CILC 2017 Oscar Garcia marchena

Spanish Fragments and Polar Verbless Clauses. Typology and corpus Distribution Oscar García Marchena LLF - Université Paris Diderot CILC 2017. Paris, 31 May 2017 Laboratoire de Linguistique Formelle Contents 1. Introduction 2. Typology of Fragments 3. Typology of Polar Verbless Clauses 4. Corpus distribution 5. Conclusions 2 1. Introduction 1.1. What are fragments? In corpora, we find different kinds of verbless utterances: - Non-predicative verbless utterances: • Markers (1) ¿Todo listo, verdad? (ADM 008A) ‘Everything ready, true?’ • Fillers (2) A : -¿Se siente culpable por no haber retirado el cartel o por haber sido su foto la que le dio en... B :-Mitad de la crisma? (NOT 034A) ‘A: -Do you feel guilty for not having retired the poster or because it was your picture that hit him… B: -In the head?’ • Calls (3) Oiga, buen hombre! (CON 241B) Listen, good man! • Existential verbless clauses (4) A:-¿Agustín dónde trabaja ? B: -Ni idea. (CON 006C) A:-‘Where does Agustin work?‘ B: -‘No idea.‘ • Polar verbless clauses: (5) -Sí que hay una cosa que me molesta. (CON 002A) ‘Yes that there is a thing that bothers me.‘ 3 1. Introduction 1.1. What are fragments? (2) Predicative Verbless Utterances: Head selects an argument (entity or propositional) No head ellipsis: verbless clauses - Evaluatives (entity or proposition) (6) ¡Qué maravilla de coche! (LUD 003B) What a wonder of car! (What a woderfun car!) - Epistemics (select proposition) (7) ¡Por supuesto que voy a ir a la fiesta! Of course that I am going to the party! (DEB 023B) - Illocutives (8) ¡Manos arriba! Hands up! - FRAGMENTS with elliptical verbal head • without correlate • with correlate (CON 026B) (9) A: -¿Vienes? B: -¿Ahora? (CON 217B) (10) A: -Entonces íbamos al bar. B: ¿Al bar? A: -Then we went to the bar. B: -To the bar? 4 1. Introduction 1.2. What are polar verbless clauses?  The head is a pro-clause, “sí”, “no” (11) A : -Te hicieron una foto con una anaconda, ¿no? B : -No, una boa. A: -They made you a picture with an anaconda, not? (i.e.: isn’t it?) B: -Not, with a boa. (12) A : -Sí que me suena. B : -¿Sí? (CON 006B) A: -Yes that it is familiar to me. B: -Yes? (i.e.: Really?) • Positive polarity (yes) or negative polarity (no) • Not elliptic • They are anaphoric to a propositional content • No elliptical head  verbless clause 5 1. Introduction 1.3. Aim. Why studying the variety of fragments and polar clauses? • Fragments are clauses with an elliptical verbal head • Verbal clauses are units with different possible syntactic configurations  In Spanish, they have a structure (SN) SV: subject-head, head only, headcomplement, head-adjunct, head-peripheric  They can also be classified by their syntactic type (declarative, exclamative, desiderative, interrogative)  each syntactic type is associated to a kind of semantic content (propositonal, outcome, ...) (Marandin, 2008) 6 1. Introduction 1.3. Aim. Why studying the variety of fragments and polar clauses? The fragment is interpreted by its association with preceding clauses (13) A: -Tu fumas? B-Desde los catorce años.  “Yo fumo desde los catorce años” A: -‘Do you smoke?’ B: – Since I am 14. ‘I smoke since I am 14’ The fragment may recover the content of the source and adds it to its own content (adapting deictic forms) But this is not the only way fragments recover their information: (14) A: -Tú fumas mucho? B:-Solo los fines de semana. A: -’Do you smoke a lot?’ B: -’Only on weekends.’  Interpreted content is not “I smoke a lot but only on weekends", but “I smoke only on weekends"  fragments have a second way of recovering content. The phrase composing the fragment is interpreted as replacing a part of the source ("mucho") → its correlate  The fragment recovers the content of the source except for the correlate 7 1. Introduction 1.3. Aim. Why study the variety of fragments and polar clauses? It is necessary to know which fragments have this property of coreference and which ones do not, because this property induces a difference in their interpretation. Furthermore, it is interesting because classifying types can allow us to predict the speech act they realize: answer, precision (12), correction (13), etc. (15) A: -Ha venido la novia de Luis. B: -Su exnovia. A: -’Luis’ girlfriend came.’ B: -’Her ex-girlfriend.’ As for polar clauses, having the coreference property, they can also perform different speech acts, like agreement (14) or acknowledgement (15): (16) A: - Vienes? B: –Sí. (17) Cuanta gente! –Sí A: -’Are you coming?’ B: -’Yes.’ A: -’What a lot of people.’ B: -’Yes.’ 8 1. Introduction 1.4. Our corpus We use the spoken Spanish CORLEC (Corpus oral de Referencia de la Lengua Española Contemporánea. (Marcos Marín 1992) • • • • • 1 078 780 words 63 291 utterances Available on line at http://www.lllf.uam.es/ESP/Corlec.html Transcribed with punctuation Distributed in 17 genres: religion, law, television…. → that we classify as either monologic or dialogic • We extracted 2 275 verbless clauses and 5 159 fradgments (Marchena 2015) 9 2. Typology of Fragments 2.1. Existing classifications : Schlangen, 2000 Schlangen 2000’s types -illocutionary acts and textual relationships: Question-answer Elaboration Correction & Contrast Continuation Explanation Plan-related relations Comment Narration A: -Who came to the party ? B: -Peter. A: -I talked to Peter. B: -Peter Miller ? A: -Are they in the cupboard ? B: -No, in the fridge. A: -I am free on Monday. And on Wednesday. A: -Peter left early. Exams. A: -Let’s meet on Monday. At two o’clock. A: -I talked to Peter. Awesome ! A: -He went to Italy. And then to Spain. •Types with and without a correlate are mixted in Schlangen's types •Types with a correlate: question-answer, elaboration, correction & contrast •Types without a correlate: continuation, explanation, plan-related relations, narration 10 2. Typology of Fragments 2.2 Existing classifications prior to this work: Fernandez and Ginzburg, 2002 Mixed criteria: speech act (answer), composition (repeated), syntactic (sluice, modifier), semantic (factual) 11 2. Typology of Fragments 2.3. Classification of fragments in the CORLEC Corpus (Garcia Marchena, 2015) Classifying fragments according to syntactic properties allows to draw 2 conclusions: 1. There is a limited number of speech acts that they can realize  Covering Preceding typologies Speech acts realized by fragments with correlates: Answer Agreement A: -When did he come? B: -Today. A: -¿Cuándo vino? B: -Hoy. A: -¿Se fue con María? B: -Con María. A: -Did he go wuth Maria? B: -With Maria. Acknowledgement A: -¿Se fue con Pedro? B: -¡Con Pedro! Check question A: -¿Se fue con Pedro? B: -¿Con Pedro? Correction Precision A: -Did he go with Pedro? B: -With P! A: -Did he go with Pedro? B: -With P? A: -¿Se fue con Pedro? B: -Con Laura. A: -Did he go with Pedro? B: -With L. A: -¿Se fue con Pedro? B: -Con Pedro y con María. A: -Did he go with Pedro? B: -With Pedro and with Maria. Precision question A: -¿Se fue con Pedro? B: -¿El hermano de Luis? ¿Cándo? A: -Did he go with Pedro? B: -Luis’s brother ? When? Fragments without correlates constitute a continuation of the content of the source. This “continuation” can be refined for coherence relations 12  Schlangen 2000’s: continuation, explanation, plan-related relations, narration  Semantic distinctions 2. Typology of Fragments 2.3. Classification of fragments in the CORLEC Corpus (Garcia Marchena, 2015) 2. the speech act is determined by syntactic and semantic properties: • Syntactic type • Illocutive value (asserting, questionning) • Identity of referent Source Interrogative Quest decla Assert Decla • • • • • Answer Agreement Acknowledgement Check question Correction asserting target = referent ≠ referent answer agreement correction acknowledgement A: -¿Cuándo vino? B: -Hoy. A: -¿Se fue con María? B: -Con María. A: -¿Se fue con Pedro? B: -¡Con Pedro! A: -¿Se fue con Pedro? B: -¿Con Pedro? A: -¿Se fue con Pedro? B: -Con Laura. questionning target = referent ≠ referent answer + incertainty check question correction A: -When did he come? B: -Today. A: -Did he go with Maria? B: -With Maria. A: Did he go with Pedro? B: -With Pedro! A: -Did he go with Pedro? B: -With Pedro? A: -Did he go with Pedro? B: -With Laura. 