The Prosody-Syntax Interface in Catalan

Transcription

1 The Prosody-Syntax Interface in Catalan Dissertation eingereicht bei der Humanwissenschaftlichen Fakultät der Universität Potsdam (Germany) Ingo Feldhausen (ingo.feldhausen AT gmx DOT de) June 2008

2 Abstract The dissertation presents a theoretical and experimental investigation of the prosodysyntax interface in complex sentence structures (preverbal subjects, object clauses and (embedded) clitic left- and right-dislocations) in Catalan. It presents new findings based on experimental data and proposes an analysis based on stochastic OT (Boersma & Hayes 2001) which accounts for the variation in the intonational grouping of the complex structures examined. The empirical data in the present work is completely new as it investigates the prosody of structures which prior work has never looked at. Four intonational experiments were conducted. The first on simple SVO structures (DP object), the second on complex SVO structures (sentential object). The third experiment is concerned with clitic left- and right-dislocations. The fourth experiment compares preverbal subjects and left-dislocations. These experiments show several interesting results that are closely connected. It is shown that phrasing tendencies change with structural differences. The most common phrasing in simple SVO structures is (S)(VO). However, (SV)(O) is possible if the object is branching. Moreover, the number of (SV)(O) phrasings increases significantly if the object is not only branching but also sentential. Sentential objects intensify the effect of branching objects. Object clauses typically are separated from the matrix clause by a prosodic break. No break, however, precedes the object clause when it contains a left-dislocated element. An embedded left-dislocation phrases with the matrix clause, while the remainder of the embedded clause is parsed in a separate prosodic phrase, i.e. (Matrix clause + emb. left-dislocation)(emb. clause). The phrasing cue for left-dislocations thus consists only of a strong right boundary. This constitutes empirical counterevidence to the claim that dislocations are minimally and exhaustively contained in a prosodic phrase (cf. Frascarelli 2000:63). Left-dislocations in general phrase with preceding material in Catalan. Clitic rightdislocations are almost always separated from the preceding main clause by a prosodic boundary. When discussing preverbal subjects and left-dislocations in one study, the status of preverbal subjects must be considered. It is argued that preverbal subjects are not automatically leftdislocated. The experimental data show that non-given preverbal subjects can phrase with following material, while given subjects show the obligatory right boundary typical of leftdislocations. A comprehensive theoretical approach must account for the variation in the data. For this reason, the analysis here is grounded in the framework of Stochastic Optimality Theory (Boersma & Hayes 2001), a variant of Optimality Theory (Prince & Smolensky 1993/2004). A nonprobabilistic constraint ranking cannot account for the variation. The analysis includes three important constraints, which reflect the prosody-syntax interface: 1) ALIGN-CP,L, a constraint which accounts for the boundary preceding the embedded clause; 2) ALIGN-TOP(IC),R, a constraint which accounts for the obligatory right boundary of dislocations (i.e. topics); and 3) ALIGN-vP,R, a constraint which aligns the right edge of vp to the right edge of a prosodic phrase. This constraint, together with the syntactic analysis of clitic right-dislocations adopted in this thesis, accounts for the intonational break between a sentence and a right-dislocation. It is argued that the syntactic position of right-dislocations is within the main clause (i.e. below TP; Cecchetto 1999; Villalba 1999a,b, 2000). The argumentation is based on an entirely new test dealing with obviation effects (i.e. subjunctive disjoint reference effects) and on two traditional tests (namely licensing of NPIs, and binding properties). Assuming remnant movement of vp to a clause-internal FocP (along the lines of Villalba 2000) accounts for the surface order. ii

3 Erklärung (Declaration in German) Ich erkläre hiermit, dass ich an keiner anderen Hochschule ein Promotionsverfahren eröffnet habe. Zudem erkläre ich hiermit, dass die Arbeit selbständig und ohne unzulässige Hilfe Dritter verfasst wurde und bei der Abfassung nur die in der Dissertation angegebenen Hilfsmittel benutzt sowie alle wörtlich oder inhaltlich übernommenen Stellen als solche gekennzeichnet wurden. Des Weiteren erkläre ich, dass die Dissertation in der gegenwärtigen oder einer anderen Fassung bei keiner anderen Fakultät einer wissenschaftlichen Hochschule zur Begutachtung im Rahmen eines Promotionsverfahrens vorgelegen hat. (Ingo Feldhausen) iii

4 Danksagung (Acknowledgements in German) Diese Arbeit wäre nie ohne das Engagement und die Unterstützung von Gisbert Fanselow und Kerstin Blume begonnen worden. Dank ihnen hatte ich die wunderbare und außerordentliche Chance in Potsdam zu promovieren und die linguistische Welt Berlins zu betreten. Gisbert Fanselow, mein ursprünglicher Betreuer, hat weit im Voraus und lange vor mir erkannt, dass die Dissertation primär prosodischer Natur sein würde. Ich begann in Potsdam, um ursprünglich eine Arbeit anzufertigen, bei der es um die Interaktion von Syntax und Informationsstruktur bei katalanischen Linksdislokationen geht. Nun, etwas Syntax und Informationsstruktur ist in der vorliegenden Arbeit noch zu finden. Ich danke ihm dafür, mich mit meiner Arbeit den Weg gehen gelassen zu haben, der sich im Laufe der Zeit herauskristallisiert hat. Zudem danke ich ihm dafür, meinen Fokus auf meine beiden jetzigen Betreuerinnen gerichtet zu haben: Caroline Féry und Pilar Prieto. Caroline Féry war seit meiner Ankunft an der Universität Potsdam an meiner Arbeit interessiert und sie war zuerst meine Zweitbetreuerin. Als ich sie nach der Trennung von Gisbert Fanselow fragte, ob sie meine Erstbetreuerin werden möchte, willigte sie umgehend und ohne zu zögern ein. Ihre Kommentare zu ersten schriftlichen Versionen der Arbeit waren hart. Aber genau das war nötig, um meine Arbeit weiter voranzubringen. Ihre Kritik war konstruktiv. Ihre Kommentare waren sachlich genau und konkret. Sie zwangen mich klarer zu werden, die Grundlagen der Arbeit viel deutlicher herauszustellen und die Analyse grundlegend zu modifizieren. Dafür möchte ich ihr zutiefst danken. Ebenso dankbar bin ich ihr dafür, meine Fragen immer umgehend beantwortet und meine Texte immer unverzüglich gelesen zu haben. Pilar Prieto lernte ich während meines Forschungsaufenthaltes in Barcelona des Jahres 2005 kennen. Sie empfing mich mit offenen Armen und unterstützte mich auf meinem Weg in die Welt der Intonation im Allgemeinen sowie des Katalanischen im Spezifischen. Ich vergesse nicht, wie oft sie mir den Unterschied zwischen Betonung und Akzent erklärt hat. Sie versorgte mich zudem regelmäßig mit Literatur. Ihre Kommentare zu diversen Kapiteln meiner Dissertation sowie zu meinem ersten Manuskript über die Prosodie katalanischer Dislokationen verhalfen mir zu erkennen, was genau das Besondere an meinen Ergebnissen ist und wie ich sie zu kommunizieren habe. Ich möchte Pilar Prieto zutiefst dafür danken, die Zweitbetreuung meiner Dissertation übernommen zu haben. Sie ist einfach die Expertin, wenn es um katalanische Intonation geht. Shinishiro Ishihara ist mein inoffizieller Betreuer. Als Kollege in Potsdam hat er von Anfang an jede einzelne Entwicklungsstufe meiner Arbeit mitbekommen und miterlebt. Er hat sich durchweg und immer geduldig meine Zweifel und Klagen angehört, wenn es so schien, als ob nichts funktionieren würde. Es machte ihm nichts aus, mit einem verzweifelten Ingo, der gerade in Katalonien war, 90 Minuten am Telefon zu verbringen, um mit ihm die nächsten Schritte seines Dissertationsprojektes zu besprechen. Während meiner gesamten Promotionsphase war er für mich da und versorgte mich stets mit praktischen und konkreten Ratschlägen, wenn ich sie brauchte. Ich bin mehr als dankbar, dass er im Laufe der Zeit ein Freund wurde. Ich bin mir auch ziemlich sicher, dass er nach all der Zeit mehr über katalanische Intonation weiß als ich. Ich weiß nicht, wie ich ihm je für alles danken kann. Xavier Villalba war mein Mentor in Barcelona und ermöglichte mir den Forschungsaufenthalt an der Universitat Autònoma de Barcelona (Katalonien, Spanien). Von Anfang an unterstütze und motivierte er mich. Er war immer bereit mir mit Grammatikalitätsurteilen und aufschlussreichen Kommentaren beiseite zu stehen. Zudem half er mir, die Daten für mein Dislokations-Experiment zu eruieren. Ich hatte das große Glück seine iv

5 wunderbare Familie kennenlernen zu können und ich möchte ihnen allen danken, dass ihre Haustür für mich offen steht. Eine einzige Sache stimmte mich während der Zeit allerdings traurig: das war die kleine Maus in der Hölle. Christoph Gabriel gilt mein ganz besonderer Dank dafür, dass er mich angetrieben hat, endlich die Arbeit niederzuschreiben. Bei jedem Treffen, bei jedem Telefonat sollte ich über mein Vorankommen berichten. Natürlich musste ich ab und zu einige Kleinigkeiten schöner färben als sie waren (er möge mir dafür verzeihen) allerdings ist die Arbeit jetzt wirklich fertig. Zudem möchte ich ihm dafür danken, dass er eben nicht nur regelmäßig nachgehakt hat, sondern mir zugleich den Weg frei gehalten hat, so dass ich auch wirklich die Arbeit schreiben konnte. Ich weiß das sehr zu schätzen. Ariadna Benet hat unglaubliche Arbeit geleistet. Sie hat mir nicht nur bei der Konstruktion meiner Sätze für drei Experimente geholfen, sondern auch alle Kontextfragen dafür eingesprochen. Regelmäßig kam ich zu ihr mit einer Frage zum Katalanischen, regelmäßig musste sie irgendwelche Änderungen kontrollieren. Sie hat sich kein einziges Mal beschwert. Ich bin ihr für ihre Arbeit sehr dankbar. Laura J. Downing und Hubert Truckenbrodt ließen mich jeweils beide an ihrem immensen linguistischen Wissen und ihrer großen linguistischen Erfahrung teilhaben. Sie haben mir immer wieder geholfen, Sachen klarer zu sehen. Gerade in den letzten Monaten der Arbeit hatte ich das Glück, mit ihnen oft diskutieren zu können. Für diese Unterstützung möchte ich ihnen danken. Markus Meyer möchte ich für die Begleitung in all den Jahren danken. Es ist etwas ganz Besonderes, dass wir den akademischen Weg seit unserem Studium gemeinsam gehen können. Und wir sind noch lange nicht am Ziel. Mein Dank gilt außerdem Lluїsa Astruc, Joanna Blasczcak, Paul Boersma, Eva Brehm- Jurish, Danièle Clément, Francesco Costantini, Susana Cortés, Jörg Dreyer, Martin Elsig, Mara Frascarelli, Marco García García, Joachim Jacobs, Stefanie Jannedy, Frank Janßen, Sam Hellmuth, Marije Michel, Daniela Lentge, Luis López, Jürgen Meisel, Cédric Patin, Gisa Rauh, Philip Rausch, Gemma Rigau, Esther Rinke, Franziska Scholz, Lisa Selkirk, Barbara Stiebels, Maria del Mar Vanrell, Anna Vilanova, Ralf Vogel und Malte Zimmermann. Sie haben mir auf unterschiedliche Weise, aber immer konstruktiv, geholfen. Und für jede einzelne Hilfe bin ich dankbar. Des Weiteren gilt mein Dank den wunderbaren Kollegen am Departament de Filologia Catalana der UAB. Sie haben mich herzlich empfangen und mich während des Forschungsaufenthaltes unterstützt. Mein Dank gilt auch der DFG für das zweijährige Stipendium im Rahmen des Graduiertenkollegs 275 Ökonomie und Komplexität in der Sprache (Humboldt-Universität zu Berlin, Universität Potsdam). Mein Dank gilt auch meinen beiden muttersprachlichen Korrekturlesern, Jackie Adams und Sean Oconnor. Zudem gilt mein Dank den vielen Katalanen, die ich im Rahmen meiner Experimente aufgenommen habe. Ebenfalls danken möchte ich Lara A. Bilbao Melcón: Durch sie habe ich erfahren, dass es eine Sprache namens Katalanisch gibt. Nicht von linguistischer Art, dennoch von unschätzbarem Wert ist die Hilfe und Unterstützung meiner Eltern, Roswitha und Winfried Feldhausen, gewesen. Sie waren für mich da, haben mir viele Sorgen abgenommen und mir immer wieder den Rücken frei gehalten. Dafür gilt Euch beiden mein größter Dank. Mein tiefster Dank gilt auch meiner Schwester, Bettina Feldhausen, sowie meinen Freunden, die mich seit vielen, vielen Jahren treu begleiten. Nicht zuletzt gilt mein Dank jenen, die ich vergessen habe was leider nicht zu entschuldigen ist. Hier ist Platz für Deinen Namen: Es versteht sich von selbst, dass trotz der vielen und großartigen Hilfe alle Fehler in der Arbeit mein Verschulden sind. v

6 Contents Abstract Declaration Acknowledgements Contents ii iii iv vi CHAPTER 1 INTRODUCTION Foundations and Basic Assumptions Central Catalan Clitic Left- & Clitic Right-Dislocations (in Catalan) Information Structure Preverbal Subjects are not Dislocated Arguments for non-left-dislocated subjects Outline of the Empirical Results and their Theoretical Analysis Conventions for Glosses and Translations 27 CHAPTER 2 FUNDAMENTAL ASPECTS OF CATALAN INTONATION The Autosegmental-Metrical (AM) model A ToBI System for Catalan ToBI Cat_ToBI Catalan Prenuclear Accents Pitch Accent Alignment The Catalan Nuclear Accent Suprasegmental Prosodic Levels Boundary Cues in Catalan for Intonational Boundary Tones Romance boundary cues in Frota et al. (2007) The Continuation Rise The Sustained Pitch Preboundary Lengthening Pauses 62 vi

7 2.3.6 Complex Boundary Tones Organization of Boundary Cues Assumptions on Phonological Inventory of Catalan 67 CHAPTER 3 PHRASING PATTERNS IN CATALAN SVO STRUCTURES Background & Experiment Hypothesis Detailed Background of Catalan Phrasing Patterns The hypotheses General Outline of the Experiments Experiments on Simple and Complex SVO Phrasing Specific Experiment Design Results Boundary Cues Results Phrasing Patterns Simple SVO (experiment 1a) Discussion Complex SVO (experiment 1b) Discussion Conclusion An Analysis for Simple and Complex SVO Phrasing Theoretical Background The Influence of Syntactic and Prosodic Factors Accounting for Catalan simple SVO structures Prieto (2005) The necessary constraints Motivation for the constraint hierarchy Prieto s (2005) OT tables for SVO phrasing Modifications of the previous account on SVO phrasing An approach to the phrasing of complex SVO structures The necessary constraints and four different constraint rankings Reconciling long objects and MAX-BIN-END A Tentative Approach Optionality in Optimality Theory Stochastic Optimality Theory Applying stochastic Optimality Theory to the phrasing of complex sentences Conclusion 138 vii

8 CHAPTER 4 SYNTACTIC ASPECTS OF CATALAN CLLD AND CLRD Syntactic Aspects of Catalan CLLD and CLRD Three syntactic approaches to CLLD and CLRD CLLD and CLRD Asymmetries Licensing of n-words Binding Obviation effects Conclusion 159 CHAPTER 5 PROSODIC PHRASING OF CATALAN CLLD AND CLRD Background The Hypotheses The Experiment Results Results Summary Discussion A Theoretical Approach to Catalan Dislocation Structures Two further constraints: ALIGN-TOP,R and ALIGN-vP,R The constraint hierarchy Conclusion 192 CHAPTER 6 LEFT-DISLOCATIONS AND PREVERBAL SUBJECTS The Hypotheses The Experiment Results Discussion and Conclusions 201 CHAPTER 7 CONCLUSIONS AND OUTLOOK 207 References 213 Appendices 228 viii

9 Chapter 1 Introduction The present work deals with the intonational phrasing of preverbal subjects, object clauses and clitic left- and right-dislocations in the Catalan language and thus deals with complex syntactic structures. Since the late 1970s the intonational grouping of sentences as a topic of linguistic research has received more and more attention, in the scope of the prosodic hierarchy. Although the research often deals with relative clauses and parentheticals in regular intervals, the literature concentrates mostly on simple clauses. The studies of the last decade that are especially concerned with the placement of intonational boundaries in Romance languages often do not inquire into complex structures. For this reason the present study is devoted to embedded object clauses, a special type of complex sentences. The intonational grouping of such sentences is approached from two angles. Firstly, the influence of sentential objects on the phrasing of the matrix subject and verb is discussed (Chapter 3). Secondly, the grouping of the embedded clauses themselves is discussed. It is shown that embedded clauses are usually intonationally separated from the matrix clause (Chapter 3). This characteristic changes however, when the embedded clause includes a left-dislocated element. Embedded left-dislocations phrase with the preceding matrix clause (Chapter 5). Despite these main characteristics, the intonational grouping of complex sentences allows for variation. The theoretical approach proposed in the present work accounts for the data presented in this study (the approach is based on Boersma & Hayes 2001 Stochastic OT; cf. below). When discussing preverbal subjects and left-dislocations in one study, the status of preverbal subjects must be considered (cf. Chapter 1.1.4). Several linguists argue that preverbal subjects are automatically left-dislocated. There are however other linguists who interpret them differently: They argue that preverbal subjects are not automatically leftdislocated. I am one of them. While the first Chapters inquire into subjects and left-dislocations separately, Chapter 6 is dedicated to both. The empirical data in the present work is completely new. Four intonational experiments were conducted. The first on simple SVO structures, the second on complex SVO structures. The

10 Chapter 1: Introduction third experiment is concerned with clitic left- and right-dislocations. The fourth experiment compares preverbal subjects and left-dislocations. These experiments show several interesting results which are now briefly introduced. The most common phrasing in simple SVO structures is (S)(VO), as in (1a). However, (SV)(O) is possible when the object is branching, as in (1b). Moreover, the number of (SV)(O) phrasings increases significantly, when the object is not only branching but also sentential, as in (2). Sentential objects intensify the effect of branching objects. Object clauses typically are separated from the matrix clause by a prosodic break, cf. (2). No break, however, precedes the object clause when it contains a left-dislocated element: An embedded left-dislocation phrases with the matrix clause and leaves out the remainder of the embedded clause in a separate prosodic phrase, as in (3). This signals in turn a strong right boundary. For this reason left-dislocations do not build a prosodic phrase of their own. This constitutes empirical counterevidence to the claim that focus is a prosodic domain of its own. Clitic right-dislocations are almost always separated from the preceding main clause by a prosodic boundary, as in (4). (1) SVO-phrasing in simple clauses (i.e. DP object): ( S ) ( V O ) <= Prosodic Grouping a. L Àguila robà el ratolí. the.eagle steal.3sg.pst the mouse The eagle stole the mouse ( S V ) ( O ) b. L àguila robà el ratolí del meu germà. the.eagle steal.3sg.pst the mouse of.the my brother The eagle stole my brother s mouse. (2) SVO-phrasing in complex structures (i.e. sentential object): ( S V ) ( q S ) ( V O ) [La Bàrbara suposa [que l àguila robà el ratolí ] CP2 ] CP1. the B. assume.3sg that the.eagle steal.3sg.pst the mouse (3) Embedded clitic left-dislocation (CLLD): ( S V CLLD ) ( V PP ) [La Maria va dir [que les taules, les va portar al pis ] CP2 ] CP1. the M. PST.3SG say.inf that the table.pl CL.ACC PST bring.inf to.the flat Mary said that (s)he brought the tables to the flat 2

11 Chapter 1: Introduction (4) Clitic right-dislocation (CLRD): ( V PP ) ( CLRD ) Les i vaig portar al pis, les taules i. CL.ACC PST.1SG bring to-the flat the tables I brought the tables to the flat SVO represents the canonical word order in Catalan: Preverbal subjects are not inherently left-dislocated; they can be left-dislocated, but this is not always the case. Prosodic evidence for this statement is based on phrasing data. Non-branching preverbal subjects which are contextually given show a strong right-boundary, typical of left-dislocations. In contrast, nonbranching preverbal subjects that are not given clearly show the tendency for (SV)(O) phrasing when the object is long. The above empirical findings represent only the main characteristics. The actual groupings differ however in certain respects. A theoretical approach to the data must account for this variation. The analysis is grounded in the framework of Stochastic Optimality Theory (Boersma & Hayes 2001), a variant of Optimality Theory (Prince & Smolensky 1993/2004), for this reason. This model accounts for the variation, while a non-probabilistic constraint ranking is unable to do so. The analysis includes two important constraints: 1) ALIGN-CP,L, a constraint which accounts for the boundary preceding the embedded clause; and 2) ALIGN-TOP(IC),R, a constraint which accounts for the obligatory right boundary of dislocations (i.e. topics). Furthermore, although right-dislocations appear clause final, their syntactic position is within the main clause (Chapter 4). Based on remnant movement of vp to a clause-internal FocP the surface order can be accounted for. The prosodic grouping is accounted for by a third constraint: ALIGNvP,R. This constraint aligns the right edge of vp to the right edge of a prosodic phrase. Chapter 1 is structured as follows. Section 1.1 introduces the foundations and the basic assumptions of the theses are introduced which represent the starting point of the study. In Section 1.2, an outline of the empirical findings and the proposed theoretical analysis is shown by presenting the main topics of the different Chapters. Finally, the conventions for glosses and translations are given (Section 1.3). 3

12 Chapter 1: Introduction 1.1 Foundations and Basic Assumptions In this section the following foundations and basic assumptions of the present work are introduced which represent the starting point of the work. The research language Central Catalan is presented in Section In Section 1.1.2, two aspects of the component called information structure (IS) important for my work are introduced; namely: a) The relation between focus and question/answer pairs and b) Givenness. A brief description of the constructions called clitic leftdislocation (CLLD) and clitic right-dislocation (CLRD) is given in Section In Section 1.1.4, several arguments are given for the basic assumption of the thesis that preverbal subjects are not (necessarily) left-dislocated Central Catalan: Catalan is a Romance language spoken in northeastern Spain (cf. Figure 1 at the end of this section). It is spoken in the autonomous communities of Balearic Islands, Catalonia and Valencia. It has the status of a co-official language in these communities. It is the official language in Andorra. It is also spoken in the French department Pyrénées-Orientales (i.e. Rosselló region) and in the city of L Alguer located on the western coast of the Italian island of Sardinia. The language has between 7 and 10 million speakers, depending on the method of definition. The territories of the Catalan language and culture are often called Paїsos Catalans Catalan Countries. During the history of the Catalan language it has often been degraded as a mere dialect of Spanish. This view is incorrect. The degradation of the language was especially pronounced during the dictatorship ( ) of Francisco Franco. The acknowledgement of Catalan as language in its own right is not a concession to ideological and political demands but is based on linguistic data. The most noticeable differences between Spanish (Castilian) and Catalan concern the sound system, the lexicon, and other less grammatical aspects (typical in Romance languages). 1 One example of these differences is the initial f- in Latin which under certain circumstances became h- in Spanish. Later it became silent but remained orthographically: FACERE > hacer do, FOLIA > hoja leaf, FUGERE > huir flee. It is supposed that this change was due to the influence of the Basque substratum during the Roman Empire. Catalan, in 1 This paragraph is mainly based on Schlösser (2005:60f.). 4

13 Chapter 1: Introduction contrast, did not undergo the development: fer, fulla, fugir. In addition, some Latin vowels became a diphthong in Spanish: SEPTEM > siete seven, TERRA > tierra earth, GROSSUM > grueso thick, PONTEM > puente bridge. Again, Catalan did not undergo the phenomenon: set, terra, gros, pont. Furthermore, many Spanish words kept a final n after the transformation from Latin, whereas Catalan dropped the n: bé good, fi End, raó reason. This in contrast to Spanish bien, fin, razón. The Catalan lexicon has more words in common with Occitan (and French) than with Spanish. According to Schlösser (2005:61) nearly 70% correspond with Occitan, and only 20% correspond with Spanish. Some examples of the basic vocabulary that do not correspond with Spanish are: arribar arrive (sp. llegar, fr. arriver), blau blue (sp. azul, fr. bleu), cercar search (sp. buscar, fr. chercher), formatge cheese (sp. queso, fr. fromage), menjar eat (sp. comer, fr. manger), res nothing (sp. nada, fr. rien). 2 Catalan comprises a collection of dialects which are mutually intelligible. The main dialects are organized into two major dialect blocks, Eastern Catalan (català oriental) and Western Catalan (català occidental). Each block encompasses several regional varieties (cf. Figure 1). The Western Catalan block comprises the two dialects Northwestern Catalan and Valencian. The Eastern Catalan block comprises four dialects: Central Catalan, Balearic, Rossellonese, and Alguerese. The dialects can further be divided into different subdialects (e.g. Central Catalan is subdivided into Barceloní (in the area of Barcelona), Tarragoní (in the eastern half of Tarragona province), and Septentrional de transició (in the transition zone to Northern Catalan)). Central Catalan can be considered the standard pronunciation of the language and has the highest number of speakers. It is commonly spoken in densely populated regions such as the province of Barcelona, the eastern half of the province of Tarragona and most of the province of Girona. In the northern part of Girona a transition to Northern Catalan begins. Typical cities where Central Catalan is spoken in addition to Barcelona are Terrassa, Manresa, Vic, Mataró, Berga, Sant Celoni, Granollers, Sabadell, Cerdanyola, Sitges, and Igualada. The domain of Central Catalan does not equal the territory of Catalonia. The western part of Catalonia (e.g. the area around Lleida) belongs to the Western Catalan block. The most striking differences between català oriental (Eastern Catalan) and català occidental (Western Catalan) are primarily of a phonetic nature. The occidental dialects 2 For detailed information on the lexically relationship between Catalan and other Romance languages cf. Melchor & Branchadell (2002:43&51) 5

14 Chapter 1: Introduction distinguish clearly between unstressed [a] and [e]. The oriental dialects, in general, merge them into one sole vowel [ə] ( schwa ), cf. Table 1. A further distinction concerning unstressed vowels is the fact that [o] is substituted by [u] in the oriental dialects, but not in the occidental dialects (nor in Majorcan), cf. Table 2. These phenomena are comprised under the term vowel reduction. In addition to these phonetic aspects, there are also morphological, syntactic and lexical differences (cf. Melchor & Branchadell 2002:71f.) Northwestern Catalan, Valencian (Western/Occidental Catalan) mar[e] mother c[a]nço song Central Catalan, Balearic, Rossellonese (Eastern/Oriental Catalan) mar[ə] c[ə]nço Table 1: Catalan dialects and the existence of the neutral vowel [ə] (taken from Melchor & Branchadell 2002:71) Northwestern Catalan, Valencian (Western/Occidental Catalan), and Majorcan p[o]sar put ferr[o] iron Central Catalan, Balearic (not Majorcan), Rossellonese, Alguerese (Oriental Catalan) p[u]sar ferr[u] Table 2: Catalan dialects and the existence of the neutral vowel [u] (taken from Melchor & Branchadell 2002:71) Figure 1: Map of the Paїsos Catalans (without L Alguer) and their dialectal division (taken from Wikipedia) 6

15 Chapter 1: Introduction Clitic Left- & Clitic Right-Dislocations (in Catalan): Clitic Left- and Right-Dislocations are very common in Romance 3 languages. Dislocated elements are often doubled by a weak pronoun, a so-called resumptive clitic, within the clause. These clitics 4 immediately precede the verb. Due to this doubling, these constructions are called clitic left-dislocation (CLLD) and clitic right-dislocation (CLRD) respectively. Below, there are Catalan examples for CLLD and CLRD. (5) Normal word order (V-DO-PP): Vaig portar les taules al pis. PAST.1SG bring the tables to-the flat I brought the tables to the flat (6) Clitic Left-Dislocation (CLLD): Les taules i, les i vaig portar al pis. The tables CL.ACC PAST.1SG bring to-the flat I brought the tables to the flat (7) Clitic Right-Dislocation (CLRD): Les i vaig portar al pis, les taules i. CL.ACC PAST.1SG bring to-the flat the tables I brought the tables to the flat The normal word order in Catalan is SVO (cf. Section 1.2.4). 5 This order is typical for Romance languages (Hulk & Pollock 2001:3) 6. (5) reflects the canonical order for Catalan. In (6) the 3 There are similar constructions in other languages such as, for example, English or German. However, the appropriate term for these constructions in these languages might be Topicalization. There are also Left-Dislocation constructions, but they seem to be rather on a par with Romance Hanging Topic constructions. However, this issue is far beyond the goal of this paper. But cf. Frey (2004) for a comparison between English and German Topicalization and Left-Dislocation. For an overview of syntactic analyses of Topicalization cf. Feldhausen (2002) and references cited there. 4 As Anderson (2000:304) points out the notion of a clitic merges two rather different kinds of behavior. Zwicky (1977) invented two terms reflecting the differences: On the one hand there are simple clitics, on the other hand there are special clitics. Simple clitics are elements which are phonologically dependent on an adjacent word. Anderson considers them as prosodically deficient insofar as they lack some of the higher-level prosodic organization (assignment to a phonological word, for example) and must therefore be incorporated into some adjacent element of an appropriate prosodic type (Anderson 2000:305). Simple clitics in English are e.g. the reduced auxiliary `s (meaning is or has) or `d (meaning would). Special clitics are members of a class of (typically pronominal) forms, whose placement is unusual or unique in terms of the syntax of corresponding non-clitic elements. The paradigmatic exemplars of clitics in this sense are the non-subject pronominals of, e.g., French or Spanish. [ ] In most cases special clitics are also prosodically deficient elements (Anderson 2000:304f). This prosodic deficiency of special clitics is also observable in Catalan clitics of CLLD and CLRD: They lack independent stress and do not count as prosodic words, and lean (at this level of the prosodic hierarchy) on another word, which is always the following verb(al complex). Cf. also Rigau & Mascaró (2002) for more information. 5 Cf. Hernanz (2002:1022) and López (2003:198); in ditransitive structures the canonical word order is S-V-DO- IO(/PP). 7

16 Chapter 1: Introduction accusative object les taules the tables is left-dislocated and precedes the clause; in (7) it is rightdislocated and follows the clause. In each case a weak pronoun (here: les ) is placed before V. In Catalan, direct and indirect objects are obligatorily resumed by a clitic (Vallduví 2002:1233&1236). 7 If the dislocated argument is a locative (as in the case of verbs such as put), clitics are also obligatory (Bonet 1991:25). 8 Dislocated subjects do not have clitics there are no weak nominative pronouns (Vallduví 2002:1242). 9 Adjuncts can also be dislocated and corresponding resumptive pronouns are optional (Vallduví 2002:1261f.). In the prosodic experiment on CLLD and CLRD (Chapter 5) I focus mainly on dislocated objects. Consequently there is always a preverbal clitic in the TP. In the cases where adjuncts are dislocated, I use a resumptive clitic. 10 The general clitic order in Catalan is DAT-ACC-PART(itive)-LOC(ative), although there is also a lot of syncretism and dialectal variation (cf. Bonet 1991; Bonet 2002:973; Brumme 1997:135). The order in Catalan for elements in the left periphery ( Vorfeld ) is as follows: Relative-wh > as for > HTLD 11 > CLLD > wh (interrogatives, exclamatives) (cf. Villalba 2000, 2004). This is shown for Hanging Topic Left-Dislocations, left-dislocations, and wh-interrogatives in (8). The left-dislocated object la Maria precedes the wh-element quan when in (8a), whereas the left-dislocated PP d aquest tema about this subject follows the hanging topic la Maria in (8b). (8) Order of elements in the Catalan left periphery (examples from Villalba 2004:8) a. La Maria 1, quan hi 1 has parlat? the M. when LOC have-2 talk Maria, when have you talk with her? b. La Maria 1, d aquest tema 2 no en 2 parlis amb ella 1. the M. of.this subject not of.it talk-subj-2 with her Maria, do not talk with her about this subject. 6 There is a consensus among both traditional and generative grammarians that the canonical surface word order of the Romance languages is subject-verb-object (Hulk & Pollock 2001:3). 7 The Catalan clitic resumption differs from Italian clitic resumption in that the resumption in Italian is obligatory only for DOs and optional for IOs (cf. Frascarelli 2000:145) 8 Spanish, for example, does not have locative clitics (Catalan hi, French y, Italian ci), although dislocating a locative argument is acceptable but it is not resumed by a clitic. Spanish also lacks a partitive clitic (Catalan and French en, Italian ne), cf. Bonet (1991:25&57), Zubizarreta (1998:157). Thus, Spanish has fewer clitics than the other three Romance languages. 9 Vallduví (2002:1242) notes that subject agreement takes over the resumptive function of object clitics (cf. also Alexiadou & Anagnostopoulou 1998:ch.6.2). 10 Besides to the dislocation of objects and subjects, Villalba (2000:45f.) points out that different maximal projections can be dislocated in Catalan: PPs, APs, AdvPs and even CPs [±tensed] (cf. also Cinque 1990:57). Even verbal projections can dislocate (but for details cf. Vallduví 2002:ch.4.6.1). 11 HTLD = Hanging Topic Left-Dislocation 8

17 Chapter 1: Introduction One important characteristic of CLLD and CLRD constituents is that they are contextually given. What this means is introduced in the following section on information structure Information Structure: Two aspects of the component called information structure (IS) are important for my work: a) The relationship between focus and question/answer pairs and b) Givenness. These two aspects are important because they guide the construction of the experimental data of the four prosodic experiments. All the simple and complex SVO sentences in Chapter 3 are all-new sentences. The sentences of the CLLD/CLRD experiment (Chapter 5) are clearly divided in terms of focus and givenness: CLLD/CLRD is maximally salient in the common ground (i.e. given), because they are mentioned in the preceding context. CLLD and CLRD thus falling in the concept of familiarity topic. The rest of the clause constitutes the focus domain. The preverbal subjects of the experiment described in Chapter 6 are either maximally salient or part of the focus domain. In what follows I briefly illustrate what is meant by information structure. After this, background information on two relevant aspects of my work is presented and the terms all-new focus, focus domain and givenness and topic are introduced. Information structure (IS) is the phenomenon of information packaging (Chafe 1976, Vallduví 1993, Krifka 2007) that responds to the demands of the communicative situation by organizing the constituents of the sentence according to communicative needs. The organization can be understood as a structuring of the sentence by syntactic, prosodic or morphological means. These means constitute cues for the addressees to correctly interpret the intended meaning: They enable the hearers to easily identify two things. On the one hand the listeners can identify which part of the sentence is an actual contribution to their information state at the time of the utterance, and on the other hand, which part represents material already subsumed by this information state (Engdahl & Vallduví 1996:2). The clear restriction of IS regarding the aspects that respond to the temporary state of the hearer s mind has been proposed in the seminal work by Chafe (1976). The information mutually known by the speakers/hearers and which is continuously modified in communication can be called common ground (CG; Stalnaker 1974, cf. Krifka 2007:15). The basic notions of IS are Focus, Givenness, and Topic (Krifka 2007). These basic concepts are briefly presented in what follows. The concept of focus is illustrated by the relation of question- 9

18 Chapter 1: Introduction answer pairs. Such pairs are the foundation of the experimental data and thus also have a great importance for my work. In addition, the intersection of givenness and topic is likewise important: The notion of givenness subsumes the notion of topic, because all topics in my work are given. Nevertheless both notions are introduced separately. Focus: The first aspect concerns the classical pragmatic use of focus, which is to highlight the part of an answer that replies to the wh-part of a constituent question (Paul 1880, Krifka 2007:21). The question specifies the way in which the information state should develop in the immediate future. The answer connects an expression to the immediately preceding context. The expression which adds information to the immediately preceding context is the focus. The major principle of focus interpretation at work in the grammar is that focus accommodates new information (Selkirk 1984:213). 12 Focus is thus a device for information-packaging. In (9) I give a German example with different questions evoking answers which are different in respect to the actual contribution which is added to the information state of the addressee. (9) Different focus domain a. Question: What happened? Answer: [Anna hat die ZEITschrift gekauft] F A. has the journal buy.ptcp Anna bought the journal b. Question: What did Anna do? Answer: Anna [hat die ZEITschrift gekauft] F A. has the journal buy.ptcp c. Question: What did Anna buy? Answer: Anna hat [die ZEITschrift] F gekauft A. has the journal buy.ptcp d. Question: Who bought the journal? Answer: [ANna] F hat die Zeitschrift gekauft A. has the journal buy.ptcp The three examples (9a-c) differ neither with respect to word order, nor to the position of the sentence accent (marked by capitals), but their focus domain (marked by [ ] F ), however, is different. The focus domain is the part of the answer that responds to the question. 13 The typical result of failing to select the right focus is incoherent communication (Krifka 2007:21). In (9a) the complete clause is contained in the focus domain. This is called all-new focus, sentence focus 12 For a critical argument on the interpretation of the notion of focus as new information cf. Krifka (2007:29f.). 13 The different size of the focus domain can be explained by assuming with Selkirk (1984, 1995a) that the focus feature F of a constituent can project to higher syntactic nodes (cf. Focus Projection Rule, Selkirk 1995a:555; cf. also Selkirk 1984:207ff.). 10

19 Chapter 1: Introduction or whole focus reading. In (9b) only the constituent without the subject constitutes the focus domain, and in (9c) it is only the object DP. The latter case can be called narrow focus, whereas the former domain can be called broad focus. These two terms (cf. Selkirk 1984, Lambrecht 1994) are imprecise and can only be used when different focus alternatives are being discussed (Krifka 2007:31). (9d) differs with respect to the three previous examples in the location of the sentence accent, which is on the subject Anna. The obvious relation between sentence accent (i.e. prominence) and focus is expressed by the Focus Prominence Rule (FPR, Chomsky 1971; Jackendoff 1972; Zubizarreta 1998:88). According to this rule, the focused marked material must be more prominent than presupposed material: Hence, focus bears the sentence stress. 14 Givenness: Elements that are not part of the focus domain are, in general, already present in the common ground (CG) of the speakers/hearers, as is the case with CLLD and CLRD. Being present means that they are given. Givenness always refers to denotation, i.e. the denotation of the element, and the expression itself is not present in CG. Krifka (2007) offers the following definition of givenness, (10), and I adopt this definition for the present work. (10) Definition of Givenness (Krifka 2007:37): A feature X of an expression α is a Givenness feature iff X indicates whether the denotation of α is present in the CG or not, and/or indicates the degree to which it is present in the immediate CG. This definition allows us to say that an expression is given to a particular degree. It can be either maximally salient or only given in the immediately preceding CG. It is also possible that it is only given in the general CG. Krifka s (2007) notion of givenness is different from that of focus. His definition of givenness allows focused constituents to be given (which becomes relevant in the case of second occurrence focus). 15 I am not, however, concerned with the transition zones of the notions. In my experimental work only constructions with a clear boundary between givenness and focus are used (as explained at the beginning of this section). 14 Sentence stress is a form of phrasal stress. Phrasal stress is the stress assigned beyond word stress (strongest stress in a prosodic word) in syntactic collocations of words, such as phrases, clauses, and sentences (Truckenbrodt 2007). The most well-known accounts to phrasal stress are the Nuclear Stress Rule (NSR) by Chomsky & Halle (1968) and the Sentence Accent Assignment Rule (SAAR) by Gussenhoven (1983, 1992). The important role of Focus in assigning sentence stress is mostly modeled by an abstract feature F or FOC, which is assigned to a syntactic constituent. Rooth (1992) developed a theory (alternative semantics of focus) including the important role of focus. (For more information cf. Truckenbrodt (2007) and the references cited there). 15 For more details cf. Schwarzschild (1999) who developed a detailed account on the interaction of givenness and focusation. 11

20 Chapter 1: Introduction Givenness can be indicated in several ways (Krifka 2007:38). Some of them are the realization of an expression in a non-canonical position and deaccentuation (i.e. the reduction of the prosodic realization of given expressions). Typically the non-canonical position is before the canonical position. These two strategies are used in Catalan, where clitic left-dislocations as well as clitic right-dislocations are not in their canonical position (cf. Section 1.2.2). In addition, the former is accented, but it does not bear sentence stress and the latter is deaccented (cf. Chapter 5). Vallduví (1993:119) notes that Catalan has a very straightforward way to represent information packaging. Left-dislocations precede the main clause and right dislocations follow the core clause (cf. (6) and (7) in the dislocation section). Whatever is left in the core clause must be interpreted as focal (with the exception of clitics). For this reason no dislocated element can be part of the focus domain (Vallduví 2002:1253). Topic: Dislocations are often taken as topics (Alexiadou 2006:686; Frascarelli 2000; Rizzi 1997; Zubizarreta 1998) and host the functional projection TopP (Rizzi 1997, Villalba 2000 and many more). 16 According to Reinhart (1981, 1995) there are two main schools of thought for the characterization of the concept topic. The two concepts are presented in (11) (based on Frey 2004). (11) a. Aboutness concept of topic: A topic is an expression whose referent the sentence is about b. Familiarity concept of topic: A topic is that expression whose referent has been already introduced into the discourse or is for other reasons already familiar to the interlocutors The aboutness concept of topic assumes that the notion of topic refers to the object the speaker is thinking about, whereas the rest of the utterance (the comment) refers to what the speaker is thinking about the object. Hence, topic can be understood in terms of aboutness: The comment is said about the topic (Hockett 1958:201). 17 In conjunction with this concept the topic can also be new; it is only important that the sentence is about the topic (cf. Krifka 2007:41 for discussion). Due to the fact that CLLD and CLRD always must be given, I choose the familiarity concept of topic as the appropriate one. For the purpose of the study, givenness and familiarity are synonymous. 16 Cf. López (2002) for a critical assessment of the functional projections FocP and TopP. 17 The most general characteristic of predicative constructions is suggested by the terms topic and comment for their ICs [immediate constituents, I.F.]: the speaker announces a topic and then says something about it. (Hockett 1958:201). 12

21 Chapter 1: Introduction Subsuming CLLD and CLRD under the notion of givenness (or familiarity topic) has the advantage of being able to move away from the differences which exist between Catalan CLLDand CLRD constituents. Vallduví (1993) and Villalba (2000) show that there are interpretational differences between these two kinds of dislocations. 18 The differences, though, are not of importance for the work at hand because I am only interested in the main informational separation of the clause as presented in Vallduví (1993:119): Given constituents must precede or follow the core clause. The question of why a constituent is located in the left or right periphery of the clause is too detailed for this study. The intent is to make a statement about the prosodic phrasing patterns of CLLD and CLRD Preverbal Subjects are not Dislocated: In this work I assume that SVO is the normal word order in Catalan. In this section several arguments are presented supporting this assumption. In Romance, there is a long lasting and involved discussion on the canonical position of subjects. The main issue is determining if the canonical word order (WO) is SVO or VOS. In Catalan, for example, there is consensus that the order of the verb and its internal arguments is fixed: V-DO-IO(/PP) (cf. López 2003, 2008; Vallduví 1993, 2002:1230; Villalba 2000; Wheeler, Yates & Dols 1999). Opinions differ when it comes to the subject position. Hernanz (2002:1022) and López (2003:198, 2008) assume that the preverbal position is the canonical one, whereas Vallduví (1993:#5.2, 2002:1245), Rosselló (2000), and Solà (1992) assume that the canonical position is postverbal (i.e. VOS). Wheeler et al. (1999) have an intermediate position. They propose that the postverbal subject order is only compulsory with intransitive verbs (Venen trens Trains are coming ) and optional in all other cases (depending on informativeness, phonological weight, etc.). Due to this discussion, it comes as a surprise that Hulk & Pollock (2001:3) state that: [t]here is a consensus among both traditional and generative grammarians that the canonical surface word order of the Romance languages is subject-verb-object. Nevertheless, this is exactly the word order argued for in this section. 18 The main difference is that right-dislocations cannot act as links to the previous discourse like left-dislocations. This is why Vallduví (1993) replaces the notion of topic by link for CLLD and tail for CLRD. In addition, leftdislocations can be contrastive, whereas right-dislocations never can be contrastive. For detailed information of differences between Catalan CLLD and CLRD cf. Villalba (2000:60ff.,144ff.). 13

