) to back their hypothesis could also serve as evidence for the paratactic origin of epistemic parentheticals. Other examples that are compatible with both scenarios are English maybe and French peut-être ‘maybe’ (cf. Van Bogaert 2011: 303). In general, it appears plausible to assume that adverbs, particles and connectives stem from former juxtaposed main clauses.

3 Analysed texts and verbs The texts in Old, Middle and Renaissance French were analysed with the following online databases: Base de français medieval 2012 (), Dictionnaire du moyen français (), and Frantext (). For a more fine-grained analysis, we subdivided the three historical stages into five centuries (before 1200, 1200–1300, 1300–1400, 1400– 1500, 1500–1600). For each century, similar amounts (between 225,000 and 271,000 words) of text in verse and prose were examined, so that each century is documented by two comparable subcorpora. The total corpus amounts to approximately 2,500,000 words. The analysed texts cover different subjects, e.g., literature, history, religion, and a wide variety of verse or prose, e.g., epic poetry, romances, chronicles,

Origin and development of French parenthetical verbs | 171 Ontogenetic paths to the parenthetical construction

dramas, miracles, sermons and others. Some text types change or even disappear in the course of the analysed period, e.g., miracles (short religious dramas) no longer exist in Middle French. The use of verse and prose also changes. In the early period, prose is not so frequent and restricted to some text types, mainly religious (psalters, sermons, Bible translations) or didactic texts (e.g., Li Dialoge Gregoire Lo Pape). In Middle French and Renaissance texts, on the other hand, the use of verse becomes less frequent and mostly restricted to poems and dramas. The texts were searched for forms3 of the verbs croire, cuidier (both meaning ‘believe’)4, espérer ‘hope’, penser ‘think’ and sembler ‘seem’5. These are basically epistemic verbs, either directly referring to the mental state of belief (croire, cuidier, penser) or including an evaluative component (espérer) or indirectly referring to belief by expressing a perception (sembler). It is well known that in Modern French these verbs can, in the first person singular (croire, cuidier, espérer, penser) or in the third person singular preceded by a personal pronoun (sembler) of the indicative present, be used as parenthetical verbs, with the pragmatic function of mitigating the proposition expressed in the host sentence (cf. Blanche-Benveniste 1989; Blanche-Benveniste and Willems 2007; Schneider 2007). As explained in the first section, we took into account parenthetical occurrences with and without connectives and with and without object pronoun. The search yielded 301 parenthetical occurrences: sembler ‘seem’, with 40% of the occurrences, is the most frequent one; croire ‘believe’ accounts for 28% of the occurrences, cuidier ‘believe’ for 19%, espérer ‘hope’ for 10% and penser ‘think’ for merely 3%. As most research in historical linguistics and especially in historical pragmatics, our study has to cope with a fundamental difficulty: the data used is written literary language, but basic assumptions and considerations regarding parenthetical verbs derive from research on contemporary spoken language. Although written language has its origin in spoken language, heuristically it makes sense to consider written texts and spoken texts as separate, independ-

|| 3 We searched the forms crei, croi, croy, crois, croys, cuid, quid, quide, cuide, quides, cuides, cuis, quis, cuit, quit, senble, semble, samble, sanble, espeir, espeire, espoir, espoire, suppose, pense, panse and excluded afterwards the non-corresponding results. The BFM corpus platform TXM also includes a lexical tool to verify that no graphical variant has been forgotten. 4 The verb cuidier (from Latin cogitare 'think') is frequent in Old French, where it takes a great part of today's occurrences of croire. After 1300, it starts to disappear (cf. Shirt 1975: 353; Féron 2005: 15). It therefore has no direct equivalent in contemporary French. 5 We also looked for the verb supposer 'suppose' in the corpus, but the search produced only three parenthetical occurrences in Middle French.

172 | Stefan Schneider and Julie Glikman ent entities with their specific regularities (cf. Söll 1974: 12). On the one hand, written language is not a perfect norm that spoken language tries to emulate with more or less success; spoken texts are not reductions or elliptical implementations of written templates (cf. Maas 2010: 117). On the other hand, it would be problematic to see written texts as mere transcriptions of oral language. Due to historically inherited conventions and traditions and because of particular conditions of use, written language is partly governed by its own set of rules. Hence, written language is conditioned both by its specific regularities and by contemporary spoken language. In the debate on the emergence of parenthetical verbs, the distinction between oral and written language often gets blurred, occasionally calling into question the validity of assumptions and conclusions. This is particularly evident in the explanation proposed by Thompson and Mulac (1991a, 1991b). Although their hypothesis is meant to explain the synchronic origin of the parenthetical use of I think and other epistemic expressions in spoken American English, they adopt as starting point the structure I think that p ([VERB [COMP XYZ]CLAUSE]SENTENCE). However, as shown by their own researches and by Kärkkäinen (2003), the complementizer that after I think is almost absent from spoken language. The distinction between the actual mode of transmission (phonic vs. graphic) on the one hand and the intended use (spoken vs. written) on the other hand (cf. Söll 1974: 11–23; Koch and Oesterreicher 2011: 3) reveals the main difference between the two types of literary texts in our corpus. Both verse and prose texts are transmitted to us in the graphic mode; their intended use, however, is not the same. During most of the historical period considered, the texts in verse were partially, if not exclusively, composed for oral presentation (often by persons distinct from the author) before a small audience, occasionally accompanied by music, whereas the texts in prose were intended for individual reading (cf. Ayres-Bennett 1996: 59; Rohr 1998: 22–27). Therefore, we can assume that the texts in verse, in spite of the limitations imposed by meter and rhyme, contain traces that reveal their intended use as spoken language. The analysed data for the Old French period suggest that parenthetical verbs occur more often in verse than in prose (126 occurrences in verse compared to 17 in prose; cf. also Féron 2005: 19) and more often in romances (verse and prose) and miracles (verse and prose), with 74 occurrences in romances and 42 occurrences in miracles compared to 27 occurrences in all the remaining texts types. In the Middle French texts, we find the same asymmetry between verse and prose: 96 parenthetical verbs occur in verse and only 19 in prose. This tendency, however, does not continue in the Renaissance texts, in which we registered 15 parenthetical verbs in verse and 28 in prose.

Origin and development of French parenthetical verbs | 173 Ontogenetic paths to the parenthetical construction

4 Parenthetical constructions Concerning the verbs croire, cuidier, espérer and penser, we identified five constructions of parenthetical usage. The construction called ce+V contains the demonstrative pronoun ce ‘it, this’ in the function of a direct object and occurs without a subject pronoun and without a connective. This construction is no longer found after 1500, due to the obligatory presence of an overt subject. The second construction, ce+V+je, includes both the demonstrative pronoun ce and the postposed subject pronoun je ‘I’. In our corpora, this construction appears only after 1400. The next construction, je+V, has the same form as the parenthetical constructions in Modern French: no connective, no object pronoun but a subject pronoun. This construction appears in all periods in our corpus. In Old French and Middle French the subject pronoun is not always expressed, therefore there is also V as fourth construction type. The fifth construction, comme+je+V, comprises a connective. If this construction type includes an object pronoun, it is the anaphoric pronoun le ‘it, him’ and never the demonstrative one. Occasionally, the subject pronoun je ‘I’ may not be expressed. The five constructions are representative of the evolution of the French verb phrase: from the non-obligatory expression of the subject pronoun in Old French (but with the verb being unacceptable in sentence-initial position) to its obligatory expression during Middle French and Renaissance French and to the position before the verb, with the object at the same time moving to the position immediately after the verb (cf. Marchello-Nizia 1995). The evolution is accompanied by a decrease in the use of the demonstrative ce as object pronoun (cf. Brunot and Bruneau 1956: 247). Tables 1–4 show the constructions and their respective occurrences for the four verbs in each period:

174 | Stefan Schneider and Julie Glikman Infinitive Type before 1200 1200–1300 1300–1400 1400–1500 1500–1600 Total

croire ce+V 9 6 3 4 22

ce+V+je

je+V

V 1

6 2 8

3 6 9 2 20

2 3

comme+je+V 9 3 5 8 6 31

Total 18 13 14 29 10 84

Table 1: Parenthetical constructions with croire ‘believe’

The verb croire ‘believe’ occurs in all construction types. As mentioned earlier, the ce+V construction is no longer found after 1500, whereas ce+V+je appears very late. The comme+je+V construction is well represented across all periods. In the first period, it occurs as often as the ce+V construction. Infinitive Type before 1200 1200–1300 1300–1400 1400–1500 1500–1600 Total

cuidier ce+V 12 10

ce+V+je

je+V 1 22

V 1

comme+V+je 1 7 1

Total 15 39 1 1

1

9

56

1 22

0

24

Table 2: Parenthetical constructions with cuidier ‘believe’

The verb cuidier disappears rapidly after the Old French period, which explains the low number of occurrences after 1300. However, it is used a lot in Old French and presents more parenthetical constructions for this stage than croire. Interestingly, the comme+je+V construction is less important than for the other verbs and the je+V construction is the most frequent one during the 1200–1300 period.

Origin and development of French parenthetical verbs | 175 Ontogenetic paths to the parenthetical construction

Infinitive Type before 1200 1200–1300 1300–1400 1400–1500 1500–1600 Total

espérer ce+V 1

1

ce+V+je

0

V 3 4 9

je+V

0

16

5

Total 7 4 14

6 14

6 31

comme+je+V 3

Table 3: Parenthetical constructions with espérer ‘hope’

The verb espérer, although common in Modern French, does not appear very often in our data. The occurrences mainly come from the types V6 and comme+je+V. Infinitive Type before 1200 1200–1300 1300–1400 1400–1500 1500–1600 Total

penser ce+V

ce+V+je

je+V

V

comme+je+V

Total

0

4 4

4 5 9

4 0

4

1 1

Table 4: Parenthetical constructions with penser ‘think’

The parenthetical occurrences with the verb penser appear only after 1400, although this verb does exist in the earlier period in other constructions. Due to their specific syntax, the constructions with the verb sembler ‘seem’ have been classified separately. We identified three constructions:

|| 6 For Marchello-Nizia (personal communication), the form espoir should be interpreted as an adverb in the 12th century.

176 | Stefan Schneider and Julie Glikman Infinitive Type before 1200 1200–1300 1300–1400 1400–1500 1500–1600 Total

ce+me+V 5 34 24 15 15 93

me+V 1 1 2 4

sembler comme+il+me+V 1 6 7 6 4 24

Total 7 40 32 21 21 121

Table 5: Parenthetical constructions with sembler ‘seem’

In the first construction, ce+me+V (which includes two cases of ce semble occurring after 1400), there is the demonstrative pronoun ce ‘it, this’, in the function of the subject of the impersonal verb, and the indirect object pronoun me ‘to me’ (where it occurs)7. In these cases, like in the constructions ce+V for the other verbs, there is no connective, but there is the demonstrative pronoun before the parenthetical, which may be interpreted as an anaphoric (or cataphoric) reference to the host sentence8. Within the second construction, me+V, without demonstrative pronoun, we also included a single occurrence with the impersonal subject pronoun il ‘it’, although it is doubtful whether the other occurrences actually are results of the omission of il or of ce. The third construction, comme+il+me+V, comprises the structures introduced by a connective, with or without an overt subject pronoun (before 1300, there is only one occurrence with il). When the subject pronoun is overt, it is always the impersonal pronoun il, never ce. Furthermore, the first person pronoun always occurs as tonic moi ‘me’ in Old French and partly in Middle French, too, except in occurrences with an overt subject pronoun. It is relevant to mention that the subject pronoun found in our data is usually ce, and not il. The pronoun il can appear with impersonal verbs in other constructions. In Old French, the same verbs may appear with il or ce as subject, but ce is usually considered preserving its demonstrative value, whereas il is considered to be a ‘real’ impersonal pronoun (cf. Buridant 2000: 140 and Mazziotta et al. 2012: 52). Summing up, there are two constructions, ce+V for croire, cuidier, espérer and penser and ce+me+V for sembler, which have no connective, but whose || 7 We took into account only the first person singular pronoun, but the verb also appears with other personal pronouns. 8 We distinguish between connectives such as complementizers or conjunctions that join clauses within a complex sentence and anaphoric or cataphoric referring devices such as personal pronouns.

Origin and development of French parenthetical verbs | 177 Ontogenetic paths to the parenthetical construction

demonstrative pronoun (object or subject) indicates an anaphoric (or cataphoric) reference (see section 5.3). Then, there is a construction ce+V+je that might represent a stage in the development towards the obligatory expression of the subject (also see section 5.3). The constructions V, je+V and me+V are similar insofar as they appear without a connective and without a demonstrative pronoun (see section 5.1). Finally, we identified the constructions comme+je+V and comme+il+me+V, which have a connective (see section 5.2).

5 The three possible sources in the light of the data 5.1 Superordinate clauses In Old French, asyndetic subordination is quite common. It can be found with different verbs, persons and tenses (cf. Glikman 2009: 182–193, 196–225). Therefore, it is possible to find epistemic verbs that may occur as superordinate clauses with or without complementizer and as medial or final parentheticals (cf. Franzén 1939; Glikman 2009: 158–163, 182–193, 2012: 74f.): (12) (Old French, Tristan en prose, p. 69, Ménard 1987) Si m’aït

Diex, fait li

vallés,

je cuit

que vous le

desfendrés

If me.helps God says the young.man I think that you him will.defend encore hui still

mout plus mauvaisement que vous ne quidiés!

today much more worse

than you not think

‘With the help of God, says the young man, I think that you will defend him still today much worse than you think!’ (13) (Old French, Gautier de Coinci, Miracles de Nostre Dame 4, v. 154, Koenig 1970) Se Diex m’aït,

je cuit

mar

ne la vielt

herbregier li

doyens.

If God me.helps I think unfortunately not her wants to.house the senior With the help of God, I think the senior unfortunately does not want to house her’ (14) (Old French, Jean Renart, Escoufle, v. 6146, Sweetser 1974) Li cuens n’avoit, je cuit, que sis Chevaliers avoec lui the count not.had I think than his knights

with

‘The count only had, I think, his knights with him there’

laiens.

him there

178 | Stefan Schneider and Julie Glikman The examples above stem from the same period and illustrate the coexistence of different developmental layers (see the layering principle in Hopper 1991: 22-24). As in Thompson and Mulac (1991b), example (13) represents the intermediate stage that provides the bridging context. The absence of the complementizer in the example enables the reanalysis and syntactic ambiguity of the superordinate je cuit. Thompson and Mulac (1991a, 1991b) repeatedly refer to quantity as evidence for their hypothesis. Thompson and Mulac (1991a: 244) compare the use of think and guess, the two most frequent epistemic verbs in their data, with and without that: 91% of the occurrences of think and even 99% of the occurrences of guess are without the complementizer that. These proportions are never reached in our French data. In the Old French data, 49% out of a total of 351 occurrences of the forms ce cuit, je cuit, ce croy, je croy (including the constructions with a nominal complement) are followed by the complementizer que (35% before 1200, 55% from 1200 to 1300). Most importantly, the cases providing the bridging context as in (13) are not very frequent: in the Old French data, we registered only 30 cases in which ce/je cuit/croi/espeir or il/ce me semble occur in sentence-initial position without complementizer, compared to 187 cases with complementizer (ce/je cuit/croi/espeir que or il/ce me semble que). In the Middle French data, even 67% out of a total of 227 occurrences of je croy ‘I believe’ are followed by the complementizer que. We are of course aware of the fact that the data in Thompson and Mulac (1991a, 1991b) stem from spoken contemporary English, whereas our data stem from, mostly literary, written old stages of French, but the differences in the proportions are certainly remarkable. In Middle French and Renaissance French, however, complementizer omission is highly exceptional or even excluded, which means that there is no bridging context anymore. Moreover, our data clearly show a phenomenon that has been mentioned repeatedly (cf. Shirt 1975: 353; Féron 2005: 15): after 1300, the verb cuidier disappears (with the form in the first person singular starting the process; cf. Moignet 1959: 556). In Modern French, asyndetic subordination is not impossible in spoken language. No corpus studies exist on this topic, but the omission of the complementizer seems to vary according to the variety; e.g., it seems more common in Canadian French than in French spoken in France (cf. Martineau 1988; Schneider 2007: 174; Avanzi 2012: 272–290).

5.2 Subordinate clauses As can be seen in tables 1–5, the French constructions comme+je+V and comme+il+me+V are attested throughout the whole period and for all verbs

Origin and development of French parenthetical verbs | 179 Ontogenetic paths to the parenthetical construction

under examination. Zink (1997: 179) confirms their existence in Middle French texts of the 14th century. In our data, their percentages of the respective totals of parentheticals vary from 30% in the period before 1200, to 17% in the period 1200–1300, to 30% in the period 1300–1400, to 25% in the period 1400–1500, and to 48% in Renaissance French. There is therefore a marked increase of their frequency in the period 1500–1600. So from the beginning, comme+je+V and comme+il+me+V play a consistent role, although they represent the dominant constructional pattern only at the end of the period considered. Moreover, data from another study suggest that they have not yet reached the peak of their development. The phenomenon is particularly evident in the case of croire ‘believe’: in a corpus of French travel journals, diaries and letters from the 17th and 18th centuries, the parentheticals with comme je crois ‘as I believe’ occur three times more often than those with je crois ‘I believe’ (cf. Schneider 2012: 93). In the following centuries, however, the use of comme je crois diminishes rapidly; today this parenthetical is almost absent from spoken and written French (cf. Schneider 2012: 97). Throughout the period studied, the comme+je+V and comme+il+me+V constructions manifest a considerable variety and do not reach the fixation and formulaic character of the other constructions. Before 1200 and between 1200 and 1300 we mostly find parentheticals similar to si cum ie croi ‘so as I believe’ and si con moi samble ‘so as it seems to me’, with an exceptional ainssi comme il me samble, which also has the meaning ‘so as it seems to me’. In the Middle French period between 1300 and 1500, the parentheticals following the models comme je croy and comme il me semble are as frequent as those with si comme. In addition, there are some parentheticals of the type ainsy comme je croy ‘so as I believe’ and an elaborate comme certainement je croy ‘as certainly I believe’. In the Renaissance period, the construction seems to have found its definitive form. Except for two cases, the introductory ainsi or si has been eliminated. All comme+je+V and comme+il+me+V parentheticals follow the same model. Summing up, one gets the impression that throughout the period under consideration the construction passes through a phase of instability with several variants being used interchangeably before the repertoire is narrowed down to one variant. Analysing the data separately for each verb and taking into account their various forms, one notices that the periods in which the same form appears in both constructions (je+V and comme+je+V, me+V and comme+il+me+V) vary considerably. Only croire ‘believe’ occurs in both constructions consistently for a long period, that is, from 1200 to 1600. In the case of cuidier ‘believe’, the time of co-occurrence lasts mainly from 1200 to 1300. According to tables 1–5, espérer ‘hope’ seems to appear in both constructions for a certain period. The analysis

180 | Stefan Schneider and Julie Glikman of its forms, however, shows that in the comme+je+V construction the verb takes a subject pronoun (e.g., si cum je espeir), whereas it occurs without subject pronoun in the V construction (e.g., espoir). That means that the simple omission of si cum would not be sufficient to derive the parenthetical espoir. For penser ‘think’ there is practically no co-occurrence of je+V and comme+je+V, if one excludes a single attestation of je pense in the Renaissance period. Finally, in the case of sembler ‘seem’, the me+V construction does have limited importance: we found three attestations for me semble and only one for il me semble distributed over the entire period studied, compared to 24 attestations of comme+il+me+V. Moreover, here too, the comme+il+me+V construction has a subject pronoun, whereas the me+V construction occurs mostly without the subject pronoun il. In brief, only for croire and cuidier there are periods in which the same verb form consistently occurs in both constructions. As is the case with Middle English, there have not been any studies and consequently there is no data concerning the possibility to omit comme ‘as’ in epistemic adverbial clauses. Summing up, there are some reasons why it seems unlikely that the je+V construction developed from comme+je+V and the me+V construction from comme+il+me+V by the omission of comme ‘as’.