13 2. Typology of Fragments 2.3. Classification of fragments in the CORLEC Corpus (Garcia Marchena, 2015) Speech acts realised by fragments can be defined as a set of synsem properties Fragments subtypes are predictable Source value Speech act Assert decla Quest Decla Answer Agreement Acknowledg + Check Q Correction + + Interr + Target value Assert Quest + + + + + Referent Same Diff + + + + 14 2. Typology of Fragments 2.3. Classification of fragments in the CORLEC Corpus (Garcia Marchena, 2015) Other subtypes, such as Precision and imPrecision, result from semantic relationships between source and target -Target is a hyponyme of source : Precision A : -¿Se fue con un coche? B : -Con un Alfa Romeo. A: -’Did he leave with a car?’ B: -’With an Alfa Romeo.’ -Target is a hiperonym of source  Imprecision /Partial negative Answer (Krifka 2001) A: -Se fue con María ? B : -Con una chica. A: -Did he go with Maria? B:-With a girl. 15 3. Typology of Polar Clauses asserting target questionning target = polarity ≠ polarity = polarity ≠ polarity Questionning agreement diss / ackno Asserting acknow diss / ack check Q check Q Source Agreement: A : -¿Nos vamos a casa ? B : -Sí. A: -’Are we going home?’ B: -’Yes’. A : -¿No nos vamos a casa ? B : -No. A : -¿No nos vamos a casa ? B : -No Dissension: Dissension or acknowledgement: Acknowledgement: A : -¿Nos vamos a casa ? B : -No. A : -¿No nos vamos a casa ? B : -Sí. A : -¿No se fueron a casa ? B : -No. A : -Se fueron a casa. B : -Sí. Dissension or acknowlegement: A : -No se fueron a casa. B : -Sí. Check question: Acknowledgement question: A : -¿No se fueron a casa ? B : -No? A: -¿No se fueron a casa ? B : -¿Sí ? A : -Vino Maria. B : -¿Sí? A : -¿No se fueron a casa ? B : -¿No ? A: -Vino Maria, ¿no? 16 3. Typology of Polar Clauses Properties of subtypes of polar clauses Speech act Agreement Dissension Acknowledg Correction Check Q Acknow Q Source value Assert Quest + + + + + + Target value Assert Quest + + + + + + Polarity Possitive Negative Polarity Same Diff + + + + + + 17 3. Typology of polar clauses Conclusions to the typology: -Fragments and polar clauses can be classified according to their syntactic and coreference properties. -Previous classifications can be subsumed by this one -The speech act they realize is determined by the syntactic properties of the source and the target of the ellipsis  The speech act of fragments can be predicted 18 4. Corpus distribution Frequencies in corpus of spoken Spanish CORLEC (Corpus oral de Referencia de la Lengua Española Contemporánea. (Marcos Marín 1992) • • • • • 1 078 780 words 63 291 utterances We have extracted 3729 fragments and 445 polar clauses Verbless clause length average : 8,75 words / utterance Fragment length average : 4,47 words / utterance • Classification in 17 genres (monologic or dialogic) 19 4.1 fragments distribution Frequencies in corpus CORLEC: • Fragments with and without correlates Monologic Dialogic TOTAL Without correlate 169 1144 1313 With correlate 275 2141 2416 TOTAL 444 3285 3729 • 7,13% of the (46 068) utterances in dialogic genres are fragments • 2,5% of the (17 223) utterances in monologic genres are fragments  Fragments are more frequent in dialogic genres, but are also present in dialogic ones  Fragments with correlates are more frequent than without • The proportion of fragments in the corpus is: • Without correlates: 35,21 % of the fragments • With correlates: 63,79 % 20 4.1 fragments distribution Subtypes of fragments by genre GENRE Dialogic Administrative Sport Publicity Debate High School Games Interviews Informal S-total S-moyenne Monologic Religion Instructions Documentary University Science Law Politics Technique News S-total S-moyenne TOTAL Moyenne Genre No correlate Correlate Amin Dep Pub Deb Edu Lud Ent Conv 48 17 26 129 149 108 180 487 1144 143 40 40 95 121 343 309 212 981 2141 267,63 Rel Ins Doc Hum Cie Jur Pol Tec Not 8 6 12 11 25 20 16 50 21 169 18,78 1313 77,24 15 5 17 31 89 32 20 46 20 275 30,56 2416 142,12 21 4. 