22 Chapter 1: Introduction The approaches to preverbal subjects can be classified as being of two kinds: One group assumes that preverbal subjects are always left-dislocated (e.g. Alexiadou & Anagnostopoulou , Barbosa 1995, Contreras 1991, Ordoñez & Treviño 1999, Solà 1992, and Vallduví 1993, 2002:1245), whereas the second group assumes that preverbal subjects can, but need not be, leftdislocated (Belletti 1990; Cardinaletti 1997, Costa 2004:ch.2; Gutiérrez-Bravo 2007; López 2003, 2008:ch.7; Pires 2007; Sheehan 2006) Arguments for non-left-dislocated preverbal subjects The syntactic literature on the status of preverbal subjects is endowed with many arguments for both groups. In this section, it is shown that preverbal subjects are not leftdislocated, but rather part of the focus domain. I mainly rely on the work by Sheehan (2006) and López (2008). One of the most recent works on the status of preverbal subjects by Sheehan (2006), discusses at length that (Romance) preverbal subjects are not automatically CLLD (Sheehan 2006:ch.2). She draws the conclusion that it is certainly the case that preverbal subjects are CLLD on some occasions in Romance, it need not be the case that they are always CLLD (Sheehan 2006:72), and adds that preverbal subjects in fact cannot always be CLLD. López (2008) draws in principal the same conclusion. 20 In this section, the following arguments for why preverbal subjects cannot always be clitic left-dislocated are introduced: i. Preverbal subjects are in an A-position (López 2008:109, Sheehan 2006:88) ii. All-new contexts (López 2008:109, Sheehan 2006:75) iii. Islandhood (Sheehan 2006:84) iv. Subjects present for disambiguation (Sheehan 2006:82) v. Non-referential QPs as preverbal subjects (Sheehan 2006:76) vi. Unambiguous wide scope (Sheehan 2006:55) (i) Preverbal Subjects are in an A-Position: López (2008) states that an experiencer argument can interfere with NP-raising in every Romance language, cf. (12), (13). In (12a) the subject Joan raises out of the infinitival complement clauses to the A-position of the raising verb semblar seem. If an experience argument intervenes between the moved subject and the matrix verb (12b), an intervention effect 19 A&A (1998) = Alexiadou & Anagnostopoulou (1998) 20 However, there are at least two differences between López (2008) and Sheehan (2006). A) López (2008) takes CLLD as a result of movement, whereas Sheehan (2006) assumes a base-generation approach, like Cinque (1990). B) López (2008:108) argues in the discussion on preverbal subjects, that if they are dislocated, they are very likely HTLD and not CLLD. Nevertheless, they are dislocated. 14

23 Chapter 1: Introduction arises. As López (2007) shows, such an intervention effect does not appear with a clear instance of dislocation, (13). Due to the finite verb és is in the embedded clause, an analysis in terms of raising to subject is not possible. Hence, the subject Juan must be dislocated and the presence of the dative experience em me does not change the grammaticality of the sentence. (12) a. El Joan sembla ser intel.ligent [CAT] the J. seem be.inf intelligent John seems to be intelligent b. *El Joan em sembla ser intel.ligent [CAT] the J. CL.DAT seem be.inf intelligent John seems to me to be intelligent (13) El Joan em sembla que és intel.ligent [CAT] the J. CL.DAT seem that be intelligent John, it seems to me that he is intelligent. López (2008) sees the contrast between (12) and (13) as evidence that an A-movement of the subject exists (from Spec,v to Spec,T). Sheehan (2006:88) also states that many preverbal subjects must be in an A-position. Relying on Suñer (2002) she says that Spanish ad sensum agreement supports this view. Suñer (2002) shows that true CLLD subjects allow optional ad sensum agreement. Ad sensum agreement refers to the fact that subject and verb differ in number as in (14): Whereas the predicate estaban presionados be pressured has 3pl agreement, the subject el jurado the jury is grammatically singular but semantically plural. (14) El jurado, María nos aseguró que estaban the jury.3sg M. us assured.3sg that were.3pl presionados pressured.3pl [SP] The jury, Maria assured us that they felt pressured. (15) El jurado *estaban presionados / estaba presionado [SP] the jury.3sg were.3pl pressured.3pl were.3sg pressured.3sg The jury felt pressured. As can be seen in (15) ad sensum agreement is not possible with all preverbal subjects. In (15) the preverbal subject is directly adjacent to the finite verb. According to Sheehan, Suñer (2002) draws the conclusion from these data, that ad sensum agreement is only possible with CLLD subjects. If, as in (15), ad sensum agreement is not possible, the preverbal subject is not an instance of CLLD. 15

24 Chapter 1: Introduction (ii) All-new contexts: Vallduví (1993:119) notes that Catalan has a very straightforward representation of information packaging (term by Chafe 1976): All links are left-detached, all tails are rightdetached, and whatever is left in the core clause (under the lowest IP) must be interpreted as focal (with the exception of clitics). The general idea behind this description is that no dislocated element can be part of the focus domain because it is not part of the rheme (Vallduví 2002:1253). Consequently, as López (2008:109) notes, a preverbal subject cannot be dislocated in an all-focus sentence. 21 A sentence counts as all-new/all-focus if it is an appropriate answer to the question What happens? (cf. Krifka 2007:23). The word order SVO in the answer is obligatory in the Catalan all-focus sentence in (16), taken from López (2008:109). (16) [Context: What happened?] [CAT] a. (que) el Joan finalement va portar els llibres SVO that the J. finally PAST bring.inf the book.pl (that) John finally brought the books. b. #(que) els llibres finalement els va portar el Joan CLLD that the books finally PAST bring.inf the J. (that) John finally brought the books. The concept that when a whole sentence is forced by the context to be a rheme, no constituent can be dislocated is also taken up by Sheehan (2006:75), citing Costa (2001). Costa (2001) shows the same pattern for European Portuguese, (17). (17) [Context: What happened?] [EP] a. O Pedro partiu o braço SVO the P. broke the arm Pedro broke his arm. b. #Partiu o Pedro o braço VSO broke the P. the arm Pedro broke his arm. c. #O braço, o Pedro partiu-o CLLD the arm the P. broke-cl The arm, Pedro broke it. 21 The all-focus context presented here constitutes counter evidence to the assumption that subjects naturally are topics (van Oosten 1986, Horn 1989:ch.7), where topic is understood in the sense of theme conflating with the notion of old information (Daneš 1970). 16

25 Chapter 1: Introduction (iii) Islandhood: Müller & Sternefeld (1993:485) show that topics in German create an island for whextraction. Gutiérrez-Bravo (2007) cites an argument from Goodall (2001) as evidence that preverbal subjects are not fronted topics based on the islandhood of embedded topics. Whereas fronted topics in Spanish create islands for extraction, (18a), preverbal subjects do not, (18b). (18) a. *A quién crees [que el premio se lo to whom think.2sg that the prize CL.DAT CL.ACC dieron] gave.3pl [SP] Lit. Who do you think that the prize they gave to? b. A quién crees [que Juan le dio el premio] [SP] to whom think.2sg that J. CL.DAT gave.3sg the prize Who do you think that Juan gave the prize to? (Examples taken from Sheehan 2006:84; bold marking by I.F.) According to Sheehan (2006:85), approaches in which the preverbal subject occupies an A-position do not have any difficulty in explaining the data in (18): The CLLD element in (18a) is in an A-bar-position and has thus a different status to the preverbal position of the embedded subject in (18b). The former blocks A-bar-movement of another more embedded element, whereas the preverbal subject, being in an A-position, does not block A-bar extraction. If the preverbal subject is treated like being in an A-bar position (as do A&A 1998) it should block whextraction, contrary to fact. (iv) Subjects present for disambiguation: Subject pronouns in Catalan are seldom used: The phonetically null form is the unmarked case regarding frequency and distribution (Vallduví 1993:99). Nevertheless, subject pronouns are used in the following circumstances. A) When used without a verb, (19); B) When accented (mostly in contrastive contexts), (20); C) In combination with mateix (same), i.e. myself, yourself,, (21); and D) When verb forms are ambiguous, i.e. when two possible antecedents are present, and the subject interpretation must be unambiguous, (22). (Examples taken from Brumme 1997:123). (19) Qui vol gelat? - Jo. [CAT] who want ice-cream I Who wants ice-cream? - Me 17

26 Chapter 1: Introduction (20) Jo compra pa i tu portes mantega [CAT] I buy bread and you get.2sg butter I will buy bread and you ll get butter. (21) M ho va dir ella mateix [CAT] me.cl PST say.inf she -self She said it to me herself. (22) Si pro hagués arribat primer [CAT] if have.subj.pst.1/3ps arrive.ptcp first If I / (S)he had arrived first, In the past indicative (indicatiu imperfet), present subjunctive (subjuntiu present), past subjunctive (subjuntiu imperfet), and conditional (condicional) of Catalan, 1 st and 3 rd person singular verb endings are all syncretic. In cases such as (22) the subject pronoun is used in order to disambiguate between the possible antecedents, (23). (23) a. Si jo hagués arribat primer [CAT] if I have.subj.pst.1ps arrive.ptcp first If I had arrived first, b. Si ell hagués arribat primer [CAT] if he have.subj.pst.3ps arrive.ptcp first If he had arrived first, Sheehan (2006:82f.) shows similar examples for Italian (taken from Cardinaletti 1997). In Italian present subjunctive, 1 st, 2 nd, and 3 rd person singular verb endings are syncretic; hence the ambiguity is three way. As for disambiguation, the 2 nd person singular requires an overt pronoun. Sheehan argues that the 2 nd person overt pronoun is not a CLLD topic, but rather is merely required for functional/pragmatics reasons to differentiate between ambiguous verb-forms. (Sheehan 2006:83). The same is true for (23). A CLLD approach to overt preverbal subjects cannot account for non-topical preverbal subjects whose only function is disambiguation Cf. Mayol (2006:76) for a further function of the use of overt pronouns, in which they could help to select the less accessible antecedent. 18

27 Chapter 1: Introduction (v) non-referential QP as preverbal subjects: Cinque (1990) claimed that robustly non-referential QPs or bare NPs in Italian, Spanish, EP, or French are not possible as CLLD with a resumptive clitic (cf. also Raposo 1986, Kato & Raposo 2005), cf. (24). Sheehan (2006:77) states that similar phenomena are true with respect to subject left-dislocation in Brazilian Portuguese and French, (25). (24) a. *Livros do Tintim, li-os antes de adormecer [EP] books of.the Tintin read:1sg-cl.them before of sleep:inf.1sg Lit. Books about Tintin, I read them before going to sleep. b. *Poucos candidatos, ouvi-os falar na televisão [EP] few candidates heard-them speak on.the TV Few candidates, I heard them speak on the television. (Examples from Sheehan 2006:48; originally from Raposo 1986) (25) a. *Ninguém i ele i gosta de chorar [BP] nobody he likes of cry.inf Nobody likes crying. b. *Personne, il n aime pleurer [French] nobody he not likes cry.inf Nobody likes crying. (Examples from Sheehan 2006:77) In contrast to (25) preverbal non-referential QPs, however, are normal in Romance NSLs, as can be seen in (26) for Spanish and Catalan. 23 The conclusion is that if elements which cannot be topics appear as preverbal subjects, they cannot be left-dislocated. (26) a. Nadie quiere ser político [SP] Nobody wants be.inf politician Nobody wants to be a politician. (Example from Sheehan 2006:77) b. Ningú vol ser el president del CF Badalona [CAT] Nobody wants be.inf the president of.the CF Badalona Nobody wants to be the president of CF Badalona. 23 In contrast to the general assumption that QPs are not tolerated as topics, A&A (1998:508) argue that CLLD of QPs is possible (giving evidence from Italian). Cf. Sheehan (2006:78) for a short discussion and rejection of clitic left-dislocated QPs. 19

28 Chapter 1: Introduction (vi) Unambiguous wide scope Finally, I present the argument by Sheehan (2006) against the claim that Quantifier Phrases (QPs) in preverbal subject position have unambiguous wide scope (A&A 1998:504ff), i.e. they fail to reconstruct. If Alexiadou & Anagnostopoulou (1998) are correct with their claim that preverbal subjects in Null Subject Languages (NSLs) are always CLLD in an A -position, then preverbal subjects should display the same syntactic and semantic properties as other CLLD elements. 24 Sheehan (2006:50) comes up with several relevant properties of CLLD as diagnostics in an empirical evaluation of the status of preverbal subjects. One of them is the unambiguous wide scope (i.e. failure to reconstruct) of CLLD elements. 25 According to Cinque (1990:66) CLLD elements do not reconstruct. For this reason, they always take wide scope over IP internal elements. This means that they are not ambiguous between a narrow scope reading and a wide scope reading of the CLLD element. A&A (1998:504ff.) claim that preverbal QPs (Quantifier Phrase) in Greek and Spanish always take wide scope over other quantifiers in the clause. This suggests that they are also CLLD, as in the Greek example (27a). (27b) shows the QP in postverbal position, where the subject can have either narrow or wide scope (cf. A&A 1998:505). (27) a. Kapios fititis sitihiothetise kathe arthro wide scope only some student filed every article Some (particular) student filed every single article. b. Sitihiothetise kapios fititis kathe arthro ambiguous filed some student every article i) Every article was filed by some student (or other). ii) Some (particular) student filed every single article. (Examples taken from Sheehan 2006:55) (28) a. Algún estudiante sacó prestado todos los libros Some student took borrowed all the books i)?? All the books were borrowed by some student. narrow scope subject ii) Some (specific) student borrowed all the books. wide scope subject 24 Due to the fact that A&A (1998) consider CLLD as base-generated in the peripheral position (cf. Cinque 1999), Sheehan also assumes that CLLD constituents are base-generated in their surface position. She does so in order to facilitate a comparison between the pattern of preverbal subjects and the approach taken in A&A (1998). 25 In addition to the diagnostic introduced here, Sheehan (2006:50ff.) mentions free positioning with respect to other fronted elements and restriction to referential XPs. As for the latter, (26) shows that non-referential QPs are normal in Romance NSLs. 20

29 Chapter 1: Introduction b. Algún estudiante sacó prestado cada libro Some student took lent each book i) Each book was borrowed by some student (or other). narrow scope subject ii) Some (particular) student borrowed each book. wide scope subject (taken from Sheehan 2006:55) The Spanish examples in (28) show a slightly different pattern. Whereas the marginality of the narrow scope reading in (28a) is by and large in line with A&A, the possible narrow scope reading in (28b) is not. It is Suñer (2002) who claims that preverbal subjects are ambiguous in Spanish and thus refutes A&A s claim. Sheehan (2006:56ff.) gives further examples from other NSLs showing that the preverbal subject position is often potentially ambiguous in scope. She concludes that there is little empirical support for A&A s claim outside of Greek and that, like in the Spanish examples, scope seems to depend on pragmatic and semantic factors relating to the particular quantifiers used. As for the general claim that CLLD does not reconstruct, Zubizarreta (1998:114) shows that Spanish CLLD reconstructs to an intermediate position, a position which is higher than its base position. López (2008:96) translates her argument into Catalan, (29). (29) a. El seu i fill, cada mare i haurà d accompanyar=lo el primer the her child each mother must.fut of.accompany=cl.acc the first dia d escola day of.school Each mother must accompany her child on the first day of school b. *El seu i fill, haurà d accompanyar=lo cada mare i el primer the her child must.fut of.accompany=cl.acc each mother the first dia d escola day of.school Each mother must accompany her child on the first day of school (Examples from López 2008:96) The subject quantifier cada each in the preverbal position can bind the possessor seu her in the CLLD, (29a). This is an indication that CLLD must reconstruct (López 2008:96) at least below the high subject position. That CLLD does not reconstruct to its initial merge position is shown in (29b), due to the fact binding is impossible here. López takes this as evidence that CLLD elements reconstruct to an intermediate position, where they are c-commanded by the preverbal subject on the one hand, but c-command the postverbal subject on the other hand. 21

30 Chapter 1: Introduction In summary, based on Sheehan (2006:ch.2) and López (2008:109) I have shown that there are good reasons to claim that preverbal subjects are not intrinsically CLLD constituents. For this reason I take for granted that preverbal subjects can be non-left-dislocated as well as being leftdislocated. That preverbal subjects can have two functions is also supported by two possible structural positions for them. Pires (2007) argues that Brazilian Portuguese allows preverbal subjects to be realized either as left-dislocated elements (i.e. topics) or as arguments internal to the clause, i.e. in [Spec,TP]. Consequently, the functional division is mirrored by the structural position. Even Rosselló (2000:105), who assumes that preverbal subjects are always leftdislocated, mentions that in Catalan preverbal subjects may merge in the Inflection Phrase or in the C-domain and that only in the latter position would they have a position similar to true leftdislocated elements. 1.2 Outline of the Empirical Results and their Theoretical Analysis Chapter 2 introduces fundamental aspects of Catalan intonation. The Cat_ToBI system (Prieto, Aguilar, Mascaró, Torres & Vanrell 2007) is taken as the basic model for transcription and is explained in detail. 26 The two prosodic levels Intermediate Phrase (ip) and Intonational Phrase (IntP, ɩ) are assumed. The term prosodic phrase is used as a hypernym for ip and IntP. There are five different boundary cues used for detecting intonational boundaries: continuation rise, sustained pitch, preboundary lengthening, pauses, and complex boundary tones. Only the last two cues indicate ɩ-boundaries, while the first three represent ip-boundaries. Each recorded sentence was analyzed based on these boundary cues. For this reason each cue is explained in detail. Several pitch tracks of recorded examples are given in order to illustrate the boundary cues. Chapter 3 is devoted to objects and their influence on the prosodic grouping of the (matrix) subject and verb. It is shown that sentential objects increase the number of matrix (SV) phrasing. The finding is modeled in the framework of stochastic Optimality Theory (Boersma & Hayes 2001) in order to account for different possible groupings. Previous research shows that in simple Catalan SVO structures, there is a tendency to produce (SV)(O) phrasings if the subject is short and the object long (i.e. number of prosodic words in Prieto 2005; number of syllables in 26 Prieto (2009) presents a strongly modified version of the intonational phonology of Catalan. These modifications are not considered in the present study. For information on the modifications, cf. Chapter 2, footnote 1. 22

31 Chapter 1: Introduction D Imperio et al. 2005). Based on this, the hypothesis is set up that sentential objects compared to DP objects increase the number of (SV) phrasings. Although the (SV) phrasing is in line with Prieto (2005), her analysis cannot account for the variation in phrasing. Two experiments on SVO-phrasing in Catalan with ten native speakers of Central Catalan were conducted. One with O being branching and non-branching (simple SVO experiment), one with O being sentential (complex SVO experiment). The first experiment has the goal to reproduce the findings of D Imperio et al. (2005). The second experiment has the goal to test the hypothesis. I carefully controlled for the all-new status of the SVO structure via a context causing the structure to be entirely new, (30). (30) Context (question): No et trobes bé? Em sembla que estàs de mal humor. Què ha passat? No you feel.2sg good Me seem that be.2sg of bad mood What have.3sg happen.ptcp Do you feel bad? You seem to be in a bad mood. What happened? Target sentence (complex SVO): La Bàrbara suposa que l Àguila robà el ratolí. the B. assume that the.eagle steal.3sg.pst the mouse Barbara assumes that the eagle stole the mouse The results of the first experiment show that the main phrasing pattern is (S)(VO), where the subject is prosodically separated from the following predicate. Nevertheless, the grouping (SV)(O) is possible, when the subject is short and the object branching. The results of the complex SVO experiment show that the matrix subject is phrased with (at least) the following matrix verb in around 66% of the examples, whereas it is phrased alone only in around 34% of the examples. This clearly supports the hypothesis. In addition, the embedded clause (including the complementizer que) is prosodically separated from the matrix clause in around 80% of the examples. The embedded clause is internally separated by a prosodic break after the embedded subject (100%). The most common prosodic grouping is (SV)(qS)(VO), where the matrix subject and the matrix verb are phrased together, followed by the group complementizer + embedded subject, and by the group embedded verb + embedded object. The next three most common groupings are (S)(V)(qS)(VO), (SVqS)(VO), and (S)(VqS)(VO). My analysis is based on Prieto s (2005) account for simple SVO and includes her three constraints MIN-N-PHRASES >> MAX-BIN-END >> ALIGN-XP,R. Due to the importance of Prieto s work, her approach and the constraints used are explained in detail. A short overview of 23

32 Chapter 1: Introduction my modifications is given here. Prieto (2005) accounts for the (SV) phrasing by both the binarity constraint MAX-BIN-END, which demands that the prosodic phrase that bears sentence stress consists of maximally two prosodic words, and the constraint for avoiding boundaries MIN-N- PHRASES, which allows the subject to be not right-aligned (as demanded by ALIGN-XP,R). These constraints and their strict ranking cannot, however, explain the variation in phrasing. Thus, I depart from Prieto (2005) in three important aspects. Firstly, MIN-N-PHRASES and MAX-BIN-END are re-ranked in the reverse order (MAX-BIN-END >> MIN-N-PHRASES). The re-ranking still accounts for her results (cf. Prieto 2005:216) and it enables to maintain the same constraint order for complex SVO structures. Secondly, a new constraint is assumed: ALIGN-CP,L. This constraint aligns the left edge of a CP with the left edge of a prosodic phrase and thus accounts for the pattern that the embedded clause is in general prosodically separated from the matrix clause. This constraint is based on Gussenhoven (2004:167), who introduces the OT constraint ALIGN(S,ɩ), which aligns the right edge of every sentence with the right edge of an Intonational Phrase. It is additionally based on de Lacy (2003:60), who argues that both constraints, ALIGN-LEFT and ALIGN-RIGHT, are present in every grammar. Thirdly, the analysis is modeled in a stochastic OT framework (Boersma & Hayes 2001), where the constraints are ranked on a continuous ranking scale. This model is explained in detail. A short distance between the constraints causes a less fixed order between them. I argue that ALIGN-CP,L, MIN-N-PHRASES, and ALIGN-XP,R are very close. The proposed general constraint hierarchy is given in (31). (31) MAX-BIN-END >> ALIGN-CP,L >> MIN-N-PHRASES >> ALIGN-XP,R Due to these factors the actual ranking of the constraints will sometimes be the reverse of their normal ranking. Consequently, the second, third or fourth best candidate is able to win. Using this approach, the four most common groupings of the data (which grasp 80% of the data) can be modeled. Table 3 pictures the ranking of the most common pattern (SV)(qS)(VO). 24

33 Chapter 1: Introduction Table 3: MAX-BIN-END ALIGN-CP, L MINIMIZE-N- PHRASES => a. 56% (SV)(qS)(VO) 3 Sm b. 24% (S)(V)(qS)(VO) 4! c. 10% (SVqS)(VO) *! 2 Sm d. 10% (S)(VqS)(VO) *! 3 e. (SVqSVO) *! * 1 Sm,Se f. (SVqSV)(O) *! 2 Sm,Se actual ranking for the most common phrasing pattern (SV)(qS)(VO) (matrix subject = Sm; embedded subject = Se) ALIGN-XP,R Chapter 4 is devoted to syntactic aspects of clitic left- and clitic right-dislocations (CLLD and CLRD respectively). It sets up the syntactic assumptions underlying my optimality theoretic approach to the prosodic phrasing of these constructions (Chapter 5). I argue for a clause-internal analysis of Catalan CLRD based on CLLD/CLRD asymmetries and adopt the analysis by Villalba (2000). In such an approach the CLRD constituent occupies a position below TP and above vp (and does not move to CP). In order to appear at the right on the surface structure, remnant movement of vp into an internal FocP is assumed, as is illustrated in (32a,b). The CLRD constituent is inside TP, but preceded by vp, which hosts the sentence accent. (32c) is relevant for the analysis in Chapter 5. (32) a. TP NegP Neg IntFocP vp i IntTopP CLRD vp i b. vp] CLRD] TP] Syntactic Structure c. ( ) p (CLRD) p Prosodic Phrasing 25

34 Chapter 1: Introduction The CLLD/CLRD asymmetries are highlighted in Villalba (1996, 1999a,b, 2000) and Cecchetto (1999), but are refuted by Samek-Lodovici (2006) 27 amongst others. I show that Catalan data still speak in favor of Villalba s and Cecchetto s assumption. Three tests for asymmetry are discussed. Two of them found in SL (2006): licensing of N- words (such as negative polarity items, NPI) and binding properties (Cecchetto s 1999 antireconstruction effects ). These two tests are discussed because they appear to be controversial in the literature on Italian and Catalan. According to SL, the tests show that dislocated elements in Italian are not c-commanded by T or any other material within TP (such as a licenser for NPIs) and thus support his view that right-dislocated constituents must be external to the clause main TP. I show that the Catalan counterparts behave contrariwise: Catalan CLRD is c-commanded by T and this can only be modeled by a clause-internal analysis. The third test is a completely new test. This test is based on obviation effects. Left-dislocations lead to a disappearance of obviation effects in subjunctive complement clauses (Costantini 2005b). Right-dislocations, as is shown, keep obviation effects. The pattern is explained by assuming a clause-internal analysis of CLRD. Chapter 5 deals with the prosodic phrasing of CLLD and CLRD in Catalan. It shows that embedded clitic left-dislocations are typically not preceded by a prosodic phrase boundary (80% of the data), while clitic left-dislocations are, in general, obligatorily followed by a prosodic boundary. Clitic right-dislocations are obligatorily separated from the preceding main clause by a boundary. The appearance of CLLD in embedded contexts is a topic that intonational research has, in general, not been of much concerned until now. I present experimental data that cannot be explained by Frascarelli (2000) and Prieto (2005). The relevant data concerns complex CLLD structures presented in Villalba (2000) and López (2003, 2008): left-dislocation out of clitic leftdislocations and embedded left-dislocations. In order to account for the data, a new constraint is introduced: ALIGN-TOP(IC),R. This constraint demands that the right edge of a CLLD and CLRD constituent is aligned with the right edge of a prosodic phrase. This accounts for the obligatory right boundary of dislocations (i.e. topics) and allows the constituent to phrase with preceding material. In order to account for the clause-internal position of CLRD constituents, a further constraint is introduced: ALIGN-vP,R. This constraint aligns the right edge of vp to the right edge of a prosodic phrase. Due to the movement of vp into the internal FocP, the right boundary of the prosodic phrase precedes the 27 SL (2006) = Samek-Lodovici (2006) 26

35 Chapter 1: Introduction right-dislocated element, as in (32c) above. The two constraints are added to the OT-approach established in Chapter 3. Chapter 6 is dedicated to the prosodic difference between CLLD and preverbal subjects. A fourth experiment comparing left-dislocations and non-branching, non-given preverbal subjects was conducted. The results show that non-given preverbal subjects are less often separated by a prosodic phrase from following material than left-dislocations. This finding supports the constraint ALIGN-TOP(IC),R and thus supports the basic assumption addressed in Chapter 1 showing that preverbal subjects are not automatically left-dislocated. The data further proves that givenness overrides the effects of branchingness and constituent length (if the subject is non-branching): Given non-branching subjects do not show the tendency for (SV)(O) when the object is long. The percentage of clear boundaries after the given (i.e. dislocated) subject comes to 88%. The percentage of clear boundaries after new non-branchings subjects only comes to 67%. This means that the number of (SV)(O) phrasings of new subjects is much higher and thus in line with Prieto (2005). No analysis is presented in this Chapter because the analysis established in the preceding Chapters already accounts for the data. Chapter 7 summarizes the contribution of this work to the area of prosody, syntax, and the prosody-syntax interface. It concludes the work by offering several directions for further research. 1.3 Conventions for Glosses and Translations The interlinear glosses used throughout this work are based on the Leipzig Glossing Rules (LGR) 28, a standard set of conventions proposed to the community of linguists. Hence, my glosses are left-aligned vertically, word by word, with the example. If segmental morphemes are separated in the examples, they are done so by hyphens ( - ), both in the example and in the gloss. If clitic boundaries are marked, they are marked with an equals sign ( = ), both in the object language and in the gloss. Grammatical morphemes are generally rendered by abbreviated grammatical category labels which are printed in small capitals ( CAPITALS ). A list of abbreviations is given at the end of this section. Instead of category labels a word from the metalanguage can also be used ( 1PL or we ; ART or the ). When a single object-language element

36 Chapter 1: Introduction is rendered by several meta-language elements (words or abbreviations), these are separated by periods (. ; e.g.: 2PL.PST ). Person and number are not separated by a period ( 1PL ). Deviations from the LGR: The LGR is not applied entirely in every example. I include a minimum of information along the lines proposed in López (2008:17): Clitics (i.e. weak pronouns) in Catalan may appear in an invariant form (e.g. the adverbial pronouns en or hi) or they may inflect for φ-features (number, gender, and person). I include the minimum of information in order that a clitic can be easily identified with the associate: the Catalan clitic for the masculine, plural direct object, els, is glossed only as CL.ACC, so the reader sees it is associated with the accusative argument. Additionally, clitics and their associates are co-indexed in the original example. This becomes important in cases where the dislocated element is not marked with Case (i.e. where they are merely glossed by CL ). Verbal morphology may inflect for tense, person, number, aspect, and mood. However, if a verb is infinite, it is glossed by.inf, otherwise it is finite (Subjunctive is glossed by SBJ; aspect is not of relevance here). If I do not indicate tense on a finite verb, it should be understood to be present. Likewise, if person and number are not indicated on the finite verb, it should be understood to be third singular (3SG). Determiners, for example, are not glossed for grammatical morphemes. The indefinite, plural article for masculine nouns uns is rendered by the word form a of the meta-language English. The English translations are nearly always in a neutral word order. As a native speaker of German, I do not try to reproduce the information structure of the original example (mostly Catalan) in the English translation. In general it is difficult to translate the information structure. For example is the Romance CLLD identical to that of an English topicalization (as proposed in Rochemont 1989:154ff.); or is CLRD better translated into an as for -construction (e.g. Frascarelli 2000:160:ex.206b) or into a cleft-construction (e.g. Cecchetto 1999:58:ex.40)?! I find the translation into neutral word order is a prudent approach, and is the least misleading choice. With respect to quoted references and page numbers (as previous in the preceding paragraph), they should be understood as follows: Prieto et al. (2007:20&37) => Page 20 and page 37 of Prieto et al. (2007) Frascarelli (2000:160:ex.206b)=> Example 206b on page 160 of Frascarelli (2000) Astruc (2005:154:Table 4.3) => Table 4.3 on page 154 of Astruc (2005) 28

37 Chapter 1: Introduction List of Abbreviations: 1 first person 2 second person 3 third person ACC accusative CL clitic COND conditional COMP complementizer DAT dative F feminine FOC focus FUT future IND indicative INF infinitive LOC locative M masculine NEG negation, negative NOM nominative PL plural PRF perfect PRS present PST past PTCP participle REFL reflexive SBJ subjunctive SG singular TOP topic 29

38 Chapter 2 Fundamental Aspects of Catalan Intonation This chapter deals with the theoretical foundations assumed throughout the present study of Catalan prosody and intonation. The Cat_ToBI system (Prieto, Aguilar, Mascaró, Torres & Vanrell 2007) is taken as the basic model for transcription 1 and as in Cat_ToBI I assume the two prosodic levels Intermediate Phrase (ip) and Intonational Phrase (IntP, ɩ). The term prosodic phrase is used as a hypernym for ip and IntP. There are five different boundary cues used for detecting intonational boundaries: continuation rise, sustained pitch, preboundary lengthening, pauses, and complex boundary tones. Only the last two cues indicate ɩ-boundaries. Experimental data is not presented here, but there are several pitch tracks of recorded examples in order to illustrate the boundary cues. This chapter starts with a general introduction to the phonology of intonation and then quickly turns to the case of Catalan. In Section 2.1 the autosegmental-metrical (AM) model is presented, including the prosodic hierarchy and a brief introduction to its relation to syntax. The ToBI transcription system is presented in Section 2.2. While Section briefly introduces the main assumptions of the original ToBI system, in Section the Catalan ToBI system (Cat_ToBI) is introduced. The relevant prenuclear and nuclear accents as well as the assumed prosodic levels, intermediate phrase and intonational phrase, are presented. The next section (2.3) is exclusively devoted to the five boundary cues assumed for intonational phrasing in Catalan. Each single cue is explained in detail. Section 2.4 summarizes in detail the main assumptions discussed in this chapter. 1 The Cat_ToBI version of May 2007/February 2008 is taken as the reference. A strongly modified version is published online in June 2008 and will appear as Prieto (2009). The modifications are not considered in the present study. The latest versions differ from the earlier ones in two respects. There is no phrase accent category (i.e. no T-) anymore and only one type of boundary tone occurs to the right of intermediate and intonational phrase boundaries. The boundary tones now are H%, M%, and L%. The medial tone M% is also a new invention of the latest Cat_ToBI version. For example, what is transcribed in the latest version as LM% (cf. p. 14) is transcribed in the present work as L!H% (a low tone combined with a downstepped high tone).

39 Chapter 2: Fundamental Aspects of Catalan Intonation 2.1 The Autosegmental-Metrical (AM) model: 2 The autosegmental-metrical (AM) model forms the basis of each tone and break indices (ToBI) transcription system. For this reason, I explain the main assumptions of the AM model in this section. The three types of tones pitch accents, phrasal accents, and boundary tones are described. Subsequently, the main assumptions of the AM model are introduced: the autosegmental part as well as the metrical part and the prosodic hierarchy with its prosodic levels (intonational phrase, phonological phrase/intermediate phrase, prosodic word, and syllable). The main goal of intonational models is to explain the complexity and the diversity of F0 contours. The F0 contour is the fundamental frequency of an utterance, which can be measured physically in Hertz (Hz). Pitch is the perceived fundamental frequency, while intonation is the variation of the fundamental frequency when speaking. Intonational languages (such as English, Catalan, Spanish or German) use pitch to convey semantic/discoursal meaning (cf. Ladd 1996:7). Pitch indicates, for example, if the utterance is a question or a statement (in general, the former has a rising intonation, i.e. the pitch of the voice increases over time; the latter has a falling intonation, i.e. the pitch decreases with time). In these languages, pitch is not used to distinguish words as is the case in tonal languages (e.g. Chinese, Thai or Hausa) where, in contrast, the pitch within a word lexically distinguishes them. A third group of languages (such as Japanese or Swedish), called pitch accent languages, has an intermediate status: these languages have properties of both intonational and tonal languages (Hall 2000:271; Jun 2005). There are three main groups of models for intonation: I) hierarchical models (superpositional modals; cf. Fujisaki 1988, Möbius 1993 and others), II) tone sequence models (e.g. Pierrehumbert 1980 and subsequent work), and III) the British School of intonation description (e.g. Crystal 1969, Halliday 1967 and others). Hierarchical models are descriptions where accentual contours are superimposed on phrasal intonational contours ( Contour Interaction models, term by Ladd 1983). Tone sequence models assume that the string of tones contains lexical and phrasal tones (cf. Bruce 1977); hence there are no different tiers for different tones. Tonal information can be exactly localized on single syllables and phrase edges. In the British School pitch modulation is captured as pitch configurations. Configurations such as fall, rise, rise-fall and so on are taken as the primitives, in contrast to tone sequence models, where 2 This chapter is based mainly on Gussenhoven (2004:#7) 31

40 Chapter 2: Fundamental Aspects of Catalan Intonation single tones are combined in order to represent a rise, a fall etc. 3 In the British School the most relevant tonal information is localized on the nucleus. 4 The present study is based on the tone sequence model of intonation. For this reason, I focus only on this model in the further description. The tone sequence model emerged from previous American structuralist approaches from the 1940s (e.g. Pike 1945, Trager and Smith 1951; cf. Gussenhoven 2004:#7.2.4 for details). It combines autosegmental phonology and metrical phonology for tonal association with texts (cf. Pierrehumbert 1980, Ladd 1996, Gussenhoven 2004) and it represents the intonation of an utterance as a linear order of abstract (phonological) entities. It generates the pitch contour from a succession of phonologically distinct tones located in one single tier. There are three types of tonal units: pitch accents, phrasal accents and boundary tones. 5 They are briefly introduced here. 6 A short illustration is presented in example (1) and (2) as well as in Figure 1 below. A more detailed presentation of the tones follows in Sections 2.2 and 2.3. Pitch accents associate with metrically strong syllables of a (prosodic) word. The tone which associates with the stressed syllable is marked by an asterisk * (cf. for example L* in (1); for details on association cf. Pierrehumbert 1980:11 and Arvaniti et al. 2000). Pitch accents are strictly locally determined, do not interact with each other (i.e. they are independent from each other), and are categorically distinct. It is assumed that only two tones, namely H(igh) and L(ow), suffice to describe a language (Bruce 1977, Pierrehumbert 1980, Gussenhoven 2004). The tones can either be monotonal, i.e. they consist of a single High tone (H) or a single Low tone (L), or they can be bi- or tritonal. Bitonal pitch accents are a combination of two low and high tones (e.g. LH or HL) and can either have a leading tone before T* (e.g. HL*) or a trailing tone after T* (e.g. L*H). Tritonal pitch accents are a combination of three tones (e.g. LHL). Boundary tones mark the edge of an intonational phrase (IntP or ɩ; cf. presentation below (4)). They can appear at the beginning or end of the IntP. In contrast to pitch accents, they are independent of stressed syllables, since the end (or the beginning) of an IntP does not always coincide with a stressed syllable. Boundary tones serve as a demarcation; they are assigned for structural reasons (i.e. to signal the prosodic constituent IntP) and not for prominence reasons. 3 Cf. Ladd (1996:82) for attempts at converting Pierrehumbert s (1980) and British School s nuclear tones. 4 Cf. Grice & Bauman (2007) and references cited there for more information on the British School. 5 In anticipation of later conclusions, it should be mentioned here that the term phrase accent will be replaced by the term ip-boundary tone and the term boundary tone will be replaced by IntP-boundary tone. 6 The presentation partly follows the one in Kügler (2007:5). 32

41 Chapter 2: Fundamental Aspects of Catalan Intonation Boundary tones are labeled with the diacritic % in order to differentiate them from other tones. A single low boundary tone is represented as L% and a high one as H%. Just as with pitch accents, boundary tones can be monotonal (as in English (Pierrehumbert 1980) or German (Féry 1993) or Catalan (Prieto et al. 2007)) or they can be complex, i.e. consist of a sequence of tones (as in Seoul Korean (Jun 1993) or Bengali (Hayes & Lahiri 1991)). The separation into pitch accents and boundary tones already existed in Trager and Smith (1951) who used the term juncture phonemes for boundary tones (cf. Gussenhoven 2004:126). Phrase accents are currently taken as tones which signal a further boundary, namely the boundary of a prosodic constituent below IntP (Grice et al. 2000). In the present study, this constituent is the intermediate phrase (ip) and thus a phrase accent is called ip-boundary tone in the remainder of the work. The general idea of having two tones signaling boundaries goes back to Pierrehumbert (1980), inspired by Bruce (1977). Pierrehumbert (1980) took the phrase accent at that time to account for the course of the pitch curve between the nuclear accent and the boundary tone (cf. (1)). She detected that in many contours two targets could be identified after the nuclear accent (the last pitch accent). Example (1), from Gussenhoven (2004:127), shows that the L* pitch accent is followed by two High tones, one after the accented Manitowac and one at the end of the sentence. The phrase accent (H-) is placed where the detectable change occurred in the contour. 7 Phrase accents are labeled with the diacritic - in order to separate them from the other tones. A single low phrasal accent is represented as L- and a high one as H-. (1) Example of two postnuclear tones signaling boundaries (Gussenhoven 2004:127 (originally from Pierrehumbert 1980)) {[Does Manitowac have a bowling alley ]} L* H- H% Pierrehumbert s seminal thesis in 1980 presents the starting point of modern intonational phonology (in terms of an autosegmental-metrical tone sequence model). Her finite-stategrammar of well-formed intonational phrases (Figure 1) shows firstly, that the tones are in a linear order and secondly, how tones are allowed to combine in order to generate a well-formed intonational phrase (in American English). Her grammar separates the phonological representation from its phonetic implementation. According to Figure 1 a well-formed IntP in 7 This assumption was changed in Beckman & Pierrehumbert (1986) and is presented below. 33

42 Chapter 2: Fundamental Aspects of Catalan Intonation American English can start with a boundary tone T% (which can be High or Low). This tone is followed by one or more pitch accents, which are obligatory. After that there is a phrasal accent T- (High or Low) preceding the terminal boundary tone T% (High or Low), which signals the end of the IntP. Figure 1 thus shows the generative nature of the AM model quite nicely: there is a limited set of elements (i.e. tones) and based on these elements rules can generate an infinite number of structures. Figure 1: Finite-state-grammar of well formed American English intonation phrases (Pierrehumbert 1980:29, figure taken from Ladd 1996:81). Pierrehumbert s (1980) original description of American English was somewhat revised in Beckman & Pierrehumbert (1986). The model was consolidated as a general theory of intonational representation in Pierrehumbert & Beckman (1988). During this evolution some tones were modified, changed or deleted. 8 One important change is the reanalysis of the phrase accent H-: Beckman & Pierrehumbert (1986) introduced the intermediate phrase (ip) as an additional constituent of the prosodic hierarchy, which is rather intonationally defined like the IntP (whereas the phonological phrase level was defined on segmental processes which only apply within this constituent (cf. N&V 1986/2007:ch.6)). They reanalyzed the tone T- as a boundary tone of ip. As a result, ips ended in T- and IntPs in T-T%, since the right edge of every 8 For details, cf. Ladd (1996:273ff.) and Gussenhoven (2004:130ff) or the original papers directly. 34

43 Chapter 2: Fundamental Aspects of Catalan Intonation IntP coincides with a right edge of an ip. This is shown in (2). The high contour on postnuclear syllables (fat lines) is still determined by H- and H%, but the former phrase accent H- of (1) is no longer located between L* and H% but combines with H% to the complex boundary tone H-H% which is placed at the end of the sentence (i.e. at the end of IntP). (2) Reanalyzing of H- as an ip-boundary tone {[Does Manitowac have a bowling alley ]} L* H-H% The autosegmental and the metrical side of the AM model: Pierrehumbert s (1980) work is based on both the models of metrical phonology (Liberman & Prince 1977) and of autosegmental phonology (Leben 1973, 1975; Goldsmith 1976). The term Autosegmental-Metrical (AM) Model has been customary since the mid-1990s (a term coined by Ladd 1996), and the model is currently one of the most used phonological frameworks for representing intonation (Hualde 2003, Prieto 2003, Jun 2005, Grice & Baumann 2007). It combines the two abovementioned phonological areas in the following way. Due to the fact that this model has separate tiers for tones (H,L) and for segmental phonemes, it is autosegmental (cf. Gussenhoven 2004:#3.3, Hall 2000:#5). Tones are not taken as inherent properties of a segment, but as entities which behave phonologically independently of their segments. Tones are autonomous segments; in other words, they are autosegments. The autonomous segmentation of tones and segments can be seen in (4) by the two different tiers for segments and for tones (segmental structure and tonal structure). The element in the segmental structure to which tone (T) associates is called the Tone Bearing Unit (TBU). Languages differ in whether a vowel / consonant or the mora (μ), the syllable (σ), or the foot (F) counts as the unit to which tone is assigned. For example, Catalan and Spanish assign tones to metrically strong syllables (σ s ) and in Japanese the mora is the relevant unit. The representations in (3) show the advantage of the mutual independence of the tone and the segmental element, here the syllable. 35

44 Chapter 2: Fundamental Aspects of Catalan Intonation (3) Different associations between segmental elements and tones (taken from Gussenhoven 2004:29) a. σ σ b. σ c. σ σ T T T T T one-to-one contouring multiple association Example (3a) is a non-autosegmental representation of tones. If only this kind of representation were needed, a tone could never be independent of the segment, because it would be an integral part of the syllable (or the vowel in it). Consequently, in deletions or insertions of segments, the tone would also be deleted or inserted. However, Leben (1973) and Goldsmith (1976) have given good arguments for the autosegmental status of tones (in tone languages; cf. also Hall 2000:#6.3). For this reason, autosegmental phonology allows for representations (3b,c): a syllable may disappear while the tone remains (3b) or a tone may disappear while the syllable remains (3c). The latter represents the continuation of the pitch value of a tone on an adjacent TBU. The contour tone (3b) frequently arises at the edges of words or phrases (cf. Gussenhoven 2004:28f). Although the autosegmental framework was originally developed for tonal languages, Pierrehumbert (1980) as well as Goldsmith (1980) propose to adapt it to English, an intonational language. Due to the influence of Pierrehumbert s (1980) AM model, the autosegmental representation of tones in intonational languages is nowadays quite common (e.g. Féry 1993, Uhmann 1991 for German; Beckman et al for Spanish; cf. also the references for language specific ToBI systems cited below). The model is metrical because the utterance is divided into phrases and relative prominence is assigned to elements within the phrase. The division is characterized by a set of phonological constituents which are hierarchically organized (4). The hierarchy is called Prosodic Hierarchy (Selkirk 1984, Nespor&Vogel 1986&2007) and is described quite nicely by Truckenbrodt (1999): In this theory, a hierarchical, layered representation structures a string of phonological segments, grouping segments into syllables, syllables into feet, and feet into prosodic words; the layers above the prosodic word are the phonological phrase, the intonational phrase, and the utterance (Truckenbrodt 1999:220). Tones, which are organized into pitch accents, phrasal accents and boundary tones, make references in several ways to these prosodic constituents. 36