5.3 Juxtaposed sentences Tables 1–5 show that from before 1200 until 1600 there are parentheticals with the pronoun ce ‘it, this’. Buridant (2000: 756) and Glikman (2009: 118, 122, 158, 166) mention the use of the pronoun ço (from Latin hoc ‘this, that’) and later ce in Old French parentheticals. According to Zink (1997: 87–92), the pronoun ce is also common in Middle French parentheticals. Féron (2005: 19) examined the texts in verse from the period before 1200 in the Base textuelle de l’ancien français () and found 64 parentheticals with ce cuit and ce croi compared to 7 parentheticals with je cuit and je croi. In the texts from the first half of the 13th century the share of parentheticals without ce increases: these texts contain 17 cases with ce cuit and ce croi and 24 with je cuit (but no case with je croi). In our data, the proportion of the three constructions ce+V, ce+V+je and ce+me+V of the total of parentheticals is always very high, but it decreases slightly during the period under scrutiny: 57% in the period before 1200, 52% in the period 1200–1300, 44% in the period 1300–1400, 53% in the period 1400– 1500, 40% in Renaissance French. There are, however, notable differences between the five verbs under examination: for penser ‘think’ and espérer ‘hope’ the construction is either non-existent or marginal, with croire ‘believe’ and

Origin and development of French parenthetical verbs | 181 Ontogenetic paths to the parenthetical construction

cuidier ‘believe’ it plays a significant role, for sembler ‘seem’ it is by far the most frequent construction. After 1500, the importance of the construction diminishes markedly: except for two cases of ce croy je by the same author in the same text, there are only examples with sembler. In fact, the parenthetical ce me semble continues to be used until the end of the 19th century (cf. Schneider 2012: 97). So, the constructions with ce play a prominent role throughout the whole period considered, but, contrary to the constructions comme+je+V and comme+il+me+V, they seem to have already reached the peak of their development and are slowly losing ground. Throughout the period studied, the form of the constructions with ce remains essentially stable. Before 1200 and between 1200 and 1300, they do not exhibit any variation at all. In the Middle French period between 1300 and 1500, the variation is limited to a case of ce croy bien ‘this I believe well’, a case of ce que je croy ‘this that I believe’ and two cases of ce semble ‘this seems’. In the Renaissance period, there is again no variation, apart from a single case of ce semble. Comparing ce+V with ce+V+je, only croire ‘believe’ is attested in both constructions, the other verbs are restricted to one construction. In the case of croire, the construction ce+V+je appears only after 1400, which is why the construction seems to be later than ce+V. Regarding the stage [PRO VERB]SENTENCE [XYZ]SENTENCE of the hypothesised scenario, there is at least one attestation in the texts before 1200, which suggests that such a structure could have existed. In the example (with our translation), ce crei ‘I believe’, preceded by two forms of address, only serves as epistemic introduction to the following sentence: (15) (Old French, La Chanson de Roland, v. 1006, l. 79, Segre 2003) / Dist Oliver: “ Sire cumpainz, ce says Oliver

crei, /

De Sarrazins purum

bataille aveir.”/

Sir companion this I.believe of Saracens we.could battle

have

‘Says Oliver: “Sir companion, this Ibelieve, from the Saracens we could have battle.”‘

The pronoun ce clearly cataphorically refers to De Sarrazins purum bataille aveir. The stage [XYZ]SENTENCE [PRO VERB]SENTENCE, in which the verb with the pronoun paratactically follows a sentence, is still recognisable in the following Middle French example: (16) (Middle French, Guillaume de Machaut, Le dit dou lyon, p. 164, Hœpffner 1911) / Et

si joliement couvert / D’un fin drap de soie tout vert

and so beautifully covered by.a thin cloth of silk all green

182 | Stefan Schneider and Julie Glikman

/ Qu’on

ne porroit mieus souhaidier, / Ce croy,

that.one not could better wish

se Dieus me puist aidier. /

this I.believe if

God me could help

‘and so beautifully covered by a thin green cloth of silk that one could not better wish, this I believe, with the help of God’

Whereas the pronoun ce anaphorically refers to the preceding sentence (or sentences), the autonomy of ce croy is underlined by the fact that it has the formulaic expression se Dieus me puist aidier depending on it. The constructions with the pronoun ce share a certain period of coexistence with the constructions je+V and me+V. But again, there are considerable differences between the single verbs. In the case of croire ‘believe’, this period lasts from 1200 to 1600, in the case of cuidier ‘believe’, the coexistence lasts until 1300. For espérer ‘hope’ and penser ‘think’ there is practically no coexistence, if we exclude one example of ce espeir bien in the texts before 1200 and a single attestation of je pense in the Renaissance period. As we have seen before, in the case of sembler ‘seem’, the me+V construction is not significant: there are three occurrences of me semble and only one for il me semble distributed over the entire period under examination, compared to 93 occurrences of ce+me+V. Hence, only for croire and cuidier there are periods in which the two constructions coexist with fairly numerous examples.

6 Conclusion Our search in Old, Middle and Renaissance French texts yielded 301 parenthetical tokens with five verbs. The parenthetical occurrences of the verbs croire, cuidier, espérer and penser can be assigned to five constructions: ce+V, ce+V+je, je+V, V and comme+je+V. The examples involving the fifth verb analysed, sembler, were assigned to three constructions: ce+me+V, me+V and comme+il+me+V. In summary, there are ce-parentheticals, commeparentheticals and parentheticals identical or similar to those found in Modern French, which for convenience we call modern parentheticals. Ceparentheticals by far outnumber the others, comprising 50% of all instances, comme-parentheticals account for 27% and modern parentheticals for 23%. There are, however, important differences between the verbs, the aforementioned disproportion being largely due to the fact that ce-parentheticals with sembler continue to play an important role even during the Renaissance period. As far as their chronological distribution is concerned, there is a reversal in the proportions: in the period before 1200, ce-parentheticals account for 57% of

Origin and development of French parenthetical verbs | 183 Ontogenetic paths to the parenthetical construction

the total, comme-parentheticals for 30% and modern parentheticals for 13%; in the Renaissance period, ce-parentheticals account for 40% of the total, commeparentheticals for 48% and modern parentheticals for 12%. So, whereas in the early period, ce-parentheticals are the dominant constructional frame, in the Renaissance period, comme-parentheticals are dominant. The reversal in the relative importance of ce-parentheticals and comme-parentheticals is even more evident if we exclude those with sembler: in the early period, ce-parentheticals account for 55% of the total, comme-parentheticals for 32% and modern parentheticals for 13%, in the Renaissance period, ce-parentheticals account for 10% of the total, comme-parentheticals for 76% and modern parentheticals for 14%. As is clear from the percentages above, in general the three types of parentheticals coexist throughout the whole period. They also coexist within each of the five historical periods defined. There are, however, some important specifications to be made. First of all, the V construction, without the subject pronoun je ‘I’ (e.g., croy ‘I believe’), is no longer used in Renaissance French since the expression of the subject pronoun has become obligatory9. Secondly, for the ce+V+je construction, we found data only for the period after 1400. Moreover, as mentioned repeatedly, the verb cuidier practically disappears after the Old French period. Finally, for the verb sembler, modern parentheticals play a marginal role during the whole period. In the period before 1200, comme-parentheticals coexist with ceparentheticals and co-occur in the works by the same author. In the period 1200–1300, ce-parentheticals, comme-parentheticals and modern parentheticals all occur in the texts by the same author. In the period 1300–1400, the three types of parentheticals also occur interchangeably in the works by the same author (e.g., Eustache Deschamps uses ce croi, ce semble, si com j’espoir, si com moy semble and je croy). During the period 1400–1500, the three parentheticals co-occur, albeit in more restricted ways: those authors employing all of them only use ce me semble as ce-parenthetical. Interestingly, ce-parentheticals and modern parentheticals with croire ‘believe’ co-occur during the periods 1300– 1400 and 1400–1500 in the texts by the same author (e.g., Christine de Pizan writes ce croy, ce croy je, croy and je croy). During the Renaissance period, the three types of parentheticals all occur, but never simultaneously in texts by the same author. To cite an example, Du Bellay uses comme je croy, ce croy je and ce me semble, but not modern parentheticals.

|| 9 The fact that parenthetical verbs appear with subject pronouns during the same time in which subject pronouns are becoming obligatory in general suggests that the parenthetical verbs in this period behave like other verbs.

184 | Stefan Schneider and Julie Glikman Concerning the emergence and development of modern parentheticals, we considered three developmental paths or sources: superordinate clauses (see section 5.1), subordinate clauses (see section 5.2) and juxtaposed sentences (see section 5.3). The last two sources can roughly be equated with commeparentheticals and ce-parentheticals. As a first option, we took into account superordinate clauses. As shown in section 5.1, in Old French, the verb forms examined may occur in sentence-initial position without the complementizer que. Most of the time, the superordinate clauses indeed comprise forms like je croi or je cuit that lead directly to the forms of modern parentheticals. The hypothesis has to cope, however, with two difficulties. Firstly, in Old French, the forms we examined are considerably less frequent in sentence-initial position without the complementizer que than their Modern English equivalents: 30 cases without complementizer compared to 187 cases with complementizer. Even if we bear in mind that the data cited by Thompson and Mulac (1991a, 1991b) are from spoken language, whereas ours are from written language, the difference in the proportions remains huge. Secondly, there is a chronological difficulty: modern parentheticals already occur in median position in the 13th century (55% of all parentheticals during the period 1200–1300). The transition from the superordinate clause to the parenthetical clause must then have occurred much earlier. Our data do not support the hypothesis that modern parentheticals derive from subordinate clauses by deletion of the connective comme ‘as’. From the early period onward, comme-parentheticals play an important role, although they represent the dominant constructional pattern only at the end of the time considered. Comme-parentheticals show considerable constructional variation and never become as fixed and formulaic as the other parentheticals. In fact, during the period examined, this construction is also common with a third person pronoun (e.g., comme il lui semble ‘as it seems to him’), which never developed into a parenthetical item. Moreover, there are modern-style parentheticals (espoir, me semble) which cannot be derived from the corresponding commeparentheticals by the mere omission of comme. Finally, there are no data indicating the possibility of omitting comme ‘as’ in epistemic adverbial clauses. Another possible source of modern parentheticals are juxtaposed sentences. According to our data, ce-parentheticals have their origin in juxtaposed sentences. These parentheticals are of course well represented, especially with ce croi, ce cuit and ce me semble. The data show how this type of parenthetical progressively moves from its original position at the periphery of the host sentence into a sentence-medial position: in the period before 1200, 57% of its occurrences are in initial or final position, compared to only 31% in the period 1200–1300. This suggests an integration of a former juxtaposed sentence. But if

Origin and development of French parenthetical verbs | 185 Ontogenetic paths to the parenthetical construction

ce-parentheticals originated modern parentheticals we have to explain the transition from ce croi to je croi. Both types of parentheticals co-occur from the early period. In the corpora examined there are no clues indicating an evolution through intermediate states as the following: ce croi → ce croi je → je ce croi → je crois. The construction je ce croi is not attested. Hence, the direct replacement of ce with je seems more plausible. In general, the transition from ce croi to je croi fits in well with what we know about the preferred argument structure of Old and Modern French. Old French is a language with optional subject and obligatory object (cf. Detges 2009; Rouquier and Marchello-Nizia 2012), whereas in Modern French the object is optional and the subject must be expressed. It certainly is too early for a solution of the riddle. Further studies that include other parenthetical verbs and constructions need to be undertaken. These should include direct speech parentheticals (cf. Marchello-Nizia 2012). Furthermore, a more detailed analysis of the differences between texts in verse and texts in prose might also provide new insights. We can, however, evaluate the probability of the three hypotheses in the light of our current knowledge: subordinate clauses provide the less probable source for modern parentheticals, superordinate clauses are more probable as source, while juxtaposed sentences seem to offer the most plausible explanation.

References Avanzi, Mathieu. 2012. L’interface prosodie/syntaxe en français. Dislocations, incises et asyndètes. Bruxelles: Lang. Ayres-Bennett, Wendy. 1996. A history of the French language through texts. London: Routledge. Base de français médiéval (BFM). 2012. Lyon: ENS de Lyon, Laboratoire ICAR. . Bazzanella, Carla. 2003. Discourse markers and politeness in Old Italian. In Gudrun Held (ed.). Partikeln und Höflichkeit. Frankfurt am Main: Lang. 247–268. Beowulf. In George Ph. Krapp & Elliot van Kirk Dobbie (eds.). 1931–1952. The Anglo-Saxon poetic records: A collective edition. New York: Columbia University Press. Blanche-Benveniste, Claire. 1989. Constructions verbales en incise et rection faible des verbes. Recherches sur le français parlé 9. 53–73. Blanche-Benveniste, Claire & Dominique Willems. 2007. Un nouveau regard sur les verbes faibles. Bulletin de la Société de Linguistique de Paris 102. 217–254. Blasco Ferrer, Edoardo. 1984. Storia linguistica della Sardegna. Tübingen: Niemeyer. Bodelot, Colette. 2009. Adverbialisation de la séquence proposition + interrogatif ou exclamatif. In Michèle Fruyt & Sophie Van Laer (eds.). Adverbes et évolution linguistique en latin. Paris: L’Harmattan. 81–100. Boye, Kasper & Peter Harder. 2007. Complement-taking predicates: Usage and linguistic structure. Studies in Language 31. 569–606.

186 | Stefan Schneider and Julie Glikman Brinton, Laurel J. 1996. Pragmatic markers in English: Grammaticalization and discourse functions. Berlin: Mouton De Gruyter. Brinton, Laurel J. 2008. The comment clause in English: Syntactic origins and pragmatic development. Cambridge: Cambridge University Press. Brinton, Laurel J. & Elizabeth C. Traugott. 2005. Lexicalization and language change. Cambridge: Cambridge University Press. Brunot, Ferdinand & Charles Bruneau. 1956. Précis de grammaire historique de la langue française. 4th edn. Paris: Masson et Cie. Buridant, Claude. 2000. Grammaire nouvelle de l’ancien français. Paris: SEDES. Chruschina, Silvio. 2010. Aspetti morfologici e sintattici degli avverbi in siciliano. In Jacopo Garzonio (ed.). Quaderni di lavoro ASIt n.11: Studi sui dialetti della Sicilia. Padova: Unipress. 19–39. Cicero M. Tullius. Epistulae ad Familiares. Louis C. Purser (ed.). 1901–1902. Oxford: Clarendon Press. . Company Company, Concepción. 2006. Subjectification of verbs into discourse markers: semantic-pragmatic change only? Belgian Journal of Linguistics 20. 97–121. Dante Alighieri. Vita nuova. Giorgio Bárberi Squarotti, Sergio Cecchin, Angelo Jacomuzzi & Maria Gabriella Stassi (eds.). 1983. Torino: UTET. Detges, Ulrich. 2009. How useful is case morphology? The loss of the Old French two-case system within a theory of Preferred Argument Structure. In Jóhanna Barðdal & Shobhana L. Chelliah (eds.). The role of semantic, pragmatic, and discourse factors in the development of case. Amsterdam: Benjamins. 93–120. Dictionnaire du moyen français (DMF). 2012. Nancy: ATILF, CNRS and Université de Lorraine. . Evans, Nicholas & David Wilkins. 2000. In the mind’s ear: The semantic extensions of perception verbs in Australian languages. Language 76. 546–592. Féron, Corinne. 2005. Modalisation et verbes d’opinion: quelques remarques sur croire, cuidier et penser dans La Queste del Saint Graal. L’information grammaticale 104. 15–21. Fischer, Olga. 2007. Morphosyntactic change: functional and formal perspectives. Oxford: Oxford University Press. Frantext. 2013. Nancy: ATILF, CNRS and Université de Lorraine. . Franzén, Torsten. 1939. Étude sur la syntaxe des pronoms personnels sujets en ancien français. Uppsala: Almqvist & Wiksell. Gautier de Coinci. Miracles de Nostre Dame. V. Frédéric Koenig (ed.). 1970. Genève: Droz. . Geoffrey Chaucer. The Canterbury Tales. In Larry D. Benson (ed.). 1987. The Riverside Chaucer. 3rd edition. Boston: Houghton Mifflin. Glikman, Julie. 2009. Parataxe et subordination en ancien français. Système syntaxique, variante et variation. Université Paris Ouest – Universität Potsdam: Doctoral thesis. Glikman, Julie. 2012. Les incises en croire et cuidier en ancien français. In Julie Glikman & Mathieu Avanzi (eds.). Entre rection et incidence: des constructions verbales atypiques? Études sur je crois, je pense et autres parenthétiques (= Linx 61). Nanterre: Université Paris Ouest Nanterre La Défense. 71–85. Gorrell, J. Hendren. 1895. Indirect discourse in Anglo-Saxon. Publications of the Modern Language Association of America 10. 342–485. Guillaume de Machaut. Le dit dou lyon. In Ernest Hœpffner (ed.). 1911. Oeuvres de Guillaume de Machaut. Tome deuxième. Paris: Didot.