1 Fragments In some genres, fragments without correlates are slightly more frequent than fragments with correlates (news, technique, debate, administrative)  In the rest, they are less frequent. The difference is often small (politics, law, documentary, instruction, religión, interviews) mainly monologic  But in some genres the difference is big (informal conversation, games, high school, publicity)  mainly dialogic 22 4.2 Fragments with correlates Subtypes by genre Text type Answer Administrative 17 Sport 22 Publicity 31 Debate 22 High School 129 Games 132 Interviews 69 Informal 237 Religion 4 Instructions 0 Documentary1 University 4 Science 14 Law 8 Politics 5 Technique 14 News 8 TOTAL 717 % de PARTIE 27,39% Agreem 8 1 6 10 15 20 18 125 1 1 2 0 20 4 1 5 1 238 9,09% Correct 0 0 3 8 19 11 2 51 0 0 2 0 10 4 0 1 2 113 4,32% Acknow 0 1 18 14 40 29 12 60 1 0 3 2 1 9 2 9 1 202 7,72% Check Q 2 0 0 4 11 5 6 72 0 0 0 0 1 1 0 0 0 102 3,90% Preciss 4 14 39 51 116 89 64 322 7 4 8 26 36 11 13 14 5 823 31,44% Precis Q 9 3 16 26 53 52 53 174 3 0 4 1 8 4 1 12 4 423 16,16% total 40 41 113 135 383 338 224 1041 16 5 20 33 90 41 22 55 21 2618 100,00% 23 4.2 fragments with correlates Subtypes by genre 24 4.2 fragments with correlates Subtypes by genre ► Fragments with correlates are much more frequent in dialogic than in monologic genres ► The most comment types are precisions and answers (around 30%), followed by precision questions (16%) ► The rarest in the corpus are check questions and corrections. Most of them are found in informal conversation ► Informal conversation, being the biggest part of the corpus, concentrates the highest frequencies ► Answers are frequent in high school, in games and in interviews ► Agreements and corrections are more common in the high school genre, and there is also a significant proportion in the monologic science genre ► Acknowledgements are frequent in the genres high school and games ► Precision and precision questions are more frequent in high school, games and interviews ► These 3 genres concentrate the greatest frequencies of fragments with correlates ► They are rare in some dialogic genres (administrative and sport) and also in the monologic genres, they are particularly rare in instructions and religion ► Among the monologic genres, they are more frequent in the genres science, technique, law and university 25 4. 2 Polar clauses Distribution by genres Genre Dialogic Administrative Sport Publicity Debate High School Games Interviews Informal S-total S-moyenne Monologic Religion Instructions Documentary University Science Law Politics Technique News S-total S-moyenne TOTAL Moyenne Agreement Dissen Check question Ack Q TOTAL si no no ¿si? ¿no? ¿no? 0 3 3 17 21 14 63 64 185 23,13 0 1 0 1 2 0 2 10 16 2,00 3 4 0 10 6 0 18 56 97 12,13 0 1 2 1 10 2 5 13 34 4,25 1 1 0 2 2 0 1 6 13 1,63 0 0 0 3 6 1 3 8 21 2,63 4 10 5 34 47 17 92 157 366 45,75 0 22 3 3 1 1 2 2 2 36 4,00 221 13,00 0 1 1 1 0 0 0 0 0 3 0,33 19 1,12 0 25 1 1 0 0 2 1 0 30 3,33 127 7,47 0 5 0 0 1 0 1 0 0 7 0,78 41 2,41 0 0 0 0 0 0 0 0 0 0 0,00 13 0,76 0 3 0 0 0 0 0 0 0 3 0,33 24 1,41 0 56 5 5 2 1 5 3 2 79 8,78 445 26,18 26 4. 2 Polar clauses 27 4.2 Polar clauses  Polar clauses are more common in dialogic genres (average of 45/genre, vs. 9/genre in monologic (366 items vs. 79)  Agreements are the most common type (240 items), followed by dissent (127)  Questionning polar clauses are less frequent (54 check questions vs. 24 acknowledgement questions)  Among monologic genres, they are frequent in instructions and there is none in religión  Among dialogic ones, appart from informal conversation, they are frequent in interviews, high school and debate 28 5. Conclusions Main conclusions to the typology:  Preceding classifications can be subsumed by this one Fragments can be classified according to their syntactic and coreference properties. The speech act they realize is determined by the syntactic properties of the source and the target of the ellipsis  The speech act of fragments can be predicted 29 5. Conclusions Main conclusions to the corpus distribution:  Fragments with and without correlate tend to be used more often in dialogic genres, specially in informal conversation  In monologic genres, they are used in similar proportions  In 2 dialogic genres (administration and debate) there is a slight preference for fragments without a correlate Precisions and Answers are the most common types with a correlate They are more frequent in high school, games and interviews genres Polar clauses are more common in dialogic genres They are mostly agreements and dissents In monologic genres, they are frequent in instructions In monologic genres, they are frequent in interviews, high school and debate 30 Thank you [email protected] Laboratoire de Linguistique Formelle 31