45 Chapter 2: Fundamental Aspects of Catalan Intonation (4) Prosodic Hierarchy (adopted from Gussenhoven 2004:124:1) υ Utterance ι ι Intonational Phrase φ φ φ Phonological Phrase ω ω ω ω ω Prosodic Word F F F F F Foot σ σ σ σ σ σ σ Syllable tu: mε ni kʊks spɔɪl ðə brɔθ H* L*+H H* H*+L Lι segmental structure tonal structure In (4) one possible pronunciation of the English saying Too many cooks spoil the broth is given and it illustrates the different levels of the prosodic hierarchy. 9 It is adopted from Gussenhoven (2004:124) but it differs from the original by using two different lines (dashed and solid) in order to represent the difference between constituency lines and association lines (cf. Brentari & Bosch 1990:2f). Dashed lines represent association lines. Association refers to the relationship between an autosegment and the metrical position that licenses it (Beckman 2006:1). Solid lines represent prosodic constituency. They connect the constituents of the prosodic hierarchy and thus represent category membership. The different constituents of the prosodic hierarchy are motivated by 9 Some phonologists (such as Gussenhoven 2004 and Féry 2004:144 (and references cited there)) assume that the Strict Layer Hypothesis (Selkirk 1984, N&V 1986/2007) can be violated under certain circumstances (e.g. Roca 1997). In the case of the definite article in (4), the direct connection of the syllable with the level of the phonological phrase is such a violation. Strict Layer Hypothesis (SLH, Nespor&Vogel 1986:7): i. A given nonterminal unit of the prosodic hierarchy, X p, is composed of one or more units of the immediately lower category X p-1. ii. A unit of a given level of the hierarchy is exhaustively contained in the superordinate unit of which it is a part. iii. iv. The hierarchical structures of prosodic phonology are n-ary branching. The relative prominence relation defined for sister nodes is such that one node is assigned the value strong (s) and all other nodes are assigned the value (w). 37

46 Chapter 2: Fundamental Aspects of Catalan Intonation representing domains in which certain phonological processes apply (N&V 1986/2007). In what follows, the different levels starting from the syllable are introduced. I concentrate on the main aspects of these prosodic constituents and do not present a detailed explanation of the phonological processes motivating them (for details, cf. N&V 1986/2007). Metrical phonology deals with prominence relations. In intonational languages a certain syllable (σ) within a word is stressed, whereas the others are unstressed. Hence, the former is more prominent than the latter. This pattern is defined lexically. The prominence relations are clarified by indicating adjacent syllables according to their relative stress pattern as strong ( s ) or weak ( w ). Consequently, the syllable σ s is more stressed than σ w. Only strong syllables are anchor points for pitch accents. A syllable consists of a sequence of speech sounds and is typically made up of a nucleus (for the most part a vowel) with optional initial elements ( syllable onset ) and optional final elements ( coda ), mostly consonants (Hall 2000:#8.6). The foot (F) is motivated as a phonological constituent due to word stress rules (cf. N&V 1986/2007:#3.2 and references cited there; Hall 2000:277ff.). A foot consists of a stressed (i.e. strong) and an unstressed (i.e. weak) adjacent syllable, which builds the underlying rhythm (trochee: strongweak; iamb: weak-strong). The level of prosodic word or phonological word (PW, ω) dominates the foot level and is thus composed of one or more feet. Also the unit of ω is motivated on the grounds of being a domain where certain phonological processes apply (cf. N&V 1986/2007:#6 for details). According to Vigário (2003:22) one of the most intuitive diagnostics 10 for the prosodic word domain in Portuguese (and many other languages) is primary word stress. It is generally accepted that the prosodic word must bear one and only one primary stress; consequently, elements which cannot bear primary word stress do not count as a prosodic word. These elements are called clitics and they are incorporated into the prosodic word, which hence is an agglutinating entity: Any element that is not dominated by a prosodic word node must be adjoined to the following prosodic word (Vigário 2003:210&263). For Catalan, a language that has lexical stress, the basic idea is by and large clear; a stressed non-compound lexical word together with any adjacent clitic counts as a prosodic word. Elements in Catalan which cannot bear primary word stress are inter alia prepositions (which are 10 For a comprehensive overview of diagnostics for the prosodic word cf. Vigário (2003:22). 38

47 Chapter 2: Fundamental Aspects of Catalan Intonation often also asyllabic), determiners, and (weak) pronouns. 11 Strong pronouns are stressed; hence they are not clitics. In general, it is easy to check if an element is unstressed in Central Catalan due to the property of vowel reduction: A reduced vowel is unstressed (on the other hand, full vowels are always stressed; cf. tables 1 and 2 in Section 1.1.1). Clitics cannot be utterances because utterances need stress (e.g.: Whom did you see? - *La / Ella / La viu Her / She / I saw her ). Definite articles, for example, are clitics since their vowel is reduced ([əl 'nεn] el nen the child ) and the vowel is often epithetic (l amic the friend ; the /l/ cannot be a ω, because there is no vowel). Prosodic words in Catalan consist of at least a moraic trochee (Cabré 1993) 12 and the primary word stress falls on the rightmost stressable element (Wheeler 2004:1). The phonological phrase (PhP, φ or ϕ) or the intermediate phrase (ip) is the level above the prosodic word. The domain of PhP is often linked with syntactical lexical projections (N&V 1986/2007:168, Truckenbrodt 1999). The term intermediate phrase stems from Beckman & Pierrehumbert (1986) and is similar to the constituent N&V (1986/2007) referred to as phonological phrase. 13 However, whereas the unit of PhP is rather based on postlexical phonological rules (such as Raddoppiamento Sintattico, cf. N&V 1986/2007:165) and thus constitutes a domain where phonological processes apply, the ip is principally motivated by intonational aspects. 14 Due to the fact that until now there has been no evidence in Catalan for phonological processes applying in this domain, I use the term intermediate phrase in the remainder of the work. The ip groups (phonological) words into one phrase which have at least one accented syllable, i.e. each ip contains at least one pitch accent. Phrasal accents, i.e. ipboundary tones, (T-) signal the boundary of an ip. Sometimes there is another level assumed inbetween the ip and the phonological word: the accentual phrase. Beckman & Pierrehumbert (1986) find clear evidence for this constituent in Japanese, but less compelling evidence in English. Since this unit does not seem to be relevant in Catalan (cf. Cat_ToBI), I have left it aside. The intonational phrase (IntP, ι) is formed by grouping together one or more ips. The IntP constitutes the domain of an intonational contour and its edges are positions where pauses are 11 There are two types of pronouns in Romance. In general, the first type takes a position adjacent to the verb and the second type occupies the canonical object position. The first type is called weak or clitic pronoun because they are not stressed and they phonologically depend on the verb. The second type is called strong pronoun, and they are stressed. (cf. Fernández Soriano 1993:13&22; cf. also introduction to clitic left- and right-dislocations in Chapter 1) 12 But cf. Wheeler (2005:277) who states that there is a discussion on whether the Catalan foot is an iamb or a trochee. 13 Hayes & Lahiri (1991), for example, suggest that the intermediate phrase is equivalent to N&V s (1986/2007) phonological phrase (cf. also Ladd 1996:93). 14 Cf. Section

48 Chapter 2: Fundamental Aspects of Catalan Intonation able to appear (N&V 1986/2007:188; Hall 2000:310). The role of pauses led researchers to postulate a systematic relation between certain syntactic constructions and intonational phrases. Parenthetical expressions and nonrestrictive relative clauses, for example, are delimited by pauses and thus seem to form an intonational domain of their own (N&V 1986/2007:188). Much attention has also been paid to syntactic (root) clauses as an important factor that plays a role in the formation of IntPs (cf. Downing 1970, N&V 1986/2007, Truckenbrodt 2005; cf. also Selkirk s 2005 Comma Phrase ). 15 On the other hand, there are also phonetic indices such as the so-called continuation rise (cf. below) and semantic conditions (cf. Selkirk s 2005:43 sense unit ) on intonational phrasing (Selkirk 1984:287ff., Selkirk 2005). Furthermore, IntPs can vary with factors such as the rate of speech or the length of constituents (N&V 1986/2007:193). Exactly as with intermediate phrases, IntPs are delimited by (initial and final) boundary tones. The tonal unit is labeled by the symbol % (i.e. T%) (cf. Beckman & Pierrehumbert 1986, Selkirk 2000:566). Initial boundary tones are optional in English as well as in Catalan and Spanish. The highest unit in the prosodic hierarchy is the Utterance (U, υ). It is composed of at least one IntP if the sentence consists of only one IntP. However, the utterance can also consist of more than one sentence. The utterance is not of interest in the present study and for this reason I refer to N&V (1986/2007:#8) for details. The syntactic grounding of prosodic constituents: The last three units, the utterance, the intonational phrase, and the intermediate phrase (phonological phrase), are all located above the word level. The literature on prosodic grouping above the word level mainly concentrates on IntP and ip and, as shortly mentioned above, syntactic structure influences this prosodic structure. Chomsky & Halle (1968) assumed that the phonological information is contained within or at least derived from syntactic trees of the surface structure. The theoretical framework at present (that Selkirk 2005:31 described as the theory of the syntactic grounding of prosodic categories ), though, assumes that the relation between syntax and phonology is indirect. It proceeds from the assumption that the hierarchy of prosodic constituents is separate from the S-structure trees, but that its prosodic constituents show systematic relations to syntactic constituent structure. These systematic relations are characterized by Selkirk (1986 and subsequent work), N&V (1986/2007), and Truckenbrodt (1995, 1999, 2005). Selkirk (1986) and N&V (1986/2007) present approaches for constructing 15 The role of root sentences is explained in the analysis of Chapter 3. 40

49 Chapter 2: Fundamental Aspects of Catalan Intonation prosodic constituents: relation-based mapping (N&V 1986/2007), (1), and edge-based mapping (Selkirk 1986), (2). Truckenbrodt (1995, 1999) presents an account that calls for a syntactic constituent to be contained within a prosodic constituent (Wrap theory). In this study, the edgebased (or end-based) theory and the Wrap theory are assumed. (1) Relation-based Mapping (Nespor & Vogel 1986/2007) A phonological phrase contains a lexical head and elements on its non-recursive side that are not themselves lexical heads. (2) Edge-based Mapping (Selkirk 1986) The right / left boundary of a prosodic constituent of the category C corresponds to the right / left boundary of a morphosyntactic category X. Whereas the relation-based theory concentrates on syntactic heads, the edge-based theory concentrates on (edges of) maximal projections. In Selkirk s approach, the right or the left edge of a certain syntactic constituent type matches up (i.e. aligns) with the right or left edge of a corresponding prosodic constituent type. The function of Truckenbrodt s (1999) Wrap constraint is to suppress the effect of Selkirk s alignment constraints because it prevents syntactic XPs from being split up into multiple prosodic constituents. The exact application of these theories is explained in detail in Section 3.4. At that time, non-syntactic factors that influence the prosodic phrasing and that help to account for nonisomorphism between syntactic constituents and prosodic structure are also discussed. To summarize, the AM model assumes that intonation has a phonological organization couched in a prosodic hierarchy. Intonation is described as a sequence of distinctive tonal units (High (H) and Low (L), and their combinations), and its contour is represented linearly by an autosegmental string of tones. These tones are associated either with a specific syllable (pitch accents) or with a specific location in a phrase (ip-boundary tones, IntP-boundary tones). This association marks the prominence relations among the words and the prosodic groupings of an utterance. Through phonetic realization rules (cf. Pierrehumbert 1980:25ff., Gussenhoven 2004:#7.2.4, Kügler 2007:7), the phonological representation of tones is mapped onto an actual phonetic representation, i.e. a F0 contour. Both the phonological representation and the phonetic realization are language specific This passage is mainly based on Jun (2005:2). 41

50 Chapter 2: Fundamental Aspects of Catalan Intonation 2.2 A ToBI System for Catalan The AM models of the 1980s were instantiated in the 1990s by a new transcription system, called ToBI. In this section, the main tenets of the original ToBI transcription system for English are presented (#2.2.1). The remainder of Chapter 2 concentrates only on Catalan. Section presents the Catalan version of the ToBI system and its main tenets are explained in detail. At the end, the two suprasegmental levels, intermediate phrase and intonational phrase, are shortly introduced. This introduction sets up the basis for the detailed presentation of the boundary cues in Section ToBI: In 1992 Silverman et al. (among the authors were also J. Pierrehumbert and M. Beckman) replaced the by and large abstract tonal grammar of Beckman & Pierrehumbert (1986) with a practically oriented transcription system called ToBI ( Tone and Break Indices ). The prosodic structure of a language is not only defined by its tonal patterns (intonational phonology), but also by the degree of juncture (i.e. prosodic separation) between any two adjacent words (Jun 2005:2). Both aspects are reflected in the ToBI system. The syllable To in ToBI represents the part of the intonational phonology, i.e. distinctive tonal events. The syllable BI represents the part of the perceived prosodic separation. As for the latter, ToBI annotates prosodic boundary strength (degree of juncture) by using a five-point scale from 0 for the boundary between a word and a cliticized form up to 4 for signaling a boundary between intonational phrases (Ladd 1996:96; Gussenhoven 2004:132). The five possible boundary strengths on the ToBI break index tier are shown in Table 1. Break Index Description 0 no word boundary 1 word boundary 2 strong juncture with no tonal markings 3 intermediate phrase boundary 4 intonational phrase boundary Table 1: The ToBI break index system 42

51 Chapter 2: Fundamental Aspects of Catalan Intonation The numerical break indices 3 and 4 represent those parts of the hierarchy of the prosodic groupings ( prosodic hierarchy ) that are relevant to my study. For this reason, they are the only break indices used throughout the remainder of this work. The three remaining break indices are not annotated. BI 3 is used to signal an ip-boundary and it appears on the break index tier any time a phrase tone is indicated on the tonal tier (cf. Ladd 1996:96f). The same applies to BI 4 with respect to an IntP-boundary. The prosodic model in ToBI is a phonological model, not a phonetic model; ToBI is consequently language specific, and the system of one language is not appropriate for describing another language s intonational system. Originally, ToBI was a transcription system for mainstream American English. Over the years, it has become very well known and many language specific versions have been developed since the early 1990s based on the principles and annotation conventions of the 1992 model (for German: G_ToBI, Reyelt et al. 1996, Grice et al. 2001/2005; for Slovak: Sk_ToBI, Rusko et al. 2007; for Spanish: SP_ToBI, Mendoza-Denton 1999, Beckman et al. 2002, Sosa 2003; for Korean: K_ToBI, Jun 2000; for an overview cf. Jun 2005). Due to this development, the original ToBI was renamed in (M)AE_ToBI ((Mainstream) American English ToBI) following the general practice of researchers of putting the initials of the language in front of the term ToBI. Four layers of labeling ( tiers ), aligned with the appropriate speech signals were specified in the original ToBI conventions: words, tones, break indices, and miscellaneous information. Since the information that can be labeled is not fixed, the quantity and types of tiers mirror the language specific prosodic system as well as the interests of the researchers (Jun 2005:3) Cat_ToBI: A ToBI system of prosodic transcription for (all varieties of) Catalan is relatively new. Pilar Prieto and colleagues established a version for Catalan in recent years and published it online in 2007 (Prieto et al. 2007). It is called Cat_ToBI ( Catalan Tone and Breaks Indices ) and is a proposal for the prosodic annotation of prosodic and intonational information of spoken language. In earlier work (Prieto 1995, Estebas-Vilaplana 2000, Astruc 2005) it was assumed that the English ToBI system developed by Silverman et al. (1992) generally also works for Catalan if some required modifications are considered. Like ToBI systems in general, Cat_ToBI is also based on the autosegmental-metrical model and it gives information on the language specific 43

52 Chapter 2: Fundamental Aspects of Catalan Intonation tonal patterns and the degree of juncture between adjacent words. In what follows, I describe the Catalan tonal system and the prosodic levels marked by the numerical break indices 3 and 4. Central Catalan has five pitch accents. Besides the two simple tones L* ( low ) and H* ( high ) there are three bitonal accents 17 : H+L*, L*+H, and L+H* (Prieto et al. 2007; cf. Figure 2). 18 The H tones of the pitch accents can acquire etiquettes for upstep ( T) and downstep (!T) and delayed peak (>H*). This minute difference possible for a single pitch accent is presented in Figure 2 in the last column ( distribution ). The symbol σ s represents the strong, i.e. stressed, syllable and σ w represents weak, i.e. unstressed, syllables. Tonic syllables are accented strong syllables and weak syllables following tonic syllables are so-called post-tonic syllables. The phonetic difference of the three bitonal accents in Figure 2 is signaled by both the different association of the star * with σ s and by the different leading and trailing tones. The difference between the two rising tones L+H* and L*+H lies on the position where the rise starts (cf. Figure 2). In L+H* the tonal movement starts at the beginning of the stressed syllable ( early rise ), whereas in L*+H the tonal movement starts at the end of the stressed syllable and continues during the following syllable ( late rise ). According to Prieto et al. (2007:10), L+H* can be realized as L+>H* (cf. below) in prenuclear positions of declarative sentences. In this case, the peak is realized on the following unstressed syllable, i.e. the peak is delayed. 19 The falling tone H+L* is characterized by a fall which starts at the beginning of the stressed syllable (cf. Figure 2). L* is realized as a low tone sustained throughout the production of the stressed syllable, whereas H* is realized as a high tone throughout the production of σ s. The etiquettes concern only high tones. In the case of a downstep (!H), the high tone is realized lower than the preceding high tone, whereas in the case of an upstep ( H), the high tone is realized higher with respect to the preceding one. Finally, >H* indicates that the tonal peak is displaced after the stressed syllable (delayed peak; Prieto et al. 2007). For specific examples of the described tones in Figure 2, see Prieto et al. (2007). 17 In Prieto (2009) L+>H* constitutes the fourth bitonal pitch accent. 18 There is a fourth bitonal accent mentioned in Cat_ToBI, namely H+L*. Due to the fact that this tone is detected only for the Balearic Catalan variety of Mallorca, I have left this tone aside. In addition, H* can also be upstepped ( ) or downstepped (!) and L+!H* is a variant of L+H*. I have left these variations aside and refer to Prieto et al. (2007) for detailed information. Astruc (2005:xviii) includes further bitonal accents. Since I rely mainly on Cat_ToBI as established in Prieto et al. (2007) I have also left them aside. 19 Cf. discussion of Figure 3 below. 44

53 Chapter 2: Fundamental Aspects of Catalan Intonation representation tone name distribution L* low Nuclear accent of absolute questions, which rise, and of neutral declaratives H* high H*: Nuclear accent of partial questions and of exhortatives!h*: Prenuclear accent of declaratives H*: Nuclear accent of absolute and partial questions L+H* early rising accent L+H*: Prenuclear accent of declaratives (often L+>H*), and of imperatives L+H*: Nuclear accent of imperatives and emphatic declaratives and of vocatives L+!H*: Nuclear accent of partial questions L*+H late rising accent (Initial) Prenuclear accent of absolute questions, which rise, and of exhortative phrases H+L* (early) falling accent Nuclear accent of falling absolute questions and of vocatives Figure 2: Catalan Pitch Accents (Cat_ToBI, based on Prieto et al. 2007) Figure 2 also gives information about the distribution of the tones described in Cat_ToBI (cf. last column). In italicized letters are those accents that are relevant for questions. The distribution of the tones which are relevant for declarative sentences according to Prieto et al. (2007) are bold faced. L* is the typical nuclear accent (the tone on the rightmost stressable syllable) for neutral declaratives and L+H* for emphatic declaratives. Typical prenuclear accents are (!)H* and the early rising tone L+H*. The latter is often realized as L+>H*. In addition to these explicitly mentioned declarative prenuclear accents, two further prenuclear tones can be added by looking at the transcription of the declarative examples in Cat_ToBI: the falling accent H+L* (cf. example 1a3Eiv (Prieto et al. 2007:4)) and the late rising accent L*+H (cf. example 1a2-StEul (Prieto et al. 2007:10)). 45

54 Chapter 2: Fundamental Aspects of Catalan Intonation Catalan Prenuclear Accents - Pitch accent alignment As mentioned above, Prieto et al. (2007) say that the peak of L+H* can also appear after the stressed syllable. The symbol > represents this displacement. Now, the question arises as to how one can differentiate between L+>H* and L*+H. In both cases the peak is realized after the stressed syllable. The difference in alignment of H is not new for Catalan. Although in Cat_ToBI (Prieto et al. 2007) there is no explicit explanation of the difference given, Prieto et al. (2005) have already discussed alignment contrasts of prenuclear rising accents (in Catalan, but also in Neapolitan and Pisa Italian). Whereas for Spanish the two-way alignment contrast between early rising accent and late rising accent can be captured by L+H* and L*+H respectively (Face 2001, Hualde 2002), this transcription cannot capture existing Catalan contrasts. Prieto et al. (2005) state that there are four different realizations of rising LH pitch accents in Catalan (Figure 3): a rise with delayed peak (a), a rise with peak aligned to the end of the syllable (b), a rise with peak aligned to the end of the prosodic word 20 (c), and a post-tonic rise (d). The difference can be accounted for by encoding the alignment phonologically by means of phonological anchoring, which is an extension of the concept of secondary association established in Pierrehumbert & Beckman (1988). The star * represents the so-called primary association, which encodes the basic affiliation between the whole accent and its tone bearing unit (TBU). Secondary association encodes the exact alignment based on metrical grounds. According to Prieto et al. (2005) the Catalan rise can either be anchored to prosodic edges on the level of syllables (σ) or of prosodic words (ω). (a) Rise with delayed peak L+>H* (b) rise with peak aligned to end of σ (c) rise with peak aligned to end of ω (d) post-tonic rise L*+H Figure 3: Schematic representation of Catalan rising LH pitch accents (based on Prieto et al. 2005:372) This is proved wrong in Prieto (2007). 21 The schematic representation in Figure 3 shows differences in the height of the accents, but this is irrelevant. There are no such height differences in the original figure (cf. Prieto et al. 2005:372). 46

55 Chapter 2: Fundamental Aspects of Catalan Intonation The difference between L+>H* and L*+H lies mainly in two properties: The former is perceived as a high accent (starting from low), whereas the latter is perceived as a low accent (rising to high). The contour of the former should start rising within the domain of σ s, (cf. Figure 3(a)), whereas the contour of the latter rises only afterwards (cf. Figure 3(d)). The accents (a), (b), and (c) in Figure 3 are all transcribed as L+H* basically, and only accent (b) and (c) are notated in Prieto et al. (2005) with further information about the exact phonological anchoring. (5) Transcription of Catalan prenuclear rising accents (Prieto et al. 2005:381): (a) L+H*, rise with delayed peak (b) L+H*]σ, rise with peak associated to the edge of the syllable (c) L+H*]ω, rise with peak associated to the edge of the prosodic word (d) L*+H, post-tonic rise The representations of the first three accents shown in (5) are different from those established in Cat_ToBI. Accent (a) and (b) are comprised under L+H*, whereas the rise with the delayed peak can be represented as L+>H*. If I transcribe a delayed peak, I use the Cat_ToBI transcription throughout the work 22 ; otherwise I simply use L+H*. Accent (c) does not exist in Cat_ToBI. Cat_ToBI follows the conclusion drawn in Prieto (2007), that there are no word edge tones in Catalan (cf. Section 2.3.2) The Catalan Nuclear Accent In Cat_ToBI there are two nuclear accents mentioned: L* (as the nuclear accent of neutral declaratives, and of absolute questions, which rise) and H+L* (as the nuclear accent of falling absolute questions and of vocatives). The nuclear accent lies on the last stressed syllable of the utterance in broad focus declarative sentences (e.g. Vallduví 1993, Estebas-Vilaplana 2003b) which is typical for Romance languages (e.g. Ordoñez 1997 and Zubizarreta 1998 for Spanish; Costa 1998 for Portuguese). Astruc (2005:159ff.) discusses the Catalan nuclear accent broadly, since there is still an ongoing discussion as to how to analyze the nuclear accent of declarative sentences (Bonet 1984, Prieto 2002b; Estebas-Vilaplana 2000, 2003b). The accent can be analyzed either as a low target L* (Bonet 1984, Prieto 1999, 2002a,b), as a high target, which is strongly downstepped, 22 The transcription L+>H* has the advantage that the exact position of the peak is not determined. As the reader sees in Section 2.3.2, Prieto (2007) says that the peak of a rising accent depends on the length of the word and does not (automatically) align with the word edge. For this reason, neither the notation L+H*]σ nor the notation L+H*]ω (cf. Prieto et al. 2005) is appropriate, because it determines the exact position by secondary association. 47

56 Chapter 2: Fundamental Aspects of Catalan Intonation i.e.!h* (Estebas-Vilaplana 2000) or as a bitonal accent H+L* (Astruc 1999). The nuclear accents differ with respect to the sentence type (declarative, question, imperative) and H+L* is mostly assumed to be the nuclear accent of falling absolute questions (Prieto et al. 2007:17, Astruc 2005:167). However, Astruc (2005) takes the nuclear accent to be bitonal; the phonological representation of the nuclear accent is H+L* which can receive specific phonetic realizations. According to Astruc (2005:159f.) there are different phonetic realizations which can be labeled as Contour A and Contour B, cf. Figure 4. The former shows a smooth and gradual fall spanning the intonational unit and the nucleus (p.159) and the latter shows a fall which has been already completed at the onset of the nucleus, and the nucleus is thus already low (p.160). Figure 4 illustrates these contours based on Catalan sentential adverbs with four syllables (1-4). Measurement point 5 represents the sentence boundary. The speaker can vary the scaling of F0 on the fourth syllable and thus he can vary the slope of the fall. This free variation is without a specific semantic function (Astruc 2005:160). These scaling differences cannot be accounted for by simply assuming L* but by assuming H+L* with secondary association of L*. Figure 4: The two main patterns of nuclear accents in Catalan (accented sentential adverbs; taken from Astruc 2005:161) The phonological representation of the nuclear accent as H+L* has the additional advantage of accounting for the similarities of both multi-accent and one-accent words. As Prieto (2002a:403, 2002b:191) says, in one-accent words the rise-fall pattern of the contour is concentrated on this syllable, as can be seen in Figure 5. In multi-accent words, though, the risefall pattern is spread over the whole word. Nevertheless, both cases sound perceptually similar (Astruc 2005:173). The pitch contour in Figure 5 can easily be accounted for by H+L*, whereas L* cannot explain the rise. 48

57 Chapter 2: Fundamental Aspects of Catalan Intonation Figure 5: Pitch trace of one-accented word (taken from Prieto 2002a:403) There are further advantages of the bitonal representation mentioned in Astruc (2005:175). What is of relevance to the present study is the fact that the nuclear accent of declaratives in Cat_ToBI is represented by L* and in Astruc (2005) as H+L*. The latter accent, however, is already included in the phonological inventory of Catalan (cf. Figure 2). It is the nuclear accent of falling absolute questions and of vocatives (Prieto et al. 2007:17&36). 23 Due to the fact that the topic of the study at hand is not an investigation into nuclear pitch accents, only the tone L* is used as the nuclear accent of Catalan declaratives Suprasegmental Prosodic Levels Prieto et al. (2007) claim two levels above the prosodic word (ω) in the prosodic hierarchy: intermediate phrase (ip) and intonational phrase (IntP, ɩ) (cf. Prieto et al. 2007:20&37). The authors follow Beckman & Pierrehumbert (1986:288) by assuming that socalled phrasal accents signal the boundary of an ip, whereas so-called boundary tones signal the boundary of an IntP. These tones count as separate units and each type of domain has thus its own edge tone. This view is adopted in (MAE-)ToBI and intonational phrases then end in a combined tone T-T% (cf. also Section 2.3.6). In Cat_ToBI, there are three ip-boundary tones: H-, L-, and!h- (p.20) and three ι-boundary tones: H%, L%, and!h% (p.23). The demonstration of combined phrasal accents and boundary tones in Cat_ToBI is limited to examples where they appear at the end of an utterance and they are mostly demonstrated by questions. However, example (4) shows that ɩ-boundaries can also appear clause internally; this is also found in previous work on ips and IntPs (e.g. Nespor&Vogel 1986, Frascarelli 2000 and Astruc 2005 among others). My work concentrates mainly on clause-internal intonational boundaries, namely boundaries after S or V in the experiments on SVO phrasing (Chapter 3), boundaries after clitic left-dislocations (Chapter 5), and boundaries after preverbal dislocated and non-dislocated 23 Vocatives (as Anna in Anna, your meal is ready ) are so-called extra-sentential elements (cf. Astruc 2005; among them are also adverbs as obviously in Obviously, she is right ; appositions as the king of England in Henry VIII, the king of England ; dislocated phrases as those girls in They acted well, those girls ; and parentheses as as we agreed in The paper, as we agreed, will be translated ). 49

58 Chapter 2: Fundamental Aspects of Catalan Intonation subjects (Chapter 6). An example of a clause internal high ip-boundary (continuation rise) is given below in Figure 6. An example of a clause-internal complex boundary is given in Section 2.3 (cf. Figure 19), where these kinds of boundaries are discussed in more detail. Figure 6: Continuation Rise (H-) in Catalan in the utterance Na Maria menja mandarines Maria eats mandarines (taken from Prieto et al. 2007) 24 The continuation rise represents the typical end of a prosodic boundary (cf. Section 2.3, Figure 9). In Figure 6, it is located clause internally, at the end of the subject Maria, and it is an ip-boundary, which receives the numerical break index value 3. The rise of the prenuclear accent L+H* continues rising until the end of the subject, where H- is located. The second tone is H+L* and triggers the F0-contour to fall. The nuclear accent, on the last stressed syllable of the utterance ( ri of mandarines), is the low tone L*. At the end of the utterance the combined low boundary tone L-L% signals the end of the sentence. Prieto et al. (2007) give seven possible combinations of ip- & ɩ-boundaries, which are partly characterized by! for downstep. The combined boundary tones are L-L%, H-H%, H-L%, L-H%, LH-L%,!H-!H%, and H-!H%. An example of the first boundary tone, which is very common in declaratives, is given in Figure 6. For detailed descriptions and appropriate F0- contours I refer to Prieto et al. (2007:23ff.). For this study, it is important that such complex boundary tones signal an IntP-edge (cf. explanation of Figure 19). 24 Actually, Figure 6 is an example of Balearic Catalan. The contour, however, would be the same in Central Catalan. 50

59 Chapter 2: Fundamental Aspects of Catalan Intonation In (Cat_)ToBI the terms phrasal accents and boundary tones are used. Due to the fact that both are associated with phrasal boundaries (the former signals an ip-boundary, the latter an IntP-boundary), I use the terms ip-boundary tone for what was called phrasal accents and ι- boundary tone (or IntP-boundary tone) for what was called boundary tones throughout this work. In my nomenclature, the term (intonational) boundary tone comprises both ip-boundary tone as well as ι-boundary tone. If I use the term boundary tone (without the preceding specification), I either make no specific claim about the exact level of phrasing or I simply refer to both levels at the same time (intermediate phrase and intonational phrase). The rest of this chapter is devoted to the presentation of boundary cues for the two prosodic levels in Catalan. 2.3 Boundary Cues in Catalan for Intonational Boundary Tones In Section 2.3, I present those boundary cues of Catalan, which are considered in the present work and based on which I analyzed the recorded data. It is important to note that the boundary cues are phonetic/phonological and as such they are not part of the grammar that accounts for the prosodic phrasing. The grammar-internal construction of the prosodic boundaries is done by the syntactic constraints of the syntax-phonology interface and by certain nonsyntactic constraints (cf. Chapter 3, Sections and 3.4.2). In 2.3.1, I present the main boundary cues of Romance languages of Frota, D Imperio, Elordieta, Prieto & Vigário (2007) and I start to concentrate on Catalan boundary cues by giving the frequency of the cues detected in the Catalan corpus of Frota et al. (2007). After that, I present the following cues in greater detail: continuation rise (2.3.2), sustained pitch (2.3.3), preboundary lengthening (2.3.4), pause (2.3.5), and complex boundary tones (2.3.6). Section deals with the organization of these boundary cues, concluding that the first three cues signal only an ip-boundary, whereas the last two cues signal an ɩ-boundary Romance boundary cues in Frota, D Imperio, Elordieta, Prieto & Vigário (2007) Frota et al. (2007) is in all probability the most important work on the phonetics and phonology of intonational boundaries in Central Catalan, in European Portuguese, Italian and Spanish. According to them, phrase boundaries in theses languages show one or more of the following boundary cues (cf. Frota et al. 2007:134): 51

60 Chapter 2: Fundamental Aspects of Catalan Intonation a. Continuation rise (the preboundary stretch is realized as a rise from/on the last stressed syllable into the boundary syllable) b. Sustained pitch (the preboundary stretch is realized as a rise on the last stressed syllable followed by a high plateau up to the boundary) c. The boundary is signaled by a High tone d. The boundary is signaled by a Low tone e. There is a pitch reset after the boundary at the beginning of the following phrase f. The F0 drops to the speaker s base level at the boundary g. There is a preboundary lengthening, and h. A pause is present at the phrase boundary (a pause is defined as a stretch of silence) Frota et al. (2007) use a corpus of 239 Catalan SVO sentences. These sentences are only a subset of the whole database. The authors decided to choose only uncontroversial cases of intonational phrasing, i.e. they did not included unclear or difficult cases, where boundaries were hard(er) to detect. In their 239 sentences not all listed boundary cues were detected. For Catalan they give the following frequency of the boundary cues (Table 2; cf. Frota et al. 2007:135): Cont. Rise Sustained Pitch Boundary Tone Pitch Reset Drop Base Level PB Length. H L Catalan Table 2: Frequency (in %) of boundary cues of the Catalan corpus in Frota et al. (2007) Pause In the Catalan corpus of Frota et al. (2007) 100% of the sentences are marked by a High tone which is realized as a continuation rise. For this reason there is neither a sustained pitch nor a Low boundary tone to be found in Frota et al. (2007). Both cues, though, are nevertheless possible in Catalan: Cat_ToBI states L- and L% as normal boundary tones (cf. Prieto et al. 2007:20&23) and a sustained pitch is found in my data (cf. Figure 13 below). According to Frota et al. (2007) prosodic breaks in Romance are predominantly marked by a High boundary tone and the preboundary stretch is predominantly realized as a continuation rise. This general statement can also be maintained for Catalan. In addition to these phonological and phonetic cues, preboundary lengthening and pauses are also relevant in Catalan. The former can be seen with a 52

61 Chapter 2: Fundamental Aspects of Catalan Intonation frequency rate of 100% in the corpus. The latter is shown below. I do not consider the remaining two cues: pitch reset 25 and the drop of the speaker s base level The Continuation Rise Estebas-Vilaplana (2000:115) gives a clear example of a continuation rise ( abrupt rise in her terms). In a subject with two anchor points for accents the second accent can be compared with the first one (cf. Figure 7). Les meves amigues eren bones nenes "My friends were good girls" Figure 7: Intonation indication for prosodic boundary (taken from Estebas-Vilaplana 2000:115; shaded grey and underlining signal stressed syllables) The continuation rise can be seen on the subject noun amigues friends (fem., pl.) in Figure 7. The F0 contour of the first pitch accent, which is located on the possessive pronoun meves my, rises and reaches its peak on the accented syllable (L+H*). After the peak the F0 contour descends due to the fact that the second accent is also L+H*: The F0 contour has to be low at the beginning of this accent. Even though the F0 contour on mi of amigues friends (fem., pl.) rises from low to high as on me of meves my, it does not descend after the accented syllable. The F0 contour continues to rise on the posttonic syllable gues of amigues friends. This rise into the 25 Pitch reset is also called simply reset and interrupts the downstep effect. A downstep lowers the F0-realization of the material following a pitch accent. In (i) the downstep can be seen between pitch accent 1, 2, and 3 and also between accent 4 and 5. There is no downstep between accent 3 and 4, though. It is often assumed that a prosodic boundary causes such a cancellation of the downstep effect (Selkirk & Tateishi 1991 among others): If there is a downstep in the realization of adjacent pitch accents, there is no prosodic break in-between them. However, if the downstep is interrupted by a reset, a prosodic boundary is probable. Figure (i) demonstrates a total reset since accent 4 is as high as accent 1. If accent 4 is higher than accent 3 but lower than accent 1, one speaks of partial reset. According to Frota et al. (2007), the number of pitch resets is only nearly one third of the 239 examined Catalan sentences. Only ratios of 0.90 or higher between the peak of the first accent of the first phrase and the peak of the first accent of the second phrase (in my example accent 4) were considered cases of reset (Frota et al. 2007:135) Figure (i): Downstep ( ) and pitch reset ( ) (taken from Ishihara 2007b:5) 53

62 Chapter 2: Fundamental Aspects of Catalan Intonation preboundary syllable is called continuation rise and indicates an intonational boundary (cf. Figure 12). It can, but need not, be obligatorily followed by a pause. Bolinger (1984:404) although not using the term continuation rise describes its characteristic well by saying that [t]here is something in the intonation itself that signals or can be used to signal incompleteness, and it is obviously the terminal rise. Hence, something is unfinished and there has to be a continuation. Previously, Delattre (1965:25) spoke of continuation for describing such incompleteness effects. 26 In my work, the High boundary tone (continuation rise) is one main cue to detecting prosodic breaks. For this reason, I shall discuss this criterion in more detail here. In Prieto (2007), the alignment of the H tonal target relative to the end of the word is examined. Rising accents (LH) are typical for Catalan (prenuclear) accents (cf. Cat_ToBI: the typical prenuclear rising accent of Catalan declaratives is L+H*). According to Estebas-Vilaplana (2003), such rises consist of a combination of L* pitch accent plus a High word edge tone, i.e. that H aligns with the right edge of a word. Prieto (2007) shows that the H location is sensitive to the position of the accented syllable within the given word though. She discovers a consistent trend: there is no strict word anchoring effect of H, but the peaks of word-final accented syllables (oxytonic words) are located after the end of the word (i.e. on the following word), whereas the peaks of word-medial accented syllables (in her case paroxytonic words) are located before the end of the word. (6) a. Compraven talls buy.3pl.past pieces They bought pieces b. Comprà ventalls buy.3sg.past fans S/he bought fans 26 Delattre (1965:20) introduces the terms minor continuation and major continuation for describing the intonational signals which serve as a clue to degrees of subordination within a given utterance (cf. also Bolinger 1984:410). 54

63 Chapter 2: Fundamental Aspects of Catalan Intonation Figure 8: Schematic diagram of the F0 peak location with respect to the end of the syllable (and the end of the word) in paroxytonic and oxytonic words (taken from Prieto 2007:13) Figure 8 (taken from Prieto 2007:13) is a schematic diagram of the differences in F0 peak locations with respect to the end of the syllable (and the end of the word) in the paroxytonic (word medial condition) and oxytonic words (word final condition) of the sentences in example (6). The thick vertical line in Figure 8 signals the word boundary; the time value provides the mean distance between syllable boundary and peak (for compraven talls) and between word boundary and peak (for comprà ventalls). Prieto (2007) concludes that the data [ ] does not support the idea that prenuclear pitch accents in Catalan contain an obligatory wordedge tone H that aligns with the right edge of a word. We entertain the possibility that the alignment effects found by Estebas-Vilaplana might have been conditioned by the presence of a phrase break H- (Prieto 2007:13). One can conclude from this discussion that a High tone at the end of a paroxytonic word is likely to signal an intonational boundary (i.e. ip and ɩ). This difference between the High tone of a bitonal rising accent and the high boundary tone is of great importance to the boundary criterion. If a high realization of F0 at the end of a paroxytonic word is very likely to be a boundary tone, the high F0 contour at the end of a proparoxytonic word is a boundary tone in all probability. Let me recap the idea: if lexical stress in a word is on the last syllable it is hard to tell the difference between L+H* and L+H* H-. The reason for this is that a) the rising tone has a high contour after the stressed (and accented) syllable (L+H* presents an early rise) and b) the high boundary tone H- also causes a high F0 contour. If the accented syllable is the last syllable before a boundary, it is unclear whether the high F0 contour stems from the rising contour tone or from a 55

64 Chapter 2: Fundamental Aspects of Catalan Intonation high boundary tone. Boundary tones, however, are independent from stressed syllables; they only mark the boundary. As a consequence, in order to tell if a high contour before a boundary stems from a continuation rise, one has to tear apart the two tones. The only tone that can be moved away from the boundary is the pitch accent since it is directly connected with the stressed syllable and independent from the boundary. For this reason, I chose proparoxytonic words (paraules esdrúixoles) in the target positions of my experiments. Their stress is located on the antepenultimate syllable so that the stress is followed by two unstressed syllables (σ w ), which separate the accent from the boundary. Based on proparoxytonic words, it is expected that the peak of a bitonal L+H* accent differs from the high realization of the High boundary tone, so that a continuation rise or a sustained pitch can easily be detected. In Figure 9 an abstract proparoxytonic word is schematically modeled. The diagonal line represents the F0 contour of this word. The L+H* accent is realized on the stressed syllable (i.e. on the strong syllable σ s ). According to Prieto (2007), the peak of the pitch accent is neither realized on σ s nor does it align with the word boundary (cf. the thick rightmost line after σ w2 marked by ω). It is located on the first posttonic syllable (σ w1, where σ w = weak syllable), which is marked in Figure 9 by the end of the thick part of the diagonal line. The rise does not end on σ w1, though, but continues until the word boundary after σ w2. The high contour on σ w2 is caused by a High boundary tone. 27 H- L+H* σ s σ w1 σ w2 ω. Figure 9: Schematic diagram of a continuation rise on a proparoxytonic word If there is no High boundary tone, the F0 contour has the pattern as schematized in Figure 10. Due to the fact that the following word (word 2 ) starts with a rising tone, F0 has to descend in order to reach the Low target of the next pitch accent. Since weak syllables are not anchor points for pitch accents, σ w2 cannot influence the contour and F0 is not high at the word edge. 27 In Selkirk (1984:288) the continuation rise is taken as a phonetic cue for an intonational phrase. I take this rise as simply signaling an intermediate phrase break. Only if the continuation rise is paired with a cue for IntPs (cf. below), it is located at an IntP boundary. 56

65 Chapter 2: Fundamental Aspects of Catalan Intonation L+H* L+H* σ s σ w1 σ w2 σ s σ w word 1 word 2 Figure 10: Schematic diagram of a proparoxytonic word without High boundary tone A real example for the schematic diagram in Figure 10 is given in Figure 11. The (left-dislocated) DP L alfàbrega d Algèria The basil from Algeria, that is presented in Figure 11, is cut out from the sentence L alfàbrega d Algèria, la Bàrbara, la va posar a l hivernacle fa un mes Barbara put the basil from Algeria into the green house one month ago word boundary Pitch (Hz) (delayed) peak L+>H* L+H* H- continuation rise 3 L al fà bre ga d Al gè ri a Time (s) Figure 11: Waveform and intonation contour of the utterance L alfàbrega d Algèria The basil from Algeria of speaker AT (sentence 02a_OO_S_AT) Both nouns in Figure 11 are proparoxytonic words. The prenuclear rising accent L+H* is associated with the strong syllable fà. of the first word L al.fà.bre.ga the basil and its peak is delayed (here marked by > ). As assumed before (cf. Figure 10) the delayed peak is not located at the end of the word, if there is no continuation rise. The peak in Figure 11 is on the first posttonic syllable and the pitch track starts falling due to the low leading tone of the second noun. Consequently, there is no ip-boundary between the two prosodic words L alfàbrega The basil 57

66 Chapter 2: Fundamental Aspects of Catalan Intonation and d Algèria from Algeria. However, there is a continuation rise on d Algèria from Algeria, marked by H- and the break index The Sustained Pitch In Figure 12 a sustained pitch is compared to a continuation rise. The difference to the continuation rise is that a sustained pitch rises up to a certain level and then keeps the fundamental frequency until the end; in other words, a sustained pitch creates a plateau. Such a pitch track can be labeled by L+H*!H-. The ip-boundary tone H- is downstepped in order to signal that F0 remains high but does not continue rising as in the case of a continuation rise. Continuation rise Sustained pitch Figure 12: Schematic example of a continuation rise and a sustained pitch on the word Badalona (taken from Frota et al. 2007:135) A sustained pitch as well as a continuation rise signals a prosodic boundary because the posttonic syllables cannot influence the fundamental frequency. The height at the end of the word is caused by a boundary tone. According to Frota et al. (2007), the sustained pitch is frequent in Italian but is rare in the two Romance languages of European Portuguese and Spanish and never occurs in the Catalan corpus (Frota et al. 2007). But there are good reasons to assume that Catalan has sustained pitch accents. Firstly, the non-existence of a sustained pitch in Frota et al. (2007) might well be a corpus effect because they only chose sentences with clear instances of boundaries and it might be the case that examples of a sustained pitch had been sorted out. Secondly, in my data there are 58

67 Chapter 2: Fundamental Aspects of Catalan Intonation some instances of a sustained pitch, cf. Figure 13. Thirdly, Cat_ToBI gives several examples of such a tone (which is called to sostingut ). Figure 13 shows a sustained pitch associated with the subject of the embedded clause que l àguila robà el ratolí that the eagle stole the mouse of speaker GM. The stressed syllable of L À.gui.la the eagle is associated with a late rising tone L*+H. 28 The contour rises on the first posttonic syllable gui and also reaches its peak there. The height of the pitch is sustained until the end of the second posttonic syllable, i.e. the end of the word Sustained pitch Pitch (Hz) 100 L*+H!H- 3 que l à gui la roba el ratolí Time (s) Figure 13: Waveform and intonation contour of the utterance que l àguila robà el ratolí that the eagle stole the mouse of speaker GM (sentence 33_Emb_GM): Sustained pitch associated with the subject l àguila the eagle Preboundary lengthening Lengthening refers to the longer duration of syllables which precede a boundary. Their length can be opposed to either the duration of similar syllables, which are not followed by a boundary, in the same utterance or to the duration of the same syllables of the same word in a different sentence, where the word is not before a boundary. In both cases, the target word is uttered by the same speaker. The duration of the preboundary syllables is significantly longer and this signals a boundary (cf. Estebas-Vilaplana 2000:120; Astruc 2005:153 for Catalan; Medina 28 Even though the early rising tone L+H* is the typical prenuclear accent in Catalan declaratives (Cat_ToBI p.10), there are also instances of the late rising tone L*+H to be found as prenuclear accent in declaratives in Cat_ToBI (cf. p. 10, example 1a2-StEul(N.S.)). 59