Origin and development of French parenthetical verbs | 187 Ontogenetic paths to the parenthetical construction

Hofmann, Johann Baptist. 1926. Lateinische Umgangssprache. Heidelberg: Carl Winters Universitätsbuchhandlung. Hopper, Paul J. 1991. On some principles of grammaticization. In Elizabeth C. Traugott & Bernd Heine (eds.). Approaches to grammaticalization. Vol. 1: Focus on theoretical and methodological issues. Amsterdam - Philadelphia: Benjamins. 17–35. Iliescu, Maria & Victoria Popovici. 2013. Rumänische Grammatik. Hamburg: Buske. Jean Renart. Escoufle. Franklin P. Sweetser (ed.). 1974. Genève: Droz. . Kärkkäinen, Elise. 2003. Epistemic stance in English conversation. A description of its interactional functions, with a focus on I think. Amsterdam/Philadelphia: Benjamins. Keesing, Roger. 1988. Melanesian Pidgin and the oceanic substrate. Stanford: Stanford University Press. Koch, Peter & Wulf Oesterreicher. 2011. Gesprochene Sprache in der Romania: Französisch, Italienisch, Spanisch. 2nd edn. Berlin/New York: De Gruyter. La Chanson de Roland. Cesare Segre (ed.). 2003. Genève: Droz. Lausberg, Heinrich. 1960. Handbuch der literarischen Rhetorik. Eine Grundlegung der Literaturwissenschaft. München: Hueber. Maas, Utz. 2010. Literat und orat Grundbegriffe der Analyse geschriebener und gesprochener Sprache. Grazer Linguistische Studien 73. 21–150. Marchello-Nizia, Christiane. 1995. L’évolution du français. Paris: Armand Colin. Marchello-Nizia, Christiane. 2012. L’oral représenté en français médiéval: un accès construit à une face cachée des langues mortes. In Céline Guillot, Bernard Combettes, Alexei Lavrentiev, Evelyne Oppermann-Marsaux & Sophie Prévost (eds.). Le changement en français. Études de linguistique diachronique. Bern: Lang. 247–264. Martineau, France. 1988. Variable deletion of que in the spoken French of Ottawa-Hull. In David Birdsong & Jean-Pierre Montreuil (eds.). Advances in Romance linguistics. Dortrecht: Foris. 275–287. Mazziotta, Nicolas, Beatrice Bischof, Julie Glikman & Thomas M. Rainsford. 2012. Ce sujet dans les “constructions impersonnelles” du Roman de Tristan de Béroul. L’information grammaticale 132. 48–52. Moignet, Gérard. 1959. Essai sur le mode subjonctif en latin postclassique et en ancien français. Paris: Presses universitaires de France. Pitau, Massimo. 1972. Grammatica del sardo-nuorese: il più conservativo dei parlari neolatini. Bologna: Pàtron. Rohr, Rupprecht. 1998. Lesen und Verstehen französischer und provenzalischer mittelalterlicher Dichtung. Frankfurt am Main: Haag + Herchen. Ross, John R. 1973. Slifting. In Maurice Gross, Morris Halle & Marcel-Paul Schützenberger (eds.). The formal analysis of natural languages. Proceedings of the first international conference. Den Haag: Mouton. 133–169. Rouquier, Magali & Christiane Marchello-Nizia. 2012. De (S)OV à SVO en français: où et quand? L’ordre des constituants propositionnels dans La Passion de Clermont et La Vie de saint Alexis. In Monique Dufresne (ed.). Constructions en changement. Hommage à Paul Hirschbühler. Quèbec: Presses de l’Université Laval. 111–155. Schneider, Stefan. 2007. Reduced parenthetical clauses as mitigators. A corpus study of spoken French, Italian and Spanish. Amsterdam/Philadelphia: Benjamins. Schneider, Stefan. 2011. Die Entstehung einer neuen Kategorie: Reduzierte parenthetische Teilsätze im Alt-, Mittel- und Neufranzösischen. In Eva Mayerthaler, Claudia E. Pichler &

188 | Stefan Schneider and Julie Glikman Christian Winkler (eds.). Was grammatische Kategorien miteinander machen. Form und Funktion in romanischen Sprachen von Morphosyntax bis Pragmatik. Festschrift für Ulrich Wandruszka. Tübingen: Narr. 225–244. Schneider, Stefan. 2012. Les clauses parenthétiques dans les textes de la Nouvelle-France du 17e et du 18e siècle: aspects pragmatiques, syntaxiques et diachroniques. In Julie Glikman & Mathieu Avanzi (eds.). Entre rection et incidence: des constructions verbales atypiques? Études sur je crois, je pense et autres parenthétiques (= Linx 61). Nanterre: Université Paris Ouest Nanterre La Défense. 87–102. Schwyzer, Eduard. 1939. Die Parenthese im engern und weitern Sinne. Abhandlungen der Preußischen Akademie der Wissenschaften, Jahrgang 1939, Philosophisch-historische Klasse, Nr.6. Berlin: Verlag der Akademie der Wissenschaften. Shirt, David J. 1975. Les verba cogitandi dans les constructions interrogatives en ancien français. Revue de Linguistique Romane 39. 351–380. Söll, Ludwig. 1974. Gesprochenes und geschriebenes Französisch. Berlin: Schmidt. Thompson, Sandra A. & Anthony Mulac. 1991a. The discourse conditions for the use of the complementizer that in conversational English. Journal of Pragmatics 15. 237–251. Thompson, Sandra A. & Anthony Mulac. 1991b. A quantitative perspective on the grammaticization of epistemic parentheticals in English. In Elisabeth C. Traugott & Bernd Heine (eds.). Approaches to grammaticalization. Vol. 2: Focus on types of grammatical markers. Amsterdam/Philadelphia: Benjamins. 313–329. Tristan en prose. Philippe Ménard (ed.). 1987. Genève: Droz. . Van Bogaert, Julie. 2011. I think and other complement-taking mental predicates: A case of and for constructional grammaticalization. Linguistics 49. 295–332. Waltereit, Richard. 1999. Reanalyse als metonymischer Prozess. In Jürgen Lang & Ingrid Neumann-Holzschuh (eds.). Reanalyse und Grammatikalisierung in den romanischen Sprachen. Tübingen: Niemeyer. 19–29. Zafiu, Rodica. 2013. Modality and evidentiality. In Gabriela Pană Dindelegan (ed.). The grammar of Romanian. Oxford: Oxford University Press. 575–584. Zink, Gaston. 1997. Morphosyntaxe du pronom personnel (non réfléchi) en moyen français (XIVe-XVe siècles). Genève: Droz.

Ditte Boeg Thomsen

Ontogenetic paths to the parenthetical construction Abstract: In children’s acquisition of double-clause constructions, clauses in parenthetical position are special in allowing different interpretations of clauserelationship structure. Children may categorize parenthetical clauses as variants of matrix clauses, adverbial clauses or independent main clauses or as autonomous chunks. Spontaneous-speech analyses of English and German acquisition take internal inflexibility as evidence of chunk acquisition (Brandt, Lieven & Tomasello 2010; Diessel & Tomasello 2001), but inflexibility in parentchild interactions may be inaccurate as chunk-status indicator since these contexts may encourage stereotyped viewpoint talk. The present study uses a kindergarten corpus to compare parenthetical clauses and possible source constructions in Danish. Analyses of distribution, flexibility, pronunciation and formal marking suggest that clauses with parenthetical verbs are used more flexibly in children’s group conversations than in one-on-one interactions with a caregiver. In Danish, the parenthetical construction most likely develops as an extension of the complement-clause construction, supported by children’s schemas for object-first clauses. Keywords: L1 acquisition, complement clauses, parenthetical verbs, viewpoint constructions, child-language corpora

|| Ditte Boeg Thomsen: Department of Nordic Studies and Linguistics, University of Copenhagen. Copenhagen, Denmark.

1 Introduction When children in the process of building up their grammatical systems encounter constructions with parenthetical clauses such as that’s an ostrich I think, they have to find out how to treat these structurally ambiguous constructions, in which the relationship between the two finite clauses can be interpreted in different ways. As linguists deliberate whether parenthetical clauses are best analysed as variants of matrix clauses in complement-clause constructions (I think that that’s an ostrich, cf. Thompson & Mulac 1991), adverbial clauses

190 | Ditte Boeg Thomsen (that’s an ostrich as I think, cf. Brinton 2008) or independent object-first clauses (that’s an ostrich. That I think, cf. Fischer 2007), children must also decide which of their developing and malleable schemas they should employ to process and produce the parenthetical construction. A child’s analysis may change at different stages of development, and individual acquisition could equally well recapitulate diachronic development or follow independent paths (Diessel & Tomasello 2001). Indeed, research on German and English acquisition has suggested that concrete clauses with complement-taking predicates (CTPs) such as I think or ich glaube (‘I believe’) are acquired early in development as autonomous chunks with adverbial distribution (Brandt, Lieven & Tomasello 2010, Diessel & Tomasello 2001). If that is the case, parenthetical use of CTPs would be primary, not dependent on any higher schemas, and if a relationship to other schemas – such as the complement-clause construction – should at any point arise, this abstraction would be a secondary development, irrelevant to a child’s felicitous use of the parenthetical clauses themselves. However, previous studies have used the internal inflexibility of a CTP clause as the main criterion for categorization as unanalysed chunk, whereas other types of evidence (such as pronunciation, formal marking and distribution) are treated more cursorily. Further, the parent-child interactions in the corpora used may encourage more stereotyped viewpoint talk than interactional settings with more participants, entailing a risk of underestimating the flexibility of children’s CTP clauses. It is therefore worth asking whether parenthetical clauses will still appear to be acquired as autonomous chunks, in the sense of internally inflexible underanalysed strings independent of more general schemas if we examine: 1) the temporal relationships and frequency balances between occurrence of clauses in parenthetical position and their possible sources (matrix clauses, independent object-first clauses, adverbial clauses), and 2) the degree of schema flexibility and solidity for these source constructions themselves in child speech when observed in a corpus of group conversations between peers. If children produce parenthetical clauses by extending a more general schema, the parenthetical clauses should be expected to occur at a point in development where the source construction is already used in a flexible manner. This hypothesis will here be examined in another Germanic language, Danish, where different possible ontogenetic origins of parenthetical clauses (as chunks, matrix clauses, adverbial clauses or independent object-first clauses) will be evaluated by means of a spontaneous-speech analysis in a 131-hour cor-

Ontogenetic paths to the parenthetical construction | 191

pus of group conversations between children aged 1;10–6;7 years1. Particular attention will be given to the status of clauses with two highly frequent parenthetical verbs, tro (‘think, believe’) and synes (‘think, find’), when they occur in sentence-initial (matrix-clause-like) position and in sentence-medial or -final (parenthetical) position at different age stages. Specifically, I examine details of pronunciation (does phonetic reduction or fusion suggest chunk status?) and formal marking (do word-order and complementizer cues point to embedding or non-embedding?) in addition to the traditionally privileged question of diversity (is the CTP clause a fixed string, or does it have flexible word order and slots for a variety of subjects and inflectable predicates?). The findings from this spontaneous-speech analysis suggest that Danish children treat CTP clauses with parenthetical verbs as less chunky than suggested for English and German children, and that use of these clauses in parenthetical (sentence-final or -medial) position only occurs as a rare and sophisticated phenomenon produced by children with flexible schemas for both complementation and object-first clauses. This surprising result could be explained by cross-linguistic differences as well as differences between the corpora used. The structure of the paper is as follows. Section 2 presents the parenthetical construction and potentially related constructions in Danish, and Section 3 summarizes research on child acquisition of parenthetical clauses in English and German, focusing on the chunk hypothesis. Section 4 presents an investigation of Danish children’s use of parenthetical clauses, with subsections examining questions of chunk status, relationship to matrix clauses, object-first clauses and adverbial clauses. Section 5 compares the results from Danish acquisition with findings from other languages and discusses possible explanations for the differences found. First, however, a terminological clarification is required. In this paper, “parenthetical construction” is used as a purely formal label registering sentence-internal ordering of elements with no reference to discourse-pragmatic function: It designates a construction where a finite clause with a complementtaking predicate accompanies another finite clause in either medial or final position without being introduced by a conjunction. Expressions with sentenceinitial CTP clauses such as I think it’s an ostrich are thus not considered instances of the parenthetical construction. As for “parenthetical verb”, the term here

|| 1 This paper follows the acquisition literature convention of giving children’s ages in (years; months), i.e. the youngest children in this study are 1 year and 10 months of age, the oldest 6 years and 7 months.

192 | Ditte Boeg Thomsen designates the subclass of CTPs which may be used in the parenthetical construction, i.e. it is a label referring to a distributional possibility for the verb, with no reference to its concrete position in a specific utterance. The term “parenthetical clause”, on the other hand, specifies the concrete clause-medial or clause-final position of a CTP clause.

2 The parenthetical construction in Danish and children’s possible paths to it In Danish, as in other European languages, clauses with a subject and a finite verb can be used parenthetically, i.e. at the end of another clause or inserted into it. The parenthetical clause is never introduced by a conjunction, the order of subject and verb is always VS (opposed to basic Danish SV order), and the verb must belong to a group of complement-taking verbs designating cognition, perception, emotion, utterance or appearance (Hansen & Heltoft 2011: 1150ff, 1147). This study aims to evaluate different possible ontogenetic origins of the parenthetical construction in Danish child language. One possibility is that children’s acquisition should resemble historical development where diachronic research in other languages has pointed out the complement-clause construction (1) (cf. Thompson & Mulac 1991), independent transitive constructions with object-first structure (2) (cf. Fischer 2007), or constructions with adverbial clauses (3) (cf. Brinton 2008) as plausible historical sources. (1) I think that she planted lilacs  she planted lilacs, I think (2) She planted lilacs. That I think  she planted lilacs, I think (3) She planted lilacs as I think  she planted lilacs, I think However, whichever construction the parenthetical construction has developed as a variant of in a specific language, children acquiring that language could follow any of the three routes if their language presents them with all three constructions, making sense of the structurally ambiguous clauses in their input by comparing them to already mastered constructions. On the other hand, children could also merely interpret the parenthetical constructions as simple clauses with monomorphemic modifiers as suggested by previous ontogenetic research. Children acquiring the same language will not necessarily make the

Ontogenetic paths to the parenthetical construction | 193

same analyses, and an individual child could certainly also approach the problem from more angles. Danish presents children with all three possible source constructions. It has complement-clause constructions where the embedded status of the complement clause may be pointed out by means of complementizers or embeddedclause word order. Short independent clauses with object-verb-subject (OVS) structure where verb and subject are positioned after a pronominal object (as in (2): that I think) are also highly frequent in conversation. As for adverbial clauses as in 3 (as I think), they do exist, but they are not very frequent (Jensen 2003), and the most frequently used parenthetical verbs tro (‘think, believe’) and synes (‘think, find’) are typically not used by adults in adverbial comparison clauses of this type, making it highly unlikely that children should take such a detour to analyse parenthetical clauses with those verbs. Nevertheless, all four routes are available for Danish children and will therefore be examined here after a presentation of previous research on children’s use of parenthetical clauses in other languages.

3 Earlier research on children’s acquisition: The chunk hypothesis Children’s use of parenthetical clauses has primarily been investigated as part of research on children’s use of constructions with complement clauses (Brandt et al. 2010, Diessel & Tomasello 2001). A few words on the relationship between the complement-clause construction and parenthetical clauses are therefore useful. In an uncontroversial complement-clause construction such as otherwise he will discover that we have crept in, the complement-clause that we have crept in functions as an object for discover in the CTP-clause otherwise he will discover, the only difference from a simple transitive clause such as otherwise he will discover us being that the object is sentential instead of nominal. What a Danish boy at age 5;7 asserted with this utterance (in Danish) was the predication of the CTP clause, whereas the predication of the complement clause was only presented as part of a hypothesized mental event. In English and German, children below school age appear to build up at least a semi-abstract schema for the complement-clause construction with its embedding of one predication in another. They use more and more different verbs and subjects in the CTP clause, indicating a flexible schema, and over the years they come to formally mark the complement clause as embedded – with

194 | Ditte Boeg Thomsen complementizers and (in German) embedded-clause word order (Brandt et al. 2010: 595). So, German preschoolers may produce utterances such as (4), English-speaking preschoolers such as (5) (4) Nee, ich glaub, dass es nicht dring macht No I believe that it not ‘dring’ makes ‘No, I believe that it doesn’t make “dring”’ (ex 24a in Brandt et al. 2010: 601).

(boy 4;9)

(5) Do you think they’ll know who wrote the letter then? (ex 60 in Diessel & Tomasello 2001: 131)

(boy 4;1)

A recurring question, however, is how to analyse constructions with highly frequent CTP clauses such as I think, you know or ich glaube (‘I believe’) in children’s speech. When they occur in clause-initial position as in (6), they formally resemble typical complement-clause constructions in which the complement clause acts as an object for the predicate in the CTP clause. (6) I think my daddy took it (ex 32n in Diessel & Tomasello 2001: 112)

(girl 3;7)

On the other hand, a line of special features makes it doubtful that these CTP clauses actually function as matrix clauses, embedding the complement clause. First of all, the predicates in these clauses are used inflexibly in present tense with 1st or 2nd person subjects. Secondly, they accompany clauses that are not formally marked as subordinate, lacking complementizers and (in German) exhibiting declarative word order. Thirdly, these clauses display deviant distribution compared to other predicating verbs, occurring also in clause-medial and -final (i.e. parenthetical) position relative to their supposed objects. Fourthly, constructions with these specific CTP clauses appear ontogenetically early compared to other complement-clause constructions. The findings of internal fixedness and external positional freedom of the CTP clauses and nonsubordination of the supposed complement clauses lead Diessel & Tomasello (2001) and Brandt et al. (2010) to conclude that children treat these clauses with parenthetical verbs as unanalysed word-like entities. These chunks can be moved around freely in another clause (cf. Diessel & Tomasello 2001: 107), which is never subordinate to the parenthetical clause, but rather modified by it as by an adverb (see 7): (7) (I think) my daddy (I think) took it (I think)

Ontogenetic paths to the parenthetical construction | 195

The chunk analysis is also loosely supported by allusions to reduced pronunciation, potentially making the CTP clauses morphologically opaque: (did/do) you know > y’know; (did/do) you mean > y’mean; I guess > guess (Diessel & Tomasello 2001: 107). In these analyses, children’s use of clauses in parenthetical position is not derived from anything else, such as primary use in clause-initial matrix-clause position, in simple independent clauses or as adverbial clauses. On the contrary, it is suggested that parenthetical clauses such as I think and ich glaube are acquired and used as unanalysed chunks with modifying function whether in initial, medial or final position. Such constructions would be syntactically simpler for the child because they would not involve integration of two clauses, but only the modification of one clause with a “monomorphemic” modifier, and they would be conceptually simpler by expressing only one predication instead of two. This would further explain why children appear to use constructions with parenthetical verbs earlier than uncontroversial complement-clause constructions. Using “a clause” in parenthetical position is thus nothing special or challenging since it is merely equivalent to using an epistemic adverb in medial or final position (es ist (vielleicht) besser so (vielleicht), it’s (perhaps) better that way (perhaps)). This proposed developmental scenario is convincing, but it would still benefit from a sharpening of the criteria for categorizing CTP clauses as unanalysed chunks. Low flexibility and high frequency in spontaneous speech are thus taken as indications of chunk status, but there are as yet no thresholds to assess how much flexibility is needed to help a child access the parts of a supposedly fixed expression. Indeed, a recent experiment by Brandt, Verhagen, Lieven & Tomasello (2011) suggests that the high frequency of a string such as ich glaube does not preclude analysis of its internal constituency, at least for 5-year-olds. As for deviant pronunciation indicating non-transparency of the potential chunks, none of the studies presents phonetic analyses, and it is impossible to know how often clauses with parenthetical verbs are phonetically reduced to a degree that makes them morphologically opaque. Regarding the clues to (non-) embedding of the “complement clauses” in the CTP clauses, the German study focuses on the difference between embedded-clause word order and mainclause word order and demonstrates that children are able to formally embed complement-clauses from their initial use of the construction, but that they do so very rarely with epistemic 1st-person CTP clauses as ich glaube ‘I believe’. The study of English does not provide quantified information on the balance between presence and absence of complementizers. Finally and, in this context, most importantly, none of the studies provides information on the balance in use of CTP clauses in clause-initial (i.e. matrix-

196 | Ditte Boeg Thomsen clause-like) vs. clause-medial and final (i.e. adverb-like) position. Diessel & Tomasello list clause order as one of their coded features (2001: 110), but only mention it cursorily in their data analysis. It is thus unclear if the highly frequent CTP clauses occur in both initial and parenthetical positions at the same point in development for children, and whether adverbial distribution should indeed count as an argument in favour of chunk analysis. All in all, it is still unclear whether children do come to use clauses in parenthetical position as unanalysed chunks, or whether they derive a schema for a parenthetical construction from other already acquired schemas such as the complement-clause construction, the object-first construction or constructions with adverbial clauses. To establish which paths children pursue to the parenthetical construction – or possibly just to use epistemic chunks – this paper therefore inspects the temporal relationship between children’s use of parenthetical verbs (i.e. CTPs with a distributional potential for occurring in parenthetical position) in different constructions in a longitudinal spontaneous-speech corpus. The language investigated is another Germanic language, Danish.