68 Chapter 2: Fundamental Aspects of Catalan Intonation Murillo 2005 for Spanish). Figure 14 and Figure 15 illustrate the different length of the three preboundary syllables of the word L alfàbrega the basil of speaker AT. Figure 14 is the same as Figure 11 with the difference that the length of the relevant syllables is added. Figure 15 shows the noun in a position right before an ι-boundary break (marked with H% and break index 4) Pitch (Hz) L+>H* L+H* H- 3 L al fà bre ga d Al gè ri a Time (s) Figure 14: Waveform and intonation contour of the utterance L alfàbrega d Algèria The basil from Algeria of speaker AT (sentence 02a_OO_S_AT): Length of non-preboundary syllables Pitch (Hz) 100 L+H* H-H% 4 L àl- -fà- -bre- -ga Time (s) Figure 15: Waveform and intonation contour of the utterance L alfàbrega The basil of speaker AT (sentence 01a_O_S_AT): Length of preboundary syllables 60

69 Chapter 2: Fundamental Aspects of Catalan Intonation Whereas the last syllable ga has a duration of only 92 milliseconds in the non-preboundary position in Figure 14, it has a duration of 168 ms in Figure 15, where it directly precedes the boundary. This is a difference of 80 ms. The pattern is similar for the first posttonic syllable bre: it is longer in the preboundary case. However, in these examples the accented syllable (fà) does not show any significant difference. The longer duration of the syllables in the preboundary case (cf. Figure 15) has the consequence that the whole word has a longer duration. This is illustrated in Figure 16 & Figure 17. In the former figure, the verb(al complex) va comptar counted is shown in a preboundary position and is uttered with a length of 568ms by speaker MO. In the latter figure, it is in a non-preboundary position and is uttered with a length of only 406ms (by the same speaker). 29 The length of the whole verbal complex reflects the length of its syllables: va (123ms in Figure 16 vs. 110ms in Figure 17), comp- (287ms vs. 180ms) and tar (157ms vs. 114ms) va comp- -tar Time (s) Figure 16: Length of syllables in seconds of the verb va comptar counted of speaker MO in an utterance with a long object (sentence 22_SVOO_MO) va comp- -tar Time (s) Figure 17: Length of syllables in seconds of verb va comptar counted of speaker MO in utterance with a short object (sentence 21_SVO_MO) 29 Cf. appendix for the results of each single speaker. 61

70 Chapter 2: Fundamental Aspects of Catalan Intonation Pauses Pauses are defined as a stretch of silence (Frota et al. 2007) or as a major F0 break (Estebas-Vilaplana 2000:118). Here, I distinguish between two kinds of pauses (Figure 18). The first kind of pause comprises stretch of silences which are visible in F0 due to an interruption of the pitch track. The data in the appendix is marked with (P) for pause if the stretch is longer than 100ms. The second kind of pause is a so-called audible pause. In general, this type of pause comprises breaks which can be perceived audibly as a pause, but which are not visible in the pitch track. In addition, stretch of silences that are shorter than 100ms are also considered as audible pauses. The latter kind of pause also comprises the tonal extension mentioned in Astruc (2005:153). Speakers often drawl the syllable before the boundary until it becomes a short silent pause. In order to decide when an extreme lengthening becomes a pause proper, she sets the cutoff point at 100ms. She considers such a duration enough for a period of silence to be perceived as a pause, i.e. as a major prosodic break in this type of read, pre-planned, non spontaneous data visible pause Pitch (Hz) 150 audible pause 100 L+H* H-H% L+H* H-H% 4 4 L Al fà bre ga la Bàr ba ra la va posar a l h Time (s) Figure 18: Waveform and intonation contour of the utterance L alfàbrega, la Bàrbara, la va posar (a l hivernacle fa un mes) Barbara put the basil (in the green house one month ago) of speaker AT (sentence 01a_O_S_AT) 62

71 Chapter 2: Fundamental Aspects of Catalan Intonation Figure 18 illustrates both kinds of pauses. The two IntP breaks are both marked by H-H% and by the break index 4. The first IntP break (the one after the left-dislocated DP L alfàbrega the basil ) is an example of an audible pause. There is no visible stretch of silence, but a pause can clearly be perceived. The second IntP break (the one after the preverbal subject la Bàrbara) is an example of a visible pause. The stretch of silence has a length of 107ms and it is visible in the oscillogram (the encoding of the acoustic signal) as well as in the pitch track. The notion major break which is used by Estebas-Vilaplana (2000) and Astruc (2005) signals that pauses are a very special boundary cue. I discuss this in more detail in section by concluding that they distinguish between ips and IntPs, exactly as a complex boundary tone Complex Boundary Tones Based on Beckman & Pierrehumbert s (1986:288) reanalysis of the phrase accent T- as an ip-boundary tone, the ToBI transcription system (Silverman et al. 1992) represents T- in IntPfinal position together with the ι-boundary tone T% as a complex symbol T-T% (cf. also Ladd 1996:98 and Gussenhoven 2004:132). As such, it is the standard assumption for English that a complex boundary tone signals a boundary of an intonational phrase (IntP, ι). Cat_ToBI takes this assumption as the basis for Catalan (Cat_ToBI: # , # ) and I consequently follow this assumption in the present work. 30 For example, the complex tone L-L% is the typical tone at the end of declarative sentences. A sustained pitch preceding an ι-boundary is associated with (!)H-!H%. A continuation rise preceding an ι-boundary is associated with H-H%. In Cat_ToBI the tone H-H% is described for questions (e.g. absolute questions (interrogatives absolutes de tipus ascendent) and for exhortative questions). However, this tone is also typical for leftdislocations (cf. Chapter 5), which surely are not questions. For this reason, I assume that H- H% s function is not only to signal questions, but that it can also appear clause-internally without an interrogative function. 30 However, in the experiment data there are many cases where it is doubtful that a complex tone automatically signals an ι-boundary. It is doubtful because the perception of these breaks is not as strong as in the case of clear instances of an IntP break. It might thus rather be the case that the ip-boundary tone might also be complex. In addition, there are examples in Cat_ToBI, with tritonal boundary tones (LH-L%; cf. also Vanrell et al. 2008). Nevertheless, this work follows the assumption that complex boundary tones signal an IntP-boundary. Here, it has to be mentioned again that Prieto (2009) as well as the latest version of Cat_ToBI, published online in June 2008, do not longer assume phrasal accents. 63

72 Chapter 2: Fundamental Aspects of Catalan Intonation A very common complex boundary tone in my data is L-H%. This tone is described in Cat_ToBI (p. 28) as one which expresses apparentness / obviousness in a soft and smooth manner and the tone is typical in declaratives expressing obviousness and in imperatives with a nuance of insistence. Although this tone appears also every now and then in all-new sentences (cf. Chapter 3), it is more common in the experiments including left-dislocations (Chapter 5 and 6). This can be explained with the given status of these constituents (cf. Chapter 1) and by repeating them a certain kind of obviousness is expressed. Figure 19 is an example of L-H%. The left-dislocated DP L alfàbrega the basil has an early rising pitch accent on the stressed syllable L al. The peak is reached between the stressed and the first posttonic syllable. After the peak, the pitch falls until the beginning of the preboundary syllable ga and then rises until the end of the word. This fallrise is expressed by L-H% Pitch (Hz) 100 L+H* L-H % 4 L al fà bre ga Time (s) Figure 19: Waveform and intonation contour of the word L alfàbrega basil of speaker MO (sentence 01a_O_S_MO): The ι-boundary tone (BI 4) consists of a Low ip-boundary tone L- and a High tone H% 64

73 Chapter 2: Fundamental Aspects of Catalan Intonation Organization of Boundary Cues Only the last two criteria, pauses and complex boundary tones, indicate a boundary at the intonational phrase level. They are therefore very important for the decision on which level a prosodic break is located. Complex boundary tones constitute, for theory-internal reasons, an IntP break and pauses are mentioned throughout the literature on Catalan intonation as one clear cue for an IntP break. Estebas-Vilaplana (2000:118) states that the absence of a major F0 break shows that a prosodic break can be classified as an ip-boundary ( PhP-boundary in her terminology). Astruc (2005:154:Table 4.3) classifies pauses as signaling the ToBI break index 4 (i.e. IntP) and in Cat_ToBI it is mentioned that the IntP manifests itself only with a pause (Prieto et al. 2007:20). In general, neither in Frota et al. (2007) nor in Cat_ToBI (Prieto et al. 2007) are the two prosodic levels intermediate phrase (ip) and intonational phrase (IntP, ι) clearly distinguished by boundary cues (beside pauses). In addition, there are no detailed studies yet on Catalan, which try to characterize the different prosodic levels, either by phonetic properties or by (segmental) phonological rules that apply in the specific domain. 31 One important instance for distinguishing between the two levels mentioned is the transcriber, for it is the transcriber who assigns the boundaries to a certain level of intonational grouping. Prieto et al. (2007:20) note El criteri per a diferenciar els dos tipus de frontera es basa en el judici auditiu del transcriptor ( The criterion to discriminate the two types of boundaries is based on the auditory judgment of the transcriber (translation: I.F.); cf. also Cat_ToBI p.36:#3.3). As for this study, I take the first three mentioned boundary cues (continuation rise, sustained pitch, and preboundary lengthening) as cues for the ip-level. My assumption is based on the following reasoning: due to the facts that (a) according to the Strict Layer Hypothesis an ι- boundary always coincides with an ip-boundary and (b) a continuation rise occurs at the level of ip (Prieto 2005, Prieto et al. 2007), each continuation rise preceding an ι-boundary can be reduced to the lower ip-boundary. This is similar to the sustained pitch and preboundary lengthening. The two remaining boundary cues, pauses and complex boundary tones, are used as cues for boundaries of the intonational phrase level. As described, pauses including visible and audible pauses and complex boundary tones are consequently taken as signaling ι-boundaries. If one of the ip-boundary cues is combined with either a pause or with a complex boundary tone, 31 Prieto (2008) clearly states that there are no arguments for prosodic constituency based on blocking or triggering of postlexical rules. 65

74 Chapter 2: Fundamental Aspects of Catalan Intonation they automatically precede an ι-boundary. Table 3 gives a short summary of the relevant boundary cues. Prosodic constituent intermediate phrase (ip) Intonational Phrase (IntP, ι) Boundary cues continuation rise sustained pitch preboundary lengthening pause (visible stretch of silence & audible pause) complex boundary tones Table 3: List of assumed prosodic levels (between prosodic word and utterance) and the corresponding boundary cues One final important note with respect to the organization of the boundary cues is necessary here. As mentioned, the boundary cues described in Frota et al. (2007) account for both prosodic levels. I clearly separate them in order to distinguish between ip and IntP. However, as my data show, it is impossible to say that a specific syntactic/pragmatic constituent is associated with a certain kind of prosodic constituent. This means that, for example, the right edge of leftdislocations does not align always with either an ι-boundary or an ip-boundary. Left-dislocations align with both kinds of boundaries (cf. also Astruc 2005 for right-dislocations). This optionality might be a reason for why in studies such as Frota et al. (2007), no decision is made as to the correspondence between boundary cues and prosodic constituents. Furthermore, this optionality is the reason for Prieto (2009) to assume only one type of boundary tones, while still assuming the two prosodic levels ip and IntP. In the present work, I follow older proposals (and hence not Prieto 2009) and continue assuming two different boundary tones; but the described variation is the reason for introducing the term prosodic phrase as a hypernym for intermediate and intonational phrases (cf. Elordieta et al. 2005:116, fn.2 for an analogous reasoning for Spanish). Prosodic phrase thus refers to both levels of intonational grouping, ip and IntP. The goal of this work is not to establish a detailed phonetic analysis in order to see how the Prosodic Hierarchy is implemented in Catalan. Rather, the aim is to decide whether a prosodic break is present or not. Moreover, it is shown that a break is in fact present at both levels of phrasing. 66

75 Chapter 2: Fundamental Aspects of Catalan Intonation 2.4 Assumptions on Phonological Inventory of Catalan To conclude this chapter, I summarize the fundamentals and assumptions I have based my work on. The foundation of my analysis is the Cat_ToBI system established in Prieto et al. (2007), which is itself based on the Autosegmental-Metrical (AM) Model (Pierrehumbert 1980, Ladd 1996). Consequently, I assume that tones (High, Low) are located on a different tier than prosodic constituents, but they do associate with them (autosegmental part). The phonological constituents are organized hierarchically in the so-called Prosodic Hierarchy (metrical part). Just as Prieto et al. (2007), I assume two levels between the prosodic word (ω) and the Utterance (U): Intermediate Phrase (ip) and Intonational Phrase (IntP, ι). However, I introduce the term prosodic phrase as a hypernym for ip and IntP, since there is often no need to distinguish between the two levels. As usual in ToBI systems, I assume three types of accents: Pitch Accents (which associate with metrical strong syllables), ip-boundary tones (which associate with the level of the Intermediate Phrase), and ι-boundary tones (which associate with Intonational Phrases). I use the term boundary tone as a hypernym for ip-boundary tone and ɩ-boundary tone. Like Prieto et al. (2007), I assume five pitch accents: the two monotonal accents L* ( low ) and H* ( high ) and three bitonal accents: H+L*, L*+H, and L+H* (Prieto et al. 2007:4; cf. Figure 2). In cases where a delayed peak is transcribed, I use the Cat_ToBI representation L+>H*. The Catalan prenuclear accents are L*+H and L+H* (including their variants with >,!, or ). As the nuclear accent of neutral declaratives, I use L* in accordance with Cat_ToBI. As for boundary tones, I assume the tones stated in Cat_ToBI for the intermediate phrase and the intonational phrase, namely H-, L-, and!h-, and H%, L%, and!h%. I also assume that the combination of these tones (like H-H%, L-L%, L-H%, and L-!H%) signals a boundary on the IntP level. I take the three boundary cues continuation rise, sustained pitch, and preboundary lengthening as cues for the ip-level. Pauses and complex boundary tones are used as cues for boundaries of the intonational phrase level. Pauses include visible and audible pauses and complex boundary tones are consequently taken as signaling ι-boundaries. If one of the ip-boundary cues is combined with either a pause or with a complex boundary tone, they automatically precede an ι-boundary. 67

76 Chapter 3 Phrasing Patterns in Catalan SVO Structures This chapter is devoted to objects and their influence on the prosodic grouping of the (matrix) subject and verb in Catalan. It is hypothesized that sentential objects compared to DP objects increase the single group phrasing of (matrix) SV. Catalan is described as a language where eurhythmic constraints play a role and cause a short subject to phrase together with the verb when the object is long (i.e. (SV)(O), Prieto 2005, D Imperio et al. 2005). In order to compare the different objects two experiments were conducted: the first for simple SVO structures and the second for complex SVO structures, in which the object is sentential. The research shows that the percentage of (SV) phrasing increases (by about 20%) when the object is sentential. It further shows that the (SV) grouping is only a root phenomenon. The (SV) grouping does not show up in embedded clauses. If the embedded object clause consists of a short subject and a long object, the embedded subject never phrases with the embedded verb. In addition, the embedded clause is in three quarters of the data prosodically separated from the matrix clause and the object clause itself is not phrased in a single constituent. The findings can be accounted for using an optimality theoretic (OT) framework based on Prieto s (2005) three main constraints. I depart from Prieto (2005), though, by assuming a new constraint which accounts for the boundary preceding the embedded clause (ALIGN-CP,L) and by assuming that there is a probabilistic constraint ranking where the constraints are situated on a continuous ranking scale (Stochastic OT, Boersma & Hayes 2001). In that way, I can account for the variation in the grouping of complex SVO structures. The research on object clauses is especially important due to the literature on Phrasing in Romance of the last decade not inquiring seriously into complex structures. These studies have mainly been concerned with simple SVO structures. Furthermore, the authors, who do examine complex structures, are mostly concerned with non-restrictive relatives, nominal appositives, parenthetical expressions or they are concerned with the comparison of right boundaries of root clauses (Downing 1970, Nespor & Vogel 1986/2007, Selkirk 2005 among others). They are

77 Chapter 3: Phrasing Patterns in Catalan SVO Structures hardly concerned with clauses which are part of the verb s argument structure. Truckenbrodt (2005), though, does inquire into subject and object clauses but his study is based on only one speaker. This chapter is structured as follows: In Section 3.1 background information on the research on Romance SVO during the last decade is given. This brief introduction quickly moves to a concentration on Catalan. Section 3.2 provides the reader with the general outline of the conducted experiments in the present study. Section 3.3 concentrates on the simple and complex SVO experiments. First, the specific experiment design is presented; then the results of the detected boundary cues are given. Finally, the results of the prosodic phrasing of these structures are presented. In Section 3.4 the theoretical approach to the phrasing of SVO and of embedded SVO is given. 3.1 Background & Experiment Hypothesis In this section the general phrasing pattern of SVO in Romance is introduced followed by a detailed description of Catalan (Section 3.1.1). Finally, the two hypotheses underlying the experimental research are introduced (Section 3.1.2). Since the introduction of the Prosodic Hierarchy (Selkirk 1984, Nespor & Vogel 1986/2007) the placement of intonational boundaries and their determining factors have been discussed. During the last decade several studies concerning the prosodic grouping in Romance sentences have been published (for Catalan Prieto 1997, Prieto 2005; for European Portuguese (EP) Frota 2000; for Italian D Imperio 2001, 2002; for Spanish Prieto 2006, Elordieta et al (also for EP); for all four languages D Imperio et al. 2005, Frota et al. 2007). The most recent works, especially, mainly concentrate on neutral declarative SVO utterances. Neutral means that the SVO sentences are taken to be all-new sentences. The sentences count as answers to the question What happened? (cf. Section for more detailed information on information structure and the relation between focus and question-answer pairs). By speaking of neutral declarative SVO utterances it is implicitly assumed that preverbal subjects are not taken as topics otherwise they could not be neutral (cf. Section for arguments that preverbal subjects are not automatically left-dislocated). 69

78 Chapter 3: Phrasing Patterns in Catalan SVO Structures In D Imperio et al. (2005), a crosslinguistic study of prosodic phrasing in Catalan, Spanish, Italian, and European Portuguese is presented. They mention three factors that can influence the phrasing pattern: syntactic branchingness, length of constituents and speech rate (D Imperio et al. 2005:60f). In their study they concentrate on only the first two factors and recorded two speakers of each language. D Imperio et al. (2005) come to the conclusion that the Romance varieties examined display different intonational phrasing patterns. The most common phrasing pattern in Catalan and Spanish is (S)(VO). Catalan has the tendency to phrase utterances in prosodic constituents of similar length. For this reason, an (SV)(O) phrasing is often produced with non-branching subjects and long objects. In Spanish, the (SV)(O) phrasing also appears only in cases with a long branching object (D Imperio et al. 2005:81, Prieto 2006:55): According to the study in D Imperio et al. (SV)(O) in Spanish is much less common than in Catalan and statistically not significant. Elordieta et al. (2005:130) draw the same conclusion. In Prieto (2006:55), the (SV)(O) phrasing in Spanish appears to be a more normal case though. This goes along with Nibert (2000), who concludes that (SV)(O) is the default type of phrasing in Spanish. It appears that in Spanish (S)(VO) and (SV)(O) phrasing patterns are both possible (depending on weight conditions), and that there are individual differences among speakers (Prieto 2006:55). In Italian and in Standard European Portuguese (SEP, the variety spoken in Lisbon) the (SVO) phrasing is the usual phrasing (D Imperio et al. 2005:83). (S)(VO) phrasings are also possible, but they seem to be limited according to the nature of S (long/branching). The nature of the object does not seem to play a role (cf. also Elordieta et al. 2005:120). In contrast to SEP, the (S)(VO) phrasing is the pervasive pattern in Northern European Portuguese (NEP, D Imperio et al. 2005:83). Although in Italian the (SV)(O) phrasing might show up sometimes, this pattern nevertheless is rarely allowed or tends to be strongly avoided (D Imperio et al 2005:76). In European Portuguese, (SV)(O) is unattested (D Imperio et al 2005:83, Elordieta et al. 2005:130). Consequently, the (SV)(O) pattern has the tendency to show up only in Catalan and Spanish, even though it does not constitute the pervasive phrasing pattern in these two languages. In (1) I present the main phrasing patterns according to D Imperio et al (2005). 70

79 Chapter 3: Phrasing Patterns in Catalan SVO Structures (1) Main phrasing patterns according to D Imperio et al (2005): NEP: (S)(VO) SEP: (SVO) Italian: (SVO) Spanish: (S)(VO) [and (SV)(O)] Catalan: (S)(VO) [and (SV)(O)] Detailed Background of Catalan Phrasing Patterns For the study of Catalan in D Imperio et al. (2005), two speakers from Central Catalan who read a corpus of 124 sentences three times were examined. The factor of length was controlled for in terms of number of syllables (cf. Jun & Fougeron 2000), and not in terms of number of feet or prosodic words (as in e.g. Selkirk 2000, Prieto 2005). In order to display the factor of branchingness, a PP or an AP was added to the argument noun (i.e. subject N/object N). Both speakers produced the sentences with two prosodic phrases. The boundary in-between was signaled by a clear F0 continuation rise. Catalan is not only examined in D Imperio et al. (2005), but also in Prieto (2005). Whereas in the former study Catalan is only one language among others, the latter focuses only on Catalan. In addition, whereas the former study is presented without any theoretical analysis, Prieto (2005) offers a theoretical analysis for the phrasing pattern. Since the results are similar in both studies, I present only the broader study of Prieto (2005) in detail. The outline and the results are presented here. The analysis is presented in detail in Section 3.4. The data in Prieto (2005) has two different sources. On the one hand, four native speakers of Central Catalan (who were all linguists) were asked to judge the phrasing divisions of the utterances which appeared in her article. The consultants were asked to judge whether the phrasings would be grammatical at a normal rate of speech and also asked to interpret the sentences as out-of-the-blue conveying new information, i.e. to take them as broad focus sentences. On the other hand, the data from a production experiment (published in Elordieta et al. 2003) were used. Two speakers of Central Catalan read the SVO target sentences three times. They were told to read the sentences at a normal speech rate as neutral declarative sentences (i.e. new information reading). Consequently, the sentences do not include topics or (narrow) focus constituents. The study of D Imperio et al (2005) is also based on this production experiment. The general phrasing pattern in Catalan is sensitive to constituent length, and verbal heads can be phrased together with preceding subjects when the DP objects are long (Prieto 2005:214). 71

80 Chapter 3: Phrasing Patterns in Catalan SVO Structures If the DP object is short, or if the subject is long, the verb is phrased with the object. This leads to the following phrasing pattern in unmarked Catalan declarative sentences (where p = phonological phrase): (2) (S)p (VO)p single word subjects, short verbal projection (SV)p (O)p single word subjects, but long objects (2ω) (S)p (VO)p long subject (>2ω), short objects 1 As the reader might have noticed, the terms prosodic phrasing and phonological phrase have been used to describe the relevant prosodic constituent. These terms are the original notations used in D Imperio et al (2005) and Prieto (2005) respectively. I continue to use the term prosodic phrase as a hypernym for ip and IntP The hypotheses The main aspect for my study on SVO phrasing in Catalan originates in the interesting and atypical finding of D Imperio et al. (2005) and Prieto (2005) that the subject can phrase together with the verb, if the DP object is long. It is common knowledge that objects can also be sentential. Hence, the question comes up as to how phrasing decisions are made (by Catalans) when the object is a complete clause. I set up the following hypothesis: Hypothesis 1: Compared to DP objects sentential objects increase the single group phrasing of (matrix) SV In order to be able to compare the influence of DP objects and sentential objects, two things are important. First, the percentage value of the (SV) groupings in D Imperio et al. (2005) is needed. In that study the (SV)(O) grouping adds up to 33.1% (D Imperio et al. 2005:68f). 2 The hypothesis is validated, if sentential objects increase the value of (SV) phrasing, i.e. the value has to be significantly higher than 33.1%. Second, it is also important to compare the utterances of the same group of speakers. For this reason, I repeated the SVO experiment with DP objects. In addition to this reason, there are two further arguments for repeating the SVO experiment: number of speakers and controlled material. The results of D Imperio et al. (2005) and Prieto (2005) are based on only two native speakers of Central Catalan. In order to verify the findings, I 1 Prieto (2005) does not include cases of a long subject together with a long object. 2 The average is calculated from table 3, 4, and 5 in D Imperio et al. (2005:68f.). 72

81 Chapter 3: Phrasing Patterns in Catalan SVO Structures conducted the experiment with ten speakers. Additionally, I carefully controlled for the all-new status of the SVO structure. Instead of asking the participants to read the sentences as all-new sentences, I introduced the target sentences in a context which caused the SVO structure to be entirely new (for more information structural background cf. Section 1.1.3). The same conditions were used in the experiment with sentential objects. The latter experiment has a further important research question: are embedded object clauses prosodically separated from the matrix clause? Downing (1970) argues that only the edges of root clauses (clauses that are not embedded) 3 are obligatorily aligned with larger prosodic boundaries meaning that the left edge of an embedded clause does not have an obligatory boundary. This position is adopted in Nespor & Vogel (1986/2007:189). Truckenbrodt (2005) picks up the question if an embedded clause triggers intonational boundaries at its left (or right) edges. He did not find any evidence of an intonational phrase boundary preceding an embedded object clause. Based on these works, I set up a second hypothesis: Hypothesis 2: Sentential objects are not obligatorily separated by prosodic means from the preceding matrix clause 3.2 General Outline of the Experiments Section 3.2 provides the reader with the general outline of three experiments. Two of them are presented in Chapter 3 (simple SVO and complex SVO experiment); the third is presented in Chapter 6 (CLLD vs. S experiment). Section 3.2 can be seen as a brief appended exposition giving detailed background information about the three experiments. The specific experiment designs of the three parts are not presented in this section. The specific experiment design of the simple and complex SVO experiments is given in Section 3.3. This design is relevant for the two hypotheses presented in the preceding section (i.e. Section 3.1.2). The specific design of the CLLD vs. S experiment is given in Section Any CP contained in another CP is called an embedded clause. A root clause is any CP that is not contained in a higher CP (cf. also Downing 1970:29ff.). A simple sentence such as [Peter sleeps] is a root clause; it is not contained in a higher CP. In a complex root clause such as [Mary supposes [that Peter sleeps]] the sentence [that Peter sleeps] is an embedded clause, since it is contained in a higher CP. Furthermore, the term matrix clause refers to the root clause minus its embedded clause, i.e. [Mary supposes [ ]]. Throughout my work, I use the term root clause or root context in its narrow sense: I refer to simple sentences/cps which do not dominate another CP (i.e. which do not have embedded clauses). The term embedded context refers to root clauses in its widest sense, i.e. CPs that dominate complement clauses. 73

82 Chapter 3: Phrasing Patterns in Catalan SVO Structures The location of the experiment was Berlin and Hamburg (Germany). A hard disk recorder, Marantz CDR310, was used and the subjects were recorded in a quiet room. I had 10 subjects: 7 female and 3 male speakers, ranging in age from 20 to 39 years old (Ø 26 years old), cf. Table 1. All speakers are native speakers of Central Catalan. None of the subjects are linguists. All subjects were totally naïve as to the purpose of the experiment. speaker sex age profession origin and L1 parents L1 AT f 22 student Barcelona : Cat. (Cat.) : (Switzerland) CB f 24 student Barcelona : Spanish (Cat. & Spanish) : Swedish CP f 23 student Barcelona : Cat. (Cat.) : Cat. DS m 31 employee Barcelona : Cat. (retail trade) (Cat.) : Cat. GM f 23 student Barcelona : Cat. (Cat.) : Cat. GV m 20 student Barcelona : Cat. (Cat. & Spanish) : Cat. IS f 25 employee Manresa : Cat. (Cat.) : Cat. MM f 26 self-employed Terrassa : Cat. (Cat.) : Cat. MO f 39 student Barcelona : Cat. (Cat. & Spanish) : Spanish RS m 26 student Sabadell :Cat.+Spanish (Cat. & Spanish) : Cat.+Spanish Table 1: Detailed information of recorded subjects The experiment consisted of three different parts. 4 They are listed in (3). (3) Parts of the Experiment: Experiment 1a: Simple SVO phrasing (12 target sentences (3 scenarios x 4 conditions)) Experiment 1b: Complex SVO phrasing (12 target sentences (3x4)) Experiment 1c: CLLD vs. S (i.e. preverbal subjects; 36 target sentences (3x12)) Fillers: (24 target sentences) 4 Actually, I had two further parts (i.e. 5 parts altogether). I do not include these two other parts here. For this reason they are taken as pure filler clauses. 74

83 Chapter 3: Phrasing Patterns in Catalan SVO Structures Altogether I used 84 different sentences in the experiment. For each single part the sentences of the two other parts counted as filler clauses (as well as the pure filler clauses). In experiment 1a and 1b, 12 sentences (14%) of the data presented were target sentences respectively, while the other 72 sentences (86%) counted as filler sentences. In experiment 1c, 36 sentences (43%) of the data presented were target sentences and 48 sentences (57%) of the data were filler sentences. The target sentences were presented with the filler sentences in a pseudo-randomized order. The sentences of the different parts were mixed so that two sentences of one part never appeared in a row. As will be explained in the relevant sections, the three experiments consist of different contextual scenarios. The scenarios represent a story for the target sentences; they do not reflect the different conditions (such as length of constituents). It was also controlled that two sentences of the same scenario never appeared in a row (independently of the filler sentences inbetween). Material: The material was controlled with respect to word stress (proparoxytonic words; Cat. paraules esdrúixoles ), with respect to the sounds in the target words, their syllables, the number of prosodic words and the context. The target words were controlled in two further ways. First, I tried to use only sonorant sounds to guarantee the vocal fold vibration. Vowels are per definition sonorant. Sonorant consonants are nasals, vibrants, approximants, and glides (Hall 2000:22). Target words such as número number and Àgueda (Catalan female name) clearly show this. Both words are completely sonorant (<g> and <d> are approximants in Catalan, i.e. [ɣ] and [ð] respectively). The only exception among the target words is the sound [ʃ] (i.e. <x>) in brúixola compass. This sound is sufficiently away from the word s end though. The following sonorants (including the two nuclei of the post stress syllables) allow the vocal folds to vibrate and are able to show the desired fundamental frequency. Second, in most cases proparoxytone words were used where the nucleus of the unstressed syllables is separated by (voiced) consonants. Open syllables with a voiced onset (i.e. CV-syllables) were also used. 5 Target words such as nú.me.ro number and À.gue.da (Catalan female name) are corresponding examples. In Catalan, diphthongs count as one syllable, but not all adjacent vowels are automatically a diphthong. There are vowel clusters which count as a 5 For Catalan, the ideal rhythmic pattern appears to consist of stresses separated by at least one mora and not more than three. (Wheeler 2005:125) 75

84 Chapter 3: Phrasing Patterns in Catalan SVO Structures hiatus (separate pronunciation of two adjacent vowels) and count as two syllables (e.g. ti.a aunt, Nú.ri.a Núria, Al.gè.ri.a Algeria, à.vi.a grandmother ). 6 Many proparoxytone words in Catalan end in a hiatus. In general, I tried to avoid these words, since they bear the risk that some speakers pronounce the hiatus as a diphthong (cf. Cabré & Prieto 2004). If so, a proparoxytone word such as à.vi.a would be uttered as a paroxytone word à.via, with the stress on the penultimate syllable. Nevertheless, some target positions in my data are filled by proparoxytone word with a hiatus which is possibly a diphtong (such as es.glé.si.a church or Al.gè.ri.a Algeria ). These two aspects enable one to profit in the best way from the distance between the boundary and the stressed syllable. Not only the anchor position for the pitch accent (i.e. the stressed syllable) is far away from the boundary, but the characteristics of the sounds and syllables in between allow the F0 contour to be tracked in an optimal way. This is important for detecting boundary tones. Finally, the context was carefully controlled. The context is the relevant environment which is responsible either for a certain element to be given (i.e. mentioned in the context) or to be non-given (i.e. not mentioned in the context), or for a certain sentence to be completely new (which is relevant for the study on SVO and embedded SVO structures). 7 Different scenarios can represent the same context. If three different scenarios force the target sentence to be entirely new, they represent the same context (cf. Section 3.3.1). The way the context is controlled for each single part of the experiment as well as the types of structures studied are explained in the sections where the specific experiment design is introduced. Procedure: The subjects were placed in a quiet room in front of a computer and the data was presented in a PowerPoint file. The context and the target sentence were presented together on a single slide; consequently, there were 84 slides. The subjects were told to read out the target sentences at a normal speech rate only after they understood the suitable context (question). The context (question) was presented in two different ways. First, it was visually presented on the computer screen. The subjects were told to read the sentences to themselves in order to understand them. Second, they listened to the sentences. The context was spoken out loud at a 6 Cabré & Prieto (2004) argue that the vowel clusters in these contexts are all diphthongs, in contrast to the norm. 7 Cf. Section for basics on information structure. 76

85 Chapter 3: Phrasing Patterns in Catalan SVO Structures normal speech rate by a native speaker of Central Catalan. 8 To summarize, after having read the context, the subjects pressed a button in order to hear the context. They were told to read the target sentence (i.e. the answer to the context question) directly after hearing the context (question). This procedure has several advantages: the subjects can read and hear the context; by reading the context, the subjects are able to understand it correctly; by hearing the context, the subjects were put in a more conversational-like situation, despite the unnatural, laboratory situation. In addition, speakers who did not read the context completely at least had to hear the context completely; and finally, by recording the context in advance, it was possible to control the way the context was uttered. Due to this, I could be sure that the context fits not only with respect to the content, but also with respect to the pronunciation. The subjects were told to read the sentences aloud in a conversational style without being given any specific instructions regarding the phrasing. The subjects were additionally told to think carefully about the sentences, since no commas were put in the data. Punctuation marks such as commas would cause the speaker to produce a pause more or less automatically. In order to avoid this influence commas were left out. Since left-dislocations in Catalan (cf. experiment 1c (CLLD vs. S)) are normally separated by a comma, the presented sentences could be difficult to understand. A short practice session at the beginning of the experiment was included because the procedure may have been slightly complex. During that period the subjects had to go through five context questions and their appropriate target sentences, so that subjects could become accustomed to the procedure. After that, the real experiment began. The sentences were recorded directly as.wav files (sample rate Hz) and F 0 tracks were analyzed using PRAAT version (developed by Boersma & Weenink , University of Amsterdam). Subsequently, a prosodic analysis of the spoken utterances was carried out. In order to determine the phrase boundaries the sentences were analyzed acoustically and instrumentally. By using PRAAT, I obtained the pitch tracks and the corresponding spectrograms and I aligned the suprasegmental events with the segmental string. The location of a phrase break, if any, was carried out by carefully listening to the sound files repeatedly and by observing the fundamental frequency and the spectrogram. In order to ensure that my judgments of Catalan boundary tones were adequate, two different Catalan phonologists (Pilar Prieto and 8 I am very grateful to Ariadna Benet for lending me her voice and for recording all the contexts not only in an excellent manner, but also within a short time period. 77

86 Chapter 3: Phrasing Patterns in Catalan SVO Structures Lluїsa Astruc) were asked to determine the boundary tones of a set of sentences. After receiving their responses, my judgments were compared with those of the native speakers. The conclusion was reached that I was able to continue the work. 3.3 Experiments on Simple and Complex SVO Phrasing The goal of the simple and complex SVO experiment is to compare the influence of the different objects on the prosodic grouping of the (matrix) subject and verb. Additionally, the grouping of embedded clauses themselves is discussed. The experiments are conducted to test the two hypotheses described in Section The study validates both hypotheses. It is shown that the percentage of (SV) phrasing increases (by about 20%) when the object is sentential. Furthermore, even though Catalan shows a great tendency to separate the embedded object clause from the matrix clause, sentential objects are not obligatorily separated. 20% of the object clauses are phrased with preceding material. The simple SVO experiment shows that (S)(VO) is the predominant grouping in Catalan and that this grouping is less robust if the object is long and the subject short. The results clearly support the findings of D Imperio et al. (2005) and Prieto (2005). In this section, the specific experiment design of experiments 1a (simple SVO) and 1b (complex SVO) is presented (Section 3.3.1). After that, the detected boundaries and their frequencies are described (Section 3.3.2). In Section the results of the phrasing patterns are given. The results of simple SVO structures are presented first followed by the results of complex SVO structures Specific Experiment Design According to D Imperio et al (2005:71) a clear effect of length is found in Catalan phrasing decisions and the branching effect is clearly prosodic (i.e. the number of prosodic words; cf. Prieto 2005:#3.4). Based on these findings, three different scenarios (representing the same information structural context) for testing the phrasing of broad focus declarative clauses were constructed. There are two groups of clauses: root clauses (simple SVO) and embedded clauses (complex SVO). In (4), one complete scenario (which is called scenario A) is presented; including the context question and the different conditions for the target sentences. The three scenarios display all the same information structural context. They cause the target sentences to 78

87 Chapter 3: Phrasing Patterns in Catalan SVO Structures be entirely new (as described in Section 1.1.3): no constituent of the target sentence is given in the context question. The different conditions for the target sentence are explained below. The two further scenarios (called scenario B and C) are shown in (5). They are not presented as detailed as scenario A. The parentheses in (5) indicate which constituents do not always appear, but depend on length requirements (similar to the pattern in (4); cf. appendix A & B for a complete overview). (4) Scenario A: Example for target sentences of (embedded) SVO sentences Context (question): No et trobes bé? Em sembla que estàs de mal humor. Què ha passat? No you feel.2sg good Me seem that be.2sg of bad mood What have happen.ptcp Do you feel bad? You seem to be in a bad mood. What happened? Target sentences: a. Condition: short S / short O L àguila robà el ratolí. the.eagle steal.pst the mouse The eagle stole the mouse b. Condition: short S / long O L àguila robà el ratolí del meu germà. c. Condition: long S / short O La meva gran àguila robà el ratolí. d. Condition: long S / long O La meva gran àguila robà el ratolí del meu germà. the my big eagle steal.pst the mouse of.the my brother My big eagle stole my brother s mouse. e. Condition: short S / short O (in the embedded clause) La Bàrbara suposa que l àguila robà el ratolí. f. Condition: short S / long O (in the embedded clause) La Bàrbara suposa que l àguila robà el ratolí del meu germà. g. Condition: long S / short O (in the embedded clause) La Bàrbara suposa que la meva àguila robà el ratolí. h. Condition: long S / long O (in the embedded clause) La Bàrbara suposa que la meva àguila robà el ratolí del meu germà. the B. assume that Barbara assumes that 79

88 Chapter 3: Phrasing Patterns in Catalan SVO Structures (5) Two further scenarios for (simple and complex) SVO 1. Scenario B: Context (question): El Pedro no es troba bé? Em sembla que està molt furiós. Què ha passat? the P. not REFL feel good Me seem that be very angry What have happen.ptcp Does Pedro feel bad? He seems to be very angry. What happened? Target sentences: (La Sílvia no va mencionar que) l (àvia) Àngela havia comptat les errades the S. not PST mention that the.grandma A. have count.ptcp the mistakes (de les frases). of the sentences (Silvia did not say that) (grandma) Angela has counted the mistakes (of the clauses) 2. Scenario C: Context (question): No m agrada que tota la família estigui sota arrest domiciliari. A més, em not me.like that all the family be.sbj under arrest home Moreover me sembla que la mare està especialment nerviosa. Què ha passat? seem that the mother be especially nervous What have happen.ptcp I don t like that the whole family is put under house arrest. Moreover, the mother seems to be very nervous. What happened? Target sentences: (El pare va dir que) la (teva tia) Amèlia se n'ha anat a (la ciutat de) Màlaga. the father PST say that the your aunt A. REFL CL.PST go.ptcp to the city of M. (Father said that) (your aunt) Amelia left for (the city of) Màlaga. (4a) consists of a short subject, a short verb, and a short object (condition short S / short O). (4b) contains a short subject, the (short) verb, and a long branching object (condition short S / long O). In (4c) the subject is long and branching, while the object is short (condition long S / short O). (4d) presents the case where both the subject and the object are long and branching (condition long S / long O). The exact patterns are repeated in (4e-h) with the only difference that the clauses are embedded in the matrix clause La Bàrbara suposa que Barbara assumes that. A short constituent consists of three or four syllables (l à.gui.la eagle, la.bàr.ba.ra Barbara ). A long constituent consists of at least six syllables. In (4) the long subject has seven syllables (la.me.va.gra.nà.gui.la my big eagle ) and the long object has eight syllables (el.ra.to.lí.del.meu.ger.mà the mouse of my brother ). The branching condition is ambiguous. Each branching constituent is syntactically and prosodically branching but only the latter one plays a role in Catalan (D Imperio et al. 2005:71). Each branching constituent consists of at least 80

89 Chapter 3: Phrasing Patterns in Catalan SVO Structures two prosodic words. The branching subject and object in (4) include three prosodic words each. Short constituents are at the same time non-branching, and long constituents are at the same time branching. 9 In addition, the matrix subject in the embedded contexts is always short and nonbranching. The context (question) in (4) introduced the participants to a context. The context ends in a question for which the target sentences function as answers. The context questions What happened? causes the target sentence(s) to be understood as an all new / broad focus answer. The àguila-scenario described above contains four examples for the root SVO pattern (cf. 4a-d) as well as four examples for the embedded SVO pattern (cf. 4e-h). There are three different scenarios, thus 24 different sentences (12 simple SVO sentences; 12 complex SVO sentences). Each single sentence was uttered by ten speakers, so that there are 240 sentences for part 1a (simple SVO phrasing; 120 sentences) and part 1b (complex SVO; 120 sentences) Results Boundary Cues In Chapter 2, several boundary cues were introduced for detecting intonational boundaries. The following tables and figures quantify the different boundary types and discuss some difficulties in the boundary judgment. Table 2 shows the percentages (and the absolute number) of the total number of ip- and IntP-boundaries after the preverbal subject. Table 3 shows the number of the different realizations of the two intonational boundaries. Figure 1 illustrates the distribution of ip- and IntP-boundaries with respect to the length of the syntactic constituents. Figure 2 shows an example of an unclear boundary. Table 2 indicates the summarized numbers for ip- and IntP-boundaries. There are altogether 361 boundaries in the data (cf. row 4). 56.2% of them are realized as ip-boundaries and the remaining 43.8% are realized as IntP-boundaries. The percentage values are practically the same for the two individual experiments (cf. row 2 and 3). 9 Cf. D Imperio et al. (2005) for an experiment which sets the two factors apart. 81

90 Chapter 3: Phrasing Patterns in Catalan SVO Structures SVO (106) Emb. SVO (255) Total (361) IntP 42.5% (45) 44.3% (113) 43.8% (158) ip 57.5% (61) 55.7% (142) 56.2% (203) Table 2: Percentages (and absolute number) of the total number of IntP- and ip-boundaries after the preverbal subject The total of IntP- and ip-boundaries of Table 2 (row 4) is split up in Table 3 below in order to show how the intonational boundaries are realized in detail. Due to this, row 1 of Table 3 gives the different possible realizations of boundaries at the two levels. Additionally, the factor length is given (last column of Table 3). The factor length is not split up for the two experiments (cf. row 2 and 3); only the total value of 86% is given (row 4). In what follows, the values of the different IntP- and ip-boundaries are presented in detail (row 2, 3, and 4). After that, a brief note on the factor length is given. Length, however, is discussed in detail later in this section (cf. Figure 3 below). IntP ip Length Pauses (P) Complex Cont. Sust. visible audible Boundary Rise Pitch H% L% H% L% (P) no (P) H-!H- L- SVO (106) 4.7% (5) 0% 13.2% (14) 0% 6.6% (7) 17.9% (19) 46.2% (49) 4.7% (5) 6.6% (7) Emb. SVO (255) 6% (15) 0.4% (1) 9.4% (24) 0.4% (1) 5.1% (13) 23.1% (59) 40.8% (104) 7.5% (19) 7.5% (19) Total (361) 5.5% (20) 0.3% (1) 10.5% (38) 0.3% (1) 5.5% (20) 21.6% (78) 42.4% (153) 6.6% (24) 7.2% (26) 86% Table 3: Percentages (and absolute number) of the type of the post-subject boundary realizations in the simple and complex SVO experiments The second and third row present the number of the realizations of the different types (in percent and absolute numbers) for simple and complex SVO structures, respectively. The last row summarizes the number of both experiments. Due to the fact that the pattern of the experiments is similar, only the last row is explained. As for the IntP-level, the most common marking of the 82