4 Spontaneous-speech analysis: Danish acquisition This study aims to evaluate different possible ontogenetic origins of the parenthetical construction in Danish child language. As mentioned in section 1 above, Danish presents children with various plausible source constructions, and after a presentation of general characteristics of development of parenthetical clauses in Danish ontogeny, this section compares acquisition of parenthetical clauses with 1) matrix clauses (4.3), 2) object-first clauses (4.4), and 3) adverbial clauses (4.5). Questions of chunk status will be treated in the examination of matrix clauses.

4.1 Method All analyses were carried out in the Trørød Corpus, a 131-hour video corpus consisting of group conversations between Danish children in their kindergarten north of Copenhagen (Boeg Thomsen 2012a). 66 children born between 2002 and 2010 participate (age range 1;10–6;7), 31 girls, 35 boys. Seven are bilingual (two with English as their other first language, four with Faeroese and one with

Ontogenetic paths to the parenthetical construction | 197

French). Table 1 shows the proportion of hours recorded at each of four agelevels. Age group 1–2 years 3 years 4 years 5–6 years

Recordings 32 hours 32 hours 31 hours 36 hours

Table 1: Amount of recordings at different age levels

A 100-hour subpart of the corpus is longitudinal, with 20 children being recorded 6 hours a year (2 hours at 4-month intervals), starting when they begin in kindergarten (at 1;10–2;3 years) and ending when they leave (at 5;7–6;7). The recordings consist of 1-hour interactions between peer groups of four children engaged in everyday activities such as drawing, having lunch, playing board games, playing with a doll’s house and looking in books in a secluded room in their daycare institution. The majority of the corpus consists of pure peer conversations with no scaffolding from adults, the exception being sessions with the youngest who cannot be left alone in the room. This type of corpus data thus differs in various ways from the ones used in previous studies of children’s use of parenthetical clauses: (1) Subjects and sampling: Many more children contribute to the Trørød Corpus (66 children vs. 7 in Diessel & Tomasello (2001) and 4 in Brandt et al. (2010)), and one can therefore hope for more representativity. On the other hand, the sampling is much sparser than desirable, especially with potentially rare constructions (cf. Brandt et al. (2010)) whose primary subject was recorded 5–20 hours a month, the other three 0.5–1 hour a month, and Diessel & Tomasello (2001) whose 7 subjects were recorded 2– 4 hours a month). (2) Age range: Though still restricted to preschool children, by including data from children up to 6;7 years the age range in the Danish corpus is wider than the range in the English corpus, 1;2–5;2 years (Diessel & Tomasello 2001), but comparable to the range in the German corpus, 2;0–7;0 years (3 of 4 children in Brandt et al. (2010); however, the majority of their analyses are restricted to a subject whose last data point is at age 5;0).

198 | Ditte Boeg Thomsen (3) Interactional setting: The Trørød Corpus consists of group conversations between peers as opposed to the child-parent interactions in the corpora used in the German and English studies. This implies two things: a) that CDS analysis is excluded, but that we see children’s own abilities to structure interaction without scaffolding; b) that several equal interlocutors participate in the conversation instead of a child and a parent with predetermined hierarchical role differences. This final feature makes it expectable that the Trørød Corpus will contain more varied perspective-marking language than corpora with 1:1-interactions between a child and its mother. Firstly, there are lots of disagreements and processes of finding allies among the four children in each group, leading to much reference to own and others’ attitudes. Secondly, there are more perspectives available in the room, facilitating talk about other people’s invisible mental states, simply because third persons with readable reactions are actually present and relevant in the room, not just a speaker and a hearer. This varied landscape of perspectives in group interactions is advantageous for investigations of children’s use of parenthetical clauses and the constructions they may be related to since the CTPs used in all of them primarily designate mental states and utterance events. The corpus only consists of video files with no transcription or tagging. Therefore, all utterances with 1) parenthetical clauses, 2) complement clauses, 3) object-first clauses, and 4) adverbial clauses with parenthetical verbs have been excerpted directly from the video files and transcribed in Danish standard orthography. In addition, all clauses with tro (‘believe’) and synes (‘think, find’) were transcribed in a semi-narrow phonetic transcription.

4.2 General characteristics The spontaneous speech analysis shows that use of CTP clauses in parenthetical position is a relatively rare phenomenon in Danish child speech. During 131 hours of conversation, only 74 utterances with parenthetical clauses occur. For comparison, 4552 utterances with complement clauses occur in the same sessions. As seen in table 2, the vast majority of parenthetical clauses occur in the oldest age group (5–6 years: 36 h, 45 utt.) whereas they are exceedingly rare among the 1–3-year-olds (64 h, 13 utt.). This rise may simply reflect a general increase in use of double-clause constructions over the preschool years, for if the parenthetical clauses are subsumed under the complement-construction (here: only for the purpose of numerical comparison), they account for a stable

Ontogenetic paths to the parenthetical construction | 199

low share of 1–2% across these age stages. Table 2 also specifies the position of the parenthetical clauses, revealing that use of a parenthetical clause in final position is much more widespread than use in medial position. Age group (recordings) 1–2 years (32 h) 3 years (32 h) 4 years (31 h) 5–6 years (36 h) All children (131 h)

No. of types 3 6 5 7 10

No. of tokens (medial) 1 2 2 2 7

No. of tokens (final) 4 6 14 43 67

No. of all CTP constructions 225 688 1238 2475 4626

Table 2: Parenthetical clauses in different age groups

The class of parenthetical verbs is small: Only 10 different CTPs are found in parenthetical clauses and only 3 of those in medial clauses. Again, for comparison, 83 different CTPs are found in initial CTP clauses. Table 3 displays the predicates and the amount of tokens in different positions. Predicate Cognition tro (think, believe) synes (think, find) mene (mean) vide (know) huske (remember)

Medial tokens

Final tokens

5

27 14 5 3 1

1

Utterance sige (say) spørge (ask)

10 2

Perception se (see) høre (hear)

3 1

Pretense lege (play, pretend)

1

1

Table 3: Complement-taking predicates (CTPs) occurring in the parenthetical construction

The predicates predominantly occur in present tense (85%) with some exceptions in past and perfect tenses, especially with the two utterance verbs. As for the subjects used in parenthetical clauses, they are almost restricted to pronouns pointing to participants in the speech event: 1.SG is by far most frequent

200 | Ditte Boeg Thomsen (76%), followed by 2.SG (12%), 1.PL (4%) and 2.PL (4%). For the verb sige, ‘say’, however, half of the utterances present 3rd-person subjects (3.SG.M, S.3G.F, different proper names). The most inflexibly used predicates in parenthetical clauses are the four cognitive verbs tro ‘think, believe’, synes ‘think, find’, mene ‘mean’ and vide ‘know’ which exclusively occur in present tense and only with speechparticipant subjects (1st and 2nd person). Such clauses might therefore be suspected of being fixed strings which children acquire as unanalysed chunks.

4.3 Comparison with the complement-clause construction To investigate the hypothesis that the parenthetical construction is derived ontogenetically from the complement-clause construction, children’s use of CTPs in parenthetical clauses was compared to use in sentence-initial CTP clauses. In order to ascertain that constructions with sentence-initial CTP clauses should indeed count as complement-clause constructions analyses of flexibility, pronunciation, formal marking and overgeneralization were also carried out. For each of the children followed longitudinally, their first use of a specific CTP in parenthetical position was compared to use in sentence-initial position. Table 3 shows the age points for first use of a CTP in initial and parenthetical clauses, respectively, as well as the distance in months between these first occurrences. In the one case where this number is negative, it means that the child used the CTP in a parenthetical clause first. The table only compares occurrences for the three CTPs which most frequently occur in parenthetical position (≥ 10 parenthetical clauses): tro (‘think, believe’), synes (‘think, find’) and sige (‘say’) with 32, 14 and 10 parenthetical tokens, respectively.

Ontogenetic paths to the parenthetical construction | 201

Age at first initial occurrence

Age at first parenthetical occurrence

Delay of parenthetical use (months)

tro (think, believe) Eva Ketil Jane Janus Christina Christa Jasper

3;5 3;1 2;10 3;7 3;4 3;3 3;7

2;10 4;2 4;6 4;7 5;1 5;3 5;4

-7 13 20 12 21 24 21

synes (think, find) Kira Carl Janus

3;2 3;2 3;11

4;9 5;0 6;0

19 22 25

sige (say) Olga Hilde Janus Jasper

2;5 2;9 4;3 3;7

2;9 4;1 5;0 6;0

4 16 9 29

Table 4: Comparison between first uses in initial and parenthetical clauses

The pattern emerging from this comparison is clear: On average, use of a CTP in a parenthetical clause only occurs 16 months after a child’s first attested use of the same verb sentence-initially. Only one child (Eva at 2;10) produces a parenthetical clause with a CTP before having used it in an initial CTP clause. For all other children, sentence-initial use precedes parenthetical use, on average with 10 sentence-initial tokens per child before the first parenthetical token. It should, however, be repeated that the sampling is not very dense (6 hours per child per year) and that it is therefore more likely than not that children use CTPs in both initial and parenthetical clauses before they show up in the corpus. Further, the infrequency of the parenthetical clauses compared to the initial CTP clauses makes the former less likely to be picked up in the samples which may thus make them appear more delayed in comparison to the latter than they actually are (cf. Tomasello & Stahl 2004). The temporal relationship between CTPs in these constructions should therefore be checked in corpora with denser sampling. Nevertheless, that the infrequency of parenthetical clauses in itself is indeed a developmental phenomenon and not just a reflection of infrequency in children’s input is made likely by a comparison of the proportion of parentheti-

202 | Ditte Boeg Thomsen cals in child speech and child-directed speech: In a 12-hour subpart of the Danish acquisition corpus Odense Twin Corpus (Basbøll et al. 2002), parenthetical clauses accounted for 20% of the 132 double-clause constructions with jeg tror/tror jeg (‘I think/believe’) produced by six adults (Boeg Thomsen 2012b) whereas they only account for 8% of the 402 double-clause constructions with jeg tror/tror jeg (‘I think/believe’) produced by the children in this study (χ2 = 14.3851, df = 1, p = 0.00017). The trend in these data thus speaks against the idea that children generally acquire a certain set of highly frequent CTP clauses as chunks which may be moved around freely as adverbs in another clause. On the contrary, the hypothesis that children slowly develop schemas for the parenthetical construction as expansions of their complement-clause-construction schemas is made likely: Only CTPs that the children are comfortable using in finite clauses accompanying other finite clauses sentence-initially are introduced in parenthetical clauses. However, whether utterances with CTP clauses in initial position are indeed instances of the complement-clause construction is debatable. As argued by Brandt et al. (2010) and Diessel & Tomasello (2001), highly frequent CTP clauses such as ich glaube and you know in initial position may still be unanalysed chunks. Even if they are not freely movable, they may still exhibit other traits making their status as matrix clauses doubtful such as inflexibility as fixed strings, strong reduction and formal marking of their supposed complement clauses as non-embedded main clauses. These traits are therefore examined for the two most frequent Danish CTPs in parenthetical clauses when they appear in sentence-initial clauses. Flexibility The corpus contains 558 utterances with an initial CTP clause with tro ‘think, believe’, and 285 with synes ‘think, find’. As expected, the majority refer to attitudes present in the speech situation, i.e. they are in present tense and have 1st or 2nd person subjects. We might therefore suspect four clause types to be fixed unanalysed strings: jeg tror ‘I believe’ (66% of initial tro-clauses), du tror ‘you believe’ (6% of initial tro-clauses), jeg synes ‘I think’ (68% of initial synesclauses) and du synes ‘you think’ (13% of initial synes-clauses). However, a couple of factors speak against it. First of all, Danish is a verb-second language where the order of verb and subject is switched from SV to VS whenever a nonsubject element occurs in the first position of the clause, P1, and when P1 is empty in polar questions. This means that all of the four frequent combinations occur in two orders (the VS orders are the ones found in parenthetical position):

Ontogenetic paths to the parenthetical construction | 203

jeg tror tror jeg du tror tror du I believe believe I you believe believe you 311 utt. 59 utt. 2 utt. 32 utt. jeg synes synes jeg du synes synes du I think think I you think think you 171 utt. 23 utt. 11 utt. 26 utt. For children encountering these clauses in their input or producing them themselves as e.g. the boy in examples (8) and (9), this variable order of the elements must be a strong incentive to analyse the expressions and recognize their identities as clauses. (8) Nu tror jeg vi skal læse den i dag Now believe I we will read it today ‘Now I think we’ll read it today’ (9) Jeg tror det var Batman I believe it was Batman ‘I think it was Batman’

(boy 3;9)

(same boy, 3 minutes later)

A further argument against regarding these frequent tro/synes clauses as fixed strings for children is that the predicates are also widely used with other subjects and in other tenses, as displayed in Diagrams 1 and 2.

204 204 | | Ditte DitteBoeg BoegThomsen Thomsen

Present Present Present addressee addressee addressee attitude; attitude;58; 58; attitude; 58; 10% 10% 10%

Present Present Present speaker speaker speaker attitude; attitude; attitude; 370; 370;67% 67% 370; 67%

Non-speech Non-speech Non-speech situation situation situation attitude attitude(3rd (3rd attitude (3rd person); person);33; 33;Non-speech person); 33; Non-speech Non-speech 6% 6% 6% situation situation situation attitude attitude attitude (past); (past);94; 94; (past); 94; 17% 17% 17%

Tro Tro (think, (think,believe) believe)

Diagram Diagram11

Present Present Present addressee addressee addressee attitude; attitude;55; 55; attitude; 55; 19% 19% 19%

Non-speech Non-speech Non-speech situation situation situation attitude attitude attitude (past); (past);15; 15; (past); 15; 5% 5% 5% Non-speech Non-speech Non-speech situation situation situation attitude attitude(3rd (3rd attitude (3rd person); person);19; 19; person); 19; 7% 7% 7%

Present Present Present speaker speaker speaker attitude; attitude; attitude; 194; 194;69% 69% 194; 69%

Synes Synes(think, (think,find) find)

Diagram Diagram22 Diagrams Diagrams11&&2: 2:Diversity DiversityininCTP CTPclauses clauseswith withtro troand andsynes synes

Diagrams Diagrams 11 and and 22 show show the the proportions proportions of of CTP CTP clauses clauses with with different different subjects subjects and and tenses tenses in in the the whole whole child child corpus corpus across across age age groups. groups. There There are, are, however, however, important importantdifferences differencesbetween betweenage agegroups groupswhen whenititcomes comesto todiversity diversityin inthe theCTP CTP 3-year-olds,almost almostonly onlyuse usetro tro‘think, ‘think, clauses. clauses.The Theyoungest youngestage agegroups, groups,the the11––3-year-olds, believe’ believe’in inpresent presenttense tensewith with1.SG 1.SGsubjects subjects(81%). (81%).For Forthe theoldest oldestage agegroup group(5-6 (5-6

Ontogenetic paths to the parenthetical construction | 205

years), on the other hand, this share has fallen to 61%, while no less than 30% refer to non-speech-situation attitudes: 8% have 3rd person subjects, 22% other tenses. It is at this age that children usually begin to use these same CTPs in parenthetical clauses. Whether or not children initially acquire highly frequent CTP clauses as chunks or not, they only introduce those clauses in the parenthetical construction at a point when they appear to have analysed the expressions as clauses with slots for different subjects and inflectable predicates. In a game with plenty of uncertainty, one child for instance refers to his own present belief with a parenthetical clause and continues to refer to this same past belief with a complement-clause construction, pointing to the relationship between the constructions: (10) Det er en struds tror jeg it is an ostrich think I ‘It’s an ostrich I think’ (11) Jeg troede det var en struds I thought it was an ostrich ‘I thought it was an ostrich’

(boy 5;4)

(same boy, 14 secs later)

Pronunciation A further argument adduced in favour of a chunk analysis has been phonetic reduction (Diessel & Tomasello 2001: 107). If children pronounce the morphemes in the CTP clause so strongly reduced or co-articulated that the morphemic structure of the clause becomes opaque, this may be evidence of the expressions behaving like autonomous expressions without internal structure (cf. Bybee 2006). To investigate this possibility, all CTP clauses with tro or synes whether in initial or parenthetical position were transcribed in IPA and compared with distinct target pronunciation (in modern standard Copenhagen Danish). A test sample of 43 tro clauses was also transcribed by another phonetician (intertranscriber agreement 91.9 %). Only the results for the most frequent combinations (1.SG subject, present tense) are presented here: 400 instances of jeg tror (‘I think, believe’) and 204 instances of jeg synes (‘I think, find’). Details of transcription and categorization can be found in Boeg Thomsen (2012b) which treats a subset (316) of the tro clauses analysed here. The phonetic analysis revealed that the predicate tror was almost always pronounced distinctly as [to] (86%, including slight variations owing to children’s individual general phonological strategies for coping with /tr/-clusters). The few reductions found were weak and unsystematic. For synes, there was

206 | Ditte Boeg Thomsen more divergence from distinct pronunciation (distinct [syns] accounting for 62% of the instances), among other things due to more co-articulation with the subject pronoun. For the subject pronoun jeg [j] a large continuum of forms is found – from fully distinct [j] to fully reduced [Ø]. When jeg is weakly reduced as in the categories Narrow/rounded [ja] [jæ] [j] [j] [jö], Centralized [j] and Loss of [j] ([]), it is still recognizable as an instance of the subject pronoun. In contrast, when jeg is strongly reduced as in the categories Schwa [], Glide [j] and Zero [Ø] or strongly assimilated as in the category Co-articulated [] [t] [] [ta] [] [], the presence of a subject pronoun may be opaque to the child. Diagrams 3 and 4 show the distribution of jeg-pronunciations in different age groups for each of the two predicates (the two first columns in Diagram 4 (synes) should be regarded with caution as they build on very few utterances, 24 and 29 respectively). For tro ‘think, believe’, we see a large amount of “unrecognizable” subject pronouns in the youngest age group: for the 2-to-3-year-olds 23% of the jegpronunciations fall in the categories Schwa and Zero. At a first glance, this lack of an explicit subject might be taken to indicate that (some of) the youngest children do not recognize jeg tror or tror jeg as clauses. However, as examined in Boeg Thomsen (2012b), individual children vary between distinct and reduced pronunciations, and the strong reductions often occur in repetition contexts where the child has just produced the full form. When the child produces both forms, the strong reductions cannot be taken as evidence of non-transparency (see Boeg Thomsen 2012b for a full discussion).

Diagram 3: Pronunciation of jeg as subject for tror (distributed on age groups)

Ontogenetic paths to the parenthetical construction | 207

Pronunciation of jeg 'I' in jeg synes /synes jeg 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%

Co-articulated Zero Glide Schwa Loss of [j] Centralized Narrow/rounded Distinct 2-3 years

4 years

5-6 years

(24 utt.)

(29 utt.)

(147 utt.)

Diagram 4: Pronunciation of jeg as subject for synes (distributed on age groups)

As for synes, there is less reduction than for tro, but here, on the other hand, we see some amount of co-articulation. This fusion is present in VS instances of the expression (synes jeg) where the alveolar fricative /s/ and the palatal glide /j/ are fused in the palatal fricative [], so that distinct [synsj] is replaced with [syn]. When elements of the clause are thus fused together and pronounced differently from other contexts, this could be a classic cue to status as autonomous chunk (Bybee 2006, Wray 2002). Again, however, it is important to note 1) that distinct pronunciation is still the dominating norm, 2) that co-articulated instances of synes jeg appear to be an advanced phenomenon, only occurring in older preschoolers, the 4-to-6-year-olds (with one exception at 3;9). The phonetic analysis thus cannot support a hypothesis that children acquire [syn] as an opaque chunk; rather, co-articulated pronunciation would seem to depend on fluency and be a signal of advanced familiarity with the expression (cf. Schachtenhaufen 2013 on phonetic reduction as a mark of competency). And again, co-articulated pronunciation alternates with fully distinct pronunciation for all children who use it, making it more likely that fusion depends on other factors than opacity, such as speech tempo or pragmatic backgrounding (Rischel [1992] 2009). This, however, remains to be investigated. Comparing children’s pronunciation of jeg tror ‘I believe’ with adult pronunciation provides a final surprising argument against the hypothesis that reduced pronunciation in the input should make the morphemic make-up of the CTP clauses opaque to children. A comparison with pronunciation of CTP clauses with jeg tror in Danish child-directed speech indicates that Danish 5–6-yearolds generally pronounce the subject jeg more distinctly than Danish adults do

208 | Ditte Boeg Thomsen in child-directed speech2. Diagram 5 shows that 5–6-year-olds produce significantly more distinct pronunciations [j] than adults and also significantly less opaque pronunciations (Schwa [], Glide [j] and Zero [Ø]).

Pronunciation of jeg ('I') in jeg tror : 5-6-year-olds vs. adults 100% 90% 80%

Zero

70%

Glide

60%

Schwa

50%

Loss of [j]

40%

Centralized

30%

Narrow/rounded Distinct

20% 10% 0% 5-6 years

adult-CDS

(216 utt.)

(119 utt.)

Distinct

Opaque

Adults

48 of 119

40%

27 of 119

23%

5–6 years

129 of 216

60%

23 of 216

11%

Χ2

11.57166, p Φ

4.2 Pauses Number and duration of the pauses before and after the parentheticals are compared to the number and duration of the pauses before and after Φs and ιs in the control. Figure 4 presents a logarithm of the duration of the pauses before (left) and after (right) the indicated phrase types. The graph only presents the durations where a pause occurred. The rate of pauses after indicated types is analysed and presented separately.

Fig. 7: Duration (in log scale) of the pauses before (left) and after (right) each type

Are reporting clauses special cases of parentheticals? | 313 Mapping to prosody 311

The first impression we get from Figure 7 is that pauses occur both before and after all phrases of interest. For all parentheticals, the general tendency is that the pauses before the tested items are shorter than the pauses that come after. The pauses that occur before and after the interjections are the longest in duraAre reporting clauses special cases of parentheticals? | 313 tion. The pauses that come after the appositives are the longest among XP parentheticals. Pauses surrounding ιs are considerably longer than the pauses that The first impression we get from Figure 7 is that pauses occur both before and surround Φ types. after all phrases of interest. For all parentheticals, the general tendency is that Pause duration results are inthan line finalafter. syllable duration values; ι > the pauses before the tested items are shorter the with pauses the that come The pauses that occur and after the interjections the longest in dura-internal variation. Within parenthetical > before Φ. Parenthetical typesareshow a large tion. The pauses that come after the appositives are the longest among XP parparentheticals there is a verbal parenthetical > XP parenthetical ordering, espeentheticals. Pauses surrounding ιs are considerably longer than the pauses that surround Φ types. cially in the cases of pauses that follow the parentheticals. Excluding the set of Pause duration results are in line with the final syllable duration values; ι > inter, the pauses that come before the parentheticals seem to be closer in duraparenthetical > Φ. Parenthetical types show a large internal variation. Within parentheticals is a verbal > XP parenthetical tion to thethere pauses thatparenthetical come before Φ types. ordering, especially in the cases of pauses that follow the parentheticals. Excluding the set of We fit two more models as before, this time predicting the pauses on the inter, the pauses that come before the parentheticals seem to be closer in durasides from the types. tion to the pauses thatphrase come before Φ types.We include speaker and the item length as the ranWe fit two more models as before, this time predicting the pauses on the dom variables. The model in Figure 8 and Table 5 estimates the pause durations sides from the phrase types. We include speaker and the item length as the ranin the occurrences four5 estimates main groups. dom variables. The model inbefore Figure 8 the and Table the pause durations in the occurrences before the four main groups.

Fig. 8: Estimates of the duration of the pauses that occur before ι, Φ, verbal and XP parenthetical

Fig. 8: Estimates of the duration of the pauses that occur before ι, Φ, verbal and XP parenthetical

314 | Güliz Güneş and Çağrı Çöltekin 312 Estimate

SE

t value

Fixed effects ι

0.139

0.011

12.583

Φ

0.013

0.009

1.423

0.038

0.011

3.399

0.012

1.695

314 | Güliz Güneş and Çağrı Çöltekin Par.(verbal)

Par.(XP)

0.020 Estimate

Random effects s.d.

Fixed effects

SE

t value

ι

Length

Intercept 0.139

0.0051 0.011

12.583

Φ

Speaker

0.013 Intercept

0.009 0.0089

1.423

Par.(verbal)

0.038

0.011

3.399

Par.(XP)

0.020

0.012

1.695

Table 5: Estimates of the duration of the pauses that occur before ι, Φ, verbal and XP parentheRandom effects s.d. tical Length Intercept 0.0051 Speaker

Intercept

0.0089

The5:model fittheon theof the pauses that the phrases show that Table Estimates of duration pauses that occurare beforeobserved ι, Φ, verbal andbefore XP parenthetical pauses that follow ι are remarkably longer than the others. Pre-ι pause duration is followed by pre-verbal parenthetical pause duration, which is followed by the The model fit on the pauses that are observed before the phrases show that pauses that follow ι are remarkably longerbefore than theXP others. Pre-ι pause duration duration of pauses that come parentheticals. Pre-Φ pause duration is is followed by pre-verbal parenthetical pause duration, which is followed by the the shortest. duration of pauses that come before XP parentheticals. Pre-Φ pause duration is The model presented in Figure 9 and Table 6 below estimates of pause durathe shortest. presented in Figure 9 and 6 belowparenthetical estimates of pauseand dura-XP parenthetical. tionThe inmodel the occurrences after ι, Table Φ, verbal

tion in the occurrences after ι, Φ, verbal parenthetical and XP parenthetical.

Fig. 9: Estimates of the duration of the pauses that occur after ι, Φ, verbal and XP parenthetical

Fig. 9: Estimates of the duration of the pauses that occur after ι, Φ, verbal and XP parenthetical

Mapping to prosody | 313 Are reporting clauses special cases of parentheticals? 315

Estimate

SE

t value

Fixed effects ι

0.140

0.009

15.382

Φ

0.010

0.007

1.420

Par.(verbal)

0.068

0.012

5.636

Par.(XP)

0.036

0.013

2.867

Random effects s.d. Length

Intercept

0.0000

Speaker Intercept

0.0066

Table 6: Estimates of the duration of the pauses that occur after ι, Φ, verbal and XP parenthetical

Pauses that occur on the right edge of the analysed items support the previously attested order, i.e. ι > verbal parenthetical > XP parenthetical > Φ. This time the differences are clearer, and the estimates of the parameters are more certain. The results of this model show that pauses that occur after the parentheticals tested are longer than the pauses that occur before them. For the sake of completion, Figure 10 and Table 7 present estimates of detailed model parameters for pause durations before and after the phrase types.

Fig. 10: Estimates of the duration of the pauses that occur before (left) and after (right) all types

316 | Güliz Güneş and Çağrı Çöltekin 314 Before Estimate

SE

After t value

Estimate

SE

t value

Fixed effects ι

0.139

0.011

12.315

0.140

0.009

15.473

Φ-ar

0.011

0.011

1.040

0.009

0.008

1.158

Φ-app

0.016

0.013

1.215

0.015

0.014

1.017

finnon

0.030

0.022

1.397

0.136

0.026

5.142

adfin

0.050

0.015

3.422

0.046

0.016

2.825

com

0.020

0.018

1.114

0.063

0.022

2.886

adper

0.006

0.016

0.378

0.014

0.018

0.801

appo

0.053

0.020

2.645

0.083

0.022

3.829

admit

0.005

0.022

0.242

0.015

0.026

0.560

Random effects s.d. Length

Intercept

0.0061

0.140

Speaker

Intercept

0.0089

0.009

Table 7: Estimates of the duration of the pauses that occur before (left) and after (right) all types

The detailed models support the above conclusions, with the previously observed outliers. Now, we focus on the frequency of pauses rather than their duration. Figure 11 presents the number of pauses before and after each phrase type.

Mapping to prosody | 317 315 Are reporting clauses special cases of parentheticals?

Fig. 11: The rate of pauses before (left) and after (right) constituents for all types

The results of the distribution of the pauses that occur before and after the items analysed showed an ordering similar to the one of the results of the duration of the pauses. Among the parenthetical set, inter and voca exhibit the highest number of pauses on their right edge, which is at least as high as the number of pauses that come after ι boundaries. When we compare the verbal parentheticals to XP parentheticals (excluding inter and voca), we see that verbal parentheticals exhibit higher occurrences of pauses, which were also longer. Pauses on both sides of the Φ boundaries are shorter and fewer in number. The pauses surrounding the XP parentheticals are closer to Φ boundaries in duration and distribution. However, in contrast to their right edge, appositives exhibit a higher amount of pauses on their left edge. The ordering is shown in (13): (13) ι > verbal parenthetical > XP parenthetical > Φ In addition to the ordering we observed in (13), another conclusion of this subsection is that the pauses that come before each parenthetical type are shorter and fewer in number, whereas the pauses that come after each type are longer and more frequent.

318 | Güliz Güneş and Çağrı Çöltekin 316

4.3 Final rise The degree of rise of the final syllables of Φs and ιs in the control data is compared to the degree of rise of the final syllables of each parenthetical in the test data. In the case of ι-final rise, the set ‘ι’ is divided into two groups; (i) on nucleus-ι (ι-n), where the ι-final item itself is the nucleus and the ι ends with a high plateau (8a), and (ii) post-nucleus ι boundary (ι-pn), where the ι-final Φ is more than one word and the right edge comes after the nucleus (8b).

Fig. 12: The degree of rise in the final syllables

The box plot on Figure 12 represents the difference between the mean pitch value of the final and penultimate syllables for all types. The first impression we get from these graphs is again similar to the previous cues. The set of Φs shows the least degree of final rise. Within the set of parentheticals, verbal parentheticals seem to show a higher rise in comparison to XP parentheticals. Again, inter exhibits a very different trend than the other verbal parentheticals. Specifically, it exhibits the lowest degree of pitch difference, which indicates in most cases there is no rise but fall indicating a low boundary, L%. This becomes more visible when the individual F0 contours of inter are examined. The other exceptional case for its category was voca in the set of XP parentheticals. Similar to the

Mapping to prosody | 319 317 Are reporting clauses special cases of parentheticals?

results attained from the previous cues, voca exhibits a variation in its group and bears a higher final rise. Excluding the type inter, the final rise of ιs and verbal parentheticals are alike. Similarly, excluding the type voca, final rise of Φs and XP parentheticals are closer to each other. We observe that there is some variation in the case of ι based on whether ι shares its last word/syllable with the last syllable of the nucleus (ι-n), or whether it follows the nucleus (ι-pn). Accordingly, the former bears a smaller magnitude of final rise in comparison to the latter. This variation is expected considering the transmission from the low levelled pitch level of the post nuclear area to a H% boundary which is triggered by comma intonation. The same variation is also observed in the cases of Φ-final rise. Specifically, the phonological phrases that bear only one word (Φ-ar-ω) show a smaller magnitude of final rise, whereas phonological phrases that contain more than one word (Φ-ar-ω) exhibit a higher rise. This variation is due to the difference in the pitch register across the head and non-head parts of the Φs that contain more than one word. We also fit a model that takes length as well as the speaker variation as random effects (against systematic pitch-range variation due to speakers). First, a model that only distinguishes ι, Φ and two main parenthetical subdivisions is presented in Figure 13 and Table 8.

Fig. 13: Estimates of the final F0 rise of ι, Φ, verbal parenthetical and XP parenthetical

320 | Güliz Güneş and Çağrı Çöltekin 318 Estimate

SE

t value

Fixed effects ι

2.106

0.557

3.782

Φ

1.700

0.519

3.275

Par.(verbal)

2.469

0.577

4.277

Par.(XP)

1.556

0.576

2.700

Random effects s.d. Length

Intercept

0.6159

Speaker

Intercept

0.7905

Table 8: Estimates of the final F0 rise of ι, Φ, verbal parenthetical and XP parenthetical

All phrase types indicate a rise from penultimate syllable to final syllable. Verbal parentheticals show the highest rise, followed by ι, XP parentheticals and Φ, with a rather small difference between the XP parentheticals and Φ. This supports a ‘verbal par > ι > XP par >≈ Φ’ ordering. The model below shows the case of detailed grouping:

Fig. 14: Estimates of the amount of the final F0 rise for all types

Mapping to prosody | 319 Are reporting clauses special cases of parentheticals? 321

Estimate

SE

t value 4.246

Fixed effects ι

2.255

0.531

Φ-ar

1.811

0.503

3.601

Φ-app

1.250

0.584

2.140

Finnon

2.850

0.822

3.467

Adfin

2.721

0.645

4.221

Com

1.931

0.716

2.698

Adper

2.199

0.645

3.409

Appo

1.663

0.712

2.337

Admit

0.117

0.816

0.144

Random effects s.d. Length

Intercept

0.5007

Speaker

Intercept

0.7907

Table 9: Estimates of the amount of the final F0 rise for all types

The main trend is similar to the results above. Although the majority of verbal parentheticals exhibit high rise (which is higher than the ι condition), comment clauses diverge and exhibit a rise that is closer to Φ condition. The ordering of the four major groups in terms of the final rise is shown in (14): (14) verbal par > ι > XP par ≈ Φ

4.4 Initial lowering In the control, the mean F0 of the initial syllable of the Φ, where the L is observed, is subtracted from the mean F0 of the final syllable of the preceding Φ, where the H- is observed. The same procedure is also applied to ι boundaries. Note that, for this condition, only the non-initial Φs and ιs are calculated - i.e. in Φ cases, sentence initial syllables are excluded. In ι cases, only the ι-initial syllables of the non-initial sentences are analysed. In the test set, the mean F0 of the initial syllable of the parentheticals is subtracted from the mean F0 of the final syllable of the host (non-)constituent that immediately precedes the parenthetical. The graph in Figure 15 presents the mean difference of initial lowering.

322 | | Güliz GülizGüneş Güneşand andÇağrı ÇağrıÇöltekin Çöltekin 320 322

Fig.15: 15:The Thedifference difference(in (insemitones) semitones)of ofthe themean meanF0 F0of ofthe theinitial initialsyllables, syllables,and andthe themean meanF0 F0of of Fig. thefinal finalsyllables syllablesof ofthe theitems itemsthat thatimmediately immediatelyprecede precedethem them the

Thegraph graphshows showsaadegree degreeof of lowering loweringfor forall allgroups. groups.However, However,there theredoes doesnot not The seem to be a consistent difference across all types. In fact, the figure shows that seem to be a consistent difference across all types. In fact, the figure shows that forall alltypes, types,the theinitial initialfall fallvalues valuesare arecentred centredbetween between11and and22semitones semitonesfor forall all for cases. cases. Figure16 16and andTable Table10 10present presentthe theresults resultswe weobtain obtainwhen whenwe weinvestigate investigateininFigure itiallowering loweringwith withaamodel modelwith withfour fourgroups groups(also (alsoincluding includingthe theι-n/ι-pn ι-n/ι-pndistincdistincitial tion): tion):

Fig.16: 16:Estimates Estimatesofofinitial initiallowering loweringfor forι-n, ι-n,ι-pn, ι-pn,verbal verbalparenthetical, parenthetical,XP XPparenthetical, parenthetical,and andΦΦ Fig.

Are reporting clauses special cases of parentheticals? | 323 Mapping to prosody | 323 321 Are reporting clauses special cases of parentheticals?

Estimate

SE

t value

Estimate 0.290

SE 0.328

t0.886 value

-0.811 0.290 1.561

0.333 0.328 0.246 0.333 0.304

-2.438 0.886 6.349

Fixed effects ι-n Fixedι-pn effects ι-n Φ ι-pn Par.(verbal) Φ Par.(XP)

-0.811 1.525 1.561 1.308

Par.(verbal) Random effects s.d. Par.(XP) Length Intercept

1.525

0.246 0.323 0.304

1.308 0.0000

0.323

Random effects s.d. Intercept Speaker Length Intercept

0.4084 0.0000

-2.438 5.024 6.349 4.052 5.024 4.052

TableSpeaker 10: Estimates ofIntercept initial lowering for ι-n, ι-pn, verbal parenthetical, XP parenthetical, and Φ 0.4084 Table Estimates of initial lowering for the ι-n, ι-pn, verbal parenthetical, XP parenthetical, and Φ While10:the base level (ι-n) shows least differences in pitch, the post-nuclear

ι (ι-pn) shows the most. The initial syllable of the ι that follows ι-pn exhibits highWhile the base levelsyllable (ι-n) shows the least differences pitch, thetone. post-nuclear er F0 than the final of the preceding ι, which in bears a H% This mayι (ι-pn) shows the most. The initial syllable of the ι that follows ι-pn exhibits highbe due to two reasons: Either the rise on the right edge of ι-pn cases is not as er F0 than the final syllable of the preceding ι, which bears a H% tone. This may high, or the ιs that succeed the ιs that end with a post-nuclear area start higher be due to two reasons: Either the right edge of ι-pn casesF0is level not as (e.g. higher than the right edgethe of rise ι-pnon cases) and not low. Φ-initial is high, thesame ιs that theXP ιs that end with a post-nuclear area start higher almostorthe as succeed verbal and parentheticals. Φs, verbal parentheticals and (e.g. higher than the rightsubstantial edge of ι-pn cases) and not low. Φ-initial F0 level is XP parentheticals exhibit lowering on their left edges, which is conalmost the same as verbal and XP parentheticals. Φs, verbal parentheticals and siderably lower than both of the ι-initial cases. The last model below shows the XP parentheticals exhibit substantial lowering on their left edges, which is concase when we apply the model to all subgroups. siderably lower than both of the ι-initial cases. The last model below shows the case when we apply the model to all subgroups.