91 Chapter 3: Phrasing Patterns in Catalan SVO Structures boundary results from complex boundary tones. 10 Some of them are not followed by a pause (21.6% of all 361 boundaries). High boundary tones (H%) that are not followed by a pause are also appear (10.5%). Low boundary tones (L%) are virtually never realized sentence-internally (0.3%). As for the ip-level, the continuation rise (H-) is the most common realization (42.4%). It is also the most common one with respect to both levels. In addition, the total number of continuation rises increases if the high IntP-boundary tones (H%) are also considered. They represent continuation rises at the IntP-level. Thus, at least 58.4% ( ) of the boundaries are realized by a continuation rise. 11 This corresponds to the findings in Frota et al. (2007), who state that prosodic breaks in Romance are predominantly marked by a High boundary tone. The preboundary stretch tends to be realized as a continuation rise (p. 135). The factor length (last column of Table 3) has to be considered separately from the different types of realizations. Whereas the latter are distributed complementarily (e.g. either the tone is realized as a sustained pitch or as a continuation rise), length overlaps all realizations (i.e. the sustained pitch as well as the continuation rise can additionally be marked by a long preboundary syllable). As Table 3 indicates, 86% of the boundaries are marked by longer syllables. I return to the results of the length measurement of preboundary and non-preboundary syllables (cf. Figure 3 below) later. Until now, nothing has been said concerning the distribution of the boundaries. For this reason, the relation between constituent length and the type of boundary (IntP vs. ip) is considered in the following paragraphs. I concentrate on the post-subject position of simple SVO structures. Experiment 1a (simple SVO) has four different conditions (short S / short O; short S / long O; long S / short O; and long S / long O) with thirty sentences each. The conditions are given in the x-axis of Figure 1. The y-axis shows the number of boundary realizations (in absolute numbers). The four different lines in the diagram represent IntP-boundaries, ipboundaries, unclear cases, and cases where no boundary appeared. 10 Recall that only boundary tones that consist of different tones (such as the most common realized tone LH%) are considered to constitute this class (cf. Chapter 2). 11 If the High tone of the complex boundary tone LH% is also considered as a continuation rise, the total number of continuation rises increases further. 83

92 Chapter 3: Phrasing Patterns in Catalan SVO Structures Figure 1: Relation between length of constituents and intonational boundary at the post-subject position Figure 1 clearly shows the tendency for the increasing number of IntP-boundaries when the subject is long. In the short subject conditions the boundary after the subject is realized only 5+6 (11) times as an IntP-boundary (i.e. 18%). In the long subject conditions, though, the boundary is in (33) cases of this type (i.e. 55%). This is an increase of 300%. The number of ipboundaries is very high in the short S / short O condition (21 realizations; i.e. 70%), and decreases in the three further conditions, but still remains relatively high (13, 11, and 12 realizations, i.e. around 43% on average). As for the two remaining cases (unclear and no boundary), they reach an interesting number only in the short S / long O condition. In around 37% (4+7 (11)) of the cases there are either no or unclear boundary tones, whereas the highest percentage in the other conditions comes to only around 13% (1+3 (4)). By looking at this aspect from the reverse side, it can be said that the clear separation of the subject from the following verb comes to only 63% (i.e realizations) in that condition, whereas the average number of the three other conditions comes to 91%. Short S / long O is the condition where Catalan shows a tendency for the (SV)(O) grouping and the post-subject boundary realizations are reduced due to balance effects of the length of prosodic constituents (D Imperio et al. 2005, Prieto 2005, Frota et al. 2007). The interesting percent value of that condition in Figure 1 thus reflects the tendency of Catalan to not place a boundary after the subject if the object is long. 84

93 Chapter 3: Phrasing Patterns in Catalan SVO Structures Figure 1 (but also Table 2 and Table 3) shows a further important aspect: there is variation with respect to the nature of the boundaries. In some instances, the breaks dividing the subject from the predicate are straightforward intonational phrases, but in some cases they are clear instances of intermediate phrases. In the long subject conditions, there is a slight preference for IntPs (55%), while Table 2 indicates a slight preference for ips (56%). Nevertheless, the number of IntPs and ips seems to be distributed equally at a percent value of 50%. This result represents empirical evidence for the notion prosodic phrase as a hypernym for intonational phrase and intermediate phrase (cf. Chapter 2). Although most of the junctures could clearly be classified as being an intonational boundary or not, some instances were difficult to judge. Unclear cases arise due to the fact that an ambiguity between delayed peak and continuation rise is possible. Even though the schematic diagrams of a continuation rise and a delayed peak in Chapter 2 (cf. figure 9 & figure 10) are clear, the actual situation in uttered sentences is not always so clear. Prieto (2007:8) shows that the peak delay is larger in words with antepenultimate stress than in words with stress closer to the end of the word. In Figure 2 the ambiguity between a continuation rise and a delayed peak is illustrated by?? in the tonal tier. The H peak is located seconds after the end of the accented syllable, whereas it is located only seconds before the end of the word. The proximity to the word edge can lead one to the conclusion that there is a phrase break H- present. On the other hand, since the H peak is not totally located at the word edge, it is possible to conclude that it is the delayed peak of the rising tone and not a phrase break. Hence, an ambiguity arises. This ambiguity is marked by? in the data in the appendix. Unclear cases are judged as presenting no boundary. This decision has an influence on the (SVO) phrasing in particular and is discussed in Section

94 Chapter 3: Phrasing Patterns in Catalan SVO Structures >H or H-? Pitch (Hz) L*+H?? L*+H L* L-L L Àn ge la va comp tar les e rra des S V O Time (s) Figure 2: Ambiguity between delayed peak and continuation rise in the F0 contour of L Àngela va comptar les errades Angela counted the mistakes of speaker GM. The ambiguity is marked with?? in the tonal tier. In addition, in a few cases it was difficult to judge if the detected break should have been categorized as an ip- or an IntP-boundary. This difficulty arose when it was unclear if a continuation rise was followed by an audible pause or not. Due to the fact that there is an intonational boundary in either case, such cases were judged as being an ip-boundary. In the appendix, the corresponding instances of H- are marked by the preceding tag (4?) (representing a possible break index 4). Finally, I return to syllable length. The bar diagrams (Figure 3 and Figure 4) illustrate the different length of preboundary and non-preboundary syllables of both experiments (1a and 1b). The first diagram considers length in milliseconds (Figure 3), while the second diagram gives the corresponding percent values (Figure 4). The height of the bar represents the magnitude of the corresponding variable. For the measurement, only syllables consisting of two pronounced segments were considered irrespective of being stressed or not, (6) In the case of infinitives the word-final /r/ is not uttered such that tar (of comptar count ) is pronounced as [ta], nar (of mencionar mention ) as [na] and dir say as [dı]. The infinitives as well as the past tense form robà stole have word-final stress. Not considered were the closed syllable nat of se n ha anat did go and the word-final material lia of Amèlia and via of Sílvia. The latter two have the described possibility of not uttering the hiat between the two vowels. If the hiat is not uttered what is the typical case according to Cabré & Prieto (2004) the wordfinal syllable consists of three segments or, if the hiat is uttered, the ultimate syllable has just one segment. No case fulfils the criteria of two phonetically realized segments. 86

95 Chapter 3: Phrasing Patterns in Catalan SVO Structures (6) Syllables considered for length measurement [consonant + vowel (CV)]: [lə] <la> (Àngela, àguila eagle ) [ta] <tar> (comptar count ) [ba] <bà> (robà stole ) [na] <nar> (mencionar mention ) [rə] <ra> (Bàrbara) [sə] <sa> (suposa supposes ) [rə] <re> (pare father ) [dı] <dir> (dir say ) The two diagrams (Figure 3 and Figure 4) show an important finding: there is a clear difference between preboundary and non-preboundary syllables. Preboundary syllables have an average length of 208ms (100%), while non-preboundary syllables have a length of 119ms (57%). Thus, they are 89ms shorter than syllables that are followed by a break. This corresponds to a percent value of 43%. The average for all ten speakers is given in the last column, indicated by (Ø). Figure 3: Length in milliseconds of preboundary and non-preboundary CV-syllables 87

96 Chapter 3: Phrasing Patterns in Catalan SVO Structures Figure 4: Length in percentages of preboundary and non-preboundary CV-syllables Additionally, the average length of preboundary and non-preboundary syllables for each speaker (black and grey bar respectively) is given in Figure 3 and Figure 4. By looking at single speakers, two extremes can be seen. On the one hand, for four speakers (CP, DS, GV, and MO) the difference comes to 50% or more. Their preboundary syllables are more than 100ms longer than non-preboundary syllables. On the other hand, one speaker (MM) shows only a very small difference. The difference comes to 20% and non-preboundary syllables are 31ms shorter. No other speaker has such a small difference between the syllable types. CB, for example, has a difference of 61ms. As mentioned before (cf. last column of Table 3), the boundary cue length of the preboundary syllable appears additionally to the boundary tones. The clear difference between preboundary and non-preboundary syllables can be seen by all ten speakers irrespective of the actual realization (such as H-, H% or LH%) of the produced boundary tones. 88

97 Chapter 3: Phrasing Patterns in Catalan SVO Structures Results Phrasing Patterns In this section the results of the intonational grouping of simple and complex SVO structures are presented. First, the results of the simple SVO experiment (1a) are described and then the results of the complex SVO experiment (1b). Both sections are followed by a discussion Simple SVO (experiment 1a) Figure 5 illustrates the total result of experiment 1a (3 scenarios x 4 sentences x 10 speakers = 120 sentences). The parentheses ( ) represent the prosodic grouping. In Catalan, there are four possible groupings (S)(VO), (SVO), (S)(V)(O), and (SV)(O) although speakers clearly prefer to place a phrase break after the subject. A mean of around 80% of the cases in normal speech rate are instances of (S)(VO). The (SVO) 13 grouping comes to 15%, (S)(V)(O) to 3.33% and (SV)(O) to only 0.83%. Figure 5: Prosodic grouping of simple SVO structures in Catalan In the following figures, examples of the F0 contour of each grouping are given. Figure 6 illustrates the (S)(VO) grouping. The post-subject boundary is marked by a continuation rise. It can clearly be seen on l àguila eagle. The rise starts in the proparoxytonic syllable l à and rises 13 For comments on the relatively high percentage of (SVO) groupings, cf. discussion below. 89

98 Chapter 3: Phrasing Patterns in Catalan SVO Structures steadily until the end of the subject, where the ip-boundary tone H- is located. V and O are not separated by a boundary Pitch (Hz) 50 L*+>H H- L*+>H L* L-L% 3 4 L À- -gui- -la ro- -bà el ra- -to- -lí S V O Time (s) Figure 6: Catalan (S)(VO) phrasing Waveform and intonation contour of the utterance L àguila robà el ratolí The eagle stole the mouse of speaker GV (sentence 29_SVO_GV) Figure 7 illustrates the (SVO) phrasing. There is no break either after the subject or after the verb. The peak of the subject s pitch accent is delayed and located on the first posttonic syllable. At the word boundary, there is no special tonal movement Pitch (Hz) L+>H*!H* L+H* H+L* L-L% 1 4 L Àn- -ge- -la va comp- -tar les e- -rra- -des S V O Time (s) Figure 7: Catalan (SVO) phrasing Waveform and intonation contour of the utterance L Àngela va comptar les errades Angela counted the mistakes of speaker MO (sentence 21_SVO_MO) 90

99 Chapter 3: Phrasing Patterns in Catalan SVO Structures The (S)(V)(O) phrasing is shown in Figure 8. The break after the long subject is indicated by sustained pitch (!H-). The peak of the preboundary pitch accent is delayed and reaches its highest point at the end of the first posttonic syllable. On the second posttonic syllable, the pitch does not rise any further but sustains at the same level. The break after the verb, separating the verb and the object, is marked by an audible pause (H%) Pitch (Hz) 50 L+>H*!H* L+>H*!H- L+H* H% L-L% La te- -va tia A- -mè- -li -a se n ha a- -nat a Ma- -la- -ga S V O Time (s) Figure 8: Catalan (S)(V)(O) phrasing Waveform and intonation contour of the utterance La teva tia Amèlia se n ha anat a Malaga Your aunt Amèlia went to Malaga of speaker RS (sentence 39_SSVO_RS) The last of the four groupings, (SV)(O), is presented in Figure 9. The subject is not separated from the verb by any tonal movement or a pause. The only break in the structure is located after the verb and is marked by a continuation rise (H-). 91

100 Chapter 3: Phrasing Patterns in Catalan SVO Structures Pitch (Hz) L+>H* H+L* H- L+>H* H+L* L-L% L Àn- -ge- -la va comp- -tar les e- -rra- -des de les fra- -ses S V (O (PP)) Time (s) Figure 9: Catalan (SV)(O) phrasing Waveform and intonation contour of the utterance L Àngela va comptar les errades de les frases Angela counted the mistakes of the sentences of speaker MO (sentence 22_SVOO_MO) Whereas Figure 5 gives the total results of all sentences irrespective of the different conditions, Figure 10 gives the percentages of the realizations of the different groupings for each condition (short S / short O; short S / long O; long S / short O; and long S / long O). Figure 10: Percentages of simple SVO phrasing patterns in the four different conditions 92

101 Chapter 3: Phrasing Patterns in Catalan SVO Structures As can be seen, (S)(VO) is the predominant phrasing across all conditions, with a value of mostly above 80%. The most robust (S)(VO) realizations arise in the long S / long O condition; nearly all examples are realized with a break after the subject (97%). Interestingly, only in the short S / long O condition is its value comparatively low. It comes to only 57%. In general, this condition deviates in many aspects from the other three conditions: besides the high number of non-(s)(vo) groupings, it is the only condition where the (SV)(O) grouping shows up and where all four groupings are realized. Besides the (S)(VO) grouping, all other groupings come to their highest number of realizations ((SVO): 33%; (S)(V)(O): 7%; and (SV)(O): 3%). Thus one can conclude that the condition where the subject is short and the object is long is the only condition that influences the phrasing decisions the most. It weakens the predominance of the (S)(VO) grouping Discussion The goal of the simple SVO experiment was to set the basis for the complex SVO experiment and to repeat the study of D Imperio et al. (2005) with more speakers and a clear information structural context. The two hypotheses in Section concern only sentential objects and as such they are not discussed here. The phrasing pattern found in part 1a is in line with the findings of D Imperio et al. (2005) and Prieto (2005): The most common phrasing in Catalan is (S)(VO), and (SV)(O) arises when the object is long. As for the first finding, D Imperio et al. (2005:71) ascribe the exclusive trigger of the (S)(VO) pattern to a long subject (i.e. branching S). This conclusion cannot be exclusively drawn here, but the data do not present counterevidence either. On the one hand, (S)(VO) comes to more than 80% in the long S conditions. However, in the short S / short O condition the subject is also phrased separately in more than 80% of the cases. Thus, a long subject cannot have such an influence because a short subject is also mostly phrased separately. On the other hand, the object seems to play an important role; this is the case in both long O conditions. Across the long S conditions, a long object increases the number of (S)(VO) groupings. Whereas in long S / short O (S)(VO) comes to 83%, it comes to 97% in the long S / long O condition. If the subject is short, though, a long object decreases the number of (S)(VO) groupings (57%). 93

102 Chapter 3: Phrasing Patterns in Catalan SVO Structures As for the second finding, (SV)(O) shows up only in the short S / long O condition. This is in line with D Imperio et al. (2005:70) and Prieto (2005:214). D Imperio et al. (2005:71) say that the tendency for S to phrase with the following V is a clear effect of length. In their data, the (SV)(O) phrasing comes to 33.1% (D Imperio et al. 2005:68f). However, according to my data, (SV)(O) is realized only very rarely and does not come to such a high percentage although it also shows up only in the short S / long O condition. The difference might appear because the sentences were uttered only at a normal speech rate in the simple SVO experiment, while D Imperio et al. (2005) examined different rates of speech. They show that (SV)(O) appears more often at a fast speech rate. The aforementioned length effect, nevertheless, has an important influence in the present data. Although it does not significantly increase the SV grouping, it weakens the otherwise robust (S)(VO) phrasing and renders possible (higher numbers of) different phrasings. One interesting aspect concerns the high percentage of the (SVO) grouping in the short S / long O condition. The (SVO) grouping is very atypical for Catalan (D Imperio et al. 2005; Prieto 2005). They found only some sporadic (SVO) groupings by one single speaker (D Imperio et al. 2005:68). The question arises as to why (SVO) can come to 33% in general and why it comes to that number in the short S / long O condition in particular. The cause can be explained easily; nearly all instances of (SVO) arise due to an unclear post-subject boundary. It was not possible to judge if there was a boundary after the subject or not (cf. below). Unclear boundaries are judged as if there were no boundary. In my data, only two instances of (SVO) out of 18 are clear instances of (SVO). Nearly 90% of the cases are unclear. If the data is reinspected closely and detailed, it might turn out that some unclear boundaries are boundaries, while some unclear boundaries are not boundaries. Consequently, the percentage of 15% for (SVO) might be smaller. As for the second part of the question, (SVO) arises in the short S / long O condition in particular, because of the length effect. The long object causes the need for a post-subject boundary to be not as strong as normal. If the boundary is not as strong anymore, it can either disappear completely or it can simply become less clear. Due to the fact that unclear cases were dealt with as if there is no boundary, the phrasing (S unclear VO) turns out to be (SVO). Thus the number of unclear cases increases, especially in the short S / long O condition. 94

103 Chapter 3: Phrasing Patterns in Catalan SVO Structures Complex SVO (experiment 1b) Figure 11 shows the results of experiment 1b (3 scenarios x 4 sentences x 10 speakers = 120 sentences). The bar diagram illustrates the percentages given in the table below the diagram. The first line of the table indicates the four main groupings as well as the general number of miscellaneous groupings ( misc. ). Figure 11: Prosodic grouping of complex SVO structures in Catalan (given in percentages) The general order is SVqSVO. S stands for subject, V for verb, and O for object, cf. (7a). The object, however, is sentential and also consists of a subject, verb, and object, cf. (7b,c). A new symbol has been added: q stands for que that, representing the complementizer. Consequently, SV preceding q represents the matrix subject and the matrix verb. SVO following q represents the embedded clause. This is illustrated in (7c). (7) Notation of complex SVO structures a. [S V O] b. [S V [O]] c. [S V [qsvo]] The abbreviation SVqSVO simply reflects the word order, independently of the length and branchingness of their constituents. As before, the parentheses ( ) represent the prosodic 95

104 Chapter 3: Phrasing Patterns in Catalan SVO Structures grouping. Due to the fact that in these complex clauses at least one clear boundary has been found, there are no complete sentences judged as unclear (i.e. the matrix clause and the embedded clause are never grouped into one single prosodic phrase). However, there are examples where some boundaries remained unclear. Again, these cases have been treated as if the unclear boundaries were not boundaries at all. They are marked with? in the tables in the appendix, which show the phrasing decisions of each single speaker. The four different main phrasing possibilities are (SV)(qS)(VO), (S)(V)(qS)(VO), (S)(V)(qSVO), and (S)(VqS)(VO). They are not the only possibilities though. The last column in the table ( misc. ) includes all other prosodic groupings which were uttered. They are briefly mentioned here. There are eight realized phrasings subsumed under the term miscellaneous : (S)(V)(qSVO) [five times], (SV)(qS)(V)(O) [five times], (SV)(qSVO) [four times], (SVqS)(V)(O) [two times], (S)(VqSVO) [one time], (SV)(qSV)(O) [one time], (S)(VqS)(V)(O) [one time], and (S)(V)(qSV)(O) [one time]. As the number of appearances in the square brackets indicates, these phrasings are not very common. For this reason, I have left them aside. However, the reader should note that the possible phrasings in contexts with embedded object clauses can be very large. Including the previously listed eight phrasings, there are 12 different prosodic groupings altogether. Thus, compared to the root context, the number of possible phrasings is considerably larger. The light grey bars indicate the number of the four main groupings and they include the number of the miscellaneous groupings. As can be seen, the four main groupings come to more than 83% of the data, while the other eight groupings come to less than 17% together. For this reason, I concentrate on only the four main groupings from now on. The dark grey bars represent the corresponding percentage value without the miscellaneous groupings. The percentage values in Figure 11 reveal that the most common phrasing pattern is (SV)(qS)(VO), where the matrix subject and the matrix verb are phrased together, followed by the group complementizer + embedded subject, followed by the group embedded verb + embedded object. This group comes to 46.67% (i.e. 56% without misc.) of the 120 sentences. The corresponding F0 contour is shown in Figure 12. The second favoured phrasing is (S)(V)(qS)(VO), which comes to 20% (i.e. 24%). This phrasing pattern differs from the former in that the matrix subject and matrix verb are phrased separately (cf. Figure 13). These two groupings are followed by (SVqS)(VO), which comes to 8.33% (i.e. 10%). Here, the matrix clause is phrased together with the complementizer and the embedded subject. The embedded 96

105 Chapter 3: Phrasing Patterns in Catalan SVO Structures verb and the embedded object create another prosodic group (cf. Figure 14). 14 The next grouping is (S)(VqS)(VO), which also comes to 8.33% (i.e. 10%). It is characterized by phrasing the matrix verb, the complementizer and the embedded subject together (cf. Figure 15). Figure 12 shows the pitch track of the most common phrasing pattern (SV)(qS)(VO). There is no boundary after the matrix subject. The pitch does not rise after the peak of the pitch accent. In the following figures for the pitch tracks, the break index 1, which signals a word boundary, is used in order to signal that there is no prosodic break in the ip- or IntP-level (break index 3 and 4 respectively). There is a break, though, after the verb. This break is signaled by an audible pause and the high tone is downstepped (!H%). The next break is located after the long subject. There is a complex boundary tone (LH%) followed by a visible pause. The embedded verb and object are not separated by a break Pitch (Hz) L+H* L+H* H+L*!H% L*+>H L+H* LH% H+L* L-L% La Sil- -via no va men- -cio- -nar quel à- -via Àn- -ge- -la ha--via comp- -tat les e- -rra- -des S V COMP S V O Time (s) Figure 12: (SV)(qS)(VO) phrasing of sentence La Silvia no va mencionar que l àvia Àngela havia comptat les errades Silvia did not say that grandmother Angela has counted the mistakes of speaker MO (sentence 27_Emb_MO) 14 Most of the (SVqS)(VO) groupings evolve from the fact that some boundaries have been classified as unclear (cf. appendix: corresponding examples in 25 (speaker MM), 26 (speaker RS), 36 (speaker MM), etc.) 97

106 Chapter 3: Phrasing Patterns in Catalan SVO Structures Pitch (Hz) L+H* LH% H+L*!H% L+H* LH% H+L* H+L* L-L% La Síl- -vi- -a no va men- -cio- -nar que l Àn- -ge- -la ha- -via comp- -tat les e--rre--des de les fra- -ses S V COMP S V (O (PP)) Time (s) Figure 13: (S)(V)(qS)(VO) phrasing of sentence La Sílvia no va mencionar que l Àngela havia comptat les erredes de les frases Silvia did not say that Angela has counted the mistakes of the sentences of speaker MO (sentence 26_Emb_MO) Figure 13 shows the pitch track of the second most common grouping, (S)(V)(qS)(VO). Here, the matrix subject and matrix verb are separated by a break. This can be clearly seen by the complex boundary tone LH% and its very high rise. There are two further breaks: one after the matrix verb (an audible pause,!h%) and one after the embedded subject (the complex boundary tone LH%). Again, the embedded verb and the embedded object are not separated by an intonational boundary. The next grouping is (SVqS)(VO) and its pitch track is given in Figure 14. There is just one break after the embedded subject. It is marked by the complex boundary tone LH%. There is no boundary after the matrix subject. The delayed peak is located on the following word. There is no break between the matrix verb and the complementizer either. The interruption of the pitch track and its high starting point on the complementizer is caused by a consonantal effect on F0 (cf. Gussenhoven 2004:7). The raised F0 is a result of a pitch perturbation of [k]. 98

107 Chapter 3: Phrasing Patterns in Catalan SVO Structures Pitch (Hz) L+H*!H* L+H* L-H% L-L% El pa- -re va dir quel A- -mè- -li- -a se n ha a- -nat a la ci- -u- -tat de Ma- -la- -ga S V COMP S V (O (PP)) Time (s) Figure 14: (SVqS)(VO) phrasing of the sentence El pare va dir que l Amèlia se n ha anat a la ciutat de Màlaga The father said that Amèlia went to the city of Malaga of speaker CP (sentence 42_Emb_CP) The fourth grouping, (S)(VqS)(VO), is illustrated in Figure 15. There are two sentenceinternal main breaks marked by the complex boundary tone LH%: one after the matrix subject; the other one after the embedded subject. One phonetic characteristic should be noted here. Interestingly, if the end of the word is marked by a complex boundary tone, the peak of the pitch accent is not delayed but aligns with the right side of the strong syllable. By inspecting the three preceding pitch tracks, it can be seen that they show the same pattern. On the other hand, if the pitch accent is not followed by a complex boundary tone, a delayed peak is possible. 15 There is no break between the matrix verb and the complementizer. There is a transition between the verb s pitch accent and the low leading tone of the embedded subject. There is no special tonal movement around the complementizer. There is also no break between the embedded verb and object. 15 This is not a generalization: it is a description of the data. Further research has to be done in order to generalize this statement. 99

108 Chapter 3: Phrasing Patterns in Catalan SVO Structures Pitch (Hz) 50 L+H* L-H% L+>H* L+H* L-H% L+>H* L+H* H+L* L* L-L% La Bàr- -ba- -ra su- -po- -sa que l à- -gui- -la ro- -bà el ra- -to- -lí del meu ger- -mà S V COMP S V (O (PP)) Time (s) Figure 15: (S)(VqS)(VO) phrasing of the sentence La Bàrbara suposa que l àguila robà el ratolí del meu germà Barbara supposes that the eagle stole my brother s mouse of speaker DS (34_Emb_DS) After the description of the general results of the complex SVO experiment, the results for each condition are given. Figure 16 presents the four conditions with the corresponding percentage values in the table. The line diagram illustrates the values. If the values for each condition are added, the total is not 100%. This is due to the fact that the difference represents the percentage of the miscellaneous groupings. Nevertheless, they are not listed in the figure. As shown in Figure 16, while the general order of the four groupings with respect to the frequency (cf. Figure 11) is by and large reflected in three conditions, it is perturbed in the short S / long O condition. The predominant grouping (SV)(qS)(VO) is the most common grouping in the three conditions short S / short O (~47%), long S / short O (60%), and long S / long O (60%). However, it is not the most common in the condition short S / long O (only 20%). It is (S)(V)(qS)(VO) that is realized most often (30%). In addition, the two groupings (SVqS)(VO) and (S)(VqS)(VO) have both the same relatively high percentage, while they never have the same frequency in the other three conditions. Thus, as in the simple SVO pattern, the condition short S / long O changes the general phrasing pattern. 100

109 Chapter 3: Phrasing Patterns in Catalan SVO Structures Figure 16: Percentages of complex SVO phrasing patterns in the four different conditions One note with respect to the miscellaneous groupings should be mentioned here. While the main groupings never separate the embedded verb and object with a prosodic break (i.e. (VO)), there are some miscellaneous groupings that do (i.e. (V)(O)). Interestingly, they only do in the two conditions with a long object. There are five realizations in the short S / long O condition and four realizations in the long S / long O condition. Still, the number of these groupings is very low and is not considered anymore. In what follows, four important characteristics of the prosodic grouping of complex SVO structures are more closely described (leaving the miscellaneous phrasings aside and thus referring to the values of the dark grey bars of Figure 11 (unless mentioned otherwise): 1) Separation of the object clause, 2) Phrasing of matrix subject and verb, 3) Phrasing of embedded subject, and 4) Phrasing of embedded verb and object. 1. Separation of object clause: The two most common groupings place a boundary between the matrix verb and the object clause. Thus, the total number of separated object clauses comes to 80% (56% + 24%). Even in the short S / long O condition, the object clause is prosodically separated from matrix material at least 50% of the time (cf. Figure 16), although the number is smaller. Among these breaks 60% are realized as an ip-boundary and 40% as an IntP-boundary. This is shown in Table 4. In both groupings ((SV)(qS)(VO) and (S)(V)(qS)(VO)), the boundary 101

110 Chapter 3: Phrasing Patterns in Catalan SVO Structures is most often realized on the ip-level. The number of IntP-boundaries in the (S)(V)(qS)(VO) grouping comes to only 30%, while it comes to 45% in the other grouping. Table 4: (SV)(q (S)(V)(q Total ip 31 (55%) 17 (70%) 48 (60%) IntP 25 (45%) 7 (30%) 32 (40%) Percentages of boundary types immediately preceding the embedded clause (all ten speakers, all three scenarios) The variation with respect to the realized boundary tone seems to be typical for Catalan. Experiment 1a (simple SVO) shows the same pattern. Furthermore, Astruc (2005) also shows that sentential adverbs are separated by (low) ip-boundary tones (32.7%) nearly as often as by (low) ɩ-boundary tones (33.2%; cf. Astruc 2005:154: table 4.3). What do previous studies say about boundaries separating the embedded clause from the matrix clause? Downing (1970) exclusively devotes his attention to obligatory boundary insertion (his OBI convention). He argues that obligatory boundaries are inserted at the leftmost and rightmost edge of root sentences (Downing 1970:31) but not of embedded sentences since they are not root clauses. The difference between (8a) and (8b) is that in the former the complement is a direct quotation, whereas in the latter example the complement expresses reported speech. (8) (Examples taken from Downing 1970:89) a. Ann said, / I ll make you some sandwiches b. Ann said (that) she would make us some sandwiches. Downing argues that complements of certain quotative verbs (such as say, remark, tell (someone), ) are identical to root clauses and thus are realized with obligatory boundaries, i.e. set off by pauses (Downing 1970:89). The complement in (8b) is simply an ordinary complement and is not obligatorily set off by a pause (Downing 1970:90). However, it can be optionally set off by a pause (Downing 1970:113ff.). Nespor & Vogel (1986/2007:189) follow the idea by Downing (1970) that root sentences delimit an intonational phrase, while those sentences that are not root sentences do not. By restructuring the intonational phrase, though, it is possible to place an intonational phrase boundary after the matrix verb. One factor for restructuring is length. N&V (1986/2007:194f.) 102

111 Chapter 3: Phrasing Patterns in Catalan SVO Structures illustrate the tendency for intonational phrases to be grouped in constituents of a more or less uniform, average length by giving examples (9), where (9a) can be grouped as (9b). (9) (Examples 13a and 13d from N&V 1986/2007:194f.) 16 a. [ I [Jennifer]ϕ [discovered]ϕ [that her attic]ϕ [had been invaded]ϕ [last winter]ϕ [by a family]ϕ [of squirrels]ϕ] I b. [ I [Jennifer]ϕ [discovered]ϕ] I [ I [that her attic]ϕ [had been invaded]ϕ [last winter]ϕ] I [ I [by a family]ϕ [of squirrels]ϕ] I In his work on subject and object clauses (inter alia), Truckenbrodt (2005) achieves different results. Subject clauses (in the Vorfeld) are systematically separated from the matrix clause and object clauses (in the Nachfeld) are not separated from the matrix clause. Due to the fact that the subject clause is located in the Vorfeld, the relevant boundary is on its right. All three cited studies agree on the non-existence of obligatory IntP-boundaries preceding object clauses. My results match insofar their findings, because only 80% of the clauses are separated and among them there are only 40% IntP-boundaries. However, ip-boundaries have to be considered, too. Thus, even though 80% does not reflect obligatoriness, Catalan clearly tends to separate object clauses from preceding material. 2. Phrasing of matrix subject and verb: As Figure 11 and Figure 16 show, there are two groupings in which the matrix subject is phrased with the matrix verb: (SV)(qS)(VO) and (SVqS)(VO). Thus the total number of S and V together in a group comes to 66% (56% + 10%). In the other two groupings the matrix subject is phrased alone. The total number of separated matrix subjects comes to 34% (24% + 10%). 3. Phrasing of embedded subject: The embedded subject is phrased separately from the matrix material and from the embedded verb and object in 80% of the groupings (i.e. only in the two main groupings; 56% + 24%). The embedded subject is not phrased alone, though, because the complementizer is part of that prosodic group. The complementizer is unstressed and consequently counts as a clitic. In the remaining 20%, the embedded subject phrases with matrix material (cf. (SVqS)(VO) and (S)(VqS)(VO)). It never groups with following material. Thus in 16 In example (9) of N&V (1986/2007), there is always a prosodic boundary between the matrix clause and the embedded clause. However, this prosodic break is a phonological phrase, constructed by a syntactic algorithm (N&V 1986/2007:168). 103

112 Chapter 3: Phrasing Patterns in Catalan SVO Structures all groupings (i.e. 100%), the embedded subject is followed by a boundary. Consequently, the tendency for (SV) does not exist in embedded clauses. 4. Phrasing of embedded verb and object: In all four main groupings, the embedded verb phrases together with the embedded object (i.e. 100%). There is never a boundary in-between them, nor does the verb phrase with the preceding subject Discussion In Section two hypotheses were formulated. The first one said that sentential objects increase the number of the single group phrasing of matrix SV. The second one said that sentential objects are not obligatorily separated by prosodic means from the preceding clause. As for the first hypothesis, it has been clearly fulfilled. On the one hand, the number of (SV) phrasing in simple SVO structures (cf. experiment 1a) is very low. On the other hand, the matrix (SV) phrasing in complex structures comes to 56%. Hypothesis 1 is also validated when comparing the value of 56% to the percentage value of 33.1% in D Imperio et al. (2005). The (SV) grouping increases by 20%. Consequently, the length of the object (i.e. its characteristic to be sentential) has a precise effect on the phrasing of the (matrix) subject and verb. The tendency of these two constituents to phrase together increases considerably; this clearly supports the findings for Catalan in Prieto (2005) and D Imperio et al. (2005). The length of the internal constituents of the object clause plays insofar a role, that in the short S / long O condition the number of matrix (SV) decreases. Otherwise, the most common grouping persistently is the (SV)(qS)(VO) grouping. Nevertheless, the two most common groupings are further on (SV)(qS)(VO) and (S)(V)(qS)(VO). For this reason, the effect of the object clause internal object is not considered in the analysis (Section 3.4). The analysis of complex SVO phrasing deals only with the four main groupings. As for the second hypothesis, it can also be taken as fulfilled. The results concerning the separation of object clauses show that the boundary preceding the embedded clause is not obligatory, since 20% of the object clauses are not preceded by a boundary. Nevertheless, Catalan has a significant tendency to place a boundary before the embedded object clause. Thus any approach should consider that there rather is a boundary. The results further show that there is a great variability for prosodic constituents to have an ip- or ɩ-boundary tone. By using the 104

113 Chapter 3: Phrasing Patterns in Catalan SVO Structures hypernym prosodic phrase all boundary tones can be considered though. This variation of boundary tones, which seems to be normal in Catalan, might be the reason why the Catalan boundary cues described in Frota et al. (2007) and presented in Chapter 2 are not specific as to the exact level of the prosodic hierarchy Conclusion I conducted two experiments on phrasing of SVO. In general, my findings support the view put forward in D Imperio et al. (2005) and Prieto (2005) that the subject tends to be phrased with the verb if the object is long. In the first experiment (simple SVO), the sentence is a root clause in its narrow sense; the object is a DP. The subject and the object were modified for length (i.e. long/branching vs. short/non-branching). In the second experiment (complex SVO), the object is sentential. The assumption was that a sentential object counts as a long object with respect to the phrasing behavior of (matrix) SV. In the first experiment, around 80% of the clauses were instances of (S)(VO). In addition, although (SV)(O) phrasing was detected under the long object condition, this pattern was realized rather seldomly by the ten speakers (~1%). This contrasts with the average percentage of 33.1% in D Imperio et al. (2005:68). However, experiment 1b showed that a sentential object leads to a (SV) phrasing in 56% of the cases, thus supporting hypothesis 1. I modified the length of the embedded subject and the embedded object. This modification does not play a role in the phrasing of complex SVO it is rather the fact that a sentential object itself is long. In 80% of the examples, the embedded clause is prosodically separated from the matrix clause. By that, hypothesis 2 can be taken as validated. Nevertheless, Catalan shows a great tendency to separate the embedded object clause from the matrix clause. The results of experiment 1b show that there are some main characteristics concerning phrasing. They are listed here: i. the embedded object is phrased with the embedded verb ii. the embedded subject is phrased separately from the embedded verb and object iii. in most cases the embedded subject is phrased alone iv. the embedded clause is normally separated from the matrix clause v. the matrix subject has the tendency to phrase with the matrix verb Any approach on the phrasing of complex SVO structures in Catalan has to account for these characteristics. In Section 3.4, a corresponding approach is given. 105

114 Chapter 3: Phrasing Patterns in Catalan SVO Structures 3.4 An analysis for SVO and embedded SVO phrasing This section offers an analysis of the prosodic phrasing of simple and complex Catalan SVO structures. The analysis is grounded in a variant of Optimality Theory (Prince & Smolensky 1993/2004) called Stochastic OT (Boersma & Hayes 2001). The approach is based on Prieto s (2005) account for simple SVO and includes her three constraints MAX-BIN-END >> MIN-N- PHRASES >> ALIGN-XP,R. A new constraint ALIGN-CP,L is assumed which is ranked below MAX-BIN-END but higher than MIN-N-PHRASES and ALIGN-XP,R, as shown in (10). The constraint ALIGN-CP,L aligns the left edge of a CP with the left edge of a prosodic phrase and thus accounts for the pattern that the embedded clause is in general prosodically separated from the matrix clause. (10) MAX-BIN-END >> ALIGN-CP,L >> MIN-N-PHRASES >> ALIGN-XP,R By using the stochastic OT framework (Boersma & Hayes 2001), in which the constraints are ranked on a continuous ranking scale, the model can account for the variation expressed by the four most common groupings of the data (which grasp 80% of the data). I argue that ALIGN-CP,L, MIN-N-PHRASES, and ALIGN-XP,R overlap and the actual ranking of the constraints will sometimes be the reverse of their normal ranking. The different rankings derived from the underlying form in (10) account for the data. Section 3.4 is structured in the following way. The theoretical background for boundary placement is presented in Section 3.4.1, where the fundamental syntactic and phonological factors (such as the alignment and wrapping constraints as well as two eurhythmic constraints) are introduced in detail. In Section the theoretical approach for the phrasing of simple SVO structures is presented. Since I adopt in great part (with some modifications) the approach of Prieto (2005), it is introduced in detail. Her approach constitutes the basis for accounting for more complex structures. In Section 3.4.3, the analysis of complex SVO structures is offered. First, the necessary constraint rankings for the four groupings are given. Second, optionality in OT and the stochastic OT model are introduced. Finally, it is shown how the necessary rankings can be modeled in stochastic OT. 106

115 Chapter 3: Phrasing Patterns in Catalan SVO Structures Theoretical Background The Influence of Syntactic and Prosodic Factors There has been evidence in the linguistic literature that syntactic and phonological factors determine the boundary placement. This section introduces in detail the theoretical background of the four constraints relevant for my approach and used in Prieto (2005): ALIGN-XP,R, WRAP-XP, MAX-BIN, and *P-PHRASE (i.e. MIN-N-PHRASES). One main aspect of syntactic factors is the relation of certain XPs to prosodic structure (Selkirk s 1986, 1995b, 2000 End-Based Theory, Truckenbrodt s 1995, 1999, 2005 Wrap constraints; cf. below). 17 A further factor mentioned in the literature is syntactic heaviness (N&V 1986/2007, Kanerva 1990, Frascarelli 2000), in the sense that a constituent that is syntactically complex / syntactically branching has a greater tendency to be phrased independently. Another factor is certain syntactic constituents, i.e. certain type of constructions which form prosodic domains on their own, for example in the case of parenthetical expressions, nonrestrictive relative clauses, and vocatives (cf. Selkirk 2005, N&V 1986/2007, Astruc 2005 among others). Here I introduce the alignment constraints of the end-based (sometimes also edge-based) theory and the wrapping constraint. The influential proposal of the end-based theory by Selkirk (1986, 1995b, 2000) specifies a set of parameters for mapping between syntactic structure and prosodic structure. It predicts anchor points where both structures coincide. The interface constraints require that the edge of a maximal projection in the surface syntactic structure (i.e. XP or Xmax) coincide with (i.e. aligns with) the edge of a prosodic constituent. The edges can either be on the right side or on the left side of both types of constituents. This is shown in (11a) and (11b) respectively. An example of the prosodic structure derived by right alignment is shown in (12). A right prosodic boundary is introduced at the right edge of XP 1 and of XP 2. (11) Selkirk s (1986:389) End parameter settings for phonological phrases a. ] Xmax b. Xmax [ (12) Derived prosodic structure by right alignment Prosodic structure: ( )ϕ ( )ϕ Syntactic structure: [ XP 2 X 1 ] XP1 17 Recently, the relation of phases (CP, vp) to prosodic structure has been highlighted (Ishihara 2004, 2007a; Kratzer & Selkirk 2007). 107

116 Chapter 3: Phrasing Patterns in Catalan SVO Structures Languages are characterized as having either right- or left-alignment (examples for right edges: Kisseberth & Abasheikh 1974 and Selkirk 1986 for Chi Mwi:ni; left edges: Selkirk & Shen 1990 for Chinese, Selkirk & Tateishi 1991 for Japanese). In general, it is assumed that it is the phonological phrase that coincides with XP. Gussenhoven (2004:167) and Truckenbrodt (2005) broaden this idea and establish a relation between a sentence/cp and the intonational phrase. In the 1990s, Selkirk s (1986:389) original approach was generalized to the format of Generalized Alignment (McCarthy & Prince 1993) in Optimality Theory (OT, Prince & Smolensky 1993/2004). 18 This is shown in (13). (13) Alignment Constraints a. ALIGN-XP,R: ALIGN (XP, R; ϕ, R) Align the right edge of a syntactic XP to the right edge of ϕ. b. ALIGN-XP,L: ALIGN (XP, L; ϕ, L) Align the left edge of a syntactic XP to the left edge of ϕ. The constraints in (13) are taken as ranked and violable constraints (Selkirk 1995b). In addition, they are universal in this theory. This becomes important for the new constraint ALIGN-CP,L (Section ). De Lacy (2003) explores the notion of universality by concluding that in all grammars all constraints are present. He states that a grammar cannot chose between, for example, ALIGN (XP,Left; ϕ,left) and ALIGN (XP,Right; ϕ,right), but that both constraints are present in every grammar (de Lacy 2003:60). He gives evidence from the Polynesian language Māori that there is both right and left alignment in one single language. However, it is possible that a constraint in a given language is rendered inactive (de Lacy 2003:70ff.) by higher ranked constraints, so that the language behaves as if there were only either left-alignment or rightalignment. In addition to the alignment constraints, there is a second class of constraints on the syntax-prosody interface: the Wrapping constraints (Truckenbrodt 1995, 1999, 2005). The function of such constraints is to suppress the effect of alignment constraints. Departing from a proposal by Hale & Selkirk (1987), Truckenbrodt (1999) argues that alignment constraints are not enough to describe the relation between syntax and phonology. He shows for the Native- American language Tohono O odham (formerly known as Papago) that lexical XPs dominated 18 The reader should already have knowledge of how Optimality Theory (OT) works. Otherwise, I refer to the original work by Prince & Smolensky (1993/2004) or to introduction books such as Archangeli & Langendoen (1997), Kager (1999), or McCarthy (2002). 108

117 Chapter 3: Phrasing Patterns in Catalan SVO Structures from VP are not separated by phonological phrase boundaries, i.e. no p-boundaries are inserted internal to VP (Truckenbrodt 1999:234): Tohono O odham displays the phrasing shown in (15b) and not the phrasing (15a), which is derived by ALIGN-XP,R. He argues for a constraint WRAP- XP (14) that demands that for each lexical XP 19 there must be a p-phrase that contains the XP (Truckenbrodt 2007:439). This constraint suppresses the right boundary after XP 2 in (15a,b). (14) The Wrapping constraint (Truckenbrodt 1999:228) WRAP-XP Each XP is contained in a phonological phrase WRAP-XP is violated if a syntactic constituent XP is separated into distinct prosodic constituents. WRAP-XP and ALIGN-XP are thus inherently in conflict with each other. This is illustrated in (15). (15) Reaction of WRAP-XP to structures derived by ALIGN-XP,R (cf.truckenbrodt 1999:229) a. * ( ) ϕ ( )ϕ [ XP 2 X 1 ] XP1 b. ( )ϕ [ XP 2 X 1 ] XP1 c. ( )ϕ [X 1 XP 2 ] XP1 d. * ( )ϕ ( )ϕ [ X 1 XP 2 XP 3 ] XP1 Structures (15a,b) are left-branching. The former violates WRAP-XP because there is a prosodic boundary to the right of XP 2 (favoured by ALIGN-XP,R). XP 1 is thus not contained in a single phonological phrase. The latter structure does not violate WRAP-XP, but it violates ALIGN-XP,R because the favoured right boundary is suppressed. The right-branching structure in (15c) does not lead to a conflict between WRAP-XP and ALIGN-XP,R because no constraint is violated: XP 2 and XP 1 are right-aligned and both maximal projections are contained in a single phonological phrase. Right-branching structures only lead to a violation of WRAP-XP when the head has multiple complements, as is the case in (15d). ALIGN-XP,R demands a boundary after XP 2 and for this reason XP 1 is no longer contained in a single phonological phrase. 19 Truckenbrodt (1999:226) formulates the Lexical Category Condition (LCC): Constraints relating syntactic and prosodic categories apply to lexical syntactic elements and their projections, but not to functional elements and their projections, or to empty syntactic elements and their projections. 109