324 | Güliz Güneş and Çağrı Çöltekin 322

Fig. 17: Estimates of the initial lowering of all sub-types

Estimate

SE

t value 0.883

Fixed effects ι-n

0.290

0.328

ι-pn

-0.811

0.334

-2.432

Φ-ar-ω

1.598

0.264

6.048

Φ-ar-ωs

1.412

0.435

3.250

Φ-app-ω

1.399

0.480

2.913

Φ-app-ωs

1.664

0.480

3.464

adfin

1.428

0.379

3.769

com

1.667

0.455

3.661

finnon

1.553

0.554

2.805

admit

1.647

0.554

2.975

adper

1.192

0.417

2.857

appo

1.218

0.512

2.378

Random effects s.d. Length

Intercept

0.0000

Speaker

Intercept

0.4080

Table 11: Estimates of the initial lowering of all sub-types

We observe that the difference between Φ types is not substantial, and that all subtypes of parentheticals resemble the Φ condition. The information we gather from initial lowering is not sufficient to attribute these tendencies to left edge

Mapping to prosody | 323 Are reporting clauses special cases of parentheticals? 325

marking, since we cannot be sure if the difference is a result of the variation in the low start on the left edge of the constituents, or the higher/lower end on the right edge of the preceding items of the corresponding constituents. Considering the initial lowering results, one cannot conclude that there is a pitch reset in the case of parentheticals. Nor can we conclude that left edge F0 is employed to mark a difference between the left edges of Φs and ιs in Turkish. Precisely how to draw generalizations from these results is not yet clear to us, and hence must remain an issue for further research.

4.5 An overview of the results The tables below summarize the results. Table 12 lists the properties of parentheticals in two main groups: verbal parentheticals and XP parentheticals.14 Table 13 presents the results in terms of the category (Φ or ι) to which the subtypes of parentheticals are closer.15 Type of parenthetical Type of measure

Type of edge

Verbal

XP

Final syllable duration

Right edge

< ι, > Φ

< ι, > Φ

=ι

< ι, ≈ Φ

Pre-parenthetical syllable duration Left edge Pause length before

Left edge

≈ι

< ι, > / ≈ Φ

Pause length after

Right edge

≈ι

< ι, > Φ

The amount of final rise

Right edge

>ι

< ι, ≈ Φ

The amount of initial lowering

Left edge

-

-

Table 12: Summary of the results with main groups of parentheticals

|| 14 We thank the anonymous reviewer for the suggestion of this table for a summary of the findings. 15 Any sub-type of parentheticals that do not pattern with the other members of its group is italicized.

Type of measure

Phonological Phrase

Intonational Phrase

Final rise

admit, appo, adper, com

finnon, adfin

Final syllable duration

admit, appo, adper, com

finnon, adfin

Pause length after

admit, adper

finnon, adfin, com,

Pre-par. final syllable duration

admit, adper, com

finnon, adfin, appo

Pause length before

admit, adper, com

finnon, adfin, appo

Initial lowering

-

-

appo

cues

Left edge

Right edge cues

326 | Güliz Güneş and Çağrı Çöltekin 324

Table 13: Summary of the results with subgroups of parentheticals

5 Discussion The orderings of the parenthetical types with respect to ι and Φ boundaries in section 4 encode two important observations. First, verbal parentheticals exhibit ‘stronger’ boundaries that are similar to ι edges. Second, XP parentheticals exhibit ‘weaker’ boundaries that are similar, if not identical to, Φ edges. This dichotomy both supports and contradicts the theories of syntax-prosody mapping. That finite clausal parentheticals are prosodically parsed closer to intonation phrases supports the idea of clause-to-ι mapping. It also supports the assumption that parenthetical structures are prosodically isolated. However, XP parentheticals exhibit the properties closer to phonological phrase-hood (i.e. they exhibit prosodic integration). This indicates either (i) that there is not such a tight mapping between syntax and prosody in Turkish, or (ii) that the mapping constraints are ranked and can, in certain environments, override one another. Mapping appears to be overridden in two ways in Turkish. First, Turkish parentheticals are shown to exhibit instances of both prosodic isolation and integration. This indicates that the prosodic realization of syntactic isolation is not highly ranked. More important is to prosodically mark the clause-hood and phrase-hood of the target syntactic structures. Thus, regardless of whether they are extra-sentential or not, root clauses are parsed as ιs, and structures that do not exhibit the properties of root clause-hood tend to be parsed as Φs. Ranked higher than constituent-to-constituent mapping (and consequently syntactic isolation) is pragmatic relation. If a parenthetical is pragmatically isolated then it is parsed as an ι, regardless of its syntactic type or level of syn-

Mapping to prosody | 325 Are reporting clauses special cases of parentheticals? 327

tactic isolation. This was observed with vocatives16 and interjections, which exhibit longer final syllables, and longer pauses on their right edge. The final syllable of the host part that immediately precedes them is also longer than XP parentheticals in the test set, and Φs in the control set. Final rise values also provide evidence of their ι-hood. While vocatives exhibit the highest magnitude of final rise (H%), interjections exhibited the lowest values of final pitch rise, which is even lower than Φs’ condition. We claim that this is due to a low tone that marks the right edge of the intonation phrase (L%). It is not surprising to observe falling intonation instead of a rising comma intonation in such cases. Since interjections are not related to the content of the discourse, and since they are, in fact, interruptions not only to the host syntactic structures but also to the host discourse structure, they do not necessitate the use of comma intonation, which typically marks a continuation of the ongoing discourse. Therefore, we conclude that pragmatically isolated parentheticals such as clausal interjections and vocatives, the latter of which are XPs, are parsed as ιs regardless of their syntactic properties. In a similar fashion semantic/pragmatic properties of comment clauses override their syntactic properties in mapping to prosody, yielding them to be parsed as Φs and not ιs. In this case, we observe that when a parenthetical presents a speaker’s stand towards the truth of the entire host proposition, then it is prosodically integrated regardless of its syntactic makeup. Güneş (2013) observes that yanılmıyorsam ‘if I am not mistaken’, a clausal parenthetical, is prosodically integrated. This is not surprising considering its function as a comment clause. In fact, based on the syntactic data and their pragmatic behaviour, Griffiths and Güneş (2014) conclude that regardless of their clausal syntax and undominated (parenthetical) syntactic behaviour, comment clauses in Turkish do not exhibit root clausal properties. In this regard, they suggest, these structures should be analysed on par with non-clausal parentheticals. Our findings support the syntactic-pragmatic analysis of these authors. This phenomenon that we observe in Turkish comment clauses is also observed in other languages (cf. Reinhart 1983, Reis 2000, Dehé and Wichmann 2010, Dehé 2014). Another parenthetical type that does not seem to follow the generalization made in this study is nominal appositives. We observe that appositives exhibit stronger isolation cues especially on their left edge. This is not surprising con|| 16 See Göksel and Pöchtrager (2013) for various prosodic realizations of a wider range of vocatives in Turkish. One should note that the types of vocatives investigated by these authors are not the same as the ones that are analysed in this paper. These authors investigate noninterpolating vocatives that convey meanings such as surprise, calling, checking for identity, and so on (ibid.:92).

328 | Güliz Güneş and Çağrı Çöltekin 326 sidering the linear position of appositives and the way they modify their anchors. The appositives that are tested in this study are constituent-modifying parentheticals, which must immediately linearly follow their anchors (Griffiths and Güneş 2014).17 Functionally, appositives provide an alternative referent for their anchor, while syntactically, they and their anchors share the same maximal projection (ibid.). Their linear position and syntactic-semantic similarity with the anchor forces a stronger prosodic boundary in the juncture of the appositive and its anchor, which acts as a parser that separates these syntacticsemantic likes. One can envisage this as a prosodic strategy that ensures that Richard’s (2001) Distinctness Condition on Linearization, or some constraint similar to it, is satisfied. However, the presence of a stronger left edge boundary does not engender total prosodic isolation of appositives, as they do not exhibit ι-level properties on their right edge. That they are parsed as Φs is further supported by Griffiths and Güneş (2014). These authors note that, while prosodically isolated parentheticals such as ki clauses (9b) cannot occupy the nucleus of their hosts, appositives can. Similarly, while parentheticals that are parsed as ιs cannot occupy post-nuclear area of their host-ι, appositives can. In this respect, appositives exhibit the same prosodic properties as their anchors, and as other arguments of their host. Therefore, we consider appositives to be parsed as Φs, but with a more pronounced left edge. Our results show that pauses are employed to mark the edges of both Φ and ι. However, the duration of the pauses displays variation. While verbal parentheticals exhibit similar values to the edges of ιs, XP parentheticals are found to be closer to the pauses surrounding Φs. Note that, in all cases, a pause that precedes a parenthetical is shorter than the pause that follows it. This fact may be analysed in two ways. First, although their prosodic category type matches with their syntax (XP to Φ and clause to ι), parentheticals inherit some properties of their paratactic nature. The resulting prosodic form may be represented as a hybrid prosodic category type, which has slightly stronger boundaries than their already existing correlates, call it “Φpar” for XP parentheticals, and “ιpar” for verbal parentheticals, where the hierarchical order of types is Φ < Φpar < ιpar < ι. Alternatively, instead of postulating a new category type, one can analyse this observation as a case of prosodic recursion, where the prosodic unit created by recursion is marked by a greater degree of prominence than a unit not created by recursion on its right edge. To illustrate the

|| 17 Their correlate structure in English is constituent modifying appositives; i.e. namelyparentheticals. In Turkish, instead of namely an Arabic loan yani is optionally used. For a detailed syntactic analysis of yani parentheticals see Griffiths and Güneş (2014).

Mapping to prosody | 329 327 Are reporting clauses special cases of parentheticals?

latter analysis, we postulate the structure in (15) for verbal parentheticals, interjections and vocatives.18 (15) Utterance ι-non-min ι-min host

ι ι-min

host

verbal-parentheticals vocatives interjections

The structure in (15) is a recursive prosodic structure in which the non-terminal prosodic type ι-non-min displays a ‘more prominent’ right edge. This prominence is realized on the right edge of the parenthetical, which is the rightbranching daughter of ι-non-min. The structural position of the parenthetical in (15) thus explains why pause durations before the parenthetical are shorter than those that occur after the parenthetical: the former marks the edge of an ‘atomic’ prosodic unit, while the latter marks the edge of a prosodic unit built from self-similar units (in this case, the self-similar units are ιs). Longer pauses are not reserved for marking verbal parentheticals. XP parentheticals also exhibit pauses on their right edge, yet these pauses are not as long as the pauses observed in the ι boundary condition, even though they were longer than the pauses that linearly precede XP parentheticals. In fact, the pauses observed after XP parentheticals are longer, but still closer to the duration of the pauses that follow Φ boundaries. Keeping in mind that the XP parentheticals are also similar to the Φ condition in terms of final syllable lengthening, final rise, and pre-parenthetical host-final syllable durations, we claim that sentence-medial XP parentheticals are not the immediate daughters of any ι (neither a ι-non-min, nor a ι-min). For XP-parentheticals, we postulate the structure in (16).

|| 18 ι-non-min refers to any ι that dominates another ι. ι-min refers to any ι that does not dominate another ι. See Elfner (2012) for cross-linguistic evidence that necessitates the use of minimal and non-minimal projections.

330 | Güliz Güneş and Çağrı Çöltekin 328 (16)

ι Φ-non-min Φ-min host

Φ Φ-min

host

XP-par.

Like (15), (16) is a recursive structure. Again, recursion explains the disparate durations of the pauses on each side. The only difference between (15) and (16) is that, in (16), the non-terminal unit created by recursion is a Φ, whereas in (15) it is an ι. Thus, we expect – and do – observe that the right edge of Φ-non-min displays the properties of a Φ-boundary, albeit one that is more prominent than the boundary observed on an ‘atomic’ Φ on its right edge. The argument propounded above accords with the idea of recursive prosodic levels, where the more recursive layers there are, the stronger the boundaries are marked (Kawahara 2012; Itô and Mester 2009, 2012). What is novel about our argument is the claim that recursion is encoded in only pause duration, and that other cues, together with the duration of pauses, are employed only to mark a prosodic unit as a ω or Φ or ι. Note that a non-recursive model cannot explain the data adequately since it would predict equal boundary strength on both edges of any category.

6 Conclusion In conclusion, verbal parentheticals exhibit ι-level properties on both edges. XP parentheticals, on the other hand, exhibit phonological phrase-level properties on both edges. Clausal parentheticals are prosodically isolated, supporting the theories of syntax-prosody mapping, while XP parentheticals are prosodically integrated, partially supporting syntax-prosody mapping theories. The latter result supports theories that assume XP-to-phonological phrase matching, but not those that predict the prosodic isolation with parentheticals. We conclude that Turkish marks constituent-to-constituent matching of syntax and prosody more faithfully than the mapping of syntactic isolation. Additionally, mapping of pragmatic isolation is highly favoured. Specifically, pragmatically isolated parentheticals such as vocatives or interjections are prosodically isolated, regardless of their syntactic make-up. In this sense our findings are on par with the findings of Güneş (2013). We discussed the prosodic structure of Turkish

Mapping to prosody | 329 Are reporting clauses special cases of parentheticals? 331

parentheticals and proposed a representation that favours the recursion of various prosodic category types. Both types of parentheticals are found to exhibit relatively longer pauses on their right edges. To account for the unequal distribution and duration of the pauses, we have proposed a recursive prosodic representation. Overall, our experiment has revealed various aspects of the prosody of parenthetical insertions in Turkish. However, our study is exploratory and leaves several questions for future inquiry. Additionally, the results reported here were based on the controlled productions of four speakers; for a more elaborate account of the parentheticals in Turkish, experiments with more speakers should be carried out, as well as the analyses of spontaneous speech.

References Arnold, Doug. 2007. Non-restrictive relatives are not orphans. Journal of Linguistics 43:2. 272– 309. Asher, Nicholas. 2000. Truth Conditional Discourse Semantics for Parentheticals. Journal of Semantics 17. 31–50. Bates, Douglas, Martin Maechler & Ben Bloker. 2013. lme4: Linear mixed-effects models using S4 classes. R package version 0.999999-2. [online] http://CRAN.Rproject.org/package=lme4 (accessed 27 November 2013). Boersma, Paul & David Weenink. 2011. Praat: doing phonetics by computer. Version 5.3.02. [online] http://www.praat.org/. (accessed 7 November 2011) Bolinger, Dwight. 1989. Intonation and its uses: Melody in grammar and discourse. Stanford: Uni. Press. Burton-Roberts, Noël. 2006. Parentheticals. In K. Brown et al. (eds.). Encyclopaedia of Language and Linguistics. 2nd edn. Amsterdam: Elsevier. 179–182. Corver, Nobert & Craig Thiersch. 2001. Remarks on parentheticals. In Marc van Oostendorp & Elena Anagnostopoulou (eds.). Progress in grammar: Articles at the 20th anniversary of the comparison of grammatical models group in Tilburg. Utrecht: Roquade. Dehé, Nicole. 2007. The relation between syntactic and prosodic parenthesis. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals, Amsterdam John Benjamins. Dehé, Nicole & Kavalova, Y. 2007. Parentheticals: An introduction. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam/Philadelphia: John Benjamins. 1–22. Dehé, Nicole. 2009. Clausal parentheticals, intonational phrasing, and prosodic theory. In Journal of Linguistics 45. 569–615. Dehé, Nicole. 2014. Parentheticals in Spoken English: the syntax prosody relation. Cambridge: C.U.P. Dehé, Nicole & Anne Wichmann. 2010. The multifunctionality of epistemic parentheticals in discourse: prosodic cues to the semantic-pragmatic boundary. In Functions of Language 17 (1). 1–28. de Vries, Mark. 2006. Reported Direct Speech in Dutch. In Linguistics in the Netherlands 23. 212–223.

332 | Güliz Güneş and Çağrı Çöltekin 330 de Vries, Mark. 2007. Invisible constituents? Parentheses as B-merged adverbial phrases. In Nicole Dehé & Yordanka Kavalova (eds.). Parentheticals. Amsterdam/ Philadelphia: John Benjamins. 203–234. de Vries, Mark. 2008. The representation of language within language: a syntactico-pragmatic typology of direct speech. Studia Linguistica 62. 39–77. de Vries, Mark. 2012. Unconventional Mergers. In M. Uribe-Etxebarria, and V.Valmala (eds.). Ways of Structure Building. Oxford: Oxford University Press. 143–166. Elfner, Emily. 2012. Syntax-Prosody Interactions in Irish. University of Massachusetts, Amherst: Doctoral dissertation. Göksel, Aslı & Markus Pöchtrager. 2013. The vocative and its kin: marking function through prosody. In P. Noel Aziz Hanna & B. Sonnenhauser (eds.). Vocative! Addressing between System and Performance. De Gruyter. 87–108. Griffiths, James. this volume. Speaker and quote reduced parenthetical clauses. Griffiths, James & Güliz Güneş. 2014. Ki issues in Turkish. In M. E. Kluck, D. Ott & M. de Vries (eds.). Parenthesis and Ellipsis: Cross-Linguistic and Theoretical Perspectives. Berlin: de Gruyter. Gussenhoven, Carlos. 2004. The phonology of tone and intonation. Cambridge: CUP. Güneş, Güliz. 2013. On the Role of Prosodic Constituency in Turkish. In Proceedings of WAFL8 MIT Working Papers in Linguistics. Cambridge, MA. Güneş, Güliz. 2014. Constraints on syntax-prosody correspondence: the case of clausal and subclausal parentheticals in Turkish. Lingua 150. 278–314. Haegeman, Liliane. 1991/2009. Parenthetical Adverbs: The radical orphanage approach. In B. Shaer, Cook, P., Frey W. & Maienborn, C. (eds.). Dislocated Elements in Discourse. New York: Routledge. 331–347. İpek, Canan & Sun-Ah Jun. 2013. Towards a model of intonational phonology of Turkish: Neutral intonation. In Proceedings of Meetings of Acoustics 19, 060230. The Acoustical Society of America. Itô, Junko & Armin Mester. 2009. The onset of the prosodic word. In S. Parker (ed.). Phonological Argumentation. London: Equinox. Itô, Junko & Armin Mester. 2012. Recursive prosodic phrasing in Japanese. In T. Borowsky, S. Kawahara, T. Shinya & M. Sugahara (eds.). Prosody Matters: Essays in Honor of Elisabeth Selkirk. Sheffield: Equinox. 208–303. Kabak, Barış & Anthi Revithiadou. 2009. An interface approach to prosodic word recursion. In J. Grijzenhout & B. Kabak (eds.). Phonological Domains: Universals and Deviations. Berlin/New York: de Gruyter. Kabak, Barış & Irene Vogel. 2001. The phonological word and stress assignment in Turkish. In Phonology 18. 315–360. Kaltenböck, Gunther. 2009. English comment clauses: position, prosody, and scope. In Arbeiten aus Anglistik und Amerikanistik 34 (1). 49–75. Kamali, Beste. 2011. Topics at the PF interface of Turkish. UK: Harvard University: Doctoral thesis. Kan, Seda. 2009. Prosodic Domains and the syntax-prosody mapping in Turkish. Istanbul: Boğaziçi University: MA thesis. Kawahara, Shigeto. 2012. The intonation of nominal parentheticals in Japanese. In: T. Borowsky, S. Kawahara, T. Shinya and M. Sugahara (eds.). Prosody Matters: Essays in Honor of Elisabeth Selkirk. Sheffield: Equinox. 304–340.