118 Chapter 3: Phrasing Patterns in Catalan SVO Structures Today the most widely accepted theory of the syntax-phonology interface is based on the two concepts expressed by the constraints ALIGN-XP,R/ ALIGN-XP,L and WRAP-XP. However, recent work has shown that phonological factors, such as constituent weight and speech rate, also play a major role in sentence phonology. This is also important in the present study. Constituent weight reflects the heaviness factor on the phonological side. A heavy prosodic constituent has a greater tendency to be phrased independently (N&V 1986/2007, Zec & Inkelas 1990, Ghini 1993 among others). Length can be measured in terms of number of syllables of a constituent along the level of syllable structure (Jun 2003, D Imperio et al. 2005, Elordieta et al. 2005), or in terms of prosodic words of a constituent along the level of prosodic words (Prieto 2005, Prieto 2006). Thus, constituent weight means prosodic branchingness and it concerns the question of whether a prosodic constituent of a certain level contains more than one prosodic constituent of the immediately lower level (Elordieta et al. 2005). As for IntPs, this would mean that it contains two or more intermediate phrases. Ghini (1993) was the first to establish constraints which consider balance and weight of constituents. He reanalyzed N&V s (1986) syntactic branching conditions in terms of average weight and balance of prosodic constituents. N&V (1986/2007:172f.) argued for the phonological phrase that a syntactic head can restructure with (i.e. build a ϕ together with) an adjacent complement if this complement is non-branching (16a,b). If it is branching, the head and its first complement on the recursive side have to remain in separate phonological phrases. This is marked by // in (16c). 20 (16) Restructuring of phonological phrase (cf. N&V 1986/2007:172&173) a. (Se prenderá)ϕ (qualcosa)ϕ (prenderá)ϕ (tordi)ϕ If he catches something, he will catch thrushes b. (Se prenderá qualcosa)ϕ (prenderá tordi)ϕ c. (Porterá)ϕ // (due tigri)ϕ (fuori dalla gabbia)ϕ He will take two tigers out of the cage Ghini (1993:51) argues that there is an isomorphism between syntax and prosodic phrasing. It is not the first non-branching complement but the head of the complement which can phrase together with the preceding V head. N&V (1986) cannot account for the example in (17) because the complement is branching. Ghini (1993), though, is able to account for the phrasing in (17b) because he considers eurhythmic principles such as uniformity (a string is parsed into units of 20 Nespor &Vogel also give examples for restructuring of intonational phrases (N&V 1986/2007:194). 110

119 Chapter 3: Phrasing Patterns in Catalan SVO Structures same length) and average weight (the ideal length of a p-phrase at a normal speech rate comprises two prosodic words). By referring to prosodic words, Ghini (1993:52) moves away from syntactic branchingness and clearly concentrates on prosodic factors. (17) Restructuring of branching objects (cf. Ghini 1993:51) a. (Comprerá)ϕ (mappe)ϕ (di cittá)ϕ (molto antiche)ϕ He will buy maps of very old towns b. (Comprerá mappe)ϕ (di cittá molto antiche)ϕ He will buy maps of very old towns A version of Ghini s (1993) constraints was adopted in Selkirk (2000:244) and Sandalo & Truckenbrodt (2002:295), who introduced size constraints for prosodic constituents for English and Brazilian Portuguese (BP) respectively. In BP, for example, a verb is phrased together with a single-word object as in (18a), but it is phrased separately if the object contains two lexical words (i.e. two prosodic words; N = noun, A = adjective), as in (18b). The constraint ALIGN-XP,R does not help because it cannot derive the internal boundary in the latter example: V is not a maximal projection. (18) Phrasing of V and O in Brazilian Portuguese (Sandalo & Truckenbrodt 2002:293): a. ( V N ) Vendeu livros sold books He sold books b. ( V ) ( N A ) Vendeu livros novos sold books new he sold new books Based on Ghini s (1993) eurhythmic principles, Sandalo & Truckenbrodt (2002:295) formulate the constraint MAX-BIN (19), which is a ranked and violable constraint in OT. MAX-BIN together with WRAP-XP accounts for the (V)(NA) phrasing in (18b). MAX-BIN inserts the boundary after V. Other groupings such as (VN)(A) and (V)(N)(A) are ruled out by WRAP-XP. (19) Prosodic Binarity constraint of Sandalo & Truckenbrodt (2002:295) MAX-BIN P-Phrases consist of maximally two prosodic words. 111

120 Chapter 3: Phrasing Patterns in Catalan SVO Structures The next non-syntactic factor influencing phrasing decisions concerns the rate of speech (N&V 1986/2007, Frascarelli 2000, Jun 2003, among others). The faster a sentence is uttered, the longer the IntPs of this utterance are, i.e. the utterance is less likely to be broken down into several IntPs (N&V 1986/2007:195). Let us consider example (20). According to N&V (1986/2007), (20a) is uttered at a fairly rapid tempo. The sentence contains just one single IntP. However, if the speech rate is reduced it is more likely that (20b) or (20c) is uttered, where the sentence contains two or three IntPs respectively. (20) Influence of rate of speech on phrasing (cf. N&V 1986/2007:194) a. (ι My friend s baby hamster always looks for food in the corners of its cage)ι b. (ι My friend s baby hamster)ι (ι always looks for food in the corners of its cage)ι c. (ι My friend s baby hamster)ι (ι always looks for food)ι (ι in the corners of its cage)ι The examples in (20) show that there is more structure (i.e. there are more IntPs) in the slower versions of the sentence, whereas there is less structure in (20a). Prince & Smolensky (1993:25, fn.13) set up a family of constraints called *STRUC that ensures that structure is constructed minimally. The phenomenon in (20) can be caught by a constraint *P-PHRASE, (21a), which is part of the family of *STRUC and that seeks to avoid phonological phrases altogether (cf. Truckenbrodt 1999:228, 2002:274). Féry (2007) has a corresponding version for a ban on intonational phrases: *I-PHRASE (21b). (21) Constraints punishing P-Phrases and IntPs (taken from Féry 2007) a. *P-PHRASE: No phonological phrase b. *I-PHRASE: No intonational phrase These constraints have in general the effect of punishing additional structure. Furthermore, as Truckenbrodt (1999:228) notes, alignment constraints that are ranked below *P-PHRASE are rendered inactive because a boundary favoured by ALIGN is punished by *P-PHRASE. Thus the universality argued for in de Lacy (2003) does not exclude languages where, for example, ALIGN- XP,L never shows up. In these languages ALIGN-XP,L is simply ranked below *P-PHRASE. At this point, I end the introduction to the theoretical background. For more information, cf. Truckenbrodt (2007) and references cited there. The next section is devoted to the relevant constraints for phrasing in Catalan. 112

121 Chapter 3: Phrasing Patterns in Catalan SVO Structures Accounting for Catalan simple SVO structures Prieto (2005) This section presents the analysis of the phrasing pattern of simple SVO structures in Catalan. I argue that Prieto s (2005) constraint MIN-N-PHRASES (a version of *P-PHRASE) can be violated cumulatively, that it is ranked below MAX-BIN-END (a version of MAX-BIN) and finally, that it is also active at normal speech rate. The results of the simple SVO experiment and of Prieto (2005) are similar. For this reason, the approach established in Prieto (2005) is adopted for my results. Furthermore, Prieto s (2005) approach underlies the approaches of the other three experiments (phrasing of embedded object clauses, phrasing of clitic left- and right dislocation, and phrasing of preverbal subjects). Due to the important role that Prieto s (2005) analysis possesses in the present study, it is presented in detail here: first, the necessary constraints are introduced (Section ) followed by a motivation for the constraint hierarchy (Section ). After that, some OT tables of Prieto (2005) are presented to illustrate her account. Finally, the modifications I propose for Prieto s (2005) account are introduced (Section ) The necessary constraints: Prieto s (2005) OT account is based on the four previously introduced constraints: the two syntactic constraints ALIGN-XP,R and WRAP-XP and modified versions of the two non-syntactic constraints MAX-BIN and *P-PHRASE. 21 The syntactic constraints have the same function as described before. As for *P-PHRASE, Prieto s version of it is called MIN-N-PHRASES and its goal is to minimize the number of phonological phrases in an utterance in rapid speech ((22), cf. Prieto 2005:216). When presenting data on SVO phrasing, the effect of this constraint is illustrated. (22) Prieto s (2005) constraint for avoiding p-phrases MIN-N-PHRASES (RAPID SPEECH) Minimize the number of phrases (rapid speech) 21 Prieto (2005) has further constraints. First, there is MIN-UTT that demands that an utterance consisting of only two prosodic words is parsed into one single phonological phrase (Prieto 2005:213). Due to the fact that all the examples used in this study are longer than two prosodic words, MIN-UTT does not have any influence and is thus not considered here. Second, there is NO-CLASH that prohibits two immediately adjacent stress syllables (Prieto 2005:219, 2008). I do not consider this constraint either because the target words of my data are proparoxytonic words. Consequently there is no clash with following material. 113

122 Chapter 3: Phrasing Patterns in Catalan SVO Structures As for MAX-BIN, it is explained in detail here. Prieto (2005:205) argues that Catalan plays a stronger version of this constraint. The prosodic binarity is restricted to the end of sentences because in Catalan there are longer p-phrases in non-final positions than in final positions. For this reason, the single prosodic grouping of long subject constituents is possible, (23), and, in null subject clauses, the verb and following material of the complement that is not sentence final can form a long phonological phrase (24). In (23) the subject consists of three prosodic words and is not split up. Even if the subject consisted of four or five prosodic words, it would constitute one single prosodic phrase (Prieto 2005:216). The resulting prosodic structure is given in (23b). (23) Phrasing of subjects with more than two prosodic words (Prieto 2005:217) ω ω ω ω a. (Els veїns catalans de l Ebre)ϕ (s enfaden)ϕ The neighbours catalan of the.ebre (river) REFL.get_angry The Catalan neighbours from the Ebre river get angry b. Resulting prosodic structure (Prieto 2005:217): (S)ϕ (V)ϕ Example (24) is a null subject clause. The verb comprava bought is followed by the object mapes de Barcelona maps of Barcelona, which consists of two prosodic words (ω). The object is followed by the adjunct PP per a l Anna for Anna. The difference between (24a) and (24b) is that in the latter example the verb is phrased together with the object. In this case, the prosodic phrase consists of three prosodic words. However, this long phrase is not sentence final and thus is fully acceptable in Catalan. The syntactic structure and its prosodic grouping are given in (24c). (24) Complement with 2 ω and additional adjunct PP (cf. Prieto 2005:204) ω ω ω ω a. (Comprava)ϕ (mapes de Barcelona)ϕ (per a l Anna)ϕ buy.pst.1/3sg map.pl of B. for to the.anna I / (s)he used to buy Barcelona maps for Anna. ω ω ω ω b. (Comprava mapes de Barcelona)ϕ (per a l Anna)ϕ c. Syntactic structure and resulting prosodic structure (Prieto 2005:204): [[V [NP [PP] PP ] NP [PP] PP ] VP ] IP/CP (V NP)ϕ (PP)ϕ In order to account for the pattern that non-final phonological phrases can contain more than two prosodic words, Prieto (2005) has invented the constraint MAX-BIN-END (25). 114

123 Chapter 3: Phrasing Patterns in Catalan SVO Structures (25) Restricted prosodic binarity in Catalan (Prieto 2005:205) MAX-BIN-END P-Phrases containing the main stress of the utterance consist of maximally two prosodic words. Since the main stress of all-new utterances is on the last stressed syllable, MAX-BIN-END concentrates on the sentence final phonological phrase. The constraint sets up only a maximal limit of prosodic words. If the sentence final phrase consists of only one prosodic word (as is the case in (23) and (24)), it is not violated. In summary, the four constraints that are relevant to the phrasing of SVO in Prieto (2005) are ALIGN-XP,R (as defined in (13)), WRAP-XP (as defined in (14)), MIN-N-PHRASES (as defined in (22)), and MAX-BIN-END (as defined in (25)). These four constraints are ranked in the following way (26): (26) hierarchy of constraints (Prieto 2005): MIN-N-PHRASES >> MAX-BIN-END >> ALIGN-XP,R >> WRAP-XP Motivation for the constraint hierarchy: In the next paragraphs the motivation given in Prieto (2005) for this ranking is presented. In Section , however, the order of the constraints in (26) is changed and MAX-BIN-END is argued to be the highest ranked constraint. ALIGN-XP,R is ranked higher than WRAP-XP because of data such as that given in (24). The adjunct PP is phrased separately in (24a) and (24c). According to the syntactic structure given in (24c), the VP (which contains the object and the adjunct PP) is split up. The right boundary after the object NP is introduced by ALIGN-XP,R and it is not suppressed by WRAP-XP. If WRAP-XP were ranked higher, there would be no boundary after the object. The same pattern also arises in double object constructions (27). (27) Prosodic grouping of double object constructions in Catalan (Prieto 2005:204) (Va donar el llibre)ϕ (a la Maria)ϕ PST give.inf the book to the M. (S)he gave the book to Mary 115

124 Chapter 3: Phrasing Patterns in Catalan SVO Structures The dative argument is prosodically separated from the accusative argument. The right boundary after the accusative argument is introduced by ALIGN-XP,R and it is not suppressed by WRAP-XP. These examples are reminiscent of the abstract pattern given in (15d), where the effect of WRAP- XP on structures derived by ALIGN-XP,R is shown. Due to the fact that the boundary after the accusative argument (i.e. after XP 2 in (15d)) results in an acceptable structure, one can conclude that ALIGN-XP,R is more dominant than WRAP-XP, cf. Table 5. If the ranking were the reverse, there would be no boundary between the accusative and the dative argument contrary to the facts (cf. Table 6). [V [NP] [PP]] ALIGN-XP,R WRAP-XP (Va donar el llibre)p (a la Maria)p * (Va donar el llibre a la Maria)p *! Table 5: Catalan double object constructions Right candidate is chosen when ALIGN-XP,R >> WRAP-XP Table 6: [V [NP] [PP]] WRAP-XP ALIGN-XP,R (Va donar el llibre)p (a la Maria)p *! (Va donar el llibre a la Maria)p * Ranking WRAP-XP >> ALIGN-XP,R wrong result for Catalan double object constructions A high ranked alignment constraint, though, prevents the wrapping constraint from having an effect, because boundaries can never be suppressed. Despite this, WRAP-XP is kept in the hierarchy, because there are cases where a low ranked wrapping constraint is able to select the optimal candidate. An example follows immediately, while presenting Prieto s (2005) data for motivating MAX-BIN-END to be ranked higher than ALIGN-XP,R and WRAP-XP. Example (28) shows the phrasing pattern of a null subject clause with an object consisting of three prosodic words. The example is reminiscent of Ghini s (1993) findings (cf. (17)) because the head noun of the object phrases together with the preceding verb. (28) Complement with 3 ω (cf. Prieto 2005:203) ω ω ω ω a. (Comprava mapes)ϕ (de la Barcelona antiga)ϕ buy.pst.1/3sg map.pl of the B. old I / (s)he used to buy maps of old Barcelona. 116

125 Chapter 3: Phrasing Patterns in Catalan SVO Structures Due to the (VN) phrasing, the sentence final prosodic phrase consists of only two prosodic words in accordance with MAX-BIN-END. The constraint ALIGN-XP,R is not able to account for this boundary since there is no edge of a maximal projection at the right of the head noun mapes maps. Prieto (2005:206) says that by considering MAX-BIN-END as higher ranked than WRAP- XP, the correct grouping can be derived. This is shown in Table 7. [V [N [PP [AP]]]] MAX-BIN-END ALIGN-XP,R WRAP-XP (Comprava mapes de la Bcn antiga)p *! (Comprava)p (mapes de la Bcn antiga)p *! * (Comprava mapes de la Bcn)p (antiga)p ***! (Comprava mapes)p (de la Bcn antiga)p ** (Comprava)p (mapes de la Bcn)p (antiga)p ***! Table 7: Dominant role of MAX-BIN-END (cf. Prieto 2005:207) Table 7 shows several things. Firstly, the high ranked constraint MAX-BIN-END is violated by the first two candidates (ωωωω)ϕ and (ω)ϕ (ωωω)ϕ. If WRAP-XP were ranked higher than MAX-BIN- END, the first candidate would wrongly win. Secondly, due to the fact that ALIGN-XP,R is not violated by any candidate, the decision is passed down to the subordinate constraint WRAP-XP. Consequently, the wrapping constraint is important in the evaluation process of the optimal candidate. The third candidate, (ωωω)ϕ (ω)ϕ, as well as the fifth candidate, (ω)ϕ (ωω)ϕ (ω)ϕ, violates WRAP-XP three times: the PP, the NP, and the VP are not contained in a ϕ-phrase. The fourth candidate, (ωω)ϕ (ωω)ϕ, violates WRAP-XP only twice: only the NP and the VP are not wrapped, whereas the PP is wrapped. For this reason, the candidate that groups the object head noun with the verb is the winning candidate. Thirdly, there is no clear evidence for MAX-BIN- END to be ranked higher than ALIGN-XP,R. If ALIGN-XP,R were ranked higher than MAX-BIN- END, the same result would be obtained: Candidate 4 would still win. This interchangeability exists in almost all tables in Prieto (2005) and one could conclude that the two constraints are not ordered. However, there is one table that motivates the dominant role of MAX-BIN-END. Due to the fact that the data stems from SVO constructions, I postpone giving the evidence until the relevant table is considered during the presentation of the SVO analysis. Finally, evidence is needed for MIN-N-PHRASES to be ranked higher than MAX-BIN-END. Unfortunately, there is no evidence given in Prieto (2005). There is only one table in which MIN- N-PHRASES is used and in this table both constraints are interchangeable. The data of this table also stem from SVO constructions and I postpone presenting the table until later. 117

126 Chapter 3: Phrasing Patterns in Catalan SVO Structures Prieto s (2005) OT tables for SVO phrasing: In (2) I presented the results of Prieto (2005) that have to be explained. For the sake of convenience, I repeat them here as (29). (29) Catalan SVO phrasing pattern Results of Prieto (2005) a. (S)ϕ (VO)ϕ single word subjects, small verbal projection b. (SV)ϕ (O)ϕ single word subjects, but long objects (2ω) c. (S)ϕ (VO)ϕ long subject (>2ω), short objects The ranking in (26) accounts for the groupings in (29). This is illustrated by the original tables from Prieto (2005:19f:ex.36,38). The (S)(VO)-phrasing of (29a) is shown in Table 8 for sentence (30). The three lowest constraints of (26) explain the pattern. Candidate c. is the winner since it does not violate any constraint. MAX-BIN-END is respected because the VP consists of two prosodic words, ALIGN-XP,R is respected because there is a boundary after the subject and after the VP, and finally WRAP-XP is respected because no lexical XP is split. MIN-N-PHRASES is not of relevance here, since all candidates have the same number of prosodic phrases. ω ω ω (30) La nena demana els regals the girl want.3sg the presents The little girl wants the presents Table 8: Catalan (S)(VO)-phrasing of the sentence: La nena demana els regals ( The little girl wants the presents ), taken from Prieto (2005:215) Table 9 illustrates the (SV)(O)-phrasing of (29b) for sentence (31). The (SV) phrasing in Catalan shows up in normal and fast speech rate (Prieto 2005:214) in the long branching object condition. Due to the speech rate, the constraint MIN-N-PHRASES is active and, according to Prieto (2005:216), it penalizes output forms with more than two prosodic words (cf. Section for a comment on the minimal restriction to two prosodic words). 118

127 Chapter 3: Phrasing Patterns in Catalan SVO Structures ω ω ω ω (31) La nena demana els regals de Reis the girl wants the presents of (three) kings The little girl wants the Christmas presents. Table 9: Catalan (SV)(O)-phrasing of the sentence La nena demana els regals de Reis ( The little girl wants the Christmas presents ), taken from Prieto (2005:216:38) 22. The optimal candidate d wins because it does not violate MIN-N-PHRASES or MAX-BIN-END and it violates WRAP-XP less often than candidate c. All of the other candidates violate higher ranked constraints. Table 9 shows that MIN-N-PHRASES has to be ranked higher than ALIGN-XP,R in order to allow for the suppression of the boundary after the subject. If it were ranked below ALIGN-XP,R, there would be no chance for the (SV) phrasing. WRAP-XP cannot be used for the suppression, because independent evidence has shown that it is to be ranked lower than ALIGN- XP,R. Furthermore, as in Table 7, the subordinate constraint WRAP-XP is responsible for the decision of the best candidate. Candidate c and d are only different with respect to the amount of WRAP-XP violations. In addition, the data constitute evidence for MAX-BIN-END, because longer p-phrases appear in non-final position. It prevents the verb from being phrased together with the object (cf. candidate a and b). If MIN-N-PHRASES did not exist in Table 9, candidate f would win, which represents an (S)(V)(O)-phrasing. And in fact, candidate f wins, if the speaker speaks slowly (Prieto 2005:214f.). In situations of slow speech rate, MIN-N-PHRASES is not active and the subject and the verb are phrased separately. Candidate f wins over candidate d, because it does not violate ALIGN-XP,R. This is illustrated in Table 10 for sentence (31). 22 In the original table, candidate f had two violations of WRAP-XP: VP and NP. However, the NP does not violate WRAP-XP; this is why I have changed it here. 119

128 Chapter 3: Phrasing Patterns in Catalan SVO Structures Table 10: Catalan (S)(V)(O)-phrasing of the sentence La nena demana els regals de Reis ( The little girl wants the Christmas presents ), taken from Prieto (2005:216:37) The last of the three groupings in (29) can also be derived by Prieto s (2005) basic hierarchy of constraints. This is pictured in Table 11 for sentence (23a), repeated here for the sake of convenience. (23a) ω ω ω ω (Els veїns catalans de l Ebre)ϕ (s enfaden)ϕ The neighbours catalan of the.ebre (river) REFL.get_angry The Catalan neighbours from the Ebre river get angry Table 11: Catalan (S)(VO) phrasing of constructions with long subjects, shown for the sentence Els veїns catalans de l Ebre s enfaden The Catalan neighbours of the Ebre river get angry (Prieto 2005:217) The winning candidate d is the only one that prevents the subject from being split up and satisfies all constraints. It is again WRAP-XP that makes the decision for the optimal candidate. Candidate e, which also satisfies MAX-BIN-END and ALIGN-XP,R violates WRAP-XP twice. To conclude, Prieto s (2005) approach for Catalan SVO-phrasing is able to explain the pattern in (29). Nevertheless, there are some aspects that are worth to be discussed. This happens in the next section. 120

129 Chapter 3: Phrasing Patterns in Catalan SVO Structures Modifications of the previous account on SVO phrasing In this section, some modifications of Prieto s (2005) approach are presented due to a closer inspection of the constraint MIN-N-PHRASES. I propose that the constraint can be violated cumulatively, that it is ranked below MAX-BIN-END and finally, that it is also active at normal speech rate. After that, the impossible (S)(VO) phrasing in Table 9 is discussed. First, there is no evidence for MIN-N-PHRASES to be ranked higher than MAX-BIN-END. The reverse order is also possible, as shown in Table 12. The results of Table 9 are not changed; candidate d still wins. Moreover, Table 9 is the only table in Prieto (2005) where MIN-N- PHRASES is active. Thus nothing speaks against a re-ranking of the constraints like in (32). From now on, therefore, I assume the ranking in (32). La nena demana els regals de Reis. ω ω ω ω. MAX- BIN-END MIN-N- PHRASES ALIGN- XP,R WRAP- XP a. ( )ϕ *! * b. ( )ϕ( )ϕ *! c. ( )ϕ( )ϕ * VP, NP! d. ( )ϕ( )ϕ * VP e. ( )ϕ( )ϕ( )ϕ *! VP, NP f. ( )ϕ( )ϕ( )ϕ *! VP Table 12: Catalan (SV)(O)-phrasing according to re-ranked constraints (cf. (32)) (32) re-ranking of the constraints in Prieto (2005): MAX-BIN-END >> MIN-N-PHRASES >> ALIGN-XP,R >> WRAP-XP Alternatively, one can conclude that MIN-N-PHRASES and MAX-BIN-END are not ordered because no data justify any of the two rankings. However, there are two arguments for ranking MAX-BIN- END higher than MIN-N-PHRASES. a) The first argument stems from data of sentential objects and is presented in the analysis of embedded SVO (cf. Table 17). b) There is some vagueness with respect to the application of the constraint MIN-N-PHRASES. The restriction of MIN-N-PHRASES to penalize only output forms with more than two phonological phrases shall be reconsidered here. It is unclear why only single violations are allowed and not multiple (cumulative) violations. In Table 9 candidates e and f violate the constraint because they consist of three phonological phrases. But the violation is marked by only one single star. Candidates that consist of one or two phrases do not violate the constraint (candidates a-e). It is unclear why the critical value is two (and not just one or even three or four), i.e. it is unclear why MIN-N-PHRASES should not be 121

130 Chapter 3: Phrasing Patterns in Catalan SVO Structures violated by output forms with one or two phonological phrases. If cumulative violations of this type of constraint were allowed in Prieto (2005) (as they are in Truckenbrodt 1999, Elordieta et al or Féry 2007), though, candidate a would wrongly win. It would evoke only a single violation because it consists of one phrase, whereas the other candidates would evoke a double or a triple violation. However, if MAX-BIN-END is ranked higher than MIN-N-PHRASES, the latter constraint could be violated cumulatively. Table 13 is the same as Table 12 with the difference that MIN-N-PHRASES is cumulative. As can be seen, candidate a is not the winning candidate because it violates the higher ranked constraint MAX-BIN-END. Candidate b is sorted out for the same reason. Candidate d wins for the same reasons as before, with the difference that it violates MIN-N-PHRASES twice but so does candidate c. La nena demana els regals de Reis. ω ω ω ω. MAX- BIN-END MIN-N- PHRASES ALIGN- XP,R WRAP- XP a. ( )ϕ *! 1 * b. ( )ϕ( )ϕ *! 2 c. ( )ϕ( )ϕ 2 * VP, NP! d. ( )ϕ( )ϕ 2 * VP e. ( )ϕ( )ϕ( )ϕ 3! VP, NP f. ( )ϕ( )ϕ( )ϕ 3! VP Table 13: Catalan (SV)(O)-phrasing with a re-ranked cumulative constraint MIN-N-PHRASES From now on, it is assumed that MIN-N-PHRASES is ranked below MAX-BIN-END and that it can be violated cumulatively. Before proceeding with the (S)(VO) phrasing of long objects, one final note with respect to MIN-N-PHRASES is necessary. The definition in (22) says that the number of (phonological) phrases shall be minimized in rapid speech (cf. Prieto 2005:216). However, the (SV)(O) phrasing is also possible at a normal speech rate (cf. Prieto 2005:214). Hence, the limitation of (22) to fast speech rate would give an incorrect picture. For this reason, I take MIN- N-PHRASES to be active in normal and fast speech rate. Second, Prieto (2005) cannot account for the (S)(VO) phrasing with long objects. Due to the strict ranking of the constraints, candidate b can never win (cf. Table 9, Table 12 and Table 10). Either (SV)(O) or (S)(V)(O) wins. But my data and the data of D Imperio et al. (2005) show that (S)(VO) phrasing is possible in the Catalan long object condition. In my data (S)(VO) comes to 57% and in D Imperio et al. (2005:68:Table 4) to approximately 38%. Prieto (2005) does not claim that this phrasing is not possible, but it is not considered in the approach. 122

131 Chapter 3: Phrasing Patterns in Catalan SVO Structures In which way can the rankings in (26) or (32) account for the most common grouping? In short, they cannot. The strict ranking of the constraints impedes the corresponding candidate from winning (cf. Table 9 to Table 10) and the variation is left unaccounted for. For the (S)(VO) candidate to win, the order has to be ALIGN-XP,R >> MIN-N-PHRASES >> MAX-BIN-END >> WRAP-XP, as illustrated in Table 14. The importance of a high ranked alignment constraint is also shown in Elordieta et al. (2005:137) for Spanish (S)(VO) phrasing in the long object condition. In Section , the framework of stochastic OT for modeling variation is introduced. At that time, I address the current problem. La nena demana els regals de Reis. ω ω ω ω. ALIGN- MIN-N- PHRASES MAX- BIN-END WRAP- XP XP,R a. ( )ϕ *! 1 * b. ( )ϕ( )ϕ 2 *! c. ( )ϕ( )ϕ *! 2 VP, NP d. ( )ϕ( )ϕ *! 2 VP e. ( )ϕ( )ϕ( )ϕ 3! VP, NP f. ( )ϕ( )ϕ( )ϕ 3! VP Table 14: Catalan (S)(VO) phrasing in the long object condition The results of the simple SVO experiment by and large reflect the findings of Prieto (2005) and D Imperio et al. (2005). (S)(VO) is the most common phrasing pattern in Catalan. For this reason, I simply adopt Prieto s (2005) analysis for my data. The only change proposed is the reranking of her two highest constraints, MIN-N-PHRASES and MAX-BIN-END, as presented in (32). Furthermore, the prosodic constituent of ALIGN-XP,R is taken to be the prosodic phrase and not the phonological phrase as originally done in Prieto (2005) An approach to the phrasing of complex SVO structures The results of the complex SVO (Part 1b) show that there is certain optionality. Often there are several variants possible for grouping complex sentences, but one of them is clearly preferred. The analysis of complex sentences concentrates on the four main groupings that represent 80% of the data. As a first step, the necessary constraints are introduced and it is shown that each grouping has a different constraint ranking (Section ). Next, it is proposed that the four different rankings are generated from a single underlying form. This is modeled in the 123

132 Chapter 3: Phrasing Patterns in Catalan SVO Structures framework of stochastic OT (Boersma 1998, Boersma & Hayes 2001). The general topic of optionality in OT is approached in Section , while the specific model of stochastic OT is introduced in detail in Section In the following section ( ) it is argued that the new constraint ALIGN-CP,L, as well as the two constraints MIN-N-PHRASES and ALIGN-XP,R overlap. Due to this, the actual ranking of the constraints will sometimes be the reverse of their normal ranking and the different rankings can therefore be derived from the underlying ranking MAX- BIN-END >> ALIGN-CP,L >> MIN-N-PHRASES >> ALIGN-XP,R. It is shown that the frequencies of the four main groupings are adequately modeled in stochastic OT. Section concludes Section The necessary constraints and four different constraint rankings: The approach to the prosodic phrasing of SVO structures with a sentential object includes the three highest constraints of (32). 23 Thus the foundation of my approach is based on the proposal by Prieto (2005). However, I depart from Prieto (2005) in two respects. Firstly, I assume a new constraint, called ALIGN-CP,L. This constraint aligns the left edge of a CP with the left edge of a prosodic phrase. Secondly, I claim that the three constraints ALIGN-CP,L, MIN-N- PHRASES, and ALIGN-XP,R. are overlapping constraints (in stochastic OT terms) and consequently show free variation. Based on the proposal by Downing (1970), who argues that intonational phrases obligatorily align with root sentences (including object clauses but excluding adverbial and different preposed, postposed, and interposed constituents), Gussenhoven (2004:167) introduces the OT constraint ALIGN(S,ɩ), which aligns the right edge of every S(entence) with the right edge of IntP. This constraint reflects the influence of the morpho-syntactic structure on prosodic structure. It is shown in the preceding sections that there is often a prosodic break between the matrix clause and the embedded complement clause (80%). Hence, at least in Catalan, object clauses can be excluded from the root sentence (in the sense of Downing 1970), thus arguing that the cases without a prosodic break are caused by a higher ranked constraint. I propose the constraint ALIGN-CP,L (33), which accounts for the prosodic break preceding the embedded clause. 23 In contrast to its role in the approach to SVO phrasing, the constraint WRAP-XP does not have any effect in the phrasing of complex clauses and is thus ignored here. 124

133 Chapter 3: Phrasing Patterns in Catalan SVO Structures (33) ALIGN-CP,L: ALIGN(CP,Left;PrP,Left), or align the left edge of a CP to the left edge of a prosodic phrase (PrP). The possibility of having both alignment of the right edge and alignment of the left edge of a prosodic constituent in one and the same language, argued for in de Lacy (2003), is also applied in Gussenhoven (2004). He introduces for French two constraints that are responsible for the pitch accent distribution either at the beginning or at the end of a phonological phrase (ϕ), (Gussenhoven 2004:255): ALIGN(ϕ, H*,Rt): Align the right edge of every ϕ with a pitch accent. ALIGN(ϕ, H*, Left): Align the left edge of every ϕ with a pitch accent In addition, he introduces for English the left-hand counterpart ALIGN(XP,Left;ϕ,Left) to the usual ALIGN(XP,Right;ϕ,Right), (Gussenhoven 2004:285): ALIGN(XP, ϕ, Left): Align the left edge of every XP with the left edge of ϕ. ALIGN(XP, ϕ, Right): Align the right edge of every XP with the right edge of ϕ. Based on the empirical evidence in my Catalan data and due to the theoretic claims of de Lacy (2003) and the work by Gussenhoven (2004), it is plausible to have a constraint ALIGN- CP,L and I conclude for the present work that it is active in Catalan. I propose the following four different rankings to account for the four different groupings (34). The small letters below the constraints shall help to identify the change in order. In addition, they signal the selection point in the stochastic OT approach to be developed (cf. Section ). 125

134 Chapter 3: Phrasing Patterns in Catalan SVO Structures (34) Four different constraint orders for the main groupings a. Ranking for (SV)(qS)(VO) [56%] MAX-BIN-END >> ALIGN-CP,L >> MIN-N-PHRASES >> ALIGN-XP,R a b c d b. Ranking for (S)(V)(qS)(VO) [24%] MAX-BIN-END >> ALIGN-CP,L >> ALIGN-XP,R >> MIN-N-PHRASES a b d c c. Ranking for (SVqS)(VO) [10%] MAX-BIN-END >> MIN-N-PHRASES >> ALIGN-XP,R >> ALIGN-CP,L a c d b d. Ranking for (S)(VqS)(VO) [10%] MAX-BIN-END >> ALIGN-XP,R >> MIN-N-PHRASES >> ALIGN-CP,L a d c b Ranking (34a), illustrated in Table 15, accounts for the most common phrasing pattern of embedded SVO structures: (SV)(qS)(VO). The ranking is similar to the one for simple SVO structures in (32). a b c d MAX-BIN-END ALIGN-CP, L MINIMIZE-N- PHRASES a. 56% (SV)(qS)(VO) 3 Sm b. 24% (S)(V)(qS)(VO) 4! c. 10% (SVqS)(VO) *! 2 Sm d. 10% (S)(VqS)(VO) *! 3 e. (SVqSVO) *! * 1 Sm,Se f. (SVqSV)(O) *! 2 Sm,Se Table 15: actual ranking for the most common phrasing pattern (SV)(qS)(VO) ALIGN-XP,R Table 15 shows the competition between the four main groupings (candidate a d) and two hypothetical phrasings (e and f). As for the highest ranked constraint, MAX-BIN-END 24, it is violated only by candidate e and thus the evaluation is passed onto the next constraint ALIGN- 24 The careful reader might object that MAX-BIN-END is violated when the object is long. Please cf. Section below for a discussion. 126

135 Chapter 3: Phrasing Patterns in Catalan SVO Structures CP,L. This constraint is violated by candidates c, d, e, and f, because these candidates do not have a boundary in-between matrix V and the complementizer q. But there are still two possible winners, a and b. The next constraint, MIN-N-PHRASES, is able to decide between them. Candidate a wins, because it violates the constraint less severely than candidate b. The lowest constraint, ALIGN-XP,R, is not of importance here, since the decision has already been made ( Sm stands for matrix subject; Se stands for embedded subject). The ranking in (34b) accounts for the second most frequent phrasing pattern (S)(V)(qS)(VO) and is pictured in Table 16. a b d c MAX-BIN-END ALIGN-CP, L ALIGN-XP,R MINIMIZE-N- PHRASES a. 56% (SV)(qS)(VO) Sm! 3 b. 24% (S)(V)(qS)(VO) 4 c. 10% (SVqS)(VO) *! Sm 2 d. 10% (S)(VqS)(VO) *! 3 e. (SVqSVO) *! * Sm,Se 1 f. (SVqSV)(O) *! Sm,Se 2 Table 16: actual ranking for the phrasing pattern (S)(V)(qS)(VO) The difference between Table 15 and Table 16 is the order of MIN-N-PHRASES and ALIGN-XP,R. Since the latter is ranked higher in Table 16, it is the constraint which decides between candidates a and b. Candidate b wins, because it does not violate ALIGN-XP,R whereas candidate a violates this constraint because the matrix subject does not right-align with a prosodic boundary. The ranking in (34c) leads to the winning candidate c, representing the third most common prosodic grouping, Table 17. The difference between Table 15 and Table 17 is made up of the position of ALIGN-CP,L. It is ranked behind MIN-N-PHRASES and ALIGN-XP,R. After the evaluation process has passed onto MIN-N-PHRASES, two candidates remain: c and f. Thus ALIGN-XP,R has the possibility of selecting a candidate and does so. Candidate c wins since candidate f violates ALIGN-XP,R once more. Not only does the matrix subject not right-align with a prosodic phrase, but neither does the embedded subject. 127

136 Chapter 3: Phrasing Patterns in Catalan SVO Structures a c d b MAX-BIN-END MINIMIZE-N- PHRASES a. 56% (SV)(qS)(VO) 3! Sm b. 24% (S)(V)(qS)(VO) 4! c. 10% (SVqS)(VO) 2 Sm * d. 10% (S)(VqS)(VO) 3! * e. (SVqSVO) *! 1 Sm,Se * f. (SVqSV)(O) 2 Sm,Se! * Table 17: actual ranking for the phrasing pattern (SVqS)(VO) ALIGN-XP,R ALIGN-CP, L a d c b MAX-BIN-END ALIGN-XP,R Table 17 additionally shows the importance of MAX-BIN-END being ranked higher than MIN-N- PHRASES. If it were lower than MIN-N-PHRASES, candidate e would wrongly win. However, in complex sentences the single phrasing of the whole clause is never attested in my data. Thus such a phrasing is very unlikely, if not impossible. In order to account for this, MAX-BIN-END has to be ranked higher than MIN-N-PHRASES. Finally, the fourth most frequent grouping pattern is represented by the winning candidate of the ranking in (34d), Table 18. Due to the fact that ALIGN-CP,L is the lowest ranked constraint and MIN-N-PHRASES is ranked lower than ALIGN-XP,R, two candidates, b and d, survive the evaluation until MIN-N-PHRASES. Since candidate b violates the constraint once more than candidate d, the latter wins. MINIMIZE-N- PHRASES a. 56% (SV)(qS)(VO) Sm! 3 b. 24% (S)(V)(qS)(VO) 4! c. 10% (SVqS)(VO) Sm! 2 * d. 10% (S)(VqS)(VO) 3 * e. (SVqSVO) *! Sm,Se 1 * f. (SVqSV)(O) Sm,Se! 2 * Table 18: actual ranking for the phrasing pattern (S)(VqS)(VO) ALIGN-CP, L 128

137 Chapter 3: Phrasing Patterns in Catalan SVO Structures The four tables show that a different order among the three constraints ALIGN-CP,L, MIN-N- PHRASES, and ALIGN-XP,R is responsible for the change of the winning candidate. However, it would be inappropriate to assume that there are four different grammars in Central Catalan for the different groupings. It is rather the case that the situation is one where one phonological input has several outputs. Such a situation is normally described by the terms variation or optionality. I take (34a), here repeated as (35), as the underlying form for the different groupings. (35) Underlying constraint hierarchy for embedded SVO phrasing MAX-BIN-END >> ALIGN-CP,L >> MIN-N-PHRASES >> ALIGN-XP,R a b c d Two aspects speak in favour of (35). Firstly, this ranking has exactly the same order of constraints established for simple SVO structures in (32) besides the enlargement by placing ALIGN-CP,L after MAX-BIN-END. Thus the analysis is an extension of the approach by Prieto (2005) and, in addition, it reflects the close relationship between the two constructions. Secondly, the ranking accounts for the most common phrasing pattern and it is common practice to take that form as underlying which has the greater distribution (e.g. cf. Hall s 2000:64 argumentation for /C/ as the phoneme of the two allophones [C] and [x] in German). Before proceeding with presentating the stochastic OT approach, which accounts for the detected variation, a brief appended exposition of the long O / MAX-BIN-END topic is necessary Reconciling long objects and MAX-BIN-END A Tentative Approach 25 The careful reader might object that MAX-BIN-END is violated in the long O conditions of the complex SVO experiment (Part 1b) in Table 15 (and in the following three tables): If the embedded object is long and phrases with the preceding verb, the sentence final prosodic phrase would consist of more than two prosodic words. In order to capture the sentences of the long O conditions, I propose the following modification of the grammar. One further constraint, namely IDENT-vP (36), is added to the basic hierarchy, as shown in (37). Please note that the modification complicate the proposed theory and cannot be more than a tentative approach at this point in time. 25 I would like to thank H. Truckenbrodt for fundamental proposals leading to this section. 129

138 Chapter 3: Phrasing Patterns in Catalan SVO Structures (36) IDENT-vP Identify the right and the left edge of vp to the right and the left edge of a prosodic phrase (37) Extended Constraint Hierarchy IDENT-vP >> MAX-BIN-END >> ALIGN-CP,L >> MIN-N-PHRASES >> ALIGN-XP,R a b c d The constraint IDENT-vP is a strongly restricted version of WRAP-XP. While WRAP-XP is satisfied when the maximal projection is contained in a prosodic phrase, IDENT-vP calls for a prosodic boundary precisely at both edges of the maximal projection vp, as in (38). (38) Phrasing derived by IDENT-vP ( ) Prosodic Structure [ vp ] Syntactic Structure IDENT-vP directly accounts for the (VO) grouping in the embedded clauses: there is a boundary preceding V, and a boundary following O, and no boundary separates the verb and the object. Most importantly, the constraint is not sensitive to the number of prosodic words in the sentence final prosodic phrase, in contrast to MAX-BIN-END. IDENT-vP, however, demands the minimally and exhaustive phrasing of any vp, i.e. the embedded as well as the matrix vp. The structure shown in (39a) would therefore be a result of this constraint. (39) Recursive phrasing of vp (incorrect) a. ( ( ) ) Prosodic Structure b. [ vp1 [ vp2 ] ] Syntactic Structure Grouping (39a) indicates a recursive prosodic structure. Recursivity, however, violates the Strict Layer Hypothesis (SLH, N&V 1986/2007:7). The structure shown in (39a) can easily be circumvented by assuming it is important to keep the derived prosodic structure. The application of IDENT-vP is cyclic. Each vp (i.e. phase) constitutes a cycle. The embedded vp (i.e. vp 2 ) is merged before the matrix vp (i.e. vp 1 ). Only the first phase (i.e. the embedded vp) shows an effect, due to the fact that previously derived structure cannot be deleted. The matrix vp would demand the deletion of the embedded vp boundaries in order to respect the SLH. But this is not an option because structure cannot be deleted. 130

139 Chapter 3: Phrasing Patterns in Catalan SVO Structures The effect of the newly added constraint is illustrated in Table 19 for the most common grouping, i.e. (SV)(qS)(VO). It is assumed that the object of the embedded clause consists of two prosodic words. For this reason MAX-BIN-END is violated by all four main groupings. a b c d SVqSVO IDENT-vP MAX-BIN-END ALIGN-CP,LMIN-N-PHRASES ALIGN-XP,R (SV)(qS)(VO) * *** * (S)(V)(qS)(VO) * ****! (SVqS)(VO) * *! ** * (S)(VqS)(VO) * *! *** (SVqSVO) *! * * * ** (SVqSV)(O) *! * ** ** (SV)(qS)(V)(O) *! **** * Table 19: actual ranking for (SV)(qS)(VO) when the object consists of two prosodic words. The basic hierarchy is modified by the higher ranked constraint IDENT-vP Table 19 clearly shows the effect of IDENT-vP. The constraint eliminates all the candidates that do not phrase the embedded verb and object in one single group. The last three candidates fatally violate IDENT-vP. The four main groupings do not violate IDENT-vP. Similarly to Table 15 above, the decision for the best candidate is made among the four main groupings. There is only one difference to Table 15: all four groupings violate MAX-BIN-END, while none of them violates MAX-BIN-END in Table 15. Due to the collective violation the evaluation is passed onto the lower ranked constraints. The evaluation process now equals the process in Table 15 and the most common grouping is the winning candidate in Table 19, cf. candidate a. The last candidate in Table 19 is similar to the first candidate. It differs, though, from the winning candidate in having a boundary that separates embedded V and O. This in turn causes a fatal violation of IDENT-vP, even though MAX-BIN-END is not violated. In conclusion, the introduction of a further constraint helps to resolve the problem of the long object pattern. Nevertheless, this constraint might cause problems when analyzing simple SVO structures. In these structures, verb and object are, in fact, sometimes phrased separately (cf. Section and D Imperio et al. 2005). I leave this question open for further research. 131

140 Chapter 3: Phrasing Patterns in Catalan SVO Structures Optionality in Optimality Theory: In this section, several proposals to capture variation are introduced. The stochastic OT model (Boersma & Hayes 2001), which underlies my approach, is presented in Section The question arises as to how to capture such gradient but robust facts in a generative model. This situation is exactly the one that forced Anttila (1997) to reconcile variation with Generative Phonology and to propose a model for deriving variation from principles of Universal Grammar. 26 The need for reconciliation is clearly described in Pierrehumbert (2001). Whereas work on sound structure outside of generative linguistics has used probabilistic models for capturing gradient optionality for many decades, classical generative models are non-probabilistic and for this reason any given sequence is either grammatical or completely impossible (Pierrehumbert 2001:195). The non-generative work has shown that the cognitive representation of sound structure is probabilistic and that frequencies play a crucial role in several areas: in the acquisition of phonological and phonetic competence, in speech production and in speech perception, and also in long-term mental representations (Pierrehumbert 2001:195). At the present time, there are several models for capturing variation in Generative Phonology. In what follows, several further models are briefly introduced. 27 A first attempt to model variation comes from Reynolds & Nagy (1994). They assume a variable ranking with a floating constraint. A floating constraint can intervene between ranked constraints in a given domain. A second model comes from Anttila (1997). He proposes a partial ordered grammar (or stratified grammar Anttila 2002:230), where two tied constraints, A and B, are freely ranked. 28 Then, in some productions A outranks B, whereas in other productions B outranks A. If the tied constraints are ranked high enough in the constraint hierarchy, variation is observed. The different rankings of the tied constraints translate the partial ordered grammar into several totally ranked grammars (Anttila 2007:527). The relative frequency is derived by a quantitative interpretation of these grammars: If a candidate wins by n [totally ranked, I.F.] grammars and t is the total number of grammars, then the candidate s probability of occurrence is n/t (Anttila 2007:524). The underlying cause of variation, as Pierrehumbert (2001:201) points out, is attributed to the mind of the individual speaker and is an intrinsic part of linguistic competence. A further approach has been developed by Truckenbrodt (2002) for accounting for variation in p- 26 Anttila (1997) is concerned with variation in Finnish genitive plurals (i.e. the variable inflectional behavior of polysyllabic Finnish nouns) and not with the prosodic phrasing of Catalan clauses, though. 27 Cf. Anttila (2002, 2007) and Gabriel (2007:247) for a detailed overview. 28 The term free ranking was introduced by Itô & Mester (1997). 132

141 Chapter 3: Phrasing Patterns in Catalan SVO Structures phrasing in Bengali. He argues that variation in complex forms of p-phrasing may be inherited from simpler forms via a process called output-to-output faithfulness (OO-faithfulness). An additional approach is presented by the stochastic OT model proposed in Boersma (1998) and Boersma & Hayes (2001), and this is the one I assume in the present study. They refine and extend Anttila s (1997) approach by providing each constraint with a probability distribution on a continuous ranking scale. In what follows, the main features of this approach are sketched Stochastic Optimality Theory: Boersma s (1998) Functional Phonology and Boersma & Hayes (2001) Gradual Learning Algorithm (GLA) can easily capture frequency-dependent variation. Their approaches are variants of OT and are subsumed under the term stochastic OT. Due to the fact that experiment 1b provided data quantified for the number of realizations of each pattern, the stochastic OT approach is the appropriate framework. Two notions are important for this model: the continuous ranking scale and the stochastic candidate evaluation. In classic OT, a ranking C 1 >> C 2 >> C 3 is interpreted non-probabilistic, meaning that a constraint C 1 dominates a constraint C 2 that dominates a constraint C 3. It is not intended that constraint C 2 and C 3 have a shorter distance between each other than C 1 and C 2. This idea, though, plays a central role in stochastic OT. Instead of assuming a strict ranking scale, a continuous ranking scale is assumed. Constraints have a certain ranking value and higher values correspond to higher-ranked constraints and lower values to lower-ranked constraints (40). The example numbers added for the ranking values reflect the relation between the constraints. Here it is merely important that a shorter distance between two constraints implies that the relative ranking of the constraints is less fixed. The less fixed order is the crucial aspect in accounting for variation. Boersma (1998) and Boersma & Hayes (2001:47) suggest that constraints are not single points, but that they act as if they are associated with ranges of values (42). This happens due to a temporarily perturbation of the position of each constraint by a random positive or negative value at evaluation time (i.e. the time when the candidates in an OT table have to be evaluated in order to determine a winner). This is illustrated in (41). The concrete value that is used for a single constraint is called the selection point. This point is given by the black line in (42); the line is marked with b at the end, meaning that this is the selection point for constraint b. The selection point can differ from the ranking value (dotted line). The latter is the center of the 133

142 Chapter 3: Phrasing Patterns in Catalan SVO Structures range, i.e. the value more permanently associated with the constraint (e.g. example value 100 in (41) and (42)). (40) Constraints with a given ranking value on a continuous ranking scale a MBE b AL-CP,L c M-N-P d AL-XP,R (high ranked) (low ranked) (41) Perturbation of constraint by a random positive or negative value b AL-CP,L (42) Range of value and selection point b AL-CP,L 100 (Ranking value) b selection point A selection point near the center (i.e. the ranking value) is more probable than a selection point far away from the center. This is so because constraint ranges are interpreted as probability distributions. By that they can account for noisy events that are described with a normal (=Gaussian) distribution (cf. Boersma & Hayes 2001:48). A normal distribution has a single peak in the center and declines towards zero on each side. For this reason values become less probable the farther away they are from the center. The grey box around the center in (42) describes the standard deviation, in which most of the values drawn from a normal distribution are located. In stochastic OT, every constraint has the same standard deviation. The less fixed order of given constraints becomes important when their distance is relatively short. If the distance is short enough, two (or more) constraints overlap, i.e. their ranges covered by the selection points overlap. This is pictured in (43) for MIN-N-PHRASES (M-N-P) and ALIGN-XP,R (AL-XP,R). Due to the fact that at evaluation time it is possible to choose the 134

143 Chapter 3: Phrasing Patterns in Catalan SVO Structures selection points from anywhere within the two given constraints, the ranking of the constraints most often results in the normal ranking order (43a), but sometimes it will be the reverse of the normal order (43b). In the former case, the selection point is taken from the upper part of MIN- N-PHRASES and from the lower part of ALIGN-XP,R. In the latter case, the selection point is taken from the lowest part of MIN-N-PHRASES and from the upper part of ALIGN-XP,R. (43) Overlapping constraints a. Normal ranking a MBE b AL-CP,L c M-N-P d AL-XP,R a b c d b. Reverse of normal ranking a MBE b AL-CP,L c M-N-P d AL-XP,R a b d c Free variation arises due to overlapping constraints because they can generate multiple output forms from a single underlying form. The more the constraints overlap, the more probable a reverse ranking is. This means that in a certain percentage of the evaluations (depending on the amount of overlap) ALIGN-XP,R will outrank MIN-N-PHRASES in (43), although the latter constraint has a higher ranking value. When this happens, the second best candidate wins. Constraint ranges are hence interpreted as probability distributions. (44) Strict ranking of constraints a MBE b AL-CP,L c M-N-P d AL-XP,R a b c d 135

144 Chapter 3: Phrasing Patterns in Catalan SVO Structures The strictness of classical OT rankings is a special case included in continuous ranking scales. It is illustrated in (44). As can be seen, it appears when the constraints do not overlap (i.e. when the distance between the constraints is very large so that the probability of deviant ranking becomes very low; cf. Boersma & Hayes 2001:50). In order to know what ranking value the necessary constraints have, Boersma & Hayes (2001) developed the Gradual Learning Algorithm (GLA), an algorithm for learning optimalitytheoretic constraint ranking. The GLA requires two kinds of inputs: OT constraints and the frequencies of distribution. The second input enables the model to include the results stemming from empirical data. Now the process of learning an appropriate constraint ranking consists merely of finding a workable set of ranking values on a continuous scale. The GLA calculates the location of the constraints relative to each other. 29, Applying stochastic Optimality Theory to the phrasing of complex sentences: The data examined in the complex SVO experiment can be accounted for by a stochastic model. This is shown here. As a first step, the constraints that overlap are illustrated. Second, the application of the Gradual Learning Algorithm gives the ranking values for deriving the frequency effects. The underlying constraint hierarchy has been given in (35). The four different orders of these constraints have been given in (34). As can be seen, whereas MAX-BIN-END is always the highest ranked constraint, the remaining three constraints vary their positions. I claim now that the relative order of the lower three constraints never changes (thus giving (35)), but that the distance between them is so short that their ranges overlap. The corresponding Catalan grammar is pictured in (45). (45) Constraint Hierarchy for Catalan clauses with sentential objects a MBE b AL-CP,L c M-N-P d AL-XP,R 29 For a complete description of the GLA process of learning cf. Boersma & Hayes (2001:51ff.). 30 In Boersma & Hayes (2001), the empirical application of the GLA is illustrated with examples of free variation of glottal stop and glides in Ilokano (an Austronesian language of the northern Philippines), of output frequency in the Finnish genitive plural, and of gradient well-formedness judgments of English light and dark /l/. 136

145 Chapter 3: Phrasing Patterns in Catalan SVO Structures The differences of the height and the length of the constraint ranges in (45) have no meaning, but are used only for the sake of clarity. The constraints have still the same standard deviation. The area of overlap of the three constraints induces that any order between the ALIGN-CP,L, MIN-N- PHRASES, and ALIGN-XP,R can be generated. This is exactly what is needed for deriving the four different rankings of (34). 31 As an example, the order of the selection for the fourth most grouping (S)(VqS)(VO) is illustrated in (46). For the sake of convenience, the corresponding constraint ranking (34d) is repeated below. (46) Order of selection points for the phrasing (S)(VqS)(VO) a MBE b AL-CP,L c M-N-P d AL-XP,R a d c b (34d) MAX-BIN-END >> ALIGN-XP,R >> MIN-N-PHRASES >> ALIGN-CP,L a d c b Now, the application of the GLA endows the hierarchy with concrete ranking values for the constraints. Possible values are given in (47), and the frequency prediction (compared to the empirical data) is given in (48). Due to the fact that the ranking values and the frequency predictions differ with each run, they cannot be more than possible values. (47) Ranking values proposed by GLA a. MAX-BIN-END b. ALIGN-CP,L c. MIN-N-PHRASES d. ALIGN-XP,R There are six possible permutations, although only four are needed to account for the variation. As the reader might easily control, the two remaining permutations (a >> c >> b >> d and a >> d >> b >> c) generate existing groupings and they therefore do not pose a problem for my approach. 137

146 Chapter 3: Phrasing Patterns in Catalan SVO Structures (48) GLA frequency prediction frequency prediction empirical results (SV)(qS)(VO) 54.33% ( ) 56% a >> b >> c >> d (S)(V)(qS)(VO) 25.13% ( ) 24% a >> b >> d >> c (SVqS)(VO) 11.31% ( ) 10% a >> c >> d >> b (S)(VqS)(VO) 09.14% (91.395) 10% a >> d >> c >> b The application of the learning algorithm shows that the proposed approach is realizable and that the clear tendencies of the empirical frequency values can be captured. However, the shorter distance between the three lowest constraints is rather more important than the exact ranking values proposed by the GLA. Finally, I return to the (S)(VO) phrasing in the long branching object condition of simple SVO structures. Prieto s (2005) approach as well as the suggested reranking in (32) cannot account for this grouping. As the data of the simple SVO experiment and the results of D Imperio et al. (2005) show, there is variation with respect to the groupings. The data practically call for a stochastic account of simple SVO structures. Table 14 shows that the ranking of the constraints should be ALIGN-XP,R >> MIN-N-PHRASES >> MAX-BIN-END >> WRAP-XP. In the approach to complex SVO structures, it was suggested that all constraints except for MAX-BIN-END overlap. In order to account for (S)(VO), the constraint MAX-BIN-END must have a closer distance to the other constraints so that it overlaps with them. Then its selection point can appear in certain cases below the selection points of ALIGN-XP,R and MIN-N-PHRASES and the (S)(VO) candidate wins. 32 A closer distance of MAX-BIN-END is also proposed for the analysis of left-dislocations in Catalan (Chapter 5) Conclusion With respect to the analysis of the prosodic phrasing pattern of simple SVO structures in Catalan, I adopt by and large the proposal of Prieto (2005) even though I hardly found (SV)(O) realizations. I deviate from Prieto (2005) by re-ranking her two highest constraints MIN-N- PHRASES and MAX-BIN-END in the reverse order (cf. (32)). Her two further constraints, ALIGN- XP,R and WRAP-XP, have been left unchanged. The re-ranking enables the welcoming possibility of maintaining the same constraint order for the most common phrasing pattern of complex SVO structures with a sentential object: (SV)(qS)(VO). However, a further constraint has to be added 32 However, I will not deepen this suggestion here, but leave it for further research. 138

147 Chapter 3: Phrasing Patterns in Catalan SVO Structures after MAX-BIN-END: ALIGN-CP,L. This constraint accounts for the pattern that the embedded clause is in general prosodically separated from the matrix clause (in 80% of the data in the complex SVO experiment). The results of the complex SVO experiment have further shown that there is variation in terms of the prosodic groupings of structures with embedded complement clauses. Although the variation is broadly diversified (12 different groupings), the four most common groupings already represent 80% of the data. The variation is modeled in the stochastic OT framework (Boersma & Hayes 2001) and ALIGN-CP,L, MIN-N-PHRASES, and ALIGN-XP,R are taken to overlap on the continuous ranking scale. The general constraint hierarchy I propose is as follows: MAX-BIN-END >> ALIGN-CP,L >> MIN-N-PHRASES >> ALIGN-XP,R Exactly as in the account of Prieto (2005), the (SV) phrasing of the matrix clause is not derived from a specific constraint reflecting the length of the object. It is derived from the interaction of several constraints. In Prieto (2005), it is the interaction of MAX-BIN-END and MIN-N-PHRASES, which outrank ALIGN-XP,R. In my approach, it depends on the actual position of ALIGN-XP,R and on the higher ranked constraints. 139

148 Chapter 4 Some Syntactic Aspects of Catalan CLLD and CLRD In the present chapter, a clause-internal analysis of Catalan clitic right-dislocations (CLRD) (i.e. a position below TP and above vp)) is argued. These arguments support Villalba s (2000) syntactic approach, which underlies my optimality theoretic approach to prosodic phrasing of clitic left-dislocations (CLLD) and CLRD (Chapter 5). The approach by Villalba (2000) is shortly introduced (as well as two further approaches to the syntax of dislocations): CLLD is taken to host TopP in the C-domain, while CLRD is taken to host an internal TopP projection below TP. Subjects are taken to host the specifier of TP. While this chapter concentrates only on syntactic aspects of CLLD and CLRD, their prosodic characteristics are presented in Chapter 5. The introduced difference between preverbal subjects and CLLD (cf. Chapter 1) is revisited from an intonational side in Chapter Syntactic Aspects of Catalan CLLD and CLRD In 4.1.1, the most common syntactic approaches to CLLD and CLRD are briefly introduced. In 4.1.2, I argue for a clause-internal analysis of Catalan CLRD based on CLLD/CLRD asymmetries. These asymmetries were highlighted in Villalba (1996, 1999a,b, 2000) and Cecchetto (1999), but refuted by Samek-Lodovici (2006) amongst others. I show that Catalan data still speak in favor of Villalba s and Cecchetto s assumption. Section concludes this chapter Three syntactic approaches to CLLD and CLRD For clitic left-dislocation (CLLD), it is common to assume that the dislocated constituent is placed in the C-domain, i.e. the (complex of) functional projection(s) above TP 1. The 1 I use TP (tense phrase) as a cover term for inflectional phrase (IP), i.e. TP is a syntactical, functional projection accommodating tense and agreement features of the sentence (cf. Chomsky 1995:377). One important reason for doing so is the fact that the abbreviation IP is used in both syntactic work (as just mentioned) and phono-prosodical work, where it stands for Intonational Phrase. In order to avoid confusion, I use TP for the former and IntP for the

149 Chapter 4: Syntactic Aspects of CLLD and CLRD functional category C is part of the core functional complex (CFC), made up of C-T-v-V (cf. Chomsky 1986a:169; Giorgi 1987). Chomsky (2005:9) says that C is shorthand for the region that Rizzi (1997) calls the left periphery. Rizzi (1997) splits the CP up into four functional projections: ForceP-Foc(us)P-Top(ic)P-Fin(ite)P. The C-domain is labeled in Rizzi (1997) as the complementizer layer, where clause-type properties (i.e. Force) and information structural aspects (topic-focus) are expressed. 2 In earlier days, CLRD was assumed to mirror CLLD: whereas a CLLD XP is leftadjoined to a certain node, a CLRD XP is simply right-adjoined to the same node ((1); cf. Vallduví 1993:104). The mirror hypothesis (Cecchetto s 1999 term) assumes that the characteristics and properties of CLLD structures can be simply transferred to CLRD structures. (1) Mirror hypothesis by Vallduví (1993:104; structure slightly modified by I.F.) 3 : left-detachments right-detachments a. TP b. TP CLLD TP TP CLRD cl t cl t As (1) shows, Vallduví (1993) assumes an adjunction-to-tp(/ip) analysis, which was quite common to assume for topicalization around the 1980 s (cf. Baltin 1982:18; Lasnik & Saito 1992:77f.; and Rochemont 1978, 1989). Nevertheless, there had already been approaches assuming a position in the CP (Authier 1992:330; Watanabe 1993:529). The two latter accounts propose a CP-iteration for topic structures, which can be seen as predecessors for Rizzi s (1997) Split-CP analysis. 4 Villalba (1996, 1999b, 2000), Cecchetto (1999), and López (2003, 2008), however, show that there are asymmetries between CLLD and CLRD which cannot be explained by the mirror hypothesis. As a consequence, they assume the CLRD constituent to be below T : Spec of a VP peripheral Topic Phrase in Cecchetto (1999:59), Spec of the Internal Topic Phrase in Villalba latter. When quoting other authors who use the functional projection IP, I also use TP and explain in a footnote their original notation. 2 In addition, Rizzi (1997) assumes two further layers: the inflectional layer (i.e. IP), which is mainly concerned with the licensing of morphological features; and the lexical layer (i.e. vp/vp), which is mainly concerned with thetaassignment. 3 Vallduví (1993) originally uses the notion IP. 4 Cf. Feldhausen (2002) for an overview of pre-rizzi (1997) approaches to topicalization (in English and German). 141

150 Chapter 4: Syntactic Aspects of CLLD and CLRD (2000), and Spec of vp in López (2003, 2008). 5 In addition, these authors assume that dislocations are derived by movement (in line with Cinque 1977, Rizzi 1997, Grohmann 2003, and Belletti 2005). The constituent moves from its base position (e.g. from the argument position of the verb if the dislocation is an object or the subject) to the topic position below TP. All three authors take this topic position as an intermediate position for CLLD (cf. also Postal 1991:15). Consequently, if the topic constituent constitutes a right-dislocation, it remains in the internal topic position. If the topic constituent constitutes a left-dislocation, it moves further up from the internal topic position to a position in the C-domain. I demonstrate the medial topic position with Villalba (1996, 1999a, 2000), as in (2). The internal Topic Phrase (IntTopP) is located above vp and below TP 6, NegP, and an internal Focus Phrase. The latter functional projection hosts the constituent which bears main prominence (for suggesting an internal FocP cf. Belletti & Shlonsky 1995, Ambar 1999). CLLD, in contrast, occupies the specifier position of an external Topic Phrase (ExtTopP), which is located in the C- domain. (2) C-domain of Villalba (2000:218) TP-domain of Villalba (2000:221&233) (slightly modified by I.F.) CP TP ExtTopP NegP CLLD ExtTop Neg IntFocP ExtTop TP Focus IntTopP CLRD vp The correct word order of the CLRD element is derived in both Villalba (1996, 1999a, 2000:232f.) and Cecchetto (1999:57f.) by movement of the material following the CLRD into the internal Focus Phrase. 7 However, not all authors assuming movement also assume an internal position for CLRD. Kayne (1994), Zubizarreta (1998:121), and more recently Samek-Lodovici (2004, 2006), for example, assume a clause-external position for right-dislocations. They are derived from leftdislocations via remnant movement of the following TP (i.e. IP): in the Italian sentence L ho 5 Cf. Cardinaletti (2002) and Samek-Lodovici (2005, 2006) for a critique of such a medial position for CLRD; and cf. Section below. 6 Villalba (1996, 1999a, 2000) originally uses the notion IP. 7 Cf. López (2006, 2008:83ff.) for a PF-analysis of how the CLRD constituent moves to the right. 142

151 Chapter 4: Syntactic Aspects of CLLD and CLRD visto, Gianni I saw John in (3), the dislocated element (Gianni) has been moved leftward to [Spec,Top]. After that in order to derive a right-dislocation (from the momentary leftdislocation) the IP (i.e. the remnant) moves into the specifier of a higher maximal projection (XP). (3) Clause-external CLRD and remnant movement (Samek-Lodovici 2006:840) In addition, not all authors assuming remnant movement assume movement of the dislocated element. Frascarelli (2000, 2004), for example, assumes that right-dislocations are base-generated in a Topic projection of the C-domain (cf. Frascarelli 2000:139,159ff; hence right-dislocations in her analysis are right-hand topics). Her base-generation account for dislocations is in line with Cinque 1990, Anagnostopoulou 1997, Suñer 2006, Alexiadou 2006, and De Cat 2002, As can be seen in the different syntactic approaches to CLLD and CLRD, these constituents host almost always a functional projection TopP. This projection is inevitably connected to the information structural notion topic (as presented in Chapter 1). In what follows, I provide evidence for an asymmetry between CLLD and CLRD. As previously shown, there is an ongoing discussion of the right analysis for CLRD. The presented evidence supports the clause-internal analysis for Catalan CLRD. This result is important for the analysis of the prosodic patterns of dislocation structures in Chapter 5. 8 The question as to whether topics are base generated in their surface position or if they are dislocations, i.e. moved from an TP-internal position, has become an important topic (cf. De Cat 2002:99ff; Frascarelli 2000:137,159ff; López 2008:175ff.; Villalba 2000:233ff.) and researchers have to take a stand. Interestingly, early accounts, such as Lasnik & Saito (1992) assume both base generation and movement of topics. Lasnik & Saito (1992:77f.) analyze (English) embedded and matrix topicalization as movement into an adjoined IP, whereas English left-dislocation is only possible in matrix clauses and base-generated in a Topic Phrase. As for the comparison with Romance CLLD, Rochemont (1989:154ff.) notes that CLLD behaves exactly like topicalization in English in its syntactic effects. However, English topicalization does not allow for resumptive pronouns. Cecchetto (1999:57:fn.22), for example, assumes movement for dislocated arguments, whereas base-generation for dislocated PPs. 143

152 Chapter 4: Syntactic Aspects of CLLD and CLRD CLLD and CLRD Asymmetries As for asymmetry, several tests can be found in Cecchetto (1999) and Villalba (2000). 9 Some of these tests can also be found in Samek-Lodovici (2006) 10, but he argues against asymmetry. In what follows, I discuss only three tests. Two of them are found in SL (2006): licensing of N-words (such as negative polarity items, NPI) and binding properties (Cecchetto s 1999 antireconstruction effects ). These two tests are discussed, because they seem to be controversial in the literature on Italian and Catalan: According to SL, the tests show that dislocated elements in Italian are not c-commanded by T or any other material within TP 11 (such as a licenser for NPIs) and thus support his view that right-dislocated constituents have to be external to the clause main TP. I show that the Catalan counterparts behave contrariwise: Catalan CLRD is c-commanded by T and this can only be modeled by a clause-internal analysis. The third test is a completely new test. This test is based on obviation effects. Left-dislocations lead to a disappearance of obviation effects in subjunctive complement clauses (Costantini 2005b). Right-dislocations, as is shown, keep obviation effects. The pattern is explained by assuming a clause-internal analysis of CLRD Licensing of n-words (SL 2006:844f) Negative words (n-words) such as ningú nobody/anybody or negative polarity items (NPI) such as mai ever and res anything are only licensed when they are in the scope of negation. For this reason they have to be overtly licensed by an element which c-commands the 9 Cecchetto (1999) provides four tests showing asymmetries between CLLD and CLRD and thus argues for a clauseinternal analysis and against the mirror hypothesis (and partly against Kayne s double topicalization analysis presented in Cecchetto 1999). The tests rely on antireconstruction effects (Cecchetto 1999:42), exploiting an argument-adjunct asymmetry of constituents following a noun; on ECP effects (Cecchetto 1999:44), exploiting a subject-object asymmetry arising when a CLLD constituent appears between a wh-word and the following clause; on the Right Roof Constraint (Cecchetto 1999:46), exploiting the idea of clause-boundedness of CLRD constituents; and finally on Aux-to-COMP constructions (Cecchetto 1999:47), exploiting the idea of intervening dislocations after gerundival adverbials in COMP. Villalba (2000) provides several tests, also arguing against the mirror hypothesis and Kayne s (1994) analysis. Among the tests are the criteria of boundedness (Villalba 2000:186; Cecchetto s Right Roof Constraint test bases on Villalba s idea); island effects (Villalba 2000:188), exploiting the idea that CLRD does not show island effects; licensing of NPIs (Villalba 2000:189; equals the test of SL 2006); Principle C effects/antireconstruction effect (Villalba 2000:190), exploiting the idea that a principle C violation can be compensated by left-dislocating the constituent with the R-expression; Quantifier binding of pronouns (Villalba 2000:191), exploiting the idea null pronouns receive a bound variable interpretation; dislocation out of dislocations (Villalba 2000:192), exploiting the idea that a constituent from within a right-dislocate can be left-dislocated whereas the reverse is not true (cf. Chapter 5 for examples of this construction). 10 Abbreviated as SL from now on. 11 SL (2006) originally uses the notion IP. 144

153 Chapter 4: Syntactic Aspects of CLLD and CLRD negative element. The licenser (mostly the sentential neg-marker no no ) can be within or higher than T. The examples presented in this section rely strongly on the examples used by Samek- Lodovici (SL). I present his original Italian examples immediately followed by their corresponding Catalan translations. Due to the fact that the Italian and the Catalan examples are so similar, only one translation is given, which is valid for both languages. Concerning the basic pattern of n-words, there are no differences between Italian and Catalan. As (4) shows, the negative marker (the licenser) is obligatory and cannot be left out. In addition, n-words cannot be right-dislocated alone, as in (5), but they may easily appear in longer dislocations, where the negmarker is also present, as in (6) and (7). (4) Neg-marker cannot be left out a. Gianni *(non) ha mai visto nessuno. ITAL John not has ever seen anybody.(sl 2006:844:15) John has never seen anybody b. El Joan *(no) ha vist mai ningú. CAT. the John not has see.ptcp2 ever anybody (5) Right-dislocated n-word a. *Non l ha invitato GIANNI, nessuno. ITAL. not him/it has invited John anybody (SL 2006:844:16) John didn t invite anybody b. *No l ha convidat el Joan, a ningú. CAT. not CL.have invite.ptcp2 the John to anybody (6) Right-dislocated neg-marker and n- word a. Lo sappiamo GIÀ, che non avete incontrato nessuno. ITAL. (we) it know already, that (you) not have met anybody (SL 2006:844:18a) We already knew that you haven t met anybody b. Ja ho sabíem, que no havies trobat ningú. CAT. already cl know.1pl.pst that not have.2sg.pst find.ptcp anybody (7) Right-dislocated neg-marker and n- word a. Lo è sembrato CARLO, non lavorare mai. ITAL. it is seemed Carl, not to-work ever (SL 2006:844:18b) It was Carl who seemed to never work 145

154 Chapter 4: Syntactic Aspects of CLLD and CLRD b.??ho sembla en Carlo, de no treballar mai. CAT. It seem.3sg the C. of not work.inf ever Carl seems to never work The Catalan example in (7) is odd. This judgement, however, has nothing to do with the NPI but stems from dislocating the infinitive. As shown in (4), if there is no neg-marker licensing the n-word (or NPI), the sentence is ungrammatical. As SL (2006:844) states, this is also the case if the n-word is right-dislocated (5), where the n-word is outside the c-commanding domain of its licenser (as proposed by the clauseexternal analysis). 12 Based on the argumentation for (4) to (6), examples (8) and (9) are ungrammatical due to the lack of a neg-marker. In (8) and (9) the n-word is part of a sentential right-dislocation, but the only available licenser has been omitted. Thus, the ungrammaticality can only be caused by the failure of licensing (SL 2006:845). (8) Right-dislocated n-word without licenser a. *Lo sappiamo GIÀ, che avete incontrato nessuno. ITAL. (we) it know already, that (you) have met anybody (SL 2006:845 (SL 2004)) We already know that you haven t met anybody b. *Ja ho sabíem, que havies trobat ningú. CAT. already it know.1pl.pst that have.2sg.pst find.ptcp anybody We already know that you haven t met anybody (9) Right-dislocated n-word without licenser a. *Lo è sembrato CARLO, lavorare mai. ITAL. it is seemed Carl, to-work ever (SL 2006:845:19b) It was Carl who seemed to never work b. *Ho sembla en Carlo, de treballar mai. CAT. cl seem.3sg the C. of work.inf ever It was Carl who seemed to never work Furthermore, if the neg-marker is located in the matrix clause, and no dislocation occurs, the neg-marker can easily license into the embedded clause as shown in (10) and (11). 12 However, it is objected that the ungrammaticality of a right-dislocated n-word can be simply due to the fact that n- words cannot act as topics. If this is the case, (5) says nothing about the relation between licenser and n-word. 146

155 Chapter 4: Syntactic Aspects of CLLD and CLRD (10) Neg-marker licenses into an embedded clause a. Non ho voglia di vedere NESSUNO. ITAL. (I) not have wish to see anybody (SL 2006:846:23) I have no wish to see ANYBODY b. No volia veure NINGÚ. CAT. not want.1sg.pst see.inf anybody I have no wish to see ANYBODY (11) Neg-marker licenses into an embedded clause a. Non desidero mangiare NULLA. ITAL. (I) not wish to-eat anything (SL 2006:846:24) I do not wish to eat ANYTHING b. No vull menjar RES. CAT. not wish.1sg eat.inf anything I do not wish to eat ANYTHING Up until now, there are no differences in the pattern of the licensing of n-words between Italian and Catalan. Examples (4) to (11) demonstrate the basic properties of n-words and the conclusion is that they are the same for both languages. SL (2006:845) now introduces the crucial test. It has been seen that the n-word must be c- commanded by its licenser, be it (the licenser) in the same clause or in a matrix clause. Examples (12) and (13) present the Italian examples from SL (2006:845). The (a) sentences in both examples involve an n-word licensed by a neg-marker within the dislocated clause. These sentences are grammatical (cf. also (6) and (7)). The (b) sentences are identical to the (a) sentence but differ in one respect: the neg-marker is located in the non-dislocated part of the sentences. As for the crucial test, if the RD constituent is low in the hierarchy (i.e. clause-internal analysis) the neg-marker should be able to c-command the n-word, and the (b) sentences should be grammatical. If, on the other hand, the RD constituent is high (i.e. clause-external analysis) the licensing should fail, because the RD constituent is higher than Neg/T. (12) Italian data for the test: Different position of the neg-marker a. Ne ho davvero VOGLIA, di non vedere nessuno per qualche giorno. (I) of-it have definitely wish of not to-see anybody for.a.few days I definitely DO wish not to see anybody for a few days (SL 2006:845:20a) b. *Non ne ho VOGLIA, di vedere nessuno per qualche giorno. (I) not of-it have wish of to-see anybody for a-few days I definitely do NOT wish to see anybody for a few days (SL 2006:845:20b) 147

156 Chapter 4: Syntactic Aspects of CLLD and CLRD (13) Italian data for the test: Different position of the neg-marker a. (Non lo desidero,) lo PRETENDO, di non mangiare nulla per qualche giorno. (I) not it wish, (I) it demand, of not to-eat anything for a-few days I do not wish, I DEMAND not to eat anything for a few days (SL 2006:845:21a) b. *Non lo DESIDERO, di mangiare nulla per qualche giorno (lo PRETENDO). (I) not it wish, of to-eat anything for a-few days ((I) it demand) I do not WISH to eat anything for a few days, (I DEMAND it) (SL 2006:846:21b) As can be seen, the (b) examples of (12) and (13) are ungrammatical: The licensing of the n- words fails. Thus, the RD constituent should be higher than Neg/T, supporting the clauseexternal analysis. In (14) and (15), the corresponding Catalan examples are given. The Catalan sentences show that the pattern which holds in Italian does not hold in Catalan thus showing a clear contrast between these two languages. In (14) and (15) the b.-examples might be somehow marked, but they are certainly not ungrammatical. (14) Catalan data for the test: Different position of the neg-marker a. Certament ho VULL, de no veure ningú /res (durant uns dies). certainly CL want.1sg of not see.inf nobody /nothing during a day.pl I definitely WISH not to see anybody/anything (for a few days) b. Certament no ho VOLIA, de veure ningú /res (durant uns dies). certainly not CL want.1sg.pst of see.inf nobody /nothing during a day.pl I definitely did not WISH to see anybody for a few days (15) Catalan data for the test: Different position of the neg-marker a. (No ho desitjo,) ho PRETENC, de no menjar res (durant uns dies). not CL wish.1sg CL want.1sg of not eat.inf nothing during a. day.pl (I do not wish it,) I WANT not to eat anything (for a few days) b. No ho VULL, de menjar res durant uns dies (ho PRETENC). not CL want.1sg of eat.inf nothing during a. day.pl CL want.1sg 'I do not WISH to eat anything for a few days (I demand it)' (16) Further Catalan data: Neg-marker licenses into RD No ho necessito, de tenir cap cotxe, però m'agradaria. not CL need.1sg of have.inf no car but me.like.3sg.cond 'I don't NEED to have a car, but I would like it.' (17) Further Catalan data: Neg-marker licenses into RD (Villalba 2000:189) a. La Maria no és responsable de res. the M. not be responsible of anything Maria is not responsible for anything. 148

157 Chapter 4: Syntactic Aspects of CLLD and CLRD b. La Maria no ho és, responsable de res. the M. not CL be responsible of anything c. *Responsable de res, la Maria no ho és. responsible of anything the M. not CL be Example (16) is a further Catalan example showing that having the neg-marker in the nondislocated part is completely fine. These data lead to the conclusion that the neg-marker can license an RD constituent. Villalba (2000:189) also noticed this asymmetry between CLLD and CLRD in Catalan, cf. (17) and uses it as one argument against the mirror hypothesis. Hence, the assumption of SL (2006) that CLLD and CLRD are in the same structural position above TP cannot be maintained at least for Catalan since NPIs in the CLRD constituent are licensed. The pattern of Catalan NPI-licensing is problematic for any analysis which assumes the RD constituent to be hierarchically higher than TP Binding With respect to LF-reconstruction, Freidin (1986) points out that the following two examples behave quite differently. Whereas in (18a) reconstruction takes place and the pronoun he cannot take John as an antecedent, reconstruction in (18b) is not obligatory. John can act as the antecedent for the pronoun. (18) Complement / adjunct asymmetry a. Which claim [that John i was asleep] was he *i/j willing to discuss b. Which claim [that John i made] was he i/j willing to discuss Lebeaux (1988) traces the different behavior to a difference between complement and adjunct. In (18a) the constituent that John was asleep is a complement of claim and must appear at the level of D-structure. At this level a principle C violation arises since the r-expression John is c-commanded by the pronoun he. In (18b) the CP that John made is a relative clause and has adjunct status. Due to the fact that it is an adjunct, it does not appear at D-structure and can be adjoined after reconstruction occurs. He never c-commands John and can thus be coreferent with the r-expression. Chomsky (1995:204f.) carries Lebeaux s analysis over to a minimalist framework, keeping the difference between complement and adjunct. Cecchetto (1999:42) develops a test for the position of RD in Italian based on the asymmetry between complements and adjuncts. He comes to the conclusion that whereas CLLD 149

158 Chapter 4: Syntactic Aspects of CLLD and CLRD displays an argument-adjunct asymmetry, CLRD does not. He explains this by the clause-internal analysis of CLRD. Samek-Lodovici (2006) repeats this test and comes to a contrary result: both complement and adjunct clauses in CLLD and CLRD display the argument-adjunct asymmetry. For this reason, he assumes a clause-external analysis of CLRD, as shown in (3). SL (2006) constructs sentences in which the matrix clause involves the null subject pro and in which the dislocated object noun is followed by a CP containing a definite subject, as in (21) and (22) below. Before presenting the data, though, the idea of the test is introduced here. The CP following the dislocated object is either a complement or an adjunct (i.e. a relative clause). The clause-external analysis predicts that in the case of the complement CP, coreference between pro and the definite subject is not possible. The complement CP reconstructs and the null subject pro thus c-commands the subject s copy left behind, as shown in (19a; adopted from SL 2006:841:9). The arrow illustrates c-commanding. (19) Reconstruction of complement CP vs. late insertion of adjunct a. Complement: [ TP pro clitic aux V [det N [ CP that Subj ]]] b. Adjunct: [ TP pro clitic aux V [det N ]] In the case of the relative clause, the CP is an adjunct and can be inserted late in the derivation, i.e. after the dislocation of the object (Chomsky 1995:204f.). Since the late inserted relative clause does not reconstruct, pro never c-commands the subject of that CP. The corresponding configuration is (19b), where no copy of the CP is found after object N. The pattern with CLLD is comparable with the asymmetry shown in (19) because only the complement CP reconstructs into a position below the matrix subject, whereas the adjunct CP remains high (cf. Villalba 2000:190 for further antireconstruction effects with respect to CLLD). The clause-internal analysis predicts that there is no asymmetry between complement CPs and adjunct CPs. The right-dislocation is below TP and the matrix subject always c-commands the r-expression, independent of being reconstructed or not (20). (20) Predictions of the clause-internal analysis a. complement: [ TP pro clitic aux [det N [ CP that Subj ]] V [det N [ CP that Subj ]]] b. adjunct: [ TP pro clitic aux [det N [ CP that Subj ]] V [det N]] 150

159 Chapter 4: Syntactic Aspects of CLLD and CLRD For Italian, Samek-Lodovici (2006) shows that asymmetry exists and a clause-external analysis is to be preferred. Right-dislocated nominal complements are not possible (21), whereas rightdislocated nominal adjuncts are possible (22). (21) Italian data: Nominal complements a. *pro i non le mantiene quasi MAI, le promesse che Berlusconi i sarà onesto. (he) not them keeps almost ever, the promises that B. will-be honest Berlusconi almost NEVER keeps the promises that he will be honest. (SL 2006:841:10b) b. *pro i non le rivela certo ai GIORNALI, le prove che il (he) not them reveals certainly to-the newspapers, the evidence that the procuratore-capo di Palermo i viola la legge. chief public prosecutor of Palermo breaks the law Palermo s chief public prosecutor does not reveal the evidence that he breaks the law to the NEWSPAPERS. (SL 2006:841:11b) (22) Italian data: Nominal adjuncts a. pro i non le mantiene quasi MAI, le promesse che Berlusconi i fa in (he) not them keeps almost ever, the promises that Berlusconi makes in campagna elettorale. campaign electoral Berlusconi almost NEVER keeps the promises that he makes during the electoral campaign. (SL 2006:841:10a) b. pro i non le rivela certo ai GIORNALI, le prove che il (he) not them reveals certainly to-the newspapers, the evidence that the procuratore-capo di Palermo i trova durante un inchiesta. chief public prosecutor of Palermo finds during an.investigation Palermo s chief public prosecutor does not reveal to the NEWSPAPERS the evidence that he collects during an investigation. (SL 2006:841:11a) This findings contrast with Cecchetto (1999). SL (2006:843:fn.2) states that the odd status of Cecchetto s original example does not come from binding effects but might be caused by the quantificational nature of the indefinite subject, i.e. Cecchetto used an incorrect example. I now turn to the Catalan pattern. The following examples show that Catalan contrasts with Italian: both right-dislocated nominal complements and adjuncts are ungrammatical, (23) and (24) respectively. 151

160 Chapter 4: Syntactic Aspects of CLLD and CLRD (23) Catalan data: Nominal complements a. *pro i no les manté gairebé MAI, les promeses que (he) not CL keep almost ever, the promise.pl that el president i serà honest. the president be.fut honest The president almost never keeps the promises that he will be honest. b. *pro i no les revela pas als DIARIS, les proves que (he) not CL reveal certainly to.the newspaper.pl the evidence.pl that el president i viola la llei. the president break the law The president does not reveal the evidence that he breaks the law to the newspapers. (24) Catalan data: Nominal adjuncts a. *pro i no les manté gairebé MAI, les promeses que el president i (he) not CL keep almost ever, the promise.pl that the president fa en campanya electoral make in campaign electoral The president almost never keeps the promises that he makes during the electoral campaign. b.?*pro i no les revela pas als DIARIS, les proves que (he) not CL reveal certainly to.the newspaper.pl the evidence.pl that el president i troba durant una instrucció. the president find during an investigation The president does not reveal the evidence that he collects during an investigation to the newspapers. The data sharply contrasts with the Italian examples presented in SL (2006). The ungrammaticality of the Catalan data in (23) and (24) speaks in favor of a clause-internal analysis of CLRD, along the lines presented in (20). Villalba (1999b:244) has already shown that there is no complement-adjunct asymmetry in CLRD constituents. One conclusion can be drawn at once: Catalan CLRD is clause-internal. The second conclusion concerns the Italian results and can be stated for my purposes as follows: Italian and Catalan have to be distinguished with respect to the position of CLRD. However, Villalba (1999b) argues that LF-reconstruction is neither necessary nor adequate and that the Split-Topic Hypothesis (Villalba 2000) is able to account for the (Catalan) data. He argues that counterexamples derive from independent factors as discourse context and modality and that by assuming an independently motivated level, such as Zubizarreta s (1998) Assertion Structure, binding facts receive a proper treatment. If the Italian binding facts in examples (21) and (22) can be accounted for in terms of an independently motivated level, it might be possible to maintain the clause-internal analysis for Italian as 152

161 Chapter 4: Syntactic Aspects of CLLD and CLRD Cecchetto (1999) proposed. I am not able to deal with this topic in the current study and will leave it open for further research Obviation effects A completely new test for the structural position of CLLD and CLRD is based on obviation effects in subjunctive complement clauses. This test also argues for a clause-internal analysis of CLRD. Complement subjunctive clauses in Romance languages such as the example in (25) are subject to an obligatory disjoint reference holding between the subjunctive subject and the matrix subject. (25) Subjunctive disjoint reference a. *Jo i vull que jo/pro i vagi a veure aquesta pel.lícula. I want.1sg that I/pro go.1sg.sbj to see.inf this movie b. Jo i vull PRO i anar a veure aquesta pel.lícula. I want.1sg PRO go.inf to see.inf this movie I want to go and see this movie. CAT. Whereas the embedded subject jo/pro in (25a) cannot be coreferent with the matrix clause subject jo I, the infinitive's subject in (25b) must pick its reference from the matrix argument. The effect in (25a) is called "Subjunctive Disjoint Reference" (Kempchinsky 1987) or "Obviation" (Farkas 1992, Luján 1999, Costantini 2005a). Now, it is possible for a left-dislocation to appear between the matrix clause and the embedded clause, i.e. embedded left-dislocation (cf. Baltin 1982:19, Authier 1992:329, Lasnik & Saito 1992:76). Costantini detects an interaction between obviation and CLLD, which has not been considered so far in linguistic research. He mentions, in his thorough overview of approaches to obviation, that embedded left-dislocations may affect obviation in some cases (Costantini 2005b:129). His examples are presented in (26) and (27). 13 (26) (Costantini 2005b:129:64) a. Gianni i spera che pro?i/j abbia fatto pochi errori all esame di linguistica. G. hopes that has(subj) made few mistakes at-the exam of linguistics Gianni hopes that he has made few mistakes on the linguistics exam ITAL. b. Gianni i spera che, [all esame di linguistica], pro i/j abbia fatto pochi errori. Gianni hopes that at-the exam of linguistics has(subj) made few mistakes Gianni hopes that he has made few mistakes on the linguistics exam ITAL. 13 Cf. Costantini (2005a:ch.3.2.1) for examples where the clause with the embedded subject is itself dislocated. 153