Mapping to prosody | 333 331 Are reporting clauses special cases of parentheticals?

Kluck, Marlies. 2011. Sentence Amalgamation. Netherlands: University of Groningen: PhD thesis. Ladd, D. Robert. 1996. Intonational Phonology. Cambridge: Cambridge University Press. Nespor, Marina & Irene Vogel. 1986. Prosodic phonology. Dordrecht: Foris. Özge, Umut & Cem Bozşahin. 2010. Intonation in the grammar of Turkish. In Lingua 120:132– 175. Peters, Jörg. 2006. Syntactic and prosodic parenthesis. Proceedings of the International Conference on Speech Prosody. 2–5 May 2006. Dresden. Peterson, Peter. 1999. On the boundaries of syntax. In Peter Collins & David Lee (eds.). The Clause in English. Amsterdam: John Benjamins. 229–250. Pierrehumbert, Janet. 1980. The Phonetics and Phonology of English Intonation. MIT dissertation. Potts, Christopher. 2005. The Logic of Conventional Implicatures. Oxford: OUP. R Development Core Team. 1993–2011. R: A language and environment for statistical computing. R Foundation for Statistical Computing. Vienna, Austria. Reinhart, Tanya. 1983. Point of View in Language – The Use of Parentheticals. In G. Rauh (ed.). Essays on Deixis. Tübingen: Müller & Bass. 169–194. Reis, Marga. 2000. Wh-movement and integrated parenthetical constructions. In J.-W. Zwart & A. Werner (eds.). Studies in Comparative Germanic Syntax, Proceedings from the 15th workshop on Comparative Germanic Syntax. Amsterdam: John Benjamins. 3–40 Richards, Norvin. 2001. A distinctness condition on linearization. In K. Megerdoomian & L. A. Bar-el (eds.). Proceedings of WCCFL 20. Somerville, Mass: Cascadilla Press. 470–483. Selkirk, Elisabeth. 1986. On derived domains in sentence phonology. In Phonology Yearbook 3. 371-405. Selkirk, Elisabeth. 2005. Comments on intonational phrasing. In S. Frota, M. Vigario & M. J. Freitas (eds.). Prosodies. Berlin: de Gruyter. 11–58. Selkirk, Elisabeth. 2009. On clause and intonational phrase in Japanese: the syntactic grounding of prosodic constituent structure. Gengo Kenkyu 136. 35–74. Selkirk, Elisabeth. 2011. The syntax-phonology interface. In J. Goldsmith, R. Riggle & A. Yu (eds.). The Handbook of Phonological Theory. 2nd edn. Oxford: Blackwell Publishing. 435– 484. Truckenbrodt, Hubert. 1999. Phonological Phrases: Their Relation to Syntax, Focus, and Prominence. MIT: Ph.D. thesis. van Maastricht, Lieke. 2011. Reporting and Comment Clauses: A cross-linguistic study. University of Groningen: MA thesis. Wagner, Michael. 2005. Prosody and recursion. MIT: Ph.D. thesis. Wichmann, Anne. 2001. Spoken parentheticals. In Karin Aijmer (ed.). A Wealth of English. Studies in honour of Göran Kjellmer. Göteborg: Acta Universitatis Gothoburgensis. 177– 193.

Silvana Abalada and Aida Cardoso

The Vocative in European Portuguese: Prosodic Effects of its Position in the Sentence Abstract: In this paper, we argue that the vocative in European Portuguese shows prosodic effects correlated to its position in the sentence. In order to prove this, we have designed a corpus and analysed a set of parameters, such as acoustic measures (e.g. relative length, intensity, and f0 measures), phrasing, and intonation. Our data shows that the vocative’s position plays a crucial part in its prosodic behaviour, since the vocative has distinctive prosodic characteristics depending on its position. The asymmetry found in the behaviour of vocatives allows us to draw a distinction between initial and non-initial (medial and final) vocatives, since the former share characteristics with isolated vocatives and the latter form a homogeneous group. Moreover, we also considered the possible effects of communicative function in the realization of the vocative and we found that there are, in fact, different pragmatic values associated with initial and non-initial vocatives. 1 Keywords: vocative, position, pragmatic functions, prosody, European Portuguese

|| Silvana Abalada: Center of Linguistics, University of Lisbon. Lisbon, Portugal. Aida Cardoso: Center of Linguistics, University of Lisbon. Lisbon, Portugal.

This paper discusses the prosodic realization of the vocative in European Portuguese (EP) considering different positions in the sentence: initial, medial, and final. Traditionally, the vocative is defined as an element with a calling function that may be uttered with different degrees of emphasis (Cunha & Cintra 2000: 160). From this perspective, the vocative is a syntactic function often associated with an “exclamatory intonation” and described as being “isolated from the sentence” (Cunha & Cintra 2000: 161). In generative grammar, the vocative is defined as a constituent occurring in a peripheral position which does not trig|| 1 We would like to thank Vera Cabarrão, who was a co-author of early versions of the work presented in this paper.

336 | Silvana Abalada and Aida Cardoso 334 ger verbal agreement and which is used in contexts of calling and interpellation (Mateus et al. 2003: 457). Furthermore, constituents with the syntactic function of vocative frequently occur in imperatives, interrogatives, and exclamatory sentences. More recent studies focus on the relation between the vocative and the sentence in order to investigate the relation between syntax and prosody in the realization of vocatives. In this context, the vocative is analysed alongside a set of elements named “extra-sentential”, “peripheral”, or “parenthetical” (Astruc 2003, 2005; Astruc‐Aguilera & Nolan 2007; Dehé 2007, 2009; Peters 2006; Prieto 2002; Wichmann 2000). Considering previous descriptions, our main goal is to verify whether there is a correlation between prosody and the position of the vocative, taking into account a set of parameters, such as acoustic measures (e.g. relative length, intensity, and f0 measures), phrasing, and intonation. Hence, we hypothesize that there are, in fact, prosodic effects of the sentence position in vocatives in EP and that these effects allow us to draw a distinction between initial and noninitial vocatives. Moreover, we will also consider the possible effects of communicative function in the realization of the vocative.

1 Theoretical Background Vocatives have been traditionally considered a type of parenthetical alongside a heterogeneous group of elements, typically including (full or elliptical) clauses, non-restrictive relatives, comment clauses, reporting verbs, nominal appositions, question tags, dislocated elements, and sentence adverbials, among others. Regardless of the theoretical framework, authors generally agree upon the fact that these elements vary in syntactic category, function, and position, as well as in length and complexity. Crucially, the acknowledgment of the heterogeneous nature of parenthetical elements has lead, especially in more recent works (e.g. Astruc 2003, 2005; Astruc-Aguilera & Nolan 2007; Dehé 2007, 2009; Prieto 2002; Wichmann 2000), to a distinctive analysis of specific types of parentheticals, as opposed to a conjoint analysis of all types of parentheticals. Nevertheless, we can still find a set of properties in the studies focussing on a conjoint analysis of different types of parentheticals that are relevant to the study of vocatives. Some of the most important properties include forming an independent tonal unit set off by pauses and having a lower and narrower pitch than the utterance (Armstrong & Ward 1931; Crystal 1969; Nespor & Vogel 1986; Bolinger 1989). Other features include diminished volume, increased tempo,

The Vocative in European Portuguese | 335 337

and rising intonational contours (see Dehé & Kavalova 2007 for an overview). However, some studies have pointed out that we cannot blindly rely on the typical features described above, because the realization of parentheticals may also be affected by length, position, syntactic weight, as well as the communicative intention of the speaker (Bolinger 1989; Dehé 2007, 2009; Peters 2006; Wichmann 2000). For example, Peters (2006) and Dehé (2007, 2009) emphasize that the prosodic realization of parentheticals is affected by their length in terms of the number of syllables, since parentheticals with a lower number of syllables show a strong tendency to be prosodically integrated in the sentence (contrarily to what is shown for longer parentheticals). Crucially, the idea that different variables can influence the prosodic realization of parentheticals can be seen as strong evidence against the traditional idea of a one-to-one relation between syntax and prosody (Nespor & Vogel 1986). Building on this idea, the literature that has analysed the specific properties of vocatives for different languages has taken into account the influence of position and/or the communicative function of the vocative in its prosodic realization. For Catalan, Prieto (2002) shows evidence that vocatives can occur in the initial, medial, and final position of the sentence (what is remarking, however, is that medial vocatives occur in more restrictive contexts). Crucially, the author states that the intonational realization of the vocative depends on its position. In fact, initial vocatives are intonationally independent from the sentence, evidencing a behaviour similar to the one of isolated vocatives: a rising movement over the stressed syllable, which is sustained over the post-tonic syllable, followed by a gradual fall. In contrast, medial and final vocatives tend to have the same intonational pattern of the sentence: a rising movement over the stressed syllable followed by a falling movement2. In spite of these differences, what all vocatives have in common is that they are realized in a pitch lower than the one of the sentence. Regarding isolated vocatives, they are in general described as constructions characterized by a sustained pitch across the syllables, which in turn have higher duration values. Nevertheless, Prieto (2002, 2014) notes that the prosodic realization of the vocative is different according to its specific pragmatic function. Hence, (i) when the speaker has the intention of calling the hearer (greeting), the vocative is characterized by a rising pitch accent on the nuclear syllable and an immediate fall to sustained pitch until the end of the contour, thus being produced with a sustained mid boundary tone (L+H*

|| 2 Please note that this specific intonational contour is described by Prieto (2002) for vocatives in declarative sentences in Catalan.

338 | Silvana Abalada and Aida Cardoso 336 MM%); and (ii) when the speaker has the intention of demanding something from the hearer (insisting call), the vocative is characterized by a rising pitch accent on the nuclear syllable, immediately followed by a gradual and slow rise and then a fall during the post-tonic syllable (L+H* HL%). As the author points out, the insisting call can be realized with some of the intonational contours described for imperatives, since they can share the pragmatic functions of command and request. Lastly, it should be mentioned that the acousticphonetic features of isolated vocatives can evidence different levels of emphasis and insistence (Prieto 2002). As for Castilian Spanish, Estebas-Vilaplana & Prieto (2014) found an intonational contour associated with calling in isolated vocatives that is characterized by a sustained mid pitch (L+H* M%). Similarly to what is described for Catalan, the insisting call, which is characterized by L+H* HL%, differs from the calling contour mainly in the boundary tone. In other studies, in which Catalan is compared with British English (Astruc 2005; Astruc‐Aguilera & Nolan 2007), the authors also argue that the position of vocatives can play a role in their prosodic realization in both languages. In this context, a relevant distinction can be made between initial and non-initial elements. Thus, initial vocatives have independent intonational contours, meaning that the vocative’s intonational contour is not integrated in the intonational contour of the sentence. Nevertheless, the authors found 13% of cases that are an exception to this pattern. As far as medial and final vocatives are concerned, they tend to form intonational units – that can be compared with Pierrehumbert’s intermediate phrase (Pierrehumbert 1980) – and present final rising movements. In spite of the similar behaviour previously described for both languages, final vocatives in Catalan and in British English differ in terms of prominence, since in British English vocatives are mostly deaccented and in Catalan they are accented in 50% of the cases. What is important is that the cases in which vocatives are accented in British English always have the function of getting the hearer’s attention. In fact, it is proposed that there is a close relation between the role of vocatives in the information structure and their prosodic form. For EP, Frota (2014: 74) identifies two types of isolated vocatives, the vocative chant (greeting) and the low vocative chant (insisting call), described as follows: In EP, there are two variants of the calling contour: the sustained pitch variant (the typical vocative chant), characterized by high pitch on the nuclear syllable and a downward step into the first post-tonic syllable after which the pitch level is sustained until the end of the contour; the low pitch variant (low vocative chant), also characterized by high pitch on

337 The Vocative in European Portuguese | 339 the nuclear syllable, immediately followed by a gradual fall until the end of contour. The two variants are pragmatically distinct: the low vocative chant expresses an insisting impatient call that would be pragmatically inadequate if used as an instance of a greeting or first call.

Given the properties explained above, the author proposes the intonational contour (L+)H* !H% for vocative chant (greeting) and (L+)H* L% for low vocative chant (insisting call), pointing out that the main difference between the two types of vocatives in EP relies on the boundary tone, similarly to what has been described for Catalan (Prieto 2002, 2014) and Castilian Spanish (EstebasVilapana & Prieto, 2010). Hence, while the low vocative chant evidences a final low boundary, as many other types of sentences in EP, the vocative chant presents a more distinctive boundary tone. Besides that, the vocative chant has one more distinctive property: the lengthening of the nuclear syllable and, especially, of the boundary syllable. On this matter, the higher duration values of this type of vocative also allow the split-up of the nuclear syllable in a specific context, namely when no post-tonic syllable is available. For the same reason, the lengthening found in the boundary syllable blocks post-tonic phonetic vowel reduction as well as vowel deletion. Another important aspect related to vocatives in EP is their pragmatic function, as signalled for other languages, namely for Catalan, British English, and Castilian Spanish. As can be seen in the excerpt above, Frota (2014) attributes distinctive pragmatic values to the two types of isolated vocatives (greeting and insisting call), which are not only not interchangeable, but also play a crucial role in their respective prosodic realization. In this respect, it is relevant that vocatives in EP can also be associated with pragmatic values of command and request. Frota (2014) makes a distinctive analysis for requests and commands with one word and with more than one word. In the case of the former, the author associates %H L* L% to requests and H*+L L% to commands. As for the latter, Frota (2014) associates H* L* L% with requests and H*+L L% and L*+H L% with commands. Finally, it is worth noting that the pragmatic values of command and request are often associated with imperatives. This is the case in Catalan (Prieto 2002, 2014) and EP (Falé & Faria 2007; Frota 2014). Even though imperatives are not extensively studied in EP, Falé & Faria (2007) focus on their phonetic features and describe them as having an initial rise followed by a falling movement of large amplitude, as well as having higher values of f0 (particularly on the last pre-stressed vowel and on the last stressed vowel) than the ones found in declaratives. Apart from this description, the authors also make a distinction, once

340 | Silvana Abalada and Aida Cardoso 338 again, between imperatives with the value of command and with the value of request, since commands have higher values of f0 than requests.

2 Methodology Our data consists of a specifically designed corpus of 108 sentences (54 targetsentences and 54 distractors), which was read twice by two informants for a total of 432 readings. Both informants are female (CA and TA) native speakers of standard EP, with ages between 25 and 30 years old, and with no background in linguistics. As for the design of the corpus, we considered for the target-sentences: (i) the number of syllables of the sentence, (ii) the number of syllables and the stressed syllable of the vocative, and (iii) the position of the vocative in the sentence. Every possible value of each variable has an equal number of items in the corpus. As for the number of syllables of the sentence, we included sentences with five (cf. (1)), seven (cf. (2)), and ten (cf. (3)) syllables in the corpus. (1)

Vê os animais. Look at (imperative-2sg) the animals ‘Look at the animals.’

(2) Distribui os gelados. Hand out (imperative-2sg) the ice creams ‘Hand out the ice creams.’ (3) Não te esqueças de ligar o rádio. No you (clitic-2sg) forget (imperative-2sg) to turn on the radio ‘Don’t forget to turn on the radio.’

Concerning the number of syllables of the vocative, we selected proper nouns with one (“Gil”), two (“João” and “Mário”), and three (“Valdemar”, “Maria”, and “Mónica”) syllables. With regard to the position of the stressed syllable of the vocative, we selected proper nouns stressed on the last (“Gil”, “João”, and “Valdemar”), the penultimate (“Mário” and “Maria”), and the antepenultimate (“Mónica”) syllables. These variables would allow us to observe whether there is a correlation between the number of syllables and the position of the stressed syllable of the vocative and its prosodic realization.

The Vocative in European Portuguese | 339 341

Lastly, as for the position of the vocative in the sentence, we considered the initial (cf. (4)), the medial (cf. (5)), and the final (cf. (6)) position. On this matter, we define medial vocative as an interpolated constituent with the syntactic function of vocative that occurs between the verb and its internal argument. Nevertheless, sentences with medial vocatives are not widely accepted by all native speakers. However, we considered this type of sentences in our analysis, firstly because it is accepted by a large group of speakers and secondly because even speakers with a more restricted grammar accept such distribution in sentences that evidence syntactic weight effects3. Note that Prieto (2002) has already pointed out that medial vocatives occur in more restrictive contexts in Catalan. (4) Maria, distribui os gelados. Maria (vocative) hand out (imperative-2sg) the ice creams ‘Maria, hand out the ice creams.’ (5) Distribui, Maria, os gelados. Hand out (imperative-2sg) Maria (vocative) the ice creams ‘Hand out, Maria, the ice creams.’ (6) Distribui os gelados, Maria. Hand out (imperative-2sg) the ice creams Maria (vocative) ‘Hand out the ice creams, Maria.’

What should also be mentioned is that all target-sentences of the corpus were built as imperative sentences. The motivation behind this decision was twofold: firstly, it would allow us to analyse the prosodic realization of vocatives in a more restrictive set of pragmatic values; secondly, it would avoid the influence of any other factors besides the position of the vocative. Regarding the distractors, we designed neutral declarative sentences without a marked word order and without focalized constituents. Additionally, we ensured that these sentences had the same number of syllables of the targetsentences and we took the lexical material selected for the vocatives (e.g. proper

|| 3 For this reason, we find that, in our corpus, speakers show a more positive response towards a sentence like “Não te esqueças, Maria, de ligar o radio.” (‘Don’t forget, Maria, to turn on the radio.’) than towards a sentence like “Vê, Maria, os animais.” (‘Look, Maria, at the animals.’).

342 | Silvana Abalada and Aida Cardoso 340 nouns) and used it in the distractor sentences as subjects, in order to neutralize any obvious differences between target-sentences and distractors. The data collection was carried out with the help of an elicited reading task motivated by a situational context4, following the methodology applied in the Atles interactiu de l’entonació del català (Prieto & Cabré 2008). The readings were recorded as wave files, with 16 bit mono at 16 kHz sampling rate, using a Rode NT3 microphone, a Soundcraft M4 mixer, an Apple Macbook Pro, and the recording software Apple Logic Pro 9. Concerning the prosodic annotation, each vocative present in the targetsentences (216 readings by both informants) was prosodically analysed in Praat, version 5.2.10. (Boersma & Weenink 2009), following the conventions defined in Towards a P_ToBI (Viana & Frota 2007), in the autosegmental framework. Hence, we annotated a set of acoustic measures, as well as break indices and intonational contours. The data annotation was performed by three researchers and those cases of tonal and break indices assignment in which the researchers did not agree were marked and subjected to discussion, after which a final decision was taken. As far as the acoustic measures are concerned, we extracted from each vocative: (i) the relative length (in seconds), (ii) the maximum of intensity (in decibels), and (iii) the minimum, maximum, and range of f0 (in semitones). While the length and f0 measures were manually extracted using Praat, the maximum of intensity was automatically extracted with a script5. It should also be mentioned that (i) the relative length of the vocative is the quotient of the vocative length by the sentence length and that (ii) the range of f0 is the difference between the maximum and the minimum of f0. Regarding the break indices, we annotated the perceived degree of boundary strength following the inventory of MAE_ToBI (Beckman et al. 2005). Hence, we considered break indices of levels “0 (very close inter-word juncture)”, “1 (ordinary phrase-internal word end)”, “3 (intermediate phrase end, with phrase accent)”, and “4 (intonational phrase end, with boundary tone)” (Beckman et al. 2005: 23). However, the literature states that in EP there are no intermediate || 4 The elicitation context presented to the informants was the following: “Imagine que vai fazer uma audição para uma série televisiva, cujo enredo envolve um grupo de amigos. Nessa audição, o director de casting irá pedir lhe que leia um conjunto de deixas da forma mais natural possível.” (‘Imagine that you are auditioning for a TV series. The plot of the show involves a group of friends. During the audition, the cast director will ask you to read a set of cues as naturally as possible’). 5 We used the script “get-intensity” created by Shigeto Kawahara (http://www.rci.rutgers.edu/~kawahara/scripts/get_ intensity_minmax.praat).

341 The Vocative in European Portuguese | 343 phrases, but minor intonational phrases instead, defined as having a narrower pitch range and a final lengthening not as big as the major intonational phrase’s boundary (Frota 2000; Viana & Frota 2007). Accordingly, break indices of level 3 indicate a minor intonational phrase end in EP (cf. Viana & Frota 2007). Therefore, we follow Towards a P_ToBI (Viana & Frota 2007) on this matter by using the conventions related to the marking of minor and major intonational phrases with 3 and 4, respectively, and the symbols “-” and “%” in the tone tier to code the different boundary strengths. On the other hand, concerning intonation, we annotated each vocative by adopting the tonal inventory described for EP, in works such as Frota (2000, 2014) and Viana & Frota (2007), and taking into account that the nuclear contours described for EP are, in general, equally found in minor and major intonational phrases (Viana & Frota 2007). Lastly, it is worth mentioning that the data collected from the prosodic annotation was statistically analysed using SPSS (Statistical Package for the Social Sciences), version 18.0.0. Concerning acoustic measures, mean values were calculated and Mann-Whitney (U) (for two independent samples) and KruskalWallis (H) (for more than two independent samples) tests were performed. The option for non-parametric tests (instead of reporting the results of a variance analysis) was due to the fact that we tested for normality and performed an ANOVA, but we didn’t have homogeneity of variance. We also performed a Bonferroni correction on data tested with Kruskal-Wallis, since this correction allows us to control the overall Type I error rate when multiple significance tests are carried out (Field 2009). In the case of break indices and intonational contours, crosstabs and Chi-square (χ2) tests were performed. On this matter, we assume that the Chi-square test is the best way to test categorical data (Field 2009). Since our sample is relatively small, we also took into account the Fisher’s exact test, in order to get a more accurate result. For this reason, we chose to report the results of both Pearson Chi-square and Fisher’s exact test.

3 Data Analysis Recovering our initial hypothesis, our data shows a correlation between prosody and the position of the vocative, since vocatives in different positions have distinctive prosodic features. First of all, an aspect that stands out regarding acoustic measures is that initial vocatives consistently stand apart from medial and final vocatives, namely by having the higher mean values in many of the analysed acoustic parameters.

344 | Silvana Abalada and Aida Cardoso 342 As can be seen in table 1, they present higher levels of intensity, as well as of f0 (maximum and range). Significantly, initial vocatives differ considerably from the others in almost every acoustic feature. As far as the maximum of intensity is concerned, we found that initial vocatives are statistically different from medial (U = 1923.5, p = .008) and final (U = 89, p < .001) vocatives. The same can be stated for f0 measures. In fact, the statistical analysis shows significant differences (i) for maximum of f0 between initial and medial vocatives (U = 1877, p = .004), and initial and final vocatives (U = 221.5, p < .001); (ii) for range of f0 between initial and medial vocatives (U = 1716, p < .001), and initial and final vocatives (U = 1971, p = .013). Although initial vocatives do not present the higher values of relative length and of minimum of f0, they still show significant differences for both parameters. As far as relative length is concerned, we found differences between initial and medial vocatives (U = 1492.5, p < .001), and initial and final vocatives (U = 1552, p < .001). For minimum of f0, the statistical analysis revealed differences only between initial and final vocatives (U = 219, p < .001).

Relative Length (sec)

Initial Vocative

Medial Vocative

Final Vocative

0.263

0.22

0.311

84.238

79.721

Maximum of Intensity (dB) 85.158 Maximum of f0 (ST)

10.175

9.044

5.358

Minimum of f0 (ST)

5.281

5.378

1.498

Range of f0 (ST)

4.894

3.666

3.863

Table 1: Mean values of the acoustic measures of the vocative by its position.

These results can be interpreted in the light of cross-linguistic theoretical claims that attribute a specific communicative function to vocatives (especially isolated and initial vocatives), namely a greeting or insisting calling function that can also be related to requests and commands (Prieto 2002, 2014; Astruc 2003, 2005; Astruc-Aguilera & Nolan 2007; Estebas-Vilaplana & Prieto, 2014; Frota, 2014). On this matter, the higher levels of intensity found in our initial vocatives reflect the presence of emphasis in the sense of Ladd (2008)6. This means that, || 6 For Ladd (2008), phonetic emphasis is defined as a paralinguistic device that can modify the realization of accent patterns in gradients. Although Ladd (2008) relates emphasis to the disambiguation of focus accent patterns, we extend this idea here to a gradient and phonetic way of marking pragmatic values in vocatives. Importantly, the author states that emphasis usually implies higher values of a range of f0 and intensity, among other acoustic correlates.

The Vocative in European Portuguese | 343 345

in these specific contexts, intensity functions as a gradient feature that contributes to the marking of the pragmatic values of greeting, insisting call, and command conveyed by vocatives. Related to the issue of emphasis, Prieto (2002) has already pointed out that this is a relevant feature in the realization and perception of vocatives, since the acoustic-phonetic features of the isolated vocatives can evidence different levels of emphasis and insistence. On the other hand, the higher levels of f0 that distinguish initial vocatives from medial and final vocatives may reflect a specificity found in imperatives (and particularly in commands) in EP by Falé & Faria (2007). Note that the vocatives in our corpus always occur in imperative sentences, which may, perhaps, have influenced the production of vocatives, and particularly initial vocatives, as commands. Regarding phrasing, the initial vocatives show a higher tendency to form an independent tonal unit (cf. table 2). Crucially, and even though they form minor intonational phrases in 59.7%, they are the only ones constituting major intonational phrases (34.7%). Such behaviour indicates that initial vocatives tend to phrase in a more independent way in relation to the sentence, when compared with the behaviour of the other vocatives, since the initial vocative forms either a minor or a major intonational phrase in 94.4% of the cases. Nevertheless, in 5.6% of cases, vocatives are integrated in the intonational phrase of the sentence. The distinct behaviour of the initial vocatives reflects significant differences between initial and medial (right boundary) vocatives (χ2 (3) = 66.515, p < .001; Fisher’s exact test (two-tailed) = 74.402, p < .001), and initial and final vocatives (χ2 (3) = 45.200, p < .001; Fisher’s exact test (two-tailed) = 53.302, p < .001). Break Indices Initial Vocative Medial

Left

Vocative

Boundary Right

0

1

3

4

-

5.6% (N = 4)

59.7% (N = 43)

34.7% (N = 25)

9.7% (N = 7)

9.7% (N = 7)

79.2% (N = 57)

1.4% (N = 1)

29.2% (N = 21)

40.3% (N = 29) 26.4% (N = 19)

4.2% (N = 3)

18.1% (N = 13)

22.2% (N = 16) 59.7% (N = 43)

-

Boundary Final Vocative

Table 2: Break indices by position of the vocative.

Concerning intonation, and since vocatives have been associated in the literature with specific intonational contours, we analysed the distribution of intonational contours of the vocative by its position in the sentence. In this regard,

346 | Silvana Abalada and Aida Cardoso 344 once again the results indicate statistical differences between initial and medial vocatives (χ2 (7) = 88.305, p < .001; Fisher’s exact test (two-tailed) = 104.955, p < .001), and initial and final vocatives (χ2 (5) = 96.880, p < .001; Fisher’s exact test (two-tailed) = 117.485, p < .001). Furthermore, as table 3 shows, there are some intonational contours that only occur in the initial vocative: vocative chant (greeting) – (L+)H* !H- / (L+)H* !H% –, low vocative chant (insisting call) – (L+)H* L- / (L+)H* L% –, and one word command – H*+L L- / H*+L L%. Intonational Contours H*/H* (L+)H* !H-/ (L+)H* L- / H*+L L- / H+L*/H+L* L*/ L* L-/ H-/H* (L+)H* !H% (L+)H* L% H*+L L% L-/ H+L* H% Initial

15.3%

Vocative (N = 11) Medial

13.9%

L*+H

L* L%

L+H*/ DeL+H* H- accented

L% 34.7%

16.7%

6.9%

26.4%

(N = 25)

(N = 12)

(N = 5)

(N = 19)

-

-

-

Vocative (N =

-

-

-

-

22.2%

43.1%

13.9%

13.9%

-

(N = 16)

(N = 31)

(N = 10) (N = 10)

10) Final Vocative

-

-

-

-

37.5%

25.0%

(N = 27)

(N = 18)

-

-

37.5% (N = 27)

Table 3: Intonational contours by position of the vocative.

The crucial point concerning the intonation of the initial vocatives is that we found the intonational contours described for isolated vocatives in EP by Frota (2014). Note that Prieto (2002) has already described a similar situation for Catalan, where initial vocatives present the same intonational contours as isolated vocatives. In fact, we found the intonational contours associated in EP with vocative chant (greeting): (L+)H* !H% (cf. figure 1), as well as with low vocative chant (insisting call): (L+)H* L% (cf. figure 2) (Frota, 2014). Furthermore, we also found a split-up of the nuclear syllable in 22% of the 36 possible contexts examined for this phenomenon – i.e. in vocatives where no post-tonic syllable is available (Frota, 2014) –, showing that the position of the stressed syllable influences the realization of the vocative. Interestingly, this phenomenon is present in our corpus regardless of the number of syllables (one, two, or three) of the vocative. In spite of having a low percentage (6.9%), the intonational contour that can be associated with one word command (H*+L L%) in EP (Frota 2014) occurs in our data. Thus, and although in a less significant way than with acoustic measures, intonational contours also seem to be indicators of the close relation between vocatives (particularly in imperatives) and a command value.

345 The The Vocative Vocative in in European European Portuguese Portuguese | | 347 347

Fig. Fig. 1: 1: Initial Initial vocative vocative forming forming aa major major intonational intonational phrase, phrase, with with an an intonational intonational contour contour assoassociated ciated with with vocative vocative chant chant (greeting), (greeting), and and with with aa split-up split-up of of the the nuclear nuclear syllable syllable (‘João, (‘João, hand hand out out the the ice ice creams.’). creams.’).

Fig.2: Fig.2: Initial Initial vocative vocative forming forming aa minor minor intonational intonational phrase phrase and and with with an an intonational intonational contour contour associated associated with with low low vocative vocative chant chant (insisting (insisting call) call) (‘Mário, (‘Mário, look look at at the the animals.’). animals.’).

348 | Silvana Abalada and Aida Cardoso 346 Focusing our attention on medial vocatives, it becomes very apparent that they do not exhibit a unified behaviour, since they do not show the same pattern across all acoustic parameters. For instance, they present intermediate values with maximal intensity and f0 when compared with initial and final vocatives on the one hand, and they present the lowest values in the case of f0 range on the other hand (cf. table 1). This last aspect can be related to the narrower pitch range feature that is frequently associated with interpolated parentheticals in the traditional literature (Armstrong & Ward 1931; Crystal 1969; Nespor & Vogel 1986; Bolinger 1989). Although medial vocatives present significant differences to initial and final vocatives in almost all the parameters, it is significant that this is not the case for the range of f0, where medial and final vocatives do not show significant differences (U = 2134.5, p = .068). Indeed, this finding points to a closer relation between medial and final vocatives. Contrarily to initial vocatives, in what regards phrasing, medial vocatives do not show such a strong tendency to form an independent tonal unit, as proved by the absence of major intonational phrases in this position (cf. table 2). This means that break indices of level 4 were never attested both on the left and on the right boundaries of the same vocative7. The fact that medial vocatives never form a major intonational phrase can be interpreted as a cue to the tendency of medial vocatives to be integrated in the compound IP-domain of the sentence, in the account by Frota (2000, 2014). These findings suggest that our data is not in total agreement with the unified vision of parentheticals that can be found in more traditional works (e.g. Nespor & Vogel 1986). Additionally, our data reveals a phrasing asymmetry between left and right boundaries of the medial vocative, since there is a higher percentage of break indices of 3 and 4 on the left boundary (79.2% and 1.4%, respectively, for a total of 80.6%) than on the right one (26.4% and 4.2%, respectively, for a total of 30.6%). Crucially, this asymmetry is statistically significant (χ2 (3) = 40.444, p < .001; Fisher’s exact test (two-tailed) = 41.597, p < .001) and suggests that medial vocatives tend to be more associated with the material occurring at the end of the sentence. The intonational contours found in medial vocatives show, once again, a more diversified behaviour than the ones found in initial and final vocatives, since they present low (L*/ L* L- / L* L%), high (H* / H* H- / H* H%), and rising (L+H*/ L+H* H- and L*+H) pitch accents, as can be seen in table 3. As Wichmann (2000) has already reported, the same type of behaviour also characterizes interpolated constituents in British English. It is worth noting that there is a

|| 7 Additionally, cases in which a vocative presents break indices of level 3 on one edge of the vocative and 4 on the other are also rare (only 2) cases of a total of 72 medial vocatives).

347 The The Vocative Vocative in in European European Portuguese Portuguese | | 349 higher frequency of low pitch accents (43.1%) in medial vocatives which can be related to the lower and narrower pitch range revealed by the acoustic analysis (cf. figure 3). In spite of their lower frequency, rising pitch accents are important in the characterization of interpolated vocatives, because in general they suggest continuity and allow to draw parallels with interpolated parentheticals in EP (Frota 2000, 2014).

Fig. Fig. 3: 3: Medial Medial vocative vocative with with aa low low pitch pitch accent accent and and aa lower lower and and narrower narrower pitch pitch range range than than the the sentence sentence (‘Hand (‘Hand out, out, João, João, the the ice ice creams.’). creams.’).

As for final vocatives, they unsurprisingly present the lowest values in almost all the acoustic measures, with the exception of f00 range (Mean = 3.863) and length (Mean = 0.311), as shown in table 1. Nevertheless, regarding the range of f00, final vocatives do not statistically differ from medial vocatives (U = 2134.5, p = .068), which present the lowest values for this parameter, as mentioned before. The higher values of length found in final vocatives were expected, since they are a consequence of the well-documented final lengthening preceding an intonational phrase boundary (see Frota (2000), for EP). Moreover, the lowest values of f00 maximum (Mean = 5.358) and minimum (Mean = 1.498) can, to some extent, be related to an aspect reported for Catalan by Prieto (2002), i.e. the fact that although both medial and final parenthetical elements evidence a lower pitch than the sentence, the latter have a lower pitch than the former.

350 | Silvana Abalada and Aida Cardoso 348 In terms of phrasing, it is important to say that there are no break indices of level 4 preceding the final vocative. This means that final vocatives never form a major intonational phrase and in this respect behave similarly as medial vocatives. Once again, our data seems to be in line with the literature about noninitial parenthetical elements in the sense that they do not necessarily form an independent phrase (Astruc 2003, 2005; Astruc-Aguilera & Nolan 2007). Moreover, our data also shows a balance between total integration of the final vocative in the sentence (18.1% occurrences of break indices of level 0 and 22.2% of level 1, for a total of 40.3%) and the formation of minor intonational phrases (59.7%). In terms of intonation, we can conclude that final vocatives are characterized by a unified behaviour, since we only find low (L* L%) and falling (H+L* L%) intonational contours associated with this position. Crucially, this fact is similar to what was reported for medial vocatives, but presents a contrast to what was described for initial vocatives in the sense that the medial and final vocatives are realized with a high percentage of low intonational contours, whereas initial vocatives are not realized with low intonational contours at all. Moreover, the final vocative is the only one to present a relevant percentage of deaccented realizations (37.5%). This observation is in line with Bolinger’s (1989) claim that final parentheticals may be deaccented. What is more, almost all final vocatives that are integrated in the sentence are deaccented (from the 29 cases of integrated final vocatives, 27 are deaccented, amounting to a percentage of 93.1%) (cf. figure 4). In spite of the relation between vocative integration and deaccentuation, 62.5% of the totality of final vocatives are accented, which means that, regarding prominence, final vocatives in EP are closer to what is described for vocatives in Catalan, but differ more from the results found for vocatives in British English (Astruc 2005; Astruc‐Aguilera & Nolan 2007).

The Vocative Vocative in in European European Portuguese Portuguese | | 349 351 The 351

Fig. 4: 4: Deaccented Deaccented and and integrated integrated final final vocative vocative (‘Don’t (‘Don’t forget forget to to turn turn on on the the radio, radio, Maria.’). Maria.’). Fig.

Additionally, we we also also considered considered it it relevant relevant to to test test the the effect effect of of the the number number of of Additionally, syllables of of the the vocative vocative on on the the break break indices. indices. On On this this matter, matter, the the data data shows shows aa syllables correlation between between the the number number of of syllables syllables and and the the break break indices, indices, since since signifisignificorrelation 22 (3) = 11.827, p cant differences between vocatives with one and three syllables (χ cant differences between vocatives with one and three syllables (χ (3) = 11.827, p = .008; .008; Fisher’s Fisher’s exact exact test test (two-tailed) (two-tailed) == 10.305, 10.305, pp = = .014) .014) and and between between vocatives vocatives = 2 2 (3) == 23.088, 23.088, pp