162 Chapter 4: Syntactic Aspects of CLLD and CLRD (27) (Costantini 2005b:130:fn.19) a. En Joan no es pensa que pro *i/j hagi fet molts errors a l examen. the Joan not it(cl) thinks that has(subj) made many mistakes at the.exam John doesn t think he has made many mistakes on the exam CAT. b. En Joan no es pensa que [a l examen de lingüística computacional] The J. not it(cl) thinks that at the.exam of linguistics computational pro i/j hi hagi fet molts errors. CL has(subj) made many mistakes John doesn t think he has made many mistakes on the computational linguistics exam CAT. The normal word order of verbs taking a subjunctive complement clause is shown in the (a) examples of (26) for Italian and of (27) for Catalan. The typical obviation effect is visible: the embedded (null) subject is disjoint in reference from the matrix subject (or it tends to be disjoint as marked by? in the Italian example). In examples (b) the adjunct all esame /a l examen ( ) on the exam is locally left-dislocated (i.e. preceding the embedded clause). Since clitics in Italian are obligatory only for left-dislocated direct objects (SL 2006:847), there is no clitic of the left-dislocation in the embedded clause in (26b). Now, if embedded CLLD applies, the coreferential properties change. Obviation is affected by CLLD and disappears: coreference between the embedded null subject and the matrix subject is acceptable. In order to combine obviation and embedded CLLD, one has to pay attention to two restrictions on the matrix verb: a) for obviation to appear the matrix verb has to be a volitional/desiderative verb (Kempchinsky 1985, Farkas 1992); b) the matrix verb has to be a bridge verb in order to allow for extraction (Erteschick-Shir 1973, Authier 1992, Müller 1995). The following examples are constructed along these lines a) Semantic Class of the Matrix Verb: Kempchinsky (1987) and Farkas (1992) observe that obviation is restricted to a subclass of predicates taking subjunctive complements, namely desiderative/manipulative or volitional verbs. Verbs of doubt and denial (i) and factive-emotive verbs (ii), for example, do not trigger obviation. (i) Ana i duda que pro i/j sea la persona más apta para el puesto A. doubt.3sg that be.3sg.subj the person best suited for the job Ana doubts that s/he is the best suited person for the job (Kempchinsky 1987:126) (ii) Ana i lamenta que pro i/j tenga tanto trabajo A. regret.3sg that have.3sg.subj so-much work Ana regrets that s/he has so much work (Kempchinsky 1987:126f) b) Bridge Contexts and embedded CLLD: Embedded topicalization is possible if it is confined to bridge contexts (Müller 1995:351, Authier 1992:333f). This means that the matrix verb has to be a bridge verb, since only these verbs easily allow for extraction from their 154

163 Chapter 4: Syntactic Aspects of CLLD and CLRD I present a new set of data including CLRD now. The reason for taking CLRD into account is the question as to whether CLRD has the same consequences for obviation as CLLD. The idea is that if CLRD constituents are structurally in the same position as CLLD constituents, as assumed by the clause-external hypothesis, obviation should likewise disappear. If CLRD does not affect obviation, it is possible to conclude that they are not in the same position (an analysis for the data is presented below). In (28) and (29) the corresponding CLRD structures to (26) and (27) are represented, respectively. (28) Obviation and CLRD (ITAL.) Gianni i spera che pro?i/j abbia fatto pochi errori, [all esame di linguistica]. Gianni hope that have.sbj make.ptcp few mistake.pl at.the.exam of linguistics Gianni hopes that he has made few mistakes on the linguistics exam (29) Obviation and CLRD (CAT) En Joan i no es pensa que pro *?i/j hi 1 hagi fet molts errors, [a l examen the J. not CL think that CL have.sbj make.ptcp many mistake.pl at the.exam de lingüística computacional] 1. of linguistics computational John doesn t think he has made many mistakes on the computational linguistics exam The configurations in (28) and (29) show that right-dislocations do not affect obviation as CLLD does. The obviation is as clear as in the examples without dislocations, (26a) and (27b). Two further examples (one for Italian, (30), and one for Catalan, (31)) are given. They also show that CLRD does not influence obviation. 15 sentential complements into the left-periphery of the embedded complement or the matrix clause. Verbs such as think, believe, say, conclude, ask, tell, report, announce, know, hope fall under the class of bridge verbs, whereas verbs such as doubt, shout, worry, want, understand are non-bridge verbs. Since Erteschik-Shir s (1973) introduction of the term bridge verb, the discussion about this class of verbs is by and large concerned with the question of whether a verb will join it or not. For Erteschick-Shir it is the notion of semantic dominance on which the condition on extraction is based (Erteschick-Shir 1973:8). A clause or a phrase is semantically dominant if it is not presupposed and does not have contextual reference (Erteschik-Shir 1977:9), i.e. the speaker wants to draw attention to the semantic content of this constituent. Extraction can occur only out of clauses or phrases which can be considered dominant [...]. A matrix which is subordinate (i.e., where the embedded clause is dominant and allows extraction) will be called a bridge. Extraction out of an island is therefore possible only across a bridge. (Erteschik- Shir 1977:50). Hence, the semantic weight (Featherston 2004) is of relevance and bridge verbs are then assumed to be semantically light. The relevance of semantic weight is also detectable in Kiparsky & Kiparsky (1970), Cattell (1978), and Erteschick-Shir & Lappin (1979) (cf. Featherston 2004:183). In his work on the relation between bridge verbs and (Germanic) V2-verbs, Featherston (2004:205) comes to the conclusion that the [b]ridge feature is a continuum and not a categorical distinction: there is no absolute group of bridge verbs, only better and worse ones. 15 I would like to thank Francesco Costantini and Gemma Rigau for helping me with the data. Both mentioned independently, that the data and the binding relations are rather subtle and that it is sometimes hard to make a decision. The judgments on the Catalan data, in addition, seem to be dependent on social factors such as age and regional factors such as dialect. Nevertheless, I conclude that generally speaking it seems that the coreference is easier to get if there is a constituent in the left periphery. 155

164 Chapter 4: Syntactic Aspects of CLLD and CLRD (30) Obviation and CLLD/CLRD 16 a. Gianni i pensa che pro *i/j sia andato nel 1991 ad Amburgo G. think that have go.ptcp in 1991 to Hamburg (ma non si ricorda l'anno esatto) (but not REFL remember the.year precise) John thinks that he went to Hamburg in 1991 (but he doesn t remember the precise year) b. CLLD: Gianni i pensa che [ad Amburgo], pro i/j ci sia andato nel 1991 G. think that to Hamburg, CL have go.ptcp in 1991 (ma non si ricorda l'anno esatto) (but not REFL remember the.year precise c. CLRD: Gianni i pensa che pro *i/j ci sia andato nel 1991, [ad Amburgo], G. think that CL have go.ptcp in 1991 to Hamburg, (ma non si ricorda l'anno esatto) (but not REFL remember the.year precise (31) Obviation and CLLD/CLRD a. En Joan espera que pro *i/j no digués res mal dit the J. hope that not say.sbj nothing bad say.ptcp al congrés de la societat sociològica. at.the congress of the society sociologic John hopes that he has said nothing bad at the congress of the society of sociology. b. CLLD: En Joan espera que [al congrés de la societat sociològica] 1 the J. hope that at.the congress of the society sociologic pro i/j no hi 1 digués res mal dit. not CL say.sbj nothing bad say.ptcp c. CLRD: En Joan espera que pro *i/j no hi 1 digués res mal dit, the J. hope that not CL say.sbj nothing bad say.ptcp [al congrés de la societat sociològica] 1. at.the congress of the society sociologic As before, the data in (30) and (31) show that there is no obviation effect in the (b) examples, the one with an embedded CLLD, while coreference is not possible in the examples without any dislocation and with right-dislocation. I thus conclude that it seems that coreference is easier to 16 I owe the examples to F. Costantini (p.c.). 156

165 Chapter 4: Syntactic Aspects of CLLD and CLRD achieve if there is a constituent in the left periphery. Since CLRD does not affect obviation, it is likely that CLRD constituents are not located in the left periphery. As the examples additionally show, the pattern is similar in Catalan and Italian. Thus, the data do not support a clause-external analysis for Italian. Now it is time to implement the findings into a theoretical approach. By doing this, I show that a clause-internal analysis can easily account for the obviation data. Up until now, there is neither an approach for the CLLD pattern of Costantini (2005b) nor one which accounts for the CLRD data. For this reason, I present an analysis below. The analysis is based on Luján s (1999) approach on obviation in general. A theoretical approach to the influence of CLLD on obviation Luján (1999) proposes a binding-theoretical approach to obviation. Binding-theoretical approaches assume that the binding domain of the null subject in the embedded subjunctive clause is extended to the whole sentence. Then, according to Principle B, pro cannot be bound by the matrix subject since the latter is part of the binding domain of pro. Binding-theoretical approaches differ in the claim of which properties are responsible for the extension of the binding domain (cf. Costantini 2005b:99). Luján (1999) assumes that a) pronouns universally undergo LF-movement in order to define their reference, and b) clausal complements are marked with Case and the Spanish complementizer que that bears that abstract Case feature (Luján 1999:106). Obviative clausal complements are assumed to be included in a simple CP structure, cf. (32), whereas non-obviative clausal complements have a double CP structure. 17 The complementizer que that saturates the matrix verb s Case feature (accusative in (32)). At LF the null pronoun has to head-move to that complementizer due to interpretational properties (cf. Hestvik 1992). By LF-adjoining to que the pronoun absorbs the Case feature. The consequence of adopting the Case feature of the matrix verb is that the pronoun has to be interpreted in the domain of the main clause (Extension of the binding domain). 17 Luján s (1999) approach is founded on the observation that ECM structures and Obviation of the subject pronoun in a subjunctive complement have the same range of restrictions. Therefore, she assumes the same syntactic structure for obviative subjunctives such as ECM structures, namely a double-strata CP : [ CP [ CP, whereas non-ecm structures are only provided with a simple CP structure. Furthermore, ECM structures involve movement. 157

166 Chapter 4: Syntactic Aspects of CLLD and CLRD (32) Binding Domain Extension ACC Binding domain extension Juan quiere [ CP que+(él) k [ IP t k venga]] J. want.3sg that (he) come.3sg.subj Juan wants that he comes The extension in (32) takes place only because the sentential complement involves a simple CP structure. Non-obviative clausal complements, in contrast, are assumed to be included in a double CP structure, cf. (33). The complement s subject pronoun has, as before, to LF-adjoin to the closest C head. In double CP structures this C head is empty and it is bound by the overt operator-like complementizer que which is located in the higher CP (Luján 1999:111). The matrix verb s Case feature is located in the higher CP. The LF-adjoined pro cannot absorb the Case feature since it is unavailable in the lower CP and, thus, pro does not extend the domain of interpretation. Consequently, the matrix subject is not part of pro s binding domain and, according to Principle B, the main clause subject can act as its antecedent. (33) No Extension of the Binding Domain Case No binding domain extension Ana duda [ CP que [ CP Ø C +(pro i/j ) k [ IP t k sea la persona más apta para el puesto]]] A. doubt.3sg that be.3sg.subj the person best suited for the job Ana doubts that s/he is the best suited person for the job This approach to obviation can be modified for the CLLD and CLRD pattern. If Luján (1999) is correct and obviation is induced by the combination of Case requirements of the matrix verb and LF-adjunction of pronouns, then CLLD is supposed to interrupt the binding domain extension exemplified in (32). I assume that this happens in the following way. CLLD, as normally assumed, is located in the C-domain of the clause. I assume with Rizzi (1997) that the CP is split into the two functional projections ForceP and FinP. The complementizer is generated in FinP and then moves up to ForceP. At LF the null subject of the embedded subjunctive has to move to the C-domain and adjoins to the closest head in C, namely Fin. Since Fin and Force together represent C, the pronoun moves further up to the complementizer que. This allows for the binding domain extension in (32). I now assume that if a further functional projection appears 158

167 Chapter 4: Syntactic Aspects of CLLD and CLRD between Fin and Force, pro cannot reach que in Force. Such a functional projection is TopP, which appears only when it is needed. In the case of left-dislocated constituents, TopP is needed and this intervention impedes the extension of the binding domain: The only option for pro is to stay at the empty head Fin. Similar to the non-obviation pattern in (33), the LF-adjoined pro cannot absorb the Case feature, which is located in Force, and thus pro does not extend the domain of interpretation. Consequently, the matrix subject is not part of pro s binding domain and, according to Principle B, the main clause subject can act as its antecedent (34). (34) Case No binding domain extension Gianni i spera [ ForceP che [ TopP all esame di linguistica [ FinP Ø+(pro i/j ) k [ IP t k abbia fatto pochi errori]]]]. Gianni hopes that at-the exam of linguistics has(subj) made few mistakes Gianni hopes that he has made few mistakes on the linguistics exam As for CLRD, if a CLRD constituent were structurally in the same position as a CLLD constituent it should also hinder pro from reaching the case assigned complementizer in ForceP. The data on the interaction of obviation and CLRD has shown that this is not the case. In contrast, if one assumes a clause-internal analysis of CLRD, the right-dislocation does not impede the LF-movement of the pronoun from Fin to Force and as such does not hinder the extension of the binding domain Conclusion In Section 4.1.2, the different syntactic approaches to clitic left-dislocation (CLLD) and clitic right-dislocation (CLRD) are introduced: the mirror hypothesis (Vallduví 1993), the Split- Topic Hypothesis (Villalba 1996, 1999, 2000; but also Cecchetto 1999, and López 2003, 2008), and a version of the mirror hypothesis which assumes remnant movement in the case of CLRD (Samek-Lodovici 2006; but also Kayne 1994, Zubizarreta 1998). Section concentrated on three tests dealing with the CLLD/CLRD. The two tests by Samek-Lodovici (2006; licensing of n-words and binding effects) showed that Catalan behaves differently from Italian and does not support a clause-external analysis. The third test has not been considered so far in linguistic research and is based on the interaction of obviation and CLLD. I showed that whereas CLLD causes obviation to disappear, CLRD does not do so. By assuming that the position for CLRD is 159

168 Chapter 4: Syntactic Aspects of CLLD and CLRD clause-internal, it could account for the difference. Based on the three tests, I conclude that a clause-internal analysis for Catalan is the best choice, exactly as argued by Villalba (1996, 1999a,b, 2000) and López (2003, 2008). It is not discussed if CLLD and CLRD constituents are derived by movement or by base generation in their surface position. Due to the fact that there is no base generation approach that assumes a clause-internal position for CLRD, I adopt the analysis as presented by Villalba (2000). As a result, I also assume a movement approach to CLLD and CLRD. 160

169 Chapter 5 Prosodic Phrasing of Catalan CLLD and CLRD 1 This chapter deals with the prosodic phrasing of CLLD and CLRD in Catalan. The results of the experiment show that embedded clitic left-dislocations are typically not preceded by a prosodic phrase boundary, while clitic left-dislocations in general are obligatorily followed by a prosodic boundary. Clitic right-dislocations are obligatorily separated from the preceding main clause by a boundary. The appearance of CLLD in embedded contexts is a topic that has not been addressed by intonational research in general so far. I present experimental data that cannot be explained by Frascarelli (2000) and Prieto (2005). The relevant data concerns complex CLLD structures presented in Villalba (2000) and López (2003, 2008): left-dislocation out of clitic leftdislocations and embedded left-dislocations. The hypotheses for the experiment are based on Frascarelli (2000). Her work constitutes a logical point of departure, because she presents the only approach including CLLD and CLRD but does not consider embedded clauses. I show that the formulation of the Topic Prosodic Domain (Frascarelli 2000:63) is too restricted for the Catalan data. The stochastic OT analysis I present is based on the modified version of Prieto s (2005) analysis (cf. Chapter 3). In order to account for dislocations the constraint ALIGN-TOP,R, is also introduced, which accounts for the obligatory right boundary. This constraint does not demand a boundary to the left and thus enables a prosodic grouping with preceding material. By considering branching and non-branching dislocations in the experimental data, I am able to show that the restructuring of non-branching topics does not occur in Catalan. Nonbranching topics are immediately followed by prosodic phrase boundary even at a fast speech rate although a weakening of the boundaries takes place. The chapter is structured in the following way. In Section 5.1 background information on dislocations in several Romance languages is given. The hypotheses are presented in Section This chapter circulated previously as Feldhausen (2006), an early manuscript presenting first results. The conclusions drawn in Feldhausen (2006) and here do not differ, although the analysis is slightly different.

170 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD The experiment design is described in Section 5.3. In Section 5.4 the results of the experiment are illustrated and described. The theoretical approach to dislocations is given in Section Background Previous studies on Catalan (Bonet 1984, Recasens 1993, Prieto 2002a, Astruc 2005) agree with respect to the prosodic behavior of clitic left- and right-dislocations. CLLD constituents are given completely independent contours. Generally, they are accented and end with a continuation rise (Prieto 2002:411, Astruc 2005:61). Right-dislocations are detached from the preceding clause and have a very low pitch without any perceivable prominence. Astruc (2005:ch.3) shows that right-dislocations are indeed unaccented. 2 Prieto (2002a:410f.) highlights that the main clause of both types of dislocations has the same intonational characteristics as a neutral declarative: rising prenuclear accents followed by a falling/low nuclear accent. The nuclear accent is on the last stressed syllable and constitutes next to the low boundary tones the progressive decline until the end (cf. also Prieto et al. 2007:10). Right-dislocations in Spanish, for example, are described in Zubizarreta (1998:154ff.) as being accented, but their pitch range is subordinate to that of the main clause. As in Catalan, right-dislocations are detached from the main clause and constitute a prosodic phrase of their own, which is more likely the Intonational Phrase than the intermediate phrase. Similar patterns are found in Lambrecht (1981) for French. He says that left-dislocations are accented, while right-dislocations are deaccented. In contrast to Catalan, Lambrecht (1981) states that right-dislocations are integrated into the same prosodic unit as the main clause. Ladd (1996:141f.) states that French right-dislocations copy the last tone of the matrix sentence. In declaratives the right-dislocation is low, whereas it is high in questions. 3 Thus, the intonation of right-dislocations depends entirely on the intonation of the matrix clause. Frascarelli (2000) presents a detailed prosodic and syntactic analysis of Topic and Focus in Italian. Left- and right-dislocations in Italian are consistently separated from the main clause on the Intonational Phrase level (Frascarelli 2000:34&63). Left-dislocations are accented and their pitch accent differs according to their discourse roles (L*+H for aboutness/shifting topics; 2 Catalan right-dislocations occur not only with declarative sentences, but also with interrogatives. When occurring with interrogatives they seem to be accented and possess a contour that duplicates the contour of the nuclear accent of the main clause, but with a lower pitch (Bonet 1984:34, Recasens 1993:214). 3 This pattern is reminiscent of the duplication in Catalan as described in Bonet (1984:34) and Recasens (1993:214). 162

171 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD H* for contrastive topics; and L* for familiar topics; cf. Frascarelli & Hinterhölzl 2007). Frascarelli & Trecci (2006) note that right-dislocated familiar topics also bear the low pitch accent L* (but they are taken by the authors as being destressed ). Some important aspects of Frascarelli s (2000) analysis are presented in greater detail in the next section. All of these studies are similarly concerned with CLLD and CLRD in simple clauses (i.e. root clauses without embedded clauses). As far as I know, there has been no intonational study undertaken with respect to embedded left-dislocations. Only Frascarelli & Trecci (2006) include some embedded clauses in their study. However, they are mainly concerned with the use and position of subjects and their tonal realization and do not investigate into their phrasing pattern The Hypotheses The hypotheses for the experiment are set up based on the work by Frascarelli (2000). For this reason, the relevant features of this approach are sketched before the hypotheses are presented. After that, it is explained how the hypotheses can be tested. Two formalizations of Frascarelli s (2000) approach are important for the present work: the Topic Prosodic Domain and Topic Restructuring. The first formalization deals with the prosodic domain of topics, (1). It is useful in assisting the PF component to recognize Topic constituents and to translate them correctly into a prosodic structure. (1) Topic Prosodic Domain (Frascarelli 2000:63): A Topic is minimally and exhaustively contained in an I[ntonational Phrase, I.F.]. (2) (Topic) TP Formalization (1) says that topics must be minimally and exhaustively contained in an Intonational Phrase, i.e. they have an immediate left and right boundary. In (2) an abstract schema of (1) is given. Corresponding Italian examples with left-dislocations (left-hand topics in Frascarelli s terms) are given in (3a) and (4). It is correctly predicted by (1) that there is an intonational phrase boundary on both sides of the topic constituent in (3a) and (4). 4 The majority of the subjects in the study of Frascarelli & Trecci (2006) are realized with a low tone L* (and are thus familiar topics) and they appear in any clausal type. Aboutness-shift topic subjects, in contrast, are seldom present in embedded clauses. 163

172 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD (3) Italian Non-Branching Topics (Frascarelli 2000:48) 5 a. [[questo libro] ϕ ] I [[[k]onosco] ϕ [l autore] ϕ [che l ha scritto] ϕ ] I this book know-1sg the author that it have-3sg write-pp I know the author who wrote this book b. [[questo libro] ϕ [[h]onosco] ϕ [l autore] ϕ [che l ha scritto] ϕ ] I this book know-1sg the author that it have-3sg write-pp I know the author who wrote this book (4) Italian Branching Topics (Frascarelli 2000:47) [[gli amici] ϕ [di Sara] ϕ ] I [[[dʒ]ianni] ϕ [é partito] ϕ [senza neanche salutarli] ϕ ] I the friends of Sara Gianni be-3sg leave-pp without even to say good-byeto.them Gianni left without saying good-bye to Sara s friends In (3b) a restructuring effect can be seen. Frascarelli notes that non-branching topics, in contrast to branching topics, generally restructure into adjacent intonational phrases when the speech rate increases (Frascarelli 2000:48). For this reason, (3b) is possible. The right intonational phrase boundary of the topic disappears and the topic is part of the intonational phrase of the main clause. This process is formalized by (5). (5) Topic Restructuring (Frascarelli 2000:63) If non-branching, a Topic may restructure into the adjacent constituent, on either side. [[ ]φ ] IntP [[ ]φ [ ]φ ] IntP => [[ ]φ [ ]φ [ ]φ ] IntP Topic Sentence Topic Sentence The topic restructuring process can override the minimal and exhaustive phrasing of topics (cf. (1)). The formalization correctly predicts for Italian that non-branching dislocations can be incorporated in a preceding (or following) intonational phrase, whereas branching CLLDs do not reconstruct and thus do not phrase together with other elements. Clearly, the two formalizations work fine for CLLD and CLRD in simple clauses. However, it is worth pointing out that the left boundary of CLLD constituents comes naturally in a simple clause, due to the fact that nothing precedes the topic constituent. This can be seen in the 5 Evidence for IntP-boundaries in (Toscanian) Italian comes from spirantisation (N&V 198/2007): Plosives are realized as fricatives when situated between two vowels. The [h] in (3b) is caused by spirantisation and signals that there is no IntP-boundary preceding the word. In (3a), however, the word is preceded by such a boundary. For this reason, spirantisation cannot apply and the underlying phoneme /k/ is realized as [k]. Spirantisation is a domain span rule (N&V 1986/2007), and the target position must not be at the edge of a prosodic constituent. 164

173 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD schema in (2): There is no material preceding (Topic). Consequently, one important question arises: What happens when CLLD structures are embedded? The formulation of the Topic Prosodic Domain as given in Frascarelli (2000) can be applied to embedded CLLD even though it was originally developed for simple clauses. It predicts that embedded CLLD is also minimally and exhaustively contained in a prosodic phrase: According to (1), only the grouping in (6a) is a valid phrasing. According to (5), the groupings (6b,c) are invalid, if the topic is branching. They are valid, if the topic is non-branching. V stands for the matrix verb, stands for additional material preceding the matrix verb. (6) a.) V (Topic) YP b.) ( V Topic) YP c.) V (Topic YP) Previous studies on Catalan show that, in general, left-dislocations end with a continuation rise (Prieto 2002:411, Astruc 2005:61). I claim that this boundary is obligatory and that dislocations do not restructure. Furthermore, based on Frascarelli s (2000) formalizations and the role of embedded dislocations, I claim that Catalan dislocations do not have an obligatory left boundary. Consequently, only the groupings (6a,b) are possible. The two following hypotheses are set up: Hypothesis 1: Left-dislocations only have a right boundary and this boundary is obligatory. Hypothesis 2: Catalan left-dislocations do not restructure 6 Hypothesis 2 is derived from hypothesis 1. If the right boundary is obligatory, restructuring is not possible. The crucial test for hypothesis 1 concerns cases where embedded CLLD (branching and non-branching) phrases with preceding material of the matrix clause at normal speech rate. At a fast speech rate, branching topics should phrase with preceding material. Thus, if there are branching dislocations in data that phrase with preceding material at normal and fast speech rate one can be sure that the grouping is not a result of the restructuring process. The grouping is rather a result of the right boundary of the dislocations. For this reason, I set up data consisting of cases of embedded CLLD; these are mainly instances of so called left-dislocations out of CLLD 6 The relevant notion for restructuring is the prosodic phrase. Thus, the (SV)(O) phrasing of Prieto (2005) can be taken as a result of a restructuring process of (S)(VO). 165

174 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD (cf. below). Hypothesis 2 is based on (5). The crucial test for hypothesis 2 concerns cases with non-branching dislocations. If they do not restructure at fast speech rate, the findings can be interpreted to indicate that Catalan dislocations do not restructure. However, if they do restructure, it is of interest if branching topics also restructure. If there are instances of restructuring branching topics, (5) cannot be maintained for Catalan. A further hypothesis can be constructed from the special prediction of hypothesis 1 that left-dislocations do not behave like (preverbal) subjects. If the right boundary of left-dislocations is obligatory, no dislocation should phrase with following material. If the matrix subject is a leftdislocation, matrix (SV) should not be possible (recall that the complex SVO experiment (Chapter 3) shows that matrix (SV) phrasing increases when the object is sentential). The balance effects described in Prieto (2005) do not appear. Hypothesis 3 states as follows: Hypothesis 3: The obligatory right boundary causes that left-dislocations do not phrase with following material if the object is long (branching or sentential) The crucial test for hypothesis 3 concerns cases with non-local left-dislocations (cf. below): If left-dislocations acted like preverbal subjects, there should be instances where non-branching non-local left-dislocations phrase with the following matrix material. Corresponding examples for the three hypotheses are presented now. Branching and non-branching topics are given in (7) and (8). Non-branching topics consist of one prosodic word, such as (7). Branching topics consist of two different groups. The topics of the first group are made up of two prosodic words (8a), while the topics of the second group are made up of at least three prosodic words, (8b). (7) Non-branching topics ω Les taules i, les i vaig portar al pis. The tables CL.ACC PAST.1SG bring to-the flat I brought the tables to the flat (8) Branching topics ω ω a. Les taules i de Barcelona, les i vaig portar al pis. The tables of B. CL.ACC PAST.1SG bring to-the flat I brought the tables from Barcelona to the flat 166

175 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD ω ω ω ω ω b. Als veïns catalans de l altre costat de l Ebre, to.the neighbour.pl catalan of the.other shore of the.ebre els volen robar l aigua. CL want.3pl steal.inf the.water They want to steal the water from the Catalan neighbours from the other side of the Ebre river A left-dislocation between a matrix clause and an embedded clause is called an embedded left-dislocation (cf. Baltin 1982:19, Authier 1992:329, Lasnik & Saito 1992:76). An example is given in (9). The DP les taules the tables is left-dislocated and embedded. (9) Embedded CLLD [La Maria va dir [que les taules, les va portar al pis ] CP2 ] CP1. the M. PST.3SG say.inf that the table.pl CL.ACC PST bring.inf to.the flat Mary said that (s)he brought the tables to the flat The embedded clause (CP 2 ) is a sentential complement of the matrix verb, for which I used the assertive predicate dir say in the experiment. Dir is a weak intensional predicate generally selecting the indicative mood for its argument clause (cf. Farkas 1992, Quer 2001). Structure (9) is not the only one that can test hypothesis 1. López (2003:196, 2008:123) shows that it is possible in Catalan to left-dislocate a constituent out of an embedded CLLD constituent. 7 An example is given in (10). (10) (CL)LD extracted from embedded CLLD (adopted from López 2003:196) [ α Del seu avi] la Maria diu que [ β les històries t(α)] of.the her grandfather the Maria say that the story.pl la Joana les coneix totes t(β) the Joana CL.ACC know all.pl Maria says that Joana knows all of her grandfather s stories In a first step the complex DP [ β les històries del seu avi] in (10) is left-dislocated to the leftperiphery of the embedded clause. The accusative clitic les appears in front of the embedded verb. In a second step, the PP constituent [ α del seu avi] inside that DP is extracted and then holds 7 This is possible in spite of claims in the literature that dislocations are opaque domains for extraction. Cf. López (2008:122f.) for further information. 167

176 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD a position in the left-periphery of the matrix clause. The PP, however, is not doubled by a clitic (cf. López 2008:122f. for details). In both (9) and (10) there is material preceding the (embedded) left-dislocation which represents the foundation for testing hypothesis 1. Most of the relevant examples in the experiment for testing the hypothesis are constructed along the lines of (10). Before proceeding with the experiment description, two further terms have to be introduced: local and non-local dislocation. Local CLLD means that an element is moved to the left periphery of the same clause. This is the case for both left-dislocations in simple clause, such as (7) and (8), and embedded leftdislocations, such as (9) and (10). Instances of multiple dislocations (or iterative dislocations) as in (11) are also local. Local CLRD means that the constituent is moved to the right of the same clause, cf. (13b) below. (11) (Local) Iterative CLLD Amb en Pere 1, del llibre 2, n 2 hi 1 va parlar ahir. with the Pere of.the book CL-CL.LOC PST talk.inf yesterday (S)he talked with Pere about the book. Non-local CLLD means that an element is moved out of an embedded clause up to the leftperiphery of the matrix-clause. This is the case of the PP in (10). It is also the case of the DP les taules the tables in (12). Non-local CLRD is not possible, because CLRD is clause-bounded and therefore always local (cf. Villalba 2000:266). (12) Non-local CLLD Les taules [el Joel va dir [que les va portar al pis. CP1 ] TP the table.pl the J. PST say.inf that CL.ACC PST bring.inf to.the flat The tables, Maria said that Joel brought from home 168

177 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD 5.3 The experiment To examine the prosodic structure of CLLD and CLRD a production experiment with 12 (9+3) subjects based on scripted speech was conducted. Scripted speech is produced on the basis of written material which is read out (Gussenhoven 2004:10). The first nine subjects were recorded at normal speech rate, while the last three subjects were recorded at fast speech rate. The locus of the experiment of the first nine subjects in Table 1 was the Universitat Autònoma de Barcelona (UAB) in Bellaterra (Spain). The subjects were recorded in a quiet room and a DAT recorder was used. Among the 9 subjects 8 7 were female and 2 were male, ranging in age from years old (Ø 29.3 years old). 4 of the subjects were linguists and hence not totally naïve to the purpose of the experiment 9. The locus of the last three subjects in Table 1 was Berlin (Germany). All three subjects were female, non-linguists, and ranged in age from old (Ø 24.3 years old). All 12 subjects were native speakers of Central Catalan. speaker sex age profession origin and L1 parents L1 1_SA f 28 employee Barcelona : Cat. (Cat.) : Cat 2_PP f 39 linguist Figueres : Cat. (Cat.) : Spanish 3_AV f 24 PhD student Manresa : Cat. (Cat.) : Cat. 4_NC m 33 linguist Barcelona : Spanish (Cat. & Spanish) : Cat. 5_CA f 22 student Barcelona : Cat. (Cat.) : Spanish 6_CC f 22 student Barcelona : Cat. (Cat.) : Cat. 7_AM f 22 student Cabrianes : Cat. (Cat.) : Cat. 11_SP f 39 linguist St. Felíu de Guíxols : Cat. (Cat.) : Cat. 12_XV m 35 linguist Barcelona : Spanish (Cat. & Spanish) : Spanish YH f 21 student Barcelona : Cat. (Cat. & Spanish) : Spanish RP f 26 PhD student Barcelona : Cat. (Cat.) : Cat MM f 26 self-employed Terrassa : Cat. (Cat.) : Cat. Table 1: Detailed information of recorded subjects (numbered speakers recorded in Barcelona, June 2005; remaining three speakers recorded in Berlin, January 2008) 8 Originally, 12 speakers were recorded, but three had to be eliminated due to the bad quality of the recording and to a strongly emphatic pronunciation. 9 The results indicate that there is no difference between linguists and non-linguists. 169

178 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD Material: A total of 32 different basic sentences was used (cf. appendix). 20 of these sentences included CLLD constituents; 8 sentences included CLRD constituents, and 4 sentences included both CLLD and CLRD. Each sentence was uttered by all 12 speakers (cf. Table 1). Thus, the experiment contained 384 sentences altogether. 288 of the sentences were uttered at a normal speech rate (9 speakers x 32 sentences); 96 sentences were uttered at a fast speech rate (3 x 32). All recorded data was controlled with respect to information structure. Each target sentence was introduced by a context question. The context question purposely mentioned the dislocated constituent of the target clauses in order to guarantee that the dislocated part was given (cf. Chapter 1). An example is given in (13). The accusative object les taules the tables is explicitly mentioned in context, and then dislocated to the left and to the right in (13a,b) respectively. (13) Example for target sentences with CLLD and CLRD 10 Context (question): Què vas fer amb les taules? what PST.2SG do.inf with the table.pl What did you do with the tables? 10 In each basic sentence, a comma was placed after each CLLD constituent and before each CLRD constituent in contrast to the three experiments described in Chapter 3 (Part 1a, 1b, and 1c). The normative tradition says that for Catalan la dislocació a l esquerra CLLD can be orthographically separated by a comma from the rest of the sentence, whereas la dislocació a la dreta CLRD must be orthographically separated. This rule is unwritten but widely accepted. Yet, in texts one can find clear instances of CLRD without comma. It is well known, that this punctuation mark indicates a slight pause or a break between parts of a sentence (OAL, p.229). An important question arises from this: whether or not this orthographical symbol influences the subject of the experiment and causes him/her to produce a pause. In the experiment conducted, this seemed not to be the case. To anticipate a finding of the experiment Part C (CLLD vs. S) in Chapter 6: Speakers also produce many clear IntP breaks even when there is no punctuation mark in the stimuli. Furthermore, despite the comma in the CLLD/CLRD experiment, speakers vary between IntP- and ip-boundaries similar to the pattern of the experiments described in the third chapter. In addition, speakers are in general influenced by their orthographical education and might read the punctuation marks even when they are not there. One final note to the evolution of the comma: In the system of the ancient Greeks a small dot at the bottom of the line was used for a short pause in reading, a small dot at mid-level was used to mark a medium pause, whereas a dot at the top of the line was used to mark the end of a sentence. The comma punctuation mark at present did not evolve from this Greek dot but from a diagonal slash (/) called lat. virgula (which is the diminutive of lat. virga rod, stick ). The virgula was used from the 13 th century on and was changed into the comma around 1700 AD. Some Romance languages kept the Latin origin word (fr. virgule, it. virgola), whereas the term in Catalan and Spanish (coma), but also in German (Komma) and in English (comma) stems from the Greek word kómma blow, cut (a derivation from gr. kóptein hit, push, cut off ). (cf. dtv-ds, p.109; Kluge 2002:513) 170

179 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD Target sentences: a. CLLD: Les taules i, les i vaig portar al pis. The tables CL.ACC PAST.1SG bring to-the flat I brought the tables to the flat b. CLRD: Les i vaig portar al pis, les taules i. CL.ACC PAST.1SG bring to-the flat the tables I brought the tables to the flat In order to test the hypotheses, the material was intermittently controlled for prosodic and syntactic aspects. The prosodic and syntactic aspects may also overlap each other. 18 basic sentences are controlled for the prosodic structure of the dislocations. They include branching and non-branching topics, as illustrated in (7) and (8): 4 sentences include non-branching leftdislocations and 8 sentences include branching left-dislocations, whereas 2 sentences include non-branching right-dislocations and 4 sentences include branching right-dislocations. As for the syntactic structure of the dislocations, there are two main groups. One group includes embedded left-dislocations, the other group includes non-embedded dislocations. With respect to embedded left-dislocation, 6 basic sentences are controlled for it (two of them are simple embedded leftdislocations, as in (9); four of them are left-dislocations out of embedded left-dislocation structures, as in (10)). 11 The other sentences do not have embedded left-dislocations. They include either local or non-local dislocations, as in (13a,b) and (12) respectively. 12 Procedure: The subjects were recorded in a quiet room where they were sat down at a table in front of the recording microphone. The data was presented on printed out paper sheets. 13 The subjects were told to read out each single sentence only after they were asked the suitable context question. In addition, they were told to read out the sentences in conversational style without been given any specific instructions regarding the phrasing. The subjects in Spain were told to 11 Ultimately, only one of the simple embedded LD clauses was used, i.e. five basic clauses were used for the results. Only the matrix clause of the excluded sentence contained a raising verb and thus, was much shorter than the matrix clause of all other examples. 12 All CLRD sentences are local (i.e. 8 basic sentences). 6 CLLD sentences are local, 6 non-local (The LD out of CLLD sentences are not included). 4 sentences include multiple dislocations, as in (11) (CLLD: 2 sentences; CLRD 2 sentences). 4 sentences include left- and right-dislocations at the same time. 13 The target sentences were not randomized and were not interspersed with fillers. 171

180 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD read out the sentences at a normal speech rate, while the subjects in Berlin read them at a fast speech rate. The sentences were digitized (wav files, sample rate Hz) and F 0 tracks were analyzed using PRAAT (Boersma & Weenink 2008). Subsequently, a prosodic analysis of the spoken sentences was conducted along the lines described in Section 3.2. The phrase boundaries were determined on the basis of intonational evidence along the criteria set up in Chapter Results In this section the results of the experiment are described and illustrated by various figures and intonation contours. In 5.4.1, it is shown that embedded CLLD is typically not preceded by a boundary tone (cf. Figure 1). Then, it is shown that left-dislocations are almost always followed by an intonational boundary (i.e. ip and IntP; cf. Figure 4). After that, it is illustrated that right dislocations are immediately preceded by intonational boundaries at normal and fast speech rate (cf. Figure 5). Finally, a phonetic characteristic of right-dislocations is presented: The longer the dislocation (i.e. the more prosodic words), the more probable is a high realization of the contour with respect to the low boundary tone of the preceding main clause (cf. Figure 8). Section summarizes the findings, while the findings are discussed in depth in Section Results Figure 1 illustrates that embedded CLLD is typically not preceded by a boundary tone. The figure gives the number (in percent) of boundaries occurring between the matrix verb and the embedded CLLD constituent (including the complementizer). The left bar gives the percentage values for normal speech rate, the right bar for fast speech rate. At normal speech rate, in 64% of the cases no boundaries were placed. In 20% of the cases it is unclear if there is a boundary or not. Thus, in 84% of the cases, embedded CLLD is not preceded by a clear prosodic break. The number of clear boundaries only comes to 16% (14% ip-boundary, 2% IntP-boundary). At fast speech rate, the instances of no or unclear boundaries also represents the majority (66%), while the number of clear boundaries comes to 34%. Interestingly, the number of no or unclear 172

181 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD boundaries is smaller than at normal speech rate. Nevertheless, clear boundaries are not typical at fast speech rate either. Figure 1: Percentages of boundary types preceding embedded CLLD constituents Figure 2 and Figure 3 give the pitch contour of a sentence where a dislocation is extracted from an embedded dislocation, as in (14). The embedded dislocation is constituted by three prosodic words, i.e. it branches. Figure 2 shows the contour at fast speech rate, while Figure 3 shows the contour at normal speech rate. (14) embedded CLLD with three prosodic words ω ω ω [De Bordeus] 1 la Maria diu que [l ampolla de vi rosat t 1 ] 2 of B. the M. say that the.bottle of wine rose la 2 va comprar el Joel the PST buy.inf the J. Mary said that Joel bought the bottle of red wine from Bordeaux 173

182 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD Pitch (Hz) L+>H* H- L+>H*!H* L+>H* L+H* L+H*!H% L+H* H+L* L* L-L% De Bor- -de- -us la Ma- -ria diu que l am- -po--lla de vi ro- -sat la va com- -prar el Jo- -el LD S V COMP CLLD cl V S Figure 2: Time (s) Catalan (LD)(SVqCLLD)(VS) phrasing at fast speech rate of the sentence De Bordeus la Maria diu que l ampolla de vi rosat la va comprar el Joel Mary said that Joel bought the bottle of red wine from Bordeus of speaker YH (sentence 12b_YH) Pitch (Hz) 75 L*H H% L+>H* L+H* H* L+H* H% H*!H* H+L* L-L% De Bordeus la Maria diu que l ampolla de vi rosat la va comprar el Joel LD S VCOMP CLLD cl V S Time (s) Figure 3: Catalan (LD)(SVqCLLD)(VS) phrasing at normal speech rate of the sentence De Bordeus la Maria diu que l ampolla de vi rosat la va comprar el Joel Mary said that Joel bought the bottle of red wine from Bordeus of speaker YH (sentence bcn_pers3_12b) 174

183 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD Figure 2 and Figure 3 show that the embedded left-dislocations are not preceded by a boundary tone. There is no tonal movement signaling a boundary. The rise of F0 on the complementizer que that in both figures is a consonantal effect and cannot be attributed to any boundary tone. Thus, the local CLLD constituent is phrased together with the preceding verb. The matrix subject is also part of that prosodic group, because there is no boundary following the subject. With respect to the right boundary of dislocations, they clearly appear in both figures. The non-local dislocation De Bordeus from Bordeaux is marked by H- in Figure 2 and by H% in Figure 3. The local CLLD l ampolla de vi rosat the bottle of red wine is marked in both figures by the IntPboundary tone H%. In sum, Catalan shows a clear tendency not to place a boundary before embedded leftdislocations. The language does so, irrespective of any branchingness. The figures further indicate that Catalan has boundaries marking the end of a left-dislocation. This aspect is presented in greater detail now. Figure 4 shows that left-dislocations are almost always followed by an intonational boundary (i.e. ip and IntP) at normal and fast speech rate (left and right bar respectively). The figure gives the value in %. The figure combines branching and non-branching CLLD constituents, because their values do not differ significantly. 14 For the same reason, the figure does not distinguish between local and non-local dislocations either. 15 The number of IntPboundaries decreases significantly at fast speech rate, while the number of ip-boundaries increases. At normal speech rate 65,5% of the boundaries directly after the CLLD constituent are realized on the IntP-level, while 33,5% are realized on the ip-level. Therefore almost all leftdislocations, irrespective of their being branching or not, are followed by a clear prosodic break (i.e. 99% (65,5% + 33,5%)). At fast speech rate only 10% of the breaks are IntP-boundaries, while 69% of them are ip-boundaries. In contrast to normal speech rate, the number of no or unclear boundaries rise. They come to about one fifth (21%) of the cases. Nevertheless, nearly 80% of the left-dislocations at fast speech rate are marked by a clear right boundary. Non-local non-branching left-dislocations (cf. hypothesis 3) are always (100%) followed by ip-boundaries (at fast speech rate). 14 Percentage values of branching and non-branching topics at normal speech rate (data of multiple CLLD excluded): Branching (IntP: 63%; ip: 37%); Non-Branching (IntP: 72%; ip: 28%). 15 At normal speech rate, local non-branching dislocations have 67% IntP-boundaries and 33% ip-boundaries. Nonlocal non-branching dislocations have 78% IntP- boundaries and 22% ip-boundaries. At fast speech rate, the number of ip-boundaries of non-local non-branching dislocations sums up to 100%. 175

184 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD Figure 4: Percentage values of boundary types immediately following branching and non-branching CLLD constituents (local and non-local) Figure 2 and Figure 3, above, also illustrate the obligatory right boundary. Although Figure 4 does not include embedded CLLD, the obligatoriness of the right boundary is valid for any leftdislocation. The two figures show that there is a clear boundary after both dislocations. An ipand IntP-boundary after De Bordeus from Bordeaux and an IntP-boundary after l ampolla de vi rosat the bottle of red wine. Figure 5 indicates that right dislocations are immediately preceded by intonational boundaries at normal and fast speech rate (left and right bar respectively). 16 The figure indicates the value as a percentage. While the percentage of IntP- and ip-boundaries is more or less balanced at normal speech rate (53% vs. 47% respectively), the number of ip-boundaries increases significantly at fast speech rate (88%). There are even some unclear instances (8%). The figure combines, as before, branching and non-branching CLRD constituents. However, there is a difference between branching and non-branching constituents. While the boundaries preceding branching right-dislocations are mostly realized on the IntP-level (64%), the boundaries preceding non-branching right-dislocations are mostly ip-boundaries (67%). 16 The criterion for the CLRD-preceding boundary must be briefly explained. The boundary is located directly after the constituent bearing sentence accent and thus marks the end of the main clause. The boundary is main clause-final (but recall that the boundary cannot be located at the end of TP, since CLRD in Catalan is clause-internal (cf. Chapter 4)). The main clause-final boundary is categorized as an ip-boundary, when the sentence without the RDconstituent would be perceived as having an abbreviated, or chopped-off ending. It is categorized as an IntPboundary when the sentence without RD is perceived as not being chopped off (i.e. having a normal end). It is also categorized as an IntP-boundary when there is a pause intervening between the main clause and the right-dislocation. 176

185 Chapter 5: Prosodic Phrasing of Catalan CLLD and CLRD Figure 5: Percentage values of boundary types immediately preceding branching and non-branching CLRD constituents Figure 6 and Figure 7 present the intonation contour of corresponding examples. Both figures show that the contour of the main clause ends in a low boundary tone. While there is only an ip-boundary tone in Figure 6, there is an IntP-boundary tone preceding the branching dislocation in Figure Pitch (Hz) H* H* L+H* L- L-L% El vam comprar a Barcelona el llibre cl V PP CLRD Time (s) Figure 6: CLRD-Phrasing of dislocation with one prosodic word separated by an ip-boundary Waveform and intonation contour of the sentence El vam comprar a Barcelona, el llibre We bought the book in Barcelona of speaker 12xa (sentence bcn_pers12_14e) 177